diff --git a/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_vnops_os.c b/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_vnops_os.c index 2b45f7ae3b30..c0d68ea0532f 100644 --- a/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_vnops_os.c +++ b/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_vnops_os.c @@ -1,6218 +1,6217 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2012, 2015 by Delphix. All rights reserved. * Copyright (c) 2014 Integros [integros.com] * Copyright 2017 Nexenta Systems, Inc. */ /* Portions Copyright 2007 Jeremy Teo */ /* Portions Copyright 2010 Robert Milkowski */ #include #include #include #include #include #include #include #include #include #include #if __FreeBSD_version >= 1300102 #include #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifndef VN_OPEN_INVFS #define VN_OPEN_INVFS 0x0 #endif VFS_SMR_DECLARE; #if __FreeBSD_version >= 1300047 #define vm_page_wire_lock(pp) #define vm_page_wire_unlock(pp) #else #define vm_page_wire_lock(pp) vm_page_lock(pp) #define vm_page_wire_unlock(pp) vm_page_unlock(pp) #endif #ifdef DEBUG_VFS_LOCKS #define VNCHECKREF(vp) \ VNASSERT((vp)->v_holdcnt > 0 && (vp)->v_usecount > 0, vp, \ ("%s: wrong ref counts", __func__)); #else #define VNCHECKREF(vp) #endif /* * Programming rules. * * Each vnode op performs some logical unit of work. To do this, the ZPL must * properly lock its in-core state, create a DMU transaction, do the work, * record this work in the intent log (ZIL), commit the DMU transaction, * and wait for the intent log to commit if it is a synchronous operation. * Moreover, the vnode ops must work in both normal and log replay context. * The ordering of events is important to avoid deadlocks and references * to freed memory. The example below illustrates the following Big Rules: * * (1) A check must be made in each zfs thread for a mounted file system. * This is done avoiding races using ZFS_ENTER(zfsvfs). * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros * can return EIO from the calling function. * * (2) VN_RELE() should always be the last thing except for zil_commit() * (if necessary) and ZFS_EXIT(). This is for 3 reasons: * First, if it's the last reference, the vnode/znode * can be freed, so the zp may point to freed memory. Second, the last * reference will call zfs_zinactive(), which may induce a lot of work -- * pushing cached pages (which acquires range locks) and syncing out * cached atime changes. Third, zfs_zinactive() may require a new tx, * which could deadlock the system if you were already holding one. * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC(). * * (3) All range locks must be grabbed before calling dmu_tx_assign(), * as they can span dmu_tx_assign() calls. * * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to * dmu_tx_assign(). This is critical because we don't want to block * while holding locks. * * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This * reduces lock contention and CPU usage when we must wait (note that if * throughput is constrained by the storage, nearly every transaction * must wait). * * Note, in particular, that if a lock is sometimes acquired before * the tx assigns, and sometimes after (e.g. z_lock), then failing * to use a non-blocking assign can deadlock the system. The scenario: * * Thread A has grabbed a lock before calling dmu_tx_assign(). * Thread B is in an already-assigned tx, and blocks for this lock. * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open() * forever, because the previous txg can't quiesce until B's tx commits. * * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT, * then drop all locks, call dmu_tx_wait(), and try again. On subsequent * calls to dmu_tx_assign(), pass TXG_NOTHROTTLE in addition to TXG_NOWAIT, * to indicate that this operation has already called dmu_tx_wait(). * This will ensure that we don't retry forever, waiting a short bit * each time. * * (5) If the operation succeeded, generate the intent log entry for it * before dropping locks. This ensures that the ordering of events * in the intent log matches the order in which they actually occurred. * During ZIL replay the zfs_log_* functions will update the sequence * number to indicate the zil transaction has replayed. * * (6) At the end of each vnode op, the DMU tx must always commit, * regardless of whether there were any errors. * * (7) After dropping all locks, invoke zil_commit(zilog, foid) * to ensure that synchronous semantics are provided when necessary. * * In general, this is how things should be ordered in each vnode op: * * ZFS_ENTER(zfsvfs); // exit if unmounted * top: * zfs_dirent_lookup(&dl, ...) // lock directory entry (may VN_HOLD()) * rw_enter(...); // grab any other locks you need * tx = dmu_tx_create(...); // get DMU tx * dmu_tx_hold_*(); // hold each object you might modify * error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT); * if (error) { * rw_exit(...); // drop locks * zfs_dirent_unlock(dl); // unlock directory entry * VN_RELE(...); // release held vnodes * if (error == ERESTART) { * waited = B_TRUE; * dmu_tx_wait(tx); * dmu_tx_abort(tx); * goto top; * } * dmu_tx_abort(tx); // abort DMU tx * ZFS_EXIT(zfsvfs); // finished in zfs * return (error); // really out of space * } * error = do_real_work(); // do whatever this VOP does * if (error == 0) * zfs_log_*(...); // on success, make ZIL entry * dmu_tx_commit(tx); // commit DMU tx -- error or not * rw_exit(...); // drop locks * zfs_dirent_unlock(dl); // unlock directory entry * VN_RELE(...); // release held vnodes * zil_commit(zilog, foid); // synchronous when necessary * ZFS_EXIT(zfsvfs); // finished in zfs * return (error); // done, report error */ /* ARGSUSED */ static int zfs_open(vnode_t **vpp, int flag, cred_t *cr) { znode_t *zp = VTOZ(*vpp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) && ((flag & FAPPEND) == 0)) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EPERM)); } /* Keep a count of the synchronous opens in the znode */ if (flag & (FSYNC | FDSYNC)) atomic_inc_32(&zp->z_sync_cnt); ZFS_EXIT(zfsvfs); return (0); } /* ARGSUSED */ static int zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr) { znode_t *zp = VTOZ(vp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); /* Decrement the synchronous opens in the znode */ if ((flag & (FSYNC | FDSYNC)) && (count == 1)) atomic_dec_32(&zp->z_sync_cnt); ZFS_EXIT(zfsvfs); return (0); } /* ARGSUSED */ static int zfs_ioctl(vnode_t *vp, ulong_t com, intptr_t data, int flag, cred_t *cred, int *rvalp) { loff_t off; int error; switch (com) { case _FIOFFS: { return (0); /* * The following two ioctls are used by bfu. Faking out, * necessary to avoid bfu errors. */ } case _FIOGDIO: case _FIOSDIO: { return (0); } case F_SEEK_DATA: case F_SEEK_HOLE: { off = *(offset_t *)data; /* offset parameter is in/out */ error = zfs_holey(VTOZ(vp), com, &off); if (error) return (error); *(offset_t *)data = off; return (0); } } return (SET_ERROR(ENOTTY)); } static vm_page_t page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes) { vm_object_t obj; vm_page_t pp; int64_t end; /* * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE * aligned boundaries, if the range is not aligned. As a result a * DEV_BSIZE subrange with partially dirty data may get marked as clean. * It may happen that all DEV_BSIZE subranges are marked clean and thus * the whole page would be considered clean despite have some * dirty data. * For this reason we should shrink the range to DEV_BSIZE aligned * boundaries before calling vm_page_clear_dirty. */ end = rounddown2(off + nbytes, DEV_BSIZE); off = roundup2(off, DEV_BSIZE); nbytes = end - off; obj = vp->v_object; zfs_vmobject_assert_wlocked_12(obj); #if __FreeBSD_version < 1300050 for (;;) { if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL && pp->valid) { if (vm_page_xbusied(pp)) { /* * Reference the page before unlocking and * sleeping so that the page daemon is less * likely to reclaim it. */ vm_page_reference(pp); vm_page_lock(pp); zfs_vmobject_wunlock(obj); vm_page_busy_sleep(pp, "zfsmwb", true); zfs_vmobject_wlock(obj); continue; } vm_page_sbusy(pp); } else if (pp != NULL) { ASSERT(!pp->valid); pp = NULL; } if (pp != NULL) { ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); vm_object_pip_add(obj, 1); pmap_remove_write(pp); if (nbytes != 0) vm_page_clear_dirty(pp, off, nbytes); } break; } #else vm_page_grab_valid_unlocked(&pp, obj, OFF_TO_IDX(start), VM_ALLOC_NOCREAT | VM_ALLOC_SBUSY | VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY); if (pp != NULL) { ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); vm_object_pip_add(obj, 1); pmap_remove_write(pp); if (nbytes != 0) vm_page_clear_dirty(pp, off, nbytes); } #endif return (pp); } static void page_unbusy(vm_page_t pp) { vm_page_sunbusy(pp); #if __FreeBSD_version >= 1300041 vm_object_pip_wakeup(pp->object); #else vm_object_pip_subtract(pp->object, 1); #endif } #if __FreeBSD_version > 1300051 static vm_page_t page_hold(vnode_t *vp, int64_t start) { vm_object_t obj; vm_page_t m; obj = vp->v_object; vm_page_grab_valid_unlocked(&m, obj, OFF_TO_IDX(start), VM_ALLOC_NOCREAT | VM_ALLOC_WIRED | VM_ALLOC_IGN_SBUSY | VM_ALLOC_NOBUSY); return (m); } #else static vm_page_t page_hold(vnode_t *vp, int64_t start) { vm_object_t obj; vm_page_t pp; obj = vp->v_object; zfs_vmobject_assert_wlocked(obj); for (;;) { if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL && pp->valid) { if (vm_page_xbusied(pp)) { /* * Reference the page before unlocking and * sleeping so that the page daemon is less * likely to reclaim it. */ vm_page_reference(pp); vm_page_lock(pp); zfs_vmobject_wunlock(obj); vm_page_busy_sleep(pp, "zfsmwb", true); zfs_vmobject_wlock(obj); continue; } ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); vm_page_wire_lock(pp); vm_page_hold(pp); vm_page_wire_unlock(pp); } else pp = NULL; break; } return (pp); } #endif static void page_unhold(vm_page_t pp) { vm_page_wire_lock(pp); #if __FreeBSD_version >= 1300035 vm_page_unwire(pp, PQ_ACTIVE); #else vm_page_unhold(pp); #endif vm_page_wire_unlock(pp); } /* * When a file is memory mapped, we must keep the IO data synchronized * between the DMU cache and the memory mapped pages. What this means: * * On Write: If we find a memory mapped page, we write to *both* * the page and the dmu buffer. */ void update_pages(znode_t *zp, int64_t start, int len, objset_t *os) { vm_object_t obj; struct sf_buf *sf; vnode_t *vp = ZTOV(zp); caddr_t va; int off; ASSERT3P(vp->v_mount, !=, NULL); obj = vp->v_object; ASSERT3P(obj, !=, NULL); off = start & PAGEOFFSET; zfs_vmobject_wlock_12(obj); #if __FreeBSD_version >= 1300041 vm_object_pip_add(obj, 1); #endif for (start &= PAGEMASK; len > 0; start += PAGESIZE) { vm_page_t pp; int nbytes = imin(PAGESIZE - off, len); if ((pp = page_busy(vp, start, off, nbytes)) != NULL) { zfs_vmobject_wunlock_12(obj); va = zfs_map_page(pp, &sf); (void) dmu_read(os, zp->z_id, start + off, nbytes, va + off, DMU_READ_PREFETCH); zfs_unmap_page(sf); zfs_vmobject_wlock_12(obj); page_unbusy(pp); } len -= nbytes; off = 0; } #if __FreeBSD_version >= 1300041 vm_object_pip_wakeup(obj); #else vm_object_pip_wakeupn(obj, 0); #endif zfs_vmobject_wunlock_12(obj); } /* * Read with UIO_NOCOPY flag means that sendfile(2) requests * ZFS to populate a range of page cache pages with data. * * NOTE: this function could be optimized to pre-allocate * all pages in advance, drain exclusive busy on all of them, * map them into contiguous KVA region and populate them * in one single dmu_read() call. */ int mappedread_sf(znode_t *zp, int nbytes, zfs_uio_t *uio) { vnode_t *vp = ZTOV(zp); objset_t *os = zp->z_zfsvfs->z_os; struct sf_buf *sf; vm_object_t obj; vm_page_t pp; int64_t start; caddr_t va; int len = nbytes; int error = 0; ASSERT3U(zfs_uio_segflg(uio), ==, UIO_NOCOPY); ASSERT3P(vp->v_mount, !=, NULL); obj = vp->v_object; ASSERT3P(obj, !=, NULL); ASSERT0(zfs_uio_offset(uio) & PAGEOFFSET); zfs_vmobject_wlock_12(obj); for (start = zfs_uio_offset(uio); len > 0; start += PAGESIZE) { int bytes = MIN(PAGESIZE, len); pp = vm_page_grab_unlocked(obj, OFF_TO_IDX(start), VM_ALLOC_SBUSY | VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY); if (vm_page_none_valid(pp)) { zfs_vmobject_wunlock_12(obj); va = zfs_map_page(pp, &sf); error = dmu_read(os, zp->z_id, start, bytes, va, DMU_READ_PREFETCH); if (bytes != PAGESIZE && error == 0) bzero(va + bytes, PAGESIZE - bytes); zfs_unmap_page(sf); zfs_vmobject_wlock_12(obj); #if __FreeBSD_version >= 1300081 if (error == 0) { vm_page_valid(pp); vm_page_activate(pp); vm_page_do_sunbusy(pp); } else { zfs_vmobject_wlock(obj); if (!vm_page_wired(pp) && pp->valid == 0 && vm_page_busy_tryupgrade(pp)) vm_page_free(pp); else vm_page_sunbusy(pp); zfs_vmobject_wunlock(obj); } #else vm_page_do_sunbusy(pp); vm_page_lock(pp); if (error) { if (pp->wire_count == 0 && pp->valid == 0 && !vm_page_busied(pp)) vm_page_free(pp); } else { pp->valid = VM_PAGE_BITS_ALL; vm_page_activate(pp); } vm_page_unlock(pp); #endif } else { ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); vm_page_do_sunbusy(pp); } if (error) break; zfs_uio_advance(uio, bytes); len -= bytes; } zfs_vmobject_wunlock_12(obj); return (error); } /* * When a file is memory mapped, we must keep the IO data synchronized * between the DMU cache and the memory mapped pages. What this means: * * On Read: We "read" preferentially from memory mapped pages, * else we default from the dmu buffer. * * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when * the file is memory mapped. */ int mappedread(znode_t *zp, int nbytes, zfs_uio_t *uio) { vnode_t *vp = ZTOV(zp); vm_object_t obj; int64_t start; int len = nbytes; int off; int error = 0; ASSERT3P(vp->v_mount, !=, NULL); obj = vp->v_object; ASSERT3P(obj, !=, NULL); start = zfs_uio_offset(uio); off = start & PAGEOFFSET; zfs_vmobject_wlock_12(obj); for (start &= PAGEMASK; len > 0; start += PAGESIZE) { vm_page_t pp; uint64_t bytes = MIN(PAGESIZE - off, len); if ((pp = page_hold(vp, start))) { struct sf_buf *sf; caddr_t va; zfs_vmobject_wunlock_12(obj); va = zfs_map_page(pp, &sf); error = vn_io_fault_uiomove(va + off, bytes, GET_UIO_STRUCT(uio)); zfs_unmap_page(sf); zfs_vmobject_wlock_12(obj); page_unhold(pp); } else { zfs_vmobject_wunlock_12(obj); error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl), uio, bytes); zfs_vmobject_wlock_12(obj); } len -= bytes; off = 0; if (error) break; } zfs_vmobject_wunlock_12(obj); return (error); } int zfs_write_simple(znode_t *zp, const void *data, size_t len, loff_t pos, size_t *presid) { int error = 0; ssize_t resid; error = vn_rdwr(UIO_WRITE, ZTOV(zp), __DECONST(void *, data), len, pos, UIO_SYSSPACE, IO_SYNC, kcred, NOCRED, &resid, curthread); if (error) { return (SET_ERROR(error)); } else if (presid == NULL) { if (resid != 0) { error = SET_ERROR(EIO); } } else { *presid = resid; } return (error); } void zfs_zrele_async(znode_t *zp) { vnode_t *vp = ZTOV(zp); objset_t *os = ITOZSB(vp)->z_os; VN_RELE_ASYNC(vp, dsl_pool_zrele_taskq(dmu_objset_pool(os))); } static int zfs_dd_callback(struct mount *mp, void *arg, int lkflags, struct vnode **vpp) { int error; *vpp = arg; error = vn_lock(*vpp, lkflags); if (error != 0) vrele(*vpp); return (error); } static int zfs_lookup_lock(vnode_t *dvp, vnode_t *vp, const char *name, int lkflags) { znode_t *zdp = VTOZ(dvp); zfsvfs_t *zfsvfs __unused = zdp->z_zfsvfs; int error; int ltype; if (zfsvfs->z_replay == B_FALSE) ASSERT_VOP_LOCKED(dvp, __func__); if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) { ASSERT3P(dvp, ==, vp); vref(dvp); ltype = lkflags & LK_TYPE_MASK; if (ltype != VOP_ISLOCKED(dvp)) { if (ltype == LK_EXCLUSIVE) vn_lock(dvp, LK_UPGRADE | LK_RETRY); else /* if (ltype == LK_SHARED) */ vn_lock(dvp, LK_DOWNGRADE | LK_RETRY); /* * Relock for the "." case could leave us with * reclaimed vnode. */ if (VN_IS_DOOMED(dvp)) { vrele(dvp); return (SET_ERROR(ENOENT)); } } return (0); } else if (name[0] == '.' && name[1] == '.' && name[2] == 0) { /* * Note that in this case, dvp is the child vnode, and we * are looking up the parent vnode - exactly reverse from * normal operation. Unlocking dvp requires some rather * tricky unlock/relock dance to prevent mp from being freed; * use vn_vget_ino_gen() which takes care of all that. * * XXX Note that there is a time window when both vnodes are * unlocked. It is possible, although highly unlikely, that * during that window the parent-child relationship between * the vnodes may change, for example, get reversed. * In that case we would have a wrong lock order for the vnodes. * All other filesystems seem to ignore this problem, so we * do the same here. * A potential solution could be implemented as follows: * - using LK_NOWAIT when locking the second vnode and retrying * if necessary * - checking that the parent-child relationship still holds * after locking both vnodes and retrying if it doesn't */ error = vn_vget_ino_gen(dvp, zfs_dd_callback, vp, lkflags, &vp); return (error); } else { error = vn_lock(vp, lkflags); if (error != 0) vrele(vp); return (error); } } /* * Lookup an entry in a directory, or an extended attribute directory. * If it exists, return a held vnode reference for it. * * IN: dvp - vnode of directory to search. * nm - name of entry to lookup. * pnp - full pathname to lookup [UNUSED]. * flags - LOOKUP_XATTR set if looking for an attribute. * rdir - root directory vnode [UNUSED]. * cr - credentials of caller. * ct - caller context * * OUT: vpp - vnode of located entry, NULL if not found. * * RETURN: 0 on success, error code on failure. * * Timestamps: * NA */ /* ARGSUSED */ static int zfs_lookup(vnode_t *dvp, const char *nm, vnode_t **vpp, struct componentname *cnp, int nameiop, cred_t *cr, kthread_t *td, int flags, boolean_t cached) { znode_t *zdp = VTOZ(dvp); znode_t *zp; zfsvfs_t *zfsvfs = zdp->z_zfsvfs; #if __FreeBSD_version > 1300124 seqc_t dvp_seqc; #endif int error = 0; /* * Fast path lookup, however we must skip DNLC lookup * for case folding or normalizing lookups because the * DNLC code only stores the passed in name. This means * creating 'a' and removing 'A' on a case insensitive * file system would work, but DNLC still thinks 'a' * exists and won't let you create it again on the next * pass through fast path. */ if (!(flags & LOOKUP_XATTR)) { if (dvp->v_type != VDIR) { return (SET_ERROR(ENOTDIR)); } else if (zdp->z_sa_hdl == NULL) { return (SET_ERROR(EIO)); } } DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, const char *, nm); ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zdp); #if __FreeBSD_version > 1300124 dvp_seqc = vn_seqc_read_notmodify(dvp); #endif *vpp = NULL; if (flags & LOOKUP_XATTR) { /* * If the xattr property is off, refuse the lookup request. */ if (!(zfsvfs->z_flags & ZSB_XATTR)) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EOPNOTSUPP)); } /* * We don't allow recursive attributes.. * Maybe someday we will. */ if (zdp->z_pflags & ZFS_XATTR) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EINVAL)); } if ((error = zfs_get_xattrdir(VTOZ(dvp), &zp, cr, flags))) { ZFS_EXIT(zfsvfs); return (error); } *vpp = ZTOV(zp); /* * Do we have permission to get into attribute directory? */ error = zfs_zaccess(zp, ACE_EXECUTE, 0, B_FALSE, cr); if (error) { vrele(ZTOV(zp)); } ZFS_EXIT(zfsvfs); return (error); } /* * Check accessibility of directory if we're not coming in via * VOP_CACHEDLOOKUP. */ if (!cached) { #ifdef NOEXECCHECK if ((cnp->cn_flags & NOEXECCHECK) != 0) { cnp->cn_flags &= ~NOEXECCHECK; } else #endif if ((error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr))) { ZFS_EXIT(zfsvfs); return (error); } } if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EILSEQ)); } /* * First handle the special cases. */ if ((cnp->cn_flags & ISDOTDOT) != 0) { /* * If we are a snapshot mounted under .zfs, return * the vp for the snapshot directory. */ if (zdp->z_id == zfsvfs->z_root && zfsvfs->z_parent != zfsvfs) { struct componentname cn; vnode_t *zfsctl_vp; int ltype; ZFS_EXIT(zfsvfs); ltype = VOP_ISLOCKED(dvp); VOP_UNLOCK1(dvp); error = zfsctl_root(zfsvfs->z_parent, LK_SHARED, &zfsctl_vp); if (error == 0) { cn.cn_nameptr = "snapshot"; cn.cn_namelen = strlen(cn.cn_nameptr); cn.cn_nameiop = cnp->cn_nameiop; cn.cn_flags = cnp->cn_flags & ~ISDOTDOT; cn.cn_lkflags = cnp->cn_lkflags; error = VOP_LOOKUP(zfsctl_vp, vpp, &cn); vput(zfsctl_vp); } vn_lock(dvp, ltype | LK_RETRY); return (error); } } if (zfs_has_ctldir(zdp) && strcmp(nm, ZFS_CTLDIR_NAME) == 0) { ZFS_EXIT(zfsvfs); if ((cnp->cn_flags & ISLASTCN) != 0 && nameiop != LOOKUP) return (SET_ERROR(ENOTSUP)); error = zfsctl_root(zfsvfs, cnp->cn_lkflags, vpp); return (error); } /* * The loop is retry the lookup if the parent-child relationship * changes during the dot-dot locking complexities. */ for (;;) { uint64_t parent; error = zfs_dirlook(zdp, nm, &zp); if (error == 0) *vpp = ZTOV(zp); ZFS_EXIT(zfsvfs); if (error != 0) break; error = zfs_lookup_lock(dvp, *vpp, nm, cnp->cn_lkflags); if (error != 0) { /* * If we've got a locking error, then the vnode * got reclaimed because of a force unmount. * We never enter doomed vnodes into the name cache. */ *vpp = NULL; return (error); } if ((cnp->cn_flags & ISDOTDOT) == 0) break; ZFS_ENTER(zfsvfs); if (zdp->z_sa_hdl == NULL) { error = SET_ERROR(EIO); } else { error = sa_lookup(zdp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent)); } if (error != 0) { ZFS_EXIT(zfsvfs); vput(ZTOV(zp)); break; } if (zp->z_id == parent) { ZFS_EXIT(zfsvfs); break; } vput(ZTOV(zp)); } if (error != 0) *vpp = NULL; /* Translate errors and add SAVENAME when needed. */ if (cnp->cn_flags & ISLASTCN) { switch (nameiop) { case CREATE: case RENAME: if (error == ENOENT) { error = EJUSTRETURN; cnp->cn_flags |= SAVENAME; break; } fallthrough; case DELETE: if (error == 0) cnp->cn_flags |= SAVENAME; break; } } #if __FreeBSD_version > 1300124 if ((cnp->cn_flags & ISDOTDOT) != 0) { /* * FIXME: zfs_lookup_lock relocks vnodes and does nothing to * handle races. In particular different callers may end up * with different vnodes and will try to add conflicting * entries to the namecache. * * While finding different result may be acceptable in face * of concurrent modification, adding conflicting entries * trips over an assert in the namecache. * * Ultimately let an entry through once everything settles. */ if (!vn_seqc_consistent(dvp, dvp_seqc)) { cnp->cn_flags &= ~MAKEENTRY; } } #endif /* Insert name into cache (as non-existent) if appropriate. */ if (zfsvfs->z_use_namecache && !zfsvfs->z_replay && error == ENOENT && (cnp->cn_flags & MAKEENTRY) != 0) cache_enter(dvp, NULL, cnp); /* Insert name into cache if appropriate. */ if (zfsvfs->z_use_namecache && !zfsvfs->z_replay && error == 0 && (cnp->cn_flags & MAKEENTRY)) { if (!(cnp->cn_flags & ISLASTCN) || (nameiop != DELETE && nameiop != RENAME)) { cache_enter(dvp, *vpp, cnp); } } return (error); } /* * Attempt to create a new entry in a directory. If the entry * already exists, truncate the file if permissible, else return * an error. Return the vp of the created or trunc'd file. * * IN: dvp - vnode of directory to put new file entry in. * name - name of new file entry. * vap - attributes of new file. * excl - flag indicating exclusive or non-exclusive mode. * mode - mode to open file with. * cr - credentials of caller. * flag - large file flag [UNUSED]. * ct - caller context * vsecp - ACL to be set * * OUT: vpp - vnode of created or trunc'd entry. * * RETURN: 0 on success, error code on failure. * * Timestamps: * dvp - ctime|mtime updated if new entry created * vp - ctime|mtime always, atime if new */ /* ARGSUSED */ int zfs_create(znode_t *dzp, const char *name, vattr_t *vap, int excl, int mode, znode_t **zpp, cred_t *cr, int flag, vsecattr_t *vsecp) { znode_t *zp; zfsvfs_t *zfsvfs = dzp->z_zfsvfs; zilog_t *zilog; objset_t *os; dmu_tx_t *tx; int error; ksid_t *ksid; uid_t uid; gid_t gid = crgetgid(cr); uint64_t projid = ZFS_DEFAULT_PROJID; zfs_acl_ids_t acl_ids; boolean_t fuid_dirtied; uint64_t txtype; #ifdef DEBUG_VFS_LOCKS vnode_t *dvp = ZTOV(dzp); #endif /* * If we have an ephemeral id, ACL, or XVATTR then * make sure file system is at proper version */ ksid = crgetsid(cr, KSID_OWNER); if (ksid) uid = ksid_getid(ksid); else uid = crgetuid(cr); if (zfsvfs->z_use_fuids == B_FALSE && (vsecp || (vap->va_mask & AT_XVATTR) || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) return (SET_ERROR(EINVAL)); ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(dzp); os = zfsvfs->z_os; zilog = zfsvfs->z_log; if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EILSEQ)); } if (vap->va_mask & AT_XVATTR) { if ((error = secpolicy_xvattr(ZTOV(dzp), (xvattr_t *)vap, crgetuid(cr), cr, vap->va_type)) != 0) { ZFS_EXIT(zfsvfs); return (error); } } *zpp = NULL; if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr)) vap->va_mode &= ~S_ISVTX; error = zfs_dirent_lookup(dzp, name, &zp, ZNEW); if (error) { ZFS_EXIT(zfsvfs); return (error); } ASSERT3P(zp, ==, NULL); /* * Create a new file object and update the directory * to reference it. */ if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) { goto out; } /* * We only support the creation of regular files in * extended attribute directories. */ if ((dzp->z_pflags & ZFS_XATTR) && (vap->va_type != VREG)) { error = SET_ERROR(EINVAL); goto out; } if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, vsecp, &acl_ids)) != 0) goto out; if (S_ISREG(vap->va_mode) || S_ISDIR(vap->va_mode)) projid = zfs_inherit_projid(dzp); if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, projid)) { zfs_acl_ids_free(&acl_ids); error = SET_ERROR(EDQUOT); goto out; } getnewvnode_reserve_(); tx = dmu_tx_create(os); dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + ZFS_SA_BASE_ATTR_SIZE); fuid_dirtied = zfsvfs->z_fuid_dirty; if (fuid_dirtied) zfs_fuid_txhold(zfsvfs, tx); dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, acl_ids.z_aclp->z_acl_bytes); } error = dmu_tx_assign(tx, TXG_WAIT); if (error) { zfs_acl_ids_free(&acl_ids); dmu_tx_abort(tx); getnewvnode_drop_reserve(); ZFS_EXIT(zfsvfs); return (error); } zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); if (fuid_dirtied) zfs_fuid_sync(zfsvfs, tx); (void) zfs_link_create(dzp, name, zp, tx, ZNEW); txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap); zfs_log_create(zilog, tx, txtype, dzp, zp, name, vsecp, acl_ids.z_fuidp, vap); zfs_acl_ids_free(&acl_ids); dmu_tx_commit(tx); getnewvnode_drop_reserve(); out: VNCHECKREF(dvp); if (error == 0) { *zpp = zp; } if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) zil_commit(zilog, 0); ZFS_EXIT(zfsvfs); return (error); } /* * Remove an entry from a directory. * * IN: dvp - vnode of directory to remove entry from. * name - name of entry to remove. * cr - credentials of caller. * ct - caller context * flags - case flags * * RETURN: 0 on success, error code on failure. * * Timestamps: * dvp - ctime|mtime * vp - ctime (if nlink > 0) */ /*ARGSUSED*/ static int zfs_remove_(vnode_t *dvp, vnode_t *vp, const char *name, cred_t *cr) { znode_t *dzp = VTOZ(dvp); znode_t *zp; znode_t *xzp; zfsvfs_t *zfsvfs = dzp->z_zfsvfs; zilog_t *zilog; uint64_t xattr_obj; uint64_t obj = 0; dmu_tx_t *tx; boolean_t unlinked; uint64_t txtype; int error; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(dzp); zp = VTOZ(vp); ZFS_VERIFY_ZP(zp); zilog = zfsvfs->z_log; xattr_obj = 0; xzp = NULL; if ((error = zfs_zaccess_delete(dzp, zp, cr))) { goto out; } /* * Need to use rmdir for removing directories. */ if (vp->v_type == VDIR) { error = SET_ERROR(EPERM); goto out; } vnevent_remove(vp, dvp, name, ct); obj = zp->z_id; /* are there any extended attributes? */ error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &xattr_obj, sizeof (xattr_obj)); if (error == 0 && xattr_obj) { error = zfs_zget(zfsvfs, xattr_obj, &xzp); ASSERT0(error); } /* * We may delete the znode now, or we may put it in the unlinked set; * it depends on whether we're the last link, and on whether there are * other holds on the vnode. So we dmu_tx_hold() the right things to * allow for either case. */ tx = dmu_tx_create(zfsvfs->z_os); dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); zfs_sa_upgrade_txholds(tx, zp); zfs_sa_upgrade_txholds(tx, dzp); if (xzp) { dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE); } /* charge as an update -- would be nice not to charge at all */ dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); /* * Mark this transaction as typically resulting in a net free of space */ dmu_tx_mark_netfree(tx); error = dmu_tx_assign(tx, TXG_WAIT); if (error) { dmu_tx_abort(tx); ZFS_EXIT(zfsvfs); return (error); } /* * Remove the directory entry. */ error = zfs_link_destroy(dzp, name, zp, tx, ZEXISTS, &unlinked); if (error) { dmu_tx_commit(tx); goto out; } if (unlinked) { zfs_unlinked_add(zp, tx); vp->v_vflag |= VV_NOSYNC; } /* XXX check changes to linux vnops */ txtype = TX_REMOVE; zfs_log_remove(zilog, tx, txtype, dzp, name, obj, unlinked); dmu_tx_commit(tx); out: if (xzp) vrele(ZTOV(xzp)); if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) zil_commit(zilog, 0); ZFS_EXIT(zfsvfs); return (error); } static int zfs_lookup_internal(znode_t *dzp, const char *name, vnode_t **vpp, struct componentname *cnp, int nameiop) { zfsvfs_t *zfsvfs = dzp->z_zfsvfs; int error; cnp->cn_nameptr = __DECONST(char *, name); cnp->cn_namelen = strlen(name); cnp->cn_nameiop = nameiop; cnp->cn_flags = ISLASTCN | SAVENAME; cnp->cn_lkflags = LK_EXCLUSIVE | LK_RETRY; cnp->cn_cred = kcred; - cnp->cn_thread = curthread; if (zfsvfs->z_use_namecache && !zfsvfs->z_replay) { struct vop_lookup_args a; a.a_gen.a_desc = &vop_lookup_desc; a.a_dvp = ZTOV(dzp); a.a_vpp = vpp; a.a_cnp = cnp; error = vfs_cache_lookup(&a); } else { error = zfs_lookup(ZTOV(dzp), name, vpp, cnp, nameiop, kcred, curthread, 0, B_FALSE); } #ifdef ZFS_DEBUG if (error) { printf("got error %d on name %s on op %d\n", error, name, nameiop); kdb_backtrace(); } #endif return (error); } int zfs_remove(znode_t *dzp, const char *name, cred_t *cr, int flags) { vnode_t *vp; int error; struct componentname cn; if ((error = zfs_lookup_internal(dzp, name, &vp, &cn, DELETE))) return (error); error = zfs_remove_(ZTOV(dzp), vp, name, cr); vput(vp); return (error); } /* * Create a new directory and insert it into dvp using the name * provided. Return a pointer to the inserted directory. * * IN: dvp - vnode of directory to add subdir to. * dirname - name of new directory. * vap - attributes of new directory. * cr - credentials of caller. * ct - caller context * flags - case flags * vsecp - ACL to be set * * OUT: vpp - vnode of created directory. * * RETURN: 0 on success, error code on failure. * * Timestamps: * dvp - ctime|mtime updated * vp - ctime|mtime|atime updated */ /*ARGSUSED*/ int zfs_mkdir(znode_t *dzp, const char *dirname, vattr_t *vap, znode_t **zpp, cred_t *cr, int flags, vsecattr_t *vsecp) { znode_t *zp; zfsvfs_t *zfsvfs = dzp->z_zfsvfs; zilog_t *zilog; uint64_t txtype; dmu_tx_t *tx; int error; ksid_t *ksid; uid_t uid; gid_t gid = crgetgid(cr); zfs_acl_ids_t acl_ids; boolean_t fuid_dirtied; ASSERT3U(vap->va_type, ==, VDIR); /* * If we have an ephemeral id, ACL, or XVATTR then * make sure file system is at proper version */ ksid = crgetsid(cr, KSID_OWNER); if (ksid) uid = ksid_getid(ksid); else uid = crgetuid(cr); if (zfsvfs->z_use_fuids == B_FALSE && ((vap->va_mask & AT_XVATTR) || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) return (SET_ERROR(EINVAL)); ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(dzp); zilog = zfsvfs->z_log; if (dzp->z_pflags & ZFS_XATTR) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EINVAL)); } if (zfsvfs->z_utf8 && u8_validate(dirname, strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EILSEQ)); } if (vap->va_mask & AT_XVATTR) { if ((error = secpolicy_xvattr(ZTOV(dzp), (xvattr_t *)vap, crgetuid(cr), cr, vap->va_type)) != 0) { ZFS_EXIT(zfsvfs); return (error); } } if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, NULL, &acl_ids)) != 0) { ZFS_EXIT(zfsvfs); return (error); } /* * First make sure the new directory doesn't exist. * * Existence is checked first to make sure we don't return * EACCES instead of EEXIST which can cause some applications * to fail. */ *zpp = NULL; if ((error = zfs_dirent_lookup(dzp, dirname, &zp, ZNEW))) { zfs_acl_ids_free(&acl_ids); ZFS_EXIT(zfsvfs); return (error); } ASSERT3P(zp, ==, NULL); if ((error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr))) { zfs_acl_ids_free(&acl_ids); ZFS_EXIT(zfsvfs); return (error); } if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, zfs_inherit_projid(dzp))) { zfs_acl_ids_free(&acl_ids); ZFS_EXIT(zfsvfs); return (SET_ERROR(EDQUOT)); } /* * Add a new entry to the directory. */ getnewvnode_reserve_(); tx = dmu_tx_create(zfsvfs->z_os); dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); fuid_dirtied = zfsvfs->z_fuid_dirty; if (fuid_dirtied) zfs_fuid_txhold(zfsvfs, tx); if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, acl_ids.z_aclp->z_acl_bytes); } dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + ZFS_SA_BASE_ATTR_SIZE); error = dmu_tx_assign(tx, TXG_WAIT); if (error) { zfs_acl_ids_free(&acl_ids); dmu_tx_abort(tx); getnewvnode_drop_reserve(); ZFS_EXIT(zfsvfs); return (error); } /* * Create new node. */ zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); if (fuid_dirtied) zfs_fuid_sync(zfsvfs, tx); /* * Now put new name in parent dir. */ (void) zfs_link_create(dzp, dirname, zp, tx, ZNEW); *zpp = zp; txtype = zfs_log_create_txtype(Z_DIR, NULL, vap); zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, NULL, acl_ids.z_fuidp, vap); zfs_acl_ids_free(&acl_ids); dmu_tx_commit(tx); getnewvnode_drop_reserve(); if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) zil_commit(zilog, 0); ZFS_EXIT(zfsvfs); return (0); } #if __FreeBSD_version < 1300124 static void cache_vop_rmdir(struct vnode *dvp, struct vnode *vp) { cache_purge(dvp); cache_purge(vp); } #endif /* * Remove a directory subdir entry. If the current working * directory is the same as the subdir to be removed, the * remove will fail. * * IN: dvp - vnode of directory to remove from. * name - name of directory to be removed. * cwd - vnode of current working directory. * cr - credentials of caller. * ct - caller context * flags - case flags * * RETURN: 0 on success, error code on failure. * * Timestamps: * dvp - ctime|mtime updated */ /*ARGSUSED*/ static int zfs_rmdir_(vnode_t *dvp, vnode_t *vp, const char *name, cred_t *cr) { znode_t *dzp = VTOZ(dvp); znode_t *zp = VTOZ(vp); zfsvfs_t *zfsvfs = dzp->z_zfsvfs; zilog_t *zilog; dmu_tx_t *tx; int error; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(dzp); ZFS_VERIFY_ZP(zp); zilog = zfsvfs->z_log; if ((error = zfs_zaccess_delete(dzp, zp, cr))) { goto out; } if (vp->v_type != VDIR) { error = SET_ERROR(ENOTDIR); goto out; } vnevent_rmdir(vp, dvp, name, ct); tx = dmu_tx_create(zfsvfs->z_os); dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); zfs_sa_upgrade_txholds(tx, zp); zfs_sa_upgrade_txholds(tx, dzp); dmu_tx_mark_netfree(tx); error = dmu_tx_assign(tx, TXG_WAIT); if (error) { dmu_tx_abort(tx); ZFS_EXIT(zfsvfs); return (error); } error = zfs_link_destroy(dzp, name, zp, tx, ZEXISTS, NULL); if (error == 0) { uint64_t txtype = TX_RMDIR; zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT, B_FALSE); } dmu_tx_commit(tx); cache_vop_rmdir(dvp, vp); out: if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) zil_commit(zilog, 0); ZFS_EXIT(zfsvfs); return (error); } int zfs_rmdir(znode_t *dzp, const char *name, znode_t *cwd, cred_t *cr, int flags) { struct componentname cn; vnode_t *vp; int error; if ((error = zfs_lookup_internal(dzp, name, &vp, &cn, DELETE))) return (error); error = zfs_rmdir_(ZTOV(dzp), vp, name, cr); vput(vp); return (error); } /* * Read as many directory entries as will fit into the provided * buffer from the given directory cursor position (specified in * the uio structure). * * IN: vp - vnode of directory to read. * uio - structure supplying read location, range info, * and return buffer. * cr - credentials of caller. * ct - caller context * flags - case flags * * OUT: uio - updated offset and range, buffer filled. * eofp - set to true if end-of-file detected. * * RETURN: 0 on success, error code on failure. * * Timestamps: * vp - atime updated * * Note that the low 4 bits of the cookie returned by zap is always zero. * This allows us to use the low range for "special" directory entries: * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, * we use the offset 2 for the '.zfs' directory. */ /* ARGSUSED */ static int zfs_readdir(vnode_t *vp, zfs_uio_t *uio, cred_t *cr, int *eofp, int *ncookies, ulong_t **cookies) { znode_t *zp = VTOZ(vp); iovec_t *iovp; edirent_t *eodp; dirent64_t *odp; zfsvfs_t *zfsvfs = zp->z_zfsvfs; objset_t *os; caddr_t outbuf; size_t bufsize; zap_cursor_t zc; zap_attribute_t zap; uint_t bytes_wanted; uint64_t offset; /* must be unsigned; checks for < 1 */ uint64_t parent; int local_eof; int outcount; int error; uint8_t prefetch; boolean_t check_sysattrs; uint8_t type; int ncooks; ulong_t *cooks = NULL; int flags = 0; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0) { ZFS_EXIT(zfsvfs); return (error); } /* * If we are not given an eof variable, * use a local one. */ if (eofp == NULL) eofp = &local_eof; /* * Check for valid iov_len. */ if (GET_UIO_STRUCT(uio)->uio_iov->iov_len <= 0) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EINVAL)); } /* * Quit if directory has been removed (posix) */ if ((*eofp = zp->z_unlinked) != 0) { ZFS_EXIT(zfsvfs); return (0); } error = 0; os = zfsvfs->z_os; offset = zfs_uio_offset(uio); prefetch = zp->z_zn_prefetch; /* * Initialize the iterator cursor. */ if (offset <= 3) { /* * Start iteration from the beginning of the directory. */ zap_cursor_init(&zc, os, zp->z_id); } else { /* * The offset is a serialized cursor. */ zap_cursor_init_serialized(&zc, os, zp->z_id, offset); } /* * Get space to change directory entries into fs independent format. */ iovp = GET_UIO_STRUCT(uio)->uio_iov; bytes_wanted = iovp->iov_len; if (zfs_uio_segflg(uio) != UIO_SYSSPACE || zfs_uio_iovcnt(uio) != 1) { bufsize = bytes_wanted; outbuf = kmem_alloc(bufsize, KM_SLEEP); odp = (struct dirent64 *)outbuf; } else { bufsize = bytes_wanted; outbuf = NULL; odp = (struct dirent64 *)iovp->iov_base; } eodp = (struct edirent *)odp; if (ncookies != NULL) { /* * Minimum entry size is dirent size and 1 byte for a file name. */ ncooks = zfs_uio_resid(uio) / (sizeof (struct dirent) - sizeof (((struct dirent *)NULL)->d_name) + 1); cooks = malloc(ncooks * sizeof (ulong_t), M_TEMP, M_WAITOK); *cookies = cooks; *ncookies = ncooks; } /* * If this VFS supports the system attribute view interface; and * we're looking at an extended attribute directory; and we care * about normalization conflicts on this vfs; then we must check * for normalization conflicts with the sysattr name space. */ #ifdef TODO check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm && (flags & V_RDDIR_ENTFLAGS); #else check_sysattrs = 0; #endif /* * Transform to file-system independent format */ outcount = 0; while (outcount < bytes_wanted) { ino64_t objnum; ushort_t reclen; off64_t *next = NULL; /* * Special case `.', `..', and `.zfs'. */ if (offset == 0) { (void) strcpy(zap.za_name, "."); zap.za_normalization_conflict = 0; objnum = zp->z_id; type = DT_DIR; } else if (offset == 1) { (void) strcpy(zap.za_name, ".."); zap.za_normalization_conflict = 0; objnum = parent; type = DT_DIR; } else if (offset == 2 && zfs_show_ctldir(zp)) { (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); zap.za_normalization_conflict = 0; objnum = ZFSCTL_INO_ROOT; type = DT_DIR; } else { /* * Grab next entry. */ if ((error = zap_cursor_retrieve(&zc, &zap))) { if ((*eofp = (error == ENOENT)) != 0) break; else goto update; } if (zap.za_integer_length != 8 || zap.za_num_integers != 1) { cmn_err(CE_WARN, "zap_readdir: bad directory " "entry, obj = %lld, offset = %lld\n", (u_longlong_t)zp->z_id, (u_longlong_t)offset); error = SET_ERROR(ENXIO); goto update; } objnum = ZFS_DIRENT_OBJ(zap.za_first_integer); /* * MacOS X can extract the object type here such as: * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer); */ type = ZFS_DIRENT_TYPE(zap.za_first_integer); if (check_sysattrs && !zap.za_normalization_conflict) { #ifdef TODO zap.za_normalization_conflict = xattr_sysattr_casechk(zap.za_name); #else panic("%s:%u: TODO", __func__, __LINE__); #endif } } if (flags & V_RDDIR_ACCFILTER) { /* * If we have no access at all, don't include * this entry in the returned information */ znode_t *ezp; if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0) goto skip_entry; if (!zfs_has_access(ezp, cr)) { vrele(ZTOV(ezp)); goto skip_entry; } vrele(ZTOV(ezp)); } if (flags & V_RDDIR_ENTFLAGS) reclen = EDIRENT_RECLEN(strlen(zap.za_name)); else reclen = DIRENT64_RECLEN(strlen(zap.za_name)); /* * Will this entry fit in the buffer? */ if (outcount + reclen > bufsize) { /* * Did we manage to fit anything in the buffer? */ if (!outcount) { error = SET_ERROR(EINVAL); goto update; } break; } if (flags & V_RDDIR_ENTFLAGS) { /* * Add extended flag entry: */ eodp->ed_ino = objnum; eodp->ed_reclen = reclen; /* NOTE: ed_off is the offset for the *next* entry */ next = &(eodp->ed_off); eodp->ed_eflags = zap.za_normalization_conflict ? ED_CASE_CONFLICT : 0; (void) strncpy(eodp->ed_name, zap.za_name, EDIRENT_NAMELEN(reclen)); eodp = (edirent_t *)((intptr_t)eodp + reclen); } else { /* * Add normal entry: */ odp->d_ino = objnum; odp->d_reclen = reclen; odp->d_namlen = strlen(zap.za_name); /* NOTE: d_off is the offset for the *next* entry. */ next = &odp->d_off; strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1); odp->d_type = type; dirent_terminate(odp); odp = (dirent64_t *)((intptr_t)odp + reclen); } outcount += reclen; ASSERT3S(outcount, <=, bufsize); /* Prefetch znode */ if (prefetch) dmu_prefetch(os, objnum, 0, 0, 0, ZIO_PRIORITY_SYNC_READ); skip_entry: /* * Move to the next entry, fill in the previous offset. */ if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { zap_cursor_advance(&zc); offset = zap_cursor_serialize(&zc); } else { offset += 1; } /* Fill the offset right after advancing the cursor. */ if (next != NULL) *next = offset; if (cooks != NULL) { *cooks++ = offset; ncooks--; KASSERT(ncooks >= 0, ("ncookies=%d", ncooks)); } } zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ /* Subtract unused cookies */ if (ncookies != NULL) *ncookies -= ncooks; if (zfs_uio_segflg(uio) == UIO_SYSSPACE && zfs_uio_iovcnt(uio) == 1) { iovp->iov_base += outcount; iovp->iov_len -= outcount; zfs_uio_resid(uio) -= outcount; } else if ((error = zfs_uiomove(outbuf, (long)outcount, UIO_READ, uio))) { /* * Reset the pointer. */ offset = zfs_uio_offset(uio); } update: zap_cursor_fini(&zc); if (zfs_uio_segflg(uio) != UIO_SYSSPACE || zfs_uio_iovcnt(uio) != 1) kmem_free(outbuf, bufsize); if (error == ENOENT) error = 0; ZFS_ACCESSTIME_STAMP(zfsvfs, zp); zfs_uio_setoffset(uio, offset); ZFS_EXIT(zfsvfs); if (error != 0 && cookies != NULL) { free(*cookies, M_TEMP); *cookies = NULL; *ncookies = 0; } return (error); } /* * Get the requested file attributes and place them in the provided * vattr structure. * * IN: vp - vnode of file. * vap - va_mask identifies requested attributes. * If AT_XVATTR set, then optional attrs are requested * flags - ATTR_NOACLCHECK (CIFS server context) * cr - credentials of caller. * * OUT: vap - attribute values. * * RETURN: 0 (always succeeds). */ /* ARGSUSED */ static int zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr) { znode_t *zp = VTOZ(vp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; int error = 0; uint32_t blksize; u_longlong_t nblocks; uint64_t mtime[2], ctime[2], crtime[2], rdev; xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ xoptattr_t *xoap = NULL; boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; sa_bulk_attr_t bulk[4]; int count = 0; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid); SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16); if (vp->v_type == VBLK || vp->v_type == VCHR) SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL, &rdev, 8); if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) { ZFS_EXIT(zfsvfs); return (error); } /* * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. * Also, if we are the owner don't bother, since owner should * always be allowed to read basic attributes of file. */ if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) && (vap->va_uid != crgetuid(cr))) { if ((error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0, skipaclchk, cr))) { ZFS_EXIT(zfsvfs); return (error); } } /* * Return all attributes. It's cheaper to provide the answer * than to determine whether we were asked the question. */ vap->va_type = IFTOVT(zp->z_mode); vap->va_mode = zp->z_mode & ~S_IFMT; vn_fsid(vp, vap); vap->va_nodeid = zp->z_id; vap->va_nlink = zp->z_links; if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp) && zp->z_links < ZFS_LINK_MAX) vap->va_nlink++; vap->va_size = zp->z_size; if (vp->v_type == VBLK || vp->v_type == VCHR) vap->va_rdev = zfs_cmpldev(rdev); vap->va_seq = zp->z_seq; vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */ vap->va_filerev = zp->z_seq; /* * Add in any requested optional attributes and the create time. * Also set the corresponding bits in the returned attribute bitmap. */ if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) { if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { xoap->xoa_archive = ((zp->z_pflags & ZFS_ARCHIVE) != 0); XVA_SET_RTN(xvap, XAT_ARCHIVE); } if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { xoap->xoa_readonly = ((zp->z_pflags & ZFS_READONLY) != 0); XVA_SET_RTN(xvap, XAT_READONLY); } if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { xoap->xoa_system = ((zp->z_pflags & ZFS_SYSTEM) != 0); XVA_SET_RTN(xvap, XAT_SYSTEM); } if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { xoap->xoa_hidden = ((zp->z_pflags & ZFS_HIDDEN) != 0); XVA_SET_RTN(xvap, XAT_HIDDEN); } if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { xoap->xoa_nounlink = ((zp->z_pflags & ZFS_NOUNLINK) != 0); XVA_SET_RTN(xvap, XAT_NOUNLINK); } if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { xoap->xoa_immutable = ((zp->z_pflags & ZFS_IMMUTABLE) != 0); XVA_SET_RTN(xvap, XAT_IMMUTABLE); } if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { xoap->xoa_appendonly = ((zp->z_pflags & ZFS_APPENDONLY) != 0); XVA_SET_RTN(xvap, XAT_APPENDONLY); } if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { xoap->xoa_nodump = ((zp->z_pflags & ZFS_NODUMP) != 0); XVA_SET_RTN(xvap, XAT_NODUMP); } if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { xoap->xoa_opaque = ((zp->z_pflags & ZFS_OPAQUE) != 0); XVA_SET_RTN(xvap, XAT_OPAQUE); } if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { xoap->xoa_av_quarantined = ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0); XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); } if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { xoap->xoa_av_modified = ((zp->z_pflags & ZFS_AV_MODIFIED) != 0); XVA_SET_RTN(xvap, XAT_AV_MODIFIED); } if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) && vp->v_type == VREG) { zfs_sa_get_scanstamp(zp, xvap); } if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0); XVA_SET_RTN(xvap, XAT_REPARSE); } if (XVA_ISSET_REQ(xvap, XAT_GEN)) { xoap->xoa_generation = zp->z_gen; XVA_SET_RTN(xvap, XAT_GEN); } if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { xoap->xoa_offline = ((zp->z_pflags & ZFS_OFFLINE) != 0); XVA_SET_RTN(xvap, XAT_OFFLINE); } if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { xoap->xoa_sparse = ((zp->z_pflags & ZFS_SPARSE) != 0); XVA_SET_RTN(xvap, XAT_SPARSE); } if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) { xoap->xoa_projinherit = ((zp->z_pflags & ZFS_PROJINHERIT) != 0); XVA_SET_RTN(xvap, XAT_PROJINHERIT); } if (XVA_ISSET_REQ(xvap, XAT_PROJID)) { xoap->xoa_projid = zp->z_projid; XVA_SET_RTN(xvap, XAT_PROJID); } } ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime); ZFS_TIME_DECODE(&vap->va_mtime, mtime); ZFS_TIME_DECODE(&vap->va_ctime, ctime); ZFS_TIME_DECODE(&vap->va_birthtime, crtime); sa_object_size(zp->z_sa_hdl, &blksize, &nblocks); vap->va_blksize = blksize; vap->va_bytes = nblocks << 9; /* nblocks * 512 */ if (zp->z_blksz == 0) { /* * Block size hasn't been set; suggest maximal I/O transfers. */ vap->va_blksize = zfsvfs->z_max_blksz; } ZFS_EXIT(zfsvfs); return (0); } /* * Set the file attributes to the values contained in the * vattr structure. * * IN: zp - znode of file to be modified. * vap - new attribute values. * If AT_XVATTR set, then optional attrs are being set * flags - ATTR_UTIME set if non-default time values provided. * - ATTR_NOACLCHECK (CIFS context only). * cr - credentials of caller. * ct - caller context * * RETURN: 0 on success, error code on failure. * * Timestamps: * vp - ctime updated, mtime updated if size changed. */ /* ARGSUSED */ int zfs_setattr(znode_t *zp, vattr_t *vap, int flags, cred_t *cr) { vnode_t *vp = ZTOV(zp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; objset_t *os = zfsvfs->z_os; zilog_t *zilog; dmu_tx_t *tx; vattr_t oldva; xvattr_t tmpxvattr; uint_t mask = vap->va_mask; uint_t saved_mask = 0; uint64_t saved_mode; int trim_mask = 0; uint64_t new_mode; uint64_t new_uid, new_gid; uint64_t xattr_obj; uint64_t mtime[2], ctime[2]; uint64_t projid = ZFS_INVALID_PROJID; znode_t *attrzp; int need_policy = FALSE; int err, err2; zfs_fuid_info_t *fuidp = NULL; xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ xoptattr_t *xoap; zfs_acl_t *aclp; boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; boolean_t fuid_dirtied = B_FALSE; sa_bulk_attr_t bulk[7], xattr_bulk[7]; int count = 0, xattr_count = 0; if (mask == 0) return (0); if (mask & AT_NOSET) return (SET_ERROR(EINVAL)); ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); zilog = zfsvfs->z_log; /* * Make sure that if we have ephemeral uid/gid or xvattr specified * that file system is at proper version level */ if (zfsvfs->z_use_fuids == B_FALSE && (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) || ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) || (mask & AT_XVATTR))) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EINVAL)); } if (mask & AT_SIZE && vp->v_type == VDIR) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EISDIR)); } if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EINVAL)); } /* * If this is an xvattr_t, then get a pointer to the structure of * optional attributes. If this is NULL, then we have a vattr_t. */ xoap = xva_getxoptattr(xvap); xva_init(&tmpxvattr); /* * Immutable files can only alter immutable bit and atime */ if ((zp->z_pflags & ZFS_IMMUTABLE) && ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) || ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EPERM)); } /* * Note: ZFS_READONLY is handled in zfs_zaccess_common. */ /* * Verify timestamps doesn't overflow 32 bits. * ZFS can handle large timestamps, but 32bit syscalls can't * handle times greater than 2039. This check should be removed * once large timestamps are fully supported. */ if (mask & (AT_ATIME | AT_MTIME)) { if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) || ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EOVERFLOW)); } } if (xoap != NULL && (mask & AT_XVATTR)) { if (XVA_ISSET_REQ(xvap, XAT_CREATETIME) && TIMESPEC_OVERFLOW(&vap->va_birthtime)) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EOVERFLOW)); } if (XVA_ISSET_REQ(xvap, XAT_PROJID)) { if (!dmu_objset_projectquota_enabled(os) || (!S_ISREG(zp->z_mode) && !S_ISDIR(zp->z_mode))) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EOPNOTSUPP)); } projid = xoap->xoa_projid; if (unlikely(projid == ZFS_INVALID_PROJID)) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EINVAL)); } if (projid == zp->z_projid && zp->z_pflags & ZFS_PROJID) projid = ZFS_INVALID_PROJID; else need_policy = TRUE; } if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT) && (xoap->xoa_projinherit != ((zp->z_pflags & ZFS_PROJINHERIT) != 0)) && (!dmu_objset_projectquota_enabled(os) || (!S_ISREG(zp->z_mode) && !S_ISDIR(zp->z_mode)))) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EOPNOTSUPP)); } } attrzp = NULL; aclp = NULL; if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EROFS)); } /* * First validate permissions */ if (mask & AT_SIZE) { /* * XXX - Note, we are not providing any open * mode flags here (like FNDELAY), so we may * block if there are locks present... this * should be addressed in openat(). */ /* XXX - would it be OK to generate a log record here? */ err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); if (err) { ZFS_EXIT(zfsvfs); return (err); } } if (mask & (AT_ATIME|AT_MTIME) || ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) || XVA_ISSET_REQ(xvap, XAT_READONLY) || XVA_ISSET_REQ(xvap, XAT_ARCHIVE) || XVA_ISSET_REQ(xvap, XAT_OFFLINE) || XVA_ISSET_REQ(xvap, XAT_SPARSE) || XVA_ISSET_REQ(xvap, XAT_CREATETIME) || XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) { need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0, skipaclchk, cr); } if (mask & (AT_UID|AT_GID)) { int idmask = (mask & (AT_UID|AT_GID)); int take_owner; int take_group; /* * NOTE: even if a new mode is being set, * we may clear S_ISUID/S_ISGID bits. */ if (!(mask & AT_MODE)) vap->va_mode = zp->z_mode; /* * Take ownership or chgrp to group we are a member of */ take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); take_group = (mask & AT_GID) && zfs_groupmember(zfsvfs, vap->va_gid, cr); /* * If both AT_UID and AT_GID are set then take_owner and * take_group must both be set in order to allow taking * ownership. * * Otherwise, send the check through secpolicy_vnode_setattr() * */ if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || ((idmask == AT_UID) && take_owner) || ((idmask == AT_GID) && take_group)) { if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0, skipaclchk, cr) == 0) { /* * Remove setuid/setgid for non-privileged users */ secpolicy_setid_clear(vap, vp, cr); trim_mask = (mask & (AT_UID|AT_GID)); } else { need_policy = TRUE; } } else { need_policy = TRUE; } } oldva.va_mode = zp->z_mode; zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid); if (mask & AT_XVATTR) { /* * Update xvattr mask to include only those attributes * that are actually changing. * * the bits will be restored prior to actually setting * the attributes so the caller thinks they were set. */ if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { if (xoap->xoa_appendonly != ((zp->z_pflags & ZFS_APPENDONLY) != 0)) { need_policy = TRUE; } else { XVA_CLR_REQ(xvap, XAT_APPENDONLY); XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY); } } if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) { if (xoap->xoa_projinherit != ((zp->z_pflags & ZFS_PROJINHERIT) != 0)) { need_policy = TRUE; } else { XVA_CLR_REQ(xvap, XAT_PROJINHERIT); XVA_SET_REQ(&tmpxvattr, XAT_PROJINHERIT); } } if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { if (xoap->xoa_nounlink != ((zp->z_pflags & ZFS_NOUNLINK) != 0)) { need_policy = TRUE; } else { XVA_CLR_REQ(xvap, XAT_NOUNLINK); XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK); } } if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { if (xoap->xoa_immutable != ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) { need_policy = TRUE; } else { XVA_CLR_REQ(xvap, XAT_IMMUTABLE); XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE); } } if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { if (xoap->xoa_nodump != ((zp->z_pflags & ZFS_NODUMP) != 0)) { need_policy = TRUE; } else { XVA_CLR_REQ(xvap, XAT_NODUMP); XVA_SET_REQ(&tmpxvattr, XAT_NODUMP); } } if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { if (xoap->xoa_av_modified != ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) { need_policy = TRUE; } else { XVA_CLR_REQ(xvap, XAT_AV_MODIFIED); XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED); } } if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { if ((vp->v_type != VREG && xoap->xoa_av_quarantined) || xoap->xoa_av_quarantined != ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) { need_policy = TRUE; } else { XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED); XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED); } } if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EPERM)); } if (need_policy == FALSE && (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) || XVA_ISSET_REQ(xvap, XAT_OPAQUE))) { need_policy = TRUE; } } if (mask & AT_MODE) { if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) { err = secpolicy_setid_setsticky_clear(vp, vap, &oldva, cr); if (err) { ZFS_EXIT(zfsvfs); return (err); } trim_mask |= AT_MODE; } else { need_policy = TRUE; } } if (need_policy) { /* * If trim_mask is set then take ownership * has been granted or write_acl is present and user * has the ability to modify mode. In that case remove * UID|GID and or MODE from mask so that * secpolicy_vnode_setattr() doesn't revoke it. */ if (trim_mask) { saved_mask = vap->va_mask; vap->va_mask &= ~trim_mask; if (trim_mask & AT_MODE) { /* * Save the mode, as secpolicy_vnode_setattr() * will overwrite it with ova.va_mode. */ saved_mode = vap->va_mode; } } err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp); if (err) { ZFS_EXIT(zfsvfs); return (err); } if (trim_mask) { vap->va_mask |= saved_mask; if (trim_mask & AT_MODE) { /* * Recover the mode after * secpolicy_vnode_setattr(). */ vap->va_mode = saved_mode; } } } /* * secpolicy_vnode_setattr, or take ownership may have * changed va_mask */ mask = vap->va_mask; if ((mask & (AT_UID | AT_GID)) || projid != ZFS_INVALID_PROJID) { err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &xattr_obj, sizeof (xattr_obj)); if (err == 0 && xattr_obj) { err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp); if (err == 0) { err = vn_lock(ZTOV(attrzp), LK_EXCLUSIVE); if (err != 0) vrele(ZTOV(attrzp)); } if (err) goto out2; } if (mask & AT_UID) { new_uid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp); if (new_uid != zp->z_uid && zfs_id_overquota(zfsvfs, DMU_USERUSED_OBJECT, new_uid)) { if (attrzp) vput(ZTOV(attrzp)); err = SET_ERROR(EDQUOT); goto out2; } } if (mask & AT_GID) { new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid, cr, ZFS_GROUP, &fuidp); if (new_gid != zp->z_gid && zfs_id_overquota(zfsvfs, DMU_GROUPUSED_OBJECT, new_gid)) { if (attrzp) vput(ZTOV(attrzp)); err = SET_ERROR(EDQUOT); goto out2; } } if (projid != ZFS_INVALID_PROJID && zfs_id_overquota(zfsvfs, DMU_PROJECTUSED_OBJECT, projid)) { if (attrzp) vput(ZTOV(attrzp)); err = SET_ERROR(EDQUOT); goto out2; } } tx = dmu_tx_create(os); if (mask & AT_MODE) { uint64_t pmode = zp->z_mode; uint64_t acl_obj; new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED && !(zp->z_pflags & ZFS_ACL_TRIVIAL)) { err = SET_ERROR(EPERM); goto out; } if ((err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))) goto out; if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) { /* * Are we upgrading ACL from old V0 format * to V1 format? */ if (zfsvfs->z_version >= ZPL_VERSION_FUID && zfs_znode_acl_version(zp) == ZFS_ACL_VERSION_INITIAL) { dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, aclp->z_acl_bytes); } else { dmu_tx_hold_write(tx, acl_obj, 0, aclp->z_acl_bytes); } } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) { dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, aclp->z_acl_bytes); } dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); } else { if (((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) || (projid != ZFS_INVALID_PROJID && !(zp->z_pflags & ZFS_PROJID))) dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); else dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); } if (attrzp) { dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE); } fuid_dirtied = zfsvfs->z_fuid_dirty; if (fuid_dirtied) zfs_fuid_txhold(zfsvfs, tx); zfs_sa_upgrade_txholds(tx, zp); err = dmu_tx_assign(tx, TXG_WAIT); if (err) goto out; count = 0; /* * Set each attribute requested. * We group settings according to the locks they need to acquire. * * Note: you cannot set ctime directly, although it will be * updated as a side-effect of calling this function. */ if (projid != ZFS_INVALID_PROJID && !(zp->z_pflags & ZFS_PROJID)) { /* * For the existed object that is upgraded from old system, * its on-disk layout has no slot for the project ID attribute. * But quota accounting logic needs to access related slots by * offset directly. So we need to adjust old objects' layout * to make the project ID to some unified and fixed offset. */ if (attrzp) err = sa_add_projid(attrzp->z_sa_hdl, tx, projid); if (err == 0) err = sa_add_projid(zp->z_sa_hdl, tx, projid); if (unlikely(err == EEXIST)) err = 0; else if (err != 0) goto out; else projid = ZFS_INVALID_PROJID; } if (mask & (AT_UID|AT_GID|AT_MODE)) mutex_enter(&zp->z_acl_lock); SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, &zp->z_pflags, sizeof (zp->z_pflags)); if (attrzp) { if (mask & (AT_UID|AT_GID|AT_MODE)) mutex_enter(&attrzp->z_acl_lock); SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags, sizeof (attrzp->z_pflags)); if (projid != ZFS_INVALID_PROJID) { attrzp->z_projid = projid; SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, SA_ZPL_PROJID(zfsvfs), NULL, &attrzp->z_projid, sizeof (attrzp->z_projid)); } } if (mask & (AT_UID|AT_GID)) { if (mask & AT_UID) { SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, &new_uid, sizeof (new_uid)); zp->z_uid = new_uid; if (attrzp) { SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, SA_ZPL_UID(zfsvfs), NULL, &new_uid, sizeof (new_uid)); attrzp->z_uid = new_uid; } } if (mask & AT_GID) { SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, &new_gid, sizeof (new_gid)); zp->z_gid = new_gid; if (attrzp) { SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, SA_ZPL_GID(zfsvfs), NULL, &new_gid, sizeof (new_gid)); attrzp->z_gid = new_gid; } } if (!(mask & AT_MODE)) { SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &new_mode, sizeof (new_mode)); new_mode = zp->z_mode; } err = zfs_acl_chown_setattr(zp); ASSERT0(err); if (attrzp) { vn_seqc_write_begin(ZTOV(attrzp)); err = zfs_acl_chown_setattr(attrzp); vn_seqc_write_end(ZTOV(attrzp)); ASSERT0(err); } } if (mask & AT_MODE) { SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &new_mode, sizeof (new_mode)); zp->z_mode = new_mode; ASSERT3P(aclp, !=, NULL); err = zfs_aclset_common(zp, aclp, cr, tx); ASSERT0(err); if (zp->z_acl_cached) zfs_acl_free(zp->z_acl_cached); zp->z_acl_cached = aclp; aclp = NULL; } if (mask & AT_ATIME) { ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime); SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, &zp->z_atime, sizeof (zp->z_atime)); } if (mask & AT_MTIME) { ZFS_TIME_ENCODE(&vap->va_mtime, mtime); SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, sizeof (mtime)); } if (projid != ZFS_INVALID_PROJID) { zp->z_projid = projid; SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PROJID(zfsvfs), NULL, &zp->z_projid, sizeof (zp->z_projid)); } /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */ if (mask & AT_SIZE && !(mask & AT_MTIME)) { SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, sizeof (mtime)); SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, sizeof (ctime)); zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime); } else if (mask != 0) { SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, sizeof (ctime)); zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime); if (attrzp) { SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, sizeof (ctime)); zfs_tstamp_update_setup(attrzp, STATE_CHANGED, mtime, ctime); } } /* * Do this after setting timestamps to prevent timestamp * update from toggling bit */ if (xoap && (mask & AT_XVATTR)) { if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) xoap->xoa_createtime = vap->va_birthtime; /* * restore trimmed off masks * so that return masks can be set for caller. */ if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) { XVA_SET_REQ(xvap, XAT_APPENDONLY); } if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) { XVA_SET_REQ(xvap, XAT_NOUNLINK); } if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) { XVA_SET_REQ(xvap, XAT_IMMUTABLE); } if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) { XVA_SET_REQ(xvap, XAT_NODUMP); } if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) { XVA_SET_REQ(xvap, XAT_AV_MODIFIED); } if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) { XVA_SET_REQ(xvap, XAT_AV_QUARANTINED); } if (XVA_ISSET_REQ(&tmpxvattr, XAT_PROJINHERIT)) { XVA_SET_REQ(xvap, XAT_PROJINHERIT); } if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) ASSERT3S(vp->v_type, ==, VREG); zfs_xvattr_set(zp, xvap, tx); } if (fuid_dirtied) zfs_fuid_sync(zfsvfs, tx); if (mask != 0) zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp); if (mask & (AT_UID|AT_GID|AT_MODE)) mutex_exit(&zp->z_acl_lock); if (attrzp) { if (mask & (AT_UID|AT_GID|AT_MODE)) mutex_exit(&attrzp->z_acl_lock); } out: if (err == 0 && attrzp) { err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk, xattr_count, tx); ASSERT0(err2); } if (attrzp) vput(ZTOV(attrzp)); if (aclp) zfs_acl_free(aclp); if (fuidp) { zfs_fuid_info_free(fuidp); fuidp = NULL; } if (err) { dmu_tx_abort(tx); } else { err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); dmu_tx_commit(tx); } out2: if (os->os_sync == ZFS_SYNC_ALWAYS) zil_commit(zilog, 0); ZFS_EXIT(zfsvfs); return (err); } /* * We acquire all but fdvp locks using non-blocking acquisitions. If we * fail to acquire any lock in the path we will drop all held locks, * acquire the new lock in a blocking fashion, and then release it and * restart the rename. This acquire/release step ensures that we do not * spin on a lock waiting for release. On error release all vnode locks * and decrement references the way tmpfs_rename() would do. */ static int zfs_rename_relock(struct vnode *sdvp, struct vnode **svpp, struct vnode *tdvp, struct vnode **tvpp, const struct componentname *scnp, const struct componentname *tcnp) { zfsvfs_t *zfsvfs; struct vnode *nvp, *svp, *tvp; znode_t *sdzp, *tdzp, *szp, *tzp; const char *snm = scnp->cn_nameptr; const char *tnm = tcnp->cn_nameptr; int error; VOP_UNLOCK1(tdvp); if (*tvpp != NULL && *tvpp != tdvp) VOP_UNLOCK1(*tvpp); relock: error = vn_lock(sdvp, LK_EXCLUSIVE); if (error) goto out; sdzp = VTOZ(sdvp); error = vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT); if (error != 0) { VOP_UNLOCK1(sdvp); if (error != EBUSY) goto out; error = vn_lock(tdvp, LK_EXCLUSIVE); if (error) goto out; VOP_UNLOCK1(tdvp); goto relock; } tdzp = VTOZ(tdvp); /* * Before using sdzp and tdzp we must ensure that they are live. * As a porting legacy from illumos we have two things to worry * about. One is typical for FreeBSD and it is that the vnode is * not reclaimed (doomed). The other is that the znode is live. * The current code can invalidate the znode without acquiring the * corresponding vnode lock if the object represented by the znode * and vnode is no longer valid after a rollback or receive operation. * z_teardown_lock hidden behind ZFS_ENTER and ZFS_EXIT is the lock * that protects the znodes from the invalidation. */ zfsvfs = sdzp->z_zfsvfs; ASSERT3P(zfsvfs, ==, tdzp->z_zfsvfs); ZFS_ENTER(zfsvfs); /* * We can not use ZFS_VERIFY_ZP() here because it could directly return * bypassing the cleanup code in the case of an error. */ if (tdzp->z_sa_hdl == NULL || sdzp->z_sa_hdl == NULL) { ZFS_EXIT(zfsvfs); VOP_UNLOCK1(sdvp); VOP_UNLOCK1(tdvp); error = SET_ERROR(EIO); goto out; } /* * Re-resolve svp to be certain it still exists and fetch the * correct vnode. */ error = zfs_dirent_lookup(sdzp, snm, &szp, ZEXISTS); if (error != 0) { /* Source entry invalid or not there. */ ZFS_EXIT(zfsvfs); VOP_UNLOCK1(sdvp); VOP_UNLOCK1(tdvp); if ((scnp->cn_flags & ISDOTDOT) != 0 || (scnp->cn_namelen == 1 && scnp->cn_nameptr[0] == '.')) error = SET_ERROR(EINVAL); goto out; } svp = ZTOV(szp); /* * Re-resolve tvp, if it disappeared we just carry on. */ error = zfs_dirent_lookup(tdzp, tnm, &tzp, 0); if (error != 0) { ZFS_EXIT(zfsvfs); VOP_UNLOCK1(sdvp); VOP_UNLOCK1(tdvp); vrele(svp); if ((tcnp->cn_flags & ISDOTDOT) != 0) error = SET_ERROR(EINVAL); goto out; } if (tzp != NULL) tvp = ZTOV(tzp); else tvp = NULL; /* * At present the vnode locks must be acquired before z_teardown_lock, * although it would be more logical to use the opposite order. */ ZFS_EXIT(zfsvfs); /* * Now try acquire locks on svp and tvp. */ nvp = svp; error = vn_lock(nvp, LK_EXCLUSIVE | LK_NOWAIT); if (error != 0) { VOP_UNLOCK1(sdvp); VOP_UNLOCK1(tdvp); if (tvp != NULL) vrele(tvp); if (error != EBUSY) { vrele(nvp); goto out; } error = vn_lock(nvp, LK_EXCLUSIVE); if (error != 0) { vrele(nvp); goto out; } VOP_UNLOCK1(nvp); /* * Concurrent rename race. * XXX ? */ if (nvp == tdvp) { vrele(nvp); error = SET_ERROR(EINVAL); goto out; } vrele(*svpp); *svpp = nvp; goto relock; } vrele(*svpp); *svpp = nvp; if (*tvpp != NULL) vrele(*tvpp); *tvpp = NULL; if (tvp != NULL) { nvp = tvp; error = vn_lock(nvp, LK_EXCLUSIVE | LK_NOWAIT); if (error != 0) { VOP_UNLOCK1(sdvp); VOP_UNLOCK1(tdvp); VOP_UNLOCK1(*svpp); if (error != EBUSY) { vrele(nvp); goto out; } error = vn_lock(nvp, LK_EXCLUSIVE); if (error != 0) { vrele(nvp); goto out; } vput(nvp); goto relock; } *tvpp = nvp; } return (0); out: return (error); } /* * Note that we must use VRELE_ASYNC in this function as it walks * up the directory tree and vrele may need to acquire an exclusive * lock if a last reference to a vnode is dropped. */ static int zfs_rename_check(znode_t *szp, znode_t *sdzp, znode_t *tdzp) { zfsvfs_t *zfsvfs; znode_t *zp, *zp1; uint64_t parent; int error; zfsvfs = tdzp->z_zfsvfs; if (tdzp == szp) return (SET_ERROR(EINVAL)); if (tdzp == sdzp) return (0); if (tdzp->z_id == zfsvfs->z_root) return (0); zp = tdzp; for (;;) { ASSERT(!zp->z_unlinked); if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0) break; if (parent == szp->z_id) { error = SET_ERROR(EINVAL); break; } if (parent == zfsvfs->z_root) break; if (parent == sdzp->z_id) break; error = zfs_zget(zfsvfs, parent, &zp1); if (error != 0) break; if (zp != tdzp) VN_RELE_ASYNC(ZTOV(zp), dsl_pool_zrele_taskq( dmu_objset_pool(zfsvfs->z_os))); zp = zp1; } if (error == ENOTDIR) panic("checkpath: .. not a directory\n"); if (zp != tdzp) VN_RELE_ASYNC(ZTOV(zp), dsl_pool_zrele_taskq(dmu_objset_pool(zfsvfs->z_os))); return (error); } #if __FreeBSD_version < 1300124 static void cache_vop_rename(struct vnode *fdvp, struct vnode *fvp, struct vnode *tdvp, struct vnode *tvp, struct componentname *fcnp, struct componentname *tcnp) { cache_purge(fvp); if (tvp != NULL) cache_purge(tvp); cache_purge_negative(tdvp); } #endif /* * Move an entry from the provided source directory to the target * directory. Change the entry name as indicated. * * IN: sdvp - Source directory containing the "old entry". * snm - Old entry name. * tdvp - Target directory to contain the "new entry". * tnm - New entry name. * cr - credentials of caller. * ct - caller context * flags - case flags * * RETURN: 0 on success, error code on failure. * * Timestamps: * sdvp,tdvp - ctime|mtime updated */ /*ARGSUSED*/ static int zfs_rename_(vnode_t *sdvp, vnode_t **svpp, struct componentname *scnp, vnode_t *tdvp, vnode_t **tvpp, struct componentname *tcnp, cred_t *cr, int log) { zfsvfs_t *zfsvfs; znode_t *sdzp, *tdzp, *szp, *tzp; zilog_t *zilog = NULL; dmu_tx_t *tx; const char *snm = scnp->cn_nameptr; const char *tnm = tcnp->cn_nameptr; int error = 0; bool want_seqc_end __maybe_unused = false; /* Reject renames across filesystems. */ if ((*svpp)->v_mount != tdvp->v_mount || ((*tvpp) != NULL && (*svpp)->v_mount != (*tvpp)->v_mount)) { error = SET_ERROR(EXDEV); goto out; } if (zfsctl_is_node(tdvp)) { error = SET_ERROR(EXDEV); goto out; } /* * Lock all four vnodes to ensure safety and semantics of renaming. */ error = zfs_rename_relock(sdvp, svpp, tdvp, tvpp, scnp, tcnp); if (error != 0) { /* no vnodes are locked in the case of error here */ return (error); } tdzp = VTOZ(tdvp); sdzp = VTOZ(sdvp); zfsvfs = tdzp->z_zfsvfs; zilog = zfsvfs->z_log; /* * After we re-enter ZFS_ENTER() we will have to revalidate all * znodes involved. */ ZFS_ENTER(zfsvfs); if (zfsvfs->z_utf8 && u8_validate(tnm, strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { error = SET_ERROR(EILSEQ); goto unlockout; } /* If source and target are the same file, there is nothing to do. */ if ((*svpp) == (*tvpp)) { error = 0; goto unlockout; } if (((*svpp)->v_type == VDIR && (*svpp)->v_mountedhere != NULL) || ((*tvpp) != NULL && (*tvpp)->v_type == VDIR && (*tvpp)->v_mountedhere != NULL)) { error = SET_ERROR(EXDEV); goto unlockout; } /* * We can not use ZFS_VERIFY_ZP() here because it could directly return * bypassing the cleanup code in the case of an error. */ if (tdzp->z_sa_hdl == NULL || sdzp->z_sa_hdl == NULL) { error = SET_ERROR(EIO); goto unlockout; } szp = VTOZ(*svpp); tzp = *tvpp == NULL ? NULL : VTOZ(*tvpp); if (szp->z_sa_hdl == NULL || (tzp != NULL && tzp->z_sa_hdl == NULL)) { error = SET_ERROR(EIO); goto unlockout; } /* * This is to prevent the creation of links into attribute space * by renaming a linked file into/outof an attribute directory. * See the comment in zfs_link() for why this is considered bad. */ if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) { error = SET_ERROR(EINVAL); goto unlockout; } /* * If we are using project inheritance, means if the directory has * ZFS_PROJINHERIT set, then its descendant directories will inherit * not only the project ID, but also the ZFS_PROJINHERIT flag. Under * such case, we only allow renames into our tree when the project * IDs are the same. */ if (tdzp->z_pflags & ZFS_PROJINHERIT && tdzp->z_projid != szp->z_projid) { error = SET_ERROR(EXDEV); goto unlockout; } /* * Must have write access at the source to remove the old entry * and write access at the target to create the new entry. * Note that if target and source are the same, this can be * done in a single check. */ if ((error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))) goto unlockout; if ((*svpp)->v_type == VDIR) { /* * Avoid ".", "..", and aliases of "." for obvious reasons. */ if ((scnp->cn_namelen == 1 && scnp->cn_nameptr[0] == '.') || sdzp == szp || (scnp->cn_flags | tcnp->cn_flags) & ISDOTDOT) { error = EINVAL; goto unlockout; } /* * Check to make sure rename is valid. * Can't do a move like this: /usr/a/b to /usr/a/b/c/d */ if ((error = zfs_rename_check(szp, sdzp, tdzp))) goto unlockout; } /* * Does target exist? */ if (tzp) { /* * Source and target must be the same type. */ if ((*svpp)->v_type == VDIR) { if ((*tvpp)->v_type != VDIR) { error = SET_ERROR(ENOTDIR); goto unlockout; } else { cache_purge(tdvp); if (sdvp != tdvp) cache_purge(sdvp); } } else { if ((*tvpp)->v_type == VDIR) { error = SET_ERROR(EISDIR); goto unlockout; } } } vn_seqc_write_begin(*svpp); vn_seqc_write_begin(sdvp); if (*tvpp != NULL) vn_seqc_write_begin(*tvpp); if (tdvp != *tvpp) vn_seqc_write_begin(tdvp); #if __FreeBSD_version >= 1300102 want_seqc_end = true; #endif vnevent_rename_src(*svpp, sdvp, scnp->cn_nameptr, ct); if (tzp) vnevent_rename_dest(*tvpp, tdvp, tnm, ct); /* * notify the target directory if it is not the same * as source directory. */ if (tdvp != sdvp) { vnevent_rename_dest_dir(tdvp, ct); } tx = dmu_tx_create(zfsvfs->z_os); dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE); dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); if (sdzp != tdzp) { dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE); zfs_sa_upgrade_txholds(tx, tdzp); } if (tzp) { dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE); zfs_sa_upgrade_txholds(tx, tzp); } zfs_sa_upgrade_txholds(tx, szp); dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); error = dmu_tx_assign(tx, TXG_WAIT); if (error) { dmu_tx_abort(tx); goto unlockout; } if (tzp) /* Attempt to remove the existing target */ error = zfs_link_destroy(tdzp, tnm, tzp, tx, 0, NULL); if (error == 0) { error = zfs_link_create(tdzp, tnm, szp, tx, ZRENAMING); if (error == 0) { szp->z_pflags |= ZFS_AV_MODIFIED; error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs), (void *)&szp->z_pflags, sizeof (uint64_t), tx); ASSERT0(error); error = zfs_link_destroy(sdzp, snm, szp, tx, ZRENAMING, NULL); if (error == 0) { zfs_log_rename(zilog, tx, TX_RENAME, sdzp, snm, tdzp, tnm, szp); /* * Update path information for the target vnode */ vn_renamepath(tdvp, *svpp, tnm, strlen(tnm)); } else { /* * At this point, we have successfully created * the target name, but have failed to remove * the source name. Since the create was done * with the ZRENAMING flag, there are * complications; for one, the link count is * wrong. The easiest way to deal with this * is to remove the newly created target, and * return the original error. This must * succeed; fortunately, it is very unlikely to * fail, since we just created it. */ VERIFY0(zfs_link_destroy(tdzp, tnm, szp, tx, ZRENAMING, NULL)); } } if (error == 0) { cache_vop_rename(sdvp, *svpp, tdvp, *tvpp, scnp, tcnp); } } dmu_tx_commit(tx); unlockout: /* all 4 vnodes are locked, ZFS_ENTER called */ ZFS_EXIT(zfsvfs); if (want_seqc_end) { vn_seqc_write_end(*svpp); vn_seqc_write_end(sdvp); if (*tvpp != NULL) vn_seqc_write_end(*tvpp); if (tdvp != *tvpp) vn_seqc_write_end(tdvp); want_seqc_end = false; } VOP_UNLOCK1(*svpp); VOP_UNLOCK1(sdvp); out: /* original two vnodes are locked */ MPASS(!want_seqc_end); if (error == 0 && zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) zil_commit(zilog, 0); if (*tvpp != NULL) VOP_UNLOCK1(*tvpp); if (tdvp != *tvpp) VOP_UNLOCK1(tdvp); return (error); } int zfs_rename(znode_t *sdzp, const char *sname, znode_t *tdzp, const char *tname, cred_t *cr, int flags) { struct componentname scn, tcn; vnode_t *sdvp, *tdvp; vnode_t *svp, *tvp; int error; svp = tvp = NULL; sdvp = ZTOV(sdzp); tdvp = ZTOV(tdzp); error = zfs_lookup_internal(sdzp, sname, &svp, &scn, DELETE); if (sdzp->z_zfsvfs->z_replay == B_FALSE) VOP_UNLOCK1(sdvp); if (error != 0) goto fail; VOP_UNLOCK1(svp); vn_lock(tdvp, LK_EXCLUSIVE | LK_RETRY); error = zfs_lookup_internal(tdzp, tname, &tvp, &tcn, RENAME); if (error == EJUSTRETURN) tvp = NULL; else if (error != 0) { VOP_UNLOCK1(tdvp); goto fail; } error = zfs_rename_(sdvp, &svp, &scn, tdvp, &tvp, &tcn, cr, 0); fail: if (svp != NULL) vrele(svp); if (tvp != NULL) vrele(tvp); return (error); } /* * Insert the indicated symbolic reference entry into the directory. * * IN: dvp - Directory to contain new symbolic link. * link - Name for new symlink entry. * vap - Attributes of new entry. * cr - credentials of caller. * ct - caller context * flags - case flags * * RETURN: 0 on success, error code on failure. * * Timestamps: * dvp - ctime|mtime updated */ /*ARGSUSED*/ int zfs_symlink(znode_t *dzp, const char *name, vattr_t *vap, const char *link, znode_t **zpp, cred_t *cr, int flags) { znode_t *zp; dmu_tx_t *tx; zfsvfs_t *zfsvfs = dzp->z_zfsvfs; zilog_t *zilog; uint64_t len = strlen(link); int error; zfs_acl_ids_t acl_ids; boolean_t fuid_dirtied; uint64_t txtype = TX_SYMLINK; ASSERT3S(vap->va_type, ==, VLNK); ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(dzp); zilog = zfsvfs->z_log; if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EILSEQ)); } if (len > MAXPATHLEN) { ZFS_EXIT(zfsvfs); return (SET_ERROR(ENAMETOOLONG)); } if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, NULL, &acl_ids)) != 0) { ZFS_EXIT(zfsvfs); return (error); } /* * Attempt to lock directory; fail if entry already exists. */ error = zfs_dirent_lookup(dzp, name, &zp, ZNEW); if (error) { zfs_acl_ids_free(&acl_ids); ZFS_EXIT(zfsvfs); return (error); } if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) { zfs_acl_ids_free(&acl_ids); ZFS_EXIT(zfsvfs); return (error); } if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, 0 /* projid */)) { zfs_acl_ids_free(&acl_ids); ZFS_EXIT(zfsvfs); return (SET_ERROR(EDQUOT)); } getnewvnode_reserve_(); tx = dmu_tx_create(zfsvfs->z_os); fuid_dirtied = zfsvfs->z_fuid_dirty; dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + ZFS_SA_BASE_ATTR_SIZE + len); dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, acl_ids.z_aclp->z_acl_bytes); } if (fuid_dirtied) zfs_fuid_txhold(zfsvfs, tx); error = dmu_tx_assign(tx, TXG_WAIT); if (error) { zfs_acl_ids_free(&acl_ids); dmu_tx_abort(tx); getnewvnode_drop_reserve(); ZFS_EXIT(zfsvfs); return (error); } /* * Create a new object for the symlink. * for version 4 ZPL datasets the symlink will be an SA attribute */ zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); if (fuid_dirtied) zfs_fuid_sync(zfsvfs, tx); if (zp->z_is_sa) error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs), __DECONST(void *, link), len, tx); else zfs_sa_symlink(zp, __DECONST(char *, link), len, tx); zp->z_size = len; (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), &zp->z_size, sizeof (zp->z_size), tx); /* * Insert the new object into the directory. */ (void) zfs_link_create(dzp, name, zp, tx, ZNEW); zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link); *zpp = zp; zfs_acl_ids_free(&acl_ids); dmu_tx_commit(tx); getnewvnode_drop_reserve(); if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) zil_commit(zilog, 0); ZFS_EXIT(zfsvfs); return (error); } /* * Return, in the buffer contained in the provided uio structure, * the symbolic path referred to by vp. * * IN: vp - vnode of symbolic link. * uio - structure to contain the link path. * cr - credentials of caller. * ct - caller context * * OUT: uio - structure containing the link path. * * RETURN: 0 on success, error code on failure. * * Timestamps: * vp - atime updated */ /* ARGSUSED */ static int zfs_readlink(vnode_t *vp, zfs_uio_t *uio, cred_t *cr, caller_context_t *ct) { znode_t *zp = VTOZ(vp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; int error; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); if (zp->z_is_sa) error = sa_lookup_uio(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs), uio); else error = zfs_sa_readlink(zp, uio); ZFS_ACCESSTIME_STAMP(zfsvfs, zp); ZFS_EXIT(zfsvfs); return (error); } /* * Insert a new entry into directory tdvp referencing svp. * * IN: tdvp - Directory to contain new entry. * svp - vnode of new entry. * name - name of new entry. * cr - credentials of caller. * * RETURN: 0 on success, error code on failure. * * Timestamps: * tdvp - ctime|mtime updated * svp - ctime updated */ /* ARGSUSED */ int zfs_link(znode_t *tdzp, znode_t *szp, const char *name, cred_t *cr, int flags) { znode_t *tzp; zfsvfs_t *zfsvfs = tdzp->z_zfsvfs; zilog_t *zilog; dmu_tx_t *tx; int error; uint64_t parent; uid_t owner; ASSERT3S(ZTOV(tdzp)->v_type, ==, VDIR); ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(tdzp); zilog = zfsvfs->z_log; /* * POSIX dictates that we return EPERM here. * Better choices include ENOTSUP or EISDIR. */ if (ZTOV(szp)->v_type == VDIR) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EPERM)); } ZFS_VERIFY_ZP(szp); /* * If we are using project inheritance, means if the directory has * ZFS_PROJINHERIT set, then its descendant directories will inherit * not only the project ID, but also the ZFS_PROJINHERIT flag. Under * such case, we only allow hard link creation in our tree when the * project IDs are the same. */ if (tdzp->z_pflags & ZFS_PROJINHERIT && tdzp->z_projid != szp->z_projid) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EXDEV)); } if (szp->z_pflags & (ZFS_APPENDONLY | ZFS_IMMUTABLE | ZFS_READONLY)) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EPERM)); } /* Prevent links to .zfs/shares files */ if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), &parent, sizeof (uint64_t))) != 0) { ZFS_EXIT(zfsvfs); return (error); } if (parent == zfsvfs->z_shares_dir) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EPERM)); } if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EILSEQ)); } /* * We do not support links between attributes and non-attributes * because of the potential security risk of creating links * into "normal" file space in order to circumvent restrictions * imposed in attribute space. */ if ((szp->z_pflags & ZFS_XATTR) != (tdzp->z_pflags & ZFS_XATTR)) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EINVAL)); } owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER); if (owner != crgetuid(cr) && secpolicy_basic_link(ZTOV(szp), cr) != 0) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EPERM)); } if ((error = zfs_zaccess(tdzp, ACE_ADD_FILE, 0, B_FALSE, cr))) { ZFS_EXIT(zfsvfs); return (error); } /* * Attempt to lock directory; fail if entry already exists. */ error = zfs_dirent_lookup(tdzp, name, &tzp, ZNEW); if (error) { ZFS_EXIT(zfsvfs); return (error); } tx = dmu_tx_create(zfsvfs->z_os); dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, name); zfs_sa_upgrade_txholds(tx, szp); zfs_sa_upgrade_txholds(tx, tdzp); error = dmu_tx_assign(tx, TXG_WAIT); if (error) { dmu_tx_abort(tx); ZFS_EXIT(zfsvfs); return (error); } error = zfs_link_create(tdzp, name, szp, tx, 0); if (error == 0) { uint64_t txtype = TX_LINK; zfs_log_link(zilog, tx, txtype, tdzp, szp, name); } dmu_tx_commit(tx); if (error == 0) { vnevent_link(ZTOV(szp), ct); } if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) zil_commit(zilog, 0); ZFS_EXIT(zfsvfs); return (error); } /* * Free or allocate space in a file. Currently, this function only * supports the `F_FREESP' command. However, this command is somewhat * misnamed, as its functionality includes the ability to allocate as * well as free space. * * IN: ip - inode of file to free data in. * cmd - action to take (only F_FREESP supported). * bfp - section of file to free/alloc. * flag - current file open mode flags. * offset - current file offset. * cr - credentials of caller. * * RETURN: 0 on success, error code on failure. * * Timestamps: * ip - ctime|mtime updated */ /* ARGSUSED */ int zfs_space(znode_t *zp, int cmd, flock64_t *bfp, int flag, offset_t offset, cred_t *cr) { zfsvfs_t *zfsvfs = ZTOZSB(zp); uint64_t off, len; int error; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); if (cmd != F_FREESP) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EINVAL)); } /* * Callers might not be able to detect properly that we are read-only, * so check it explicitly here. */ if (zfs_is_readonly(zfsvfs)) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EROFS)); } if (bfp->l_len < 0) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EINVAL)); } /* * Permissions aren't checked on Solaris because on this OS * zfs_space() can only be called with an opened file handle. * On Linux we can get here through truncate_range() which * operates directly on inodes, so we need to check access rights. */ if ((error = zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr))) { ZFS_EXIT(zfsvfs); return (error); } off = bfp->l_start; len = bfp->l_len; /* 0 means from off to end of file */ error = zfs_freesp(zp, off, len, flag, TRUE); ZFS_EXIT(zfsvfs); return (error); } /*ARGSUSED*/ static void zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct) { znode_t *zp = VTOZ(vp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; int error; ZFS_TEARDOWN_INACTIVE_ENTER_READ(zfsvfs); if (zp->z_sa_hdl == NULL) { /* * The fs has been unmounted, or we did a * suspend/resume and this file no longer exists. */ ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs); vrecycle(vp); return; } if (zp->z_unlinked) { /* * Fast path to recycle a vnode of a removed file. */ ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs); vrecycle(vp); return; } if (zp->z_atime_dirty && zp->z_unlinked == 0) { dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); zfs_sa_upgrade_txholds(tx, zp); error = dmu_tx_assign(tx, TXG_WAIT); if (error) { dmu_tx_abort(tx); } else { (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs), (void *)&zp->z_atime, sizeof (zp->z_atime), tx); zp->z_atime_dirty = 0; dmu_tx_commit(tx); } } ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs); } CTASSERT(sizeof (struct zfid_short) <= sizeof (struct fid)); CTASSERT(sizeof (struct zfid_long) <= sizeof (struct fid)); /*ARGSUSED*/ static int zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct) { znode_t *zp = VTOZ(vp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; uint32_t gen; uint64_t gen64; uint64_t object = zp->z_id; zfid_short_t *zfid; int size, i, error; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &gen64, sizeof (uint64_t))) != 0) { ZFS_EXIT(zfsvfs); return (error); } gen = (uint32_t)gen64; size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; fidp->fid_len = size; zfid = (zfid_short_t *)fidp; zfid->zf_len = size; for (i = 0; i < sizeof (zfid->zf_object); i++) zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); /* Must have a non-zero generation number to distinguish from .zfs */ if (gen == 0) gen = 1; for (i = 0; i < sizeof (zfid->zf_gen); i++) zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); if (size == LONG_FID_LEN) { uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); zfid_long_t *zlfid; zlfid = (zfid_long_t *)fidp; for (i = 0; i < sizeof (zlfid->zf_setid); i++) zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); /* XXX - this should be the generation number for the objset */ for (i = 0; i < sizeof (zlfid->zf_setgen); i++) zlfid->zf_setgen[i] = 0; } ZFS_EXIT(zfsvfs); return (0); } static int zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, caller_context_t *ct) { znode_t *zp; zfsvfs_t *zfsvfs; switch (cmd) { case _PC_LINK_MAX: *valp = MIN(LONG_MAX, ZFS_LINK_MAX); return (0); case _PC_FILESIZEBITS: *valp = 64; return (0); case _PC_MIN_HOLE_SIZE: *valp = (int)SPA_MINBLOCKSIZE; return (0); case _PC_ACL_EXTENDED: #if 0 /* POSIX ACLs are not implemented for ZFS on FreeBSD yet. */ zp = VTOZ(vp); zfsvfs = zp->z_zfsvfs; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); *valp = zfsvfs->z_acl_type == ZFSACLTYPE_POSIX ? 1 : 0; ZFS_EXIT(zfsvfs); #else *valp = 0; #endif return (0); case _PC_ACL_NFS4: zp = VTOZ(vp); zfsvfs = zp->z_zfsvfs; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); *valp = zfsvfs->z_acl_type == ZFS_ACLTYPE_NFSV4 ? 1 : 0; ZFS_EXIT(zfsvfs); return (0); case _PC_ACL_PATH_MAX: *valp = ACL_MAX_ENTRIES; return (0); default: return (EOPNOTSUPP); } } static int zfs_getpages(struct vnode *vp, vm_page_t *ma, int count, int *rbehind, int *rahead) { znode_t *zp = VTOZ(vp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; objset_t *os = zp->z_zfsvfs->z_os; zfs_locked_range_t *lr; vm_object_t object; off_t start, end, obj_size; uint_t blksz; int pgsin_b, pgsin_a; int error; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); start = IDX_TO_OFF(ma[0]->pindex); end = IDX_TO_OFF(ma[count - 1]->pindex + 1); /* * Lock a range covering all required and optional pages. * Note that we need to handle the case of the block size growing. */ for (;;) { blksz = zp->z_blksz; lr = zfs_rangelock_tryenter(&zp->z_rangelock, rounddown(start, blksz), roundup(end, blksz) - rounddown(start, blksz), RL_READER); if (lr == NULL) { if (rahead != NULL) { *rahead = 0; rahead = NULL; } if (rbehind != NULL) { *rbehind = 0; rbehind = NULL; } break; } if (blksz == zp->z_blksz) break; zfs_rangelock_exit(lr); } object = ma[0]->object; zfs_vmobject_wlock(object); obj_size = object->un_pager.vnp.vnp_size; zfs_vmobject_wunlock(object); if (IDX_TO_OFF(ma[count - 1]->pindex) >= obj_size) { if (lr != NULL) zfs_rangelock_exit(lr); ZFS_EXIT(zfsvfs); return (zfs_vm_pagerret_bad); } pgsin_b = 0; if (rbehind != NULL) { pgsin_b = OFF_TO_IDX(start - rounddown(start, blksz)); pgsin_b = MIN(*rbehind, pgsin_b); } pgsin_a = 0; if (rahead != NULL) { pgsin_a = OFF_TO_IDX(roundup(end, blksz) - end); if (end + IDX_TO_OFF(pgsin_a) >= obj_size) pgsin_a = OFF_TO_IDX(round_page(obj_size) - end); pgsin_a = MIN(*rahead, pgsin_a); } /* * NB: we need to pass the exact byte size of the data that we expect * to read after accounting for the file size. This is required because * ZFS will panic if we request DMU to read beyond the end of the last * allocated block. */ error = dmu_read_pages(os, zp->z_id, ma, count, &pgsin_b, &pgsin_a, MIN(end, obj_size) - (end - PAGE_SIZE)); if (lr != NULL) zfs_rangelock_exit(lr); ZFS_ACCESSTIME_STAMP(zfsvfs, zp); ZFS_EXIT(zfsvfs); if (error != 0) return (zfs_vm_pagerret_error); VM_CNT_INC(v_vnodein); VM_CNT_ADD(v_vnodepgsin, count + pgsin_b + pgsin_a); if (rbehind != NULL) *rbehind = pgsin_b; if (rahead != NULL) *rahead = pgsin_a; return (zfs_vm_pagerret_ok); } #ifndef _SYS_SYSPROTO_H_ struct vop_getpages_args { struct vnode *a_vp; vm_page_t *a_m; int a_count; int *a_rbehind; int *a_rahead; }; #endif static int zfs_freebsd_getpages(struct vop_getpages_args *ap) { return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_rbehind, ap->a_rahead)); } static int zfs_putpages(struct vnode *vp, vm_page_t *ma, size_t len, int flags, int *rtvals) { znode_t *zp = VTOZ(vp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; zfs_locked_range_t *lr; dmu_tx_t *tx; struct sf_buf *sf; vm_object_t object; vm_page_t m; caddr_t va; size_t tocopy; size_t lo_len; vm_ooffset_t lo_off; vm_ooffset_t off; uint_t blksz; int ncount; int pcount; int err; int i; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); object = vp->v_object; pcount = btoc(len); ncount = pcount; KASSERT(ma[0]->object == object, ("mismatching object")); KASSERT(len > 0 && (len & PAGE_MASK) == 0, ("unexpected length")); for (i = 0; i < pcount; i++) rtvals[i] = zfs_vm_pagerret_error; off = IDX_TO_OFF(ma[0]->pindex); blksz = zp->z_blksz; lo_off = rounddown(off, blksz); lo_len = roundup(len + (off - lo_off), blksz); lr = zfs_rangelock_enter(&zp->z_rangelock, lo_off, lo_len, RL_WRITER); zfs_vmobject_wlock(object); if (len + off > object->un_pager.vnp.vnp_size) { if (object->un_pager.vnp.vnp_size > off) { int pgoff; len = object->un_pager.vnp.vnp_size - off; ncount = btoc(len); if ((pgoff = (int)len & PAGE_MASK) != 0) { /* * If the object is locked and the following * conditions hold, then the page's dirty * field cannot be concurrently changed by a * pmap operation. */ m = ma[ncount - 1]; vm_page_assert_sbusied(m); KASSERT(!pmap_page_is_write_mapped(m), ("zfs_putpages: page %p is not read-only", m)); vm_page_clear_dirty(m, pgoff, PAGE_SIZE - pgoff); } } else { len = 0; ncount = 0; } if (ncount < pcount) { for (i = ncount; i < pcount; i++) { rtvals[i] = zfs_vm_pagerret_bad; } } } zfs_vmobject_wunlock(object); if (ncount == 0) goto out; if (zfs_id_overblockquota(zfsvfs, DMU_USERUSED_OBJECT, zp->z_uid) || zfs_id_overblockquota(zfsvfs, DMU_GROUPUSED_OBJECT, zp->z_gid) || (zp->z_projid != ZFS_DEFAULT_PROJID && zfs_id_overblockquota(zfsvfs, DMU_PROJECTUSED_OBJECT, zp->z_projid))) { goto out; } tx = dmu_tx_create(zfsvfs->z_os); dmu_tx_hold_write(tx, zp->z_id, off, len); dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); zfs_sa_upgrade_txholds(tx, zp); err = dmu_tx_assign(tx, TXG_WAIT); if (err != 0) { dmu_tx_abort(tx); goto out; } if (zp->z_blksz < PAGE_SIZE) { for (i = 0; len > 0; off += tocopy, len -= tocopy, i++) { tocopy = len > PAGE_SIZE ? PAGE_SIZE : len; va = zfs_map_page(ma[i], &sf); dmu_write(zfsvfs->z_os, zp->z_id, off, tocopy, va, tx); zfs_unmap_page(sf); } } else { err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, ma, tx); } if (err == 0) { uint64_t mtime[2], ctime[2]; sa_bulk_attr_t bulk[3]; int count = 0; SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, &zp->z_pflags, 8); zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime); err = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); ASSERT0(err); /* * XXX we should be passing a callback to undirty * but that would make the locking messier */ zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0, NULL, NULL); zfs_vmobject_wlock(object); for (i = 0; i < ncount; i++) { rtvals[i] = zfs_vm_pagerret_ok; vm_page_undirty(ma[i]); } zfs_vmobject_wunlock(object); VM_CNT_INC(v_vnodeout); VM_CNT_ADD(v_vnodepgsout, ncount); } dmu_tx_commit(tx); out: zfs_rangelock_exit(lr); if ((flags & (zfs_vm_pagerput_sync | zfs_vm_pagerput_inval)) != 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) zil_commit(zfsvfs->z_log, zp->z_id); ZFS_EXIT(zfsvfs); return (rtvals[0]); } #ifndef _SYS_SYSPROTO_H_ struct vop_putpages_args { struct vnode *a_vp; vm_page_t *a_m; int a_count; int a_sync; int *a_rtvals; }; #endif static int zfs_freebsd_putpages(struct vop_putpages_args *ap) { return (zfs_putpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_sync, ap->a_rtvals)); } #ifndef _SYS_SYSPROTO_H_ struct vop_bmap_args { struct vnode *a_vp; daddr_t a_bn; struct bufobj **a_bop; daddr_t *a_bnp; int *a_runp; int *a_runb; }; #endif static int zfs_freebsd_bmap(struct vop_bmap_args *ap) { if (ap->a_bop != NULL) *ap->a_bop = &ap->a_vp->v_bufobj; if (ap->a_bnp != NULL) *ap->a_bnp = ap->a_bn; if (ap->a_runp != NULL) *ap->a_runp = 0; if (ap->a_runb != NULL) *ap->a_runb = 0; return (0); } #ifndef _SYS_SYSPROTO_H_ struct vop_open_args { struct vnode *a_vp; int a_mode; struct ucred *a_cred; struct thread *a_td; }; #endif static int zfs_freebsd_open(struct vop_open_args *ap) { vnode_t *vp = ap->a_vp; znode_t *zp = VTOZ(vp); int error; error = zfs_open(&vp, ap->a_mode, ap->a_cred); if (error == 0) vnode_create_vobject(vp, zp->z_size, ap->a_td); return (error); } #ifndef _SYS_SYSPROTO_H_ struct vop_close_args { struct vnode *a_vp; int a_fflag; struct ucred *a_cred; struct thread *a_td; }; #endif static int zfs_freebsd_close(struct vop_close_args *ap) { return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred)); } #ifndef _SYS_SYSPROTO_H_ struct vop_ioctl_args { struct vnode *a_vp; ulong_t a_command; caddr_t a_data; int a_fflag; struct ucred *cred; struct thread *td; }; #endif static int zfs_freebsd_ioctl(struct vop_ioctl_args *ap) { return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data, ap->a_fflag, ap->a_cred, NULL)); } static int ioflags(int ioflags) { int flags = 0; if (ioflags & IO_APPEND) flags |= FAPPEND; if (ioflags & IO_NDELAY) flags |= FNONBLOCK; if (ioflags & IO_SYNC) flags |= (FSYNC | FDSYNC | FRSYNC); return (flags); } #ifndef _SYS_SYSPROTO_H_ struct vop_read_args { struct vnode *a_vp; struct uio *a_uio; int a_ioflag; struct ucred *a_cred; }; #endif static int zfs_freebsd_read(struct vop_read_args *ap) { zfs_uio_t uio; zfs_uio_init(&uio, ap->a_uio); return (zfs_read(VTOZ(ap->a_vp), &uio, ioflags(ap->a_ioflag), ap->a_cred)); } #ifndef _SYS_SYSPROTO_H_ struct vop_write_args { struct vnode *a_vp; struct uio *a_uio; int a_ioflag; struct ucred *a_cred; }; #endif static int zfs_freebsd_write(struct vop_write_args *ap) { zfs_uio_t uio; zfs_uio_init(&uio, ap->a_uio); return (zfs_write(VTOZ(ap->a_vp), &uio, ioflags(ap->a_ioflag), ap->a_cred)); } #if __FreeBSD_version >= 1300102 /* * VOP_FPLOOKUP_VEXEC routines are subject to special circumstances, see * the comment above cache_fplookup for details. */ static int zfs_freebsd_fplookup_vexec(struct vop_fplookup_vexec_args *v) { vnode_t *vp; znode_t *zp; uint64_t pflags; vp = v->a_vp; zp = VTOZ_SMR(vp); if (__predict_false(zp == NULL)) return (EAGAIN); pflags = atomic_load_64(&zp->z_pflags); if (pflags & ZFS_AV_QUARANTINED) return (EAGAIN); if (pflags & ZFS_XATTR) return (EAGAIN); if ((pflags & ZFS_NO_EXECS_DENIED) == 0) return (EAGAIN); return (0); } #endif #if __FreeBSD_version >= 1300139 static int zfs_freebsd_fplookup_symlink(struct vop_fplookup_symlink_args *v) { vnode_t *vp; znode_t *zp; char *target; vp = v->a_vp; zp = VTOZ_SMR(vp); if (__predict_false(zp == NULL)) { return (EAGAIN); } target = atomic_load_consume_ptr(&zp->z_cached_symlink); if (target == NULL) { return (EAGAIN); } return (cache_symlink_resolve(v->a_fpl, target, strlen(target))); } #endif #ifndef _SYS_SYSPROTO_H_ struct vop_access_args { struct vnode *a_vp; accmode_t a_accmode; struct ucred *a_cred; struct thread *a_td; }; #endif static int zfs_freebsd_access(struct vop_access_args *ap) { vnode_t *vp = ap->a_vp; znode_t *zp = VTOZ(vp); accmode_t accmode; int error = 0; if (ap->a_accmode == VEXEC) { if (zfs_fastaccesschk_execute(zp, ap->a_cred) == 0) return (0); } /* * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND, */ accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND); if (accmode != 0) error = zfs_access(zp, accmode, 0, ap->a_cred); /* * VADMIN has to be handled by vaccess(). */ if (error == 0) { accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND); if (accmode != 0) { #if __FreeBSD_version >= 1300105 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid, zp->z_gid, accmode, ap->a_cred); #else error = vaccess(vp->v_type, zp->z_mode, zp->z_uid, zp->z_gid, accmode, ap->a_cred, NULL); #endif } } /* * For VEXEC, ensure that at least one execute bit is set for * non-directories. */ if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR && (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) { error = EACCES; } return (error); } #ifndef _SYS_SYSPROTO_H_ struct vop_lookup_args { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; }; #endif static int zfs_freebsd_lookup(struct vop_lookup_args *ap, boolean_t cached) { struct componentname *cnp = ap->a_cnp; char nm[NAME_MAX + 1]; ASSERT3U(cnp->cn_namelen, <, sizeof (nm)); strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof (nm))); return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop, - cnp->cn_cred, cnp->cn_thread, 0, cached)); + cnp->cn_cred, curthread, 0, cached)); } static int zfs_freebsd_cachedlookup(struct vop_cachedlookup_args *ap) { return (zfs_freebsd_lookup((struct vop_lookup_args *)ap, B_TRUE)); } #ifndef _SYS_SYSPROTO_H_ struct vop_lookup_args { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; }; #endif static int zfs_cache_lookup(struct vop_lookup_args *ap) { zfsvfs_t *zfsvfs; zfsvfs = ap->a_dvp->v_mount->mnt_data; if (zfsvfs->z_use_namecache) return (vfs_cache_lookup(ap)); else return (zfs_freebsd_lookup(ap, B_FALSE)); } #ifndef _SYS_SYSPROTO_H_ struct vop_create_args { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; struct vattr *a_vap; }; #endif static int zfs_freebsd_create(struct vop_create_args *ap) { zfsvfs_t *zfsvfs; struct componentname *cnp = ap->a_cnp; vattr_t *vap = ap->a_vap; znode_t *zp = NULL; int rc, mode; ASSERT(cnp->cn_flags & SAVENAME); vattr_init_mask(vap); mode = vap->va_mode & ALLPERMS; zfsvfs = ap->a_dvp->v_mount->mnt_data; *ap->a_vpp = NULL; rc = zfs_create(VTOZ(ap->a_dvp), cnp->cn_nameptr, vap, !EXCL, mode, &zp, cnp->cn_cred, 0 /* flag */, NULL /* vsecattr */); if (rc == 0) *ap->a_vpp = ZTOV(zp); if (zfsvfs->z_use_namecache && rc == 0 && (cnp->cn_flags & MAKEENTRY) != 0) cache_enter(ap->a_dvp, *ap->a_vpp, cnp); return (rc); } #ifndef _SYS_SYSPROTO_H_ struct vop_remove_args { struct vnode *a_dvp; struct vnode *a_vp; struct componentname *a_cnp; }; #endif static int zfs_freebsd_remove(struct vop_remove_args *ap) { ASSERT(ap->a_cnp->cn_flags & SAVENAME); return (zfs_remove_(ap->a_dvp, ap->a_vp, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_cred)); } #ifndef _SYS_SYSPROTO_H_ struct vop_mkdir_args { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; struct vattr *a_vap; }; #endif static int zfs_freebsd_mkdir(struct vop_mkdir_args *ap) { vattr_t *vap = ap->a_vap; znode_t *zp = NULL; int rc; ASSERT(ap->a_cnp->cn_flags & SAVENAME); vattr_init_mask(vap); *ap->a_vpp = NULL; rc = zfs_mkdir(VTOZ(ap->a_dvp), ap->a_cnp->cn_nameptr, vap, &zp, ap->a_cnp->cn_cred, 0, NULL); if (rc == 0) *ap->a_vpp = ZTOV(zp); return (rc); } #ifndef _SYS_SYSPROTO_H_ struct vop_rmdir_args { struct vnode *a_dvp; struct vnode *a_vp; struct componentname *a_cnp; }; #endif static int zfs_freebsd_rmdir(struct vop_rmdir_args *ap) { struct componentname *cnp = ap->a_cnp; ASSERT(cnp->cn_flags & SAVENAME); return (zfs_rmdir_(ap->a_dvp, ap->a_vp, cnp->cn_nameptr, cnp->cn_cred)); } #ifndef _SYS_SYSPROTO_H_ struct vop_readdir_args { struct vnode *a_vp; struct uio *a_uio; struct ucred *a_cred; int *a_eofflag; int *a_ncookies; ulong_t **a_cookies; }; #endif static int zfs_freebsd_readdir(struct vop_readdir_args *ap) { zfs_uio_t uio; zfs_uio_init(&uio, ap->a_uio); return (zfs_readdir(ap->a_vp, &uio, ap->a_cred, ap->a_eofflag, ap->a_ncookies, ap->a_cookies)); } #ifndef _SYS_SYSPROTO_H_ struct vop_fsync_args { struct vnode *a_vp; int a_waitfor; struct thread *a_td; }; #endif static int zfs_freebsd_fsync(struct vop_fsync_args *ap) { vop_stdfsync(ap); return (zfs_fsync(VTOZ(ap->a_vp), 0, ap->a_td->td_ucred)); } #ifndef _SYS_SYSPROTO_H_ struct vop_getattr_args { struct vnode *a_vp; struct vattr *a_vap; struct ucred *a_cred; }; #endif static int zfs_freebsd_getattr(struct vop_getattr_args *ap) { vattr_t *vap = ap->a_vap; xvattr_t xvap; ulong_t fflags = 0; int error; xva_init(&xvap); xvap.xva_vattr = *vap; xvap.xva_vattr.va_mask |= AT_XVATTR; /* Convert chflags into ZFS-type flags. */ /* XXX: what about SF_SETTABLE?. */ XVA_SET_REQ(&xvap, XAT_IMMUTABLE); XVA_SET_REQ(&xvap, XAT_APPENDONLY); XVA_SET_REQ(&xvap, XAT_NOUNLINK); XVA_SET_REQ(&xvap, XAT_NODUMP); XVA_SET_REQ(&xvap, XAT_READONLY); XVA_SET_REQ(&xvap, XAT_ARCHIVE); XVA_SET_REQ(&xvap, XAT_SYSTEM); XVA_SET_REQ(&xvap, XAT_HIDDEN); XVA_SET_REQ(&xvap, XAT_REPARSE); XVA_SET_REQ(&xvap, XAT_OFFLINE); XVA_SET_REQ(&xvap, XAT_SPARSE); error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred); if (error != 0) return (error); /* Convert ZFS xattr into chflags. */ #define FLAG_CHECK(fflag, xflag, xfield) do { \ if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \ fflags |= (fflag); \ } while (0) FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE, xvap.xva_xoptattrs.xoa_immutable); FLAG_CHECK(SF_APPEND, XAT_APPENDONLY, xvap.xva_xoptattrs.xoa_appendonly); FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK, xvap.xva_xoptattrs.xoa_nounlink); FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE, xvap.xva_xoptattrs.xoa_archive); FLAG_CHECK(UF_NODUMP, XAT_NODUMP, xvap.xva_xoptattrs.xoa_nodump); FLAG_CHECK(UF_READONLY, XAT_READONLY, xvap.xva_xoptattrs.xoa_readonly); FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM, xvap.xva_xoptattrs.xoa_system); FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN, xvap.xva_xoptattrs.xoa_hidden); FLAG_CHECK(UF_REPARSE, XAT_REPARSE, xvap.xva_xoptattrs.xoa_reparse); FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE, xvap.xva_xoptattrs.xoa_offline); FLAG_CHECK(UF_SPARSE, XAT_SPARSE, xvap.xva_xoptattrs.xoa_sparse); #undef FLAG_CHECK *vap = xvap.xva_vattr; vap->va_flags = fflags; return (0); } #ifndef _SYS_SYSPROTO_H_ struct vop_setattr_args { struct vnode *a_vp; struct vattr *a_vap; struct ucred *a_cred; }; #endif static int zfs_freebsd_setattr(struct vop_setattr_args *ap) { vnode_t *vp = ap->a_vp; vattr_t *vap = ap->a_vap; cred_t *cred = ap->a_cred; xvattr_t xvap; ulong_t fflags; uint64_t zflags; vattr_init_mask(vap); vap->va_mask &= ~AT_NOSET; xva_init(&xvap); xvap.xva_vattr = *vap; zflags = VTOZ(vp)->z_pflags; if (vap->va_flags != VNOVAL) { zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs; int error; if (zfsvfs->z_use_fuids == B_FALSE) return (EOPNOTSUPP); fflags = vap->va_flags; /* * XXX KDM * We need to figure out whether it makes sense to allow * UF_REPARSE through, since we don't really have other * facilities to handle reparse points and zfs_setattr() * doesn't currently allow setting that attribute anyway. */ if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE| UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE| UF_OFFLINE|UF_SPARSE)) != 0) return (EOPNOTSUPP); /* * Unprivileged processes are not permitted to unset system * flags, or modify flags if any system flags are set. * Privileged non-jail processes may not modify system flags * if securelevel > 0 and any existing system flags are set. * Privileged jail processes behave like privileged non-jail * processes if the PR_ALLOW_CHFLAGS permission bit is set; * otherwise, they behave like unprivileged processes. */ if (secpolicy_fs_owner(vp->v_mount, cred) == 0 || spl_priv_check_cred(cred, PRIV_VFS_SYSFLAGS) == 0) { if (zflags & (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) { error = securelevel_gt(cred, 0); if (error != 0) return (error); } } else { /* * Callers may only modify the file flags on * objects they have VADMIN rights for. */ if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0) return (error); if (zflags & (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) { return (EPERM); } if (fflags & (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) { return (EPERM); } } #define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \ if (((fflags & (fflag)) && !(zflags & (zflag))) || \ ((zflags & (zflag)) && !(fflags & (fflag)))) { \ XVA_SET_REQ(&xvap, (xflag)); \ (xfield) = ((fflags & (fflag)) != 0); \ } \ } while (0) /* Convert chflags into ZFS-type flags. */ /* XXX: what about SF_SETTABLE?. */ FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE, xvap.xva_xoptattrs.xoa_immutable); FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY, xvap.xva_xoptattrs.xoa_appendonly); FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK, xvap.xva_xoptattrs.xoa_nounlink); FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE, xvap.xva_xoptattrs.xoa_archive); FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP, xvap.xva_xoptattrs.xoa_nodump); FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY, xvap.xva_xoptattrs.xoa_readonly); FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM, xvap.xva_xoptattrs.xoa_system); FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN, xvap.xva_xoptattrs.xoa_hidden); FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE, xvap.xva_xoptattrs.xoa_reparse); FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE, xvap.xva_xoptattrs.xoa_offline); FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE, xvap.xva_xoptattrs.xoa_sparse); #undef FLAG_CHANGE } if (vap->va_birthtime.tv_sec != VNOVAL) { xvap.xva_vattr.va_mask |= AT_XVATTR; XVA_SET_REQ(&xvap, XAT_CREATETIME); } return (zfs_setattr(VTOZ(vp), (vattr_t *)&xvap, 0, cred)); } #ifndef _SYS_SYSPROTO_H_ struct vop_rename_args { struct vnode *a_fdvp; struct vnode *a_fvp; struct componentname *a_fcnp; struct vnode *a_tdvp; struct vnode *a_tvp; struct componentname *a_tcnp; }; #endif static int zfs_freebsd_rename(struct vop_rename_args *ap) { vnode_t *fdvp = ap->a_fdvp; vnode_t *fvp = ap->a_fvp; vnode_t *tdvp = ap->a_tdvp; vnode_t *tvp = ap->a_tvp; int error; ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART)); ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART)); error = zfs_rename_(fdvp, &fvp, ap->a_fcnp, tdvp, &tvp, ap->a_tcnp, ap->a_fcnp->cn_cred, 1); vrele(fdvp); vrele(fvp); vrele(tdvp); if (tvp != NULL) vrele(tvp); return (error); } #ifndef _SYS_SYSPROTO_H_ struct vop_symlink_args { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; struct vattr *a_vap; char *a_target; }; #endif static int zfs_freebsd_symlink(struct vop_symlink_args *ap) { struct componentname *cnp = ap->a_cnp; vattr_t *vap = ap->a_vap; znode_t *zp = NULL; #if __FreeBSD_version >= 1300139 char *symlink; size_t symlink_len; #endif int rc; ASSERT(cnp->cn_flags & SAVENAME); vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */ vattr_init_mask(vap); *ap->a_vpp = NULL; rc = zfs_symlink(VTOZ(ap->a_dvp), cnp->cn_nameptr, vap, ap->a_target, &zp, cnp->cn_cred, 0 /* flags */); if (rc == 0) { *ap->a_vpp = ZTOV(zp); ASSERT_VOP_ELOCKED(ZTOV(zp), __func__); #if __FreeBSD_version >= 1300139 MPASS(zp->z_cached_symlink == NULL); symlink_len = strlen(ap->a_target); symlink = cache_symlink_alloc(symlink_len + 1, M_WAITOK); if (symlink != NULL) { memcpy(symlink, ap->a_target, symlink_len); symlink[symlink_len] = '\0'; atomic_store_rel_ptr((uintptr_t *)&zp->z_cached_symlink, (uintptr_t)symlink); } #endif } return (rc); } #ifndef _SYS_SYSPROTO_H_ struct vop_readlink_args { struct vnode *a_vp; struct uio *a_uio; struct ucred *a_cred; }; #endif static int zfs_freebsd_readlink(struct vop_readlink_args *ap) { zfs_uio_t uio; int error; #if __FreeBSD_version >= 1300139 znode_t *zp = VTOZ(ap->a_vp); char *symlink, *base; size_t symlink_len; bool trycache; #endif zfs_uio_init(&uio, ap->a_uio); #if __FreeBSD_version >= 1300139 trycache = false; if (zfs_uio_segflg(&uio) == UIO_SYSSPACE && zfs_uio_iovcnt(&uio) == 1) { base = zfs_uio_iovbase(&uio, 0); symlink_len = zfs_uio_iovlen(&uio, 0); trycache = true; } #endif error = zfs_readlink(ap->a_vp, &uio, ap->a_cred, NULL); #if __FreeBSD_version >= 1300139 if (atomic_load_ptr(&zp->z_cached_symlink) != NULL || error != 0 || !trycache) { return (error); } symlink_len -= zfs_uio_resid(&uio); symlink = cache_symlink_alloc(symlink_len + 1, M_WAITOK); if (symlink != NULL) { memcpy(symlink, base, symlink_len); symlink[symlink_len] = '\0'; if (!atomic_cmpset_rel_ptr((uintptr_t *)&zp->z_cached_symlink, (uintptr_t)NULL, (uintptr_t)symlink)) { cache_symlink_free(symlink, symlink_len + 1); } } #endif return (error); } #ifndef _SYS_SYSPROTO_H_ struct vop_link_args { struct vnode *a_tdvp; struct vnode *a_vp; struct componentname *a_cnp; }; #endif static int zfs_freebsd_link(struct vop_link_args *ap) { struct componentname *cnp = ap->a_cnp; vnode_t *vp = ap->a_vp; vnode_t *tdvp = ap->a_tdvp; if (tdvp->v_mount != vp->v_mount) return (EXDEV); ASSERT(cnp->cn_flags & SAVENAME); return (zfs_link(VTOZ(tdvp), VTOZ(vp), cnp->cn_nameptr, cnp->cn_cred, 0)); } #ifndef _SYS_SYSPROTO_H_ struct vop_inactive_args { struct vnode *a_vp; struct thread *a_td; }; #endif static int zfs_freebsd_inactive(struct vop_inactive_args *ap) { vnode_t *vp = ap->a_vp; #if __FreeBSD_version >= 1300123 zfs_inactive(vp, curthread->td_ucred, NULL); #else zfs_inactive(vp, ap->a_td->td_ucred, NULL); #endif return (0); } #if __FreeBSD_version >= 1300042 #ifndef _SYS_SYSPROTO_H_ struct vop_need_inactive_args { struct vnode *a_vp; struct thread *a_td; }; #endif static int zfs_freebsd_need_inactive(struct vop_need_inactive_args *ap) { vnode_t *vp = ap->a_vp; znode_t *zp = VTOZ(vp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; int need; if (vn_need_pageq_flush(vp)) return (1); if (!ZFS_TEARDOWN_INACTIVE_TRY_ENTER_READ(zfsvfs)) return (1); need = (zp->z_sa_hdl == NULL || zp->z_unlinked || zp->z_atime_dirty); ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs); return (need); } #endif #ifndef _SYS_SYSPROTO_H_ struct vop_reclaim_args { struct vnode *a_vp; struct thread *a_td; }; #endif static int zfs_freebsd_reclaim(struct vop_reclaim_args *ap) { vnode_t *vp = ap->a_vp; znode_t *zp = VTOZ(vp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; ASSERT3P(zp, !=, NULL); #if __FreeBSD_version < 1300042 /* Destroy the vm object and flush associated pages. */ vnode_destroy_vobject(vp); #endif /* * z_teardown_inactive_lock protects from a race with * zfs_znode_dmu_fini in zfsvfs_teardown during * force unmount. */ ZFS_TEARDOWN_INACTIVE_ENTER_READ(zfsvfs); if (zp->z_sa_hdl == NULL) zfs_znode_free(zp); else zfs_zinactive(zp); ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs); vp->v_data = NULL; return (0); } #ifndef _SYS_SYSPROTO_H_ struct vop_fid_args { struct vnode *a_vp; struct fid *a_fid; }; #endif static int zfs_freebsd_fid(struct vop_fid_args *ap) { return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL)); } #ifndef _SYS_SYSPROTO_H_ struct vop_pathconf_args { struct vnode *a_vp; int a_name; register_t *a_retval; } *ap; #endif static int zfs_freebsd_pathconf(struct vop_pathconf_args *ap) { ulong_t val; int error; error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL); if (error == 0) { *ap->a_retval = val; return (error); } if (error != EOPNOTSUPP) return (error); switch (ap->a_name) { case _PC_NAME_MAX: *ap->a_retval = NAME_MAX; return (0); #if __FreeBSD_version >= 1400032 case _PC_DEALLOC_PRESENT: *ap->a_retval = 1; return (0); #endif case _PC_PIPE_BUF: if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO) { *ap->a_retval = PIPE_BUF; return (0); } return (EINVAL); default: return (vop_stdpathconf(ap)); } } /* * FreeBSD's extended attributes namespace defines file name prefix for ZFS' * extended attribute name: * * NAMESPACE PREFIX * system freebsd:system: * user (none, can be used to access ZFS fsattr(5) attributes * created on Solaris) */ static int zfs_create_attrname(int attrnamespace, const char *name, char *attrname, size_t size) { const char *namespace, *prefix, *suffix; /* We don't allow '/' character in attribute name. */ if (strchr(name, '/') != NULL) return (SET_ERROR(EINVAL)); /* We don't allow attribute names that start with "freebsd:" string. */ if (strncmp(name, "freebsd:", 8) == 0) return (SET_ERROR(EINVAL)); bzero(attrname, size); switch (attrnamespace) { case EXTATTR_NAMESPACE_USER: #if 0 prefix = "freebsd:"; namespace = EXTATTR_NAMESPACE_USER_STRING; suffix = ":"; #else /* * This is the default namespace by which we can access all * attributes created on Solaris. */ prefix = namespace = suffix = ""; #endif break; case EXTATTR_NAMESPACE_SYSTEM: prefix = "freebsd:"; namespace = EXTATTR_NAMESPACE_SYSTEM_STRING; suffix = ":"; break; case EXTATTR_NAMESPACE_EMPTY: default: return (SET_ERROR(EINVAL)); } if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix, name) >= size) { return (SET_ERROR(ENAMETOOLONG)); } return (0); } static int zfs_ensure_xattr_cached(znode_t *zp) { int error = 0; ASSERT(RW_LOCK_HELD(&zp->z_xattr_lock)); if (zp->z_xattr_cached != NULL) return (0); if (rw_write_held(&zp->z_xattr_lock)) return (zfs_sa_get_xattr(zp)); if (!rw_tryupgrade(&zp->z_xattr_lock)) { rw_exit(&zp->z_xattr_lock); rw_enter(&zp->z_xattr_lock, RW_WRITER); } if (zp->z_xattr_cached == NULL) error = zfs_sa_get_xattr(zp); rw_downgrade(&zp->z_xattr_lock); return (error); } #ifndef _SYS_SYSPROTO_H_ struct vop_getextattr { IN struct vnode *a_vp; IN int a_attrnamespace; IN const char *a_name; INOUT struct uio *a_uio; OUT size_t *a_size; IN struct ucred *a_cred; IN struct thread *a_td; }; #endif static int zfs_getextattr_dir(struct vop_getextattr_args *ap, const char *attrname) { struct thread *td = ap->a_td; struct nameidata nd; struct vattr va; vnode_t *xvp = NULL, *vp; int error, flags; error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, LOOKUP_XATTR, B_FALSE); if (error != 0) return (error); flags = FREAD; NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, xvp, td); error = vn_open_cred(&nd, &flags, 0, VN_OPEN_INVFS, ap->a_cred, NULL); vp = nd.ni_vp; NDFREE(&nd, NDF_ONLY_PNBUF); if (error != 0) return (SET_ERROR(error)); if (ap->a_size != NULL) { error = VOP_GETATTR(vp, &va, ap->a_cred); if (error == 0) *ap->a_size = (size_t)va.va_size; } else if (ap->a_uio != NULL) error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred); VOP_UNLOCK1(vp); vn_close(vp, flags, ap->a_cred, td); return (error); } static int zfs_getextattr_sa(struct vop_getextattr_args *ap, const char *attrname) { znode_t *zp = VTOZ(ap->a_vp); uchar_t *nv_value; uint_t nv_size; int error; error = zfs_ensure_xattr_cached(zp); if (error != 0) return (error); ASSERT(RW_LOCK_HELD(&zp->z_xattr_lock)); ASSERT3P(zp->z_xattr_cached, !=, NULL); error = nvlist_lookup_byte_array(zp->z_xattr_cached, attrname, &nv_value, &nv_size); if (error != 0) return (SET_ERROR(error)); if (ap->a_size != NULL) *ap->a_size = nv_size; else if (ap->a_uio != NULL) error = uiomove(nv_value, nv_size, ap->a_uio); if (error != 0) return (SET_ERROR(error)); return (0); } /* * Vnode operation to retrieve a named extended attribute. */ static int zfs_getextattr(struct vop_getextattr_args *ap) { znode_t *zp = VTOZ(ap->a_vp); zfsvfs_t *zfsvfs = ZTOZSB(zp); char attrname[EXTATTR_MAXNAMELEN+1]; int error; /* * If the xattr property is off, refuse the request. */ if (!(zfsvfs->z_flags & ZSB_XATTR)) return (SET_ERROR(EOPNOTSUPP)); error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, ap->a_cred, ap->a_td, VREAD); if (error != 0) return (SET_ERROR(error)); error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, sizeof (attrname)); if (error != 0) return (error); error = ENOENT; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp) rw_enter(&zp->z_xattr_lock, RW_READER); if (zfsvfs->z_use_sa && zp->z_is_sa) error = zfs_getextattr_sa(ap, attrname); if (error == ENOENT) error = zfs_getextattr_dir(ap, attrname); rw_exit(&zp->z_xattr_lock); ZFS_EXIT(zfsvfs); if (error == ENOENT) error = SET_ERROR(ENOATTR); return (error); } #ifndef _SYS_SYSPROTO_H_ struct vop_deleteextattr { IN struct vnode *a_vp; IN int a_attrnamespace; IN const char *a_name; IN struct ucred *a_cred; IN struct thread *a_td; }; #endif static int zfs_deleteextattr_dir(struct vop_deleteextattr_args *ap, const char *attrname) { struct thread *td = ap->a_td; struct nameidata nd; vnode_t *xvp = NULL, *vp; int error; error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, LOOKUP_XATTR, B_FALSE); if (error != 0) return (error); NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF, UIO_SYSSPACE, attrname, xvp, td); error = namei(&nd); vp = nd.ni_vp; if (error != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); return (SET_ERROR(error)); } error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd); NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if (vp == nd.ni_dvp) vrele(vp); else vput(vp); return (error); } static int zfs_deleteextattr_sa(struct vop_deleteextattr_args *ap, const char *attrname) { znode_t *zp = VTOZ(ap->a_vp); nvlist_t *nvl; int error; error = zfs_ensure_xattr_cached(zp); if (error != 0) return (error); ASSERT(RW_WRITE_HELD(&zp->z_xattr_lock)); ASSERT3P(zp->z_xattr_cached, !=, NULL); nvl = zp->z_xattr_cached; error = nvlist_remove(nvl, attrname, DATA_TYPE_BYTE_ARRAY); if (error != 0) error = SET_ERROR(error); else error = zfs_sa_set_xattr(zp); if (error != 0) { zp->z_xattr_cached = NULL; nvlist_free(nvl); } return (error); } /* * Vnode operation to remove a named attribute. */ static int zfs_deleteextattr(struct vop_deleteextattr_args *ap) { znode_t *zp = VTOZ(ap->a_vp); zfsvfs_t *zfsvfs = ZTOZSB(zp); char attrname[EXTATTR_MAXNAMELEN+1]; int error; /* * If the xattr property is off, refuse the request. */ if (!(zfsvfs->z_flags & ZSB_XATTR)) return (SET_ERROR(EOPNOTSUPP)); error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, ap->a_cred, ap->a_td, VWRITE); if (error != 0) return (SET_ERROR(error)); error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, sizeof (attrname)); if (error != 0) return (error); size_t size = 0; struct vop_getextattr_args vga = { .a_vp = ap->a_vp, .a_size = &size, .a_cred = ap->a_cred, .a_td = ap->a_td, }; error = ENOENT; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); rw_enter(&zp->z_xattr_lock, RW_WRITER); if (zfsvfs->z_use_sa && zp->z_is_sa) { error = zfs_getextattr_sa(&vga, attrname); if (error == 0) error = zfs_deleteextattr_sa(ap, attrname); } if (error == ENOENT) { error = zfs_getextattr_dir(&vga, attrname); if (error == 0) error = zfs_deleteextattr_dir(ap, attrname); } rw_exit(&zp->z_xattr_lock); ZFS_EXIT(zfsvfs); if (error == ENOENT) error = SET_ERROR(ENOATTR); return (error); } #ifndef _SYS_SYSPROTO_H_ struct vop_setextattr { IN struct vnode *a_vp; IN int a_attrnamespace; IN const char *a_name; INOUT struct uio *a_uio; IN struct ucred *a_cred; IN struct thread *a_td; }; #endif static int zfs_setextattr_dir(struct vop_setextattr_args *ap, const char *attrname) { struct thread *td = ap->a_td; struct nameidata nd; struct vattr va; vnode_t *xvp = NULL, *vp; int error, flags; error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, LOOKUP_XATTR | CREATE_XATTR_DIR, B_FALSE); if (error != 0) return (error); flags = FFLAGS(O_WRONLY | O_CREAT); NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, xvp, td); error = vn_open_cred(&nd, &flags, 0600, VN_OPEN_INVFS, ap->a_cred, NULL); vp = nd.ni_vp; NDFREE(&nd, NDF_ONLY_PNBUF); if (error != 0) return (SET_ERROR(error)); VATTR_NULL(&va); va.va_size = 0; error = VOP_SETATTR(vp, &va, ap->a_cred); if (error == 0) VOP_WRITE(vp, ap->a_uio, IO_UNIT, ap->a_cred); VOP_UNLOCK1(vp); vn_close(vp, flags, ap->a_cred, td); return (error); } static int zfs_setextattr_sa(struct vop_setextattr_args *ap, const char *attrname) { znode_t *zp = VTOZ(ap->a_vp); nvlist_t *nvl; size_t sa_size; int error; error = zfs_ensure_xattr_cached(zp); if (error != 0) return (error); ASSERT(RW_WRITE_HELD(&zp->z_xattr_lock)); ASSERT3P(zp->z_xattr_cached, !=, NULL); nvl = zp->z_xattr_cached; size_t entry_size = ap->a_uio->uio_resid; if (entry_size > DXATTR_MAX_ENTRY_SIZE) return (SET_ERROR(EFBIG)); error = nvlist_size(nvl, &sa_size, NV_ENCODE_XDR); if (error != 0) return (SET_ERROR(error)); if (sa_size > DXATTR_MAX_SA_SIZE) return (SET_ERROR(EFBIG)); uchar_t *buf = kmem_alloc(entry_size, KM_SLEEP); error = uiomove(buf, entry_size, ap->a_uio); if (error != 0) { error = SET_ERROR(error); } else { error = nvlist_add_byte_array(nvl, attrname, buf, entry_size); if (error != 0) error = SET_ERROR(error); } kmem_free(buf, entry_size); if (error == 0) error = zfs_sa_set_xattr(zp); if (error != 0) { zp->z_xattr_cached = NULL; nvlist_free(nvl); } return (error); } /* * Vnode operation to set a named attribute. */ static int zfs_setextattr(struct vop_setextattr_args *ap) { znode_t *zp = VTOZ(ap->a_vp); zfsvfs_t *zfsvfs = ZTOZSB(zp); char attrname[EXTATTR_MAXNAMELEN+1]; int error; /* * If the xattr property is off, refuse the request. */ if (!(zfsvfs->z_flags & ZSB_XATTR)) return (SET_ERROR(EOPNOTSUPP)); error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, ap->a_cred, ap->a_td, VWRITE); if (error != 0) return (SET_ERROR(error)); error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, sizeof (attrname)); if (error != 0) return (error); struct vop_deleteextattr_args vda = { .a_vp = ap->a_vp, .a_cred = ap->a_cred, .a_td = ap->a_td, }; error = ENOENT; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); rw_enter(&zp->z_xattr_lock, RW_WRITER); if (zfsvfs->z_use_sa && zp->z_is_sa && zfsvfs->z_xattr_sa) { error = zfs_setextattr_sa(ap, attrname); if (error == 0) /* * Successfully put into SA, we need to clear the one * in dir if present. */ zfs_deleteextattr_dir(&vda, attrname); } if (error) { error = zfs_setextattr_dir(ap, attrname); if (error == 0 && zp->z_is_sa) /* * Successfully put into dir, we need to clear the one * in SA if present. */ zfs_deleteextattr_sa(&vda, attrname); } rw_exit(&zp->z_xattr_lock); ZFS_EXIT(zfsvfs); return (error); } #ifndef _SYS_SYSPROTO_H_ struct vop_listextattr { IN struct vnode *a_vp; IN int a_attrnamespace; INOUT struct uio *a_uio; OUT size_t *a_size; IN struct ucred *a_cred; IN struct thread *a_td; }; #endif static int zfs_listextattr_dir(struct vop_listextattr_args *ap, const char *attrprefix) { struct thread *td = ap->a_td; struct nameidata nd; uint8_t dirbuf[sizeof (struct dirent)]; struct iovec aiov; struct uio auio; vnode_t *xvp = NULL, *vp; int error, eof; error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, LOOKUP_XATTR, B_FALSE); if (error != 0) { /* * ENOATTR means that the EA directory does not yet exist, * i.e. there are no extended attributes there. */ if (error == ENOATTR) error = 0; return (error); } NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED, UIO_SYSSPACE, ".", xvp, td); error = namei(&nd); vp = nd.ni_vp; NDFREE(&nd, NDF_ONLY_PNBUF); if (error != 0) return (SET_ERROR(error)); auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_segflg = UIO_SYSSPACE; auio.uio_td = td; auio.uio_rw = UIO_READ; auio.uio_offset = 0; size_t plen = strlen(attrprefix); do { aiov.iov_base = (void *)dirbuf; aiov.iov_len = sizeof (dirbuf); auio.uio_resid = sizeof (dirbuf); error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL); if (error != 0) break; int done = sizeof (dirbuf) - auio.uio_resid; for (int pos = 0; pos < done; ) { struct dirent *dp = (struct dirent *)(dirbuf + pos); pos += dp->d_reclen; /* * XXX: Temporarily we also accept DT_UNKNOWN, as this * is what we get when attribute was created on Solaris. */ if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN) continue; else if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0) continue; else if (strncmp(dp->d_name, attrprefix, plen) != 0) continue; uint8_t nlen = dp->d_namlen - plen; if (ap->a_size != NULL) { *ap->a_size += 1 + nlen; } else if (ap->a_uio != NULL) { /* * Format of extattr name entry is one byte for * length and the rest for name. */ error = uiomove(&nlen, 1, ap->a_uio); if (error == 0) { char *namep = dp->d_name + plen; error = uiomove(namep, nlen, ap->a_uio); } if (error != 0) { error = SET_ERROR(error); break; } } } } while (!eof && error == 0); vput(vp); return (error); } static int zfs_listextattr_sa(struct vop_listextattr_args *ap, const char *attrprefix) { znode_t *zp = VTOZ(ap->a_vp); int error; error = zfs_ensure_xattr_cached(zp); if (error != 0) return (error); ASSERT(RW_LOCK_HELD(&zp->z_xattr_lock)); ASSERT3P(zp->z_xattr_cached, !=, NULL); size_t plen = strlen(attrprefix); nvpair_t *nvp = NULL; while ((nvp = nvlist_next_nvpair(zp->z_xattr_cached, nvp)) != NULL) { ASSERT3U(nvpair_type(nvp), ==, DATA_TYPE_BYTE_ARRAY); const char *name = nvpair_name(nvp); if (plen == 0 && strncmp(name, "freebsd:", 8) == 0) continue; else if (strncmp(name, attrprefix, plen) != 0) continue; uint8_t nlen = strlen(name) - plen; if (ap->a_size != NULL) { *ap->a_size += 1 + nlen; } else if (ap->a_uio != NULL) { /* * Format of extattr name entry is one byte for * length and the rest for name. */ error = uiomove(&nlen, 1, ap->a_uio); if (error == 0) { char *namep = __DECONST(char *, name) + plen; error = uiomove(namep, nlen, ap->a_uio); } if (error != 0) { error = SET_ERROR(error); break; } } } return (error); } /* * Vnode operation to retrieve extended attributes on a vnode. */ static int zfs_listextattr(struct vop_listextattr_args *ap) { znode_t *zp = VTOZ(ap->a_vp); zfsvfs_t *zfsvfs = ZTOZSB(zp); char attrprefix[16]; int error; if (ap->a_size != NULL) *ap->a_size = 0; /* * If the xattr property is off, refuse the request. */ if (!(zfsvfs->z_flags & ZSB_XATTR)) return (SET_ERROR(EOPNOTSUPP)); error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, ap->a_cred, ap->a_td, VREAD); if (error != 0) return (SET_ERROR(error)); error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix, sizeof (attrprefix)); if (error != 0) return (error); ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); rw_enter(&zp->z_xattr_lock, RW_READER); if (zfsvfs->z_use_sa && zp->z_is_sa) error = zfs_listextattr_sa(ap, attrprefix); if (error == 0) error = zfs_listextattr_dir(ap, attrprefix); rw_exit(&zp->z_xattr_lock); ZFS_EXIT(zfsvfs); return (error); } #ifndef _SYS_SYSPROTO_H_ struct vop_getacl_args { struct vnode *vp; acl_type_t type; struct acl *aclp; struct ucred *cred; struct thread *td; }; #endif static int zfs_freebsd_getacl(struct vop_getacl_args *ap) { int error; vsecattr_t vsecattr; if (ap->a_type != ACL_TYPE_NFS4) return (EINVAL); vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT; if ((error = zfs_getsecattr(VTOZ(ap->a_vp), &vsecattr, 0, ap->a_cred))) return (error); error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt); if (vsecattr.vsa_aclentp != NULL) kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz); return (error); } #ifndef _SYS_SYSPROTO_H_ struct vop_setacl_args { struct vnode *vp; acl_type_t type; struct acl *aclp; struct ucred *cred; struct thread *td; }; #endif static int zfs_freebsd_setacl(struct vop_setacl_args *ap) { int error; vsecattr_t vsecattr; int aclbsize; /* size of acl list in bytes */ aclent_t *aaclp; if (ap->a_type != ACL_TYPE_NFS4) return (EINVAL); if (ap->a_aclp == NULL) return (EINVAL); if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES) return (EINVAL); /* * With NFSv4 ACLs, chmod(2) may need to add additional entries, * splitting every entry into two and appending "canonical six" * entries at the end. Don't allow for setting an ACL that would * cause chmod(2) to run out of ACL entries. */ if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES) return (ENOSPC); error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR); if (error != 0) return (error); vsecattr.vsa_mask = VSA_ACE; aclbsize = ap->a_aclp->acl_cnt * sizeof (ace_t); vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP); aaclp = vsecattr.vsa_aclentp; vsecattr.vsa_aclentsz = aclbsize; aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp); error = zfs_setsecattr(VTOZ(ap->a_vp), &vsecattr, 0, ap->a_cred); kmem_free(aaclp, aclbsize); return (error); } #ifndef _SYS_SYSPROTO_H_ struct vop_aclcheck_args { struct vnode *vp; acl_type_t type; struct acl *aclp; struct ucred *cred; struct thread *td; }; #endif static int zfs_freebsd_aclcheck(struct vop_aclcheck_args *ap) { return (EOPNOTSUPP); } static int zfs_vptocnp(struct vop_vptocnp_args *ap) { vnode_t *covered_vp; vnode_t *vp = ap->a_vp; zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; znode_t *zp = VTOZ(vp); int ltype; int error; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); /* * If we are a snapshot mounted under .zfs, run the operation * on the covered vnode. */ if (zp->z_id != zfsvfs->z_root || zfsvfs->z_parent == zfsvfs) { char name[MAXNAMLEN + 1]; znode_t *dzp; size_t len; error = zfs_znode_parent_and_name(zp, &dzp, name); if (error == 0) { len = strlen(name); if (*ap->a_buflen < len) error = SET_ERROR(ENOMEM); } if (error == 0) { *ap->a_buflen -= len; bcopy(name, ap->a_buf + *ap->a_buflen, len); *ap->a_vpp = ZTOV(dzp); } ZFS_EXIT(zfsvfs); return (error); } ZFS_EXIT(zfsvfs); covered_vp = vp->v_mount->mnt_vnodecovered; #if __FreeBSD_version >= 1300045 enum vgetstate vs = vget_prep(covered_vp); #else vhold(covered_vp); #endif ltype = VOP_ISLOCKED(vp); VOP_UNLOCK1(vp); #if __FreeBSD_version >= 1300045 error = vget_finish(covered_vp, LK_SHARED, vs); #else error = vget(covered_vp, LK_SHARED | LK_VNHELD, curthread); #endif if (error == 0) { #if __FreeBSD_version >= 1300123 error = VOP_VPTOCNP(covered_vp, ap->a_vpp, ap->a_buf, ap->a_buflen); #else error = VOP_VPTOCNP(covered_vp, ap->a_vpp, ap->a_cred, ap->a_buf, ap->a_buflen); #endif vput(covered_vp); } vn_lock(vp, ltype | LK_RETRY); if (VN_IS_DOOMED(vp)) error = SET_ERROR(ENOENT); return (error); } #if __FreeBSD_version >= 1400032 static int zfs_deallocate(struct vop_deallocate_args *ap) { znode_t *zp = VTOZ(ap->a_vp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; zilog_t *zilog; off_t off, len, file_sz; int error; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); /* * Callers might not be able to detect properly that we are read-only, * so check it explicitly here. */ if (zfs_is_readonly(zfsvfs)) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EROFS)); } zilog = zfsvfs->z_log; off = *ap->a_offset; len = *ap->a_len; file_sz = zp->z_size; if (off + len > file_sz) len = file_sz - off; /* Fast path for out-of-range request. */ if (len <= 0) { *ap->a_len = 0; ZFS_EXIT(zfsvfs); return (0); } error = zfs_freesp(zp, off, len, O_RDWR, TRUE); if (error == 0) { if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS || (ap->a_ioflag & IO_SYNC) != 0) zil_commit(zilog, zp->z_id); *ap->a_offset = off + len; *ap->a_len = 0; } ZFS_EXIT(zfsvfs); return (error); } #endif struct vop_vector zfs_vnodeops; struct vop_vector zfs_fifoops; struct vop_vector zfs_shareops; struct vop_vector zfs_vnodeops = { .vop_default = &default_vnodeops, .vop_inactive = zfs_freebsd_inactive, #if __FreeBSD_version >= 1300042 .vop_need_inactive = zfs_freebsd_need_inactive, #endif .vop_reclaim = zfs_freebsd_reclaim, #if __FreeBSD_version >= 1300102 .vop_fplookup_vexec = zfs_freebsd_fplookup_vexec, #endif #if __FreeBSD_version >= 1300139 .vop_fplookup_symlink = zfs_freebsd_fplookup_symlink, #endif .vop_access = zfs_freebsd_access, .vop_allocate = VOP_EINVAL, #if __FreeBSD_version >= 1400032 .vop_deallocate = zfs_deallocate, #endif .vop_lookup = zfs_cache_lookup, .vop_cachedlookup = zfs_freebsd_cachedlookup, .vop_getattr = zfs_freebsd_getattr, .vop_setattr = zfs_freebsd_setattr, .vop_create = zfs_freebsd_create, .vop_mknod = (vop_mknod_t *)zfs_freebsd_create, .vop_mkdir = zfs_freebsd_mkdir, .vop_readdir = zfs_freebsd_readdir, .vop_fsync = zfs_freebsd_fsync, .vop_open = zfs_freebsd_open, .vop_close = zfs_freebsd_close, .vop_rmdir = zfs_freebsd_rmdir, .vop_ioctl = zfs_freebsd_ioctl, .vop_link = zfs_freebsd_link, .vop_symlink = zfs_freebsd_symlink, .vop_readlink = zfs_freebsd_readlink, .vop_read = zfs_freebsd_read, .vop_write = zfs_freebsd_write, .vop_remove = zfs_freebsd_remove, .vop_rename = zfs_freebsd_rename, .vop_pathconf = zfs_freebsd_pathconf, .vop_bmap = zfs_freebsd_bmap, .vop_fid = zfs_freebsd_fid, .vop_getextattr = zfs_getextattr, .vop_deleteextattr = zfs_deleteextattr, .vop_setextattr = zfs_setextattr, .vop_listextattr = zfs_listextattr, .vop_getacl = zfs_freebsd_getacl, .vop_setacl = zfs_freebsd_setacl, .vop_aclcheck = zfs_freebsd_aclcheck, .vop_getpages = zfs_freebsd_getpages, .vop_putpages = zfs_freebsd_putpages, .vop_vptocnp = zfs_vptocnp, #if __FreeBSD_version >= 1300064 .vop_lock1 = vop_lock, .vop_unlock = vop_unlock, .vop_islocked = vop_islocked, #endif }; VFS_VOP_VECTOR_REGISTER(zfs_vnodeops); struct vop_vector zfs_fifoops = { .vop_default = &fifo_specops, .vop_fsync = zfs_freebsd_fsync, #if __FreeBSD_version >= 1300102 .vop_fplookup_vexec = zfs_freebsd_fplookup_vexec, #endif #if __FreeBSD_version >= 1300139 .vop_fplookup_symlink = zfs_freebsd_fplookup_symlink, #endif .vop_access = zfs_freebsd_access, .vop_getattr = zfs_freebsd_getattr, .vop_inactive = zfs_freebsd_inactive, .vop_read = VOP_PANIC, .vop_reclaim = zfs_freebsd_reclaim, .vop_setattr = zfs_freebsd_setattr, .vop_write = VOP_PANIC, .vop_pathconf = zfs_freebsd_pathconf, .vop_fid = zfs_freebsd_fid, .vop_getacl = zfs_freebsd_getacl, .vop_setacl = zfs_freebsd_setacl, .vop_aclcheck = zfs_freebsd_aclcheck, }; VFS_VOP_VECTOR_REGISTER(zfs_fifoops); /* * special share hidden files vnode operations template */ struct vop_vector zfs_shareops = { .vop_default = &default_vnodeops, #if __FreeBSD_version >= 1300121 .vop_fplookup_vexec = VOP_EAGAIN, #endif #if __FreeBSD_version >= 1300139 .vop_fplookup_symlink = VOP_EAGAIN, #endif .vop_access = zfs_freebsd_access, .vop_inactive = zfs_freebsd_inactive, .vop_reclaim = zfs_freebsd_reclaim, .vop_fid = zfs_freebsd_fid, .vop_pathconf = zfs_freebsd_pathconf, }; VFS_VOP_VECTOR_REGISTER(zfs_shareops); diff --git a/sys/fs/autofs/autofs_vnops.c b/sys/fs/autofs/autofs_vnops.c index 13daa950d75b..c06d6a24b9b8 100644 --- a/sys/fs/autofs/autofs_vnops.c +++ b/sys/fs/autofs/autofs_vnops.c @@ -1,714 +1,714 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2014 The FreeBSD Foundation * * This software was developed by Edward Tomasz Napierala under sponsorship * from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static int autofs_trigger_vn(struct vnode *vp, const char *path, int pathlen, struct vnode **newvp); extern struct autofs_softc *autofs_softc; static int autofs_access(struct vop_access_args *ap) { /* * Nothing to do here; the only kind of access control * needed is in autofs_mkdir(). */ return (0); } static int autofs_getattr(struct vop_getattr_args *ap) { struct vnode *vp, *newvp; struct autofs_node *anp; struct mount *mp; struct vattr *vap; int error; vp = ap->a_vp; anp = vp->v_data; mp = vp->v_mount; vap = ap->a_vap; KASSERT(ap->a_vp->v_type == VDIR, ("!VDIR")); /* * The reason we must do this is that some tree-walking software, * namely fts(3), assumes that stat(".") results will not change * between chdir("subdir") and chdir(".."), and fails with ENOENT * otherwise. */ if (autofs_mount_on_stat && autofs_cached(anp, NULL, 0) == false && autofs_ignore_thread(curthread) == false) { error = autofs_trigger_vn(vp, "", 0, &newvp); if (error != 0) return (error); if (newvp != NULL) { error = VOP_GETATTR(newvp, ap->a_vap, ap->a_cred); vput(newvp); return (error); } } vap->va_type = VDIR; vap->va_mode = 0755; vap->va_nlink = 3; /* XXX */ vap->va_uid = 0; vap->va_gid = 0; vap->va_rdev = NODEV; vap->va_fsid = mp->mnt_stat.f_fsid.val[0]; vap->va_fileid = anp->an_fileno; vap->va_size = S_BLKSIZE; vap->va_blocksize = S_BLKSIZE; vap->va_mtime = anp->an_ctime; vap->va_atime = anp->an_ctime; vap->va_ctime = anp->an_ctime; vap->va_birthtime = anp->an_ctime; vap->va_gen = 0; vap->va_flags = 0; vap->va_rdev = 0; vap->va_bytes = S_BLKSIZE; vap->va_filerev = 0; vap->va_spare = 0; return (0); } /* * Unlock the vnode, request automountd(8) action, and then lock it back. * If anything got mounted on top of the vnode, return the new filesystem's * root vnode in 'newvp', locked. */ static int autofs_trigger_vn(struct vnode *vp, const char *path, int pathlen, struct vnode **newvp) { struct autofs_node *anp; int error, lock_flags; anp = vp->v_data; /* * Release the vnode lock, so that other operations, in partcular * mounting a filesystem on top of it, can proceed. Increase use * count, to prevent the vnode from being deallocated and to prevent * filesystem from being unmounted. */ lock_flags = VOP_ISLOCKED(vp); vref(vp); VOP_UNLOCK(vp); sx_xlock(&autofs_softc->sc_lock); /* * XXX: Workaround for mounting the same thing multiple times; revisit. */ if (vp->v_mountedhere != NULL) { error = 0; goto mounted; } error = autofs_trigger(anp, path, pathlen); mounted: sx_xunlock(&autofs_softc->sc_lock); vn_lock(vp, lock_flags | LK_RETRY); vunref(vp); if (VN_IS_DOOMED(vp)) { AUTOFS_DEBUG("VIRF_DOOMED"); return (ENOENT); } if (error != 0) return (error); if (vp->v_mountedhere == NULL) { *newvp = NULL; return (0); } else { /* * If the operation that succeeded was mount, then mark * the node as non-cached. Otherwise, if someone unmounts * the filesystem before the cache times out, we will fail * to trigger. */ anp->an_cached = false; } error = VFS_ROOT(vp->v_mountedhere, lock_flags, newvp); if (error != 0) { AUTOFS_WARN("VFS_ROOT() failed with error %d", error); return (error); } return (0); } static int autofs_vget_callback(struct mount *mp, void *arg, int flags, struct vnode **vpp) { return (autofs_node_vn(arg, mp, flags, vpp)); } static int autofs_lookup(struct vop_lookup_args *ap) { struct vnode *dvp, *newvp, **vpp; struct mount *mp; struct autofs_mount *amp; struct autofs_node *anp, *child; struct componentname *cnp; int error; dvp = ap->a_dvp; vpp = ap->a_vpp; mp = dvp->v_mount; amp = VFSTOAUTOFS(mp); anp = dvp->v_data; cnp = ap->a_cnp; if (cnp->cn_flags & ISDOTDOT) { KASSERT(anp->an_parent != NULL, ("NULL parent")); /* * Note that in this case, dvp is the child vnode, and we * are looking up the parent vnode - exactly reverse from * normal operation. Unlocking dvp requires some rather * tricky unlock/relock dance to prevent mp from being freed; * use vn_vget_ino_gen() which takes care of all that. */ error = vn_vget_ino_gen(dvp, autofs_vget_callback, anp->an_parent, cnp->cn_lkflags, vpp); if (error != 0) { AUTOFS_WARN("vn_vget_ino_gen() failed with error %d", error); return (error); } return (error); } if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { vref(dvp); *vpp = dvp; return (0); } if (autofs_cached(anp, cnp->cn_nameptr, cnp->cn_namelen) == false && - autofs_ignore_thread(cnp->cn_thread) == false) { + autofs_ignore_thread(curthread) == false) { error = autofs_trigger_vn(dvp, cnp->cn_nameptr, cnp->cn_namelen, &newvp); if (error != 0) return (error); if (newvp != NULL) { /* * The target filesystem got automounted. * Let the lookup(9) go around with the same * path component. */ vput(newvp); return (ERELOOKUP); } } AUTOFS_SLOCK(amp); error = autofs_node_find(anp, cnp->cn_nameptr, cnp->cn_namelen, &child); if (error != 0) { if ((cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) { AUTOFS_SUNLOCK(amp); return (EJUSTRETURN); } AUTOFS_SUNLOCK(amp); return (ENOENT); } /* * XXX: Dropping the node here is ok, because we never remove nodes. */ AUTOFS_SUNLOCK(amp); error = autofs_node_vn(child, mp, cnp->cn_lkflags, vpp); if (error != 0) { if ((cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) return (EJUSTRETURN); return (error); } return (0); } static int autofs_mkdir(struct vop_mkdir_args *ap) { struct vnode *vp; struct autofs_node *anp; struct autofs_mount *amp; struct autofs_node *child; int error; vp = ap->a_dvp; anp = vp->v_data; amp = VFSTOAUTOFS(vp->v_mount); /* * Do not allow mkdir() if the calling thread is not * automountd(8) descendant. */ if (autofs_ignore_thread(curthread) == false) return (EPERM); AUTOFS_XLOCK(amp); error = autofs_node_new(anp, amp, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen, &child); if (error != 0) { AUTOFS_XUNLOCK(amp); return (error); } AUTOFS_XUNLOCK(amp); error = autofs_node_vn(child, vp->v_mount, LK_EXCLUSIVE, ap->a_vpp); return (error); } static int autofs_print(struct vop_print_args *ap) { struct vnode *vp; struct autofs_node *anp; vp = ap->a_vp; anp = vp->v_data; printf(" name \"%s\", fileno %d, cached %d, wildcards %d\n", anp->an_name, anp->an_fileno, anp->an_cached, anp->an_wildcards); return (0); } /* * Write out a single 'struct dirent', based on 'name' and 'fileno' arguments. */ static int autofs_readdir_one(struct uio *uio, const char *name, int fileno, size_t *reclenp) { struct dirent dirent; size_t namlen, reclen; int error; namlen = strlen(name); reclen = _GENERIC_DIRLEN(namlen); if (reclenp != NULL) *reclenp = reclen; if (uio == NULL) return (0); if (uio->uio_resid < reclen) return (EINVAL); dirent.d_fileno = fileno; dirent.d_off = uio->uio_offset + reclen; dirent.d_reclen = reclen; dirent.d_type = DT_DIR; dirent.d_namlen = namlen; memcpy(dirent.d_name, name, namlen); dirent_terminate(&dirent); error = uiomove(&dirent, reclen, uio); return (error); } static size_t autofs_dirent_reclen(const char *name) { size_t reclen; (void)autofs_readdir_one(NULL, name, -1, &reclen); return (reclen); } static int autofs_readdir(struct vop_readdir_args *ap) { struct vnode *vp, *newvp; struct autofs_mount *amp; struct autofs_node *anp, *child; struct uio *uio; size_t reclen, reclens; ssize_t initial_resid; int error; vp = ap->a_vp; amp = VFSTOAUTOFS(vp->v_mount); anp = vp->v_data; uio = ap->a_uio; initial_resid = ap->a_uio->uio_resid; KASSERT(vp->v_type == VDIR, ("!VDIR")); if (autofs_cached(anp, NULL, 0) == false && autofs_ignore_thread(curthread) == false) { error = autofs_trigger_vn(vp, "", 0, &newvp); if (error != 0) return (error); if (newvp != NULL) { error = VOP_READDIR(newvp, ap->a_uio, ap->a_cred, ap->a_eofflag, ap->a_ncookies, ap->a_cookies); vput(newvp); return (error); } } if (uio->uio_offset < 0) return (EINVAL); if (ap->a_eofflag != NULL) *ap->a_eofflag = FALSE; /* * Write out the directory entry for ".". This is conditional * on the current offset into the directory; same applies to the * other two cases below. */ if (uio->uio_offset == 0) { error = autofs_readdir_one(uio, ".", anp->an_fileno, &reclen); if (error != 0) goto out; } reclens = autofs_dirent_reclen("."); /* * Write out the directory entry for "..". */ if (uio->uio_offset <= reclens) { if (uio->uio_offset != reclens) return (EINVAL); if (anp->an_parent == NULL) { error = autofs_readdir_one(uio, "..", anp->an_fileno, &reclen); } else { error = autofs_readdir_one(uio, "..", anp->an_parent->an_fileno, &reclen); } if (error != 0) goto out; } reclens += autofs_dirent_reclen(".."); /* * Write out the directory entries for subdirectories. */ AUTOFS_SLOCK(amp); RB_FOREACH(child, autofs_node_tree, &anp->an_children) { /* * Check the offset to skip entries returned by previous * calls to getdents(). */ if (uio->uio_offset > reclens) { reclens += autofs_dirent_reclen(child->an_name); continue; } /* * Prevent seeking into the middle of dirent. */ if (uio->uio_offset != reclens) { AUTOFS_SUNLOCK(amp); return (EINVAL); } error = autofs_readdir_one(uio, child->an_name, child->an_fileno, &reclen); reclens += reclen; if (error != 0) { AUTOFS_SUNLOCK(amp); goto out; } } AUTOFS_SUNLOCK(amp); if (ap->a_eofflag != NULL) *ap->a_eofflag = TRUE; return (0); out: /* * Return error if the initial buffer was too small to do anything. */ if (uio->uio_resid == initial_resid) return (error); /* * Don't return an error if we managed to copy out some entries. */ if (uio->uio_resid < reclen) return (0); return (error); } static int autofs_reclaim(struct vop_reclaim_args *ap) { struct vnode *vp; struct autofs_node *anp; vp = ap->a_vp; anp = vp->v_data; /* * We do not free autofs_node here; instead we are * destroying them in autofs_node_delete(). */ sx_xlock(&anp->an_vnode_lock); anp->an_vnode = NULL; vp->v_data = NULL; sx_xunlock(&anp->an_vnode_lock); return (0); } struct vop_vector autofs_vnodeops = { .vop_default = &default_vnodeops, .vop_access = autofs_access, .vop_lookup = autofs_lookup, .vop_create = VOP_EOPNOTSUPP, .vop_getattr = autofs_getattr, .vop_link = VOP_EOPNOTSUPP, .vop_mkdir = autofs_mkdir, .vop_mknod = VOP_EOPNOTSUPP, .vop_print = autofs_print, .vop_read = VOP_EOPNOTSUPP, .vop_readdir = autofs_readdir, .vop_remove = VOP_EOPNOTSUPP, .vop_rename = VOP_EOPNOTSUPP, .vop_rmdir = VOP_EOPNOTSUPP, .vop_setattr = VOP_EOPNOTSUPP, .vop_symlink = VOP_EOPNOTSUPP, .vop_write = VOP_EOPNOTSUPP, .vop_reclaim = autofs_reclaim, }; VFS_VOP_VECTOR_REGISTER(autofs_vnodeops); int autofs_node_new(struct autofs_node *parent, struct autofs_mount *amp, const char *name, int namelen, struct autofs_node **anpp) { struct autofs_node *anp; if (parent != NULL) { AUTOFS_ASSERT_XLOCKED(parent->an_mount); KASSERT(autofs_node_find(parent, name, namelen, NULL) == ENOENT, ("node \"%s\" already exists", name)); } anp = uma_zalloc(autofs_node_zone, M_WAITOK | M_ZERO); if (namelen >= 0) anp->an_name = strndup(name, namelen, M_AUTOFS); else anp->an_name = strdup(name, M_AUTOFS); anp->an_fileno = atomic_fetchadd_int(&->am_last_fileno, 1); callout_init(&anp->an_callout, 1); /* * The reason for SX_NOWITNESS here is that witness(4) * cannot tell vnodes apart, so the following perfectly * valid lock order... * * vnode lock A -> autofsvlk B -> vnode lock B * * ... gets reported as a LOR. */ sx_init_flags(&anp->an_vnode_lock, "autofsvlk", SX_NOWITNESS); getnanotime(&anp->an_ctime); anp->an_parent = parent; anp->an_mount = amp; if (parent != NULL) RB_INSERT(autofs_node_tree, &parent->an_children, anp); RB_INIT(&anp->an_children); *anpp = anp; return (0); } int autofs_node_find(struct autofs_node *parent, const char *name, int namelen, struct autofs_node **anpp) { struct autofs_node *anp, find; int error; AUTOFS_ASSERT_LOCKED(parent->an_mount); if (namelen >= 0) find.an_name = strndup(name, namelen, M_AUTOFS); else find.an_name = strdup(name, M_AUTOFS); anp = RB_FIND(autofs_node_tree, &parent->an_children, &find); if (anp != NULL) { error = 0; if (anpp != NULL) *anpp = anp; } else { error = ENOENT; } free(find.an_name, M_AUTOFS); return (error); } void autofs_node_delete(struct autofs_node *anp) { struct autofs_node *parent; AUTOFS_ASSERT_XLOCKED(anp->an_mount); KASSERT(RB_EMPTY(&anp->an_children), ("have children")); callout_drain(&anp->an_callout); parent = anp->an_parent; if (parent != NULL) RB_REMOVE(autofs_node_tree, &parent->an_children, anp); sx_destroy(&anp->an_vnode_lock); free(anp->an_name, M_AUTOFS); uma_zfree(autofs_node_zone, anp); } int autofs_node_vn(struct autofs_node *anp, struct mount *mp, int flags, struct vnode **vpp) { struct vnode *vp; int error; AUTOFS_ASSERT_UNLOCKED(anp->an_mount); sx_xlock(&anp->an_vnode_lock); vp = anp->an_vnode; if (vp != NULL) { error = vget(vp, flags | LK_RETRY); if (error != 0) { AUTOFS_WARN("vget failed with error %d", error); sx_xunlock(&anp->an_vnode_lock); return (error); } if (VN_IS_DOOMED(vp)) { /* * We got forcibly unmounted. */ AUTOFS_DEBUG("doomed vnode"); sx_xunlock(&anp->an_vnode_lock); vput(vp); return (ENOENT); } *vpp = vp; sx_xunlock(&anp->an_vnode_lock); return (0); } error = getnewvnode("autofs", mp, &autofs_vnodeops, &vp); if (error != 0) { sx_xunlock(&anp->an_vnode_lock); return (error); } error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); if (error != 0) { sx_xunlock(&anp->an_vnode_lock); vdrop(vp); return (error); } vp->v_type = VDIR; if (anp->an_parent == NULL) vp->v_vflag |= VV_ROOT; vp->v_data = anp; VN_LOCK_ASHARE(vp); error = insmntque(vp, mp); if (error != 0) { AUTOFS_DEBUG("insmntque() failed with error %d", error); sx_xunlock(&anp->an_vnode_lock); return (error); } KASSERT(anp->an_vnode == NULL, ("lost race")); anp->an_vnode = vp; sx_xunlock(&anp->an_vnode_lock); *vpp = vp; return (0); } diff --git a/sys/fs/devfs/devfs_vnops.c b/sys/fs/devfs/devfs_vnops.c index 40b80596523d..9a327c02ee96 100644 --- a/sys/fs/devfs/devfs_vnops.c +++ b/sys/fs/devfs/devfs_vnops.c @@ -1,2122 +1,2122 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2000-2004 * Poul-Henning Kamp. All rights reserved. * Copyright (c) 1989, 1992-1993, 1995 * The Regents of the University of California. All rights reserved. * * This code is derived from software donated to Berkeley by * Jan-Simon Pendry. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)kernfs_vnops.c 8.15 (Berkeley) 5/21/95 * From: FreeBSD: src/sys/miscfs/kernfs/kernfs_vnops.c 1.43 * * $FreeBSD$ */ /* * TODO: * mkdir: want it ? */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static struct vop_vector devfs_vnodeops; static struct vop_vector devfs_specops; static struct fileops devfs_ops_f; #include #include #include #include #include #include static MALLOC_DEFINE(M_CDEVPDATA, "DEVFSP", "Metainfo for cdev-fp data"); struct mtx devfs_de_interlock; MTX_SYSINIT(devfs_de_interlock, &devfs_de_interlock, "devfs interlock", MTX_DEF); struct sx clone_drain_lock; SX_SYSINIT(clone_drain_lock, &clone_drain_lock, "clone events drain lock"); struct mtx cdevpriv_mtx; MTX_SYSINIT(cdevpriv_mtx, &cdevpriv_mtx, "cdevpriv lock", MTX_DEF); SYSCTL_DECL(_vfs_devfs); static int devfs_dotimes; SYSCTL_INT(_vfs_devfs, OID_AUTO, dotimes, CTLFLAG_RW, &devfs_dotimes, 0, "Update timestamps on DEVFS with default precision"); /* * Update devfs node timestamp. Note that updates are unlocked and * stat(2) could see partially updated times. */ static void devfs_timestamp(struct timespec *tsp) { time_t ts; if (devfs_dotimes) { vfs_timestamp(tsp); } else { ts = time_second; if (tsp->tv_sec != ts) { tsp->tv_sec = ts; tsp->tv_nsec = 0; } } } static int devfs_fp_check(struct file *fp, struct cdev **devp, struct cdevsw **dswp, int *ref) { *dswp = devvn_refthread(fp->f_vnode, devp, ref); if (*dswp == NULL || *devp != fp->f_data) { if (*dswp != NULL) dev_relthread(*devp, *ref); return (ENXIO); } KASSERT((*devp)->si_refcount > 0, ("devfs: un-referenced struct cdev *(%s)", devtoname(*devp))); if (*dswp == NULL) return (ENXIO); curthread->td_fpop = fp; return (0); } int devfs_get_cdevpriv(void **datap) { struct file *fp; struct cdev_privdata *p; int error; fp = curthread->td_fpop; if (fp == NULL) return (EBADF); p = fp->f_cdevpriv; if (p != NULL) { error = 0; *datap = p->cdpd_data; } else error = ENOENT; return (error); } int devfs_set_cdevpriv(void *priv, d_priv_dtor_t *priv_dtr) { struct file *fp; struct cdev_priv *cdp; struct cdev_privdata *p; int error; fp = curthread->td_fpop; if (fp == NULL) return (ENOENT); cdp = cdev2priv((struct cdev *)fp->f_data); p = malloc(sizeof(struct cdev_privdata), M_CDEVPDATA, M_WAITOK); p->cdpd_data = priv; p->cdpd_dtr = priv_dtr; p->cdpd_fp = fp; mtx_lock(&cdevpriv_mtx); if (fp->f_cdevpriv == NULL) { LIST_INSERT_HEAD(&cdp->cdp_fdpriv, p, cdpd_list); fp->f_cdevpriv = p; mtx_unlock(&cdevpriv_mtx); error = 0; } else { mtx_unlock(&cdevpriv_mtx); free(p, M_CDEVPDATA); error = EBUSY; } return (error); } void devfs_destroy_cdevpriv(struct cdev_privdata *p) { mtx_assert(&cdevpriv_mtx, MA_OWNED); KASSERT(p->cdpd_fp->f_cdevpriv == p, ("devfs_destoy_cdevpriv %p != %p", p->cdpd_fp->f_cdevpriv, p)); p->cdpd_fp->f_cdevpriv = NULL; LIST_REMOVE(p, cdpd_list); mtx_unlock(&cdevpriv_mtx); (p->cdpd_dtr)(p->cdpd_data); free(p, M_CDEVPDATA); } static void devfs_fpdrop(struct file *fp) { struct cdev_privdata *p; mtx_lock(&cdevpriv_mtx); if ((p = fp->f_cdevpriv) == NULL) { mtx_unlock(&cdevpriv_mtx); return; } devfs_destroy_cdevpriv(p); } void devfs_clear_cdevpriv(void) { struct file *fp; fp = curthread->td_fpop; if (fp == NULL) return; devfs_fpdrop(fp); } static void devfs_usecount_add(struct vnode *vp) { struct devfs_dirent *de; struct cdev *dev; mtx_lock(&devfs_de_interlock); VI_LOCK(vp); VNPASS(vp->v_type == VCHR || vp->v_type == VBAD, vp); if (VN_IS_DOOMED(vp)) { goto out_unlock; } de = vp->v_data; dev = vp->v_rdev; MPASS(de != NULL); MPASS(dev != NULL); dev->si_usecount++; de->de_usecount++; out_unlock: VI_UNLOCK(vp); mtx_unlock(&devfs_de_interlock); } static void devfs_usecount_subl(struct vnode *vp) { struct devfs_dirent *de; struct cdev *dev; mtx_assert(&devfs_de_interlock, MA_OWNED); ASSERT_VI_LOCKED(vp, __func__); VNPASS(vp->v_type == VCHR || vp->v_type == VBAD, vp); de = vp->v_data; dev = vp->v_rdev; if (de == NULL) return; if (dev == NULL) { MPASS(de->de_usecount == 0); return; } if (dev->si_usecount < de->de_usecount) panic("%s: si_usecount underflow for dev %p " "(has %ld, dirent has %d)\n", __func__, dev, dev->si_usecount, de->de_usecount); if (VN_IS_DOOMED(vp)) { dev->si_usecount -= de->de_usecount; de->de_usecount = 0; } else { if (de->de_usecount == 0) panic("%s: de_usecount underflow for dev %p\n", __func__, dev); dev->si_usecount--; de->de_usecount--; } } static void devfs_usecount_sub(struct vnode *vp) { mtx_lock(&devfs_de_interlock); VI_LOCK(vp); devfs_usecount_subl(vp); VI_UNLOCK(vp); mtx_unlock(&devfs_de_interlock); } static int devfs_usecountl(struct vnode *vp) { VNPASS(vp->v_type == VCHR, vp); mtx_assert(&devfs_de_interlock, MA_OWNED); ASSERT_VI_LOCKED(vp, __func__); return (vp->v_rdev->si_usecount); } int devfs_usecount(struct vnode *vp) { int count; VNPASS(vp->v_type == VCHR, vp); mtx_lock(&devfs_de_interlock); VI_LOCK(vp); count = devfs_usecountl(vp); VI_UNLOCK(vp); mtx_unlock(&devfs_de_interlock); return (count); } void devfs_ctty_ref(struct vnode *vp) { vrefact(vp); devfs_usecount_add(vp); } void devfs_ctty_unref(struct vnode *vp) { devfs_usecount_sub(vp); vrele(vp); } /* * On success devfs_populate_vp() returns with dmp->dm_lock held. */ static int devfs_populate_vp(struct vnode *vp) { struct devfs_dirent *de; struct devfs_mount *dmp; int locked; ASSERT_VOP_LOCKED(vp, "devfs_populate_vp"); dmp = VFSTODEVFS(vp->v_mount); if (!devfs_populate_needed(dmp)) { sx_xlock(&dmp->dm_lock); goto out_nopopulate; } locked = VOP_ISLOCKED(vp); sx_xlock(&dmp->dm_lock); DEVFS_DMP_HOLD(dmp); /* Can't call devfs_populate() with the vnode lock held. */ VOP_UNLOCK(vp); devfs_populate(dmp); sx_xunlock(&dmp->dm_lock); vn_lock(vp, locked | LK_RETRY); sx_xlock(&dmp->dm_lock); if (DEVFS_DMP_DROP(dmp)) { sx_xunlock(&dmp->dm_lock); devfs_unmount_final(dmp); return (ERESTART); } out_nopopulate: if (VN_IS_DOOMED(vp)) { sx_xunlock(&dmp->dm_lock); return (ERESTART); } de = vp->v_data; KASSERT(de != NULL, ("devfs_populate_vp: vp->v_data == NULL but vnode not doomed")); if ((de->de_flags & DE_DOOMED) != 0) { sx_xunlock(&dmp->dm_lock); return (ERESTART); } return (0); } static int devfs_vptocnp(struct vop_vptocnp_args *ap) { struct vnode *vp = ap->a_vp; struct vnode **dvp = ap->a_vpp; struct devfs_mount *dmp; char *buf = ap->a_buf; size_t *buflen = ap->a_buflen; struct devfs_dirent *dd, *de; int i, error; dmp = VFSTODEVFS(vp->v_mount); error = devfs_populate_vp(vp); if (error != 0) return (error); if (vp->v_type != VCHR && vp->v_type != VDIR) { error = ENOENT; goto finished; } dd = vp->v_data; if (vp->v_type == VDIR && dd == dmp->dm_rootdir) { *dvp = vp; vref(*dvp); goto finished; } i = *buflen; i -= dd->de_dirent->d_namlen; if (i < 0) { error = ENOMEM; goto finished; } bcopy(dd->de_dirent->d_name, buf + i, dd->de_dirent->d_namlen); *buflen = i; de = devfs_parent_dirent(dd); if (de == NULL) { error = ENOENT; goto finished; } mtx_lock(&devfs_de_interlock); *dvp = de->de_vnode; if (*dvp != NULL) { VI_LOCK(*dvp); mtx_unlock(&devfs_de_interlock); vholdl(*dvp); VI_UNLOCK(*dvp); vref(*dvp); vdrop(*dvp); } else { mtx_unlock(&devfs_de_interlock); error = ENOENT; } finished: sx_xunlock(&dmp->dm_lock); return (error); } /* * Construct the fully qualified path name relative to the mountpoint. * If a NULL cnp is provided, no '/' is appended to the resulting path. */ char * devfs_fqpn(char *buf, struct devfs_mount *dmp, struct devfs_dirent *dd, struct componentname *cnp) { int i; struct devfs_dirent *de; sx_assert(&dmp->dm_lock, SA_LOCKED); i = SPECNAMELEN; buf[i] = '\0'; if (cnp != NULL) i -= cnp->cn_namelen; if (i < 0) return (NULL); if (cnp != NULL) bcopy(cnp->cn_nameptr, buf + i, cnp->cn_namelen); de = dd; while (de != dmp->dm_rootdir) { if (cnp != NULL || i < SPECNAMELEN) { i--; if (i < 0) return (NULL); buf[i] = '/'; } i -= de->de_dirent->d_namlen; if (i < 0) return (NULL); bcopy(de->de_dirent->d_name, buf + i, de->de_dirent->d_namlen); de = devfs_parent_dirent(de); if (de == NULL) return (NULL); } return (buf + i); } static int devfs_allocv_drop_refs(int drop_dm_lock, struct devfs_mount *dmp, struct devfs_dirent *de) { int not_found; not_found = 0; if (de->de_flags & DE_DOOMED) not_found = 1; if (DEVFS_DE_DROP(de)) { KASSERT(not_found == 1, ("DEVFS de dropped but not doomed")); devfs_dirent_free(de); } if (DEVFS_DMP_DROP(dmp)) { KASSERT(not_found == 1, ("DEVFS mount struct freed before dirent")); not_found = 2; sx_xunlock(&dmp->dm_lock); devfs_unmount_final(dmp); } if (not_found == 1 || (drop_dm_lock && not_found != 2)) sx_unlock(&dmp->dm_lock); return (not_found); } static void devfs_insmntque_dtr(struct vnode *vp, void *arg) { struct devfs_dirent *de; de = (struct devfs_dirent *)arg; mtx_lock(&devfs_de_interlock); vp->v_data = NULL; de->de_vnode = NULL; mtx_unlock(&devfs_de_interlock); vgone(vp); vput(vp); } /* * devfs_allocv shall be entered with dmp->dm_lock held, and it drops * it on return. */ int devfs_allocv(struct devfs_dirent *de, struct mount *mp, int lockmode, struct vnode **vpp) { int error; struct vnode *vp; struct cdev *dev; struct devfs_mount *dmp; struct cdevsw *dsw; enum vgetstate vs; dmp = VFSTODEVFS(mp); if (de->de_flags & DE_DOOMED) { sx_xunlock(&dmp->dm_lock); return (ENOENT); } loop: DEVFS_DE_HOLD(de); DEVFS_DMP_HOLD(dmp); mtx_lock(&devfs_de_interlock); vp = de->de_vnode; if (vp != NULL) { vs = vget_prep(vp); mtx_unlock(&devfs_de_interlock); sx_xunlock(&dmp->dm_lock); vget_finish(vp, lockmode | LK_RETRY, vs); sx_xlock(&dmp->dm_lock); if (devfs_allocv_drop_refs(0, dmp, de)) { vput(vp); return (ENOENT); } else if (VN_IS_DOOMED(vp)) { mtx_lock(&devfs_de_interlock); if (de->de_vnode == vp) { de->de_vnode = NULL; vp->v_data = NULL; } mtx_unlock(&devfs_de_interlock); vput(vp); goto loop; } sx_xunlock(&dmp->dm_lock); *vpp = vp; return (0); } mtx_unlock(&devfs_de_interlock); if (de->de_dirent->d_type == DT_CHR) { if (!(de->de_cdp->cdp_flags & CDP_ACTIVE)) { devfs_allocv_drop_refs(1, dmp, de); return (ENOENT); } dev = &de->de_cdp->cdp_c; } else { dev = NULL; } error = getnewvnode("devfs", mp, &devfs_vnodeops, &vp); if (error != 0) { devfs_allocv_drop_refs(1, dmp, de); printf("devfs_allocv: failed to allocate new vnode\n"); return (error); } if (de->de_dirent->d_type == DT_CHR) { vp->v_type = VCHR; VI_LOCK(vp); dev_lock(); dev_refl(dev); /* XXX: v_rdev should be protect by vnode lock */ vp->v_rdev = dev; VNPASS(vp->v_usecount == 1, vp); /* Special casing of ttys for deadfs. Probably redundant. */ dsw = dev->si_devsw; if (dsw != NULL && (dsw->d_flags & D_TTY) != 0) vp->v_vflag |= VV_ISTTY; dev_unlock(); VI_UNLOCK(vp); if ((dev->si_flags & SI_ETERNAL) != 0) vp->v_vflag |= VV_ETERNALDEV; vp->v_op = &devfs_specops; } else if (de->de_dirent->d_type == DT_DIR) { vp->v_type = VDIR; } else if (de->de_dirent->d_type == DT_LNK) { vp->v_type = VLNK; } else { vp->v_type = VBAD; } vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_NOWITNESS); VN_LOCK_ASHARE(vp); mtx_lock(&devfs_de_interlock); vp->v_data = de; de->de_vnode = vp; mtx_unlock(&devfs_de_interlock); error = insmntque1(vp, mp, devfs_insmntque_dtr, de); if (error != 0) { (void) devfs_allocv_drop_refs(1, dmp, de); return (error); } if (devfs_allocv_drop_refs(0, dmp, de)) { vput(vp); return (ENOENT); } #ifdef MAC mac_devfs_vnode_associate(mp, de, vp); #endif sx_xunlock(&dmp->dm_lock); *vpp = vp; return (0); } static int devfs_access(struct vop_access_args *ap) { struct vnode *vp = ap->a_vp; struct devfs_dirent *de; struct proc *p; int error; de = vp->v_data; if (vp->v_type == VDIR) de = de->de_dir; error = vaccess(vp->v_type, de->de_mode, de->de_uid, de->de_gid, ap->a_accmode, ap->a_cred); if (error == 0) return (0); if (error != EACCES) return (error); p = ap->a_td->td_proc; /* We do, however, allow access to the controlling terminal */ PROC_LOCK(p); if (!(p->p_flag & P_CONTROLT)) { PROC_UNLOCK(p); return (error); } if (p->p_session->s_ttydp == de->de_cdp) error = 0; PROC_UNLOCK(p); return (error); } _Static_assert(((FMASK | FCNTLFLAGS) & (FLASTCLOSE | FREVOKE)) == 0, "devfs-only flag reuse failed"); static int devfs_close(struct vop_close_args *ap) { struct vnode *vp = ap->a_vp, *oldvp; struct thread *td = ap->a_td; struct proc *p; struct cdev *dev = vp->v_rdev; struct cdevsw *dsw; struct devfs_dirent *de = vp->v_data; int dflags, error, ref, vp_locked; /* * XXX: Don't call d_close() if we were called because of * XXX: insmntque1() failure. */ if (vp->v_data == NULL) return (0); /* * Hack: a tty device that is a controlling terminal * has a reference from the session structure. * We cannot easily tell that a character device is * a controlling terminal, unless it is the closing * process' controlling terminal. In that case, * if the reference count is 2 (this last descriptor * plus the session), release the reference from the session. */ if (de->de_usecount == 2 && td != NULL) { p = td->td_proc; PROC_LOCK(p); if (vp == p->p_session->s_ttyvp) { PROC_UNLOCK(p); oldvp = NULL; sx_xlock(&proctree_lock); if (vp == p->p_session->s_ttyvp) { SESS_LOCK(p->p_session); mtx_lock(&devfs_de_interlock); VI_LOCK(vp); if (devfs_usecountl(vp) == 2 && !VN_IS_DOOMED(vp)) { p->p_session->s_ttyvp = NULL; p->p_session->s_ttydp = NULL; oldvp = vp; } VI_UNLOCK(vp); mtx_unlock(&devfs_de_interlock); SESS_UNLOCK(p->p_session); } sx_xunlock(&proctree_lock); if (oldvp != NULL) devfs_ctty_unref(oldvp); } else PROC_UNLOCK(p); } /* * We do not want to really close the device if it * is still in use unless we are trying to close it * forcibly. Since every use (buffer, vnode, swap, cmap) * holds a reference to the vnode, and because we mark * any other vnodes that alias this device, when the * sum of the reference counts on all the aliased * vnodes descends to one, we are on last close. */ dsw = dev_refthread(dev, &ref); if (dsw == NULL) return (ENXIO); dflags = 0; mtx_lock(&devfs_de_interlock); VI_LOCK(vp); if (devfs_usecountl(vp) == 1) dflags |= FLASTCLOSE; devfs_usecount_subl(vp); mtx_unlock(&devfs_de_interlock); if (VN_IS_DOOMED(vp)) { /* Forced close. */ dflags |= FREVOKE | FNONBLOCK; } else if (dsw->d_flags & D_TRACKCLOSE) { /* Keep device updated on status. */ } else if ((dflags & FLASTCLOSE) == 0) { VI_UNLOCK(vp); dev_relthread(dev, ref); return (0); } vholdnz(vp); VI_UNLOCK(vp); vp_locked = VOP_ISLOCKED(vp); VOP_UNLOCK(vp); KASSERT(dev->si_refcount > 0, ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev))); error = dsw->d_close(dev, ap->a_fflag | dflags, S_IFCHR, td); dev_relthread(dev, ref); vn_lock(vp, vp_locked | LK_RETRY); vdrop(vp); return (error); } static int devfs_close_f(struct file *fp, struct thread *td) { int error; struct file *fpop; /* * NB: td may be NULL if this descriptor is closed due to * garbage collection from a closed UNIX domain socket. */ fpop = curthread->td_fpop; curthread->td_fpop = fp; error = vnops.fo_close(fp, td); curthread->td_fpop = fpop; /* * The f_cdevpriv cannot be assigned non-NULL value while we * are destroying the file. */ if (fp->f_cdevpriv != NULL) devfs_fpdrop(fp); return (error); } static int devfs_getattr(struct vop_getattr_args *ap) { struct vnode *vp = ap->a_vp; struct vattr *vap = ap->a_vap; struct devfs_dirent *de; struct devfs_mount *dmp; struct cdev *dev; struct timeval boottime; int error; error = devfs_populate_vp(vp); if (error != 0) return (error); dmp = VFSTODEVFS(vp->v_mount); sx_xunlock(&dmp->dm_lock); de = vp->v_data; KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp)); if (vp->v_type == VDIR) { de = de->de_dir; KASSERT(de != NULL, ("Null dir dirent in devfs_getattr vp=%p", vp)); } vap->va_uid = de->de_uid; vap->va_gid = de->de_gid; vap->va_mode = de->de_mode; if (vp->v_type == VLNK) vap->va_size = strlen(de->de_symlink); else if (vp->v_type == VDIR) vap->va_size = vap->va_bytes = DEV_BSIZE; else vap->va_size = 0; if (vp->v_type != VDIR) vap->va_bytes = 0; vap->va_blocksize = DEV_BSIZE; vap->va_type = vp->v_type; getboottime(&boottime); #define fix(aa) \ do { \ if ((aa).tv_sec <= 3600) { \ (aa).tv_sec = boottime.tv_sec; \ (aa).tv_nsec = boottime.tv_usec * 1000; \ } \ } while (0) if (vp->v_type != VCHR) { fix(de->de_atime); vap->va_atime = de->de_atime; fix(de->de_mtime); vap->va_mtime = de->de_mtime; fix(de->de_ctime); vap->va_ctime = de->de_ctime; } else { dev = vp->v_rdev; fix(dev->si_atime); vap->va_atime = dev->si_atime; fix(dev->si_mtime); vap->va_mtime = dev->si_mtime; fix(dev->si_ctime); vap->va_ctime = dev->si_ctime; vap->va_rdev = cdev2priv(dev)->cdp_inode; } vap->va_gen = 0; vap->va_flags = 0; vap->va_filerev = 0; vap->va_nlink = de->de_links; vap->va_fileid = de->de_inode; return (error); } /* ARGSUSED */ static int devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td) { struct file *fpop; int error; fpop = td->td_fpop; td->td_fpop = fp; error = vnops.fo_ioctl(fp, com, data, cred, td); td->td_fpop = fpop; return (error); } void * fiodgname_buf_get_ptr(void *fgnp, u_long com) { union { struct fiodgname_arg fgn; #ifdef COMPAT_FREEBSD32 struct fiodgname_arg32 fgn32; #endif } *fgnup; fgnup = fgnp; switch (com) { case FIODGNAME: return (fgnup->fgn.buf); #ifdef COMPAT_FREEBSD32 case FIODGNAME_32: return ((void *)(uintptr_t)fgnup->fgn32.buf); #endif default: panic("Unhandled ioctl command %ld", com); } } static int devfs_ioctl(struct vop_ioctl_args *ap) { struct fiodgname_arg *fgn; struct vnode *vpold, *vp; struct cdevsw *dsw; struct thread *td; struct session *sess; struct cdev *dev; int error, ref, i; const char *p; u_long com; vp = ap->a_vp; com = ap->a_command; td = ap->a_td; dsw = devvn_refthread(vp, &dev, &ref); if (dsw == NULL) return (ENXIO); KASSERT(dev->si_refcount > 0, ("devfs: un-referenced struct cdev *(%s)", devtoname(dev))); switch (com) { case FIODTYPE: *(int *)ap->a_data = dsw->d_flags & D_TYPEMASK; error = 0; break; case FIODGNAME: #ifdef COMPAT_FREEBSD32 case FIODGNAME_32: #endif fgn = ap->a_data; p = devtoname(dev); i = strlen(p) + 1; if (i > fgn->len) error = EINVAL; else error = copyout(p, fiodgname_buf_get_ptr(fgn, com), i); break; default: error = dsw->d_ioctl(dev, com, ap->a_data, ap->a_fflag, td); } dev_relthread(dev, ref); if (error == ENOIOCTL) error = ENOTTY; if (error == 0 && com == TIOCSCTTY) { /* * Do nothing if reassigning same control tty, or if the * control tty has already disappeared. If it disappeared, * it's because we were racing with TIOCNOTTY. TIOCNOTTY * already took care of releasing the old vnode and we have * nothing left to do. */ sx_slock(&proctree_lock); sess = td->td_proc->p_session; if (sess->s_ttyvp == vp || sess->s_ttyp == NULL) { sx_sunlock(&proctree_lock); return (0); } devfs_ctty_ref(vp); SESS_LOCK(sess); vpold = sess->s_ttyvp; sess->s_ttyvp = vp; sess->s_ttydp = cdev2priv(dev); SESS_UNLOCK(sess); sx_sunlock(&proctree_lock); /* Get rid of reference to old control tty */ if (vpold) devfs_ctty_unref(vpold); } return (error); } /* ARGSUSED */ static int devfs_kqfilter_f(struct file *fp, struct knote *kn) { struct cdev *dev; struct cdevsw *dsw; int error, ref; struct file *fpop; struct thread *td; td = curthread; fpop = td->td_fpop; error = devfs_fp_check(fp, &dev, &dsw, &ref); if (error) return (error); error = dsw->d_kqfilter(dev, kn); td->td_fpop = fpop; dev_relthread(dev, ref); return (error); } static inline int devfs_prison_check(struct devfs_dirent *de, struct thread *td) { struct cdev_priv *cdp; struct ucred *dcr; struct proc *p; int error; cdp = de->de_cdp; if (cdp == NULL) return (0); dcr = cdp->cdp_c.si_cred; if (dcr == NULL) return (0); error = prison_check(td->td_ucred, dcr); if (error == 0) return (0); /* We do, however, allow access to the controlling terminal */ p = td->td_proc; PROC_LOCK(p); if (!(p->p_flag & P_CONTROLT)) { PROC_UNLOCK(p); return (error); } if (p->p_session->s_ttydp == cdp) error = 0; PROC_UNLOCK(p); return (error); } static int devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock) { struct componentname *cnp; struct vnode *dvp, **vpp; struct thread *td; struct devfs_dirent *de, *dd; struct devfs_dirent **dde; struct devfs_mount *dmp; struct mount *mp; struct cdev *cdev; int error, flags, nameiop, dvplocked; char specname[SPECNAMELEN + 1], *pname; + td = curthread; cnp = ap->a_cnp; vpp = ap->a_vpp; dvp = ap->a_dvp; pname = cnp->cn_nameptr; - td = cnp->cn_thread; flags = cnp->cn_flags; nameiop = cnp->cn_nameiop; mp = dvp->v_mount; dmp = VFSTODEVFS(mp); dd = dvp->v_data; *vpp = NULLVP; if ((flags & ISLASTCN) && nameiop == RENAME) return (EOPNOTSUPP); if (dvp->v_type != VDIR) return (ENOTDIR); if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT)) return (EIO); error = vn_dir_check_exec(dvp, cnp); if (error != 0) return (error); if (cnp->cn_namelen == 1 && *pname == '.') { if ((flags & ISLASTCN) && nameiop != LOOKUP) return (EINVAL); *vpp = dvp; VREF(dvp); return (0); } if (flags & ISDOTDOT) { if ((flags & ISLASTCN) && nameiop != LOOKUP) return (EINVAL); de = devfs_parent_dirent(dd); if (de == NULL) return (ENOENT); dvplocked = VOP_ISLOCKED(dvp); VOP_UNLOCK(dvp); error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK, vpp); *dm_unlock = 0; vn_lock(dvp, dvplocked | LK_RETRY); return (error); } dd = dvp->v_data; de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0); while (de == NULL) { /* While(...) so we can use break */ if (nameiop == DELETE) return (ENOENT); /* * OK, we didn't have an entry for the name we were asked for * so we try to see if anybody can create it on demand. */ pname = devfs_fqpn(specname, dmp, dd, cnp); if (pname == NULL) break; cdev = NULL; DEVFS_DMP_HOLD(dmp); sx_xunlock(&dmp->dm_lock); sx_slock(&clone_drain_lock); EVENTHANDLER_INVOKE(dev_clone, td->td_ucred, pname, strlen(pname), &cdev); sx_sunlock(&clone_drain_lock); if (cdev == NULL) sx_xlock(&dmp->dm_lock); else if (devfs_populate_vp(dvp) != 0) { *dm_unlock = 0; sx_xlock(&dmp->dm_lock); if (DEVFS_DMP_DROP(dmp)) { sx_xunlock(&dmp->dm_lock); devfs_unmount_final(dmp); } else sx_xunlock(&dmp->dm_lock); dev_rel(cdev); return (ENOENT); } if (DEVFS_DMP_DROP(dmp)) { *dm_unlock = 0; sx_xunlock(&dmp->dm_lock); devfs_unmount_final(dmp); if (cdev != NULL) dev_rel(cdev); return (ENOENT); } if (cdev == NULL) break; dev_lock(); dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx]; if (dde != NULL && *dde != NULL) de = *dde; dev_unlock(); dev_rel(cdev); break; } if (de == NULL || de->de_flags & DE_WHITEOUT) { if ((nameiop == CREATE || nameiop == RENAME) && (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) { cnp->cn_flags |= SAVENAME; return (EJUSTRETURN); } return (ENOENT); } if (devfs_prison_check(de, td)) return (ENOENT); if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) { error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td); if (error) return (error); if (*vpp == dvp) { VREF(dvp); *vpp = dvp; return (0); } } error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK, vpp); *dm_unlock = 0; return (error); } static int devfs_lookup(struct vop_lookup_args *ap) { int j; struct devfs_mount *dmp; int dm_unlock; if (devfs_populate_vp(ap->a_dvp) != 0) return (ENOTDIR); dmp = VFSTODEVFS(ap->a_dvp->v_mount); dm_unlock = 1; j = devfs_lookupx(ap, &dm_unlock); if (dm_unlock == 1) sx_xunlock(&dmp->dm_lock); return (j); } static int devfs_mknod(struct vop_mknod_args *ap) { struct componentname *cnp; struct vnode *dvp, **vpp; struct devfs_dirent *dd, *de; struct devfs_mount *dmp; int error; /* * The only type of node we should be creating here is a * character device, for anything else return EOPNOTSUPP. */ if (ap->a_vap->va_type != VCHR) return (EOPNOTSUPP); dvp = ap->a_dvp; dmp = VFSTODEVFS(dvp->v_mount); cnp = ap->a_cnp; vpp = ap->a_vpp; dd = dvp->v_data; error = ENOENT; sx_xlock(&dmp->dm_lock); TAILQ_FOREACH(de, &dd->de_dlist, de_list) { if (cnp->cn_namelen != de->de_dirent->d_namlen) continue; if (de->de_dirent->d_type == DT_CHR && (de->de_cdp->cdp_flags & CDP_ACTIVE) == 0) continue; if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name, de->de_dirent->d_namlen) != 0) continue; if (de->de_flags & DE_WHITEOUT) break; goto notfound; } if (de == NULL) goto notfound; de->de_flags &= ~DE_WHITEOUT; error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp); return (error); notfound: sx_xunlock(&dmp->dm_lock); return (error); } /* ARGSUSED */ static int devfs_open(struct vop_open_args *ap) { struct thread *td = ap->a_td; struct vnode *vp = ap->a_vp; struct cdev *dev = vp->v_rdev; struct file *fp = ap->a_fp; int error, ref, vlocked; struct cdevsw *dsw; struct file *fpop; if (vp->v_type == VBLK) return (ENXIO); if (dev == NULL) return (ENXIO); /* Make this field valid before any I/O in d_open. */ if (dev->si_iosize_max == 0) dev->si_iosize_max = DFLTPHYS; dsw = dev_refthread(dev, &ref); if (dsw == NULL) return (ENXIO); if (fp == NULL && dsw->d_fdopen != NULL) { dev_relthread(dev, ref); return (ENXIO); } if (vp->v_type == VCHR) devfs_usecount_add(vp); vlocked = VOP_ISLOCKED(vp); VOP_UNLOCK(vp); fpop = td->td_fpop; td->td_fpop = fp; if (fp != NULL) { fp->f_data = dev; fp->f_vnode = vp; } if (dsw->d_fdopen != NULL) error = dsw->d_fdopen(dev, ap->a_mode, td, fp); else error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td); /* Clean up any cdevpriv upon error. */ if (error != 0) devfs_clear_cdevpriv(); td->td_fpop = fpop; vn_lock(vp, vlocked | LK_RETRY); if (error != 0 && vp->v_type == VCHR) devfs_usecount_sub(vp); dev_relthread(dev, ref); if (error != 0) { if (error == ERESTART) error = EINTR; return (error); } #if 0 /* /dev/console */ KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp")); #else if (fp == NULL) return (error); #endif if (fp->f_ops == &badfileops) finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f); return (error); } static int devfs_pathconf(struct vop_pathconf_args *ap) { switch (ap->a_name) { case _PC_FILESIZEBITS: *ap->a_retval = 64; return (0); case _PC_NAME_MAX: *ap->a_retval = NAME_MAX; return (0); case _PC_LINK_MAX: *ap->a_retval = INT_MAX; return (0); case _PC_SYMLINK_MAX: *ap->a_retval = MAXPATHLEN; return (0); case _PC_MAX_CANON: if (ap->a_vp->v_vflag & VV_ISTTY) { *ap->a_retval = MAX_CANON; return (0); } return (EINVAL); case _PC_MAX_INPUT: if (ap->a_vp->v_vflag & VV_ISTTY) { *ap->a_retval = MAX_INPUT; return (0); } return (EINVAL); case _PC_VDISABLE: if (ap->a_vp->v_vflag & VV_ISTTY) { *ap->a_retval = _POSIX_VDISABLE; return (0); } return (EINVAL); case _PC_MAC_PRESENT: #ifdef MAC /* * If MAC is enabled, devfs automatically supports * trivial non-persistant label storage. */ *ap->a_retval = 1; #else *ap->a_retval = 0; #endif return (0); case _PC_CHOWN_RESTRICTED: *ap->a_retval = 1; return (0); default: return (vop_stdpathconf(ap)); } /* NOTREACHED */ } /* ARGSUSED */ static int devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td) { struct cdev *dev; struct cdevsw *dsw; int error, ref; struct file *fpop; fpop = td->td_fpop; error = devfs_fp_check(fp, &dev, &dsw, &ref); if (error != 0) { error = vnops.fo_poll(fp, events, cred, td); return (error); } error = dsw->d_poll(dev, events, td); td->td_fpop = fpop; dev_relthread(dev, ref); return(error); } /* * Print out the contents of a special device vnode. */ static int devfs_print(struct vop_print_args *ap) { printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev)); return (0); } static int devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred, int flags, struct thread *td) { struct cdev *dev; int ioflag, error, ref; ssize_t resid; struct cdevsw *dsw; struct file *fpop; if (uio->uio_resid > DEVFS_IOSIZE_MAX) return (EINVAL); fpop = td->td_fpop; error = devfs_fp_check(fp, &dev, &dsw, &ref); if (error != 0) { error = vnops.fo_read(fp, uio, cred, flags, td); return (error); } resid = uio->uio_resid; ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT); if (ioflag & O_DIRECT) ioflag |= IO_DIRECT; foffset_lock_uio(fp, uio, flags | FOF_NOLOCK); error = dsw->d_read(dev, uio, ioflag); if (uio->uio_resid != resid || (error == 0 && resid != 0)) devfs_timestamp(&dev->si_atime); td->td_fpop = fpop; dev_relthread(dev, ref); foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF_R); return (error); } static int devfs_readdir(struct vop_readdir_args *ap) { int error; struct uio *uio; struct dirent *dp; struct devfs_dirent *dd; struct devfs_dirent *de; struct devfs_mount *dmp; off_t off; int *tmp_ncookies = NULL; if (ap->a_vp->v_type != VDIR) return (ENOTDIR); uio = ap->a_uio; if (uio->uio_offset < 0) return (EINVAL); /* * XXX: This is a temporary hack to get around this filesystem not * supporting cookies. We store the location of the ncookies pointer * in a temporary variable before calling vfs_subr.c:vfs_read_dirent() * and set the number of cookies to 0. We then set the pointer to * NULL so that vfs_read_dirent doesn't try to call realloc() on * ap->a_cookies. Later in this function, we restore the ap->a_ncookies * pointer to its original location before returning to the caller. */ if (ap->a_ncookies != NULL) { tmp_ncookies = ap->a_ncookies; *ap->a_ncookies = 0; ap->a_ncookies = NULL; } dmp = VFSTODEVFS(ap->a_vp->v_mount); if (devfs_populate_vp(ap->a_vp) != 0) { if (tmp_ncookies != NULL) ap->a_ncookies = tmp_ncookies; return (EIO); } error = 0; de = ap->a_vp->v_data; off = 0; TAILQ_FOREACH(dd, &de->de_dlist, de_list) { KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__)); if (dd->de_flags & (DE_COVERED | DE_WHITEOUT)) continue; if (devfs_prison_check(dd, uio->uio_td)) continue; if (dd->de_dirent->d_type == DT_DIR) de = dd->de_dir; else de = dd; dp = dd->de_dirent; MPASS(dp->d_reclen == GENERIC_DIRSIZ(dp)); if (dp->d_reclen > uio->uio_resid) break; dp->d_fileno = de->de_inode; /* NOTE: d_off is the offset for the *next* entry. */ dp->d_off = off + dp->d_reclen; if (off >= uio->uio_offset) { error = vfs_read_dirent(ap, dp, off); if (error) break; } off += dp->d_reclen; } sx_xunlock(&dmp->dm_lock); uio->uio_offset = off; /* * Restore ap->a_ncookies if it wasn't originally NULL in the first * place. */ if (tmp_ncookies != NULL) ap->a_ncookies = tmp_ncookies; return (error); } static int devfs_readlink(struct vop_readlink_args *ap) { struct devfs_dirent *de; de = ap->a_vp->v_data; return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio)); } static void devfs_reclaiml(struct vnode *vp) { struct devfs_dirent *de; mtx_assert(&devfs_de_interlock, MA_OWNED); de = vp->v_data; if (de != NULL) { MPASS(de->de_usecount == 0); de->de_vnode = NULL; vp->v_data = NULL; } } static int devfs_reclaim(struct vop_reclaim_args *ap) { struct vnode *vp; vp = ap->a_vp; mtx_lock(&devfs_de_interlock); devfs_reclaiml(vp); mtx_unlock(&devfs_de_interlock); return (0); } static int devfs_reclaim_vchr(struct vop_reclaim_args *ap) { struct vnode *vp; struct cdev *dev; vp = ap->a_vp; MPASS(vp->v_type == VCHR); mtx_lock(&devfs_de_interlock); VI_LOCK(vp); devfs_usecount_subl(vp); devfs_reclaiml(vp); mtx_unlock(&devfs_de_interlock); dev_lock(); dev = vp->v_rdev; vp->v_rdev = NULL; dev_unlock(); VI_UNLOCK(vp); if (dev != NULL) dev_rel(dev); return (0); } static int devfs_remove(struct vop_remove_args *ap) { struct vnode *dvp = ap->a_dvp; struct vnode *vp = ap->a_vp; struct devfs_dirent *dd; struct devfs_dirent *de, *de_covered; struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount); ASSERT_VOP_ELOCKED(dvp, "devfs_remove"); ASSERT_VOP_ELOCKED(vp, "devfs_remove"); sx_xlock(&dmp->dm_lock); dd = ap->a_dvp->v_data; de = vp->v_data; if (de->de_cdp == NULL) { TAILQ_REMOVE(&dd->de_dlist, de, de_list); if (de->de_dirent->d_type == DT_LNK) { de_covered = devfs_find(dd, de->de_dirent->d_name, de->de_dirent->d_namlen, 0); if (de_covered != NULL) de_covered->de_flags &= ~DE_COVERED; } /* We need to unlock dvp because devfs_delete() may lock it. */ VOP_UNLOCK(vp); if (dvp != vp) VOP_UNLOCK(dvp); devfs_delete(dmp, de, 0); sx_xunlock(&dmp->dm_lock); if (dvp != vp) vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); } else { de->de_flags |= DE_WHITEOUT; sx_xunlock(&dmp->dm_lock); } return (0); } /* * Revoke is called on a tty when a terminal session ends. The vnode * is orphaned by setting v_op to deadfs so we need to let go of it * as well so that we create a new one next time around. * */ static int devfs_revoke(struct vop_revoke_args *ap) { struct vnode *vp = ap->a_vp, *vp2; struct cdev *dev; struct cdev_priv *cdp; struct devfs_dirent *de; enum vgetstate vs; u_int i; KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL")); dev = vp->v_rdev; cdp = cdev2priv(dev); dev_lock(); cdp->cdp_inuse++; dev_unlock(); vhold(vp); vgone(vp); vdrop(vp); VOP_UNLOCK(vp); loop: for (;;) { mtx_lock(&devfs_de_interlock); dev_lock(); vp2 = NULL; for (i = 0; i <= cdp->cdp_maxdirent; i++) { de = cdp->cdp_dirents[i]; if (de == NULL) continue; vp2 = de->de_vnode; if (vp2 != NULL) { dev_unlock(); vs = vget_prep(vp2); mtx_unlock(&devfs_de_interlock); if (vget_finish(vp2, LK_EXCLUSIVE, vs) != 0) goto loop; vhold(vp2); vgone(vp2); vdrop(vp2); vput(vp2); break; } } if (vp2 != NULL) { continue; } dev_unlock(); mtx_unlock(&devfs_de_interlock); break; } dev_lock(); cdp->cdp_inuse--; if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) { TAILQ_REMOVE(&cdevp_list, cdp, cdp_list); dev_unlock(); dev_rel(&cdp->cdp_c); } else dev_unlock(); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); return (0); } static int devfs_rioctl(struct vop_ioctl_args *ap) { struct vnode *vp; struct devfs_mount *dmp; int error; vp = ap->a_vp; vn_lock(vp, LK_SHARED | LK_RETRY); if (VN_IS_DOOMED(vp)) { VOP_UNLOCK(vp); return (EBADF); } dmp = VFSTODEVFS(vp->v_mount); sx_xlock(&dmp->dm_lock); VOP_UNLOCK(vp); DEVFS_DMP_HOLD(dmp); devfs_populate(dmp); if (DEVFS_DMP_DROP(dmp)) { sx_xunlock(&dmp->dm_lock); devfs_unmount_final(dmp); return (ENOENT); } error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td); sx_xunlock(&dmp->dm_lock); return (error); } static int devfs_rread(struct vop_read_args *ap) { if (ap->a_vp->v_type != VDIR) return (EINVAL); return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL)); } static int devfs_setattr(struct vop_setattr_args *ap) { struct devfs_dirent *de; struct vattr *vap; struct vnode *vp; struct thread *td; int c, error; uid_t uid; gid_t gid; vap = ap->a_vap; vp = ap->a_vp; td = curthread; if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) || (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) || (vap->va_blocksize != VNOVAL) || (vap->va_flags != VNOVAL && vap->va_flags != 0) || (vap->va_rdev != VNOVAL) || ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) { return (EINVAL); } error = devfs_populate_vp(vp); if (error != 0) return (error); de = vp->v_data; if (vp->v_type == VDIR) de = de->de_dir; c = 0; if (vap->va_uid == (uid_t)VNOVAL) uid = de->de_uid; else uid = vap->va_uid; if (vap->va_gid == (gid_t)VNOVAL) gid = de->de_gid; else gid = vap->va_gid; if (uid != de->de_uid || gid != de->de_gid) { if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid || (gid != de->de_gid && !groupmember(gid, ap->a_cred))) { error = priv_check(td, PRIV_VFS_CHOWN); if (error != 0) goto ret; } de->de_uid = uid; de->de_gid = gid; c = 1; } if (vap->va_mode != (mode_t)VNOVAL) { if (ap->a_cred->cr_uid != de->de_uid) { error = priv_check(td, PRIV_VFS_ADMIN); if (error != 0) goto ret; } de->de_mode = vap->va_mode; c = 1; } if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { error = vn_utimes_perm(vp, vap, ap->a_cred, td); if (error != 0) goto ret; if (vap->va_atime.tv_sec != VNOVAL) { if (vp->v_type == VCHR) vp->v_rdev->si_atime = vap->va_atime; else de->de_atime = vap->va_atime; } if (vap->va_mtime.tv_sec != VNOVAL) { if (vp->v_type == VCHR) vp->v_rdev->si_mtime = vap->va_mtime; else de->de_mtime = vap->va_mtime; } c = 1; } if (c) { if (vp->v_type == VCHR) vfs_timestamp(&vp->v_rdev->si_ctime); else vfs_timestamp(&de->de_mtime); } ret: sx_xunlock(&VFSTODEVFS(vp->v_mount)->dm_lock); return (error); } #ifdef MAC static int devfs_setlabel(struct vop_setlabel_args *ap) { struct vnode *vp; struct devfs_dirent *de; vp = ap->a_vp; de = vp->v_data; mac_vnode_relabel(ap->a_cred, vp, ap->a_label); mac_devfs_update(vp->v_mount, de, vp); return (0); } #endif static int devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td) { return (vnops.fo_stat(fp, sb, cred, td)); } static int devfs_symlink(struct vop_symlink_args *ap) { int i, error; struct devfs_dirent *dd; struct devfs_dirent *de, *de_covered, *de_dotdot; struct devfs_mount *dmp; error = priv_check(curthread, PRIV_DEVFS_SYMLINK); if (error) return(error); dmp = VFSTODEVFS(ap->a_dvp->v_mount); if (devfs_populate_vp(ap->a_dvp) != 0) return (ENOENT); dd = ap->a_dvp->v_data; de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen); de->de_flags = DE_USER; de->de_uid = 0; de->de_gid = 0; de->de_mode = 0755; de->de_inode = alloc_unr(devfs_inos); de->de_dir = dd; de->de_dirent->d_type = DT_LNK; i = strlen(ap->a_target) + 1; de->de_symlink = malloc(i, M_DEVFS, M_WAITOK); bcopy(ap->a_target, de->de_symlink, i); #ifdef MAC mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de); #endif de_covered = devfs_find(dd, de->de_dirent->d_name, de->de_dirent->d_namlen, 0); if (de_covered != NULL) { if ((de_covered->de_flags & DE_USER) != 0) { devfs_delete(dmp, de, DEVFS_DEL_NORECURSE); sx_xunlock(&dmp->dm_lock); return (EEXIST); } KASSERT((de_covered->de_flags & DE_COVERED) == 0, ("devfs_symlink: entry %p already covered", de_covered)); de_covered->de_flags |= DE_COVERED; } de_dotdot = TAILQ_FIRST(&dd->de_dlist); /* "." */ de_dotdot = TAILQ_NEXT(de_dotdot, de_list); /* ".." */ TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list); devfs_dir_ref_de(dmp, dd); devfs_rules_apply(dmp, de); return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp)); } static int devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td) { return (vnops.fo_truncate(fp, length, cred, td)); } static int devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred, int flags, struct thread *td) { struct cdev *dev; int error, ioflag, ref; ssize_t resid; struct cdevsw *dsw; struct file *fpop; if (uio->uio_resid > DEVFS_IOSIZE_MAX) return (EINVAL); fpop = td->td_fpop; error = devfs_fp_check(fp, &dev, &dsw, &ref); if (error != 0) { error = vnops.fo_write(fp, uio, cred, flags, td); return (error); } KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td)); ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC); if (ioflag & O_DIRECT) ioflag |= IO_DIRECT; foffset_lock_uio(fp, uio, flags | FOF_NOLOCK); resid = uio->uio_resid; error = dsw->d_write(dev, uio, ioflag); if (uio->uio_resid != resid || (error == 0 && resid != 0)) { devfs_timestamp(&dev->si_ctime); dev->si_mtime = dev->si_ctime; } td->td_fpop = fpop; dev_relthread(dev, ref); foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF_W); return (error); } static int devfs_mmap_f(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff, struct thread *td) { struct cdev *dev; struct cdevsw *dsw; struct mount *mp; struct vnode *vp; struct file *fpop; vm_object_t object; vm_prot_t maxprot; int error, ref; vp = fp->f_vnode; /* * Ensure that file and memory protections are * compatible. */ mp = vp->v_mount; if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) { maxprot = VM_PROT_NONE; if ((prot & VM_PROT_EXECUTE) != 0) return (EACCES); } else maxprot = VM_PROT_EXECUTE; if ((fp->f_flag & FREAD) != 0) maxprot |= VM_PROT_READ; else if ((prot & VM_PROT_READ) != 0) return (EACCES); /* * If we are sharing potential changes via MAP_SHARED and we * are trying to get write permission although we opened it * without asking for it, bail out. * * Note that most character devices always share mappings. * The one exception is that D_MMAP_ANON devices * (i.e. /dev/zero) permit private writable mappings. * * Rely on vm_mmap_cdev() to fail invalid MAP_PRIVATE requests * as well as updating maxprot to permit writing for * D_MMAP_ANON devices rather than doing that here. */ if ((flags & MAP_SHARED) != 0) { if ((fp->f_flag & FWRITE) != 0) maxprot |= VM_PROT_WRITE; else if ((prot & VM_PROT_WRITE) != 0) return (EACCES); } maxprot &= cap_maxprot; fpop = td->td_fpop; error = devfs_fp_check(fp, &dev, &dsw, &ref); if (error != 0) return (error); error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, dev, dsw, &foff, &object); td->td_fpop = fpop; dev_relthread(dev, ref); if (error != 0) return (error); error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object, foff, FALSE, td); if (error != 0) vm_object_deallocate(object); return (error); } dev_t dev2udev(struct cdev *x) { if (x == NULL) return (NODEV); return (cdev2priv(x)->cdp_inode); } static struct fileops devfs_ops_f = { .fo_read = devfs_read_f, .fo_write = devfs_write_f, .fo_truncate = devfs_truncate_f, .fo_ioctl = devfs_ioctl_f, .fo_poll = devfs_poll_f, .fo_kqfilter = devfs_kqfilter_f, .fo_stat = devfs_stat_f, .fo_close = devfs_close_f, .fo_chmod = vn_chmod, .fo_chown = vn_chown, .fo_sendfile = vn_sendfile, .fo_seek = vn_seek, .fo_fill_kinfo = vn_fill_kinfo, .fo_mmap = devfs_mmap_f, .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE }; /* Vops for non-CHR vnodes in /dev. */ static struct vop_vector devfs_vnodeops = { .vop_default = &default_vnodeops, .vop_access = devfs_access, .vop_getattr = devfs_getattr, .vop_ioctl = devfs_rioctl, .vop_lookup = devfs_lookup, .vop_mknod = devfs_mknod, .vop_pathconf = devfs_pathconf, .vop_read = devfs_rread, .vop_readdir = devfs_readdir, .vop_readlink = devfs_readlink, .vop_reclaim = devfs_reclaim, .vop_remove = devfs_remove, .vop_revoke = devfs_revoke, .vop_setattr = devfs_setattr, #ifdef MAC .vop_setlabel = devfs_setlabel, #endif .vop_symlink = devfs_symlink, .vop_vptocnp = devfs_vptocnp, .vop_lock1 = vop_lock, .vop_unlock = vop_unlock, .vop_islocked = vop_islocked, }; VFS_VOP_VECTOR_REGISTER(devfs_vnodeops); /* Vops for VCHR vnodes in /dev. */ static struct vop_vector devfs_specops = { .vop_default = &default_vnodeops, .vop_access = devfs_access, .vop_bmap = VOP_PANIC, .vop_close = devfs_close, .vop_create = VOP_PANIC, .vop_fsync = vop_stdfsync, .vop_getattr = devfs_getattr, .vop_ioctl = devfs_ioctl, .vop_link = VOP_PANIC, .vop_mkdir = VOP_PANIC, .vop_mknod = VOP_PANIC, .vop_open = devfs_open, .vop_pathconf = devfs_pathconf, .vop_poll = dead_poll, .vop_print = devfs_print, .vop_read = dead_read, .vop_readdir = VOP_PANIC, .vop_readlink = VOP_PANIC, .vop_reallocblks = VOP_PANIC, .vop_reclaim = devfs_reclaim_vchr, .vop_remove = devfs_remove, .vop_rename = VOP_PANIC, .vop_revoke = devfs_revoke, .vop_rmdir = VOP_PANIC, .vop_setattr = devfs_setattr, #ifdef MAC .vop_setlabel = devfs_setlabel, #endif .vop_strategy = VOP_PANIC, .vop_symlink = VOP_PANIC, .vop_vptocnp = devfs_vptocnp, .vop_write = dead_write, .vop_lock1 = vop_lock, .vop_unlock = vop_unlock, .vop_islocked = vop_islocked, }; VFS_VOP_VECTOR_REGISTER(devfs_specops); /* * Our calling convention to the device drivers used to be that we passed * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_ * flags instead since that's what open(), close() and ioctl() takes and * we don't really want vnode.h in device drivers. * We solved the source compatibility by redefining some vnode flags to * be the same as the fcntl ones and by sending down the bitwise OR of * the respective fcntl/vnode flags. These CTASSERTS make sure nobody * pulls the rug out under this. */ CTASSERT(O_NONBLOCK == IO_NDELAY); CTASSERT(O_FSYNC == IO_SYNC); diff --git a/sys/fs/ext2fs/ext2_lookup.c b/sys/fs/ext2fs/ext2_lookup.c index 68482e38feb8..3b435bf96a8a 100644 --- a/sys/fs/ext2fs/ext2_lookup.c +++ b/sys/fs/ext2fs/ext2_lookup.c @@ -1,1294 +1,1294 @@ /*- * modified for Lites 1.1 * * Aug 1995, Godmar Back (gback@cs.utah.edu) * University of Utah, Department of Computer Science */ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)ufs_lookup.c 8.6 (Berkeley) 4/1/94 * $FreeBSD$ */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include SDT_PROVIDER_DECLARE(ext2fs); /* * ext2fs trace probe: * arg0: verbosity. Higher numbers give more verbose messages * arg1: Textual message */ SDT_PROBE_DEFINE2(ext2fs, , lookup, trace, "int", "char*"); SDT_PROBE_DEFINE4(ext2fs, , trace, ext2_dirbad_error, "char*", "ino_t", "doff_t", "char*"); SDT_PROBE_DEFINE5(ext2fs, , trace, ext2_dirbadentry_error, "char*", "int", "uint32_t", "uint16_t", "uint8_t"); #ifdef INVARIANTS static int dirchk = 1; #else static int dirchk = 0; #endif static SYSCTL_NODE(_vfs, OID_AUTO, e2fs, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "EXT2FS filesystem"); SYSCTL_INT(_vfs_e2fs, OID_AUTO, dircheck, CTLFLAG_RW, &dirchk, 0, ""); /* DIRBLKSIZE in ffs is DEV_BSIZE (in most cases 512) while it is the native blocksize in ext2fs - thus, a #define is no longer appropriate */ #undef DIRBLKSIZ static u_char ext2_ft_to_dt[] = { DT_UNKNOWN, /* EXT2_FT_UNKNOWN */ DT_REG, /* EXT2_FT_REG_FILE */ DT_DIR, /* EXT2_FT_DIR */ DT_CHR, /* EXT2_FT_CHRDEV */ DT_BLK, /* EXT2_FT_BLKDEV */ DT_FIFO, /* EXT2_FT_FIFO */ DT_SOCK, /* EXT2_FT_SOCK */ DT_LNK, /* EXT2_FT_SYMLINK */ }; #define FTTODT(ft) \ ((ft) < nitems(ext2_ft_to_dt) ? ext2_ft_to_dt[(ft)] : DT_UNKNOWN) static u_char dt_to_ext2_ft[] = { EXT2_FT_UNKNOWN, /* DT_UNKNOWN */ EXT2_FT_FIFO, /* DT_FIFO */ EXT2_FT_CHRDEV, /* DT_CHR */ EXT2_FT_UNKNOWN, /* unused */ EXT2_FT_DIR, /* DT_DIR */ EXT2_FT_UNKNOWN, /* unused */ EXT2_FT_BLKDEV, /* DT_BLK */ EXT2_FT_UNKNOWN, /* unused */ EXT2_FT_REG_FILE, /* DT_REG */ EXT2_FT_UNKNOWN, /* unused */ EXT2_FT_SYMLINK, /* DT_LNK */ EXT2_FT_UNKNOWN, /* unused */ EXT2_FT_SOCK, /* DT_SOCK */ EXT2_FT_UNKNOWN, /* unused */ EXT2_FT_UNKNOWN, /* DT_WHT */ }; #define DTTOFT(dt) \ ((dt) < nitems(dt_to_ext2_ft) ? dt_to_ext2_ft[(dt)] : EXT2_FT_UNKNOWN) static int ext2_dirbadentry(struct vnode *dp, struct ext2fs_direct_2 *de, int entryoffsetinblock); static int ext2_is_dot_entry(struct componentname *cnp); static int ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp, struct componentname *cnp, ino_t *dd_ino); static int ext2_is_dot_entry(struct componentname *cnp) { if (cnp->cn_namelen <= 2 && cnp->cn_nameptr[0] == '.' && (cnp->cn_nameptr[1] == '.' || cnp->cn_nameptr[1] == '\0')) return (1); return (0); } /* * Vnode op for reading directories. */ int ext2_readdir(struct vop_readdir_args *ap) { struct vnode *vp = ap->a_vp; struct uio *uio = ap->a_uio; struct buf *bp; struct inode *ip; struct ext2fs_direct_2 *dp, *edp; u_long *cookies; struct dirent dstdp; off_t offset, startoffset; size_t readcnt, skipcnt; ssize_t startresid; u_int ncookies; int DIRBLKSIZ = VTOI(ap->a_vp)->i_e2fs->e2fs_bsize; int error; if (uio->uio_offset < 0) return (EINVAL); ip = VTOI(vp); if (ap->a_ncookies != NULL) { if (uio->uio_resid < 0) ncookies = 0; else ncookies = uio->uio_resid; if (uio->uio_offset >= ip->i_size) ncookies = 0; else if (ip->i_size - uio->uio_offset < ncookies) ncookies = ip->i_size - uio->uio_offset; ncookies = ncookies / (offsetof(struct ext2fs_direct_2, e2d_namlen) + 4) + 1; cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK); *ap->a_ncookies = ncookies; *ap->a_cookies = cookies; } else { ncookies = 0; cookies = NULL; } offset = startoffset = uio->uio_offset; startresid = uio->uio_resid; error = 0; while (error == 0 && uio->uio_resid > 0 && uio->uio_offset < ip->i_size) { error = ext2_blkatoff(vp, uio->uio_offset, NULL, &bp); if (error) break; if (bp->b_offset + bp->b_bcount > ip->i_size) readcnt = ip->i_size - bp->b_offset; else readcnt = bp->b_bcount; skipcnt = (size_t)(uio->uio_offset - bp->b_offset) & ~(size_t)(DIRBLKSIZ - 1); offset = bp->b_offset + skipcnt; dp = (struct ext2fs_direct_2 *)&bp->b_data[skipcnt]; edp = (struct ext2fs_direct_2 *)&bp->b_data[readcnt]; while (error == 0 && uio->uio_resid > 0 && dp < edp) { if (le16toh(dp->e2d_reclen) <= offsetof(struct ext2fs_direct_2, e2d_namlen) || (caddr_t)dp + le16toh(dp->e2d_reclen) > (caddr_t)edp) { error = EIO; break; } /*- * "New" ext2fs directory entries differ in 3 ways * from ufs on-disk ones: * - the name is not necessarily NUL-terminated. * - the file type field always exists and always * follows the name length field. * - the file type is encoded in a different way. * * "Old" ext2fs directory entries need no special * conversions, since they are binary compatible * with "new" entries having a file type of 0 (i.e., * EXT2_FT_UNKNOWN). Splitting the old name length * field didn't make a mess like it did in ufs, * because ext2fs uses a machine-independent disk * layout. */ dstdp.d_namlen = dp->e2d_namlen; dstdp.d_type = FTTODT(dp->e2d_type); if (offsetof(struct ext2fs_direct_2, e2d_namlen) + dstdp.d_namlen > le16toh(dp->e2d_reclen)) { error = EIO; break; } if (offset < startoffset || le32toh(dp->e2d_ino) == 0) goto nextentry; dstdp.d_fileno = le32toh(dp->e2d_ino); dstdp.d_reclen = GENERIC_DIRSIZ(&dstdp); bcopy(dp->e2d_name, dstdp.d_name, dstdp.d_namlen); /* NOTE: d_off is the offset of the *next* entry. */ dstdp.d_off = offset + le16toh(dp->e2d_reclen); dirent_terminate(&dstdp); if (dstdp.d_reclen > uio->uio_resid) { if (uio->uio_resid == startresid) error = EINVAL; else error = EJUSTRETURN; break; } /* Advance dp. */ error = uiomove((caddr_t)&dstdp, dstdp.d_reclen, uio); if (error) break; if (cookies != NULL) { KASSERT(ncookies > 0, ("ext2_readdir: cookies buffer too small")); *cookies = offset + le16toh(dp->e2d_reclen); cookies++; ncookies--; } nextentry: offset += le16toh(dp->e2d_reclen); dp = (struct ext2fs_direct_2 *)((caddr_t)dp + le16toh(dp->e2d_reclen)); } bqrelse(bp); uio->uio_offset = offset; } /* We need to correct uio_offset. */ uio->uio_offset = offset; if (error == EJUSTRETURN) error = 0; if (ap->a_ncookies != NULL) { if (error == 0) { ap->a_ncookies -= ncookies; } else { free(*ap->a_cookies, M_TEMP); *ap->a_ncookies = 0; *ap->a_cookies = NULL; } } if (error == 0 && ap->a_eofflag) *ap->a_eofflag = ip->i_size <= uio->uio_offset; return (error); } /* * Convert a component of a pathname into a pointer to a locked inode. * This is a very central and rather complicated routine. * If the file system is not maintained in a strict tree hierarchy, * this can result in a deadlock situation (see comments in code below). * * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending * on whether the name is to be looked up, created, renamed, or deleted. * When CREATE, RENAME, or DELETE is specified, information usable in * creating, renaming, or deleting a directory entry may be calculated. * If flag has LOCKPARENT or'ed into it and the target of the pathname * exists, lookup returns both the target and its parent directory locked. * When creating or renaming and LOCKPARENT is specified, the target may * not be ".". When deleting and LOCKPARENT is specified, the target may * be "."., but the caller must check to ensure it does an vrele and vput * instead of two vputs. * * Overall outline of ext2_lookup: * * search for name in directory, to found or notfound * notfound: * if creating, return locked directory, leaving info on available slots * else return error * found: * if at end of path and deleting, return information to allow delete * if at end of path and rewriting (RENAME and LOCKPARENT), lock target * inode and return info to allow rewrite * if not at end, add name to cache; if at end and neither creating * nor deleting, add name to cache */ int ext2_lookup(struct vop_cachedlookup_args *ap) { return (ext2_lookup_ino(ap->a_dvp, ap->a_vpp, ap->a_cnp, NULL)); } static int ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp, struct componentname *cnp, ino_t *dd_ino) { struct inode *dp; /* inode for directory being searched */ struct buf *bp; /* a buffer of directory entries */ struct ext2fs_direct_2 *ep; /* the current directory entry */ int entryoffsetinblock; /* offset of ep in bp's buffer */ struct ext2fs_searchslot ss; doff_t i_diroff; /* cached i_diroff value */ doff_t i_offset; /* cached i_offset value */ int numdirpasses; /* strategy for directory search */ doff_t endsearch; /* offset to end directory search */ doff_t prevoff; /* prev entry dp->i_offset */ struct vnode *pdp; /* saved dp during symlink work */ struct vnode *tdp; /* returned by VFS_VGET */ doff_t enduseful; /* pointer past last used dir slot */ u_long bmask; /* block offset mask */ int error; struct ucred *cred = cnp->cn_cred; int flags = cnp->cn_flags; int nameiop = cnp->cn_nameiop; ino_t ino, ino1; int ltype; int entry_found = 0; int DIRBLKSIZ = VTOI(vdp)->i_e2fs->e2fs_bsize; if (vpp != NULL) *vpp = NULL; dp = VTOI(vdp); bmask = VFSTOEXT2(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1; restart: bp = NULL; ss.slotoffset = -1; /* * We now have a segment name to search for, and a directory to search. * * Suppress search for slots unless creating * file and at end of pathname, in which case * we watch for a place to put the new file in * case it doesn't already exist. */ i_diroff = dp->i_diroff; ss.slotstatus = FOUND; ss.slotfreespace = ss.slotsize = ss.slotneeded = 0; if ((nameiop == CREATE || nameiop == RENAME) && (flags & ISLASTCN)) { ss.slotstatus = NONE; ss.slotneeded = EXT2_DIR_REC_LEN(cnp->cn_namelen); /* * was ss.slotneeded = (sizeof(struct direct) - MAXNAMLEN + * cnp->cn_namelen + 3) &~ 3; */ } /* * Try to lookup dir entry using htree directory index. * * If we got an error or we want to find '.' or '..' entry, * we will fall back to linear search. */ if (!ext2_is_dot_entry(cnp) && ext2_htree_has_idx(dp)) { numdirpasses = 1; entryoffsetinblock = 0; switch (ext2_htree_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen, &bp, &entryoffsetinblock, &i_offset, &prevoff, &enduseful, &ss)) { case 0: ep = (struct ext2fs_direct_2 *)((char *)bp->b_data + (i_offset & bmask)); goto foundentry; case ENOENT: i_offset = roundup2(dp->i_size, DIRBLKSIZ); goto notfound; default: /* * Something failed; just fallback to do a linear * search. */ break; } } /* * If there is cached information on a previous search of * this directory, pick up where we last left off. * We cache only lookups as these are the most common * and have the greatest payoff. Caching CREATE has little * benefit as it usually must search the entire directory * to determine that the entry does not exist. Caching the * location of the last DELETE or RENAME has not reduced * profiling time and hence has been removed in the interest * of simplicity. */ if (nameiop != LOOKUP || i_diroff == 0 || i_diroff > dp->i_size) { entryoffsetinblock = 0; i_offset = 0; numdirpasses = 1; } else { i_offset = i_diroff; if ((entryoffsetinblock = i_offset & bmask) && (error = ext2_blkatoff(vdp, (off_t)i_offset, NULL, &bp))) return (error); numdirpasses = 2; nchstats.ncs_2passes++; } prevoff = i_offset; endsearch = roundup2(dp->i_size, DIRBLKSIZ); enduseful = 0; searchloop: while (i_offset < endsearch) { /* * If necessary, get the next directory block. */ if (bp != NULL) brelse(bp); error = ext2_blkatoff(vdp, (off_t)i_offset, NULL, &bp); if (error != 0) return (error); entryoffsetinblock = 0; if (ss.slotstatus == NONE) { ss.slotoffset = -1; ss.slotfreespace = 0; } error = ext2_search_dirblock(dp, bp->b_data, &entry_found, cnp->cn_nameptr, cnp->cn_namelen, &entryoffsetinblock, &i_offset, &prevoff, &enduseful, &ss); if (error != 0) { brelse(bp); return (error); } if (entry_found) { ep = (struct ext2fs_direct_2 *)((char *)bp->b_data + (entryoffsetinblock & bmask)); foundentry: ino = le32toh(ep->e2d_ino); goto found; } } notfound: /* * If we started in the middle of the directory and failed * to find our target, we must check the beginning as well. */ if (numdirpasses == 2) { numdirpasses--; i_offset = 0; endsearch = i_diroff; goto searchloop; } if (bp != NULL) brelse(bp); /* * If creating, and at end of pathname and current * directory has not been removed, then can consider * allowing file to be created. */ if ((nameiop == CREATE || nameiop == RENAME) && (flags & ISLASTCN) && dp->i_nlink != 0) { /* * Access for write is interpreted as allowing * creation of files in the directory. */ - if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0) + if ((error = VOP_ACCESS(vdp, VWRITE, cred, curthread)) != 0) return (error); /* * Return an indication of where the new directory * entry should be put. If we didn't find a slot, * then set dp->i_count to 0 indicating * that the new slot belongs at the end of the * directory. If we found a slot, then the new entry * can be put in the range from dp->i_offset to * dp->i_offset + dp->i_count. */ if (ss.slotstatus == NONE) { dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ); dp->i_count = 0; enduseful = dp->i_offset; } else { dp->i_offset = ss.slotoffset; dp->i_count = ss.slotsize; if (enduseful < ss.slotoffset + ss.slotsize) enduseful = ss.slotoffset + ss.slotsize; } dp->i_endoff = roundup2(enduseful, DIRBLKSIZ); /* * We return with the directory locked, so that * the parameters we set up above will still be * valid if we actually decide to do a direnter(). * We return ni_vp == NULL to indicate that the entry * does not currently exist; we leave a pointer to * the (locked) directory inode in ndp->ni_dvp. * The pathname buffer is saved so that the name * can be obtained later. * * NB - if the directory is unlocked, then this * information cannot be used. */ cnp->cn_flags |= SAVENAME; return (EJUSTRETURN); } /* * Insert name into cache (as non-existent) if appropriate. */ if ((cnp->cn_flags & MAKEENTRY) != 0) cache_enter(vdp, NULL, cnp); return (ENOENT); found: if (dd_ino != NULL) *dd_ino = ino; if (numdirpasses == 2) nchstats.ncs_pass2++; /* * Check that directory length properly reflects presence * of this entry. */ if (entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen) > dp->i_size) { ext2_dirbad(dp, i_offset, "i_size too small"); dp->i_size = entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen); dp->i_flag |= IN_CHANGE | IN_UPDATE; } brelse(bp); /* * Found component in pathname. * If the final component of path name, save information * in the cache as to where the entry was found. */ if ((flags & ISLASTCN) && nameiop == LOOKUP) dp->i_diroff = rounddown2(i_offset, DIRBLKSIZ); /* * If deleting, and at end of pathname, return * parameters which can be used to remove file. */ if (nameiop == DELETE && (flags & ISLASTCN)) { if (flags & LOCKPARENT) ASSERT_VOP_ELOCKED(vdp, __FUNCTION__); /* * Write access to directory required to delete files. */ - if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0) + if ((error = VOP_ACCESS(vdp, VWRITE, cred, curthread)) != 0) return (error); /* * Return pointer to current entry in dp->i_offset, * and distance past previous entry (if there * is a previous entry in this block) in dp->i_count. * Save directory inode pointer in ndp->ni_dvp for dirremove(). * * Technically we shouldn't be setting these in the * WANTPARENT case (first lookup in rename()), but any * lookups that will result in directory changes will * overwrite these. */ dp->i_offset = i_offset; if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0) dp->i_count = 0; else dp->i_count = dp->i_offset - prevoff; if (dd_ino != NULL) return (0); if (dp->i_number == ino) { VREF(vdp); *vpp = vdp; return (0); } if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE, &tdp)) != 0) return (error); /* * If directory is "sticky", then user must own * the directory, or the file in it, else she * may not delete it (unless she's root). This * implements append-only directories. */ if ((dp->i_mode & ISVTX) && cred->cr_uid != 0 && cred->cr_uid != dp->i_uid && VTOI(tdp)->i_uid != cred->cr_uid) { vput(tdp); return (EPERM); } *vpp = tdp; return (0); } /* * If rewriting (RENAME), return the inode and the * information required to rewrite the present directory * Must get inode of directory entry to verify it's a * regular file, or empty directory. */ if (nameiop == RENAME && (flags & ISLASTCN)) { - if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0) + if ((error = VOP_ACCESS(vdp, VWRITE, cred, curthread)) != 0) return (error); /* * Careful about locking second inode. * This can only occur if the target is ".". */ dp->i_offset = i_offset; if (dp->i_number == ino) return (EISDIR); if (dd_ino != NULL) return (0); if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE, &tdp)) != 0) return (error); *vpp = tdp; cnp->cn_flags |= SAVENAME; return (0); } if (dd_ino != NULL) return (0); /* * Step through the translation in the name. We do not `vput' the * directory because we may need it again if a symbolic link * is relative to the current directory. Instead we save it * unlocked as "pdp". We must get the target inode before unlocking * the directory to insure that the inode will not be removed * before we get it. We prevent deadlock by always fetching * inodes from the root, moving down the directory tree. Thus * when following backward pointers ".." we must unlock the * parent directory before getting the requested directory. * There is a potential race condition here if both the current * and parent directories are removed before the VFS_VGET for the * inode associated with ".." returns. We hope that this occurs * infrequently since we cannot avoid this race condition without * implementing a sophisticated deadlock detection algorithm. * Note also that this simple deadlock detection scheme will not * work if the file system has any hard links other than ".." * that point backwards in the directory structure. */ pdp = vdp; if (flags & ISDOTDOT) { error = vn_vget_ino(pdp, ino, cnp->cn_lkflags, &tdp); if (VN_IS_DOOMED(pdp)) { if (error == 0) vput(tdp); error = ENOENT; } if (error) return (error); /* * Recheck that ".." entry in the vdp directory points * to the inode we looked up before vdp lock was * dropped. */ error = ext2_lookup_ino(pdp, NULL, cnp, &ino1); if (error) { vput(tdp); return (error); } if (ino1 != ino) { vput(tdp); goto restart; } *vpp = tdp; } else if (dp->i_number == ino) { VREF(vdp); /* we want ourself, ie "." */ /* * When we lookup "." we still can be asked to lock it * differently. */ ltype = cnp->cn_lkflags & LK_TYPE_MASK; if (ltype != VOP_ISLOCKED(vdp)) { if (ltype == LK_EXCLUSIVE) vn_lock(vdp, LK_UPGRADE | LK_RETRY); else /* if (ltype == LK_SHARED) */ vn_lock(vdp, LK_DOWNGRADE | LK_RETRY); } *vpp = vdp; } else { if ((error = VFS_VGET(vdp->v_mount, ino, cnp->cn_lkflags, &tdp)) != 0) return (error); *vpp = tdp; } /* * Insert name into cache if appropriate. */ if (cnp->cn_flags & MAKEENTRY) cache_enter(vdp, *vpp, cnp); return (0); } int ext2_search_dirblock(struct inode *ip, void *data, int *foundp, const char *name, int namelen, int *entryoffsetinblockp, doff_t *offp, doff_t *prevoffp, doff_t *endusefulp, struct ext2fs_searchslot *ssp) { struct vnode *vdp; struct ext2fs_direct_2 *ep, *top; uint32_t bsize = ip->i_e2fs->e2fs_bsize; int offset = *entryoffsetinblockp; int namlen; vdp = ITOV(ip); ep = (struct ext2fs_direct_2 *)((char *)data + offset); top = (struct ext2fs_direct_2 *)((char *)data + bsize); while (ep < top) { /* * Full validation checks are slow, so we only check * enough to insure forward progress through the * directory. Complete checks can be run by setting * "vfs.e2fs.dirchk" to be true. */ if (le16toh(ep->e2d_reclen) == 0 || (dirchk && ext2_dirbadentry(vdp, ep, offset))) { int i; ext2_dirbad(ip, *offp, "mangled entry"); i = bsize - (offset & (bsize - 1)); *offp += i; offset += i; continue; } /* * If an appropriate sized slot has not yet been found, * check to see if one is available. Also accumulate space * in the current block so that we can determine if * compaction is viable. */ if (ssp->slotstatus != FOUND) { int size = le16toh(ep->e2d_reclen); if (ep->e2d_ino != 0) size -= EXT2_DIR_REC_LEN(ep->e2d_namlen); else if (ext2_is_dirent_tail(ip, ep)) size -= sizeof(struct ext2fs_direct_tail); if (size > 0) { if (size >= ssp->slotneeded) { ssp->slotstatus = FOUND; ssp->slotoffset = *offp; ssp->slotsize = le16toh(ep->e2d_reclen); } else if (ssp->slotstatus == NONE) { ssp->slotfreespace += size; if (ssp->slotoffset == -1) ssp->slotoffset = *offp; if (ssp->slotfreespace >= ssp->slotneeded) { ssp->slotstatus = COMPACT; ssp->slotsize = *offp + le16toh(ep->e2d_reclen) - ssp->slotoffset; } } } } /* * Check for a name match. */ if (ep->e2d_ino != 0) { namlen = ep->e2d_namlen; if (namlen == namelen && !bcmp(name, ep->e2d_name, (unsigned)namlen)) { /* * Save directory entry's inode number and * reclen in ndp->ni_ufs area, and release * directory buffer. */ *foundp = 1; return (0); } } *prevoffp = *offp; *offp += le16toh(ep->e2d_reclen); offset += le16toh(ep->e2d_reclen); *entryoffsetinblockp = offset; if (ep->e2d_ino != 0) *endusefulp = *offp; /* * Get pointer to the next entry. */ ep = (struct ext2fs_direct_2 *)((char *)data + offset); } return (0); } void ext2_dirbad(struct inode *ip, doff_t offset, char *how) { struct mount *mp; mp = ITOV(ip)->v_mount; if ((mp->mnt_flag & MNT_RDONLY) == 0) panic("ext2_dirbad: %s: bad dir ino %ju at offset %ld: %s\n", mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number, (long)offset, how); else SDT_PROBE4(ext2fs, , trace, ext2_dirbad_error, mp->mnt_stat.f_mntonname, ip->i_number, offset, how); } /* * Do consistency checking on a directory entry: * record length must be multiple of 4 * entry must fit in rest of its DIRBLKSIZ block * record must be large enough to contain entry * name is not longer than MAXNAMLEN * name must be as long as advertised, and null terminated */ /* * changed so that it confirms to ext2_check_dir_entry */ static int ext2_dirbadentry(struct vnode *dp, struct ext2fs_direct_2 *de, int entryoffsetinblock) { int DIRBLKSIZ = VTOI(dp)->i_e2fs->e2fs_bsize; char *error_msg = NULL; if (le16toh(de->e2d_reclen) < EXT2_DIR_REC_LEN(1)) error_msg = "rec_len is smaller than minimal"; else if (le16toh(de->e2d_reclen) % 4 != 0) error_msg = "rec_len % 4 != 0"; else if (le16toh(de->e2d_reclen) < EXT2_DIR_REC_LEN(de->e2d_namlen)) error_msg = "reclen is too small for name_len"; else if (entryoffsetinblock + le16toh(de->e2d_reclen)> DIRBLKSIZ) error_msg = "directory entry across blocks"; /* else LATER if (de->inode > dir->i_sb->u.ext2_sb.s_es->s_inodes_count) error_msg = "inode out of bounds"; */ if (error_msg != NULL) { SDT_PROBE5(ext2fs, , trace, ext2_dirbadentry_error, error_msg, entryoffsetinblock, le32toh(de->e2d_ino), le16toh(de->e2d_reclen), de->e2d_namlen); } return (error_msg == NULL ? 0 : 1); } /* * Insert an entry into the fresh directory block. * Initialize entry tail if the metadata_csum feature is turned on. */ static int ext2_add_first_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry, struct componentname *cnp) { struct inode *dp; struct iovec aiov; struct uio auio; char* buf = NULL; int dirblksize, error; dp = VTOI(dvp); dirblksize = dp->i_e2fs->e2fs_bsize; if (dp->i_offset & (dirblksize - 1)) panic("ext2_add_first_entry: bad directory offset"); if (EXT2_HAS_RO_COMPAT_FEATURE(dp->i_e2fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) { entry->e2d_reclen = htole16(dirblksize - sizeof(struct ext2fs_direct_tail)); buf = malloc(dirblksize, M_TEMP, M_WAITOK); memcpy(buf, entry, EXT2_DIR_REC_LEN(entry->e2d_namlen)); ext2_init_dirent_tail(EXT2_DIRENT_TAIL(buf, dirblksize)); ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)buf); auio.uio_offset = dp->i_offset; auio.uio_resid = dirblksize; aiov.iov_len = auio.uio_resid; aiov.iov_base = (caddr_t)buf; } else { entry->e2d_reclen = htole16(dirblksize); auio.uio_offset = dp->i_offset; auio.uio_resid = EXT2_DIR_REC_LEN(entry->e2d_namlen); aiov.iov_len = auio.uio_resid; aiov.iov_base = (caddr_t)entry; } auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_rw = UIO_WRITE; auio.uio_segflg = UIO_SYSSPACE; auio.uio_td = (struct thread *)0; error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred); if (error) goto out; dp->i_size = roundup2(dp->i_size, dirblksize); dp->i_flag |= IN_CHANGE; out: free(buf, M_TEMP); return (error); } /* * Write a directory entry after a call to namei, using the parameters * that it left in nameidata. The argument ip is the inode which the new * directory entry will refer to. Dvp is a pointer to the directory to * be written, which was left locked by namei. Remaining parameters * (dp->i_offset, dp->i_count) indicate how the space for the new * entry is to be obtained. */ int ext2_direnter(struct inode *ip, struct vnode *dvp, struct componentname *cnp) { struct inode *dp; struct ext2fs_direct_2 newdir; int DIRBLKSIZ = ip->i_e2fs->e2fs_bsize; int error; #ifdef INVARIANTS if ((cnp->cn_flags & SAVENAME) == 0) panic("ext2_direnter: missing name"); #endif dp = VTOI(dvp); newdir.e2d_ino = htole32(ip->i_number); if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs, EXT2F_INCOMPAT_FTYPE)) { newdir.e2d_namlen = cnp->cn_namelen; newdir.e2d_type = DTTOFT(IFTODT(ip->i_mode)); } else newdir.e2d_namlen = htole16(cnp->cn_namelen); bcopy(cnp->cn_nameptr, newdir.e2d_name, (unsigned)cnp->cn_namelen + 1); if (ext2_htree_has_idx(dp)) { error = ext2_htree_add_entry(dvp, &newdir, cnp); if (error) { dp->i_flag &= ~IN_E3INDEX; dp->i_flag |= IN_CHANGE | IN_UPDATE; } return (error); } if (EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_DIRHASHINDEX) && !ext2_htree_has_idx(dp)) { if ((dp->i_size / DIRBLKSIZ) == 1 && dp->i_offset == DIRBLKSIZ) { /* * Making indexed directory when one block is not * enough to save all entries. */ return ext2_htree_create_index(dvp, cnp, &newdir); } } /* * If dp->i_count is 0, then namei could find no * space in the directory. Here, dp->i_offset will * be on a directory block boundary and we will write the * new entry into a fresh block. */ if (dp->i_count == 0) return ext2_add_first_entry(dvp, &newdir, cnp); error = ext2_add_entry(dvp, &newdir); if (!error && dp->i_endoff && dp->i_endoff < dp->i_size) error = ext2_truncate(dvp, (off_t)dp->i_endoff, IO_SYNC, - cnp->cn_cred, cnp->cn_thread); + cnp->cn_cred, curthread); return (error); } /* * Insert an entry into the directory block. * Compact the contents. */ int ext2_add_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry) { struct ext2fs_direct_2 *ep, *nep; struct inode *dp; struct buf *bp; u_int dsize; int error, loc, newentrysize, spacefree; char *dirbuf; dp = VTOI(dvp); /* * If dp->i_count is non-zero, then namei found space * for the new entry in the range dp->i_offset to * dp->i_offset + dp->i_count in the directory. * To use this space, we may have to compact the entries located * there, by copying them together towards the beginning of the * block, leaving the free space in one usable chunk at the end. */ /* * Increase size of directory if entry eats into new space. * This should never push the size past a new multiple of * DIRBLKSIZE. * * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN. */ if (dp->i_offset + dp->i_count > dp->i_size) dp->i_size = dp->i_offset + dp->i_count; /* * Get the block containing the space for the new directory entry. */ if ((error = ext2_blkatoff(dvp, (off_t)dp->i_offset, &dirbuf, &bp)) != 0) return (error); /* * Find space for the new entry. In the simple case, the entry at * offset base will have the space. If it does not, then namei * arranged that compacting the region dp->i_offset to * dp->i_offset + dp->i_count would yield the * space. */ newentrysize = EXT2_DIR_REC_LEN(entry->e2d_namlen); ep = (struct ext2fs_direct_2 *)dirbuf; dsize = EXT2_DIR_REC_LEN(ep->e2d_namlen); spacefree = le16toh(ep->e2d_reclen) - dsize; for (loc = le16toh(ep->e2d_reclen); loc < dp->i_count; ) { nep = (struct ext2fs_direct_2 *)(dirbuf + loc); if (le32toh(ep->e2d_ino)) { /* trim the existing slot */ ep->e2d_reclen = htole16(dsize); ep = (struct ext2fs_direct_2 *)((char *)ep + dsize); } else { /* overwrite; nothing there; header is ours */ spacefree += dsize; } dsize = EXT2_DIR_REC_LEN(nep->e2d_namlen); spacefree += le16toh(nep->e2d_reclen) - dsize; loc += le16toh(nep->e2d_reclen); bcopy((caddr_t)nep, (caddr_t)ep, dsize); } /* * Update the pointer fields in the previous entry (if any), * copy in the new entry, and write out the block. */ if (ep->e2d_ino == 0) { if (spacefree + dsize < newentrysize) panic("ext2_direnter: compact1"); entry->e2d_reclen = htole16(spacefree + dsize); } else { if (spacefree < newentrysize) panic("ext2_direnter: compact2"); entry->e2d_reclen = htole16(spacefree); ep->e2d_reclen = htole16(dsize); ep = (struct ext2fs_direct_2 *)((char *)ep + dsize); } bcopy((caddr_t)entry, (caddr_t)ep, (u_int)newentrysize); ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data); if (DOINGASYNC(dvp)) { bdwrite(bp); error = 0; } else { error = bwrite(bp); } dp->i_flag |= IN_CHANGE | IN_UPDATE; return (error); } /* * Remove a directory entry after a call to namei, using * the parameters which it left in nameidata. The entry * dp->i_offset contains the offset into the directory of the * entry to be eliminated. The dp->i_count field contains the * size of the previous record in the directory. If this * is 0, the first entry is being deleted, so we need only * zero the inode number to mark the entry as free. If the * entry is not the first in the directory, we must reclaim * the space of the now empty record by adding the record size * to the size of the previous entry. */ int ext2_dirremove(struct vnode *dvp, struct componentname *cnp) { struct inode *dp; struct ext2fs_direct_2 *ep, *rep; struct buf *bp; int error; dp = VTOI(dvp); if (dp->i_count == 0) { /* * First entry in block: set d_ino to zero. */ if ((error = ext2_blkatoff(dvp, (off_t)dp->i_offset, (char **)&ep, &bp)) != 0) return (error); ep->e2d_ino = 0; ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data); error = bwrite(bp); dp->i_flag |= IN_CHANGE | IN_UPDATE; return (error); } /* * Collapse new free space into previous entry. */ if ((error = ext2_blkatoff(dvp, (off_t)(dp->i_offset - dp->i_count), (char **)&ep, &bp)) != 0) return (error); /* Set 'rep' to the entry being removed. */ if (dp->i_count == 0) rep = ep; else rep = (struct ext2fs_direct_2 *)((char *)ep + le16toh(ep->e2d_reclen)); ep->e2d_reclen += rep->e2d_reclen; ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data); if (DOINGASYNC(dvp) && dp->i_count != 0) bdwrite(bp); else error = bwrite(bp); dp->i_flag |= IN_CHANGE | IN_UPDATE; return (error); } /* * Rewrite an existing directory entry to point at the inode * supplied. The parameters describing the directory entry are * set up by a call to namei. */ int ext2_dirrewrite(struct inode *dp, struct inode *ip, struct componentname *cnp) { struct buf *bp; struct ext2fs_direct_2 *ep; struct vnode *vdp = ITOV(dp); int error; if ((error = ext2_blkatoff(vdp, (off_t)dp->i_offset, (char **)&ep, &bp)) != 0) return (error); ep->e2d_ino = htole32(ip->i_number); if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs, EXT2F_INCOMPAT_FTYPE)) ep->e2d_type = DTTOFT(IFTODT(ip->i_mode)); else ep->e2d_type = EXT2_FT_UNKNOWN; ext2_dirent_csum_set(dp, (struct ext2fs_direct_2 *)bp->b_data); error = bwrite(bp); dp->i_flag |= IN_CHANGE | IN_UPDATE; return (error); } /* * Check if a directory is empty or not. * Inode supplied must be locked. * * Using a struct dirtemplate here is not precisely * what we want, but better than using a struct direct. * * NB: does not handle corrupted directories. */ int ext2_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred) { off_t off; struct dirtemplate dbuf; struct ext2fs_direct_2 *dp = (struct ext2fs_direct_2 *)&dbuf; int error, namlen; ssize_t count; #define MINDIRSIZ (sizeof(struct dirtemplate) / 2) for (off = 0; off < ip->i_size; off += le16toh(dp->e2d_reclen)) { error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, off, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, &count, (struct thread *)0); /* * Since we read MINDIRSIZ, residual must * be 0 unless we're at end of file. */ if (error || count != 0) return (0); /* avoid infinite loops */ if (dp->e2d_reclen == 0) return (0); /* skip empty entries */ if (dp->e2d_ino == 0) continue; /* accept only "." and ".." */ namlen = dp->e2d_namlen; if (namlen > 2) return (0); if (dp->e2d_name[0] != '.') return (0); /* * At this point namlen must be 1 or 2. * 1 implies ".", 2 implies ".." if second * char is also "." */ if (namlen == 1) continue; if (dp->e2d_name[1] == '.' && le32toh(dp->e2d_ino) == parentino) continue; return (0); } return (1); } /* * Check if source directory is in the path of the target directory. * Target is supplied locked, source is unlocked. * The target is always vput before returning. */ int ext2_checkpath(struct inode *source, struct inode *target, struct ucred *cred) { struct vnode *vp; int error, namlen; struct dirtemplate dirbuf; vp = ITOV(target); if (target->i_number == source->i_number) { error = EEXIST; goto out; } if (target->i_number == EXT2_ROOTINO) { error = 0; goto out; } for (;;) { if (vp->v_type != VDIR) { error = ENOTDIR; break; } error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf, sizeof(struct dirtemplate), (off_t)0, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, NULL, NULL); if (error != 0) break; namlen = dirbuf.dotdot_type; /* like ufs little-endian */ if (namlen != 2 || dirbuf.dotdot_name[0] != '.' || dirbuf.dotdot_name[1] != '.') { error = ENOTDIR; break; } if (le32toh(dirbuf.dotdot_ino) == source->i_number) { error = EINVAL; break; } if (le32toh(dirbuf.dotdot_ino) == EXT2_ROOTINO) break; vput(vp); if ((error = VFS_VGET(vp->v_mount, le32toh(dirbuf.dotdot_ino), LK_EXCLUSIVE, &vp)) != 0) { vp = NULL; break; } } out: if (error == ENOTDIR) SDT_PROBE2(ext2fs, , lookup, trace, 1, "checkpath: .. not a directory"); if (vp != NULL) vput(vp); return (error); } diff --git a/sys/fs/ext2fs/ext2_vnops.c b/sys/fs/ext2fs/ext2_vnops.c index e9e8c07090da..f577241a457e 100644 --- a/sys/fs/ext2fs/ext2_vnops.c +++ b/sys/fs/ext2fs/ext2_vnops.c @@ -1,2350 +1,2350 @@ /*- * modified for EXT2FS support in Lites 1.1 * * Aug 1995, Godmar Back (gback@cs.utah.edu) * University of Utah, Department of Computer Science */ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)ufs_vnops.c 8.7 (Berkeley) 2/3/94 * @(#)ufs_vnops.c 8.27 (Berkeley) 5/27/95 * $FreeBSD$ */ #include "opt_suiddir.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "opt_directio.h" #include #include #include #include #include #include #include #include #include #include #include SDT_PROVIDER_DECLARE(ext2fs); /* * ext2fs trace probe: * arg0: verbosity. Higher numbers give more verbose messages * arg1: Textual message */ SDT_PROBE_DEFINE2(ext2fs, , vnops, trace, "int", "char*"); static int ext2_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *); static void ext2_itimes_locked(struct vnode *); static vop_access_t ext2_access; static int ext2_chmod(struct vnode *, int, struct ucred *, struct thread *); static int ext2_chown(struct vnode *, uid_t, gid_t, struct ucred *, struct thread *); static vop_close_t ext2_close; static vop_create_t ext2_create; static vop_fsync_t ext2_fsync; static vop_getattr_t ext2_getattr; static vop_ioctl_t ext2_ioctl; static vop_link_t ext2_link; static vop_mkdir_t ext2_mkdir; static vop_mknod_t ext2_mknod; static vop_open_t ext2_open; static vop_pathconf_t ext2_pathconf; static vop_print_t ext2_print; static vop_read_t ext2_read; static vop_readlink_t ext2_readlink; static vop_remove_t ext2_remove; static vop_rename_t ext2_rename; static vop_rmdir_t ext2_rmdir; static vop_setattr_t ext2_setattr; static vop_strategy_t ext2_strategy; static vop_symlink_t ext2_symlink; static vop_write_t ext2_write; static vop_deleteextattr_t ext2_deleteextattr; static vop_getextattr_t ext2_getextattr; static vop_listextattr_t ext2_listextattr; static vop_setextattr_t ext2_setextattr; static vop_vptofh_t ext2_vptofh; static vop_close_t ext2fifo_close; /* Global vfs data structures for ext2. */ struct vop_vector ext2_vnodeops = { .vop_default = &default_vnodeops, .vop_access = ext2_access, .vop_bmap = ext2_bmap, .vop_cachedlookup = ext2_lookup, .vop_close = ext2_close, .vop_create = ext2_create, .vop_fsync = ext2_fsync, .vop_getpages = vnode_pager_local_getpages, .vop_getpages_async = vnode_pager_local_getpages_async, .vop_getattr = ext2_getattr, .vop_inactive = ext2_inactive, .vop_ioctl = ext2_ioctl, .vop_link = ext2_link, .vop_lookup = vfs_cache_lookup, .vop_mkdir = ext2_mkdir, .vop_mknod = ext2_mknod, .vop_open = ext2_open, .vop_pathconf = ext2_pathconf, .vop_poll = vop_stdpoll, .vop_print = ext2_print, .vop_read = ext2_read, .vop_readdir = ext2_readdir, .vop_readlink = ext2_readlink, .vop_reallocblks = ext2_reallocblks, .vop_reclaim = ext2_reclaim, .vop_remove = ext2_remove, .vop_rename = ext2_rename, .vop_rmdir = ext2_rmdir, .vop_setattr = ext2_setattr, .vop_strategy = ext2_strategy, .vop_symlink = ext2_symlink, .vop_write = ext2_write, .vop_deleteextattr = ext2_deleteextattr, .vop_getextattr = ext2_getextattr, .vop_listextattr = ext2_listextattr, .vop_setextattr = ext2_setextattr, #ifdef UFS_ACL .vop_getacl = ext2_getacl, .vop_setacl = ext2_setacl, .vop_aclcheck = ext2_aclcheck, #endif /* UFS_ACL */ .vop_vptofh = ext2_vptofh, }; VFS_VOP_VECTOR_REGISTER(ext2_vnodeops); struct vop_vector ext2_fifoops = { .vop_default = &fifo_specops, .vop_access = ext2_access, .vop_close = ext2fifo_close, .vop_fsync = ext2_fsync, .vop_getattr = ext2_getattr, .vop_inactive = ext2_inactive, .vop_pathconf = ext2_pathconf, .vop_print = ext2_print, .vop_read = VOP_PANIC, .vop_reclaim = ext2_reclaim, .vop_setattr = ext2_setattr, .vop_write = VOP_PANIC, .vop_vptofh = ext2_vptofh, }; VFS_VOP_VECTOR_REGISTER(ext2_fifoops); /* * A virgin directory (no blushing please). * Note that the type and namlen fields are reversed relative to ext2. * Also, we don't use `struct odirtemplate', since it would just cause * endianness problems. */ static struct dirtemplate mastertemplate = { 0, htole16(12), 1, EXT2_FT_DIR, ".", 0, htole16(DIRBLKSIZ - 12), 2, EXT2_FT_DIR, ".." }; static struct dirtemplate omastertemplate = { 0, htole16(12), 1, EXT2_FT_UNKNOWN, ".", 0, htole16(DIRBLKSIZ - 12), 2, EXT2_FT_UNKNOWN, ".." }; static void ext2_itimes_locked(struct vnode *vp) { struct inode *ip; struct timespec ts; ASSERT_VI_LOCKED(vp, __func__); ip = VTOI(vp); if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0) return; if ((vp->v_type == VBLK || vp->v_type == VCHR)) ip->i_flag |= IN_LAZYMOD; else ip->i_flag |= IN_MODIFIED; if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { vfs_timestamp(&ts); if (ip->i_flag & IN_ACCESS) { ip->i_atime = ts.tv_sec; ip->i_atimensec = ts.tv_nsec; } if (ip->i_flag & IN_UPDATE) { ip->i_mtime = ts.tv_sec; ip->i_mtimensec = ts.tv_nsec; ip->i_modrev++; } if (ip->i_flag & IN_CHANGE) { ip->i_ctime = ts.tv_sec; ip->i_ctimensec = ts.tv_nsec; } } ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE); } void ext2_itimes(struct vnode *vp) { VI_LOCK(vp); ext2_itimes_locked(vp); VI_UNLOCK(vp); } /* * Create a regular file */ static int ext2_create(struct vop_create_args *ap) { int error; error = ext2_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode), ap->a_dvp, ap->a_vpp, ap->a_cnp); if (error != 0) return (error); if ((ap->a_cnp->cn_flags & MAKEENTRY) != 0) cache_enter(ap->a_dvp, *ap->a_vpp, ap->a_cnp); return (0); } static int ext2_open(struct vop_open_args *ap) { if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR) return (EOPNOTSUPP); /* * Files marked append-only must be opened for appending. */ if ((VTOI(ap->a_vp)->i_flags & APPEND) && (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE) return (EPERM); vnode_create_vobject(ap->a_vp, VTOI(ap->a_vp)->i_size, ap->a_td); return (0); } /* * Close called. * * Update the times on the inode. */ static int ext2_close(struct vop_close_args *ap) { struct vnode *vp = ap->a_vp; VI_LOCK(vp); if (vp->v_usecount > 1) ext2_itimes_locked(vp); VI_UNLOCK(vp); return (0); } static int ext2_access(struct vop_access_args *ap) { struct vnode *vp = ap->a_vp; struct inode *ip = VTOI(vp); accmode_t accmode = ap->a_accmode; int error; /* * Disallow write attempts on read-only file systems; * unless the file is a socket, fifo, or a block or * character device resident on the file system. */ if (accmode & VWRITE) { switch (vp->v_type) { case VDIR: case VLNK: case VREG: if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); break; default: break; } } /* If immutable bit set, nobody gets to write it. */ if ((accmode & VWRITE) && (ip->i_flags & (SF_IMMUTABLE | SF_SNAPSHOT))) return (EPERM); error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid, ap->a_accmode, ap->a_cred); return (error); } static int ext2_getattr(struct vop_getattr_args *ap) { struct vnode *vp = ap->a_vp; struct inode *ip = VTOI(vp); struct vattr *vap = ap->a_vap; ext2_itimes(vp); /* * Copy from inode table */ vap->va_fsid = dev2udev(ip->i_devvp->v_rdev); vap->va_fileid = ip->i_number; vap->va_mode = ip->i_mode & ~IFMT; vap->va_nlink = ip->i_nlink; vap->va_uid = ip->i_uid; vap->va_gid = ip->i_gid; vap->va_rdev = ip->i_rdev; vap->va_size = ip->i_size; vap->va_atime.tv_sec = ip->i_atime; vap->va_atime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_atimensec : 0; vap->va_mtime.tv_sec = ip->i_mtime; vap->va_mtime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_mtimensec : 0; vap->va_ctime.tv_sec = ip->i_ctime; vap->va_ctime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_ctimensec : 0; if (E2DI_HAS_XTIME(ip)) { vap->va_birthtime.tv_sec = ip->i_birthtime; vap->va_birthtime.tv_nsec = ip->i_birthnsec; } vap->va_flags = ip->i_flags; vap->va_gen = ip->i_gen; vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; vap->va_bytes = dbtob((u_quad_t)ip->i_blocks); vap->va_type = IFTOVT(ip->i_mode); vap->va_filerev = ip->i_modrev; return (0); } /* * Set attribute vnode op. called from several syscalls */ static int ext2_setattr(struct vop_setattr_args *ap) { struct vattr *vap = ap->a_vap; struct vnode *vp = ap->a_vp; struct inode *ip = VTOI(vp); struct ucred *cred = ap->a_cred; struct thread *td = curthread; int error; /* * Check for unsettable attributes. */ if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) || (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) || (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) || ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) { return (EINVAL); } if (vap->va_flags != VNOVAL) { /* Disallow flags not supported by ext2fs. */ if (vap->va_flags & ~(SF_APPEND | SF_IMMUTABLE | UF_NODUMP)) return (EOPNOTSUPP); if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); /* * Callers may only modify the file flags on objects they * have VADMIN rights for. */ if ((error = VOP_ACCESS(vp, VADMIN, cred, td))) return (error); /* * Unprivileged processes and privileged processes in * jail() are not permitted to unset system flags, or * modify flags if any system flags are set. * Privileged non-jail processes may not modify system flags * if securelevel > 0 and any existing system flags are set. */ if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS)) { if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND)) { error = securelevel_gt(cred, 0); if (error) return (error); } } else { if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND) || ((vap->va_flags ^ ip->i_flags) & SF_SETTABLE)) return (EPERM); } ip->i_flags = vap->va_flags; ip->i_flag |= IN_CHANGE; if (ip->i_flags & (IMMUTABLE | APPEND)) return (0); } if (ip->i_flags & (IMMUTABLE | APPEND)) return (EPERM); /* * Go through the fields and update iff not VNOVAL. */ if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) { if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); if ((error = ext2_chown(vp, vap->va_uid, vap->va_gid, cred, td)) != 0) return (error); } if (vap->va_size != VNOVAL) { /* * Disallow write attempts on read-only file systems; * unless the file is a socket, fifo, or a block or * character device resident on the file system. */ switch (vp->v_type) { case VDIR: return (EISDIR); case VLNK: case VREG: if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); break; default: break; } if ((error = ext2_truncate(vp, vap->va_size, 0, cred, td)) != 0) return (error); } if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); /* * From utimes(2): * If times is NULL, ... The caller must be the owner of * the file, have permission to write the file, or be the * super-user. * If times is non-NULL, ... The caller must be the owner of * the file or be the super-user. */ if ((error = VOP_ACCESS(vp, VADMIN, cred, td)) && ((vap->va_vaflags & VA_UTIMES_NULL) == 0 || (error = VOP_ACCESS(vp, VWRITE, cred, td)))) return (error); ip->i_flag |= IN_CHANGE | IN_MODIFIED; if (vap->va_atime.tv_sec != VNOVAL) { ip->i_flag &= ~IN_ACCESS; ip->i_atime = vap->va_atime.tv_sec; ip->i_atimensec = vap->va_atime.tv_nsec; } if (vap->va_mtime.tv_sec != VNOVAL) { ip->i_flag &= ~IN_UPDATE; ip->i_mtime = vap->va_mtime.tv_sec; ip->i_mtimensec = vap->va_mtime.tv_nsec; } if (E2DI_HAS_XTIME(ip) && vap->va_birthtime.tv_sec != VNOVAL) { ip->i_birthtime = vap->va_birthtime.tv_sec; ip->i_birthnsec = vap->va_birthtime.tv_nsec; } error = ext2_update(vp, 0); if (error) return (error); } error = 0; if (vap->va_mode != (mode_t)VNOVAL) { if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); error = ext2_chmod(vp, (int)vap->va_mode, cred, td); } return (error); } /* * Change the mode on a file. * Inode must be locked before calling. */ static int ext2_chmod(struct vnode *vp, int mode, struct ucred *cred, struct thread *td) { struct inode *ip = VTOI(vp); int error; /* * To modify the permissions on a file, must possess VADMIN * for that file. */ if ((error = VOP_ACCESS(vp, VADMIN, cred, td))) return (error); /* * Privileged processes may set the sticky bit on non-directories, * as well as set the setgid bit on a file with a group that the * process is not a member of. */ if (vp->v_type != VDIR && (mode & S_ISTXT)) { error = priv_check_cred(cred, PRIV_VFS_STICKYFILE); if (error) return (EFTYPE); } if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) { error = priv_check_cred(cred, PRIV_VFS_SETGID); if (error) return (error); } ip->i_mode &= ~ALLPERMS; ip->i_mode |= (mode & ALLPERMS); ip->i_flag |= IN_CHANGE; return (0); } /* * Perform chown operation on inode ip; * inode must be locked prior to call. */ static int ext2_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred, struct thread *td) { struct inode *ip = VTOI(vp); uid_t ouid; gid_t ogid; int error = 0; if (uid == (uid_t)VNOVAL) uid = ip->i_uid; if (gid == (gid_t)VNOVAL) gid = ip->i_gid; /* * To modify the ownership of a file, must possess VADMIN * for that file. */ if ((error = VOP_ACCESS(vp, VADMIN, cred, td))) return (error); /* * To change the owner of a file, or change the group of a file * to a group of which we are not a member, the caller must * have privilege. */ if (uid != ip->i_uid || (gid != ip->i_gid && !groupmember(gid, cred))) { error = priv_check_cred(cred, PRIV_VFS_CHOWN); if (error) return (error); } ogid = ip->i_gid; ouid = ip->i_uid; ip->i_gid = gid; ip->i_uid = uid; ip->i_flag |= IN_CHANGE; if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) { if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID) != 0) ip->i_mode &= ~(ISUID | ISGID); } return (0); } /* * Synch an open file. */ /* ARGSUSED */ static int ext2_fsync(struct vop_fsync_args *ap) { /* * Flush all dirty buffers associated with a vnode. */ vop_stdfsync(ap); return (ext2_update(ap->a_vp, ap->a_waitfor == MNT_WAIT)); } static int ext2_check_mknod_limits(dev_t dev) { unsigned maj = major(dev); unsigned min = minor(dev); if (maj > EXT2_MAJOR_MAX || min > EXT2_MINOR_MAX) return (EINVAL); return (0); } /* * Mknod vnode call */ /* ARGSUSED */ static int ext2_mknod(struct vop_mknod_args *ap) { struct vattr *vap = ap->a_vap; struct vnode **vpp = ap->a_vpp; struct inode *ip; ino_t ino; int error; if (vap->va_rdev != VNOVAL) { error = ext2_check_mknod_limits(vap->va_rdev); if (error) return (error); } error = ext2_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), ap->a_dvp, vpp, ap->a_cnp); if (error) return (error); ip = VTOI(*vpp); ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; if (vap->va_rdev != VNOVAL) ip->i_rdev = vap->va_rdev; /* * Remove inode, then reload it through VFS_VGET so it is * checked to see if it is an alias of an existing entry in * the inode cache. XXX I don't believe this is necessary now. */ (*vpp)->v_type = VNON; ino = ip->i_number; /* Save this before vgone() invalidates ip. */ vgone(*vpp); vput(*vpp); error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp); if (error) { *vpp = NULL; return (error); } return (0); } static int ext2_remove(struct vop_remove_args *ap) { struct inode *ip; struct vnode *vp = ap->a_vp; struct vnode *dvp = ap->a_dvp; int error; ip = VTOI(vp); if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || (VTOI(dvp)->i_flags & APPEND)) { error = EPERM; goto out; } error = ext2_dirremove(dvp, ap->a_cnp); if (error == 0) { ip->i_nlink--; ip->i_flag |= IN_CHANGE; } out: return (error); } /* * link vnode call */ static int ext2_link(struct vop_link_args *ap) { struct vnode *vp = ap->a_vp; struct vnode *tdvp = ap->a_tdvp; struct componentname *cnp = ap->a_cnp; struct inode *ip; int error; #ifdef INVARIANTS if ((cnp->cn_flags & HASBUF) == 0) panic("ext2_link: no name"); #endif ip = VTOI(vp); if ((nlink_t)ip->i_nlink >= EXT4_LINK_MAX) { error = EMLINK; goto out; } if (ip->i_flags & (IMMUTABLE | APPEND)) { error = EPERM; goto out; } ip->i_nlink++; ip->i_flag |= IN_CHANGE; error = ext2_update(vp, !DOINGASYNC(vp)); if (!error) error = ext2_direnter(ip, tdvp, cnp); if (error) { ip->i_nlink--; ip->i_flag |= IN_CHANGE; } out: return (error); } static int ext2_inc_nlink(struct inode *ip) { ip->i_nlink++; if (S_ISDIR(ip->i_mode) && EXT2_HAS_RO_COMPAT_FEATURE(ip->i_e2fs, EXT2F_ROCOMPAT_DIR_NLINK) && ip->i_nlink > 1) { if (ip->i_nlink >= EXT4_LINK_MAX || ip->i_nlink == 2) ip->i_nlink = 1; } else if (ip->i_nlink > EXT4_LINK_MAX) { ip->i_nlink--; return (EMLINK); } return (0); } static void ext2_dec_nlink(struct inode *ip) { if (!S_ISDIR(ip->i_mode) || ip->i_nlink > 2) ip->i_nlink--; } /* * Rename system call. * rename("foo", "bar"); * is essentially * unlink("bar"); * link("foo", "bar"); * unlink("foo"); * but ``atomically''. Can't do full commit without saving state in the * inode on disk which isn't feasible at this time. Best we can do is * always guarantee the target exists. * * Basic algorithm is: * * 1) Bump link count on source while we're linking it to the * target. This also ensure the inode won't be deleted out * from underneath us while we work (it may be truncated by * a concurrent `trunc' or `open' for creation). * 2) Link source to destination. If destination already exists, * delete it first. * 3) Unlink source reference to inode if still around. If a * directory was moved and the parent of the destination * is different from the source, patch the ".." entry in the * directory. */ static int ext2_rename(struct vop_rename_args *ap) { struct vnode *tvp = ap->a_tvp; struct vnode *tdvp = ap->a_tdvp; struct vnode *fvp = ap->a_fvp; struct vnode *fdvp = ap->a_fdvp; struct componentname *tcnp = ap->a_tcnp; struct componentname *fcnp = ap->a_fcnp; struct inode *ip, *xp, *dp; struct dirtemplate *dirbuf; int doingdirectory = 0, oldparent = 0, newparent = 0; int error = 0; u_char namlen; #ifdef INVARIANTS if ((tcnp->cn_flags & HASBUF) == 0 || (fcnp->cn_flags & HASBUF) == 0) panic("ext2_rename: no name"); #endif /* * Check for cross-device rename. */ if ((fvp->v_mount != tdvp->v_mount) || (tvp && (fvp->v_mount != tvp->v_mount))) { error = EXDEV; abortit: if (tdvp == tvp) vrele(tdvp); else vput(tdvp); if (tvp) vput(tvp); vrele(fdvp); vrele(fvp); return (error); } if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || (VTOI(tdvp)->i_flags & APPEND))) { error = EPERM; goto abortit; } /* * Renaming a file to itself has no effect. The upper layers should * not call us in that case. Temporarily just warn if they do. */ if (fvp == tvp) { SDT_PROBE2(ext2fs, , vnops, trace, 1, "rename: fvp == tvp (can't happen)"); error = 0; goto abortit; } if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0) goto abortit; dp = VTOI(fdvp); ip = VTOI(fvp); if (ip->i_nlink >= EXT4_LINK_MAX && !EXT2_HAS_RO_COMPAT_FEATURE(ip->i_e2fs, EXT2F_ROCOMPAT_DIR_NLINK)) { VOP_UNLOCK(fvp); error = EMLINK; goto abortit; } if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || (dp->i_flags & APPEND)) { VOP_UNLOCK(fvp); error = EPERM; goto abortit; } if ((ip->i_mode & IFMT) == IFDIR) { /* * Avoid ".", "..", and aliases of "." for obvious reasons. */ if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') || dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT || (ip->i_flag & IN_RENAME)) { VOP_UNLOCK(fvp); error = EINVAL; goto abortit; } ip->i_flag |= IN_RENAME; oldparent = dp->i_number; doingdirectory++; } vrele(fdvp); /* * When the target exists, both the directory * and target vnodes are returned locked. */ dp = VTOI(tdvp); xp = NULL; if (tvp) xp = VTOI(tvp); /* * 1) Bump link count while we're moving stuff * around. If we crash somewhere before * completing our work, the link count * may be wrong, but correctable. */ ext2_inc_nlink(ip); ip->i_flag |= IN_CHANGE; if ((error = ext2_update(fvp, !DOINGASYNC(fvp))) != 0) { VOP_UNLOCK(fvp); goto bad; } /* * If ".." must be changed (ie the directory gets a new * parent) then the source directory must not be in the * directory hierarchy above the target, as this would * orphan everything below the source directory. Also * the user must have write permission in the source so * as to be able to change "..". We must repeat the call * to namei, as the parent directory is unlocked by the * call to checkpath(). */ - error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread); + error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, curthread); VOP_UNLOCK(fvp); if (oldparent != dp->i_number) newparent = dp->i_number; if (doingdirectory && newparent) { if (error) /* write access check above */ goto bad; if (xp != NULL) vput(tvp); error = ext2_checkpath(ip, dp, tcnp->cn_cred); if (error) goto out; VREF(tdvp); error = relookup(tdvp, &tvp, tcnp); if (error) goto out; vrele(tdvp); dp = VTOI(tdvp); xp = NULL; if (tvp) xp = VTOI(tvp); } /* * 2) If target doesn't exist, link the target * to the source and unlink the source. * Otherwise, rewrite the target directory * entry to reference the source inode and * expunge the original entry's existence. */ if (xp == NULL) { if (dp->i_devvp != ip->i_devvp) panic("ext2_rename: EXDEV"); /* * Account for ".." in new directory. * When source and destination have the same * parent we don't fool with the link count. */ if (doingdirectory && newparent) { error = ext2_inc_nlink(dp); if (error) goto bad; dp->i_flag |= IN_CHANGE; error = ext2_update(tdvp, !DOINGASYNC(tdvp)); if (error) goto bad; } error = ext2_direnter(ip, tdvp, tcnp); if (error) { if (doingdirectory && newparent) { ext2_dec_nlink(dp); dp->i_flag |= IN_CHANGE; (void)ext2_update(tdvp, 1); } goto bad; } vput(tdvp); } else { if (xp->i_devvp != dp->i_devvp || xp->i_devvp != ip->i_devvp) panic("ext2_rename: EXDEV"); /* * Short circuit rename(foo, foo). */ if (xp->i_number == ip->i_number) panic("ext2_rename: same file"); /* * If the parent directory is "sticky", then the user must * own the parent directory, or the destination of the rename, * otherwise the destination may not be changed (except by * root). This implements append-only directories. */ if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 && tcnp->cn_cred->cr_uid != dp->i_uid && xp->i_uid != tcnp->cn_cred->cr_uid) { error = EPERM; goto bad; } /* * Target must be empty if a directory and have no links * to it. Also, ensure source and target are compatible * (both directories, or both not directories). */ if ((xp->i_mode & IFMT) == IFDIR) { if (!ext2_dirempty(xp, dp->i_number, tcnp->cn_cred)) { error = ENOTEMPTY; goto bad; } if (!doingdirectory) { error = ENOTDIR; goto bad; } cache_purge(tdvp); } else if (doingdirectory) { error = EISDIR; goto bad; } error = ext2_dirrewrite(dp, ip, tcnp); if (error) goto bad; /* * If the target directory is in the same * directory as the source directory, * decrement the link count on the parent * of the target directory. */ if (doingdirectory && !newparent) { ext2_dec_nlink(dp); dp->i_flag |= IN_CHANGE; } vput(tdvp); /* * Adjust the link count of the target to * reflect the dirrewrite above. If this is * a directory it is empty and there are * no links to it, so we can squash the inode and * any space associated with it. We disallowed * renaming over top of a directory with links to * it above, as the remaining link would point to * a directory without "." or ".." entries. */ ext2_dec_nlink(xp); if (doingdirectory) { if (xp->i_nlink > 2) panic("ext2_rename: linked directory"); error = ext2_truncate(tvp, (off_t)0, IO_SYNC, - tcnp->cn_cred, tcnp->cn_thread); + tcnp->cn_cred, curthread); xp->i_nlink = 0; } xp->i_flag |= IN_CHANGE; vput(tvp); xp = NULL; } /* * 3) Unlink the source. */ fcnp->cn_flags &= ~MODMASK; fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; VREF(fdvp); error = relookup(fdvp, &fvp, fcnp); if (error == 0) vrele(fdvp); if (fvp != NULL) { xp = VTOI(fvp); dp = VTOI(fdvp); } else { /* * From name has disappeared. IN_RENAME is not sufficient * to protect against directory races due to timing windows, * so we can't panic here. */ vrele(ap->a_fvp); return (0); } /* * Ensure that the directory entry still exists and has not * changed while the new name has been entered. If the source is * a file then the entry may have been unlinked or renamed. In * either case there is no further work to be done. If the source * is a directory then it cannot have been rmdir'ed; its link * count of three would cause a rmdir to fail with ENOTEMPTY. * The IN_RENAME flag ensures that it cannot be moved by another * rename. */ if (xp != ip) { /* * From name resolves to a different inode. IN_RENAME is * not sufficient protection against timing window races * so we can't panic here. */ } else { /* * If the source is a directory with a * new parent, the link count of the old * parent directory must be decremented * and ".." set to point to the new parent. */ if (doingdirectory && newparent) { ext2_dec_nlink(dp); dp->i_flag |= IN_CHANGE; dirbuf = malloc(dp->i_e2fs->e2fs_bsize, M_TEMP, M_WAITOK | M_ZERO); error = vn_rdwr(UIO_READ, fvp, (caddr_t)dirbuf, ip->i_e2fs->e2fs_bsize, (off_t)0, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, tcnp->cn_cred, NOCRED, NULL, NULL); if (error == 0) { /* Like ufs little-endian: */ namlen = dirbuf->dotdot_type; if (namlen != 2 || dirbuf->dotdot_name[0] != '.' || dirbuf->dotdot_name[1] != '.') { ext2_dirbad(xp, (doff_t)12, "rename: mangled dir"); } else { dirbuf->dotdot_ino = htole32(newparent); /* * dirblock 0 could be htree root, * try both csum update functions. */ ext2_dirent_csum_set(ip, (struct ext2fs_direct_2 *)dirbuf); ext2_dx_csum_set(ip, (struct ext2fs_direct_2 *)dirbuf); (void)vn_rdwr(UIO_WRITE, fvp, (caddr_t)dirbuf, ip->i_e2fs->e2fs_bsize, (off_t)0, UIO_SYSSPACE, IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, tcnp->cn_cred, NOCRED, NULL, NULL); cache_purge(fdvp); } } free(dirbuf, M_TEMP); } error = ext2_dirremove(fdvp, fcnp); if (!error) { ext2_dec_nlink(xp); xp->i_flag |= IN_CHANGE; } xp->i_flag &= ~IN_RENAME; } if (dp) vput(fdvp); if (xp) vput(fvp); vrele(ap->a_fvp); return (error); bad: if (xp) vput(ITOV(xp)); vput(ITOV(dp)); out: if (doingdirectory) ip->i_flag &= ~IN_RENAME; if (vn_lock(fvp, LK_EXCLUSIVE) == 0) { ext2_dec_nlink(ip); ip->i_flag |= IN_CHANGE; ip->i_flag &= ~IN_RENAME; vput(fvp); } else vrele(fvp); return (error); } #ifdef UFS_ACL static int ext2_do_posix1e_acl_inheritance_dir(struct vnode *dvp, struct vnode *tvp, mode_t dmode, struct ucred *cred, struct thread *td) { int error; struct inode *ip = VTOI(tvp); struct acl *dacl, *acl; acl = acl_alloc(M_WAITOK); dacl = acl_alloc(M_WAITOK); /* * Retrieve default ACL from parent, if any. */ error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td); switch (error) { case 0: /* * Retrieved a default ACL, so merge mode and ACL if * necessary. If the ACL is empty, fall through to * the "not defined or available" case. */ if (acl->acl_cnt != 0) { dmode = acl_posix1e_newfilemode(dmode, acl); ip->i_mode = dmode; *dacl = *acl; ext2_sync_acl_from_inode(ip, acl); break; } /* FALLTHROUGH */ case EOPNOTSUPP: /* * Just use the mode as-is. */ ip->i_mode = dmode; error = 0; goto out; default: goto out; } error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td); if (error == 0) error = VOP_SETACL(tvp, ACL_TYPE_DEFAULT, dacl, cred, td); switch (error) { case 0: break; case EOPNOTSUPP: /* * XXX: This should not happen, as EOPNOTSUPP above * was supposed to free acl. */ #ifdef DEBUG printf("ext2_mkdir: VOP_GETACL() but no VOP_SETACL()\n"); #endif /* DEBUG */ break; default: goto out; } out: acl_free(acl); acl_free(dacl); return (error); } static int ext2_do_posix1e_acl_inheritance_file(struct vnode *dvp, struct vnode *tvp, mode_t mode, struct ucred *cred, struct thread *td) { int error; struct inode *ip = VTOI(tvp); struct acl *acl; acl = acl_alloc(M_WAITOK); /* * Retrieve default ACL for parent, if any. */ error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td); switch (error) { case 0: /* * Retrieved a default ACL, so merge mode and ACL if * necessary. */ if (acl->acl_cnt != 0) { /* * Two possible ways for default ACL to not * be present. First, the EA can be * undefined, or second, the default ACL can * be blank. If it's blank, fall through to * the it's not defined case. */ mode = acl_posix1e_newfilemode(mode, acl); ip->i_mode = mode; ext2_sync_acl_from_inode(ip, acl); break; } /* FALLTHROUGH */ case EOPNOTSUPP: /* * Just use the mode as-is. */ ip->i_mode = mode; error = 0; goto out; default: goto out; } error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td); switch (error) { case 0: break; case EOPNOTSUPP: /* * XXX: This should not happen, as EOPNOTSUPP above was * supposed to free acl. */ printf("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() " "but no VOP_SETACL()\n"); /* panic("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() " "but no VOP_SETACL()"); */ break; default: goto out; } out: acl_free(acl); return (error); } #endif /* UFS_ACL */ /* * Mkdir system call */ static int ext2_mkdir(struct vop_mkdir_args *ap) { struct m_ext2fs *fs; struct vnode *dvp = ap->a_dvp; struct vattr *vap = ap->a_vap; struct componentname *cnp = ap->a_cnp; struct inode *ip, *dp; struct vnode *tvp; struct dirtemplate dirtemplate, *dtp; char *buf = NULL; int error, dmode; #ifdef INVARIANTS if ((cnp->cn_flags & HASBUF) == 0) panic("ext2_mkdir: no name"); #endif dp = VTOI(dvp); if ((nlink_t)dp->i_nlink >= EXT4_LINK_MAX && !EXT2_HAS_RO_COMPAT_FEATURE(dp->i_e2fs, EXT2F_ROCOMPAT_DIR_NLINK)) { error = EMLINK; goto out; } dmode = vap->va_mode & 0777; dmode |= IFDIR; /* * Must simulate part of ext2_makeinode here to acquire the inode, * but not have it entered in the parent directory. The entry is * made later after writing "." and ".." entries. */ error = ext2_valloc(dvp, dmode, cnp->cn_cred, &tvp); if (error) goto out; ip = VTOI(tvp); fs = ip->i_e2fs; ip->i_gid = dp->i_gid; #ifdef SUIDDIR { /* * if we are hacking owners here, (only do this where told to) * and we are not giving it TOO root, (would subvert quotas) * then go ahead and give it to the other user. * The new directory also inherits the SUID bit. * If user's UID and dir UID are the same, * 'give it away' so that the SUID is still forced on. */ if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) && (dp->i_mode & ISUID) && dp->i_uid) { dmode |= ISUID; ip->i_uid = dp->i_uid; } else { ip->i_uid = cnp->cn_cred->cr_uid; } } #else ip->i_uid = cnp->cn_cred->cr_uid; #endif ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; ip->i_mode = dmode; tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */ ip->i_nlink = 2; if (cnp->cn_flags & ISWHITEOUT) ip->i_flags |= UF_OPAQUE; error = ext2_update(tvp, 1); /* * Bump link count in parent directory * to reflect work done below. Should * be done before reference is created * so reparation is possible if we crash. */ ext2_inc_nlink(dp); dp->i_flag |= IN_CHANGE; error = ext2_update(dvp, !DOINGASYNC(dvp)); if (error) goto bad; /* Initialize directory with "." and ".." from static template. */ if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs, EXT2F_INCOMPAT_FTYPE)) dtp = &mastertemplate; else dtp = &omastertemplate; dirtemplate = *dtp; dirtemplate.dot_ino = htole32(ip->i_number); dirtemplate.dotdot_ino = htole32(dp->i_number); /* * note that in ext2 DIRBLKSIZ == blocksize, not DEV_BSIZE so let's * just redefine it - for this function only */ #undef DIRBLKSIZ #define DIRBLKSIZ VTOI(dvp)->i_e2fs->e2fs_bsize dirtemplate.dotdot_reclen = htole16(DIRBLKSIZ - 12); buf = malloc(DIRBLKSIZ, M_TEMP, M_WAITOK | M_ZERO); if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) { dirtemplate.dotdot_reclen = htole16(le16toh(dirtemplate.dotdot_reclen) - sizeof(struct ext2fs_direct_tail)); ext2_init_dirent_tail(EXT2_DIRENT_TAIL(buf, DIRBLKSIZ)); } memcpy(buf, &dirtemplate, sizeof(dirtemplate)); ext2_dirent_csum_set(ip, (struct ext2fs_direct_2 *)buf); error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)buf, DIRBLKSIZ, (off_t)0, UIO_SYSSPACE, IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, cnp->cn_cred, NOCRED, NULL, NULL); if (error) { ext2_dec_nlink(dp); dp->i_flag |= IN_CHANGE; goto bad; } if (DIRBLKSIZ > VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize) /* XXX should grow with balloc() */ panic("ext2_mkdir: blksize"); else { ip->i_size = DIRBLKSIZ; ip->i_flag |= IN_CHANGE; } #ifdef UFS_ACL if (dvp->v_mount->mnt_flag & MNT_ACLS) { error = ext2_do_posix1e_acl_inheritance_dir(dvp, tvp, dmode, - cnp->cn_cred, cnp->cn_thread); + cnp->cn_cred, curthread); if (error) goto bad; } #endif /* UFS_ACL */ /* Directory set up, now install its entry in the parent directory. */ error = ext2_direnter(ip, dvp, cnp); if (error) { ext2_dec_nlink(dp); dp->i_flag |= IN_CHANGE; } bad: /* * No need to do an explicit VOP_TRUNCATE here, vrele will do this * for us because we set the link count to 0. */ if (error) { ip->i_nlink = 0; ip->i_flag |= IN_CHANGE; vput(tvp); } else *ap->a_vpp = tvp; out: free(buf, M_TEMP); return (error); #undef DIRBLKSIZ #define DIRBLKSIZ DEV_BSIZE } /* * Rmdir system call. */ static int ext2_rmdir(struct vop_rmdir_args *ap) { struct vnode *vp = ap->a_vp; struct vnode *dvp = ap->a_dvp; struct componentname *cnp = ap->a_cnp; struct inode *ip, *dp; int error; ip = VTOI(vp); dp = VTOI(dvp); /* * Verify the directory is empty (and valid). * (Rmdir ".." won't be valid since * ".." will contain a reference to * the current directory and thus be * non-empty.) */ if (!ext2_dirempty(ip, dp->i_number, cnp->cn_cred)) { error = ENOTEMPTY; goto out; } if ((dp->i_flags & APPEND) || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) { error = EPERM; goto out; } /* * Delete reference to directory before purging * inode. If we crash in between, the directory * will be reattached to lost+found, */ error = ext2_dirremove(dvp, cnp); if (error) goto out; ext2_dec_nlink(dp); dp->i_flag |= IN_CHANGE; cache_purge(dvp); VOP_UNLOCK(dvp); /* * Truncate inode. The only stuff left * in the directory is "." and "..". */ ip->i_nlink = 0; error = ext2_truncate(vp, (off_t)0, IO_SYNC, cnp->cn_cred, - cnp->cn_thread); + curthread); cache_purge(ITOV(ip)); if (vn_lock(dvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { VOP_UNLOCK(vp); vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); } out: return (error); } /* * symlink -- make a symbolic link */ static int ext2_symlink(struct vop_symlink_args *ap) { struct vnode *vp, **vpp = ap->a_vpp; struct inode *ip; int len, error; error = ext2_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp, vpp, ap->a_cnp); if (error) return (error); vp = *vpp; len = strlen(ap->a_target); if (len < VFSTOEXT2(vp->v_mount)->um_e2fs->e2fs_maxsymlinklen) { ip = VTOI(vp); bcopy(ap->a_target, (char *)ip->i_shortlink, len); ip->i_size = len; ip->i_flag |= IN_CHANGE | IN_UPDATE; } else error = vn_rdwr(UIO_WRITE, vp, __DECONST(void *, ap->a_target), len, (off_t)0, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, ap->a_cnp->cn_cred, NOCRED, NULL, NULL); if (error) vput(vp); return (error); } /* * Return target name of a symbolic link */ static int ext2_readlink(struct vop_readlink_args *ap) { struct vnode *vp = ap->a_vp; struct inode *ip = VTOI(vp); int isize; isize = ip->i_size; if (isize < VFSTOEXT2(vp->v_mount)->um_e2fs->e2fs_maxsymlinklen) { uiomove((char *)ip->i_shortlink, isize, ap->a_uio); return (0); } return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred)); } /* * Calculate the logical to physical mapping if not done already, * then call the device strategy routine. * * In order to be able to swap to a file, the ext2_bmaparray() operation may not * deadlock on memory. See ext2_bmap() for details. */ static int ext2_strategy(struct vop_strategy_args *ap) { struct buf *bp = ap->a_bp; struct vnode *vp = ap->a_vp; struct bufobj *bo; daddr_t blkno; int error; if (vp->v_type == VBLK || vp->v_type == VCHR) panic("ext2_strategy: spec"); if (bp->b_blkno == bp->b_lblkno) { if (VTOI(ap->a_vp)->i_flag & IN_E4EXTENTS) error = ext4_bmapext(vp, bp->b_lblkno, &blkno, NULL, NULL); else error = ext2_bmaparray(vp, bp->b_lblkno, &blkno, NULL, NULL); bp->b_blkno = blkno; if (error) { bp->b_error = error; bp->b_ioflags |= BIO_ERROR; bufdone(bp); return (0); } if ((long)bp->b_blkno == -1) vfs_bio_clrbuf(bp); } if ((long)bp->b_blkno == -1) { bufdone(bp); return (0); } bp->b_iooffset = dbtob(bp->b_blkno); bo = VFSTOEXT2(vp->v_mount)->um_bo; BO_STRATEGY(bo, bp); return (0); } /* * Print out the contents of an inode. */ static int ext2_print(struct vop_print_args *ap) { struct vnode *vp = ap->a_vp; struct inode *ip = VTOI(vp); vn_printf(ip->i_devvp, "\tino %ju", (uintmax_t)ip->i_number); if (vp->v_type == VFIFO) fifo_printinfo(vp); printf("\n"); return (0); } /* * Close wrapper for fifos. * * Update the times on the inode then do device close. */ static int ext2fifo_close(struct vop_close_args *ap) { struct vnode *vp = ap->a_vp; VI_LOCK(vp); if (vp->v_usecount > 1) ext2_itimes_locked(vp); VI_UNLOCK(vp); return (fifo_specops.vop_close(ap)); } /* * Return POSIX pathconf information applicable to ext2 filesystems. */ static int ext2_pathconf(struct vop_pathconf_args *ap) { int error = 0; switch (ap->a_name) { case _PC_LINK_MAX: if (EXT2_HAS_RO_COMPAT_FEATURE(VTOI(ap->a_vp)->i_e2fs, EXT2F_ROCOMPAT_DIR_NLINK)) *ap->a_retval = INT_MAX; else *ap->a_retval = EXT4_LINK_MAX; break; case _PC_NAME_MAX: *ap->a_retval = NAME_MAX; break; case _PC_PIPE_BUF: if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO) *ap->a_retval = PIPE_BUF; else error = EINVAL; break; case _PC_CHOWN_RESTRICTED: *ap->a_retval = 1; break; case _PC_NO_TRUNC: *ap->a_retval = 1; break; #ifdef UFS_ACL case _PC_ACL_EXTENDED: if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS) *ap->a_retval = 1; else *ap->a_retval = 0; break; case _PC_ACL_PATH_MAX: if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS) *ap->a_retval = ACL_MAX_ENTRIES; else *ap->a_retval = 3; break; #endif /* UFS_ACL */ case _PC_MIN_HOLE_SIZE: *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize; break; case _PC_PRIO_IO: *ap->a_retval = 0; break; case _PC_SYNC_IO: *ap->a_retval = 0; break; case _PC_ALLOC_SIZE_MIN: *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize; break; case _PC_FILESIZEBITS: *ap->a_retval = 64; break; case _PC_REC_INCR_XFER_SIZE: *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize; break; case _PC_REC_MAX_XFER_SIZE: *ap->a_retval = -1; /* means ``unlimited'' */ break; case _PC_REC_MIN_XFER_SIZE: *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize; break; case _PC_REC_XFER_ALIGN: *ap->a_retval = PAGE_SIZE; break; case _PC_SYMLINK_MAX: *ap->a_retval = MAXPATHLEN; break; default: error = vop_stdpathconf(ap); break; } return (error); } /* * Vnode operation to remove a named attribute. */ static int ext2_deleteextattr(struct vop_deleteextattr_args *ap) { struct inode *ip; struct m_ext2fs *fs; int error; ip = VTOI(ap->a_vp); fs = ip->i_e2fs; if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR)) return (EOPNOTSUPP); if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) return (EOPNOTSUPP); error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, ap->a_cred, ap->a_td, VWRITE); if (error) return (error); error = ENOATTR; if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) { error = ext2_extattr_inode_delete(ip, ap->a_attrnamespace, ap->a_name); if (error != ENOATTR) return (error); } if (ip->i_facl) error = ext2_extattr_block_delete(ip, ap->a_attrnamespace, ap->a_name); return (error); } /* * Vnode operation to retrieve a named extended attribute. */ static int ext2_getextattr(struct vop_getextattr_args *ap) { struct inode *ip; struct m_ext2fs *fs; int error; ip = VTOI(ap->a_vp); fs = ip->i_e2fs; if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR)) return (EOPNOTSUPP); if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) return (EOPNOTSUPP); error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, ap->a_cred, ap->a_td, VREAD); if (error) return (error); if (ap->a_size != NULL) *ap->a_size = 0; error = ENOATTR; if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) { error = ext2_extattr_inode_get(ip, ap->a_attrnamespace, ap->a_name, ap->a_uio, ap->a_size); if (error != ENOATTR) return (error); } if (ip->i_facl) error = ext2_extattr_block_get(ip, ap->a_attrnamespace, ap->a_name, ap->a_uio, ap->a_size); return (error); } /* * Vnode operation to retrieve extended attributes on a vnode. */ static int ext2_listextattr(struct vop_listextattr_args *ap) { struct inode *ip; struct m_ext2fs *fs; int error; ip = VTOI(ap->a_vp); fs = ip->i_e2fs; if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR)) return (EOPNOTSUPP); if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) return (EOPNOTSUPP); error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, ap->a_cred, ap->a_td, VREAD); if (error) return (error); if (ap->a_size != NULL) *ap->a_size = 0; if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) { error = ext2_extattr_inode_list(ip, ap->a_attrnamespace, ap->a_uio, ap->a_size); if (error) return (error); } if (ip->i_facl) error = ext2_extattr_block_list(ip, ap->a_attrnamespace, ap->a_uio, ap->a_size); return (error); } /* * Vnode operation to set a named attribute. */ static int ext2_setextattr(struct vop_setextattr_args *ap) { struct inode *ip; struct m_ext2fs *fs; int error; ip = VTOI(ap->a_vp); fs = ip->i_e2fs; if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR)) return (EOPNOTSUPP); if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) return (EOPNOTSUPP); error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, ap->a_cred, ap->a_td, VWRITE); if (error) return (error); error = ext2_extattr_valid_attrname(ap->a_attrnamespace, ap->a_name); if (error) return (error); if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) { error = ext2_extattr_inode_set(ip, ap->a_attrnamespace, ap->a_name, ap->a_uio); if (error != ENOSPC) return (error); } error = ext2_extattr_block_set(ip, ap->a_attrnamespace, ap->a_name, ap->a_uio); return (error); } /* * Vnode pointer to File handle */ /* ARGSUSED */ static int ext2_vptofh(struct vop_vptofh_args *ap) { struct inode *ip; struct ufid *ufhp; ip = VTOI(ap->a_vp); ufhp = (struct ufid *)ap->a_fhp; ufhp->ufid_len = sizeof(struct ufid); ufhp->ufid_ino = ip->i_number; ufhp->ufid_gen = ip->i_gen; return (0); } /* * Initialize the vnode associated with a new inode, handle aliased * vnodes. */ int ext2_vinit(struct mount *mntp, struct vop_vector *fifoops, struct vnode **vpp) { struct inode *ip; struct vnode *vp; vp = *vpp; ip = VTOI(vp); vp->v_type = IFTOVT(ip->i_mode); /* * Only unallocated inodes should be of type VNON. */ if (ip->i_mode != 0 && vp->v_type == VNON) return (EINVAL); if (vp->v_type == VFIFO) vp->v_op = fifoops; if (ip->i_number == EXT2_ROOTINO) vp->v_vflag |= VV_ROOT; ip->i_modrev = init_va_filerev(); *vpp = vp; return (0); } /* * Allocate a new inode. */ static int ext2_makeinode(int mode, struct vnode *dvp, struct vnode **vpp, struct componentname *cnp) { struct inode *ip, *pdir; struct vnode *tvp; int error; pdir = VTOI(dvp); #ifdef INVARIANTS if ((cnp->cn_flags & HASBUF) == 0) panic("ext2_makeinode: no name"); #endif *vpp = NULL; if ((mode & IFMT) == 0) mode |= IFREG; error = ext2_valloc(dvp, mode, cnp->cn_cred, &tvp); if (error) { return (error); } ip = VTOI(tvp); ip->i_gid = pdir->i_gid; #ifdef SUIDDIR { /* * if we are * not the owner of the directory, * and we are hacking owners here, (only do this where told to) * and we are not giving it TOO root, (would subvert quotas) * then go ahead and give it to the other user. * Note that this drops off the execute bits for security. */ if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) && (pdir->i_mode & ISUID) && (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) { ip->i_uid = pdir->i_uid; mode &= ~07111; } else { ip->i_uid = cnp->cn_cred->cr_uid; } } #else ip->i_uid = cnp->cn_cred->cr_uid; #endif ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; ip->i_mode = mode; tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */ ip->i_nlink = 1; if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred)) { if (priv_check_cred(cnp->cn_cred, PRIV_VFS_RETAINSUGID)) ip->i_mode &= ~ISGID; } if (cnp->cn_flags & ISWHITEOUT) ip->i_flags |= UF_OPAQUE; /* * Make sure inode goes to disk before directory entry. */ error = ext2_update(tvp, !DOINGASYNC(tvp)); if (error) goto bad; #ifdef UFS_ACL if (dvp->v_mount->mnt_flag & MNT_ACLS) { error = ext2_do_posix1e_acl_inheritance_file(dvp, tvp, mode, - cnp->cn_cred, cnp->cn_thread); + cnp->cn_cred, curthread); if (error) goto bad; } #endif /* UFS_ACL */ error = ext2_direnter(ip, dvp, cnp); if (error) goto bad; *vpp = tvp; return (0); bad: /* * Write error occurred trying to update the inode * or the directory so must deallocate the inode. */ ip->i_nlink = 0; ip->i_flag |= IN_CHANGE; vput(tvp); return (error); } /* * Vnode op for reading. */ static int ext2_read(struct vop_read_args *ap) { struct vnode *vp; struct inode *ip; struct uio *uio; struct m_ext2fs *fs; struct buf *bp; daddr_t lbn, nextlbn; off_t bytesinfile; long size, xfersize, blkoffset; int error, orig_resid, seqcount; int ioflag; vp = ap->a_vp; uio = ap->a_uio; ioflag = ap->a_ioflag; seqcount = ap->a_ioflag >> IO_SEQSHIFT; ip = VTOI(vp); #ifdef INVARIANTS if (uio->uio_rw != UIO_READ) panic("%s: mode", "ext2_read"); if (vp->v_type == VLNK) { if ((int)ip->i_size < VFSTOEXT2(vp->v_mount)->um_e2fs->e2fs_maxsymlinklen) panic("%s: short symlink", "ext2_read"); } else if (vp->v_type != VREG && vp->v_type != VDIR) panic("%s: type %d", "ext2_read", vp->v_type); #endif orig_resid = uio->uio_resid; KASSERT(orig_resid >= 0, ("ext2_read: uio->uio_resid < 0")); if (orig_resid == 0) return (0); KASSERT(uio->uio_offset >= 0, ("ext2_read: uio->uio_offset < 0")); fs = ip->i_e2fs; if (uio->uio_offset < ip->i_size && uio->uio_offset >= fs->e2fs_maxfilesize) return (EOVERFLOW); for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) { if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0) break; lbn = lblkno(fs, uio->uio_offset); nextlbn = lbn + 1; size = blksize(fs, ip, lbn); blkoffset = blkoff(fs, uio->uio_offset); xfersize = fs->e2fs_fsize - blkoffset; if (uio->uio_resid < xfersize) xfersize = uio->uio_resid; if (bytesinfile < xfersize) xfersize = bytesinfile; if (lblktosize(fs, nextlbn) >= ip->i_size) error = bread(vp, lbn, size, NOCRED, &bp); else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) { error = cluster_read(vp, ip->i_size, lbn, size, NOCRED, blkoffset + uio->uio_resid, seqcount, 0, &bp); } else if (seqcount > 1) { u_int nextsize = blksize(fs, ip, nextlbn); error = breadn(vp, lbn, size, &nextlbn, &nextsize, 1, NOCRED, &bp); } else error = bread(vp, lbn, size, NOCRED, &bp); if (error) { brelse(bp); bp = NULL; break; } /* * We should only get non-zero b_resid when an I/O error * has occurred, which should cause us to break above. * However, if the short read did not cause an error, * then we want to ensure that we do not uiomove bad * or uninitialized data. */ size -= bp->b_resid; if (size < xfersize) { if (size == 0) break; xfersize = size; } error = uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio); if (error) break; vfs_bio_brelse(bp, ioflag); } /* * This can only happen in the case of an error because the loop * above resets bp to NULL on each iteration and on normal * completion has not set a new value into it. so it must have come * from a 'break' statement */ if (bp != NULL) vfs_bio_brelse(bp, ioflag); if ((error == 0 || uio->uio_resid != orig_resid) && (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) ip->i_flag |= IN_ACCESS; return (error); } static int ext2_ioctl(struct vop_ioctl_args *ap) { struct vnode *vp; int error; vp = ap->a_vp; switch (ap->a_command) { case FIOSEEKDATA: if (!(VTOI(vp)->i_flag & IN_E4EXTENTS)) { error = vn_lock(vp, LK_SHARED); if (error == 0) { error = ext2_bmap_seekdata(vp, (off_t *)ap->a_data); VOP_UNLOCK(vp); } else error = EBADF; return (error); } case FIOSEEKHOLE: return (vn_bmap_seekhole(vp, ap->a_command, (off_t *)ap->a_data, ap->a_cred)); default: return (ENOTTY); } } /* * Vnode op for writing. */ static int ext2_write(struct vop_write_args *ap) { struct vnode *vp; struct uio *uio; struct inode *ip; struct m_ext2fs *fs; struct buf *bp; daddr_t lbn; off_t osize; int blkoffset, error, flags, ioflag, resid, size, seqcount, xfersize; ioflag = ap->a_ioflag; uio = ap->a_uio; vp = ap->a_vp; seqcount = ioflag >> IO_SEQSHIFT; ip = VTOI(vp); #ifdef INVARIANTS if (uio->uio_rw != UIO_WRITE) panic("%s: mode", "ext2_write"); #endif switch (vp->v_type) { case VREG: if (ioflag & IO_APPEND) uio->uio_offset = ip->i_size; if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size) return (EPERM); /* FALLTHROUGH */ case VLNK: break; case VDIR: /* XXX differs from ffs -- this is called from ext2_mkdir(). */ if ((ioflag & IO_SYNC) == 0) panic("ext2_write: nonsync dir write"); break; default: panic("ext2_write: type %p %d (%jd,%jd)", (void *)vp, vp->v_type, (intmax_t)uio->uio_offset, (intmax_t)uio->uio_resid); } KASSERT(uio->uio_resid >= 0, ("ext2_write: uio->uio_resid < 0")); KASSERT(uio->uio_offset >= 0, ("ext2_write: uio->uio_offset < 0")); fs = ip->i_e2fs; if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->e2fs_maxfilesize) return (EFBIG); /* * Maybe this should be above the vnode op call, but so long as * file servers have no limits, I don't think it matters. */ if (vn_rlimit_fsize(vp, uio, uio->uio_td)) return (EFBIG); resid = uio->uio_resid; osize = ip->i_size; if (seqcount > BA_SEQMAX) flags = BA_SEQMAX << BA_SEQSHIFT; else flags = seqcount << BA_SEQSHIFT; if ((ioflag & IO_SYNC) && !DOINGASYNC(vp)) flags |= IO_SYNC; for (error = 0; uio->uio_resid > 0;) { lbn = lblkno(fs, uio->uio_offset); blkoffset = blkoff(fs, uio->uio_offset); xfersize = fs->e2fs_fsize - blkoffset; if (uio->uio_resid < xfersize) xfersize = uio->uio_resid; if (uio->uio_offset + xfersize > ip->i_size) vnode_pager_setsize(vp, uio->uio_offset + xfersize); /* * We must perform a read-before-write if the transfer size * does not cover the entire buffer. */ if (fs->e2fs_bsize > xfersize) flags |= BA_CLRBUF; else flags &= ~BA_CLRBUF; error = ext2_balloc(ip, lbn, blkoffset + xfersize, ap->a_cred, &bp, flags); if (error != 0) break; if ((ioflag & (IO_SYNC | IO_INVAL)) == (IO_SYNC | IO_INVAL)) bp->b_flags |= B_NOCACHE; if (uio->uio_offset + xfersize > ip->i_size) ip->i_size = uio->uio_offset + xfersize; size = blksize(fs, ip, lbn) - bp->b_resid; if (size < xfersize) xfersize = size; error = uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio); /* * If the buffer is not already filled and we encounter an * error while trying to fill it, we have to clear out any * garbage data from the pages instantiated for the buffer. * If we do not, a failed uiomove() during a write can leave * the prior contents of the pages exposed to a userland mmap. * * Note that we need only clear buffers with a transfer size * equal to the block size because buffers with a shorter * transfer size were cleared above by the call to ext2_balloc() * with the BA_CLRBUF flag set. * * If the source region for uiomove identically mmaps the * buffer, uiomove() performed the NOP copy, and the buffer * content remains valid because the page fault handler * validated the pages. */ if (error != 0 && (bp->b_flags & B_CACHE) == 0 && fs->e2fs_bsize == xfersize) vfs_bio_clrbuf(bp); vfs_bio_set_flags(bp, ioflag); /* * If IO_SYNC each buffer is written synchronously. Otherwise * if we have a severe page deficiency write the buffer * asynchronously. Otherwise try to cluster, and if that * doesn't do it then either do an async write (if O_DIRECT), * or a delayed write (if not). */ if (ioflag & IO_SYNC) { (void)bwrite(bp); } else if (vm_page_count_severe() || buf_dirty_count_severe() || (ioflag & IO_ASYNC)) { bp->b_flags |= B_CLUSTEROK; bawrite(bp); } else if (xfersize + blkoffset == fs->e2fs_fsize) { if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) { bp->b_flags |= B_CLUSTEROK; cluster_write(vp, &ip->i_clusterw, bp, ip->i_size, seqcount, 0); } else { bawrite(bp); } } else if (ioflag & IO_DIRECT) { bp->b_flags |= B_CLUSTEROK; bawrite(bp); } else { bp->b_flags |= B_CLUSTEROK; bdwrite(bp); } if (error || xfersize == 0) break; } /* * If we successfully wrote any data, and we are not the superuser * we clear the setuid and setgid bits as a precaution against * tampering. */ if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ap->a_cred) { if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID)) ip->i_mode &= ~(ISUID | ISGID); } if (error) { if (ioflag & IO_UNIT) { (void)ext2_truncate(vp, osize, ioflag & IO_SYNC, ap->a_cred, uio->uio_td); uio->uio_offset -= resid - uio->uio_resid; uio->uio_resid = resid; } } if (uio->uio_resid != resid) { ip->i_flag |= IN_CHANGE | IN_UPDATE; if (ioflag & IO_SYNC) error = ext2_update(vp, 1); } return (error); } diff --git a/sys/fs/fdescfs/fdesc_vnops.c b/sys/fs/fdescfs/fdesc_vnops.c index c5a7b86f1de5..209e3b3df4f3 100644 --- a/sys/fs/fdescfs/fdesc_vnops.c +++ b/sys/fs/fdescfs/fdesc_vnops.c @@ -1,670 +1,670 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1992, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software donated to Berkeley by * Jan-Simon Pendry. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)fdesc_vnops.c 8.9 (Berkeley) 1/21/94 * * $FreeBSD$ */ /* * /dev/fd Filesystem */ #include #include #include #include #include #include #include /* boottime */ #include #include #include #include /* Must come after sys/malloc.h */ #include #include #include #include #include #include #include #include #define NFDCACHE 4 #define FD_NHASH(ix) \ (&fdhashtbl[(ix) & fdhash]) static LIST_HEAD(fdhashhead, fdescnode) *fdhashtbl; static u_long fdhash; struct mtx fdesc_hashmtx; static vop_getattr_t fdesc_getattr; static vop_lookup_t fdesc_lookup; static vop_open_t fdesc_open; static vop_pathconf_t fdesc_pathconf; static vop_readdir_t fdesc_readdir; static vop_readlink_t fdesc_readlink; static vop_reclaim_t fdesc_reclaim; static vop_setattr_t fdesc_setattr; static struct vop_vector fdesc_vnodeops = { .vop_default = &default_vnodeops, .vop_access = VOP_NULL, .vop_getattr = fdesc_getattr, .vop_lookup = fdesc_lookup, .vop_open = fdesc_open, .vop_pathconf = fdesc_pathconf, .vop_readdir = fdesc_readdir, .vop_readlink = fdesc_readlink, .vop_reclaim = fdesc_reclaim, .vop_setattr = fdesc_setattr, }; VFS_VOP_VECTOR_REGISTER(fdesc_vnodeops); static void fdesc_insmntque_dtr(struct vnode *, void *); static void fdesc_remove_entry(struct fdescnode *); /* * Initialise cache headers */ int fdesc_init(struct vfsconf *vfsp) { mtx_init(&fdesc_hashmtx, "fdescfs_hash", NULL, MTX_DEF); fdhashtbl = hashinit(NFDCACHE, M_CACHE, &fdhash); return (0); } /* * Uninit ready for unload. */ int fdesc_uninit(struct vfsconf *vfsp) { hashdestroy(fdhashtbl, M_CACHE, fdhash); mtx_destroy(&fdesc_hashmtx); return (0); } /* * If allocating vnode fails, call this. */ static void fdesc_insmntque_dtr(struct vnode *vp, void *arg) { vgone(vp); vput(vp); } /* * Remove an entry from the hash if it exists. */ static void fdesc_remove_entry(struct fdescnode *fd) { struct fdhashhead *fc; struct fdescnode *fd2; fc = FD_NHASH(fd->fd_ix); mtx_lock(&fdesc_hashmtx); LIST_FOREACH(fd2, fc, fd_hash) { if (fd == fd2) { LIST_REMOVE(fd, fd_hash); break; } } mtx_unlock(&fdesc_hashmtx); } int fdesc_allocvp(fdntype ftype, unsigned fd_fd, int ix, struct mount *mp, struct vnode **vpp) { struct fdescmount *fmp; struct fdhashhead *fc; struct fdescnode *fd, *fd2; struct vnode *vp, *vp2; struct thread *td; int error; td = curthread; fc = FD_NHASH(ix); loop: mtx_lock(&fdesc_hashmtx); /* * If a forced unmount is progressing, we need to drop it. The flags are * protected by the hashmtx. */ fmp = mp->mnt_data; if (fmp == NULL || fmp->flags & FMNT_UNMOUNTF) { mtx_unlock(&fdesc_hashmtx); return (-1); } LIST_FOREACH(fd, fc, fd_hash) { if (fd->fd_ix == ix && fd->fd_vnode->v_mount == mp) { /* Get reference to vnode in case it's being free'd */ vp = fd->fd_vnode; VI_LOCK(vp); mtx_unlock(&fdesc_hashmtx); if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK)) goto loop; *vpp = vp; return (0); } } mtx_unlock(&fdesc_hashmtx); fd = malloc(sizeof(struct fdescnode), M_TEMP, M_WAITOK); error = getnewvnode("fdescfs", mp, &fdesc_vnodeops, &vp); if (error) { free(fd, M_TEMP); return (error); } vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); vp->v_data = fd; fd->fd_vnode = vp; fd->fd_type = ftype; fd->fd_fd = fd_fd; fd->fd_ix = ix; if (ftype == Fdesc && fmp->flags & FMNT_LINRDLNKF) vp->v_vflag |= VV_READLINK; error = insmntque1(vp, mp, fdesc_insmntque_dtr, NULL); if (error != 0) { *vpp = NULLVP; return (error); } /* Make sure that someone didn't beat us when inserting the vnode. */ mtx_lock(&fdesc_hashmtx); /* * If a forced unmount is progressing, we need to drop it. The flags are * protected by the hashmtx. */ fmp = mp->mnt_data; if (fmp == NULL || fmp->flags & FMNT_UNMOUNTF) { mtx_unlock(&fdesc_hashmtx); vgone(vp); vput(vp); *vpp = NULLVP; return (-1); } LIST_FOREACH(fd2, fc, fd_hash) { if (fd2->fd_ix == ix && fd2->fd_vnode->v_mount == mp) { /* Get reference to vnode in case it's being free'd */ vp2 = fd2->fd_vnode; VI_LOCK(vp2); mtx_unlock(&fdesc_hashmtx); error = vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK); /* Someone beat us, dec use count and wait for reclaim */ vgone(vp); vput(vp); /* If we didn't get it, return no vnode. */ if (error) vp2 = NULLVP; *vpp = vp2; return (error); } } /* If we came here, we can insert it safely. */ LIST_INSERT_HEAD(fc, fd, fd_hash); mtx_unlock(&fdesc_hashmtx); *vpp = vp; return (0); } struct fdesc_get_ino_args { fdntype ftype; unsigned fd_fd; int ix; struct file *fp; struct thread *td; }; static int fdesc_get_ino_alloc(struct mount *mp, void *arg, int lkflags, struct vnode **rvp) { struct fdesc_get_ino_args *a; struct fdescmount *fdm; struct vnode *vp; int error; a = arg; fdm = VFSTOFDESC(mp); if ((fdm->flags & FMNT_NODUP) != 0 && a->fp->f_type == DTYPE_VNODE) { vp = a->fp->f_vnode; vget(vp, lkflags | LK_RETRY); *rvp = vp; error = 0; } else { error = fdesc_allocvp(a->ftype, a->fd_fd, a->ix, mp, rvp); } fdrop(a->fp, a->td); return (error); } /* * vp is the current namei directory * ndp is the name to locate in that directory... */ static int fdesc_lookup(struct vop_lookup_args *ap) { struct vnode **vpp = ap->a_vpp; struct vnode *dvp = ap->a_dvp; struct componentname *cnp = ap->a_cnp; char *pname = cnp->cn_nameptr; - struct thread *td = cnp->cn_thread; + struct thread *td = curthread; struct file *fp; struct fdesc_get_ino_args arg; int nlen = cnp->cn_namelen; u_int fd, fd1; int error; struct vnode *fvp; if ((cnp->cn_flags & ISLASTCN) && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { error = EROFS; goto bad; } if (cnp->cn_namelen == 1 && *pname == '.') { *vpp = dvp; VREF(dvp); return (0); } if (VTOFDESC(dvp)->fd_type != Froot) { error = ENOTDIR; goto bad; } fd = 0; /* the only time a leading 0 is acceptable is if it's "0" */ if (*pname == '0' && nlen != 1) { error = ENOENT; goto bad; } while (nlen--) { if (*pname < '0' || *pname > '9') { error = ENOENT; goto bad; } fd1 = 10 * fd + *pname++ - '0'; if (fd1 < fd) { error = ENOENT; goto bad; } fd = fd1; } /* * No rights to check since 'fp' isn't actually used. */ if ((error = fget(td, fd, &cap_no_rights, &fp)) != 0) goto bad; /* Check if we're looking up ourselves. */ if (VTOFDESC(dvp)->fd_ix == FD_DESC + fd) { /* * In case we're holding the last reference to the file, the dvp * will be re-acquired. */ vhold(dvp); VOP_UNLOCK(dvp); fdrop(fp, td); /* Re-aquire the lock afterwards. */ vn_lock(dvp, LK_RETRY | LK_EXCLUSIVE); vdrop(dvp); fvp = dvp; if (VN_IS_DOOMED(dvp)) error = ENOENT; } else { /* * Unlock our root node (dvp) when doing this, since we might * deadlock since the vnode might be locked by another thread * and the root vnode lock will be obtained afterwards (in case * we're looking up the fd of the root vnode), which will be the * opposite lock order. Vhold the root vnode first so we don't * lose it. */ arg.ftype = Fdesc; arg.fd_fd = fd; arg.ix = FD_DESC + fd; arg.fp = fp; arg.td = td; error = vn_vget_ino_gen(dvp, fdesc_get_ino_alloc, &arg, LK_EXCLUSIVE, &fvp); } if (error) goto bad; *vpp = fvp; return (0); bad: *vpp = NULL; return (error); } static int fdesc_open(struct vop_open_args *ap) { struct vnode *vp = ap->a_vp; if (VTOFDESC(vp)->fd_type == Froot) return (0); /* * XXX Kludge: set td->td_proc->p_dupfd to contain the value of the file * descriptor being sought for duplication. The error return ensures * that the vnode for this device will be released by vn_open. Open * will detect this special error and take the actions in dupfdopen. * Other callers of vn_open or VOP_OPEN will simply report the * error. */ ap->a_td->td_dupfd = VTOFDESC(vp)->fd_fd; /* XXX */ return (ENODEV); } static int fdesc_pathconf(struct vop_pathconf_args *ap) { struct vnode *vp = ap->a_vp; int error; switch (ap->a_name) { case _PC_NAME_MAX: *ap->a_retval = NAME_MAX; return (0); case _PC_LINK_MAX: if (VTOFDESC(vp)->fd_type == Froot) *ap->a_retval = 2; else *ap->a_retval = 1; return (0); default: if (VTOFDESC(vp)->fd_type == Froot) return (vop_stdpathconf(ap)); vref(vp); VOP_UNLOCK(vp); error = kern_fpathconf(curthread, VTOFDESC(vp)->fd_fd, ap->a_name, ap->a_retval); vn_lock(vp, LK_SHARED | LK_RETRY); vunref(vp); return (error); } } static int fdesc_getattr(struct vop_getattr_args *ap) { struct vnode *vp = ap->a_vp; struct vattr *vap = ap->a_vap; struct timeval boottime; getboottime(&boottime); vap->va_mode = S_IRUSR|S_IXUSR|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH; vap->va_fileid = VTOFDESC(vp)->fd_ix; vap->va_uid = 0; vap->va_gid = 0; vap->va_blocksize = DEV_BSIZE; vap->va_atime.tv_sec = boottime.tv_sec; vap->va_atime.tv_nsec = 0; vap->va_mtime = vap->va_atime; vap->va_ctime = vap->va_mtime; vap->va_gen = 0; vap->va_flags = 0; vap->va_bytes = 0; vap->va_filerev = 0; switch (VTOFDESC(vp)->fd_type) { case Froot: vap->va_type = VDIR; vap->va_nlink = 2; vap->va_size = DEV_BSIZE; vap->va_rdev = NODEV; break; case Fdesc: vap->va_type = (vp->v_vflag & VV_READLINK) == 0 ? VCHR : VLNK; vap->va_nlink = 1; vap->va_size = 0; vap->va_rdev = makedev(0, vap->va_fileid); break; default: panic("fdesc_getattr"); break; } vp->v_type = vap->va_type; return (0); } static int fdesc_setattr(struct vop_setattr_args *ap) { struct vattr *vap = ap->a_vap; struct vnode *vp; struct mount *mp; struct file *fp; struct thread *td = curthread; cap_rights_t rights; unsigned fd; int error; /* * Can't mess with the root vnode */ if (VTOFDESC(ap->a_vp)->fd_type == Froot) return (EACCES); fd = VTOFDESC(ap->a_vp)->fd_fd; /* * Allow setattr where there is an underlying vnode. */ error = getvnode(td, fd, cap_rights_init_one(&rights, CAP_EXTATTR_SET), &fp); if (error) { /* * getvnode() returns EINVAL if the file descriptor is not * backed by a vnode. Silently drop all changes except * chflags(2) in this case. */ if (error == EINVAL) { if (vap->va_flags != VNOVAL) error = EOPNOTSUPP; else error = 0; } return (error); } vp = fp->f_vnode; if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) == 0) { vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); error = VOP_SETATTR(vp, ap->a_vap, ap->a_cred); VOP_UNLOCK(vp); vn_finished_write(mp); } fdrop(fp, td); return (error); } #define UIO_MX _GENERIC_DIRLEN(10) /* number of symbols in INT_MAX printout */ static int fdesc_readdir(struct vop_readdir_args *ap) { struct fdescmount *fmp; struct uio *uio = ap->a_uio; struct filedesc *fdp; struct dirent d; struct dirent *dp = &d; int error, i, off, fcnt; if (VTOFDESC(ap->a_vp)->fd_type != Froot) panic("fdesc_readdir: not dir"); fmp = VFSTOFDESC(ap->a_vp->v_mount); if (ap->a_ncookies != NULL) *ap->a_ncookies = 0; off = (int)uio->uio_offset; if (off != uio->uio_offset || off < 0 || (u_int)off % UIO_MX != 0 || uio->uio_resid < UIO_MX) return (EINVAL); i = (u_int)off / UIO_MX; fdp = uio->uio_td->td_proc->p_fd; error = 0; fcnt = i - 2; /* The first two nodes are `.' and `..' */ FILEDESC_SLOCK(fdp); while (i < fdp->fd_nfiles + 2 && uio->uio_resid >= UIO_MX) { bzero((caddr_t)dp, UIO_MX); switch (i) { case 0: /* `.' */ case 1: /* `..' */ dp->d_fileno = i + FD_ROOT; dp->d_namlen = i + 1; dp->d_reclen = UIO_MX; bcopy("..", dp->d_name, dp->d_namlen); dp->d_type = DT_DIR; dirent_terminate(dp); break; default: if (fdp->fd_ofiles[fcnt].fde_file == NULL) break; dp->d_namlen = sprintf(dp->d_name, "%d", fcnt); dp->d_reclen = UIO_MX; dp->d_type = (fmp->flags & FMNT_LINRDLNKF) == 0 ? DT_CHR : DT_LNK; dp->d_fileno = i + FD_DESC; dirent_terminate(dp); break; } /* NOTE: d_off is the offset of the *next* entry. */ dp->d_off = UIO_MX * (i + 1); if (dp->d_namlen != 0) { /* * And ship to userland */ FILEDESC_SUNLOCK(fdp); error = uiomove(dp, UIO_MX, uio); if (error) goto done; FILEDESC_SLOCK(fdp); } i++; fcnt++; } FILEDESC_SUNLOCK(fdp); done: uio->uio_offset = i * UIO_MX; return (error); } static int fdesc_reclaim(struct vop_reclaim_args *ap) { struct vnode *vp; struct fdescnode *fd; vp = ap->a_vp; fd = VTOFDESC(vp); fdesc_remove_entry(fd); free(vp->v_data, M_TEMP); vp->v_data = NULL; return (0); } static int fdesc_readlink(struct vop_readlink_args *va) { struct vnode *vp, *vn; struct thread *td; struct uio *uio; struct file *fp; char *freepath, *fullpath; size_t pathlen; int lockflags, fd_fd; int error; freepath = NULL; vn = va->a_vp; if (VTOFDESC(vn)->fd_type != Fdesc) panic("fdesc_readlink: not fdescfs link"); fd_fd = ((struct fdescnode *)vn->v_data)->fd_fd; lockflags = VOP_ISLOCKED(vn); VOP_UNLOCK(vn); td = curthread; error = fget_cap(td, fd_fd, &cap_no_rights, &fp, NULL); if (error != 0) goto out; switch (fp->f_type) { case DTYPE_VNODE: vp = fp->f_vnode; error = vn_fullpath(vp, &fullpath, &freepath); break; default: fullpath = "anon_inode:[unknown]"; break; } if (error == 0) { uio = va->a_uio; pathlen = strlen(fullpath); error = uiomove(fullpath, pathlen, uio); } if (freepath != NULL) free(freepath, M_TEMP); fdrop(fp, td); out: vn_lock(vn, lockflags | LK_RETRY); return (error); } diff --git a/sys/fs/fuse/fuse_internal.c b/sys/fs/fuse/fuse_internal.c index ffa4a40095b2..cf11daaed68f 100644 --- a/sys/fs/fuse/fuse_internal.c +++ b/sys/fs/fuse/fuse_internal.c @@ -1,1337 +1,1336 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 2007-2009 Google Inc. and Amit Singh * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions are * met: * * * Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * * Redistributions in binary form must reproduce the above * copyright notice, this list of conditions and the following disclaimer * in the documentation and/or other materials provided with the * distribution. * * Neither the name of Google Inc. nor the names of its * contributors may be used to endorse or promote products derived from * this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. * * Copyright (C) 2005 Csaba Henk. * All rights reserved. * * Copyright (c) 2019 The FreeBSD Foundation * * Portions of this software were developed by BFF Storage Systems, LLC under * sponsorship from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "fuse.h" #include "fuse_file.h" #include "fuse_internal.h" #include "fuse_io.h" #include "fuse_ipc.h" #include "fuse_node.h" #include "fuse_file.h" SDT_PROVIDER_DECLARE(fusefs); /* * Fuse trace probe: * arg0: verbosity. Higher numbers give more verbose messages * arg1: Textual message */ SDT_PROBE_DEFINE2(fusefs, , internal, trace, "int", "char*"); #ifdef ZERO_PAD_INCOMPLETE_BUFS static int isbzero(void *buf, size_t len); #endif counter_u64_t fuse_lookup_cache_hits; counter_u64_t fuse_lookup_cache_misses; SYSCTL_COUNTER_U64(_vfs_fusefs_stats, OID_AUTO, lookup_cache_hits, CTLFLAG_RD, &fuse_lookup_cache_hits, "number of positive cache hits in lookup"); SYSCTL_COUNTER_U64(_vfs_fusefs_stats, OID_AUTO, lookup_cache_misses, CTLFLAG_RD, &fuse_lookup_cache_misses, "number of cache misses in lookup"); int fuse_internal_get_cached_vnode(struct mount* mp, ino_t ino, int flags, struct vnode **vpp) { struct bintime now; struct thread *td = curthread; uint64_t nodeid = ino; int error; *vpp = NULL; error = vfs_hash_get(mp, fuse_vnode_hash(nodeid), flags, td, vpp, fuse_vnode_cmp, &nodeid); if (error) return error; /* * Check the entry cache timeout. We have to do this within fusefs * instead of by using cache_enter_time/cache_lookup because those * routines are only intended to work with pathnames, not inodes */ if (*vpp != NULL) { getbinuptime(&now); if (bintime_cmp(&(VTOFUD(*vpp)->entry_cache_timeout), &now, >)){ counter_u64_add(fuse_lookup_cache_hits, 1); return 0; } else { /* Entry cache timeout */ counter_u64_add(fuse_lookup_cache_misses, 1); cache_purge(*vpp); vput(*vpp); *vpp = NULL; } } return 0; } SDT_PROBE_DEFINE0(fusefs, , internal, access_vadmin); /* Synchronously send a FUSE_ACCESS operation */ int fuse_internal_access(struct vnode *vp, accmode_t mode, struct thread *td, struct ucred *cred) { int err = 0; uint32_t mask = F_OK; int dataflags; int vtype; struct mount *mp; struct fuse_dispatcher fdi; struct fuse_access_in *fai; struct fuse_data *data; mp = vnode_mount(vp); vtype = vnode_vtype(vp); data = fuse_get_mpdata(mp); dataflags = data->dataflags; if (mode == 0) return 0; if (mode & VMODIFY_PERMS && vfs_isrdonly(mp)) { switch (vp->v_type) { case VDIR: /* FALLTHROUGH */ case VLNK: /* FALLTHROUGH */ case VREG: return EROFS; default: break; } } /* Unless explicitly permitted, deny everyone except the fs owner. */ if (!(dataflags & FSESS_DAEMON_CAN_SPY)) { if (fuse_match_cred(data->daemoncred, cred)) return EPERM; } if (dataflags & FSESS_DEFAULT_PERMISSIONS) { struct vattr va; fuse_internal_getattr(vp, &va, cred, td); return vaccess(vp->v_type, va.va_mode, va.va_uid, va.va_gid, mode, cred); } if (mode & VADMIN) { /* * The FUSE protocol doesn't have an equivalent of VADMIN, so * it's a bug if we ever reach this point with that bit set. */ SDT_PROBE0(fusefs, , internal, access_vadmin); } if (fsess_not_impl(mp, FUSE_ACCESS)) return 0; if ((mode & (VWRITE | VAPPEND)) != 0) mask |= W_OK; if ((mode & VREAD) != 0) mask |= R_OK; if ((mode & VEXEC) != 0) mask |= X_OK; fdisp_init(&fdi, sizeof(*fai)); fdisp_make_vp(&fdi, FUSE_ACCESS, vp, td, cred); fai = fdi.indata; fai->mask = mask; err = fdisp_wait_answ(&fdi); fdisp_destroy(&fdi); if (err == ENOSYS) { fsess_set_notimpl(mp, FUSE_ACCESS); err = 0; } return err; } /* * Cache FUSE attributes from attr, in attribute cache associated with vnode * 'vp'. Optionally, if argument 'vap' is not NULL, store a copy of the * converted attributes there as well. * * If the nominal attribute cache TTL is zero, do not cache on the 'vp' (but do * return the result to the caller). */ void fuse_internal_cache_attrs(struct vnode *vp, struct fuse_attr *attr, uint64_t attr_valid, uint32_t attr_valid_nsec, struct vattr *vap, bool from_server) { struct mount *mp; struct fuse_vnode_data *fvdat; struct fuse_data *data; struct vattr *vp_cache_at; mp = vnode_mount(vp); fvdat = VTOFUD(vp); data = fuse_get_mpdata(mp); ASSERT_VOP_ELOCKED(vp, "fuse_internal_cache_attrs"); fuse_validity_2_bintime(attr_valid, attr_valid_nsec, &fvdat->attr_cache_timeout); if (vnode_isreg(vp) && fvdat->cached_attrs.va_size != VNOVAL && attr->size != fvdat->cached_attrs.va_size) { if ( data->cache_mode == FUSE_CACHE_WB && fvdat->flag & FN_SIZECHANGE) { const char *msg; /* * The server changed the file's size even though we're * using writeback cacheing and and we have outstanding * dirty writes! That's a server bug. */ if (fuse_libabi_geq(data, 7, 23)) { msg = "writeback cache incoherent!." "To prevent data corruption, disable " "the writeback cache according to your " "FUSE server's documentation."; } else { msg = "writeback cache incoherent!." "To prevent data corruption, disable " "the writeback cache by setting " "vfs.fusefs.data_cache_mode to 0 or 1."; } fuse_warn(data, FSESS_WARN_WB_CACHE_INCOHERENT, msg); } if (fuse_vnode_attr_cache_valid(vp) && data->cache_mode != FUSE_CACHE_UC) { /* * The server changed the file's size even though we * have it cached and our cache has not yet expired. * That's a bug. */ fuse_warn(data, FSESS_WARN_CACHE_INCOHERENT, "cache incoherent! " "To prevent " "data corruption, disable the data cache " "by mounting with -o direct_io, or as " "directed otherwise by your FUSE server's " "documentation."); } } /* Fix our buffers if the filesize changed without us knowing */ if (vnode_isreg(vp) && attr->size != fvdat->cached_attrs.va_size) { (void)fuse_vnode_setsize(vp, attr->size, from_server); fvdat->cached_attrs.va_size = attr->size; } if (attr_valid > 0 || attr_valid_nsec > 0) vp_cache_at = &(fvdat->cached_attrs); else if (vap != NULL) vp_cache_at = vap; else return; vattr_null(vp_cache_at); vp_cache_at->va_fsid = mp->mnt_stat.f_fsid.val[0]; vp_cache_at->va_fileid = attr->ino; vp_cache_at->va_mode = attr->mode & ~S_IFMT; vp_cache_at->va_nlink = attr->nlink; vp_cache_at->va_uid = attr->uid; vp_cache_at->va_gid = attr->gid; vp_cache_at->va_rdev = attr->rdev; vp_cache_at->va_size = attr->size; /* XXX on i386, seconds are truncated to 32 bits */ vp_cache_at->va_atime.tv_sec = attr->atime; vp_cache_at->va_atime.tv_nsec = attr->atimensec; vp_cache_at->va_mtime.tv_sec = attr->mtime; vp_cache_at->va_mtime.tv_nsec = attr->mtimensec; vp_cache_at->va_ctime.tv_sec = attr->ctime; vp_cache_at->va_ctime.tv_nsec = attr->ctimensec; if (fuse_libabi_geq(data, 7, 9) && attr->blksize > 0) vp_cache_at->va_blocksize = attr->blksize; else vp_cache_at->va_blocksize = PAGE_SIZE; vp_cache_at->va_type = IFTOVT(attr->mode); vp_cache_at->va_bytes = attr->blocks * S_BLKSIZE; vp_cache_at->va_flags = 0; if (vap != vp_cache_at && vap != NULL) memcpy(vap, vp_cache_at, sizeof(*vap)); } /* fsync */ int fuse_internal_fsync_callback(struct fuse_ticket *tick, struct uio *uio) { if (tick->tk_aw_ohead.error == ENOSYS) { fsess_set_notimpl(tick->tk_data->mp, fticket_opcode(tick)); } return 0; } int fuse_internal_fsync(struct vnode *vp, struct thread *td, int waitfor, bool datasync) { struct fuse_fsync_in *ffsi = NULL; struct fuse_dispatcher fdi; struct fuse_filehandle *fufh; struct fuse_vnode_data *fvdat = VTOFUD(vp); struct mount *mp = vnode_mount(vp); int op = FUSE_FSYNC; int err = 0; if (fsess_not_impl(vnode_mount(vp), (vnode_vtype(vp) == VDIR ? FUSE_FSYNCDIR : FUSE_FSYNC))) { return 0; } if (vnode_isdir(vp)) op = FUSE_FSYNCDIR; if (fsess_not_impl(mp, op)) return 0; fdisp_init(&fdi, sizeof(*ffsi)); /* * fsync every open file handle for this file, because we can't be sure * which file handle the caller is really referring to. */ LIST_FOREACH(fufh, &fvdat->handles, next) { fdi.iosize = sizeof(*ffsi); if (ffsi == NULL) fdisp_make_vp(&fdi, op, vp, td, NULL); else fdisp_refresh_vp(&fdi, op, vp, td, NULL); ffsi = fdi.indata; ffsi->fh = fufh->fh_id; ffsi->fsync_flags = 0; if (datasync) ffsi->fsync_flags = FUSE_FSYNC_FDATASYNC; if (waitfor == MNT_WAIT) { err = fdisp_wait_answ(&fdi); } else { fuse_insert_callback(fdi.tick, fuse_internal_fsync_callback); fuse_insert_message(fdi.tick, false); } if (err == ENOSYS) { /* ENOSYS means "success, and don't call again" */ fsess_set_notimpl(mp, op); err = 0; break; } } fdisp_destroy(&fdi); return err; } /* Asynchronous invalidation */ SDT_PROBE_DEFINE3(fusefs, , internal, invalidate_entry, "struct vnode*", "struct fuse_notify_inval_entry_out*", "char*"); int fuse_internal_invalidate_entry(struct mount *mp, struct uio *uio) { struct fuse_notify_inval_entry_out fnieo; struct componentname cn; struct vnode *dvp, *vp; char name[PATH_MAX]; int err; if ((err = uiomove(&fnieo, sizeof(fnieo), uio)) != 0) return (err); if (fnieo.namelen >= sizeof(name)) return (EINVAL); if ((err = uiomove(name, fnieo.namelen, uio)) != 0) return (err); name[fnieo.namelen] = '\0'; /* fusefs does not cache "." or ".." entries */ if (strncmp(name, ".", sizeof(".")) == 0 || strncmp(name, "..", sizeof("..")) == 0) return (0); if (fnieo.parent == FUSE_ROOT_ID) err = VFS_ROOT(mp, LK_SHARED, &dvp); else err = fuse_internal_get_cached_vnode( mp, fnieo.parent, LK_SHARED, &dvp); SDT_PROBE3(fusefs, , internal, invalidate_entry, dvp, &fnieo, name); /* * If dvp is not in the cache, then it must've been reclaimed. And * since fuse_vnop_reclaim does a cache_purge, name's entry must've * been invalidated already. So we can safely return if dvp == NULL */ if (err != 0 || dvp == NULL) return (err); /* * XXX we can't check dvp's generation because the FUSE invalidate * entry message doesn't include it. Worse case is that we invalidate * an entry that didn't need to be invalidated. */ cn.cn_nameiop = LOOKUP; cn.cn_flags = 0; /* !MAKEENTRY means free cached entry */ - cn.cn_thread = curthread; cn.cn_cred = curthread->td_ucred; cn.cn_lkflags = LK_SHARED; cn.cn_pnbuf = NULL; cn.cn_nameptr = name; cn.cn_namelen = fnieo.namelen; err = cache_lookup(dvp, &vp, &cn, NULL, NULL); MPASS(err == 0); fuse_vnode_clear_attr_cache(dvp); vput(dvp); return (0); } SDT_PROBE_DEFINE2(fusefs, , internal, invalidate_inode, "struct vnode*", "struct fuse_notify_inval_inode_out *"); int fuse_internal_invalidate_inode(struct mount *mp, struct uio *uio) { struct fuse_notify_inval_inode_out fniio; struct vnode *vp; int err; if ((err = uiomove(&fniio, sizeof(fniio), uio)) != 0) return (err); if (fniio.ino == FUSE_ROOT_ID) err = VFS_ROOT(mp, LK_EXCLUSIVE, &vp); else err = fuse_internal_get_cached_vnode(mp, fniio.ino, LK_SHARED, &vp); SDT_PROBE2(fusefs, , internal, invalidate_inode, vp, &fniio); if (err != 0 || vp == NULL) return (err); /* * XXX we can't check vp's generation because the FUSE invalidate * entry message doesn't include it. Worse case is that we invalidate * an inode that didn't need to be invalidated. */ /* * Flush and invalidate buffers if off >= 0. Technically we only need * to flush and invalidate the range of offsets [off, off + len), but * for simplicity's sake we do everything. */ if (fniio.off >= 0) fuse_io_invalbuf(vp, curthread); fuse_vnode_clear_attr_cache(vp); vput(vp); return (0); } /* mknod */ int fuse_internal_mknod(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct vattr *vap) { struct fuse_data *data; struct fuse_mknod_in fmni; size_t insize; data = fuse_get_mpdata(dvp->v_mount); fmni.mode = MAKEIMODE(vap->va_type, vap->va_mode); fmni.rdev = vap->va_rdev; if (fuse_libabi_geq(data, 7, 12)) { insize = sizeof(fmni); fmni.umask = curthread->td_proc->p_pd->pd_cmask; } else { insize = FUSE_COMPAT_MKNOD_IN_SIZE; } return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKNOD, &fmni, insize, vap->va_type)); } /* readdir */ int fuse_internal_readdir(struct vnode *vp, struct uio *uio, off_t startoff, struct fuse_filehandle *fufh, struct fuse_iov *cookediov, int *ncookies, u_long *cookies) { int err = 0; struct fuse_dispatcher fdi; struct fuse_read_in *fri = NULL; int fnd_start; if (uio_resid(uio) == 0) return 0; fdisp_init(&fdi, 0); /* * Note that we DO NOT have a UIO_SYSSPACE here (so no need for p2p * I/O). */ /* * fnd_start is set non-zero once the offset in the directory gets * to the startoff. This is done because directories must be read * from the beginning (offset == 0) when fuse_vnop_readdir() needs * to do an open of the directory. * If it is not set non-zero here, it will be set non-zero in * fuse_internal_readdir_processdata() when uio_offset == startoff. */ fnd_start = 0; if (uio->uio_offset == startoff) fnd_start = 1; while (uio_resid(uio) > 0) { fdi.iosize = sizeof(*fri); if (fri == NULL) fdisp_make_vp(&fdi, FUSE_READDIR, vp, NULL, NULL); else fdisp_refresh_vp(&fdi, FUSE_READDIR, vp, NULL, NULL); fri = fdi.indata; fri->fh = fufh->fh_id; fri->offset = uio_offset(uio); fri->size = MIN(uio->uio_resid, fuse_get_mpdata(vp->v_mount)->max_read); if ((err = fdisp_wait_answ(&fdi))) break; if ((err = fuse_internal_readdir_processdata(uio, startoff, &fnd_start, fri->size, fdi.answ, fdi.iosize, cookediov, ncookies, &cookies))) break; } fdisp_destroy(&fdi); return ((err == -1) ? 0 : err); } /* * Return -1 to indicate that this readdir is finished, 0 if it copied * all the directory data read in and it may be possible to read more * and greater than 0 for a failure. */ int fuse_internal_readdir_processdata(struct uio *uio, off_t startoff, int *fnd_start, size_t reqsize, void *buf, size_t bufsize, struct fuse_iov *cookediov, int *ncookies, u_long **cookiesp) { int err = 0; int oreclen; size_t freclen; struct dirent *de; struct fuse_dirent *fudge; u_long *cookies; cookies = *cookiesp; if (bufsize < FUSE_NAME_OFFSET) return -1; for (;;) { if (bufsize < FUSE_NAME_OFFSET) { err = -1; break; } fudge = (struct fuse_dirent *)buf; freclen = FUSE_DIRENT_SIZE(fudge); if (bufsize < freclen) { /* * This indicates a partial directory entry at the * end of the directory data. */ err = -1; break; } #ifdef ZERO_PAD_INCOMPLETE_BUFS if (isbzero(buf, FUSE_NAME_OFFSET)) { err = -1; break; } #endif if (!fudge->namelen || fudge->namelen > MAXNAMLEN) { err = EINVAL; break; } oreclen = GENERIC_DIRSIZ((struct pseudo_dirent *) &fudge->namelen); if (oreclen > uio_resid(uio)) { /* Out of space for the dir so we are done. */ err = -1; break; } /* * Don't start to copy the directory entries out until * the requested offset in the directory is found. */ if (*fnd_start != 0) { fiov_adjust(cookediov, oreclen); bzero(cookediov->base, oreclen); de = (struct dirent *)cookediov->base; de->d_fileno = fudge->ino; de->d_off = fudge->off; de->d_reclen = oreclen; de->d_type = fudge->type; de->d_namlen = fudge->namelen; memcpy((char *)cookediov->base + sizeof(struct dirent) - MAXNAMLEN - 1, (char *)buf + FUSE_NAME_OFFSET, fudge->namelen); dirent_terminate(de); err = uiomove(cookediov->base, cookediov->len, uio); if (err) break; if (cookies != NULL) { if (*ncookies == 0) { err = -1; break; } *cookies = fudge->off; cookies++; (*ncookies)--; } } else if (startoff == fudge->off) *fnd_start = 1; buf = (char *)buf + freclen; bufsize -= freclen; uio_setoffset(uio, fudge->off); } *cookiesp = cookies; return err; } /* remove */ int fuse_internal_remove(struct vnode *dvp, struct vnode *vp, struct componentname *cnp, enum fuse_opcode op) { struct fuse_dispatcher fdi; nlink_t nlink; int err = 0; fdisp_init(&fdi, cnp->cn_namelen + 1); - fdisp_make_vp(&fdi, op, dvp, cnp->cn_thread, cnp->cn_cred); + fdisp_make_vp(&fdi, op, dvp, curthread, cnp->cn_cred); memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen); ((char *)fdi.indata)[cnp->cn_namelen] = '\0'; err = fdisp_wait_answ(&fdi); fdisp_destroy(&fdi); if (err) return (err); /* * Access the cached nlink even if the attr cached has expired. If * it's inaccurate, the worst that will happen is: * 1) We'll recycle the vnode even though the file has another link we * don't know about, costing a bit of cpu time, or * 2) We won't recycle the vnode even though all of its links are gone. * It will linger around until vnlru reclaims it, costing a bit of * temporary memory. */ nlink = VTOFUD(vp)->cached_attrs.va_nlink--; /* * Purge the parent's attribute cache because the daemon * should've updated its mtime and ctime. */ fuse_vnode_clear_attr_cache(dvp); /* NB: nlink could be zero if it was never cached */ if (nlink <= 1 || vnode_vtype(vp) == VDIR) { fuse_internal_vnode_disappear(vp); } else { cache_purge(vp); fuse_vnode_update(vp, FN_CTIMECHANGE); } return err; } /* rename */ int fuse_internal_rename(struct vnode *fdvp, struct componentname *fcnp, struct vnode *tdvp, struct componentname *tcnp) { struct fuse_dispatcher fdi; struct fuse_rename_in *fri; int err = 0; fdisp_init(&fdi, sizeof(*fri) + fcnp->cn_namelen + tcnp->cn_namelen + 2); - fdisp_make_vp(&fdi, FUSE_RENAME, fdvp, tcnp->cn_thread, tcnp->cn_cred); + fdisp_make_vp(&fdi, FUSE_RENAME, fdvp, curthread, tcnp->cn_cred); fri = fdi.indata; fri->newdir = VTOI(tdvp); memcpy((char *)fdi.indata + sizeof(*fri), fcnp->cn_nameptr, fcnp->cn_namelen); ((char *)fdi.indata)[sizeof(*fri) + fcnp->cn_namelen] = '\0'; memcpy((char *)fdi.indata + sizeof(*fri) + fcnp->cn_namelen + 1, tcnp->cn_nameptr, tcnp->cn_namelen); ((char *)fdi.indata)[sizeof(*fri) + fcnp->cn_namelen + tcnp->cn_namelen + 1] = '\0'; err = fdisp_wait_answ(&fdi); fdisp_destroy(&fdi); return err; } /* strategy */ /* entity creation */ void fuse_internal_newentry_makerequest(struct mount *mp, uint64_t dnid, struct componentname *cnp, enum fuse_opcode op, void *buf, size_t bufsize, struct fuse_dispatcher *fdip) { fdip->iosize = bufsize + cnp->cn_namelen + 1; - fdisp_make(fdip, op, mp, dnid, cnp->cn_thread, cnp->cn_cred); + fdisp_make(fdip, op, mp, dnid, curthread, cnp->cn_cred); memcpy(fdip->indata, buf, bufsize); memcpy((char *)fdip->indata + bufsize, cnp->cn_nameptr, cnp->cn_namelen); ((char *)fdip->indata)[bufsize + cnp->cn_namelen] = '\0'; } int fuse_internal_newentry_core(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, enum vtype vtyp, struct fuse_dispatcher *fdip) { int err = 0; struct fuse_entry_out *feo; struct mount *mp = vnode_mount(dvp); if ((err = fdisp_wait_answ(fdip))) { return err; } feo = fdip->answ; if ((err = fuse_internal_checkentry(feo, vtyp))) { return err; } err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vtyp); if (err) { - fuse_internal_forget_send(mp, cnp->cn_thread, cnp->cn_cred, + fuse_internal_forget_send(mp, curthread, cnp->cn_cred, feo->nodeid, 1); return err; } /* * Purge the parent's attribute cache because the daemon should've * updated its mtime and ctime */ fuse_vnode_clear_attr_cache(dvp); fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid, feo->attr_valid_nsec, NULL, true); return err; } int fuse_internal_newentry(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, enum fuse_opcode op, void *buf, size_t bufsize, enum vtype vtype) { int err; struct fuse_dispatcher fdi; struct mount *mp = vnode_mount(dvp); fdisp_init(&fdi, 0); fuse_internal_newentry_makerequest(mp, VTOI(dvp), cnp, op, buf, bufsize, &fdi); err = fuse_internal_newentry_core(dvp, vpp, cnp, vtype, &fdi); fdisp_destroy(&fdi); return err; } /* entity destruction */ int fuse_internal_forget_callback(struct fuse_ticket *ftick, struct uio *uio) { fuse_internal_forget_send(ftick->tk_data->mp, curthread, NULL, ((struct fuse_in_header *)ftick->tk_ms_fiov.base)->nodeid, 1); return 0; } void fuse_internal_forget_send(struct mount *mp, struct thread *td, struct ucred *cred, uint64_t nodeid, uint64_t nlookup) { struct fuse_dispatcher fdi; struct fuse_forget_in *ffi; /* * KASSERT(nlookup > 0, ("zero-times forget for vp #%llu", * (long long unsigned) nodeid)); */ fdisp_init(&fdi, sizeof(*ffi)); fdisp_make(&fdi, FUSE_FORGET, mp, nodeid, td, cred); ffi = fdi.indata; ffi->nlookup = nlookup; fuse_insert_message(fdi.tick, false); fdisp_destroy(&fdi); } /* Fetch the vnode's attributes from the daemon*/ int fuse_internal_do_getattr(struct vnode *vp, struct vattr *vap, struct ucred *cred, struct thread *td) { struct fuse_dispatcher fdi; struct fuse_vnode_data *fvdat = VTOFUD(vp); struct fuse_getattr_in *fgai; struct fuse_attr_out *fao; off_t old_filesize = fvdat->cached_attrs.va_size; struct timespec old_ctime = fvdat->cached_attrs.va_ctime; struct timespec old_mtime = fvdat->cached_attrs.va_mtime; enum vtype vtyp; int err; fdisp_init(&fdi, sizeof(*fgai)); fdisp_make_vp(&fdi, FUSE_GETATTR, vp, td, cred); fgai = fdi.indata; /* * We could look up a file handle and set it in fgai->fh, but that * involves extra runtime work and I'm unaware of any file systems that * care. */ fgai->getattr_flags = 0; if ((err = fdisp_wait_answ(&fdi))) { if (err == ENOENT) fuse_internal_vnode_disappear(vp); goto out; } fao = (struct fuse_attr_out *)fdi.answ; vtyp = IFTOVT(fao->attr.mode); if (fvdat->flag & FN_SIZECHANGE) fao->attr.size = old_filesize; if (fvdat->flag & FN_CTIMECHANGE) { fao->attr.ctime = old_ctime.tv_sec; fao->attr.ctimensec = old_ctime.tv_nsec; } if (fvdat->flag & FN_MTIMECHANGE) { fao->attr.mtime = old_mtime.tv_sec; fao->attr.mtimensec = old_mtime.tv_nsec; } fuse_internal_cache_attrs(vp, &fao->attr, fao->attr_valid, fao->attr_valid_nsec, vap, true); if (vtyp != vnode_vtype(vp)) { fuse_internal_vnode_disappear(vp); err = ENOENT; } out: fdisp_destroy(&fdi); return err; } /* Read a vnode's attributes from cache or fetch them from the fuse daemon */ int fuse_internal_getattr(struct vnode *vp, struct vattr *vap, struct ucred *cred, struct thread *td) { struct vattr *attrs; if ((attrs = VTOVA(vp)) != NULL) { *vap = *attrs; /* struct copy */ return 0; } return fuse_internal_do_getattr(vp, vap, cred, td); } void fuse_internal_vnode_disappear(struct vnode *vp) { struct fuse_vnode_data *fvdat = VTOFUD(vp); ASSERT_VOP_ELOCKED(vp, "fuse_internal_vnode_disappear"); fvdat->flag |= FN_REVOKED; cache_purge(vp); } /* fuse start/stop */ SDT_PROBE_DEFINE2(fusefs, , internal, init_done, "struct fuse_data*", "struct fuse_init_out*"); int fuse_internal_init_callback(struct fuse_ticket *tick, struct uio *uio) { int err = 0; struct fuse_data *data = tick->tk_data; struct fuse_init_out *fiio; if ((err = tick->tk_aw_ohead.error)) { goto out; } if ((err = fticket_pull(tick, uio))) { goto out; } fiio = fticket_resp(tick)->base; data->fuse_libabi_major = fiio->major; data->fuse_libabi_minor = fiio->minor; if (!fuse_libabi_geq(data, 7, 4)) { /* * With a little work we could support servers as old as 7.1. * But there would be little payoff. */ SDT_PROBE2(fusefs, , internal, trace, 1, "userpace version too low"); err = EPROTONOSUPPORT; goto out; } if (fuse_libabi_geq(data, 7, 5)) { if (fticket_resp(tick)->len == sizeof(struct fuse_init_out) || fticket_resp(tick)->len == FUSE_COMPAT_22_INIT_OUT_SIZE) { data->max_write = fiio->max_write; if (fiio->flags & FUSE_ASYNC_READ) data->dataflags |= FSESS_ASYNC_READ; if (fiio->flags & FUSE_POSIX_LOCKS) data->dataflags |= FSESS_POSIX_LOCKS; if (fiio->flags & FUSE_EXPORT_SUPPORT) data->dataflags |= FSESS_EXPORT_SUPPORT; if (fiio->flags & FUSE_NO_OPEN_SUPPORT) data->dataflags |= FSESS_NO_OPEN_SUPPORT; if (fiio->flags & FUSE_NO_OPENDIR_SUPPORT) data->dataflags |= FSESS_NO_OPENDIR_SUPPORT; /* * Don't bother to check FUSE_BIG_WRITES, because it's * redundant with max_write */ /* * max_background and congestion_threshold are not * implemented */ } else { err = EINVAL; } } else { /* Old fixed values */ data->max_write = 4096; } if (fuse_libabi_geq(data, 7, 6)) data->max_readahead_blocks = fiio->max_readahead / maxbcachebuf; if (!fuse_libabi_geq(data, 7, 7)) fsess_set_notimpl(data->mp, FUSE_INTERRUPT); if (!fuse_libabi_geq(data, 7, 8)) { fsess_set_notimpl(data->mp, FUSE_BMAP); fsess_set_notimpl(data->mp, FUSE_DESTROY); } if (fuse_libabi_geq(data, 7, 23) && fiio->time_gran >= 1 && fiio->time_gran <= 1000000000) data->time_gran = fiio->time_gran; else data->time_gran = 1; if (!fuse_libabi_geq(data, 7, 23)) data->cache_mode = fuse_data_cache_mode; else if (fiio->flags & FUSE_WRITEBACK_CACHE) data->cache_mode = FUSE_CACHE_WB; else data->cache_mode = FUSE_CACHE_WT; if (!fuse_libabi_geq(data, 7, 24)) fsess_set_notimpl(data->mp, FUSE_LSEEK); if (!fuse_libabi_geq(data, 7, 28)) fsess_set_notimpl(data->mp, FUSE_COPY_FILE_RANGE); out: if (err) { fdata_set_dead(data); } FUSE_LOCK(); data->dataflags |= FSESS_INITED; SDT_PROBE2(fusefs, , internal, init_done, data, fiio); wakeup(&data->ticketer); FUSE_UNLOCK(); return 0; } void fuse_internal_send_init(struct fuse_data *data, struct thread *td) { struct fuse_init_in *fiii; struct fuse_dispatcher fdi; fdisp_init(&fdi, sizeof(*fiii)); fdisp_make(&fdi, FUSE_INIT, data->mp, 0, td, NULL); fiii = fdi.indata; fiii->major = FUSE_KERNEL_VERSION; fiii->minor = FUSE_KERNEL_MINOR_VERSION; /* * fusefs currently reads ahead no more than one cache block at a time. * See fuse_read_biobackend */ fiii->max_readahead = maxbcachebuf; /* * Unsupported features: * FUSE_FILE_OPS: No known FUSE server or client supports it * FUSE_ATOMIC_O_TRUNC: our VFS cannot support it * FUSE_DONT_MASK: unlike Linux, FreeBSD always applies the umask, even * when default ACLs are in use. * FUSE_SPLICE_WRITE, FUSE_SPLICE_MOVE, FUSE_SPLICE_READ: FreeBSD * doesn't have splice(2). * FUSE_FLOCK_LOCKS: not yet implemented * FUSE_HAS_IOCTL_DIR: not yet implemented * FUSE_AUTO_INVAL_DATA: not yet implemented * FUSE_DO_READDIRPLUS: not yet implemented * FUSE_READDIRPLUS_AUTO: not yet implemented * FUSE_ASYNC_DIO: not yet implemented * FUSE_PARALLEL_DIROPS: not yet implemented * FUSE_HANDLE_KILLPRIV: not yet implemented * FUSE_POSIX_ACL: not yet implemented * FUSE_ABORT_ERROR: not yet implemented * FUSE_CACHE_SYMLINKS: not yet implemented * FUSE_MAX_PAGES: not yet implemented */ fiii->flags = FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT | FUSE_NO_OPENDIR_SUPPORT; fuse_insert_callback(fdi.tick, fuse_internal_init_callback); fuse_insert_message(fdi.tick, false); fdisp_destroy(&fdi); } /* * Send a FUSE_SETATTR operation with no permissions checks. If cred is NULL, * send the request with root credentials */ int fuse_internal_setattr(struct vnode *vp, struct vattr *vap, struct thread *td, struct ucred *cred) { struct fuse_vnode_data *fvdat; struct fuse_dispatcher fdi; struct fuse_setattr_in *fsai; struct mount *mp; pid_t pid = td->td_proc->p_pid; struct fuse_data *data; int dataflags; int err = 0; enum vtype vtyp; int sizechanged = -1; uint64_t newsize = 0; mp = vnode_mount(vp); fvdat = VTOFUD(vp); data = fuse_get_mpdata(mp); dataflags = data->dataflags; fdisp_init(&fdi, sizeof(*fsai)); fdisp_make_vp(&fdi, FUSE_SETATTR, vp, td, cred); if (!cred) { fdi.finh->uid = 0; fdi.finh->gid = 0; } fsai = fdi.indata; fsai->valid = 0; if (vap->va_uid != (uid_t)VNOVAL) { fsai->uid = vap->va_uid; fsai->valid |= FATTR_UID; } if (vap->va_gid != (gid_t)VNOVAL) { fsai->gid = vap->va_gid; fsai->valid |= FATTR_GID; } if (vap->va_size != VNOVAL) { struct fuse_filehandle *fufh = NULL; /*Truncate to a new value. */ fsai->size = vap->va_size; sizechanged = 1; newsize = vap->va_size; fsai->valid |= FATTR_SIZE; fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid); if (fufh) { fsai->fh = fufh->fh_id; fsai->valid |= FATTR_FH; } VTOFUD(vp)->flag &= ~FN_SIZECHANGE; } if (vap->va_atime.tv_sec != VNOVAL) { fsai->atime = vap->va_atime.tv_sec; fsai->atimensec = vap->va_atime.tv_nsec; fsai->valid |= FATTR_ATIME; if (vap->va_vaflags & VA_UTIMES_NULL) fsai->valid |= FATTR_ATIME_NOW; } if (vap->va_mtime.tv_sec != VNOVAL) { fsai->mtime = vap->va_mtime.tv_sec; fsai->mtimensec = vap->va_mtime.tv_nsec; fsai->valid |= FATTR_MTIME; if (vap->va_vaflags & VA_UTIMES_NULL) fsai->valid |= FATTR_MTIME_NOW; } else if (fvdat->flag & FN_MTIMECHANGE) { fsai->mtime = fvdat->cached_attrs.va_mtime.tv_sec; fsai->mtimensec = fvdat->cached_attrs.va_mtime.tv_nsec; fsai->valid |= FATTR_MTIME; } if (fuse_libabi_geq(data, 7, 23) && fvdat->flag & FN_CTIMECHANGE) { fsai->ctime = fvdat->cached_attrs.va_ctime.tv_sec; fsai->ctimensec = fvdat->cached_attrs.va_ctime.tv_nsec; fsai->valid |= FATTR_CTIME; } if (vap->va_mode != (mode_t)VNOVAL) { fsai->mode = vap->va_mode & ALLPERMS; fsai->valid |= FATTR_MODE; } if (!fsai->valid) { goto out; } if ((err = fdisp_wait_answ(&fdi))) goto out; vtyp = IFTOVT(((struct fuse_attr_out *)fdi.answ)->attr.mode); if (vnode_vtype(vp) != vtyp) { if (vnode_vtype(vp) == VNON && vtyp != VNON) { SDT_PROBE2(fusefs, , internal, trace, 1, "FUSE: Dang! " "vnode_vtype is VNON and vtype isn't."); } else { /* * STALE vnode, ditch * * The vnode has changed its type "behind our back". * There's nothing really we can do, so let us just * force an internal revocation and tell the caller to * try again, if interested. */ fuse_internal_vnode_disappear(vp); err = EAGAIN; } } if (err == 0) { struct fuse_attr_out *fao = (struct fuse_attr_out*)fdi.answ; fuse_vnode_undirty_cached_timestamps(vp); fuse_internal_cache_attrs(vp, &fao->attr, fao->attr_valid, fao->attr_valid_nsec, NULL, false); } out: fdisp_destroy(&fdi); return err; } /* * FreeBSD clears the SUID and SGID bits on any write by a non-root user. */ void fuse_internal_clear_suid_on_write(struct vnode *vp, struct ucred *cred, struct thread *td) { struct fuse_data *data; struct mount *mp; struct vattr va; int dataflags; mp = vnode_mount(vp); data = fuse_get_mpdata(mp); dataflags = data->dataflags; ASSERT_VOP_LOCKED(vp, __func__); if (dataflags & FSESS_DEFAULT_PERMISSIONS) { if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID)) { fuse_internal_getattr(vp, &va, cred, td); if (va.va_mode & (S_ISUID | S_ISGID)) { mode_t mode = va.va_mode & ~(S_ISUID | S_ISGID); /* Clear all vattr fields except mode */ vattr_null(&va); va.va_mode = mode; /* * Ignore fuse_internal_setattr's return value, * because at this point the write operation has * already succeeded and we don't want to return * failing status for that. */ (void)fuse_internal_setattr(vp, &va, td, NULL); } } } } #ifdef ZERO_PAD_INCOMPLETE_BUFS static int isbzero(void *buf, size_t len) { int i; for (i = 0; i < len; i++) { if (((char *)buf)[i]) return (0); } return (1); } #endif void fuse_internal_init(void) { fuse_lookup_cache_misses = counter_u64_alloc(M_WAITOK); fuse_lookup_cache_hits = counter_u64_alloc(M_WAITOK); } void fuse_internal_destroy(void) { counter_u64_free(fuse_lookup_cache_hits); counter_u64_free(fuse_lookup_cache_misses); } diff --git a/sys/fs/fuse/fuse_node.c b/sys/fs/fuse/fuse_node.c index c296a3b3c330..937f80ffc9ce 100644 --- a/sys/fs/fuse/fuse_node.c +++ b/sys/fs/fuse/fuse_node.c @@ -1,516 +1,516 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 2007-2009 Google Inc. and Amit Singh * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions are * met: * * * Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * * Redistributions in binary form must reproduce the above * copyright notice, this list of conditions and the following disclaimer * in the documentation and/or other materials provided with the * distribution. * * Neither the name of Google Inc. nor the names of its * contributors may be used to endorse or promote products derived from * this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. * * Copyright (C) 2005 Csaba Henk. * All rights reserved. * * Copyright (c) 2019 The FreeBSD Foundation * * Portions of this software were developed by BFF Storage Systems, LLC under * sponsorship from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "fuse.h" #include "fuse_node.h" #include "fuse_internal.h" #include "fuse_io.h" #include "fuse_ipc.h" SDT_PROVIDER_DECLARE(fusefs); /* * Fuse trace probe: * arg0: verbosity. Higher numbers give more verbose messages * arg1: Textual message */ SDT_PROBE_DEFINE2(fusefs, , node, trace, "int", "char*"); MALLOC_DEFINE(M_FUSEVN, "fuse_vnode", "fuse vnode private data"); static int sysctl_fuse_cache_mode(SYSCTL_HANDLER_ARGS); static counter_u64_t fuse_node_count; SYSCTL_COUNTER_U64(_vfs_fusefs_stats, OID_AUTO, node_count, CTLFLAG_RD, &fuse_node_count, "Count of FUSE vnodes"); int fuse_data_cache_mode = FUSE_CACHE_WT; /* * DEPRECATED * This sysctl is no longer needed as of fuse protocol 7.23. Individual * servers can select the cache behavior they need for each mountpoint: * - writethrough: the default * - writeback: set FUSE_WRITEBACK_CACHE in fuse_init_out.flags * - uncached: set FOPEN_DIRECT_IO for every file * The sysctl is retained primarily for use by jails supporting older FUSE * protocols. It may be removed entirely once FreeBSD 11.3 and 12.0 are EOL. */ SYSCTL_PROC(_vfs_fusefs, OID_AUTO, data_cache_mode, CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, &fuse_data_cache_mode, 0, sysctl_fuse_cache_mode, "I", "Zero: disable caching of FUSE file data; One: write-through caching " "(default); Two: write-back caching (generally unsafe)"); static int sysctl_fuse_cache_mode(SYSCTL_HANDLER_ARGS) { int val, error; val = *(int *)arg1; error = sysctl_handle_int(oidp, &val, 0, req); if (error || !req->newptr) return (error); switch (val) { case FUSE_CACHE_UC: case FUSE_CACHE_WT: case FUSE_CACHE_WB: *(int *)arg1 = val; break; default: return (EDOM); } return (0); } static void fuse_vnode_init(struct vnode *vp, struct fuse_vnode_data *fvdat, uint64_t nodeid, enum vtype vtyp) { fvdat->nid = nodeid; LIST_INIT(&fvdat->handles); vattr_null(&fvdat->cached_attrs); if (nodeid == FUSE_ROOT_ID) { vp->v_vflag |= VV_ROOT; } vp->v_type = vtyp; vp->v_data = fvdat; cluster_init_vn(&fvdat->clusterw); counter_u64_add(fuse_node_count, 1); } void fuse_vnode_destroy(struct vnode *vp) { struct fuse_vnode_data *fvdat = vp->v_data; vp->v_data = NULL; KASSERT(LIST_EMPTY(&fvdat->handles), ("Destroying fuse vnode with open files!")); free(fvdat, M_FUSEVN); counter_u64_add(fuse_node_count, -1); } int fuse_vnode_cmp(struct vnode *vp, void *nidp) { return (VTOI(vp) != *((uint64_t *)nidp)); } SDT_PROBE_DEFINE3(fusefs, , node, stale_vnode, "struct vnode*", "enum vtype", "uint64_t"); static int fuse_vnode_alloc(struct mount *mp, struct thread *td, uint64_t nodeid, enum vtype vtyp, struct vnode **vpp) { struct fuse_data *data; struct fuse_vnode_data *fvdat; struct vnode *vp2; int err = 0; data = fuse_get_mpdata(mp); if (vtyp == VNON) { return EINVAL; } *vpp = NULL; err = vfs_hash_get(mp, fuse_vnode_hash(nodeid), LK_EXCLUSIVE, td, vpp, fuse_vnode_cmp, &nodeid); if (err) return (err); if (*vpp) { if ((*vpp)->v_type != vtyp) { /* * STALE vnode! This probably indicates a buggy * server, but it could also be the result of a race * between FUSE_LOOKUP and another client's * FUSE_UNLINK/FUSE_CREATE */ SDT_PROBE3(fusefs, , node, stale_vnode, *vpp, vtyp, nodeid); fuse_internal_vnode_disappear(*vpp); lockmgr((*vpp)->v_vnlock, LK_RELEASE, NULL); *vpp = NULL; return (EAGAIN); } MPASS((*vpp)->v_data != NULL); MPASS(VTOFUD(*vpp)->nid == nodeid); SDT_PROBE2(fusefs, , node, trace, 1, "vnode taken from hash"); return (0); } fvdat = malloc(sizeof(*fvdat), M_FUSEVN, M_WAITOK | M_ZERO); switch (vtyp) { case VFIFO: err = getnewvnode("fuse", mp, &fuse_fifoops, vpp); break; default: err = getnewvnode("fuse", mp, &fuse_vnops, vpp); break; } if (err) { free(fvdat, M_FUSEVN); return (err); } lockmgr((*vpp)->v_vnlock, LK_EXCLUSIVE, NULL); fuse_vnode_init(*vpp, fvdat, nodeid, vtyp); err = insmntque(*vpp, mp); ASSERT_VOP_ELOCKED(*vpp, "fuse_vnode_alloc"); if (err) { lockmgr((*vpp)->v_vnlock, LK_RELEASE, NULL); free(fvdat, M_FUSEVN); *vpp = NULL; return (err); } /* Disallow async reads for fifos because UFS does. I don't know why */ if (data->dataflags & FSESS_ASYNC_READ && vtyp != VFIFO) VN_LOCK_ASHARE(*vpp); err = vfs_hash_insert(*vpp, fuse_vnode_hash(nodeid), LK_EXCLUSIVE, td, &vp2, fuse_vnode_cmp, &nodeid); if (err) { lockmgr((*vpp)->v_vnlock, LK_RELEASE, NULL); free(fvdat, M_FUSEVN); *vpp = NULL; return (err); } if (vp2 != NULL) { *vpp = vp2; return (0); } ASSERT_VOP_ELOCKED(*vpp, "fuse_vnode_alloc"); return (0); } int fuse_vnode_get(struct mount *mp, struct fuse_entry_out *feo, uint64_t nodeid, struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, enum vtype vtyp) { - struct thread *td = (cnp != NULL ? cnp->cn_thread : curthread); + struct thread *td = curthread; /* * feo should only be NULL for the root directory, which (when libfuse * is used) always has generation 0 */ uint64_t generation = feo ? feo->generation : 0; int err = 0; err = fuse_vnode_alloc(mp, td, nodeid, vtyp, vpp); if (err) { return err; } if (dvp != NULL) { MPASS(cnp && (cnp->cn_flags & ISDOTDOT) == 0); MPASS(cnp && !(cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.')); fuse_vnode_setparent(*vpp, dvp); } if (dvp != NULL && cnp != NULL && (cnp->cn_flags & MAKEENTRY) != 0 && feo != NULL && (feo->entry_valid != 0 || feo->entry_valid_nsec != 0)) { struct timespec timeout; ASSERT_VOP_LOCKED(*vpp, "fuse_vnode_get"); ASSERT_VOP_LOCKED(dvp, "fuse_vnode_get"); fuse_validity_2_timespec(feo, &timeout); cache_enter_time(dvp, *vpp, cnp, &timeout, NULL); } VTOFUD(*vpp)->generation = generation; /* * In userland, libfuse uses cached lookups for dot and dotdot entries, * thus it does not really bump the nlookup counter for forget. * Follow the same semantic and avoid the bump in order to keep * nlookup counters consistent. */ if (cnp == NULL || ((cnp->cn_flags & ISDOTDOT) == 0 && (cnp->cn_namelen != 1 || cnp->cn_nameptr[0] != '.'))) VTOFUD(*vpp)->nlookup++; return 0; } /* * Called for every fusefs vnode open to initialize the vnode (not * fuse_filehandle) for use */ void fuse_vnode_open(struct vnode *vp, int32_t fuse_open_flags, struct thread *td) { if (vnode_vtype(vp) == VREG) vnode_create_vobject(vp, 0, td); } int fuse_vnode_savesize(struct vnode *vp, struct ucred *cred, pid_t pid) { struct fuse_vnode_data *fvdat = VTOFUD(vp); struct thread *td = curthread; struct fuse_filehandle *fufh = NULL; struct fuse_dispatcher fdi; struct fuse_setattr_in *fsai; int err = 0; ASSERT_VOP_ELOCKED(vp, "fuse_io_extend"); if (fuse_isdeadfs(vp)) { return EBADF; } if (vnode_vtype(vp) == VDIR) { return EISDIR; } if (vfs_isrdonly(vnode_mount(vp))) { return EROFS; } if (cred == NULL) { cred = td->td_ucred; } fdisp_init(&fdi, sizeof(*fsai)); fdisp_make_vp(&fdi, FUSE_SETATTR, vp, td, cred); fsai = fdi.indata; fsai->valid = 0; /* Truncate to a new value. */ MPASS((fvdat->flag & FN_SIZECHANGE) != 0); fsai->size = fvdat->cached_attrs.va_size; fsai->valid |= FATTR_SIZE; fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid); if (fufh) { fsai->fh = fufh->fh_id; fsai->valid |= FATTR_FH; } err = fdisp_wait_answ(&fdi); fdisp_destroy(&fdi); if (err == 0) fvdat->flag &= ~FN_SIZECHANGE; return err; } /* * Adjust the vnode's size to a new value. * * If the new value came from the server, such as from a FUSE_GETATTR * operation, set `from_server` true. But if it came from a local operation, * such as write(2) or truncate(2), set `from_server` false. */ int fuse_vnode_setsize(struct vnode *vp, off_t newsize, bool from_server) { struct fuse_vnode_data *fvdat = VTOFUD(vp); struct vattr *attrs; off_t oldsize; size_t iosize; struct buf *bp = NULL; int err = 0; ASSERT_VOP_ELOCKED(vp, "fuse_vnode_setsize"); iosize = fuse_iosize(vp); oldsize = fvdat->cached_attrs.va_size; fvdat->cached_attrs.va_size = newsize; if ((attrs = VTOVA(vp)) != NULL) attrs->va_size = newsize; if (newsize < oldsize) { daddr_t lbn; err = vtruncbuf(vp, newsize, fuse_iosize(vp)); if (err) goto out; if (newsize % iosize == 0) goto out; /* * Zero the contents of the last partial block. * Sure seems like vtruncbuf should do this for us. */ lbn = newsize / iosize; bp = getblk(vp, lbn, iosize, PCATCH, 0, 0); if (!bp) { err = EINTR; goto out; } if (!(bp->b_flags & B_CACHE)) goto out; /* Nothing to do */ MPASS(bp->b_flags & B_VMIO); vfs_bio_clrbuf(bp); bp->b_dirtyend = MIN(bp->b_dirtyend, newsize - lbn * iosize); } else if (from_server && newsize > oldsize && oldsize != VNOVAL) { /* * The FUSE server changed the file size behind our back. We * should invalidate the entire cache. */ daddr_t left_lbn, end_lbn; left_lbn = oldsize / iosize; end_lbn = howmany(newsize, iosize); v_inval_buf_range(vp, 0, end_lbn, iosize); } out: if (bp) brelse(bp); vnode_pager_setsize(vp, newsize); return err; } /* Get the current, possibly dirty, size of the file */ int fuse_vnode_size(struct vnode *vp, off_t *filesize, struct ucred *cred, struct thread *td) { struct fuse_vnode_data *fvdat = VTOFUD(vp); int error = 0; if (!(fvdat->flag & FN_SIZECHANGE) && (!fuse_vnode_attr_cache_valid(vp) || fvdat->cached_attrs.va_size == VNOVAL)) error = fuse_internal_do_getattr(vp, NULL, cred, td); if (!error) *filesize = fvdat->cached_attrs.va_size; return error; } void fuse_vnode_undirty_cached_timestamps(struct vnode *vp) { struct fuse_vnode_data *fvdat = VTOFUD(vp); fvdat->flag &= ~(FN_MTIMECHANGE | FN_CTIMECHANGE); } /* Update a fuse file's cached timestamps */ void fuse_vnode_update(struct vnode *vp, int flags) { struct fuse_vnode_data *fvdat = VTOFUD(vp); struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp)); struct timespec ts; vfs_timestamp(&ts); if (data->time_gran > 1) ts.tv_nsec = rounddown(ts.tv_nsec, data->time_gran); if (flags & FN_MTIMECHANGE) fvdat->cached_attrs.va_mtime = ts; if (flags & FN_CTIMECHANGE) fvdat->cached_attrs.va_ctime = ts; fvdat->flag |= flags; } void fuse_node_init(void) { fuse_node_count = counter_u64_alloc(M_WAITOK); } void fuse_node_destroy(void) { counter_u64_free(fuse_node_count); } diff --git a/sys/fs/fuse/fuse_vnops.c b/sys/fs/fuse/fuse_vnops.c index 9aafbad990c5..e1c04f1583ba 100644 --- a/sys/fs/fuse/fuse_vnops.c +++ b/sys/fs/fuse/fuse_vnops.c @@ -1,2728 +1,2728 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 2007-2009 Google Inc. and Amit Singh * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions are * met: * * * Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * * Redistributions in binary form must reproduce the above * copyright notice, this list of conditions and the following disclaimer * in the documentation and/or other materials provided with the * distribution. * * Neither the name of Google Inc. nor the names of its * contributors may be used to endorse or promote products derived from * this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. * * Copyright (C) 2005 Csaba Henk. * All rights reserved. * * Copyright (c) 2019 The FreeBSD Foundation * * Portions of this software were developed by BFF Storage Systems, LLC under * sponsorship from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "fuse.h" #include "fuse_file.h" #include "fuse_internal.h" #include "fuse_ipc.h" #include "fuse_node.h" #include "fuse_io.h" #include /* Maximum number of hardlinks to a single FUSE file */ #define FUSE_LINK_MAX UINT32_MAX SDT_PROVIDER_DECLARE(fusefs); /* * Fuse trace probe: * arg0: verbosity. Higher numbers give more verbose messages * arg1: Textual message */ SDT_PROBE_DEFINE2(fusefs, , vnops, trace, "int", "char*"); /* vnode ops */ static vop_access_t fuse_vnop_access; static vop_advlock_t fuse_vnop_advlock; static vop_bmap_t fuse_vnop_bmap; static vop_close_t fuse_fifo_close; static vop_close_t fuse_vnop_close; static vop_copy_file_range_t fuse_vnop_copy_file_range; static vop_create_t fuse_vnop_create; static vop_deleteextattr_t fuse_vnop_deleteextattr; static vop_fdatasync_t fuse_vnop_fdatasync; static vop_fsync_t fuse_vnop_fsync; static vop_getattr_t fuse_vnop_getattr; static vop_getextattr_t fuse_vnop_getextattr; static vop_inactive_t fuse_vnop_inactive; static vop_ioctl_t fuse_vnop_ioctl; static vop_link_t fuse_vnop_link; static vop_listextattr_t fuse_vnop_listextattr; static vop_lookup_t fuse_vnop_lookup; static vop_mkdir_t fuse_vnop_mkdir; static vop_mknod_t fuse_vnop_mknod; static vop_open_t fuse_vnop_open; static vop_pathconf_t fuse_vnop_pathconf; static vop_read_t fuse_vnop_read; static vop_readdir_t fuse_vnop_readdir; static vop_readlink_t fuse_vnop_readlink; static vop_reclaim_t fuse_vnop_reclaim; static vop_remove_t fuse_vnop_remove; static vop_rename_t fuse_vnop_rename; static vop_rmdir_t fuse_vnop_rmdir; static vop_setattr_t fuse_vnop_setattr; static vop_setextattr_t fuse_vnop_setextattr; static vop_strategy_t fuse_vnop_strategy; static vop_symlink_t fuse_vnop_symlink; static vop_write_t fuse_vnop_write; static vop_getpages_t fuse_vnop_getpages; static vop_print_t fuse_vnop_print; static vop_vptofh_t fuse_vnop_vptofh; struct vop_vector fuse_fifoops = { .vop_default = &fifo_specops, .vop_access = fuse_vnop_access, .vop_close = fuse_fifo_close, .vop_fsync = fuse_vnop_fsync, .vop_getattr = fuse_vnop_getattr, .vop_inactive = fuse_vnop_inactive, .vop_pathconf = fuse_vnop_pathconf, .vop_print = fuse_vnop_print, .vop_read = VOP_PANIC, .vop_reclaim = fuse_vnop_reclaim, .vop_setattr = fuse_vnop_setattr, .vop_write = VOP_PANIC, .vop_vptofh = fuse_vnop_vptofh, }; VFS_VOP_VECTOR_REGISTER(fuse_fifoops); struct vop_vector fuse_vnops = { .vop_allocate = VOP_EINVAL, .vop_default = &default_vnodeops, .vop_access = fuse_vnop_access, .vop_advlock = fuse_vnop_advlock, .vop_bmap = fuse_vnop_bmap, .vop_close = fuse_vnop_close, .vop_copy_file_range = fuse_vnop_copy_file_range, .vop_create = fuse_vnop_create, .vop_deleteextattr = fuse_vnop_deleteextattr, .vop_fsync = fuse_vnop_fsync, .vop_fdatasync = fuse_vnop_fdatasync, .vop_getattr = fuse_vnop_getattr, .vop_getextattr = fuse_vnop_getextattr, .vop_inactive = fuse_vnop_inactive, .vop_ioctl = fuse_vnop_ioctl, .vop_link = fuse_vnop_link, .vop_listextattr = fuse_vnop_listextattr, .vop_lookup = fuse_vnop_lookup, .vop_mkdir = fuse_vnop_mkdir, .vop_mknod = fuse_vnop_mknod, .vop_open = fuse_vnop_open, .vop_pathconf = fuse_vnop_pathconf, /* * TODO: implement vop_poll after upgrading to protocol 7.21. * FUSE_POLL was added in protocol 7.11, but it's kind of broken until * 7.21, which adds the ability for the client to choose which poll * events it wants, and for a client to deregister a file handle */ .vop_read = fuse_vnop_read, .vop_readdir = fuse_vnop_readdir, .vop_readlink = fuse_vnop_readlink, .vop_reclaim = fuse_vnop_reclaim, .vop_remove = fuse_vnop_remove, .vop_rename = fuse_vnop_rename, .vop_rmdir = fuse_vnop_rmdir, .vop_setattr = fuse_vnop_setattr, .vop_setextattr = fuse_vnop_setextattr, .vop_strategy = fuse_vnop_strategy, .vop_symlink = fuse_vnop_symlink, .vop_write = fuse_vnop_write, .vop_getpages = fuse_vnop_getpages, .vop_print = fuse_vnop_print, .vop_vptofh = fuse_vnop_vptofh, }; VFS_VOP_VECTOR_REGISTER(fuse_vnops); uma_zone_t fuse_pbuf_zone; /* Check permission for extattr operations, much like extattr_check_cred */ static int fuse_extattr_check_cred(struct vnode *vp, int ns, struct ucred *cred, struct thread *td, accmode_t accmode) { struct mount *mp = vnode_mount(vp); struct fuse_data *data = fuse_get_mpdata(mp); int default_permissions = data->dataflags & FSESS_DEFAULT_PERMISSIONS; /* * Kernel-invoked always succeeds. */ if (cred == NOCRED) return (0); /* * Do not allow privileged processes in jail to directly manipulate * system attributes. */ switch (ns) { case EXTATTR_NAMESPACE_SYSTEM: if (default_permissions) { return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM)); } return (0); case EXTATTR_NAMESPACE_USER: if (default_permissions) { return (fuse_internal_access(vp, accmode, td, cred)); } return (0); default: return (EPERM); } } /* Get a filehandle for a directory */ static int fuse_filehandle_get_dir(struct vnode *vp, struct fuse_filehandle **fufhp, struct ucred *cred, pid_t pid) { if (fuse_filehandle_get(vp, FREAD, fufhp, cred, pid) == 0) return 0; return fuse_filehandle_get(vp, FEXEC, fufhp, cred, pid); } /* Send FUSE_FLUSH for this vnode */ static int fuse_flush(struct vnode *vp, struct ucred *cred, pid_t pid, int fflag) { struct fuse_flush_in *ffi; struct fuse_filehandle *fufh; struct fuse_dispatcher fdi; struct thread *td = curthread; struct mount *mp = vnode_mount(vp); int err; if (fsess_not_impl(vnode_mount(vp), FUSE_FLUSH)) return 0; err = fuse_filehandle_getrw(vp, fflag, &fufh, cred, pid); if (err) return err; fdisp_init(&fdi, sizeof(*ffi)); fdisp_make_vp(&fdi, FUSE_FLUSH, vp, td, cred); ffi = fdi.indata; ffi->fh = fufh->fh_id; /* * If the file has a POSIX lock then we're supposed to set lock_owner. * If not, then lock_owner is undefined. So we may as well always set * it. */ ffi->lock_owner = td->td_proc->p_pid; err = fdisp_wait_answ(&fdi); if (err == ENOSYS) { fsess_set_notimpl(mp, FUSE_FLUSH); err = 0; } fdisp_destroy(&fdi); return err; } /* Close wrapper for fifos. */ static int fuse_fifo_close(struct vop_close_args *ap) { return (fifo_specops.vop_close(ap)); } /* Send FUSE_LSEEK for this node */ static int fuse_vnop_do_lseek(struct vnode *vp, struct thread *td, struct ucred *cred, pid_t pid, off_t *offp, int whence) { struct fuse_dispatcher fdi; struct fuse_filehandle *fufh; struct fuse_lseek_in *flsi; struct fuse_lseek_out *flso; struct mount *mp = vnode_mount(vp); int err; ASSERT_VOP_LOCKED(vp, __func__); err = fuse_filehandle_getrw(vp, FREAD, &fufh, cred, pid); if (err) return (err); fdisp_init(&fdi, sizeof(*flsi)); fdisp_make_vp(&fdi, FUSE_LSEEK, vp, td, cred); flsi = fdi.indata; flsi->fh = fufh->fh_id; flsi->offset = *offp; flsi->whence = whence; err = fdisp_wait_answ(&fdi); if (err == ENOSYS) { fsess_set_notimpl(mp, FUSE_LSEEK); } else if (err == 0) { fsess_set_impl(mp, FUSE_LSEEK); flso = fdi.answ; *offp = flso->offset; } fdisp_destroy(&fdi); return (err); } /* struct vnop_access_args { struct vnode *a_vp; #if VOP_ACCESS_TAKES_ACCMODE_T accmode_t a_accmode; #else int a_mode; #endif struct ucred *a_cred; struct thread *a_td; }; */ static int fuse_vnop_access(struct vop_access_args *ap) { struct vnode *vp = ap->a_vp; int accmode = ap->a_accmode; struct ucred *cred = ap->a_cred; struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp)); int err; if (fuse_isdeadfs(vp)) { if (vnode_isvroot(vp)) { return 0; } return ENXIO; } if (!(data->dataflags & FSESS_INITED)) { if (vnode_isvroot(vp)) { if (priv_check_cred(cred, PRIV_VFS_ADMIN) || (fuse_match_cred(data->daemoncred, cred) == 0)) { return 0; } } return EBADF; } if (vnode_islnk(vp)) { return 0; } err = fuse_internal_access(vp, accmode, ap->a_td, ap->a_cred); return err; } /* * struct vop_advlock_args { * struct vop_generic_args a_gen; * struct vnode *a_vp; * void *a_id; * int a_op; * struct flock *a_fl; * int a_flags; * } */ static int fuse_vnop_advlock(struct vop_advlock_args *ap) { struct vnode *vp = ap->a_vp; struct flock *fl = ap->a_fl; struct thread *td = curthread; struct ucred *cred = td->td_ucred; pid_t pid = td->td_proc->p_pid; struct fuse_filehandle *fufh; struct fuse_dispatcher fdi; struct fuse_lk_in *fli; struct fuse_lk_out *flo; enum fuse_opcode op; int dataflags, err; int flags = ap->a_flags; dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags; if (fuse_isdeadfs(vp)) { return ENXIO; } switch(ap->a_op) { case F_GETLK: op = FUSE_GETLK; break; case F_SETLK: if (flags & F_WAIT) op = FUSE_SETLKW; else op = FUSE_SETLK; break; case F_UNLCK: op = FUSE_SETLK; break; default: return EINVAL; } if (!(dataflags & FSESS_POSIX_LOCKS)) return vop_stdadvlock(ap); /* FUSE doesn't properly support flock until protocol 7.17 */ if (flags & F_FLOCK) return vop_stdadvlock(ap); vn_lock(vp, LK_SHARED | LK_RETRY); err = fuse_filehandle_get_anyflags(vp, &fufh, cred, pid); if (err) goto out; fdisp_init(&fdi, sizeof(*fli)); fdisp_make_vp(&fdi, op, vp, td, cred); fli = fdi.indata; fli->fh = fufh->fh_id; fli->owner = td->td_proc->p_pid; fli->lk.start = fl->l_start; if (fl->l_len != 0) fli->lk.end = fl->l_start + fl->l_len - 1; else fli->lk.end = INT64_MAX; fli->lk.type = fl->l_type; fli->lk.pid = td->td_proc->p_pid; err = fdisp_wait_answ(&fdi); fdisp_destroy(&fdi); if (err == 0 && op == FUSE_GETLK) { flo = fdi.answ; fl->l_type = flo->lk.type; fl->l_pid = flo->lk.pid; if (flo->lk.type != F_UNLCK) { fl->l_start = flo->lk.start; if (flo->lk.end == INT64_MAX) fl->l_len = 0; else fl->l_len = flo->lk.end - flo->lk.start + 1; fl->l_start = flo->lk.start; } } out: VOP_UNLOCK(vp); return err; } /* { struct vnode *a_vp; daddr_t a_bn; struct bufobj **a_bop; daddr_t *a_bnp; int *a_runp; int *a_runb; } */ static int fuse_vnop_bmap(struct vop_bmap_args *ap) { struct vnode *vp = ap->a_vp; struct bufobj **bo = ap->a_bop; struct thread *td = curthread; struct mount *mp; struct fuse_dispatcher fdi; struct fuse_bmap_in *fbi; struct fuse_bmap_out *fbo; struct fuse_data *data; struct fuse_vnode_data *fvdat = VTOFUD(vp); uint64_t biosize; off_t fsize; daddr_t lbn = ap->a_bn; daddr_t *pbn = ap->a_bnp; int *runp = ap->a_runp; int *runb = ap->a_runb; int error = 0; int maxrun; if (fuse_isdeadfs(vp)) { return ENXIO; } mp = vnode_mount(vp); data = fuse_get_mpdata(mp); biosize = fuse_iosize(vp); maxrun = MIN(vp->v_mount->mnt_iosize_max / biosize - 1, data->max_readahead_blocks); if (bo != NULL) *bo = &vp->v_bufobj; /* * The FUSE_BMAP operation does not include the runp and runb * variables, so we must guess. Report nonzero contiguous runs so * cluster_read will combine adjacent reads. It's worthwhile to reduce * upcalls even if we don't know the true physical layout of the file. * * FUSE file systems may opt out of read clustering in two ways: * * mounting with -onoclusterr * * Setting max_readahead <= maxbcachebuf during FUSE_INIT */ if (runb != NULL) *runb = MIN(lbn, maxrun); if (runp != NULL && maxrun == 0) *runp = 0; else if (runp != NULL) { /* * If the file's size is cached, use that value to calculate * runp, even if the cache is expired. runp is only advisory, * and the risk of getting it wrong is not worth the cost of * another upcall. */ if (fvdat->cached_attrs.va_size != VNOVAL) fsize = fvdat->cached_attrs.va_size; else error = fuse_vnode_size(vp, &fsize, td->td_ucred, td); if (error == 0) *runp = MIN(MAX(0, fsize / (off_t)biosize - lbn - 1), maxrun); else *runp = 0; } if (fsess_maybe_impl(mp, FUSE_BMAP)) { fdisp_init(&fdi, sizeof(*fbi)); fdisp_make_vp(&fdi, FUSE_BMAP, vp, td, td->td_ucred); fbi = fdi.indata; fbi->block = lbn; fbi->blocksize = biosize; error = fdisp_wait_answ(&fdi); if (error == ENOSYS) { fdisp_destroy(&fdi); fsess_set_notimpl(mp, FUSE_BMAP); error = 0; } else { fbo = fdi.answ; if (error == 0 && pbn != NULL) *pbn = fbo->block; fdisp_destroy(&fdi); return error; } } /* If the daemon doesn't support BMAP, make up a sensible default */ if (pbn != NULL) *pbn = lbn * btodb(biosize); return (error); } /* struct vop_close_args { struct vnode *a_vp; int a_fflag; struct ucred *a_cred; struct thread *a_td; }; */ static int fuse_vnop_close(struct vop_close_args *ap) { struct vnode *vp = ap->a_vp; struct ucred *cred = ap->a_cred; int fflag = ap->a_fflag; struct thread *td = ap->a_td; pid_t pid = td->td_proc->p_pid; int err = 0; if (fuse_isdeadfs(vp)) return 0; if (vnode_isdir(vp)) return 0; if (fflag & IO_NDELAY) return 0; err = fuse_flush(vp, cred, pid, fflag); /* TODO: close the file handle, if we're sure it's no longer used */ if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) { fuse_vnode_savesize(vp, cred, td->td_proc->p_pid); } return err; } /* struct vop_copy_file_range_args { struct vop_generic_args a_gen; struct vnode *a_invp; off_t *a_inoffp; struct vnode *a_outvp; off_t *a_outoffp; size_t *a_lenp; unsigned int a_flags; struct ucred *a_incred; struct ucred *a_outcred; struct thread *a_fsizetd; } */ static int fuse_vnop_copy_file_range(struct vop_copy_file_range_args *ap) { struct vnode *invp = ap->a_invp; struct vnode *outvp = ap->a_outvp; struct mount *mp = vnode_mount(invp); struct fuse_dispatcher fdi; struct fuse_filehandle *infufh, *outfufh; struct fuse_copy_file_range_in *fcfri; struct ucred *incred = ap->a_incred; struct ucred *outcred = ap->a_outcred; struct fuse_write_out *fwo; struct thread *td; struct uio io; pid_t pid; int err; if (mp != vnode_mount(outvp)) goto fallback; if (incred->cr_uid != outcred->cr_uid) goto fallback; if (incred->cr_groups[0] != outcred->cr_groups[0]) goto fallback; if (fsess_not_impl(mp, FUSE_COPY_FILE_RANGE)) goto fallback; if (ap->a_fsizetd == NULL) td = curthread; else td = ap->a_fsizetd; pid = td->td_proc->p_pid; /* Lock both vnodes, avoiding risk of deadlock. */ do { err = vn_lock(outvp, LK_EXCLUSIVE); if (invp == outvp) break; if (err == 0) { err = vn_lock(invp, LK_SHARED | LK_NOWAIT); if (err == 0) break; VOP_UNLOCK(outvp); err = vn_lock(invp, LK_SHARED); if (err == 0) VOP_UNLOCK(invp); } } while (err == 0); if (err != 0) return (err); err = fuse_filehandle_getrw(invp, FREAD, &infufh, incred, pid); if (err) goto unlock; err = fuse_filehandle_getrw(outvp, FWRITE, &outfufh, outcred, pid); if (err) goto unlock; if (ap->a_fsizetd) { io.uio_offset = *ap->a_outoffp; io.uio_resid = *ap->a_lenp; err = vn_rlimit_fsize(outvp, &io, ap->a_fsizetd); if (err) goto unlock; } fdisp_init(&fdi, sizeof(*fcfri)); fdisp_make_vp(&fdi, FUSE_COPY_FILE_RANGE, invp, td, incred); fcfri = fdi.indata; fcfri->fh_in = infufh->fh_id; fcfri->off_in = *ap->a_inoffp; fcfri->nodeid_out = VTOI(outvp); fcfri->fh_out = outfufh->fh_id; fcfri->off_out = *ap->a_outoffp; fcfri->len = *ap->a_lenp; fcfri->flags = 0; err = fdisp_wait_answ(&fdi); if (err == 0) { fwo = fdi.answ; *ap->a_lenp = fwo->size; *ap->a_inoffp += fwo->size; *ap->a_outoffp += fwo->size; fuse_internal_clear_suid_on_write(outvp, outcred, td); } fdisp_destroy(&fdi); unlock: if (invp != outvp) VOP_UNLOCK(invp); VOP_UNLOCK(outvp); if (err == ENOSYS) { fsess_set_notimpl(mp, FUSE_COPY_FILE_RANGE); fallback: err = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp, ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags, ap->a_incred, ap->a_outcred, ap->a_fsizetd); } return (err); } static void fdisp_make_mknod_for_fallback( struct fuse_dispatcher *fdip, struct componentname *cnp, struct vnode *dvp, uint64_t parentnid, struct thread *td, struct ucred *cred, mode_t mode, enum fuse_opcode *op) { struct fuse_mknod_in *fmni; fdisp_init(fdip, sizeof(*fmni) + cnp->cn_namelen + 1); *op = FUSE_MKNOD; fdisp_make(fdip, *op, vnode_mount(dvp), parentnid, td, cred); fmni = fdip->indata; fmni->mode = mode; fmni->rdev = 0; memcpy((char *)fdip->indata + sizeof(*fmni), cnp->cn_nameptr, cnp->cn_namelen); ((char *)fdip->indata)[sizeof(*fmni) + cnp->cn_namelen] = '\0'; } /* struct vnop_create_args { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; struct vattr *a_vap; }; */ static int fuse_vnop_create(struct vop_create_args *ap) { struct vnode *dvp = ap->a_dvp; struct vnode **vpp = ap->a_vpp; struct componentname *cnp = ap->a_cnp; struct vattr *vap = ap->a_vap; - struct thread *td = cnp->cn_thread; + struct thread *td = curthread; struct ucred *cred = cnp->cn_cred; struct fuse_data *data; struct fuse_create_in *fci; struct fuse_entry_out *feo; struct fuse_open_out *foo; struct fuse_dispatcher fdi, fdi2; struct fuse_dispatcher *fdip = &fdi; struct fuse_dispatcher *fdip2 = NULL; int err; struct mount *mp = vnode_mount(dvp); data = fuse_get_mpdata(mp); uint64_t parentnid = VTOFUD(dvp)->nid; mode_t mode = MAKEIMODE(vap->va_type, vap->va_mode); enum fuse_opcode op; int flags; if (fuse_isdeadfs(dvp)) return ENXIO; /* FUSE expects sockets to be created with FUSE_MKNOD */ if (vap->va_type == VSOCK) return fuse_internal_mknod(dvp, vpp, cnp, vap); /* * VOP_CREATE doesn't tell us the open(2) flags, so we guess. Only a * writable mode makes sense, and we might as well include readability * too. */ flags = O_RDWR; bzero(&fdi, sizeof(fdi)); if (vap->va_type != VREG) return (EINVAL); if (fsess_not_impl(mp, FUSE_CREATE) || vap->va_type == VSOCK) { /* Fallback to FUSE_MKNOD/FUSE_OPEN */ fdisp_make_mknod_for_fallback(fdip, cnp, dvp, parentnid, td, cred, mode, &op); } else { /* Use FUSE_CREATE */ size_t insize; op = FUSE_CREATE; fdisp_init(fdip, sizeof(*fci) + cnp->cn_namelen + 1); fdisp_make(fdip, op, vnode_mount(dvp), parentnid, td, cred); fci = fdip->indata; fci->mode = mode; fci->flags = O_CREAT | flags; if (fuse_libabi_geq(data, 7, 12)) { insize = sizeof(*fci); fci->umask = td->td_proc->p_pd->pd_cmask; } else { insize = sizeof(struct fuse_open_in); } memcpy((char *)fdip->indata + insize, cnp->cn_nameptr, cnp->cn_namelen); ((char *)fdip->indata)[insize + cnp->cn_namelen] = '\0'; } err = fdisp_wait_answ(fdip); if (err) { if (err == ENOSYS && op == FUSE_CREATE) { fsess_set_notimpl(mp, FUSE_CREATE); fdisp_destroy(fdip); fdisp_make_mknod_for_fallback(fdip, cnp, dvp, parentnid, td, cred, mode, &op); err = fdisp_wait_answ(fdip); } if (err) goto out; } feo = fdip->answ; if ((err = fuse_internal_checkentry(feo, vap->va_type))) { goto out; } if (op == FUSE_CREATE) { foo = (struct fuse_open_out*)(feo + 1); } else { /* Issue a separate FUSE_OPEN */ struct fuse_open_in *foi; fdip2 = &fdi2; fdisp_init(fdip2, sizeof(*foi)); fdisp_make(fdip2, FUSE_OPEN, vnode_mount(dvp), feo->nodeid, td, cred); foi = fdip2->indata; foi->flags = flags; err = fdisp_wait_answ(fdip2); if (err) goto out; foo = fdip2->answ; } err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vap->va_type); if (err) { struct fuse_release_in *fri; uint64_t nodeid = feo->nodeid; uint64_t fh_id = foo->fh; fdisp_init(fdip, sizeof(*fri)); fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred); fri = fdip->indata; fri->fh = fh_id; fri->flags = flags; fuse_insert_callback(fdip->tick, fuse_internal_forget_callback); fuse_insert_message(fdip->tick, false); goto out; } ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create"); fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid, feo->attr_valid_nsec, NULL, true); fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, td, cred, foo); fuse_vnode_open(*vpp, foo->open_flags, td); /* * Purge the parent's attribute cache because the daemon should've * updated its mtime and ctime */ fuse_vnode_clear_attr_cache(dvp); cache_purge_negative(dvp); out: if (fdip2) fdisp_destroy(fdip2); fdisp_destroy(fdip); return err; } /* struct vnop_fdatasync_args { struct vop_generic_args a_gen; struct vnode * a_vp; struct thread * a_td; }; */ static int fuse_vnop_fdatasync(struct vop_fdatasync_args *ap) { struct vnode *vp = ap->a_vp; struct thread *td = ap->a_td; int waitfor = MNT_WAIT; int err = 0; if (fuse_isdeadfs(vp)) { return 0; } if ((err = vop_stdfdatasync_buf(ap))) return err; return fuse_internal_fsync(vp, td, waitfor, true); } /* struct vnop_fsync_args { struct vop_generic_args a_gen; struct vnode * a_vp; int a_waitfor; struct thread * a_td; }; */ static int fuse_vnop_fsync(struct vop_fsync_args *ap) { struct vnode *vp = ap->a_vp; struct thread *td = ap->a_td; int waitfor = ap->a_waitfor; int err = 0; if (fuse_isdeadfs(vp)) { return 0; } if ((err = vop_stdfsync(ap))) return err; return fuse_internal_fsync(vp, td, waitfor, false); } /* struct vnop_getattr_args { struct vnode *a_vp; struct vattr *a_vap; struct ucred *a_cred; struct thread *a_td; }; */ static int fuse_vnop_getattr(struct vop_getattr_args *ap) { struct vnode *vp = ap->a_vp; struct vattr *vap = ap->a_vap; struct ucred *cred = ap->a_cred; struct thread *td = curthread; int err = 0; int dataflags; dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags; /* Note that we are not bailing out on a dead file system just yet. */ if (!(dataflags & FSESS_INITED)) { if (!vnode_isvroot(vp)) { fdata_set_dead(fuse_get_mpdata(vnode_mount(vp))); err = ENOTCONN; return err; } else { goto fake; } } err = fuse_internal_getattr(vp, vap, cred, td); if (err == ENOTCONN && vnode_isvroot(vp)) { /* see comment in fuse_vfsop_statfs() */ goto fake; } else { return err; } fake: bzero(vap, sizeof(*vap)); vap->va_type = vnode_vtype(vp); return 0; } /* struct vnop_inactive_args { struct vnode *a_vp; }; */ static int fuse_vnop_inactive(struct vop_inactive_args *ap) { struct vnode *vp = ap->a_vp; struct thread *td = curthread; struct fuse_vnode_data *fvdat = VTOFUD(vp); struct fuse_filehandle *fufh, *fufh_tmp; int need_flush = 1; LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) { if (need_flush && vp->v_type == VREG) { if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) { fuse_vnode_savesize(vp, NULL, 0); } if ((fvdat->flag & FN_REVOKED) != 0) fuse_io_invalbuf(vp, td); else fuse_io_flushbuf(vp, MNT_WAIT, td); need_flush = 0; } fuse_filehandle_close(vp, fufh, td, NULL); } if ((fvdat->flag & FN_REVOKED) != 0) vrecycle(vp); return 0; } /* struct vnop_ioctl_args { struct vnode *a_vp; u_long a_command; caddr_t a_data; int a_fflag; struct ucred *a_cred; struct thread *a_td; }; */ static int fuse_vnop_ioctl(struct vop_ioctl_args *ap) { struct vnode *vp = ap->a_vp; struct mount *mp = vnode_mount(vp); struct ucred *cred = ap->a_cred; off_t *offp; pid_t pid = ap->a_td->td_proc->p_pid; int err; switch (ap->a_command) { case FIOSEEKDATA: case FIOSEEKHOLE: /* Call FUSE_LSEEK, if we can, or fall back to vop_stdioctl */ if (fsess_maybe_impl(mp, FUSE_LSEEK)) { int whence; offp = ap->a_data; if (ap->a_command == FIOSEEKDATA) whence = SEEK_DATA; else whence = SEEK_HOLE; vn_lock(vp, LK_SHARED | LK_RETRY); err = fuse_vnop_do_lseek(vp, ap->a_td, cred, pid, offp, whence); VOP_UNLOCK(vp); } if (fsess_not_impl(mp, FUSE_LSEEK)) err = vop_stdioctl(ap); break; default: /* TODO: implement FUSE_IOCTL */ err = ENOTTY; break; } return (err); } /* struct vnop_link_args { struct vnode *a_tdvp; struct vnode *a_vp; struct componentname *a_cnp; }; */ static int fuse_vnop_link(struct vop_link_args *ap) { struct vnode *vp = ap->a_vp; struct vnode *tdvp = ap->a_tdvp; struct componentname *cnp = ap->a_cnp; struct vattr *vap = VTOVA(vp); struct fuse_dispatcher fdi; struct fuse_entry_out *feo; struct fuse_link_in fli; int err; if (fuse_isdeadfs(vp)) { return ENXIO; } if (vnode_mount(tdvp) != vnode_mount(vp)) { return EXDEV; } /* * This is a seatbelt check to protect naive userspace filesystems from * themselves and the limitations of the FUSE IPC protocol. If a * filesystem does not allow attribute caching, assume it is capable of * validating that nlink does not overflow. */ if (vap != NULL && vap->va_nlink >= FUSE_LINK_MAX) return EMLINK; fli.oldnodeid = VTOI(vp); fdisp_init(&fdi, 0); fuse_internal_newentry_makerequest(vnode_mount(tdvp), VTOI(tdvp), cnp, FUSE_LINK, &fli, sizeof(fli), &fdi); if ((err = fdisp_wait_answ(&fdi))) { goto out; } feo = fdi.answ; err = fuse_internal_checkentry(feo, vnode_vtype(vp)); if (!err) { /* * Purge the parent's attribute cache because the daemon * should've updated its mtime and ctime */ fuse_vnode_clear_attr_cache(tdvp); fuse_internal_cache_attrs(vp, &feo->attr, feo->attr_valid, feo->attr_valid_nsec, NULL, true); } out: fdisp_destroy(&fdi); return err; } struct fuse_lookup_alloc_arg { struct fuse_entry_out *feo; struct componentname *cnp; uint64_t nid; enum vtype vtyp; }; /* Callback for vn_get_ino */ static int fuse_lookup_alloc(struct mount *mp, void *arg, int lkflags, struct vnode **vpp) { struct fuse_lookup_alloc_arg *flaa = arg; return fuse_vnode_get(mp, flaa->feo, flaa->nid, NULL, vpp, flaa->cnp, flaa->vtyp); } SDT_PROBE_DEFINE3(fusefs, , vnops, cache_lookup, "int", "struct timespec*", "struct timespec*"); /* struct vnop_lookup_args { struct vnodeop_desc *a_desc; struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; }; */ int fuse_vnop_lookup(struct vop_lookup_args *ap) { struct vnode *dvp = ap->a_dvp; struct vnode **vpp = ap->a_vpp; struct componentname *cnp = ap->a_cnp; - struct thread *td = cnp->cn_thread; + struct thread *td = curthread; struct ucred *cred = cnp->cn_cred; int nameiop = cnp->cn_nameiop; int flags = cnp->cn_flags; int wantparent = flags & (LOCKPARENT | WANTPARENT); int islastcn = flags & ISLASTCN; struct mount *mp = vnode_mount(dvp); struct fuse_data *data = fuse_get_mpdata(mp); int default_permissions = data->dataflags & FSESS_DEFAULT_PERMISSIONS; int err = 0; int lookup_err = 0; struct vnode *vp = NULL; struct fuse_dispatcher fdi; bool did_lookup = false; struct fuse_entry_out *feo = NULL; enum vtype vtyp; /* vnode type of target */ off_t filesize; /* filesize of target */ uint64_t nid; if (fuse_isdeadfs(dvp)) { *vpp = NULL; return ENXIO; } if (!vnode_isdir(dvp)) return ENOTDIR; if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP)) return EROFS; if ((cnp->cn_flags & NOEXECCHECK) != 0) cnp->cn_flags &= ~NOEXECCHECK; else if ((err = fuse_internal_access(dvp, VEXEC, td, cred))) return err; if (flags & ISDOTDOT) { KASSERT(VTOFUD(dvp)->flag & FN_PARENT_NID, ("Looking up .. is TODO")); nid = VTOFUD(dvp)->parent_nid; if (nid == 0) return ENOENT; /* .. is obviously a directory */ vtyp = VDIR; filesize = 0; } else if (cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.') { nid = VTOI(dvp); /* . is obviously a directory */ vtyp = VDIR; filesize = 0; } else { struct timespec now, timeout; int ncpticks; /* here to accomodate for API contract */ err = cache_lookup(dvp, vpp, cnp, &timeout, &ncpticks); getnanouptime(&now); SDT_PROBE3(fusefs, , vnops, cache_lookup, err, &timeout, &now); switch (err) { case -1: /* positive match */ if (timespeccmp(&timeout, &now, >)) { counter_u64_add(fuse_lookup_cache_hits, 1); } else { /* Cache timeout */ counter_u64_add(fuse_lookup_cache_misses, 1); bintime_clear( &VTOFUD(*vpp)->entry_cache_timeout); cache_purge(*vpp); if (dvp != *vpp) vput(*vpp); else vrele(*vpp); *vpp = NULL; break; } return 0; case 0: /* no match in cache */ counter_u64_add(fuse_lookup_cache_misses, 1); break; case ENOENT: /* negative match */ getnanouptime(&now); if (timespeccmp(&timeout, &now, <=)) { /* Cache timeout */ cache_purge_negative(dvp); break; } /* fall through */ default: return err; } nid = VTOI(dvp); fdisp_init(&fdi, cnp->cn_namelen + 1); fdisp_make(&fdi, FUSE_LOOKUP, mp, nid, td, cred); memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen); ((char *)fdi.indata)[cnp->cn_namelen] = '\0'; lookup_err = fdisp_wait_answ(&fdi); did_lookup = true; if (!lookup_err) { /* lookup call succeeded */ feo = (struct fuse_entry_out *)fdi.answ; nid = feo->nodeid; if (nid == 0) { /* zero nodeid means ENOENT and cache it */ struct timespec timeout; fdi.answ_stat = ENOENT; lookup_err = ENOENT; if (cnp->cn_flags & MAKEENTRY) { fuse_validity_2_timespec(feo, &timeout); cache_enter_time(dvp, *vpp, cnp, &timeout, NULL); } } else if (nid == FUSE_ROOT_ID) { lookup_err = EINVAL; } vtyp = IFTOVT(feo->attr.mode); filesize = feo->attr.size; } if (lookup_err && (!fdi.answ_stat || lookup_err != ENOENT)) { fdisp_destroy(&fdi); return lookup_err; } } /* lookup_err, if non-zero, must be ENOENT at this point */ if (lookup_err) { /* Entry not found */ if ((nameiop == CREATE || nameiop == RENAME) && islastcn) { if (default_permissions) err = fuse_internal_access(dvp, VWRITE, td, cred); else err = 0; if (!err) { /* * Set the SAVENAME flag to hold onto the * pathname for use later in VOP_CREATE or * VOP_RENAME. */ cnp->cn_flags |= SAVENAME; err = EJUSTRETURN; } } else { err = ENOENT; } } else { /* Entry was found */ if (flags & ISDOTDOT) { struct fuse_lookup_alloc_arg flaa; flaa.nid = nid; flaa.feo = feo; flaa.cnp = cnp; flaa.vtyp = vtyp; err = vn_vget_ino_gen(dvp, fuse_lookup_alloc, &flaa, 0, &vp); *vpp = vp; } else if (nid == VTOI(dvp)) { vref(dvp); *vpp = dvp; } else { struct fuse_vnode_data *fvdat; err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp, &vp, cnp, vtyp); if (err) goto out; *vpp = vp; fvdat = VTOFUD(vp); MPASS(feo != NULL); fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid, feo->attr_valid_nsec, NULL, true); fuse_validity_2_bintime(feo->entry_valid, feo->entry_valid_nsec, &fvdat->entry_cache_timeout); if ((nameiop == DELETE || nameiop == RENAME) && islastcn && default_permissions) { struct vattr dvattr; err = fuse_internal_access(dvp, VWRITE, td, cred); if (err != 0) goto out; /* * if the parent's sticky bit is set, check * whether we're allowed to remove the file. * Need to figure out the vnode locking to make * this work. */ fuse_internal_getattr(dvp, &dvattr, cred, td); if ((dvattr.va_mode & S_ISTXT) && fuse_internal_access(dvp, VADMIN, td, cred) && fuse_internal_access(*vpp, VADMIN, td, cred)) { err = EPERM; goto out; } } if (islastcn && ( (nameiop == DELETE) || (nameiop == RENAME && wantparent))) { cnp->cn_flags |= SAVENAME; } } } out: if (err) { if (vp != NULL && dvp != vp) vput(vp); else if (vp != NULL) vrele(vp); *vpp = NULL; } if (did_lookup) fdisp_destroy(&fdi); return err; } /* struct vnop_mkdir_args { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; struct vattr *a_vap; }; */ static int fuse_vnop_mkdir(struct vop_mkdir_args *ap) { struct vnode *dvp = ap->a_dvp; struct vnode **vpp = ap->a_vpp; struct componentname *cnp = ap->a_cnp; struct vattr *vap = ap->a_vap; struct fuse_mkdir_in fmdi; if (fuse_isdeadfs(dvp)) { return ENXIO; } fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode); fmdi.umask = curthread->td_proc->p_pd->pd_cmask; return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi, sizeof(fmdi), VDIR)); } /* struct vnop_mknod_args { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; struct vattr *a_vap; }; */ static int fuse_vnop_mknod(struct vop_mknod_args *ap) { struct vnode *dvp = ap->a_dvp; struct vnode **vpp = ap->a_vpp; struct componentname *cnp = ap->a_cnp; struct vattr *vap = ap->a_vap; if (fuse_isdeadfs(dvp)) return ENXIO; return fuse_internal_mknod(dvp, vpp, cnp, vap); } /* struct vop_open_args { struct vnode *a_vp; int a_mode; struct ucred *a_cred; struct thread *a_td; int a_fdidx; / struct file *a_fp; }; */ static int fuse_vnop_open(struct vop_open_args *ap) { struct vnode *vp = ap->a_vp; int a_mode = ap->a_mode; struct thread *td = ap->a_td; struct ucred *cred = ap->a_cred; pid_t pid = td->td_proc->p_pid; struct fuse_vnode_data *fvdat; if (fuse_isdeadfs(vp)) return ENXIO; if (vp->v_type == VCHR || vp->v_type == VBLK || vp->v_type == VFIFO) return (EOPNOTSUPP); if ((a_mode & (FREAD | FWRITE | FEXEC)) == 0) return EINVAL; fvdat = VTOFUD(vp); if (fuse_filehandle_validrw(vp, a_mode, cred, pid)) { fuse_vnode_open(vp, 0, td); return 0; } return fuse_filehandle_open(vp, a_mode, NULL, td, cred); } static int fuse_vnop_pathconf(struct vop_pathconf_args *ap) { struct vnode *vp = ap->a_vp; struct mount *mp; switch (ap->a_name) { case _PC_FILESIZEBITS: *ap->a_retval = 64; return (0); case _PC_NAME_MAX: *ap->a_retval = NAME_MAX; return (0); case _PC_LINK_MAX: *ap->a_retval = MIN(LONG_MAX, FUSE_LINK_MAX); return (0); case _PC_SYMLINK_MAX: *ap->a_retval = MAXPATHLEN; return (0); case _PC_NO_TRUNC: *ap->a_retval = 1; return (0); case _PC_MIN_HOLE_SIZE: /* * The FUSE protocol provides no mechanism for a server to * report _PC_MIN_HOLE_SIZE. It's a protocol bug. Instead, * return EINVAL if the server does not support FUSE_LSEEK, or * 1 if it does. */ mp = vnode_mount(vp); if (!fsess_is_impl(mp, FUSE_LSEEK) && !fsess_not_impl(mp, FUSE_LSEEK)) { off_t offset = 0; /* Issue a FUSE_LSEEK to find out if it's implemented */ fuse_vnop_do_lseek(vp, curthread, curthread->td_ucred, curthread->td_proc->p_pid, &offset, SEEK_DATA); } if (fsess_is_impl(mp, FUSE_LSEEK)) { *ap->a_retval = 1; return (0); } else { /* * Probably FUSE_LSEEK is not implemented. It might * be, if the FUSE_LSEEK above returned an error like * EACCES, but in that case we can't tell, so it's * safest to report EINVAL anyway. */ return (EINVAL); } default: return (vop_stdpathconf(ap)); } } /* struct vnop_read_args { struct vnode *a_vp; struct uio *a_uio; int a_ioflag; struct ucred *a_cred; }; */ static int fuse_vnop_read(struct vop_read_args *ap) { struct vnode *vp = ap->a_vp; struct uio *uio = ap->a_uio; int ioflag = ap->a_ioflag; struct ucred *cred = ap->a_cred; pid_t pid = curthread->td_proc->p_pid; if (fuse_isdeadfs(vp)) { return ENXIO; } if (VTOFUD(vp)->flag & FN_DIRECTIO) { ioflag |= IO_DIRECT; } return fuse_io_dispatch(vp, uio, ioflag, cred, pid); } /* struct vnop_readdir_args { struct vnode *a_vp; struct uio *a_uio; struct ucred *a_cred; int *a_eofflag; int *a_ncookies; u_long **a_cookies; }; */ static int fuse_vnop_readdir(struct vop_readdir_args *ap) { struct vnode *vp = ap->a_vp; struct uio *uio = ap->a_uio; struct ucred *cred = ap->a_cred; struct fuse_filehandle *fufh = NULL; struct fuse_iov cookediov; int err = 0; u_long *cookies; off_t startoff; ssize_t tresid; int ncookies; bool closefufh = false; pid_t pid = curthread->td_proc->p_pid; if (ap->a_eofflag) *ap->a_eofflag = 0; if (fuse_isdeadfs(vp)) { return ENXIO; } if ( /* XXXIP ((uio_iovcnt(uio) > 1)) || */ (uio_resid(uio) < sizeof(struct dirent))) { return EINVAL; } tresid = uio->uio_resid; startoff = uio->uio_offset; err = fuse_filehandle_get_dir(vp, &fufh, cred, pid); if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) { /* * nfsd will do VOP_READDIR without first doing VOP_OPEN. We * must implicitly open the directory here */ err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred); if (err == 0) { /* * When a directory is opened, it must be read from * the beginning. Hopefully, the "startoff" still * exists as an offset cookie for the directory. * If not, it will read the entire directory without * returning any entries and just return eof. */ uio->uio_offset = 0; } closefufh = true; } if (err) return (err); if (ap->a_ncookies != NULL) { ncookies = uio->uio_resid / (offsetof(struct dirent, d_name) + 4) + 1; cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK); *ap->a_ncookies = ncookies; *ap->a_cookies = cookies; } else { ncookies = 0; cookies = NULL; } #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1) fiov_init(&cookediov, DIRCOOKEDSIZE); err = fuse_internal_readdir(vp, uio, startoff, fufh, &cookediov, &ncookies, cookies); fiov_teardown(&cookediov); if (closefufh) fuse_filehandle_close(vp, fufh, curthread, cred); if (ap->a_ncookies != NULL) { if (err == 0) { *ap->a_ncookies -= ncookies; } else { free(*ap->a_cookies, M_TEMP); *ap->a_ncookies = 0; *ap->a_cookies = NULL; } } if (err == 0 && tresid == uio->uio_resid) *ap->a_eofflag = 1; return err; } /* struct vnop_readlink_args { struct vnode *a_vp; struct uio *a_uio; struct ucred *a_cred; }; */ static int fuse_vnop_readlink(struct vop_readlink_args *ap) { struct vnode *vp = ap->a_vp; struct uio *uio = ap->a_uio; struct ucred *cred = ap->a_cred; struct fuse_dispatcher fdi; int err; if (fuse_isdeadfs(vp)) { return ENXIO; } if (!vnode_islnk(vp)) { return EINVAL; } fdisp_init(&fdi, 0); err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred); if (err) { goto out; } if (((char *)fdi.answ)[0] == '/' && fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) { char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname; err = uiomove(mpth, strlen(mpth), uio); } if (!err) { err = uiomove(fdi.answ, fdi.iosize, uio); } out: fdisp_destroy(&fdi); return err; } /* struct vnop_reclaim_args { struct vnode *a_vp; }; */ static int fuse_vnop_reclaim(struct vop_reclaim_args *ap) { struct vnode *vp = ap->a_vp; struct thread *td = curthread; struct fuse_vnode_data *fvdat = VTOFUD(vp); struct fuse_filehandle *fufh, *fufh_tmp; if (!fvdat) { panic("FUSE: no vnode data during recycling"); } LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) { printf("FUSE: vnode being reclaimed with open fufh " "(type=%#x)", fufh->fufh_type); fuse_filehandle_close(vp, fufh, td, NULL); } if (!fuse_isdeadfs(vp) && fvdat->nlookup > 0) { fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp), fvdat->nlookup); } cache_purge(vp); vfs_hash_remove(vp); fuse_vnode_destroy(vp); return 0; } /* struct vnop_remove_args { struct vnode *a_dvp; struct vnode *a_vp; struct componentname *a_cnp; }; */ static int fuse_vnop_remove(struct vop_remove_args *ap) { struct vnode *dvp = ap->a_dvp; struct vnode *vp = ap->a_vp; struct componentname *cnp = ap->a_cnp; int err; if (fuse_isdeadfs(vp)) { return ENXIO; } if (vnode_isdir(vp)) { return EPERM; } err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK); return err; } /* struct vnop_rename_args { struct vnode *a_fdvp; struct vnode *a_fvp; struct componentname *a_fcnp; struct vnode *a_tdvp; struct vnode *a_tvp; struct componentname *a_tcnp; }; */ static int fuse_vnop_rename(struct vop_rename_args *ap) { struct vnode *fdvp = ap->a_fdvp; struct vnode *fvp = ap->a_fvp; struct componentname *fcnp = ap->a_fcnp; struct vnode *tdvp = ap->a_tdvp; struct vnode *tvp = ap->a_tvp; struct componentname *tcnp = ap->a_tcnp; struct fuse_data *data; bool newparent = fdvp != tdvp; bool isdir = fvp->v_type == VDIR; int err = 0; if (fuse_isdeadfs(fdvp)) { return ENXIO; } if (fvp->v_mount != tdvp->v_mount || (tvp && fvp->v_mount != tvp->v_mount)) { SDT_PROBE2(fusefs, , vnops, trace, 1, "cross-device rename"); err = EXDEV; goto out; } cache_purge(fvp); /* * FUSE library is expected to check if target directory is not * under the source directory in the file system tree. * Linux performs this check at VFS level. */ /* * If source is a directory, and it will get a new parent, user must * have write permission to it, so ".." can be modified. */ data = fuse_get_mpdata(vnode_mount(tdvp)); if (data->dataflags & FSESS_DEFAULT_PERMISSIONS && isdir && newparent) { err = fuse_internal_access(fvp, VWRITE, - tcnp->cn_thread, tcnp->cn_cred); + curthread, tcnp->cn_cred); if (err) goto out; } sx_xlock(&data->rename_lock); err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp); if (err == 0) { if (tdvp != fdvp) fuse_vnode_setparent(fvp, tdvp); if (tvp != NULL) fuse_vnode_setparent(tvp, NULL); } sx_unlock(&data->rename_lock); if (tvp != NULL && tvp != fvp) { cache_purge(tvp); } if (vnode_isdir(fvp)) { if ((tvp != NULL) && vnode_isdir(tvp)) { cache_purge(tdvp); } cache_purge(fdvp); } out: if (tdvp == tvp) { vrele(tdvp); } else { vput(tdvp); } if (tvp != NULL) { vput(tvp); } vrele(fdvp); vrele(fvp); return err; } /* struct vnop_rmdir_args { struct vnode *a_dvp; struct vnode *a_vp; struct componentname *a_cnp; } *ap; */ static int fuse_vnop_rmdir(struct vop_rmdir_args *ap) { struct vnode *dvp = ap->a_dvp; struct vnode *vp = ap->a_vp; int err; if (fuse_isdeadfs(vp)) { return ENXIO; } if (VTOFUD(vp) == VTOFUD(dvp)) { return EINVAL; } err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR); return err; } /* struct vnop_setattr_args { struct vnode *a_vp; struct vattr *a_vap; struct ucred *a_cred; struct thread *a_td; }; */ static int fuse_vnop_setattr(struct vop_setattr_args *ap) { struct vnode *vp = ap->a_vp; struct vattr *vap = ap->a_vap; struct ucred *cred = ap->a_cred; struct thread *td = curthread; struct mount *mp; struct fuse_data *data; struct vattr old_va; int dataflags; int err = 0, err2; accmode_t accmode = 0; bool checkperm; bool drop_suid = false; gid_t cr_gid; mp = vnode_mount(vp); data = fuse_get_mpdata(mp); dataflags = data->dataflags; checkperm = dataflags & FSESS_DEFAULT_PERMISSIONS; if (cred->cr_ngroups > 0) cr_gid = cred->cr_groups[0]; else cr_gid = 0; if (fuse_isdeadfs(vp)) { return ENXIO; } if (vap->va_uid != (uid_t)VNOVAL) { if (checkperm) { /* Only root may change a file's owner */ err = priv_check_cred(cred, PRIV_VFS_CHOWN); if (err) { /* As a special case, allow the null chown */ err2 = fuse_internal_getattr(vp, &old_va, cred, td); if (err2) return (err2); if (vap->va_uid != old_va.va_uid) return err; else accmode |= VADMIN; drop_suid = true; } else accmode |= VADMIN; } else accmode |= VADMIN; } if (vap->va_gid != (gid_t)VNOVAL) { if (checkperm && priv_check_cred(cred, PRIV_VFS_CHOWN)) drop_suid = true; if (checkperm && !groupmember(vap->va_gid, cred)) { /* * Non-root users may only chgrp to one of their own * groups */ err = priv_check_cred(cred, PRIV_VFS_CHOWN); if (err) { /* As a special case, allow the null chgrp */ err2 = fuse_internal_getattr(vp, &old_va, cred, td); if (err2) return (err2); if (vap->va_gid != old_va.va_gid) return err; accmode |= VADMIN; } else accmode |= VADMIN; } else accmode |= VADMIN; } if (vap->va_size != VNOVAL) { switch (vp->v_type) { case VDIR: return (EISDIR); case VLNK: case VREG: if (vfs_isrdonly(mp)) return (EROFS); break; default: /* * According to POSIX, the result is unspecified * for file types other than regular files, * directories and shared memory objects. We * don't support shared memory objects in the file * system, and have dubious support for truncating * symlinks. Just ignore the request in other cases. */ return (0); } /* Don't set accmode. Permission to trunc is checked upstack */ } if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { if (vap->va_vaflags & VA_UTIMES_NULL) accmode |= VWRITE; else accmode |= VADMIN; } if (drop_suid) { if (vap->va_mode != (mode_t)VNOVAL) vap->va_mode &= ~(S_ISUID | S_ISGID); else { err = fuse_internal_getattr(vp, &old_va, cred, td); if (err) return (err); vap->va_mode = old_va.va_mode & ~(S_ISUID | S_ISGID); } } if (vap->va_mode != (mode_t)VNOVAL) { /* Only root may set the sticky bit on non-directories */ if (checkperm && vp->v_type != VDIR && (vap->va_mode & S_ISTXT) && priv_check_cred(cred, PRIV_VFS_STICKYFILE)) return EFTYPE; if (checkperm && (vap->va_mode & S_ISGID)) { err = fuse_internal_getattr(vp, &old_va, cred, td); if (err) return (err); if (!groupmember(old_va.va_gid, cred)) { err = priv_check_cred(cred, PRIV_VFS_SETGID); if (err) return (err); } } accmode |= VADMIN; } if (vfs_isrdonly(mp)) return EROFS; if (checkperm) { err = fuse_internal_access(vp, accmode, td, cred); } else { err = 0; } if (err) return err; else return fuse_internal_setattr(vp, vap, td, cred); } /* struct vnop_strategy_args { struct vnode *a_vp; struct buf *a_bp; }; */ static int fuse_vnop_strategy(struct vop_strategy_args *ap) { struct vnode *vp = ap->a_vp; struct buf *bp = ap->a_bp; if (!vp || fuse_isdeadfs(vp)) { bp->b_ioflags |= BIO_ERROR; bp->b_error = ENXIO; bufdone(bp); return 0; } /* * VOP_STRATEGY always returns zero and signals error via bp->b_ioflags. * fuse_io_strategy sets bp's error fields */ (void)fuse_io_strategy(vp, bp); return 0; } /* struct vnop_symlink_args { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; struct vattr *a_vap; char *a_target; }; */ static int fuse_vnop_symlink(struct vop_symlink_args *ap) { struct vnode *dvp = ap->a_dvp; struct vnode **vpp = ap->a_vpp; struct componentname *cnp = ap->a_cnp; const char *target = ap->a_target; struct fuse_dispatcher fdi; int err; size_t len; if (fuse_isdeadfs(dvp)) { return ENXIO; } /* * Unlike the other creator type calls, here we have to create a message * where the name of the new entry comes first, and the data describing * the entry comes second. * Hence we can't rely on our handy fuse_internal_newentry() routine, * but put together the message manually and just call the core part. */ len = strlen(target) + 1; fdisp_init(&fdi, len + cnp->cn_namelen + 1); fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL); memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen); ((char *)fdi.indata)[cnp->cn_namelen] = '\0'; memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len); err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi); fdisp_destroy(&fdi); return err; } /* struct vnop_write_args { struct vnode *a_vp; struct uio *a_uio; int a_ioflag; struct ucred *a_cred; }; */ static int fuse_vnop_write(struct vop_write_args *ap) { struct vnode *vp = ap->a_vp; struct uio *uio = ap->a_uio; int ioflag = ap->a_ioflag; struct ucred *cred = ap->a_cred; pid_t pid = curthread->td_proc->p_pid; if (fuse_isdeadfs(vp)) { return ENXIO; } if (VTOFUD(vp)->flag & FN_DIRECTIO) { ioflag |= IO_DIRECT; } return fuse_io_dispatch(vp, uio, ioflag, cred, pid); } static daddr_t fuse_gbp_getblkno(struct vnode *vp, vm_ooffset_t off) { const int biosize = fuse_iosize(vp); return (off / biosize); } static int fuse_gbp_getblksz(struct vnode *vp, daddr_t lbn, long *blksz) { off_t filesize; int err; const int biosize = fuse_iosize(vp); err = fuse_vnode_size(vp, &filesize, NULL, NULL); if (err) { /* This will turn into a SIGBUS */ return (EIO); } else if ((off_t)lbn * biosize >= filesize) { *blksz = 0; } else if ((off_t)(lbn + 1) * biosize > filesize) { *blksz = filesize - (off_t)lbn *biosize; } else { *blksz = biosize; } return (0); } /* struct vnop_getpages_args { struct vnode *a_vp; vm_page_t *a_m; int a_count; int a_reqpage; }; */ static int fuse_vnop_getpages(struct vop_getpages_args *ap) { struct vnode *vp = ap->a_vp; if (!fsess_opt_mmap(vnode_mount(vp))) { SDT_PROBE2(fusefs, , vnops, trace, 1, "called on non-cacheable vnode??\n"); return (VM_PAGER_ERROR); } return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind, ap->a_rahead, fuse_gbp_getblkno, fuse_gbp_getblksz)); } static const char extattr_namespace_separator = '.'; /* struct vop_getextattr_args { struct vop_generic_args a_gen; struct vnode *a_vp; int a_attrnamespace; const char *a_name; struct uio *a_uio; size_t *a_size; struct ucred *a_cred; struct thread *a_td; }; */ static int fuse_vnop_getextattr(struct vop_getextattr_args *ap) { struct vnode *vp = ap->a_vp; struct uio *uio = ap->a_uio; struct fuse_dispatcher fdi; struct fuse_getxattr_in *get_xattr_in; struct fuse_getxattr_out *get_xattr_out; struct mount *mp = vnode_mount(vp); struct thread *td = ap->a_td; struct ucred *cred = ap->a_cred; char *prefix; char *attr_str; size_t len; int err; if (fuse_isdeadfs(vp)) return (ENXIO); if (fsess_not_impl(mp, FUSE_GETXATTR)) return EOPNOTSUPP; err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD); if (err) return err; /* Default to looking for user attributes. */ if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM) prefix = EXTATTR_NAMESPACE_SYSTEM_STRING; else prefix = EXTATTR_NAMESPACE_USER_STRING; len = strlen(prefix) + sizeof(extattr_namespace_separator) + strlen(ap->a_name) + 1; fdisp_init(&fdi, len + sizeof(*get_xattr_in)); fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred); get_xattr_in = fdi.indata; /* * Check to see whether we're querying the available size or * issuing the actual request. If we pass in 0, we get back struct * fuse_getxattr_out. If we pass in a non-zero size, we get back * that much data, without the struct fuse_getxattr_out header. */ if (uio == NULL) get_xattr_in->size = 0; else get_xattr_in->size = uio->uio_resid; attr_str = (char *)fdi.indata + sizeof(*get_xattr_in); snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator, ap->a_name); err = fdisp_wait_answ(&fdi); if (err != 0) { if (err == ENOSYS) { fsess_set_notimpl(mp, FUSE_GETXATTR); err = EOPNOTSUPP; } goto out; } get_xattr_out = fdi.answ; if (ap->a_size != NULL) *ap->a_size = get_xattr_out->size; if (uio != NULL) err = uiomove(fdi.answ, fdi.iosize, uio); out: fdisp_destroy(&fdi); return (err); } /* struct vop_setextattr_args { struct vop_generic_args a_gen; struct vnode *a_vp; int a_attrnamespace; const char *a_name; struct uio *a_uio; struct ucred *a_cred; struct thread *a_td; }; */ static int fuse_vnop_setextattr(struct vop_setextattr_args *ap) { struct vnode *vp = ap->a_vp; struct uio *uio = ap->a_uio; struct fuse_dispatcher fdi; struct fuse_setxattr_in *set_xattr_in; struct mount *mp = vnode_mount(vp); struct thread *td = ap->a_td; struct ucred *cred = ap->a_cred; char *prefix; size_t len; char *attr_str; int err; if (fuse_isdeadfs(vp)) return (ENXIO); if (fsess_not_impl(mp, FUSE_SETXATTR)) return EOPNOTSUPP; if (vfs_isrdonly(mp)) return EROFS; /* Deleting xattrs must use VOP_DELETEEXTATTR instead */ if (ap->a_uio == NULL) { /* * If we got here as fallback from VOP_DELETEEXTATTR, then * return EOPNOTSUPP. */ if (fsess_not_impl(mp, FUSE_REMOVEXATTR)) return (EOPNOTSUPP); else return (EINVAL); } err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VWRITE); if (err) return err; /* Default to looking for user attributes. */ if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM) prefix = EXTATTR_NAMESPACE_SYSTEM_STRING; else prefix = EXTATTR_NAMESPACE_USER_STRING; len = strlen(prefix) + sizeof(extattr_namespace_separator) + strlen(ap->a_name) + 1; fdisp_init(&fdi, len + sizeof(*set_xattr_in) + uio->uio_resid); fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred); set_xattr_in = fdi.indata; set_xattr_in->size = uio->uio_resid; attr_str = (char *)fdi.indata + sizeof(*set_xattr_in); snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator, ap->a_name); err = uiomove((char *)fdi.indata + sizeof(*set_xattr_in) + len, uio->uio_resid, uio); if (err != 0) { goto out; } err = fdisp_wait_answ(&fdi); if (err == ENOSYS) { fsess_set_notimpl(mp, FUSE_SETXATTR); err = EOPNOTSUPP; } if (err == ERESTART) { /* Can't restart after calling uiomove */ err = EINTR; } out: fdisp_destroy(&fdi); return (err); } /* * The Linux / FUSE extended attribute list is simply a collection of * NUL-terminated strings. The FreeBSD extended attribute list is a single * byte length followed by a non-NUL terminated string. So, this allows * conversion of the Linux / FUSE format to the FreeBSD format in place. * Linux attribute names are reported with the namespace as a prefix (e.g. * "user.attribute_name"), but in FreeBSD they are reported without the * namespace prefix (e.g. "attribute_name"). So, we're going from: * * user.attr_name1\0user.attr_name2\0 * * to: * * attr_name1attr_name2 * * Where "" is a single byte number of characters in the attribute name. * * Args: * prefix - exattr namespace prefix string * list, list_len - input list with namespace prefixes * bsd_list, bsd_list_len - output list compatible with bsd vfs */ static int fuse_xattrlist_convert(char *prefix, const char *list, int list_len, char *bsd_list, int *bsd_list_len) { int len, pos, dist_to_next, prefix_len; pos = 0; *bsd_list_len = 0; prefix_len = strlen(prefix); while (pos < list_len && list[pos] != '\0') { dist_to_next = strlen(&list[pos]) + 1; if (bcmp(&list[pos], prefix, prefix_len) == 0 && list[pos + prefix_len] == extattr_namespace_separator) { len = dist_to_next - (prefix_len + sizeof(extattr_namespace_separator)) - 1; if (len >= EXTATTR_MAXNAMELEN) return (ENAMETOOLONG); bsd_list[*bsd_list_len] = len; memcpy(&bsd_list[*bsd_list_len + 1], &list[pos + prefix_len + sizeof(extattr_namespace_separator)], len); *bsd_list_len += len + 1; } pos += dist_to_next; } return (0); } /* * List extended attributes * * The FUSE_LISTXATTR operation is based on Linux's listxattr(2) syscall, which * has a number of differences compared to its FreeBSD equivalent, * extattr_list_file: * * - FUSE_LISTXATTR returns all extended attributes across all namespaces, * whereas listxattr(2) only returns attributes for a single namespace * - FUSE_LISTXATTR prepends each attribute name with "namespace." * - If the provided buffer is not large enough to hold the result, * FUSE_LISTXATTR should return ERANGE, whereas listxattr is expected to * return as many results as will fit. */ /* struct vop_listextattr_args { struct vop_generic_args a_gen; struct vnode *a_vp; int a_attrnamespace; struct uio *a_uio; size_t *a_size; struct ucred *a_cred; struct thread *a_td; }; */ static int fuse_vnop_listextattr(struct vop_listextattr_args *ap) { struct vnode *vp = ap->a_vp; struct uio *uio = ap->a_uio; struct fuse_dispatcher fdi; struct fuse_listxattr_in *list_xattr_in; struct fuse_listxattr_out *list_xattr_out; struct mount *mp = vnode_mount(vp); struct thread *td = ap->a_td; struct ucred *cred = ap->a_cred; char *prefix; char *bsd_list = NULL; char *linux_list; int bsd_list_len; int linux_list_len; int err; if (fuse_isdeadfs(vp)) return (ENXIO); if (fsess_not_impl(mp, FUSE_LISTXATTR)) return EOPNOTSUPP; err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD); if (err) return err; /* * Add space for a NUL and the period separator if enabled. * Default to looking for user attributes. */ if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM) prefix = EXTATTR_NAMESPACE_SYSTEM_STRING; else prefix = EXTATTR_NAMESPACE_USER_STRING; fdisp_init(&fdi, sizeof(*list_xattr_in)); fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred); /* * Retrieve Linux / FUSE compatible list size. */ list_xattr_in = fdi.indata; list_xattr_in->size = 0; err = fdisp_wait_answ(&fdi); if (err != 0) { if (err == ENOSYS) { fsess_set_notimpl(mp, FUSE_LISTXATTR); err = EOPNOTSUPP; } goto out; } list_xattr_out = fdi.answ; linux_list_len = list_xattr_out->size; if (linux_list_len == 0) { if (ap->a_size != NULL) *ap->a_size = linux_list_len; goto out; } /* * Retrieve Linux / FUSE compatible list values. */ fdisp_refresh_vp(&fdi, FUSE_LISTXATTR, vp, td, cred); list_xattr_in = fdi.indata; list_xattr_in->size = linux_list_len; err = fdisp_wait_answ(&fdi); if (err == ERANGE) { /* * Race detected. The attribute list must've grown since the * first FUSE_LISTXATTR call. Start over. Go all the way back * to userland so we can process signals, if necessary, before * restarting. */ err = ERESTART; goto out; } else if (err != 0) goto out; linux_list = fdi.answ; /* FUSE doesn't allow the server to return more data than requested */ if (fdi.iosize > linux_list_len) { struct fuse_data *data = fuse_get_mpdata(mp); fuse_warn(data, FSESS_WARN_LSEXTATTR_LONG, "server returned " "more extended attribute data than requested; " "should've returned ERANGE instead."); } else { /* But returning less data is fine */ linux_list_len = fdi.iosize; } /* * Retrieve the BSD compatible list values. * The Linux / FUSE attribute list format isn't the same * as FreeBSD's format. So we need to transform it into * FreeBSD's format before giving it to the user. */ bsd_list = malloc(linux_list_len, M_TEMP, M_WAITOK); err = fuse_xattrlist_convert(prefix, linux_list, linux_list_len, bsd_list, &bsd_list_len); if (err != 0) goto out; if (ap->a_size != NULL) *ap->a_size = bsd_list_len; if (uio != NULL) err = uiomove(bsd_list, bsd_list_len, uio); out: free(bsd_list, M_TEMP); fdisp_destroy(&fdi); return (err); } /* struct vop_deleteextattr_args { struct vop_generic_args a_gen; struct vnode *a_vp; int a_attrnamespace; const char *a_name; struct ucred *a_cred; struct thread *a_td; }; */ static int fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap) { struct vnode *vp = ap->a_vp; struct fuse_dispatcher fdi; struct mount *mp = vnode_mount(vp); struct thread *td = ap->a_td; struct ucred *cred = ap->a_cred; char *prefix; size_t len; char *attr_str; int err; if (fuse_isdeadfs(vp)) return (ENXIO); if (fsess_not_impl(mp, FUSE_REMOVEXATTR)) return EOPNOTSUPP; if (vfs_isrdonly(mp)) return EROFS; err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VWRITE); if (err) return err; /* Default to looking for user attributes. */ if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM) prefix = EXTATTR_NAMESPACE_SYSTEM_STRING; else prefix = EXTATTR_NAMESPACE_USER_STRING; len = strlen(prefix) + sizeof(extattr_namespace_separator) + strlen(ap->a_name) + 1; fdisp_init(&fdi, len); fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred); attr_str = fdi.indata; snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator, ap->a_name); err = fdisp_wait_answ(&fdi); if (err == ENOSYS) { fsess_set_notimpl(mp, FUSE_REMOVEXATTR); err = EOPNOTSUPP; } fdisp_destroy(&fdi); return (err); } /* struct vnop_print_args { struct vnode *a_vp; }; */ static int fuse_vnop_print(struct vop_print_args *ap) { struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp); printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n", (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid, (uintmax_t)fvdat->nlookup, fvdat->flag); return 0; } /* * Get an NFS filehandle for a FUSE file. * * This will only work for FUSE file systems that guarantee the uniqueness of * nodeid:generation, which most don't. */ /* vop_vptofh { IN struct vnode *a_vp; IN struct fid *a_fhp; }; */ static int fuse_vnop_vptofh(struct vop_vptofh_args *ap) { struct vnode *vp = ap->a_vp; struct fuse_vnode_data *fvdat = VTOFUD(vp); struct fuse_fid *fhp = (struct fuse_fid *)(ap->a_fhp); _Static_assert(sizeof(struct fuse_fid) <= sizeof(struct fid), "FUSE fid type is too big"); struct mount *mp = vnode_mount(vp); struct fuse_data *data = fuse_get_mpdata(mp); struct vattr va; int err; if (!(data->dataflags & FSESS_EXPORT_SUPPORT)) return EOPNOTSUPP; err = fuse_internal_getattr(vp, &va, curthread->td_ucred, curthread); if (err) return err; /*ip = VTOI(ap->a_vp);*/ /*ufhp = (struct ufid *)ap->a_fhp;*/ fhp->len = sizeof(struct fuse_fid); fhp->nid = fvdat->nid; if (fvdat->generation <= UINT32_MAX) fhp->gen = fvdat->generation; else return EOVERFLOW; return (0); } diff --git a/sys/fs/msdosfs/msdosfs_lookup.c b/sys/fs/msdosfs/msdosfs_lookup.c index 9940800a72c2..d9b52539e602 100644 --- a/sys/fs/msdosfs/msdosfs_lookup.c +++ b/sys/fs/msdosfs/msdosfs_lookup.c @@ -1,1067 +1,1067 @@ /* $FreeBSD$ */ /* $NetBSD: msdosfs_lookup.c,v 1.37 1997/11/17 15:36:54 ws Exp $ */ /*- * SPDX-License-Identifier: BSD-4-Clause * * Copyright (C) 1994, 1995, 1997 Wolfgang Solfrank. * Copyright (C) 1994, 1995, 1997 TooLs GmbH. * All rights reserved. * Original code by Paul Popelka (paulp@uts.amdahl.com) (see below). * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by TooLs GmbH. * 4. The name of TooLs GmbH may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ /*- * Written by Paul Popelka (paulp@uts.amdahl.com) * * You can do anything you want with this software, just don't say you wrote * it, and don't remove this notice. * * This software is provided "as is". * * The author supplies this software to be publicly redistributed on the * understanding that the author is not responsible for the correct * functioning of this software in any circumstances and is not liable for * any damages caused by this software. * * October 1992 */ #include #include #include #include #include #include #include #include #include #include #include int msdosfs_lookup(struct vop_cachedlookup_args *ap) { return (msdosfs_lookup_ino(ap->a_dvp, ap->a_vpp, ap->a_cnp, NULL, NULL)); } struct deget_dotdot { u_long cluster; int blkoff; }; static int msdosfs_deget_dotdot(struct mount *mp, void *arg, int lkflags, struct vnode **rvp) { struct deget_dotdot *dd_arg; struct denode *rdp; struct msdosfsmount *pmp; int error; pmp = VFSTOMSDOSFS(mp); dd_arg = arg; error = deget(pmp, dd_arg->cluster, dd_arg->blkoff, LK_EXCLUSIVE, &rdp); if (error == 0) *rvp = DETOV(rdp); return (error); } /* * When we search a directory the blocks containing directory entries are * read and examined. The directory entries contain information that would * normally be in the inode of a unix filesystem. This means that some of * a directory's contents may also be in memory resident denodes (sort of * an inode). This can cause problems if we are searching while some other * process is modifying a directory. To prevent one process from accessing * incompletely modified directory information we depend upon being the * sole owner of a directory block. bread/brelse provide this service. * This being the case, when a process modifies a directory it must first * acquire the disk block that contains the directory entry to be modified. * Then update the disk block and the denode, and then write the disk block * out to disk. This way disk blocks containing directory entries and in * memory denode's will be in synch. */ int msdosfs_lookup_ino(struct vnode *vdp, struct vnode **vpp, struct componentname *cnp, daddr_t *scnp, u_long *blkoffp) { struct mbnambuf nb; daddr_t bn; int error; int slotcount; int slotoffset = 0; int frcn; u_long cluster; u_long blkoff; int diroff; int blsize; int isadir; /* ~0 if found direntry is a directory */ daddr_t scn; /* starting cluster number */ struct vnode *pdp; struct denode *dp; struct denode *tdp; struct msdosfsmount *pmp; struct buf *bp = NULL; struct direntry *dep = NULL; struct deget_dotdot dd_arg; u_char dosfilename[12]; int flags = cnp->cn_flags; int nameiop = cnp->cn_nameiop; int unlen; uint64_t inode1; int wincnt = 1; int chksum = -1, chksum_ok; int olddos = 1; #ifdef MSDOSFS_DEBUG printf("msdosfs_lookup(): looking for %s\n", cnp->cn_nameptr); #endif dp = VTODE(vdp); pmp = dp->de_pmp; #ifdef MSDOSFS_DEBUG printf("msdosfs_lookup(): vdp %p, dp %p, Attr %02x\n", vdp, dp, dp->de_Attributes); #endif restart: if (vpp != NULL) *vpp = NULL; /* * If they are going after the . or .. entry in the root directory, * they won't find it. DOS filesystems don't have them in the root * directory. So, we fake it. deget() is in on this scam too. */ if ((vdp->v_vflag & VV_ROOT) && cnp->cn_nameptr[0] == '.' && (cnp->cn_namelen == 1 || (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.'))) { isadir = ATTR_DIRECTORY; scn = MSDOSFSROOT; #ifdef MSDOSFS_DEBUG printf("msdosfs_lookup(): looking for . or .. in root directory\n"); #endif cluster = MSDOSFSROOT; blkoff = MSDOSFSROOT_OFS; goto foundroot; } switch (unix2dosfn((const u_char *)cnp->cn_nameptr, dosfilename, cnp->cn_namelen, 0, pmp)) { case 0: return (EINVAL); case 1: break; case 2: wincnt = winSlotCnt((const u_char *)cnp->cn_nameptr, cnp->cn_namelen, pmp) + 1; break; case 3: olddos = 0; wincnt = winSlotCnt((const u_char *)cnp->cn_nameptr, cnp->cn_namelen, pmp) + 1; break; } if (pmp->pm_flags & MSDOSFSMNT_SHORTNAME) { wincnt = 1; olddos = 1; } unlen = winLenFixup(cnp->cn_nameptr, cnp->cn_namelen); /* * Suppress search for slots unless creating * file and at end of pathname, in which case * we watch for a place to put the new file in * case it doesn't already exist. */ slotcount = wincnt; if ((nameiop == CREATE || nameiop == RENAME) && (flags & ISLASTCN)) slotcount = 0; #ifdef MSDOSFS_DEBUG printf("msdosfs_lookup(): dos version of filename %s, length %ld\n", dosfilename, cnp->cn_namelen); #endif /* * Search the directory pointed at by vdp for the name pointed at * by cnp->cn_nameptr. */ tdp = NULL; mbnambuf_init(&nb); /* * The outer loop ranges over the clusters that make up the * directory. Note that the root directory is different from all * other directories. It has a fixed number of blocks that are not * part of the pool of allocatable clusters. So, we treat it a * little differently. The root directory starts at "cluster" 0. */ diroff = 0; for (frcn = 0;; frcn++) { error = pcbmap(dp, frcn, &bn, &cluster, &blsize); if (error) { if (error == E2BIG) break; return (error); } error = bread(pmp->pm_devvp, bn, blsize, NOCRED, &bp); if (error) { return (error); } for (blkoff = 0; blkoff < blsize; blkoff += sizeof(struct direntry), diroff += sizeof(struct direntry)) { dep = (struct direntry *)(bp->b_data + blkoff); /* * If the slot is empty and we are still looking * for an empty then remember this one. If the * slot is not empty then check to see if it * matches what we are looking for. If the slot * has never been filled with anything, then the * remainder of the directory has never been used, * so there is no point in searching it. */ if (dep->deName[0] == SLOT_EMPTY || dep->deName[0] == SLOT_DELETED) { /* * Drop memory of previous long matches */ chksum = -1; mbnambuf_init(&nb); if (slotcount < wincnt) { slotcount++; slotoffset = diroff; } if (dep->deName[0] == SLOT_EMPTY) { brelse(bp); goto notfound; } } else { /* * If there wasn't enough space for our winentries, * forget about the empty space */ if (slotcount < wincnt) slotcount = 0; /* * Check for Win95 long filename entry */ if (dep->deAttributes == ATTR_WIN95) { if (pmp->pm_flags & MSDOSFSMNT_SHORTNAME) continue; chksum = win2unixfn(&nb, (struct winentry *)dep, chksum, pmp); continue; } chksum = winChkName(&nb, (const u_char *)cnp->cn_nameptr, unlen, chksum, pmp); if (chksum == -2) { chksum = -1; continue; } /* * Ignore volume labels (anywhere, not just * the root directory). */ if (dep->deAttributes & ATTR_VOLUME) { chksum = -1; continue; } /* * Check for a checksum or name match */ chksum_ok = (chksum == winChksum(dep->deName)); if (!chksum_ok && (!olddos || bcmp(dosfilename, dep->deName, 11))) { chksum = -1; continue; } #ifdef MSDOSFS_DEBUG printf("msdosfs_lookup(): match blkoff %d, diroff %d\n", blkoff, diroff); #endif /* * Remember where this directory * entry came from for whoever did * this lookup. */ dp->de_fndoffset = diroff; if (chksum_ok && nameiop == RENAME) { /* * Target had correct long name * directory entries, reuse them * as needed. */ dp->de_fndcnt = wincnt - 1; } else { /* * Long name directory entries * not present or corrupt, can only * reuse dos directory entry. */ dp->de_fndcnt = 0; } goto found; } } /* for (blkoff = 0; .... */ /* * Release the buffer holding the directory cluster just * searched. */ brelse(bp); } /* for (frcn = 0; ; frcn++) */ notfound: /* * We hold no disk buffers at this point. */ /* * Fixup the slot description to point to the place where * we might put the new DOS direntry (putting the Win95 * long name entries before that) */ if (!slotcount) { slotcount = 1; slotoffset = diroff; } if (wincnt > slotcount) slotoffset += sizeof(struct direntry) * (wincnt - slotcount); /* * If we get here we didn't find the entry we were looking for. But * that's ok if we are creating or renaming and are at the end of * the pathname and the directory hasn't been removed. */ #ifdef MSDOSFS_DEBUG printf("msdosfs_lookup(): op %d, refcnt %ld\n", nameiop, dp->de_refcnt); printf(" slotcount %d, slotoffset %d\n", slotcount, slotoffset); #endif if ((nameiop == CREATE || nameiop == RENAME) && (flags & ISLASTCN) && dp->de_refcnt != 0) { /* * Access for write is interpreted as allowing * creation of files in the directory. */ - error = VOP_ACCESS(vdp, VWRITE, cnp->cn_cred, cnp->cn_thread); + error = VOP_ACCESS(vdp, VWRITE, cnp->cn_cred, curthread); if (error) return (error); /* * Return an indication of where the new directory * entry should be put. */ dp->de_fndoffset = slotoffset; dp->de_fndcnt = wincnt - 1; /* * We return with the directory locked, so that * the parameters we set up above will still be * valid if we actually decide to do a direnter(). * We return ni_vp == NULL to indicate that the entry * does not currently exist; we leave a pointer to * the (locked) directory inode in ndp->ni_dvp. * The pathname buffer is saved so that the name * can be obtained later. * * NB - if the directory is unlocked, then this * information cannot be used. */ cnp->cn_flags |= SAVENAME; return (EJUSTRETURN); } #if 0 /* * Insert name into cache (as non-existent) if appropriate. * * XXX Negative caching is broken for msdosfs because the name * cache doesn't understand peculiarities such as case insensitivity * and 8.3 filenames. Hence, it may not invalidate all negative * entries if a file with this name is later created. */ if ((cnp->cn_flags & MAKEENTRY) != 0) cache_enter(vdp, *vpp, cnp); #endif return (ENOENT); found: /* * NOTE: We still have the buffer with matched directory entry at * this point. */ isadir = dep->deAttributes & ATTR_DIRECTORY; scn = getushort(dep->deStartCluster); if (FAT32(pmp)) { scn |= getushort(dep->deHighClust) << 16; if (scn == pmp->pm_rootdirblk) { /* * There should actually be 0 here. * Just ignore the error. */ scn = MSDOSFSROOT; } } if (isadir) { cluster = scn; if (cluster == MSDOSFSROOT) blkoff = MSDOSFSROOT_OFS; else blkoff = 0; } else if (cluster == MSDOSFSROOT) blkoff = diroff; /* * Now release buf to allow deget to read the entry again. * Reserving it here and giving it to deget could result * in a deadlock. */ brelse(bp); bp = NULL; foundroot: /* * If we entered at foundroot, then we are looking for the . or .. * entry of the filesystems root directory. isadir and scn were * setup before jumping here. And, bp is already null. */ if (FAT32(pmp) && scn == MSDOSFSROOT) scn = pmp->pm_rootdirblk; if (scnp != NULL) { *scnp = cluster; *blkoffp = blkoff; return (0); } /* * If deleting, and at end of pathname, return * parameters which can be used to remove file. */ if (nameiop == DELETE && (flags & ISLASTCN)) { /* * Don't allow deleting the root. */ if (blkoff == MSDOSFSROOT_OFS) return (EBUSY); /* * Write access to directory required to delete files. */ - error = VOP_ACCESS(vdp, VWRITE, cnp->cn_cred, cnp->cn_thread); + error = VOP_ACCESS(vdp, VWRITE, cnp->cn_cred, curthread); if (error) return (error); /* * Return pointer to current entry in dp->i_offset. * Save directory inode pointer in ndp->ni_dvp for dirremove(). */ if (dp->de_StartCluster == scn && isadir) { /* "." */ VREF(vdp); *vpp = vdp; return (0); } error = deget(pmp, cluster, blkoff, LK_EXCLUSIVE, &tdp); if (error) return (error); *vpp = DETOV(tdp); return (0); } /* * If rewriting (RENAME), return the inode and the * information required to rewrite the present directory * Must get inode of directory entry to verify it's a * regular file, or empty directory. */ if (nameiop == RENAME && (flags & ISLASTCN)) { if (blkoff == MSDOSFSROOT_OFS) return (EBUSY); - error = VOP_ACCESS(vdp, VWRITE, cnp->cn_cred, cnp->cn_thread); + error = VOP_ACCESS(vdp, VWRITE, cnp->cn_cred, curthread); if (error) return (error); /* * Careful about locking second inode. * This can only occur if the target is ".". */ if (dp->de_StartCluster == scn && isadir) return (EISDIR); if ((error = deget(pmp, cluster, blkoff, LK_EXCLUSIVE, &tdp)) != 0) return (error); *vpp = DETOV(tdp); cnp->cn_flags |= SAVENAME; return (0); } /* * Step through the translation in the name. We do not `vput' the * directory because we may need it again if a symbolic link * is relative to the current directory. Instead we save it * unlocked as "pdp". We must get the target inode before unlocking * the directory to insure that the inode will not be removed * before we get it. We prevent deadlock by always fetching * inodes from the root, moving down the directory tree. Thus * when following backward pointers ".." we must unlock the * parent directory before getting the requested directory. */ pdp = vdp; if (flags & ISDOTDOT) { dd_arg.cluster = cluster; dd_arg.blkoff = blkoff; error = vn_vget_ino_gen(vdp, msdosfs_deget_dotdot, &dd_arg, cnp->cn_lkflags, vpp); if (error != 0) { *vpp = NULL; return (error); } /* * Recheck that ".." still points to the inode we * looked up before pdp lock was dropped. */ error = msdosfs_lookup_ino(pdp, NULL, cnp, &scn, &blkoff); if (error) { vput(*vpp); *vpp = NULL; return (error); } if (FAT32(pmp) && scn == MSDOSFSROOT) scn = pmp->pm_rootdirblk; inode1 = scn * pmp->pm_bpcluster + blkoff; if (VTODE(*vpp)->de_inode != inode1) { vput(*vpp); goto restart; } } else if (dp->de_StartCluster == scn && isadir) { VREF(vdp); /* we want ourself, ie "." */ *vpp = vdp; } else { if ((error = deget(pmp, cluster, blkoff, LK_EXCLUSIVE, &tdp)) != 0) return (error); *vpp = DETOV(tdp); } /* * Insert name into cache if appropriate. */ if (cnp->cn_flags & MAKEENTRY) cache_enter(vdp, *vpp, cnp); return (0); } /* * dep - directory entry to copy into the directory * ddep - directory to add to * depp - return the address of the denode for the created directory entry * if depp != 0 * cnp - componentname needed for Win95 long filenames */ int createde(struct denode *dep, struct denode *ddep, struct denode **depp, struct componentname *cnp) { int error; u_long dirclust, diroffset; struct direntry *ndep; struct msdosfsmount *pmp = ddep->de_pmp; struct buf *bp; daddr_t bn; int blsize; #ifdef MSDOSFS_DEBUG printf("createde(dep %p, ddep %p, depp %p, cnp %p)\n", dep, ddep, depp, cnp); #endif /* * If no space left in the directory then allocate another cluster * and chain it onto the end of the file. There is one exception * to this. That is, if the root directory has no more space it * can NOT be expanded. extendfile() checks for and fails attempts * to extend the root directory. We just return an error in that * case. */ if (ddep->de_fndoffset >= ddep->de_FileSize) { diroffset = ddep->de_fndoffset + sizeof(struct direntry) - ddep->de_FileSize; dirclust = de_clcount(pmp, diroffset); error = extendfile(ddep, dirclust, 0, 0, DE_CLEAR); if (error) { (void)detrunc(ddep, ddep->de_FileSize, 0, NOCRED); return error; } /* * Update the size of the directory */ ddep->de_FileSize += de_cn2off(pmp, dirclust); } /* * We just read in the cluster with space. Copy the new directory * entry in. Then write it to disk. NOTE: DOS directories * do not get smaller as clusters are emptied. */ error = pcbmap(ddep, de_cluster(pmp, ddep->de_fndoffset), &bn, &dirclust, &blsize); if (error) return error; diroffset = ddep->de_fndoffset; if (dirclust != MSDOSFSROOT) diroffset &= pmp->pm_crbomask; if ((error = bread(pmp->pm_devvp, bn, blsize, NOCRED, &bp)) != 0) { brelse(bp); return error; } ndep = bptoep(pmp, bp, ddep->de_fndoffset); DE_EXTERNALIZE(ndep, dep); /* * Now write the Win95 long name */ if (ddep->de_fndcnt > 0) { uint8_t chksum = winChksum(ndep->deName); const u_char *un = (const u_char *)cnp->cn_nameptr; int unlen = cnp->cn_namelen; int cnt = 1; while (--ddep->de_fndcnt >= 0) { if (!(ddep->de_fndoffset & pmp->pm_crbomask)) { if (DOINGASYNC(DETOV(ddep))) bdwrite(bp); else if ((error = bwrite(bp)) != 0) return error; ddep->de_fndoffset -= sizeof(struct direntry); error = pcbmap(ddep, de_cluster(pmp, ddep->de_fndoffset), &bn, 0, &blsize); if (error) return error; error = bread(pmp->pm_devvp, bn, blsize, NOCRED, &bp); if (error) { return error; } ndep = bptoep(pmp, bp, ddep->de_fndoffset); } else { ndep--; ddep->de_fndoffset -= sizeof(struct direntry); } if (!unix2winfn(un, unlen, (struct winentry *)ndep, cnt++, chksum, pmp)) break; } } if (DOINGASYNC(DETOV(ddep))) bdwrite(bp); else if ((error = bwrite(bp)) != 0) return error; /* * If they want us to return with the denode gotten. */ if (depp) { if (dep->de_Attributes & ATTR_DIRECTORY) { dirclust = dep->de_StartCluster; if (FAT32(pmp) && dirclust == pmp->pm_rootdirblk) dirclust = MSDOSFSROOT; if (dirclust == MSDOSFSROOT) diroffset = MSDOSFSROOT_OFS; else diroffset = 0; } return (deget(pmp, dirclust, diroffset, LK_EXCLUSIVE, depp)); } return 0; } /* * Be sure a directory is empty except for "." and "..". Return 1 if empty, * return 0 if not empty or error. */ int dosdirempty(struct denode *dep) { int blsize; int error; u_long cn; daddr_t bn; struct buf *bp; struct msdosfsmount *pmp = dep->de_pmp; struct direntry *dentp; /* * Since the filesize field in directory entries for a directory is * zero, we just have to feel our way through the directory until * we hit end of file. */ for (cn = 0;; cn++) { if ((error = pcbmap(dep, cn, &bn, 0, &blsize)) != 0) { if (error == E2BIG) return (1); /* it's empty */ return (0); } error = bread(pmp->pm_devvp, bn, blsize, NOCRED, &bp); if (error) { return (0); } for (dentp = (struct direntry *)bp->b_data; (char *)dentp < bp->b_data + blsize; dentp++) { if (dentp->deName[0] != SLOT_DELETED && (dentp->deAttributes & ATTR_VOLUME) == 0) { /* * In dos directories an entry whose name * starts with SLOT_EMPTY (0) starts the * beginning of the unused part of the * directory, so we can just return that it * is empty. */ if (dentp->deName[0] == SLOT_EMPTY) { brelse(bp); return (1); } /* * Any names other than "." and ".." in a * directory mean it is not empty. */ if (bcmp(dentp->deName, ". ", 11) && bcmp(dentp->deName, ".. ", 11)) { brelse(bp); #ifdef MSDOSFS_DEBUG printf("dosdirempty(): entry found %02x, %02x\n", dentp->deName[0], dentp->deName[1]); #endif return (0); /* not empty */ } } } brelse(bp); } /* NOTREACHED */ } /* * Check to see if the directory described by target is in some * subdirectory of source. This prevents something like the following from * succeeding and leaving a bunch or files and directories orphaned. mv * /a/b/c /a/b/c/d/e/f Where c and f are directories. * * source - the inode for /a/b/c * target - the inode for /a/b/c/d/e/f * * Returns 0 if target is NOT a subdirectory of source. * Otherwise returns a non-zero error number. */ int doscheckpath(struct denode *source, struct denode *target, daddr_t *wait_scn) { daddr_t scn; struct msdosfsmount *pmp; struct direntry *ep; struct denode *dep; struct buf *bp = NULL; int error = 0; *wait_scn = 0; pmp = target->de_pmp; lockmgr_assert(&pmp->pm_checkpath_lock, KA_XLOCKED); KASSERT(pmp == source->de_pmp, ("doscheckpath: source and target on different filesystems")); if ((target->de_Attributes & ATTR_DIRECTORY) == 0 || (source->de_Attributes & ATTR_DIRECTORY) == 0) return (ENOTDIR); if (target->de_StartCluster == source->de_StartCluster) return (EEXIST); if (target->de_StartCluster == MSDOSFSROOT || (FAT32(pmp) && target->de_StartCluster == pmp->pm_rootdirblk)) return (0); dep = target; vget(DETOV(dep), LK_EXCLUSIVE); for (;;) { if ((dep->de_Attributes & ATTR_DIRECTORY) == 0) { error = ENOTDIR; break; } scn = dep->de_StartCluster; error = bread(pmp->pm_devvp, cntobn(pmp, scn), pmp->pm_bpcluster, NOCRED, &bp); if (error != 0) break; ep = (struct direntry *)bp->b_data + 1; if ((ep->deAttributes & ATTR_DIRECTORY) == 0 || bcmp(ep->deName, ".. ", 11) != 0) { error = ENOTDIR; brelse(bp); break; } scn = getushort(ep->deStartCluster); if (FAT32(pmp)) scn |= getushort(ep->deHighClust) << 16; brelse(bp); if (scn == source->de_StartCluster) { error = EINVAL; break; } if (scn == MSDOSFSROOT) break; if (FAT32(pmp) && scn == pmp->pm_rootdirblk) { /* * scn should be 0 in this case, * but we silently ignore the error. */ break; } vput(DETOV(dep)); dep = NULL; /* NOTE: deget() clears dep on error */ error = deget(pmp, scn, 0, LK_EXCLUSIVE | LK_NOWAIT, &dep); if (error != 0) { *wait_scn = scn; break; } } #ifdef MSDOSFS_DEBUG if (error == ENOTDIR) printf("doscheckpath(): .. not a directory?\n"); #endif if (dep != NULL) vput(DETOV(dep)); return (error); } /* * Read in the disk block containing the directory entry (dirclu, dirofs) * and return the address of the buf header, and the address of the * directory entry within the block. */ int readep(struct msdosfsmount *pmp, u_long dirclust, u_long diroffset, struct buf **bpp, struct direntry **epp) { int error; daddr_t bn; int blsize; blsize = pmp->pm_bpcluster; if (dirclust == MSDOSFSROOT && de_blk(pmp, diroffset + blsize) > pmp->pm_rootdirsize) blsize = de_bn2off(pmp, pmp->pm_rootdirsize) & pmp->pm_crbomask; bn = detobn(pmp, dirclust, diroffset); if ((error = bread(pmp->pm_devvp, bn, blsize, NOCRED, bpp)) != 0) { brelse(*bpp); *bpp = NULL; return (error); } if (epp) *epp = bptoep(pmp, *bpp, diroffset); return (0); } /* * Read in the disk block containing the directory entry dep came from and * return the address of the buf header, and the address of the directory * entry within the block. */ int readde(struct denode *dep, struct buf **bpp, struct direntry **epp) { return (readep(dep->de_pmp, dep->de_dirclust, dep->de_diroffset, bpp, epp)); } /* * Remove a directory entry. At this point the file represented by the * directory entry to be removed is still full length until no one has it * open. When the file no longer being used msdosfs_inactive() is called * and will truncate the file to 0 length. When the vnode containing the * denode is needed for some other purpose by VFS it will call * msdosfs_reclaim() which will remove the denode from the denode cache. * * pdep directory where the entry is removed * dep file to be removed */ int removede(struct denode *pdep, struct denode *dep) { int error; struct direntry *ep; struct buf *bp; daddr_t bn; int blsize; struct msdosfsmount *pmp = pdep->de_pmp; u_long offset = pdep->de_fndoffset; #ifdef MSDOSFS_DEBUG printf("removede(): filename %s, dep %p, offset %08lx\n", dep->de_Name, dep, offset); #endif dep->de_refcnt--; offset += sizeof(struct direntry); do { offset -= sizeof(struct direntry); error = pcbmap(pdep, de_cluster(pmp, offset), &bn, 0, &blsize); if (error) return error; error = bread(pmp->pm_devvp, bn, blsize, NOCRED, &bp); if (error) { return error; } ep = bptoep(pmp, bp, offset); /* * Check whether, if we came here the second time, i.e. * when underflowing into the previous block, the last * entry in this block is a longfilename entry, too. */ if (ep->deAttributes != ATTR_WIN95 && offset != pdep->de_fndoffset) { brelse(bp); break; } offset += sizeof(struct direntry); while (1) { /* * We are a bit aggressive here in that we delete any Win95 * entries preceding this entry, not just the ones we "own". * Since these presumably aren't valid anyway, * there should be no harm. */ offset -= sizeof(struct direntry); ep--->deName[0] = SLOT_DELETED; if ((pmp->pm_flags & MSDOSFSMNT_NOWIN95) || !(offset & pmp->pm_crbomask) || ep->deAttributes != ATTR_WIN95) break; } if (DOINGASYNC(DETOV(pdep))) bdwrite(bp); else if ((error = bwrite(bp)) != 0) return error; } while (!(pmp->pm_flags & MSDOSFSMNT_NOWIN95) && !(offset & pmp->pm_crbomask) && offset); return 0; } /* * Create a unique DOS name in dvp */ int uniqdosname(struct denode *dep, struct componentname *cnp, u_char *cp) { struct msdosfsmount *pmp = dep->de_pmp; struct direntry *dentp; int gen; int blsize; u_long cn; daddr_t bn; struct buf *bp; int error; if (pmp->pm_flags & MSDOSFSMNT_SHORTNAME) return (unix2dosfn((const u_char *)cnp->cn_nameptr, cp, cnp->cn_namelen, 0, pmp) ? 0 : EINVAL); for (gen = 1;; gen++) { /* * Generate DOS name with generation number */ if (!unix2dosfn((const u_char *)cnp->cn_nameptr, cp, cnp->cn_namelen, gen, pmp)) return gen == 1 ? EINVAL : EEXIST; /* * Now look for a dir entry with this exact name */ for (cn = error = 0; !error; cn++) { if ((error = pcbmap(dep, cn, &bn, 0, &blsize)) != 0) { if (error == E2BIG) /* EOF reached and not found */ return 0; return error; } error = bread(pmp->pm_devvp, bn, blsize, NOCRED, &bp); if (error) { return error; } for (dentp = (struct direntry *)bp->b_data; (char *)dentp < bp->b_data + blsize; dentp++) { if (dentp->deName[0] == SLOT_EMPTY) { /* * Last used entry and not found */ brelse(bp); return 0; } /* * Ignore volume labels and Win95 entries */ if (dentp->deAttributes & ATTR_VOLUME) continue; if (!bcmp(dentp->deName, cp, 11)) { error = EEXIST; break; } } brelse(bp); } } } diff --git a/sys/fs/msdosfs/msdosfs_vnops.c b/sys/fs/msdosfs/msdosfs_vnops.c index 5f778724786d..44026c270f1c 100644 --- a/sys/fs/msdosfs/msdosfs_vnops.c +++ b/sys/fs/msdosfs/msdosfs_vnops.c @@ -1,2010 +1,2010 @@ /* $FreeBSD$ */ /* $NetBSD: msdosfs_vnops.c,v 1.68 1998/02/10 14:10:04 mrg Exp $ */ /*- * SPDX-License-Identifier: BSD-4-Clause * * Copyright (C) 1994, 1995, 1997 Wolfgang Solfrank. * Copyright (C) 1994, 1995, 1997 TooLs GmbH. * All rights reserved. * Original code by Paul Popelka (paulp@uts.amdahl.com) (see below). * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by TooLs GmbH. * 4. The name of TooLs GmbH may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ /*- * Written by Paul Popelka (paulp@uts.amdahl.com) * * You can do anything you want with this software, just don't say you wrote * it, and don't remove this notice. * * This software is provided "as is". * * The author supplies this software to be publicly redistributed on the * understanding that the author is not responsible for the correct * functioning of this software in any circumstances and is not liable for * any damages caused by this software. * * October 1992 */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * Prototypes for MSDOSFS vnode operations */ static vop_create_t msdosfs_create; static vop_mknod_t msdosfs_mknod; static vop_open_t msdosfs_open; static vop_close_t msdosfs_close; static vop_access_t msdosfs_access; static vop_getattr_t msdosfs_getattr; static vop_setattr_t msdosfs_setattr; static vop_read_t msdosfs_read; static vop_write_t msdosfs_write; static vop_fsync_t msdosfs_fsync; static vop_remove_t msdosfs_remove; static vop_link_t msdosfs_link; static vop_rename_t msdosfs_rename; static vop_mkdir_t msdosfs_mkdir; static vop_rmdir_t msdosfs_rmdir; static vop_symlink_t msdosfs_symlink; static vop_readdir_t msdosfs_readdir; static vop_bmap_t msdosfs_bmap; static vop_getpages_t msdosfs_getpages; static vop_strategy_t msdosfs_strategy; static vop_print_t msdosfs_print; static vop_pathconf_t msdosfs_pathconf; static vop_vptofh_t msdosfs_vptofh; /* * Some general notes: * * In the ufs filesystem the inodes, superblocks, and indirect blocks are * read/written using the vnode for the filesystem. Blocks that represent * the contents of a file are read/written using the vnode for the file * (including directories when they are read/written as files). This * presents problems for the dos filesystem because data that should be in * an inode (if dos had them) resides in the directory itself. Since we * must update directory entries without the benefit of having the vnode * for the directory we must use the vnode for the filesystem. This means * that when a directory is actually read/written (via read, write, or * readdir, or seek) we must use the vnode for the filesystem instead of * the vnode for the directory as would happen in ufs. This is to insure we * retrieve the correct block from the buffer cache since the hash value is * based upon the vnode address and the desired block number. */ /* * Create a regular file. On entry the directory to contain the file being * created is locked. We must release before we return. We must also free * the pathname buffer pointed at by cnp->cn_pnbuf, always on error, or * only if the SAVESTART bit in cn_flags is clear on success. */ static int msdosfs_create(struct vop_create_args *ap) { struct componentname *cnp = ap->a_cnp; struct denode ndirent; struct denode *dep; struct denode *pdep = VTODE(ap->a_dvp); struct timespec ts; int error; #ifdef MSDOSFS_DEBUG printf("msdosfs_create(cnp %p, vap %p\n", cnp, ap->a_vap); #endif /* * If this is the root directory and there is no space left we * can't do anything. This is because the root directory can not * change size. */ if (pdep->de_StartCluster == MSDOSFSROOT && pdep->de_fndoffset >= pdep->de_FileSize) { error = ENOSPC; goto bad; } /* * Create a directory entry for the file, then call createde() to * have it installed. NOTE: DOS files are always executable. We * use the absence of the owner write bit to make the file * readonly. */ #ifdef DIAGNOSTIC if ((cnp->cn_flags & HASBUF) == 0) panic("msdosfs_create: no name"); #endif memset(&ndirent, 0, sizeof(ndirent)); error = uniqdosname(pdep, cnp, ndirent.de_Name); if (error) goto bad; ndirent.de_Attributes = ATTR_ARCHIVE; ndirent.de_LowerCase = 0; ndirent.de_StartCluster = 0; ndirent.de_FileSize = 0; ndirent.de_pmp = pdep->de_pmp; ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE; vfs_timestamp(&ts); DETIMES(&ndirent, &ts, &ts, &ts); error = createde(&ndirent, pdep, &dep, cnp); if (error) goto bad; *ap->a_vpp = DETOV(dep); if ((cnp->cn_flags & MAKEENTRY) != 0) cache_enter(ap->a_dvp, *ap->a_vpp, cnp); return (0); bad: return (error); } static int msdosfs_mknod(struct vop_mknod_args *ap) { return (EINVAL); } static int msdosfs_open(struct vop_open_args *ap) { struct denode *dep = VTODE(ap->a_vp); vnode_create_vobject(ap->a_vp, dep->de_FileSize, ap->a_td); return 0; } static int msdosfs_close(struct vop_close_args *ap) { struct vnode *vp = ap->a_vp; struct denode *dep = VTODE(vp); struct timespec ts; VI_LOCK(vp); if (vp->v_usecount > 1) { vfs_timestamp(&ts); DETIMES(dep, &ts, &ts, &ts); } VI_UNLOCK(vp); return 0; } static int msdosfs_access(struct vop_access_args *ap) { struct vnode *vp = ap->a_vp; struct denode *dep = VTODE(ap->a_vp); struct msdosfsmount *pmp = dep->de_pmp; mode_t file_mode; accmode_t accmode = ap->a_accmode; file_mode = S_IRWXU|S_IRWXG|S_IRWXO; file_mode &= (vp->v_type == VDIR ? pmp->pm_dirmask : pmp->pm_mask); /* * Disallow writing to directories and regular files if the * filesystem is read-only. */ if (accmode & VWRITE) { switch (vp->v_type) { case VREG: case VDIR: if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); break; default: break; } } return (vaccess(vp->v_type, file_mode, pmp->pm_uid, pmp->pm_gid, ap->a_accmode, ap->a_cred)); } static int msdosfs_getattr(struct vop_getattr_args *ap) { struct denode *dep = VTODE(ap->a_vp); struct msdosfsmount *pmp = dep->de_pmp; struct vattr *vap = ap->a_vap; mode_t mode; struct timespec ts; u_long dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry); uint64_t fileid; vfs_timestamp(&ts); DETIMES(dep, &ts, &ts, &ts); vap->va_fsid = dev2udev(pmp->pm_dev); /* * The following computation of the fileid must be the same as that * used in msdosfs_readdir() to compute d_fileno. If not, pwd * doesn't work. */ if (dep->de_Attributes & ATTR_DIRECTORY) { fileid = (uint64_t)cntobn(pmp, dep->de_StartCluster) * dirsperblk; if (dep->de_StartCluster == MSDOSFSROOT) fileid = 1; } else { fileid = (uint64_t)cntobn(pmp, dep->de_dirclust) * dirsperblk; if (dep->de_dirclust == MSDOSFSROOT) fileid = (uint64_t)roottobn(pmp, 0) * dirsperblk; fileid += (uoff_t)dep->de_diroffset / sizeof(struct direntry); } vap->va_fileid = fileid; mode = S_IRWXU|S_IRWXG|S_IRWXO; if (dep->de_Attributes & ATTR_READONLY) mode &= ~(S_IWUSR|S_IWGRP|S_IWOTH); vap->va_mode = mode & (ap->a_vp->v_type == VDIR ? pmp->pm_dirmask : pmp->pm_mask); vap->va_uid = pmp->pm_uid; vap->va_gid = pmp->pm_gid; vap->va_nlink = 1; vap->va_rdev = NODEV; vap->va_size = dep->de_FileSize; fattime2timespec(dep->de_MDate, dep->de_MTime, 0, 0, &vap->va_mtime); vap->va_ctime = vap->va_mtime; if (pmp->pm_flags & MSDOSFSMNT_LONGNAME) { fattime2timespec(dep->de_ADate, 0, 0, 0, &vap->va_atime); fattime2timespec(dep->de_CDate, dep->de_CTime, dep->de_CHun, 0, &vap->va_birthtime); } else { vap->va_atime = vap->va_mtime; vap->va_birthtime.tv_sec = -1; vap->va_birthtime.tv_nsec = 0; } vap->va_flags = 0; if (dep->de_Attributes & ATTR_ARCHIVE) vap->va_flags |= UF_ARCHIVE; if (dep->de_Attributes & ATTR_HIDDEN) vap->va_flags |= UF_HIDDEN; if (dep->de_Attributes & ATTR_READONLY) vap->va_flags |= UF_READONLY; if (dep->de_Attributes & ATTR_SYSTEM) vap->va_flags |= UF_SYSTEM; vap->va_gen = 0; vap->va_blocksize = pmp->pm_bpcluster; vap->va_bytes = (dep->de_FileSize + pmp->pm_crbomask) & ~pmp->pm_crbomask; vap->va_type = ap->a_vp->v_type; vap->va_filerev = dep->de_modrev; return (0); } static int msdosfs_setattr(struct vop_setattr_args *ap) { struct vnode *vp = ap->a_vp; struct denode *dep = VTODE(ap->a_vp); struct msdosfsmount *pmp = dep->de_pmp; struct vattr *vap = ap->a_vap; struct ucred *cred = ap->a_cred; struct thread *td = curthread; int error = 0; #ifdef MSDOSFS_DEBUG printf("msdosfs_setattr(): vp %p, vap %p, cred %p\n", ap->a_vp, vap, cred); #endif /* * Check for unsettable attributes. */ if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) || (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) || (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) || (vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) { #ifdef MSDOSFS_DEBUG printf("msdosfs_setattr(): returning EINVAL\n"); printf(" va_type %d, va_nlink %llx, va_fsid %llx, va_fileid %llx\n", vap->va_type, (unsigned long long)vap->va_nlink, (unsigned long long)vap->va_fsid, (unsigned long long)vap->va_fileid); printf(" va_blocksize %lx, va_rdev %llx, va_bytes %llx, va_gen %lx\n", vap->va_blocksize, (unsigned long long)vap->va_rdev, (unsigned long long)vap->va_bytes, vap->va_gen); printf(" va_uid %x, va_gid %x\n", vap->va_uid, vap->va_gid); #endif return (EINVAL); } /* * We don't allow setting attributes on the root directory. * The special case for the root directory is because before * FAT32, the root directory didn't have an entry for itself * (and was otherwise special). With FAT32, the root * directory is not so special, but still doesn't have an * entry for itself. */ if (vp->v_vflag & VV_ROOT) return (EINVAL); if (vap->va_flags != VNOVAL) { if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); if (cred->cr_uid != pmp->pm_uid) { error = priv_check_cred(cred, PRIV_VFS_ADMIN); if (error) return (error); } /* * We are very inconsistent about handling unsupported * attributes. We ignored the access time and the * read and execute bits. We were strict for the other * attributes. */ if (vap->va_flags & ~(UF_ARCHIVE | UF_HIDDEN | UF_READONLY | UF_SYSTEM)) return EOPNOTSUPP; if (vap->va_flags & UF_ARCHIVE) dep->de_Attributes |= ATTR_ARCHIVE; else dep->de_Attributes &= ~ATTR_ARCHIVE; if (vap->va_flags & UF_HIDDEN) dep->de_Attributes |= ATTR_HIDDEN; else dep->de_Attributes &= ~ATTR_HIDDEN; /* We don't allow changing the readonly bit on directories. */ if (vp->v_type != VDIR) { if (vap->va_flags & UF_READONLY) dep->de_Attributes |= ATTR_READONLY; else dep->de_Attributes &= ~ATTR_READONLY; } if (vap->va_flags & UF_SYSTEM) dep->de_Attributes |= ATTR_SYSTEM; else dep->de_Attributes &= ~ATTR_SYSTEM; dep->de_flag |= DE_MODIFIED; } if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) { uid_t uid; gid_t gid; if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); uid = vap->va_uid; if (uid == (uid_t)VNOVAL) uid = pmp->pm_uid; gid = vap->va_gid; if (gid == (gid_t)VNOVAL) gid = pmp->pm_gid; if (cred->cr_uid != pmp->pm_uid || uid != pmp->pm_uid || (gid != pmp->pm_gid && !groupmember(gid, cred))) { error = priv_check_cred(cred, PRIV_VFS_CHOWN); if (error) return (error); } if (uid != pmp->pm_uid || gid != pmp->pm_gid) return EINVAL; } if (vap->va_size != VNOVAL) { switch (vp->v_type) { case VDIR: return (EISDIR); case VREG: /* * Truncation is only supported for regular files, * Disallow it if the filesystem is read-only. */ if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); break; default: /* * According to POSIX, the result is unspecified * for file types other than regular files, * directories and shared memory objects. We * don't support any file types except regular * files and directories in this file system, so * this (default) case is unreachable and can do * anything. Keep falling through to detrunc() * for now. */ break; } error = detrunc(dep, vap->va_size, 0, cred); if (error) return error; } if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); error = vn_utimes_perm(vp, vap, cred, td); if (error != 0) return (error); if ((pmp->pm_flags & MSDOSFSMNT_NOWIN95) == 0 && vap->va_atime.tv_sec != VNOVAL) { dep->de_flag &= ~DE_ACCESS; timespec2fattime(&vap->va_atime, 0, &dep->de_ADate, NULL, NULL); } if (vap->va_mtime.tv_sec != VNOVAL) { dep->de_flag &= ~DE_UPDATE; timespec2fattime(&vap->va_mtime, 0, &dep->de_MDate, &dep->de_MTime, NULL); } /* * We don't set the archive bit when modifying the time of * a directory to emulate the Windows/DOS behavior. */ if (vp->v_type != VDIR) dep->de_Attributes |= ATTR_ARCHIVE; dep->de_flag |= DE_MODIFIED; } /* * DOS files only have the ability to have their writability * attribute set, so we use the owner write bit to set the readonly * attribute. */ if (vap->va_mode != (mode_t)VNOVAL) { if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); if (cred->cr_uid != pmp->pm_uid) { error = priv_check_cred(cred, PRIV_VFS_ADMIN); if (error) return (error); } if (vp->v_type != VDIR) { /* We ignore the read and execute bits. */ if (vap->va_mode & S_IWUSR) dep->de_Attributes &= ~ATTR_READONLY; else dep->de_Attributes |= ATTR_READONLY; dep->de_Attributes |= ATTR_ARCHIVE; dep->de_flag |= DE_MODIFIED; } } return (deupdat(dep, 0)); } static int msdosfs_read(struct vop_read_args *ap) { int error = 0; int blsize; int isadir; ssize_t orig_resid; u_int n; u_long diff; u_long on; daddr_t lbn; daddr_t rablock; int rasize; int seqcount; struct buf *bp; struct vnode *vp = ap->a_vp; struct denode *dep = VTODE(vp); struct msdosfsmount *pmp = dep->de_pmp; struct uio *uio = ap->a_uio; /* * If they didn't ask for any data, then we are done. */ orig_resid = uio->uio_resid; if (orig_resid == 0) return (0); /* * The caller is supposed to ensure that * uio->uio_offset >= 0 and uio->uio_resid >= 0. * We don't need to check for large offsets as in ffs because * dep->de_FileSize <= MSDOSFS_FILESIZE_MAX < OFF_MAX, so large * offsets cannot cause overflow even in theory. */ seqcount = ap->a_ioflag >> IO_SEQSHIFT; isadir = dep->de_Attributes & ATTR_DIRECTORY; do { if (uio->uio_offset >= dep->de_FileSize) break; lbn = de_cluster(pmp, uio->uio_offset); rablock = lbn + 1; blsize = pmp->pm_bpcluster; on = uio->uio_offset & pmp->pm_crbomask; /* * If we are operating on a directory file then be sure to * do i/o with the vnode for the filesystem instead of the * vnode for the directory. */ if (isadir) { /* convert cluster # to block # */ error = pcbmap(dep, lbn, &lbn, 0, &blsize); if (error == E2BIG) { error = EINVAL; break; } else if (error) break; error = bread(pmp->pm_devvp, lbn, blsize, NOCRED, &bp); } else if (de_cn2off(pmp, rablock) >= dep->de_FileSize) { error = bread(vp, lbn, blsize, NOCRED, &bp); } else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) { error = cluster_read(vp, dep->de_FileSize, lbn, blsize, NOCRED, on + uio->uio_resid, seqcount, 0, &bp); } else if (seqcount > 1) { rasize = blsize; error = breadn(vp, lbn, blsize, &rablock, &rasize, 1, NOCRED, &bp); } else { error = bread(vp, lbn, blsize, NOCRED, &bp); } if (error) { brelse(bp); break; } diff = pmp->pm_bpcluster - on; n = diff > uio->uio_resid ? uio->uio_resid : diff; diff = dep->de_FileSize - uio->uio_offset; if (diff < n) n = diff; diff = blsize - bp->b_resid; if (diff < n) n = diff; error = vn_io_fault_uiomove(bp->b_data + on, (int) n, uio); brelse(bp); } while (error == 0 && uio->uio_resid > 0 && n != 0); if (!isadir && (error == 0 || uio->uio_resid != orig_resid) && (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) dep->de_flag |= DE_ACCESS; return (error); } /* * Write data to a file or directory. */ static int msdosfs_write(struct vop_write_args *ap) { int n; int croffset; ssize_t resid; u_long osize; int error = 0; u_long count; int seqcount; daddr_t bn, lastcn; struct buf *bp; int ioflag = ap->a_ioflag; struct uio *uio = ap->a_uio; struct vnode *vp = ap->a_vp; struct vnode *thisvp; struct denode *dep = VTODE(vp); struct msdosfsmount *pmp = dep->de_pmp; struct ucred *cred = ap->a_cred; #ifdef MSDOSFS_DEBUG printf("msdosfs_write(vp %p, uio %p, ioflag %x, cred %p\n", vp, uio, ioflag, cred); printf("msdosfs_write(): diroff %lu, dirclust %lu, startcluster %lu\n", dep->de_diroffset, dep->de_dirclust, dep->de_StartCluster); #endif switch (vp->v_type) { case VREG: if (ioflag & IO_APPEND) uio->uio_offset = dep->de_FileSize; thisvp = vp; break; case VDIR: return EISDIR; default: panic("msdosfs_write(): bad file type"); } /* * This is needed (unlike in ffs_write()) because we extend the * file outside of the loop but we don't want to extend the file * for writes of 0 bytes. */ if (uio->uio_resid == 0) return (0); /* * The caller is supposed to ensure that * uio->uio_offset >= 0 and uio->uio_resid >= 0. */ if ((uoff_t)uio->uio_offset + uio->uio_resid > MSDOSFS_FILESIZE_MAX) return (EFBIG); /* * If they've exceeded their filesize limit, tell them about it. */ if (vn_rlimit_fsize(vp, uio, uio->uio_td)) return (EFBIG); /* * If the offset we are starting the write at is beyond the end of * the file, then they've done a seek. Unix filesystems allow * files with holes in them, DOS doesn't so we must fill the hole * with zeroed blocks. */ if (uio->uio_offset > dep->de_FileSize) { error = deextend(dep, uio->uio_offset, cred); if (error) return (error); } /* * Remember some values in case the write fails. */ resid = uio->uio_resid; osize = dep->de_FileSize; /* * If we write beyond the end of the file, extend it to its ultimate * size ahead of the time to hopefully get a contiguous area. */ if (uio->uio_offset + resid > osize) { count = de_clcount(pmp, uio->uio_offset + resid) - de_clcount(pmp, osize); error = extendfile(dep, count, NULL, NULL, 0); if (error && (error != ENOSPC || (ioflag & IO_UNIT))) goto errexit; lastcn = dep->de_fc[FC_LASTFC].fc_frcn; } else lastcn = de_clcount(pmp, osize) - 1; seqcount = ioflag >> IO_SEQSHIFT; do { if (de_cluster(pmp, uio->uio_offset) > lastcn) { error = ENOSPC; break; } croffset = uio->uio_offset & pmp->pm_crbomask; n = min(uio->uio_resid, pmp->pm_bpcluster - croffset); if (uio->uio_offset + n > dep->de_FileSize) { dep->de_FileSize = uio->uio_offset + n; /* The object size needs to be set before buffer is allocated */ vnode_pager_setsize(vp, dep->de_FileSize); } bn = de_cluster(pmp, uio->uio_offset); if ((uio->uio_offset & pmp->pm_crbomask) == 0 && (de_cluster(pmp, uio->uio_offset + uio->uio_resid) > de_cluster(pmp, uio->uio_offset) || uio->uio_offset + uio->uio_resid >= dep->de_FileSize)) { /* * If either the whole cluster gets written, * or we write the cluster from its start beyond EOF, * then no need to read data from disk. */ bp = getblk(thisvp, bn, pmp->pm_bpcluster, 0, 0, 0); /* * This call to vfs_bio_clrbuf() ensures that * even if vn_io_fault_uiomove() below faults, * garbage from the newly instantiated buffer * is not exposed to the userspace via mmap(). */ vfs_bio_clrbuf(bp); /* * Do the bmap now, since pcbmap needs buffers * for the FAT table. (see msdosfs_strategy) */ if (bp->b_blkno == bp->b_lblkno) { error = pcbmap(dep, bp->b_lblkno, &bn, 0, 0); if (error) bp->b_blkno = -1; else bp->b_blkno = bn; } if (bp->b_blkno == -1) { brelse(bp); if (!error) error = EIO; /* XXX */ break; } } else { /* * The block we need to write into exists, so read it in. */ error = bread(thisvp, bn, pmp->pm_bpcluster, cred, &bp); if (error) { break; } } /* * Should these vnode_pager_* functions be done on dir * files? */ /* * Copy the data from user space into the buf header. */ error = vn_io_fault_uiomove(bp->b_data + croffset, n, uio); if (error) { brelse(bp); break; } /* Prepare for clustered writes in some else clauses. */ if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) bp->b_flags |= B_CLUSTEROK; /* * If IO_SYNC, then each buffer is written synchronously. * Otherwise, if we have a severe page deficiency then * write the buffer asynchronously. Otherwise, if on a * cluster boundary then write the buffer asynchronously, * combining it with contiguous clusters if permitted and * possible, since we don't expect more writes into this * buffer soon. Otherwise, do a delayed write because we * expect more writes into this buffer soon. */ if (ioflag & IO_SYNC) (void)bwrite(bp); else if (vm_page_count_severe() || buf_dirty_count_severe()) bawrite(bp); else if (n + croffset == pmp->pm_bpcluster) { if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) cluster_write(vp, &dep->de_clusterw, bp, dep->de_FileSize, seqcount, 0); else bawrite(bp); } else bdwrite(bp); dep->de_flag |= DE_UPDATE; } while (error == 0 && uio->uio_resid > 0); /* * If the write failed and they want us to, truncate the file back * to the size it was before the write was attempted. */ errexit: if (error) { if (ioflag & IO_UNIT) { detrunc(dep, osize, ioflag & IO_SYNC, NOCRED); uio->uio_offset -= resid - uio->uio_resid; uio->uio_resid = resid; } else { detrunc(dep, dep->de_FileSize, ioflag & IO_SYNC, NOCRED); if (uio->uio_resid != resid) error = 0; } } else if (ioflag & IO_SYNC) error = deupdat(dep, 1); return (error); } /* * Flush the blocks of a file to disk. */ static int msdosfs_fsync(struct vop_fsync_args *ap) { struct vnode *devvp; int allerror, error; vop_stdfsync(ap); /* * If the syncing request comes from fsync(2), sync the entire * FAT and any other metadata that happens to be on devvp. We * need this mainly for the FAT. We write the FAT sloppily, and * syncing it all now is the best we can easily do to get all * directory entries associated with the file (not just the file) * fully synced. The other metadata includes critical metadata * for all directory entries, but only in the MNT_ASYNC case. We * will soon sync all metadata in the file's directory entry. * Non-critical metadata for associated directory entries only * gets synced accidentally, as in most file systems. */ if (ap->a_waitfor != MNT_NOWAIT) { devvp = VTODE(ap->a_vp)->de_pmp->pm_devvp; vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); allerror = VOP_FSYNC(devvp, MNT_WAIT, ap->a_td); VOP_UNLOCK(devvp); } else allerror = 0; error = deupdat(VTODE(ap->a_vp), ap->a_waitfor != MNT_NOWAIT); if (allerror == 0) allerror = error; return (allerror); } static int msdosfs_remove(struct vop_remove_args *ap) { struct denode *dep = VTODE(ap->a_vp); struct denode *ddep = VTODE(ap->a_dvp); int error; if (ap->a_vp->v_type == VDIR) error = EPERM; else error = removede(ddep, dep); #ifdef MSDOSFS_DEBUG printf("msdosfs_remove(), dep %p, v_usecount %d\n", dep, ap->a_vp->v_usecount); #endif return (error); } /* * DOS filesystems don't know what links are. */ static int msdosfs_link(struct vop_link_args *ap) { return (EOPNOTSUPP); } /* * Renames on files require moving the denode to a new hash queue since the * denode's location is used to compute which hash queue to put the file * in. Unless it is a rename in place. For example "mv a b". * * What follows is the basic algorithm: * * if (file move) { * if (dest file exists) { * remove dest file * } * if (dest and src in same directory) { * rewrite name in existing directory slot * } else { * write new entry in dest directory * update offset and dirclust in denode * move denode to new hash chain * clear old directory entry * } * } else { * directory move * if (dest directory exists) { * if (dest is not empty) { * return ENOTEMPTY * } * remove dest directory * } * if (dest and src in same directory) { * rewrite name in existing entry * } else { * be sure dest is not a child of src directory * write entry in dest directory * update "." and ".." in moved directory * clear old directory entry for moved directory * } * } * * On entry: * source's parent directory is unlocked * source file or directory is unlocked * destination's parent directory is locked * destination file or directory is locked if it exists * * On exit: * all denodes should be released */ static int msdosfs_rename(struct vop_rename_args *ap) { struct vnode *fdvp, *fvp, *tdvp, *tvp, *vp; struct componentname *fcnp, *tcnp; struct denode *fdip, *fip, *tdip, *tip, *nip; u_char toname[12], oldname[11]; u_long from_diroffset, to_diroffset; bool checkpath_locked, doingdirectory, newparent; u_char to_count; int error; u_long cn, pcl, blkoff; daddr_t bn, wait_scn, scn; struct msdosfsmount *pmp; struct mount *mp; struct direntry *dotdotp; struct buf *bp; tdvp = ap->a_tdvp; fvp = ap->a_fvp; fdvp = ap->a_fdvp; tvp = ap->a_tvp; tcnp = ap->a_tcnp; fcnp = ap->a_fcnp; pmp = VFSTOMSDOSFS(fdvp->v_mount); #ifdef DIAGNOSTIC if ((tcnp->cn_flags & HASBUF) == 0 || (fcnp->cn_flags & HASBUF) == 0) panic("msdosfs_rename: no name"); #endif /* * Check for cross-device rename. */ mp = fvp->v_mount; if (fvp->v_mount != tdvp->v_mount || (tvp != NULL && fvp->v_mount != tvp->v_mount)) { error = EXDEV; goto abortit; } /* * If source and dest are the same, do nothing. */ if (tvp == fvp) { error = 0; goto abortit; } /* * When the target exists, both the directory * and target vnodes are passed locked. */ VOP_UNLOCK(tdvp); if (tvp != NULL && tvp != tdvp) VOP_UNLOCK(tvp); checkpath_locked = false; relock: doingdirectory = newparent = false; error = vn_lock(fdvp, LK_EXCLUSIVE); if (error != 0) goto releout; if (vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { VOP_UNLOCK(fdvp); error = vn_lock(tdvp, LK_EXCLUSIVE); if (error != 0) goto releout; VOP_UNLOCK(tdvp); goto relock; } error = msdosfs_lookup_ino(fdvp, NULL, fcnp, &scn, &blkoff); if (error != 0) { VOP_UNLOCK(fdvp); VOP_UNLOCK(tdvp); goto releout; } error = deget(pmp, scn, blkoff, LK_EXCLUSIVE | LK_NOWAIT, &nip); if (error != 0) { VOP_UNLOCK(fdvp); VOP_UNLOCK(tdvp); if (error != EBUSY) goto releout; error = deget(pmp, scn, blkoff, LK_EXCLUSIVE, &nip); if (error != 0) goto releout; vp = fvp; fvp = DETOV(nip); VOP_UNLOCK(fvp); vrele(vp); goto relock; } vrele(fvp); fvp = DETOV(nip); from_diroffset = fdip->de_fndoffset; error = msdosfs_lookup_ino(tdvp, NULL, tcnp, &scn, &blkoff); if (error != 0 && error != EJUSTRETURN) { VOP_UNLOCK(fdvp); VOP_UNLOCK(tdvp); VOP_UNLOCK(fvp); goto releout; } if (error == EJUSTRETURN && tvp != NULL) { vrele(tvp); tvp = NULL; } if (error == 0) { nip = NULL; error = deget(pmp, scn, blkoff, LK_EXCLUSIVE | LK_NOWAIT, &nip); if (tvp != NULL) { vrele(tvp); tvp = NULL; } if (error != 0) { VOP_UNLOCK(fdvp); VOP_UNLOCK(tdvp); VOP_UNLOCK(fvp); if (error != EBUSY) goto releout; error = deget(pmp, scn, blkoff, LK_EXCLUSIVE, &nip); if (error != 0) goto releout; vput(DETOV(nip)); goto relock; } tvp = DETOV(nip); } fdip = VTODE(fdvp); fip = VTODE(fvp); tdip = VTODE(tdvp); tip = tvp != NULL ? VTODE(tvp) : NULL; /* * Remember direntry place to use for destination */ to_diroffset = tdip->de_fndoffset; to_count = tdip->de_fndcnt; /* * Be sure we are not renaming ".", "..", or an alias of ".". This * leads to a crippled directory tree. It's pretty tough to do a * "ls" or "pwd" with the "." directory entry missing, and "cd .." * doesn't work if the ".." entry is missing. */ if ((fip->de_Attributes & ATTR_DIRECTORY) != 0) { /* * Avoid ".", "..", and aliases of "." for obvious reasons. */ if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') || fdip == fip || (fcnp->cn_flags & ISDOTDOT) != 0 || (tcnp->cn_flags & ISDOTDOT) != 0) { error = EINVAL; goto unlock; } doingdirectory = true; } /* * If ".." must be changed (ie the directory gets a new * parent) then the source directory must not be in the * directory hierarchy above the target, as this would * orphan everything below the source directory. Also * the user must have write permission in the source so * as to be able to change "..". We must repeat the call * to namei, as the parent directory is unlocked by the * call to doscheckpath(). */ - error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread); + error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, curthread); if (fdip->de_StartCluster != tdip->de_StartCluster) newparent = true; if (doingdirectory && newparent) { if (error != 0) /* write access check above */ goto unlock; lockmgr(&pmp->pm_checkpath_lock, LK_EXCLUSIVE, NULL); checkpath_locked = true; error = doscheckpath(fip, tdip, &wait_scn); if (wait_scn != 0) { lockmgr(&pmp->pm_checkpath_lock, LK_RELEASE, NULL); checkpath_locked = false; VOP_UNLOCK(fdvp); VOP_UNLOCK(tdvp); VOP_UNLOCK(fvp); if (tvp != NULL && tvp != tdvp) VOP_UNLOCK(tvp); error = deget(pmp, wait_scn, 0, LK_EXCLUSIVE, &nip); if (error == 0) { vput(DETOV(nip)); goto relock; } } if (error != 0) goto unlock; if ((tcnp->cn_flags & SAVESTART) == 0) panic("msdosfs_rename: lost to startdir"); } if (tip != NULL) { /* * Target must be empty if a directory and have no links * to it. Also, ensure source and target are compatible * (both directories, or both not directories). */ if ((tip->de_Attributes & ATTR_DIRECTORY) != 0) { if (!dosdirempty(tip)) { error = ENOTEMPTY; goto unlock; } if (!doingdirectory) { error = ENOTDIR; goto unlock; } cache_purge(tdvp); } else if (doingdirectory) { error = EISDIR; goto unlock; } error = msdosfs_lookup_ino(tdvp, NULL, tcnp, &scn, &blkoff); MPASS(error == 0); error = removede(tdip, tip); if (error != 0) goto unlock; vput(tvp); tvp = NULL; tip = NULL; } /* * Convert the filename in tcnp into a dos filename. We copy this * into the denode and directory entry for the destination * file/directory. */ error = uniqdosname(tdip, tcnp, toname); if (error != 0) goto unlock; /* * First write a new entry in the destination * directory and mark the entry in the source directory * as deleted. Then move the denode to the correct hash * chain for its new location in the filesystem. And, if * we moved a directory, then update its .. entry to point * to the new parent directory. */ memcpy(oldname, fip->de_Name, 11); memcpy(fip->de_Name, toname, 11); /* update denode */ error = msdosfs_lookup_ino(tdvp, NULL, tcnp, &scn, &blkoff); MPASS(error == EJUSTRETURN); error = createde(fip, tdip, NULL, tcnp); if (error != 0) { memcpy(fip->de_Name, oldname, 11); goto unlock; } /* * If fip is for a directory, then its name should always * be "." since it is for the directory entry in the * directory itself (msdosfs_lookup() always translates * to the "." entry so as to get a unique denode, except * for the root directory there are different * complications). However, we just corrupted its name * to pass the correct name to createde(). Undo this. */ if ((fip->de_Attributes & ATTR_DIRECTORY) != 0) memcpy(fip->de_Name, oldname, 11); fip->de_refcnt++; error = msdosfs_lookup_ino(fdvp, NULL, fcnp, &scn, &blkoff); MPASS(error == 0); error = removede(fdip, fip); if (error != 0) { /* XXX should downgrade to ro here, fs is corrupt */ goto unlock; } if (!doingdirectory) { error = pcbmap(tdip, de_cluster(pmp, to_diroffset), 0, &fip->de_dirclust, 0); if (error != 0) { /* * XXX should downgrade to ro here, * fs is corrupt */ goto unlock; } if (fip->de_dirclust == MSDOSFSROOT) fip->de_diroffset = to_diroffset; else fip->de_diroffset = to_diroffset & pmp->pm_crbomask; } reinsert(fip); /* * If we moved a directory to a new parent directory, then we must * fixup the ".." entry in the moved directory. */ if (doingdirectory && newparent) { cn = fip->de_StartCluster; if (cn == MSDOSFSROOT) { /* this should never happen */ panic("msdosfs_rename(): updating .. in root directory?"); } else bn = cntobn(pmp, cn); error = bread(pmp->pm_devvp, bn, pmp->pm_bpcluster, NOCRED, &bp); if (error != 0) { /* XXX should downgrade to ro here, fs is corrupt */ goto unlock; } dotdotp = (struct direntry *)bp->b_data + 1; pcl = tdip->de_StartCluster; if (FAT32(pmp) && pcl == pmp->pm_rootdirblk) pcl = MSDOSFSROOT; putushort(dotdotp->deStartCluster, pcl); if (FAT32(pmp)) putushort(dotdotp->deHighClust, pcl >> 16); if (DOINGASYNC(fvp)) bdwrite(bp); else if ((error = bwrite(bp)) != 0) { /* XXX should downgrade to ro here, fs is corrupt */ goto unlock; } } /* * The msdosfs lookup is case insensitive. Several aliases may * be inserted for a single directory entry. As a consequnce, * name cache purge done by lookup for fvp when DELETE op for * namei is specified, might be not enough to expunge all * namecache entries that were installed for this direntry. */ cache_purge(fvp); unlock: if (checkpath_locked) lockmgr(&pmp->pm_checkpath_lock, LK_RELEASE, NULL); vput(fdvp); vput(fvp); if (tvp != NULL) { if (tvp != tdvp) vput(tvp); else vrele(tvp); } vput(tdvp); return (error); releout: MPASS(!checkpath_locked); vrele(tdvp); if (tvp != NULL) vrele(tvp); vrele(fdvp); vrele(fvp); return (error); abortit: if (tdvp == tvp) vrele(tdvp); else vput(tdvp); if (tvp != NULL) vput(tvp); vrele(fdvp); vrele(fvp); return (error); } static struct { struct direntry dot; struct direntry dotdot; } dosdirtemplate = { { ". ", /* the . entry */ ATTR_DIRECTORY, /* file attribute */ 0, /* reserved */ 0, { 0, 0 }, { 0, 0 }, /* create time & date */ { 0, 0 }, /* access date */ { 0, 0 }, /* high bits of start cluster */ { 210, 4 }, { 210, 4 }, /* modify time & date */ { 0, 0 }, /* startcluster */ { 0, 0, 0, 0 } /* filesize */ }, { ".. ", /* the .. entry */ ATTR_DIRECTORY, /* file attribute */ 0, /* reserved */ 0, { 0, 0 }, { 0, 0 }, /* create time & date */ { 0, 0 }, /* access date */ { 0, 0 }, /* high bits of start cluster */ { 210, 4 }, { 210, 4 }, /* modify time & date */ { 0, 0 }, /* startcluster */ { 0, 0, 0, 0 } /* filesize */ } }; static int msdosfs_mkdir(struct vop_mkdir_args *ap) { struct componentname *cnp = ap->a_cnp; struct denode *dep; struct denode *pdep = VTODE(ap->a_dvp); struct direntry *denp; struct msdosfsmount *pmp = pdep->de_pmp; struct buf *bp; u_long newcluster, pcl; int bn; int error; struct denode ndirent; struct timespec ts; /* * If this is the root directory and there is no space left we * can't do anything. This is because the root directory can not * change size. */ if (pdep->de_StartCluster == MSDOSFSROOT && pdep->de_fndoffset >= pdep->de_FileSize) { error = ENOSPC; goto bad2; } /* * Allocate a cluster to hold the about to be created directory. */ error = clusteralloc(pmp, 0, 1, CLUST_EOFE, &newcluster, NULL); if (error) goto bad2; memset(&ndirent, 0, sizeof(ndirent)); ndirent.de_pmp = pmp; ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE; vfs_timestamp(&ts); DETIMES(&ndirent, &ts, &ts, &ts); /* * Now fill the cluster with the "." and ".." entries. And write * the cluster to disk. This way it is there for the parent * directory to be pointing at if there were a crash. */ bn = cntobn(pmp, newcluster); /* always succeeds */ bp = getblk(pmp->pm_devvp, bn, pmp->pm_bpcluster, 0, 0, 0); memset(bp->b_data, 0, pmp->pm_bpcluster); memcpy(bp->b_data, &dosdirtemplate, sizeof dosdirtemplate); denp = (struct direntry *)bp->b_data; putushort(denp[0].deStartCluster, newcluster); putushort(denp[0].deCDate, ndirent.de_CDate); putushort(denp[0].deCTime, ndirent.de_CTime); denp[0].deCHundredth = ndirent.de_CHun; putushort(denp[0].deADate, ndirent.de_ADate); putushort(denp[0].deMDate, ndirent.de_MDate); putushort(denp[0].deMTime, ndirent.de_MTime); pcl = pdep->de_StartCluster; /* * Although the root directory has a non-magic starting cluster * number for FAT32, chkdsk and fsck_msdosfs still require * references to it in dotdot entries to be magic. */ if (FAT32(pmp) && pcl == pmp->pm_rootdirblk) pcl = MSDOSFSROOT; putushort(denp[1].deStartCluster, pcl); putushort(denp[1].deCDate, ndirent.de_CDate); putushort(denp[1].deCTime, ndirent.de_CTime); denp[1].deCHundredth = ndirent.de_CHun; putushort(denp[1].deADate, ndirent.de_ADate); putushort(denp[1].deMDate, ndirent.de_MDate); putushort(denp[1].deMTime, ndirent.de_MTime); if (FAT32(pmp)) { putushort(denp[0].deHighClust, newcluster >> 16); putushort(denp[1].deHighClust, pcl >> 16); } if (DOINGASYNC(ap->a_dvp)) bdwrite(bp); else if ((error = bwrite(bp)) != 0) goto bad; /* * Now build up a directory entry pointing to the newly allocated * cluster. This will be written to an empty slot in the parent * directory. */ #ifdef DIAGNOSTIC if ((cnp->cn_flags & HASBUF) == 0) panic("msdosfs_mkdir: no name"); #endif error = uniqdosname(pdep, cnp, ndirent.de_Name); if (error) goto bad; ndirent.de_Attributes = ATTR_DIRECTORY; ndirent.de_LowerCase = 0; ndirent.de_StartCluster = newcluster; ndirent.de_FileSize = 0; error = createde(&ndirent, pdep, &dep, cnp); if (error) goto bad; *ap->a_vpp = DETOV(dep); return (0); bad: clusterfree(pmp, newcluster, NULL); bad2: return (error); } static int msdosfs_rmdir(struct vop_rmdir_args *ap) { struct vnode *vp = ap->a_vp; struct vnode *dvp = ap->a_dvp; struct componentname *cnp = ap->a_cnp; struct denode *ip, *dp; int error; ip = VTODE(vp); dp = VTODE(dvp); /* * Verify the directory is empty (and valid). * (Rmdir ".." won't be valid since * ".." will contain a reference to * the current directory and thus be * non-empty.) */ error = 0; if (!dosdirempty(ip)) { error = ENOTEMPTY; goto out; } /* * Delete the entry from the directory. For dos filesystems this * gets rid of the directory entry on disk, the in memory copy * still exists but the de_refcnt is <= 0. This prevents it from * being found by deget(). When the vput() on dep is done we give * up access and eventually msdosfs_reclaim() will be called which * will remove it from the denode cache. */ error = removede(dp, ip); if (error) goto out; /* * This is where we decrement the link count in the parent * directory. Since dos filesystems don't do this we just purge * the name cache. */ cache_purge(dvp); /* * Truncate the directory that is being deleted. */ error = detrunc(ip, (u_long)0, IO_SYNC, cnp->cn_cred); cache_purge(vp); out: return (error); } /* * DOS filesystems don't know what symlinks are. */ static int msdosfs_symlink(struct vop_symlink_args *ap) { return (EOPNOTSUPP); } static int msdosfs_readdir(struct vop_readdir_args *ap) { struct mbnambuf nb; int error = 0; int diff; long n; int blsize; long on; u_long cn; u_long dirsperblk; long bias = 0; daddr_t bn, lbn; struct buf *bp; struct denode *dep = VTODE(ap->a_vp); struct msdosfsmount *pmp = dep->de_pmp; struct direntry *dentp; struct dirent dirbuf; struct uio *uio = ap->a_uio; u_long *cookies = NULL; int ncookies = 0; off_t offset, off; int chksum = -1; #ifdef MSDOSFS_DEBUG printf("msdosfs_readdir(): vp %p, uio %p, cred %p, eofflagp %p\n", ap->a_vp, uio, ap->a_cred, ap->a_eofflag); #endif /* * msdosfs_readdir() won't operate properly on regular files since * it does i/o only with the filesystem vnode, and hence can * retrieve the wrong block from the buffer cache for a plain file. * So, fail attempts to readdir() on a plain file. */ if ((dep->de_Attributes & ATTR_DIRECTORY) == 0) return (ENOTDIR); /* * To be safe, initialize dirbuf */ memset(dirbuf.d_name, 0, sizeof(dirbuf.d_name)); /* * If the user buffer is smaller than the size of one dos directory * entry or the file offset is not a multiple of the size of a * directory entry, then we fail the read. */ off = offset = uio->uio_offset; if (uio->uio_resid < sizeof(struct direntry) || (offset & (sizeof(struct direntry) - 1))) return (EINVAL); if (ap->a_ncookies) { ncookies = uio->uio_resid / 16; cookies = malloc(ncookies * sizeof(u_long), M_TEMP, M_WAITOK); *ap->a_cookies = cookies; *ap->a_ncookies = ncookies; } dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry); /* * If they are reading from the root directory then, we simulate * the . and .. entries since these don't exist in the root * directory. We also set the offset bias to make up for having to * simulate these entries. By this I mean that at file offset 64 we * read the first entry in the root directory that lives on disk. */ if (dep->de_StartCluster == MSDOSFSROOT || (FAT32(pmp) && dep->de_StartCluster == pmp->pm_rootdirblk)) { #if 0 printf("msdosfs_readdir(): going after . or .. in root dir, offset %d\n", offset); #endif bias = 2 * sizeof(struct direntry); if (offset < bias) { for (n = (int)offset / sizeof(struct direntry); n < 2; n++) { dirbuf.d_fileno = FAT32(pmp) ? (uint64_t)cntobn(pmp, pmp->pm_rootdirblk) * dirsperblk : 1; dirbuf.d_type = DT_DIR; switch (n) { case 0: dirbuf.d_namlen = 1; dirbuf.d_name[0] = '.'; break; case 1: dirbuf.d_namlen = 2; dirbuf.d_name[0] = '.'; dirbuf.d_name[1] = '.'; break; } dirbuf.d_reclen = GENERIC_DIRSIZ(&dirbuf); /* NOTE: d_off is the offset of the *next* entry. */ dirbuf.d_off = offset + sizeof(struct direntry); dirent_terminate(&dirbuf); if (uio->uio_resid < dirbuf.d_reclen) goto out; error = uiomove(&dirbuf, dirbuf.d_reclen, uio); if (error) goto out; offset += sizeof(struct direntry); off = offset; if (cookies) { *cookies++ = offset; if (--ncookies <= 0) goto out; } } } } mbnambuf_init(&nb); off = offset; while (uio->uio_resid > 0) { lbn = de_cluster(pmp, offset - bias); on = (offset - bias) & pmp->pm_crbomask; n = min(pmp->pm_bpcluster - on, uio->uio_resid); diff = dep->de_FileSize - (offset - bias); if (diff <= 0) break; n = min(n, diff); error = pcbmap(dep, lbn, &bn, &cn, &blsize); if (error) break; error = bread(pmp->pm_devvp, bn, blsize, NOCRED, &bp); if (error) { return (error); } n = min(n, blsize - bp->b_resid); if (n == 0) { brelse(bp); return (EIO); } /* * Convert from dos directory entries to fs-independent * directory entries. */ for (dentp = (struct direntry *)(bp->b_data + on); (char *)dentp < bp->b_data + on + n; dentp++, offset += sizeof(struct direntry)) { #if 0 printf("rd: dentp %08x prev %08x crnt %08x deName %02x attr %02x\n", dentp, prev, crnt, dentp->deName[0], dentp->deAttributes); #endif /* * If this is an unused entry, we can stop. */ if (dentp->deName[0] == SLOT_EMPTY) { brelse(bp); goto out; } /* * Skip deleted entries. */ if (dentp->deName[0] == SLOT_DELETED) { chksum = -1; mbnambuf_init(&nb); continue; } /* * Handle Win95 long directory entries */ if (dentp->deAttributes == ATTR_WIN95) { if (pmp->pm_flags & MSDOSFSMNT_SHORTNAME) continue; chksum = win2unixfn(&nb, (struct winentry *)dentp, chksum, pmp); continue; } /* * Skip volume labels */ if (dentp->deAttributes & ATTR_VOLUME) { chksum = -1; mbnambuf_init(&nb); continue; } /* * This computation of d_fileno must match * the computation of va_fileid in * msdosfs_getattr. */ if (dentp->deAttributes & ATTR_DIRECTORY) { cn = getushort(dentp->deStartCluster); if (FAT32(pmp)) { cn |= getushort(dentp->deHighClust) << 16; if (cn == MSDOSFSROOT) cn = pmp->pm_rootdirblk; } if (cn == MSDOSFSROOT && !FAT32(pmp)) dirbuf.d_fileno = 1; else dirbuf.d_fileno = cntobn(pmp, cn) * dirsperblk; dirbuf.d_type = DT_DIR; } else { dirbuf.d_fileno = (uoff_t)offset / sizeof(struct direntry); dirbuf.d_type = DT_REG; } if (chksum != winChksum(dentp->deName)) { dirbuf.d_namlen = dos2unixfn(dentp->deName, (u_char *)dirbuf.d_name, dentp->deLowerCase | ((pmp->pm_flags & MSDOSFSMNT_SHORTNAME) ? (LCASE_BASE | LCASE_EXT) : 0), pmp); mbnambuf_init(&nb); } else mbnambuf_flush(&nb, &dirbuf); chksum = -1; dirbuf.d_reclen = GENERIC_DIRSIZ(&dirbuf); /* NOTE: d_off is the offset of the *next* entry. */ dirbuf.d_off = offset + sizeof(struct direntry); dirent_terminate(&dirbuf); if (uio->uio_resid < dirbuf.d_reclen) { brelse(bp); goto out; } error = uiomove(&dirbuf, dirbuf.d_reclen, uio); if (error) { brelse(bp); goto out; } if (cookies) { *cookies++ = offset + sizeof(struct direntry); if (--ncookies <= 0) { brelse(bp); goto out; } } off = offset + sizeof(struct direntry); } brelse(bp); } out: /* Subtract unused cookies */ if (ap->a_ncookies) *ap->a_ncookies -= ncookies; uio->uio_offset = off; /* * Set the eofflag (NFS uses it) */ if (ap->a_eofflag) { if (dep->de_FileSize - (offset - bias) <= 0) *ap->a_eofflag = 1; else *ap->a_eofflag = 0; } return (error); } /*- * a_vp - pointer to the file's vnode * a_bn - logical block number within the file (cluster number for us) * a_bop - where to return the bufobj of the special file containing the fs * a_bnp - where to return the "physical" block number corresponding to a_bn * (relative to the special file; units are blocks of size DEV_BSIZE) * a_runp - where to return the "run past" a_bn. This is the count of logical * blocks whose physical blocks (together with a_bn's physical block) * are contiguous. * a_runb - where to return the "run before" a_bn. */ static int msdosfs_bmap(struct vop_bmap_args *ap) { struct fatcache savefc; struct denode *dep; struct mount *mp; struct msdosfsmount *pmp; struct vnode *vp; daddr_t runbn; u_long cn; int bnpercn, error, maxio, maxrun, run; vp = ap->a_vp; dep = VTODE(vp); pmp = dep->de_pmp; if (ap->a_bop != NULL) *ap->a_bop = &pmp->pm_devvp->v_bufobj; if (ap->a_bnp == NULL) return (0); if (ap->a_runp != NULL) *ap->a_runp = 0; if (ap->a_runb != NULL) *ap->a_runb = 0; cn = ap->a_bn; if (cn != ap->a_bn) return (EFBIG); error = pcbmap(dep, cn, ap->a_bnp, NULL, NULL); if (error != 0 || (ap->a_runp == NULL && ap->a_runb == NULL)) return (error); /* * Prepare to back out updates of the fatchain cache after the one * for the first block done by pcbmap() above. Without the backout, * then whenever the caller doesn't do i/o to all of the blocks that * we find, the single useful cache entry would be too far in advance * of the actual i/o to work for the next sequential i/o. Then the * FAT would be searched from the beginning. With the backout, the * FAT is searched starting at most a few blocks early. This wastes * much less time. Time is also wasted finding more blocks than the * caller will do i/o to. This is necessary because the runlength * parameters are output-only. */ savefc = dep->de_fc[FC_LASTMAP]; mp = vp->v_mount; maxio = mp->mnt_iosize_max / mp->mnt_stat.f_iosize; bnpercn = de_cn2bn(pmp, 1); if (ap->a_runp != NULL) { maxrun = ulmin(maxio - 1, pmp->pm_maxcluster - cn); for (run = 1; run <= maxrun; run++) { if (pcbmap(dep, cn + run, &runbn, NULL, NULL) != 0 || runbn != *ap->a_bnp + run * bnpercn) break; } *ap->a_runp = run - 1; } if (ap->a_runb != NULL) { maxrun = ulmin(maxio - 1, cn); for (run = 1; run < maxrun; run++) { if (pcbmap(dep, cn - run, &runbn, NULL, NULL) != 0 || runbn != *ap->a_bnp - run * bnpercn) break; } *ap->a_runb = run - 1; } dep->de_fc[FC_LASTMAP] = savefc; return (0); } SYSCTL_NODE(_vfs, OID_AUTO, msdosfs, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "msdos filesystem"); static int use_buf_pager = 1; SYSCTL_INT(_vfs_msdosfs, OID_AUTO, use_buf_pager, CTLFLAG_RWTUN, &use_buf_pager, 0, "Use buffer pager instead of bmap"); static daddr_t msdosfs_gbp_getblkno(struct vnode *vp, vm_ooffset_t off) { return (de_cluster(VTODE(vp)->de_pmp, off)); } static int msdosfs_gbp_getblksz(struct vnode *vp, daddr_t lbn, long *sz) { *sz = VTODE(vp)->de_pmp->pm_bpcluster; return (0); } static int msdosfs_getpages(struct vop_getpages_args *ap) { if (use_buf_pager) return (vfs_bio_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_rbehind, ap->a_rahead, msdosfs_gbp_getblkno, msdosfs_gbp_getblksz)); return (vnode_pager_generic_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_rbehind, ap->a_rahead, NULL, NULL)); } static int msdosfs_strategy(struct vop_strategy_args *ap) { struct buf *bp = ap->a_bp; struct denode *dep = VTODE(ap->a_vp); struct bufobj *bo; int error = 0; daddr_t blkno; /* * If we don't already know the filesystem relative block number * then get it using pcbmap(). If pcbmap() returns the block * number as -1 then we've got a hole in the file. DOS filesystems * don't allow files with holes, so we shouldn't ever see this. */ if (bp->b_blkno == bp->b_lblkno) { error = pcbmap(dep, bp->b_lblkno, &blkno, 0, 0); bp->b_blkno = blkno; if (error) { bp->b_error = error; bp->b_ioflags |= BIO_ERROR; bufdone(bp); return (0); } if ((long)bp->b_blkno == -1) vfs_bio_clrbuf(bp); } if (bp->b_blkno == -1) { bufdone(bp); return (0); } /* * Read/write the block from/to the disk that contains the desired * file block. */ bp->b_iooffset = dbtob(bp->b_blkno); bo = dep->de_pmp->pm_bo; BO_STRATEGY(bo, bp); return (0); } static int msdosfs_print(struct vop_print_args *ap) { struct denode *dep = VTODE(ap->a_vp); printf("\tstartcluster %lu, dircluster %lu, diroffset %lu, ", dep->de_StartCluster, dep->de_dirclust, dep->de_diroffset); printf("on dev %s\n", devtoname(dep->de_pmp->pm_dev)); return (0); } static int msdosfs_pathconf(struct vop_pathconf_args *ap) { struct msdosfsmount *pmp = VTODE(ap->a_vp)->de_pmp; switch (ap->a_name) { case _PC_FILESIZEBITS: *ap->a_retval = 32; return (0); case _PC_LINK_MAX: *ap->a_retval = 1; return (0); case _PC_NAME_MAX: *ap->a_retval = pmp->pm_flags & MSDOSFSMNT_LONGNAME ? WIN_MAXLEN : 12; return (0); case _PC_CHOWN_RESTRICTED: *ap->a_retval = 1; return (0); case _PC_NO_TRUNC: *ap->a_retval = 0; return (0); default: return (vop_stdpathconf(ap)); } /* NOTREACHED */ } static int msdosfs_vptofh(struct vop_vptofh_args *ap) { struct denode *dep; struct defid *defhp; dep = VTODE(ap->a_vp); defhp = (struct defid *)ap->a_fhp; defhp->defid_len = sizeof(struct defid); defhp->defid_dirclust = dep->de_dirclust; defhp->defid_dirofs = dep->de_diroffset; /* defhp->defid_gen = dep->de_gen; */ return (0); } /* Global vfs data structures for msdosfs */ struct vop_vector msdosfs_vnodeops = { .vop_default = &default_vnodeops, .vop_access = msdosfs_access, .vop_bmap = msdosfs_bmap, .vop_getpages = msdosfs_getpages, .vop_cachedlookup = msdosfs_lookup, .vop_open = msdosfs_open, .vop_close = msdosfs_close, .vop_create = msdosfs_create, .vop_fsync = msdosfs_fsync, .vop_fdatasync = vop_stdfdatasync_buf, .vop_getattr = msdosfs_getattr, .vop_inactive = msdosfs_inactive, .vop_link = msdosfs_link, .vop_lookup = vfs_cache_lookup, .vop_mkdir = msdosfs_mkdir, .vop_mknod = msdosfs_mknod, .vop_pathconf = msdosfs_pathconf, .vop_print = msdosfs_print, .vop_read = msdosfs_read, .vop_readdir = msdosfs_readdir, .vop_reclaim = msdosfs_reclaim, .vop_remove = msdosfs_remove, .vop_rename = msdosfs_rename, .vop_rmdir = msdosfs_rmdir, .vop_setattr = msdosfs_setattr, .vop_strategy = msdosfs_strategy, .vop_symlink = msdosfs_symlink, .vop_write = msdosfs_write, .vop_vptofh = msdosfs_vptofh, }; VFS_VOP_VECTOR_REGISTER(msdosfs_vnodeops); diff --git a/sys/fs/nfsclient/nfs_clvnops.c b/sys/fs/nfsclient/nfs_clvnops.c index 3e444780eb24..4a1ab9552a15 100644 --- a/sys/fs/nfsclient/nfs_clvnops.c +++ b/sys/fs/nfsclient/nfs_clvnops.c @@ -1,4446 +1,4446 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Rick Macklem at The University of Guelph. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from nfs_vnops.c 8.16 (Berkeley) 5/27/95 */ #include __FBSDID("$FreeBSD$"); /* * vnode op calls for Sun NFS version 2, 3 and 4 */ #include "opt_inet.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef KDTRACE_HOOKS #include dtrace_nfsclient_accesscache_flush_probe_func_t dtrace_nfscl_accesscache_flush_done_probe; uint32_t nfscl_accesscache_flush_done_id; dtrace_nfsclient_accesscache_get_probe_func_t dtrace_nfscl_accesscache_get_hit_probe, dtrace_nfscl_accesscache_get_miss_probe; uint32_t nfscl_accesscache_get_hit_id; uint32_t nfscl_accesscache_get_miss_id; dtrace_nfsclient_accesscache_load_probe_func_t dtrace_nfscl_accesscache_load_done_probe; uint32_t nfscl_accesscache_load_done_id; #endif /* !KDTRACE_HOOKS */ /* Defs */ #define TRUE 1 #define FALSE 0 extern struct nfsstatsv1 nfsstatsv1; extern int nfsrv_useacl; extern int nfscl_debuglevel; MALLOC_DECLARE(M_NEWNFSREQ); static vop_read_t nfsfifo_read; static vop_write_t nfsfifo_write; static vop_close_t nfsfifo_close; static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *, struct thread *); static vop_lookup_t nfs_lookup; static vop_create_t nfs_create; static vop_mknod_t nfs_mknod; static vop_open_t nfs_open; static vop_pathconf_t nfs_pathconf; static vop_close_t nfs_close; static vop_access_t nfs_access; static vop_getattr_t nfs_getattr; static vop_setattr_t nfs_setattr; static vop_read_t nfs_read; static vop_fsync_t nfs_fsync; static vop_remove_t nfs_remove; static vop_link_t nfs_link; static vop_rename_t nfs_rename; static vop_mkdir_t nfs_mkdir; static vop_rmdir_t nfs_rmdir; static vop_symlink_t nfs_symlink; static vop_readdir_t nfs_readdir; static vop_strategy_t nfs_strategy; static int nfs_lookitup(struct vnode *, char *, int, struct ucred *, struct thread *, struct nfsnode **); static int nfs_sillyrename(struct vnode *, struct vnode *, struct componentname *); static vop_access_t nfsspec_access; static vop_readlink_t nfs_readlink; static vop_print_t nfs_print; static vop_advlock_t nfs_advlock; static vop_advlockasync_t nfs_advlockasync; static vop_getacl_t nfs_getacl; static vop_setacl_t nfs_setacl; static vop_advise_t nfs_advise; static vop_allocate_t nfs_allocate; static vop_deallocate_t nfs_deallocate; static vop_copy_file_range_t nfs_copy_file_range; static vop_ioctl_t nfs_ioctl; static vop_getextattr_t nfs_getextattr; static vop_setextattr_t nfs_setextattr; static vop_listextattr_t nfs_listextattr; static vop_deleteextattr_t nfs_deleteextattr; static vop_lock1_t nfs_lock; /* * Global vfs data structures for nfs */ static struct vop_vector newnfs_vnodeops_nosig = { .vop_default = &default_vnodeops, .vop_access = nfs_access, .vop_advlock = nfs_advlock, .vop_advlockasync = nfs_advlockasync, .vop_close = nfs_close, .vop_create = nfs_create, .vop_fsync = nfs_fsync, .vop_getattr = nfs_getattr, .vop_getpages = ncl_getpages, .vop_putpages = ncl_putpages, .vop_inactive = ncl_inactive, .vop_link = nfs_link, .vop_lock1 = nfs_lock, .vop_lookup = nfs_lookup, .vop_mkdir = nfs_mkdir, .vop_mknod = nfs_mknod, .vop_open = nfs_open, .vop_pathconf = nfs_pathconf, .vop_print = nfs_print, .vop_read = nfs_read, .vop_readdir = nfs_readdir, .vop_readlink = nfs_readlink, .vop_reclaim = ncl_reclaim, .vop_remove = nfs_remove, .vop_rename = nfs_rename, .vop_rmdir = nfs_rmdir, .vop_setattr = nfs_setattr, .vop_strategy = nfs_strategy, .vop_symlink = nfs_symlink, .vop_write = ncl_write, .vop_getacl = nfs_getacl, .vop_setacl = nfs_setacl, .vop_advise = nfs_advise, .vop_allocate = nfs_allocate, .vop_deallocate = nfs_deallocate, .vop_copy_file_range = nfs_copy_file_range, .vop_ioctl = nfs_ioctl, .vop_getextattr = nfs_getextattr, .vop_setextattr = nfs_setextattr, .vop_listextattr = nfs_listextattr, .vop_deleteextattr = nfs_deleteextattr, }; VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops_nosig); static int nfs_vnodeops_bypass(struct vop_generic_args *a) { return (vop_sigdefer(&newnfs_vnodeops_nosig, a)); } struct vop_vector newnfs_vnodeops = { .vop_default = &default_vnodeops, .vop_bypass = nfs_vnodeops_bypass, }; VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops); static struct vop_vector newnfs_fifoops_nosig = { .vop_default = &fifo_specops, .vop_access = nfsspec_access, .vop_close = nfsfifo_close, .vop_fsync = nfs_fsync, .vop_getattr = nfs_getattr, .vop_inactive = ncl_inactive, .vop_pathconf = nfs_pathconf, .vop_print = nfs_print, .vop_read = nfsfifo_read, .vop_reclaim = ncl_reclaim, .vop_setattr = nfs_setattr, .vop_write = nfsfifo_write, }; VFS_VOP_VECTOR_REGISTER(newnfs_fifoops_nosig); static int nfs_fifoops_bypass(struct vop_generic_args *a) { return (vop_sigdefer(&newnfs_fifoops_nosig, a)); } struct vop_vector newnfs_fifoops = { .vop_default = &default_vnodeops, .vop_bypass = nfs_fifoops_bypass, }; VFS_VOP_VECTOR_REGISTER(newnfs_fifoops); static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct vattr *vap); static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, int namelen, struct ucred *cred, struct thread *td); static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td); static int nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp, struct sillyrename *sp); /* * Global variables */ SYSCTL_DECL(_vfs_nfs); static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO; SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); static int nfs_prime_access_cache = 0; SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW, &nfs_prime_access_cache, 0, "Prime NFS ACCESS cache when fetching attributes"); static int newnfs_commit_on_close = 0; SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW, &newnfs_commit_on_close, 0, "write+commit on close, else only write"); static int nfs_clean_pages_on_close = 1; SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW, &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close"); int newnfs_directio_enable = 0; SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW, &newnfs_directio_enable, 0, "Enable NFS directio"); int nfs_keep_dirty_on_error; SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW, &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned"); /* * This sysctl allows other processes to mmap a file that has been opened * O_DIRECT by a process. In general, having processes mmap the file while * Direct IO is in progress can lead to Data Inconsistencies. But, we allow * this by default to prevent DoS attacks - to prevent a malicious user from * opening up files O_DIRECT preventing other users from mmap'ing these * files. "Protected" environments where stricter consistency guarantees are * required can disable this knob. The process that opened the file O_DIRECT * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not * meaningful. */ int newnfs_directio_allow_mmap = 1; SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW, &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens"); static uint64_t nfs_maxalloclen = 64 * 1024 * 1024; SYSCTL_U64(_vfs_nfs, OID_AUTO, maxalloclen, CTLFLAG_RW, &nfs_maxalloclen, 0, "NFS max allocate/deallocate length"); #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \ | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \ | NFSACCESS_DELETE | NFSACCESS_LOOKUP) /* * SMP Locking Note : * The list of locks after the description of the lock is the ordering * of other locks acquired with the lock held. * np->n_mtx : Protects the fields in the nfsnode. VM Object Lock VI_MTX (acquired indirectly) * nmp->nm_mtx : Protects the fields in the nfsmount. rep->r_mtx * ncl_iod_mutex : Global lock, protects shared nfsiod state. * nfs_reqq_mtx : Global lock, protects the nfs_reqq list. nmp->nm_mtx rep->r_mtx * rep->r_mtx : Protects the fields in an nfsreq. */ static int nfs_lock(struct vop_lock1_args *ap) { struct vnode *vp; struct nfsnode *np; u_quad_t nsize; int error, lktype; bool onfault; vp = ap->a_vp; lktype = ap->a_flags & LK_TYPE_MASK; error = VOP_LOCK1_APV(&default_vnodeops, ap); if (error != 0 || vp->v_op != &newnfs_vnodeops) return (error); np = VTONFS(vp); if (np == NULL) return (0); NFSLOCKNODE(np); if ((np->n_flag & NVNSETSZSKIP) == 0 || (lktype != LK_SHARED && lktype != LK_EXCLUSIVE && lktype != LK_UPGRADE && lktype != LK_TRYUPGRADE)) { NFSUNLOCKNODE(np); return (0); } onfault = (ap->a_flags & LK_EATTR_MASK) == LK_NOWAIT && (ap->a_flags & LK_INIT_MASK) == LK_CANRECURSE && (lktype == LK_SHARED || lktype == LK_EXCLUSIVE); if (onfault && vp->v_vnlock->lk_recurse == 0) { /* * Force retry in vm_fault(), to make the lock request * sleepable, which allows us to piggy-back the * sleepable call to vnode_pager_setsize(). */ NFSUNLOCKNODE(np); VOP_UNLOCK(vp); return (EBUSY); } if ((ap->a_flags & LK_NOWAIT) != 0 || (lktype == LK_SHARED && vp->v_vnlock->lk_recurse > 0)) { NFSUNLOCKNODE(np); return (0); } if (lktype == LK_SHARED) { NFSUNLOCKNODE(np); VOP_UNLOCK(vp); ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK); ap->a_flags |= LK_EXCLUSIVE; error = VOP_LOCK1_APV(&default_vnodeops, ap); if (error != 0 || vp->v_op != &newnfs_vnodeops) return (error); if (vp->v_data == NULL) goto downgrade; MPASS(vp->v_data == np); NFSLOCKNODE(np); if ((np->n_flag & NVNSETSZSKIP) == 0) { NFSUNLOCKNODE(np); goto downgrade; } } np->n_flag &= ~NVNSETSZSKIP; nsize = np->n_size; NFSUNLOCKNODE(np); vnode_pager_setsize(vp, nsize); downgrade: if (lktype == LK_SHARED) { ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK); ap->a_flags |= LK_DOWNGRADE; (void)VOP_LOCK1_APV(&default_vnodeops, ap); } return (0); } static int nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td, struct ucred *cred, u_int32_t *retmode) { int error = 0, attrflag, i, lrupos; u_int32_t rmode; struct nfsnode *np = VTONFS(vp); struct nfsvattr nfsva; error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag, &rmode, NULL); if (attrflag) (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); if (!error) { lrupos = 0; NFSLOCKNODE(np); for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { if (np->n_accesscache[i].uid == cred->cr_uid) { np->n_accesscache[i].mode = rmode; np->n_accesscache[i].stamp = time_second; break; } if (i > 0 && np->n_accesscache[i].stamp < np->n_accesscache[lrupos].stamp) lrupos = i; } if (i == NFS_ACCESSCACHESIZE) { np->n_accesscache[lrupos].uid = cred->cr_uid; np->n_accesscache[lrupos].mode = rmode; np->n_accesscache[lrupos].stamp = time_second; } NFSUNLOCKNODE(np); if (retmode != NULL) *retmode = rmode; KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0); } else if (NFS_ISV4(vp)) { error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); } #ifdef KDTRACE_HOOKS if (error != 0) KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0, error); #endif return (error); } /* * nfs access vnode op. * For nfs version 2, just return ok. File accesses may fail later. * For nfs version 3, use the access rpc to check accessibility. If file modes * are changed on the server, accesses might still fail later. */ static int nfs_access(struct vop_access_args *ap) { struct vnode *vp = ap->a_vp; int error = 0, i, gotahit; u_int32_t mode, wmode, rmode; int v34 = NFS_ISV34(vp); struct nfsnode *np = VTONFS(vp); /* * Disallow write attempts on filesystems mounted read-only; * unless the file is a socket, fifo, or a block or character * device resident on the filesystem. */ if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS | VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL | VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) { switch (vp->v_type) { case VREG: case VDIR: case VLNK: return (EROFS); default: break; } } /* * For nfs v3 or v4, check to see if we have done this recently, and if * so return our cached result instead of making an ACCESS call. * If not, do an access rpc, otherwise you are stuck emulating * ufs_access() locally using the vattr. This may not be correct, * since the server may apply other access criteria such as * client uid-->server uid mapping that we do not know about. */ if (v34) { if (ap->a_accmode & VREAD) mode = NFSACCESS_READ; else mode = 0; if (vp->v_type != VDIR) { if (ap->a_accmode & VWRITE) mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); if (ap->a_accmode & VAPPEND) mode |= NFSACCESS_EXTEND; if (ap->a_accmode & VEXEC) mode |= NFSACCESS_EXECUTE; if (ap->a_accmode & VDELETE) mode |= NFSACCESS_DELETE; } else { if (ap->a_accmode & VWRITE) mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); if (ap->a_accmode & VAPPEND) mode |= NFSACCESS_EXTEND; if (ap->a_accmode & VEXEC) mode |= NFSACCESS_LOOKUP; if (ap->a_accmode & VDELETE) mode |= NFSACCESS_DELETE; if (ap->a_accmode & VDELETE_CHILD) mode |= NFSACCESS_MODIFY; } /* XXX safety belt, only make blanket request if caching */ if (nfsaccess_cache_timeout > 0) { wmode = NFSACCESS_READ | NFSACCESS_MODIFY | NFSACCESS_EXTEND | NFSACCESS_EXECUTE | NFSACCESS_DELETE | NFSACCESS_LOOKUP; } else { wmode = mode; } /* * Does our cached result allow us to give a definite yes to * this request? */ gotahit = 0; NFSLOCKNODE(np); for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) { if (time_second < (np->n_accesscache[i].stamp + nfsaccess_cache_timeout) && (np->n_accesscache[i].mode & mode) == mode) { NFSINCRGLOBAL(nfsstatsv1.accesscache_hits); gotahit = 1; } break; } } NFSUNLOCKNODE(np); #ifdef KDTRACE_HOOKS if (gotahit != 0) KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp, ap->a_cred->cr_uid, mode); else KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp, ap->a_cred->cr_uid, mode); #endif if (gotahit == 0) { /* * Either a no, or a don't know. Go to the wire. */ NFSINCRGLOBAL(nfsstatsv1.accesscache_misses); error = nfs34_access_otw(vp, wmode, ap->a_td, ap->a_cred, &rmode); if (!error && (rmode & mode) != mode) error = EACCES; } return (error); } else { if ((error = nfsspec_access(ap)) != 0) { return (error); } /* * Attempt to prevent a mapped root from accessing a file * which it shouldn't. We try to read a byte from the file * if the user is root and the file is not zero length. * After calling nfsspec_access, we should have the correct * file size cached. */ NFSLOCKNODE(np); if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD) && VTONFS(vp)->n_size > 0) { struct iovec aiov; struct uio auio; char buf[1]; NFSUNLOCKNODE(np); aiov.iov_base = buf; aiov.iov_len = 1; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_offset = 0; auio.uio_resid = 1; auio.uio_segflg = UIO_SYSSPACE; auio.uio_rw = UIO_READ; auio.uio_td = ap->a_td; if (vp->v_type == VREG) error = ncl_readrpc(vp, &auio, ap->a_cred); else if (vp->v_type == VDIR) { char* bp; bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); aiov.iov_base = bp; aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; error = ncl_readdirrpc(vp, &auio, ap->a_cred, ap->a_td); free(bp, M_TEMP); } else if (vp->v_type == VLNK) error = ncl_readlinkrpc(vp, &auio, ap->a_cred); else error = EACCES; } else NFSUNLOCKNODE(np); return (error); } } /* * nfs open vnode op * Check to see if the type is ok * and that deletion is not in progress. * For paged in text files, you will need to flush the page cache * if consistency is lost. */ /* ARGSUSED */ static int nfs_open(struct vop_open_args *ap) { struct vnode *vp = ap->a_vp; struct nfsnode *np = VTONFS(vp); struct vattr vattr; int error; int fmode = ap->a_mode; struct ucred *cred; vm_object_t obj; if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) return (EOPNOTSUPP); /* * For NFSv4, we need to do the Open Op before cache validation, * so that we conform to RFC3530 Sec. 9.3.1. */ if (NFS_ISV4(vp)) { error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td); if (error) { error = nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); return (error); } } /* * Now, if this Open will be doing reading, re-validate/flush the * cache, so that Close/Open coherency is maintained. */ NFSLOCKNODE(np); if (np->n_flag & NMODIFIED) { NFSUNLOCKNODE(np); error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); if (error == EINTR || error == EIO) { if (NFS_ISV4(vp)) (void) nfsrpc_close(vp, 0, ap->a_td); return (error); } NFSLOCKNODE(np); np->n_attrstamp = 0; KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); if (vp->v_type == VDIR) np->n_direofoffset = 0; NFSUNLOCKNODE(np); error = VOP_GETATTR(vp, &vattr, ap->a_cred); if (error) { if (NFS_ISV4(vp)) (void) nfsrpc_close(vp, 0, ap->a_td); return (error); } NFSLOCKNODE(np); np->n_mtime = vattr.va_mtime; if (NFS_ISV4(vp)) np->n_change = vattr.va_filerev; } else { NFSUNLOCKNODE(np); error = VOP_GETATTR(vp, &vattr, ap->a_cred); if (error) { if (NFS_ISV4(vp)) (void) nfsrpc_close(vp, 0, ap->a_td); return (error); } NFSLOCKNODE(np); if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) || NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { if (vp->v_type == VDIR) np->n_direofoffset = 0; NFSUNLOCKNODE(np); error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); if (error == EINTR || error == EIO) { if (NFS_ISV4(vp)) (void) nfsrpc_close(vp, 0, ap->a_td); return (error); } NFSLOCKNODE(np); np->n_mtime = vattr.va_mtime; if (NFS_ISV4(vp)) np->n_change = vattr.va_filerev; } } /* * If the object has >= 1 O_DIRECT active opens, we disable caching. */ if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) { if (np->n_directio_opens == 0) { NFSUNLOCKNODE(np); error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); if (error) { if (NFS_ISV4(vp)) (void) nfsrpc_close(vp, 0, ap->a_td); return (error); } NFSLOCKNODE(np); np->n_flag |= NNONCACHE; } np->n_directio_opens++; } /* If opened for writing via NFSv4.1 or later, mark that for pNFS. */ if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0) np->n_flag |= NWRITEOPENED; /* * If this is an open for writing, capture a reference to the * credentials, so they can be used by ncl_putpages(). Using * these write credentials is preferable to the credentials of * whatever thread happens to be doing the VOP_PUTPAGES() since * the write RPCs are less likely to fail with EACCES. */ if ((fmode & FWRITE) != 0) { cred = np->n_writecred; np->n_writecred = crhold(ap->a_cred); } else cred = NULL; NFSUNLOCKNODE(np); if (cred != NULL) crfree(cred); vnode_create_vobject(vp, vattr.va_size, ap->a_td); /* * If the text file has been mmap'd, flush any dirty pages to the * buffer cache and then... * Make sure all writes are pushed to the NFS server. If this is not * done, the modify time of the file can change while the text * file is being executed. This will cause the process that is * executing the text file to be terminated. */ if (vp->v_writecount <= -1) { if ((obj = vp->v_object) != NULL && vm_object_mightbedirty(obj)) { VM_OBJECT_WLOCK(obj); vm_object_page_clean(obj, 0, 0, OBJPC_SYNC); VM_OBJECT_WUNLOCK(obj); } /* Now, flush the buffer cache. */ ncl_flush(vp, MNT_WAIT, curthread, 0, 0); /* And, finally, make sure that n_mtime is up to date. */ np = VTONFS(vp); NFSLOCKNODE(np); np->n_mtime = np->n_vattr.na_mtime; NFSUNLOCKNODE(np); } return (0); } /* * nfs close vnode op * What an NFS client should do upon close after writing is a debatable issue. * Most NFS clients push delayed writes to the server upon close, basically for * two reasons: * 1 - So that any write errors may be reported back to the client process * doing the close system call. By far the two most likely errors are * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. * 2 - To put a worst case upper bound on cache inconsistency between * multiple clients for the file. * There is also a consistency problem for Version 2 of the protocol w.r.t. * not being able to tell if other clients are writing a file concurrently, * since there is no way of knowing if the changed modify time in the reply * is only due to the write for this client. * (NFS Version 3 provides weak cache consistency data in the reply that * should be sufficient to detect and handle this case.) * * The current code does the following: * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers * for NFS Version 3 - flush dirty buffers to the server but don't invalidate * or commit them (this satisfies 1 and 2 except for the * case where the server crashes after this close but * before the commit RPC, which is felt to be "good * enough". Changing the last argument to ncl_flush() to * a 1 would force a commit operation, if it is felt a * commit is necessary now. * for NFS Version 4 - flush the dirty buffers and commit them, if * nfscl_mustflush() says this is necessary. * It is necessary if there is no write delegation held, * in order to satisfy open/close coherency. * If the file isn't cached on local stable storage, * it may be necessary in order to detect "out of space" * errors from the server, if the write delegation * issued by the server doesn't allow the file to grow. */ /* ARGSUSED */ static int nfs_close(struct vop_close_args *ap) { struct vnode *vp = ap->a_vp; struct nfsnode *np = VTONFS(vp); struct nfsvattr nfsva; struct ucred *cred; int error = 0, ret, localcred = 0; int fmode = ap->a_fflag; if (NFSCL_FORCEDISM(vp->v_mount)) return (0); /* * During shutdown, a_cred isn't valid, so just use root. */ if (ap->a_cred == NOCRED) { cred = newnfs_getcred(); localcred = 1; } else { cred = ap->a_cred; } if (vp->v_type == VREG) { /* * Examine and clean dirty pages, regardless of NMODIFIED. * This closes a major hole in close-to-open consistency. * We want to push out all dirty pages (and buffers) on * close, regardless of whether they were dirtied by * mmap'ed writes or via write(). */ if (nfs_clean_pages_on_close && vp->v_object) { VM_OBJECT_WLOCK(vp->v_object); vm_object_page_clean(vp->v_object, 0, 0, 0); VM_OBJECT_WUNLOCK(vp->v_object); } NFSLOCKNODE(np); if (np->n_flag & NMODIFIED) { NFSUNLOCKNODE(np); if (NFS_ISV3(vp)) { /* * Under NFSv3 we have dirty buffers to dispose of. We * must flush them to the NFS server. We have the option * of waiting all the way through the commit rpc or just * waiting for the initial write. The default is to only * wait through the initial write so the data is in the * server's cache, which is roughly similar to the state * a standard disk subsystem leaves the file in on close(). * * We cannot clear the NMODIFIED bit in np->n_flag due to * potential races with other processes, and certainly * cannot clear it if we don't commit. * These races occur when there is no longer the old * traditional vnode locking implemented for Vnode Ops. */ int cm = newnfs_commit_on_close ? 1 : 0; error = ncl_flush(vp, MNT_WAIT, ap->a_td, cm, 0); /* np->n_flag &= ~NMODIFIED; */ } else if (NFS_ISV4(vp)) { if (nfscl_mustflush(vp) != 0) { int cm = newnfs_commit_on_close ? 1 : 0; error = ncl_flush(vp, MNT_WAIT, ap->a_td, cm, 0); /* * as above w.r.t races when clearing * NMODIFIED. * np->n_flag &= ~NMODIFIED; */ } } else { error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); } NFSLOCKNODE(np); } /* * Invalidate the attribute cache in all cases. * An open is going to fetch fresh attrs any way, other procs * on this node that have file open will be forced to do an * otw attr fetch, but this is safe. * --> A user found that their RPC count dropped by 20% when * this was commented out and I can't see any requirement * for it, so I've disabled it when negative lookups are * enabled. (What does this have to do with negative lookup * caching? Well nothing, except it was reported by the * same user that needed negative lookup caching and I wanted * there to be a way to disable it to see if it * is the cause of some caching/coherency issue that might * crop up.) */ if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) { np->n_attrstamp = 0; KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); } if (np->n_flag & NWRITEERR) { np->n_flag &= ~NWRITEERR; error = np->n_error; } NFSUNLOCKNODE(np); } if (NFS_ISV4(vp)) { /* * Get attributes so "change" is up to date. */ if (error == 0 && nfscl_mustflush(vp) != 0 && vp->v_type == VREG && (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) { ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva, NULL); if (!ret) { np->n_change = nfsva.na_filerev; (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 0); } } /* * and do the close. */ ret = nfsrpc_close(vp, 0, ap->a_td); if (!error && ret) error = ret; if (error) error = nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); } if (newnfs_directio_enable) KASSERT((np->n_directio_asyncwr == 0), ("nfs_close: dirty unflushed (%d) directio buffers\n", np->n_directio_asyncwr)); if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) { NFSLOCKNODE(np); KASSERT((np->n_directio_opens > 0), ("nfs_close: unexpectedly value (0) of n_directio_opens\n")); np->n_directio_opens--; if (np->n_directio_opens == 0) np->n_flag &= ~NNONCACHE; NFSUNLOCKNODE(np); } if (localcred) NFSFREECRED(cred); return (error); } /* * nfs getattr call from vfs. */ static int nfs_getattr(struct vop_getattr_args *ap) { struct vnode *vp = ap->a_vp; struct thread *td = curthread; /* XXX */ struct nfsnode *np = VTONFS(vp); int error = 0; struct nfsvattr nfsva; struct vattr *vap = ap->a_vap; struct vattr vattr; /* * Update local times for special files. */ NFSLOCKNODE(np); if (np->n_flag & (NACC | NUPD)) np->n_flag |= NCHG; NFSUNLOCKNODE(np); /* * First look in the cache. */ if (ncl_getattrcache(vp, &vattr) == 0) { ncl_copy_vattr(vap, &vattr); /* * Get the local modify time for the case of a write * delegation. */ nfscl_deleggetmodtime(vp, &vap->va_mtime); return (0); } if (NFS_ISV34(vp) && nfs_prime_access_cache && nfsaccess_cache_timeout > 0) { NFSINCRGLOBAL(nfsstatsv1.accesscache_misses); nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL); if (ncl_getattrcache(vp, ap->a_vap) == 0) { nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime); return (0); } } error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL); if (!error) error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0); if (!error) { /* * Get the local modify time for the case of a write * delegation. */ nfscl_deleggetmodtime(vp, &vap->va_mtime); } else if (NFS_ISV4(vp)) { error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); } return (error); } /* * nfs setattr call. */ static int nfs_setattr(struct vop_setattr_args *ap) { struct vnode *vp = ap->a_vp; struct nfsnode *np = VTONFS(vp); struct thread *td = curthread; /* XXX */ struct vattr *vap = ap->a_vap; int error = 0; u_quad_t tsize; #ifndef nolint tsize = (u_quad_t)0; #endif /* * Setting of flags and marking of atimes are not supported. */ if (vap->va_flags != VNOVAL) return (EOPNOTSUPP); /* * Disallow write attempts if the filesystem is mounted read-only. */ if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL || vap->va_birthtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && (vp->v_mount->mnt_flag & MNT_RDONLY)) return (EROFS); if (vap->va_size != VNOVAL) { switch (vp->v_type) { case VDIR: return (EISDIR); case VCHR: case VBLK: case VSOCK: case VFIFO: if (vap->va_mtime.tv_sec == VNOVAL && vap->va_atime.tv_sec == VNOVAL && vap->va_birthtime.tv_sec == VNOVAL && vap->va_mode == (mode_t)VNOVAL && vap->va_uid == (uid_t)VNOVAL && vap->va_gid == (gid_t)VNOVAL) return (0); vap->va_size = VNOVAL; break; default: /* * Disallow write attempts if the filesystem is * mounted read-only. */ if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); /* * We run vnode_pager_setsize() early (why?), * we must set np->n_size now to avoid vinvalbuf * V_SAVE races that might setsize a lower * value. */ NFSLOCKNODE(np); tsize = np->n_size; NFSUNLOCKNODE(np); error = ncl_meta_setsize(vp, td, vap->va_size); NFSLOCKNODE(np); if (np->n_flag & NMODIFIED) { tsize = np->n_size; NFSUNLOCKNODE(np); error = ncl_vinvalbuf(vp, vap->va_size == 0 ? 0 : V_SAVE, td, 1); if (error != 0) { vnode_pager_setsize(vp, tsize); return (error); } /* * Call nfscl_delegmodtime() to set the modify time * locally, as required. */ nfscl_delegmodtime(vp); } else NFSUNLOCKNODE(np); /* * np->n_size has already been set to vap->va_size * in ncl_meta_setsize(). We must set it again since * nfs_loadattrcache() could be called through * ncl_meta_setsize() and could modify np->n_size. */ NFSLOCKNODE(np); np->n_vattr.na_size = np->n_size = vap->va_size; NFSUNLOCKNODE(np); } } else { NFSLOCKNODE(np); if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NMODIFIED) && vp->v_type == VREG) { NFSUNLOCKNODE(np); error = ncl_vinvalbuf(vp, V_SAVE, td, 1); if (error == EINTR || error == EIO) return (error); } else NFSUNLOCKNODE(np); } error = nfs_setattrrpc(vp, vap, ap->a_cred, td); if (error && vap->va_size != VNOVAL) { NFSLOCKNODE(np); np->n_size = np->n_vattr.na_size = tsize; vnode_pager_setsize(vp, tsize); NFSUNLOCKNODE(np); } return (error); } /* * Do an nfs setattr rpc. */ static int nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred, struct thread *td) { struct nfsnode *np = VTONFS(vp); int error, ret, attrflag, i; struct nfsvattr nfsva; if (NFS_ISV34(vp)) { NFSLOCKNODE(np); for (i = 0; i < NFS_ACCESSCACHESIZE; i++) np->n_accesscache[i].stamp = 0; np->n_flag |= NDELEGMOD; NFSUNLOCKNODE(np); KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp); } error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag, NULL); if (attrflag) { ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); if (ret && !error) error = ret; } if (error && NFS_ISV4(vp)) error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid); return (error); } /* * nfs lookup call, one step at a time... * First look in cache * If not found, unlock the directory nfsnode and do the rpc */ static int nfs_lookup(struct vop_lookup_args *ap) { struct componentname *cnp = ap->a_cnp; struct vnode *dvp = ap->a_dvp; struct vnode **vpp = ap->a_vpp; struct mount *mp = dvp->v_mount; int flags = cnp->cn_flags; struct vnode *newvp; struct nfsmount *nmp; struct nfsnode *np, *newnp; int error = 0, attrflag, dattrflag, ltype, ncticks; - struct thread *td = cnp->cn_thread; + struct thread *td = curthread; struct nfsfh *nfhp; struct nfsvattr dnfsva, nfsva; struct vattr vattr; struct timespec nctime; uint32_t openmode; *vpp = NULLVP; if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) return (EROFS); if (dvp->v_type != VDIR) return (ENOTDIR); nmp = VFSTONFS(mp); np = VTONFS(dvp); /* For NFSv4, wait until any remove is done. */ NFSLOCKNODE(np); while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) { np->n_flag |= NREMOVEWANT; (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0); } NFSUNLOCKNODE(np); error = vn_dir_check_exec(dvp, cnp); if (error != 0) return (error); error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks); if (error > 0 && error != ENOENT) return (error); if (error == -1) { /* * Lookups of "." are special and always return the * current directory. cache_lookup() already handles * associated locking bookkeeping, etc. */ if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { /* XXX: Is this really correct? */ if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) cnp->cn_flags |= SAVENAME; return (0); } /* * We only accept a positive hit in the cache if the * change time of the file matches our cached copy. * Otherwise, we discard the cache entry and fallback * to doing a lookup RPC. We also only trust cache * entries for less than nm_nametimeo seconds. * * To better handle stale file handles and attributes, * clear the attribute cache of this node if it is a * leaf component, part of an open() call, and not * locally modified before fetching the attributes. * This should allow stale file handles to be detected * here where we can fall back to a LOOKUP RPC to * recover rather than having nfs_open() detect the * stale file handle and failing open(2) with ESTALE. */ newvp = *vpp; newnp = VTONFS(newvp); if (!(nmp->nm_flag & NFSMNT_NOCTO) && (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && !(newnp->n_flag & NMODIFIED)) { NFSLOCKNODE(newnp); newnp->n_attrstamp = 0; KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); NFSUNLOCKNODE(newnp); } if (nfscl_nodeleg(newvp, 0) == 0 || ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) && VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 && timespeccmp(&vattr.va_ctime, &nctime, ==))) { NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits); if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) cnp->cn_flags |= SAVENAME; return (0); } cache_purge(newvp); if (dvp != newvp) vput(newvp); else vrele(newvp); *vpp = NULLVP; } else if (error == ENOENT) { if (VN_IS_DOOMED(dvp)) return (ENOENT); /* * We only accept a negative hit in the cache if the * modification time of the parent directory matches * the cached copy in the name cache entry. * Otherwise, we discard all of the negative cache * entries for this directory. We also only trust * negative cache entries for up to nm_negnametimeo * seconds. */ if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) && VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 && timespeccmp(&vattr.va_mtime, &nctime, ==)) { NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits); return (ENOENT); } cache_purge_negative(dvp); } /* * If this an NFSv4.1/4.2 mount using the "oneopenown" mount * option, it is possible to do the Open operation in the same * compound as Lookup, so long as delegations are not being * issued. This saves doing a separate RPC for Open. */ openmode = 0; NFSLOCKMNT(nmp); if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp) && (nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0 && (!NFSMNT_RDONLY(mp) || (flags & OPENWRITE) == 0) && (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN)) { if ((flags & OPENREAD) != 0) openmode |= NFSV4OPEN_ACCESSREAD; if ((flags & OPENWRITE) != 0) openmode |= NFSV4OPEN_ACCESSWRITE; } NFSUNLOCKMNT(nmp); newvp = NULLVP; NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses); error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, NULL, openmode); if (dattrflag) (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); if (error) { if (newvp != NULLVP) { vput(newvp); *vpp = NULLVP; } if (error != ENOENT) { if (NFS_ISV4(dvp)) error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); return (error); } /* The requested file was not found. */ if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && (flags & ISLASTCN)) { /* * XXX: UFS does a full VOP_ACCESS(dvp, * VWRITE) here instead of just checking * MNT_RDONLY. */ if (mp->mnt_flag & MNT_RDONLY) return (EROFS); cnp->cn_flags |= SAVENAME; return (EJUSTRETURN); } if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) { /* * Cache the modification time of the parent * directory from the post-op attributes in * the name cache entry. The negative cache * entry will be ignored once the directory * has changed. Don't bother adding the entry * if the directory has already changed. */ NFSLOCKNODE(np); if (timespeccmp(&np->n_vattr.na_mtime, &dnfsva.na_mtime, ==)) { NFSUNLOCKNODE(np); cache_enter_time(dvp, NULL, cnp, &dnfsva.na_mtime, NULL); } else NFSUNLOCKNODE(np); } return (ENOENT); } /* * Handle RENAME case... */ if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { free(nfhp, M_NFSFH); return (EISDIR); } error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, LK_EXCLUSIVE); if (error) return (error); newvp = NFSTOV(np); if (attrflag) (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 0, 1); *vpp = newvp; cnp->cn_flags |= SAVENAME; return (0); } if (flags & ISDOTDOT) { ltype = NFSVOPISLOCKED(dvp); error = vfs_busy(mp, MBF_NOWAIT); if (error != 0) { vfs_ref(mp); NFSVOPUNLOCK(dvp); error = vfs_busy(mp, 0); NFSVOPLOCK(dvp, ltype | LK_RETRY); vfs_rel(mp); if (error == 0 && VN_IS_DOOMED(dvp)) { vfs_unbusy(mp); error = ENOENT; } if (error != 0) return (error); } NFSVOPUNLOCK(dvp); error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, cnp->cn_lkflags); if (error == 0) newvp = NFSTOV(np); vfs_unbusy(mp); if (newvp != dvp) NFSVOPLOCK(dvp, ltype | LK_RETRY); if (VN_IS_DOOMED(dvp)) { if (error == 0) { if (newvp == dvp) vrele(newvp); else vput(newvp); } error = ENOENT; } if (error != 0) return (error); if (attrflag) (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 0, 1); } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { free(nfhp, M_NFSFH); VREF(dvp); newvp = dvp; if (attrflag) (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 0, 1); } else { error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, cnp->cn_lkflags); if (error) return (error); newvp = NFSTOV(np); if (attrflag) (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 0, 1); else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && !(np->n_flag & NMODIFIED)) { /* * Flush the attribute cache when opening a * leaf node to ensure that fresh attributes * are fetched in nfs_open() since we did not * fetch attributes from the LOOKUP reply. */ NFSLOCKNODE(np); np->n_attrstamp = 0; KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); NFSUNLOCKNODE(np); } } if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) cnp->cn_flags |= SAVENAME; if ((cnp->cn_flags & MAKEENTRY) && dvp != newvp && (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) && attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0)) cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime); *vpp = newvp; return (0); } /* * nfs read call. * Just call ncl_bioread() to do the work. */ static int nfs_read(struct vop_read_args *ap) { struct vnode *vp = ap->a_vp; switch (vp->v_type) { case VREG: return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred)); case VDIR: return (EISDIR); default: return (EOPNOTSUPP); } } /* * nfs readlink call */ static int nfs_readlink(struct vop_readlink_args *ap) { struct vnode *vp = ap->a_vp; if (vp->v_type != VLNK) return (EINVAL); return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred)); } /* * Do a readlink rpc. * Called by ncl_doio() from below the buffer cache. */ int ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) { int error, ret, attrflag; struct nfsvattr nfsva; error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva, &attrflag, NULL); if (attrflag) { ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); if (ret && !error) error = ret; } if (error && NFS_ISV4(vp)) error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); return (error); } /* * nfs read rpc call * Ditto above */ int ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) { int error, ret, attrflag; struct nfsvattr nfsva; struct nfsmount *nmp; nmp = VFSTONFS(vp->v_mount); error = EIO; attrflag = 0; if (NFSHASPNFS(nmp)) error = nfscl_doiods(vp, uiop, NULL, NULL, NFSV4OPEN_ACCESSREAD, 0, cred, uiop->uio_td); NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error); if (error != 0) error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, &attrflag, NULL); if (attrflag) { ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); if (ret && !error) error = ret; } if (error && NFS_ISV4(vp)) error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); return (error); } /* * nfs write call */ int ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, int *iomode, int *must_commit, int called_from_strategy) { struct nfsvattr nfsva; int error, attrflag, ret; struct nfsmount *nmp; nmp = VFSTONFS(vp->v_mount); error = EIO; attrflag = 0; if (NFSHASPNFS(nmp)) error = nfscl_doiods(vp, uiop, iomode, must_commit, NFSV4OPEN_ACCESSWRITE, 0, cred, uiop->uio_td); NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error); if (error != 0) error = nfsrpc_write(vp, uiop, iomode, must_commit, cred, uiop->uio_td, &nfsva, &attrflag, NULL, called_from_strategy); if (attrflag) { if (VTONFS(vp)->n_flag & ND_NFSV4) ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1, 1); else ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); if (ret && !error) error = ret; } if (DOINGASYNC(vp)) *iomode = NFSWRITE_FILESYNC; if (error && NFS_ISV4(vp)) error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); return (error); } /* * nfs mknod rpc * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the * mode set to specify the file type and the size field for rdev. */ static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct vattr *vap) { struct nfsvattr nfsva, dnfsva; struct vnode *newvp = NULL; struct nfsnode *np = NULL, *dnp; struct nfsfh *nfhp; struct vattr vattr; int error = 0, attrflag, dattrflag; u_int32_t rdev; if (vap->va_type == VCHR || vap->va_type == VBLK) rdev = vap->va_rdev; else if (vap->va_type == VFIFO || vap->va_type == VSOCK) rdev = 0xffffffff; else return (EOPNOTSUPP); if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) return (error); error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap, - rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva, + rdev, vap->va_type, cnp->cn_cred, curthread, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, NULL); if (!error) { if (!nfhp) (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, - cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, + cnp->cn_namelen, cnp->cn_cred, curthread, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, NULL, 0); if (nfhp) error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, - cnp->cn_thread, &np, NULL, LK_EXCLUSIVE); + curthread, &np, NULL, LK_EXCLUSIVE); } if (dattrflag) (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); if (!error) { newvp = NFSTOV(np); if (attrflag != 0) { error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 0, 1); if (error != 0) vput(newvp); } } if (!error) { *vpp = newvp; } else if (NFS_ISV4(dvp)) { - error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid, + error = nfscl_maperr(curthread, error, vap->va_uid, vap->va_gid); } dnp = VTONFS(dvp); NFSLOCKNODE(dnp); dnp->n_flag |= NMODIFIED; if (!dattrflag) { dnp->n_attrstamp = 0; KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); } NFSUNLOCKNODE(dnp); return (error); } /* * nfs mknod vop * just call nfs_mknodrpc() to do the work. */ /* ARGSUSED */ static int nfs_mknod(struct vop_mknod_args *ap) { return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap)); } static struct mtx nfs_cverf_mtx; MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex", MTX_DEF); static nfsquad_t nfs_get_cverf(void) { static nfsquad_t cverf; nfsquad_t ret; static int cverf_initialized = 0; mtx_lock(&nfs_cverf_mtx); if (cverf_initialized == 0) { cverf.lval[0] = arc4random(); cverf.lval[1] = arc4random(); cverf_initialized = 1; } else cverf.qval++; ret = cverf; mtx_unlock(&nfs_cverf_mtx); return (ret); } /* * nfs file create call */ static int nfs_create(struct vop_create_args *ap) { struct vnode *dvp = ap->a_dvp; struct vattr *vap = ap->a_vap; struct componentname *cnp = ap->a_cnp; struct nfsnode *np = NULL, *dnp; struct vnode *newvp = NULL; struct nfsmount *nmp; struct nfsvattr dnfsva, nfsva; struct nfsfh *nfhp; nfsquad_t cverf; int error = 0, attrflag, dattrflag, fmode = 0; struct vattr vattr; /* * Oops, not for me.. */ if (vap->va_type == VSOCK) return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) return (error); if (vap->va_vaflags & VA_EXCLUSIVE) fmode |= O_EXCL; dnp = VTONFS(dvp); nmp = VFSTONFS(dvp->v_mount); again: /* For NFSv4, wait until any remove is done. */ NFSLOCKNODE(dnp); while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) { dnp->n_flag |= NREMOVEWANT; (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0); } NFSUNLOCKNODE(dnp); cverf = nfs_get_cverf(); error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen, - vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, + vap, cverf, fmode, cnp->cn_cred, curthread, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, NULL); if (!error) { if (nfhp == NULL) (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, - cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, + cnp->cn_namelen, cnp->cn_cred, curthread, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, NULL, 0); if (nfhp != NULL) error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, - cnp->cn_thread, &np, NULL, LK_EXCLUSIVE); + curthread, &np, NULL, LK_EXCLUSIVE); } if (dattrflag) (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); if (!error) { newvp = NFSTOV(np); if (attrflag == 0) error = nfsrpc_getattr(newvp, cnp->cn_cred, - cnp->cn_thread, &nfsva, NULL); + curthread, &nfsva, NULL); if (error == 0) error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 0, 1); } if (error) { if (newvp != NULL) { vput(newvp); newvp = NULL; } if (NFS_ISV34(dvp) && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) { fmode &= ~O_EXCL; goto again; } } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) { if (nfscl_checksattr(vap, &nfsva)) { error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred, - cnp->cn_thread, &nfsva, &attrflag, NULL); + curthread, &nfsva, &attrflag, NULL); if (error && (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL)) { /* try again without setting uid/gid */ vap->va_uid = (uid_t)VNOVAL; vap->va_gid = (uid_t)VNOVAL; error = nfsrpc_setattr(newvp, vap, NULL, - cnp->cn_cred, cnp->cn_thread, &nfsva, + cnp->cn_cred, curthread, &nfsva, &attrflag, NULL); } if (attrflag) (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 0, 1); if (error != 0) vput(newvp); } } if (!error) { if ((cnp->cn_flags & MAKEENTRY) && attrflag) { if (dvp != newvp) cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL); else printf("nfs_create: bogus NFS server returned " "the directory as the new file object\n"); } *ap->a_vpp = newvp; } else if (NFS_ISV4(dvp)) { - error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid, + error = nfscl_maperr(curthread, error, vap->va_uid, vap->va_gid); } NFSLOCKNODE(dnp); dnp->n_flag |= NMODIFIED; if (!dattrflag) { dnp->n_attrstamp = 0; KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); } NFSUNLOCKNODE(dnp); return (error); } /* * nfs file remove call * To try and make nfs semantics closer to ufs semantics, a file that has * other processes using the vnode is renamed instead of removed and then * removed later on the last close. * - If v_usecount > 1 * If a rename is not already in the works * call nfs_sillyrename() to set it up * else * do the remove rpc */ static int nfs_remove(struct vop_remove_args *ap) { struct vnode *vp = ap->a_vp; struct vnode *dvp = ap->a_dvp; struct componentname *cnp = ap->a_cnp; struct nfsnode *np = VTONFS(vp); int error = 0; struct vattr vattr; KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name")); KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount")); if (vp->v_type == VDIR) error = EPERM; else if (vrefcnt(vp) == 1 || (np->n_sillyrename && VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && vattr.va_nlink > 1)) { /* * Purge the name cache so that the chance of a lookup for * the name succeeding while the remove is in progress is * minimized. Without node locking it can still happen, such * that an I/O op returns ESTALE, but since you get this if * another host removes the file.. */ cache_purge(vp); /* * throw away biocache buffers, mainly to avoid * unnecessary delayed writes later. */ - error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1); + error = ncl_vinvalbuf(vp, 0, curthread, 1); if (error != EINTR && error != EIO) /* Do the rpc */ error = nfs_removerpc(dvp, vp, cnp->cn_nameptr, - cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread); + cnp->cn_namelen, cnp->cn_cred, curthread); /* * Kludge City: If the first reply to the remove rpc is lost.. * the reply to the retransmitted request will be ENOENT * since the file was in fact removed * Therefore, we cheat and return success. */ if (error == ENOENT) error = 0; } else if (!np->n_sillyrename) error = nfs_sillyrename(dvp, vp, cnp); NFSLOCKNODE(np); np->n_attrstamp = 0; NFSUNLOCKNODE(np); KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); return (error); } /* * nfs file remove rpc called from nfs_inactive */ int ncl_removeit(struct sillyrename *sp, struct vnode *vp) { /* * Make sure that the directory vnode is still valid. * XXX we should lock sp->s_dvp here. */ if (sp->s_dvp->v_type == VBAD) return (0); return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen, sp->s_cred, NULL)); } /* * Nfs remove rpc, called from nfs_remove() and ncl_removeit(). */ static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, int namelen, struct ucred *cred, struct thread *td) { struct nfsvattr dnfsva; struct nfsnode *dnp = VTONFS(dvp); int error = 0, dattrflag; NFSLOCKNODE(dnp); dnp->n_flag |= NREMOVEINPROG; NFSUNLOCKNODE(dnp); error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva, &dattrflag, NULL); NFSLOCKNODE(dnp); if ((dnp->n_flag & NREMOVEWANT)) { dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG); NFSUNLOCKNODE(dnp); wakeup((caddr_t)dnp); } else { dnp->n_flag &= ~NREMOVEINPROG; NFSUNLOCKNODE(dnp); } if (dattrflag) (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); NFSLOCKNODE(dnp); dnp->n_flag |= NMODIFIED; if (!dattrflag) { dnp->n_attrstamp = 0; KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); } NFSUNLOCKNODE(dnp); if (error && NFS_ISV4(dvp)) error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); return (error); } /* * nfs file rename call */ static int nfs_rename(struct vop_rename_args *ap) { struct vnode *fvp = ap->a_fvp; struct vnode *tvp = ap->a_tvp; struct vnode *fdvp = ap->a_fdvp; struct vnode *tdvp = ap->a_tdvp; struct componentname *tcnp = ap->a_tcnp; struct componentname *fcnp = ap->a_fcnp; struct nfsnode *fnp = VTONFS(ap->a_fvp); struct nfsnode *tdnp = VTONFS(ap->a_tdvp); struct nfsv4node *newv4 = NULL; int error; KASSERT((tcnp->cn_flags & HASBUF) != 0 && (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name")); /* Check for cross-device rename */ if ((fvp->v_mount != tdvp->v_mount) || (tvp && (fvp->v_mount != tvp->v_mount))) { error = EXDEV; goto out; } if (fvp == tvp) { printf("nfs_rename: fvp == tvp (can't happen)\n"); error = 0; goto out; } if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0) goto out; /* * We have to flush B_DELWRI data prior to renaming * the file. If we don't, the delayed-write buffers * can be flushed out later after the file has gone stale * under NFSV3. NFSV2 does not have this problem because * ( as far as I can tell ) it flushes dirty buffers more * often. * * Skip the rename operation if the fsync fails, this can happen * due to the server's volume being full, when we pushed out data * that was written back to our cache earlier. Not checking for * this condition can result in potential (silent) data loss. */ - error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread); + error = VOP_FSYNC(fvp, MNT_WAIT, curthread); NFSVOPUNLOCK(fvp); if (!error && tvp) - error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread); + error = VOP_FSYNC(tvp, MNT_WAIT, curthread); if (error) goto out; /* * If the tvp exists and is in use, sillyrename it before doing the * rename of the new file over it. * XXX Can't sillyrename a directory. */ if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename && tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { vput(tvp); tvp = NULL; } error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen, tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, - tcnp->cn_thread); + curthread); if (error == 0 && NFS_ISV4(tdvp)) { /* * For NFSv4, check to see if it is the same name and * replace the name, if it is different. */ newv4 = malloc( sizeof (struct nfsv4node) + tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK); NFSLOCKNODE(tdnp); NFSLOCKNODE(fnp); if (fnp->n_v4 != NULL && fvp->v_type == VREG && (fnp->n_v4->n4_namelen != tcnp->cn_namelen || NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen) || tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen || NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, tdnp->n_fhp->nfh_len))) { #ifdef notdef { char nnn[100]; int nnnl; nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99; bcopy(tcnp->cn_nameptr, nnn, nnnl); nnn[nnnl] = '\0'; printf("ren replace=%s\n",nnn); } #endif free(fnp->n_v4, M_NFSV4NODE); fnp->n_v4 = newv4; newv4 = NULL; fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len; fnp->n_v4->n4_namelen = tcnp->cn_namelen; NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, tdnp->n_fhp->nfh_len); NFSBCOPY(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen); } NFSUNLOCKNODE(tdnp); NFSUNLOCKNODE(fnp); if (newv4 != NULL) free(newv4, M_NFSV4NODE); } if (fvp->v_type == VDIR) { if (tvp != NULL && tvp->v_type == VDIR) cache_purge(tdvp); cache_purge(fdvp); } out: if (tdvp == tvp) vrele(tdvp); else vput(tdvp); if (tvp) vput(tvp); vrele(fdvp); vrele(fvp); /* * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. */ if (error == ENOENT) error = 0; return (error); } /* * nfs file rename rpc called from nfs_remove() above */ static int nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp, struct sillyrename *sp) { return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen, sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred, - scnp->cn_thread)); + curthread)); } /* * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). */ static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td) { struct nfsvattr fnfsva, tnfsva; struct nfsnode *fdnp = VTONFS(fdvp); struct nfsnode *tdnp = VTONFS(tdvp); int error = 0, fattrflag, tattrflag; error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp, tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag, &tattrflag, NULL, NULL); NFSLOCKNODE(fdnp); fdnp->n_flag |= NMODIFIED; if (fattrflag != 0) { NFSUNLOCKNODE(fdnp); (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1); } else { fdnp->n_attrstamp = 0; NFSUNLOCKNODE(fdnp); KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp); } NFSLOCKNODE(tdnp); tdnp->n_flag |= NMODIFIED; if (tattrflag != 0) { NFSUNLOCKNODE(tdnp); (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1); } else { tdnp->n_attrstamp = 0; NFSUNLOCKNODE(tdnp); KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); } if (error && NFS_ISV4(fdvp)) error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); return (error); } /* * nfs hard link create call */ static int nfs_link(struct vop_link_args *ap) { struct vnode *vp = ap->a_vp; struct vnode *tdvp = ap->a_tdvp; struct componentname *cnp = ap->a_cnp; struct nfsnode *np, *tdnp; struct nfsvattr nfsva, dnfsva; int error = 0, attrflag, dattrflag; /* * Push all writes to the server, so that the attribute cache * doesn't get "out of sync" with the server. * XXX There should be a better way! */ - VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread); + VOP_FSYNC(vp, MNT_WAIT, curthread); error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen, - cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag, + cnp->cn_cred, curthread, &dnfsva, &nfsva, &attrflag, &dattrflag, NULL); tdnp = VTONFS(tdvp); NFSLOCKNODE(tdnp); tdnp->n_flag |= NMODIFIED; if (dattrflag != 0) { NFSUNLOCKNODE(tdnp); (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1); } else { tdnp->n_attrstamp = 0; NFSUNLOCKNODE(tdnp); KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); } if (attrflag) (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); else { np = VTONFS(vp); NFSLOCKNODE(np); np->n_attrstamp = 0; NFSUNLOCKNODE(np); KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); } /* * If negative lookup caching is enabled, I might as well * add an entry for this node. Not necessary for correctness, * but if negative caching is enabled, then the system * must care about lookup caching hit rate, so... */ if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 && (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { if (tdvp != vp) cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL); else printf("nfs_link: bogus NFS server returned " "the directory as the new link\n"); } if (error && NFS_ISV4(vp)) - error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0, + error = nfscl_maperr(curthread, error, (uid_t)0, (gid_t)0); return (error); } /* * nfs symbolic link create call */ static int nfs_symlink(struct vop_symlink_args *ap) { struct vnode *dvp = ap->a_dvp; struct vattr *vap = ap->a_vap; struct componentname *cnp = ap->a_cnp; struct nfsvattr nfsva, dnfsva; struct nfsfh *nfhp; struct nfsnode *np = NULL, *dnp; struct vnode *newvp = NULL; int error = 0, attrflag, dattrflag, ret; vap->va_type = VLNK; error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen, - ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, + ap->a_target, vap, cnp->cn_cred, curthread, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, NULL); if (nfhp) { - ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, + ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, curthread, &np, NULL, LK_EXCLUSIVE); if (!ret) newvp = NFSTOV(np); else if (!error) error = ret; } if (newvp != NULL) { if (attrflag) (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 0, 1); } else if (!error) { /* * If we do not have an error and we could not extract the * newvp from the response due to the request being NFSv2, we * have to do a lookup in order to obtain a newvp to return. */ error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, - cnp->cn_cred, cnp->cn_thread, &np); + cnp->cn_cred, curthread, &np); if (!error) newvp = NFSTOV(np); } if (error) { if (newvp) vput(newvp); if (NFS_ISV4(dvp)) - error = nfscl_maperr(cnp->cn_thread, error, + error = nfscl_maperr(curthread, error, vap->va_uid, vap->va_gid); } else { *ap->a_vpp = newvp; } dnp = VTONFS(dvp); NFSLOCKNODE(dnp); dnp->n_flag |= NMODIFIED; if (dattrflag != 0) { NFSUNLOCKNODE(dnp); (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); } else { dnp->n_attrstamp = 0; NFSUNLOCKNODE(dnp); KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); } /* * If negative lookup caching is enabled, I might as well * add an entry for this node. Not necessary for correctness, * but if negative caching is enabled, then the system * must care about lookup caching hit rate, so... */ if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { if (dvp != newvp) cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL); else printf("nfs_symlink: bogus NFS server returned " "the directory as the new file object\n"); } return (error); } /* * nfs make dir call */ static int nfs_mkdir(struct vop_mkdir_args *ap) { struct vnode *dvp = ap->a_dvp; struct vattr *vap = ap->a_vap; struct componentname *cnp = ap->a_cnp; struct nfsnode *np = NULL, *dnp; struct vnode *newvp = NULL; struct vattr vattr; struct nfsfh *nfhp; struct nfsvattr nfsva, dnfsva; int error = 0, attrflag, dattrflag, ret; if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) return (error); vap->va_type = VDIR; error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, - vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp, + vap, cnp->cn_cred, curthread, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, NULL); dnp = VTONFS(dvp); NFSLOCKNODE(dnp); dnp->n_flag |= NMODIFIED; if (dattrflag != 0) { NFSUNLOCKNODE(dnp); (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); } else { dnp->n_attrstamp = 0; NFSUNLOCKNODE(dnp); KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); } if (nfhp) { - ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, + ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, curthread, &np, NULL, LK_EXCLUSIVE); if (!ret) { newvp = NFSTOV(np); if (attrflag) (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 0, 1); } else if (!error) error = ret; } if (!error && newvp == NULL) { error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, - cnp->cn_cred, cnp->cn_thread, &np); + cnp->cn_cred, curthread, &np); if (!error) { newvp = NFSTOV(np); if (newvp->v_type != VDIR) error = EEXIST; } } if (error) { if (newvp) vput(newvp); if (NFS_ISV4(dvp)) - error = nfscl_maperr(cnp->cn_thread, error, + error = nfscl_maperr(curthread, error, vap->va_uid, vap->va_gid); } else { /* * If negative lookup caching is enabled, I might as well * add an entry for this node. Not necessary for correctness, * but if negative caching is enabled, then the system * must care about lookup caching hit rate, so... */ if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && dattrflag != 0) { if (dvp != newvp) cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, &dnfsva.na_ctime); else printf("nfs_mkdir: bogus NFS server returned " "the directory that the directory was " "created in as the new file object\n"); } *ap->a_vpp = newvp; } return (error); } /* * nfs remove directory call */ static int nfs_rmdir(struct vop_rmdir_args *ap) { struct vnode *vp = ap->a_vp; struct vnode *dvp = ap->a_dvp; struct componentname *cnp = ap->a_cnp; struct nfsnode *dnp; struct nfsvattr dnfsva; int error, dattrflag; if (dvp == vp) return (EINVAL); error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, - cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL); + cnp->cn_cred, curthread, &dnfsva, &dattrflag, NULL); dnp = VTONFS(dvp); NFSLOCKNODE(dnp); dnp->n_flag |= NMODIFIED; if (dattrflag != 0) { NFSUNLOCKNODE(dnp); (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); } else { dnp->n_attrstamp = 0; NFSUNLOCKNODE(dnp); KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); } cache_purge(dvp); cache_purge(vp); if (error && NFS_ISV4(dvp)) - error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0, + error = nfscl_maperr(curthread, error, (uid_t)0, (gid_t)0); /* * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. */ if (error == ENOENT) error = 0; return (error); } /* * nfs readdir call */ static int nfs_readdir(struct vop_readdir_args *ap) { struct vnode *vp = ap->a_vp; struct nfsnode *np = VTONFS(vp); struct uio *uio = ap->a_uio; ssize_t tresid, left; int error = 0; struct vattr vattr; if (ap->a_eofflag != NULL) *ap->a_eofflag = 0; if (vp->v_type != VDIR) return(EPERM); /* * First, check for hit on the EOF offset cache */ NFSLOCKNODE(np); if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && (np->n_flag & NMODIFIED) == 0) { NFSUNLOCKNODE(np); if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) { NFSLOCKNODE(np); if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) || !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { NFSUNLOCKNODE(np); NFSINCRGLOBAL(nfsstatsv1.direofcache_hits); if (ap->a_eofflag != NULL) *ap->a_eofflag = 1; return (0); } else NFSUNLOCKNODE(np); } } else NFSUNLOCKNODE(np); /* * NFS always guarantees that directory entries don't straddle * DIRBLKSIZ boundaries. As such, we need to limit the size * to an exact multiple of DIRBLKSIZ, to avoid copying a partial * directory entry. */ left = uio->uio_resid % DIRBLKSIZ; if (left == uio->uio_resid) return (EINVAL); uio->uio_resid -= left; /* * Call ncl_bioread() to do the real work. */ tresid = uio->uio_resid; error = ncl_bioread(vp, uio, 0, ap->a_cred); if (!error && uio->uio_resid == tresid) { NFSINCRGLOBAL(nfsstatsv1.direofcache_misses); if (ap->a_eofflag != NULL) *ap->a_eofflag = 1; } /* Add the partial DIRBLKSIZ (left) back in. */ uio->uio_resid += left; return (error); } /* * Readdir rpc call. * Called from below the buffer cache by ncl_doio(). */ int ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, struct thread *td) { struct nfsvattr nfsva; nfsuint64 *cookiep, cookie; struct nfsnode *dnp = VTONFS(vp); struct nfsmount *nmp = VFSTONFS(vp->v_mount); int error = 0, eof, attrflag; KASSERT(uiop->uio_iovcnt == 1 && (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, ("nfs readdirrpc bad uio")); /* * If there is no cookie, assume directory was stale. */ ncl_dircookie_lock(dnp); NFSUNLOCKNODE(dnp); cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); if (cookiep) { cookie = *cookiep; ncl_dircookie_unlock(dnp); } else { ncl_dircookie_unlock(dnp); return (NFSERR_BAD_COOKIE); } if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) (void)ncl_fsinfo(nmp, vp, cred, td); error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva, &attrflag, &eof, NULL); if (attrflag) (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); if (!error) { /* * We are now either at the end of the directory or have filled * the block. */ if (eof) { NFSLOCKNODE(dnp); dnp->n_direofoffset = uiop->uio_offset; NFSUNLOCKNODE(dnp); } else { if (uiop->uio_resid > 0) printf("EEK! readdirrpc resid > 0\n"); ncl_dircookie_lock(dnp); NFSUNLOCKNODE(dnp); cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); *cookiep = cookie; ncl_dircookie_unlock(dnp); } } else if (NFS_ISV4(vp)) { error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); } return (error); } /* * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc(). */ int ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, struct thread *td) { struct nfsvattr nfsva; nfsuint64 *cookiep, cookie; struct nfsnode *dnp = VTONFS(vp); struct nfsmount *nmp = VFSTONFS(vp->v_mount); int error = 0, attrflag, eof; KASSERT(uiop->uio_iovcnt == 1 && (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, ("nfs readdirplusrpc bad uio")); /* * If there is no cookie, assume directory was stale. */ ncl_dircookie_lock(dnp); NFSUNLOCKNODE(dnp); cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); if (cookiep) { cookie = *cookiep; ncl_dircookie_unlock(dnp); } else { ncl_dircookie_unlock(dnp); return (NFSERR_BAD_COOKIE); } if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) (void)ncl_fsinfo(nmp, vp, cred, td); error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva, &attrflag, &eof, NULL); if (attrflag) (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); if (!error) { /* * We are now either at end of the directory or have filled the * the block. */ if (eof) { NFSLOCKNODE(dnp); dnp->n_direofoffset = uiop->uio_offset; NFSUNLOCKNODE(dnp); } else { if (uiop->uio_resid > 0) printf("EEK! readdirplusrpc resid > 0\n"); ncl_dircookie_lock(dnp); NFSUNLOCKNODE(dnp); cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); *cookiep = cookie; ncl_dircookie_unlock(dnp); } } else if (NFS_ISV4(vp)) { error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); } return (error); } /* * Silly rename. To make the NFS filesystem that is stateless look a little * more like the "ufs" a remove of an active vnode is translated to a rename * to a funny looking filename that is removed by nfs_inactive on the * nfsnode. There is the potential for another process on a different client * to create the same funny name between the nfs_lookitup() fails and the * nfs_rename() completes, but... */ static int nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) { struct sillyrename *sp; struct nfsnode *np; int error; short pid; unsigned int lticks; cache_purge(dvp); np = VTONFS(vp); KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir")); sp = malloc(sizeof (struct sillyrename), M_NEWNFSREQ, M_WAITOK); sp->s_cred = crhold(cnp->cn_cred); sp->s_dvp = dvp; VREF(dvp); /* * Fudge together a funny name. * Changing the format of the funny name to accommodate more * sillynames per directory. * The name is now changed to .nfs...4, where ticks is * CPU ticks since boot. */ - pid = cnp->cn_thread->td_proc->p_pid; + pid = curthread->td_proc->p_pid; lticks = (unsigned int)ticks; for ( ; ; ) { sp->s_namlen = sprintf(sp->s_name, ".nfs.%08x.%04x4.4", lticks, pid); if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, - cnp->cn_thread, NULL)) + curthread, NULL)) break; lticks++; } error = nfs_renameit(dvp, vp, cnp, sp); if (error) goto bad; error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, - cnp->cn_thread, &np); + curthread, &np); np->n_sillyrename = sp; return (0); bad: vrele(sp->s_dvp); crfree(sp->s_cred); free(sp, M_NEWNFSREQ); return (error); } /* * Look up a file name and optionally either update the file handle or * allocate an nfsnode, depending on the value of npp. * npp == NULL --> just do the lookup * *npp == NULL --> allocate a new nfsnode and make sure attributes are * handled too * *npp != NULL --> update the file handle in the vnode */ static int nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred, struct thread *td, struct nfsnode **npp) { struct vnode *newvp = NULL, *vp; struct nfsnode *np, *dnp = VTONFS(dvp); struct nfsfh *nfhp, *onfhp; struct nfsvattr nfsva, dnfsva; struct componentname cn; int error = 0, attrflag, dattrflag; u_int hash; error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, NULL, 0); if (dattrflag) (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); if (npp && !error) { if (*npp != NULL) { np = *npp; vp = NFSTOV(np); /* * For NFSv4, check to see if it is the same name and * replace the name, if it is different. */ if (np->n_v4 != NULL && nfsva.na_type == VREG && (np->n_v4->n4_namelen != len || NFSBCMP(name, NFS4NODENAME(np->n_v4), len) || dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, dnp->n_fhp->nfh_len))) { #ifdef notdef { char nnn[100]; int nnnl; nnnl = (len < 100) ? len : 99; bcopy(name, nnn, nnnl); nnn[nnnl] = '\0'; printf("replace=%s\n",nnn); } #endif free(np->n_v4, M_NFSV4NODE); np->n_v4 = malloc( sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len + len - 1, M_NFSV4NODE, M_WAITOK); np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; np->n_v4->n4_namelen = len; NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, dnp->n_fhp->nfh_len); NFSBCOPY(name, NFS4NODENAME(np->n_v4), len); } hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT); onfhp = np->n_fhp; /* * Rehash node for new file handle. */ vfs_hash_rehash(vp, hash); np->n_fhp = nfhp; if (onfhp != NULL) free(onfhp, M_NFSFH); newvp = NFSTOV(np); } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) { free(nfhp, M_NFSFH); VREF(dvp); newvp = dvp; } else { cn.cn_nameptr = name; cn.cn_namelen = len; error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td, &np, NULL, LK_EXCLUSIVE); if (error) return (error); newvp = NFSTOV(np); } if (!attrflag && *npp == NULL) { if (newvp == dvp) vrele(newvp); else vput(newvp); return (ENOENT); } if (attrflag) (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 0, 1); } if (npp && *npp == NULL) { if (error) { if (newvp) { if (newvp == dvp) vrele(newvp); else vput(newvp); } } else *npp = np; } if (error && NFS_ISV4(dvp)) error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); return (error); } /* * Nfs Version 3 and 4 commit rpc */ int ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred, struct thread *td) { struct nfsvattr nfsva; struct nfsmount *nmp = VFSTONFS(vp->v_mount); struct nfsnode *np; struct uio uio; int error, attrflag; np = VTONFS(vp); error = EIO; attrflag = 0; if (NFSHASPNFS(nmp) && (np->n_flag & NDSCOMMIT) != 0) { uio.uio_offset = offset; uio.uio_resid = cnt; error = nfscl_doiods(vp, &uio, NULL, NULL, NFSV4OPEN_ACCESSWRITE, 1, cred, td); if (error != 0) { NFSLOCKNODE(np); np->n_flag &= ~NDSCOMMIT; NFSUNLOCKNODE(np); } } if (error != 0) { mtx_lock(&nmp->nm_mtx); if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) { mtx_unlock(&nmp->nm_mtx); return (0); } mtx_unlock(&nmp->nm_mtx); error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva, &attrflag, NULL); } if (attrflag != 0) (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); if (error != 0 && NFS_ISV4(vp)) error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); return (error); } /* * Strategy routine. * For async requests when nfsiod(s) are running, queue the request by * calling ncl_asyncio(), otherwise just all ncl_doio() to do the * request. */ static int nfs_strategy(struct vop_strategy_args *ap) { struct buf *bp; struct vnode *vp; struct ucred *cr; bp = ap->a_bp; vp = ap->a_vp; KASSERT(bp->b_vp == vp, ("missing b_getvp")); KASSERT(!(bp->b_flags & B_DONE), ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno) bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize / DEV_BSIZE); if (bp->b_iocmd == BIO_READ) cr = bp->b_rcred; else cr = bp->b_wcred; /* * If the op is asynchronous and an i/o daemon is waiting * queue the request, wake it up and wait for completion * otherwise just do it ourselves. */ if ((bp->b_flags & B_ASYNC) == 0 || ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread)) (void) ncl_doio(vp, bp, cr, curthread, 1); return (0); } /* * fsync vnode op. Just call ncl_flush() with commit == 1. */ /* ARGSUSED */ static int nfs_fsync(struct vop_fsync_args *ap) { if (ap->a_vp->v_type != VREG) { /* * For NFS, metadata is changed synchronously on the server, * so there is nothing to flush. Also, ncl_flush() clears * the NMODIFIED flag and that shouldn't be done here for * directories. */ return (0); } return (ncl_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1, 0)); } /* * Flush all the blocks associated with a vnode. * Walk through the buffer pool and push any dirty pages * associated with the vnode. * If the called_from_renewthread argument is TRUE, it has been called * from the NFSv4 renew thread and, as such, cannot block indefinitely * waiting for a buffer write to complete. */ int ncl_flush(struct vnode *vp, int waitfor, struct thread *td, int commit, int called_from_renewthread) { struct nfsnode *np = VTONFS(vp); struct buf *bp; int i; struct buf *nbp; struct nfsmount *nmp = VFSTONFS(vp->v_mount); int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos; int passone = 1, trycnt = 0; u_quad_t off, endoff, toff; struct ucred* wcred = NULL; struct buf **bvec = NULL; struct bufobj *bo; #ifndef NFS_COMMITBVECSIZ #define NFS_COMMITBVECSIZ 20 #endif struct buf *bvec_on_stack[NFS_COMMITBVECSIZ]; u_int bvecsize = 0, bveccount; if (called_from_renewthread != 0) slptimeo = hz; if (nmp->nm_flag & NFSMNT_INT) slpflag = PCATCH; if (!commit) passone = 0; bo = &vp->v_bufobj; /* * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the * server, but has not been committed to stable storage on the server * yet. On the first pass, the byte range is worked out and the commit * rpc is done. On the second pass, ncl_writebp() is called to do the * job. */ again: off = (u_quad_t)-1; endoff = 0; bvecpos = 0; if (NFS_ISV34(vp) && commit) { if (bvec != NULL && bvec != bvec_on_stack) free(bvec, M_TEMP); /* * Count up how many buffers waiting for a commit. */ bveccount = 0; BO_LOCK(bo); TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { if (!BUF_ISLOCKED(bp) && (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) == (B_DELWRI | B_NEEDCOMMIT)) bveccount++; } /* * Allocate space to remember the list of bufs to commit. It is * important to use M_NOWAIT here to avoid a race with nfs_write. * If we can't get memory (for whatever reason), we will end up * committing the buffers one-by-one in the loop below. */ if (bveccount > NFS_COMMITBVECSIZ) { /* * Release the vnode interlock to avoid a lock * order reversal. */ BO_UNLOCK(bo); bvec = (struct buf **) malloc(bveccount * sizeof(struct buf *), M_TEMP, M_NOWAIT); BO_LOCK(bo); if (bvec == NULL) { bvec = bvec_on_stack; bvecsize = NFS_COMMITBVECSIZ; } else bvecsize = bveccount; } else { bvec = bvec_on_stack; bvecsize = NFS_COMMITBVECSIZ; } TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { if (bvecpos >= bvecsize) break; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { nbp = TAILQ_NEXT(bp, b_bobufs); continue; } if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != (B_DELWRI | B_NEEDCOMMIT)) { BUF_UNLOCK(bp); nbp = TAILQ_NEXT(bp, b_bobufs); continue; } BO_UNLOCK(bo); bremfree(bp); /* * Work out if all buffers are using the same cred * so we can deal with them all with one commit. * * NOTE: we are not clearing B_DONE here, so we have * to do it later on in this routine if we intend to * initiate I/O on the bp. * * Note: to avoid loopback deadlocks, we do not * assign b_runningbufspace. */ if (wcred == NULL) wcred = bp->b_wcred; else if (wcred != bp->b_wcred) wcred = NOCRED; vfs_busy_pages(bp, 1); BO_LOCK(bo); /* * bp is protected by being locked, but nbp is not * and vfs_busy_pages() may sleep. We have to * recalculate nbp. */ nbp = TAILQ_NEXT(bp, b_bobufs); /* * A list of these buffers is kept so that the * second loop knows which buffers have actually * been committed. This is necessary, since there * may be a race between the commit rpc and new * uncommitted writes on the file. */ bvec[bvecpos++] = bp; toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + bp->b_dirtyoff; if (toff < off) off = toff; toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); if (toff > endoff) endoff = toff; } BO_UNLOCK(bo); } if (bvecpos > 0) { /* * Commit data on the server, as required. * If all bufs are using the same wcred, then use that with * one call for all of them, otherwise commit each one * separately. */ if (wcred != NOCRED) retv = ncl_commit(vp, off, (int)(endoff - off), wcred, td); else { retv = 0; for (i = 0; i < bvecpos; i++) { off_t off, size; bp = bvec[i]; off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + bp->b_dirtyoff; size = (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); retv = ncl_commit(vp, off, (int)size, bp->b_wcred, td); if (retv) break; } } if (retv == NFSERR_STALEWRITEVERF) ncl_clearcommit(vp->v_mount); /* * Now, either mark the blocks I/O done or mark the * blocks dirty, depending on whether the commit * succeeded. */ for (i = 0; i < bvecpos; i++) { bp = bvec[i]; bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); if (retv) { /* * Error, leave B_DELWRI intact */ vfs_unbusy_pages(bp); brelse(bp); } else { /* * Success, remove B_DELWRI ( bundirty() ). * * b_dirtyoff/b_dirtyend seem to be NFS * specific. We should probably move that * into bundirty(). XXX */ bufobj_wref(bo); bp->b_flags |= B_ASYNC; bundirty(bp); bp->b_flags &= ~B_DONE; bp->b_ioflags &= ~BIO_ERROR; bp->b_dirtyoff = bp->b_dirtyend = 0; bufdone(bp); } } } /* * Start/do any write(s) that are required. */ loop: BO_LOCK(bo); TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { if (waitfor != MNT_WAIT || passone) continue; error = BUF_TIMELOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo); if (error == 0) { BUF_UNLOCK(bp); goto loop; } if (error == ENOLCK) { error = 0; goto loop; } if (called_from_renewthread != 0) { /* * Return EIO so the flush will be retried * later. */ error = EIO; goto done; } if (newnfs_sigintr(nmp, td)) { error = EINTR; goto done; } if (slpflag == PCATCH) { slpflag = 0; slptimeo = 2 * hz; } goto loop; } if ((bp->b_flags & B_DELWRI) == 0) panic("nfs_fsync: not dirty"); if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) { BUF_UNLOCK(bp); continue; } BO_UNLOCK(bo); bremfree(bp); bp->b_flags |= B_ASYNC; bwrite(bp); if (newnfs_sigintr(nmp, td)) { error = EINTR; goto done; } goto loop; } if (passone) { passone = 0; BO_UNLOCK(bo); goto again; } if (waitfor == MNT_WAIT) { while (bo->bo_numoutput) { error = bufobj_wwait(bo, slpflag, slptimeo); if (error) { BO_UNLOCK(bo); if (called_from_renewthread != 0) { /* * Return EIO so that the flush will be * retried later. */ error = EIO; goto done; } error = newnfs_sigintr(nmp, td); if (error) goto done; if (slpflag == PCATCH) { slpflag = 0; slptimeo = 2 * hz; } BO_LOCK(bo); } } if (bo->bo_dirty.bv_cnt != 0 && commit) { BO_UNLOCK(bo); goto loop; } /* * Wait for all the async IO requests to drain */ BO_UNLOCK(bo); NFSLOCKNODE(np); while (np->n_directio_asyncwr > 0) { np->n_flag |= NFSYNCWAIT; error = newnfs_msleep(td, &np->n_directio_asyncwr, &np->n_mtx, slpflag | (PRIBIO + 1), "nfsfsync", 0); if (error) { if (newnfs_sigintr(nmp, td)) { NFSUNLOCKNODE(np); error = EINTR; goto done; } } } NFSUNLOCKNODE(np); } else BO_UNLOCK(bo); if (NFSHASPNFS(nmp)) { nfscl_layoutcommit(vp, td); /* * Invalidate the attribute cache, since writes to a DS * won't update the size attribute. */ NFSLOCKNODE(np); np->n_attrstamp = 0; } else NFSLOCKNODE(np); if (np->n_flag & NWRITEERR) { error = np->n_error; np->n_flag &= ~NWRITEERR; } if (commit && bo->bo_dirty.bv_cnt == 0 && bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0) np->n_flag &= ~NMODIFIED; NFSUNLOCKNODE(np); done: if (bvec != NULL && bvec != bvec_on_stack) free(bvec, M_TEMP); if (error == 0 && commit != 0 && waitfor == MNT_WAIT && (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 || np->n_directio_asyncwr != 0)) { if (trycnt++ < 5) { /* try, try again... */ passone = 1; wcred = NULL; bvec = NULL; bvecsize = 0; goto again; } vn_printf(vp, "ncl_flush failed"); error = called_from_renewthread != 0 ? EIO : EBUSY; } return (error); } /* * NFS advisory byte-level locks. */ static int nfs_advlock(struct vop_advlock_args *ap) { struct vnode *vp = ap->a_vp; struct ucred *cred; struct nfsnode *np = VTONFS(ap->a_vp); struct proc *p = (struct proc *)ap->a_id; struct thread *td = curthread; /* XXX */ struct vattr va; int ret, error; u_quad_t size; struct nfsmount *nmp; error = NFSVOPLOCK(vp, LK_SHARED); if (error != 0) return (EBADF); if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) { if (vp->v_type != VREG) { error = EINVAL; goto out; } if ((ap->a_flags & F_POSIX) != 0) cred = p->p_ucred; else cred = td->td_ucred; NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); if (VN_IS_DOOMED(vp)) { error = EBADF; goto out; } /* * If this is unlocking a write locked region, flush and * commit them before unlocking. This is required by * RFC3530 Sec. 9.3.2. */ if (ap->a_op == F_UNLCK && nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id, ap->a_flags)) (void) ncl_flush(vp, MNT_WAIT, td, 1, 0); /* * Mark NFS node as might have acquired a lock. * This is separate from NHASBEENLOCKED, because it must * be done before the nfsrpc_advlock() call, which might * add a nfscllock structure to the client state. * It is used to check for the case where a nfscllock * state structure cannot exist for the file. * Only done for "oneopenown" NFSv4.1/4.2 mounts. */ nmp = VFSTONFS(vp->v_mount); if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp)) { NFSLOCKNODE(np); np->n_flag |= NMIGHTBELOCKED; NFSUNLOCKNODE(np); } /* * Loop around doing the lock op, while a blocking lock * must wait for the lock op to succeed. */ do { ret = nfsrpc_advlock(vp, np->n_size, ap->a_op, ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags); if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && ap->a_op == F_SETLK) { NFSVOPUNLOCK(vp); error = nfs_catnap(PZERO | PCATCH, ret, "ncladvl"); if (error) return (EINTR); NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); if (VN_IS_DOOMED(vp)) { error = EBADF; goto out; } } } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && ap->a_op == F_SETLK); if (ret == NFSERR_DENIED) { error = EAGAIN; goto out; } else if (ret == EINVAL || ret == EBADF || ret == EINTR) { error = ret; goto out; } else if (ret != 0) { error = EACCES; goto out; } /* * Now, if we just got a lock, invalidate data in the buffer * cache, as required, so that the coherency conforms with * RFC3530 Sec. 9.3.2. */ if (ap->a_op == F_SETLK) { if ((np->n_flag & NMODIFIED) == 0) { np->n_attrstamp = 0; KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); ret = VOP_GETATTR(vp, &va, cred); } if ((np->n_flag & NMODIFIED) || ret || np->n_change != va.va_filerev) { (void) ncl_vinvalbuf(vp, V_SAVE, td, 1); np->n_attrstamp = 0; KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); ret = VOP_GETATTR(vp, &va, cred); if (!ret) { np->n_mtime = va.va_mtime; np->n_change = va.va_filerev; } } /* Mark that a file lock has been acquired. */ NFSLOCKNODE(np); np->n_flag |= NHASBEENLOCKED; NFSUNLOCKNODE(np); } } else if (!NFS_ISV4(vp)) { if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { size = VTONFS(vp)->n_size; NFSVOPUNLOCK(vp); error = lf_advlock(ap, &(vp->v_lockf), size); } else { if (nfs_advlock_p != NULL) error = nfs_advlock_p(ap); else { NFSVOPUNLOCK(vp); error = ENOLCK; } } if (error == 0 && ap->a_op == F_SETLK) { error = NFSVOPLOCK(vp, LK_SHARED); if (error == 0) { /* Mark that a file lock has been acquired. */ NFSLOCKNODE(np); np->n_flag |= NHASBEENLOCKED; NFSUNLOCKNODE(np); NFSVOPUNLOCK(vp); } } return (error); } else error = EOPNOTSUPP; out: NFSVOPUNLOCK(vp); return (error); } /* * NFS advisory byte-level locks. */ static int nfs_advlockasync(struct vop_advlockasync_args *ap) { struct vnode *vp = ap->a_vp; u_quad_t size; int error; if (NFS_ISV4(vp)) return (EOPNOTSUPP); error = NFSVOPLOCK(vp, LK_SHARED); if (error) return (error); if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { size = VTONFS(vp)->n_size; NFSVOPUNLOCK(vp); error = lf_advlockasync(ap, &(vp->v_lockf), size); } else { NFSVOPUNLOCK(vp); error = EOPNOTSUPP; } return (error); } /* * Print out the contents of an nfsnode. */ static int nfs_print(struct vop_print_args *ap) { struct vnode *vp = ap->a_vp; struct nfsnode *np = VTONFS(vp); printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid, (uintmax_t)np->n_vattr.na_fsid); if (vp->v_type == VFIFO) fifo_printinfo(vp); printf("\n"); return (0); } /* * This is the "real" nfs::bwrite(struct buf*). * We set B_CACHE if this is a VMIO buffer. */ int ncl_writebp(struct buf *bp, int force __unused, struct thread *td) { int oldflags, rtval; if (bp->b_flags & B_INVAL) { brelse(bp); return (0); } oldflags = bp->b_flags; bp->b_flags |= B_CACHE; /* * Undirty the bp. We will redirty it later if the I/O fails. */ bundirty(bp); bp->b_flags &= ~B_DONE; bp->b_ioflags &= ~BIO_ERROR; bp->b_iocmd = BIO_WRITE; bufobj_wref(bp->b_bufobj); curthread->td_ru.ru_oublock++; /* * Note: to avoid loopback deadlocks, we do not * assign b_runningbufspace. */ vfs_busy_pages(bp, 1); BUF_KERNPROC(bp); bp->b_iooffset = dbtob(bp->b_blkno); bstrategy(bp); if ((oldflags & B_ASYNC) != 0) return (0); rtval = bufwait(bp); if (oldflags & B_DELWRI) reassignbuf(bp); brelse(bp); return (rtval); } /* * nfs special file access vnode op. * Essentially just get vattr and then imitate iaccess() since the device is * local to the client. */ static int nfsspec_access(struct vop_access_args *ap) { struct vattr *vap; struct ucred *cred = ap->a_cred; struct vnode *vp = ap->a_vp; accmode_t accmode = ap->a_accmode; struct vattr vattr; int error; /* * Disallow write attempts on filesystems mounted read-only; * unless the file is a socket, fifo, or a block or character * device resident on the filesystem. */ if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { switch (vp->v_type) { case VREG: case VDIR: case VLNK: return (EROFS); default: break; } } vap = &vattr; error = VOP_GETATTR(vp, vap, cred); if (error) goto out; error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid, accmode, cred); out: return error; } /* * Read wrapper for fifos. */ static int nfsfifo_read(struct vop_read_args *ap) { struct nfsnode *np = VTONFS(ap->a_vp); int error; /* * Set access flag. */ NFSLOCKNODE(np); np->n_flag |= NACC; vfs_timestamp(&np->n_atim); NFSUNLOCKNODE(np); error = fifo_specops.vop_read(ap); return error; } /* * Write wrapper for fifos. */ static int nfsfifo_write(struct vop_write_args *ap) { struct nfsnode *np = VTONFS(ap->a_vp); /* * Set update flag. */ NFSLOCKNODE(np); np->n_flag |= NUPD; vfs_timestamp(&np->n_mtim); NFSUNLOCKNODE(np); return(fifo_specops.vop_write(ap)); } /* * Close wrapper for fifos. * * Update the times on the nfsnode then do fifo close. */ static int nfsfifo_close(struct vop_close_args *ap) { struct vnode *vp = ap->a_vp; struct nfsnode *np = VTONFS(vp); struct vattr vattr; struct timespec ts; NFSLOCKNODE(np); if (np->n_flag & (NACC | NUPD)) { vfs_timestamp(&ts); if (np->n_flag & NACC) np->n_atim = ts; if (np->n_flag & NUPD) np->n_mtim = ts; np->n_flag |= NCHG; if (vrefcnt(vp) == 1 && (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { VATTR_NULL(&vattr); if (np->n_flag & NACC) vattr.va_atime = np->n_atim; if (np->n_flag & NUPD) vattr.va_mtime = np->n_mtim; NFSUNLOCKNODE(np); (void)VOP_SETATTR(vp, &vattr, ap->a_cred); goto out; } } NFSUNLOCKNODE(np); out: return (fifo_specops.vop_close(ap)); } /* * Just call ncl_writebp() with the force argument set to 1. * * NOTE: B_DONE may or may not be set in a_bp on call. */ static int nfs_bwrite(struct buf *bp) { return (ncl_writebp(bp, 1, curthread)); } struct buf_ops buf_ops_newnfs = { .bop_name = "buf_ops_nfs", .bop_write = nfs_bwrite, .bop_strategy = bufstrategy, .bop_sync = bufsync, .bop_bdflush = bufbdflush, }; static int nfs_getacl(struct vop_getacl_args *ap) { int error; if (ap->a_type != ACL_TYPE_NFS4) return (EOPNOTSUPP); error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, NULL); if (error > NFSERR_STALE) { (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); error = EPERM; } return (error); } static int nfs_setacl(struct vop_setacl_args *ap) { int error; if (ap->a_type != ACL_TYPE_NFS4) return (EOPNOTSUPP); error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, NULL); if (error > NFSERR_STALE) { (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); error = EPERM; } return (error); } /* * VOP_ADVISE for NFS. * Just return 0 for any errors, since it is just a hint. */ static int nfs_advise(struct vop_advise_args *ap) { struct thread *td = curthread; struct nfsmount *nmp; uint64_t len; int error; /* * First do vop_stdadvise() to handle the buffer cache. */ error = vop_stdadvise(ap); if (error != 0) return (error); if (ap->a_start < 0 || ap->a_end < 0) return (0); if (ap->a_end == OFF_MAX) len = 0; else if (ap->a_end < ap->a_start) return (0); else len = ap->a_end - ap->a_start + 1; nmp = VFSTONFS(ap->a_vp->v_mount); mtx_lock(&nmp->nm_mtx); if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || (NFSHASPNFS(nmp) && (nmp->nm_privflag & NFSMNTP_IOADVISETHRUMDS) == 0) || (nmp->nm_privflag & NFSMNTP_NOADVISE) != 0) { mtx_unlock(&nmp->nm_mtx); return (0); } mtx_unlock(&nmp->nm_mtx); error = nfsrpc_advise(ap->a_vp, ap->a_start, len, ap->a_advice, td->td_ucred, td); if (error == NFSERR_NOTSUPP) { mtx_lock(&nmp->nm_mtx); nmp->nm_privflag |= NFSMNTP_NOADVISE; mtx_unlock(&nmp->nm_mtx); } return (0); } /* * nfs allocate call */ static int nfs_allocate(struct vop_allocate_args *ap) { struct vnode *vp = ap->a_vp; struct thread *td = curthread; struct nfsvattr nfsva; struct nfsmount *nmp; off_t alen; int attrflag, error, ret; attrflag = 0; nmp = VFSTONFS(vp->v_mount); mtx_lock(&nmp->nm_mtx); if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION && (nmp->nm_privflag & NFSMNTP_NOALLOCATE) == 0) { mtx_unlock(&nmp->nm_mtx); /* * Flush first to ensure that the allocate adds to the * file's allocation on the server. */ error = ncl_flush(vp, MNT_WAIT, td, 1, 0); if (error == 0) { alen = *ap->a_len; if ((uint64_t)alen > nfs_maxalloclen) alen = nfs_maxalloclen; error = nfsrpc_allocate(vp, *ap->a_offset, alen, &nfsva, &attrflag, td->td_ucred, td, NULL); } if (error == 0) { *ap->a_offset += alen; *ap->a_len -= alen; } else if (error == NFSERR_NOTSUPP) { mtx_lock(&nmp->nm_mtx); nmp->nm_privflag |= NFSMNTP_NOALLOCATE; mtx_unlock(&nmp->nm_mtx); error = EINVAL; } } else { mtx_unlock(&nmp->nm_mtx); error = EINVAL; } if (attrflag != 0) { ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); if (error == 0 && ret != 0) error = ret; } if (error != 0) error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); return (error); } /* * nfs deallocate call */ static int nfs_deallocate(struct vop_deallocate_args *ap) { struct vnode *vp = ap->a_vp; struct thread *td = curthread; struct nfsvattr nfsva; struct nfsmount *nmp; struct nfsnode *np; off_t tlen, mlen; int attrflag, error, ret; bool clipped; error = 0; attrflag = 0; nmp = VFSTONFS(vp->v_mount); np = VTONFS(vp); mtx_lock(&nmp->nm_mtx); if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION && (nmp->nm_privflag & NFSMNTP_NODEALLOCATE) == 0) { mtx_unlock(&nmp->nm_mtx); tlen = omin(OFF_MAX - *ap->a_offset, *ap->a_len); NFSCL_DEBUG(4, "dealloc: off=%jd len=%jd maxfilesize=%ju\n", (intmax_t)*ap->a_offset, (intmax_t)tlen, (uintmax_t)nmp->nm_maxfilesize); if ((uint64_t)*ap->a_offset >= nmp->nm_maxfilesize) { /* Avoid EFBIG error return from the NFSv4.2 server. */ *ap->a_len = 0; return (0); } clipped = false; if ((uint64_t)*ap->a_offset + tlen > nmp->nm_maxfilesize) tlen = nmp->nm_maxfilesize - *ap->a_offset; if ((uint64_t)*ap->a_offset < np->n_size) { /* Limit the len to nfs_maxalloclen before EOF. */ mlen = omin((off_t)np->n_size - *ap->a_offset, tlen); if ((uint64_t)mlen > nfs_maxalloclen) { NFSCL_DEBUG(4, "dealloc: tlen maxalloclen\n"); tlen = nfs_maxalloclen; clipped = true; } } if (error == 0) error = ncl_vinvalbuf(vp, V_SAVE, td, 1); if (error == 0) { vnode_pager_purge_range(vp, *ap->a_offset, *ap->a_offset + tlen); error = nfsrpc_deallocate(vp, *ap->a_offset, tlen, &nfsva, &attrflag, ap->a_cred, td, NULL); NFSCL_DEBUG(4, "dealloc: rpc=%d\n", error); } if (error == 0) { NFSCL_DEBUG(4, "dealloc: attrflag=%d na_size=%ju\n", attrflag, (uintmax_t)nfsva.na_size); if (attrflag != 0) { if ((uint64_t)*ap->a_offset < nfsva.na_size) *ap->a_offset += omin((off_t) nfsva.na_size - *ap->a_offset, tlen); } if (clipped && tlen < *ap->a_len) *ap->a_len -= tlen; else *ap->a_len = 0; } else if (error == NFSERR_NOTSUPP) { mtx_lock(&nmp->nm_mtx); nmp->nm_privflag |= NFSMNTP_NODEALLOCATE; mtx_unlock(&nmp->nm_mtx); } } else { mtx_unlock(&nmp->nm_mtx); error = EIO; } /* * If the NFS server cannot perform the Deallocate operation, just call * vop_stddeallocate() to perform it. */ if (error != 0 && error != NFSERR_FBIG && error != NFSERR_INVAL) { error = vop_stddeallocate(ap); NFSCL_DEBUG(4, "dealloc: stddeallocate=%d\n", error); } if (attrflag != 0) { ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); if (error == 0 && ret != 0) error = ret; } if (error != 0) error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); return (error); } /* * nfs copy_file_range call */ static int nfs_copy_file_range(struct vop_copy_file_range_args *ap) { struct vnode *invp = ap->a_invp; struct vnode *outvp = ap->a_outvp; struct mount *mp; struct nfsvattr innfsva, outnfsva; struct vattr *vap; struct uio io; struct nfsmount *nmp; size_t len, len2; int error, inattrflag, outattrflag, ret, ret2; off_t inoff, outoff; bool consecutive, must_commit, tryoutcred; ret = ret2 = 0; nmp = VFSTONFS(invp->v_mount); mtx_lock(&nmp->nm_mtx); /* NFSv4.2 Copy is not permitted for infile == outfile. */ if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || (nmp->nm_privflag & NFSMNTP_NOCOPY) != 0 || invp == outvp) { mtx_unlock(&nmp->nm_mtx); error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp, ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags, ap->a_incred, ap->a_outcred, ap->a_fsizetd); return (error); } mtx_unlock(&nmp->nm_mtx); /* Lock both vnodes, avoiding risk of deadlock. */ do { mp = NULL; error = vn_start_write(outvp, &mp, V_WAIT); if (error == 0) { error = vn_lock(outvp, LK_EXCLUSIVE); if (error == 0) { error = vn_lock(invp, LK_SHARED | LK_NOWAIT); if (error == 0) break; VOP_UNLOCK(outvp); if (mp != NULL) vn_finished_write(mp); mp = NULL; error = vn_lock(invp, LK_SHARED); if (error == 0) VOP_UNLOCK(invp); } } if (mp != NULL) vn_finished_write(mp); } while (error == 0); if (error != 0) return (error); /* * Do the vn_rlimit_fsize() check. Should this be above the VOP layer? */ io.uio_offset = *ap->a_outoffp; io.uio_resid = *ap->a_lenp; error = vn_rlimit_fsize(outvp, &io, ap->a_fsizetd); /* * Flush the input file so that the data is up to date before * the copy. Flush writes for the output file so that they * do not overwrite the data copied to the output file by the Copy. * Set the commit argument for both flushes so that the data is on * stable storage before the Copy RPC. This is done in case the * server reboots during the Copy and needs to be redone. */ if (error == 0) error = ncl_flush(invp, MNT_WAIT, curthread, 1, 0); if (error == 0) error = ncl_flush(outvp, MNT_WAIT, curthread, 1, 0); /* Do the actual NFSv4.2 RPC. */ len = *ap->a_lenp; mtx_lock(&nmp->nm_mtx); if ((nmp->nm_privflag & NFSMNTP_NOCONSECUTIVE) == 0) consecutive = true; else consecutive = false; mtx_unlock(&nmp->nm_mtx); inoff = *ap->a_inoffp; outoff = *ap->a_outoffp; tryoutcred = true; must_commit = false; if (error == 0) { vap = &VTONFS(invp)->n_vattr.na_vattr; error = VOP_GETATTR(invp, vap, ap->a_incred); if (error == 0) { /* * Clip "len" at va_size so that RFC compliant servers * will not reply NFSERR_INVAL. * Setting "len == 0" for the RPC would be preferred, * but some Linux servers do not support that. */ if (inoff >= vap->va_size) *ap->a_lenp = len = 0; else if (inoff + len > vap->va_size) *ap->a_lenp = len = vap->va_size - inoff; } else error = 0; } /* * len will be set to 0 upon a successful Copy RPC. * As such, this only loops when the Copy RPC needs to be retried. */ while (len > 0 && error == 0) { inattrflag = outattrflag = 0; len2 = len; if (tryoutcred) error = nfsrpc_copy_file_range(invp, ap->a_inoffp, outvp, ap->a_outoffp, &len2, ap->a_flags, &inattrflag, &innfsva, &outattrflag, &outnfsva, ap->a_outcred, consecutive, &must_commit); else error = nfsrpc_copy_file_range(invp, ap->a_inoffp, outvp, ap->a_outoffp, &len2, ap->a_flags, &inattrflag, &innfsva, &outattrflag, &outnfsva, ap->a_incred, consecutive, &must_commit); if (inattrflag != 0) ret = nfscl_loadattrcache(&invp, &innfsva, NULL, NULL, 0, 1); if (outattrflag != 0) ret2 = nfscl_loadattrcache(&outvp, &outnfsva, NULL, NULL, 1, 1); if (error == 0) { if (consecutive == false) { if (len2 == len) { mtx_lock(&nmp->nm_mtx); nmp->nm_privflag |= NFSMNTP_NOCONSECUTIVE; mtx_unlock(&nmp->nm_mtx); } else error = NFSERR_OFFLOADNOREQS; } *ap->a_lenp = len2; len = 0; if (len2 > 0 && must_commit && error == 0) error = ncl_commit(outvp, outoff, *ap->a_lenp, ap->a_outcred, curthread); if (error == 0 && ret != 0) error = ret; if (error == 0 && ret2 != 0) error = ret2; } else if (error == NFSERR_OFFLOADNOREQS && consecutive) { /* * Try consecutive == false, which is ok only if all * bytes are copied. * If only some bytes were copied when consecutive * is false, there is no way to know which bytes * still need to be written. */ consecutive = false; error = 0; } else if (error == NFSERR_ACCES && tryoutcred) { /* Try again with incred. */ tryoutcred = false; error = 0; } if (error == NFSERR_STALEWRITEVERF) { /* * Server rebooted, so do it all again. */ *ap->a_inoffp = inoff; *ap->a_outoffp = outoff; len = *ap->a_lenp; must_commit = false; error = 0; } } VOP_UNLOCK(invp); VOP_UNLOCK(outvp); if (mp != NULL) vn_finished_write(mp); if (error == NFSERR_NOTSUPP || error == NFSERR_OFFLOADNOREQS || error == NFSERR_ACCES) { /* * Unlike the NFSv4.2 Copy, vn_generic_copy_file_range() can * use a_incred for the read and a_outcred for the write, so * try this for NFSERR_ACCES failures for the Copy. * For NFSERR_NOTSUPP and NFSERR_OFFLOADNOREQS, the Copy can * never succeed, so disable it. */ if (error != NFSERR_ACCES) { /* Can never do Copy on this mount. */ mtx_lock(&nmp->nm_mtx); nmp->nm_privflag |= NFSMNTP_NOCOPY; mtx_unlock(&nmp->nm_mtx); } *ap->a_inoffp = inoff; *ap->a_outoffp = outoff; error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp, ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags, ap->a_incred, ap->a_outcred, ap->a_fsizetd); } else if (error != 0) *ap->a_lenp = 0; if (error != 0) error = nfscl_maperr(curthread, error, (uid_t)0, (gid_t)0); return (error); } /* * nfs ioctl call */ static int nfs_ioctl(struct vop_ioctl_args *ap) { struct vnode *vp = ap->a_vp; struct nfsvattr nfsva; struct nfsmount *nmp; int attrflag, content, error, ret; bool eof = false; /* shut up compiler. */ if (vp->v_type != VREG) return (ENOTTY); nmp = VFSTONFS(vp->v_mount); if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION) { error = vop_stdioctl(ap); return (error); } /* Do the actual NFSv4.2 RPC. */ switch (ap->a_command) { case FIOSEEKDATA: content = NFSV4CONTENT_DATA; break; case FIOSEEKHOLE: content = NFSV4CONTENT_HOLE; break; default: return (ENOTTY); } error = vn_lock(vp, LK_SHARED); if (error != 0) return (EBADF); attrflag = 0; if (*((off_t *)ap->a_data) >= VTONFS(vp)->n_size) error = ENXIO; else { /* * Flush all writes, so that the server is up to date. * Although a Commit is not required, the commit argument * is set so that, for a pNFS File/Flexible File Layout * server, the LayoutCommit will be done to ensure the file * size is up to date on the Metadata Server. */ error = ncl_flush(vp, MNT_WAIT, ap->a_td, 1, 0); if (error == 0) error = nfsrpc_seek(vp, (off_t *)ap->a_data, &eof, content, ap->a_cred, &nfsva, &attrflag); /* If at eof for FIOSEEKDATA, return ENXIO. */ if (eof && error == 0 && content == NFSV4CONTENT_DATA) error = ENXIO; } if (attrflag != 0) { ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); if (error == 0 && ret != 0) error = ret; } NFSVOPUNLOCK(vp); if (error != 0) error = ENXIO; return (error); } /* * nfs getextattr call */ static int nfs_getextattr(struct vop_getextattr_args *ap) { struct vnode *vp = ap->a_vp; struct nfsmount *nmp; struct ucred *cred; struct thread *td = ap->a_td; struct nfsvattr nfsva; ssize_t len; int attrflag, error, ret; nmp = VFSTONFS(vp->v_mount); mtx_lock(&nmp->nm_mtx); if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 || ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) { mtx_unlock(&nmp->nm_mtx); return (EOPNOTSUPP); } mtx_unlock(&nmp->nm_mtx); cred = ap->a_cred; if (cred == NULL) cred = td->td_ucred; /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */ attrflag = 0; error = nfsrpc_getextattr(vp, ap->a_name, ap->a_uio, &len, &nfsva, &attrflag, cred, td); if (attrflag != 0) { ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); if (error == 0 && ret != 0) error = ret; } if (error == 0 && ap->a_size != NULL) *ap->a_size = len; switch (error) { case NFSERR_NOTSUPP: case NFSERR_OPILLEGAL: mtx_lock(&nmp->nm_mtx); nmp->nm_privflag |= NFSMNTP_NOXATTR; mtx_unlock(&nmp->nm_mtx); error = EOPNOTSUPP; break; case NFSERR_NOXATTR: case NFSERR_XATTR2BIG: error = ENOATTR; break; default: error = nfscl_maperr(td, error, 0, 0); break; } return (error); } /* * nfs setextattr call */ static int nfs_setextattr(struct vop_setextattr_args *ap) { struct vnode *vp = ap->a_vp; struct nfsmount *nmp; struct ucred *cred; struct thread *td = ap->a_td; struct nfsvattr nfsva; int attrflag, error, ret; nmp = VFSTONFS(vp->v_mount); mtx_lock(&nmp->nm_mtx); if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 || ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) { mtx_unlock(&nmp->nm_mtx); return (EOPNOTSUPP); } mtx_unlock(&nmp->nm_mtx); if (ap->a_uio->uio_resid < 0) return (EINVAL); cred = ap->a_cred; if (cred == NULL) cred = td->td_ucred; /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */ attrflag = 0; error = nfsrpc_setextattr(vp, ap->a_name, ap->a_uio, &nfsva, &attrflag, cred, td); if (attrflag != 0) { ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); if (error == 0 && ret != 0) error = ret; } switch (error) { case NFSERR_NOTSUPP: case NFSERR_OPILLEGAL: mtx_lock(&nmp->nm_mtx); nmp->nm_privflag |= NFSMNTP_NOXATTR; mtx_unlock(&nmp->nm_mtx); error = EOPNOTSUPP; break; case NFSERR_NOXATTR: case NFSERR_XATTR2BIG: error = ENOATTR; break; default: error = nfscl_maperr(td, error, 0, 0); break; } return (error); } /* * nfs listextattr call */ static int nfs_listextattr(struct vop_listextattr_args *ap) { struct vnode *vp = ap->a_vp; struct nfsmount *nmp; struct ucred *cred; struct thread *td = ap->a_td; struct nfsvattr nfsva; size_t len, len2; uint64_t cookie; int attrflag, error, ret; bool eof; nmp = VFSTONFS(vp->v_mount); mtx_lock(&nmp->nm_mtx); if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 || ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) { mtx_unlock(&nmp->nm_mtx); return (EOPNOTSUPP); } mtx_unlock(&nmp->nm_mtx); cred = ap->a_cred; if (cred == NULL) cred = td->td_ucred; /* Loop around doing List Extended Attribute RPCs. */ eof = false; cookie = 0; len2 = 0; error = 0; while (!eof && error == 0) { len = nmp->nm_rsize; attrflag = 0; error = nfsrpc_listextattr(vp, &cookie, ap->a_uio, &len, &eof, &nfsva, &attrflag, cred, td); if (attrflag != 0) { ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); if (error == 0 && ret != 0) error = ret; } if (error == 0) { len2 += len; if (len2 > SSIZE_MAX) error = ENOATTR; } } if (error == 0 && ap->a_size != NULL) *ap->a_size = len2; switch (error) { case NFSERR_NOTSUPP: case NFSERR_OPILLEGAL: mtx_lock(&nmp->nm_mtx); nmp->nm_privflag |= NFSMNTP_NOXATTR; mtx_unlock(&nmp->nm_mtx); error = EOPNOTSUPP; break; case NFSERR_NOXATTR: case NFSERR_XATTR2BIG: error = ENOATTR; break; default: error = nfscl_maperr(td, error, 0, 0); break; } return (error); } /* * nfs setextattr call */ static int nfs_deleteextattr(struct vop_deleteextattr_args *ap) { struct vnode *vp = ap->a_vp; struct nfsmount *nmp; struct nfsvattr nfsva; int attrflag, error, ret; nmp = VFSTONFS(vp->v_mount); mtx_lock(&nmp->nm_mtx); if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION || (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 || ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) { mtx_unlock(&nmp->nm_mtx); return (EOPNOTSUPP); } mtx_unlock(&nmp->nm_mtx); /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */ attrflag = 0; error = nfsrpc_rmextattr(vp, ap->a_name, &nfsva, &attrflag, ap->a_cred, ap->a_td); if (attrflag != 0) { ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); if (error == 0 && ret != 0) error = ret; } switch (error) { case NFSERR_NOTSUPP: case NFSERR_OPILLEGAL: mtx_lock(&nmp->nm_mtx); nmp->nm_privflag |= NFSMNTP_NOXATTR; mtx_unlock(&nmp->nm_mtx); error = EOPNOTSUPP; break; case NFSERR_NOXATTR: case NFSERR_XATTR2BIG: error = ENOATTR; break; default: error = nfscl_maperr(ap->a_td, error, 0, 0); break; } return (error); } /* * Return POSIX pathconf information applicable to nfs filesystems. */ static int nfs_pathconf(struct vop_pathconf_args *ap) { struct nfsv3_pathconf pc; struct nfsvattr nfsva; struct vnode *vp = ap->a_vp; struct nfsmount *nmp; struct thread *td = curthread; off_t off; bool eof; int attrflag, error; if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX || ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED || ap->a_name == _PC_NO_TRUNC)) || (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) { /* * Since only the above 4 a_names are returned by the NFSv3 * Pathconf RPC, there is no point in doing it for others. * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can * be used for _PC_NFS4_ACL as well. */ error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva, &attrflag, NULL); if (attrflag != 0) (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); if (error != 0) return (error); } else { /* * For NFSv2 (or NFSv3 when not one of the above 4 a_names), * just fake them. */ pc.pc_linkmax = NFS_LINK_MAX; pc.pc_namemax = NFS_MAXNAMLEN; pc.pc_notrunc = 1; pc.pc_chownrestricted = 1; pc.pc_caseinsensitive = 0; pc.pc_casepreserving = 1; error = 0; } switch (ap->a_name) { case _PC_LINK_MAX: #ifdef _LP64 *ap->a_retval = pc.pc_linkmax; #else *ap->a_retval = MIN(LONG_MAX, pc.pc_linkmax); #endif break; case _PC_NAME_MAX: *ap->a_retval = pc.pc_namemax; break; case _PC_PIPE_BUF: if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO) *ap->a_retval = PIPE_BUF; else error = EINVAL; break; case _PC_CHOWN_RESTRICTED: *ap->a_retval = pc.pc_chownrestricted; break; case _PC_NO_TRUNC: *ap->a_retval = pc.pc_notrunc; break; case _PC_ACL_NFS4: if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 && NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL)) *ap->a_retval = 1; else *ap->a_retval = 0; break; case _PC_ACL_PATH_MAX: if (NFS_ISV4(vp)) *ap->a_retval = ACL_MAX_ENTRIES; else *ap->a_retval = 3; break; case _PC_PRIO_IO: *ap->a_retval = 0; break; case _PC_SYNC_IO: *ap->a_retval = 0; break; case _PC_ALLOC_SIZE_MIN: *ap->a_retval = vp->v_mount->mnt_stat.f_bsize; break; case _PC_FILESIZEBITS: if (NFS_ISV34(vp)) *ap->a_retval = 64; else *ap->a_retval = 32; break; case _PC_REC_INCR_XFER_SIZE: *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; break; case _PC_REC_MAX_XFER_SIZE: *ap->a_retval = -1; /* means ``unlimited'' */ break; case _PC_REC_MIN_XFER_SIZE: *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; break; case _PC_REC_XFER_ALIGN: *ap->a_retval = PAGE_SIZE; break; case _PC_SYMLINK_MAX: *ap->a_retval = NFS_MAXPATHLEN; break; case _PC_MIN_HOLE_SIZE: /* Only some NFSv4.2 servers support Seek for Holes. */ *ap->a_retval = 0; nmp = VFSTONFS(vp->v_mount); if (NFS_ISV4(vp) && nmp->nm_minorvers == NFSV42_MINORVERSION) { /* * NFSv4.2 doesn't have an attribute for hole size, * so all we can do is see if the Seek operation is * supported and then use f_iosize as a "best guess". */ mtx_lock(&nmp->nm_mtx); if ((nmp->nm_privflag & NFSMNTP_SEEKTESTED) == 0) { mtx_unlock(&nmp->nm_mtx); off = 0; attrflag = 0; error = nfsrpc_seek(vp, &off, &eof, NFSV4CONTENT_HOLE, td->td_ucred, &nfsva, &attrflag); if (attrflag != 0) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); mtx_lock(&nmp->nm_mtx); if (error == NFSERR_NOTSUPP) nmp->nm_privflag |= NFSMNTP_SEEKTESTED; else nmp->nm_privflag |= NFSMNTP_SEEKTESTED | NFSMNTP_SEEK; error = 0; } if ((nmp->nm_privflag & NFSMNTP_SEEK) != 0) *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; mtx_unlock(&nmp->nm_mtx); } break; default: error = vop_stdpathconf(ap); break; } return (error); } diff --git a/sys/fs/nfsserver/nfs_nfsdport.c b/sys/fs/nfsserver/nfs_nfsdport.c index 0f678cd5640e..5fa6853446ae 100644 --- a/sys/fs/nfsserver/nfs_nfsdport.c +++ b/sys/fs/nfsserver/nfs_nfsdport.c @@ -1,7068 +1,7062 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Rick Macklem at The University of Guelph. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * */ #include __FBSDID("$FreeBSD$"); #include #include /* * Functions that perform the vfs operations required by the routines in * nfsd_serv.c. It is hoped that this change will make the server more * portable. */ #include #include #include #include #include #include #include FEATURE(nfsd, "NFSv4 server"); extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1; extern int nfsrv_useacl; extern int newnfs_numnfsd; extern struct mount nfsv4root_mnt; extern struct nfsrv_stablefirst nfsrv_stablefirst; extern void (*nfsd_call_servertimer)(void); extern SVCPOOL *nfsrvd_pool; extern struct nfsv4lock nfsd_suspend_lock; extern struct nfsclienthashhead *nfsclienthash; extern struct nfslockhashhead *nfslockhash; extern struct nfssessionhash *nfssessionhash; extern int nfsrv_sessionhashsize; extern struct nfsstatsv1 nfsstatsv1; extern struct nfslayouthash *nfslayouthash; extern int nfsrv_layouthashsize; extern struct mtx nfsrv_dslock_mtx; extern int nfs_pnfsiothreads; extern struct nfsdontlisthead nfsrv_dontlisthead; extern volatile int nfsrv_dontlistlen; extern volatile int nfsrv_devidcnt; extern int nfsrv_maxpnfsmirror; extern uint32_t nfs_srvmaxio; extern int nfs_bufpackets; extern u_long sb_max_adj; struct vfsoptlist nfsv4root_opt, nfsv4root_newopt; NFSDLOCKMUTEX; NFSSTATESPINLOCK; struct nfsrchash_bucket nfsrchash_table[NFSRVCACHE_HASHSIZE]; struct nfsrchash_bucket nfsrcahash_table[NFSRVCACHE_HASHSIZE]; struct mtx nfsrc_udpmtx; struct mtx nfs_v4root_mutex; struct mtx nfsrv_dontlistlock_mtx; struct mtx nfsrv_recalllock_mtx; struct nfsrvfh nfs_rootfh, nfs_pubfh; int nfs_pubfhset = 0, nfs_rootfhset = 0; struct proc *nfsd_master_proc = NULL; int nfsd_debuglevel = 0; static pid_t nfsd_master_pid = (pid_t)-1; static char nfsd_master_comm[MAXCOMLEN + 1]; static struct timeval nfsd_master_start; static uint32_t nfsv4_sysid = 0; static fhandle_t zerofh; static int nfssvc_srvcall(struct thread *, struct nfssvc_args *, struct ucred *); int nfsrv_enable_crossmntpt = 1; static int nfs_commit_blks; static int nfs_commit_miss; extern int nfsrv_issuedelegs; extern int nfsrv_dolocallocks; extern int nfsd_enable_stringtouid; extern struct nfsdevicehead nfsrv_devidhead; static int nfsrv_createiovec(int, struct mbuf **, struct mbuf **, struct iovec **); static int nfsrv_createiovec_extpgs(int, int, struct mbuf **, struct mbuf **, struct iovec **); static int nfsrv_createiovecw(int, struct mbuf *, char *, struct iovec **, int *); static void nfsrv_pnfscreate(struct vnode *, struct vattr *, struct ucred *, NFSPROC_T *); static void nfsrv_pnfsremovesetup(struct vnode *, NFSPROC_T *, struct vnode **, int *, char *, fhandle_t *); static void nfsrv_pnfsremove(struct vnode **, int, char *, fhandle_t *, NFSPROC_T *); static int nfsrv_proxyds(struct vnode *, off_t, int, struct ucred *, struct thread *, int, struct mbuf **, char *, struct mbuf **, struct nfsvattr *, struct acl *, off_t *, int, bool *); static int nfsrv_setextattr(struct vnode *, struct nfsvattr *, NFSPROC_T *); static int nfsrv_readdsrpc(fhandle_t *, off_t, int, struct ucred *, NFSPROC_T *, struct nfsmount *, struct mbuf **, struct mbuf **); static int nfsrv_writedsrpc(fhandle_t *, off_t, int, struct ucred *, NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct mbuf **, char *, int *); static int nfsrv_allocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *, NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *); static int nfsrv_deallocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *, NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *); static int nfsrv_setacldsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct acl *, int *); static int nfsrv_setattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct nfsvattr *, int *); static int nfsrv_getattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, struct vnode *, struct nfsmount *, struct nfsvattr *); static int nfsrv_seekdsrpc(fhandle_t *, off_t *, int, bool *, struct ucred *, NFSPROC_T *, struct nfsmount *); static int nfsrv_putfhname(fhandle_t *, char *); static int nfsrv_pnfslookupds(struct vnode *, struct vnode *, struct pnfsdsfile *, struct vnode **, NFSPROC_T *); static void nfsrv_pnfssetfh(struct vnode *, struct pnfsdsfile *, char *, char *, struct vnode *, NFSPROC_T *); static int nfsrv_dsremove(struct vnode *, char *, struct ucred *, NFSPROC_T *); static int nfsrv_dssetacl(struct vnode *, struct acl *, struct ucred *, NFSPROC_T *); static int nfsrv_pnfsstatfs(struct statfs *, struct mount *); int nfs_pnfsio(task_fn_t *, void *); SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "NFS server"); SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW, &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points"); SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks, 0, ""); SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss, 0, ""); SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW, &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations"); SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW, &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files"); SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel, 0, "Debug level for NFS server"); SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, CTLFLAG_RW, &nfsd_enable_stringtouid, 0, "Enable nfsd to accept numeric owner_names"); static int nfsrv_pnfsgetdsattr = 1; SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsgetdsattr, CTLFLAG_RW, &nfsrv_pnfsgetdsattr, 0, "When set getattr gets DS attributes via RPC"); /* * nfsrv_dsdirsize can only be increased and only when the nfsd threads are * not running. * The dsN subdirectories for the increased values must have been created * on all DS servers before this increase is done. */ u_int nfsrv_dsdirsize = 20; static int sysctl_dsdirsize(SYSCTL_HANDLER_ARGS) { int error, newdsdirsize; newdsdirsize = nfsrv_dsdirsize; error = sysctl_handle_int(oidp, &newdsdirsize, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (newdsdirsize <= nfsrv_dsdirsize || newdsdirsize > 10000 || newnfs_numnfsd != 0) return (EINVAL); nfsrv_dsdirsize = newdsdirsize; return (0); } SYSCTL_PROC(_vfs_nfsd, OID_AUTO, dsdirsize, CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(nfsrv_dsdirsize), sysctl_dsdirsize, "IU", "Number of dsN subdirs on the DS servers"); /* * nfs_srvmaxio can only be increased and only when the nfsd threads are * not running. The setting must be a power of 2, with the current limit of * 1Mbyte. */ static int sysctl_srvmaxio(SYSCTL_HANDLER_ARGS) { int error; u_int newsrvmaxio; uint64_t tval; newsrvmaxio = nfs_srvmaxio; error = sysctl_handle_int(oidp, &newsrvmaxio, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (newsrvmaxio == nfs_srvmaxio) return (0); if (newsrvmaxio < nfs_srvmaxio) { printf("nfsd: vfs.nfsd.srvmaxio can only be increased\n"); return (EINVAL); } if (newsrvmaxio > 1048576) { printf("nfsd: vfs.nfsd.srvmaxio cannot be > 1Mbyte\n"); return (EINVAL); } if ((newsrvmaxio & (newsrvmaxio - 1)) != 0) { printf("nfsd: vfs.nfsd.srvmaxio must be a power of 2\n"); return (EINVAL); } /* * Check that kern.ipc.maxsockbuf is large enough for * newsrviomax, given the setting of vfs.nfs.bufpackets. */ if ((newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets > sb_max_adj) { /* * Suggest vfs.nfs.bufpackets * maximum RPC message for * sb_max_adj. */ tval = (newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets; /* * Convert suggested sb_max_adj value to a suggested * sb_max value, which is what is set via kern.ipc.maxsockbuf. * Perform the inverse calculation of (from uipc_sockbuf.c): * sb_max_adj = (u_quad_t)sb_max * MCLBYTES / * (MSIZE + MCLBYTES); * XXX If the calculation of sb_max_adj from sb_max changes, * this calculation must be changed as well. */ tval *= (MSIZE + MCLBYTES); /* Brackets for readability. */ tval += MCLBYTES - 1; /* Round up divide. */ tval /= MCLBYTES; printf("nfsd: set kern.ipc.maxsockbuf to a minimum of " "%ju to support %ubyte NFS I/O\n", (uintmax_t)tval, newsrvmaxio); return (EINVAL); } NFSD_LOCK(); if (newnfs_numnfsd != 0) { NFSD_UNLOCK(); printf("nfsd: cannot set vfs.nfsd.srvmaxio when nfsd " "threads are running\n"); return (EINVAL); } nfs_srvmaxio = newsrvmaxio; NFSD_UNLOCK(); return (0); } SYSCTL_PROC(_vfs_nfsd, OID_AUTO, srvmaxio, CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0, sysctl_srvmaxio, "IU", "Maximum I/O size in bytes"); #define MAX_REORDERED_RPC 16 #define NUM_HEURISTIC 1031 #define NHUSE_INIT 64 #define NHUSE_INC 16 #define NHUSE_MAX 2048 static struct nfsheur { struct vnode *nh_vp; /* vp to match (unreferenced pointer) */ off_t nh_nextoff; /* next offset for sequential detection */ int nh_use; /* use count for selection */ int nh_seqcount; /* heuristic */ } nfsheur[NUM_HEURISTIC]; /* * Heuristic to detect sequential operation. */ static struct nfsheur * nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp) { struct nfsheur *nh; int hi, try; /* Locate best candidate. */ try = 32; hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC; nh = &nfsheur[hi]; while (try--) { if (nfsheur[hi].nh_vp == vp) { nh = &nfsheur[hi]; break; } if (nfsheur[hi].nh_use > 0) --nfsheur[hi].nh_use; hi = (hi + 1) % NUM_HEURISTIC; if (nfsheur[hi].nh_use < nh->nh_use) nh = &nfsheur[hi]; } /* Initialize hint if this is a new file. */ if (nh->nh_vp != vp) { nh->nh_vp = vp; nh->nh_nextoff = uio->uio_offset; nh->nh_use = NHUSE_INIT; if (uio->uio_offset == 0) nh->nh_seqcount = 4; else nh->nh_seqcount = 1; } /* Calculate heuristic. */ if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) || uio->uio_offset == nh->nh_nextoff) { /* See comments in vfs_vnops.c:sequential_heuristic(). */ nh->nh_seqcount += howmany(uio->uio_resid, 16384); if (nh->nh_seqcount > IO_SEQMAX) nh->nh_seqcount = IO_SEQMAX; } else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC * imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) { /* Probably a reordered RPC, leave seqcount alone. */ } else if (nh->nh_seqcount > 1) { nh->nh_seqcount /= 2; } else { nh->nh_seqcount = 0; } nh->nh_use += NHUSE_INC; if (nh->nh_use > NHUSE_MAX) nh->nh_use = NHUSE_MAX; return (nh); } /* * Get attributes into nfsvattr structure. */ int nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, struct nfsrv_descript *nd, struct thread *p, int vpislocked, nfsattrbit_t *attrbitp) { int error, gotattr, lockedit = 0; struct nfsvattr na; if (vpislocked == 0) { /* * When vpislocked == 0, the vnode is either exclusively * locked by this thread or not locked by this thread. * As such, shared lock it, if not exclusively locked. */ if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) { lockedit = 1; NFSVOPLOCK(vp, LK_SHARED | LK_RETRY); } } /* * Acquire the Change, Size, TimeAccess, TimeModify and SpaceUsed * attributes, as required. * This needs to be done for regular files if: * - non-NFSv4 RPCs or * - when attrbitp == NULL or * - an NFSv4 RPC with any of the above attributes in attrbitp. * A return of 0 for nfsrv_proxyds() indicates that it has acquired * these attributes. nfsrv_proxyds() will return an error if the * server is not a pNFS one. */ gotattr = 0; if (vp->v_type == VREG && nfsrv_devidcnt > 0 && (attrbitp == NULL || (nd->nd_flag & ND_NFSV4) == 0 || NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_CHANGE) || NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE) || NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS) || NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEMODIFY) || NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEUSED))) { error = nfsrv_proxyds(vp, 0, 0, nd->nd_cred, p, NFSPROC_GETATTR, NULL, NULL, NULL, &na, NULL, NULL, 0, NULL); if (error == 0) gotattr = 1; } error = VOP_GETATTR(vp, &nvap->na_vattr, nd->nd_cred); if (lockedit != 0) NFSVOPUNLOCK(vp); /* * If we got the Change, Size and Modify Time from the DS, * replace them. */ if (gotattr != 0) { nvap->na_atime = na.na_atime; nvap->na_mtime = na.na_mtime; nvap->na_filerev = na.na_filerev; nvap->na_size = na.na_size; nvap->na_bytes = na.na_bytes; } NFSD_DEBUG(4, "nfsvno_getattr: gotattr=%d err=%d chg=%ju\n", gotattr, error, (uintmax_t)na.na_filerev); NFSEXITCODE(error); return (error); } /* * Get a file handle for a vnode. */ int nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p) { int error; NFSBZERO((caddr_t)fhp, sizeof(fhandle_t)); fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid; error = VOP_VPTOFH(vp, &fhp->fh_fid); NFSEXITCODE(error); return (error); } /* * Perform access checking for vnodes obtained from file handles that would * refer to files already opened by a Unix client. You cannot just use * vn_writechk() and VOP_ACCESSX() for two reasons. * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write * case. * 2 - The owner is to be given access irrespective of mode bits for some * operations, so that processes that chmod after opening a file don't * break. */ int nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred, struct nfsexstuff *exp, struct thread *p, int override, int vpislocked, u_int32_t *supportedtypep) { struct vattr vattr; int error = 0, getret = 0; if (vpislocked == 0) { if (NFSVOPLOCK(vp, LK_SHARED) != 0) { error = EPERM; goto out; } } if (accmode & VWRITE) { /* Just vn_writechk() changed to check rdonly */ /* * Disallow write attempts on read-only file systems; * unless the file is a socket or a block or character * device resident on the file system. */ if (NFSVNO_EXRDONLY(exp) || (vp->v_mount->mnt_flag & MNT_RDONLY)) { switch (vp->v_type) { case VREG: case VDIR: case VLNK: error = EROFS; default: break; } } /* * If there's shared text associated with * the inode, try to free it up once. If * we fail, we can't allow writing. */ if (VOP_IS_TEXT(vp) && error == 0) error = ETXTBSY; } if (error != 0) { if (vpislocked == 0) NFSVOPUNLOCK(vp); goto out; } /* * Should the override still be applied when ACLs are enabled? */ error = VOP_ACCESSX(vp, accmode, cred, p); if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) { /* * Try again with VEXPLICIT_DENY, to see if the test for * deletion is supported. */ error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p); if (error == 0) { if (vp->v_type == VDIR) { accmode &= ~(VDELETE | VDELETE_CHILD); accmode |= VWRITE; error = VOP_ACCESSX(vp, accmode, cred, p); } else if (supportedtypep != NULL) { *supportedtypep &= ~NFSACCESS_DELETE; } } } /* * Allow certain operations for the owner (reads and writes * on files that are already open). */ if (override != NFSACCCHK_NOOVERRIDE && (error == EPERM || error == EACCES)) { if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT)) error = 0; else if (override & NFSACCCHK_ALLOWOWNER) { getret = VOP_GETATTR(vp, &vattr, cred); if (getret == 0 && cred->cr_uid == vattr.va_uid) error = 0; } } if (vpislocked == 0) NFSVOPUNLOCK(vp); out: NFSEXITCODE(error); return (error); } /* * Set attribute(s) vnop. */ int nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred, struct thread *p, struct nfsexstuff *exp) { u_quad_t savsize = 0; int error, savedit; time_t savbtime; /* * If this is an exported file system and a pNFS service is running, * don't VOP_SETATTR() of size for the MDS file system. */ savedit = 0; error = 0; if (vp->v_type == VREG && (vp->v_mount->mnt_flag & MNT_EXPORTED) != 0 && nfsrv_devidcnt != 0 && nvap->na_vattr.va_size != VNOVAL && nvap->na_vattr.va_size > 0) { savsize = nvap->na_vattr.va_size; nvap->na_vattr.va_size = VNOVAL; if (nvap->na_vattr.va_uid != (uid_t)VNOVAL || nvap->na_vattr.va_gid != (gid_t)VNOVAL || nvap->na_vattr.va_mode != (mode_t)VNOVAL || nvap->na_vattr.va_atime.tv_sec != VNOVAL || nvap->na_vattr.va_mtime.tv_sec != VNOVAL) savedit = 1; else savedit = 2; } if (savedit != 2) error = VOP_SETATTR(vp, &nvap->na_vattr, cred); if (savedit != 0) nvap->na_vattr.va_size = savsize; if (error == 0 && (nvap->na_vattr.va_uid != (uid_t)VNOVAL || nvap->na_vattr.va_gid != (gid_t)VNOVAL || nvap->na_vattr.va_size != VNOVAL || nvap->na_vattr.va_mode != (mode_t)VNOVAL || nvap->na_vattr.va_atime.tv_sec != VNOVAL || nvap->na_vattr.va_mtime.tv_sec != VNOVAL)) { /* Never modify birthtime on a DS file. */ savbtime = nvap->na_vattr.va_birthtime.tv_sec; nvap->na_vattr.va_birthtime.tv_sec = VNOVAL; /* For a pNFS server, set the attributes on the DS file. */ error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETATTR, NULL, NULL, NULL, nvap, NULL, NULL, 0, NULL); nvap->na_vattr.va_birthtime.tv_sec = savbtime; if (error == ENOENT) error = 0; } NFSEXITCODE(error); return (error); } /* * Set up nameidata for a lookup() call and do it. */ int nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp, struct vnode *dp, int islocked, struct nfsexstuff *exp, struct vnode **retdirp) { struct componentname *cnp = &ndp->ni_cnd; int i; struct iovec aiov; struct uio auio; int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen; int error = 0; char *cp; *retdirp = NULL; cnp->cn_nameptr = cnp->cn_pnbuf; ndp->ni_lcf = 0; /* * Extract and set starting directory. */ if (dp->v_type != VDIR) { if (islocked) vput(dp); else vrele(dp); nfsvno_relpathbuf(ndp); error = ENOTDIR; goto out1; } if (islocked) NFSVOPUNLOCK(dp); VREF(dp); *retdirp = dp; if (NFSVNO_EXRDONLY(exp)) cnp->cn_flags |= RDONLY; ndp->ni_segflg = UIO_SYSSPACE; if (nd->nd_flag & ND_PUBLOOKUP) { ndp->ni_loopcnt = 0; if (cnp->cn_pnbuf[0] == '/') { vrele(dp); /* * Check for degenerate pathnames here, since lookup() * panics on them. */ for (i = 1; i < ndp->ni_pathlen; i++) if (cnp->cn_pnbuf[i] != '/') break; if (i == ndp->ni_pathlen) { error = NFSERR_ACCES; goto out; } dp = rootvnode; VREF(dp); } } else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) || (nd->nd_flag & ND_NFSV4) == 0) { /* * Only cross mount points for NFSv4 when doing a * mount while traversing the file system above * the mount point, unless nfsrv_enable_crossmntpt is set. */ cnp->cn_flags |= NOCROSSMOUNT; } /* * Initialize for scan, set ni_startdir and bump ref on dp again * because lookup() will dereference ni_startdir. */ - cnp->cn_thread = curthread; ndp->ni_startdir = dp; ndp->ni_rootdir = rootvnode; ndp->ni_topdir = NULL; if (!lockleaf) cnp->cn_flags |= LOCKLEAF; for (;;) { cnp->cn_nameptr = cnp->cn_pnbuf; /* * Call lookup() to do the real work. If an error occurs, * ndp->ni_vp and ni_dvp are left uninitialized or NULL and * we do not have to dereference anything before returning. * In either case ni_startdir will be dereferenced and NULLed * out. */ error = lookup(ndp); if (error) break; /* * Check for encountering a symbolic link. Trivial * termination occurs if no symlink encountered. */ if ((cnp->cn_flags & ISSYMLINK) == 0) { if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) nfsvno_relpathbuf(ndp); if (ndp->ni_vp && !lockleaf) NFSVOPUNLOCK(ndp->ni_vp); break; } /* * Validate symlink */ if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1) NFSVOPUNLOCK(ndp->ni_dvp); if (!(nd->nd_flag & ND_PUBLOOKUP)) { error = EINVAL; goto badlink2; } if (ndp->ni_loopcnt++ >= MAXSYMLINKS) { error = ELOOP; goto badlink2; } if (ndp->ni_pathlen > 1) cp = uma_zalloc(namei_zone, M_WAITOK); else cp = cnp->cn_pnbuf; aiov.iov_base = cp; aiov.iov_len = MAXPATHLEN; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_offset = 0; auio.uio_rw = UIO_READ; auio.uio_segflg = UIO_SYSSPACE; auio.uio_td = NULL; auio.uio_resid = MAXPATHLEN; error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred); if (error) { badlink1: if (ndp->ni_pathlen > 1) uma_zfree(namei_zone, cp); badlink2: vrele(ndp->ni_dvp); vput(ndp->ni_vp); break; } linklen = MAXPATHLEN - auio.uio_resid; if (linklen == 0) { error = ENOENT; goto badlink1; } if (linklen + ndp->ni_pathlen >= MAXPATHLEN) { error = ENAMETOOLONG; goto badlink1; } /* * Adjust or replace path */ if (ndp->ni_pathlen > 1) { NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen); uma_zfree(namei_zone, cnp->cn_pnbuf); cnp->cn_pnbuf = cp; } else cnp->cn_pnbuf[linklen] = '\0'; ndp->ni_pathlen += linklen; /* * Cleanup refs for next loop and check if root directory * should replace current directory. Normally ni_dvp * becomes the new base directory and is cleaned up when * we loop. Explicitly null pointers after invalidation * to clarify operation. */ vput(ndp->ni_vp); ndp->ni_vp = NULL; if (cnp->cn_pnbuf[0] == '/') { vrele(ndp->ni_dvp); ndp->ni_dvp = ndp->ni_rootdir; VREF(ndp->ni_dvp); } ndp->ni_startdir = ndp->ni_dvp; ndp->ni_dvp = NULL; } if (!lockleaf) cnp->cn_flags &= ~LOCKLEAF; out: if (error) { nfsvno_relpathbuf(ndp); ndp->ni_vp = NULL; ndp->ni_dvp = NULL; ndp->ni_startdir = NULL; } else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) { ndp->ni_dvp = NULL; } out1: NFSEXITCODE2(error, nd); return (error); } /* * Set up a pathname buffer and return a pointer to it and, optionally * set a hash pointer. */ void nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp) { struct componentname *cnp = &ndp->ni_cnd; cnp->cn_flags |= (NOMACCHECK | HASBUF); cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK); if (hashpp != NULL) *hashpp = NULL; *bufpp = cnp->cn_pnbuf; } /* * Release the above path buffer, if not released by nfsvno_namei(). */ void nfsvno_relpathbuf(struct nameidata *ndp) { if ((ndp->ni_cnd.cn_flags & HASBUF) == 0) panic("nfsrelpath"); uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf); ndp->ni_cnd.cn_flags &= ~HASBUF; } /* * Readlink vnode op into an mbuf list. */ int nfsvno_readlink(struct vnode *vp, struct ucred *cred, int maxextsiz, struct thread *p, struct mbuf **mpp, struct mbuf **mpendp, int *lenp) { struct iovec *iv; struct uio io, *uiop = &io; struct mbuf *mp, *mp3; int len, tlen, error = 0; len = NFS_MAXPATHLEN; if (maxextsiz > 0) uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz, &mp3, &mp, &iv); else uiop->uio_iovcnt = nfsrv_createiovec(len, &mp3, &mp, &iv); uiop->uio_iov = iv; uiop->uio_offset = 0; uiop->uio_resid = len; uiop->uio_rw = UIO_READ; uiop->uio_segflg = UIO_SYSSPACE; uiop->uio_td = NULL; error = VOP_READLINK(vp, uiop, cred); free(iv, M_TEMP); if (error) { m_freem(mp3); *lenp = 0; goto out; } if (uiop->uio_resid > 0) { len -= uiop->uio_resid; tlen = NFSM_RNDUP(len); if (tlen == 0) { m_freem(mp3); mp3 = mp = NULL; } else if (tlen != NFS_MAXPATHLEN || tlen != len) mp = nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len); } *lenp = len; *mpp = mp3; *mpendp = mp; out: NFSEXITCODE(error); return (error); } /* * Create an mbuf chain and an associated iovec that can be used to Read * or Getextattr of data. * Upon success, return pointers to the first and last mbufs in the chain * plus the malloc'd iovec and its iovlen. */ static int nfsrv_createiovec(int len, struct mbuf **mpp, struct mbuf **mpendp, struct iovec **ivp) { struct mbuf *m, *m2 = NULL, *m3; struct iovec *iv; int i, left, siz; left = len; m3 = NULL; /* * Generate the mbuf list with the uio_iov ref. to it. */ i = 0; while (left > 0) { NFSMGET(m); MCLGET(m, M_WAITOK); m->m_len = 0; siz = min(M_TRAILINGSPACE(m), left); left -= siz; i++; if (m3) m2->m_next = m; else m3 = m; m2 = m; } *ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK); m = m3; left = len; i = 0; while (left > 0) { if (m == NULL) panic("nfsrv_createiovec iov"); siz = min(M_TRAILINGSPACE(m), left); if (siz > 0) { iv->iov_base = mtod(m, caddr_t) + m->m_len; iv->iov_len = siz; m->m_len += siz; left -= siz; iv++; i++; } m = m->m_next; } *mpp = m3; *mpendp = m2; return (i); } /* * Create an mbuf chain and an associated iovec that can be used to Read * or Getextattr of data. * Upon success, return pointers to the first and last mbufs in the chain * plus the malloc'd iovec and its iovlen. * Same as above, but creates ext_pgs mbuf(s). */ static int nfsrv_createiovec_extpgs(int len, int maxextsiz, struct mbuf **mpp, struct mbuf **mpendp, struct iovec **ivp) { struct mbuf *m, *m2 = NULL, *m3; struct iovec *iv; int i, left, pgno, siz; left = len; m3 = NULL; /* * Generate the mbuf list with the uio_iov ref. to it. */ i = 0; while (left > 0) { siz = min(left, maxextsiz); m = mb_alloc_ext_plus_pages(siz, M_WAITOK); left -= siz; i += m->m_epg_npgs; if (m3 != NULL) m2->m_next = m; else m3 = m; m2 = m; } *ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK); m = m3; left = len; i = 0; pgno = 0; while (left > 0) { if (m == NULL) panic("nfsvno_createiovec_extpgs iov"); siz = min(PAGE_SIZE, left); if (siz > 0) { iv->iov_base = (void *)PHYS_TO_DMAP(m->m_epg_pa[pgno]); iv->iov_len = siz; m->m_len += siz; if (pgno == m->m_epg_npgs - 1) m->m_epg_last_len = siz; left -= siz; iv++; i++; pgno++; } if (pgno == m->m_epg_npgs && left > 0) { m = m->m_next; if (m == NULL) panic("nfsvno_createiovec_extpgs iov"); pgno = 0; } } *mpp = m3; *mpendp = m2; return (i); } /* * Read vnode op call into mbuf list. */ int nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred, int maxextsiz, struct thread *p, struct mbuf **mpp, struct mbuf **mpendp) { struct mbuf *m; struct iovec *iv; int error = 0, len, tlen, ioflag = 0; struct mbuf *m3; struct uio io, *uiop = &io; struct nfsheur *nh; /* * Attempt to read from a DS file. A return of ENOENT implies * there is no DS file to read. */ error = nfsrv_proxyds(vp, off, cnt, cred, p, NFSPROC_READDS, mpp, NULL, mpendp, NULL, NULL, NULL, 0, NULL); if (error != ENOENT) return (error); len = NFSM_RNDUP(cnt); if (maxextsiz > 0) uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz, &m3, &m, &iv); else uiop->uio_iovcnt = nfsrv_createiovec(len, &m3, &m, &iv); uiop->uio_iov = iv; uiop->uio_offset = off; uiop->uio_resid = len; uiop->uio_rw = UIO_READ; uiop->uio_segflg = UIO_SYSSPACE; uiop->uio_td = NULL; nh = nfsrv_sequential_heuristic(uiop, vp); ioflag |= nh->nh_seqcount << IO_SEQSHIFT; /* XXX KDM make this more systematic? */ nfsstatsv1.srvbytes[NFSV4OP_READ] += uiop->uio_resid; error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred); free(iv, M_TEMP); if (error) { m_freem(m3); *mpp = NULL; goto out; } nh->nh_nextoff = uiop->uio_offset; tlen = len - uiop->uio_resid; cnt = cnt < tlen ? cnt : tlen; tlen = NFSM_RNDUP(cnt); if (tlen == 0) { m_freem(m3); m3 = m = NULL; } else if (len != tlen || tlen != cnt) m = nfsrv_adj(m3, len - tlen, tlen - cnt); *mpp = m3; *mpendp = m; out: NFSEXITCODE(error); return (error); } /* * Create the iovec for the mbuf chain passed in as an argument. * The "cp" argument is where the data starts within the first mbuf in * the chain. It returns the iovec and the iovcnt. */ static int nfsrv_createiovecw(int retlen, struct mbuf *m, char *cp, struct iovec **ivpp, int *iovcntp) { struct mbuf *mp; struct iovec *ivp; int cnt, i, len; /* * Loop through the mbuf chain, counting how many mbufs are a * part of this write operation, so the iovec size is known. */ cnt = 0; len = retlen; mp = m; i = mtod(mp, caddr_t) + mp->m_len - cp; while (len > 0) { if (i > 0) { len -= i; cnt++; } mp = mp->m_next; if (!mp) { if (len > 0) return (EBADRPC); } else i = mp->m_len; } /* Now, create the iovec. */ mp = m; *ivpp = ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, M_WAITOK); *iovcntp = cnt; i = mtod(mp, caddr_t) + mp->m_len - cp; len = retlen; while (len > 0) { if (mp == NULL) panic("nfsrv_createiovecw"); if (i > 0) { i = min(i, len); ivp->iov_base = cp; ivp->iov_len = i; ivp++; len -= i; } mp = mp->m_next; if (mp) { i = mp->m_len; cp = mtod(mp, caddr_t); } } return (0); } /* * Write vnode op from an mbuf list. */ int nfsvno_write(struct vnode *vp, off_t off, int retlen, int *stable, struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p) { struct iovec *iv; int cnt, ioflags, error; struct uio io, *uiop = &io; struct nfsheur *nh; /* * Attempt to write to a DS file. A return of ENOENT implies * there is no DS file to write. */ error = nfsrv_proxyds(vp, off, retlen, cred, p, NFSPROC_WRITEDS, &mp, cp, NULL, NULL, NULL, NULL, 0, NULL); if (error != ENOENT) { *stable = NFSWRITE_FILESYNC; return (error); } if (*stable == NFSWRITE_UNSTABLE) ioflags = IO_NODELOCKED; else ioflags = (IO_SYNC | IO_NODELOCKED); error = nfsrv_createiovecw(retlen, mp, cp, &iv, &cnt); if (error != 0) return (error); uiop->uio_iov = iv; uiop->uio_iovcnt = cnt; uiop->uio_resid = retlen; uiop->uio_rw = UIO_WRITE; uiop->uio_segflg = UIO_SYSSPACE; NFSUIOPROC(uiop, p); uiop->uio_offset = off; nh = nfsrv_sequential_heuristic(uiop, vp); ioflags |= nh->nh_seqcount << IO_SEQSHIFT; /* XXX KDM make this more systematic? */ nfsstatsv1.srvbytes[NFSV4OP_WRITE] += uiop->uio_resid; error = VOP_WRITE(vp, uiop, ioflags, cred); if (error == 0) nh->nh_nextoff = uiop->uio_offset; free(iv, M_TEMP); NFSEXITCODE(error); return (error); } /* * Common code for creating a regular file (plus special files for V2). */ int nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp, struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp, int32_t *cverf, NFSDEV_T rdev, struct nfsexstuff *exp) { u_quad_t tempsize; int error; struct thread *p = curthread; error = nd->nd_repstat; if (!error && ndp->ni_vp == NULL) { if (nvap->na_type == VREG || nvap->na_type == VSOCK) { vrele(ndp->ni_startdir); error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); /* For a pNFS server, create the data file on a DS. */ if (error == 0 && nvap->na_type == VREG) { /* * Create a data file on a DS for a pNFS server. * This function just returns if not * running a pNFS DS or the creation fails. */ nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr, nd->nd_cred, p); } VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, false); nfsvno_relpathbuf(ndp); if (!error) { if (*exclusive_flagp) { *exclusive_flagp = 0; NFSVNO_ATTRINIT(nvap); nvap->na_atime.tv_sec = cverf[0]; nvap->na_atime.tv_nsec = cverf[1]; error = VOP_SETATTR(ndp->ni_vp, &nvap->na_vattr, nd->nd_cred); if (error != 0) { vput(ndp->ni_vp); ndp->ni_vp = NULL; error = NFSERR_NOTSUPP; } } } /* * NFS V2 Only. nfsrvd_mknod() does this for V3. * (This implies, just get out on an error.) */ } else if (nvap->na_type == VCHR || nvap->na_type == VBLK || nvap->na_type == VFIFO) { if (nvap->na_type == VCHR && rdev == 0xffffffff) nvap->na_type = VFIFO; if (nvap->na_type != VFIFO && (error = priv_check_cred(nd->nd_cred, PRIV_VFS_MKNOD_DEV))) { vrele(ndp->ni_startdir); nfsvno_relpathbuf(ndp); vput(ndp->ni_dvp); goto out; } nvap->na_rdev = rdev; error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, false); nfsvno_relpathbuf(ndp); vrele(ndp->ni_startdir); if (error) goto out; } else { vrele(ndp->ni_startdir); nfsvno_relpathbuf(ndp); vput(ndp->ni_dvp); error = ENXIO; goto out; } *vpp = ndp->ni_vp; } else { /* * Handle cases where error is already set and/or * the file exists. * 1 - clean up the lookup * 2 - iff !error and na_size set, truncate it */ vrele(ndp->ni_startdir); nfsvno_relpathbuf(ndp); *vpp = ndp->ni_vp; if (ndp->ni_dvp == *vpp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); if (!error && nvap->na_size != VNOVAL) { error = nfsvno_accchk(*vpp, VWRITE, nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, NFSACCCHK_VPISLOCKED, NULL); if (!error) { tempsize = nvap->na_size; NFSVNO_ATTRINIT(nvap); nvap->na_size = tempsize; error = VOP_SETATTR(*vpp, &nvap->na_vattr, nd->nd_cred); } } if (error) vput(*vpp); } out: NFSEXITCODE(error); return (error); } /* * Do a mknod vnode op. */ int nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred, struct thread *p) { int error = 0; enum vtype vtyp; vtyp = nvap->na_type; /* * Iff doesn't exist, create it. */ if (ndp->ni_vp) { vrele(ndp->ni_startdir); nfsvno_relpathbuf(ndp); vput(ndp->ni_dvp); vrele(ndp->ni_vp); error = EEXIST; goto out; } if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) { vrele(ndp->ni_startdir); nfsvno_relpathbuf(ndp); vput(ndp->ni_dvp); error = NFSERR_BADTYPE; goto out; } if (vtyp == VSOCK) { vrele(ndp->ni_startdir); error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, false); nfsvno_relpathbuf(ndp); } else { if (nvap->na_type != VFIFO && (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV))) { vrele(ndp->ni_startdir); nfsvno_relpathbuf(ndp); vput(ndp->ni_dvp); goto out; } error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, false); nfsvno_relpathbuf(ndp); vrele(ndp->ni_startdir); /* * Since VOP_MKNOD returns the ni_vp, I can't * see any reason to do the lookup. */ } out: NFSEXITCODE(error); return (error); } /* * Mkdir vnode op. */ int nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid, struct ucred *cred, struct thread *p, struct nfsexstuff *exp) { int error = 0; if (ndp->ni_vp != NULL) { if (ndp->ni_dvp == ndp->ni_vp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); vrele(ndp->ni_vp); nfsvno_relpathbuf(ndp); error = EEXIST; goto out; } error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, false); nfsvno_relpathbuf(ndp); out: NFSEXITCODE(error); return (error); } /* * symlink vnode op. */ int nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp, int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p, struct nfsexstuff *exp) { int error = 0; if (ndp->ni_vp) { vrele(ndp->ni_startdir); nfsvno_relpathbuf(ndp); if (ndp->ni_dvp == ndp->ni_vp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); vrele(ndp->ni_vp); error = EEXIST; goto out; } error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr, pathcp); /* * Although FreeBSD still had the lookup code in * it for 7/current, there doesn't seem to be any * point, since VOP_SYMLINK() returns the ni_vp. * Just vput it for v2. */ VOP_VPUT_PAIR(ndp->ni_dvp, &ndp->ni_vp, !not_v2 && error == 0); vrele(ndp->ni_startdir); nfsvno_relpathbuf(ndp); out: NFSEXITCODE(error); return (error); } /* * Parse symbolic link arguments. * This function has an ugly side effect. It will malloc() an area for * the symlink and set iov_base to point to it, only if it succeeds. * So, if it returns with uiop->uio_iov->iov_base != NULL, that must * be FREE'd later. */ int nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap, struct thread *p, char **pathcpp, int *lenp) { u_int32_t *tl; char *pathcp = NULL; int error = 0, len; struct nfsv2_sattr *sp; *pathcpp = NULL; *lenp = 0; if ((nd->nd_flag & ND_NFSV3) && (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p))) goto nfsmout; NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); len = fxdr_unsigned(int, *tl); if (len > NFS_MAXPATHLEN || len <= 0) { error = EBADRPC; goto nfsmout; } pathcp = malloc(len + 1, M_TEMP, M_WAITOK); error = nfsrv_mtostr(nd, pathcp, len); if (error) goto nfsmout; if (nd->nd_flag & ND_NFSV2) { NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR); nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode); } *pathcpp = pathcp; *lenp = len; NFSEXITCODE2(0, nd); return (0); nfsmout: if (pathcp) free(pathcp, M_TEMP); NFSEXITCODE2(error, nd); return (error); } /* * Remove a non-directory object. */ int nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred, struct thread *p, struct nfsexstuff *exp) { struct vnode *vp, *dsdvp[NFSDEV_MAXMIRRORS]; int error = 0, mirrorcnt; char fname[PNFS_FILENAME_LEN + 1]; fhandle_t fh; vp = ndp->ni_vp; dsdvp[0] = NULL; if (vp->v_type == VDIR) error = NFSERR_ISDIR; else if (is_v4) error = nfsrv_checkremove(vp, 1, NULL, (nfsquad_t)((u_quad_t)0), p); if (error == 0) nfsrv_pnfsremovesetup(vp, p, dsdvp, &mirrorcnt, fname, &fh); if (!error) error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd); if (error == 0 && dsdvp[0] != NULL) nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p); if (ndp->ni_dvp == vp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); vput(vp); if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0) nfsvno_relpathbuf(ndp); NFSEXITCODE(error); return (error); } /* * Remove a directory. */ int nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred, struct thread *p, struct nfsexstuff *exp) { struct vnode *vp; int error = 0; vp = ndp->ni_vp; if (vp->v_type != VDIR) { error = ENOTDIR; goto out; } /* * No rmdir "." please. */ if (ndp->ni_dvp == vp) { error = EINVAL; goto out; } /* * The root of a mounted filesystem cannot be deleted. */ if (vp->v_vflag & VV_ROOT) error = EBUSY; out: if (!error) error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd); if (ndp->ni_dvp == vp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); vput(vp); if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0) nfsvno_relpathbuf(ndp); NFSEXITCODE(error); return (error); } /* * Rename vnode op. */ int nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp, u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p) { struct vnode *fvp, *tvp, *tdvp, *dsdvp[NFSDEV_MAXMIRRORS]; int error = 0, mirrorcnt; char fname[PNFS_FILENAME_LEN + 1]; fhandle_t fh; dsdvp[0] = NULL; fvp = fromndp->ni_vp; if (ndstat) { vrele(fromndp->ni_dvp); vrele(fvp); error = ndstat; goto out1; } tdvp = tondp->ni_dvp; tvp = tondp->ni_vp; if (tvp != NULL) { if (fvp->v_type == VDIR && tvp->v_type != VDIR) { error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST; goto out; } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST; goto out; } if (tvp->v_type == VDIR && tvp->v_mountedhere) { error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; goto out; } /* * A rename to '.' or '..' results in a prematurely * unlocked vnode on FreeBSD5, so I'm just going to fail that * here. */ if ((tondp->ni_cnd.cn_namelen == 1 && tondp->ni_cnd.cn_nameptr[0] == '.') || (tondp->ni_cnd.cn_namelen == 2 && tondp->ni_cnd.cn_nameptr[0] == '.' && tondp->ni_cnd.cn_nameptr[1] == '.')) { error = EINVAL; goto out; } } if (fvp->v_type == VDIR && fvp->v_mountedhere) { error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; goto out; } if (fvp->v_mount != tdvp->v_mount) { error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; goto out; } if (fvp == tdvp) { error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL; goto out; } if (fvp == tvp) { /* * If source and destination are the same, there is nothing to * do. Set error to -1 to indicate this. */ error = -1; goto out; } if (ndflag & ND_NFSV4) { if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) { error = nfsrv_checkremove(fvp, 0, NULL, (nfsquad_t)((u_quad_t)0), p); NFSVOPUNLOCK(fvp); } else error = EPERM; if (tvp && !error) error = nfsrv_checkremove(tvp, 1, NULL, (nfsquad_t)((u_quad_t)0), p); } else { /* * For NFSv2 and NFSv3, try to get rid of the delegation, so * that the NFSv4 client won't be confused by the rename. * Since nfsd_recalldelegation() can only be called on an * unlocked vnode at this point and fvp is the file that will * still exist after the rename, just do fvp. */ nfsd_recalldelegation(fvp, p); } if (error == 0 && tvp != NULL) { nfsrv_pnfsremovesetup(tvp, p, dsdvp, &mirrorcnt, fname, &fh); NFSD_DEBUG(4, "nfsvno_rename: pnfsremovesetup" " dsdvp=%p\n", dsdvp[0]); } out: if (!error) { error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp, &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp, &tondp->ni_cnd); } else { if (tdvp == tvp) vrele(tdvp); else vput(tdvp); if (tvp) vput(tvp); vrele(fromndp->ni_dvp); vrele(fvp); if (error == -1) error = 0; } /* * If dsdvp[0] != NULL, it was set up by nfsrv_pnfsremovesetup() and * if the rename succeeded, the DS file for the tvp needs to be * removed. */ if (error == 0 && dsdvp[0] != NULL) { nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p); NFSD_DEBUG(4, "nfsvno_rename: pnfsremove\n"); } vrele(tondp->ni_startdir); nfsvno_relpathbuf(tondp); out1: vrele(fromndp->ni_startdir); nfsvno_relpathbuf(fromndp); NFSEXITCODE(error); return (error); } /* * Link vnode op. */ int nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred, struct thread *p, struct nfsexstuff *exp) { struct vnode *xp; int error = 0; xp = ndp->ni_vp; if (xp != NULL) { error = EEXIST; } else { xp = ndp->ni_dvp; if (vp->v_mount != xp->v_mount) error = EXDEV; } if (!error) { NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); if (!VN_IS_DOOMED(vp)) error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd); else error = EPERM; if (ndp->ni_dvp == vp) { vrele(ndp->ni_dvp); NFSVOPUNLOCK(vp); } else { vref(vp); VOP_VPUT_PAIR(ndp->ni_dvp, &vp, true); } } else { if (ndp->ni_dvp == ndp->ni_vp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); if (ndp->ni_vp) vrele(ndp->ni_vp); } nfsvno_relpathbuf(ndp); NFSEXITCODE(error); return (error); } /* * Do the fsync() appropriate for the commit. */ int nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred, struct thread *td) { int error = 0; /* * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of * file is done. At this time VOP_FSYNC does not accept offset and * byte count parameters so call VOP_FSYNC the whole file for now. * The same is true for NFSv4: RFC 3530 Sec. 14.2.3. * File systems that do not use the buffer cache (as indicated * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC(). */ if (cnt == 0 || cnt > MAX_COMMIT_COUNT || (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) { /* * Give up and do the whole thing */ if (vp->v_object && vm_object_mightbedirty(vp->v_object)) { VM_OBJECT_WLOCK(vp->v_object); vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC); VM_OBJECT_WUNLOCK(vp->v_object); } error = VOP_FSYNC(vp, MNT_WAIT, td); } else { /* * Locate and synchronously write any buffers that fall * into the requested range. Note: we are assuming that * f_iosize is a power of 2. */ int iosize = vp->v_mount->mnt_stat.f_iosize; int iomask = iosize - 1; struct bufobj *bo; daddr_t lblkno; /* * Align to iosize boundary, super-align to page boundary. */ if (off & iomask) { cnt += off & iomask; off &= ~(u_quad_t)iomask; } if (off & PAGE_MASK) { cnt += off & PAGE_MASK; off &= ~(u_quad_t)PAGE_MASK; } lblkno = off / iosize; if (vp->v_object && vm_object_mightbedirty(vp->v_object)) { VM_OBJECT_WLOCK(vp->v_object); vm_object_page_clean(vp->v_object, off, off + cnt, OBJPC_SYNC); VM_OBJECT_WUNLOCK(vp->v_object); } bo = &vp->v_bufobj; BO_LOCK(bo); while (cnt > 0) { struct buf *bp; /* * If we have a buffer and it is marked B_DELWRI we * have to lock and write it. Otherwise the prior * write is assumed to have already been committed. * * gbincore() can return invalid buffers now so we * have to check that bit as well (though B_DELWRI * should not be set if B_INVAL is set there could be * a race here since we haven't locked the buffer). */ if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) { if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) { BO_LOCK(bo); continue; /* retry */ } if ((bp->b_flags & (B_DELWRI|B_INVAL)) == B_DELWRI) { bremfree(bp); bp->b_flags &= ~B_ASYNC; bwrite(bp); ++nfs_commit_miss; } else BUF_UNLOCK(bp); BO_LOCK(bo); } ++nfs_commit_blks; if (cnt < iosize) break; cnt -= iosize; ++lblkno; } BO_UNLOCK(bo); } NFSEXITCODE(error); return (error); } /* * Statfs vnode op. */ int nfsvno_statfs(struct vnode *vp, struct statfs *sf) { struct statfs *tsf; int error; tsf = NULL; if (nfsrv_devidcnt > 0) { /* For a pNFS service, get the DS numbers. */ tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK | M_ZERO); error = nfsrv_pnfsstatfs(tsf, vp->v_mount); if (error != 0) { free(tsf, M_TEMP); tsf = NULL; } } error = VFS_STATFS(vp->v_mount, sf); if (error == 0) { if (tsf != NULL) { sf->f_blocks = tsf->f_blocks; sf->f_bavail = tsf->f_bavail; sf->f_bfree = tsf->f_bfree; sf->f_bsize = tsf->f_bsize; } /* * Since NFS handles these values as unsigned on the * wire, there is no way to represent negative values, * so set them to 0. Without this, they will appear * to be very large positive values for clients like * Solaris10. */ if (sf->f_bavail < 0) sf->f_bavail = 0; if (sf->f_ffree < 0) sf->f_ffree = 0; } free(tsf, M_TEMP); NFSEXITCODE(error); return (error); } /* * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but * must handle nfsrv_opencheck() calls after any other access checks. */ void nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp, nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp, int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create, NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, struct nfsexstuff *exp, struct vnode **vpp) { struct vnode *vp = NULL; u_quad_t tempsize; struct nfsexstuff nes; struct thread *p = curthread; if (ndp->ni_vp == NULL) nd->nd_repstat = nfsrv_opencheck(clientid, stateidp, stp, NULL, nd, p, nd->nd_repstat); if (!nd->nd_repstat) { if (ndp->ni_vp == NULL) { vrele(ndp->ni_startdir); nd->nd_repstat = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); /* For a pNFS server, create the data file on a DS. */ if (nd->nd_repstat == 0) { /* * Create a data file on a DS for a pNFS server. * This function just returns if not * running a pNFS DS or the creation fails. */ nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr, cred, p); } VOP_VPUT_PAIR(ndp->ni_dvp, nd->nd_repstat == 0 ? &ndp->ni_vp : NULL, false); nfsvno_relpathbuf(ndp); if (!nd->nd_repstat) { if (*exclusive_flagp) { *exclusive_flagp = 0; NFSVNO_ATTRINIT(nvap); nvap->na_atime.tv_sec = cverf[0]; nvap->na_atime.tv_nsec = cverf[1]; nd->nd_repstat = VOP_SETATTR(ndp->ni_vp, &nvap->na_vattr, cred); if (nd->nd_repstat != 0) { vput(ndp->ni_vp); ndp->ni_vp = NULL; nd->nd_repstat = NFSERR_NOTSUPP; } else NFSSETBIT_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS); } else { nfsrv_fixattr(nd, ndp->ni_vp, nvap, aclp, p, attrbitp, exp); } } vp = ndp->ni_vp; } else { if (ndp->ni_startdir) vrele(ndp->ni_startdir); nfsvno_relpathbuf(ndp); vp = ndp->ni_vp; if (create == NFSV4OPEN_CREATE) { if (ndp->ni_dvp == vp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); } if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) { if (ndp->ni_cnd.cn_flags & RDONLY) NFSVNO_SETEXRDONLY(&nes); else NFSVNO_EXINIT(&nes); nd->nd_repstat = nfsvno_accchk(vp, VWRITE, cred, &nes, p, NFSACCCHK_NOOVERRIDE, NFSACCCHK_VPISLOCKED, NULL); nd->nd_repstat = nfsrv_opencheck(clientid, stateidp, stp, vp, nd, p, nd->nd_repstat); if (!nd->nd_repstat) { tempsize = nvap->na_size; NFSVNO_ATTRINIT(nvap); nvap->na_size = tempsize; nd->nd_repstat = VOP_SETATTR(vp, &nvap->na_vattr, cred); } } else if (vp->v_type == VREG) { nd->nd_repstat = nfsrv_opencheck(clientid, stateidp, stp, vp, nd, p, nd->nd_repstat); } } } else { if (ndp->ni_cnd.cn_flags & HASBUF) nfsvno_relpathbuf(ndp); if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) { vrele(ndp->ni_startdir); if (ndp->ni_dvp == ndp->ni_vp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); if (ndp->ni_vp) vput(ndp->ni_vp); } } *vpp = vp; NFSEXITCODE2(0, nd); } /* * Updates the file rev and sets the mtime and ctime * to the current clock time, returning the va_filerev and va_Xtime * values. * Return ESTALE to indicate the vnode is VIRF_DOOMED. */ int nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap, struct nfsrv_descript *nd, struct thread *p) { struct vattr va; VATTR_NULL(&va); vfs_timestamp(&va.va_mtime); if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) { NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); if (VN_IS_DOOMED(vp)) return (ESTALE); } (void) VOP_SETATTR(vp, &va, nd->nd_cred); (void) nfsvno_getattr(vp, nvap, nd, p, 1, NULL); return (0); } /* * Glue routine to nfsv4_fillattr(). */ int nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp, struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp, struct ucred *cred, struct thread *p, int isdgram, int reterr, int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno) { struct statfs *sf; int error; sf = NULL; if (nfsrv_devidcnt > 0 && (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEAVAIL) || NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEFREE) || NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACETOTAL))) { sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK | M_ZERO); error = nfsrv_pnfsstatfs(sf, mp); if (error != 0) { free(sf, M_TEMP); sf = NULL; } } error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror, attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root, mounted_on_fileno, sf); free(sf, M_TEMP); NFSEXITCODE2(0, nd); return (error); } /* Since the Readdir vnode ops vary, put the entire functions in here. */ /* * nfs readdir service * - mallocs what it thinks is enough to read * count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR * - calls VOP_READDIR() * - loops around building the reply * if the output generated exceeds count break out of loop * The NFSM_CLGET macro is used here so that the reply will be packed * tightly in mbuf clusters. * - it trims out records with d_fileno == 0 * this doesn't matter for Unix clients, but they might confuse clients * for other os'. * - it trims out records with d_type == DT_WHT * these cannot be seen through NFS (unless we extend the protocol) * The alternate call nfsrvd_readdirplus() does lookups as well. * PS: The NFS protocol spec. does not clarify what the "count" byte * argument is a count of.. just name strings and file id's or the * entire reply rpc or ... * I tried just file name and id sizes and it confused the Sun client, * so I am using the full rpc size now. The "paranoia.." comment refers * to including the status longwords that are not a part of the dir. * "entry" structures, but are in the rpc. */ int nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram, struct vnode *vp, struct nfsexstuff *exp) { struct dirent *dp; u_int32_t *tl; int dirlen; char *cpos, *cend, *rbuf; struct nfsvattr at; int nlen, error = 0, getret = 1; int siz, cnt, fullsiz, eofflag, ncookies; u_int64_t off, toff, verf __unused; u_long *cookies = NULL, *cookiep; struct uio io; struct iovec iv; int is_ufs; struct thread *p = curthread; if (nd->nd_repstat) { nfsrv_postopattr(nd, getret, &at); goto out; } if (nd->nd_flag & ND_NFSV2) { NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); off = fxdr_unsigned(u_quad_t, *tl++); } else { NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED); off = fxdr_hyper(tl); tl += 2; verf = fxdr_hyper(tl); tl += 2; } toff = off; cnt = fxdr_unsigned(int, *tl); if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0) cnt = NFS_SRVMAXDATA(nd); siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1)); fullsiz = siz; if (nd->nd_flag & ND_NFSV3) { nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); #if 0 /* * va_filerev is not sufficient as a cookie verifier, * since it is not supposed to change when entries are * removed/added unless that offset cookies returned to * the client are no longer valid. */ if (!nd->nd_repstat && toff && verf != at.na_filerev) nd->nd_repstat = NFSERR_BAD_COOKIE; #endif } if (!nd->nd_repstat && vp->v_type != VDIR) nd->nd_repstat = NFSERR_NOTDIR; if (nd->nd_repstat == 0 && cnt == 0) { if (nd->nd_flag & ND_NFSV2) /* NFSv2 does not have NFSERR_TOOSMALL */ nd->nd_repstat = EPERM; else nd->nd_repstat = NFSERR_TOOSMALL; } if (!nd->nd_repstat) nd->nd_repstat = nfsvno_accchk(vp, VEXEC, nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, NFSACCCHK_VPISLOCKED, NULL); if (nd->nd_repstat) { vput(vp); if (nd->nd_flag & ND_NFSV3) nfsrv_postopattr(nd, getret, &at); goto out; } is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0; rbuf = malloc(siz, M_TEMP, M_WAITOK); again: eofflag = 0; if (cookies) { free(cookies, M_TEMP); cookies = NULL; } iv.iov_base = rbuf; iv.iov_len = siz; io.uio_iov = &iv; io.uio_iovcnt = 1; io.uio_offset = (off_t)off; io.uio_resid = siz; io.uio_segflg = UIO_SYSSPACE; io.uio_rw = UIO_READ; io.uio_td = NULL; nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies, &cookies); off = (u_int64_t)io.uio_offset; if (io.uio_resid) siz -= io.uio_resid; if (!cookies && !nd->nd_repstat) nd->nd_repstat = NFSERR_PERM; if (nd->nd_flag & ND_NFSV3) { getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); if (!nd->nd_repstat) nd->nd_repstat = getret; } /* * Handles the failed cases. nd->nd_repstat == 0 past here. */ if (nd->nd_repstat) { vput(vp); free(rbuf, M_TEMP); if (cookies) free(cookies, M_TEMP); if (nd->nd_flag & ND_NFSV3) nfsrv_postopattr(nd, getret, &at); goto out; } /* * If nothing read, return eof * rpc reply */ if (siz == 0) { vput(vp); if (nd->nd_flag & ND_NFSV2) { NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); } else { nfsrv_postopattr(nd, getret, &at); NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); txdr_hyper(at.na_filerev, tl); tl += 2; } *tl++ = newnfs_false; *tl = newnfs_true; free(rbuf, M_TEMP); free(cookies, M_TEMP); goto out; } /* * Check for degenerate cases of nothing useful read. * If so go try again */ cpos = rbuf; cend = rbuf + siz; dp = (struct dirent *)cpos; cookiep = cookies; /* * For some reason FreeBSD's ufs_readdir() chooses to back the * directory offset up to a block boundary, so it is necessary to * skip over the records that precede the requested offset. This * requires the assumption that file offset cookies monotonically * increase. */ while (cpos < cend && ncookies > 0 && (dp->d_fileno == 0 || dp->d_type == DT_WHT || (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) { cpos += dp->d_reclen; dp = (struct dirent *)cpos; cookiep++; ncookies--; } if (cpos >= cend || ncookies == 0) { siz = fullsiz; toff = off; goto again; } vput(vp); /* * If cnt > MCLBYTES and the reply will not be saved, use * ext_pgs mbufs for TLS. * For NFSv4.0, we do not know for sure if the reply will * be saved, so do not use ext_pgs mbufs for NFSv4.0. */ if (cnt > MCLBYTES && siz > MCLBYTES && (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS && (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4) nd->nd_flag |= ND_EXTPG; /* * dirlen is the size of the reply, including all XDR and must * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate * if the XDR should be included in "count", but to be safe, we do. * (Include the two booleans at the end of the reply in dirlen now.) */ if (nd->nd_flag & ND_NFSV3) { nfsrv_postopattr(nd, getret, &at); NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); txdr_hyper(at.na_filerev, tl); dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED; } else { dirlen = 2 * NFSX_UNSIGNED; } /* Loop through the records and build reply */ while (cpos < cend && ncookies > 0) { nlen = dp->d_namlen; if (dp->d_fileno != 0 && dp->d_type != DT_WHT && nlen <= NFS_MAXNAMLEN) { if (nd->nd_flag & ND_NFSV3) dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen)); else dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen)); if (dirlen > cnt) { eofflag = 0; break; } /* * Build the directory record xdr from * the dirent entry. */ if (nd->nd_flag & ND_NFSV3) { NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); *tl++ = newnfs_true; *tl++ = 0; } else { NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = newnfs_true; } *tl = txdr_unsigned(dp->d_fileno); (void) nfsm_strtom(nd, dp->d_name, nlen); if (nd->nd_flag & ND_NFSV3) { NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = 0; } else NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(*cookiep); } cpos += dp->d_reclen; dp = (struct dirent *)cpos; cookiep++; ncookies--; } if (cpos < cend) eofflag = 0; NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = newnfs_false; if (eofflag) *tl = newnfs_true; else *tl = newnfs_false; free(rbuf, M_TEMP); free(cookies, M_TEMP); out: NFSEXITCODE2(0, nd); return (0); nfsmout: vput(vp); NFSEXITCODE2(error, nd); return (error); } /* * Readdirplus for V3 and Readdir for V4. */ int nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram, struct vnode *vp, struct nfsexstuff *exp) { struct dirent *dp; u_int32_t *tl; int dirlen; char *cpos, *cend, *rbuf; struct vnode *nvp; fhandle_t nfh; struct nfsvattr nva, at, *nvap = &nva; struct mbuf *mb0, *mb1; struct nfsreferral *refp; int nlen, r, error = 0, getret = 1, usevget = 1; int siz, cnt, fullsiz, eofflag, ncookies, entrycnt; caddr_t bpos0, bpos1; u_int64_t off, toff, verf; u_long *cookies = NULL, *cookiep; nfsattrbit_t attrbits, rderrbits, savbits; struct uio io; struct iovec iv; struct componentname cn; int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls; struct mount *mp, *new_mp; uint64_t mounted_on_fileno; struct thread *p = curthread; int bextpg0, bextpg1, bextpgsiz0, bextpgsiz1; if (nd->nd_repstat) { nfsrv_postopattr(nd, getret, &at); goto out; } NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED); off = fxdr_hyper(tl); toff = off; tl += 2; verf = fxdr_hyper(tl); tl += 2; siz = fxdr_unsigned(int, *tl++); cnt = fxdr_unsigned(int, *tl); /* * Use the server's maximum data transfer size as the upper bound * on reply datalen. */ if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0) cnt = NFS_SRVMAXDATA(nd); /* * siz is a "hint" of how much directory information (name, fileid, * cookie) should be in the reply. At least one client "hints" 0, * so I set it to cnt for that case. I also round it up to the * next multiple of DIRBLKSIZ. * Since the size of a Readdirplus directory entry reply will always * be greater than a directory entry returned by VOP_READDIR(), it * does not make sense to read more than NFS_SRVMAXDATA() via * VOP_READDIR(). */ if (siz <= 0) siz = cnt; else if (siz > NFS_SRVMAXDATA(nd)) siz = NFS_SRVMAXDATA(nd); siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1)); if (nd->nd_flag & ND_NFSV4) { error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); if (error) goto nfsmout; NFSSET_ATTRBIT(&savbits, &attrbits); NFSCLRNOTFILLABLE_ATTRBIT(&attrbits, nd); NFSZERO_ATTRBIT(&rderrbits); NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR); } else { NFSZERO_ATTRBIT(&attrbits); } fullsiz = siz; nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); #if 0 if (!nd->nd_repstat) { if (off && verf != at.na_filerev) { /* * va_filerev is not sufficient as a cookie verifier, * since it is not supposed to change when entries are * removed/added unless that offset cookies returned to * the client are no longer valid. */ if (nd->nd_flag & ND_NFSV4) { nd->nd_repstat = NFSERR_NOTSAME; } else { nd->nd_repstat = NFSERR_BAD_COOKIE; } } } #endif if (!nd->nd_repstat && vp->v_type != VDIR) nd->nd_repstat = NFSERR_NOTDIR; if (!nd->nd_repstat && cnt == 0) nd->nd_repstat = NFSERR_TOOSMALL; if (!nd->nd_repstat) nd->nd_repstat = nfsvno_accchk(vp, VEXEC, nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, NFSACCCHK_VPISLOCKED, NULL); if (nd->nd_repstat) { vput(vp); if (nd->nd_flag & ND_NFSV3) nfsrv_postopattr(nd, getret, &at); goto out; } is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0; is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0; rbuf = malloc(siz, M_TEMP, M_WAITOK); again: eofflag = 0; if (cookies) { free(cookies, M_TEMP); cookies = NULL; } iv.iov_base = rbuf; iv.iov_len = siz; io.uio_iov = &iv; io.uio_iovcnt = 1; io.uio_offset = (off_t)off; io.uio_resid = siz; io.uio_segflg = UIO_SYSSPACE; io.uio_rw = UIO_READ; io.uio_td = NULL; nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies, &cookies); off = (u_int64_t)io.uio_offset; if (io.uio_resid) siz -= io.uio_resid; getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); if (!cookies && !nd->nd_repstat) nd->nd_repstat = NFSERR_PERM; if (!nd->nd_repstat) nd->nd_repstat = getret; if (nd->nd_repstat) { vput(vp); if (cookies) free(cookies, M_TEMP); free(rbuf, M_TEMP); if (nd->nd_flag & ND_NFSV3) nfsrv_postopattr(nd, getret, &at); goto out; } /* * If nothing read, return eof * rpc reply */ if (siz == 0) { vput(vp); if (nd->nd_flag & ND_NFSV3) nfsrv_postopattr(nd, getret, &at); NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); txdr_hyper(at.na_filerev, tl); tl += 2; *tl++ = newnfs_false; *tl = newnfs_true; free(cookies, M_TEMP); free(rbuf, M_TEMP); goto out; } /* * Check for degenerate cases of nothing useful read. * If so go try again */ cpos = rbuf; cend = rbuf + siz; dp = (struct dirent *)cpos; cookiep = cookies; /* * For some reason FreeBSD's ufs_readdir() chooses to back the * directory offset up to a block boundary, so it is necessary to * skip over the records that precede the requested offset. This * requires the assumption that file offset cookies monotonically * increase. */ while (cpos < cend && ncookies > 0 && (dp->d_fileno == 0 || dp->d_type == DT_WHT || (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) || ((nd->nd_flag & ND_NFSV4) && ((dp->d_namlen == 1 && dp->d_name[0] == '.') || (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) { cpos += dp->d_reclen; dp = (struct dirent *)cpos; cookiep++; ncookies--; } if (cpos >= cend || ncookies == 0) { siz = fullsiz; toff = off; goto again; } /* * Busy the file system so that the mount point won't go away * and, as such, VFS_VGET() can be used safely. */ mp = vp->v_mount; vfs_ref(mp); NFSVOPUNLOCK(vp); nd->nd_repstat = vfs_busy(mp, 0); vfs_rel(mp); if (nd->nd_repstat != 0) { vrele(vp); free(cookies, M_TEMP); free(rbuf, M_TEMP); if (nd->nd_flag & ND_NFSV3) nfsrv_postopattr(nd, getret, &at); goto out; } /* * Check to see if entries in this directory can be safely acquired * via VFS_VGET() or if a switch to VOP_LOOKUP() is required. * ZFS snapshot directories need VOP_LOOKUP(), so that any * automount of the snapshot directory that is required will * be done. * This needs to be done here for NFSv4, since NFSv4 never does * a VFS_VGET() for "." or "..". */ if (is_zfs == 1) { r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp); if (r == EOPNOTSUPP) { usevget = 0; cn.cn_nameiop = LOOKUP; cn.cn_lkflags = LK_SHARED | LK_RETRY; cn.cn_cred = nd->nd_cred; - cn.cn_thread = p; } else if (r == 0) vput(nvp); } /* * If the reply is likely to exceed MCLBYTES and the reply will * not be saved, use ext_pgs mbufs for TLS. * It is difficult to predict how large each entry will be and * how many entries have been read, so just assume the directory * entries grow by a factor of 4 when attributes are included. * For NFSv4.0, we do not know for sure if the reply will * be saved, so do not use ext_pgs mbufs for NFSv4.0. */ if (cnt > MCLBYTES && siz > MCLBYTES / 4 && (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS && (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4) nd->nd_flag |= ND_EXTPG; /* * Save this position, in case there is an error before one entry * is created. */ mb0 = nd->nd_mb; bpos0 = nd->nd_bpos; bextpg0 = nd->nd_bextpg; bextpgsiz0 = nd->nd_bextpgsiz; /* * Fill in the first part of the reply. * dirlen is the reply length in bytes and cannot exceed cnt. * (Include the two booleans at the end of the reply in dirlen now, * so we recognize when we have exceeded cnt.) */ if (nd->nd_flag & ND_NFSV3) { dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED; nfsrv_postopattr(nd, getret, &at); } else { dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED; } NFSM_BUILD(tl, u_int32_t *, NFSX_VERF); txdr_hyper(at.na_filerev, tl); /* * Save this position, in case there is an empty reply needed. */ mb1 = nd->nd_mb; bpos1 = nd->nd_bpos; bextpg1 = nd->nd_bextpg; bextpgsiz1 = nd->nd_bextpgsiz; /* Loop through the records and build reply */ entrycnt = 0; while (cpos < cend && ncookies > 0 && dirlen < cnt) { nlen = dp->d_namlen; if (dp->d_fileno != 0 && dp->d_type != DT_WHT && nlen <= NFS_MAXNAMLEN && ((nd->nd_flag & ND_NFSV3) || nlen > 2 || (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.')) || (nlen == 1 && dp->d_name[0] != '.'))) { /* * Save the current position in the reply, in case * this entry exceeds cnt. */ mb1 = nd->nd_mb; bpos1 = nd->nd_bpos; bextpg1 = nd->nd_bextpg; bextpgsiz1 = nd->nd_bextpgsiz; /* * For readdir_and_lookup get the vnode using * the file number. */ nvp = NULL; refp = NULL; r = 0; at_root = 0; needs_unbusy = 0; new_mp = mp; mounted_on_fileno = (uint64_t)dp->d_fileno; if ((nd->nd_flag & ND_NFSV3) || NFSNONZERO_ATTRBIT(&savbits)) { if (nd->nd_flag & ND_NFSV4) refp = nfsv4root_getreferral(NULL, vp, dp->d_fileno); if (refp == NULL) { if (usevget) r = VFS_VGET(mp, dp->d_fileno, LK_SHARED, &nvp); else r = EOPNOTSUPP; if (r == EOPNOTSUPP) { if (usevget) { usevget = 0; cn.cn_nameiop = LOOKUP; cn.cn_lkflags = LK_SHARED | LK_RETRY; cn.cn_cred = nd->nd_cred; - cn.cn_thread = p; } cn.cn_nameptr = dp->d_name; cn.cn_namelen = nlen; cn.cn_flags = ISLASTCN | NOFOLLOW | LOCKLEAF; if (nlen == 2 && dp->d_name[0] == '.' && dp->d_name[1] == '.') cn.cn_flags |= ISDOTDOT; if (NFSVOPLOCK(vp, LK_SHARED) != 0) { nd->nd_repstat = EPERM; break; } if ((vp->v_vflag & VV_ROOT) != 0 && (cn.cn_flags & ISDOTDOT) != 0) { vref(vp); nvp = vp; r = 0; } else { r = VOP_LOOKUP(vp, &nvp, &cn); if (vp != nvp) NFSVOPUNLOCK(vp); } } /* * For NFSv4, check to see if nvp is * a mount point and get the mount * point vnode, as required. */ if (r == 0 && nfsrv_enable_crossmntpt != 0 && (nd->nd_flag & ND_NFSV4) != 0 && nvp->v_type == VDIR && nvp->v_mountedhere != NULL) { new_mp = nvp->v_mountedhere; r = vfs_busy(new_mp, 0); vput(nvp); nvp = NULL; if (r == 0) { r = VFS_ROOT(new_mp, LK_SHARED, &nvp); needs_unbusy = 1; if (r == 0) at_root = 1; } } } /* * If we failed to look up the entry, then it * has become invalid, most likely removed. */ if (r != 0) { if (needs_unbusy) vfs_unbusy(new_mp); goto invalid; } KASSERT(refp != NULL || nvp != NULL, ("%s: undetected lookup error", __func__)); if (refp == NULL && ((nd->nd_flag & ND_NFSV3) || NFSNONZERO_ATTRBIT(&attrbits))) { r = nfsvno_getfh(nvp, &nfh, p); if (!r) r = nfsvno_getattr(nvp, nvap, nd, p, 1, &attrbits); if (r == 0 && is_zfs == 1 && nfsrv_enable_crossmntpt != 0 && (nd->nd_flag & ND_NFSV4) != 0 && nvp->v_type == VDIR && vp->v_mount != nvp->v_mount) { /* * For a ZFS snapshot, there is a * pseudo mount that does not set * v_mountedhere, so it needs to * be detected via a different * mount structure. */ at_root = 1; if (new_mp == mp) new_mp = nvp->v_mount; } } /* * If we failed to get attributes of the entry, * then just skip it for NFSv3 (the traditional * behavior in the old NFS server). * For NFSv4 the behavior is controlled by * RDATTRERROR: we either ignore the error or * fail the request. * Note that RDATTRERROR is never set for NFSv3. */ if (r != 0) { if (!NFSISSET_ATTRBIT(&attrbits, NFSATTRBIT_RDATTRERROR)) { vput(nvp); if (needs_unbusy != 0) vfs_unbusy(new_mp); if ((nd->nd_flag & ND_NFSV3)) goto invalid; nd->nd_repstat = r; break; } } } /* * Build the directory record xdr */ if (nd->nd_flag & ND_NFSV3) { NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); *tl++ = newnfs_true; *tl++ = 0; *tl = txdr_unsigned(dp->d_fileno); dirlen += nfsm_strtom(nd, dp->d_name, nlen); NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = 0; *tl = txdr_unsigned(*cookiep); nfsrv_postopattr(nd, 0, nvap); dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1); dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR); if (nvp != NULL) vput(nvp); } else { NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); *tl++ = newnfs_true; *tl++ = 0; *tl = txdr_unsigned(*cookiep); dirlen += nfsm_strtom(nd, dp->d_name, nlen); if (nvp != NULL) { supports_nfsv4acls = nfs_supportsnfsv4acls(nvp); NFSVOPUNLOCK(nvp); } else supports_nfsv4acls = 0; if (refp != NULL) { dirlen += nfsrv_putreferralattr(nd, &savbits, refp, 0, &nd->nd_repstat); if (nd->nd_repstat) { if (nvp != NULL) vrele(nvp); if (needs_unbusy != 0) vfs_unbusy(new_mp); break; } } else if (r) { dirlen += nfsvno_fillattr(nd, new_mp, nvp, nvap, &nfh, r, &rderrbits, nd->nd_cred, p, isdgram, 0, supports_nfsv4acls, at_root, mounted_on_fileno); } else { dirlen += nfsvno_fillattr(nd, new_mp, nvp, nvap, &nfh, r, &attrbits, nd->nd_cred, p, isdgram, 0, supports_nfsv4acls, at_root, mounted_on_fileno); } if (nvp != NULL) vrele(nvp); dirlen += (3 * NFSX_UNSIGNED); } if (needs_unbusy != 0) vfs_unbusy(new_mp); if (dirlen <= cnt) entrycnt++; } invalid: cpos += dp->d_reclen; dp = (struct dirent *)cpos; cookiep++; ncookies--; } vrele(vp); vfs_unbusy(mp); /* * If dirlen > cnt, we must strip off the last entry. If that * results in an empty reply, report NFSERR_TOOSMALL. */ if (dirlen > cnt || nd->nd_repstat) { if (!nd->nd_repstat && entrycnt == 0) nd->nd_repstat = NFSERR_TOOSMALL; if (nd->nd_repstat) { nfsm_trimtrailing(nd, mb0, bpos0, bextpg0, bextpgsiz0); if (nd->nd_flag & ND_NFSV3) nfsrv_postopattr(nd, getret, &at); } else nfsm_trimtrailing(nd, mb1, bpos1, bextpg1, bextpgsiz1); eofflag = 0; } else if (cpos < cend) eofflag = 0; if (!nd->nd_repstat) { NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = newnfs_false; if (eofflag) *tl = newnfs_true; else *tl = newnfs_false; } free(cookies, M_TEMP); free(rbuf, M_TEMP); out: NFSEXITCODE2(0, nd); return (0); nfsmout: vput(vp); NFSEXITCODE2(error, nd); return (error); } /* * Get the settable attributes out of the mbuf list. * (Return 0 or EBADRPC) */ int nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p) { u_int32_t *tl; struct nfsv2_sattr *sp; int error = 0, toclient = 0; switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) { case ND_NFSV2: NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR); /* * Some old clients didn't fill in the high order 16bits. * --> check the low order 2 bytes for 0xffff */ if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff) nvap->na_mode = nfstov_mode(sp->sa_mode); if (sp->sa_uid != newnfs_xdrneg1) nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid); if (sp->sa_gid != newnfs_xdrneg1) nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid); if (sp->sa_size != newnfs_xdrneg1) nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size); if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) { #ifdef notyet fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime); #else nvap->na_atime.tv_sec = fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec); nvap->na_atime.tv_nsec = 0; #endif } if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1) fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime); break; case ND_NFSV3: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (*tl == newnfs_true) { NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); nvap->na_mode = nfstov_mode(*tl); } NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (*tl == newnfs_true) { NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); nvap->na_uid = fxdr_unsigned(uid_t, *tl); } NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (*tl == newnfs_true) { NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); nvap->na_gid = fxdr_unsigned(gid_t, *tl); } NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (*tl == newnfs_true) { NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); nvap->na_size = fxdr_hyper(tl); } NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); switch (fxdr_unsigned(int, *tl)) { case NFSV3SATTRTIME_TOCLIENT: NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); fxdr_nfsv3time(tl, &nvap->na_atime); toclient = 1; break; case NFSV3SATTRTIME_TOSERVER: vfs_timestamp(&nvap->na_atime); nvap->na_vaflags |= VA_UTIMES_NULL; break; } NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); switch (fxdr_unsigned(int, *tl)) { case NFSV3SATTRTIME_TOCLIENT: NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); fxdr_nfsv3time(tl, &nvap->na_mtime); nvap->na_vaflags &= ~VA_UTIMES_NULL; break; case NFSV3SATTRTIME_TOSERVER: vfs_timestamp(&nvap->na_mtime); if (!toclient) nvap->na_vaflags |= VA_UTIMES_NULL; break; } break; case ND_NFSV4: error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p); } nfsmout: NFSEXITCODE2(error, nd); return (error); } /* * Handle the setable attributes for V4. * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise. */ int nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p) { u_int32_t *tl; int attrsum = 0; int i, j; int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0; int moderet, toclient = 0; u_char *cp, namestr[NFSV4_SMALLSTR + 1]; uid_t uid; gid_t gid; u_short mode, mask; /* Same type as va_mode. */ struct vattr va; error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup); if (error) goto nfsmout; NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); attrsize = fxdr_unsigned(int, *tl); /* * Loop around getting the setable attributes. If an unsupported * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return. */ if (retnotsup) { nd->nd_repstat = NFSERR_ATTRNOTSUPP; bitpos = NFSATTRBIT_MAX; } else { bitpos = 0; } moderet = 0; for (; bitpos < NFSATTRBIT_MAX; bitpos++) { if (attrsum > attrsize) { error = NFSERR_BADXDR; goto nfsmout; } if (NFSISSET_ATTRBIT(attrbitp, bitpos)) switch (bitpos) { case NFSATTRBIT_SIZE: NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); if (vp != NULL && vp->v_type != VREG) { error = (vp->v_type == VDIR) ? NFSERR_ISDIR : NFSERR_INVAL; goto nfsmout; } nvap->na_size = fxdr_hyper(tl); attrsum += NFSX_HYPER; break; case NFSATTRBIT_ACL: error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize, p); if (error) goto nfsmout; if (aceerr && !nd->nd_repstat) nd->nd_repstat = aceerr; attrsum += aclsize; break; case NFSATTRBIT_ARCHIVE: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (!nd->nd_repstat) nd->nd_repstat = NFSERR_ATTRNOTSUPP; attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_HIDDEN: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (!nd->nd_repstat) nd->nd_repstat = NFSERR_ATTRNOTSUPP; attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_MIMETYPE: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); i = fxdr_unsigned(int, *tl); error = nfsm_advance(nd, NFSM_RNDUP(i), -1); if (error) goto nfsmout; if (!nd->nd_repstat) nd->nd_repstat = NFSERR_ATTRNOTSUPP; attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i)); break; case NFSATTRBIT_MODE: moderet = NFSERR_INVAL; /* Can't do MODESETMASKED. */ NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); nvap->na_mode = nfstov_mode(*tl); attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_OWNER: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); j = fxdr_unsigned(int, *tl); if (j < 0) { error = NFSERR_BADXDR; goto nfsmout; } if (j > NFSV4_SMALLSTR) cp = malloc(j + 1, M_NFSSTRING, M_WAITOK); else cp = namestr; error = nfsrv_mtostr(nd, cp, j); if (error) { if (j > NFSV4_SMALLSTR) free(cp, M_NFSSTRING); goto nfsmout; } if (!nd->nd_repstat) { nd->nd_repstat = nfsv4_strtouid(nd, cp, j, &uid); if (!nd->nd_repstat) nvap->na_uid = uid; } if (j > NFSV4_SMALLSTR) free(cp, M_NFSSTRING); attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j)); break; case NFSATTRBIT_OWNERGROUP: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); j = fxdr_unsigned(int, *tl); if (j < 0) { error = NFSERR_BADXDR; goto nfsmout; } if (j > NFSV4_SMALLSTR) cp = malloc(j + 1, M_NFSSTRING, M_WAITOK); else cp = namestr; error = nfsrv_mtostr(nd, cp, j); if (error) { if (j > NFSV4_SMALLSTR) free(cp, M_NFSSTRING); goto nfsmout; } if (!nd->nd_repstat) { nd->nd_repstat = nfsv4_strtogid(nd, cp, j, &gid); if (!nd->nd_repstat) nvap->na_gid = gid; } if (j > NFSV4_SMALLSTR) free(cp, M_NFSSTRING); attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j)); break; case NFSATTRBIT_SYSTEM: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (!nd->nd_repstat) nd->nd_repstat = NFSERR_ATTRNOTSUPP; attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_TIMEACCESSSET: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); attrsum += NFSX_UNSIGNED; if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) { NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); fxdr_nfsv4time(tl, &nvap->na_atime); toclient = 1; attrsum += NFSX_V4TIME; } else { vfs_timestamp(&nvap->na_atime); nvap->na_vaflags |= VA_UTIMES_NULL; } break; case NFSATTRBIT_TIMEBACKUP: NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); if (!nd->nd_repstat) nd->nd_repstat = NFSERR_ATTRNOTSUPP; attrsum += NFSX_V4TIME; break; case NFSATTRBIT_TIMECREATE: NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); fxdr_nfsv4time(tl, &nvap->na_btime); attrsum += NFSX_V4TIME; break; case NFSATTRBIT_TIMEMODIFYSET: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); attrsum += NFSX_UNSIGNED; if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) { NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); fxdr_nfsv4time(tl, &nvap->na_mtime); nvap->na_vaflags &= ~VA_UTIMES_NULL; attrsum += NFSX_V4TIME; } else { vfs_timestamp(&nvap->na_mtime); if (!toclient) nvap->na_vaflags |= VA_UTIMES_NULL; } break; case NFSATTRBIT_MODESETMASKED: NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); mode = fxdr_unsigned(u_short, *tl++); mask = fxdr_unsigned(u_short, *tl); /* * vp == NULL implies an Open/Create operation. * This attribute can only be used for Setattr and * only for NFSv4.1 or higher. * If moderet != 0, a mode attribute has also been * specified and this attribute cannot be done in the * same Setattr operation. */ if ((nd->nd_flag & ND_NFSV41) == 0) nd->nd_repstat = NFSERR_ATTRNOTSUPP; else if ((mode & ~07777) != 0 || (mask & ~07777) != 0 || vp == NULL) nd->nd_repstat = NFSERR_INVAL; else if (moderet == 0) moderet = VOP_GETATTR(vp, &va, nd->nd_cred); if (moderet == 0) nvap->na_mode = (mode & mask) | (va.va_mode & ~mask); else nd->nd_repstat = moderet; attrsum += 2 * NFSX_UNSIGNED; break; default: nd->nd_repstat = NFSERR_ATTRNOTSUPP; /* * set bitpos so we drop out of the loop. */ bitpos = NFSATTRBIT_MAX; break; } } /* * some clients pad the attrlist, so we need to skip over the * padding. */ if (attrsum > attrsize) { error = NFSERR_BADXDR; } else { attrsize = NFSM_RNDUP(attrsize); if (attrsum < attrsize) error = nfsm_advance(nd, attrsize - attrsum, -1); } nfsmout: NFSEXITCODE2(error, nd); return (error); } /* * Check/setup export credentials. */ int nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp, struct ucred *credanon, bool testsec) { int error; /* * Check/setup credentials. */ if (nd->nd_flag & ND_GSS) exp->nes_exflag &= ~MNT_EXPORTANON; /* * Check to see if the operation is allowed for this security flavor. */ error = 0; if (testsec) { error = nfsvno_testexp(nd, exp); if (error != 0) goto out; } /* * Check to see if the file system is exported V4 only. */ if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) { error = NFSERR_PROGNOTV4; goto out; } /* * Now, map the user credentials. * (Note that ND_AUTHNONE will only be set for an NFSv3 * Fsinfo RPC. If set for anything else, this code might need * to change.) */ if (NFSVNO_EXPORTED(exp)) { if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) || NFSVNO_EXPORTANON(exp) || (nd->nd_flag & ND_AUTHNONE) != 0) { nd->nd_cred->cr_uid = credanon->cr_uid; nd->nd_cred->cr_gid = credanon->cr_gid; crsetgroups(nd->nd_cred, credanon->cr_ngroups, credanon->cr_groups); } else if ((nd->nd_flag & ND_GSS) == 0) { /* * If using AUTH_SYS, call nfsrv_getgrpscred() to see * if there is a replacement credential with a group * list set up by "nfsuserd -manage-gids". * If there is no replacement, nfsrv_getgrpscred() * simply returns its argument. */ nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred); } } out: NFSEXITCODE2(error, nd); return (error); } /* * Check exports. */ int nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp, struct ucred **credp) { int error; error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp, &exp->nes_numsecflavor, exp->nes_secflavors); if (error) { if (nfs_rootfhset) { exp->nes_exflag = 0; exp->nes_numsecflavor = 0; error = 0; } } else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor > MAXSECFLAVORS) { printf("nfsvno_checkexp: numsecflavors out of range\n"); exp->nes_numsecflavor = 0; error = EACCES; } NFSEXITCODE(error); return (error); } /* * Get a vnode for a file handle and export stuff. */ int nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam, int lktype, struct vnode **vpp, struct nfsexstuff *exp, struct ucred **credp) { int error; *credp = NULL; exp->nes_numsecflavor = 0; error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp); if (error != 0) /* Make sure the server replies ESTALE to the client. */ error = ESTALE; if (nam && !error) { error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp, &exp->nes_numsecflavor, exp->nes_secflavors); if (error) { if (nfs_rootfhset) { exp->nes_exflag = 0; exp->nes_numsecflavor = 0; error = 0; } else { vput(*vpp); } } else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor > MAXSECFLAVORS) { printf("nfsvno_fhtovp: numsecflavors out of range\n"); exp->nes_numsecflavor = 0; error = EACCES; vput(*vpp); } } NFSEXITCODE(error); return (error); } /* * nfsd_fhtovp() - convert a fh to a vnode ptr * - look up fsid in mount list (if not found ret error) * - get vp and export rights by calling nfsvno_fhtovp() * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon * for AUTH_SYS * - if mpp != NULL, return the mount point so that it can * be used for vn_finished_write() by the caller */ void nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype, struct vnode **vpp, struct nfsexstuff *exp, struct mount **mpp, int startwrite, int nextop) { struct mount *mp, *mpw; struct ucred *credanon; fhandle_t *fhp; int error; if (mpp != NULL) *mpp = NULL; *vpp = NULL; fhp = (fhandle_t *)nfp->nfsrvfh_data; mp = vfs_busyfs(&fhp->fh_fsid); if (mp == NULL) { nd->nd_repstat = ESTALE; goto out; } if (startwrite) { mpw = mp; error = vn_start_write(NULL, &mpw, V_WAIT); if (error != 0) { mpw = NULL; vfs_unbusy(mp); nd->nd_repstat = ESTALE; goto out; } if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp))) lktype = LK_EXCLUSIVE; } else mpw = NULL; nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp, &credanon); vfs_unbusy(mp); /* * For NFSv4 without a pseudo root fs, unexported file handles * can be returned, so that Lookup works everywhere. */ if (!nd->nd_repstat && exp->nes_exflag == 0 && !(nd->nd_flag & ND_NFSV4)) { vput(*vpp); *vpp = NULL; nd->nd_repstat = EACCES; } /* * Personally, I've never seen any point in requiring a * reserved port#, since only in the rare case where the * clients are all boxes with secure system privileges, * does it provide any enhanced security, but... some people * believe it to be useful and keep putting this code back in. * (There is also some "security checker" out there that * complains if the nfs server doesn't enforce this.) * However, note the following: * RFC3530 (NFSv4) specifies that a reserved port# not be * required. * RFC2623 recommends that, if a reserved port# is checked for, * that there be a way to turn that off--> ifdef'd. */ #ifdef NFS_REQRSVPORT if (!nd->nd_repstat) { struct sockaddr_in *saddr; struct sockaddr_in6 *saddr6; saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *); saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *); if (!(nd->nd_flag & ND_NFSV4) && ((saddr->sin_family == AF_INET && ntohs(saddr->sin_port) >= IPPORT_RESERVED) || (saddr6->sin6_family == AF_INET6 && ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) { vput(*vpp); nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); } } #endif /* NFS_REQRSVPORT */ /* * Check/setup credentials. */ if (!nd->nd_repstat) { nd->nd_saveduid = nd->nd_cred->cr_uid; nd->nd_repstat = nfsd_excred(nd, exp, credanon, nfsrv_checkwrongsec(nd, nextop, (*vpp)->v_type)); if (nd->nd_repstat) vput(*vpp); } if (credanon != NULL) crfree(credanon); if (nd->nd_repstat) { vn_finished_write(mpw); *vpp = NULL; } else if (mpp != NULL) { *mpp = mpw; } out: NFSEXITCODE2(0, nd); } /* * glue for fp. */ static int fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp) { struct filedesc *fdp; struct file *fp; int error = 0; fdp = p->td_proc->p_fd; if (fd < 0 || fd >= fdp->fd_nfiles || (fp = fdp->fd_ofiles[fd].fde_file) == NULL) { error = EBADF; goto out; } *fpp = fp; out: NFSEXITCODE(error); return (error); } /* * Called from nfssvc() to update the exports list. Just call * vfs_export(). This has to be done, since the v4 root fake fs isn't * in the mount list. */ int nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p) { struct nfsex_args *nfsexargp = (struct nfsex_args *)argp; int error = 0; struct nameidata nd; fhandle_t fh; error = vfs_export(&nfsv4root_mnt, &nfsexargp->export); if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0) nfs_rootfhset = 0; else if (error == 0) { if (nfsexargp->fspec == NULL) { error = EPERM; goto out; } /* * If fspec != NULL, this is the v4root path. */ NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, nfsexargp->fspec, p); if ((error = namei(&nd)) != 0) goto out; error = nfsvno_getfh(nd.ni_vp, &fh, p); vrele(nd.ni_vp); if (!error) { nfs_rootfh.nfsrvfh_len = NFSX_MYFH; NFSBCOPY((caddr_t)&fh, nfs_rootfh.nfsrvfh_data, sizeof (fhandle_t)); nfs_rootfhset = 1; } } out: NFSEXITCODE(error); return (error); } /* * This function needs to test to see if the system is near its limit * for memory allocation via malloc() or mget() and return True iff * either of these resources are near their limit. * XXX (For now, this is just a stub.) */ int nfsrv_testmalloclimit = 0; int nfsrv_mallocmget_limit(void) { static int printmesg = 0; static int testval = 1; if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) { if ((printmesg++ % 100) == 0) printf("nfsd: malloc/mget near limit\n"); return (1); } return (0); } /* * BSD specific initialization of a mount point. */ void nfsd_mntinit(void) { static int inited = 0; if (inited) return; inited = 1; nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED); TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist); TAILQ_INIT(&nfsv4root_mnt.mnt_lazyvnodelist); nfsv4root_mnt.mnt_export = NULL; TAILQ_INIT(&nfsv4root_opt); TAILQ_INIT(&nfsv4root_newopt); nfsv4root_mnt.mnt_opt = &nfsv4root_opt; nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt; nfsv4root_mnt.mnt_nvnodelistsize = 0; nfsv4root_mnt.mnt_lazyvnodelistsize = 0; } /* * Get a vnode for a file handle, without checking exports, etc. */ struct vnode * nfsvno_getvp(fhandle_t *fhp) { struct mount *mp; struct vnode *vp; int error; mp = vfs_busyfs(&fhp->fh_fsid); if (mp == NULL) return (NULL); error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp); vfs_unbusy(mp); if (error) return (NULL); return (vp); } /* * Do a local VOP_ADVLOCK(). */ int nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first, u_int64_t end, struct thread *td) { int error = 0; struct flock fl; u_int64_t tlen; if (nfsrv_dolocallocks == 0) goto out; ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked"); fl.l_whence = SEEK_SET; fl.l_type = ftype; fl.l_start = (off_t)first; if (end == NFS64BITSSET) { fl.l_len = 0; } else { tlen = end - first; fl.l_len = (off_t)tlen; } /* * For FreeBSD8, the l_pid and l_sysid must be set to the same * values for all calls, so that all locks will be held by the * nfsd server. (The nfsd server handles conflicts between the * various clients.) * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024 * bytes, so it can't be put in l_sysid. */ if (nfsv4_sysid == 0) nfsv4_sysid = nlm_acquire_next_sysid(); fl.l_pid = (pid_t)0; fl.l_sysid = (int)nfsv4_sysid; if (ftype == F_UNLCK) error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl, (F_POSIX | F_REMOTE)); else error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl, (F_POSIX | F_REMOTE)); out: NFSEXITCODE(error); return (error); } /* * Check the nfsv4 root exports. */ int nfsvno_v4rootexport(struct nfsrv_descript *nd) { struct ucred *credanon; int error = 0, numsecflavor, secflavors[MAXSECFLAVORS], i; uint64_t exflags; error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags, &credanon, &numsecflavor, secflavors); if (error) { error = NFSERR_PROGUNAVAIL; goto out; } if (credanon != NULL) crfree(credanon); for (i = 0; i < numsecflavor; i++) { if (secflavors[i] == AUTH_SYS) nd->nd_flag |= ND_EXAUTHSYS; else if (secflavors[i] == RPCSEC_GSS_KRB5) nd->nd_flag |= ND_EXGSS; else if (secflavors[i] == RPCSEC_GSS_KRB5I) nd->nd_flag |= ND_EXGSSINTEGRITY; else if (secflavors[i] == RPCSEC_GSS_KRB5P) nd->nd_flag |= ND_EXGSSPRIVACY; } /* And set ND_EXxx flags for TLS. */ if ((exflags & MNT_EXTLS) != 0) { nd->nd_flag |= ND_EXTLS; if ((exflags & MNT_EXTLSCERT) != 0) nd->nd_flag |= ND_EXTLSCERT; if ((exflags & MNT_EXTLSCERTUSER) != 0) nd->nd_flag |= ND_EXTLSCERTUSER; } out: NFSEXITCODE(error); return (error); } /* * Nfs server pseudo system call for the nfsd's */ /* * MPSAFE */ static int nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap) { struct file *fp; struct nfsd_addsock_args sockarg; struct nfsd_nfsd_args nfsdarg; struct nfsd_nfsd_oargs onfsdarg; struct nfsd_pnfsd_args pnfsdarg; struct vnode *vp, *nvp, *curdvp; struct pnfsdsfile *pf; struct nfsdevice *ds, *fds; cap_rights_t rights; int buflen, error, ret; char *buf, *cp, *cp2, *cp3; char fname[PNFS_FILENAME_LEN + 1]; if (uap->flag & NFSSVC_NFSDADDSOCK) { error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg)); if (error) goto out; /* * Since we don't know what rights might be required, * pretend that we need them all. It is better to be too * careful than too reckless. */ error = fget(td, sockarg.sock, cap_rights_init_one(&rights, CAP_SOCK_SERVER), &fp); if (error != 0) goto out; if (fp->f_type != DTYPE_SOCKET) { fdrop(fp, td); error = EPERM; goto out; } error = nfsrvd_addsock(fp); fdrop(fp, td); } else if (uap->flag & NFSSVC_NFSDNFSD) { if (uap->argp == NULL) { error = EINVAL; goto out; } if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) { error = copyin(uap->argp, &onfsdarg, sizeof(onfsdarg)); if (error == 0) { nfsdarg.principal = onfsdarg.principal; nfsdarg.minthreads = onfsdarg.minthreads; nfsdarg.maxthreads = onfsdarg.maxthreads; nfsdarg.version = 1; nfsdarg.addr = NULL; nfsdarg.addrlen = 0; nfsdarg.dnshost = NULL; nfsdarg.dnshostlen = 0; nfsdarg.dspath = NULL; nfsdarg.dspathlen = 0; nfsdarg.mdspath = NULL; nfsdarg.mdspathlen = 0; nfsdarg.mirrorcnt = 1; } } else error = copyin(uap->argp, &nfsdarg, sizeof(nfsdarg)); if (error) goto out; if (nfsdarg.addrlen > 0 && nfsdarg.addrlen < 10000 && nfsdarg.dnshostlen > 0 && nfsdarg.dnshostlen < 10000 && nfsdarg.dspathlen > 0 && nfsdarg.dspathlen < 10000 && nfsdarg.mdspathlen > 0 && nfsdarg.mdspathlen < 10000 && nfsdarg.mirrorcnt >= 1 && nfsdarg.mirrorcnt <= NFSDEV_MAXMIRRORS && nfsdarg.addr != NULL && nfsdarg.dnshost != NULL && nfsdarg.dspath != NULL && nfsdarg.mdspath != NULL) { NFSD_DEBUG(1, "addrlen=%d dspathlen=%d dnslen=%d" " mdspathlen=%d mirrorcnt=%d\n", nfsdarg.addrlen, nfsdarg.dspathlen, nfsdarg.dnshostlen, nfsdarg.mdspathlen, nfsdarg.mirrorcnt); cp = malloc(nfsdarg.addrlen + 1, M_TEMP, M_WAITOK); error = copyin(nfsdarg.addr, cp, nfsdarg.addrlen); if (error != 0) { free(cp, M_TEMP); goto out; } cp[nfsdarg.addrlen] = '\0'; /* Ensure nul term. */ nfsdarg.addr = cp; cp = malloc(nfsdarg.dnshostlen + 1, M_TEMP, M_WAITOK); error = copyin(nfsdarg.dnshost, cp, nfsdarg.dnshostlen); if (error != 0) { free(nfsdarg.addr, M_TEMP); free(cp, M_TEMP); goto out; } cp[nfsdarg.dnshostlen] = '\0'; /* Ensure nul term. */ nfsdarg.dnshost = cp; cp = malloc(nfsdarg.dspathlen + 1, M_TEMP, M_WAITOK); error = copyin(nfsdarg.dspath, cp, nfsdarg.dspathlen); if (error != 0) { free(nfsdarg.addr, M_TEMP); free(nfsdarg.dnshost, M_TEMP); free(cp, M_TEMP); goto out; } cp[nfsdarg.dspathlen] = '\0'; /* Ensure nul term. */ nfsdarg.dspath = cp; cp = malloc(nfsdarg.mdspathlen + 1, M_TEMP, M_WAITOK); error = copyin(nfsdarg.mdspath, cp, nfsdarg.mdspathlen); if (error != 0) { free(nfsdarg.addr, M_TEMP); free(nfsdarg.dnshost, M_TEMP); free(nfsdarg.dspath, M_TEMP); free(cp, M_TEMP); goto out; } cp[nfsdarg.mdspathlen] = '\0'; /* Ensure nul term. */ nfsdarg.mdspath = cp; } else { nfsdarg.addr = NULL; nfsdarg.addrlen = 0; nfsdarg.dnshost = NULL; nfsdarg.dnshostlen = 0; nfsdarg.dspath = NULL; nfsdarg.dspathlen = 0; nfsdarg.mdspath = NULL; nfsdarg.mdspathlen = 0; nfsdarg.mirrorcnt = 1; } error = nfsrvd_nfsd(td, &nfsdarg); free(nfsdarg.addr, M_TEMP); free(nfsdarg.dnshost, M_TEMP); free(nfsdarg.dspath, M_TEMP); free(nfsdarg.mdspath, M_TEMP); } else if (uap->flag & NFSSVC_PNFSDS) { error = copyin(uap->argp, &pnfsdarg, sizeof(pnfsdarg)); if (error == 0 && (pnfsdarg.op == PNFSDOP_DELDSSERVER || pnfsdarg.op == PNFSDOP_FORCEDELDS)) { cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); error = copyinstr(pnfsdarg.dspath, cp, PATH_MAX + 1, NULL); if (error == 0) error = nfsrv_deldsserver(pnfsdarg.op, cp, td); free(cp, M_TEMP); } else if (error == 0 && pnfsdarg.op == PNFSDOP_COPYMR) { cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); buflen = sizeof(*pf) * NFSDEV_MAXMIRRORS; buf = malloc(buflen, M_TEMP, M_WAITOK); error = copyinstr(pnfsdarg.mdspath, cp, PATH_MAX + 1, NULL); NFSD_DEBUG(4, "pnfsdcopymr cp mdspath=%d\n", error); if (error == 0 && pnfsdarg.dspath != NULL) { cp2 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); error = copyinstr(pnfsdarg.dspath, cp2, PATH_MAX + 1, NULL); NFSD_DEBUG(4, "pnfsdcopymr cp dspath=%d\n", error); } else cp2 = NULL; if (error == 0 && pnfsdarg.curdspath != NULL) { cp3 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); error = copyinstr(pnfsdarg.curdspath, cp3, PATH_MAX + 1, NULL); NFSD_DEBUG(4, "pnfsdcopymr cp curdspath=%d\n", error); } else cp3 = NULL; curdvp = NULL; fds = NULL; if (error == 0) error = nfsrv_mdscopymr(cp, cp2, cp3, buf, &buflen, fname, td, &vp, &nvp, &pf, &ds, &fds); NFSD_DEBUG(4, "nfsrv_mdscopymr=%d\n", error); if (error == 0) { if (pf->dsf_dir >= nfsrv_dsdirsize) { printf("copymr: dsdir out of range\n"); pf->dsf_dir = 0; } NFSD_DEBUG(4, "copymr: buflen=%d\n", buflen); error = nfsrv_copymr(vp, nvp, ds->nfsdev_dsdir[pf->dsf_dir], ds, pf, (struct pnfsdsfile *)buf, buflen / sizeof(*pf), td->td_ucred, td); vput(vp); vput(nvp); if (fds != NULL && error == 0) { curdvp = fds->nfsdev_dsdir[pf->dsf_dir]; ret = vn_lock(curdvp, LK_EXCLUSIVE); if (ret == 0) { nfsrv_dsremove(curdvp, fname, td->td_ucred, td); NFSVOPUNLOCK(curdvp); } } NFSD_DEBUG(4, "nfsrv_copymr=%d\n", error); } free(cp, M_TEMP); free(cp2, M_TEMP); free(cp3, M_TEMP); free(buf, M_TEMP); } } else { error = nfssvc_srvcall(td, uap, td->td_ucred); } out: NFSEXITCODE(error); return (error); } static int nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred) { struct nfsex_args export; struct nfsex_oldargs oexp; struct file *fp = NULL; int stablefd, i, len; struct nfsd_clid adminrevoke; struct nfsd_dumplist dumplist; struct nfsd_dumpclients *dumpclients; struct nfsd_dumplocklist dumplocklist; struct nfsd_dumplocks *dumplocks; struct nameidata nd; vnode_t vp; int error = EINVAL, igotlock; struct proc *procp; gid_t *grps; static int suspend_nfsd = 0; if (uap->flag & NFSSVC_PUBLICFH) { NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t)); error = copyin(uap->argp, &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t)); if (!error) nfs_pubfhset = 1; } else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) == (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) { error = copyin(uap->argp,(caddr_t)&export, sizeof (struct nfsex_args)); if (!error) { grps = NULL; if (export.export.ex_ngroups > NGROUPS_MAX || export.export.ex_ngroups < 0) error = EINVAL; else if (export.export.ex_ngroups > 0) { grps = malloc(export.export.ex_ngroups * sizeof(gid_t), M_TEMP, M_WAITOK); error = copyin(export.export.ex_groups, grps, export.export.ex_ngroups * sizeof(gid_t)); export.export.ex_groups = grps; } else export.export.ex_groups = NULL; if (!error) error = nfsrv_v4rootexport(&export, cred, p); free(grps, M_TEMP); } } else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) == NFSSVC_V4ROOTEXPORT) { error = copyin(uap->argp,(caddr_t)&oexp, sizeof (struct nfsex_oldargs)); if (!error) { memset(&export.export, 0, sizeof(export.export)); export.export.ex_flags = (uint64_t)oexp.export.ex_flags; export.export.ex_root = oexp.export.ex_root; export.export.ex_uid = oexp.export.ex_anon.cr_uid; export.export.ex_ngroups = oexp.export.ex_anon.cr_ngroups; export.export.ex_groups = NULL; if (export.export.ex_ngroups > XU_NGROUPS || export.export.ex_ngroups < 0) error = EINVAL; else if (export.export.ex_ngroups > 0) { export.export.ex_groups = malloc( export.export.ex_ngroups * sizeof(gid_t), M_TEMP, M_WAITOK); for (i = 0; i < export.export.ex_ngroups; i++) export.export.ex_groups[i] = oexp.export.ex_anon.cr_groups[i]; } export.export.ex_addr = oexp.export.ex_addr; export.export.ex_addrlen = oexp.export.ex_addrlen; export.export.ex_mask = oexp.export.ex_mask; export.export.ex_masklen = oexp.export.ex_masklen; export.export.ex_indexfile = oexp.export.ex_indexfile; export.export.ex_numsecflavors = oexp.export.ex_numsecflavors; if (export.export.ex_numsecflavors >= MAXSECFLAVORS || export.export.ex_numsecflavors < 0) error = EINVAL; else { for (i = 0; i < export.export.ex_numsecflavors; i++) export.export.ex_secflavors[i] = oexp.export.ex_secflavors[i]; } export.fspec = oexp.fspec; if (error == 0) error = nfsrv_v4rootexport(&export, cred, p); free(export.export.ex_groups, M_TEMP); } } else if (uap->flag & NFSSVC_NOPUBLICFH) { nfs_pubfhset = 0; error = 0; } else if (uap->flag & NFSSVC_STABLERESTART) { error = copyin(uap->argp, (caddr_t)&stablefd, sizeof (int)); if (!error) error = fp_getfvp(p, stablefd, &fp, &vp); if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE)) error = EBADF; if (!error && newnfs_numnfsd != 0) error = EPERM; if (!error) { nfsrv_stablefirst.nsf_fp = fp; nfsrv_setupstable(p); } } else if (uap->flag & NFSSVC_ADMINREVOKE) { error = copyin(uap->argp, (caddr_t)&adminrevoke, sizeof (struct nfsd_clid)); if (!error) error = nfsrv_adminrevoke(&adminrevoke, p); } else if (uap->flag & NFSSVC_DUMPCLIENTS) { error = copyin(uap->argp, (caddr_t)&dumplist, sizeof (struct nfsd_dumplist)); if (!error && (dumplist.ndl_size < 1 || dumplist.ndl_size > NFSRV_MAXDUMPLIST)) error = EPERM; if (!error) { len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size; dumpclients = malloc(len, M_TEMP, M_WAITOK | M_ZERO); nfsrv_dumpclients(dumpclients, dumplist.ndl_size); error = copyout(dumpclients, dumplist.ndl_list, len); free(dumpclients, M_TEMP); } } else if (uap->flag & NFSSVC_DUMPLOCKS) { error = copyin(uap->argp, (caddr_t)&dumplocklist, sizeof (struct nfsd_dumplocklist)); if (!error && (dumplocklist.ndllck_size < 1 || dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST)) error = EPERM; if (!error) error = nfsrv_lookupfilename(&nd, dumplocklist.ndllck_fname, p); if (!error) { len = sizeof (struct nfsd_dumplocks) * dumplocklist.ndllck_size; dumplocks = malloc(len, M_TEMP, M_WAITOK | M_ZERO); nfsrv_dumplocks(nd.ni_vp, dumplocks, dumplocklist.ndllck_size, p); vput(nd.ni_vp); error = copyout(dumplocks, dumplocklist.ndllck_list, len); free(dumplocks, M_TEMP); } } else if (uap->flag & NFSSVC_BACKUPSTABLE) { procp = p->td_proc; PROC_LOCK(procp); nfsd_master_pid = procp->p_pid; bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1); nfsd_master_start = procp->p_stats->p_start; nfsd_master_proc = procp; PROC_UNLOCK(procp); } else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) { NFSLOCKV4ROOTMUTEX(); if (suspend_nfsd == 0) { /* Lock out all nfsd threads */ do { igotlock = nfsv4_lock(&nfsd_suspend_lock, 1, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); } while (igotlock == 0 && suspend_nfsd == 0); suspend_nfsd = 1; } NFSUNLOCKV4ROOTMUTEX(); error = 0; } else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) { NFSLOCKV4ROOTMUTEX(); if (suspend_nfsd != 0) { nfsv4_unlock(&nfsd_suspend_lock, 0); suspend_nfsd = 0; } NFSUNLOCKV4ROOTMUTEX(); error = 0; } NFSEXITCODE(error); return (error); } /* * Check exports. * Returns 0 if ok, 1 otherwise. */ int nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp) { int i; /* * Allow NFSv3 Fsinfo per RFC2623. */ if (((nd->nd_flag & ND_NFSV4) != 0 || nd->nd_procnum != NFSPROC_FSINFO) && ((NFSVNO_EXTLS(exp) && (nd->nd_flag & ND_TLS) == 0) || (NFSVNO_EXTLSCERT(exp) && (nd->nd_flag & ND_TLSCERT) == 0) || (NFSVNO_EXTLSCERTUSER(exp) && (nd->nd_flag & ND_TLSCERTUSER) == 0))) { if ((nd->nd_flag & ND_NFSV4) != 0) return (NFSERR_WRONGSEC); else if ((nd->nd_flag & ND_TLS) == 0) return (NFSERR_AUTHERR | AUTH_NEEDS_TLS); else return (NFSERR_AUTHERR | AUTH_NEEDS_TLS_MUTUAL_HOST); } /* * This seems odd, but allow the case where the security flavor * list is empty. This happens when NFSv4 is traversing non-exported * file systems. Exported file systems should always have a non-empty * security flavor list. */ if (exp->nes_numsecflavor == 0) return (0); for (i = 0; i < exp->nes_numsecflavor; i++) { /* * The tests for privacy and integrity must be first, * since ND_GSS is set for everything but AUTH_SYS. */ if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P && (nd->nd_flag & ND_GSSPRIVACY)) return (0); if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I && (nd->nd_flag & ND_GSSINTEGRITY)) return (0); if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 && (nd->nd_flag & ND_GSS)) return (0); if (exp->nes_secflavors[i] == AUTH_SYS && (nd->nd_flag & ND_GSS) == 0) return (0); } if ((nd->nd_flag & ND_NFSV4) != 0) return (NFSERR_WRONGSEC); return (NFSERR_AUTHERR | AUTH_TOOWEAK); } /* * Calculate a hash value for the fid in a file handle. */ uint32_t nfsrv_hashfh(fhandle_t *fhp) { uint32_t hashval; hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0); return (hashval); } /* * Calculate a hash value for the sessionid. */ uint32_t nfsrv_hashsessionid(uint8_t *sessionid) { uint32_t hashval; hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0); return (hashval); } /* * Signal the userland master nfsd to backup the stable restart file. */ void nfsrv_backupstable(void) { struct proc *procp; if (nfsd_master_proc != NULL) { procp = pfind(nfsd_master_pid); /* Try to make sure it is the correct process. */ if (procp == nfsd_master_proc && procp->p_stats->p_start.tv_sec == nfsd_master_start.tv_sec && procp->p_stats->p_start.tv_usec == nfsd_master_start.tv_usec && strcmp(procp->p_comm, nfsd_master_comm) == 0) kern_psignal(procp, SIGUSR2); else nfsd_master_proc = NULL; if (procp != NULL) PROC_UNLOCK(procp); } } /* * Create a DS data file for nfsrv_pnfscreate(). Called for each mirror. * The arguments are in a structure, so that they can be passed through * taskqueue for a kernel process to execute this function. */ struct nfsrvdscreate { int done; int inprog; struct task tsk; struct ucred *tcred; struct vnode *dvp; NFSPROC_T *p; struct pnfsdsfile *pf; int err; fhandle_t fh; struct vattr va; struct vattr createva; }; int nfsrv_dscreate(struct vnode *dvp, struct vattr *vap, struct vattr *nvap, fhandle_t *fhp, struct pnfsdsfile *pf, struct pnfsdsattr *dsa, char *fnamep, struct ucred *tcred, NFSPROC_T *p, struct vnode **nvpp) { struct vnode *nvp; struct nameidata named; struct vattr va; char *bufp; u_long *hashp; struct nfsnode *np; struct nfsmount *nmp; int error; NFSNAMEICNDSET(&named.ni_cnd, tcred, CREATE, LOCKPARENT | LOCKLEAF | SAVESTART | NOCACHE); nfsvno_setpathbuf(&named, &bufp, &hashp); named.ni_cnd.cn_lkflags = LK_EXCLUSIVE; - named.ni_cnd.cn_thread = p; named.ni_cnd.cn_nameptr = bufp; if (fnamep != NULL) { strlcpy(bufp, fnamep, PNFS_FILENAME_LEN + 1); named.ni_cnd.cn_namelen = strlen(bufp); } else named.ni_cnd.cn_namelen = nfsrv_putfhname(fhp, bufp); NFSD_DEBUG(4, "nfsrv_dscreate: dvp=%p fname=%s\n", dvp, bufp); /* Create the date file in the DS mount. */ error = NFSVOPLOCK(dvp, LK_EXCLUSIVE); if (error == 0) { error = VOP_CREATE(dvp, &nvp, &named.ni_cnd, vap); vref(dvp); VOP_VPUT_PAIR(dvp, error == 0 ? &nvp : NULL, false); if (error == 0) { /* Set the ownership of the file. */ error = VOP_SETATTR(nvp, nvap, tcred); NFSD_DEBUG(4, "nfsrv_dscreate:" " setattr-uid=%d\n", error); if (error != 0) vput(nvp); } if (error != 0) printf("pNFS: pnfscreate failed=%d\n", error); } else printf("pNFS: pnfscreate vnlock=%d\n", error); if (error == 0) { np = VTONFS(nvp); nmp = VFSTONFS(nvp->v_mount); if (strcmp(nvp->v_mount->mnt_vfc->vfc_name, "nfs") != 0 || nmp->nm_nam->sa_len > sizeof( struct sockaddr_in6) || np->n_fhp->nfh_len != NFSX_MYFH) { printf("Bad DS file: fstype=%s salen=%d" " fhlen=%d\n", nvp->v_mount->mnt_vfc->vfc_name, nmp->nm_nam->sa_len, np->n_fhp->nfh_len); error = ENOENT; } /* Set extattrs for the DS on the MDS file. */ if (error == 0) { if (dsa != NULL) { error = VOP_GETATTR(nvp, &va, tcred); if (error == 0) { dsa->dsa_filerev = va.va_filerev; dsa->dsa_size = va.va_size; dsa->dsa_atime = va.va_atime; dsa->dsa_mtime = va.va_mtime; dsa->dsa_bytes = va.va_bytes; } } if (error == 0) { NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, NFSX_MYFH); NFSBCOPY(nmp->nm_nam, &pf->dsf_sin, nmp->nm_nam->sa_len); NFSBCOPY(named.ni_cnd.cn_nameptr, pf->dsf_filename, sizeof(pf->dsf_filename)); } } else printf("pNFS: pnfscreate can't get DS" " attr=%d\n", error); if (nvpp != NULL && error == 0) *nvpp = nvp; else vput(nvp); } nfsvno_relpathbuf(&named); return (error); } /* * Start up the thread that will execute nfsrv_dscreate(). */ static void start_dscreate(void *arg, int pending) { struct nfsrvdscreate *dsc; dsc = (struct nfsrvdscreate *)arg; dsc->err = nfsrv_dscreate(dsc->dvp, &dsc->createva, &dsc->va, &dsc->fh, dsc->pf, NULL, NULL, dsc->tcred, dsc->p, NULL); dsc->done = 1; NFSD_DEBUG(4, "start_dscreate: err=%d\n", dsc->err); } /* * Create a pNFS data file on the Data Server(s). */ static void nfsrv_pnfscreate(struct vnode *vp, struct vattr *vap, struct ucred *cred, NFSPROC_T *p) { struct nfsrvdscreate *dsc, *tdsc = NULL; struct nfsdevice *ds, *tds, *fds; struct mount *mp; struct pnfsdsfile *pf, *tpf; struct pnfsdsattr dsattr; struct vattr va; struct vnode *dvp[NFSDEV_MAXMIRRORS]; struct nfsmount *nmp; fhandle_t fh; uid_t vauid; gid_t vagid; u_short vamode; struct ucred *tcred; int dsdir[NFSDEV_MAXMIRRORS], error, i, mirrorcnt, ret; int failpos, timo; /* Get a DS server directory in a round-robin order. */ mirrorcnt = 1; mp = vp->v_mount; ds = fds = NULL; NFSDDSLOCK(); /* * Search for the first entry that handles this MDS fs, but use the * first entry for all MDS fs's otherwise. */ TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) { if (tds->nfsdev_nmp != NULL) { if (tds->nfsdev_mdsisset == 0 && ds == NULL) ds = tds; else if (tds->nfsdev_mdsisset != 0 && fsidcmp( &mp->mnt_stat.f_fsid, &tds->nfsdev_mdsfsid) == 0) { ds = fds = tds; break; } } } if (ds == NULL) { NFSDDSUNLOCK(); NFSD_DEBUG(4, "nfsrv_pnfscreate: no srv\n"); return; } i = dsdir[0] = ds->nfsdev_nextdir; ds->nfsdev_nextdir = (ds->nfsdev_nextdir + 1) % nfsrv_dsdirsize; dvp[0] = ds->nfsdev_dsdir[i]; tds = TAILQ_NEXT(ds, nfsdev_list); if (nfsrv_maxpnfsmirror > 1 && tds != NULL) { TAILQ_FOREACH_FROM(tds, &nfsrv_devidhead, nfsdev_list) { if (tds->nfsdev_nmp != NULL && ((tds->nfsdev_mdsisset == 0 && fds == NULL) || (tds->nfsdev_mdsisset != 0 && fds != NULL && fsidcmp(&mp->mnt_stat.f_fsid, &tds->nfsdev_mdsfsid) == 0))) { dsdir[mirrorcnt] = i; dvp[mirrorcnt] = tds->nfsdev_dsdir[i]; mirrorcnt++; if (mirrorcnt >= nfsrv_maxpnfsmirror) break; } } } /* Put at end of list to implement round-robin usage. */ TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list); TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list); NFSDDSUNLOCK(); dsc = NULL; if (mirrorcnt > 1) tdsc = dsc = malloc(sizeof(*dsc) * (mirrorcnt - 1), M_TEMP, M_WAITOK | M_ZERO); tpf = pf = malloc(sizeof(*pf) * nfsrv_maxpnfsmirror, M_TEMP, M_WAITOK | M_ZERO); error = nfsvno_getfh(vp, &fh, p); if (error == 0) error = VOP_GETATTR(vp, &va, cred); if (error == 0) { /* Set the attributes for "vp" to Setattr the DS vp. */ vauid = va.va_uid; vagid = va.va_gid; vamode = va.va_mode; VATTR_NULL(&va); va.va_uid = vauid; va.va_gid = vagid; va.va_mode = vamode; va.va_size = 0; } else printf("pNFS: pnfscreate getfh+attr=%d\n", error); NFSD_DEBUG(4, "nfsrv_pnfscreate: cruid=%d crgid=%d\n", cred->cr_uid, cred->cr_gid); /* Make data file name based on FH. */ tcred = newnfs_getcred(); /* * Create the file on each DS mirror, using kernel process(es) for the * additional mirrors. */ failpos = -1; for (i = 0; i < mirrorcnt - 1 && error == 0; i++, tpf++, tdsc++) { tpf->dsf_dir = dsdir[i]; tdsc->tcred = tcred; tdsc->p = p; tdsc->pf = tpf; tdsc->createva = *vap; NFSBCOPY(&fh, &tdsc->fh, sizeof(fh)); tdsc->va = va; tdsc->dvp = dvp[i]; tdsc->done = 0; tdsc->inprog = 0; tdsc->err = 0; ret = EIO; if (nfs_pnfsiothreads != 0) { ret = nfs_pnfsio(start_dscreate, tdsc); NFSD_DEBUG(4, "nfsrv_pnfscreate: nfs_pnfsio=%d\n", ret); } if (ret != 0) { ret = nfsrv_dscreate(dvp[i], vap, &va, &fh, tpf, NULL, NULL, tcred, p, NULL); if (ret != 0) { KASSERT(error == 0, ("nfsrv_dscreate err=%d", error)); if (failpos == -1 && nfsds_failerr(ret)) failpos = i; else error = ret; } } } if (error == 0) { tpf->dsf_dir = dsdir[mirrorcnt - 1]; error = nfsrv_dscreate(dvp[mirrorcnt - 1], vap, &va, &fh, tpf, &dsattr, NULL, tcred, p, NULL); if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(error)) { failpos = mirrorcnt - 1; error = 0; } } timo = hz / 50; /* Wait for 20msec. */ if (timo < 1) timo = 1; /* Wait for kernel task(s) to complete. */ for (tdsc = dsc, i = 0; i < mirrorcnt - 1; i++, tdsc++) { while (tdsc->inprog != 0 && tdsc->done == 0) tsleep(&tdsc->tsk, PVFS, "srvdcr", timo); if (tdsc->err != 0) { if (failpos == -1 && nfsds_failerr(tdsc->err)) failpos = i; else if (error == 0) error = tdsc->err; } } /* * If failpos has been set, that mirror has failed, so it needs * to be disabled. */ if (failpos >= 0) { nmp = VFSTONFS(dvp[failpos]->v_mount); NFSLOCKMNT(nmp); if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | NFSMNTP_CANCELRPCS)) == 0) { nmp->nm_privflag |= NFSMNTP_CANCELRPCS; NFSUNLOCKMNT(nmp); ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p); NFSD_DEBUG(4, "dscreatfail fail=%d ds=%p\n", failpos, ds); if (ds != NULL) nfsrv_killrpcs(nmp); NFSLOCKMNT(nmp); nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; wakeup(nmp); } NFSUNLOCKMNT(nmp); } NFSFREECRED(tcred); if (error == 0) { ASSERT_VOP_ELOCKED(vp, "nfsrv_pnfscreate vp"); NFSD_DEBUG(4, "nfsrv_pnfscreate: mirrorcnt=%d maxmirror=%d\n", mirrorcnt, nfsrv_maxpnfsmirror); /* * For all mirrors that couldn't be created, fill in the * *pf structure, but with an IP address == 0.0.0.0. */ tpf = pf + mirrorcnt; for (i = mirrorcnt; i < nfsrv_maxpnfsmirror; i++, tpf++) { *tpf = *pf; tpf->dsf_sin.sin_family = AF_INET; tpf->dsf_sin.sin_len = sizeof(struct sockaddr_in); tpf->dsf_sin.sin_addr.s_addr = 0; tpf->dsf_sin.sin_port = 0; } error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*pf) * nfsrv_maxpnfsmirror, (char *)pf, p); if (error == 0) error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p); if (error != 0) printf("pNFS: pnfscreate setextattr=%d\n", error); } else printf("pNFS: pnfscreate=%d\n", error); free(pf, M_TEMP); free(dsc, M_TEMP); } /* * Get the information needed to remove the pNFS Data Server file from the * Metadata file. Upon success, ddvp is set non-NULL to the locked * DS directory vnode. The caller must unlock *ddvp when done with it. */ static void nfsrv_pnfsremovesetup(struct vnode *vp, NFSPROC_T *p, struct vnode **dvpp, int *mirrorcntp, char *fname, fhandle_t *fhp) { struct vattr va; struct ucred *tcred; char *buf; int buflen, error; dvpp[0] = NULL; /* If not an exported regular file or not a pNFS server, just return. */ if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 || nfsrv_devidcnt == 0) return; /* Check to see if this is the last hard link. */ tcred = newnfs_getcred(); error = VOP_GETATTR(vp, &va, tcred); NFSFREECRED(tcred); if (error != 0) { printf("pNFS: nfsrv_pnfsremovesetup getattr=%d\n", error); return; } if (va.va_nlink > 1) return; error = nfsvno_getfh(vp, fhp, p); if (error != 0) { printf("pNFS: nfsrv_pnfsremovesetup getfh=%d\n", error); return; } buflen = 1024; buf = malloc(buflen, M_TEMP, M_WAITOK); /* Get the directory vnode for the DS mount and the file handle. */ error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, dvpp, NULL, NULL, fname, NULL, NULL, NULL, NULL, NULL); free(buf, M_TEMP); if (error != 0) printf("pNFS: nfsrv_pnfsremovesetup getsockmnt=%d\n", error); } /* * Remove a DS data file for nfsrv_pnfsremove(). Called for each mirror. * The arguments are in a structure, so that they can be passed through * taskqueue for a kernel process to execute this function. */ struct nfsrvdsremove { int done; int inprog; struct task tsk; struct ucred *tcred; struct vnode *dvp; NFSPROC_T *p; int err; char fname[PNFS_FILENAME_LEN + 1]; }; static int nfsrv_dsremove(struct vnode *dvp, char *fname, struct ucred *tcred, NFSPROC_T *p) { struct nameidata named; struct vnode *nvp; char *bufp; u_long *hashp; int error; error = NFSVOPLOCK(dvp, LK_EXCLUSIVE); if (error != 0) return (error); named.ni_cnd.cn_nameiop = DELETE; named.ni_cnd.cn_lkflags = LK_EXCLUSIVE | LK_RETRY; named.ni_cnd.cn_cred = tcred; - named.ni_cnd.cn_thread = p; named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME; nfsvno_setpathbuf(&named, &bufp, &hashp); named.ni_cnd.cn_nameptr = bufp; named.ni_cnd.cn_namelen = strlen(fname); strlcpy(bufp, fname, NAME_MAX); NFSD_DEBUG(4, "nfsrv_pnfsremove: filename=%s\n", bufp); error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd); NFSD_DEBUG(4, "nfsrv_pnfsremove: aft LOOKUP=%d\n", error); if (error == 0) { error = VOP_REMOVE(dvp, nvp, &named.ni_cnd); vput(nvp); } NFSVOPUNLOCK(dvp); nfsvno_relpathbuf(&named); if (error != 0) printf("pNFS: nfsrv_pnfsremove failed=%d\n", error); return (error); } /* * Start up the thread that will execute nfsrv_dsremove(). */ static void start_dsremove(void *arg, int pending) { struct nfsrvdsremove *dsrm; dsrm = (struct nfsrvdsremove *)arg; dsrm->err = nfsrv_dsremove(dsrm->dvp, dsrm->fname, dsrm->tcred, dsrm->p); dsrm->done = 1; NFSD_DEBUG(4, "start_dsremove: err=%d\n", dsrm->err); } /* * Remove a pNFS data file from a Data Server. * nfsrv_pnfsremovesetup() must have been called before the MDS file was * removed to set up the dvp and fill in the FH. */ static void nfsrv_pnfsremove(struct vnode **dvp, int mirrorcnt, char *fname, fhandle_t *fhp, NFSPROC_T *p) { struct ucred *tcred; struct nfsrvdsremove *dsrm, *tdsrm; struct nfsdevice *ds; struct nfsmount *nmp; int failpos, i, ret, timo; tcred = newnfs_getcred(); dsrm = NULL; if (mirrorcnt > 1) dsrm = malloc(sizeof(*dsrm) * mirrorcnt - 1, M_TEMP, M_WAITOK); /* * Remove the file on each DS mirror, using kernel process(es) for the * additional mirrors. */ failpos = -1; for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) { tdsrm->tcred = tcred; tdsrm->p = p; tdsrm->dvp = dvp[i]; strlcpy(tdsrm->fname, fname, PNFS_FILENAME_LEN + 1); tdsrm->inprog = 0; tdsrm->done = 0; tdsrm->err = 0; ret = EIO; if (nfs_pnfsiothreads != 0) { ret = nfs_pnfsio(start_dsremove, tdsrm); NFSD_DEBUG(4, "nfsrv_pnfsremove: nfs_pnfsio=%d\n", ret); } if (ret != 0) { ret = nfsrv_dsremove(dvp[i], fname, tcred, p); if (failpos == -1 && nfsds_failerr(ret)) failpos = i; } } ret = nfsrv_dsremove(dvp[mirrorcnt - 1], fname, tcred, p); if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(ret)) failpos = mirrorcnt - 1; timo = hz / 50; /* Wait for 20msec. */ if (timo < 1) timo = 1; /* Wait for kernel task(s) to complete. */ for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) { while (tdsrm->inprog != 0 && tdsrm->done == 0) tsleep(&tdsrm->tsk, PVFS, "srvdsrm", timo); if (failpos == -1 && nfsds_failerr(tdsrm->err)) failpos = i; } /* * If failpos has been set, that mirror has failed, so it needs * to be disabled. */ if (failpos >= 0) { nmp = VFSTONFS(dvp[failpos]->v_mount); NFSLOCKMNT(nmp); if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | NFSMNTP_CANCELRPCS)) == 0) { nmp->nm_privflag |= NFSMNTP_CANCELRPCS; NFSUNLOCKMNT(nmp); ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p); NFSD_DEBUG(4, "dsremovefail fail=%d ds=%p\n", failpos, ds); if (ds != NULL) nfsrv_killrpcs(nmp); NFSLOCKMNT(nmp); nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; wakeup(nmp); } NFSUNLOCKMNT(nmp); } /* Get rid all layouts for the file. */ nfsrv_freefilelayouts(fhp); NFSFREECRED(tcred); free(dsrm, M_TEMP); } /* * Generate a file name based on the file handle and put it in *bufp. * Return the number of bytes generated. */ static int nfsrv_putfhname(fhandle_t *fhp, char *bufp) { int i; uint8_t *cp; const uint8_t *hexdigits = "0123456789abcdef"; cp = (uint8_t *)fhp; for (i = 0; i < sizeof(*fhp); i++) { bufp[2 * i] = hexdigits[(*cp >> 4) & 0xf]; bufp[2 * i + 1] = hexdigits[*cp++ & 0xf]; } bufp[2 * i] = '\0'; return (2 * i); } /* * Update the Metadata file's attributes from the DS file when a Read/Write * layout is returned. * Basically just call nfsrv_proxyds() with procedure == NFSPROC_LAYOUTRETURN * so that it does a nfsrv_getattrdsrpc() and nfsrv_setextattr() on the DS file. */ int nfsrv_updatemdsattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p) { struct ucred *tcred; int error; /* Do this as root so that it won't fail with EACCES. */ tcred = newnfs_getcred(); error = nfsrv_proxyds(vp, 0, 0, tcred, p, NFSPROC_LAYOUTRETURN, NULL, NULL, NULL, nap, NULL, NULL, 0, NULL); NFSFREECRED(tcred); return (error); } /* * Set the NFSv4 ACL on the DS file to the same ACL as the MDS file. */ static int nfsrv_dssetacl(struct vnode *vp, struct acl *aclp, struct ucred *cred, NFSPROC_T *p) { int error; error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETACL, NULL, NULL, NULL, NULL, aclp, NULL, 0, NULL); return (error); } static int nfsrv_proxyds(struct vnode *vp, off_t off, int cnt, struct ucred *cred, struct thread *p, int ioproc, struct mbuf **mpp, char *cp, struct mbuf **mpp2, struct nfsvattr *nap, struct acl *aclp, off_t *offp, int content, bool *eofp) { struct nfsmount *nmp[NFSDEV_MAXMIRRORS], *failnmp; fhandle_t fh[NFSDEV_MAXMIRRORS]; struct vnode *dvp[NFSDEV_MAXMIRRORS]; struct nfsdevice *ds; struct pnfsdsattr dsattr; struct opnfsdsattr odsattr; char *buf; int buflen, error, failpos, i, mirrorcnt, origmircnt, trycnt; NFSD_DEBUG(4, "in nfsrv_proxyds\n"); /* * If not a regular file, not exported or not a pNFS server, * just return ENOENT. */ if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 || nfsrv_devidcnt == 0) return (ENOENT); buflen = 1024; buf = malloc(buflen, M_TEMP, M_WAITOK); error = 0; /* * For Getattr, get the Change attribute (va_filerev) and size (va_size) * from the MetaData file's extended attribute. */ if (ioproc == NFSPROC_GETATTR) { error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", &buflen, buf, p); if (error == 0) { if (buflen == sizeof(odsattr)) { NFSBCOPY(buf, &odsattr, buflen); nap->na_filerev = odsattr.dsa_filerev; nap->na_size = odsattr.dsa_size; nap->na_atime = odsattr.dsa_atime; nap->na_mtime = odsattr.dsa_mtime; /* * Fake na_bytes by rounding up na_size. * Since we don't know the block size, just * use BLKDEV_IOSIZE. */ nap->na_bytes = (odsattr.dsa_size + BLKDEV_IOSIZE - 1) & ~(BLKDEV_IOSIZE - 1); } else if (buflen == sizeof(dsattr)) { NFSBCOPY(buf, &dsattr, buflen); nap->na_filerev = dsattr.dsa_filerev; nap->na_size = dsattr.dsa_size; nap->na_atime = dsattr.dsa_atime; nap->na_mtime = dsattr.dsa_mtime; nap->na_bytes = dsattr.dsa_bytes; } else error = ENXIO; } if (error == 0) { /* * If nfsrv_pnfsgetdsattr is 0 or nfsrv_checkdsattr() * returns 0, just return now. nfsrv_checkdsattr() * returns 0 if there is no Read/Write layout * plus either an Open/Write_access or Write * delegation issued to a client for the file. */ if (nfsrv_pnfsgetdsattr == 0 || nfsrv_checkdsattr(vp, p) == 0) { free(buf, M_TEMP); return (error); } } /* * Clear ENOATTR so the code below will attempt to do a * nfsrv_getattrdsrpc() to get the attributes and (re)create * the extended attribute. */ if (error == ENOATTR) error = 0; } origmircnt = -1; trycnt = 0; tryagain: if (error == 0) { buflen = 1024; if (ioproc == NFSPROC_READDS && NFSVOPISLOCKED(vp) == LK_EXCLUSIVE) printf("nfsrv_proxyds: Readds vp exclusively locked\n"); error = nfsrv_dsgetsockmnt(vp, LK_SHARED, buf, &buflen, &mirrorcnt, p, dvp, fh, NULL, NULL, NULL, NULL, NULL, NULL, NULL); if (error == 0) { for (i = 0; i < mirrorcnt; i++) nmp[i] = VFSTONFS(dvp[i]->v_mount); } else printf("pNFS: proxy getextattr sockaddr=%d\n", error); } else printf("pNFS: nfsrv_dsgetsockmnt=%d\n", error); if (error == 0) { failpos = -1; if (origmircnt == -1) origmircnt = mirrorcnt; /* * If failpos is set to a mirror#, then that mirror has * failed and will be disabled. For Read, Getattr and Seek, the * function only tries one mirror, so if that mirror has * failed, it will need to be retried. As such, increment * tryitagain for these cases. * For Write, Setattr and Setacl, the function tries all * mirrors and will not return an error for the case where * one mirror has failed. For these cases, the functioning * mirror(s) will have been modified, so a retry isn't * necessary. These functions will set failpos for the * failed mirror#. */ if (ioproc == NFSPROC_READDS) { error = nfsrv_readdsrpc(fh, off, cnt, cred, p, nmp[0], mpp, mpp2); if (nfsds_failerr(error) && mirrorcnt > 1) { /* * Setting failpos will cause the mirror * to be disabled and then a retry of this * read is required. */ failpos = 0; error = 0; trycnt++; } } else if (ioproc == NFSPROC_WRITEDS) error = nfsrv_writedsrpc(fh, off, cnt, cred, p, vp, &nmp[0], mirrorcnt, mpp, cp, &failpos); else if (ioproc == NFSPROC_SETATTR) error = nfsrv_setattrdsrpc(fh, cred, p, vp, &nmp[0], mirrorcnt, nap, &failpos); else if (ioproc == NFSPROC_SETACL) error = nfsrv_setacldsrpc(fh, cred, p, vp, &nmp[0], mirrorcnt, aclp, &failpos); else if (ioproc == NFSPROC_SEEKDS) { error = nfsrv_seekdsrpc(fh, offp, content, eofp, cred, p, nmp[0]); if (nfsds_failerr(error) && mirrorcnt > 1) { /* * Setting failpos will cause the mirror * to be disabled and then a retry of this * read is required. */ failpos = 0; error = 0; trycnt++; } } else if (ioproc == NFSPROC_ALLOCATE) error = nfsrv_allocatedsrpc(fh, off, *offp, cred, p, vp, &nmp[0], mirrorcnt, &failpos); else if (ioproc == NFSPROC_DEALLOCATE) error = nfsrv_deallocatedsrpc(fh, off, *offp, cred, p, vp, &nmp[0], mirrorcnt, &failpos); else { error = nfsrv_getattrdsrpc(&fh[mirrorcnt - 1], cred, p, vp, nmp[mirrorcnt - 1], nap); if (nfsds_failerr(error) && mirrorcnt > 1) { /* * Setting failpos will cause the mirror * to be disabled and then a retry of this * getattr is required. */ failpos = mirrorcnt - 1; error = 0; trycnt++; } } ds = NULL; if (failpos >= 0) { failnmp = nmp[failpos]; NFSLOCKMNT(failnmp); if ((failnmp->nm_privflag & (NFSMNTP_FORCEDISM | NFSMNTP_CANCELRPCS)) == 0) { failnmp->nm_privflag |= NFSMNTP_CANCELRPCS; NFSUNLOCKMNT(failnmp); ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, failnmp, p); NFSD_DEBUG(4, "dsldsnmp fail=%d ds=%p\n", failpos, ds); if (ds != NULL) nfsrv_killrpcs(failnmp); NFSLOCKMNT(failnmp); failnmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; wakeup(failnmp); } NFSUNLOCKMNT(failnmp); } for (i = 0; i < mirrorcnt; i++) NFSVOPUNLOCK(dvp[i]); NFSD_DEBUG(4, "nfsrv_proxyds: aft RPC=%d trya=%d\n", error, trycnt); /* Try the Read/Getattr again if a mirror was deleted. */ if (ds != NULL && trycnt > 0 && trycnt < origmircnt) goto tryagain; } else { /* Return ENOENT for any Extended Attribute error. */ error = ENOENT; } free(buf, M_TEMP); NFSD_DEBUG(4, "nfsrv_proxyds: error=%d\n", error); return (error); } /* * Get the DS mount point, fh and directory from the "pnfsd.dsfile" extended * attribute. * newnmpp - If it points to a non-NULL nmp, that is the destination and needs * to be checked. If it points to a NULL nmp, then it returns * a suitable destination. * curnmp - If non-NULL, it is the source mount for the copy. */ int nfsrv_dsgetsockmnt(struct vnode *vp, int lktype, char *buf, int *buflenp, int *mirrorcntp, NFSPROC_T *p, struct vnode **dvpp, fhandle_t *fhp, char *devid, char *fnamep, struct vnode **nvpp, struct nfsmount **newnmpp, struct nfsmount *curnmp, int *ippos, int *dsdirp) { struct vnode *dvp, *nvp = NULL, **tdvpp; struct mount *mp; struct nfsmount *nmp, *newnmp; struct sockaddr *sad; struct sockaddr_in *sin; struct nfsdevice *ds, *tds, *fndds; struct pnfsdsfile *pf; uint32_t dsdir; int error, fhiszero, fnd, gotone, i, mirrorcnt; ASSERT_VOP_LOCKED(vp, "nfsrv_dsgetsockmnt vp"); *mirrorcntp = 1; tdvpp = dvpp; if (nvpp != NULL) *nvpp = NULL; if (dvpp != NULL) *dvpp = NULL; if (ippos != NULL) *ippos = -1; if (newnmpp != NULL) newnmp = *newnmpp; else newnmp = NULL; mp = vp->v_mount; error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", buflenp, buf, p); mirrorcnt = *buflenp / sizeof(*pf); if (error == 0 && (mirrorcnt < 1 || mirrorcnt > NFSDEV_MAXMIRRORS || *buflenp != sizeof(*pf) * mirrorcnt)) error = ENOATTR; pf = (struct pnfsdsfile *)buf; /* If curnmp != NULL, check for a match in the mirror list. */ if (curnmp != NULL && error == 0) { fnd = 0; for (i = 0; i < mirrorcnt; i++, pf++) { sad = (struct sockaddr *)&pf->dsf_sin; if (nfsaddr2_match(sad, curnmp->nm_nam)) { if (ippos != NULL) *ippos = i; fnd = 1; break; } } if (fnd == 0) error = ENXIO; } gotone = 0; pf = (struct pnfsdsfile *)buf; NFSD_DEBUG(4, "nfsrv_dsgetsockmnt: mirrorcnt=%d err=%d\n", mirrorcnt, error); for (i = 0; i < mirrorcnt && error == 0; i++, pf++) { fhiszero = 0; sad = (struct sockaddr *)&pf->dsf_sin; sin = &pf->dsf_sin; dsdir = pf->dsf_dir; if (dsdir >= nfsrv_dsdirsize) { printf("nfsrv_dsgetsockmnt: dsdir=%d\n", dsdir); error = ENOATTR; } else if (nvpp != NULL && newnmp != NULL && nfsaddr2_match(sad, newnmp->nm_nam)) error = EEXIST; if (error == 0) { if (ippos != NULL && curnmp == NULL && sad->sa_family == AF_INET && sin->sin_addr.s_addr == 0) *ippos = i; if (NFSBCMP(&zerofh, &pf->dsf_fh, sizeof(zerofh)) == 0) fhiszero = 1; /* Use the socket address to find the mount point. */ fndds = NULL; NFSDDSLOCK(); /* Find a match for the IP address. */ TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { if (ds->nfsdev_nmp != NULL) { dvp = ds->nfsdev_dvp; nmp = VFSTONFS(dvp->v_mount); if (nmp != ds->nfsdev_nmp) printf("different2 nmp %p %p\n", nmp, ds->nfsdev_nmp); if (nfsaddr2_match(sad, nmp->nm_nam)) { fndds = ds; break; } } } if (fndds != NULL && newnmpp != NULL && newnmp == NULL) { /* Search for a place to make a mirror copy. */ TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) { if (tds->nfsdev_nmp != NULL && fndds != tds && ((tds->nfsdev_mdsisset == 0 && fndds->nfsdev_mdsisset == 0) || (tds->nfsdev_mdsisset != 0 && fndds->nfsdev_mdsisset != 0 && fsidcmp(&tds->nfsdev_mdsfsid, &mp->mnt_stat.f_fsid) == 0))) { *newnmpp = tds->nfsdev_nmp; break; } } if (tds != NULL) { /* * Move this entry to the end of the * list, so it won't be selected as * easily the next time. */ TAILQ_REMOVE(&nfsrv_devidhead, tds, nfsdev_list); TAILQ_INSERT_TAIL(&nfsrv_devidhead, tds, nfsdev_list); } } NFSDDSUNLOCK(); if (fndds != NULL) { dvp = fndds->nfsdev_dsdir[dsdir]; if (lktype != 0 || fhiszero != 0 || (nvpp != NULL && *nvpp == NULL)) { if (fhiszero != 0) error = vn_lock(dvp, LK_EXCLUSIVE); else if (lktype != 0) error = vn_lock(dvp, lktype); else error = vn_lock(dvp, LK_SHARED); /* * If the file handle is all 0's, try to * do a Lookup against the DS to acquire * it. * If dvpp == NULL or the Lookup fails, * unlock dvp after the call. */ if (error == 0 && (fhiszero != 0 || (nvpp != NULL && *nvpp == NULL))) { error = nfsrv_pnfslookupds(vp, dvp, pf, &nvp, p); if (error == 0) { if (fhiszero != 0) nfsrv_pnfssetfh( vp, pf, devid, fnamep, nvp, p); if (nvpp != NULL && *nvpp == NULL) { *nvpp = nvp; *dsdirp = dsdir; } else vput(nvp); } if (error != 0 || lktype == 0) NFSVOPUNLOCK(dvp); } } if (error == 0) { gotone++; NFSD_DEBUG(4, "gotone=%d\n", gotone); if (devid != NULL) { NFSBCOPY(fndds->nfsdev_deviceid, devid, NFSX_V4DEVICEID); devid += NFSX_V4DEVICEID; } if (dvpp != NULL) *tdvpp++ = dvp; if (fhp != NULL) NFSBCOPY(&pf->dsf_fh, fhp++, NFSX_MYFH); if (fnamep != NULL && gotone == 1) strlcpy(fnamep, pf->dsf_filename, sizeof(pf->dsf_filename)); } else NFSD_DEBUG(4, "nfsrv_dsgetsockmnt " "err=%d\n", error); } } } if (error == 0 && gotone == 0) error = ENOENT; NFSD_DEBUG(4, "eo nfsrv_dsgetsockmnt: gotone=%d err=%d\n", gotone, error); if (error == 0) *mirrorcntp = gotone; else { if (gotone > 0 && dvpp != NULL) { /* * If the error didn't occur on the first one and * dvpp != NULL, the one(s) prior to the failure will * have locked dvp's that need to be unlocked. */ for (i = 0; i < gotone; i++) { NFSVOPUNLOCK(*dvpp); *dvpp++ = NULL; } } /* * If it found the vnode to be copied from before a failure, * it needs to be vput()'d. */ if (nvpp != NULL && *nvpp != NULL) { vput(*nvpp); *nvpp = NULL; } } return (error); } /* * Set the extended attribute for the Change attribute. */ static int nfsrv_setextattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p) { struct pnfsdsattr dsattr; int error; ASSERT_VOP_ELOCKED(vp, "nfsrv_setextattr vp"); dsattr.dsa_filerev = nap->na_filerev; dsattr.dsa_size = nap->na_size; dsattr.dsa_atime = nap->na_atime; dsattr.dsa_mtime = nap->na_mtime; dsattr.dsa_bytes = nap->na_bytes; error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p); if (error != 0) printf("pNFS: setextattr=%d\n", error); return (error); } static int nfsrv_readdsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred, NFSPROC_T *p, struct nfsmount *nmp, struct mbuf **mpp, struct mbuf **mpendp) { uint32_t *tl; struct nfsrv_descript *nd; nfsv4stateid_t st; struct mbuf *m, *m2; int error = 0, retlen, tlen, trimlen; NFSD_DEBUG(4, "in nfsrv_readdsrpc\n"); nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); *mpp = NULL; /* * Use a stateid where other is an alternating 01010 pattern and * seqid is 0xffffffff. This value is not defined as special by * the RFC and is used by the FreeBSD NFS server to indicate an * MDS->DS proxy operation. */ st.other[0] = 0x55555555; st.other[1] = 0x55555555; st.other[2] = 0x55555555; st.seqid = 0xffffffff; nfscl_reqstart(nd, NFSPROC_READDS, nmp, (u_int8_t *)fhp, sizeof(*fhp), NULL, NULL, 0, 0); nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED * 3); txdr_hyper(off, tl); *(tl + 2) = txdr_unsigned(len); error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) { free(nd, M_TEMP); return (error); } if (nd->nd_repstat == 0) { NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); NFSM_STRSIZ(retlen, len); if (retlen > 0) { /* Trim off the pre-data XDR from the mbuf chain. */ m = nd->nd_mrep; while (m != NULL && m != nd->nd_md) { if (m->m_next == nd->nd_md) { m->m_next = NULL; m_freem(nd->nd_mrep); nd->nd_mrep = m = nd->nd_md; } else m = m->m_next; } if (m == NULL) { printf("nfsrv_readdsrpc: busted mbuf list\n"); error = ENOENT; goto nfsmout; } /* * Now, adjust first mbuf so that any XDR before the * read data is skipped over. */ trimlen = nd->nd_dpos - mtod(m, char *); if (trimlen > 0) { m->m_len -= trimlen; NFSM_DATAP(m, trimlen); } /* * Truncate the mbuf chain at retlen bytes of data, * plus XDR padding that brings the length up to a * multiple of 4. */ tlen = NFSM_RNDUP(retlen); do { if (m->m_len >= tlen) { m->m_len = tlen; tlen = 0; m2 = m->m_next; m->m_next = NULL; m_freem(m2); break; } tlen -= m->m_len; m = m->m_next; } while (m != NULL); if (tlen > 0) { printf("nfsrv_readdsrpc: busted mbuf list\n"); error = ENOENT; goto nfsmout; } *mpp = nd->nd_mrep; *mpendp = m; nd->nd_mrep = NULL; } } else error = nd->nd_repstat; nfsmout: /* If nd->nd_mrep is already NULL, this is a no-op. */ m_freem(nd->nd_mrep); free(nd, M_TEMP); NFSD_DEBUG(4, "nfsrv_readdsrpc error=%d\n", error); return (error); } /* * Do a write RPC on a DS data file, using this structure for the arguments, * so that this function can be executed by a separate kernel process. */ struct nfsrvwritedsdorpc { int done; int inprog; struct task tsk; fhandle_t fh; off_t off; int len; struct nfsmount *nmp; struct ucred *cred; NFSPROC_T *p; struct mbuf *m; int err; }; static int nfsrv_writedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, int len, struct nfsvattr *nap, struct mbuf *m, struct ucred *cred, NFSPROC_T *p) { uint32_t *tl; struct nfsrv_descript *nd; nfsattrbit_t attrbits; nfsv4stateid_t st; int commit, error, retlen; nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); nfscl_reqstart(nd, NFSPROC_WRITE, nmp, (u_int8_t *)fhp, sizeof(fhandle_t), NULL, NULL, 0, 0); /* * Use a stateid where other is an alternating 01010 pattern and * seqid is 0xffffffff. This value is not defined as special by * the RFC and is used by the FreeBSD NFS server to indicate an * MDS->DS proxy operation. */ st.other[0] = 0x55555555; st.other[1] = 0x55555555; st.other[2] = 0x55555555; st.seqid = 0xffffffff; nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED); txdr_hyper(off, tl); tl += 2; /* * Do all writes FileSync, since the server doesn't hold onto dirty * buffers. Since clients should be accessing the DS servers directly * using the pNFS layouts, this just needs to work correctly as a * fallback. */ *tl++ = txdr_unsigned(NFSWRITE_FILESYNC); *tl = txdr_unsigned(len); NFSD_DEBUG(4, "nfsrv_writedsdorpc: len=%d\n", len); /* Put data in mbuf chain. */ nd->nd_mb->m_next = m; /* Set nd_mb and nd_bpos to end of data. */ while (m->m_next != NULL) m = m->m_next; nd->nd_mb = m; nfsm_set(nd, m->m_len); NFSD_DEBUG(4, "nfsrv_writedsdorpc: lastmb len=%d\n", m->m_len); /* Do a Getattr for the attributes that change upon writing. */ NFSZERO_ATTRBIT(&attrbits); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_GETATTR); (void) nfsrv_putattrbit(nd, &attrbits); error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) { free(nd, M_TEMP); return (error); } NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft writerpc=%d\n", nd->nd_repstat); /* Get rid of weak cache consistency data for now. */ if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == (ND_NFSV4 | ND_V4WCCATTR)) { error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); NFSD_DEBUG(4, "nfsrv_writedsdorpc: wcc attr=%d\n", error); if (error != 0) goto nfsmout; /* * Get rid of Op# and status for next op. */ NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); if (*++tl != 0) nd->nd_flag |= ND_NOMOREDATA; } if (nd->nd_repstat == 0) { NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF); retlen = fxdr_unsigned(int, *tl++); commit = fxdr_unsigned(int, *tl); if (commit != NFSWRITE_FILESYNC) error = NFSERR_IO; NFSD_DEBUG(4, "nfsrv_writedsdorpc:retlen=%d commit=%d err=%d\n", retlen, commit, error); } else error = nd->nd_repstat; /* We have no use for the Write Verifier since we use FileSync. */ /* * Get the Change, Size, Access Time and Modify Time attributes and set * on the Metadata file, so its attributes will be what the file's * would be if it had been written. */ if (error == 0) { NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); } NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft loadattr=%d\n", error); nfsmout: m_freem(nd->nd_mrep); free(nd, M_TEMP); NFSD_DEBUG(4, "nfsrv_writedsdorpc error=%d\n", error); return (error); } /* * Start up the thread that will execute nfsrv_writedsdorpc(). */ static void start_writedsdorpc(void *arg, int pending) { struct nfsrvwritedsdorpc *drpc; drpc = (struct nfsrvwritedsdorpc *)arg; drpc->err = nfsrv_writedsdorpc(drpc->nmp, &drpc->fh, drpc->off, drpc->len, NULL, drpc->m, drpc->cred, drpc->p); drpc->done = 1; NFSD_DEBUG(4, "start_writedsdorpc: err=%d\n", drpc->err); } static int nfsrv_writedsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred, NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct mbuf **mpp, char *cp, int *failposp) { struct nfsrvwritedsdorpc *drpc, *tdrpc = NULL; struct nfsvattr na; struct mbuf *m; int error, i, offs, ret, timo; NFSD_DEBUG(4, "in nfsrv_writedsrpc\n"); KASSERT(*mpp != NULL, ("nfsrv_writedsrpc: NULL mbuf chain")); drpc = NULL; if (mirrorcnt > 1) tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, M_WAITOK); /* Calculate offset in mbuf chain that data starts. */ offs = cp - mtod(*mpp, char *); NFSD_DEBUG(4, "nfsrv_writedsrpc: mcopy offs=%d len=%d\n", offs, len); /* * Do the write RPC for every DS, using a separate kernel process * for every DS except the last one. */ error = 0; for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { tdrpc->done = 0; NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); tdrpc->off = off; tdrpc->len = len; tdrpc->nmp = *nmpp; tdrpc->cred = cred; tdrpc->p = p; tdrpc->inprog = 0; tdrpc->err = 0; tdrpc->m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK); ret = EIO; if (nfs_pnfsiothreads != 0) { ret = nfs_pnfsio(start_writedsdorpc, tdrpc); NFSD_DEBUG(4, "nfsrv_writedsrpc: nfs_pnfsio=%d\n", ret); } if (ret != 0) { ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, NULL, tdrpc->m, cred, p); if (nfsds_failerr(ret) && *failposp == -1) *failposp = i; else if (error == 0 && ret != 0) error = ret; } nmpp++; fhp++; } m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK); ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, &na, m, cred, p); if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) *failposp = mirrorcnt - 1; else if (error == 0 && ret != 0) error = ret; if (error == 0) error = nfsrv_setextattr(vp, &na, p); NFSD_DEBUG(4, "nfsrv_writedsrpc: aft setextat=%d\n", error); tdrpc = drpc; timo = hz / 50; /* Wait for 20msec. */ if (timo < 1) timo = 1; for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { /* Wait for RPCs on separate threads to complete. */ while (tdrpc->inprog != 0 && tdrpc->done == 0) tsleep(&tdrpc->tsk, PVFS, "srvwrds", timo); if (nfsds_failerr(tdrpc->err) && *failposp == -1) *failposp = i; else if (error == 0 && tdrpc->err != 0) error = tdrpc->err; } free(drpc, M_TEMP); return (error); } /* * Do a allocate RPC on a DS data file, using this structure for the arguments, * so that this function can be executed by a separate kernel process. */ struct nfsrvallocatedsdorpc { int done; int inprog; struct task tsk; fhandle_t fh; off_t off; off_t len; struct nfsmount *nmp; struct ucred *cred; NFSPROC_T *p; int err; }; static int nfsrv_allocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p) { uint32_t *tl; struct nfsrv_descript *nd; nfsattrbit_t attrbits; nfsv4stateid_t st; int error; nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); nfscl_reqstart(nd, NFSPROC_ALLOCATE, nmp, (u_int8_t *)fhp, sizeof(fhandle_t), NULL, NULL, 0, 0); /* * Use a stateid where other is an alternating 01010 pattern and * seqid is 0xffffffff. This value is not defined as special by * the RFC and is used by the FreeBSD NFS server to indicate an * MDS->DS proxy operation. */ st.other[0] = 0x55555555; st.other[1] = 0x55555555; st.other[2] = 0x55555555; st.seqid = 0xffffffff; nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED); txdr_hyper(off, tl); tl += 2; txdr_hyper(len, tl); tl += 2; NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: len=%jd\n", (intmax_t)len); *tl = txdr_unsigned(NFSV4OP_GETATTR); NFSGETATTR_ATTRBIT(&attrbits); nfsrv_putattrbit(nd, &attrbits); error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) { free(nd, M_TEMP); return (error); } NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft allocaterpc=%d\n", nd->nd_repstat); if (nd->nd_repstat == 0) { NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); } else error = nd->nd_repstat; NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft loadattr=%d\n", error); nfsmout: m_freem(nd->nd_mrep); free(nd, M_TEMP); NFSD_DEBUG(4, "nfsrv_allocatedsdorpc error=%d\n", error); return (error); } /* * Start up the thread that will execute nfsrv_allocatedsdorpc(). */ static void start_allocatedsdorpc(void *arg, int pending) { struct nfsrvallocatedsdorpc *drpc; drpc = (struct nfsrvallocatedsdorpc *)arg; drpc->err = nfsrv_allocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off, drpc->len, NULL, drpc->cred, drpc->p); drpc->done = 1; NFSD_DEBUG(4, "start_allocatedsdorpc: err=%d\n", drpc->err); } static int nfsrv_allocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred, NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, int *failposp) { struct nfsrvallocatedsdorpc *drpc, *tdrpc = NULL; struct nfsvattr na; int error, i, ret, timo; NFSD_DEBUG(4, "in nfsrv_allocatedsrpc\n"); drpc = NULL; if (mirrorcnt > 1) tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, M_WAITOK); /* * Do the allocate RPC for every DS, using a separate kernel process * for every DS except the last one. */ error = 0; for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { tdrpc->done = 0; NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); tdrpc->off = off; tdrpc->len = len; tdrpc->nmp = *nmpp; tdrpc->cred = cred; tdrpc->p = p; tdrpc->inprog = 0; tdrpc->err = 0; ret = EIO; if (nfs_pnfsiothreads != 0) { ret = nfs_pnfsio(start_allocatedsdorpc, tdrpc); NFSD_DEBUG(4, "nfsrv_allocatedsrpc: nfs_pnfsio=%d\n", ret); } if (ret != 0) { ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, NULL, cred, p); if (nfsds_failerr(ret) && *failposp == -1) *failposp = i; else if (error == 0 && ret != 0) error = ret; } nmpp++; fhp++; } ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p); if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) *failposp = mirrorcnt - 1; else if (error == 0 && ret != 0) error = ret; if (error == 0) error = nfsrv_setextattr(vp, &na, p); NFSD_DEBUG(4, "nfsrv_allocatedsrpc: aft setextat=%d\n", error); tdrpc = drpc; timo = hz / 50; /* Wait for 20msec. */ if (timo < 1) timo = 1; for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { /* Wait for RPCs on separate threads to complete. */ while (tdrpc->inprog != 0 && tdrpc->done == 0) tsleep(&tdrpc->tsk, PVFS, "srvalds", timo); if (nfsds_failerr(tdrpc->err) && *failposp == -1) *failposp = i; else if (error == 0 && tdrpc->err != 0) error = tdrpc->err; } free(drpc, M_TEMP); return (error); } /* * Do a deallocate RPC on a DS data file, using this structure for the * arguments, so that this function can be executed by a separate kernel * process. */ struct nfsrvdeallocatedsdorpc { int done; int inprog; struct task tsk; fhandle_t fh; off_t off; off_t len; struct nfsmount *nmp; struct ucred *cred; NFSPROC_T *p; int err; }; static int nfsrv_deallocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p) { uint32_t *tl; struct nfsrv_descript *nd; nfsattrbit_t attrbits; nfsv4stateid_t st; int error; nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); nfscl_reqstart(nd, NFSPROC_DEALLOCATE, nmp, (u_int8_t *)fhp, sizeof(fhandle_t), NULL, NULL, 0, 0); /* * Use a stateid where other is an alternating 01010 pattern and * seqid is 0xffffffff. This value is not defined as special by * the RFC and is used by the FreeBSD NFS server to indicate an * MDS->DS proxy operation. */ st.other[0] = 0x55555555; st.other[1] = 0x55555555; st.other[2] = 0x55555555; st.seqid = 0xffffffff; nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED); txdr_hyper(off, tl); tl += 2; txdr_hyper(len, tl); tl += 2; NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: len=%jd\n", (intmax_t)len); /* Do a Getattr for the attributes that change upon writing. */ NFSZERO_ATTRBIT(&attrbits); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); *tl = txdr_unsigned(NFSV4OP_GETATTR); nfsrv_putattrbit(nd, &attrbits); error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) { free(nd, M_TEMP); return (error); } NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft deallocaterpc=%d\n", nd->nd_repstat); /* Get rid of weak cache consistency data for now. */ if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == (ND_NFSV4 | ND_V4WCCATTR)) { error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: wcc attr=%d\n", error); if (error != 0) goto nfsmout; /* * Get rid of Op# and status for next op. */ NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); if (*++tl != 0) nd->nd_flag |= ND_NOMOREDATA; } if (nd->nd_repstat == 0) { NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); } else error = nd->nd_repstat; NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft loadattr=%d\n", error); nfsmout: m_freem(nd->nd_mrep); free(nd, M_TEMP); NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc error=%d\n", error); return (error); } /* * Start up the thread that will execute nfsrv_deallocatedsdorpc(). */ static void start_deallocatedsdorpc(void *arg, int pending) { struct nfsrvdeallocatedsdorpc *drpc; drpc = (struct nfsrvdeallocatedsdorpc *)arg; drpc->err = nfsrv_deallocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off, drpc->len, NULL, drpc->cred, drpc->p); drpc->done = 1; NFSD_DEBUG(4, "start_deallocatedsdorpc: err=%d\n", drpc->err); } static int nfsrv_deallocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred, NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, int *failposp) { struct nfsrvdeallocatedsdorpc *drpc, *tdrpc = NULL; struct nfsvattr na; int error, i, ret, timo; NFSD_DEBUG(4, "in nfsrv_deallocatedsrpc\n"); drpc = NULL; if (mirrorcnt > 1) tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, M_WAITOK); /* * Do the deallocate RPC for every DS, using a separate kernel process * for every DS except the last one. */ error = 0; for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { tdrpc->done = 0; NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); tdrpc->off = off; tdrpc->len = len; tdrpc->nmp = *nmpp; tdrpc->cred = cred; tdrpc->p = p; tdrpc->inprog = 0; tdrpc->err = 0; ret = EIO; if (nfs_pnfsiothreads != 0) { ret = nfs_pnfsio(start_deallocatedsdorpc, tdrpc); NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: nfs_pnfsio=%d\n", ret); } if (ret != 0) { ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len, NULL, cred, p); if (nfsds_failerr(ret) && *failposp == -1) *failposp = i; else if (error == 0 && ret != 0) error = ret; } nmpp++; fhp++; } ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p); if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) *failposp = mirrorcnt - 1; else if (error == 0 && ret != 0) error = ret; if (error == 0) error = nfsrv_setextattr(vp, &na, p); NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: aft setextat=%d\n", error); tdrpc = drpc; timo = hz / 50; /* Wait for 20msec. */ if (timo < 1) timo = 1; for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { /* Wait for RPCs on separate threads to complete. */ while (tdrpc->inprog != 0 && tdrpc->done == 0) tsleep(&tdrpc->tsk, PVFS, "srvalds", timo); if (nfsds_failerr(tdrpc->err) && *failposp == -1) *failposp = i; else if (error == 0 && tdrpc->err != 0) error = tdrpc->err; } free(drpc, M_TEMP); return (error); } static int nfsrv_setattrdsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap, struct nfsvattr *dsnap) { uint32_t *tl; struct nfsrv_descript *nd; nfsv4stateid_t st; nfsattrbit_t attrbits; int error; NFSD_DEBUG(4, "in nfsrv_setattrdsdorpc\n"); nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); /* * Use a stateid where other is an alternating 01010 pattern and * seqid is 0xffffffff. This value is not defined as special by * the RFC and is used by the FreeBSD NFS server to indicate an * MDS->DS proxy operation. */ st.other[0] = 0x55555555; st.other[1] = 0x55555555; st.other[2] = 0x55555555; st.seqid = 0xffffffff; nfscl_reqstart(nd, NFSPROC_SETATTR, nmp, (u_int8_t *)fhp, sizeof(*fhp), NULL, NULL, 0, 0); nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); nfscl_fillsattr(nd, &nap->na_vattr, vp, NFSSATTR_FULL, 0); /* Do a Getattr for the attributes that change due to writing. */ NFSZERO_ATTRBIT(&attrbits); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_GETATTR); (void) nfsrv_putattrbit(nd, &attrbits); error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) { free(nd, M_TEMP); return (error); } NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattrrpc=%d\n", nd->nd_repstat); /* Get rid of weak cache consistency data for now. */ if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == (ND_NFSV4 | ND_V4WCCATTR)) { error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: wcc attr=%d\n", error); if (error != 0) goto nfsmout; /* * Get rid of Op# and status for next op. */ NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); if (*++tl != 0) nd->nd_flag |= ND_NOMOREDATA; } error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); if (error != 0) goto nfsmout; if (nd->nd_repstat != 0) error = nd->nd_repstat; /* * Get the Change, Size, Access Time and Modify Time attributes and set * on the Metadata file, so its attributes will be what the file's * would be if it had been written. */ if (error == 0) { NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); } NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattr loadattr=%d\n", error); nfsmout: m_freem(nd->nd_mrep); free(nd, M_TEMP); NFSD_DEBUG(4, "nfsrv_setattrdsdorpc error=%d\n", error); return (error); } struct nfsrvsetattrdsdorpc { int done; int inprog; struct task tsk; fhandle_t fh; struct nfsmount *nmp; struct vnode *vp; struct ucred *cred; NFSPROC_T *p; struct nfsvattr na; struct nfsvattr dsna; int err; }; /* * Start up the thread that will execute nfsrv_setattrdsdorpc(). */ static void start_setattrdsdorpc(void *arg, int pending) { struct nfsrvsetattrdsdorpc *drpc; drpc = (struct nfsrvsetattrdsdorpc *)arg; drpc->err = nfsrv_setattrdsdorpc(&drpc->fh, drpc->cred, drpc->p, drpc->vp, drpc->nmp, &drpc->na, &drpc->dsna); drpc->done = 1; } static int nfsrv_setattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct nfsvattr *nap, int *failposp) { struct nfsrvsetattrdsdorpc *drpc, *tdrpc = NULL; struct nfsvattr na; int error, i, ret, timo; NFSD_DEBUG(4, "in nfsrv_setattrdsrpc\n"); drpc = NULL; if (mirrorcnt > 1) tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, M_WAITOK); /* * Do the setattr RPC for every DS, using a separate kernel process * for every DS except the last one. */ error = 0; for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { tdrpc->done = 0; tdrpc->inprog = 0; NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); tdrpc->nmp = *nmpp; tdrpc->vp = vp; tdrpc->cred = cred; tdrpc->p = p; tdrpc->na = *nap; tdrpc->err = 0; ret = EIO; if (nfs_pnfsiothreads != 0) { ret = nfs_pnfsio(start_setattrdsdorpc, tdrpc); NFSD_DEBUG(4, "nfsrv_setattrdsrpc: nfs_pnfsio=%d\n", ret); } if (ret != 0) { ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, &na); if (nfsds_failerr(ret) && *failposp == -1) *failposp = i; else if (error == 0 && ret != 0) error = ret; } nmpp++; fhp++; } ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, &na); if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) *failposp = mirrorcnt - 1; else if (error == 0 && ret != 0) error = ret; if (error == 0) error = nfsrv_setextattr(vp, &na, p); NFSD_DEBUG(4, "nfsrv_setattrdsrpc: aft setextat=%d\n", error); tdrpc = drpc; timo = hz / 50; /* Wait for 20msec. */ if (timo < 1) timo = 1; for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { /* Wait for RPCs on separate threads to complete. */ while (tdrpc->inprog != 0 && tdrpc->done == 0) tsleep(&tdrpc->tsk, PVFS, "srvsads", timo); if (nfsds_failerr(tdrpc->err) && *failposp == -1) *failposp = i; else if (error == 0 && tdrpc->err != 0) error = tdrpc->err; } free(drpc, M_TEMP); return (error); } /* * Do a Setattr of an NFSv4 ACL on the DS file. */ static int nfsrv_setacldsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, struct vnode *vp, struct nfsmount *nmp, struct acl *aclp) { struct nfsrv_descript *nd; nfsv4stateid_t st; nfsattrbit_t attrbits; int error; NFSD_DEBUG(4, "in nfsrv_setacldsdorpc\n"); nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); /* * Use a stateid where other is an alternating 01010 pattern and * seqid is 0xffffffff. This value is not defined as special by * the RFC and is used by the FreeBSD NFS server to indicate an * MDS->DS proxy operation. */ st.other[0] = 0x55555555; st.other[1] = 0x55555555; st.other[2] = 0x55555555; st.seqid = 0xffffffff; nfscl_reqstart(nd, NFSPROC_SETACL, nmp, (u_int8_t *)fhp, sizeof(*fhp), NULL, NULL, 0, 0); nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); NFSZERO_ATTRBIT(&attrbits); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_ACL); /* * The "vp" argument to nfsv4_fillattr() is only used for vnode_type(), * so passing in the metadata "vp" will be ok, since it is of * the same type (VREG). */ nfsv4_fillattr(nd, NULL, vp, aclp, NULL, NULL, 0, &attrbits, NULL, NULL, 0, 0, 0, 0, 0, NULL); error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) { free(nd, M_TEMP); return (error); } NFSD_DEBUG(4, "nfsrv_setacldsdorpc: aft setaclrpc=%d\n", nd->nd_repstat); error = nd->nd_repstat; m_freem(nd->nd_mrep); free(nd, M_TEMP); return (error); } struct nfsrvsetacldsdorpc { int done; int inprog; struct task tsk; fhandle_t fh; struct nfsmount *nmp; struct vnode *vp; struct ucred *cred; NFSPROC_T *p; struct acl *aclp; int err; }; /* * Start up the thread that will execute nfsrv_setacldsdorpc(). */ static void start_setacldsdorpc(void *arg, int pending) { struct nfsrvsetacldsdorpc *drpc; drpc = (struct nfsrvsetacldsdorpc *)arg; drpc->err = nfsrv_setacldsdorpc(&drpc->fh, drpc->cred, drpc->p, drpc->vp, drpc->nmp, drpc->aclp); drpc->done = 1; } static int nfsrv_setacldsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct acl *aclp, int *failposp) { struct nfsrvsetacldsdorpc *drpc, *tdrpc = NULL; int error, i, ret, timo; NFSD_DEBUG(4, "in nfsrv_setacldsrpc\n"); drpc = NULL; if (mirrorcnt > 1) tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, M_WAITOK); /* * Do the setattr RPC for every DS, using a separate kernel process * for every DS except the last one. */ error = 0; for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { tdrpc->done = 0; tdrpc->inprog = 0; NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); tdrpc->nmp = *nmpp; tdrpc->vp = vp; tdrpc->cred = cred; tdrpc->p = p; tdrpc->aclp = aclp; tdrpc->err = 0; ret = EIO; if (nfs_pnfsiothreads != 0) { ret = nfs_pnfsio(start_setacldsdorpc, tdrpc); NFSD_DEBUG(4, "nfsrv_setacldsrpc: nfs_pnfsio=%d\n", ret); } if (ret != 0) { ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, aclp); if (nfsds_failerr(ret) && *failposp == -1) *failposp = i; else if (error == 0 && ret != 0) error = ret; } nmpp++; fhp++; } ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, aclp); if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) *failposp = mirrorcnt - 1; else if (error == 0 && ret != 0) error = ret; NFSD_DEBUG(4, "nfsrv_setacldsrpc: aft setextat=%d\n", error); tdrpc = drpc; timo = hz / 50; /* Wait for 20msec. */ if (timo < 1) timo = 1; for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { /* Wait for RPCs on separate threads to complete. */ while (tdrpc->inprog != 0 && tdrpc->done == 0) tsleep(&tdrpc->tsk, PVFS, "srvacds", timo); if (nfsds_failerr(tdrpc->err) && *failposp == -1) *failposp = i; else if (error == 0 && tdrpc->err != 0) error = tdrpc->err; } free(drpc, M_TEMP); return (error); } /* * Getattr call to the DS for the attributes that change due to writing. */ static int nfsrv_getattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap) { struct nfsrv_descript *nd; int error; nfsattrbit_t attrbits; NFSD_DEBUG(4, "in nfsrv_getattrdsrpc\n"); nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); nfscl_reqstart(nd, NFSPROC_GETATTR, nmp, (u_int8_t *)fhp, sizeof(fhandle_t), NULL, NULL, 0, 0); NFSZERO_ATTRBIT(&attrbits); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); (void) nfsrv_putattrbit(nd, &attrbits); error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) { free(nd, M_TEMP); return (error); } NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft getattrrpc=%d\n", nd->nd_repstat); if (nd->nd_repstat == 0) { error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); /* * We can only save the updated values in the extended * attribute if the vp is exclusively locked. * This should happen when any of the following operations * occur on the vnode: * Close, Delegreturn, LayoutCommit, LayoutReturn * As such, the updated extended attribute should get saved * before nfsrv_checkdsattr() returns 0 and allows the cached * attributes to be returned without calling this function. */ if (error == 0 && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) { error = nfsrv_setextattr(vp, nap, p); NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft setextat=%d\n", error); } } else error = nd->nd_repstat; m_freem(nd->nd_mrep); free(nd, M_TEMP); NFSD_DEBUG(4, "nfsrv_getattrdsrpc error=%d\n", error); return (error); } /* * Seek call to a DS. */ static int nfsrv_seekdsrpc(fhandle_t *fhp, off_t *offp, int content, bool *eofp, struct ucred *cred, NFSPROC_T *p, struct nfsmount *nmp) { uint32_t *tl; struct nfsrv_descript *nd; nfsv4stateid_t st; int error; NFSD_DEBUG(4, "in nfsrv_seekdsrpc\n"); /* * Use a stateid where other is an alternating 01010 pattern and * seqid is 0xffffffff. This value is not defined as special by * the RFC and is used by the FreeBSD NFS server to indicate an * MDS->DS proxy operation. */ st.other[0] = 0x55555555; st.other[1] = 0x55555555; st.other[2] = 0x55555555; st.seqid = 0xffffffff; nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); nfscl_reqstart(nd, NFSPROC_SEEKDS, nmp, (u_int8_t *)fhp, sizeof(fhandle_t), NULL, NULL, 0, 0); nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + NFSX_UNSIGNED); txdr_hyper(*offp, tl); tl += 2; *tl = txdr_unsigned(content); error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) { free(nd, M_TEMP); return (error); } NFSD_DEBUG(4, "nfsrv_seekdsrpc: aft seekrpc=%d\n", nd->nd_repstat); if (nd->nd_repstat == 0) { NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED + NFSX_HYPER); if (*tl++ == newnfs_true) *eofp = true; else *eofp = false; *offp = fxdr_hyper(tl); } else error = nd->nd_repstat; nfsmout: m_freem(nd->nd_mrep); free(nd, M_TEMP); NFSD_DEBUG(4, "nfsrv_seekdsrpc error=%d\n", error); return (error); } /* * Get the device id and file handle for a DS file. */ int nfsrv_dsgetdevandfh(struct vnode *vp, NFSPROC_T *p, int *mirrorcntp, fhandle_t *fhp, char *devid) { int buflen, error; char *buf; buflen = 1024; buf = malloc(buflen, M_TEMP, M_WAITOK); error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, NULL, fhp, devid, NULL, NULL, NULL, NULL, NULL, NULL); free(buf, M_TEMP); return (error); } /* * Do a Lookup against the DS for the filename. */ static int nfsrv_pnfslookupds(struct vnode *vp, struct vnode *dvp, struct pnfsdsfile *pf, struct vnode **nvpp, NFSPROC_T *p) { struct nameidata named; struct ucred *tcred; char *bufp; u_long *hashp; struct vnode *nvp; int error; tcred = newnfs_getcred(); named.ni_cnd.cn_nameiop = LOOKUP; named.ni_cnd.cn_lkflags = LK_SHARED | LK_RETRY; named.ni_cnd.cn_cred = tcred; - named.ni_cnd.cn_thread = p; named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME; nfsvno_setpathbuf(&named, &bufp, &hashp); named.ni_cnd.cn_nameptr = bufp; named.ni_cnd.cn_namelen = strlen(pf->dsf_filename); strlcpy(bufp, pf->dsf_filename, NAME_MAX); NFSD_DEBUG(4, "nfsrv_pnfslookupds: filename=%s\n", bufp); error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd); NFSD_DEBUG(4, "nfsrv_pnfslookupds: aft LOOKUP=%d\n", error); NFSFREECRED(tcred); nfsvno_relpathbuf(&named); if (error == 0) *nvpp = nvp; NFSD_DEBUG(4, "eo nfsrv_pnfslookupds=%d\n", error); return (error); } /* * Set the file handle to the correct one. */ static void nfsrv_pnfssetfh(struct vnode *vp, struct pnfsdsfile *pf, char *devid, char *fnamep, struct vnode *nvp, NFSPROC_T *p) { struct nfsnode *np; int ret = 0; np = VTONFS(nvp); NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, NFSX_MYFH); /* * We can only do a vn_set_extattr() if the vnode is exclusively * locked and vn_start_write() has been done. If devid != NULL or * fnamep != NULL or the vnode is shared locked, vn_start_write() * may not have been done. * If not done now, it will be done on a future call. */ if (devid == NULL && fnamep == NULL && NFSVOPISLOCKED(vp) == LK_EXCLUSIVE) ret = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*pf), (char *)pf, p); NFSD_DEBUG(4, "eo nfsrv_pnfssetfh=%d\n", ret); } /* * Cause RPCs waiting on "nmp" to fail. This is called for a DS mount point * when the DS has failed. */ void nfsrv_killrpcs(struct nfsmount *nmp) { /* * Call newnfs_nmcancelreqs() to cause * any RPCs in progress on the mount point to * fail. * This will cause any process waiting for an * RPC to complete while holding a vnode lock * on the mounted-on vnode (such as "df" or * a non-forced "umount") to fail. * This will unlock the mounted-on vnode so * a forced dismount can succeed. * The NFSMNTP_CANCELRPCS flag should be set when this function is * called. */ newnfs_nmcancelreqs(nmp); } /* * Sum up the statfs info for each of the DSs, so that the client will * receive the total for all DSs. */ static int nfsrv_pnfsstatfs(struct statfs *sf, struct mount *mp) { struct statfs *tsf; struct nfsdevice *ds; struct vnode **dvpp, **tdvpp, *dvp; uint64_t tot; int cnt, error = 0, i; if (nfsrv_devidcnt <= 0) return (ENXIO); dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK); tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK); /* Get an array of the dvps for the DSs. */ tdvpp = dvpp; i = 0; NFSDDSLOCK(); /* First, search for matches for same file system. */ TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { if (ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 && fsidcmp(&ds->nfsdev_mdsfsid, &mp->mnt_stat.f_fsid) == 0) { if (++i > nfsrv_devidcnt) break; *tdvpp++ = ds->nfsdev_dvp; } } /* * If no matches for same file system, total all servers not assigned * to a file system. */ if (i == 0) { TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { if (ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset == 0) { if (++i > nfsrv_devidcnt) break; *tdvpp++ = ds->nfsdev_dvp; } } } NFSDDSUNLOCK(); cnt = i; /* Do a VFS_STATFS() for each of the DSs and sum them up. */ tdvpp = dvpp; for (i = 0; i < cnt && error == 0; i++) { dvp = *tdvpp++; error = VFS_STATFS(dvp->v_mount, tsf); if (error == 0) { if (sf->f_bsize == 0) { if (tsf->f_bsize > 0) sf->f_bsize = tsf->f_bsize; else sf->f_bsize = 8192; } if (tsf->f_blocks > 0) { if (sf->f_bsize != tsf->f_bsize) { tot = tsf->f_blocks * tsf->f_bsize; sf->f_blocks += (tot / sf->f_bsize); } else sf->f_blocks += tsf->f_blocks; } if (tsf->f_bfree > 0) { if (sf->f_bsize != tsf->f_bsize) { tot = tsf->f_bfree * tsf->f_bsize; sf->f_bfree += (tot / sf->f_bsize); } else sf->f_bfree += tsf->f_bfree; } if (tsf->f_bavail > 0) { if (sf->f_bsize != tsf->f_bsize) { tot = tsf->f_bavail * tsf->f_bsize; sf->f_bavail += (tot / sf->f_bsize); } else sf->f_bavail += tsf->f_bavail; } } } free(tsf, M_TEMP); free(dvpp, M_TEMP); return (error); } /* * Set an NFSv4 acl. */ int nfsrv_setacl(struct vnode *vp, NFSACL_T *aclp, struct ucred *cred, NFSPROC_T *p) { int error; if (nfsrv_useacl == 0 || nfs_supportsnfsv4acls(vp) == 0) { error = NFSERR_ATTRNOTSUPP; goto out; } /* * With NFSv4 ACLs, chmod(2) may need to add additional entries. * Make sure it has enough room for that - splitting every entry * into two and appending "canonical six" entries at the end. * Cribbed out of kern/vfs_acl.c - Rick M. */ if (aclp->acl_cnt > (ACL_MAX_ENTRIES - 6) / 2) { error = NFSERR_ATTRNOTSUPP; goto out; } error = VOP_SETACL(vp, ACL_TYPE_NFS4, aclp, cred, p); if (error == 0) { error = nfsrv_dssetacl(vp, aclp, cred, p); if (error == ENOENT) error = 0; } out: NFSEXITCODE(error); return (error); } /* * Seek vnode op call (actually it is a VOP_IOCTL()). * This function is called with the vnode locked, but unlocks and vrele()s * the vp before returning. */ int nfsvno_seek(struct nfsrv_descript *nd, struct vnode *vp, u_long cmd, off_t *offp, int content, bool *eofp, struct ucred *cred, NFSPROC_T *p) { struct nfsvattr at; int error, ret; ASSERT_VOP_LOCKED(vp, "nfsvno_seek vp"); /* * Attempt to seek on a DS file. A return of ENOENT implies * there is no DS file to seek on. */ error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SEEKDS, NULL, NULL, NULL, NULL, NULL, offp, content, eofp); if (error != ENOENT) { vput(vp); return (error); } /* * Do the VOP_IOCTL() call. For the case where *offp == file_size, * VOP_IOCTL() will return ENXIO. However, the correct reply for * NFSv4.2 is *eofp == true and error == 0 for this case. */ NFSVOPUNLOCK(vp); error = VOP_IOCTL(vp, cmd, offp, 0, cred, p); *eofp = false; if (error == ENXIO || (error == 0 && cmd == FIOSEEKHOLE)) { /* Handle the cases where we might be at EOF. */ ret = nfsvno_getattr(vp, &at, nd, p, 0, NULL); if (ret == 0 && *offp == at.na_size) { *eofp = true; error = 0; } if (ret != 0 && error == 0) error = ret; } vrele(vp); NFSEXITCODE(error); return (error); } /* * Allocate vnode op call. */ int nfsvno_allocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred, NFSPROC_T *p) { int error; off_t olen; ASSERT_VOP_ELOCKED(vp, "nfsvno_allocate vp"); /* * Attempt to allocate on a DS file. A return of ENOENT implies * there is no DS file to allocate on. */ error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_ALLOCATE, NULL, NULL, NULL, NULL, NULL, &len, 0, NULL); if (error != ENOENT) return (error); /* * Do the actual VOP_ALLOCATE(), looping so long as * progress is being made, to achieve completion. */ do { olen = len; error = VOP_ALLOCATE(vp, &off, &len); if (error == 0 && len > 0 && olen > len) maybe_yield(); } while (error == 0 && len > 0 && olen > len); if (error == 0 && len > 0) error = NFSERR_IO; NFSEXITCODE(error); return (error); } /* * Deallocate vnode op call. */ int nfsvno_deallocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred, NFSPROC_T *p) { int error; off_t olen; ASSERT_VOP_ELOCKED(vp, "nfsvno_deallocate vp"); /* * Attempt to deallocate on a DS file. A return of ENOENT implies * there is no DS file to deallocate on. */ error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_DEALLOCATE, NULL, NULL, NULL, NULL, NULL, &len, 0, NULL); if (error != ENOENT) return (error); /* * Do the actual VOP_DEALLOCATE(), looping so long as * progress is being made, to achieve completion. */ do { olen = len; error = VOP_DEALLOCATE(vp, &off, &len, 0, IO_SYNC, cred); if (error == 0 && len > 0 && olen > len) maybe_yield(); } while (error == 0 && len > 0 && olen > len); if (error == 0 && len > 0) error = NFSERR_IO; NFSEXITCODE(error); return (error); } /* * Get Extended Atribute vnode op into an mbuf list. */ int nfsvno_getxattr(struct vnode *vp, char *name, uint32_t maxresp, struct ucred *cred, uint64_t flag, int maxextsiz, struct thread *p, struct mbuf **mpp, struct mbuf **mpendp, int *lenp) { struct iovec *iv; struct uio io, *uiop = &io; struct mbuf *m, *m2; int alen, error, len, tlen; size_t siz; /* First, find out the size of the extended attribute. */ error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL, &siz, cred, p); if (error != 0) return (NFSERR_NOXATTR); if (siz > maxresp - NFS_MAXXDR) return (NFSERR_XATTR2BIG); len = siz; tlen = NFSM_RNDUP(len); if (tlen > 0) { /* * If cnt > MCLBYTES and the reply will not be saved, use * ext_pgs mbufs for TLS. * For NFSv4.0, we do not know for sure if the reply will * be saved, so do not use ext_pgs mbufs for NFSv4.0. * Always use ext_pgs mbufs if ND_EXTPG is set. */ if ((flag & ND_EXTPG) != 0 || (tlen > MCLBYTES && (flag & (ND_TLS | ND_SAVEREPLY)) == ND_TLS && (flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)) uiop->uio_iovcnt = nfsrv_createiovec_extpgs(tlen, maxextsiz, &m, &m2, &iv); else uiop->uio_iovcnt = nfsrv_createiovec(tlen, &m, &m2, &iv); uiop->uio_iov = iv; } else { uiop->uio_iovcnt = 0; uiop->uio_iov = iv = NULL; m = m2 = NULL; } uiop->uio_offset = 0; uiop->uio_resid = tlen; uiop->uio_rw = UIO_READ; uiop->uio_segflg = UIO_SYSSPACE; uiop->uio_td = p; #ifdef MAC error = mac_vnode_check_getextattr(cred, vp, EXTATTR_NAMESPACE_USER, name); if (error != 0) goto out; #endif if (tlen > 0) error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop, NULL, cred, p); if (error != 0) goto out; if (uiop->uio_resid > 0) { alen = tlen; len = tlen - uiop->uio_resid; tlen = NFSM_RNDUP(len); if (alen != tlen) printf("nfsvno_getxattr: weird size read\n"); if (tlen == 0) { m_freem(m); m = m2 = NULL; } else if (alen != tlen || tlen != len) m2 = nfsrv_adj(m, alen - tlen, tlen - len); } *lenp = len; *mpp = m; *mpendp = m2; out: if (error != 0) { if (m != NULL) m_freem(m); *lenp = 0; } free(iv, M_TEMP); NFSEXITCODE(error); return (error); } /* * Set Extended attribute vnode op from an mbuf list. */ int nfsvno_setxattr(struct vnode *vp, char *name, int len, struct mbuf *m, char *cp, struct ucred *cred, struct thread *p) { struct iovec *iv; struct uio uio, *uiop = &uio; int cnt, error; error = 0; #ifdef MAC error = mac_vnode_check_setextattr(cred, vp, EXTATTR_NAMESPACE_USER, name); #endif if (error != 0) goto out; uiop->uio_rw = UIO_WRITE; uiop->uio_segflg = UIO_SYSSPACE; uiop->uio_td = p; uiop->uio_offset = 0; uiop->uio_resid = len; if (len > 0) { error = nfsrv_createiovecw(len, m, cp, &iv, &cnt); uiop->uio_iov = iv; uiop->uio_iovcnt = cnt; } else { uiop->uio_iov = iv = NULL; uiop->uio_iovcnt = 0; } if (error == 0) { error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop, cred, p); free(iv, M_TEMP); } out: NFSEXITCODE(error); return (error); } /* * Remove Extended attribute vnode op. */ int nfsvno_rmxattr(struct nfsrv_descript *nd, struct vnode *vp, char *name, struct ucred *cred, struct thread *p) { int error; /* * Get rid of any delegations. I am not sure why this is required, * but RFC-8276 says so. */ error = nfsrv_checkremove(vp, 0, nd, nd->nd_clientid, p); if (error != 0) goto out; #ifdef MAC error = mac_vnode_check_deleteextattr(cred, vp, EXTATTR_NAMESPACE_USER, name); if (error != 0) goto out; #endif error = VOP_DELETEEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, cred, p); if (error == EOPNOTSUPP) error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL, cred, p); out: NFSEXITCODE(error); return (error); } /* * List Extended Atribute vnode op into an mbuf list. */ int nfsvno_listxattr(struct vnode *vp, uint64_t cookie, struct ucred *cred, struct thread *p, u_char **bufp, uint32_t *lenp, bool *eofp) { struct iovec iv; struct uio io; int error; size_t siz; *bufp = NULL; /* First, find out the size of the extended attribute. */ error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, NULL, &siz, cred, p); if (error != 0) return (NFSERR_NOXATTR); if (siz <= cookie) { *lenp = 0; *eofp = true; goto out; } if (siz > cookie + *lenp) { siz = cookie + *lenp; *eofp = false; } else *eofp = true; /* Just choose a sanity limit of 10Mbytes for malloc(M_TEMP). */ if (siz > 10 * 1024 * 1024) { error = NFSERR_XATTR2BIG; goto out; } *bufp = malloc(siz, M_TEMP, M_WAITOK); iv.iov_base = *bufp; iv.iov_len = siz; io.uio_iovcnt = 1; io.uio_iov = &iv; io.uio_offset = 0; io.uio_resid = siz; io.uio_rw = UIO_READ; io.uio_segflg = UIO_SYSSPACE; io.uio_td = p; #ifdef MAC error = mac_vnode_check_listextattr(cred, vp, EXTATTR_NAMESPACE_USER); if (error != 0) goto out; #endif error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, &io, NULL, cred, p); if (error != 0) goto out; if (io.uio_resid > 0) siz -= io.uio_resid; *lenp = siz; out: if (error != 0) { free(*bufp, M_TEMP); *bufp = NULL; } NFSEXITCODE(error); return (error); } /* * Trim trailing data off the mbuf list being built. */ void nfsm_trimtrailing(struct nfsrv_descript *nd, struct mbuf *mb, char *bpos, int bextpg, int bextpgsiz) { vm_page_t pg; int fullpgsiz, i; if (mb->m_next != NULL) { m_freem(mb->m_next); mb->m_next = NULL; } if ((mb->m_flags & M_EXTPG) != 0) { KASSERT(bextpg >= 0 && bextpg < mb->m_epg_npgs, ("nfsm_trimtrailing: bextpg out of range")); KASSERT(bpos == (char *)(void *) PHYS_TO_DMAP(mb->m_epg_pa[bextpg]) + PAGE_SIZE - bextpgsiz, ("nfsm_trimtrailing: bextpgsiz bad!")); /* First, get rid of any pages after this position. */ for (i = mb->m_epg_npgs - 1; i > bextpg; i--) { pg = PHYS_TO_VM_PAGE(mb->m_epg_pa[i]); vm_page_unwire_noq(pg); vm_page_free(pg); } mb->m_epg_npgs = bextpg + 1; if (bextpg == 0) fullpgsiz = PAGE_SIZE - mb->m_epg_1st_off; else fullpgsiz = PAGE_SIZE; mb->m_epg_last_len = fullpgsiz - bextpgsiz; mb->m_len = m_epg_pagelen(mb, 0, mb->m_epg_1st_off); for (i = 1; i < mb->m_epg_npgs; i++) mb->m_len += m_epg_pagelen(mb, i, 0); nd->nd_bextpgsiz = bextpgsiz; nd->nd_bextpg = bextpg; } else mb->m_len = bpos - mtod(mb, char *); nd->nd_mb = mb; nd->nd_bpos = bpos; } /* * Check to see if a put file handle operation should test for * NFSERR_WRONGSEC, although NFSv3 actually returns NFSERR_AUTHERR. * When Open is the next operation, NFSERR_WRONGSEC cannot be * replied for the Open cases that use a component. Thia can * be identified by the fact that the file handle's type is VDIR. */ bool nfsrv_checkwrongsec(struct nfsrv_descript *nd, int nextop, enum vtype vtyp) { if ((nd->nd_flag & ND_NFSV4) == 0) { if (nd->nd_procnum == NFSPROC_FSINFO) return (false); return (true); } if ((nd->nd_flag & ND_LASTOP) != 0) return (false); if (nextop == NFSV4OP_PUTROOTFH || nextop == NFSV4OP_PUTFH || nextop == NFSV4OP_PUTPUBFH || nextop == NFSV4OP_RESTOREFH || nextop == NFSV4OP_LOOKUP || nextop == NFSV4OP_LOOKUPP || nextop == NFSV4OP_SECINFO || nextop == NFSV4OP_SECINFONONAME) return (false); if (nextop == NFSV4OP_OPEN && vtyp == VDIR) return (false); return (true); } extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *); /* * Called once to initialize data structures... */ static int nfsd_modevent(module_t mod, int type, void *data) { int error = 0, i; static int loaded = 0; switch (type) { case MOD_LOAD: if (loaded) goto out; newnfs_portinit(); for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) { mtx_init(&nfsrchash_table[i].mtx, "nfsrtc", NULL, MTX_DEF); mtx_init(&nfsrcahash_table[i].mtx, "nfsrtca", NULL, MTX_DEF); } mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF); mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF); mtx_init(&nfsv4root_mnt.mnt_mtx, "nfs4mnt", NULL, MTX_DEF); mtx_init(&nfsrv_dontlistlock_mtx, "nfs4dnl", NULL, MTX_DEF); mtx_init(&nfsrv_recalllock_mtx, "nfs4rec", NULL, MTX_DEF); lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0); nfsrvd_initcache(); nfsd_init(); NFSD_LOCK(); nfsrvd_init(0); NFSD_UNLOCK(); nfsd_mntinit(); #ifdef VV_DISABLEDELEG vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation; vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation; #endif nfsd_call_servertimer = nfsrv_servertimer; nfsd_call_nfsd = nfssvc_nfsd; loaded = 1; break; case MOD_UNLOAD: if (newnfs_numnfsd != 0) { error = EBUSY; break; } #ifdef VV_DISABLEDELEG vn_deleg_ops.vndeleg_recall = NULL; vn_deleg_ops.vndeleg_disable = NULL; #endif nfsd_call_servertimer = NULL; nfsd_call_nfsd = NULL; /* Clean out all NFSv4 state. */ nfsrv_throwawayallstate(curthread); /* Clean the NFS server reply cache */ nfsrvd_cleancache(); /* Free up the krpc server pool. */ if (nfsrvd_pool != NULL) svcpool_destroy(nfsrvd_pool); /* and get rid of the locks */ for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) { mtx_destroy(&nfsrchash_table[i].mtx); mtx_destroy(&nfsrcahash_table[i].mtx); } mtx_destroy(&nfsrc_udpmtx); mtx_destroy(&nfs_v4root_mutex); mtx_destroy(&nfsv4root_mnt.mnt_mtx); mtx_destroy(&nfsrv_dontlistlock_mtx); mtx_destroy(&nfsrv_recalllock_mtx); for (i = 0; i < nfsrv_sessionhashsize; i++) mtx_destroy(&nfssessionhash[i].mtx); if (nfslayouthash != NULL) { for (i = 0; i < nfsrv_layouthashsize; i++) mtx_destroy(&nfslayouthash[i].mtx); free(nfslayouthash, M_NFSDSESSION); } lockdestroy(&nfsv4root_mnt.mnt_explock); free(nfsclienthash, M_NFSDCLIENT); free(nfslockhash, M_NFSDLOCKFILE); free(nfssessionhash, M_NFSDSESSION); loaded = 0; break; default: error = EOPNOTSUPP; break; } out: NFSEXITCODE(error); return (error); } static moduledata_t nfsd_mod = { "nfsd", nfsd_modevent, NULL, }; DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY); /* So that loader and kldload(2) can find us, wherever we are.. */ MODULE_VERSION(nfsd, 1); MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1); MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1); MODULE_DEPEND(nfsd, krpc, 1, 1, 1); MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1); diff --git a/sys/fs/smbfs/smbfs_vnops.c b/sys/fs/smbfs/smbfs_vnops.c index ad04bb93962a..ac32f13977db 100644 --- a/sys/fs/smbfs/smbfs_vnops.c +++ b/sys/fs/smbfs/smbfs_vnops.c @@ -1,1382 +1,1382 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2000-2001 Boris Popov * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * Prototypes for SMBFS vnode operations */ static vop_create_t smbfs_create; static vop_mknod_t smbfs_mknod; static vop_open_t smbfs_open; static vop_close_t smbfs_close; static vop_access_t smbfs_access; static vop_getattr_t smbfs_getattr; static vop_setattr_t smbfs_setattr; static vop_read_t smbfs_read; static vop_write_t smbfs_write; static vop_fsync_t smbfs_fsync; static vop_remove_t smbfs_remove; static vop_link_t smbfs_link; static vop_lookup_t smbfs_lookup; static vop_rename_t smbfs_rename; static vop_mkdir_t smbfs_mkdir; static vop_rmdir_t smbfs_rmdir; static vop_symlink_t smbfs_symlink; static vop_readdir_t smbfs_readdir; static vop_strategy_t smbfs_strategy; static vop_print_t smbfs_print; static vop_pathconf_t smbfs_pathconf; static vop_advlock_t smbfs_advlock; static vop_getextattr_t smbfs_getextattr; struct vop_vector smbfs_vnodeops = { .vop_default = &default_vnodeops, .vop_access = smbfs_access, .vop_advlock = smbfs_advlock, .vop_close = smbfs_close, .vop_create = smbfs_create, .vop_fsync = smbfs_fsync, .vop_getattr = smbfs_getattr, .vop_getextattr = smbfs_getextattr, .vop_getpages = smbfs_getpages, .vop_inactive = smbfs_inactive, .vop_ioctl = smbfs_ioctl, .vop_link = smbfs_link, .vop_lookup = smbfs_lookup, .vop_mkdir = smbfs_mkdir, .vop_mknod = smbfs_mknod, .vop_open = smbfs_open, .vop_pathconf = smbfs_pathconf, .vop_print = smbfs_print, .vop_putpages = smbfs_putpages, .vop_read = smbfs_read, .vop_readdir = smbfs_readdir, .vop_reclaim = smbfs_reclaim, .vop_remove = smbfs_remove, .vop_rename = smbfs_rename, .vop_rmdir = smbfs_rmdir, .vop_setattr = smbfs_setattr, /* .vop_setextattr = smbfs_setextattr,*/ .vop_strategy = smbfs_strategy, .vop_symlink = smbfs_symlink, .vop_write = smbfs_write, }; VFS_VOP_VECTOR_REGISTER(smbfs_vnodeops); static int smbfs_access(ap) struct vop_access_args /* { struct vnode *a_vp; accmode_t a_accmode; struct ucred *a_cred; struct thread *a_td; } */ *ap; { struct vnode *vp = ap->a_vp; accmode_t accmode = ap->a_accmode; mode_t mpmode; struct smbmount *smp = VTOSMBFS(vp); SMBVDEBUG("\n"); if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { switch (vp->v_type) { case VREG: case VDIR: case VLNK: return EROFS; default: break; } } mpmode = vp->v_type == VREG ? smp->sm_file_mode : smp->sm_dir_mode; return (vaccess(vp->v_type, mpmode, smp->sm_uid, smp->sm_gid, ap->a_accmode, ap->a_cred)); } /* ARGSUSED */ static int smbfs_open(ap) struct vop_open_args /* { struct vnode *a_vp; int a_mode; struct ucred *a_cred; struct thread *a_td; } */ *ap; { struct vnode *vp = ap->a_vp; struct smbnode *np = VTOSMB(vp); struct smb_cred *scred; struct vattr vattr; int mode = ap->a_mode; int error, accmode; SMBVDEBUG("%s,%d\n", np->n_name, (np->n_flag & NOPEN) != 0); if (vp->v_type != VREG && vp->v_type != VDIR) { SMBFSERR("open eacces vtype=%d\n", vp->v_type); return EACCES; } if (vp->v_type == VDIR) { np->n_flag |= NOPEN; return 0; } if (np->n_flag & NMODIFIED) { if ((error = smbfs_vinvalbuf(vp, ap->a_td)) == EINTR) return error; smbfs_attr_cacheremove(vp); error = VOP_GETATTR(vp, &vattr, ap->a_cred); if (error) return error; np->n_mtime.tv_sec = vattr.va_mtime.tv_sec; } else { error = VOP_GETATTR(vp, &vattr, ap->a_cred); if (error) return error; if (np->n_mtime.tv_sec != vattr.va_mtime.tv_sec) { error = smbfs_vinvalbuf(vp, ap->a_td); if (error == EINTR) return error; np->n_mtime.tv_sec = vattr.va_mtime.tv_sec; } } if ((np->n_flag & NOPEN) != 0) return 0; /* * Use DENYNONE to give unixy semantics of permitting * everything not forbidden by permissions. Ie denial * is up to server with clients/openers needing to use * advisory locks for further control. */ accmode = SMB_SM_DENYNONE|SMB_AM_OPENREAD; if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) accmode = SMB_SM_DENYNONE|SMB_AM_OPENRW; scred = smbfs_malloc_scred(); smb_makescred(scred, ap->a_td, ap->a_cred); error = smbfs_smb_open(np, accmode, scred); if (error) { if (mode & FWRITE) return EACCES; else if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { accmode = SMB_SM_DENYNONE|SMB_AM_OPENREAD; error = smbfs_smb_open(np, accmode, scred); } } if (error == 0) { np->n_flag |= NOPEN; vnode_create_vobject(ap->a_vp, vattr.va_size, ap->a_td); } smbfs_attr_cacheremove(vp); smbfs_free_scred(scred); return error; } static int smbfs_close(ap) struct vop_close_args /* { struct vnodeop_desc *a_desc; struct vnode *a_vp; int a_fflag; struct ucred *a_cred; struct thread *a_td; } */ *ap; { struct vnode *vp = ap->a_vp; struct thread *td = ap->a_td; struct smbnode *np = VTOSMB(vp); struct smb_cred *scred; if (vp->v_type == VDIR && (np->n_flag & NOPEN) != 0 && np->n_dirseq != NULL) { scred = smbfs_malloc_scred(); smb_makescred(scred, td, ap->a_cred); smbfs_findclose(np->n_dirseq, scred); smbfs_free_scred(scred); np->n_dirseq = NULL; } return 0; } /* * smbfs_getattr call from vfs. */ static int smbfs_getattr(ap) struct vop_getattr_args /* { struct vnode *a_vp; struct vattr *a_vap; struct ucred *a_cred; } */ *ap; { struct vnode *vp = ap->a_vp; struct smbnode *np = VTOSMB(vp); struct vattr *va=ap->a_vap; struct smbfattr fattr; struct smb_cred *scred; u_quad_t oldsize; int error; SMBVDEBUG("%lx: '%s' %d\n", (long)vp, np->n_name, (vp->v_vflag & VV_ROOT) != 0); error = smbfs_attr_cachelookup(vp, va); if (!error) return 0; SMBVDEBUG("not in the cache\n"); scred = smbfs_malloc_scred(); smb_makescred(scred, curthread, ap->a_cred); oldsize = np->n_size; error = smbfs_smb_lookup(np, NULL, 0, &fattr, scred); if (error) { SMBVDEBUG("error %d\n", error); smbfs_free_scred(scred); return error; } smbfs_attr_cacheenter(vp, &fattr); smbfs_attr_cachelookup(vp, va); if (np->n_flag & NOPEN) np->n_size = oldsize; smbfs_free_scred(scred); return 0; } static int smbfs_setattr(ap) struct vop_setattr_args /* { struct vnode *a_vp; struct vattr *a_vap; struct ucred *a_cred; } */ *ap; { struct vnode *vp = ap->a_vp; struct smbnode *np = VTOSMB(vp); struct vattr *vap = ap->a_vap; struct timespec *mtime, *atime; struct smb_cred *scred; struct smb_share *ssp = np->n_mount->sm_share; struct smb_vc *vcp = SSTOVC(ssp); struct thread *td = curthread; u_quad_t tsize = 0; int isreadonly, doclose, error = 0; int old_n_dosattr; SMBVDEBUG("\n"); isreadonly = (vp->v_mount->mnt_flag & MNT_RDONLY); /* * Disallow write attempts if the filesystem is mounted read-only. */ if ((vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL || vap->va_flags != VNOVAL) && isreadonly) return EROFS; /* * We only support setting four flags. Don't allow setting others. * * We map UF_READONLY to SMB_FA_RDONLY, unlike the MacOS X version * of this code, which maps both UF_IMMUTABLE AND SF_IMMUTABLE to * SMB_FA_RDONLY. The immutable flags have different semantics * than readonly, which is the reason for the difference. */ if (vap->va_flags != VNOVAL) { if (vap->va_flags & ~(UF_HIDDEN|UF_SYSTEM|UF_ARCHIVE| UF_READONLY)) return EINVAL; } scred = smbfs_malloc_scred(); smb_makescred(scred, td, ap->a_cred); if (vap->va_size != VNOVAL) { switch (vp->v_type) { case VDIR: error = EISDIR; goto out; case VREG: break; default: error = EINVAL; goto out; } if (isreadonly) { error = EROFS; goto out; } doclose = 0; vnode_pager_setsize(vp, (u_long)vap->va_size); tsize = np->n_size; np->n_size = vap->va_size; if ((np->n_flag & NOPEN) == 0) { error = smbfs_smb_open(np, SMB_SM_DENYNONE|SMB_AM_OPENRW, scred); if (error == 0) doclose = 1; } if (error == 0) error = smbfs_smb_setfsize(np, (int64_t)vap->va_size, scred); if (doclose) smbfs_smb_close(ssp, np->n_fid, NULL, scred); if (error) { np->n_size = tsize; vnode_pager_setsize(vp, (u_long)tsize); goto out; } } if ((vap->va_flags != VNOVAL) || (vap->va_mode != (mode_t)VNOVAL)) { old_n_dosattr = np->n_dosattr; if (vap->va_mode != (mode_t)VNOVAL) { if (vap->va_mode & S_IWUSR) np->n_dosattr &= ~SMB_FA_RDONLY; else np->n_dosattr |= SMB_FA_RDONLY; } if (vap->va_flags != VNOVAL) { if (vap->va_flags & UF_HIDDEN) np->n_dosattr |= SMB_FA_HIDDEN; else np->n_dosattr &= ~SMB_FA_HIDDEN; if (vap->va_flags & UF_SYSTEM) np->n_dosattr |= SMB_FA_SYSTEM; else np->n_dosattr &= ~SMB_FA_SYSTEM; if (vap->va_flags & UF_ARCHIVE) np->n_dosattr |= SMB_FA_ARCHIVE; else np->n_dosattr &= ~SMB_FA_ARCHIVE; /* * We only support setting the immutable / readonly * bit for regular files. According to comments in * the MacOS X version of this code, supporting the * readonly bit on directories doesn't do the same * thing in Windows as in Unix. */ if (vp->v_type == VREG) { if (vap->va_flags & UF_READONLY) np->n_dosattr |= SMB_FA_RDONLY; else np->n_dosattr &= ~SMB_FA_RDONLY; } } if (np->n_dosattr != old_n_dosattr) { error = smbfs_smb_setpattr(np, np->n_dosattr, NULL, scred); if (error) goto out; } } mtime = atime = NULL; if (vap->va_mtime.tv_sec != VNOVAL) mtime = &vap->va_mtime; if (vap->va_atime.tv_sec != VNOVAL) atime = &vap->va_atime; if (mtime != atime) { if (vap->va_vaflags & VA_UTIMES_NULL) { error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td); if (error) error = VOP_ACCESS(vp, VWRITE, ap->a_cred, td); } else error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td); #if 0 if (mtime == NULL) mtime = &np->n_mtime; if (atime == NULL) atime = &np->n_atime; #endif /* * If file is opened, then we can use handle based calls. * If not, use path based ones. */ if ((np->n_flag & NOPEN) == 0) { if (vcp->vc_flags & SMBV_WIN95) { error = VOP_OPEN(vp, FWRITE, ap->a_cred, td, NULL); if (!error) { /* error = smbfs_smb_setfattrNT(np, 0, mtime, atime, scred); VOP_GETATTR(vp, &vattr, ap->a_cred); */ if (mtime) np->n_mtime = *mtime; VOP_CLOSE(vp, FWRITE, ap->a_cred, td); } } else if ((vcp->vc_sopt.sv_caps & SMB_CAP_NT_SMBS)) { error = smbfs_smb_setptime2(np, mtime, atime, 0, scred); /* error = smbfs_smb_setpattrNT(np, 0, mtime, atime, scred);*/ } else if (SMB_DIALECT(vcp) >= SMB_DIALECT_LANMAN2_0) { error = smbfs_smb_setptime2(np, mtime, atime, 0, scred); } else { error = smbfs_smb_setpattr(np, 0, mtime, scred); } } else { if (vcp->vc_sopt.sv_caps & SMB_CAP_NT_SMBS) { error = smbfs_smb_setfattrNT(np, 0, mtime, atime, scred); } else if (SMB_DIALECT(vcp) >= SMB_DIALECT_LANMAN1_0) { error = smbfs_smb_setftime(np, mtime, atime, scred); } else { /* * I have no idea how to handle this for core * level servers. The possible solution is to * update mtime after file is closed. */ SMBERROR("can't update times on an opened file\n"); } } } /* * Invalidate attribute cache in case if server doesn't set * required attributes. */ smbfs_attr_cacheremove(vp); /* invalidate cache */ VOP_GETATTR(vp, vap, ap->a_cred); np->n_mtime.tv_sec = vap->va_mtime.tv_sec; out: smbfs_free_scred(scred); return error; } /* * smbfs_read call. */ static int smbfs_read(ap) struct vop_read_args /* { struct vnode *a_vp; struct uio *a_uio; int a_ioflag; struct ucred *a_cred; } */ *ap; { struct vnode *vp = ap->a_vp; struct uio *uio = ap->a_uio; SMBVDEBUG("\n"); if (vp->v_type != VREG && vp->v_type != VDIR) return EPERM; return smbfs_readvnode(vp, uio, ap->a_cred); } static int smbfs_write(ap) struct vop_write_args /* { struct vnode *a_vp; struct uio *a_uio; int a_ioflag; struct ucred *a_cred; } */ *ap; { struct vnode *vp = ap->a_vp; struct uio *uio = ap->a_uio; SMBVDEBUG("%d,ofs=%jd,sz=%zd\n",vp->v_type, (intmax_t)uio->uio_offset, uio->uio_resid); if (vp->v_type != VREG) return (EPERM); return smbfs_writevnode(vp, uio, ap->a_cred,ap->a_ioflag); } /* * smbfs_create call * Create a regular file. On entry the directory to contain the file being * created is locked. We must release before we return. We must also free * the pathname buffer pointed at by cnp->cn_pnbuf, always on error, or * only if the SAVESTART bit in cn_flags is clear on success. */ static int smbfs_create(ap) struct vop_create_args /* { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; struct vattr *a_vap; } */ *ap; { struct vnode *dvp = ap->a_dvp; struct vattr *vap = ap->a_vap; struct vnode **vpp=ap->a_vpp; struct componentname *cnp = ap->a_cnp; struct smbnode *dnp = VTOSMB(dvp); struct vnode *vp; struct vattr vattr; struct smbfattr fattr; struct smb_cred *scred; char *name = cnp->cn_nameptr; int nmlen = cnp->cn_namelen; int error; SMBVDEBUG("\n"); *vpp = NULL; if (vap->va_type != VREG) return EOPNOTSUPP; if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) return error; scred = smbfs_malloc_scred(); - smb_makescred(scred, cnp->cn_thread, cnp->cn_cred); + smb_makescred(scred, curthread, cnp->cn_cred); error = smbfs_smb_create(dnp, name, nmlen, scred); if (error) goto out; error = smbfs_smb_lookup(dnp, name, nmlen, &fattr, scred); if (error) goto out; error = smbfs_nget(VTOVFS(dvp), dvp, name, nmlen, &fattr, &vp); if (error) goto out; *vpp = vp; if (cnp->cn_flags & MAKEENTRY) cache_enter(dvp, vp, cnp); out: smbfs_free_scred(scred); return error; } static int smbfs_remove(ap) struct vop_remove_args /* { struct vnodeop_desc *a_desc; struct vnode * a_dvp; struct vnode * a_vp; struct componentname * a_cnp; } */ *ap; { struct vnode *vp = ap->a_vp; /* struct vnode *dvp = ap->a_dvp;*/ struct componentname *cnp = ap->a_cnp; struct smbnode *np = VTOSMB(vp); struct smb_cred *scred; int error; if (vp->v_type == VDIR || (np->n_flag & NOPEN) != 0 || vrefcnt(vp) != 1) return EPERM; scred = smbfs_malloc_scred(); - smb_makescred(scred, cnp->cn_thread, cnp->cn_cred); + smb_makescred(scred, curthread, cnp->cn_cred); error = smbfs_smb_delete(np, scred); if (error == 0) np->n_flag |= NGONE; cache_purge(vp); smbfs_free_scred(scred); return error; } /* * smbfs_file rename call */ static int smbfs_rename(ap) struct vop_rename_args /* { struct vnode *a_fdvp; struct vnode *a_fvp; struct componentname *a_fcnp; struct vnode *a_tdvp; struct vnode *a_tvp; struct componentname *a_tcnp; } */ *ap; { struct vnode *fvp = ap->a_fvp; struct vnode *tvp = ap->a_tvp; struct vnode *fdvp = ap->a_fdvp; struct vnode *tdvp = ap->a_tdvp; struct componentname *tcnp = ap->a_tcnp; /* struct componentname *fcnp = ap->a_fcnp;*/ struct smb_cred *scred; u_int16_t flags = 6; int error=0; scred = NULL; /* Check for cross-device rename */ if ((fvp->v_mount != tdvp->v_mount) || (tvp && (fvp->v_mount != tvp->v_mount))) { error = EXDEV; goto out; } if (tvp && vrefcnt(tvp) > 1) { error = EBUSY; goto out; } flags = 0x10; /* verify all writes */ if (fvp->v_type == VDIR) { flags |= 2; } else if (fvp->v_type == VREG) { flags |= 1; } else { return EINVAL; } scred = smbfs_malloc_scred(); - smb_makescred(scred, tcnp->cn_thread, tcnp->cn_cred); + smb_makescred(scred, curthread, tcnp->cn_cred); /* * It seems that Samba doesn't implement SMB_COM_MOVE call... */ #ifdef notnow if (SMB_DIALECT(SSTOCN(smp->sm_share)) >= SMB_DIALECT_LANMAN1_0) { error = smbfs_smb_move(VTOSMB(fvp), VTOSMB(tdvp), tcnp->cn_nameptr, tcnp->cn_namelen, flags, scred); } else #endif { /* * We have to do the work atomicaly */ if (tvp && tvp != fvp) { error = smbfs_smb_delete(VTOSMB(tvp), scred); if (error) goto out_cacherem; VTOSMB(fvp)->n_flag |= NGONE; } error = smbfs_smb_rename(VTOSMB(fvp), VTOSMB(tdvp), tcnp->cn_nameptr, tcnp->cn_namelen, scred); } if (fvp->v_type == VDIR) { if (tvp != NULL && tvp->v_type == VDIR) cache_purge(tdvp); cache_purge(fdvp); } out_cacherem: smbfs_attr_cacheremove(fdvp); smbfs_attr_cacheremove(tdvp); out: smbfs_free_scred(scred); if (tdvp == tvp) vrele(tdvp); else vput(tdvp); if (tvp) vput(tvp); vrele(fdvp); vrele(fvp); #ifdef possible_mistake vgone(fvp); if (tvp) vgone(tvp); #endif return error; } /* * somtime it will come true... */ static int smbfs_link(ap) struct vop_link_args /* { struct vnode *a_tdvp; struct vnode *a_vp; struct componentname *a_cnp; } */ *ap; { return EOPNOTSUPP; } /* * smbfs_symlink link create call. * Sometime it will be functional... */ static int smbfs_symlink(ap) struct vop_symlink_args /* { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; struct vattr *a_vap; char *a_target; } */ *ap; { return EOPNOTSUPP; } static int smbfs_mknod(ap) struct vop_mknod_args /* { } */ *ap; { return EOPNOTSUPP; } static int smbfs_mkdir(ap) struct vop_mkdir_args /* { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; struct vattr *a_vap; } */ *ap; { struct vnode *dvp = ap->a_dvp; /* struct vattr *vap = ap->a_vap;*/ struct vnode *vp; struct componentname *cnp = ap->a_cnp; struct smbnode *dnp = VTOSMB(dvp); struct vattr vattr; struct smb_cred *scred; struct smbfattr fattr; char *name = cnp->cn_nameptr; int len = cnp->cn_namelen; int error; if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) { return error; } if ((name[0] == '.') && ((len == 1) || ((len == 2) && (name[1] == '.')))) return EEXIST; scred = smbfs_malloc_scred(); - smb_makescred(scred, cnp->cn_thread, cnp->cn_cred); + smb_makescred(scred, curthread, cnp->cn_cred); error = smbfs_smb_mkdir(dnp, name, len, scred); if (error) goto out; error = smbfs_smb_lookup(dnp, name, len, &fattr, scred); if (error) goto out; error = smbfs_nget(VTOVFS(dvp), dvp, name, len, &fattr, &vp); if (error) goto out; *ap->a_vpp = vp; out: smbfs_free_scred(scred); return error; } /* * smbfs_remove directory call */ static int smbfs_rmdir(ap) struct vop_rmdir_args /* { struct vnode *a_dvp; struct vnode *a_vp; struct componentname *a_cnp; } */ *ap; { struct vnode *vp = ap->a_vp; struct vnode *dvp = ap->a_dvp; struct componentname *cnp = ap->a_cnp; /* struct smbmount *smp = VTOSMBFS(vp);*/ struct smbnode *dnp = VTOSMB(dvp); struct smbnode *np = VTOSMB(vp); struct smb_cred *scred; int error; if (dvp == vp) return EINVAL; scred = smbfs_malloc_scred(); - smb_makescred(scred, cnp->cn_thread, cnp->cn_cred); + smb_makescred(scred, curthread, cnp->cn_cred); error = smbfs_smb_rmdir(np, scred); if (error == 0) np->n_flag |= NGONE; dnp->n_flag |= NMODIFIED; smbfs_attr_cacheremove(dvp); /* cache_purge(dvp);*/ cache_purge(vp); smbfs_free_scred(scred); return error; } /* * smbfs_readdir call */ static int smbfs_readdir(ap) struct vop_readdir_args /* { struct vnode *a_vp; struct uio *a_uio; struct ucred *a_cred; int *a_eofflag; u_long *a_cookies; int a_ncookies; } */ *ap; { struct vnode *vp = ap->a_vp; struct uio *uio = ap->a_uio; int error; if (vp->v_type != VDIR) return (EPERM); #ifdef notnow if (ap->a_ncookies) { printf("smbfs_readdir: no support for cookies now..."); return (EOPNOTSUPP); } #endif error = smbfs_readvnode(vp, uio, ap->a_cred); return error; } /* ARGSUSED */ static int smbfs_fsync(ap) struct vop_fsync_args /* { struct vnodeop_desc *a_desc; struct vnode * a_vp; struct ucred * a_cred; int a_waitfor; struct thread * a_td; } */ *ap; { /* return (smb_flush(ap->a_vp, ap->a_cred, ap->a_waitfor, ap->a_td, 1));*/ return (0); } static int smbfs_print (ap) struct vop_print_args /* { struct vnode *a_vp; } */ *ap; { struct vnode *vp = ap->a_vp; struct smbnode *np = VTOSMB(vp); if (np == NULL) { printf("no smbnode data\n"); return (0); } printf("\tname = %s, parent = %p, open = %d\n", np->n_name, np->n_parent ? np->n_parent : NULL, (np->n_flag & NOPEN) != 0); return (0); } static int smbfs_pathconf (ap) struct vop_pathconf_args /* { struct vnode *vp; int name; register_t *retval; } */ *ap; { struct smbmount *smp = VFSTOSMBFS(VTOVFS(ap->a_vp)); struct smb_vc *vcp = SSTOVC(smp->sm_share); long *retval = ap->a_retval; int error = 0; switch (ap->a_name) { case _PC_FILESIZEBITS: if (vcp->vc_sopt.sv_caps & (SMB_CAP_LARGE_READX | SMB_CAP_LARGE_WRITEX)) *retval = 64; else *retval = 32; break; case _PC_NAME_MAX: *retval = (vcp->vc_hflags2 & SMB_FLAGS2_KNOWS_LONG_NAMES) ? 255 : 12; break; case _PC_PATH_MAX: *retval = 800; /* XXX: a correct one ? */ break; case _PC_NO_TRUNC: *retval = 1; break; default: error = vop_stdpathconf(ap); } return error; } static int smbfs_strategy (ap) struct vop_strategy_args /* { struct buf *a_bp } */ *ap; { struct buf *bp=ap->a_bp; struct ucred *cr; struct thread *td; SMBVDEBUG("\n"); if (bp->b_flags & B_ASYNC) td = (struct thread *)0; else td = curthread; /* XXX */ if (bp->b_iocmd == BIO_READ) cr = bp->b_rcred; else cr = bp->b_wcred; if ((bp->b_flags & B_ASYNC) == 0 ) (void)smbfs_doio(ap->a_vp, bp, cr, td); return (0); } int smbfs_ioctl(ap) struct vop_ioctl_args /* { struct vnode *a_vp; u_long a_command; caddr_t a_data; int fflag; struct ucred *cred; struct thread *td; } */ *ap; { return ENOTTY; } static char smbfs_atl[] = "rhsvda"; static int smbfs_getextattr(struct vop_getextattr_args *ap) /* { IN struct vnode *a_vp; IN char *a_name; INOUT struct uio *a_uio; IN struct ucred *a_cred; IN struct thread *a_td; }; */ { struct vnode *vp = ap->a_vp; struct thread *td = ap->a_td; struct ucred *cred = ap->a_cred; struct uio *uio = ap->a_uio; const char *name = ap->a_name; struct smbnode *np = VTOSMB(vp); struct vattr vattr; char buf[10]; int i, attr, error; error = VOP_ACCESS(vp, VREAD, cred, td); if (error) return error; error = VOP_GETATTR(vp, &vattr, cred); if (error) return error; if (strcmp(name, "dosattr") == 0) { attr = np->n_dosattr; for (i = 0; i < 6; i++, attr >>= 1) buf[i] = (attr & 1) ? smbfs_atl[i] : '-'; buf[i] = 0; error = uiomove(buf, i, uio); } else error = EINVAL; return error; } /* * Since we expected to support F_GETLK (and SMB protocol has no such function), * it is necessary to use lf_advlock(). It would be nice if this function had * a callback mechanism because it will help to improve a level of consistency. */ int smbfs_advlock(ap) struct vop_advlock_args /* { struct vnode *a_vp; caddr_t a_id; int a_op; struct flock *a_fl; int a_flags; } */ *ap; { struct vnode *vp = ap->a_vp; struct smbnode *np = VTOSMB(vp); struct flock *fl = ap->a_fl; caddr_t id = (caddr_t)1 /* ap->a_id */; /* int flags = ap->a_flags;*/ struct thread *td = curthread; struct smb_cred *scred; u_quad_t size; off_t start, end, oadd; int error, lkop; if (vp->v_type == VDIR) { /* * SMB protocol have no support for directory locking. * Although locks can be processed on local machine, I don't * think that this is a good idea, because some programs * can work wrong assuming directory is locked. So, we just * return 'operation not supported */ return EOPNOTSUPP; } size = np->n_size; switch (fl->l_whence) { case SEEK_SET: case SEEK_CUR: start = fl->l_start; break; case SEEK_END: if (size > OFF_MAX || (fl->l_start > 0 && size > OFF_MAX - fl->l_start)) return EOVERFLOW; start = size + fl->l_start; break; default: return EINVAL; } if (start < 0) return EINVAL; if (fl->l_len < 0) { if (start == 0) return EINVAL; end = start - 1; start += fl->l_len; if (start < 0) return EINVAL; } else if (fl->l_len == 0) end = -1; else { oadd = fl->l_len - 1; if (oadd > OFF_MAX - start) return EOVERFLOW; end = start + oadd; } scred = smbfs_malloc_scred(); smb_makescred(scred, td, td->td_ucred); switch (ap->a_op) { case F_SETLK: switch (fl->l_type) { case F_WRLCK: lkop = SMB_LOCK_EXCL; break; case F_RDLCK: lkop = SMB_LOCK_SHARED; break; case F_UNLCK: lkop = SMB_LOCK_RELEASE; break; default: smbfs_free_scred(scred); return EINVAL; } error = lf_advlock(ap, &vp->v_lockf, size); if (error) break; lkop = SMB_LOCK_EXCL; error = smbfs_smb_lock(np, lkop, id, start, end, scred); if (error) { int oldtype = fl->l_type; fl->l_type = F_UNLCK; ap->a_op = F_UNLCK; lf_advlock(ap, &vp->v_lockf, size); fl->l_type = oldtype; } break; case F_UNLCK: lf_advlock(ap, &vp->v_lockf, size); error = smbfs_smb_lock(np, SMB_LOCK_RELEASE, id, start, end, scred); break; case F_GETLK: error = lf_advlock(ap, &vp->v_lockf, size); break; default: smbfs_free_scred(scred); return EINVAL; } smbfs_free_scred(scred); return error; } static int smbfs_pathcheck(struct smbmount *smp, const char *name, int nmlen, int nameiop) { static const char *badchars = "*/:<>?"; static const char *badchars83 = " +|,[]=;"; const char *cp; int i, error; /* * Backslash characters, being a path delimiter, are prohibited * within a path component even for LOOKUP operations. */ if (strchr(name, '\\') != NULL) return ENOENT; if (nameiop == LOOKUP) return 0; error = ENOENT; if (SMB_DIALECT(SSTOVC(smp->sm_share)) < SMB_DIALECT_LANMAN2_0) { /* * Name should conform 8.3 format */ if (nmlen > 12) return ENAMETOOLONG; cp = strchr(name, '.'); if (cp == NULL) return error; if (cp == name || (cp - name) > 8) return error; cp = strchr(cp + 1, '.'); if (cp != NULL) return error; for (cp = name, i = 0; i < nmlen; i++, cp++) if (strchr(badchars83, *cp) != NULL) return error; } for (cp = name, i = 0; i < nmlen; i++, cp++) if (strchr(badchars, *cp) != NULL) return error; return 0; } /* * Things go even weird without fixed inode numbers... */ int smbfs_lookup(ap) struct vop_lookup_args /* { struct vnodeop_desc *a_desc; struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; } */ *ap; { struct componentname *cnp = ap->a_cnp; - struct thread *td = cnp->cn_thread; + struct thread *td = curthread; struct vnode *dvp = ap->a_dvp; struct vnode **vpp = ap->a_vpp; struct vnode *vp; struct smbmount *smp; struct mount *mp = dvp->v_mount; struct smbnode *dnp; struct smbfattr fattr, *fap; struct smb_cred *scred; char *name = cnp->cn_nameptr; int flags = cnp->cn_flags; int nameiop = cnp->cn_nameiop; int nmlen = cnp->cn_namelen; int error, islastcn, isdot; int killit; SMBVDEBUG("\n"); if (dvp->v_type != VDIR) return ENOTDIR; if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT)) { SMBFSERR("invalid '..'\n"); return EIO; } islastcn = flags & ISLASTCN; if (islastcn && (mp->mnt_flag & MNT_RDONLY) && (nameiop != LOOKUP)) return EROFS; error = vn_dir_check_exec(dvp, cnp); if (error != 0) return error; smp = VFSTOSMBFS(mp); dnp = VTOSMB(dvp); isdot = (nmlen == 1 && name[0] == '.'); error = smbfs_pathcheck(smp, cnp->cn_nameptr, cnp->cn_namelen, nameiop); if (error) return ENOENT; error = cache_lookup(dvp, vpp, cnp, NULL, NULL); SMBVDEBUG("cache_lookup returned %d\n", error); if (error > 0) return error; if (error) { /* name was found */ struct vattr vattr; killit = 0; vp = *vpp; error = VOP_GETATTR(vp, &vattr, cnp->cn_cred); /* * If the file type on the server is inconsistent * with what it was when we created the vnode, * kill the bogus vnode now and fall through to * the code below to create a new one with the * right type. */ if (error == 0 && ((vp->v_type == VDIR && (VTOSMB(vp)->n_dosattr & SMB_FA_DIR) == 0) || (vp->v_type == VREG && (VTOSMB(vp)->n_dosattr & SMB_FA_DIR) != 0))) killit = 1; else if (error == 0 /* && vattr.va_ctime.tv_sec == VTOSMB(vp)->n_ctime*/) { if (nameiop != LOOKUP && islastcn) cnp->cn_flags |= SAVENAME; SMBVDEBUG("use cached vnode\n"); return (0); } cache_purge(vp); /* * XXX This is not quite right, if '.' is * inconsistent, we really need to start the lookup * all over again. Hopefully there is some other * guarantee that prevents this case from happening. */ if (killit && vp != dvp) vgone(vp); if (vp != dvp) vput(vp); else vrele(vp); *vpp = NULLVP; } /* * entry is not in the cache or has been expired */ error = 0; *vpp = NULLVP; scred = smbfs_malloc_scred(); smb_makescred(scred, td, cnp->cn_cred); fap = &fattr; if (flags & ISDOTDOT) { /* * In the DOTDOT case, don't go over-the-wire * in order to request attributes. We already * know it's a directory and subsequent call to * smbfs_getattr() will restore consistency. * */ SMBVDEBUG("smbfs_smb_lookup: dotdot\n"); } else if (isdot) { error = smbfs_smb_lookup(dnp, NULL, 0, fap, scred); SMBVDEBUG("result of smbfs_smb_lookup: %d\n", error); } else { error = smbfs_smb_lookup(dnp, name, nmlen, fap, scred); SMBVDEBUG("result of smbfs_smb_lookup: %d\n", error); } if (error && error != ENOENT) goto out; if (error) { /* entry not found */ /* * Handle RENAME or CREATE case... */ if ((nameiop == CREATE || nameiop == RENAME) && islastcn) { error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td); if (error) goto out; cnp->cn_flags |= SAVENAME; error = EJUSTRETURN; goto out; } error = ENOENT; goto out; }/* else { SMBVDEBUG("Found entry %s with id=%d\n", fap->entryName, fap->dirEntNum); }*/ /* * handle DELETE case ... */ if (nameiop == DELETE && islastcn) { /* delete last component */ error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td); if (error) goto out; if (isdot) { VREF(dvp); *vpp = dvp; goto out; } error = smbfs_nget(mp, dvp, name, nmlen, fap, &vp); if (error) goto out; *vpp = vp; cnp->cn_flags |= SAVENAME; goto out; } if (nameiop == RENAME && islastcn) { error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td); if (error) goto out; if (isdot) { error = EISDIR; goto out; } error = smbfs_nget(mp, dvp, name, nmlen, fap, &vp); if (error) goto out; *vpp = vp; cnp->cn_flags |= SAVENAME; goto out; } if (flags & ISDOTDOT) { mp = dvp->v_mount; error = vfs_busy(mp, MBF_NOWAIT); if (error != 0) { vfs_ref(mp); VOP_UNLOCK(dvp); error = vfs_busy(mp, 0); vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); vfs_rel(mp); if (error) { error = ENOENT; goto out; } if (VN_IS_DOOMED(dvp)) { vfs_unbusy(mp); error = ENOENT; goto out; } } VOP_UNLOCK(dvp); error = smbfs_nget(mp, dvp, name, nmlen, NULL, &vp); vfs_unbusy(mp); vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); if (VN_IS_DOOMED(dvp)) { if (error == 0) vput(vp); error = ENOENT; } if (error) goto out; *vpp = vp; } else if (isdot) { vref(dvp); *vpp = dvp; } else { error = smbfs_nget(mp, dvp, name, nmlen, fap, &vp); if (error) goto out; *vpp = vp; SMBVDEBUG("lookup: getnewvp!\n"); } if ((cnp->cn_flags & MAKEENTRY)/* && !islastcn*/) { /* VTOSMB(*vpp)->n_ctime = VTOSMB(*vpp)->n_vattr.va_ctime.tv_sec;*/ cache_enter(dvp, *vpp, cnp); } out: smbfs_free_scred(scred); return (error); } diff --git a/sys/fs/tmpfs/tmpfs_vnops.c b/sys/fs/tmpfs/tmpfs_vnops.c index d8c74cecdfe4..3b498bf27a79 100644 --- a/sys/fs/tmpfs/tmpfs_vnops.c +++ b/sys/fs/tmpfs/tmpfs_vnops.c @@ -1,1881 +1,1881 @@ /* $NetBSD: tmpfs_vnops.c,v 1.39 2007/07/23 15:41:01 jmmv Exp $ */ /*- * SPDX-License-Identifier: BSD-2-Clause-NetBSD * * Copyright (c) 2005, 2006 The NetBSD Foundation, Inc. * All rights reserved. * * This code is derived from software contributed to The NetBSD Foundation * by Julio M. Merino Vidal, developed as part of Google's Summer of Code * 2005 program. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGE. */ /* * tmpfs vnode interface. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include SYSCTL_DECL(_vfs_tmpfs); VFS_SMR_DECLARE; static volatile int tmpfs_rename_restarts; SYSCTL_INT(_vfs_tmpfs, OID_AUTO, rename_restarts, CTLFLAG_RD, __DEVOLATILE(int *, &tmpfs_rename_restarts), 0, "Times rename had to restart due to lock contention"); static int tmpfs_vn_get_ino_alloc(struct mount *mp, void *arg, int lkflags, struct vnode **rvp) { return (tmpfs_alloc_vp(mp, arg, lkflags, rvp)); } static int tmpfs_lookup1(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp) { struct tmpfs_dirent *de; struct tmpfs_node *dnode, *pnode; struct tmpfs_mount *tm; int error; /* Caller assumes responsibility for ensuring access (VEXEC). */ dnode = VP_TO_TMPFS_DIR(dvp); *vpp = NULLVP; /* We cannot be requesting the parent directory of the root node. */ MPASS(IMPLIES(dnode->tn_type == VDIR && dnode->tn_dir.tn_parent == dnode, !(cnp->cn_flags & ISDOTDOT))); TMPFS_ASSERT_LOCKED(dnode); if (dnode->tn_dir.tn_parent == NULL) { error = ENOENT; goto out; } if (cnp->cn_flags & ISDOTDOT) { tm = VFS_TO_TMPFS(dvp->v_mount); pnode = dnode->tn_dir.tn_parent; tmpfs_ref_node(pnode); error = vn_vget_ino_gen(dvp, tmpfs_vn_get_ino_alloc, pnode, cnp->cn_lkflags, vpp); tmpfs_free_node(tm, pnode); if (error != 0) goto out; } else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { VREF(dvp); *vpp = dvp; error = 0; } else { de = tmpfs_dir_lookup(dnode, NULL, cnp); if (de != NULL && de->td_node == NULL) cnp->cn_flags |= ISWHITEOUT; if (de == NULL || de->td_node == NULL) { /* * The entry was not found in the directory. * This is OK if we are creating or renaming an * entry and are working on the last component of * the path name. */ if ((cnp->cn_flags & ISLASTCN) && (cnp->cn_nameiop == CREATE || \ cnp->cn_nameiop == RENAME || (cnp->cn_nameiop == DELETE && cnp->cn_flags & DOWHITEOUT && cnp->cn_flags & ISWHITEOUT))) { error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, - cnp->cn_thread); + curthread); if (error != 0) goto out; /* * Keep the component name in the buffer for * future uses. */ cnp->cn_flags |= SAVENAME; error = EJUSTRETURN; } else error = ENOENT; } else { struct tmpfs_node *tnode; /* * The entry was found, so get its associated * tmpfs_node. */ tnode = de->td_node; /* * If we are not at the last path component and * found a non-directory or non-link entry (which * may itself be pointing to a directory), raise * an error. */ if ((tnode->tn_type != VDIR && tnode->tn_type != VLNK) && !(cnp->cn_flags & ISLASTCN)) { error = ENOTDIR; goto out; } /* * If we are deleting or renaming the entry, keep * track of its tmpfs_dirent so that it can be * easily deleted later. */ if ((cnp->cn_flags & ISLASTCN) && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, - cnp->cn_thread); + curthread); if (error != 0) goto out; /* Allocate a new vnode on the matching entry. */ error = tmpfs_alloc_vp(dvp->v_mount, tnode, cnp->cn_lkflags, vpp); if (error != 0) goto out; if ((dnode->tn_mode & S_ISTXT) && VOP_ACCESS(dvp, VADMIN, cnp->cn_cred, - cnp->cn_thread) && VOP_ACCESS(*vpp, VADMIN, - cnp->cn_cred, cnp->cn_thread)) { + curthread) && VOP_ACCESS(*vpp, VADMIN, + cnp->cn_cred, curthread)) { error = EPERM; vput(*vpp); *vpp = NULL; goto out; } cnp->cn_flags |= SAVENAME; } else { error = tmpfs_alloc_vp(dvp->v_mount, tnode, cnp->cn_lkflags, vpp); if (error != 0) goto out; } } } /* * Store the result of this lookup in the cache. Avoid this if the * request was for creation, as it does not improve timings on * emprical tests. */ if ((cnp->cn_flags & MAKEENTRY) != 0 && tmpfs_use_nc(dvp)) cache_enter(dvp, *vpp, cnp); out: /* * If there were no errors, *vpp cannot be null and it must be * locked. */ MPASS(IFF(error == 0, *vpp != NULLVP && VOP_ISLOCKED(*vpp))); return (error); } static int tmpfs_cached_lookup(struct vop_cachedlookup_args *v) { return (tmpfs_lookup1(v->a_dvp, v->a_vpp, v->a_cnp)); } static int tmpfs_lookup(struct vop_lookup_args *v) { struct vnode *dvp = v->a_dvp; struct vnode **vpp = v->a_vpp; struct componentname *cnp = v->a_cnp; int error; /* Check accessibility of requested node as a first step. */ error = vn_dir_check_exec(dvp, cnp); if (error != 0) return (error); return (tmpfs_lookup1(dvp, vpp, cnp)); } static int tmpfs_create(struct vop_create_args *v) { struct vnode *dvp = v->a_dvp; struct vnode **vpp = v->a_vpp; struct componentname *cnp = v->a_cnp; struct vattr *vap = v->a_vap; int error; MPASS(vap->va_type == VREG || vap->va_type == VSOCK); error = tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL); if (error == 0 && (cnp->cn_flags & MAKEENTRY) != 0 && tmpfs_use_nc(dvp)) cache_enter(dvp, *vpp, cnp); return (error); } static int tmpfs_mknod(struct vop_mknod_args *v) { struct vnode *dvp = v->a_dvp; struct vnode **vpp = v->a_vpp; struct componentname *cnp = v->a_cnp; struct vattr *vap = v->a_vap; if (vap->va_type != VBLK && vap->va_type != VCHR && vap->va_type != VFIFO) return (EINVAL); return (tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL)); } struct fileops tmpfs_fnops; static int tmpfs_open(struct vop_open_args *v) { struct vnode *vp; struct tmpfs_node *node; struct file *fp; int error, mode; vp = v->a_vp; mode = v->a_mode; node = VP_TO_TMPFS_NODE(vp); /* * The file is still active but all its names have been removed * (e.g. by a "rmdir $(pwd)"). It cannot be opened any more as * it is about to die. */ if (node->tn_links < 1) return (ENOENT); /* If the file is marked append-only, deny write requests. */ if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE) error = EPERM; else { error = 0; /* For regular files, the call below is nop. */ KASSERT(vp->v_type != VREG || (node->tn_reg.tn_aobj->flags & OBJ_DEAD) == 0, ("dead object")); vnode_create_vobject(vp, node->tn_size, v->a_td); } fp = v->a_fp; MPASS(fp == NULL || fp->f_data == NULL); if (error == 0 && fp != NULL && vp->v_type == VREG) { tmpfs_ref_node(node); finit_vnode(fp, mode, node, &tmpfs_fnops); } return (error); } static int tmpfs_close(struct vop_close_args *v) { struct vnode *vp = v->a_vp; /* Update node times. */ tmpfs_update(vp); return (0); } int tmpfs_fo_close(struct file *fp, struct thread *td) { struct tmpfs_node *node; node = fp->f_data; if (node != NULL) { MPASS(node->tn_type == VREG); tmpfs_free_node(node->tn_reg.tn_tmp, node); } return (vnops.fo_close(fp, td)); } /* * VOP_FPLOOKUP_VEXEC routines are subject to special circumstances, see * the comment above cache_fplookup for details. */ int tmpfs_fplookup_vexec(struct vop_fplookup_vexec_args *v) { struct vnode *vp; struct tmpfs_node *node; struct ucred *cred; mode_t all_x, mode; vp = v->a_vp; node = VP_TO_TMPFS_NODE_SMR(vp); if (__predict_false(node == NULL)) return (EAGAIN); all_x = S_IXUSR | S_IXGRP | S_IXOTH; mode = atomic_load_short(&node->tn_mode); if (__predict_true((mode & all_x) == all_x)) return (0); cred = v->a_cred; return (vaccess_vexec_smr(mode, node->tn_uid, node->tn_gid, cred)); } int tmpfs_access(struct vop_access_args *v) { struct vnode *vp = v->a_vp; accmode_t accmode = v->a_accmode; struct ucred *cred = v->a_cred; mode_t all_x = S_IXUSR | S_IXGRP | S_IXOTH; int error; struct tmpfs_node *node; MPASS(VOP_ISLOCKED(vp)); node = VP_TO_TMPFS_NODE(vp); /* * Common case path lookup. */ if (__predict_true(accmode == VEXEC && (node->tn_mode & all_x) == all_x)) return (0); switch (vp->v_type) { case VDIR: /* FALLTHROUGH */ case VLNK: /* FALLTHROUGH */ case VREG: if (accmode & VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) { error = EROFS; goto out; } break; case VBLK: /* FALLTHROUGH */ case VCHR: /* FALLTHROUGH */ case VSOCK: /* FALLTHROUGH */ case VFIFO: break; default: error = EINVAL; goto out; } if (accmode & VWRITE && node->tn_flags & IMMUTABLE) { error = EPERM; goto out; } error = vaccess(vp->v_type, node->tn_mode, node->tn_uid, node->tn_gid, accmode, cred); out: MPASS(VOP_ISLOCKED(vp)); return (error); } int tmpfs_stat(struct vop_stat_args *v) { struct vnode *vp = v->a_vp; struct stat *sb = v->a_sb; vm_object_t obj; struct tmpfs_node *node; int error; node = VP_TO_TMPFS_NODE(vp); tmpfs_update_getattr(vp); error = vop_stat_helper_pre(v); if (__predict_false(error)) return (error); sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; sb->st_ino = node->tn_id; sb->st_mode = node->tn_mode | VTTOIF(vp->v_type); sb->st_nlink = node->tn_links; sb->st_uid = node->tn_uid; sb->st_gid = node->tn_gid; sb->st_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ? node->tn_rdev : NODEV; sb->st_size = node->tn_size; sb->st_atim.tv_sec = node->tn_atime.tv_sec; sb->st_atim.tv_nsec = node->tn_atime.tv_nsec; sb->st_mtim.tv_sec = node->tn_mtime.tv_sec; sb->st_mtim.tv_nsec = node->tn_mtime.tv_nsec; sb->st_ctim.tv_sec = node->tn_ctime.tv_sec; sb->st_ctim.tv_nsec = node->tn_ctime.tv_nsec; sb->st_birthtim.tv_sec = node->tn_birthtime.tv_sec; sb->st_birthtim.tv_nsec = node->tn_birthtime.tv_nsec; sb->st_blksize = PAGE_SIZE; sb->st_flags = node->tn_flags; sb->st_gen = node->tn_gen; if (vp->v_type == VREG) { obj = node->tn_reg.tn_aobj; sb->st_blocks = (u_quad_t)obj->resident_page_count * PAGE_SIZE; } else sb->st_blocks = node->tn_size; sb->st_blocks /= S_BLKSIZE; return (vop_stat_helper_post(v, error)); } int tmpfs_getattr(struct vop_getattr_args *v) { struct vnode *vp = v->a_vp; struct vattr *vap = v->a_vap; vm_object_t obj; struct tmpfs_node *node; node = VP_TO_TMPFS_NODE(vp); tmpfs_update_getattr(vp); vap->va_type = vp->v_type; vap->va_mode = node->tn_mode; vap->va_nlink = node->tn_links; vap->va_uid = node->tn_uid; vap->va_gid = node->tn_gid; vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; vap->va_fileid = node->tn_id; vap->va_size = node->tn_size; vap->va_blocksize = PAGE_SIZE; vap->va_atime = node->tn_atime; vap->va_mtime = node->tn_mtime; vap->va_ctime = node->tn_ctime; vap->va_birthtime = node->tn_birthtime; vap->va_gen = node->tn_gen; vap->va_flags = node->tn_flags; vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ? node->tn_rdev : NODEV; if (vp->v_type == VREG) { obj = node->tn_reg.tn_aobj; vap->va_bytes = (u_quad_t)obj->resident_page_count * PAGE_SIZE; } else vap->va_bytes = node->tn_size; vap->va_filerev = 0; return (0); } int tmpfs_setattr(struct vop_setattr_args *v) { struct vnode *vp = v->a_vp; struct vattr *vap = v->a_vap; struct ucred *cred = v->a_cred; struct thread *td = curthread; int error; MPASS(VOP_ISLOCKED(vp)); ASSERT_VOP_IN_SEQC(vp); error = 0; /* Abort if any unsettable attribute is given. */ if (vap->va_type != VNON || vap->va_nlink != VNOVAL || vap->va_fsid != VNOVAL || vap->va_fileid != VNOVAL || vap->va_blocksize != VNOVAL || vap->va_gen != VNOVAL || vap->va_rdev != VNOVAL || vap->va_bytes != VNOVAL) error = EINVAL; if (error == 0 && (vap->va_flags != VNOVAL)) error = tmpfs_chflags(vp, vap->va_flags, cred, td); if (error == 0 && (vap->va_size != VNOVAL)) error = tmpfs_chsize(vp, vap->va_size, cred, td); if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL)) error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, td); if (error == 0 && (vap->va_mode != (mode_t)VNOVAL)) error = tmpfs_chmod(vp, vap->va_mode, cred, td); if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL && vap->va_atime.tv_nsec != VNOVAL) || (vap->va_mtime.tv_sec != VNOVAL && vap->va_mtime.tv_nsec != VNOVAL) || (vap->va_birthtime.tv_sec != VNOVAL && vap->va_birthtime.tv_nsec != VNOVAL))) error = tmpfs_chtimes(vp, vap, cred, td); /* * Update the node times. We give preference to the error codes * generated by this function rather than the ones that may arise * from tmpfs_update. */ tmpfs_update(vp); MPASS(VOP_ISLOCKED(vp)); return (error); } static int tmpfs_read(struct vop_read_args *v) { struct vnode *vp; struct uio *uio; struct tmpfs_node *node; vp = v->a_vp; if (vp->v_type != VREG) return (EISDIR); uio = v->a_uio; if (uio->uio_offset < 0) return (EINVAL); node = VP_TO_TMPFS_NODE(vp); tmpfs_set_accessed(VFS_TO_TMPFS(vp->v_mount), node); return (uiomove_object(node->tn_reg.tn_aobj, node->tn_size, uio)); } static int tmpfs_read_pgcache(struct vop_read_pgcache_args *v) { struct vnode *vp; struct tmpfs_node *node; vm_object_t object; off_t size; int error; vp = v->a_vp; VNPASS((vn_irflag_read(vp) & VIRF_PGREAD) != 0, vp); if (v->a_uio->uio_offset < 0) return (EINVAL); error = EJUSTRETURN; vfs_smr_enter(); node = VP_TO_TMPFS_NODE_SMR(vp); if (node == NULL) goto out_smr; MPASS(node->tn_type == VREG); MPASS(node->tn_refcount >= 1); object = node->tn_reg.tn_aobj; if (object == NULL) goto out_smr; MPASS(object->type == tmpfs_pager_type); MPASS((object->flags & (OBJ_ANON | OBJ_DEAD | OBJ_SWAP)) == OBJ_SWAP); if (!VN_IS_DOOMED(vp)) { /* size cannot become shorter due to rangelock. */ size = node->tn_size; tmpfs_set_accessed(node->tn_reg.tn_tmp, node); vfs_smr_exit(); error = uiomove_object(object, size, v->a_uio); return (error); } out_smr: vfs_smr_exit(); return (error); } static int tmpfs_write(struct vop_write_args *v) { struct vnode *vp; struct uio *uio; struct tmpfs_node *node; off_t oldsize; int error, ioflag; mode_t newmode; vp = v->a_vp; uio = v->a_uio; ioflag = v->a_ioflag; error = 0; node = VP_TO_TMPFS_NODE(vp); oldsize = node->tn_size; if (uio->uio_offset < 0 || vp->v_type != VREG) return (EINVAL); if (uio->uio_resid == 0) return (0); if (ioflag & IO_APPEND) uio->uio_offset = node->tn_size; if (uio->uio_offset + uio->uio_resid > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) return (EFBIG); if (vn_rlimit_fsize(vp, uio, uio->uio_td)) return (EFBIG); if (uio->uio_offset + uio->uio_resid > node->tn_size) { error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid, FALSE); if (error != 0) goto out; } error = uiomove_object(node->tn_reg.tn_aobj, node->tn_size, uio); node->tn_status |= TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED; node->tn_accessed = true; if (node->tn_mode & (S_ISUID | S_ISGID)) { if (priv_check_cred(v->a_cred, PRIV_VFS_RETAINSUGID)) { newmode = node->tn_mode & ~(S_ISUID | S_ISGID); vn_seqc_write_begin(vp); atomic_store_short(&node->tn_mode, newmode); vn_seqc_write_end(vp); } } if (error != 0) (void)tmpfs_reg_resize(vp, oldsize, TRUE); out: MPASS(IMPLIES(error == 0, uio->uio_resid == 0)); MPASS(IMPLIES(error != 0, oldsize == node->tn_size)); return (error); } static int tmpfs_deallocate(struct vop_deallocate_args *v) { return (tmpfs_reg_punch_hole(v->a_vp, v->a_offset, v->a_len)); } static int tmpfs_fsync(struct vop_fsync_args *v) { struct vnode *vp = v->a_vp; MPASS(VOP_ISLOCKED(vp)); tmpfs_check_mtime(vp); tmpfs_update(vp); return (0); } static int tmpfs_remove(struct vop_remove_args *v) { struct vnode *dvp = v->a_dvp; struct vnode *vp = v->a_vp; int error; struct tmpfs_dirent *de; struct tmpfs_mount *tmp; struct tmpfs_node *dnode; struct tmpfs_node *node; MPASS(VOP_ISLOCKED(dvp)); MPASS(VOP_ISLOCKED(vp)); if (vp->v_type == VDIR) { error = EISDIR; goto out; } dnode = VP_TO_TMPFS_DIR(dvp); node = VP_TO_TMPFS_NODE(vp); tmp = VFS_TO_TMPFS(vp->v_mount); de = tmpfs_dir_lookup(dnode, node, v->a_cnp); MPASS(de != NULL); /* Files marked as immutable or append-only cannot be deleted. */ if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) || (dnode->tn_flags & APPEND)) { error = EPERM; goto out; } /* Remove the entry from the directory; as it is a file, we do not * have to change the number of hard links of the directory. */ tmpfs_dir_detach(dvp, de); if (v->a_cnp->cn_flags & DOWHITEOUT) tmpfs_dir_whiteout_add(dvp, v->a_cnp); /* Free the directory entry we just deleted. Note that the node * referred by it will not be removed until the vnode is really * reclaimed. */ tmpfs_free_dirent(tmp, de); node->tn_status |= TMPFS_NODE_CHANGED; node->tn_accessed = true; error = 0; out: return (error); } static int tmpfs_link(struct vop_link_args *v) { struct vnode *dvp = v->a_tdvp; struct vnode *vp = v->a_vp; struct componentname *cnp = v->a_cnp; int error; struct tmpfs_dirent *de; struct tmpfs_node *node; MPASS(VOP_ISLOCKED(dvp)); MPASS(cnp->cn_flags & HASBUF); MPASS(dvp != vp); /* XXX When can this be false? */ node = VP_TO_TMPFS_NODE(vp); /* Ensure that we do not overflow the maximum number of links imposed * by the system. */ MPASS(node->tn_links <= TMPFS_LINK_MAX); if (node->tn_links == TMPFS_LINK_MAX) { error = EMLINK; goto out; } /* We cannot create links of files marked immutable or append-only. */ if (node->tn_flags & (IMMUTABLE | APPEND)) { error = EPERM; goto out; } /* Allocate a new directory entry to represent the node. */ error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node, cnp->cn_nameptr, cnp->cn_namelen, &de); if (error != 0) goto out; /* Insert the new directory entry into the appropriate directory. */ if (cnp->cn_flags & ISWHITEOUT) tmpfs_dir_whiteout_remove(dvp, cnp); tmpfs_dir_attach(dvp, de); /* vp link count has changed, so update node times. */ node->tn_status |= TMPFS_NODE_CHANGED; tmpfs_update(vp); error = 0; out: return (error); } /* * We acquire all but fdvp locks using non-blocking acquisitions. If we * fail to acquire any lock in the path we will drop all held locks, * acquire the new lock in a blocking fashion, and then release it and * restart the rename. This acquire/release step ensures that we do not * spin on a lock waiting for release. On error release all vnode locks * and decrement references the way tmpfs_rename() would do. */ static int tmpfs_rename_relock(struct vnode *fdvp, struct vnode **fvpp, struct vnode *tdvp, struct vnode **tvpp, struct componentname *fcnp, struct componentname *tcnp) { struct vnode *nvp; struct mount *mp; struct tmpfs_dirent *de; int error, restarts = 0; VOP_UNLOCK(tdvp); if (*tvpp != NULL && *tvpp != tdvp) VOP_UNLOCK(*tvpp); mp = fdvp->v_mount; relock: restarts += 1; error = vn_lock(fdvp, LK_EXCLUSIVE); if (error) goto releout; if (vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { VOP_UNLOCK(fdvp); error = vn_lock(tdvp, LK_EXCLUSIVE); if (error) goto releout; VOP_UNLOCK(tdvp); goto relock; } /* * Re-resolve fvp to be certain it still exists and fetch the * correct vnode. */ de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(fdvp), NULL, fcnp); if (de == NULL) { VOP_UNLOCK(fdvp); VOP_UNLOCK(tdvp); if ((fcnp->cn_flags & ISDOTDOT) != 0 || (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.')) error = EINVAL; else error = ENOENT; goto releout; } error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE | LK_NOWAIT, &nvp); if (error != 0) { VOP_UNLOCK(fdvp); VOP_UNLOCK(tdvp); if (error != EBUSY) goto releout; error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE, &nvp); if (error != 0) goto releout; VOP_UNLOCK(nvp); /* * Concurrent rename race. */ if (nvp == tdvp) { vrele(nvp); error = EINVAL; goto releout; } vrele(*fvpp); *fvpp = nvp; goto relock; } vrele(*fvpp); *fvpp = nvp; VOP_UNLOCK(*fvpp); /* * Re-resolve tvp and acquire the vnode lock if present. */ de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(tdvp), NULL, tcnp); /* * If tvp disappeared we just carry on. */ if (de == NULL && *tvpp != NULL) { vrele(*tvpp); *tvpp = NULL; } /* * Get the tvp ino if the lookup succeeded. We may have to restart * if the non-blocking acquire fails. */ if (de != NULL) { nvp = NULL; error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE | LK_NOWAIT, &nvp); if (*tvpp != NULL) vrele(*tvpp); *tvpp = nvp; if (error != 0) { VOP_UNLOCK(fdvp); VOP_UNLOCK(tdvp); if (error != EBUSY) goto releout; error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE, &nvp); if (error != 0) goto releout; VOP_UNLOCK(nvp); /* * fdvp contains fvp, thus tvp (=fdvp) is not empty. */ if (nvp == fdvp) { error = ENOTEMPTY; goto releout; } goto relock; } } tmpfs_rename_restarts += restarts; return (0); releout: vrele(fdvp); vrele(*fvpp); vrele(tdvp); if (*tvpp != NULL) vrele(*tvpp); tmpfs_rename_restarts += restarts; return (error); } static int tmpfs_rename(struct vop_rename_args *v) { struct vnode *fdvp = v->a_fdvp; struct vnode *fvp = v->a_fvp; struct componentname *fcnp = v->a_fcnp; struct vnode *tdvp = v->a_tdvp; struct vnode *tvp = v->a_tvp; struct componentname *tcnp = v->a_tcnp; char *newname; struct tmpfs_dirent *de; struct tmpfs_mount *tmp; struct tmpfs_node *fdnode; struct tmpfs_node *fnode; struct tmpfs_node *tnode; struct tmpfs_node *tdnode; int error; bool want_seqc_end; MPASS(VOP_ISLOCKED(tdvp)); MPASS(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp))); MPASS(fcnp->cn_flags & HASBUF); MPASS(tcnp->cn_flags & HASBUF); want_seqc_end = false; /* * Disallow cross-device renames. * XXX Why isn't this done by the caller? */ if (fvp->v_mount != tdvp->v_mount || (tvp != NULL && fvp->v_mount != tvp->v_mount)) { error = EXDEV; goto out; } /* If source and target are the same file, there is nothing to do. */ if (fvp == tvp) { error = 0; goto out; } /* * If we need to move the directory between entries, lock the * source so that we can safely operate on it. */ if (fdvp != tdvp && fdvp != tvp) { if (vn_lock(fdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { error = tmpfs_rename_relock(fdvp, &fvp, tdvp, &tvp, fcnp, tcnp); if (error != 0) return (error); ASSERT_VOP_ELOCKED(fdvp, "tmpfs_rename: fdvp not locked"); ASSERT_VOP_ELOCKED(tdvp, "tmpfs_rename: tdvp not locked"); if (tvp != NULL) ASSERT_VOP_ELOCKED(tvp, "tmpfs_rename: tvp not locked"); if (fvp == tvp) { error = 0; goto out_locked; } } } if (tvp != NULL) vn_seqc_write_begin(tvp); vn_seqc_write_begin(tdvp); vn_seqc_write_begin(fvp); vn_seqc_write_begin(fdvp); want_seqc_end = true; tmp = VFS_TO_TMPFS(tdvp->v_mount); tdnode = VP_TO_TMPFS_DIR(tdvp); tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp); fdnode = VP_TO_TMPFS_DIR(fdvp); fnode = VP_TO_TMPFS_NODE(fvp); de = tmpfs_dir_lookup(fdnode, fnode, fcnp); /* * Entry can disappear before we lock fdvp, * also avoid manipulating '.' and '..' entries. */ if (de == NULL) { if ((fcnp->cn_flags & ISDOTDOT) != 0 || (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.')) error = EINVAL; else error = ENOENT; goto out_locked; } MPASS(de->td_node == fnode); /* * If re-naming a directory to another preexisting directory * ensure that the target directory is empty so that its * removal causes no side effects. * Kern_rename guarantees the destination to be a directory * if the source is one. */ if (tvp != NULL) { MPASS(tnode != NULL); if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) || (tdnode->tn_flags & (APPEND | IMMUTABLE))) { error = EPERM; goto out_locked; } if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) { if (tnode->tn_size > 0) { error = ENOTEMPTY; goto out_locked; } } else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) { error = ENOTDIR; goto out_locked; } else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) { error = EISDIR; goto out_locked; } else { MPASS(fnode->tn_type != VDIR && tnode->tn_type != VDIR); } } if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) || (fdnode->tn_flags & (APPEND | IMMUTABLE))) { error = EPERM; goto out_locked; } /* * Ensure that we have enough memory to hold the new name, if it * has to be changed. */ if (fcnp->cn_namelen != tcnp->cn_namelen || bcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) { newname = malloc(tcnp->cn_namelen, M_TMPFSNAME, M_WAITOK); } else newname = NULL; /* * If the node is being moved to another directory, we have to do * the move. */ if (fdnode != tdnode) { /* * In case we are moving a directory, we have to adjust its * parent to point to the new parent. */ if (de->td_node->tn_type == VDIR) { struct tmpfs_node *n; /* * Ensure the target directory is not a child of the * directory being moved. Otherwise, we'd end up * with stale nodes. */ n = tdnode; /* * TMPFS_LOCK guaranties that no nodes are freed while * traversing the list. Nodes can only be marked as * removed: tn_parent == NULL. */ TMPFS_LOCK(tmp); TMPFS_NODE_LOCK(n); while (n != n->tn_dir.tn_parent) { struct tmpfs_node *parent; if (n == fnode) { TMPFS_NODE_UNLOCK(n); TMPFS_UNLOCK(tmp); error = EINVAL; if (newname != NULL) free(newname, M_TMPFSNAME); goto out_locked; } parent = n->tn_dir.tn_parent; TMPFS_NODE_UNLOCK(n); if (parent == NULL) { n = NULL; break; } TMPFS_NODE_LOCK(parent); if (parent->tn_dir.tn_parent == NULL) { TMPFS_NODE_UNLOCK(parent); n = NULL; break; } n = parent; } TMPFS_UNLOCK(tmp); if (n == NULL) { error = EINVAL; if (newname != NULL) free(newname, M_TMPFSNAME); goto out_locked; } TMPFS_NODE_UNLOCK(n); /* Adjust the parent pointer. */ TMPFS_VALIDATE_DIR(fnode); TMPFS_NODE_LOCK(de->td_node); de->td_node->tn_dir.tn_parent = tdnode; TMPFS_NODE_UNLOCK(de->td_node); /* * As a result of changing the target of the '..' * entry, the link count of the source and target * directories has to be adjusted. */ TMPFS_NODE_LOCK(tdnode); TMPFS_ASSERT_LOCKED(tdnode); tdnode->tn_links++; TMPFS_NODE_UNLOCK(tdnode); TMPFS_NODE_LOCK(fdnode); TMPFS_ASSERT_LOCKED(fdnode); fdnode->tn_links--; TMPFS_NODE_UNLOCK(fdnode); } } /* * Do the move: just remove the entry from the source directory * and insert it into the target one. */ tmpfs_dir_detach(fdvp, de); if (fcnp->cn_flags & DOWHITEOUT) tmpfs_dir_whiteout_add(fdvp, fcnp); if (tcnp->cn_flags & ISWHITEOUT) tmpfs_dir_whiteout_remove(tdvp, tcnp); /* * If the name has changed, we need to make it effective by changing * it in the directory entry. */ if (newname != NULL) { MPASS(tcnp->cn_namelen <= MAXNAMLEN); free(de->ud.td_name, M_TMPFSNAME); de->ud.td_name = newname; tmpfs_dirent_init(de, tcnp->cn_nameptr, tcnp->cn_namelen); fnode->tn_status |= TMPFS_NODE_CHANGED; tdnode->tn_status |= TMPFS_NODE_MODIFIED; } /* * If we are overwriting an entry, we have to remove the old one * from the target directory. */ if (tvp != NULL) { struct tmpfs_dirent *tde; /* Remove the old entry from the target directory. */ tde = tmpfs_dir_lookup(tdnode, tnode, tcnp); tmpfs_dir_detach(tdvp, tde); /* * Free the directory entry we just deleted. Note that the * node referred by it will not be removed until the vnode is * really reclaimed. */ tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), tde); } tmpfs_dir_attach(tdvp, de); if (tmpfs_use_nc(fvp)) { cache_vop_rename(fdvp, fvp, tdvp, tvp, fcnp, tcnp); } error = 0; out_locked: if (fdvp != tdvp && fdvp != tvp) VOP_UNLOCK(fdvp); out: if (want_seqc_end) { if (tvp != NULL) vn_seqc_write_end(tvp); vn_seqc_write_end(tdvp); vn_seqc_write_end(fvp); vn_seqc_write_end(fdvp); } /* * Release target nodes. * XXX: I don't understand when tdvp can be the same as tvp, but * other code takes care of this... */ if (tdvp == tvp) vrele(tdvp); else vput(tdvp); if (tvp != NULL) vput(tvp); /* Release source nodes. */ vrele(fdvp); vrele(fvp); return (error); } static int tmpfs_mkdir(struct vop_mkdir_args *v) { struct vnode *dvp = v->a_dvp; struct vnode **vpp = v->a_vpp; struct componentname *cnp = v->a_cnp; struct vattr *vap = v->a_vap; MPASS(vap->va_type == VDIR); return (tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL)); } static int tmpfs_rmdir(struct vop_rmdir_args *v) { struct vnode *dvp = v->a_dvp; struct vnode *vp = v->a_vp; int error; struct tmpfs_dirent *de; struct tmpfs_mount *tmp; struct tmpfs_node *dnode; struct tmpfs_node *node; MPASS(VOP_ISLOCKED(dvp)); MPASS(VOP_ISLOCKED(vp)); tmp = VFS_TO_TMPFS(dvp->v_mount); dnode = VP_TO_TMPFS_DIR(dvp); node = VP_TO_TMPFS_DIR(vp); /* Directories with more than two entries ('.' and '..') cannot be * removed. */ if (node->tn_size > 0) { error = ENOTEMPTY; goto out; } if ((dnode->tn_flags & APPEND) || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) { error = EPERM; goto out; } /* This invariant holds only if we are not trying to remove "..". * We checked for that above so this is safe now. */ MPASS(node->tn_dir.tn_parent == dnode); /* Get the directory entry associated with node (vp). This was * filled by tmpfs_lookup while looking up the entry. */ de = tmpfs_dir_lookup(dnode, node, v->a_cnp); MPASS(TMPFS_DIRENT_MATCHES(de, v->a_cnp->cn_nameptr, v->a_cnp->cn_namelen)); /* Check flags to see if we are allowed to remove the directory. */ if ((dnode->tn_flags & APPEND) != 0 || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) != 0) { error = EPERM; goto out; } /* Detach the directory entry from the directory (dnode). */ tmpfs_dir_detach(dvp, de); if (v->a_cnp->cn_flags & DOWHITEOUT) tmpfs_dir_whiteout_add(dvp, v->a_cnp); /* No vnode should be allocated for this entry from this point */ TMPFS_NODE_LOCK(node); node->tn_links--; node->tn_dir.tn_parent = NULL; node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; node->tn_accessed = true; TMPFS_NODE_UNLOCK(node); TMPFS_NODE_LOCK(dnode); dnode->tn_links--; dnode->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; dnode->tn_accessed = true; TMPFS_NODE_UNLOCK(dnode); if (tmpfs_use_nc(dvp)) { cache_vop_rmdir(dvp, vp); } /* Free the directory entry we just deleted. Note that the node * referred by it will not be removed until the vnode is really * reclaimed. */ tmpfs_free_dirent(tmp, de); /* Release the deleted vnode (will destroy the node, notify * interested parties and clean it from the cache). */ dnode->tn_status |= TMPFS_NODE_CHANGED; tmpfs_update(dvp); error = 0; out: return (error); } static int tmpfs_symlink(struct vop_symlink_args *v) { struct vnode *dvp = v->a_dvp; struct vnode **vpp = v->a_vpp; struct componentname *cnp = v->a_cnp; struct vattr *vap = v->a_vap; const char *target = v->a_target; #ifdef notyet /* XXX FreeBSD BUG: kern_symlink is not setting VLNK */ MPASS(vap->va_type == VLNK); #else vap->va_type = VLNK; #endif return (tmpfs_alloc_file(dvp, vpp, vap, cnp, target)); } static int tmpfs_readdir(struct vop_readdir_args *va) { struct vnode *vp; struct uio *uio; struct tmpfs_mount *tm; struct tmpfs_node *node; u_long **cookies; int *eofflag, *ncookies; ssize_t startresid; int error, maxcookies; vp = va->a_vp; uio = va->a_uio; eofflag = va->a_eofflag; cookies = va->a_cookies; ncookies = va->a_ncookies; /* This operation only makes sense on directory nodes. */ if (vp->v_type != VDIR) return (ENOTDIR); maxcookies = 0; node = VP_TO_TMPFS_DIR(vp); tm = VFS_TO_TMPFS(vp->v_mount); startresid = uio->uio_resid; /* Allocate cookies for NFS and compat modules. */ if (cookies != NULL && ncookies != NULL) { maxcookies = howmany(node->tn_size, sizeof(struct tmpfs_dirent)) + 2; *cookies = malloc(maxcookies * sizeof(**cookies), M_TEMP, M_WAITOK); *ncookies = 0; } if (cookies == NULL) error = tmpfs_dir_getdents(tm, node, uio, 0, NULL, NULL); else error = tmpfs_dir_getdents(tm, node, uio, maxcookies, *cookies, ncookies); /* Buffer was filled without hitting EOF. */ if (error == EJUSTRETURN) error = (uio->uio_resid != startresid) ? 0 : EINVAL; if (error != 0 && cookies != NULL && ncookies != NULL) { free(*cookies, M_TEMP); *cookies = NULL; *ncookies = 0; } if (eofflag != NULL) *eofflag = (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF); return (error); } static int tmpfs_readlink(struct vop_readlink_args *v) { struct vnode *vp = v->a_vp; struct uio *uio = v->a_uio; int error; struct tmpfs_node *node; MPASS(uio->uio_offset == 0); MPASS(vp->v_type == VLNK); node = VP_TO_TMPFS_NODE(vp); error = uiomove(node->tn_link_target, MIN(node->tn_size, uio->uio_resid), uio); tmpfs_set_accessed(VFS_TO_TMPFS(vp->v_mount), node); return (error); } /* * VOP_FPLOOKUP_SYMLINK routines are subject to special circumstances, see * the comment above cache_fplookup for details. * * Check tmpfs_alloc_node for tmpfs-specific synchronisation notes. */ static int tmpfs_fplookup_symlink(struct vop_fplookup_symlink_args *v) { struct vnode *vp; struct tmpfs_node *node; char *symlink; vp = v->a_vp; node = VP_TO_TMPFS_NODE_SMR(vp); if (__predict_false(node == NULL)) return (EAGAIN); if (!atomic_load_char(&node->tn_link_smr)) return (EAGAIN); symlink = atomic_load_ptr(&node->tn_link_target); if (symlink == NULL) return (EAGAIN); return (cache_symlink_resolve(v->a_fpl, symlink, node->tn_size)); } static int tmpfs_inactive(struct vop_inactive_args *v) { struct vnode *vp; struct tmpfs_node *node; vp = v->a_vp; node = VP_TO_TMPFS_NODE(vp); if (node->tn_links == 0) vrecycle(vp); else tmpfs_check_mtime(vp); return (0); } static int tmpfs_need_inactive(struct vop_need_inactive_args *ap) { struct vnode *vp; struct tmpfs_node *node; struct vm_object *obj; vp = ap->a_vp; node = VP_TO_TMPFS_NODE(vp); if (node->tn_links == 0) goto need; if (vp->v_type == VREG) { obj = vp->v_object; if (obj->generation != obj->cleangeneration) goto need; } return (0); need: return (1); } int tmpfs_reclaim(struct vop_reclaim_args *v) { struct vnode *vp; struct tmpfs_mount *tmp; struct tmpfs_node *node; bool unlock; vp = v->a_vp; node = VP_TO_TMPFS_NODE(vp); tmp = VFS_TO_TMPFS(vp->v_mount); if (vp->v_type == VREG) tmpfs_destroy_vobject(vp, node->tn_reg.tn_aobj); vp->v_object = NULL; TMPFS_LOCK(tmp); TMPFS_NODE_LOCK(node); tmpfs_free_vp(vp); /* * If the node referenced by this vnode was deleted by the user, * we must free its associated data structures (now that the vnode * is being reclaimed). */ unlock = true; if (node->tn_links == 0 && (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0) { node->tn_vpstate = TMPFS_VNODE_DOOMED; unlock = !tmpfs_free_node_locked(tmp, node, true); } if (unlock) { TMPFS_NODE_UNLOCK(node); TMPFS_UNLOCK(tmp); } MPASS(vp->v_data == NULL); return (0); } int tmpfs_print(struct vop_print_args *v) { struct vnode *vp = v->a_vp; struct tmpfs_node *node; node = VP_TO_TMPFS_NODE(vp); printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%lx, links %jd\n", node, node->tn_flags, (uintmax_t)node->tn_links); printf("\tmode 0%o, owner %d, group %d, size %jd, status 0x%x\n", node->tn_mode, node->tn_uid, node->tn_gid, (intmax_t)node->tn_size, node->tn_status); if (vp->v_type == VFIFO) fifo_printinfo(vp); printf("\n"); return (0); } int tmpfs_pathconf(struct vop_pathconf_args *v) { struct vnode *vp = v->a_vp; int name = v->a_name; long *retval = v->a_retval; int error; error = 0; switch (name) { case _PC_LINK_MAX: *retval = TMPFS_LINK_MAX; break; case _PC_SYMLINK_MAX: *retval = MAXPATHLEN; break; case _PC_NAME_MAX: *retval = NAME_MAX; break; case _PC_PIPE_BUF: if (vp->v_type == VDIR || vp->v_type == VFIFO) *retval = PIPE_BUF; else error = EINVAL; break; case _PC_CHOWN_RESTRICTED: *retval = 1; break; case _PC_NO_TRUNC: *retval = 1; break; case _PC_SYNC_IO: *retval = 1; break; case _PC_FILESIZEBITS: *retval = 64; break; default: error = vop_stdpathconf(v); } return (error); } static int tmpfs_vptofh(struct vop_vptofh_args *ap) /* vop_vptofh { IN struct vnode *a_vp; IN struct fid *a_fhp; }; */ { struct tmpfs_fid_data tfd; struct tmpfs_node *node; struct fid *fhp; node = VP_TO_TMPFS_NODE(ap->a_vp); fhp = ap->a_fhp; fhp->fid_len = sizeof(tfd); /* * Copy into fid_data from the stack to avoid unaligned pointer use. * See the comment in sys/mount.h on struct fid for details. */ tfd.tfd_id = node->tn_id; tfd.tfd_gen = node->tn_gen; memcpy(fhp->fid_data, &tfd, fhp->fid_len); return (0); } static int tmpfs_whiteout(struct vop_whiteout_args *ap) { struct vnode *dvp = ap->a_dvp; struct componentname *cnp = ap->a_cnp; struct tmpfs_dirent *de; switch (ap->a_flags) { case LOOKUP: return (0); case CREATE: de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp); if (de != NULL) return (de->td_node == NULL ? 0 : EEXIST); return (tmpfs_dir_whiteout_add(dvp, cnp)); case DELETE: tmpfs_dir_whiteout_remove(dvp, cnp); return (0); default: panic("tmpfs_whiteout: unknown op"); } } static int tmpfs_vptocnp_dir(struct tmpfs_node *tn, struct tmpfs_node *tnp, struct tmpfs_dirent **pde) { struct tmpfs_dir_cursor dc; struct tmpfs_dirent *de; for (de = tmpfs_dir_first(tnp, &dc); de != NULL; de = tmpfs_dir_next(tnp, &dc)) { if (de->td_node == tn) { *pde = de; return (0); } } return (ENOENT); } static int tmpfs_vptocnp_fill(struct vnode *vp, struct tmpfs_node *tn, struct tmpfs_node *tnp, char *buf, size_t *buflen, struct vnode **dvp) { struct tmpfs_dirent *de; int error, i; error = vn_vget_ino_gen(vp, tmpfs_vn_get_ino_alloc, tnp, LK_SHARED, dvp); if (error != 0) return (error); error = tmpfs_vptocnp_dir(tn, tnp, &de); if (error == 0) { i = *buflen; i -= de->td_namelen; if (i < 0) { error = ENOMEM; } else { bcopy(de->ud.td_name, buf + i, de->td_namelen); *buflen = i; } } if (error == 0) { if (vp != *dvp) VOP_UNLOCK(*dvp); } else { if (vp != *dvp) vput(*dvp); else vrele(vp); } return (error); } static int tmpfs_vptocnp(struct vop_vptocnp_args *ap) { struct vnode *vp, **dvp; struct tmpfs_node *tn, *tnp, *tnp1; struct tmpfs_dirent *de; struct tmpfs_mount *tm; char *buf; size_t *buflen; int error; vp = ap->a_vp; dvp = ap->a_vpp; buf = ap->a_buf; buflen = ap->a_buflen; tm = VFS_TO_TMPFS(vp->v_mount); tn = VP_TO_TMPFS_NODE(vp); if (tn->tn_type == VDIR) { tnp = tn->tn_dir.tn_parent; if (tnp == NULL) return (ENOENT); tmpfs_ref_node(tnp); error = tmpfs_vptocnp_fill(vp, tn, tn->tn_dir.tn_parent, buf, buflen, dvp); tmpfs_free_node(tm, tnp); return (error); } restart: TMPFS_LOCK(tm); restart_locked: LIST_FOREACH_SAFE(tnp, &tm->tm_nodes_used, tn_entries, tnp1) { if (tnp->tn_type != VDIR) continue; TMPFS_NODE_LOCK(tnp); tmpfs_ref_node(tnp); /* * tn_vnode cannot be instantiated while we hold the * node lock, so the directory cannot be changed while * we iterate over it. Do this to avoid instantiating * vnode for directories which cannot point to our * node. */ error = tnp->tn_vnode == NULL ? tmpfs_vptocnp_dir(tn, tnp, &de) : 0; if (error == 0) { TMPFS_NODE_UNLOCK(tnp); TMPFS_UNLOCK(tm); error = tmpfs_vptocnp_fill(vp, tn, tnp, buf, buflen, dvp); if (error == 0) { tmpfs_free_node(tm, tnp); return (0); } if (VN_IS_DOOMED(vp)) { tmpfs_free_node(tm, tnp); return (ENOENT); } TMPFS_LOCK(tm); TMPFS_NODE_LOCK(tnp); } if (tmpfs_free_node_locked(tm, tnp, false)) { goto restart; } else { KASSERT(tnp->tn_refcount > 0, ("node %p refcount zero", tnp)); if (tnp->tn_attached) { tnp1 = LIST_NEXT(tnp, tn_entries); TMPFS_NODE_UNLOCK(tnp); } else { TMPFS_NODE_UNLOCK(tnp); goto restart_locked; } } } TMPFS_UNLOCK(tm); return (ENOENT); } /* * Vnode operations vector used for files stored in a tmpfs file system. */ struct vop_vector tmpfs_vnodeop_entries = { .vop_default = &default_vnodeops, .vop_lookup = vfs_cache_lookup, .vop_cachedlookup = tmpfs_cached_lookup, .vop_create = tmpfs_create, .vop_mknod = tmpfs_mknod, .vop_open = tmpfs_open, .vop_close = tmpfs_close, .vop_fplookup_vexec = tmpfs_fplookup_vexec, .vop_fplookup_symlink = tmpfs_fplookup_symlink, .vop_access = tmpfs_access, .vop_stat = tmpfs_stat, .vop_getattr = tmpfs_getattr, .vop_setattr = tmpfs_setattr, .vop_read = tmpfs_read, .vop_read_pgcache = tmpfs_read_pgcache, .vop_write = tmpfs_write, .vop_deallocate = tmpfs_deallocate, .vop_fsync = tmpfs_fsync, .vop_remove = tmpfs_remove, .vop_link = tmpfs_link, .vop_rename = tmpfs_rename, .vop_mkdir = tmpfs_mkdir, .vop_rmdir = tmpfs_rmdir, .vop_symlink = tmpfs_symlink, .vop_readdir = tmpfs_readdir, .vop_readlink = tmpfs_readlink, .vop_inactive = tmpfs_inactive, .vop_need_inactive = tmpfs_need_inactive, .vop_reclaim = tmpfs_reclaim, .vop_print = tmpfs_print, .vop_pathconf = tmpfs_pathconf, .vop_vptofh = tmpfs_vptofh, .vop_whiteout = tmpfs_whiteout, .vop_bmap = VOP_EOPNOTSUPP, .vop_vptocnp = tmpfs_vptocnp, .vop_lock1 = vop_lock, .vop_unlock = vop_unlock, .vop_islocked = vop_islocked, }; VFS_VOP_VECTOR_REGISTER(tmpfs_vnodeop_entries); /* * Same vector for mounts which do not use namecache. */ struct vop_vector tmpfs_vnodeop_nonc_entries = { .vop_default = &tmpfs_vnodeop_entries, .vop_lookup = tmpfs_lookup, }; VFS_VOP_VECTOR_REGISTER(tmpfs_vnodeop_nonc_entries); diff --git a/sys/fs/unionfs/union_subr.c b/sys/fs/unionfs/union_subr.c index dcdb41e55258..eb7681df645b 100644 --- a/sys/fs/unionfs/union_subr.c +++ b/sys/fs/unionfs/union_subr.c @@ -1,1336 +1,1333 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1994 Jan-Simon Pendry * Copyright (c) 1994 * The Regents of the University of California. All rights reserved. * Copyright (c) 2005, 2006, 2012 Masanori Ozawa , ONGS Inc. * Copyright (c) 2006, 2012 Daichi Goto * * This code is derived from software contributed to Berkeley by * Jan-Simon Pendry. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)union_subr.c 8.20 (Berkeley) 5/20/95 * $FreeBSD$ */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define NUNIONFSNODECACHE 16 static MALLOC_DEFINE(M_UNIONFSHASH, "UNIONFS hash", "UNIONFS hash table"); MALLOC_DEFINE(M_UNIONFSNODE, "UNIONFS node", "UNIONFS vnode private part"); MALLOC_DEFINE(M_UNIONFSPATH, "UNIONFS path", "UNIONFS path private part"); static struct task unionfs_deferred_rele_task; static struct mtx unionfs_deferred_rele_lock; static STAILQ_HEAD(, unionfs_node) unionfs_deferred_rele_list = STAILQ_HEAD_INITIALIZER(unionfs_deferred_rele_list); static TASKQUEUE_DEFINE_THREAD(unionfs_rele); unsigned int unionfs_ndeferred = 0; SYSCTL_UINT(_vfs, OID_AUTO, unionfs_ndeferred, CTLFLAG_RD, &unionfs_ndeferred, 0, "unionfs deferred vnode release"); static void unionfs_deferred_rele(void *, int); /* * Initialize */ int unionfs_init(struct vfsconf *vfsp) { UNIONFSDEBUG("unionfs_init\n"); /* printed during system boot */ TASK_INIT(&unionfs_deferred_rele_task, 0, unionfs_deferred_rele, NULL); mtx_init(&unionfs_deferred_rele_lock, "uniondefr", NULL, MTX_DEF); return (0); } /* * Uninitialize */ int unionfs_uninit(struct vfsconf *vfsp) { taskqueue_quiesce(taskqueue_unionfs_rele); taskqueue_free(taskqueue_unionfs_rele); mtx_destroy(&unionfs_deferred_rele_lock); return (0); } static void unionfs_deferred_rele(void *arg __unused, int pending __unused) { STAILQ_HEAD(, unionfs_node) local_rele_list; struct unionfs_node *unp, *tunp; unsigned int ndeferred; ndeferred = 0; STAILQ_INIT(&local_rele_list); mtx_lock(&unionfs_deferred_rele_lock); STAILQ_CONCAT(&local_rele_list, &unionfs_deferred_rele_list); mtx_unlock(&unionfs_deferred_rele_lock); STAILQ_FOREACH_SAFE(unp, &local_rele_list, un_rele, tunp) { ++ndeferred; MPASS(unp->un_dvp != NULL); vrele(unp->un_dvp); free(unp, M_UNIONFSNODE); } /* We expect this function to be single-threaded, thus no atomic */ unionfs_ndeferred += ndeferred; } static struct unionfs_node_hashhead * unionfs_get_hashhead(struct vnode *dvp, char *path) { struct unionfs_node *unp; int count; char hash; hash = 0; unp = VTOUNIONFS(dvp); if (path != NULL) { for (count = 0; path[count]; count++) hash += path[count]; } return (&(unp->un_hashtbl[hash & (unp->un_hashmask)])); } /* * Get the cached vnode. */ static struct vnode * unionfs_get_cached_vnode(struct vnode *uvp, struct vnode *lvp, struct vnode *dvp, char *path) { struct unionfs_node_hashhead *hd; struct unionfs_node *unp; struct vnode *vp; KASSERT((uvp == NULLVP || uvp->v_type == VDIR), ("unionfs_get_cached_vnode: v_type != VDIR")); KASSERT((lvp == NULLVP || lvp->v_type == VDIR), ("unionfs_get_cached_vnode: v_type != VDIR")); VI_LOCK(dvp); hd = unionfs_get_hashhead(dvp, path); LIST_FOREACH(unp, hd, un_hash) { if (!strcmp(unp->un_path, path)) { vp = UNIONFSTOV(unp); VI_LOCK_FLAGS(vp, MTX_DUPOK); VI_UNLOCK(dvp); vp->v_iflag &= ~VI_OWEINACT; if (VN_IS_DOOMED(vp) || ((vp->v_iflag & VI_DOINGINACT) != 0)) { VI_UNLOCK(vp); vp = NULLVP; } else VI_UNLOCK(vp); return (vp); } } VI_UNLOCK(dvp); return (NULLVP); } /* * Add the new vnode into cache. */ static struct vnode * unionfs_ins_cached_vnode(struct unionfs_node *uncp, struct vnode *dvp, char *path) { struct unionfs_node_hashhead *hd; struct unionfs_node *unp; struct vnode *vp; KASSERT((uncp->un_uppervp==NULLVP || uncp->un_uppervp->v_type==VDIR), ("unionfs_ins_cached_vnode: v_type != VDIR")); KASSERT((uncp->un_lowervp==NULLVP || uncp->un_lowervp->v_type==VDIR), ("unionfs_ins_cached_vnode: v_type != VDIR")); VI_LOCK(dvp); hd = unionfs_get_hashhead(dvp, path); LIST_FOREACH(unp, hd, un_hash) { if (!strcmp(unp->un_path, path)) { vp = UNIONFSTOV(unp); VI_LOCK_FLAGS(vp, MTX_DUPOK); vp->v_iflag &= ~VI_OWEINACT; if (VN_IS_DOOMED(vp) || ((vp->v_iflag & VI_DOINGINACT) != 0)) { LIST_INSERT_HEAD(hd, uncp, un_hash); VI_UNLOCK(vp); vp = NULLVP; } else VI_UNLOCK(vp); VI_UNLOCK(dvp); return (vp); } } LIST_INSERT_HEAD(hd, uncp, un_hash); VI_UNLOCK(dvp); return (NULLVP); } /* * Remove the vnode. */ static void unionfs_rem_cached_vnode(struct unionfs_node *unp, struct vnode *dvp) { KASSERT((unp != NULL), ("unionfs_rem_cached_vnode: null node")); KASSERT((dvp != NULLVP), ("unionfs_rem_cached_vnode: null parent vnode")); KASSERT((unp->un_hash.le_prev != NULL), ("unionfs_rem_cached_vnode: null hash")); VI_LOCK(dvp); LIST_REMOVE(unp, un_hash); unp->un_hash.le_next = NULL; unp->un_hash.le_prev = NULL; VI_UNLOCK(dvp); } /* * Common cleanup handling for unionfs_nodeget * Upper, lower, and parent directory vnodes are expected to be referenced by * the caller. Upper and lower vnodes, if non-NULL, are also expected to be * exclusively locked by the caller. * This function will return with the caller's locks and references undone. */ static void unionfs_nodeget_cleanup(struct vnode *vp, void *arg) { struct unionfs_node *unp; /* * Lock and reset the default vnode lock; vgone() expects a locked * vnode, and we're going to reset the vnode ops. */ lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL); /* * Clear out private data and reset the vnode ops to avoid use of * unionfs vnode ops on a partially constructed vnode. */ VI_LOCK(vp); vp->v_data = NULL; vp->v_vnlock = &vp->v_lock; vp->v_op = &dead_vnodeops; VI_UNLOCK(vp); vgone(vp); vput(vp); unp = arg; if (unp->un_dvp != NULLVP) vrele(unp->un_dvp); if (unp->un_uppervp != NULLVP) vput(unp->un_uppervp); if (unp->un_lowervp != NULLVP) vput(unp->un_lowervp); if (unp->un_hashtbl != NULL) hashdestroy(unp->un_hashtbl, M_UNIONFSHASH, unp->un_hashmask); free(unp->un_path, M_UNIONFSPATH); free(unp, M_UNIONFSNODE); } /* * Make a new or get existing unionfs node. * * uppervp and lowervp should be unlocked. Because if new unionfs vnode is * locked, uppervp or lowervp is locked too. In order to prevent dead lock, * you should not lock plurality simultaneously. */ int unionfs_nodeget(struct mount *mp, struct vnode *uppervp, struct vnode *lowervp, struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct thread *td) { char *path; struct unionfs_mount *ump; struct unionfs_node *unp; struct vnode *vp; int error; int lkflags; enum vtype vt; error = 0; ump = MOUNTTOUNIONFSMOUNT(mp); lkflags = (cnp ? cnp->cn_lkflags : 0); path = (cnp ? cnp->cn_nameptr : NULL); *vpp = NULLVP; if (uppervp == NULLVP && lowervp == NULLVP) panic("unionfs_nodeget: upper and lower is null"); vt = (uppervp != NULLVP ? uppervp->v_type : lowervp->v_type); /* If it has no ISLASTCN flag, path check is skipped. */ if (cnp && !(cnp->cn_flags & ISLASTCN)) path = NULL; /* check the cache */ if (path != NULL && dvp != NULLVP && vt == VDIR) { vp = unionfs_get_cached_vnode(uppervp, lowervp, dvp, path); if (vp != NULLVP) { vref(vp); *vpp = vp; goto unionfs_nodeget_out; } } if ((uppervp == NULLVP || ump->um_uppervp != uppervp) || (lowervp == NULLVP || ump->um_lowervp != lowervp)) { /* dvp will be NULLVP only in case of root vnode. */ if (dvp == NULLVP) return (EINVAL); } unp = malloc(sizeof(struct unionfs_node), M_UNIONFSNODE, M_WAITOK | M_ZERO); error = getnewvnode("unionfs", mp, &unionfs_vnodeops, &vp); if (error != 0) { free(unp, M_UNIONFSNODE); return (error); } if (dvp != NULLVP) vref(dvp); if (uppervp != NULLVP) vref(uppervp); if (lowervp != NULLVP) vref(lowervp); if (vt == VDIR) unp->un_hashtbl = hashinit(NUNIONFSNODECACHE, M_UNIONFSHASH, &(unp->un_hashmask)); unp->un_vnode = vp; unp->un_uppervp = uppervp; unp->un_lowervp = lowervp; unp->un_dvp = dvp; if (uppervp != NULLVP) vp->v_vnlock = uppervp->v_vnlock; else vp->v_vnlock = lowervp->v_vnlock; if (path != NULL) { unp->un_path = malloc(cnp->cn_namelen + 1, M_UNIONFSPATH, M_WAITOK | M_ZERO); bcopy(cnp->cn_nameptr, unp->un_path, cnp->cn_namelen); unp->un_path[cnp->cn_namelen] = '\0'; unp->un_pathlen = cnp->cn_namelen; } vp->v_type = vt; vp->v_data = unp; if ((uppervp != NULLVP && ump->um_uppervp == uppervp) && (lowervp != NULLVP && ump->um_lowervp == lowervp)) vp->v_vflag |= VV_ROOT; vn_lock_pair(lowervp, false, uppervp, false); error = insmntque1(vp, mp, unionfs_nodeget_cleanup, unp); if (error != 0) return (error); if (lowervp != NULL && VN_IS_DOOMED(lowervp)) { vput(lowervp); unp->un_lowervp = NULL; } if (uppervp != NULL && VN_IS_DOOMED(uppervp)) { vput(uppervp); unp->un_uppervp = NULL; } if (unp->un_lowervp == NULL && unp->un_uppervp == NULL) { unionfs_nodeget_cleanup(vp, unp); return (ENOENT); } if (path != NULL && dvp != NULLVP && vt == VDIR) *vpp = unionfs_ins_cached_vnode(unp, dvp, path); if (*vpp != NULLVP) { unionfs_nodeget_cleanup(vp, unp); vp = *vpp; vref(vp); } else { if (uppervp != NULL) VOP_UNLOCK(uppervp); if (lowervp != NULL) VOP_UNLOCK(lowervp); *vpp = vp; } unionfs_nodeget_out: if (lkflags & LK_TYPE_MASK) vn_lock(vp, lkflags | LK_RETRY); return (0); } /* * Clean up the unionfs node. */ void unionfs_noderem(struct vnode *vp, struct thread *td) { struct unionfs_node *unp, *unp_t1, *unp_t2; struct unionfs_node_hashhead *hd; struct unionfs_node_status *unsp, *unsp_tmp; struct vnode *lvp; struct vnode *uvp; struct vnode *dvp; int count; /* * Use the interlock to protect the clearing of v_data to * prevent faults in unionfs_lock(). */ VI_LOCK(vp); unp = VTOUNIONFS(vp); lvp = unp->un_lowervp; uvp = unp->un_uppervp; dvp = unp->un_dvp; unp->un_lowervp = unp->un_uppervp = NULLVP; vp->v_vnlock = &(vp->v_lock); vp->v_data = NULL; vp->v_object = NULL; if (vp->v_writecount > 0) { if (uvp != NULL) VOP_ADD_WRITECOUNT(uvp, -vp->v_writecount); else if (lvp != NULL) VOP_ADD_WRITECOUNT(lvp, -vp->v_writecount); } else if (vp->v_writecount < 0) vp->v_writecount = 0; VI_UNLOCK(vp); if (lvp != NULLVP) VOP_UNLOCK(lvp); if (uvp != NULLVP) VOP_UNLOCK(uvp); if (dvp != NULLVP && unp->un_hash.le_prev != NULL) unionfs_rem_cached_vnode(unp, dvp); if (lockmgr(vp->v_vnlock, LK_EXCLUSIVE, VI_MTX(vp)) != 0) panic("the lock for deletion is unacquirable."); if (lvp != NULLVP) vrele(lvp); if (uvp != NULLVP) vrele(uvp); if (unp->un_path != NULL) { free(unp->un_path, M_UNIONFSPATH); unp->un_path = NULL; unp->un_pathlen = 0; } if (unp->un_hashtbl != NULL) { for (count = 0; count <= unp->un_hashmask; count++) { hd = unp->un_hashtbl + count; LIST_FOREACH_SAFE(unp_t1, hd, un_hash, unp_t2) { LIST_REMOVE(unp_t1, un_hash); unp_t1->un_hash.le_next = NULL; unp_t1->un_hash.le_prev = NULL; } } hashdestroy(unp->un_hashtbl, M_UNIONFSHASH, unp->un_hashmask); } LIST_FOREACH_SAFE(unsp, &(unp->un_unshead), uns_list, unsp_tmp) { LIST_REMOVE(unsp, uns_list); free(unsp, M_TEMP); } if (dvp != NULLVP) { mtx_lock(&unionfs_deferred_rele_lock); STAILQ_INSERT_TAIL(&unionfs_deferred_rele_list, unp, un_rele); mtx_unlock(&unionfs_deferred_rele_lock); taskqueue_enqueue(taskqueue_unionfs_rele, &unionfs_deferred_rele_task); } else free(unp, M_UNIONFSNODE); } /* * Get the unionfs node status. * You need exclusive lock this vnode. */ void unionfs_get_node_status(struct unionfs_node *unp, struct thread *td, struct unionfs_node_status **unspp) { struct unionfs_node_status *unsp; pid_t pid; pid = td->td_proc->p_pid; KASSERT(NULL != unspp, ("null pointer")); ASSERT_VOP_ELOCKED(UNIONFSTOV(unp), "unionfs_get_node_status"); LIST_FOREACH(unsp, &(unp->un_unshead), uns_list) { if (unsp->uns_pid == pid) { *unspp = unsp; return; } } /* create a new unionfs node status */ unsp = malloc(sizeof(struct unionfs_node_status), M_TEMP, M_WAITOK | M_ZERO); unsp->uns_pid = pid; LIST_INSERT_HEAD(&(unp->un_unshead), unsp, uns_list); *unspp = unsp; } /* * Remove the unionfs node status, if you can. * You need exclusive lock this vnode. */ void unionfs_tryrem_node_status(struct unionfs_node *unp, struct unionfs_node_status *unsp) { KASSERT(NULL != unsp, ("null pointer")); ASSERT_VOP_ELOCKED(UNIONFSTOV(unp), "unionfs_get_node_status"); if (0 < unsp->uns_lower_opencnt || 0 < unsp->uns_upper_opencnt) return; LIST_REMOVE(unsp, uns_list); free(unsp, M_TEMP); } /* * Create upper node attr. */ void unionfs_create_uppervattr_core(struct unionfs_mount *ump, struct vattr *lva, struct vattr *uva, struct thread *td) { VATTR_NULL(uva); uva->va_type = lva->va_type; uva->va_atime = lva->va_atime; uva->va_mtime = lva->va_mtime; uva->va_ctime = lva->va_ctime; switch (ump->um_copymode) { case UNIONFS_TRANSPARENT: uva->va_mode = lva->va_mode; uva->va_uid = lva->va_uid; uva->va_gid = lva->va_gid; break; case UNIONFS_MASQUERADE: if (ump->um_uid == lva->va_uid) { uva->va_mode = lva->va_mode & 077077; uva->va_mode |= (lva->va_type == VDIR ? ump->um_udir : ump->um_ufile) & 0700; uva->va_uid = lva->va_uid; uva->va_gid = lva->va_gid; } else { uva->va_mode = (lva->va_type == VDIR ? ump->um_udir : ump->um_ufile); uva->va_uid = ump->um_uid; uva->va_gid = ump->um_gid; } break; default: /* UNIONFS_TRADITIONAL */ uva->va_mode = 0777 & ~td->td_proc->p_pd->pd_cmask; uva->va_uid = ump->um_uid; uva->va_gid = ump->um_gid; break; } } /* * Create upper node attr. */ int unionfs_create_uppervattr(struct unionfs_mount *ump, struct vnode *lvp, struct vattr *uva, struct ucred *cred, struct thread *td) { struct vattr lva; int error; if ((error = VOP_GETATTR(lvp, &lva, cred))) return (error); unionfs_create_uppervattr_core(ump, &lva, uva, td); return (error); } /* * relookup * * dvp should be locked on entry and will be locked on return. * * If an error is returned, *vpp will be invalid, otherwise it will hold a * locked, referenced vnode. If *vpp == dvp then remember that only one * LK_EXCLUSIVE lock is held. */ int unionfs_relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct componentname *cn, struct thread *td, char *path, int pathlen, u_long nameiop) { int error; cn->cn_namelen = pathlen; cn->cn_pnbuf = path; cn->cn_nameiop = nameiop; cn->cn_flags = (LOCKPARENT | LOCKLEAF | HASBUF | SAVENAME | ISLASTCN); cn->cn_lkflags = LK_EXCLUSIVE; - cn->cn_thread = td; cn->cn_cred = cnp->cn_cred; cn->cn_nameptr = cn->cn_pnbuf; if (nameiop == DELETE) cn->cn_flags |= (cnp->cn_flags & (DOWHITEOUT | SAVESTART)); else if (RENAME == nameiop) cn->cn_flags |= (cnp->cn_flags & SAVESTART); else if (nameiop == CREATE) cn->cn_flags |= NOCACHE; vref(dvp); VOP_UNLOCK(dvp); if ((error = relookup(dvp, vpp, cn))) { vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); } else vrele(dvp); KASSERT((cn->cn_flags & HASBUF) != 0, ("%s: HASBUF cleared", __func__)); KASSERT((cn->cn_flags & SAVENAME) != 0, ("%s: SAVENAME cleared", __func__)); KASSERT(cn->cn_pnbuf == path, ("%s: cn_pnbuf changed", __func__)); return (error); } /* * relookup for CREATE namei operation. * * dvp is unionfs vnode. dvp should be locked. * * If it called 'unionfs_copyfile' function by unionfs_link etc, * VOP_LOOKUP information is broken. * So it need relookup in order to create link etc. */ int unionfs_relookup_for_create(struct vnode *dvp, struct componentname *cnp, struct thread *td) { struct vnode *udvp; struct vnode *vp; struct componentname cn; int error; udvp = UNIONFSVPTOUPPERVP(dvp); vp = NULLVP; error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr, cnp->cn_namelen, CREATE); if (error) return (error); if (vp != NULLVP) { if (udvp == vp) vrele(vp); else vput(vp); error = EEXIST; } return (error); } /* * relookup for DELETE namei operation. * * dvp is unionfs vnode. dvp should be locked. */ int unionfs_relookup_for_delete(struct vnode *dvp, struct componentname *cnp, struct thread *td) { struct vnode *udvp; struct vnode *vp; struct componentname cn; int error; udvp = UNIONFSVPTOUPPERVP(dvp); vp = NULLVP; error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr, cnp->cn_namelen, DELETE); if (error) return (error); if (vp == NULLVP) error = ENOENT; else { if (udvp == vp) vrele(vp); else vput(vp); } return (error); } /* * relookup for RENAME namei operation. * * dvp is unionfs vnode. dvp should be locked. */ int unionfs_relookup_for_rename(struct vnode *dvp, struct componentname *cnp, struct thread *td) { struct vnode *udvp; struct vnode *vp; struct componentname cn; int error; udvp = UNIONFSVPTOUPPERVP(dvp); vp = NULLVP; error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr, cnp->cn_namelen, RENAME); if (error) return (error); if (vp != NULLVP) { if (udvp == vp) vrele(vp); else vput(vp); } return (error); } /* * Update the unionfs_node. * * uvp is new locked upper vnode. unionfs vnode's lock will be exchanged to the * uvp's lock and lower's lock will be unlocked. */ static void unionfs_node_update(struct unionfs_node *unp, struct vnode *uvp, struct thread *td) { struct vnode *vp; struct vnode *lvp; struct vnode *dvp; unsigned count, lockrec; vp = UNIONFSTOV(unp); lvp = unp->un_lowervp; ASSERT_VOP_ELOCKED(lvp, "unionfs_node_update"); dvp = unp->un_dvp; /* * lock update */ VI_LOCK(vp); unp->un_uppervp = uvp; vp->v_vnlock = uvp->v_vnlock; VI_UNLOCK(vp); lockrec = lvp->v_vnlock->lk_recurse; for (count = 0; count < lockrec; count++) vn_lock(uvp, LK_EXCLUSIVE | LK_CANRECURSE | LK_RETRY); /* * cache update */ if (unp->un_path != NULL && dvp != NULLVP && vp->v_type == VDIR) { static struct unionfs_node_hashhead *hd; VI_LOCK(dvp); hd = unionfs_get_hashhead(dvp, unp->un_path); LIST_REMOVE(unp, un_hash); LIST_INSERT_HEAD(hd, unp, un_hash); VI_UNLOCK(dvp); } } /* * Create a new shadow dir. * * udvp should be locked on entry and will be locked on return. * * If no error returned, unp will be updated. */ int unionfs_mkshadowdir(struct unionfs_mount *ump, struct vnode *udvp, struct unionfs_node *unp, struct componentname *cnp, struct thread *td) { struct vnode *lvp; struct vnode *uvp; struct vattr va; struct vattr lva; struct nameidata nd; struct mount *mp; struct ucred *cred; struct ucred *credbk; struct uidinfo *rootinfo; int error; if (unp->un_uppervp != NULLVP) return (EEXIST); lvp = unp->un_lowervp; uvp = NULLVP; credbk = cnp->cn_cred; /* Authority change to root */ rootinfo = uifind((uid_t)0); cred = crdup(cnp->cn_cred); /* * The calls to chgproccnt() are needed to compensate for change_ruid() * calling chgproccnt(). */ chgproccnt(cred->cr_ruidinfo, 1, 0); change_euid(cred, rootinfo); change_ruid(cred, rootinfo); change_svuid(cred, (uid_t)0); uifree(rootinfo); cnp->cn_cred = cred; memset(&nd.ni_cnd, 0, sizeof(struct componentname)); NDPREINIT(&nd); if ((error = VOP_GETATTR(lvp, &lva, cnp->cn_cred))) goto unionfs_mkshadowdir_abort; if ((error = unionfs_relookup(udvp, &uvp, cnp, &nd.ni_cnd, td, cnp->cn_nameptr, cnp->cn_namelen, CREATE))) goto unionfs_mkshadowdir_abort; if (uvp != NULLVP) { if (udvp == uvp) vrele(uvp); else vput(uvp); error = EEXIST; goto unionfs_mkshadowdir_abort; } if ((error = vn_start_write(udvp, &mp, V_WAIT | PCATCH))) goto unionfs_mkshadowdir_abort; unionfs_create_uppervattr_core(ump, &lva, &va, td); error = VOP_MKDIR(udvp, &uvp, &nd.ni_cnd, &va); if (!error) { unionfs_node_update(unp, uvp, td); /* * XXX The bug which cannot set uid/gid was corrected. * Ignore errors. */ va.va_type = VNON; VOP_SETATTR(uvp, &va, nd.ni_cnd.cn_cred); } vn_finished_write(mp); unionfs_mkshadowdir_abort: cnp->cn_cred = credbk; chgproccnt(cred->cr_ruidinfo, -1, 0); crfree(cred); return (error); } /* * Create a new whiteout. * * dvp should be locked on entry and will be locked on return. */ int unionfs_mkwhiteout(struct vnode *dvp, struct componentname *cnp, struct thread *td, char *path, int pathlen) { struct vnode *wvp; struct nameidata nd; struct mount *mp; int error; wvp = NULLVP; NDPREINIT(&nd); if ((error = unionfs_relookup(dvp, &wvp, cnp, &nd.ni_cnd, td, path, pathlen, CREATE))) { return (error); } if (wvp != NULLVP) { if (dvp == wvp) vrele(wvp); else vput(wvp); return (EEXIST); } if ((error = vn_start_write(dvp, &mp, V_WAIT | PCATCH))) goto unionfs_mkwhiteout_free_out; error = VOP_WHITEOUT(dvp, &nd.ni_cnd, CREATE); vn_finished_write(mp); unionfs_mkwhiteout_free_out: return (error); } /* * Create a new vnode for create a new shadow file. * * If an error is returned, *vpp will be invalid, otherwise it will hold a * locked, referenced and opened vnode. * * unp is never updated. */ static int unionfs_vn_create_on_upper(struct vnode **vpp, struct vnode *udvp, struct unionfs_node *unp, struct vattr *uvap, struct thread *td) { struct unionfs_mount *ump; struct vnode *vp; struct vnode *lvp; struct ucred *cred; struct vattr lva; struct nameidata nd; int fmode; int error; ump = MOUNTTOUNIONFSMOUNT(UNIONFSTOV(unp)->v_mount); vp = NULLVP; lvp = unp->un_lowervp; cred = td->td_ucred; fmode = FFLAGS(O_WRONLY | O_CREAT | O_TRUNC | O_EXCL); error = 0; if ((error = VOP_GETATTR(lvp, &lva, cred)) != 0) return (error); unionfs_create_uppervattr_core(ump, &lva, uvap, td); if (unp->un_path == NULL) panic("unionfs: un_path is null"); nd.ni_cnd.cn_namelen = unp->un_pathlen; nd.ni_cnd.cn_pnbuf = unp->un_path; nd.ni_cnd.cn_nameiop = CREATE; nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | HASBUF | SAVENAME | ISLASTCN; nd.ni_cnd.cn_lkflags = LK_EXCLUSIVE; - nd.ni_cnd.cn_thread = td; nd.ni_cnd.cn_cred = cred; nd.ni_cnd.cn_nameptr = nd.ni_cnd.cn_pnbuf; NDPREINIT(&nd); vref(udvp); if ((error = relookup(udvp, &vp, &nd.ni_cnd)) != 0) goto unionfs_vn_create_on_upper_free_out2; vrele(udvp); if (vp != NULLVP) { if (vp == udvp) vrele(vp); else vput(vp); error = EEXIST; goto unionfs_vn_create_on_upper_free_out1; } if ((error = VOP_CREATE(udvp, &vp, &nd.ni_cnd, uvap)) != 0) goto unionfs_vn_create_on_upper_free_out1; if ((error = VOP_OPEN(vp, fmode, cred, td, NULL)) != 0) { vput(vp); goto unionfs_vn_create_on_upper_free_out1; } error = VOP_ADD_WRITECOUNT(vp, 1); CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d", __func__, vp, vp->v_writecount); if (error == 0) { *vpp = vp; } else { VOP_CLOSE(vp, fmode, cred, td); } unionfs_vn_create_on_upper_free_out1: VOP_UNLOCK(udvp); unionfs_vn_create_on_upper_free_out2: KASSERT((nd.ni_cnd.cn_flags & HASBUF) != 0, ("%s: HASBUF cleared", __func__)); KASSERT((nd.ni_cnd.cn_flags & SAVENAME) != 0, ("%s: SAVENAME cleared", __func__)); KASSERT(nd.ni_cnd.cn_pnbuf == unp->un_path, ("%s: cn_pnbuf changed", __func__)); return (error); } /* * Copy from lvp to uvp. * * lvp and uvp should be locked and opened on entry and will be locked and * opened on return. */ static int unionfs_copyfile_core(struct vnode *lvp, struct vnode *uvp, struct ucred *cred, struct thread *td) { char *buf; struct uio uio; struct iovec iov; off_t offset; int count; int error; int bufoffset; error = 0; memset(&uio, 0, sizeof(uio)); uio.uio_td = td; uio.uio_segflg = UIO_SYSSPACE; uio.uio_offset = 0; buf = malloc(MAXBSIZE, M_TEMP, M_WAITOK); while (error == 0) { offset = uio.uio_offset; uio.uio_iov = &iov; uio.uio_iovcnt = 1; iov.iov_base = buf; iov.iov_len = MAXBSIZE; uio.uio_resid = iov.iov_len; uio.uio_rw = UIO_READ; if ((error = VOP_READ(lvp, &uio, 0, cred)) != 0) break; if ((count = MAXBSIZE - uio.uio_resid) == 0) break; bufoffset = 0; while (bufoffset < count) { uio.uio_iov = &iov; uio.uio_iovcnt = 1; iov.iov_base = buf + bufoffset; iov.iov_len = count - bufoffset; uio.uio_offset = offset + bufoffset; uio.uio_resid = iov.iov_len; uio.uio_rw = UIO_WRITE; if ((error = VOP_WRITE(uvp, &uio, 0, cred)) != 0) break; bufoffset += (count - bufoffset) - uio.uio_resid; } uio.uio_offset = offset + bufoffset; } free(buf, M_TEMP); return (error); } /* * Copy file from lower to upper. * * If you need copy of the contents, set 1 to docopy. Otherwise, set 0 to * docopy. * * If no error returned, unp will be updated. */ int unionfs_copyfile(struct unionfs_node *unp, int docopy, struct ucred *cred, struct thread *td) { struct mount *mp; struct vnode *udvp; struct vnode *lvp; struct vnode *uvp; struct vattr uva; int error; lvp = unp->un_lowervp; uvp = NULLVP; if ((UNIONFSTOV(unp)->v_mount->mnt_flag & MNT_RDONLY)) return (EROFS); if (unp->un_dvp == NULLVP) return (EINVAL); if (unp->un_uppervp != NULLVP) return (EEXIST); udvp = VTOUNIONFS(unp->un_dvp)->un_uppervp; if (udvp == NULLVP) return (EROFS); if ((udvp->v_mount->mnt_flag & MNT_RDONLY)) return (EROFS); error = VOP_ACCESS(lvp, VREAD, cred, td); if (error != 0) return (error); if ((error = vn_start_write(udvp, &mp, V_WAIT | PCATCH)) != 0) return (error); error = unionfs_vn_create_on_upper(&uvp, udvp, unp, &uva, td); if (error != 0) { vn_finished_write(mp); return (error); } if (docopy != 0) { error = VOP_OPEN(lvp, FREAD, cred, td, NULL); if (error == 0) { error = unionfs_copyfile_core(lvp, uvp, cred, td); VOP_CLOSE(lvp, FREAD, cred, td); } } VOP_CLOSE(uvp, FWRITE, cred, td); VOP_ADD_WRITECOUNT_CHECKED(uvp, -1); CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d", __func__, uvp, uvp->v_writecount); vn_finished_write(mp); if (error == 0) { /* Reset the attributes. Ignore errors. */ uva.va_type = VNON; VOP_SETATTR(uvp, &uva, cred); } unionfs_node_update(unp, uvp, td); return (error); } /* * It checks whether vp can rmdir. (check empty) * * vp is unionfs vnode. * vp should be locked. */ int unionfs_check_rmdir(struct vnode *vp, struct ucred *cred, struct thread *td) { struct vnode *uvp; struct vnode *lvp; struct vnode *tvp; struct dirent *dp; struct dirent *edp; struct componentname cn; struct iovec iov; struct uio uio; struct vattr va; int error; int eofflag; int lookuperr; /* * The size of buf needs to be larger than DIRBLKSIZ. */ char buf[256 * 6]; ASSERT_VOP_ELOCKED(vp, "unionfs_check_rmdir"); eofflag = 0; uvp = UNIONFSVPTOUPPERVP(vp); lvp = UNIONFSVPTOLOWERVP(vp); /* check opaque */ if ((error = VOP_GETATTR(uvp, &va, cred)) != 0) return (error); if (va.va_flags & OPAQUE) return (0); /* open vnode */ #ifdef MAC if ((error = mac_vnode_check_open(cred, vp, VEXEC|VREAD)) != 0) return (error); #endif if ((error = VOP_ACCESS(vp, VEXEC|VREAD, cred, td)) != 0) return (error); if ((error = VOP_OPEN(vp, FREAD, cred, td, NULL)) != 0) return (error); uio.uio_rw = UIO_READ; uio.uio_segflg = UIO_SYSSPACE; uio.uio_td = td; uio.uio_offset = 0; #ifdef MAC error = mac_vnode_check_readdir(td->td_ucred, lvp); #endif while (!error && !eofflag) { iov.iov_base = buf; iov.iov_len = sizeof(buf); uio.uio_iov = &iov; uio.uio_iovcnt = 1; uio.uio_resid = iov.iov_len; error = VOP_READDIR(lvp, &uio, cred, &eofflag, NULL, NULL); if (error != 0) break; if (eofflag == 0 && uio.uio_resid == sizeof(buf)) { #ifdef DIAGNOSTIC panic("bad readdir response from lower FS."); #endif break; } edp = (struct dirent*)&buf[sizeof(buf) - uio.uio_resid]; for (dp = (struct dirent*)buf; !error && dp < edp; dp = (struct dirent*)((caddr_t)dp + dp->d_reclen)) { if (dp->d_type == DT_WHT || dp->d_fileno == 0 || (dp->d_namlen == 1 && dp->d_name[0] == '.') || (dp->d_namlen == 2 && !bcmp(dp->d_name, "..", 2))) continue; cn.cn_namelen = dp->d_namlen; cn.cn_pnbuf = NULL; cn.cn_nameptr = dp->d_name; cn.cn_nameiop = LOOKUP; cn.cn_flags = LOCKPARENT | LOCKLEAF | SAVENAME | RDONLY | ISLASTCN; cn.cn_lkflags = LK_EXCLUSIVE; - cn.cn_thread = td; cn.cn_cred = cred; /* * check entry in lower. * Sometimes, readdir function returns * wrong entry. */ lookuperr = VOP_LOOKUP(lvp, &tvp, &cn); if (!lookuperr) vput(tvp); else continue; /* skip entry */ /* * check entry * If it has no exist/whiteout entry in upper, * directory is not empty. */ cn.cn_flags = LOCKPARENT | LOCKLEAF | SAVENAME | RDONLY | ISLASTCN; lookuperr = VOP_LOOKUP(uvp, &tvp, &cn); if (!lookuperr) vput(tvp); /* ignore exist or whiteout entry */ if (!lookuperr || (lookuperr == ENOENT && (cn.cn_flags & ISWHITEOUT))) continue; error = ENOTEMPTY; } } /* close vnode */ VOP_CLOSE(vp, FREAD, cred, td); return (error); } #ifdef DIAGNOSTIC struct vnode * unionfs_checkuppervp(struct vnode *vp, char *fil, int lno) { struct unionfs_node *unp; unp = VTOUNIONFS(vp); #ifdef notyet if (vp->v_op != unionfs_vnodeop_p) { printf("unionfs_checkuppervp: on non-unionfs-node.\n"); #ifdef KDB kdb_enter(KDB_WHY_UNIONFS, "unionfs_checkuppervp: on non-unionfs-node.\n"); #endif panic("unionfs_checkuppervp"); } #endif return (unp->un_uppervp); } struct vnode * unionfs_checklowervp(struct vnode *vp, char *fil, int lno) { struct unionfs_node *unp; unp = VTOUNIONFS(vp); #ifdef notyet if (vp->v_op != unionfs_vnodeop_p) { printf("unionfs_checklowervp: on non-unionfs-node.\n"); #ifdef KDB kdb_enter(KDB_WHY_UNIONFS, "unionfs_checklowervp: on non-unionfs-node.\n"); #endif panic("unionfs_checklowervp"); } #endif return (unp->un_lowervp); } #endif diff --git a/sys/kern/uipc_mqueue.c b/sys/kern/uipc_mqueue.c index 5c1775a261fc..a9fb40848fbd 100644 --- a/sys/kern/uipc_mqueue.c +++ b/sys/kern/uipc_mqueue.c @@ -1,2945 +1,2945 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2005 David Xu * Copyright (c) 2016-2017 Robert N. M. Watson * All rights reserved. * * Portions of this software were developed by BAE Systems, the University of * Cambridge Computer Laboratory, and Memorial University under DARPA/AFRL * contract FA8650-15-C-7558 ("CADETS"), as part of the DARPA Transparent * Computing (TC) research program. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * */ /* * POSIX message queue implementation. * * 1) A mqueue filesystem can be mounted, each message queue appears * in mounted directory, user can change queue's permission and * ownership, or remove a queue. Manually creating a file in the * directory causes a message queue to be created in the kernel with * default message queue attributes applied and same name used, this * method is not advocated since mq_open syscall allows user to specify * different attributes. Also the file system can be mounted multiple * times at different mount points but shows same contents. * * 2) Standard POSIX message queue API. The syscalls do not use vfs layer, * but directly operate on internal data structure, this allows user to * use the IPC facility without having to mount mqueue file system. */ #include __FBSDID("$FreeBSD$"); #include "opt_capsicum.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include FEATURE(p1003_1b_mqueue, "POSIX P1003.1B message queues support"); /* * Limits and constants */ #define MQFS_NAMELEN NAME_MAX #define MQFS_DELEN (8 + MQFS_NAMELEN) /* node types */ typedef enum { mqfstype_none = 0, mqfstype_root, mqfstype_dir, mqfstype_this, mqfstype_parent, mqfstype_file, mqfstype_symlink, } mqfs_type_t; struct mqfs_node; /* * mqfs_info: describes a mqfs instance */ struct mqfs_info { struct sx mi_lock; struct mqfs_node *mi_root; struct unrhdr *mi_unrhdr; }; struct mqfs_vdata { LIST_ENTRY(mqfs_vdata) mv_link; struct mqfs_node *mv_node; struct vnode *mv_vnode; struct task mv_task; }; /* * mqfs_node: describes a node (file or directory) within a mqfs */ struct mqfs_node { char mn_name[MQFS_NAMELEN+1]; struct mqfs_info *mn_info; struct mqfs_node *mn_parent; LIST_HEAD(,mqfs_node) mn_children; LIST_ENTRY(mqfs_node) mn_sibling; LIST_HEAD(,mqfs_vdata) mn_vnodes; const void *mn_pr_root; int mn_refcount; mqfs_type_t mn_type; int mn_deleted; uint32_t mn_fileno; void *mn_data; struct timespec mn_birth; struct timespec mn_ctime; struct timespec mn_atime; struct timespec mn_mtime; uid_t mn_uid; gid_t mn_gid; int mn_mode; }; #define VTON(vp) (((struct mqfs_vdata *)((vp)->v_data))->mv_node) #define VTOMQ(vp) ((struct mqueue *)(VTON(vp)->mn_data)) #define VFSTOMQFS(m) ((struct mqfs_info *)((m)->mnt_data)) #define FPTOMQ(fp) ((struct mqueue *)(((struct mqfs_node *) \ (fp)->f_data)->mn_data)) TAILQ_HEAD(msgq, mqueue_msg); struct mqueue; struct mqueue_notifier { LIST_ENTRY(mqueue_notifier) nt_link; struct sigevent nt_sigev; ksiginfo_t nt_ksi; struct proc *nt_proc; }; struct mqueue { struct mtx mq_mutex; int mq_flags; long mq_maxmsg; long mq_msgsize; long mq_curmsgs; long mq_totalbytes; struct msgq mq_msgq; int mq_receivers; int mq_senders; struct selinfo mq_rsel; struct selinfo mq_wsel; struct mqueue_notifier *mq_notifier; }; #define MQ_RSEL 0x01 #define MQ_WSEL 0x02 struct mqueue_msg { TAILQ_ENTRY(mqueue_msg) msg_link; unsigned int msg_prio; unsigned int msg_size; /* following real data... */ }; static SYSCTL_NODE(_kern, OID_AUTO, mqueue, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "POSIX real time message queue"); static int default_maxmsg = 10; static int default_msgsize = 1024; static int maxmsg = 100; SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmsg, CTLFLAG_RW, &maxmsg, 0, "Default maximum messages in queue"); static int maxmsgsize = 16384; SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmsgsize, CTLFLAG_RW, &maxmsgsize, 0, "Default maximum message size"); static int maxmq = 100; SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmq, CTLFLAG_RW, &maxmq, 0, "maximum message queues"); static int curmq = 0; SYSCTL_INT(_kern_mqueue, OID_AUTO, curmq, CTLFLAG_RW, &curmq, 0, "current message queue number"); static int unloadable = 0; static MALLOC_DEFINE(M_MQUEUEDATA, "mqdata", "mqueue data"); static eventhandler_tag exit_tag; /* Only one instance per-system */ static struct mqfs_info mqfs_data; static uma_zone_t mqnode_zone; static uma_zone_t mqueue_zone; static uma_zone_t mvdata_zone; static uma_zone_t mqnoti_zone; static struct vop_vector mqfs_vnodeops; static struct fileops mqueueops; static unsigned mqfs_osd_jail_slot; /* * Directory structure construction and manipulation */ #ifdef notyet static struct mqfs_node *mqfs_create_dir(struct mqfs_node *parent, const char *name, int namelen, struct ucred *cred, int mode); static struct mqfs_node *mqfs_create_link(struct mqfs_node *parent, const char *name, int namelen, struct ucred *cred, int mode); #endif static struct mqfs_node *mqfs_create_file(struct mqfs_node *parent, const char *name, int namelen, struct ucred *cred, int mode); static int mqfs_destroy(struct mqfs_node *mn); static void mqfs_fileno_alloc(struct mqfs_info *mi, struct mqfs_node *mn); static void mqfs_fileno_free(struct mqfs_info *mi, struct mqfs_node *mn); static int mqfs_allocv(struct mount *mp, struct vnode **vpp, struct mqfs_node *pn); static int mqfs_prison_remove(void *obj, void *data); /* * Message queue construction and maniplation */ static struct mqueue *mqueue_alloc(const struct mq_attr *attr); static void mqueue_free(struct mqueue *mq); static int mqueue_send(struct mqueue *mq, const char *msg_ptr, size_t msg_len, unsigned msg_prio, int waitok, const struct timespec *abs_timeout); static int mqueue_receive(struct mqueue *mq, char *msg_ptr, size_t msg_len, unsigned *msg_prio, int waitok, const struct timespec *abs_timeout); static int _mqueue_send(struct mqueue *mq, struct mqueue_msg *msg, int timo); static int _mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg, int timo); static void mqueue_send_notification(struct mqueue *mq); static void mqueue_fdclose(struct thread *td, int fd, struct file *fp); static void mq_proc_exit(void *arg, struct proc *p); /* * kqueue filters */ static void filt_mqdetach(struct knote *kn); static int filt_mqread(struct knote *kn, long hint); static int filt_mqwrite(struct knote *kn, long hint); struct filterops mq_rfiltops = { .f_isfd = 1, .f_detach = filt_mqdetach, .f_event = filt_mqread, }; struct filterops mq_wfiltops = { .f_isfd = 1, .f_detach = filt_mqdetach, .f_event = filt_mqwrite, }; /* * Initialize fileno bitmap */ static void mqfs_fileno_init(struct mqfs_info *mi) { struct unrhdr *up; up = new_unrhdr(1, INT_MAX, NULL); mi->mi_unrhdr = up; } /* * Tear down fileno bitmap */ static void mqfs_fileno_uninit(struct mqfs_info *mi) { struct unrhdr *up; up = mi->mi_unrhdr; mi->mi_unrhdr = NULL; delete_unrhdr(up); } /* * Allocate a file number */ static void mqfs_fileno_alloc(struct mqfs_info *mi, struct mqfs_node *mn) { /* make sure our parent has a file number */ if (mn->mn_parent && !mn->mn_parent->mn_fileno) mqfs_fileno_alloc(mi, mn->mn_parent); switch (mn->mn_type) { case mqfstype_root: case mqfstype_dir: case mqfstype_file: case mqfstype_symlink: mn->mn_fileno = alloc_unr(mi->mi_unrhdr); break; case mqfstype_this: KASSERT(mn->mn_parent != NULL, ("mqfstype_this node has no parent")); mn->mn_fileno = mn->mn_parent->mn_fileno; break; case mqfstype_parent: KASSERT(mn->mn_parent != NULL, ("mqfstype_parent node has no parent")); if (mn->mn_parent == mi->mi_root) { mn->mn_fileno = mn->mn_parent->mn_fileno; break; } KASSERT(mn->mn_parent->mn_parent != NULL, ("mqfstype_parent node has no grandparent")); mn->mn_fileno = mn->mn_parent->mn_parent->mn_fileno; break; default: KASSERT(0, ("mqfs_fileno_alloc() called for unknown type node: %d", mn->mn_type)); break; } } /* * Release a file number */ static void mqfs_fileno_free(struct mqfs_info *mi, struct mqfs_node *mn) { switch (mn->mn_type) { case mqfstype_root: case mqfstype_dir: case mqfstype_file: case mqfstype_symlink: free_unr(mi->mi_unrhdr, mn->mn_fileno); break; case mqfstype_this: case mqfstype_parent: /* ignore these, as they don't "own" their file number */ break; default: KASSERT(0, ("mqfs_fileno_free() called for unknown type node: %d", mn->mn_type)); break; } } static __inline struct mqfs_node * mqnode_alloc(void) { return uma_zalloc(mqnode_zone, M_WAITOK | M_ZERO); } static __inline void mqnode_free(struct mqfs_node *node) { uma_zfree(mqnode_zone, node); } static __inline void mqnode_addref(struct mqfs_node *node) { atomic_add_int(&node->mn_refcount, 1); } static __inline void mqnode_release(struct mqfs_node *node) { struct mqfs_info *mqfs; int old, exp; mqfs = node->mn_info; old = atomic_fetchadd_int(&node->mn_refcount, -1); if (node->mn_type == mqfstype_dir || node->mn_type == mqfstype_root) exp = 3; /* include . and .. */ else exp = 1; if (old == exp) { int locked = sx_xlocked(&mqfs->mi_lock); if (!locked) sx_xlock(&mqfs->mi_lock); mqfs_destroy(node); if (!locked) sx_xunlock(&mqfs->mi_lock); } } /* * Add a node to a directory */ static int mqfs_add_node(struct mqfs_node *parent, struct mqfs_node *node) { KASSERT(parent != NULL, ("%s(): parent is NULL", __func__)); KASSERT(parent->mn_info != NULL, ("%s(): parent has no mn_info", __func__)); KASSERT(parent->mn_type == mqfstype_dir || parent->mn_type == mqfstype_root, ("%s(): parent is not a directory", __func__)); node->mn_info = parent->mn_info; node->mn_parent = parent; LIST_INIT(&node->mn_children); LIST_INIT(&node->mn_vnodes); LIST_INSERT_HEAD(&parent->mn_children, node, mn_sibling); mqnode_addref(parent); return (0); } static struct mqfs_node * mqfs_create_node(const char *name, int namelen, struct ucred *cred, int mode, int nodetype) { struct mqfs_node *node; node = mqnode_alloc(); strncpy(node->mn_name, name, namelen); node->mn_pr_root = cred->cr_prison->pr_root; node->mn_type = nodetype; node->mn_refcount = 1; vfs_timestamp(&node->mn_birth); node->mn_ctime = node->mn_atime = node->mn_mtime = node->mn_birth; node->mn_uid = cred->cr_uid; node->mn_gid = cred->cr_gid; node->mn_mode = mode; return (node); } /* * Create a file */ static struct mqfs_node * mqfs_create_file(struct mqfs_node *parent, const char *name, int namelen, struct ucred *cred, int mode) { struct mqfs_node *node; node = mqfs_create_node(name, namelen, cred, mode, mqfstype_file); if (mqfs_add_node(parent, node) != 0) { mqnode_free(node); return (NULL); } return (node); } /* * Add . and .. to a directory */ static int mqfs_fixup_dir(struct mqfs_node *parent) { struct mqfs_node *dir; dir = mqnode_alloc(); dir->mn_name[0] = '.'; dir->mn_type = mqfstype_this; dir->mn_refcount = 1; if (mqfs_add_node(parent, dir) != 0) { mqnode_free(dir); return (-1); } dir = mqnode_alloc(); dir->mn_name[0] = dir->mn_name[1] = '.'; dir->mn_type = mqfstype_parent; dir->mn_refcount = 1; if (mqfs_add_node(parent, dir) != 0) { mqnode_free(dir); return (-1); } return (0); } #ifdef notyet /* * Create a directory */ static struct mqfs_node * mqfs_create_dir(struct mqfs_node *parent, const char *name, int namelen, struct ucred *cred, int mode) { struct mqfs_node *node; node = mqfs_create_node(name, namelen, cred, mode, mqfstype_dir); if (mqfs_add_node(parent, node) != 0) { mqnode_free(node); return (NULL); } if (mqfs_fixup_dir(node) != 0) { mqfs_destroy(node); return (NULL); } return (node); } /* * Create a symlink */ static struct mqfs_node * mqfs_create_link(struct mqfs_node *parent, const char *name, int namelen, struct ucred *cred, int mode) { struct mqfs_node *node; node = mqfs_create_node(name, namelen, cred, mode, mqfstype_symlink); if (mqfs_add_node(parent, node) != 0) { mqnode_free(node); return (NULL); } return (node); } #endif /* * Destroy a node or a tree of nodes */ static int mqfs_destroy(struct mqfs_node *node) { struct mqfs_node *parent; KASSERT(node != NULL, ("%s(): node is NULL", __func__)); KASSERT(node->mn_info != NULL, ("%s(): node has no mn_info", __func__)); /* destroy children */ if (node->mn_type == mqfstype_dir || node->mn_type == mqfstype_root) while (! LIST_EMPTY(&node->mn_children)) mqfs_destroy(LIST_FIRST(&node->mn_children)); /* unlink from parent */ if ((parent = node->mn_parent) != NULL) { KASSERT(parent->mn_info == node->mn_info, ("%s(): parent has different mn_info", __func__)); LIST_REMOVE(node, mn_sibling); } if (node->mn_fileno != 0) mqfs_fileno_free(node->mn_info, node); if (node->mn_data != NULL) mqueue_free(node->mn_data); mqnode_free(node); return (0); } /* * Mount a mqfs instance */ static int mqfs_mount(struct mount *mp) { struct statfs *sbp; if (mp->mnt_flag & MNT_UPDATE) return (EOPNOTSUPP); mp->mnt_data = &mqfs_data; MNT_ILOCK(mp); mp->mnt_flag |= MNT_LOCAL; MNT_IUNLOCK(mp); vfs_getnewfsid(mp); sbp = &mp->mnt_stat; vfs_mountedfrom(mp, "mqueue"); sbp->f_bsize = PAGE_SIZE; sbp->f_iosize = PAGE_SIZE; sbp->f_blocks = 1; sbp->f_bfree = 0; sbp->f_bavail = 0; sbp->f_files = 1; sbp->f_ffree = 0; return (0); } /* * Unmount a mqfs instance */ static int mqfs_unmount(struct mount *mp, int mntflags) { int error; error = vflush(mp, 0, (mntflags & MNT_FORCE) ? FORCECLOSE : 0, curthread); return (error); } /* * Return a root vnode */ static int mqfs_root(struct mount *mp, int flags, struct vnode **vpp) { struct mqfs_info *mqfs; int ret; mqfs = VFSTOMQFS(mp); ret = mqfs_allocv(mp, vpp, mqfs->mi_root); return (ret); } /* * Return filesystem stats */ static int mqfs_statfs(struct mount *mp, struct statfs *sbp) { /* XXX update statistics */ return (0); } /* * Initialize a mqfs instance */ static int mqfs_init(struct vfsconf *vfc) { struct mqfs_node *root; struct mqfs_info *mi; osd_method_t methods[PR_MAXMETHOD] = { [PR_METHOD_REMOVE] = mqfs_prison_remove, }; mqnode_zone = uma_zcreate("mqnode", sizeof(struct mqfs_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); mqueue_zone = uma_zcreate("mqueue", sizeof(struct mqueue), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); mvdata_zone = uma_zcreate("mvdata", sizeof(struct mqfs_vdata), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); mqnoti_zone = uma_zcreate("mqnotifier", sizeof(struct mqueue_notifier), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); mi = &mqfs_data; sx_init(&mi->mi_lock, "mqfs lock"); /* set up the root diretory */ root = mqfs_create_node("/", 1, curthread->td_ucred, 01777, mqfstype_root); root->mn_info = mi; LIST_INIT(&root->mn_children); LIST_INIT(&root->mn_vnodes); mi->mi_root = root; mqfs_fileno_init(mi); mqfs_fileno_alloc(mi, root); mqfs_fixup_dir(root); exit_tag = EVENTHANDLER_REGISTER(process_exit, mq_proc_exit, NULL, EVENTHANDLER_PRI_ANY); mq_fdclose = mqueue_fdclose; p31b_setcfg(CTL_P1003_1B_MESSAGE_PASSING, _POSIX_MESSAGE_PASSING); mqfs_osd_jail_slot = osd_jail_register(NULL, methods); return (0); } /* * Destroy a mqfs instance */ static int mqfs_uninit(struct vfsconf *vfc) { struct mqfs_info *mi; if (!unloadable) return (EOPNOTSUPP); osd_jail_deregister(mqfs_osd_jail_slot); EVENTHANDLER_DEREGISTER(process_exit, exit_tag); mi = &mqfs_data; mqfs_destroy(mi->mi_root); mi->mi_root = NULL; mqfs_fileno_uninit(mi); sx_destroy(&mi->mi_lock); uma_zdestroy(mqnode_zone); uma_zdestroy(mqueue_zone); uma_zdestroy(mvdata_zone); uma_zdestroy(mqnoti_zone); return (0); } /* * task routine */ static void do_recycle(void *context, int pending __unused) { struct vnode *vp = (struct vnode *)context; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); vrecycle(vp); VOP_UNLOCK(vp); vdrop(vp); } /* * Allocate a vnode */ static int mqfs_allocv(struct mount *mp, struct vnode **vpp, struct mqfs_node *pn) { struct mqfs_vdata *vd; struct mqfs_info *mqfs; struct vnode *newvpp; int error; mqfs = pn->mn_info; *vpp = NULL; sx_xlock(&mqfs->mi_lock); LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) { if (vd->mv_vnode->v_mount == mp) { vhold(vd->mv_vnode); break; } } if (vd != NULL) { found: *vpp = vd->mv_vnode; sx_xunlock(&mqfs->mi_lock); error = vget(*vpp, LK_RETRY | LK_EXCLUSIVE); vdrop(*vpp); return (error); } sx_xunlock(&mqfs->mi_lock); error = getnewvnode("mqueue", mp, &mqfs_vnodeops, &newvpp); if (error) return (error); vn_lock(newvpp, LK_EXCLUSIVE | LK_RETRY); error = insmntque(newvpp, mp); if (error != 0) return (error); sx_xlock(&mqfs->mi_lock); /* * Check if it has already been allocated * while we were blocked. */ LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) { if (vd->mv_vnode->v_mount == mp) { vhold(vd->mv_vnode); sx_xunlock(&mqfs->mi_lock); vgone(newvpp); vput(newvpp); goto found; } } *vpp = newvpp; vd = uma_zalloc(mvdata_zone, M_WAITOK); (*vpp)->v_data = vd; vd->mv_vnode = *vpp; vd->mv_node = pn; TASK_INIT(&vd->mv_task, 0, do_recycle, *vpp); LIST_INSERT_HEAD(&pn->mn_vnodes, vd, mv_link); mqnode_addref(pn); switch (pn->mn_type) { case mqfstype_root: (*vpp)->v_vflag = VV_ROOT; /* fall through */ case mqfstype_dir: case mqfstype_this: case mqfstype_parent: (*vpp)->v_type = VDIR; break; case mqfstype_file: (*vpp)->v_type = VREG; break; case mqfstype_symlink: (*vpp)->v_type = VLNK; break; case mqfstype_none: KASSERT(0, ("mqfs_allocf called for null node\n")); default: panic("%s has unexpected type: %d", pn->mn_name, pn->mn_type); } sx_xunlock(&mqfs->mi_lock); return (0); } /* * Search a directory entry */ static struct mqfs_node * mqfs_search(struct mqfs_node *pd, const char *name, int len, struct ucred *cred) { struct mqfs_node *pn; const void *pr_root; sx_assert(&pd->mn_info->mi_lock, SX_LOCKED); pr_root = cred->cr_prison->pr_root; LIST_FOREACH(pn, &pd->mn_children, mn_sibling) { /* Only match names within the same prison root directory */ if ((pn->mn_pr_root == NULL || pn->mn_pr_root == pr_root) && strncmp(pn->mn_name, name, len) == 0 && pn->mn_name[len] == '\0') return (pn); } return (NULL); } /* * Look up a file or directory. */ static int mqfs_lookupx(struct vop_cachedlookup_args *ap) { struct componentname *cnp; struct vnode *dvp, **vpp; struct mqfs_node *pd; struct mqfs_node *pn; struct mqfs_info *mqfs; int nameiop, flags, error, namelen; char *pname; struct thread *td; + td = curthread; cnp = ap->a_cnp; vpp = ap->a_vpp; dvp = ap->a_dvp; pname = cnp->cn_nameptr; namelen = cnp->cn_namelen; - td = cnp->cn_thread; flags = cnp->cn_flags; nameiop = cnp->cn_nameiop; pd = VTON(dvp); pn = NULL; mqfs = pd->mn_info; *vpp = NULLVP; if (dvp->v_type != VDIR) return (ENOTDIR); - error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_thread); + error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td); if (error) return (error); /* shortcut: check if the name is too long */ if (cnp->cn_namelen >= MQFS_NAMELEN) return (ENOENT); /* self */ if (namelen == 1 && pname[0] == '.') { if ((flags & ISLASTCN) && nameiop != LOOKUP) return (EINVAL); pn = pd; *vpp = dvp; VREF(dvp); return (0); } /* parent */ if (cnp->cn_flags & ISDOTDOT) { if (dvp->v_vflag & VV_ROOT) return (EIO); if ((flags & ISLASTCN) && nameiop != LOOKUP) return (EINVAL); VOP_UNLOCK(dvp); KASSERT(pd->mn_parent, ("non-root directory has no parent")); pn = pd->mn_parent; error = mqfs_allocv(dvp->v_mount, vpp, pn); vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); return (error); } /* named node */ sx_xlock(&mqfs->mi_lock); pn = mqfs_search(pd, pname, namelen, cnp->cn_cred); if (pn != NULL) mqnode_addref(pn); sx_xunlock(&mqfs->mi_lock); /* found */ if (pn != NULL) { /* DELETE */ if (nameiop == DELETE && (flags & ISLASTCN)) { error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td); if (error) { mqnode_release(pn); return (error); } if (*vpp == dvp) { VREF(dvp); *vpp = dvp; mqnode_release(pn); return (0); } } /* allocate vnode */ error = mqfs_allocv(dvp->v_mount, vpp, pn); mqnode_release(pn); if (error == 0 && cnp->cn_flags & MAKEENTRY) cache_enter(dvp, *vpp, cnp); return (error); } /* not found */ /* will create a new entry in the directory ? */ if ((nameiop == CREATE || nameiop == RENAME) && (flags & LOCKPARENT) && (flags & ISLASTCN)) { error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td); if (error) return (error); cnp->cn_flags |= SAVENAME; return (EJUSTRETURN); } return (ENOENT); } #if 0 struct vop_lookup_args { struct vop_generic_args a_gen; struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; }; #endif /* * vnode lookup operation */ static int mqfs_lookup(struct vop_cachedlookup_args *ap) { int rc; rc = mqfs_lookupx(ap); return (rc); } #if 0 struct vop_create_args { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; struct vattr *a_vap; }; #endif /* * vnode creation operation */ static int mqfs_create(struct vop_create_args *ap) { struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount); struct componentname *cnp = ap->a_cnp; struct mqfs_node *pd; struct mqfs_node *pn; struct mqueue *mq; int error; pd = VTON(ap->a_dvp); if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir) return (ENOTDIR); mq = mqueue_alloc(NULL); if (mq == NULL) return (EAGAIN); sx_xlock(&mqfs->mi_lock); if ((cnp->cn_flags & HASBUF) == 0) panic("%s: no name", __func__); pn = mqfs_create_file(pd, cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_cred, ap->a_vap->va_mode); if (pn == NULL) { sx_xunlock(&mqfs->mi_lock); error = ENOSPC; } else { mqnode_addref(pn); sx_xunlock(&mqfs->mi_lock); error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn); mqnode_release(pn); if (error) mqfs_destroy(pn); else pn->mn_data = mq; } if (error) mqueue_free(mq); return (error); } /* * Remove an entry */ static int do_unlink(struct mqfs_node *pn, struct ucred *ucred) { struct mqfs_node *parent; struct mqfs_vdata *vd; int error = 0; sx_assert(&pn->mn_info->mi_lock, SX_LOCKED); if (ucred->cr_uid != pn->mn_uid && (error = priv_check_cred(ucred, PRIV_MQ_ADMIN)) != 0) error = EACCES; else if (!pn->mn_deleted) { parent = pn->mn_parent; pn->mn_parent = NULL; pn->mn_deleted = 1; LIST_REMOVE(pn, mn_sibling); LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) { cache_purge(vd->mv_vnode); vhold(vd->mv_vnode); taskqueue_enqueue(taskqueue_thread, &vd->mv_task); } mqnode_release(pn); mqnode_release(parent); } else error = ENOENT; return (error); } #if 0 struct vop_remove_args { struct vnode *a_dvp; struct vnode *a_vp; struct componentname *a_cnp; }; #endif /* * vnode removal operation */ static int mqfs_remove(struct vop_remove_args *ap) { struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount); struct mqfs_node *pn; int error; if (ap->a_vp->v_type == VDIR) return (EPERM); pn = VTON(ap->a_vp); sx_xlock(&mqfs->mi_lock); error = do_unlink(pn, ap->a_cnp->cn_cred); sx_xunlock(&mqfs->mi_lock); return (error); } #if 0 struct vop_inactive_args { struct vnode *a_vp; struct thread *a_td; }; #endif static int mqfs_inactive(struct vop_inactive_args *ap) { struct mqfs_node *pn = VTON(ap->a_vp); if (pn->mn_deleted) vrecycle(ap->a_vp); return (0); } #if 0 struct vop_reclaim_args { struct vop_generic_args a_gen; struct vnode *a_vp; }; #endif static int mqfs_reclaim(struct vop_reclaim_args *ap) { struct mqfs_info *mqfs = VFSTOMQFS(ap->a_vp->v_mount); struct vnode *vp = ap->a_vp; struct mqfs_node *pn; struct mqfs_vdata *vd; vd = vp->v_data; pn = vd->mv_node; sx_xlock(&mqfs->mi_lock); vp->v_data = NULL; LIST_REMOVE(vd, mv_link); uma_zfree(mvdata_zone, vd); mqnode_release(pn); sx_xunlock(&mqfs->mi_lock); return (0); } #if 0 struct vop_open_args { struct vop_generic_args a_gen; struct vnode *a_vp; int a_mode; struct ucred *a_cred; struct thread *a_td; struct file *a_fp; }; #endif static int mqfs_open(struct vop_open_args *ap) { return (0); } #if 0 struct vop_close_args { struct vop_generic_args a_gen; struct vnode *a_vp; int a_fflag; struct ucred *a_cred; struct thread *a_td; }; #endif static int mqfs_close(struct vop_close_args *ap) { return (0); } #if 0 struct vop_access_args { struct vop_generic_args a_gen; struct vnode *a_vp; accmode_t a_accmode; struct ucred *a_cred; struct thread *a_td; }; #endif /* * Verify permissions */ static int mqfs_access(struct vop_access_args *ap) { struct vnode *vp = ap->a_vp; struct vattr vattr; int error; error = VOP_GETATTR(vp, &vattr, ap->a_cred); if (error) return (error); error = vaccess(vp->v_type, vattr.va_mode, vattr.va_uid, vattr.va_gid, ap->a_accmode, ap->a_cred); return (error); } #if 0 struct vop_getattr_args { struct vop_generic_args a_gen; struct vnode *a_vp; struct vattr *a_vap; struct ucred *a_cred; }; #endif /* * Get file attributes */ static int mqfs_getattr(struct vop_getattr_args *ap) { struct vnode *vp = ap->a_vp; struct mqfs_node *pn = VTON(vp); struct vattr *vap = ap->a_vap; int error = 0; vap->va_type = vp->v_type; vap->va_mode = pn->mn_mode; vap->va_nlink = 1; vap->va_uid = pn->mn_uid; vap->va_gid = pn->mn_gid; vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; vap->va_fileid = pn->mn_fileno; vap->va_size = 0; vap->va_blocksize = PAGE_SIZE; vap->va_bytes = vap->va_size = 0; vap->va_atime = pn->mn_atime; vap->va_mtime = pn->mn_mtime; vap->va_ctime = pn->mn_ctime; vap->va_birthtime = pn->mn_birth; vap->va_gen = 0; vap->va_flags = 0; vap->va_rdev = NODEV; vap->va_bytes = 0; vap->va_filerev = 0; return (error); } #if 0 struct vop_setattr_args { struct vop_generic_args a_gen; struct vnode *a_vp; struct vattr *a_vap; struct ucred *a_cred; }; #endif /* * Set attributes */ static int mqfs_setattr(struct vop_setattr_args *ap) { struct mqfs_node *pn; struct vattr *vap; struct vnode *vp; struct thread *td; int c, error; uid_t uid; gid_t gid; td = curthread; vap = ap->a_vap; vp = ap->a_vp; if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) || (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) || (vap->va_blocksize != VNOVAL) || (vap->va_flags != VNOVAL && vap->va_flags != 0) || (vap->va_rdev != VNOVAL) || ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) { return (EINVAL); } pn = VTON(vp); error = c = 0; if (vap->va_uid == (uid_t)VNOVAL) uid = pn->mn_uid; else uid = vap->va_uid; if (vap->va_gid == (gid_t)VNOVAL) gid = pn->mn_gid; else gid = vap->va_gid; if (uid != pn->mn_uid || gid != pn->mn_gid) { /* * To modify the ownership of a file, must possess VADMIN * for that file. */ if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td))) return (error); /* * XXXRW: Why is there a privilege check here: shouldn't the * check in VOP_ACCESS() be enough? Also, are the group bits * below definitely right? */ if (((ap->a_cred->cr_uid != pn->mn_uid) || uid != pn->mn_uid || (gid != pn->mn_gid && !groupmember(gid, ap->a_cred))) && (error = priv_check(td, PRIV_MQ_ADMIN)) != 0) return (error); pn->mn_uid = uid; pn->mn_gid = gid; c = 1; } if (vap->va_mode != (mode_t)VNOVAL) { if ((ap->a_cred->cr_uid != pn->mn_uid) && (error = priv_check(td, PRIV_MQ_ADMIN))) return (error); pn->mn_mode = vap->va_mode; c = 1; } if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { /* See the comment in ufs_vnops::ufs_setattr(). */ if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td)) && ((vap->va_vaflags & VA_UTIMES_NULL) == 0 || (error = VOP_ACCESS(vp, VWRITE, ap->a_cred, td)))) return (error); if (vap->va_atime.tv_sec != VNOVAL) { pn->mn_atime = vap->va_atime; } if (vap->va_mtime.tv_sec != VNOVAL) { pn->mn_mtime = vap->va_mtime; } c = 1; } if (c) { vfs_timestamp(&pn->mn_ctime); } return (0); } #if 0 struct vop_read_args { struct vop_generic_args a_gen; struct vnode *a_vp; struct uio *a_uio; int a_ioflag; struct ucred *a_cred; }; #endif /* * Read from a file */ static int mqfs_read(struct vop_read_args *ap) { char buf[80]; struct vnode *vp = ap->a_vp; struct uio *uio = ap->a_uio; struct mqueue *mq; int len, error; if (vp->v_type != VREG) return (EINVAL); mq = VTOMQ(vp); snprintf(buf, sizeof(buf), "QSIZE:%-10ld MAXMSG:%-10ld CURMSG:%-10ld MSGSIZE:%-10ld\n", mq->mq_totalbytes, mq->mq_maxmsg, mq->mq_curmsgs, mq->mq_msgsize); buf[sizeof(buf)-1] = '\0'; len = strlen(buf); error = uiomove_frombuf(buf, len, uio); return (error); } #if 0 struct vop_readdir_args { struct vop_generic_args a_gen; struct vnode *a_vp; struct uio *a_uio; struct ucred *a_cred; int *a_eofflag; int *a_ncookies; u_long **a_cookies; }; #endif /* * Return directory entries. */ static int mqfs_readdir(struct vop_readdir_args *ap) { struct vnode *vp; struct mqfs_info *mi; struct mqfs_node *pd; struct mqfs_node *pn; struct dirent entry; struct uio *uio; const void *pr_root; int *tmp_ncookies = NULL; off_t offset; int error, i; vp = ap->a_vp; mi = VFSTOMQFS(vp->v_mount); pd = VTON(vp); uio = ap->a_uio; if (vp->v_type != VDIR) return (ENOTDIR); if (uio->uio_offset < 0) return (EINVAL); if (ap->a_ncookies != NULL) { tmp_ncookies = ap->a_ncookies; *ap->a_ncookies = 0; ap->a_ncookies = NULL; } error = 0; offset = 0; pr_root = ap->a_cred->cr_prison->pr_root; sx_xlock(&mi->mi_lock); LIST_FOREACH(pn, &pd->mn_children, mn_sibling) { entry.d_reclen = sizeof(entry); /* * Only show names within the same prison root directory * (or not associated with a prison, e.g. "." and ".."). */ if (pn->mn_pr_root != NULL && pn->mn_pr_root != pr_root) continue; if (!pn->mn_fileno) mqfs_fileno_alloc(mi, pn); entry.d_fileno = pn->mn_fileno; entry.d_off = offset + entry.d_reclen; for (i = 0; i < MQFS_NAMELEN - 1 && pn->mn_name[i] != '\0'; ++i) entry.d_name[i] = pn->mn_name[i]; entry.d_namlen = i; switch (pn->mn_type) { case mqfstype_root: case mqfstype_dir: case mqfstype_this: case mqfstype_parent: entry.d_type = DT_DIR; break; case mqfstype_file: entry.d_type = DT_REG; break; case mqfstype_symlink: entry.d_type = DT_LNK; break; default: panic("%s has unexpected node type: %d", pn->mn_name, pn->mn_type); } dirent_terminate(&entry); if (entry.d_reclen > uio->uio_resid) break; if (offset >= uio->uio_offset) { error = vfs_read_dirent(ap, &entry, offset); if (error) break; } offset += entry.d_reclen; } sx_xunlock(&mi->mi_lock); uio->uio_offset = offset; if (tmp_ncookies != NULL) ap->a_ncookies = tmp_ncookies; return (error); } #ifdef notyet #if 0 struct vop_mkdir_args { struct vnode *a_dvp; struvt vnode **a_vpp; struvt componentname *a_cnp; struct vattr *a_vap; }; #endif /* * Create a directory. */ static int mqfs_mkdir(struct vop_mkdir_args *ap) { struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount); struct componentname *cnp = ap->a_cnp; struct mqfs_node *pd = VTON(ap->a_dvp); struct mqfs_node *pn; int error; if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir) return (ENOTDIR); sx_xlock(&mqfs->mi_lock); if ((cnp->cn_flags & HASBUF) == 0) panic("%s: no name", __func__); pn = mqfs_create_dir(pd, cnp->cn_nameptr, cnp->cn_namelen, ap->a_vap->cn_cred, ap->a_vap->va_mode); if (pn != NULL) mqnode_addref(pn); sx_xunlock(&mqfs->mi_lock); if (pn == NULL) { error = ENOSPC; } else { error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn); mqnode_release(pn); } return (error); } #if 0 struct vop_rmdir_args { struct vnode *a_dvp; struct vnode *a_vp; struct componentname *a_cnp; }; #endif /* * Remove a directory. */ static int mqfs_rmdir(struct vop_rmdir_args *ap) { struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount); struct mqfs_node *pn = VTON(ap->a_vp); struct mqfs_node *pt; if (pn->mn_type != mqfstype_dir) return (ENOTDIR); sx_xlock(&mqfs->mi_lock); if (pn->mn_deleted) { sx_xunlock(&mqfs->mi_lock); return (ENOENT); } pt = LIST_FIRST(&pn->mn_children); pt = LIST_NEXT(pt, mn_sibling); pt = LIST_NEXT(pt, mn_sibling); if (pt != NULL) { sx_xunlock(&mqfs->mi_lock); return (ENOTEMPTY); } pt = pn->mn_parent; pn->mn_parent = NULL; pn->mn_deleted = 1; LIST_REMOVE(pn, mn_sibling); mqnode_release(pn); mqnode_release(pt); sx_xunlock(&mqfs->mi_lock); cache_purge(ap->a_vp); return (0); } #endif /* notyet */ /* * See if this prison root is obsolete, and clean up associated queues if it is. */ static int mqfs_prison_remove(void *obj, void *data __unused) { const struct prison *pr = obj; struct prison *tpr; struct mqfs_node *pn, *tpn; struct vnode *pr_root; pr_root = pr->pr_root; if (pr->pr_parent->pr_root == pr_root) return (0); TAILQ_FOREACH(tpr, &allprison, pr_list) { if (tpr != pr && tpr->pr_root == pr_root) return (0); } /* * No jails are rooted in this directory anymore, * so no queues should be either. */ sx_xlock(&mqfs_data.mi_lock); LIST_FOREACH_SAFE(pn, &mqfs_data.mi_root->mn_children, mn_sibling, tpn) { if (pn->mn_pr_root == pr_root) (void)do_unlink(pn, curthread->td_ucred); } sx_xunlock(&mqfs_data.mi_lock); return (0); } /* * Allocate a message queue */ static struct mqueue * mqueue_alloc(const struct mq_attr *attr) { struct mqueue *mq; if (curmq >= maxmq) return (NULL); mq = uma_zalloc(mqueue_zone, M_WAITOK | M_ZERO); TAILQ_INIT(&mq->mq_msgq); if (attr != NULL) { mq->mq_maxmsg = attr->mq_maxmsg; mq->mq_msgsize = attr->mq_msgsize; } else { mq->mq_maxmsg = default_maxmsg; mq->mq_msgsize = default_msgsize; } mtx_init(&mq->mq_mutex, "mqueue lock", NULL, MTX_DEF); knlist_init_mtx(&mq->mq_rsel.si_note, &mq->mq_mutex); knlist_init_mtx(&mq->mq_wsel.si_note, &mq->mq_mutex); atomic_add_int(&curmq, 1); return (mq); } /* * Destroy a message queue */ static void mqueue_free(struct mqueue *mq) { struct mqueue_msg *msg; while ((msg = TAILQ_FIRST(&mq->mq_msgq)) != NULL) { TAILQ_REMOVE(&mq->mq_msgq, msg, msg_link); free(msg, M_MQUEUEDATA); } mtx_destroy(&mq->mq_mutex); seldrain(&mq->mq_rsel); seldrain(&mq->mq_wsel); knlist_destroy(&mq->mq_rsel.si_note); knlist_destroy(&mq->mq_wsel.si_note); uma_zfree(mqueue_zone, mq); atomic_add_int(&curmq, -1); } /* * Load a message from user space */ static struct mqueue_msg * mqueue_loadmsg(const char *msg_ptr, size_t msg_size, int msg_prio) { struct mqueue_msg *msg; size_t len; int error; len = sizeof(struct mqueue_msg) + msg_size; msg = malloc(len, M_MQUEUEDATA, M_WAITOK); error = copyin(msg_ptr, ((char *)msg) + sizeof(struct mqueue_msg), msg_size); if (error) { free(msg, M_MQUEUEDATA); msg = NULL; } else { msg->msg_size = msg_size; msg->msg_prio = msg_prio; } return (msg); } /* * Save a message to user space */ static int mqueue_savemsg(struct mqueue_msg *msg, char *msg_ptr, int *msg_prio) { int error; error = copyout(((char *)msg) + sizeof(*msg), msg_ptr, msg->msg_size); if (error == 0 && msg_prio != NULL) error = copyout(&msg->msg_prio, msg_prio, sizeof(int)); return (error); } /* * Free a message's memory */ static __inline void mqueue_freemsg(struct mqueue_msg *msg) { free(msg, M_MQUEUEDATA); } /* * Send a message. if waitok is false, thread will not be * blocked if there is no data in queue, otherwise, absolute * time will be checked. */ int mqueue_send(struct mqueue *mq, const char *msg_ptr, size_t msg_len, unsigned msg_prio, int waitok, const struct timespec *abs_timeout) { struct mqueue_msg *msg; struct timespec ts, ts2; struct timeval tv; int error; if (msg_prio >= MQ_PRIO_MAX) return (EINVAL); if (msg_len > mq->mq_msgsize) return (EMSGSIZE); msg = mqueue_loadmsg(msg_ptr, msg_len, msg_prio); if (msg == NULL) return (EFAULT); /* O_NONBLOCK case */ if (!waitok) { error = _mqueue_send(mq, msg, -1); if (error) goto bad; return (0); } /* we allow a null timeout (wait forever) */ if (abs_timeout == NULL) { error = _mqueue_send(mq, msg, 0); if (error) goto bad; return (0); } /* send it before checking time */ error = _mqueue_send(mq, msg, -1); if (error == 0) return (0); if (error != EAGAIN) goto bad; if (abs_timeout->tv_nsec >= 1000000000 || abs_timeout->tv_nsec < 0) { error = EINVAL; goto bad; } for (;;) { getnanotime(&ts); timespecsub(abs_timeout, &ts, &ts2); if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) { error = ETIMEDOUT; break; } TIMESPEC_TO_TIMEVAL(&tv, &ts2); error = _mqueue_send(mq, msg, tvtohz(&tv)); if (error != ETIMEDOUT) break; } if (error == 0) return (0); bad: mqueue_freemsg(msg); return (error); } /* * Common routine to send a message */ static int _mqueue_send(struct mqueue *mq, struct mqueue_msg *msg, int timo) { struct mqueue_msg *msg2; int error = 0; mtx_lock(&mq->mq_mutex); while (mq->mq_curmsgs >= mq->mq_maxmsg && error == 0) { if (timo < 0) { mtx_unlock(&mq->mq_mutex); return (EAGAIN); } mq->mq_senders++; error = msleep(&mq->mq_senders, &mq->mq_mutex, PCATCH, "mqsend", timo); mq->mq_senders--; if (error == EAGAIN) error = ETIMEDOUT; } if (mq->mq_curmsgs >= mq->mq_maxmsg) { mtx_unlock(&mq->mq_mutex); return (error); } error = 0; if (TAILQ_EMPTY(&mq->mq_msgq)) { TAILQ_INSERT_HEAD(&mq->mq_msgq, msg, msg_link); } else { if (msg->msg_prio <= TAILQ_LAST(&mq->mq_msgq, msgq)->msg_prio) { TAILQ_INSERT_TAIL(&mq->mq_msgq, msg, msg_link); } else { TAILQ_FOREACH(msg2, &mq->mq_msgq, msg_link) { if (msg2->msg_prio < msg->msg_prio) break; } TAILQ_INSERT_BEFORE(msg2, msg, msg_link); } } mq->mq_curmsgs++; mq->mq_totalbytes += msg->msg_size; if (mq->mq_receivers) wakeup_one(&mq->mq_receivers); else if (mq->mq_notifier != NULL) mqueue_send_notification(mq); if (mq->mq_flags & MQ_RSEL) { mq->mq_flags &= ~MQ_RSEL; selwakeup(&mq->mq_rsel); } KNOTE_LOCKED(&mq->mq_rsel.si_note, 0); mtx_unlock(&mq->mq_mutex); return (0); } /* * Send realtime a signal to process which registered itself * successfully by mq_notify. */ static void mqueue_send_notification(struct mqueue *mq) { struct mqueue_notifier *nt; struct thread *td; struct proc *p; int error; mtx_assert(&mq->mq_mutex, MA_OWNED); nt = mq->mq_notifier; if (nt->nt_sigev.sigev_notify != SIGEV_NONE) { p = nt->nt_proc; error = sigev_findtd(p, &nt->nt_sigev, &td); if (error) { mq->mq_notifier = NULL; return; } if (!KSI_ONQ(&nt->nt_ksi)) { ksiginfo_set_sigev(&nt->nt_ksi, &nt->nt_sigev); tdsendsignal(p, td, nt->nt_ksi.ksi_signo, &nt->nt_ksi); } PROC_UNLOCK(p); } mq->mq_notifier = NULL; } /* * Get a message. if waitok is false, thread will not be * blocked if there is no data in queue, otherwise, absolute * time will be checked. */ int mqueue_receive(struct mqueue *mq, char *msg_ptr, size_t msg_len, unsigned *msg_prio, int waitok, const struct timespec *abs_timeout) { struct mqueue_msg *msg; struct timespec ts, ts2; struct timeval tv; int error; if (msg_len < mq->mq_msgsize) return (EMSGSIZE); /* O_NONBLOCK case */ if (!waitok) { error = _mqueue_recv(mq, &msg, -1); if (error) return (error); goto received; } /* we allow a null timeout (wait forever). */ if (abs_timeout == NULL) { error = _mqueue_recv(mq, &msg, 0); if (error) return (error); goto received; } /* try to get a message before checking time */ error = _mqueue_recv(mq, &msg, -1); if (error == 0) goto received; if (error != EAGAIN) return (error); if (abs_timeout->tv_nsec >= 1000000000 || abs_timeout->tv_nsec < 0) { error = EINVAL; return (error); } for (;;) { getnanotime(&ts); timespecsub(abs_timeout, &ts, &ts2); if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) { error = ETIMEDOUT; return (error); } TIMESPEC_TO_TIMEVAL(&tv, &ts2); error = _mqueue_recv(mq, &msg, tvtohz(&tv)); if (error == 0) break; if (error != ETIMEDOUT) return (error); } received: error = mqueue_savemsg(msg, msg_ptr, msg_prio); if (error == 0) { curthread->td_retval[0] = msg->msg_size; curthread->td_retval[1] = 0; } mqueue_freemsg(msg); return (error); } /* * Common routine to receive a message */ static int _mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg, int timo) { int error = 0; mtx_lock(&mq->mq_mutex); while ((*msg = TAILQ_FIRST(&mq->mq_msgq)) == NULL && error == 0) { if (timo < 0) { mtx_unlock(&mq->mq_mutex); return (EAGAIN); } mq->mq_receivers++; error = msleep(&mq->mq_receivers, &mq->mq_mutex, PCATCH, "mqrecv", timo); mq->mq_receivers--; if (error == EAGAIN) error = ETIMEDOUT; } if (*msg != NULL) { error = 0; TAILQ_REMOVE(&mq->mq_msgq, *msg, msg_link); mq->mq_curmsgs--; mq->mq_totalbytes -= (*msg)->msg_size; if (mq->mq_senders) wakeup_one(&mq->mq_senders); if (mq->mq_flags & MQ_WSEL) { mq->mq_flags &= ~MQ_WSEL; selwakeup(&mq->mq_wsel); } KNOTE_LOCKED(&mq->mq_wsel.si_note, 0); } if (mq->mq_notifier != NULL && mq->mq_receivers == 0 && !TAILQ_EMPTY(&mq->mq_msgq)) { mqueue_send_notification(mq); } mtx_unlock(&mq->mq_mutex); return (error); } static __inline struct mqueue_notifier * notifier_alloc(void) { return (uma_zalloc(mqnoti_zone, M_WAITOK | M_ZERO)); } static __inline void notifier_free(struct mqueue_notifier *p) { uma_zfree(mqnoti_zone, p); } static struct mqueue_notifier * notifier_search(struct proc *p, int fd) { struct mqueue_notifier *nt; LIST_FOREACH(nt, &p->p_mqnotifier, nt_link) { if (nt->nt_ksi.ksi_mqd == fd) break; } return (nt); } static __inline void notifier_insert(struct proc *p, struct mqueue_notifier *nt) { LIST_INSERT_HEAD(&p->p_mqnotifier, nt, nt_link); } static __inline void notifier_delete(struct proc *p, struct mqueue_notifier *nt) { LIST_REMOVE(nt, nt_link); notifier_free(nt); } static void notifier_remove(struct proc *p, struct mqueue *mq, int fd) { struct mqueue_notifier *nt; mtx_assert(&mq->mq_mutex, MA_OWNED); PROC_LOCK(p); nt = notifier_search(p, fd); if (nt != NULL) { if (mq->mq_notifier == nt) mq->mq_notifier = NULL; sigqueue_take(&nt->nt_ksi); notifier_delete(p, nt); } PROC_UNLOCK(p); } static int kern_kmq_open(struct thread *td, const char *upath, int flags, mode_t mode, const struct mq_attr *attr) { char path[MQFS_NAMELEN + 1]; struct mqfs_node *pn; struct pwddesc *pdp; struct file *fp; struct mqueue *mq; int fd, error, len, cmode; AUDIT_ARG_FFLAGS(flags); AUDIT_ARG_MODE(mode); pdp = td->td_proc->p_pd; cmode = (((mode & ~pdp->pd_cmask) & ALLPERMS) & ~S_ISTXT); mq = NULL; if ((flags & O_CREAT) != 0 && attr != NULL) { if (attr->mq_maxmsg <= 0 || attr->mq_maxmsg > maxmsg) return (EINVAL); if (attr->mq_msgsize <= 0 || attr->mq_msgsize > maxmsgsize) return (EINVAL); } error = copyinstr(upath, path, MQFS_NAMELEN + 1, NULL); if (error) return (error); /* * The first character of name must be a slash (/) character * and the remaining characters of name cannot include any slash * characters. */ len = strlen(path); if (len < 2 || path[0] != '/' || strchr(path + 1, '/') != NULL) return (EINVAL); /* * "." and ".." are magic directories, populated on the fly, and cannot * be opened as queues. */ if (strcmp(path, "/.") == 0 || strcmp(path, "/..") == 0) return (EINVAL); AUDIT_ARG_UPATH1_CANON(path); error = falloc(td, &fp, &fd, O_CLOEXEC); if (error) return (error); sx_xlock(&mqfs_data.mi_lock); pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1, td->td_ucred); if (pn == NULL) { if (!(flags & O_CREAT)) { error = ENOENT; } else { mq = mqueue_alloc(attr); if (mq == NULL) { error = ENFILE; } else { pn = mqfs_create_file(mqfs_data.mi_root, path + 1, len - 1, td->td_ucred, cmode); if (pn == NULL) { error = ENOSPC; mqueue_free(mq); } } } if (error == 0) { pn->mn_data = mq; } } else { if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) { error = EEXIST; } else { accmode_t accmode = 0; if (flags & FREAD) accmode |= VREAD; if (flags & FWRITE) accmode |= VWRITE; error = vaccess(VREG, pn->mn_mode, pn->mn_uid, pn->mn_gid, accmode, td->td_ucred); } } if (error) { sx_xunlock(&mqfs_data.mi_lock); fdclose(td, fp, fd); fdrop(fp, td); return (error); } mqnode_addref(pn); sx_xunlock(&mqfs_data.mi_lock); finit(fp, flags & (FREAD | FWRITE | O_NONBLOCK), DTYPE_MQUEUE, pn, &mqueueops); td->td_retval[0] = fd; fdrop(fp, td); return (0); } /* * Syscall to open a message queue. */ int sys_kmq_open(struct thread *td, struct kmq_open_args *uap) { struct mq_attr attr; int flags, error; if ((uap->flags & O_ACCMODE) == O_ACCMODE || uap->flags & O_EXEC) return (EINVAL); flags = FFLAGS(uap->flags); if ((flags & O_CREAT) != 0 && uap->attr != NULL) { error = copyin(uap->attr, &attr, sizeof(attr)); if (error) return (error); } return (kern_kmq_open(td, uap->path, flags, uap->mode, uap->attr != NULL ? &attr : NULL)); } /* * Syscall to unlink a message queue. */ int sys_kmq_unlink(struct thread *td, struct kmq_unlink_args *uap) { char path[MQFS_NAMELEN+1]; struct mqfs_node *pn; int error, len; error = copyinstr(uap->path, path, MQFS_NAMELEN + 1, NULL); if (error) return (error); len = strlen(path); if (len < 2 || path[0] != '/' || strchr(path + 1, '/') != NULL) return (EINVAL); if (strcmp(path, "/.") == 0 || strcmp(path, "/..") == 0) return (EINVAL); AUDIT_ARG_UPATH1_CANON(path); sx_xlock(&mqfs_data.mi_lock); pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1, td->td_ucred); if (pn != NULL) error = do_unlink(pn, td->td_ucred); else error = ENOENT; sx_xunlock(&mqfs_data.mi_lock); return (error); } typedef int (*_fgetf)(struct thread *, int, cap_rights_t *, struct file **); /* * Get message queue by giving file slot */ static int _getmq(struct thread *td, int fd, cap_rights_t *rightsp, _fgetf func, struct file **fpp, struct mqfs_node **ppn, struct mqueue **pmq) { struct mqfs_node *pn; int error; error = func(td, fd, rightsp, fpp); if (error) return (error); if (&mqueueops != (*fpp)->f_ops) { fdrop(*fpp, td); return (EBADF); } pn = (*fpp)->f_data; if (ppn) *ppn = pn; if (pmq) *pmq = pn->mn_data; return (0); } static __inline int getmq(struct thread *td, int fd, struct file **fpp, struct mqfs_node **ppn, struct mqueue **pmq) { return _getmq(td, fd, &cap_event_rights, fget, fpp, ppn, pmq); } static __inline int getmq_read(struct thread *td, int fd, struct file **fpp, struct mqfs_node **ppn, struct mqueue **pmq) { return _getmq(td, fd, &cap_read_rights, fget_read, fpp, ppn, pmq); } static __inline int getmq_write(struct thread *td, int fd, struct file **fpp, struct mqfs_node **ppn, struct mqueue **pmq) { return _getmq(td, fd, &cap_write_rights, fget_write, fpp, ppn, pmq); } static int kern_kmq_setattr(struct thread *td, int mqd, const struct mq_attr *attr, struct mq_attr *oattr) { struct mqueue *mq; struct file *fp; u_int oflag, flag; int error; AUDIT_ARG_FD(mqd); if (attr != NULL && (attr->mq_flags & ~O_NONBLOCK) != 0) return (EINVAL); error = getmq(td, mqd, &fp, NULL, &mq); if (error) return (error); oattr->mq_maxmsg = mq->mq_maxmsg; oattr->mq_msgsize = mq->mq_msgsize; oattr->mq_curmsgs = mq->mq_curmsgs; if (attr != NULL) { do { oflag = flag = fp->f_flag; flag &= ~O_NONBLOCK; flag |= (attr->mq_flags & O_NONBLOCK); } while (atomic_cmpset_int(&fp->f_flag, oflag, flag) == 0); } else oflag = fp->f_flag; oattr->mq_flags = (O_NONBLOCK & oflag); fdrop(fp, td); return (error); } int sys_kmq_setattr(struct thread *td, struct kmq_setattr_args *uap) { struct mq_attr attr, oattr; int error; if (uap->attr != NULL) { error = copyin(uap->attr, &attr, sizeof(attr)); if (error != 0) return (error); } error = kern_kmq_setattr(td, uap->mqd, uap->attr != NULL ? &attr : NULL, &oattr); if (error == 0 && uap->oattr != NULL) { bzero(oattr.__reserved, sizeof(oattr.__reserved)); error = copyout(&oattr, uap->oattr, sizeof(oattr)); } return (error); } int sys_kmq_timedreceive(struct thread *td, struct kmq_timedreceive_args *uap) { struct mqueue *mq; struct file *fp; struct timespec *abs_timeout, ets; int error; int waitok; AUDIT_ARG_FD(uap->mqd); error = getmq_read(td, uap->mqd, &fp, NULL, &mq); if (error) return (error); if (uap->abs_timeout != NULL) { error = copyin(uap->abs_timeout, &ets, sizeof(ets)); if (error != 0) goto out; abs_timeout = &ets; } else abs_timeout = NULL; waitok = !(fp->f_flag & O_NONBLOCK); error = mqueue_receive(mq, uap->msg_ptr, uap->msg_len, uap->msg_prio, waitok, abs_timeout); out: fdrop(fp, td); return (error); } int sys_kmq_timedsend(struct thread *td, struct kmq_timedsend_args *uap) { struct mqueue *mq; struct file *fp; struct timespec *abs_timeout, ets; int error, waitok; AUDIT_ARG_FD(uap->mqd); error = getmq_write(td, uap->mqd, &fp, NULL, &mq); if (error) return (error); if (uap->abs_timeout != NULL) { error = copyin(uap->abs_timeout, &ets, sizeof(ets)); if (error != 0) goto out; abs_timeout = &ets; } else abs_timeout = NULL; waitok = !(fp->f_flag & O_NONBLOCK); error = mqueue_send(mq, uap->msg_ptr, uap->msg_len, uap->msg_prio, waitok, abs_timeout); out: fdrop(fp, td); return (error); } static int kern_kmq_notify(struct thread *td, int mqd, struct sigevent *sigev) { struct filedesc *fdp; struct proc *p; struct mqueue *mq; struct file *fp, *fp2; struct mqueue_notifier *nt, *newnt = NULL; int error; AUDIT_ARG_FD(mqd); if (sigev != NULL) { if (sigev->sigev_notify != SIGEV_SIGNAL && sigev->sigev_notify != SIGEV_THREAD_ID && sigev->sigev_notify != SIGEV_NONE) return (EINVAL); if ((sigev->sigev_notify == SIGEV_SIGNAL || sigev->sigev_notify == SIGEV_THREAD_ID) && !_SIG_VALID(sigev->sigev_signo)) return (EINVAL); } p = td->td_proc; fdp = td->td_proc->p_fd; error = getmq(td, mqd, &fp, NULL, &mq); if (error) return (error); again: FILEDESC_SLOCK(fdp); fp2 = fget_locked(fdp, mqd); if (fp2 == NULL) { FILEDESC_SUNLOCK(fdp); error = EBADF; goto out; } #ifdef CAPABILITIES error = cap_check(cap_rights(fdp, mqd), &cap_event_rights); if (error) { FILEDESC_SUNLOCK(fdp); goto out; } #endif if (fp2 != fp) { FILEDESC_SUNLOCK(fdp); error = EBADF; goto out; } mtx_lock(&mq->mq_mutex); FILEDESC_SUNLOCK(fdp); if (sigev != NULL) { if (mq->mq_notifier != NULL) { error = EBUSY; } else { PROC_LOCK(p); nt = notifier_search(p, mqd); if (nt == NULL) { if (newnt == NULL) { PROC_UNLOCK(p); mtx_unlock(&mq->mq_mutex); newnt = notifier_alloc(); goto again; } } if (nt != NULL) { sigqueue_take(&nt->nt_ksi); if (newnt != NULL) { notifier_free(newnt); newnt = NULL; } } else { nt = newnt; newnt = NULL; ksiginfo_init(&nt->nt_ksi); nt->nt_ksi.ksi_flags |= KSI_INS | KSI_EXT; nt->nt_ksi.ksi_code = SI_MESGQ; nt->nt_proc = p; nt->nt_ksi.ksi_mqd = mqd; notifier_insert(p, nt); } nt->nt_sigev = *sigev; mq->mq_notifier = nt; PROC_UNLOCK(p); /* * if there is no receivers and message queue * is not empty, we should send notification * as soon as possible. */ if (mq->mq_receivers == 0 && !TAILQ_EMPTY(&mq->mq_msgq)) mqueue_send_notification(mq); } } else { notifier_remove(p, mq, mqd); } mtx_unlock(&mq->mq_mutex); out: fdrop(fp, td); if (newnt != NULL) notifier_free(newnt); return (error); } int sys_kmq_notify(struct thread *td, struct kmq_notify_args *uap) { struct sigevent ev, *evp; int error; if (uap->sigev == NULL) { evp = NULL; } else { error = copyin(uap->sigev, &ev, sizeof(ev)); if (error != 0) return (error); evp = &ev; } return (kern_kmq_notify(td, uap->mqd, evp)); } static void mqueue_fdclose(struct thread *td, int fd, struct file *fp) { struct mqueue *mq; #ifdef INVARIANTS struct filedesc *fdp; fdp = td->td_proc->p_fd; FILEDESC_LOCK_ASSERT(fdp); #endif if (fp->f_ops == &mqueueops) { mq = FPTOMQ(fp); mtx_lock(&mq->mq_mutex); notifier_remove(td->td_proc, mq, fd); /* have to wakeup thread in same process */ if (mq->mq_flags & MQ_RSEL) { mq->mq_flags &= ~MQ_RSEL; selwakeup(&mq->mq_rsel); } if (mq->mq_flags & MQ_WSEL) { mq->mq_flags &= ~MQ_WSEL; selwakeup(&mq->mq_wsel); } mtx_unlock(&mq->mq_mutex); } } static void mq_proc_exit(void *arg __unused, struct proc *p) { struct filedesc *fdp; struct file *fp; struct mqueue *mq; int i; fdp = p->p_fd; FILEDESC_SLOCK(fdp); for (i = 0; i < fdp->fd_nfiles; ++i) { fp = fget_locked(fdp, i); if (fp != NULL && fp->f_ops == &mqueueops) { mq = FPTOMQ(fp); mtx_lock(&mq->mq_mutex); notifier_remove(p, FPTOMQ(fp), i); mtx_unlock(&mq->mq_mutex); } } FILEDESC_SUNLOCK(fdp); KASSERT(LIST_EMPTY(&p->p_mqnotifier), ("mq notifiers left")); } static int mqf_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td) { struct mqueue *mq = FPTOMQ(fp); int revents = 0; mtx_lock(&mq->mq_mutex); if (events & (POLLIN | POLLRDNORM)) { if (mq->mq_curmsgs) { revents |= events & (POLLIN | POLLRDNORM); } else { mq->mq_flags |= MQ_RSEL; selrecord(td, &mq->mq_rsel); } } if (events & POLLOUT) { if (mq->mq_curmsgs < mq->mq_maxmsg) revents |= POLLOUT; else { mq->mq_flags |= MQ_WSEL; selrecord(td, &mq->mq_wsel); } } mtx_unlock(&mq->mq_mutex); return (revents); } static int mqf_close(struct file *fp, struct thread *td) { struct mqfs_node *pn; fp->f_ops = &badfileops; pn = fp->f_data; fp->f_data = NULL; sx_xlock(&mqfs_data.mi_lock); mqnode_release(pn); sx_xunlock(&mqfs_data.mi_lock); return (0); } static int mqf_stat(struct file *fp, struct stat *st, struct ucred *active_cred, struct thread *td) { struct mqfs_node *pn = fp->f_data; bzero(st, sizeof *st); sx_xlock(&mqfs_data.mi_lock); st->st_atim = pn->mn_atime; st->st_mtim = pn->mn_mtime; st->st_ctim = pn->mn_ctime; st->st_birthtim = pn->mn_birth; st->st_uid = pn->mn_uid; st->st_gid = pn->mn_gid; st->st_mode = S_IFIFO | pn->mn_mode; sx_xunlock(&mqfs_data.mi_lock); return (0); } static int mqf_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, struct thread *td) { struct mqfs_node *pn; int error; error = 0; pn = fp->f_data; sx_xlock(&mqfs_data.mi_lock); error = vaccess(VREG, pn->mn_mode, pn->mn_uid, pn->mn_gid, VADMIN, active_cred); if (error != 0) goto out; pn->mn_mode = mode & ACCESSPERMS; out: sx_xunlock(&mqfs_data.mi_lock); return (error); } static int mqf_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, struct thread *td) { struct mqfs_node *pn; int error; error = 0; pn = fp->f_data; sx_xlock(&mqfs_data.mi_lock); if (uid == (uid_t)-1) uid = pn->mn_uid; if (gid == (gid_t)-1) gid = pn->mn_gid; if (((uid != pn->mn_uid && uid != active_cred->cr_uid) || (gid != pn->mn_gid && !groupmember(gid, active_cred))) && (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN))) goto out; pn->mn_uid = uid; pn->mn_gid = gid; out: sx_xunlock(&mqfs_data.mi_lock); return (error); } static int mqf_kqfilter(struct file *fp, struct knote *kn) { struct mqueue *mq = FPTOMQ(fp); int error = 0; if (kn->kn_filter == EVFILT_READ) { kn->kn_fop = &mq_rfiltops; knlist_add(&mq->mq_rsel.si_note, kn, 0); } else if (kn->kn_filter == EVFILT_WRITE) { kn->kn_fop = &mq_wfiltops; knlist_add(&mq->mq_wsel.si_note, kn, 0); } else error = EINVAL; return (error); } static void filt_mqdetach(struct knote *kn) { struct mqueue *mq = FPTOMQ(kn->kn_fp); if (kn->kn_filter == EVFILT_READ) knlist_remove(&mq->mq_rsel.si_note, kn, 0); else if (kn->kn_filter == EVFILT_WRITE) knlist_remove(&mq->mq_wsel.si_note, kn, 0); else panic("filt_mqdetach"); } static int filt_mqread(struct knote *kn, long hint) { struct mqueue *mq = FPTOMQ(kn->kn_fp); mtx_assert(&mq->mq_mutex, MA_OWNED); return (mq->mq_curmsgs != 0); } static int filt_mqwrite(struct knote *kn, long hint) { struct mqueue *mq = FPTOMQ(kn->kn_fp); mtx_assert(&mq->mq_mutex, MA_OWNED); return (mq->mq_curmsgs < mq->mq_maxmsg); } static int mqf_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) { kif->kf_type = KF_TYPE_MQUEUE; return (0); } static struct fileops mqueueops = { .fo_read = invfo_rdwr, .fo_write = invfo_rdwr, .fo_truncate = invfo_truncate, .fo_ioctl = invfo_ioctl, .fo_poll = mqf_poll, .fo_kqfilter = mqf_kqfilter, .fo_stat = mqf_stat, .fo_close = mqf_close, .fo_chmod = mqf_chmod, .fo_chown = mqf_chown, .fo_sendfile = invfo_sendfile, .fo_fill_kinfo = mqf_fill_kinfo, .fo_flags = DFLAG_PASSABLE, }; static struct vop_vector mqfs_vnodeops = { .vop_default = &default_vnodeops, .vop_access = mqfs_access, .vop_cachedlookup = mqfs_lookup, .vop_lookup = vfs_cache_lookup, .vop_reclaim = mqfs_reclaim, .vop_create = mqfs_create, .vop_remove = mqfs_remove, .vop_inactive = mqfs_inactive, .vop_open = mqfs_open, .vop_close = mqfs_close, .vop_getattr = mqfs_getattr, .vop_setattr = mqfs_setattr, .vop_read = mqfs_read, .vop_write = VOP_EOPNOTSUPP, .vop_readdir = mqfs_readdir, .vop_mkdir = VOP_EOPNOTSUPP, .vop_rmdir = VOP_EOPNOTSUPP }; VFS_VOP_VECTOR_REGISTER(mqfs_vnodeops); static struct vfsops mqfs_vfsops = { .vfs_init = mqfs_init, .vfs_uninit = mqfs_uninit, .vfs_mount = mqfs_mount, .vfs_unmount = mqfs_unmount, .vfs_root = mqfs_root, .vfs_statfs = mqfs_statfs, }; static struct vfsconf mqueuefs_vfsconf = { .vfc_version = VFS_VERSION, .vfc_name = "mqueuefs", .vfc_vfsops = &mqfs_vfsops, .vfc_typenum = -1, .vfc_flags = VFCF_SYNTHETIC }; static struct syscall_helper_data mq_syscalls[] = { SYSCALL_INIT_HELPER(kmq_open), SYSCALL_INIT_HELPER_F(kmq_setattr, SYF_CAPENABLED), SYSCALL_INIT_HELPER_F(kmq_timedsend, SYF_CAPENABLED), SYSCALL_INIT_HELPER_F(kmq_timedreceive, SYF_CAPENABLED), SYSCALL_INIT_HELPER_F(kmq_notify, SYF_CAPENABLED), SYSCALL_INIT_HELPER(kmq_unlink), SYSCALL_INIT_LAST }; #ifdef COMPAT_FREEBSD32 #include #include #include #include #include static void mq_attr_from32(const struct mq_attr32 *from, struct mq_attr *to) { to->mq_flags = from->mq_flags; to->mq_maxmsg = from->mq_maxmsg; to->mq_msgsize = from->mq_msgsize; to->mq_curmsgs = from->mq_curmsgs; } static void mq_attr_to32(const struct mq_attr *from, struct mq_attr32 *to) { to->mq_flags = from->mq_flags; to->mq_maxmsg = from->mq_maxmsg; to->mq_msgsize = from->mq_msgsize; to->mq_curmsgs = from->mq_curmsgs; } int freebsd32_kmq_open(struct thread *td, struct freebsd32_kmq_open_args *uap) { struct mq_attr attr; struct mq_attr32 attr32; int flags, error; if ((uap->flags & O_ACCMODE) == O_ACCMODE || uap->flags & O_EXEC) return (EINVAL); flags = FFLAGS(uap->flags); if ((flags & O_CREAT) != 0 && uap->attr != NULL) { error = copyin(uap->attr, &attr32, sizeof(attr32)); if (error) return (error); mq_attr_from32(&attr32, &attr); } return (kern_kmq_open(td, uap->path, flags, uap->mode, uap->attr != NULL ? &attr : NULL)); } int freebsd32_kmq_setattr(struct thread *td, struct freebsd32_kmq_setattr_args *uap) { struct mq_attr attr, oattr; struct mq_attr32 attr32, oattr32; int error; if (uap->attr != NULL) { error = copyin(uap->attr, &attr32, sizeof(attr32)); if (error != 0) return (error); mq_attr_from32(&attr32, &attr); } error = kern_kmq_setattr(td, uap->mqd, uap->attr != NULL ? &attr : NULL, &oattr); if (error == 0 && uap->oattr != NULL) { mq_attr_to32(&oattr, &oattr32); bzero(oattr32.__reserved, sizeof(oattr32.__reserved)); error = copyout(&oattr32, uap->oattr, sizeof(oattr32)); } return (error); } int freebsd32_kmq_timedsend(struct thread *td, struct freebsd32_kmq_timedsend_args *uap) { struct mqueue *mq; struct file *fp; struct timespec32 ets32; struct timespec *abs_timeout, ets; int error; int waitok; AUDIT_ARG_FD(uap->mqd); error = getmq_write(td, uap->mqd, &fp, NULL, &mq); if (error) return (error); if (uap->abs_timeout != NULL) { error = copyin(uap->abs_timeout, &ets32, sizeof(ets32)); if (error != 0) goto out; CP(ets32, ets, tv_sec); CP(ets32, ets, tv_nsec); abs_timeout = &ets; } else abs_timeout = NULL; waitok = !(fp->f_flag & O_NONBLOCK); error = mqueue_send(mq, uap->msg_ptr, uap->msg_len, uap->msg_prio, waitok, abs_timeout); out: fdrop(fp, td); return (error); } int freebsd32_kmq_timedreceive(struct thread *td, struct freebsd32_kmq_timedreceive_args *uap) { struct mqueue *mq; struct file *fp; struct timespec32 ets32; struct timespec *abs_timeout, ets; int error, waitok; AUDIT_ARG_FD(uap->mqd); error = getmq_read(td, uap->mqd, &fp, NULL, &mq); if (error) return (error); if (uap->abs_timeout != NULL) { error = copyin(uap->abs_timeout, &ets32, sizeof(ets32)); if (error != 0) goto out; CP(ets32, ets, tv_sec); CP(ets32, ets, tv_nsec); abs_timeout = &ets; } else abs_timeout = NULL; waitok = !(fp->f_flag & O_NONBLOCK); error = mqueue_receive(mq, uap->msg_ptr, uap->msg_len, uap->msg_prio, waitok, abs_timeout); out: fdrop(fp, td); return (error); } int freebsd32_kmq_notify(struct thread *td, struct freebsd32_kmq_notify_args *uap) { struct sigevent ev, *evp; struct sigevent32 ev32; int error; if (uap->sigev == NULL) { evp = NULL; } else { error = copyin(uap->sigev, &ev32, sizeof(ev32)); if (error != 0) return (error); error = convert_sigevent32(&ev32, &ev); if (error != 0) return (error); evp = &ev; } return (kern_kmq_notify(td, uap->mqd, evp)); } static struct syscall_helper_data mq32_syscalls[] = { SYSCALL32_INIT_HELPER(freebsd32_kmq_open), SYSCALL32_INIT_HELPER_F(freebsd32_kmq_setattr, SYF_CAPENABLED), SYSCALL32_INIT_HELPER_F(freebsd32_kmq_timedsend, SYF_CAPENABLED), SYSCALL32_INIT_HELPER_F(freebsd32_kmq_timedreceive, SYF_CAPENABLED), SYSCALL32_INIT_HELPER_F(freebsd32_kmq_notify, SYF_CAPENABLED), SYSCALL32_INIT_HELPER_COMPAT(kmq_unlink), SYSCALL_INIT_LAST }; #endif static int mqinit(void) { int error; error = syscall_helper_register(mq_syscalls, SY_THR_STATIC_KLD); if (error != 0) return (error); #ifdef COMPAT_FREEBSD32 error = syscall32_helper_register(mq32_syscalls, SY_THR_STATIC_KLD); if (error != 0) return (error); #endif return (0); } static int mqunload(void) { #ifdef COMPAT_FREEBSD32 syscall32_helper_unregister(mq32_syscalls); #endif syscall_helper_unregister(mq_syscalls); return (0); } static int mq_modload(struct module *module, int cmd, void *arg) { int error = 0; error = vfs_modevent(module, cmd, arg); if (error != 0) return (error); switch (cmd) { case MOD_LOAD: error = mqinit(); if (error != 0) mqunload(); break; case MOD_UNLOAD: error = mqunload(); break; default: break; } return (error); } static moduledata_t mqueuefs_mod = { "mqueuefs", mq_modload, &mqueuefs_vfsconf }; DECLARE_MODULE(mqueuefs, mqueuefs_mod, SI_SUB_VFS, SI_ORDER_MIDDLE); MODULE_VERSION(mqueuefs, 1); diff --git a/sys/kern/vfs_cache.c b/sys/kern/vfs_cache.c index 3c79395f5d92..ae26dc70bd05 100644 --- a/sys/kern/vfs_cache.c +++ b/sys/kern/vfs_cache.c @@ -1,6062 +1,6061 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1993, 1995 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Poul-Henning Kamp of the FreeBSD Project. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vfs_cache.c 8.5 (Berkeley) 3/22/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include "opt_ktrace.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef KTRACE #include #endif #ifdef INVARIANTS #include #endif #include #include #include #ifdef DDB #include #endif #include /* * High level overview of name caching in the VFS layer. * * Originally caching was implemented as part of UFS, later extracted to allow * use by other filesystems. A decision was made to make it optional and * completely detached from the rest of the kernel, which comes with limitations * outlined near the end of this comment block. * * This fundamental choice needs to be revisited. In the meantime, the current * state is described below. Significance of all notable routines is explained * in comments placed above their implementation. Scattered thoroughout the * file are TODO comments indicating shortcomings which can be fixed without * reworking everything (most of the fixes will likely be reusable). Various * details are omitted from this explanation to not clutter the overview, they * have to be checked by reading the code and associated commentary. * * Keep in mind that it's individual path components which are cached, not full * paths. That is, for a fully cached path "foo/bar/baz" there are 3 entries, * one for each name. * * I. Data organization * * Entries are described by "struct namecache" objects and stored in a hash * table. See cache_get_hash for more information. * * "struct vnode" contains pointers to source entries (names which can be found * when traversing through said vnode), destination entries (names of that * vnode (see "Limitations" for a breakdown on the subject) and a pointer to * the parent vnode. * * The (directory vnode; name) tuple reliably determines the target entry if * it exists. * * Since there are no small locks at this time (all are 32 bytes in size on * LP64), the code works around the problem by introducing lock arrays to * protect hash buckets and vnode lists. * * II. Filesystem integration * * Filesystems participating in name caching do the following: * - set vop_lookup routine to vfs_cache_lookup * - set vop_cachedlookup to whatever can perform the lookup if the above fails * - if they support lockless lookup (see below), vop_fplookup_vexec and * vop_fplookup_symlink are set along with the MNTK_FPLOOKUP flag on the * mount point * - call cache_purge or cache_vop_* routines to eliminate stale entries as * applicable * - call cache_enter to add entries depending on the MAKEENTRY flag * * With the above in mind, there are 2 entry points when doing lookups: * - ... -> namei -> cache_fplookup -- this is the default * - ... -> VOP_LOOKUP -> vfs_cache_lookup -- normally only called by namei * should the above fail * * Example code flow how an entry is added: * ... -> namei -> cache_fplookup -> cache_fplookup_noentry -> VOP_LOOKUP -> * vfs_cache_lookup -> VOP_CACHEDLOOKUP -> ufs_lookup_ino -> cache_enter * * III. Performance considerations * * For lockless case forward lookup avoids any writes to shared areas apart * from the terminal path component. In other words non-modifying lookups of * different files don't suffer any scalability problems in the namecache. * Looking up the same file is limited by VFS and goes beyond the scope of this * file. * * At least on amd64 the single-threaded bottleneck for long paths is hashing * (see cache_get_hash). There are cases where the code issues acquire fence * multiple times, they can be combined on architectures which suffer from it. * * For locked case each encountered vnode has to be referenced and locked in * order to be handed out to the caller (normally that's namei). This * introduces significant hit single-threaded and serialization multi-threaded. * * Reverse lookup (e.g., "getcwd") fully scales provided it is fully cached -- * avoids any writes to shared areas to any components. * * Unrelated insertions are partially serialized on updating the global entry * counter and possibly serialized on colliding bucket or vnode locks. * * IV. Observability * * Note not everything has an explicit dtrace probe nor it should have, thus * some of the one-liners below depend on implementation details. * * Examples: * * # Check what lookups failed to be handled in a lockless manner. Column 1 is * # line number, column 2 is status code (see cache_fpl_status) * dtrace -n 'vfs:fplookup:lookup:done { @[arg1, arg2] = count(); }' * * # Lengths of names added by binary name * dtrace -n 'fbt::cache_enter_time:entry { @[execname] = quantize(args[2]->cn_namelen); }' * * # Same as above but only those which exceed 64 characters * dtrace -n 'fbt::cache_enter_time:entry /args[2]->cn_namelen > 64/ { @[execname] = quantize(args[2]->cn_namelen); }' * * # Who is performing lookups with spurious slashes (e.g., "foo//bar") and what * # path is it * dtrace -n 'fbt::cache_fplookup_skip_slashes:entry { @[execname, stringof(args[0]->cnp->cn_pnbuf)] = count(); }' * * V. Limitations and implementation defects * * - since it is possible there is no entry for an open file, tools like * "procstat" may fail to resolve fd -> vnode -> path to anything * - even if a filesystem adds an entry, it may get purged (e.g., due to memory * shortage) in which case the above problem applies * - hardlinks are not tracked, thus if a vnode is reachable in more than one * way, resolving a name may return a different path than the one used to * open it (even if said path is still valid) * - by default entries are not added for newly created files * - adding an entry may need to evict negative entry first, which happens in 2 * distinct places (evicting on lookup, adding in a later VOP) making it * impossible to simply reuse it * - there is a simple scheme to evict negative entries as the cache is approaching * its capacity, but it is very unclear if doing so is a good idea to begin with * - vnodes are subject to being recycled even if target inode is left in memory, * which loses the name cache entries when it perhaps should not. in case of tmpfs * names get duplicated -- kept by filesystem itself and namecache separately * - struct namecache has a fixed size and comes in 2 variants, often wasting space. * now hard to replace with malloc due to dependence on SMR. * - lack of better integration with the kernel also turns nullfs into a layered * filesystem instead of something which can take advantage of caching */ static SYSCTL_NODE(_vfs, OID_AUTO, cache, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "Name cache"); SDT_PROVIDER_DECLARE(vfs); SDT_PROBE_DEFINE3(vfs, namecache, enter, done, "struct vnode *", "char *", "struct vnode *"); SDT_PROBE_DEFINE3(vfs, namecache, enter, duplicate, "struct vnode *", "char *", "struct vnode *"); SDT_PROBE_DEFINE2(vfs, namecache, enter_negative, done, "struct vnode *", "char *"); SDT_PROBE_DEFINE2(vfs, namecache, fullpath_smr, hit, "struct vnode *", "const char *"); SDT_PROBE_DEFINE4(vfs, namecache, fullpath_smr, miss, "struct vnode *", "struct namecache *", "int", "int"); SDT_PROBE_DEFINE1(vfs, namecache, fullpath, entry, "struct vnode *"); SDT_PROBE_DEFINE3(vfs, namecache, fullpath, hit, "struct vnode *", "char *", "struct vnode *"); SDT_PROBE_DEFINE1(vfs, namecache, fullpath, miss, "struct vnode *"); SDT_PROBE_DEFINE3(vfs, namecache, fullpath, return, "int", "struct vnode *", "char *"); SDT_PROBE_DEFINE3(vfs, namecache, lookup, hit, "struct vnode *", "char *", "struct vnode *"); SDT_PROBE_DEFINE2(vfs, namecache, lookup, hit__negative, "struct vnode *", "char *"); SDT_PROBE_DEFINE2(vfs, namecache, lookup, miss, "struct vnode *", "char *"); SDT_PROBE_DEFINE2(vfs, namecache, removecnp, hit, "struct vnode *", "struct componentname *"); SDT_PROBE_DEFINE2(vfs, namecache, removecnp, miss, "struct vnode *", "struct componentname *"); SDT_PROBE_DEFINE3(vfs, namecache, purge, done, "struct vnode *", "size_t", "size_t"); SDT_PROBE_DEFINE1(vfs, namecache, purge, batch, "int"); SDT_PROBE_DEFINE1(vfs, namecache, purge_negative, done, "struct vnode *"); SDT_PROBE_DEFINE1(vfs, namecache, purgevfs, done, "struct mount *"); SDT_PROBE_DEFINE3(vfs, namecache, zap, done, "struct vnode *", "char *", "struct vnode *"); SDT_PROBE_DEFINE2(vfs, namecache, zap_negative, done, "struct vnode *", "char *"); SDT_PROBE_DEFINE2(vfs, namecache, evict_negative, done, "struct vnode *", "char *"); SDT_PROBE_DEFINE1(vfs, namecache, symlink, alloc__fail, "size_t"); SDT_PROBE_DEFINE3(vfs, fplookup, lookup, done, "struct nameidata", "int", "bool"); SDT_PROBE_DECLARE(vfs, namei, lookup, entry); SDT_PROBE_DECLARE(vfs, namei, lookup, return); static char __read_frequently cache_fast_lookup_enabled = true; /* * This structure describes the elements in the cache of recent * names looked up by namei. */ struct negstate { u_char neg_flag; u_char neg_hit; }; _Static_assert(sizeof(struct negstate) <= sizeof(struct vnode *), "the state must fit in a union with a pointer without growing it"); struct namecache { LIST_ENTRY(namecache) nc_src; /* source vnode list */ TAILQ_ENTRY(namecache) nc_dst; /* destination vnode list */ CK_SLIST_ENTRY(namecache) nc_hash;/* hash chain */ struct vnode *nc_dvp; /* vnode of parent of name */ union { struct vnode *nu_vp; /* vnode the name refers to */ struct negstate nu_neg;/* negative entry state */ } n_un; u_char nc_flag; /* flag bits */ u_char nc_nlen; /* length of name */ char nc_name[0]; /* segment name + nul */ }; /* * struct namecache_ts repeats struct namecache layout up to the * nc_nlen member. * struct namecache_ts is used in place of struct namecache when time(s) need * to be stored. The nc_dotdottime field is used when a cache entry is mapping * both a non-dotdot directory name plus dotdot for the directory's * parent. * * See below for alignment requirement. */ struct namecache_ts { struct timespec nc_time; /* timespec provided by fs */ struct timespec nc_dotdottime; /* dotdot timespec provided by fs */ int nc_ticks; /* ticks value when entry was added */ int nc_pad; struct namecache nc_nc; }; TAILQ_HEAD(cache_freebatch, namecache); /* * At least mips n32 performs 64-bit accesses to timespec as found * in namecache_ts and requires them to be aligned. Since others * may be in the same spot suffer a little bit and enforce the * alignment for everyone. Note this is a nop for 64-bit platforms. */ #define CACHE_ZONE_ALIGNMENT UMA_ALIGNOF(time_t) /* * TODO: the initial value of CACHE_PATH_CUTOFF was inherited from the * 4.4 BSD codebase. Later on struct namecache was tweaked to become * smaller and the value was bumped to retain the total size, but it * was never re-evaluated for suitability. A simple test counting * lengths during package building shows that the value of 45 covers * about 86% of all added entries, reaching 99% at 65. * * Regardless of the above, use of dedicated zones instead of malloc may be * inducing additional waste. This may be hard to address as said zones are * tied to VFS SMR. Even if retaining them, the current split should be * re-evaluated. */ #ifdef __LP64__ #define CACHE_PATH_CUTOFF 45 #define CACHE_LARGE_PAD 6 #else #define CACHE_PATH_CUTOFF 41 #define CACHE_LARGE_PAD 2 #endif #define CACHE_ZONE_SMALL_SIZE (offsetof(struct namecache, nc_name) + CACHE_PATH_CUTOFF + 1) #define CACHE_ZONE_SMALL_TS_SIZE (offsetof(struct namecache_ts, nc_nc) + CACHE_ZONE_SMALL_SIZE) #define CACHE_ZONE_LARGE_SIZE (offsetof(struct namecache, nc_name) + NAME_MAX + 1 + CACHE_LARGE_PAD) #define CACHE_ZONE_LARGE_TS_SIZE (offsetof(struct namecache_ts, nc_nc) + CACHE_ZONE_LARGE_SIZE) _Static_assert((CACHE_ZONE_SMALL_SIZE % (CACHE_ZONE_ALIGNMENT + 1)) == 0, "bad zone size"); _Static_assert((CACHE_ZONE_SMALL_TS_SIZE % (CACHE_ZONE_ALIGNMENT + 1)) == 0, "bad zone size"); _Static_assert((CACHE_ZONE_LARGE_SIZE % (CACHE_ZONE_ALIGNMENT + 1)) == 0, "bad zone size"); _Static_assert((CACHE_ZONE_LARGE_TS_SIZE % (CACHE_ZONE_ALIGNMENT + 1)) == 0, "bad zone size"); #define nc_vp n_un.nu_vp #define nc_neg n_un.nu_neg /* * Flags in namecache.nc_flag */ #define NCF_WHITE 0x01 #define NCF_ISDOTDOT 0x02 #define NCF_TS 0x04 #define NCF_DTS 0x08 #define NCF_DVDROP 0x10 #define NCF_NEGATIVE 0x20 #define NCF_INVALID 0x40 #define NCF_WIP 0x80 /* * Flags in negstate.neg_flag */ #define NEG_HOT 0x01 static bool cache_neg_evict_cond(u_long lnumcache); /* * Mark an entry as invalid. * * This is called before it starts getting deconstructed. */ static void cache_ncp_invalidate(struct namecache *ncp) { KASSERT((ncp->nc_flag & NCF_INVALID) == 0, ("%s: entry %p already invalid", __func__, ncp)); atomic_store_char(&ncp->nc_flag, ncp->nc_flag | NCF_INVALID); atomic_thread_fence_rel(); } /* * Check whether the entry can be safely used. * * All places which elide locks are supposed to call this after they are * done with reading from an entry. */ #define cache_ncp_canuse(ncp) ({ \ struct namecache *_ncp = (ncp); \ u_char _nc_flag; \ \ atomic_thread_fence_acq(); \ _nc_flag = atomic_load_char(&_ncp->nc_flag); \ __predict_true((_nc_flag & (NCF_INVALID | NCF_WIP)) == 0); \ }) /* * Like the above but also checks NCF_WHITE. */ #define cache_fpl_neg_ncp_canuse(ncp) ({ \ struct namecache *_ncp = (ncp); \ u_char _nc_flag; \ \ atomic_thread_fence_acq(); \ _nc_flag = atomic_load_char(&_ncp->nc_flag); \ __predict_true((_nc_flag & (NCF_INVALID | NCF_WIP | NCF_WHITE)) == 0); \ }) VFS_SMR_DECLARE; static SYSCTL_NODE(_vfs_cache, OID_AUTO, param, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "Name cache parameters"); static u_int __read_mostly ncsize; /* the size as computed on creation or resizing */ SYSCTL_UINT(_vfs_cache_param, OID_AUTO, size, CTLFLAG_RW, &ncsize, 0, "Total namecache capacity"); u_int ncsizefactor = 2; SYSCTL_UINT(_vfs_cache_param, OID_AUTO, sizefactor, CTLFLAG_RW, &ncsizefactor, 0, "Size factor for namecache"); static u_long __read_mostly ncnegfactor = 5; /* ratio of negative entries */ SYSCTL_ULONG(_vfs_cache_param, OID_AUTO, negfactor, CTLFLAG_RW, &ncnegfactor, 0, "Ratio of negative namecache entries"); /* * Negative entry % of namecache capacity above which automatic eviction is allowed. * * Check cache_neg_evict_cond for details. */ static u_int ncnegminpct = 3; static u_int __read_mostly neg_min; /* the above recomputed against ncsize */ SYSCTL_UINT(_vfs_cache_param, OID_AUTO, negmin, CTLFLAG_RD, &neg_min, 0, "Negative entry count above which automatic eviction is allowed"); /* * Structures associated with name caching. */ #define NCHHASH(hash) \ (&nchashtbl[(hash) & nchash]) static __read_mostly CK_SLIST_HEAD(nchashhead, namecache) *nchashtbl;/* Hash Table */ static u_long __read_mostly nchash; /* size of hash table */ SYSCTL_ULONG(_debug, OID_AUTO, nchash, CTLFLAG_RD, &nchash, 0, "Size of namecache hash table"); static u_long __exclusive_cache_line numneg; /* number of negative entries allocated */ static u_long __exclusive_cache_line numcache;/* number of cache entries allocated */ struct nchstats nchstats; /* cache effectiveness statistics */ static bool __read_mostly cache_rename_add = true; SYSCTL_BOOL(_vfs, OID_AUTO, cache_rename_add, CTLFLAG_RW, &cache_rename_add, 0, ""); static u_int __exclusive_cache_line neg_cycle; #define ncneghash 3 #define numneglists (ncneghash + 1) struct neglist { struct mtx nl_evict_lock; struct mtx nl_lock __aligned(CACHE_LINE_SIZE); TAILQ_HEAD(, namecache) nl_list; TAILQ_HEAD(, namecache) nl_hotlist; u_long nl_hotnum; } __aligned(CACHE_LINE_SIZE); static struct neglist neglists[numneglists]; static inline struct neglist * NCP2NEGLIST(struct namecache *ncp) { return (&neglists[(((uintptr_t)(ncp) >> 8) & ncneghash)]); } static inline struct negstate * NCP2NEGSTATE(struct namecache *ncp) { MPASS(atomic_load_char(&ncp->nc_flag) & NCF_NEGATIVE); return (&ncp->nc_neg); } #define numbucketlocks (ncbuckethash + 1) static u_int __read_mostly ncbuckethash; static struct mtx_padalign __read_mostly *bucketlocks; #define HASH2BUCKETLOCK(hash) \ ((struct mtx *)(&bucketlocks[((hash) & ncbuckethash)])) #define numvnodelocks (ncvnodehash + 1) static u_int __read_mostly ncvnodehash; static struct mtx __read_mostly *vnodelocks; static inline struct mtx * VP2VNODELOCK(struct vnode *vp) { return (&vnodelocks[(((uintptr_t)(vp) >> 8) & ncvnodehash)]); } static void cache_out_ts(struct namecache *ncp, struct timespec *tsp, int *ticksp) { struct namecache_ts *ncp_ts; KASSERT((ncp->nc_flag & NCF_TS) != 0 || (tsp == NULL && ticksp == NULL), ("No NCF_TS")); if (tsp == NULL) return; ncp_ts = __containerof(ncp, struct namecache_ts, nc_nc); *tsp = ncp_ts->nc_time; *ticksp = ncp_ts->nc_ticks; } #ifdef DEBUG_CACHE static int __read_mostly doingcache = 1; /* 1 => enable the cache */ SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0, "VFS namecache enabled"); #endif /* Export size information to userland */ SYSCTL_INT(_debug_sizeof, OID_AUTO, namecache, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, sizeof(struct namecache), "sizeof(struct namecache)"); /* * The new name cache statistics */ static SYSCTL_NODE(_vfs_cache, OID_AUTO, stats, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "Name cache statistics"); #define STATNODE_ULONG(name, varname, descr) \ SYSCTL_ULONG(_vfs_cache_stats, OID_AUTO, name, CTLFLAG_RD, &varname, 0, descr); #define STATNODE_COUNTER(name, varname, descr) \ static COUNTER_U64_DEFINE_EARLY(varname); \ SYSCTL_COUNTER_U64(_vfs_cache_stats, OID_AUTO, name, CTLFLAG_RD, &varname, \ descr); STATNODE_ULONG(neg, numneg, "Number of negative cache entries"); STATNODE_ULONG(count, numcache, "Number of cache entries"); STATNODE_COUNTER(heldvnodes, numcachehv, "Number of namecache entries with vnodes held"); STATNODE_COUNTER(drops, numdrops, "Number of dropped entries due to reaching the limit"); STATNODE_COUNTER(dothits, dothits, "Number of '.' hits"); STATNODE_COUNTER(dotdothis, dotdothits, "Number of '..' hits"); STATNODE_COUNTER(miss, nummiss, "Number of cache misses"); STATNODE_COUNTER(misszap, nummisszap, "Number of cache misses we do not want to cache"); STATNODE_COUNTER(posszaps, numposzaps, "Number of cache hits (positive) we do not want to cache"); STATNODE_COUNTER(poshits, numposhits, "Number of cache hits (positive)"); STATNODE_COUNTER(negzaps, numnegzaps, "Number of cache hits (negative) we do not want to cache"); STATNODE_COUNTER(neghits, numneghits, "Number of cache hits (negative)"); /* These count for vn_getcwd(), too. */ STATNODE_COUNTER(fullpathcalls, numfullpathcalls, "Number of fullpath search calls"); STATNODE_COUNTER(fullpathfail1, numfullpathfail1, "Number of fullpath search errors (ENOTDIR)"); STATNODE_COUNTER(fullpathfail2, numfullpathfail2, "Number of fullpath search errors (VOP_VPTOCNP failures)"); STATNODE_COUNTER(fullpathfail4, numfullpathfail4, "Number of fullpath search errors (ENOMEM)"); STATNODE_COUNTER(fullpathfound, numfullpathfound, "Number of successful fullpath calls"); STATNODE_COUNTER(symlinktoobig, symlinktoobig, "Number of times symlink did not fit the cache"); /* * Debug or developer statistics. */ static SYSCTL_NODE(_vfs_cache, OID_AUTO, debug, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "Name cache debugging"); #define DEBUGNODE_ULONG(name, varname, descr) \ SYSCTL_ULONG(_vfs_cache_debug, OID_AUTO, name, CTLFLAG_RD, &varname, 0, descr); #define DEBUGNODE_COUNTER(name, varname, descr) \ static COUNTER_U64_DEFINE_EARLY(varname); \ SYSCTL_COUNTER_U64(_vfs_cache_debug, OID_AUTO, name, CTLFLAG_RD, &varname, \ descr); DEBUGNODE_COUNTER(zap_bucket_relock_success, zap_bucket_relock_success, "Number of successful removals after relocking"); static long zap_bucket_fail; DEBUGNODE_ULONG(zap_bucket_fail, zap_bucket_fail, ""); static long zap_bucket_fail2; DEBUGNODE_ULONG(zap_bucket_fail2, zap_bucket_fail2, ""); static long cache_lock_vnodes_cel_3_failures; DEBUGNODE_ULONG(vnodes_cel_3_failures, cache_lock_vnodes_cel_3_failures, "Number of times 3-way vnode locking failed"); static void cache_zap_locked(struct namecache *ncp); static int vn_fullpath_hardlink(struct nameidata *ndp, char **retbuf, char **freebuf, size_t *buflen); static int vn_fullpath_any_smr(struct vnode *vp, struct vnode *rdir, char *buf, char **retbuf, size_t *buflen, size_t addend); static int vn_fullpath_any(struct vnode *vp, struct vnode *rdir, char *buf, char **retbuf, size_t *buflen); static int vn_fullpath_dir(struct vnode *vp, struct vnode *rdir, char *buf, char **retbuf, size_t *len, size_t addend); static MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries"); static inline void cache_assert_vlp_locked(struct mtx *vlp) { if (vlp != NULL) mtx_assert(vlp, MA_OWNED); } static inline void cache_assert_vnode_locked(struct vnode *vp) { struct mtx *vlp; vlp = VP2VNODELOCK(vp); cache_assert_vlp_locked(vlp); } /* * Directory vnodes with entries are held for two reasons: * 1. make them less of a target for reclamation in vnlru * 2. suffer smaller performance penalty in locked lookup as requeieing is avoided * * It will be feasible to stop doing it altogether if all filesystems start * supporting lockless lookup. */ static void cache_hold_vnode(struct vnode *vp) { cache_assert_vnode_locked(vp); VNPASS(LIST_EMPTY(&vp->v_cache_src), vp); vhold(vp); counter_u64_add(numcachehv, 1); } static void cache_drop_vnode(struct vnode *vp) { /* * Called after all locks are dropped, meaning we can't assert * on the state of v_cache_src. */ vdrop(vp); counter_u64_add(numcachehv, -1); } /* * UMA zones. */ static uma_zone_t __read_mostly cache_zone_small; static uma_zone_t __read_mostly cache_zone_small_ts; static uma_zone_t __read_mostly cache_zone_large; static uma_zone_t __read_mostly cache_zone_large_ts; char * cache_symlink_alloc(size_t size, int flags) { if (size < CACHE_ZONE_SMALL_SIZE) { return (uma_zalloc_smr(cache_zone_small, flags)); } if (size < CACHE_ZONE_LARGE_SIZE) { return (uma_zalloc_smr(cache_zone_large, flags)); } counter_u64_add(symlinktoobig, 1); SDT_PROBE1(vfs, namecache, symlink, alloc__fail, size); return (NULL); } void cache_symlink_free(char *string, size_t size) { MPASS(string != NULL); KASSERT(size < CACHE_ZONE_LARGE_SIZE, ("%s: size %zu too big", __func__, size)); if (size < CACHE_ZONE_SMALL_SIZE) { uma_zfree_smr(cache_zone_small, string); return; } if (size < CACHE_ZONE_LARGE_SIZE) { uma_zfree_smr(cache_zone_large, string); return; } __assert_unreachable(); } static struct namecache * cache_alloc_uma(int len, bool ts) { struct namecache_ts *ncp_ts; struct namecache *ncp; if (__predict_false(ts)) { if (len <= CACHE_PATH_CUTOFF) ncp_ts = uma_zalloc_smr(cache_zone_small_ts, M_WAITOK); else ncp_ts = uma_zalloc_smr(cache_zone_large_ts, M_WAITOK); ncp = &ncp_ts->nc_nc; } else { if (len <= CACHE_PATH_CUTOFF) ncp = uma_zalloc_smr(cache_zone_small, M_WAITOK); else ncp = uma_zalloc_smr(cache_zone_large, M_WAITOK); } return (ncp); } static void cache_free_uma(struct namecache *ncp) { struct namecache_ts *ncp_ts; if (__predict_false(ncp->nc_flag & NCF_TS)) { ncp_ts = __containerof(ncp, struct namecache_ts, nc_nc); if (ncp->nc_nlen <= CACHE_PATH_CUTOFF) uma_zfree_smr(cache_zone_small_ts, ncp_ts); else uma_zfree_smr(cache_zone_large_ts, ncp_ts); } else { if (ncp->nc_nlen <= CACHE_PATH_CUTOFF) uma_zfree_smr(cache_zone_small, ncp); else uma_zfree_smr(cache_zone_large, ncp); } } static struct namecache * cache_alloc(int len, bool ts) { u_long lnumcache; /* * Avoid blowout in namecache entries. * * Bugs: * 1. filesystems may end up trying to add an already existing entry * (for example this can happen after a cache miss during concurrent * lookup), in which case we will call cache_neg_evict despite not * adding anything. * 2. the routine may fail to free anything and no provisions are made * to make it try harder (see the inside for failure modes) * 3. it only ever looks at negative entries. */ lnumcache = atomic_fetchadd_long(&numcache, 1) + 1; if (cache_neg_evict_cond(lnumcache)) { lnumcache = atomic_load_long(&numcache); } if (__predict_false(lnumcache >= ncsize)) { atomic_subtract_long(&numcache, 1); counter_u64_add(numdrops, 1); return (NULL); } return (cache_alloc_uma(len, ts)); } static void cache_free(struct namecache *ncp) { MPASS(ncp != NULL); if ((ncp->nc_flag & NCF_DVDROP) != 0) { cache_drop_vnode(ncp->nc_dvp); } cache_free_uma(ncp); atomic_subtract_long(&numcache, 1); } static void cache_free_batch(struct cache_freebatch *batch) { struct namecache *ncp, *nnp; int i; i = 0; if (TAILQ_EMPTY(batch)) goto out; TAILQ_FOREACH_SAFE(ncp, batch, nc_dst, nnp) { if ((ncp->nc_flag & NCF_DVDROP) != 0) { cache_drop_vnode(ncp->nc_dvp); } cache_free_uma(ncp); i++; } atomic_subtract_long(&numcache, i); out: SDT_PROBE1(vfs, namecache, purge, batch, i); } /* * Hashing. * * The code was made to use FNV in 2001 and this choice needs to be revisited. * * Short summary of the difficulty: * The longest name which can be inserted is NAME_MAX characters in length (or * 255 at the time of writing this comment), while majority of names used in * practice are significantly shorter (mostly below 10). More importantly * majority of lookups performed find names are even shorter than that. * * This poses a problem where hashes which do better than FNV past word size * (or so) tend to come with additional overhead when finalizing the result, * making them noticeably slower for the most commonly used range. * * Consider a path like: /usr/obj/usr/src/sys/amd64/GENERIC/vnode_if.c * * When looking it up the most time consuming part by a large margin (at least * on amd64) is hashing. Replacing FNV with something which pessimizes short * input would make the slowest part stand out even more. */ /* * TODO: With the value stored we can do better than computing the hash based * on the address. */ static void cache_prehash(struct vnode *vp) { vp->v_nchash = fnv_32_buf(&vp, sizeof(vp), FNV1_32_INIT); } static uint32_t cache_get_hash(char *name, u_char len, struct vnode *dvp) { return (fnv_32_buf(name, len, dvp->v_nchash)); } static uint32_t cache_get_hash_iter_start(struct vnode *dvp) { return (dvp->v_nchash); } static uint32_t cache_get_hash_iter(char c, uint32_t hash) { return (fnv_32_buf(&c, 1, hash)); } static uint32_t cache_get_hash_iter_finish(uint32_t hash) { return (hash); } static inline struct nchashhead * NCP2BUCKET(struct namecache *ncp) { uint32_t hash; hash = cache_get_hash(ncp->nc_name, ncp->nc_nlen, ncp->nc_dvp); return (NCHHASH(hash)); } static inline struct mtx * NCP2BUCKETLOCK(struct namecache *ncp) { uint32_t hash; hash = cache_get_hash(ncp->nc_name, ncp->nc_nlen, ncp->nc_dvp); return (HASH2BUCKETLOCK(hash)); } #ifdef INVARIANTS static void cache_assert_bucket_locked(struct namecache *ncp) { struct mtx *blp; blp = NCP2BUCKETLOCK(ncp); mtx_assert(blp, MA_OWNED); } static void cache_assert_bucket_unlocked(struct namecache *ncp) { struct mtx *blp; blp = NCP2BUCKETLOCK(ncp); mtx_assert(blp, MA_NOTOWNED); } #else #define cache_assert_bucket_locked(x) do { } while (0) #define cache_assert_bucket_unlocked(x) do { } while (0) #endif #define cache_sort_vnodes(x, y) _cache_sort_vnodes((void **)(x), (void **)(y)) static void _cache_sort_vnodes(void **p1, void **p2) { void *tmp; MPASS(*p1 != NULL || *p2 != NULL); if (*p1 > *p2) { tmp = *p2; *p2 = *p1; *p1 = tmp; } } static void cache_lock_all_buckets(void) { u_int i; for (i = 0; i < numbucketlocks; i++) mtx_lock(&bucketlocks[i]); } static void cache_unlock_all_buckets(void) { u_int i; for (i = 0; i < numbucketlocks; i++) mtx_unlock(&bucketlocks[i]); } static void cache_lock_all_vnodes(void) { u_int i; for (i = 0; i < numvnodelocks; i++) mtx_lock(&vnodelocks[i]); } static void cache_unlock_all_vnodes(void) { u_int i; for (i = 0; i < numvnodelocks; i++) mtx_unlock(&vnodelocks[i]); } static int cache_trylock_vnodes(struct mtx *vlp1, struct mtx *vlp2) { cache_sort_vnodes(&vlp1, &vlp2); if (vlp1 != NULL) { if (!mtx_trylock(vlp1)) return (EAGAIN); } if (!mtx_trylock(vlp2)) { if (vlp1 != NULL) mtx_unlock(vlp1); return (EAGAIN); } return (0); } static void cache_lock_vnodes(struct mtx *vlp1, struct mtx *vlp2) { MPASS(vlp1 != NULL || vlp2 != NULL); MPASS(vlp1 <= vlp2); if (vlp1 != NULL) mtx_lock(vlp1); if (vlp2 != NULL) mtx_lock(vlp2); } static void cache_unlock_vnodes(struct mtx *vlp1, struct mtx *vlp2) { MPASS(vlp1 != NULL || vlp2 != NULL); if (vlp1 != NULL) mtx_unlock(vlp1); if (vlp2 != NULL) mtx_unlock(vlp2); } static int sysctl_nchstats(SYSCTL_HANDLER_ARGS) { struct nchstats snap; if (req->oldptr == NULL) return (SYSCTL_OUT(req, 0, sizeof(snap))); snap = nchstats; snap.ncs_goodhits = counter_u64_fetch(numposhits); snap.ncs_neghits = counter_u64_fetch(numneghits); snap.ncs_badhits = counter_u64_fetch(numposzaps) + counter_u64_fetch(numnegzaps); snap.ncs_miss = counter_u64_fetch(nummisszap) + counter_u64_fetch(nummiss); return (SYSCTL_OUT(req, &snap, sizeof(snap))); } SYSCTL_PROC(_vfs_cache, OID_AUTO, nchstats, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 0, sysctl_nchstats, "LU", "VFS cache effectiveness statistics"); static void cache_recalc_neg_min(u_int val) { neg_min = (ncsize * val) / 100; } static int sysctl_negminpct(SYSCTL_HANDLER_ARGS) { u_int val; int error; val = ncnegminpct; error = sysctl_handle_int(oidp, &val, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (val == ncnegminpct) return (0); if (val < 0 || val > 99) return (EINVAL); ncnegminpct = val; cache_recalc_neg_min(val); return (0); } SYSCTL_PROC(_vfs_cache_param, OID_AUTO, negminpct, CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0, sysctl_negminpct, "I", "Negative entry \% of namecache capacity above which automatic eviction is allowed"); #ifdef DIAGNOSTIC /* * Grab an atomic snapshot of the name cache hash chain lengths */ static SYSCTL_NODE(_debug, OID_AUTO, hashstat, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, "hash table stats"); static int sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS) { struct nchashhead *ncpp; struct namecache *ncp; int i, error, n_nchash, *cntbuf; retry: n_nchash = nchash + 1; /* nchash is max index, not count */ if (req->oldptr == NULL) return SYSCTL_OUT(req, 0, n_nchash * sizeof(int)); cntbuf = malloc(n_nchash * sizeof(int), M_TEMP, M_ZERO | M_WAITOK); cache_lock_all_buckets(); if (n_nchash != nchash + 1) { cache_unlock_all_buckets(); free(cntbuf, M_TEMP); goto retry; } /* Scan hash tables counting entries */ for (ncpp = nchashtbl, i = 0; i < n_nchash; ncpp++, i++) CK_SLIST_FOREACH(ncp, ncpp, nc_hash) cntbuf[i]++; cache_unlock_all_buckets(); for (error = 0, i = 0; i < n_nchash; i++) if ((error = SYSCTL_OUT(req, &cntbuf[i], sizeof(int))) != 0) break; free(cntbuf, M_TEMP); return (error); } SYSCTL_PROC(_debug_hashstat, OID_AUTO, rawnchash, CTLTYPE_INT|CTLFLAG_RD| CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_rawnchash, "S,int", "nchash chain lengths"); static int sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS) { int error; struct nchashhead *ncpp; struct namecache *ncp; int n_nchash; int count, maxlength, used, pct; if (!req->oldptr) return SYSCTL_OUT(req, 0, 4 * sizeof(int)); cache_lock_all_buckets(); n_nchash = nchash + 1; /* nchash is max index, not count */ used = 0; maxlength = 0; /* Scan hash tables for applicable entries */ for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) { count = 0; CK_SLIST_FOREACH(ncp, ncpp, nc_hash) { count++; } if (count) used++; if (maxlength < count) maxlength = count; } n_nchash = nchash + 1; cache_unlock_all_buckets(); pct = (used * 100) / (n_nchash / 100); error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash)); if (error) return (error); error = SYSCTL_OUT(req, &used, sizeof(used)); if (error) return (error); error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength)); if (error) return (error); error = SYSCTL_OUT(req, &pct, sizeof(pct)); if (error) return (error); return (0); } SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD| CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_nchash, "I", "nchash statistics (number of total/used buckets, maximum chain length, usage percentage)"); #endif /* * Negative entries management * * Various workloads create plenty of negative entries and barely use them * afterwards. Moreover malicious users can keep performing bogus lookups * adding even more entries. For example "make tinderbox" as of writing this * comment ends up with 2.6M namecache entries in total, 1.2M of which are * negative. * * As such, a rather aggressive eviction method is needed. The currently * employed method is a placeholder. * * Entries are split over numneglists separate lists, each of which is further * split into hot and cold entries. Entries get promoted after getting a hit. * Eviction happens on addition of new entry. */ static SYSCTL_NODE(_vfs_cache, OID_AUTO, neg, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "Name cache negative entry statistics"); SYSCTL_ULONG(_vfs_cache_neg, OID_AUTO, count, CTLFLAG_RD, &numneg, 0, "Number of negative cache entries"); static COUNTER_U64_DEFINE_EARLY(neg_created); SYSCTL_COUNTER_U64(_vfs_cache_neg, OID_AUTO, created, CTLFLAG_RD, &neg_created, "Number of created negative entries"); static COUNTER_U64_DEFINE_EARLY(neg_evicted); SYSCTL_COUNTER_U64(_vfs_cache_neg, OID_AUTO, evicted, CTLFLAG_RD, &neg_evicted, "Number of evicted negative entries"); static COUNTER_U64_DEFINE_EARLY(neg_evict_skipped_empty); SYSCTL_COUNTER_U64(_vfs_cache_neg, OID_AUTO, evict_skipped_empty, CTLFLAG_RD, &neg_evict_skipped_empty, "Number of times evicting failed due to lack of entries"); static COUNTER_U64_DEFINE_EARLY(neg_evict_skipped_missed); SYSCTL_COUNTER_U64(_vfs_cache_neg, OID_AUTO, evict_skipped_missed, CTLFLAG_RD, &neg_evict_skipped_missed, "Number of times evicting failed due to target entry disappearing"); static COUNTER_U64_DEFINE_EARLY(neg_evict_skipped_contended); SYSCTL_COUNTER_U64(_vfs_cache_neg, OID_AUTO, evict_skipped_contended, CTLFLAG_RD, &neg_evict_skipped_contended, "Number of times evicting failed due to contention"); SYSCTL_COUNTER_U64(_vfs_cache_neg, OID_AUTO, hits, CTLFLAG_RD, &numneghits, "Number of cache hits (negative)"); static int sysctl_neg_hot(SYSCTL_HANDLER_ARGS) { int i, out; out = 0; for (i = 0; i < numneglists; i++) out += neglists[i].nl_hotnum; return (SYSCTL_OUT(req, &out, sizeof(out))); } SYSCTL_PROC(_vfs_cache_neg, OID_AUTO, hot, CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 0, sysctl_neg_hot, "I", "Number of hot negative entries"); static void cache_neg_init(struct namecache *ncp) { struct negstate *ns; ncp->nc_flag |= NCF_NEGATIVE; ns = NCP2NEGSTATE(ncp); ns->neg_flag = 0; ns->neg_hit = 0; counter_u64_add(neg_created, 1); } #define CACHE_NEG_PROMOTION_THRESH 2 static bool cache_neg_hit_prep(struct namecache *ncp) { struct negstate *ns; u_char n; ns = NCP2NEGSTATE(ncp); n = atomic_load_char(&ns->neg_hit); for (;;) { if (n >= CACHE_NEG_PROMOTION_THRESH) return (false); if (atomic_fcmpset_8(&ns->neg_hit, &n, n + 1)) break; } return (n + 1 == CACHE_NEG_PROMOTION_THRESH); } /* * Nothing to do here but it is provided for completeness as some * cache_neg_hit_prep callers may end up returning without even * trying to promote. */ #define cache_neg_hit_abort(ncp) do { } while (0) static void cache_neg_hit_finish(struct namecache *ncp) { SDT_PROBE2(vfs, namecache, lookup, hit__negative, ncp->nc_dvp, ncp->nc_name); counter_u64_add(numneghits, 1); } /* * Move a negative entry to the hot list. */ static void cache_neg_promote_locked(struct namecache *ncp) { struct neglist *nl; struct negstate *ns; ns = NCP2NEGSTATE(ncp); nl = NCP2NEGLIST(ncp); mtx_assert(&nl->nl_lock, MA_OWNED); if ((ns->neg_flag & NEG_HOT) == 0) { TAILQ_REMOVE(&nl->nl_list, ncp, nc_dst); TAILQ_INSERT_TAIL(&nl->nl_hotlist, ncp, nc_dst); nl->nl_hotnum++; ns->neg_flag |= NEG_HOT; } } /* * Move a hot negative entry to the cold list. */ static void cache_neg_demote_locked(struct namecache *ncp) { struct neglist *nl; struct negstate *ns; ns = NCP2NEGSTATE(ncp); nl = NCP2NEGLIST(ncp); mtx_assert(&nl->nl_lock, MA_OWNED); MPASS(ns->neg_flag & NEG_HOT); TAILQ_REMOVE(&nl->nl_hotlist, ncp, nc_dst); TAILQ_INSERT_TAIL(&nl->nl_list, ncp, nc_dst); nl->nl_hotnum--; ns->neg_flag &= ~NEG_HOT; atomic_store_char(&ns->neg_hit, 0); } /* * Move a negative entry to the hot list if it matches the lookup. * * We have to take locks, but they may be contended and in the worst * case we may need to go off CPU. We don't want to spin within the * smr section and we can't block with it. Exiting the section means * the found entry could have been evicted. We are going to look it * up again. */ static bool cache_neg_promote_cond(struct vnode *dvp, struct componentname *cnp, struct namecache *oncp, uint32_t hash) { struct namecache *ncp; struct neglist *nl; u_char nc_flag; nl = NCP2NEGLIST(oncp); mtx_lock(&nl->nl_lock); /* * For hash iteration. */ vfs_smr_enter(); /* * Avoid all surprises by only succeeding if we got the same entry and * bailing completely otherwise. * XXX There are no provisions to keep the vnode around, meaning we may * end up promoting a negative entry for a *new* vnode and returning * ENOENT on its account. This is the error we want to return anyway * and promotion is harmless. * * In particular at this point there can be a new ncp which matches the * search but hashes to a different neglist. */ CK_SLIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) { if (ncp == oncp) break; } /* * No match to begin with. */ if (__predict_false(ncp == NULL)) { goto out_abort; } /* * The newly found entry may be something different... */ if (!(ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen && !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen))) { goto out_abort; } /* * ... and not even negative. */ nc_flag = atomic_load_char(&ncp->nc_flag); if ((nc_flag & NCF_NEGATIVE) == 0) { goto out_abort; } if (!cache_ncp_canuse(ncp)) { goto out_abort; } cache_neg_promote_locked(ncp); cache_neg_hit_finish(ncp); vfs_smr_exit(); mtx_unlock(&nl->nl_lock); return (true); out_abort: vfs_smr_exit(); mtx_unlock(&nl->nl_lock); return (false); } static void cache_neg_promote(struct namecache *ncp) { struct neglist *nl; nl = NCP2NEGLIST(ncp); mtx_lock(&nl->nl_lock); cache_neg_promote_locked(ncp); mtx_unlock(&nl->nl_lock); } static void cache_neg_insert(struct namecache *ncp) { struct neglist *nl; MPASS(ncp->nc_flag & NCF_NEGATIVE); cache_assert_bucket_locked(ncp); nl = NCP2NEGLIST(ncp); mtx_lock(&nl->nl_lock); TAILQ_INSERT_TAIL(&nl->nl_list, ncp, nc_dst); mtx_unlock(&nl->nl_lock); atomic_add_long(&numneg, 1); } static void cache_neg_remove(struct namecache *ncp) { struct neglist *nl; struct negstate *ns; cache_assert_bucket_locked(ncp); nl = NCP2NEGLIST(ncp); ns = NCP2NEGSTATE(ncp); mtx_lock(&nl->nl_lock); if ((ns->neg_flag & NEG_HOT) != 0) { TAILQ_REMOVE(&nl->nl_hotlist, ncp, nc_dst); nl->nl_hotnum--; } else { TAILQ_REMOVE(&nl->nl_list, ncp, nc_dst); } mtx_unlock(&nl->nl_lock); atomic_subtract_long(&numneg, 1); } static struct neglist * cache_neg_evict_select_list(void) { struct neglist *nl; u_int c; c = atomic_fetchadd_int(&neg_cycle, 1) + 1; nl = &neglists[c % numneglists]; if (!mtx_trylock(&nl->nl_evict_lock)) { counter_u64_add(neg_evict_skipped_contended, 1); return (NULL); } return (nl); } static struct namecache * cache_neg_evict_select_entry(struct neglist *nl) { struct namecache *ncp, *lncp; struct negstate *ns, *lns; int i; mtx_assert(&nl->nl_evict_lock, MA_OWNED); mtx_assert(&nl->nl_lock, MA_OWNED); ncp = TAILQ_FIRST(&nl->nl_list); if (ncp == NULL) return (NULL); lncp = ncp; lns = NCP2NEGSTATE(lncp); for (i = 1; i < 4; i++) { ncp = TAILQ_NEXT(ncp, nc_dst); if (ncp == NULL) break; ns = NCP2NEGSTATE(ncp); if (ns->neg_hit < lns->neg_hit) { lncp = ncp; lns = ns; } } return (lncp); } static bool cache_neg_evict(void) { struct namecache *ncp, *ncp2; struct neglist *nl; struct vnode *dvp; struct mtx *dvlp; struct mtx *blp; uint32_t hash; u_char nlen; bool evicted; nl = cache_neg_evict_select_list(); if (nl == NULL) { return (false); } mtx_lock(&nl->nl_lock); ncp = TAILQ_FIRST(&nl->nl_hotlist); if (ncp != NULL) { cache_neg_demote_locked(ncp); } ncp = cache_neg_evict_select_entry(nl); if (ncp == NULL) { counter_u64_add(neg_evict_skipped_empty, 1); mtx_unlock(&nl->nl_lock); mtx_unlock(&nl->nl_evict_lock); return (false); } nlen = ncp->nc_nlen; dvp = ncp->nc_dvp; hash = cache_get_hash(ncp->nc_name, nlen, dvp); dvlp = VP2VNODELOCK(dvp); blp = HASH2BUCKETLOCK(hash); mtx_unlock(&nl->nl_lock); mtx_unlock(&nl->nl_evict_lock); mtx_lock(dvlp); mtx_lock(blp); /* * Note that since all locks were dropped above, the entry may be * gone or reallocated to be something else. */ CK_SLIST_FOREACH(ncp2, (NCHHASH(hash)), nc_hash) { if (ncp2 == ncp && ncp2->nc_dvp == dvp && ncp2->nc_nlen == nlen && (ncp2->nc_flag & NCF_NEGATIVE) != 0) break; } if (ncp2 == NULL) { counter_u64_add(neg_evict_skipped_missed, 1); ncp = NULL; evicted = false; } else { MPASS(dvlp == VP2VNODELOCK(ncp->nc_dvp)); MPASS(blp == NCP2BUCKETLOCK(ncp)); SDT_PROBE2(vfs, namecache, evict_negative, done, ncp->nc_dvp, ncp->nc_name); cache_zap_locked(ncp); counter_u64_add(neg_evicted, 1); evicted = true; } mtx_unlock(blp); mtx_unlock(dvlp); if (ncp != NULL) cache_free(ncp); return (evicted); } /* * Maybe evict a negative entry to create more room. * * The ncnegfactor parameter limits what fraction of the total count * can comprise of negative entries. However, if the cache is just * warming up this leads to excessive evictions. As such, ncnegminpct * (recomputed to neg_min) dictates whether the above should be * applied. * * Try evicting if the cache is close to full capacity regardless of * other considerations. */ static bool cache_neg_evict_cond(u_long lnumcache) { u_long lnumneg; if (ncsize - 1000 < lnumcache) goto out_evict; lnumneg = atomic_load_long(&numneg); if (lnumneg < neg_min) return (false); if (lnumneg * ncnegfactor < lnumcache) return (false); out_evict: return (cache_neg_evict()); } /* * cache_zap_locked(): * * Removes a namecache entry from cache, whether it contains an actual * pointer to a vnode or if it is just a negative cache entry. */ static void cache_zap_locked(struct namecache *ncp) { struct nchashhead *ncpp; struct vnode *dvp, *vp; dvp = ncp->nc_dvp; vp = ncp->nc_vp; if (!(ncp->nc_flag & NCF_NEGATIVE)) cache_assert_vnode_locked(vp); cache_assert_vnode_locked(dvp); cache_assert_bucket_locked(ncp); cache_ncp_invalidate(ncp); ncpp = NCP2BUCKET(ncp); CK_SLIST_REMOVE(ncpp, ncp, namecache, nc_hash); if (!(ncp->nc_flag & NCF_NEGATIVE)) { SDT_PROBE3(vfs, namecache, zap, done, dvp, ncp->nc_name, vp); TAILQ_REMOVE(&vp->v_cache_dst, ncp, nc_dst); if (ncp == vp->v_cache_dd) { atomic_store_ptr(&vp->v_cache_dd, NULL); } } else { SDT_PROBE2(vfs, namecache, zap_negative, done, dvp, ncp->nc_name); cache_neg_remove(ncp); } if (ncp->nc_flag & NCF_ISDOTDOT) { if (ncp == dvp->v_cache_dd) { atomic_store_ptr(&dvp->v_cache_dd, NULL); } } else { LIST_REMOVE(ncp, nc_src); if (LIST_EMPTY(&dvp->v_cache_src)) { ncp->nc_flag |= NCF_DVDROP; } } } static void cache_zap_negative_locked_vnode_kl(struct namecache *ncp, struct vnode *vp) { struct mtx *blp; MPASS(ncp->nc_dvp == vp); MPASS(ncp->nc_flag & NCF_NEGATIVE); cache_assert_vnode_locked(vp); blp = NCP2BUCKETLOCK(ncp); mtx_lock(blp); cache_zap_locked(ncp); mtx_unlock(blp); } static bool cache_zap_locked_vnode_kl2(struct namecache *ncp, struct vnode *vp, struct mtx **vlpp) { struct mtx *pvlp, *vlp1, *vlp2, *to_unlock; struct mtx *blp; MPASS(vp == ncp->nc_dvp || vp == ncp->nc_vp); cache_assert_vnode_locked(vp); if (ncp->nc_flag & NCF_NEGATIVE) { if (*vlpp != NULL) { mtx_unlock(*vlpp); *vlpp = NULL; } cache_zap_negative_locked_vnode_kl(ncp, vp); return (true); } pvlp = VP2VNODELOCK(vp); blp = NCP2BUCKETLOCK(ncp); vlp1 = VP2VNODELOCK(ncp->nc_dvp); vlp2 = VP2VNODELOCK(ncp->nc_vp); if (*vlpp == vlp1 || *vlpp == vlp2) { to_unlock = *vlpp; *vlpp = NULL; } else { if (*vlpp != NULL) { mtx_unlock(*vlpp); *vlpp = NULL; } cache_sort_vnodes(&vlp1, &vlp2); if (vlp1 == pvlp) { mtx_lock(vlp2); to_unlock = vlp2; } else { if (!mtx_trylock(vlp1)) goto out_relock; to_unlock = vlp1; } } mtx_lock(blp); cache_zap_locked(ncp); mtx_unlock(blp); if (to_unlock != NULL) mtx_unlock(to_unlock); return (true); out_relock: mtx_unlock(vlp2); mtx_lock(vlp1); mtx_lock(vlp2); MPASS(*vlpp == NULL); *vlpp = vlp1; return (false); } /* * If trylocking failed we can get here. We know enough to take all needed locks * in the right order and re-lookup the entry. */ static int cache_zap_unlocked_bucket(struct namecache *ncp, struct componentname *cnp, struct vnode *dvp, struct mtx *dvlp, struct mtx *vlp, uint32_t hash, struct mtx *blp) { struct namecache *rncp; cache_assert_bucket_unlocked(ncp); cache_sort_vnodes(&dvlp, &vlp); cache_lock_vnodes(dvlp, vlp); mtx_lock(blp); CK_SLIST_FOREACH(rncp, (NCHHASH(hash)), nc_hash) { if (rncp == ncp && rncp->nc_dvp == dvp && rncp->nc_nlen == cnp->cn_namelen && !bcmp(rncp->nc_name, cnp->cn_nameptr, rncp->nc_nlen)) break; } if (rncp != NULL) { cache_zap_locked(rncp); mtx_unlock(blp); cache_unlock_vnodes(dvlp, vlp); counter_u64_add(zap_bucket_relock_success, 1); return (0); } mtx_unlock(blp); cache_unlock_vnodes(dvlp, vlp); return (EAGAIN); } static int __noinline cache_zap_locked_bucket(struct namecache *ncp, struct componentname *cnp, uint32_t hash, struct mtx *blp) { struct mtx *dvlp, *vlp; struct vnode *dvp; cache_assert_bucket_locked(ncp); dvlp = VP2VNODELOCK(ncp->nc_dvp); vlp = NULL; if (!(ncp->nc_flag & NCF_NEGATIVE)) vlp = VP2VNODELOCK(ncp->nc_vp); if (cache_trylock_vnodes(dvlp, vlp) == 0) { cache_zap_locked(ncp); mtx_unlock(blp); cache_unlock_vnodes(dvlp, vlp); return (0); } dvp = ncp->nc_dvp; mtx_unlock(blp); return (cache_zap_unlocked_bucket(ncp, cnp, dvp, dvlp, vlp, hash, blp)); } static __noinline int cache_remove_cnp(struct vnode *dvp, struct componentname *cnp) { struct namecache *ncp; struct mtx *blp; struct mtx *dvlp, *dvlp2; uint32_t hash; int error; if (cnp->cn_namelen == 2 && cnp->cn_nameptr[0] == '.' && cnp->cn_nameptr[1] == '.') { dvlp = VP2VNODELOCK(dvp); dvlp2 = NULL; mtx_lock(dvlp); retry_dotdot: ncp = dvp->v_cache_dd; if (ncp == NULL) { mtx_unlock(dvlp); if (dvlp2 != NULL) mtx_unlock(dvlp2); SDT_PROBE2(vfs, namecache, removecnp, miss, dvp, cnp); return (0); } if ((ncp->nc_flag & NCF_ISDOTDOT) != 0) { if (!cache_zap_locked_vnode_kl2(ncp, dvp, &dvlp2)) goto retry_dotdot; MPASS(dvp->v_cache_dd == NULL); mtx_unlock(dvlp); if (dvlp2 != NULL) mtx_unlock(dvlp2); cache_free(ncp); } else { atomic_store_ptr(&dvp->v_cache_dd, NULL); mtx_unlock(dvlp); if (dvlp2 != NULL) mtx_unlock(dvlp2); } SDT_PROBE2(vfs, namecache, removecnp, hit, dvp, cnp); return (1); } hash = cache_get_hash(cnp->cn_nameptr, cnp->cn_namelen, dvp); blp = HASH2BUCKETLOCK(hash); retry: if (CK_SLIST_EMPTY(NCHHASH(hash))) goto out_no_entry; mtx_lock(blp); CK_SLIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) { if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen && !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen)) break; } if (ncp == NULL) { mtx_unlock(blp); goto out_no_entry; } error = cache_zap_locked_bucket(ncp, cnp, hash, blp); if (__predict_false(error != 0)) { zap_bucket_fail++; goto retry; } counter_u64_add(numposzaps, 1); SDT_PROBE2(vfs, namecache, removecnp, hit, dvp, cnp); cache_free(ncp); return (1); out_no_entry: counter_u64_add(nummisszap, 1); SDT_PROBE2(vfs, namecache, removecnp, miss, dvp, cnp); return (0); } static int __noinline cache_lookup_dot(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct timespec *tsp, int *ticksp) { int ltype; *vpp = dvp; counter_u64_add(dothits, 1); SDT_PROBE3(vfs, namecache, lookup, hit, dvp, ".", *vpp); if (tsp != NULL) timespecclear(tsp); if (ticksp != NULL) *ticksp = ticks; vrefact(*vpp); /* * When we lookup "." we still can be asked to lock it * differently... */ ltype = cnp->cn_lkflags & LK_TYPE_MASK; if (ltype != VOP_ISLOCKED(*vpp)) { if (ltype == LK_EXCLUSIVE) { vn_lock(*vpp, LK_UPGRADE | LK_RETRY); if (VN_IS_DOOMED((*vpp))) { /* forced unmount */ vrele(*vpp); *vpp = NULL; return (ENOENT); } } else vn_lock(*vpp, LK_DOWNGRADE | LK_RETRY); } return (-1); } static int __noinline cache_lookup_dotdot(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct timespec *tsp, int *ticksp) { struct namecache_ts *ncp_ts; struct namecache *ncp; struct mtx *dvlp; enum vgetstate vs; int error, ltype; bool whiteout; MPASS((cnp->cn_flags & ISDOTDOT) != 0); if ((cnp->cn_flags & MAKEENTRY) == 0) { cache_remove_cnp(dvp, cnp); return (0); } counter_u64_add(dotdothits, 1); retry: dvlp = VP2VNODELOCK(dvp); mtx_lock(dvlp); ncp = dvp->v_cache_dd; if (ncp == NULL) { SDT_PROBE2(vfs, namecache, lookup, miss, dvp, ".."); mtx_unlock(dvlp); return (0); } if ((ncp->nc_flag & NCF_ISDOTDOT) != 0) { if (ncp->nc_flag & NCF_NEGATIVE) *vpp = NULL; else *vpp = ncp->nc_vp; } else *vpp = ncp->nc_dvp; if (*vpp == NULL) goto negative_success; SDT_PROBE3(vfs, namecache, lookup, hit, dvp, "..", *vpp); cache_out_ts(ncp, tsp, ticksp); if ((ncp->nc_flag & (NCF_ISDOTDOT | NCF_DTS)) == NCF_DTS && tsp != NULL) { ncp_ts = __containerof(ncp, struct namecache_ts, nc_nc); *tsp = ncp_ts->nc_dotdottime; } MPASS(dvp != *vpp); ltype = VOP_ISLOCKED(dvp); VOP_UNLOCK(dvp); vs = vget_prep(*vpp); mtx_unlock(dvlp); error = vget_finish(*vpp, cnp->cn_lkflags, vs); vn_lock(dvp, ltype | LK_RETRY); if (VN_IS_DOOMED(dvp)) { if (error == 0) vput(*vpp); *vpp = NULL; return (ENOENT); } if (error) { *vpp = NULL; goto retry; } return (-1); negative_success: if (__predict_false(cnp->cn_nameiop == CREATE)) { if (cnp->cn_flags & ISLASTCN) { counter_u64_add(numnegzaps, 1); cache_zap_negative_locked_vnode_kl(ncp, dvp); mtx_unlock(dvlp); cache_free(ncp); return (0); } } whiteout = (ncp->nc_flag & NCF_WHITE); cache_out_ts(ncp, tsp, ticksp); if (cache_neg_hit_prep(ncp)) cache_neg_promote(ncp); else cache_neg_hit_finish(ncp); mtx_unlock(dvlp); if (whiteout) cnp->cn_flags |= ISWHITEOUT; return (ENOENT); } /** * Lookup a name in the name cache * * # Arguments * * - dvp: Parent directory in which to search. * - vpp: Return argument. Will contain desired vnode on cache hit. * - cnp: Parameters of the name search. The most interesting bits of * the cn_flags field have the following meanings: * - MAKEENTRY: If clear, free an entry from the cache rather than look * it up. * - ISDOTDOT: Must be set if and only if cn_nameptr == ".." * - tsp: Return storage for cache timestamp. On a successful (positive * or negative) lookup, tsp will be filled with any timespec that * was stored when this cache entry was created. However, it will * be clear for "." entries. * - ticks: Return storage for alternate cache timestamp. On a successful * (positive or negative) lookup, it will contain the ticks value * that was current when the cache entry was created, unless cnp * was ".". * * Either both tsp and ticks have to be provided or neither of them. * * # Returns * * - -1: A positive cache hit. vpp will contain the desired vnode. * - ENOENT: A negative cache hit, or dvp was recycled out from under us due * to a forced unmount. vpp will not be modified. If the entry * is a whiteout, then the ISWHITEOUT flag will be set in * cnp->cn_flags. * - 0: A cache miss. vpp will not be modified. * * # Locking * * On a cache hit, vpp will be returned locked and ref'd. If we're looking up * .., dvp is unlocked. If we're looking up . an extra ref is taken, but the * lock is not recursively acquired. */ static int __noinline cache_lookup_fallback(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct timespec *tsp, int *ticksp) { struct namecache *ncp; struct mtx *blp; uint32_t hash; enum vgetstate vs; int error; bool whiteout; MPASS((cnp->cn_flags & ISDOTDOT) == 0); MPASS((cnp->cn_flags & (MAKEENTRY | NC_KEEPPOSENTRY)) != 0); retry: hash = cache_get_hash(cnp->cn_nameptr, cnp->cn_namelen, dvp); blp = HASH2BUCKETLOCK(hash); mtx_lock(blp); CK_SLIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) { if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen && !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen)) break; } if (__predict_false(ncp == NULL)) { mtx_unlock(blp); SDT_PROBE2(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr); counter_u64_add(nummiss, 1); return (0); } if (ncp->nc_flag & NCF_NEGATIVE) goto negative_success; counter_u64_add(numposhits, 1); *vpp = ncp->nc_vp; SDT_PROBE3(vfs, namecache, lookup, hit, dvp, ncp->nc_name, *vpp); cache_out_ts(ncp, tsp, ticksp); MPASS(dvp != *vpp); vs = vget_prep(*vpp); mtx_unlock(blp); error = vget_finish(*vpp, cnp->cn_lkflags, vs); if (error) { *vpp = NULL; goto retry; } return (-1); negative_success: /* * We don't get here with regular lookup apart from corner cases. */ if (__predict_true(cnp->cn_nameiop == CREATE)) { if (cnp->cn_flags & ISLASTCN) { counter_u64_add(numnegzaps, 1); error = cache_zap_locked_bucket(ncp, cnp, hash, blp); if (__predict_false(error != 0)) { zap_bucket_fail2++; goto retry; } cache_free(ncp); return (0); } } whiteout = (ncp->nc_flag & NCF_WHITE); cache_out_ts(ncp, tsp, ticksp); if (cache_neg_hit_prep(ncp)) cache_neg_promote(ncp); else cache_neg_hit_finish(ncp); mtx_unlock(blp); if (whiteout) cnp->cn_flags |= ISWHITEOUT; return (ENOENT); } int cache_lookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct timespec *tsp, int *ticksp) { struct namecache *ncp; uint32_t hash; enum vgetstate vs; int error; bool whiteout, neg_promote; u_short nc_flag; MPASS((tsp == NULL && ticksp == NULL) || (tsp != NULL && ticksp != NULL)); #ifdef DEBUG_CACHE if (__predict_false(!doingcache)) { cnp->cn_flags &= ~MAKEENTRY; return (0); } #endif if (__predict_false(cnp->cn_nameptr[0] == '.')) { if (cnp->cn_namelen == 1) return (cache_lookup_dot(dvp, vpp, cnp, tsp, ticksp)); if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') return (cache_lookup_dotdot(dvp, vpp, cnp, tsp, ticksp)); } MPASS((cnp->cn_flags & ISDOTDOT) == 0); if ((cnp->cn_flags & (MAKEENTRY | NC_KEEPPOSENTRY)) == 0) { cache_remove_cnp(dvp, cnp); return (0); } hash = cache_get_hash(cnp->cn_nameptr, cnp->cn_namelen, dvp); vfs_smr_enter(); CK_SLIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) { if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen && !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen)) break; } if (__predict_false(ncp == NULL)) { vfs_smr_exit(); SDT_PROBE2(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr); counter_u64_add(nummiss, 1); return (0); } nc_flag = atomic_load_char(&ncp->nc_flag); if (nc_flag & NCF_NEGATIVE) goto negative_success; counter_u64_add(numposhits, 1); *vpp = ncp->nc_vp; SDT_PROBE3(vfs, namecache, lookup, hit, dvp, ncp->nc_name, *vpp); cache_out_ts(ncp, tsp, ticksp); MPASS(dvp != *vpp); if (!cache_ncp_canuse(ncp)) { vfs_smr_exit(); *vpp = NULL; goto out_fallback; } vs = vget_prep_smr(*vpp); vfs_smr_exit(); if (__predict_false(vs == VGET_NONE)) { *vpp = NULL; goto out_fallback; } error = vget_finish(*vpp, cnp->cn_lkflags, vs); if (error) { *vpp = NULL; goto out_fallback; } return (-1); negative_success: if (cnp->cn_nameiop == CREATE) { if (cnp->cn_flags & ISLASTCN) { vfs_smr_exit(); goto out_fallback; } } cache_out_ts(ncp, tsp, ticksp); whiteout = (atomic_load_char(&ncp->nc_flag) & NCF_WHITE); neg_promote = cache_neg_hit_prep(ncp); if (!cache_ncp_canuse(ncp)) { cache_neg_hit_abort(ncp); vfs_smr_exit(); goto out_fallback; } if (neg_promote) { vfs_smr_exit(); if (!cache_neg_promote_cond(dvp, cnp, ncp, hash)) goto out_fallback; } else { cache_neg_hit_finish(ncp); vfs_smr_exit(); } if (whiteout) cnp->cn_flags |= ISWHITEOUT; return (ENOENT); out_fallback: return (cache_lookup_fallback(dvp, vpp, cnp, tsp, ticksp)); } struct celockstate { struct mtx *vlp[3]; struct mtx *blp[2]; }; CTASSERT((nitems(((struct celockstate *)0)->vlp) == 3)); CTASSERT((nitems(((struct celockstate *)0)->blp) == 2)); static inline void cache_celockstate_init(struct celockstate *cel) { bzero(cel, sizeof(*cel)); } static void cache_lock_vnodes_cel(struct celockstate *cel, struct vnode *vp, struct vnode *dvp) { struct mtx *vlp1, *vlp2; MPASS(cel->vlp[0] == NULL); MPASS(cel->vlp[1] == NULL); MPASS(cel->vlp[2] == NULL); MPASS(vp != NULL || dvp != NULL); vlp1 = VP2VNODELOCK(vp); vlp2 = VP2VNODELOCK(dvp); cache_sort_vnodes(&vlp1, &vlp2); if (vlp1 != NULL) { mtx_lock(vlp1); cel->vlp[0] = vlp1; } mtx_lock(vlp2); cel->vlp[1] = vlp2; } static void cache_unlock_vnodes_cel(struct celockstate *cel) { MPASS(cel->vlp[0] != NULL || cel->vlp[1] != NULL); if (cel->vlp[0] != NULL) mtx_unlock(cel->vlp[0]); if (cel->vlp[1] != NULL) mtx_unlock(cel->vlp[1]); if (cel->vlp[2] != NULL) mtx_unlock(cel->vlp[2]); } static bool cache_lock_vnodes_cel_3(struct celockstate *cel, struct vnode *vp) { struct mtx *vlp; bool ret; cache_assert_vlp_locked(cel->vlp[0]); cache_assert_vlp_locked(cel->vlp[1]); MPASS(cel->vlp[2] == NULL); MPASS(vp != NULL); vlp = VP2VNODELOCK(vp); ret = true; if (vlp >= cel->vlp[1]) { mtx_lock(vlp); } else { if (mtx_trylock(vlp)) goto out; cache_lock_vnodes_cel_3_failures++; cache_unlock_vnodes_cel(cel); if (vlp < cel->vlp[0]) { mtx_lock(vlp); mtx_lock(cel->vlp[0]); mtx_lock(cel->vlp[1]); } else { if (cel->vlp[0] != NULL) mtx_lock(cel->vlp[0]); mtx_lock(vlp); mtx_lock(cel->vlp[1]); } ret = false; } out: cel->vlp[2] = vlp; return (ret); } static void cache_lock_buckets_cel(struct celockstate *cel, struct mtx *blp1, struct mtx *blp2) { MPASS(cel->blp[0] == NULL); MPASS(cel->blp[1] == NULL); cache_sort_vnodes(&blp1, &blp2); if (blp1 != NULL) { mtx_lock(blp1); cel->blp[0] = blp1; } mtx_lock(blp2); cel->blp[1] = blp2; } static void cache_unlock_buckets_cel(struct celockstate *cel) { if (cel->blp[0] != NULL) mtx_unlock(cel->blp[0]); mtx_unlock(cel->blp[1]); } /* * Lock part of the cache affected by the insertion. * * This means vnodelocks for dvp, vp and the relevant bucketlock. * However, insertion can result in removal of an old entry. In this * case we have an additional vnode and bucketlock pair to lock. * * That is, in the worst case we have to lock 3 vnodes and 2 bucketlocks, while * preserving the locking order (smaller address first). */ static void cache_enter_lock(struct celockstate *cel, struct vnode *dvp, struct vnode *vp, uint32_t hash) { struct namecache *ncp; struct mtx *blps[2]; u_char nc_flag; blps[0] = HASH2BUCKETLOCK(hash); for (;;) { blps[1] = NULL; cache_lock_vnodes_cel(cel, dvp, vp); if (vp == NULL || vp->v_type != VDIR) break; ncp = atomic_load_consume_ptr(&vp->v_cache_dd); if (ncp == NULL) break; nc_flag = atomic_load_char(&ncp->nc_flag); if ((nc_flag & NCF_ISDOTDOT) == 0) break; MPASS(ncp->nc_dvp == vp); blps[1] = NCP2BUCKETLOCK(ncp); if ((nc_flag & NCF_NEGATIVE) != 0) break; if (cache_lock_vnodes_cel_3(cel, ncp->nc_vp)) break; /* * All vnodes got re-locked. Re-validate the state and if * nothing changed we are done. Otherwise restart. */ if (ncp == vp->v_cache_dd && (ncp->nc_flag & NCF_ISDOTDOT) != 0 && blps[1] == NCP2BUCKETLOCK(ncp) && VP2VNODELOCK(ncp->nc_vp) == cel->vlp[2]) break; cache_unlock_vnodes_cel(cel); cel->vlp[0] = NULL; cel->vlp[1] = NULL; cel->vlp[2] = NULL; } cache_lock_buckets_cel(cel, blps[0], blps[1]); } static void cache_enter_lock_dd(struct celockstate *cel, struct vnode *dvp, struct vnode *vp, uint32_t hash) { struct namecache *ncp; struct mtx *blps[2]; u_char nc_flag; blps[0] = HASH2BUCKETLOCK(hash); for (;;) { blps[1] = NULL; cache_lock_vnodes_cel(cel, dvp, vp); ncp = atomic_load_consume_ptr(&dvp->v_cache_dd); if (ncp == NULL) break; nc_flag = atomic_load_char(&ncp->nc_flag); if ((nc_flag & NCF_ISDOTDOT) == 0) break; MPASS(ncp->nc_dvp == dvp); blps[1] = NCP2BUCKETLOCK(ncp); if ((nc_flag & NCF_NEGATIVE) != 0) break; if (cache_lock_vnodes_cel_3(cel, ncp->nc_vp)) break; if (ncp == dvp->v_cache_dd && (ncp->nc_flag & NCF_ISDOTDOT) != 0 && blps[1] == NCP2BUCKETLOCK(ncp) && VP2VNODELOCK(ncp->nc_vp) == cel->vlp[2]) break; cache_unlock_vnodes_cel(cel); cel->vlp[0] = NULL; cel->vlp[1] = NULL; cel->vlp[2] = NULL; } cache_lock_buckets_cel(cel, blps[0], blps[1]); } static void cache_enter_unlock(struct celockstate *cel) { cache_unlock_buckets_cel(cel); cache_unlock_vnodes_cel(cel); } static void __noinline cache_enter_dotdot_prep(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) { struct celockstate cel; struct namecache *ncp; uint32_t hash; int len; if (atomic_load_ptr(&dvp->v_cache_dd) == NULL) return; len = cnp->cn_namelen; cache_celockstate_init(&cel); hash = cache_get_hash(cnp->cn_nameptr, len, dvp); cache_enter_lock_dd(&cel, dvp, vp, hash); ncp = dvp->v_cache_dd; if (ncp != NULL && (ncp->nc_flag & NCF_ISDOTDOT)) { KASSERT(ncp->nc_dvp == dvp, ("wrong isdotdot parent")); cache_zap_locked(ncp); } else { ncp = NULL; } atomic_store_ptr(&dvp->v_cache_dd, NULL); cache_enter_unlock(&cel); if (ncp != NULL) cache_free(ncp); } /* * Add an entry to the cache. */ void cache_enter_time(struct vnode *dvp, struct vnode *vp, struct componentname *cnp, struct timespec *tsp, struct timespec *dtsp) { struct celockstate cel; struct namecache *ncp, *n2, *ndd; struct namecache_ts *ncp_ts; struct nchashhead *ncpp; uint32_t hash; int flag; int len; KASSERT(cnp->cn_namelen <= NAME_MAX, ("%s: passed len %ld exceeds NAME_MAX (%d)", __func__, cnp->cn_namelen, NAME_MAX)); #ifdef notyet /* * Not everything doing this is weeded out yet. */ VNPASS(dvp != vp, dvp); #endif VNPASS(!VN_IS_DOOMED(dvp), dvp); VNPASS(dvp->v_type != VNON, dvp); if (vp != NULL) { VNPASS(!VN_IS_DOOMED(vp), vp); VNPASS(vp->v_type != VNON, vp); } #ifdef DEBUG_CACHE if (__predict_false(!doingcache)) return; #endif flag = 0; if (__predict_false(cnp->cn_nameptr[0] == '.')) { if (cnp->cn_namelen == 1) return; if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') { cache_enter_dotdot_prep(dvp, vp, cnp); flag = NCF_ISDOTDOT; } } ncp = cache_alloc(cnp->cn_namelen, tsp != NULL); if (ncp == NULL) return; cache_celockstate_init(&cel); ndd = NULL; ncp_ts = NULL; /* * Calculate the hash key and setup as much of the new * namecache entry as possible before acquiring the lock. */ ncp->nc_flag = flag | NCF_WIP; ncp->nc_vp = vp; if (vp == NULL) cache_neg_init(ncp); ncp->nc_dvp = dvp; if (tsp != NULL) { ncp_ts = __containerof(ncp, struct namecache_ts, nc_nc); ncp_ts->nc_time = *tsp; ncp_ts->nc_ticks = ticks; ncp_ts->nc_nc.nc_flag |= NCF_TS; if (dtsp != NULL) { ncp_ts->nc_dotdottime = *dtsp; ncp_ts->nc_nc.nc_flag |= NCF_DTS; } } len = ncp->nc_nlen = cnp->cn_namelen; hash = cache_get_hash(cnp->cn_nameptr, len, dvp); memcpy(ncp->nc_name, cnp->cn_nameptr, len); ncp->nc_name[len] = '\0'; cache_enter_lock(&cel, dvp, vp, hash); /* * See if this vnode or negative entry is already in the cache * with this name. This can happen with concurrent lookups of * the same path name. */ ncpp = NCHHASH(hash); CK_SLIST_FOREACH(n2, ncpp, nc_hash) { if (n2->nc_dvp == dvp && n2->nc_nlen == cnp->cn_namelen && !bcmp(n2->nc_name, cnp->cn_nameptr, n2->nc_nlen)) { MPASS(cache_ncp_canuse(n2)); if ((n2->nc_flag & NCF_NEGATIVE) != 0) KASSERT(vp == NULL, ("%s: found entry pointing to a different vnode (%p != %p) ; name [%s]", __func__, NULL, vp, cnp->cn_nameptr)); else KASSERT(n2->nc_vp == vp, ("%s: found entry pointing to a different vnode (%p != %p) ; name [%s]", __func__, n2->nc_vp, vp, cnp->cn_nameptr)); /* * Entries are supposed to be immutable unless in the * process of getting destroyed. Accommodating for * changing timestamps is possible but not worth it. * This should be harmless in terms of correctness, in * the worst case resulting in an earlier expiration. * Alternatively, the found entry can be replaced * altogether. */ MPASS((n2->nc_flag & (NCF_TS | NCF_DTS)) == (ncp->nc_flag & (NCF_TS | NCF_DTS))); #if 0 if (tsp != NULL) { KASSERT((n2->nc_flag & NCF_TS) != 0, ("no NCF_TS")); n2_ts = __containerof(n2, struct namecache_ts, nc_nc); n2_ts->nc_time = ncp_ts->nc_time; n2_ts->nc_ticks = ncp_ts->nc_ticks; if (dtsp != NULL) { n2_ts->nc_dotdottime = ncp_ts->nc_dotdottime; n2_ts->nc_nc.nc_flag |= NCF_DTS; } } #endif SDT_PROBE3(vfs, namecache, enter, duplicate, dvp, ncp->nc_name, vp); goto out_unlock_free; } } if (flag == NCF_ISDOTDOT) { /* * See if we are trying to add .. entry, but some other lookup * has populated v_cache_dd pointer already. */ if (dvp->v_cache_dd != NULL) goto out_unlock_free; KASSERT(vp == NULL || vp->v_type == VDIR, ("wrong vnode type %p", vp)); atomic_thread_fence_rel(); atomic_store_ptr(&dvp->v_cache_dd, ncp); } if (vp != NULL) { if (flag != NCF_ISDOTDOT) { /* * For this case, the cache entry maps both the * directory name in it and the name ".." for the * directory's parent. */ if ((ndd = vp->v_cache_dd) != NULL) { if ((ndd->nc_flag & NCF_ISDOTDOT) != 0) cache_zap_locked(ndd); else ndd = NULL; } atomic_thread_fence_rel(); atomic_store_ptr(&vp->v_cache_dd, ncp); } else if (vp->v_type != VDIR) { if (vp->v_cache_dd != NULL) { atomic_store_ptr(&vp->v_cache_dd, NULL); } } } if (flag != NCF_ISDOTDOT) { if (LIST_EMPTY(&dvp->v_cache_src)) { cache_hold_vnode(dvp); } LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src); } /* * If the entry is "negative", we place it into the * "negative" cache queue, otherwise, we place it into the * destination vnode's cache entries queue. */ if (vp != NULL) { TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst); SDT_PROBE3(vfs, namecache, enter, done, dvp, ncp->nc_name, vp); } else { if (cnp->cn_flags & ISWHITEOUT) atomic_store_char(&ncp->nc_flag, ncp->nc_flag | NCF_WHITE); cache_neg_insert(ncp); SDT_PROBE2(vfs, namecache, enter_negative, done, dvp, ncp->nc_name); } /* * Insert the new namecache entry into the appropriate chain * within the cache entries table. */ CK_SLIST_INSERT_HEAD(ncpp, ncp, nc_hash); atomic_thread_fence_rel(); /* * Mark the entry as fully constructed. * It is immutable past this point until its removal. */ atomic_store_char(&ncp->nc_flag, ncp->nc_flag & ~NCF_WIP); cache_enter_unlock(&cel); if (ndd != NULL) cache_free(ndd); return; out_unlock_free: cache_enter_unlock(&cel); cache_free(ncp); return; } /* * A variant of the above accepting flags. * * - VFS_CACHE_DROPOLD -- if a conflicting entry is found, drop it. * * TODO: this routine is a hack. It blindly removes the old entry, even if it * happens to match and it is doing it in an inefficient manner. It was added * to accomodate NFS which runs into a case where the target for a given name * may change from under it. Note this does nothing to solve the following * race: 2 callers of cache_enter_time_flags pass a different target vnode for * the same [dvp, cnp]. It may be argued that code doing this is broken. */ void cache_enter_time_flags(struct vnode *dvp, struct vnode *vp, struct componentname *cnp, struct timespec *tsp, struct timespec *dtsp, int flags) { MPASS((flags & ~(VFS_CACHE_DROPOLD)) == 0); if (flags & VFS_CACHE_DROPOLD) cache_remove_cnp(dvp, cnp); cache_enter_time(dvp, vp, cnp, tsp, dtsp); } static u_int cache_roundup_2(u_int val) { u_int res; for (res = 1; res <= val; res <<= 1) continue; return (res); } static struct nchashhead * nchinittbl(u_long elements, u_long *hashmask) { struct nchashhead *hashtbl; u_long hashsize, i; hashsize = cache_roundup_2(elements) / 2; hashtbl = malloc((u_long)hashsize * sizeof(*hashtbl), M_VFSCACHE, M_WAITOK); for (i = 0; i < hashsize; i++) CK_SLIST_INIT(&hashtbl[i]); *hashmask = hashsize - 1; return (hashtbl); } static void ncfreetbl(struct nchashhead *hashtbl) { free(hashtbl, M_VFSCACHE); } /* * Name cache initialization, from vfs_init() when we are booting */ static void nchinit(void *dummy __unused) { u_int i; cache_zone_small = uma_zcreate("S VFS Cache", CACHE_ZONE_SMALL_SIZE, NULL, NULL, NULL, NULL, CACHE_ZONE_ALIGNMENT, UMA_ZONE_ZINIT); cache_zone_small_ts = uma_zcreate("STS VFS Cache", CACHE_ZONE_SMALL_TS_SIZE, NULL, NULL, NULL, NULL, CACHE_ZONE_ALIGNMENT, UMA_ZONE_ZINIT); cache_zone_large = uma_zcreate("L VFS Cache", CACHE_ZONE_LARGE_SIZE, NULL, NULL, NULL, NULL, CACHE_ZONE_ALIGNMENT, UMA_ZONE_ZINIT); cache_zone_large_ts = uma_zcreate("LTS VFS Cache", CACHE_ZONE_LARGE_TS_SIZE, NULL, NULL, NULL, NULL, CACHE_ZONE_ALIGNMENT, UMA_ZONE_ZINIT); VFS_SMR_ZONE_SET(cache_zone_small); VFS_SMR_ZONE_SET(cache_zone_small_ts); VFS_SMR_ZONE_SET(cache_zone_large); VFS_SMR_ZONE_SET(cache_zone_large_ts); ncsize = desiredvnodes * ncsizefactor; cache_recalc_neg_min(ncnegminpct); nchashtbl = nchinittbl(desiredvnodes * 2, &nchash); ncbuckethash = cache_roundup_2(mp_ncpus * mp_ncpus) - 1; if (ncbuckethash < 7) /* arbitrarily chosen to avoid having one lock */ ncbuckethash = 7; if (ncbuckethash > nchash) ncbuckethash = nchash; bucketlocks = malloc(sizeof(*bucketlocks) * numbucketlocks, M_VFSCACHE, M_WAITOK | M_ZERO); for (i = 0; i < numbucketlocks; i++) mtx_init(&bucketlocks[i], "ncbuc", NULL, MTX_DUPOK | MTX_RECURSE); ncvnodehash = ncbuckethash; vnodelocks = malloc(sizeof(*vnodelocks) * numvnodelocks, M_VFSCACHE, M_WAITOK | M_ZERO); for (i = 0; i < numvnodelocks; i++) mtx_init(&vnodelocks[i], "ncvn", NULL, MTX_DUPOK | MTX_RECURSE); for (i = 0; i < numneglists; i++) { mtx_init(&neglists[i].nl_evict_lock, "ncnege", NULL, MTX_DEF); mtx_init(&neglists[i].nl_lock, "ncnegl", NULL, MTX_DEF); TAILQ_INIT(&neglists[i].nl_list); TAILQ_INIT(&neglists[i].nl_hotlist); } } SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL); void cache_vnode_init(struct vnode *vp) { LIST_INIT(&vp->v_cache_src); TAILQ_INIT(&vp->v_cache_dst); vp->v_cache_dd = NULL; cache_prehash(vp); } /* * Induce transient cache misses for lockless operation in cache_lookup() by * using a temporary hash table. * * This will force a fs lookup. * * Synchronisation is done in 2 steps, calling vfs_smr_synchronize each time * to observe all CPUs not performing the lookup. */ static void cache_changesize_set_temp(struct nchashhead *temptbl, u_long temphash) { MPASS(temphash < nchash); /* * Change the size. The new size is smaller and can safely be used * against the existing table. All lookups which now hash wrong will * result in a cache miss, which all callers are supposed to know how * to handle. */ atomic_store_long(&nchash, temphash); atomic_thread_fence_rel(); vfs_smr_synchronize(); /* * At this point everyone sees the updated hash value, but they still * see the old table. */ atomic_store_ptr(&nchashtbl, temptbl); atomic_thread_fence_rel(); vfs_smr_synchronize(); /* * At this point everyone sees the updated table pointer and size pair. */ } /* * Set the new hash table. * * Similarly to cache_changesize_set_temp(), this has to synchronize against * lockless operation in cache_lookup(). */ static void cache_changesize_set_new(struct nchashhead *new_tbl, u_long new_hash) { MPASS(nchash < new_hash); /* * Change the pointer first. This wont result in out of bounds access * since the temporary table is guaranteed to be smaller. */ atomic_store_ptr(&nchashtbl, new_tbl); atomic_thread_fence_rel(); vfs_smr_synchronize(); /* * At this point everyone sees the updated pointer value, but they * still see the old size. */ atomic_store_long(&nchash, new_hash); atomic_thread_fence_rel(); vfs_smr_synchronize(); /* * At this point everyone sees the updated table pointer and size pair. */ } void cache_changesize(u_long newmaxvnodes) { struct nchashhead *new_nchashtbl, *old_nchashtbl, *temptbl; u_long new_nchash, old_nchash, temphash; struct namecache *ncp; uint32_t hash; u_long newncsize; int i; newncsize = newmaxvnodes * ncsizefactor; newmaxvnodes = cache_roundup_2(newmaxvnodes * 2); if (newmaxvnodes < numbucketlocks) newmaxvnodes = numbucketlocks; new_nchashtbl = nchinittbl(newmaxvnodes, &new_nchash); /* If same hash table size, nothing to do */ if (nchash == new_nchash) { ncfreetbl(new_nchashtbl); return; } temptbl = nchinittbl(1, &temphash); /* * Move everything from the old hash table to the new table. * None of the namecache entries in the table can be removed * because to do so, they have to be removed from the hash table. */ cache_lock_all_vnodes(); cache_lock_all_buckets(); old_nchashtbl = nchashtbl; old_nchash = nchash; cache_changesize_set_temp(temptbl, temphash); for (i = 0; i <= old_nchash; i++) { while ((ncp = CK_SLIST_FIRST(&old_nchashtbl[i])) != NULL) { hash = cache_get_hash(ncp->nc_name, ncp->nc_nlen, ncp->nc_dvp); CK_SLIST_REMOVE(&old_nchashtbl[i], ncp, namecache, nc_hash); CK_SLIST_INSERT_HEAD(&new_nchashtbl[hash & new_nchash], ncp, nc_hash); } } ncsize = newncsize; cache_recalc_neg_min(ncnegminpct); cache_changesize_set_new(new_nchashtbl, new_nchash); cache_unlock_all_buckets(); cache_unlock_all_vnodes(); ncfreetbl(old_nchashtbl); ncfreetbl(temptbl); } /* * Remove all entries from and to a particular vnode. */ static void cache_purge_impl(struct vnode *vp) { struct cache_freebatch batch; struct namecache *ncp; struct mtx *vlp, *vlp2; TAILQ_INIT(&batch); vlp = VP2VNODELOCK(vp); vlp2 = NULL; mtx_lock(vlp); retry: while (!LIST_EMPTY(&vp->v_cache_src)) { ncp = LIST_FIRST(&vp->v_cache_src); if (!cache_zap_locked_vnode_kl2(ncp, vp, &vlp2)) goto retry; TAILQ_INSERT_TAIL(&batch, ncp, nc_dst); } while (!TAILQ_EMPTY(&vp->v_cache_dst)) { ncp = TAILQ_FIRST(&vp->v_cache_dst); if (!cache_zap_locked_vnode_kl2(ncp, vp, &vlp2)) goto retry; TAILQ_INSERT_TAIL(&batch, ncp, nc_dst); } ncp = vp->v_cache_dd; if (ncp != NULL) { KASSERT(ncp->nc_flag & NCF_ISDOTDOT, ("lost dotdot link")); if (!cache_zap_locked_vnode_kl2(ncp, vp, &vlp2)) goto retry; TAILQ_INSERT_TAIL(&batch, ncp, nc_dst); } KASSERT(vp->v_cache_dd == NULL, ("incomplete purge")); mtx_unlock(vlp); if (vlp2 != NULL) mtx_unlock(vlp2); cache_free_batch(&batch); } /* * Opportunistic check to see if there is anything to do. */ static bool cache_has_entries(struct vnode *vp) { if (LIST_EMPTY(&vp->v_cache_src) && TAILQ_EMPTY(&vp->v_cache_dst) && atomic_load_ptr(&vp->v_cache_dd) == NULL) return (false); return (true); } void cache_purge(struct vnode *vp) { SDT_PROBE1(vfs, namecache, purge, done, vp); if (!cache_has_entries(vp)) return; cache_purge_impl(vp); } /* * Only to be used by vgone. */ void cache_purge_vgone(struct vnode *vp) { struct mtx *vlp; VNPASS(VN_IS_DOOMED(vp), vp); if (cache_has_entries(vp)) { cache_purge_impl(vp); return; } /* * Serialize against a potential thread doing cache_purge. */ vlp = VP2VNODELOCK(vp); mtx_wait_unlocked(vlp); if (cache_has_entries(vp)) { cache_purge_impl(vp); return; } return; } /* * Remove all negative entries for a particular directory vnode. */ void cache_purge_negative(struct vnode *vp) { struct cache_freebatch batch; struct namecache *ncp, *nnp; struct mtx *vlp; SDT_PROBE1(vfs, namecache, purge_negative, done, vp); if (LIST_EMPTY(&vp->v_cache_src)) return; TAILQ_INIT(&batch); vlp = VP2VNODELOCK(vp); mtx_lock(vlp); LIST_FOREACH_SAFE(ncp, &vp->v_cache_src, nc_src, nnp) { if (!(ncp->nc_flag & NCF_NEGATIVE)) continue; cache_zap_negative_locked_vnode_kl(ncp, vp); TAILQ_INSERT_TAIL(&batch, ncp, nc_dst); } mtx_unlock(vlp); cache_free_batch(&batch); } /* * Entry points for modifying VOP operations. */ void cache_vop_rename(struct vnode *fdvp, struct vnode *fvp, struct vnode *tdvp, struct vnode *tvp, struct componentname *fcnp, struct componentname *tcnp) { ASSERT_VOP_IN_SEQC(fdvp); ASSERT_VOP_IN_SEQC(fvp); ASSERT_VOP_IN_SEQC(tdvp); if (tvp != NULL) ASSERT_VOP_IN_SEQC(tvp); cache_purge(fvp); if (tvp != NULL) { cache_purge(tvp); KASSERT(!cache_remove_cnp(tdvp, tcnp), ("%s: lingering negative entry", __func__)); } else { cache_remove_cnp(tdvp, tcnp); } /* * TODO * * Historically renaming was always purging all revelang entries, * but that's quite wasteful. In particular turns out that in many cases * the target file is immediately accessed after rename, inducing a cache * miss. * * Recode this to reduce relocking and reuse the existing entry (if any) * instead of just removing it above and allocating a new one here. */ if (cache_rename_add) { cache_enter(tdvp, fvp, tcnp); } } void cache_vop_rmdir(struct vnode *dvp, struct vnode *vp) { ASSERT_VOP_IN_SEQC(dvp); ASSERT_VOP_IN_SEQC(vp); cache_purge(vp); } #ifdef INVARIANTS /* * Validate that if an entry exists it matches. */ void cache_validate(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) { struct namecache *ncp; struct mtx *blp; uint32_t hash; hash = cache_get_hash(cnp->cn_nameptr, cnp->cn_namelen, dvp); if (CK_SLIST_EMPTY(NCHHASH(hash))) return; blp = HASH2BUCKETLOCK(hash); mtx_lock(blp); CK_SLIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) { if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen && !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen)) { if (ncp->nc_vp != vp) panic("%s: mismatch (%p != %p); ncp %p [%s] dvp %p\n", __func__, vp, ncp->nc_vp, ncp, ncp->nc_name, ncp->nc_dvp); } } mtx_unlock(blp); } #endif /* * Flush all entries referencing a particular filesystem. */ void cache_purgevfs(struct mount *mp) { struct vnode *vp, *mvp; size_t visited, purged; visited = purged = 0; /* * Somewhat wasteful iteration over all vnodes. Would be better to * support filtering and avoid the interlock to begin with. */ MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { visited++; if (!cache_has_entries(vp)) { VI_UNLOCK(vp); continue; } vholdl(vp); VI_UNLOCK(vp); cache_purge(vp); purged++; vdrop(vp); } SDT_PROBE3(vfs, namecache, purgevfs, done, mp, visited, purged); } /* * Perform canonical checks and cache lookup and pass on to filesystem * through the vop_cachedlookup only if needed. */ int vfs_cache_lookup(struct vop_lookup_args *ap) { struct vnode *dvp; int error; struct vnode **vpp = ap->a_vpp; struct componentname *cnp = ap->a_cnp; int flags = cnp->cn_flags; *vpp = NULL; dvp = ap->a_dvp; if (dvp->v_type != VDIR) return (ENOTDIR); if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) return (EROFS); error = vn_dir_check_exec(dvp, cnp); if (error != 0) return (error); error = cache_lookup(dvp, vpp, cnp, NULL, NULL); if (error == 0) return (VOP_CACHEDLOOKUP(dvp, vpp, cnp)); if (error == -1) return (0); return (error); } /* Implementation of the getcwd syscall. */ int sys___getcwd(struct thread *td, struct __getcwd_args *uap) { char *buf, *retbuf; size_t buflen; int error; buflen = uap->buflen; if (__predict_false(buflen < 2)) return (EINVAL); if (buflen > MAXPATHLEN) buflen = MAXPATHLEN; buf = uma_zalloc(namei_zone, M_WAITOK); error = vn_getcwd(buf, &retbuf, &buflen); if (error == 0) error = copyout(retbuf, uap->buf, buflen); uma_zfree(namei_zone, buf); return (error); } int vn_getcwd(char *buf, char **retbuf, size_t *buflen) { struct pwd *pwd; int error; vfs_smr_enter(); pwd = pwd_get_smr(); error = vn_fullpath_any_smr(pwd->pwd_cdir, pwd->pwd_rdir, buf, retbuf, buflen, 0); VFS_SMR_ASSERT_NOT_ENTERED(); if (error < 0) { pwd = pwd_hold(curthread); error = vn_fullpath_any(pwd->pwd_cdir, pwd->pwd_rdir, buf, retbuf, buflen); pwd_drop(pwd); } #ifdef KTRACE if (KTRPOINT(curthread, KTR_NAMEI) && error == 0) ktrnamei(*retbuf); #endif return (error); } static int kern___realpathat(struct thread *td, int fd, const char *path, char *buf, size_t size, int flags, enum uio_seg pathseg) { struct nameidata nd; char *retbuf, *freebuf; int error; if (flags != 0) return (EINVAL); NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | SAVENAME | WANTPARENT | AUDITVNODE1, pathseg, path, fd, &cap_fstat_rights, td); if ((error = namei(&nd)) != 0) return (error); error = vn_fullpath_hardlink(&nd, &retbuf, &freebuf, &size); if (error == 0) { error = copyout(retbuf, buf, size); free(freebuf, M_TEMP); } NDFREE(&nd, 0); return (error); } int sys___realpathat(struct thread *td, struct __realpathat_args *uap) { return (kern___realpathat(td, uap->fd, uap->path, uap->buf, uap->size, uap->flags, UIO_USERSPACE)); } /* * Retrieve the full filesystem path that correspond to a vnode from the name * cache (if available) */ int vn_fullpath(struct vnode *vp, char **retbuf, char **freebuf) { struct pwd *pwd; char *buf; size_t buflen; int error; if (__predict_false(vp == NULL)) return (EINVAL); buflen = MAXPATHLEN; buf = malloc(buflen, M_TEMP, M_WAITOK); vfs_smr_enter(); pwd = pwd_get_smr(); error = vn_fullpath_any_smr(vp, pwd->pwd_rdir, buf, retbuf, &buflen, 0); VFS_SMR_ASSERT_NOT_ENTERED(); if (error < 0) { pwd = pwd_hold(curthread); error = vn_fullpath_any(vp, pwd->pwd_rdir, buf, retbuf, &buflen); pwd_drop(pwd); } if (error == 0) *freebuf = buf; else free(buf, M_TEMP); return (error); } /* * This function is similar to vn_fullpath, but it attempts to lookup the * pathname relative to the global root mount point. This is required for the * auditing sub-system, as audited pathnames must be absolute, relative to the * global root mount point. */ int vn_fullpath_global(struct vnode *vp, char **retbuf, char **freebuf) { char *buf; size_t buflen; int error; if (__predict_false(vp == NULL)) return (EINVAL); buflen = MAXPATHLEN; buf = malloc(buflen, M_TEMP, M_WAITOK); vfs_smr_enter(); error = vn_fullpath_any_smr(vp, rootvnode, buf, retbuf, &buflen, 0); VFS_SMR_ASSERT_NOT_ENTERED(); if (error < 0) { error = vn_fullpath_any(vp, rootvnode, buf, retbuf, &buflen); } if (error == 0) *freebuf = buf; else free(buf, M_TEMP); return (error); } static struct namecache * vn_dd_from_dst(struct vnode *vp) { struct namecache *ncp; cache_assert_vnode_locked(vp); TAILQ_FOREACH(ncp, &vp->v_cache_dst, nc_dst) { if ((ncp->nc_flag & NCF_ISDOTDOT) == 0) return (ncp); } return (NULL); } int vn_vptocnp(struct vnode **vp, char *buf, size_t *buflen) { struct vnode *dvp; struct namecache *ncp; struct mtx *vlp; int error; vlp = VP2VNODELOCK(*vp); mtx_lock(vlp); ncp = (*vp)->v_cache_dd; if (ncp != NULL && (ncp->nc_flag & NCF_ISDOTDOT) == 0) { KASSERT(ncp == vn_dd_from_dst(*vp), ("%s: mismatch for dd entry (%p != %p)", __func__, ncp, vn_dd_from_dst(*vp))); } else { ncp = vn_dd_from_dst(*vp); } if (ncp != NULL) { if (*buflen < ncp->nc_nlen) { mtx_unlock(vlp); vrele(*vp); counter_u64_add(numfullpathfail4, 1); error = ENOMEM; SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); return (error); } *buflen -= ncp->nc_nlen; memcpy(buf + *buflen, ncp->nc_name, ncp->nc_nlen); SDT_PROBE3(vfs, namecache, fullpath, hit, ncp->nc_dvp, ncp->nc_name, vp); dvp = *vp; *vp = ncp->nc_dvp; vref(*vp); mtx_unlock(vlp); vrele(dvp); return (0); } SDT_PROBE1(vfs, namecache, fullpath, miss, vp); mtx_unlock(vlp); vn_lock(*vp, LK_SHARED | LK_RETRY); error = VOP_VPTOCNP(*vp, &dvp, buf, buflen); vput(*vp); if (error) { counter_u64_add(numfullpathfail2, 1); SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); return (error); } *vp = dvp; if (VN_IS_DOOMED(dvp)) { /* forced unmount */ vrele(dvp); error = ENOENT; SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); return (error); } /* * *vp has its use count incremented still. */ return (0); } /* * Resolve a directory to a pathname. * * The name of the directory can always be found in the namecache or fetched * from the filesystem. There is also guaranteed to be only one parent, meaning * we can just follow vnodes up until we find the root. * * The vnode must be referenced. */ static int vn_fullpath_dir(struct vnode *vp, struct vnode *rdir, char *buf, char **retbuf, size_t *len, size_t addend) { #ifdef KDTRACE_HOOKS struct vnode *startvp = vp; #endif struct vnode *vp1; size_t buflen; int error; bool slash_prefixed; VNPASS(vp->v_type == VDIR || VN_IS_DOOMED(vp), vp); VNPASS(vp->v_usecount > 0, vp); buflen = *len; slash_prefixed = true; if (addend == 0) { MPASS(*len >= 2); buflen--; buf[buflen] = '\0'; slash_prefixed = false; } error = 0; SDT_PROBE1(vfs, namecache, fullpath, entry, vp); counter_u64_add(numfullpathcalls, 1); while (vp != rdir && vp != rootvnode) { /* * The vp vnode must be already fully constructed, * since it is either found in namecache or obtained * from VOP_VPTOCNP(). We may test for VV_ROOT safely * without obtaining the vnode lock. */ if ((vp->v_vflag & VV_ROOT) != 0) { vn_lock(vp, LK_RETRY | LK_SHARED); /* * With the vnode locked, check for races with * unmount, forced or not. Note that we * already verified that vp is not equal to * the root vnode, which means that * mnt_vnodecovered can be NULL only for the * case of unmount. */ if (VN_IS_DOOMED(vp) || (vp1 = vp->v_mount->mnt_vnodecovered) == NULL || vp1->v_mountedhere != vp->v_mount) { vput(vp); error = ENOENT; SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); break; } vref(vp1); vput(vp); vp = vp1; continue; } if (vp->v_type != VDIR) { vrele(vp); counter_u64_add(numfullpathfail1, 1); error = ENOTDIR; SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); break; } error = vn_vptocnp(&vp, buf, &buflen); if (error) break; if (buflen == 0) { vrele(vp); error = ENOMEM; SDT_PROBE3(vfs, namecache, fullpath, return, error, startvp, NULL); break; } buf[--buflen] = '/'; slash_prefixed = true; } if (error) return (error); if (!slash_prefixed) { if (buflen == 0) { vrele(vp); counter_u64_add(numfullpathfail4, 1); SDT_PROBE3(vfs, namecache, fullpath, return, ENOMEM, startvp, NULL); return (ENOMEM); } buf[--buflen] = '/'; } counter_u64_add(numfullpathfound, 1); vrele(vp); *retbuf = buf + buflen; SDT_PROBE3(vfs, namecache, fullpath, return, 0, startvp, *retbuf); *len -= buflen; *len += addend; return (0); } /* * Resolve an arbitrary vnode to a pathname. * * Note 2 caveats: * - hardlinks are not tracked, thus if the vnode is not a directory this can * resolve to a different path than the one used to find it * - namecache is not mandatory, meaning names are not guaranteed to be added * (in which case resolving fails) */ static void __inline cache_rev_failed_impl(int *reason, int line) { *reason = line; } #define cache_rev_failed(var) cache_rev_failed_impl((var), __LINE__) static int vn_fullpath_any_smr(struct vnode *vp, struct vnode *rdir, char *buf, char **retbuf, size_t *buflen, size_t addend) { #ifdef KDTRACE_HOOKS struct vnode *startvp = vp; #endif struct vnode *tvp; struct mount *mp; struct namecache *ncp; size_t orig_buflen; int reason; int error; #ifdef KDTRACE_HOOKS int i; #endif seqc_t vp_seqc, tvp_seqc; u_char nc_flag; VFS_SMR_ASSERT_ENTERED(); if (!atomic_load_char(&cache_fast_lookup_enabled)) { vfs_smr_exit(); return (-1); } orig_buflen = *buflen; if (addend == 0) { MPASS(*buflen >= 2); *buflen -= 1; buf[*buflen] = '\0'; } if (vp == rdir || vp == rootvnode) { if (addend == 0) { *buflen -= 1; buf[*buflen] = '/'; } goto out_ok; } #ifdef KDTRACE_HOOKS i = 0; #endif error = -1; ncp = NULL; /* for sdt probe down below */ vp_seqc = vn_seqc_read_any(vp); if (seqc_in_modify(vp_seqc)) { cache_rev_failed(&reason); goto out_abort; } for (;;) { #ifdef KDTRACE_HOOKS i++; #endif if ((vp->v_vflag & VV_ROOT) != 0) { mp = atomic_load_ptr(&vp->v_mount); if (mp == NULL) { cache_rev_failed(&reason); goto out_abort; } tvp = atomic_load_ptr(&mp->mnt_vnodecovered); tvp_seqc = vn_seqc_read_any(tvp); if (seqc_in_modify(tvp_seqc)) { cache_rev_failed(&reason); goto out_abort; } if (!vn_seqc_consistent(vp, vp_seqc)) { cache_rev_failed(&reason); goto out_abort; } vp = tvp; vp_seqc = tvp_seqc; continue; } ncp = atomic_load_consume_ptr(&vp->v_cache_dd); if (ncp == NULL) { cache_rev_failed(&reason); goto out_abort; } nc_flag = atomic_load_char(&ncp->nc_flag); if ((nc_flag & NCF_ISDOTDOT) != 0) { cache_rev_failed(&reason); goto out_abort; } if (ncp->nc_nlen >= *buflen) { cache_rev_failed(&reason); error = ENOMEM; goto out_abort; } *buflen -= ncp->nc_nlen; memcpy(buf + *buflen, ncp->nc_name, ncp->nc_nlen); *buflen -= 1; buf[*buflen] = '/'; tvp = ncp->nc_dvp; tvp_seqc = vn_seqc_read_any(tvp); if (seqc_in_modify(tvp_seqc)) { cache_rev_failed(&reason); goto out_abort; } if (!vn_seqc_consistent(vp, vp_seqc)) { cache_rev_failed(&reason); goto out_abort; } /* * Acquire fence provided by vn_seqc_read_any above. */ if (__predict_false(atomic_load_ptr(&vp->v_cache_dd) != ncp)) { cache_rev_failed(&reason); goto out_abort; } if (!cache_ncp_canuse(ncp)) { cache_rev_failed(&reason); goto out_abort; } vp = tvp; vp_seqc = tvp_seqc; if (vp == rdir || vp == rootvnode) break; } out_ok: vfs_smr_exit(); *retbuf = buf + *buflen; *buflen = orig_buflen - *buflen + addend; SDT_PROBE2(vfs, namecache, fullpath_smr, hit, startvp, *retbuf); return (0); out_abort: *buflen = orig_buflen; SDT_PROBE4(vfs, namecache, fullpath_smr, miss, startvp, ncp, reason, i); vfs_smr_exit(); return (error); } static int vn_fullpath_any(struct vnode *vp, struct vnode *rdir, char *buf, char **retbuf, size_t *buflen) { size_t orig_buflen, addend; int error; if (*buflen < 2) return (EINVAL); orig_buflen = *buflen; vref(vp); addend = 0; if (vp->v_type != VDIR) { *buflen -= 1; buf[*buflen] = '\0'; error = vn_vptocnp(&vp, buf, buflen); if (error) return (error); if (*buflen == 0) { vrele(vp); return (ENOMEM); } *buflen -= 1; buf[*buflen] = '/'; addend = orig_buflen - *buflen; } return (vn_fullpath_dir(vp, rdir, buf, retbuf, buflen, addend)); } /* * Resolve an arbitrary vnode to a pathname (taking care of hardlinks). * * Since the namecache does not track hardlinks, the caller is expected to first * look up the target vnode with SAVENAME | WANTPARENT flags passed to namei. * * Then we have 2 cases: * - if the found vnode is a directory, the path can be constructed just by * following names up the chain * - otherwise we populate the buffer with the saved name and start resolving * from the parent */ static int vn_fullpath_hardlink(struct nameidata *ndp, char **retbuf, char **freebuf, size_t *buflen) { char *buf, *tmpbuf; struct pwd *pwd; struct componentname *cnp; struct vnode *vp; size_t addend; int error; enum vtype type; if (*buflen < 2) return (EINVAL); if (*buflen > MAXPATHLEN) *buflen = MAXPATHLEN; buf = malloc(*buflen, M_TEMP, M_WAITOK); addend = 0; vp = ndp->ni_vp; /* * Check for VBAD to work around the vp_crossmp bug in lookup(). * * For example consider tmpfs on /tmp and realpath /tmp. ni_vp will be * set to mount point's root vnode while ni_dvp will be vp_crossmp. * If the type is VDIR (like in this very case) we can skip looking * at ni_dvp in the first place. However, since vnodes get passed here * unlocked the target may transition to doomed state (type == VBAD) * before we get to evaluate the condition. If this happens, we will * populate part of the buffer and descend to vn_fullpath_dir with * vp == vp_crossmp. Prevent the problem by checking for VBAD. * * This should be atomic_load(&vp->v_type) but it is illegal to take * an address of a bit field, even if said field is sized to char. * Work around the problem by reading the value into a full-sized enum * and then re-reading it with atomic_load which will still prevent * the compiler from re-reading down the road. */ type = vp->v_type; type = atomic_load_int(&type); if (type == VBAD) { error = ENOENT; goto out_bad; } if (type != VDIR) { cnp = &ndp->ni_cnd; addend = cnp->cn_namelen + 2; if (*buflen < addend) { error = ENOMEM; goto out_bad; } *buflen -= addend; tmpbuf = buf + *buflen; tmpbuf[0] = '/'; memcpy(&tmpbuf[1], cnp->cn_nameptr, cnp->cn_namelen); tmpbuf[addend - 1] = '\0'; vp = ndp->ni_dvp; } vfs_smr_enter(); pwd = pwd_get_smr(); error = vn_fullpath_any_smr(vp, pwd->pwd_rdir, buf, retbuf, buflen, addend); VFS_SMR_ASSERT_NOT_ENTERED(); if (error < 0) { pwd = pwd_hold(curthread); vref(vp); error = vn_fullpath_dir(vp, pwd->pwd_rdir, buf, retbuf, buflen, addend); pwd_drop(pwd); } if (error != 0) goto out_bad; *freebuf = buf; return (0); out_bad: free(buf, M_TEMP); return (error); } struct vnode * vn_dir_dd_ino(struct vnode *vp) { struct namecache *ncp; struct vnode *ddvp; struct mtx *vlp; enum vgetstate vs; ASSERT_VOP_LOCKED(vp, "vn_dir_dd_ino"); vlp = VP2VNODELOCK(vp); mtx_lock(vlp); TAILQ_FOREACH(ncp, &(vp->v_cache_dst), nc_dst) { if ((ncp->nc_flag & NCF_ISDOTDOT) != 0) continue; ddvp = ncp->nc_dvp; vs = vget_prep(ddvp); mtx_unlock(vlp); if (vget_finish(ddvp, LK_SHARED | LK_NOWAIT, vs)) return (NULL); return (ddvp); } mtx_unlock(vlp); return (NULL); } int vn_commname(struct vnode *vp, char *buf, u_int buflen) { struct namecache *ncp; struct mtx *vlp; int l; vlp = VP2VNODELOCK(vp); mtx_lock(vlp); TAILQ_FOREACH(ncp, &vp->v_cache_dst, nc_dst) if ((ncp->nc_flag & NCF_ISDOTDOT) == 0) break; if (ncp == NULL) { mtx_unlock(vlp); return (ENOENT); } l = min(ncp->nc_nlen, buflen - 1); memcpy(buf, ncp->nc_name, l); mtx_unlock(vlp); buf[l] = '\0'; return (0); } /* * This function updates path string to vnode's full global path * and checks the size of the new path string against the pathlen argument. * * Requires a locked, referenced vnode. * Vnode is re-locked on success or ENODEV, otherwise unlocked. * * If vp is a directory, the call to vn_fullpath_global() always succeeds * because it falls back to the ".." lookup if the namecache lookup fails. */ int vn_path_to_global_path(struct thread *td, struct vnode *vp, char *path, u_int pathlen) { struct nameidata nd; struct vnode *vp1; char *rpath, *fbuf; int error; ASSERT_VOP_ELOCKED(vp, __func__); /* Construct global filesystem path from vp. */ VOP_UNLOCK(vp); error = vn_fullpath_global(vp, &rpath, &fbuf); if (error != 0) { vrele(vp); return (error); } if (strlen(rpath) >= pathlen) { vrele(vp); error = ENAMETOOLONG; goto out; } /* * Re-lookup the vnode by path to detect a possible rename. * As a side effect, the vnode is relocked. * If vnode was renamed, return ENOENT. */ NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE, path, td); error = namei(&nd); if (error != 0) { vrele(vp); goto out; } NDFREE(&nd, NDF_ONLY_PNBUF); vp1 = nd.ni_vp; vrele(vp); if (vp1 == vp) strcpy(path, rpath); else { vput(vp1); error = ENOENT; } out: free(fbuf, M_TEMP); return (error); } #ifdef DDB static void db_print_vpath(struct vnode *vp) { while (vp != NULL) { db_printf("%p: ", vp); if (vp == rootvnode) { db_printf("/"); vp = NULL; } else { if (vp->v_vflag & VV_ROOT) { db_printf(""); vp = vp->v_mount->mnt_vnodecovered; } else { struct namecache *ncp; char *ncn; int i; ncp = TAILQ_FIRST(&vp->v_cache_dst); if (ncp != NULL) { ncn = ncp->nc_name; for (i = 0; i < ncp->nc_nlen; i++) db_printf("%c", *ncn++); vp = ncp->nc_dvp; } else { vp = NULL; } } } db_printf("\n"); } return; } DB_SHOW_COMMAND(vpath, db_show_vpath) { struct vnode *vp; if (!have_addr) { db_printf("usage: show vpath \n"); return; } vp = (struct vnode *)addr; db_print_vpath(vp); } #endif static int cache_fast_lookup = 1; #define CACHE_FPL_FAILED -2020 void cache_fast_lookup_enabled_recalc(void) { int lookup_flag; int mac_on; #ifdef MAC mac_on = mac_vnode_check_lookup_enabled(); mac_on |= mac_vnode_check_readlink_enabled(); #else mac_on = 0; #endif lookup_flag = atomic_load_int(&cache_fast_lookup); if (lookup_flag && !mac_on) { atomic_store_char(&cache_fast_lookup_enabled, true); } else { atomic_store_char(&cache_fast_lookup_enabled, false); } } static int syscal_vfs_cache_fast_lookup(SYSCTL_HANDLER_ARGS) { int error, old; old = atomic_load_int(&cache_fast_lookup); error = sysctl_handle_int(oidp, arg1, arg2, req); if (error == 0 && req->newptr && old != atomic_load_int(&cache_fast_lookup)) cache_fast_lookup_enabled_recalc(); return (error); } SYSCTL_PROC(_vfs, OID_AUTO, cache_fast_lookup, CTLTYPE_INT|CTLFLAG_RW|CTLFLAG_MPSAFE, &cache_fast_lookup, 0, syscal_vfs_cache_fast_lookup, "IU", ""); /* * Components of nameidata (or objects it can point to) which may * need restoring in case fast path lookup fails. */ struct nameidata_outer { size_t ni_pathlen; int cn_flags; }; struct nameidata_saved { #ifdef INVARIANTS char *cn_nameptr; size_t ni_pathlen; #endif }; #ifdef INVARIANTS struct cache_fpl_debug { size_t ni_pathlen; }; #endif struct cache_fpl { struct nameidata *ndp; struct componentname *cnp; char *nulchar; struct vnode *dvp; struct vnode *tvp; seqc_t dvp_seqc; seqc_t tvp_seqc; uint32_t hash; struct nameidata_saved snd; struct nameidata_outer snd_outer; int line; enum cache_fpl_status status:8; bool in_smr; bool fsearch; bool savename; struct pwd **pwd; #ifdef INVARIANTS struct cache_fpl_debug debug; #endif }; static bool cache_fplookup_mp_supported(struct mount *mp); static bool cache_fplookup_is_mp(struct cache_fpl *fpl); static int cache_fplookup_cross_mount(struct cache_fpl *fpl); static int cache_fplookup_partial_setup(struct cache_fpl *fpl); static int cache_fplookup_skip_slashes(struct cache_fpl *fpl); static int cache_fplookup_trailingslash(struct cache_fpl *fpl); static void cache_fpl_pathlen_dec(struct cache_fpl *fpl); static void cache_fpl_pathlen_inc(struct cache_fpl *fpl); static void cache_fpl_pathlen_add(struct cache_fpl *fpl, size_t n); static void cache_fpl_pathlen_sub(struct cache_fpl *fpl, size_t n); static void cache_fpl_cleanup_cnp(struct componentname *cnp) { uma_zfree(namei_zone, cnp->cn_pnbuf); #ifdef DIAGNOSTIC cnp->cn_pnbuf = NULL; cnp->cn_nameptr = NULL; #endif } static struct vnode * cache_fpl_handle_root(struct cache_fpl *fpl) { struct nameidata *ndp; struct componentname *cnp; ndp = fpl->ndp; cnp = fpl->cnp; MPASS(*(cnp->cn_nameptr) == '/'); cnp->cn_nameptr++; cache_fpl_pathlen_dec(fpl); if (__predict_false(*(cnp->cn_nameptr) == '/')) { do { cnp->cn_nameptr++; cache_fpl_pathlen_dec(fpl); } while (*(cnp->cn_nameptr) == '/'); } return (ndp->ni_rootdir); } static void cache_fpl_checkpoint_outer(struct cache_fpl *fpl) { fpl->snd_outer.ni_pathlen = fpl->ndp->ni_pathlen; fpl->snd_outer.cn_flags = fpl->ndp->ni_cnd.cn_flags; } static void cache_fpl_checkpoint(struct cache_fpl *fpl) { #ifdef INVARIANTS fpl->snd.cn_nameptr = fpl->ndp->ni_cnd.cn_nameptr; fpl->snd.ni_pathlen = fpl->debug.ni_pathlen; #endif } static void cache_fpl_restore_partial(struct cache_fpl *fpl) { fpl->ndp->ni_cnd.cn_flags = fpl->snd_outer.cn_flags; #ifdef INVARIANTS fpl->debug.ni_pathlen = fpl->snd.ni_pathlen; #endif } static void cache_fpl_restore_abort(struct cache_fpl *fpl) { cache_fpl_restore_partial(fpl); /* * It is 0 on entry by API contract. */ fpl->ndp->ni_resflags = 0; fpl->ndp->ni_cnd.cn_nameptr = fpl->ndp->ni_cnd.cn_pnbuf; fpl->ndp->ni_pathlen = fpl->snd_outer.ni_pathlen; } #ifdef INVARIANTS #define cache_fpl_smr_assert_entered(fpl) ({ \ struct cache_fpl *_fpl = (fpl); \ MPASS(_fpl->in_smr == true); \ VFS_SMR_ASSERT_ENTERED(); \ }) #define cache_fpl_smr_assert_not_entered(fpl) ({ \ struct cache_fpl *_fpl = (fpl); \ MPASS(_fpl->in_smr == false); \ VFS_SMR_ASSERT_NOT_ENTERED(); \ }) static void cache_fpl_assert_status(struct cache_fpl *fpl) { switch (fpl->status) { case CACHE_FPL_STATUS_UNSET: __assert_unreachable(); break; case CACHE_FPL_STATUS_DESTROYED: case CACHE_FPL_STATUS_ABORTED: case CACHE_FPL_STATUS_PARTIAL: case CACHE_FPL_STATUS_HANDLED: break; } } #else #define cache_fpl_smr_assert_entered(fpl) do { } while (0) #define cache_fpl_smr_assert_not_entered(fpl) do { } while (0) #define cache_fpl_assert_status(fpl) do { } while (0) #endif #define cache_fpl_smr_enter_initial(fpl) ({ \ struct cache_fpl *_fpl = (fpl); \ vfs_smr_enter(); \ _fpl->in_smr = true; \ }) #define cache_fpl_smr_enter(fpl) ({ \ struct cache_fpl *_fpl = (fpl); \ MPASS(_fpl->in_smr == false); \ vfs_smr_enter(); \ _fpl->in_smr = true; \ }) #define cache_fpl_smr_exit(fpl) ({ \ struct cache_fpl *_fpl = (fpl); \ MPASS(_fpl->in_smr == true); \ vfs_smr_exit(); \ _fpl->in_smr = false; \ }) static int cache_fpl_aborted_early_impl(struct cache_fpl *fpl, int line) { if (fpl->status != CACHE_FPL_STATUS_UNSET) { KASSERT(fpl->status == CACHE_FPL_STATUS_PARTIAL, ("%s: converting to abort from %d at %d, set at %d\n", __func__, fpl->status, line, fpl->line)); } cache_fpl_smr_assert_not_entered(fpl); fpl->status = CACHE_FPL_STATUS_ABORTED; fpl->line = line; return (CACHE_FPL_FAILED); } #define cache_fpl_aborted_early(x) cache_fpl_aborted_early_impl((x), __LINE__) static int __noinline cache_fpl_aborted_impl(struct cache_fpl *fpl, int line) { struct nameidata *ndp; struct componentname *cnp; ndp = fpl->ndp; cnp = fpl->cnp; if (fpl->status != CACHE_FPL_STATUS_UNSET) { KASSERT(fpl->status == CACHE_FPL_STATUS_PARTIAL, ("%s: converting to abort from %d at %d, set at %d\n", __func__, fpl->status, line, fpl->line)); } fpl->status = CACHE_FPL_STATUS_ABORTED; fpl->line = line; if (fpl->in_smr) cache_fpl_smr_exit(fpl); cache_fpl_restore_abort(fpl); /* * Resolving symlinks overwrites data passed by the caller. * Let namei know. */ if (ndp->ni_loopcnt > 0) { fpl->status = CACHE_FPL_STATUS_DESTROYED; cache_fpl_cleanup_cnp(cnp); } return (CACHE_FPL_FAILED); } #define cache_fpl_aborted(x) cache_fpl_aborted_impl((x), __LINE__) static int __noinline cache_fpl_partial_impl(struct cache_fpl *fpl, int line) { KASSERT(fpl->status == CACHE_FPL_STATUS_UNSET, ("%s: setting to partial at %d, but already set to %d at %d\n", __func__, line, fpl->status, fpl->line)); cache_fpl_smr_assert_entered(fpl); fpl->status = CACHE_FPL_STATUS_PARTIAL; fpl->line = line; return (cache_fplookup_partial_setup(fpl)); } #define cache_fpl_partial(x) cache_fpl_partial_impl((x), __LINE__) static int cache_fpl_handled_impl(struct cache_fpl *fpl, int line) { KASSERT(fpl->status == CACHE_FPL_STATUS_UNSET, ("%s: setting to handled at %d, but already set to %d at %d\n", __func__, line, fpl->status, fpl->line)); cache_fpl_smr_assert_not_entered(fpl); fpl->status = CACHE_FPL_STATUS_HANDLED; fpl->line = line; return (0); } #define cache_fpl_handled(x) cache_fpl_handled_impl((x), __LINE__) static int cache_fpl_handled_error_impl(struct cache_fpl *fpl, int error, int line) { KASSERT(fpl->status == CACHE_FPL_STATUS_UNSET, ("%s: setting to handled at %d, but already set to %d at %d\n", __func__, line, fpl->status, fpl->line)); MPASS(error != 0); MPASS(error != CACHE_FPL_FAILED); cache_fpl_smr_assert_not_entered(fpl); fpl->status = CACHE_FPL_STATUS_HANDLED; fpl->line = line; fpl->dvp = NULL; fpl->tvp = NULL; fpl->savename = false; return (error); } #define cache_fpl_handled_error(x, e) cache_fpl_handled_error_impl((x), (e), __LINE__) static bool cache_fpl_terminated(struct cache_fpl *fpl) { return (fpl->status != CACHE_FPL_STATUS_UNSET); } #define CACHE_FPL_SUPPORTED_CN_FLAGS \ (NC_NOMAKEENTRY | NC_KEEPPOSENTRY | LOCKLEAF | LOCKPARENT | WANTPARENT | \ FAILIFEXISTS | FOLLOW | LOCKSHARED | SAVENAME | SAVESTART | WILLBEDIR | \ ISOPEN | NOMACCHECK | AUDITVNODE1 | AUDITVNODE2 | NOCAPCHECK | OPENREAD | \ OPENWRITE) #define CACHE_FPL_INTERNAL_CN_FLAGS \ (ISDOTDOT | MAKEENTRY | ISLASTCN) _Static_assert((CACHE_FPL_SUPPORTED_CN_FLAGS & CACHE_FPL_INTERNAL_CN_FLAGS) == 0, "supported and internal flags overlap"); static bool cache_fpl_islastcn(struct nameidata *ndp) { return (*ndp->ni_next == 0); } static bool cache_fpl_istrailingslash(struct cache_fpl *fpl) { return (*(fpl->nulchar - 1) == '/'); } static bool cache_fpl_isdotdot(struct componentname *cnp) { if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.') return (true); return (false); } static bool cache_can_fplookup(struct cache_fpl *fpl) { struct nameidata *ndp; struct componentname *cnp; struct thread *td; ndp = fpl->ndp; cnp = fpl->cnp; - td = cnp->cn_thread; + td = curthread; if (!atomic_load_char(&cache_fast_lookup_enabled)) { cache_fpl_aborted_early(fpl); return (false); } if ((cnp->cn_flags & ~CACHE_FPL_SUPPORTED_CN_FLAGS) != 0) { cache_fpl_aborted_early(fpl); return (false); } if (IN_CAPABILITY_MODE(td)) { cache_fpl_aborted_early(fpl); return (false); } if (AUDITING_TD(td)) { cache_fpl_aborted_early(fpl); return (false); } if (ndp->ni_startdir != NULL) { cache_fpl_aborted_early(fpl); return (false); } return (true); } static int cache_fplookup_dirfd(struct cache_fpl *fpl, struct vnode **vpp) { struct nameidata *ndp; int error; bool fsearch; ndp = fpl->ndp; error = fgetvp_lookup_smr(ndp->ni_dirfd, ndp, vpp, &fsearch); if (__predict_false(error != 0)) { return (cache_fpl_aborted(fpl)); } fpl->fsearch = fsearch; return (0); } static int __noinline cache_fplookup_negative_promote(struct cache_fpl *fpl, struct namecache *oncp, uint32_t hash) { struct componentname *cnp; struct vnode *dvp; cnp = fpl->cnp; dvp = fpl->dvp; cache_fpl_smr_exit(fpl); if (cache_neg_promote_cond(dvp, cnp, oncp, hash)) return (cache_fpl_handled_error(fpl, ENOENT)); else return (cache_fpl_aborted(fpl)); } /* * The target vnode is not supported, prepare for the slow path to take over. */ static int __noinline cache_fplookup_partial_setup(struct cache_fpl *fpl) { struct nameidata *ndp; struct componentname *cnp; enum vgetstate dvs; struct vnode *dvp; struct pwd *pwd; seqc_t dvp_seqc; ndp = fpl->ndp; cnp = fpl->cnp; pwd = *(fpl->pwd); dvp = fpl->dvp; dvp_seqc = fpl->dvp_seqc; if (!pwd_hold_smr(pwd)) { return (cache_fpl_aborted(fpl)); } /* * Note that seqc is checked before the vnode is locked, so by * the time regular lookup gets to it it may have moved. * * Ultimately this does not affect correctness, any lookup errors * are userspace racing with itself. It is guaranteed that any * path which ultimately gets found could also have been found * by regular lookup going all the way in absence of concurrent * modifications. */ dvs = vget_prep_smr(dvp); cache_fpl_smr_exit(fpl); if (__predict_false(dvs == VGET_NONE)) { pwd_drop(pwd); return (cache_fpl_aborted(fpl)); } vget_finish_ref(dvp, dvs); if (!vn_seqc_consistent(dvp, dvp_seqc)) { vrele(dvp); pwd_drop(pwd); return (cache_fpl_aborted(fpl)); } cache_fpl_restore_partial(fpl); #ifdef INVARIANTS if (cnp->cn_nameptr != fpl->snd.cn_nameptr) { panic("%s: cn_nameptr mismatch (%p != %p) full [%s]\n", __func__, cnp->cn_nameptr, fpl->snd.cn_nameptr, cnp->cn_pnbuf); } #endif ndp->ni_startdir = dvp; cnp->cn_flags |= MAKEENTRY; if (cache_fpl_islastcn(ndp)) cnp->cn_flags |= ISLASTCN; if (cache_fpl_isdotdot(cnp)) cnp->cn_flags |= ISDOTDOT; /* * Skip potential extra slashes parsing did not take care of. * cache_fplookup_skip_slashes explains the mechanism. */ if (__predict_false(*(cnp->cn_nameptr) == '/')) { do { cnp->cn_nameptr++; cache_fpl_pathlen_dec(fpl); } while (*(cnp->cn_nameptr) == '/'); } ndp->ni_pathlen = fpl->nulchar - cnp->cn_nameptr + 1; #ifdef INVARIANTS if (ndp->ni_pathlen != fpl->debug.ni_pathlen) { panic("%s: mismatch (%zu != %zu) nulchar %p nameptr %p [%s] ; full string [%s]\n", __func__, ndp->ni_pathlen, fpl->debug.ni_pathlen, fpl->nulchar, cnp->cn_nameptr, cnp->cn_nameptr, cnp->cn_pnbuf); } #endif return (0); } static int cache_fplookup_final_child(struct cache_fpl *fpl, enum vgetstate tvs) { struct componentname *cnp; struct vnode *tvp; seqc_t tvp_seqc; int error, lkflags; cnp = fpl->cnp; tvp = fpl->tvp; tvp_seqc = fpl->tvp_seqc; if ((cnp->cn_flags & LOCKLEAF) != 0) { lkflags = LK_SHARED; if ((cnp->cn_flags & LOCKSHARED) == 0) lkflags = LK_EXCLUSIVE; error = vget_finish(tvp, lkflags, tvs); if (__predict_false(error != 0)) { return (cache_fpl_aborted(fpl)); } } else { vget_finish_ref(tvp, tvs); } if (!vn_seqc_consistent(tvp, tvp_seqc)) { if ((cnp->cn_flags & LOCKLEAF) != 0) vput(tvp); else vrele(tvp); return (cache_fpl_aborted(fpl)); } return (cache_fpl_handled(fpl)); } /* * They want to possibly modify the state of the namecache. */ static int __noinline cache_fplookup_final_modifying(struct cache_fpl *fpl) { struct nameidata *ndp; struct componentname *cnp; enum vgetstate dvs; struct vnode *dvp, *tvp; struct mount *mp; seqc_t dvp_seqc; int error; bool docache; ndp = fpl->ndp; cnp = fpl->cnp; dvp = fpl->dvp; dvp_seqc = fpl->dvp_seqc; MPASS(*(cnp->cn_nameptr) != '/'); MPASS(cache_fpl_islastcn(ndp)); if ((cnp->cn_flags & LOCKPARENT) == 0) MPASS((cnp->cn_flags & WANTPARENT) != 0); MPASS((cnp->cn_flags & TRAILINGSLASH) == 0); MPASS(cnp->cn_nameiop == CREATE || cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME); MPASS((cnp->cn_flags & MAKEENTRY) == 0); MPASS((cnp->cn_flags & ISDOTDOT) == 0); docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE; if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME) docache = false; /* * Regular lookup nulifies the slash, which we don't do here. * Don't take chances with filesystem routines seeing it for * the last entry. */ if (cache_fpl_istrailingslash(fpl)) { return (cache_fpl_partial(fpl)); } mp = atomic_load_ptr(&dvp->v_mount); if (__predict_false(mp == NULL)) { return (cache_fpl_aborted(fpl)); } if (__predict_false(mp->mnt_flag & MNT_RDONLY)) { cache_fpl_smr_exit(fpl); /* * Original code keeps not checking for CREATE which * might be a bug. For now let the old lookup decide. */ if (cnp->cn_nameiop == CREATE) { return (cache_fpl_aborted(fpl)); } return (cache_fpl_handled_error(fpl, EROFS)); } if (fpl->tvp != NULL && (cnp->cn_flags & FAILIFEXISTS) != 0) { cache_fpl_smr_exit(fpl); return (cache_fpl_handled_error(fpl, EEXIST)); } /* * Secure access to dvp; check cache_fplookup_partial_setup for * reasoning. * * XXX At least UFS requires its lookup routine to be called for * the last path component, which leads to some level of complication * and inefficiency: * - the target routine always locks the target vnode, but our caller * may not need it locked * - some of the VOP machinery asserts that the parent is locked, which * once more may be not required * * TODO: add a flag for filesystems which don't need this. */ dvs = vget_prep_smr(dvp); cache_fpl_smr_exit(fpl); if (__predict_false(dvs == VGET_NONE)) { return (cache_fpl_aborted(fpl)); } vget_finish_ref(dvp, dvs); if (!vn_seqc_consistent(dvp, dvp_seqc)) { vrele(dvp); return (cache_fpl_aborted(fpl)); } error = vn_lock(dvp, LK_EXCLUSIVE); if (__predict_false(error != 0)) { vrele(dvp); return (cache_fpl_aborted(fpl)); } tvp = NULL; cnp->cn_flags |= ISLASTCN; if (docache) cnp->cn_flags |= MAKEENTRY; if (cache_fpl_isdotdot(cnp)) cnp->cn_flags |= ISDOTDOT; cnp->cn_lkflags = LK_EXCLUSIVE; error = VOP_LOOKUP(dvp, &tvp, cnp); switch (error) { case EJUSTRETURN: case 0: break; case ENOTDIR: case ENOENT: vput(dvp); return (cache_fpl_handled_error(fpl, error)); default: vput(dvp); return (cache_fpl_aborted(fpl)); } fpl->tvp = tvp; fpl->savename = (cnp->cn_flags & SAVENAME) != 0; if (tvp == NULL) { if ((cnp->cn_flags & SAVESTART) != 0) { ndp->ni_startdir = dvp; vrefact(ndp->ni_startdir); cnp->cn_flags |= SAVENAME; fpl->savename = true; } MPASS(error == EJUSTRETURN); if ((cnp->cn_flags & LOCKPARENT) == 0) { VOP_UNLOCK(dvp); } return (cache_fpl_handled(fpl)); } /* * There are very hairy corner cases concerning various flag combinations * and locking state. In particular here we only hold one lock instead of * two. * * Skip the complexity as it is of no significance for normal workloads. */ if (__predict_false(tvp == dvp)) { vput(dvp); vrele(tvp); return (cache_fpl_aborted(fpl)); } /* * If they want the symlink itself we are fine, but if they want to * follow it regular lookup has to be engaged. */ if (tvp->v_type == VLNK) { if ((cnp->cn_flags & FOLLOW) != 0) { vput(dvp); vput(tvp); return (cache_fpl_aborted(fpl)); } } /* * Since we expect this to be the terminal vnode it should almost never * be a mount point. */ if (__predict_false(cache_fplookup_is_mp(fpl))) { vput(dvp); vput(tvp); return (cache_fpl_aborted(fpl)); } if ((cnp->cn_flags & FAILIFEXISTS) != 0) { vput(dvp); vput(tvp); return (cache_fpl_handled_error(fpl, EEXIST)); } if ((cnp->cn_flags & LOCKLEAF) == 0) { VOP_UNLOCK(tvp); } if ((cnp->cn_flags & LOCKPARENT) == 0) { VOP_UNLOCK(dvp); } if ((cnp->cn_flags & SAVESTART) != 0) { ndp->ni_startdir = dvp; vrefact(ndp->ni_startdir); cnp->cn_flags |= SAVENAME; fpl->savename = true; } return (cache_fpl_handled(fpl)); } static int __noinline cache_fplookup_modifying(struct cache_fpl *fpl) { struct nameidata *ndp; ndp = fpl->ndp; if (!cache_fpl_islastcn(ndp)) { return (cache_fpl_partial(fpl)); } return (cache_fplookup_final_modifying(fpl)); } static int __noinline cache_fplookup_final_withparent(struct cache_fpl *fpl) { struct componentname *cnp; enum vgetstate dvs, tvs; struct vnode *dvp, *tvp; seqc_t dvp_seqc; int error; cnp = fpl->cnp; dvp = fpl->dvp; dvp_seqc = fpl->dvp_seqc; tvp = fpl->tvp; MPASS((cnp->cn_flags & (LOCKPARENT|WANTPARENT)) != 0); /* * This is less efficient than it can be for simplicity. */ dvs = vget_prep_smr(dvp); if (__predict_false(dvs == VGET_NONE)) { return (cache_fpl_aborted(fpl)); } tvs = vget_prep_smr(tvp); if (__predict_false(tvs == VGET_NONE)) { cache_fpl_smr_exit(fpl); vget_abort(dvp, dvs); return (cache_fpl_aborted(fpl)); } cache_fpl_smr_exit(fpl); if ((cnp->cn_flags & LOCKPARENT) != 0) { error = vget_finish(dvp, LK_EXCLUSIVE, dvs); if (__predict_false(error != 0)) { vget_abort(tvp, tvs); return (cache_fpl_aborted(fpl)); } } else { vget_finish_ref(dvp, dvs); } if (!vn_seqc_consistent(dvp, dvp_seqc)) { vget_abort(tvp, tvs); if ((cnp->cn_flags & LOCKPARENT) != 0) vput(dvp); else vrele(dvp); return (cache_fpl_aborted(fpl)); } error = cache_fplookup_final_child(fpl, tvs); if (__predict_false(error != 0)) { MPASS(fpl->status == CACHE_FPL_STATUS_ABORTED || fpl->status == CACHE_FPL_STATUS_DESTROYED); if ((cnp->cn_flags & LOCKPARENT) != 0) vput(dvp); else vrele(dvp); return (error); } MPASS(fpl->status == CACHE_FPL_STATUS_HANDLED); return (0); } static int cache_fplookup_final(struct cache_fpl *fpl) { struct componentname *cnp; enum vgetstate tvs; struct vnode *dvp, *tvp; seqc_t dvp_seqc; cnp = fpl->cnp; dvp = fpl->dvp; dvp_seqc = fpl->dvp_seqc; tvp = fpl->tvp; MPASS(*(cnp->cn_nameptr) != '/'); if (cnp->cn_nameiop != LOOKUP) { return (cache_fplookup_final_modifying(fpl)); } if ((cnp->cn_flags & (LOCKPARENT|WANTPARENT)) != 0) return (cache_fplookup_final_withparent(fpl)); tvs = vget_prep_smr(tvp); if (__predict_false(tvs == VGET_NONE)) { return (cache_fpl_partial(fpl)); } if (!vn_seqc_consistent(dvp, dvp_seqc)) { cache_fpl_smr_exit(fpl); vget_abort(tvp, tvs); return (cache_fpl_aborted(fpl)); } cache_fpl_smr_exit(fpl); return (cache_fplookup_final_child(fpl, tvs)); } /* * Comment from locked lookup: * Check for degenerate name (e.g. / or "") which is a way of talking about a * directory, e.g. like "/." or ".". */ static int __noinline cache_fplookup_degenerate(struct cache_fpl *fpl) { struct componentname *cnp; struct vnode *dvp; enum vgetstate dvs; int error, lkflags; #ifdef INVARIANTS char *cp; #endif fpl->tvp = fpl->dvp; fpl->tvp_seqc = fpl->dvp_seqc; cnp = fpl->cnp; dvp = fpl->dvp; #ifdef INVARIANTS for (cp = cnp->cn_pnbuf; *cp != '\0'; cp++) { KASSERT(*cp == '/', ("%s: encountered non-slash; string [%s]\n", __func__, cnp->cn_pnbuf)); } #endif if (__predict_false(cnp->cn_nameiop != LOOKUP)) { cache_fpl_smr_exit(fpl); return (cache_fpl_handled_error(fpl, EISDIR)); } MPASS((cnp->cn_flags & SAVESTART) == 0); if ((cnp->cn_flags & (LOCKPARENT|WANTPARENT)) != 0) { return (cache_fplookup_final_withparent(fpl)); } dvs = vget_prep_smr(dvp); cache_fpl_smr_exit(fpl); if (__predict_false(dvs == VGET_NONE)) { return (cache_fpl_aborted(fpl)); } if ((cnp->cn_flags & LOCKLEAF) != 0) { lkflags = LK_SHARED; if ((cnp->cn_flags & LOCKSHARED) == 0) lkflags = LK_EXCLUSIVE; error = vget_finish(dvp, lkflags, dvs); if (__predict_false(error != 0)) { return (cache_fpl_aborted(fpl)); } } else { vget_finish_ref(dvp, dvs); } return (cache_fpl_handled(fpl)); } static int __noinline cache_fplookup_noentry(struct cache_fpl *fpl) { struct nameidata *ndp; struct componentname *cnp; enum vgetstate dvs; struct vnode *dvp, *tvp; seqc_t dvp_seqc; int error; bool docache; ndp = fpl->ndp; cnp = fpl->cnp; dvp = fpl->dvp; dvp_seqc = fpl->dvp_seqc; MPASS((cnp->cn_flags & MAKEENTRY) == 0); MPASS((cnp->cn_flags & ISDOTDOT) == 0); MPASS(!cache_fpl_isdotdot(cnp)); /* * Hack: delayed name len checking. */ if (__predict_false(cnp->cn_namelen > NAME_MAX)) { cache_fpl_smr_exit(fpl); return (cache_fpl_handled_error(fpl, ENAMETOOLONG)); } if (cnp->cn_nameptr[0] == '/') { return (cache_fplookup_skip_slashes(fpl)); } if (cnp->cn_nameptr[0] == '\0') { if (fpl->tvp == NULL) { return (cache_fplookup_degenerate(fpl)); } return (cache_fplookup_trailingslash(fpl)); } if (cnp->cn_nameiop != LOOKUP) { fpl->tvp = NULL; return (cache_fplookup_modifying(fpl)); } MPASS((cnp->cn_flags & SAVESTART) == 0); /* * Only try to fill in the component if it is the last one, * otherwise not only there may be several to handle but the * walk may be complicated. */ if (!cache_fpl_islastcn(ndp)) { return (cache_fpl_partial(fpl)); } /* * Regular lookup nulifies the slash, which we don't do here. * Don't take chances with filesystem routines seeing it for * the last entry. */ if (cache_fpl_istrailingslash(fpl)) { return (cache_fpl_partial(fpl)); } /* * Secure access to dvp; check cache_fplookup_partial_setup for * reasoning. */ dvs = vget_prep_smr(dvp); cache_fpl_smr_exit(fpl); if (__predict_false(dvs == VGET_NONE)) { return (cache_fpl_aborted(fpl)); } vget_finish_ref(dvp, dvs); if (!vn_seqc_consistent(dvp, dvp_seqc)) { vrele(dvp); return (cache_fpl_aborted(fpl)); } error = vn_lock(dvp, LK_SHARED); if (__predict_false(error != 0)) { vrele(dvp); return (cache_fpl_aborted(fpl)); } tvp = NULL; /* * TODO: provide variants which don't require locking either vnode. */ cnp->cn_flags |= ISLASTCN; docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE; if (docache) cnp->cn_flags |= MAKEENTRY; cnp->cn_lkflags = LK_SHARED; if ((cnp->cn_flags & LOCKSHARED) == 0) { cnp->cn_lkflags = LK_EXCLUSIVE; } error = VOP_LOOKUP(dvp, &tvp, cnp); switch (error) { case EJUSTRETURN: case 0: break; case ENOTDIR: case ENOENT: vput(dvp); return (cache_fpl_handled_error(fpl, error)); default: vput(dvp); return (cache_fpl_aborted(fpl)); } fpl->tvp = tvp; if (!fpl->savename) { MPASS((cnp->cn_flags & SAVENAME) == 0); } if (tvp == NULL) { MPASS(error == EJUSTRETURN); if ((cnp->cn_flags & (WANTPARENT | LOCKPARENT)) == 0) { vput(dvp); } else if ((cnp->cn_flags & LOCKPARENT) == 0) { VOP_UNLOCK(dvp); } return (cache_fpl_handled(fpl)); } if (tvp->v_type == VLNK) { if ((cnp->cn_flags & FOLLOW) != 0) { vput(dvp); vput(tvp); return (cache_fpl_aborted(fpl)); } } if (__predict_false(cache_fplookup_is_mp(fpl))) { vput(dvp); vput(tvp); return (cache_fpl_aborted(fpl)); } if ((cnp->cn_flags & LOCKLEAF) == 0) { VOP_UNLOCK(tvp); } if ((cnp->cn_flags & (WANTPARENT | LOCKPARENT)) == 0) { vput(dvp); } else if ((cnp->cn_flags & LOCKPARENT) == 0) { VOP_UNLOCK(dvp); } return (cache_fpl_handled(fpl)); } static int __noinline cache_fplookup_dot(struct cache_fpl *fpl) { int error; MPASS(!seqc_in_modify(fpl->dvp_seqc)); /* * Just re-assign the value. seqc will be checked later for the first * non-dot path component in line and/or before deciding to return the * vnode. */ fpl->tvp = fpl->dvp; fpl->tvp_seqc = fpl->dvp_seqc; counter_u64_add(dothits, 1); SDT_PROBE3(vfs, namecache, lookup, hit, fpl->dvp, ".", fpl->dvp); error = 0; if (cache_fplookup_is_mp(fpl)) { error = cache_fplookup_cross_mount(fpl); } return (error); } static int __noinline cache_fplookup_dotdot(struct cache_fpl *fpl) { struct nameidata *ndp; struct componentname *cnp; struct namecache *ncp; struct vnode *dvp; struct prison *pr; u_char nc_flag; ndp = fpl->ndp; cnp = fpl->cnp; dvp = fpl->dvp; MPASS(cache_fpl_isdotdot(cnp)); /* * XXX this is racy the same way regular lookup is */ for (pr = cnp->cn_cred->cr_prison; pr != NULL; pr = pr->pr_parent) if (dvp == pr->pr_root) break; if (dvp == ndp->ni_rootdir || dvp == ndp->ni_topdir || dvp == rootvnode || pr != NULL) { fpl->tvp = dvp; fpl->tvp_seqc = vn_seqc_read_any(dvp); if (seqc_in_modify(fpl->tvp_seqc)) { return (cache_fpl_aborted(fpl)); } return (0); } if ((dvp->v_vflag & VV_ROOT) != 0) { /* * TODO * The opposite of climb mount is needed here. */ return (cache_fpl_partial(fpl)); } ncp = atomic_load_consume_ptr(&dvp->v_cache_dd); if (ncp == NULL) { return (cache_fpl_aborted(fpl)); } nc_flag = atomic_load_char(&ncp->nc_flag); if ((nc_flag & NCF_ISDOTDOT) != 0) { if ((nc_flag & NCF_NEGATIVE) != 0) return (cache_fpl_aborted(fpl)); fpl->tvp = ncp->nc_vp; } else { fpl->tvp = ncp->nc_dvp; } fpl->tvp_seqc = vn_seqc_read_any(fpl->tvp); if (seqc_in_modify(fpl->tvp_seqc)) { return (cache_fpl_partial(fpl)); } /* * Acquire fence provided by vn_seqc_read_any above. */ if (__predict_false(atomic_load_ptr(&dvp->v_cache_dd) != ncp)) { return (cache_fpl_aborted(fpl)); } if (!cache_ncp_canuse(ncp)) { return (cache_fpl_aborted(fpl)); } counter_u64_add(dotdothits, 1); return (0); } static int __noinline cache_fplookup_neg(struct cache_fpl *fpl, struct namecache *ncp, uint32_t hash) { u_char nc_flag; bool neg_promote; nc_flag = atomic_load_char(&ncp->nc_flag); MPASS((nc_flag & NCF_NEGATIVE) != 0); /* * If they want to create an entry we need to replace this one. */ if (__predict_false(fpl->cnp->cn_nameiop != LOOKUP)) { fpl->tvp = NULL; return (cache_fplookup_modifying(fpl)); } neg_promote = cache_neg_hit_prep(ncp); if (!cache_fpl_neg_ncp_canuse(ncp)) { cache_neg_hit_abort(ncp); return (cache_fpl_partial(fpl)); } if (neg_promote) { return (cache_fplookup_negative_promote(fpl, ncp, hash)); } cache_neg_hit_finish(ncp); cache_fpl_smr_exit(fpl); return (cache_fpl_handled_error(fpl, ENOENT)); } /* * Resolve a symlink. Called by filesystem-specific routines. * * Code flow is: * ... -> cache_fplookup_symlink -> VOP_FPLOOKUP_SYMLINK -> cache_symlink_resolve */ int cache_symlink_resolve(struct cache_fpl *fpl, const char *string, size_t len) { struct nameidata *ndp; struct componentname *cnp; size_t adjust; ndp = fpl->ndp; cnp = fpl->cnp; if (__predict_false(len == 0)) { return (ENOENT); } if (__predict_false(len > MAXPATHLEN - 2)) { if (cache_fpl_istrailingslash(fpl)) { return (EAGAIN); } } ndp->ni_pathlen = fpl->nulchar - cnp->cn_nameptr - cnp->cn_namelen + 1; #ifdef INVARIANTS if (ndp->ni_pathlen != fpl->debug.ni_pathlen) { panic("%s: mismatch (%zu != %zu) nulchar %p nameptr %p [%s] ; full string [%s]\n", __func__, ndp->ni_pathlen, fpl->debug.ni_pathlen, fpl->nulchar, cnp->cn_nameptr, cnp->cn_nameptr, cnp->cn_pnbuf); } #endif if (__predict_false(len + ndp->ni_pathlen > MAXPATHLEN)) { return (ENAMETOOLONG); } if (__predict_false(ndp->ni_loopcnt++ >= MAXSYMLINKS)) { return (ELOOP); } adjust = len; if (ndp->ni_pathlen > 1) { bcopy(ndp->ni_next, cnp->cn_pnbuf + len, ndp->ni_pathlen); } else { if (cache_fpl_istrailingslash(fpl)) { adjust = len + 1; cnp->cn_pnbuf[len] = '/'; cnp->cn_pnbuf[len + 1] = '\0'; } else { cnp->cn_pnbuf[len] = '\0'; } } bcopy(string, cnp->cn_pnbuf, len); ndp->ni_pathlen += adjust; cache_fpl_pathlen_add(fpl, adjust); cnp->cn_nameptr = cnp->cn_pnbuf; fpl->nulchar = &cnp->cn_nameptr[ndp->ni_pathlen - 1]; fpl->tvp = NULL; return (0); } static int __noinline cache_fplookup_symlink(struct cache_fpl *fpl) { struct mount *mp; struct nameidata *ndp; struct componentname *cnp; struct vnode *dvp, *tvp; int error; ndp = fpl->ndp; cnp = fpl->cnp; dvp = fpl->dvp; tvp = fpl->tvp; if (cache_fpl_islastcn(ndp)) { if ((cnp->cn_flags & FOLLOW) == 0) { return (cache_fplookup_final(fpl)); } } mp = atomic_load_ptr(&dvp->v_mount); if (__predict_false(mp == NULL)) { return (cache_fpl_aborted(fpl)); } /* * Note this check races against setting the flag just like regular * lookup. */ if (__predict_false((mp->mnt_flag & MNT_NOSYMFOLLOW) != 0)) { cache_fpl_smr_exit(fpl); return (cache_fpl_handled_error(fpl, EACCES)); } error = VOP_FPLOOKUP_SYMLINK(tvp, fpl); if (__predict_false(error != 0)) { switch (error) { case EAGAIN: return (cache_fpl_partial(fpl)); case ENOENT: case ENAMETOOLONG: case ELOOP: cache_fpl_smr_exit(fpl); return (cache_fpl_handled_error(fpl, error)); default: return (cache_fpl_aborted(fpl)); } } if (*(cnp->cn_nameptr) == '/') { fpl->dvp = cache_fpl_handle_root(fpl); fpl->dvp_seqc = vn_seqc_read_any(fpl->dvp); if (seqc_in_modify(fpl->dvp_seqc)) { return (cache_fpl_aborted(fpl)); } /* * The main loop assumes that ->dvp points to a vnode belonging * to a filesystem which can do lockless lookup, but the absolute * symlink can be wandering off to one which does not. */ mp = atomic_load_ptr(&fpl->dvp->v_mount); if (__predict_false(mp == NULL)) { return (cache_fpl_aborted(fpl)); } if (!cache_fplookup_mp_supported(mp)) { cache_fpl_checkpoint(fpl); return (cache_fpl_partial(fpl)); } } return (0); } static int cache_fplookup_next(struct cache_fpl *fpl) { struct componentname *cnp; struct namecache *ncp; struct vnode *dvp, *tvp; u_char nc_flag; uint32_t hash; int error; cnp = fpl->cnp; dvp = fpl->dvp; hash = fpl->hash; if (__predict_false(cnp->cn_nameptr[0] == '.')) { if (cnp->cn_namelen == 1) { return (cache_fplookup_dot(fpl)); } if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') { return (cache_fplookup_dotdot(fpl)); } } MPASS(!cache_fpl_isdotdot(cnp)); CK_SLIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) { if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen && !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen)) break; } if (__predict_false(ncp == NULL)) { return (cache_fplookup_noentry(fpl)); } tvp = atomic_load_ptr(&ncp->nc_vp); nc_flag = atomic_load_char(&ncp->nc_flag); if ((nc_flag & NCF_NEGATIVE) != 0) { return (cache_fplookup_neg(fpl, ncp, hash)); } if (!cache_ncp_canuse(ncp)) { return (cache_fpl_partial(fpl)); } fpl->tvp = tvp; fpl->tvp_seqc = vn_seqc_read_any(tvp); if (seqc_in_modify(fpl->tvp_seqc)) { return (cache_fpl_partial(fpl)); } counter_u64_add(numposhits, 1); SDT_PROBE3(vfs, namecache, lookup, hit, dvp, ncp->nc_name, tvp); error = 0; if (cache_fplookup_is_mp(fpl)) { error = cache_fplookup_cross_mount(fpl); } return (error); } static bool cache_fplookup_mp_supported(struct mount *mp) { MPASS(mp != NULL); if ((mp->mnt_kern_flag & MNTK_FPLOOKUP) == 0) return (false); return (true); } /* * Walk up the mount stack (if any). * * Correctness is provided in the following ways: * - all vnodes are protected from freeing with SMR * - struct mount objects are type stable making them always safe to access * - stability of the particular mount is provided by busying it * - relationship between the vnode which is mounted on and the mount is * verified with the vnode sequence counter after busying * - association between root vnode of the mount and the mount is protected * by busy * * From that point on we can read the sequence counter of the root vnode * and get the next mount on the stack (if any) using the same protection. * * By the end of successful walk we are guaranteed the reached state was * indeed present at least at some point which matches the regular lookup. */ static int __noinline cache_fplookup_climb_mount(struct cache_fpl *fpl) { struct mount *mp, *prev_mp; struct mount_pcpu *mpcpu, *prev_mpcpu; struct vnode *vp; seqc_t vp_seqc; vp = fpl->tvp; vp_seqc = fpl->tvp_seqc; VNPASS(vp->v_type == VDIR || vp->v_type == VBAD, vp); mp = atomic_load_ptr(&vp->v_mountedhere); if (__predict_false(mp == NULL)) { return (0); } prev_mp = NULL; for (;;) { if (!vfs_op_thread_enter_crit(mp, mpcpu)) { if (prev_mp != NULL) vfs_op_thread_exit_crit(prev_mp, prev_mpcpu); return (cache_fpl_partial(fpl)); } if (prev_mp != NULL) vfs_op_thread_exit_crit(prev_mp, prev_mpcpu); if (!vn_seqc_consistent(vp, vp_seqc)) { vfs_op_thread_exit_crit(mp, mpcpu); return (cache_fpl_partial(fpl)); } if (!cache_fplookup_mp_supported(mp)) { vfs_op_thread_exit_crit(mp, mpcpu); return (cache_fpl_partial(fpl)); } vp = atomic_load_ptr(&mp->mnt_rootvnode); if (vp == NULL) { vfs_op_thread_exit_crit(mp, mpcpu); return (cache_fpl_partial(fpl)); } vp_seqc = vn_seqc_read_any(vp); if (seqc_in_modify(vp_seqc)) { vfs_op_thread_exit_crit(mp, mpcpu); return (cache_fpl_partial(fpl)); } prev_mp = mp; prev_mpcpu = mpcpu; mp = atomic_load_ptr(&vp->v_mountedhere); if (mp == NULL) break; } vfs_op_thread_exit_crit(prev_mp, prev_mpcpu); fpl->tvp = vp; fpl->tvp_seqc = vp_seqc; return (0); } static int __noinline cache_fplookup_cross_mount(struct cache_fpl *fpl) { struct mount *mp; struct mount_pcpu *mpcpu; struct vnode *vp; seqc_t vp_seqc; vp = fpl->tvp; vp_seqc = fpl->tvp_seqc; VNPASS(vp->v_type == VDIR || vp->v_type == VBAD, vp); mp = atomic_load_ptr(&vp->v_mountedhere); if (__predict_false(mp == NULL)) { return (0); } if (!vfs_op_thread_enter_crit(mp, mpcpu)) { return (cache_fpl_partial(fpl)); } if (!vn_seqc_consistent(vp, vp_seqc)) { vfs_op_thread_exit_crit(mp, mpcpu); return (cache_fpl_partial(fpl)); } if (!cache_fplookup_mp_supported(mp)) { vfs_op_thread_exit_crit(mp, mpcpu); return (cache_fpl_partial(fpl)); } vp = atomic_load_ptr(&mp->mnt_rootvnode); if (__predict_false(vp == NULL)) { vfs_op_thread_exit_crit(mp, mpcpu); return (cache_fpl_partial(fpl)); } vp_seqc = vn_seqc_read_any(vp); vfs_op_thread_exit_crit(mp, mpcpu); if (seqc_in_modify(vp_seqc)) { return (cache_fpl_partial(fpl)); } mp = atomic_load_ptr(&vp->v_mountedhere); if (__predict_false(mp != NULL)) { /* * There are possibly more mount points on top. * Normally this does not happen so for simplicity just start * over. */ return (cache_fplookup_climb_mount(fpl)); } fpl->tvp = vp; fpl->tvp_seqc = vp_seqc; return (0); } /* * Check if a vnode is mounted on. */ static bool cache_fplookup_is_mp(struct cache_fpl *fpl) { struct vnode *vp; vp = fpl->tvp; return ((vn_irflag_read(vp) & VIRF_MOUNTPOINT) != 0); } /* * Parse the path. * * The code was originally copy-pasted from regular lookup and despite * clean ups leaves performance on the table. Any modifications here * must take into account that in case off fallback the resulting * nameidata state has to be compatible with the original. */ /* * Debug ni_pathlen tracking. */ #ifdef INVARIANTS static void cache_fpl_pathlen_add(struct cache_fpl *fpl, size_t n) { fpl->debug.ni_pathlen += n; KASSERT(fpl->debug.ni_pathlen <= PATH_MAX, ("%s: pathlen overflow to %zd\n", __func__, fpl->debug.ni_pathlen)); } static void cache_fpl_pathlen_sub(struct cache_fpl *fpl, size_t n) { fpl->debug.ni_pathlen -= n; KASSERT(fpl->debug.ni_pathlen <= PATH_MAX, ("%s: pathlen underflow to %zd\n", __func__, fpl->debug.ni_pathlen)); } static void cache_fpl_pathlen_inc(struct cache_fpl *fpl) { cache_fpl_pathlen_add(fpl, 1); } static void cache_fpl_pathlen_dec(struct cache_fpl *fpl) { cache_fpl_pathlen_sub(fpl, 1); } #else static void cache_fpl_pathlen_add(struct cache_fpl *fpl, size_t n) { } static void cache_fpl_pathlen_sub(struct cache_fpl *fpl, size_t n) { } static void cache_fpl_pathlen_inc(struct cache_fpl *fpl) { } static void cache_fpl_pathlen_dec(struct cache_fpl *fpl) { } #endif static void cache_fplookup_parse(struct cache_fpl *fpl) { struct nameidata *ndp; struct componentname *cnp; struct vnode *dvp; char *cp; uint32_t hash; ndp = fpl->ndp; cnp = fpl->cnp; dvp = fpl->dvp; /* * Find the end of this path component, it is either / or nul. * * Store / as a temporary sentinel so that we only have one character * to test for. Pathnames tend to be short so this should not be * resulting in cache misses. * * TODO: fix this to be word-sized. */ KASSERT(&cnp->cn_nameptr[fpl->debug.ni_pathlen - 1] == fpl->nulchar, ("%s: mismatch between pathlen (%zu) and nulchar (%p != %p), string [%s]\n", __func__, fpl->debug.ni_pathlen, &cnp->cn_nameptr[fpl->debug.ni_pathlen - 1], fpl->nulchar, cnp->cn_pnbuf)); KASSERT(*fpl->nulchar == '\0', ("%s: expected nul at %p; string [%s]\n", __func__, fpl->nulchar, cnp->cn_pnbuf)); hash = cache_get_hash_iter_start(dvp); *fpl->nulchar = '/'; for (cp = cnp->cn_nameptr; *cp != '/'; cp++) { KASSERT(*cp != '\0', ("%s: encountered unexpected nul; string [%s]\n", __func__, cnp->cn_nameptr)); hash = cache_get_hash_iter(*cp, hash); continue; } *fpl->nulchar = '\0'; fpl->hash = cache_get_hash_iter_finish(hash); cnp->cn_namelen = cp - cnp->cn_nameptr; cache_fpl_pathlen_sub(fpl, cnp->cn_namelen); #ifdef INVARIANTS /* * cache_get_hash only accepts lengths up to NAME_MAX. This is fine since * we are going to fail this lookup with ENAMETOOLONG (see below). */ if (cnp->cn_namelen <= NAME_MAX) { if (fpl->hash != cache_get_hash(cnp->cn_nameptr, cnp->cn_namelen, dvp)) { panic("%s: mismatched hash for [%s] len %ld", __func__, cnp->cn_nameptr, cnp->cn_namelen); } } #endif /* * Hack: we have to check if the found path component's length exceeds * NAME_MAX. However, the condition is very rarely true and check can * be elided in the common case -- if an entry was found in the cache, * then it could not have been too long to begin with. */ ndp->ni_next = cp; } static void cache_fplookup_parse_advance(struct cache_fpl *fpl) { struct nameidata *ndp; struct componentname *cnp; ndp = fpl->ndp; cnp = fpl->cnp; cnp->cn_nameptr = ndp->ni_next; KASSERT(*(cnp->cn_nameptr) == '/', ("%s: should have seen slash at %p ; buf %p [%s]\n", __func__, cnp->cn_nameptr, cnp->cn_pnbuf, cnp->cn_pnbuf)); cnp->cn_nameptr++; cache_fpl_pathlen_dec(fpl); } /* * Skip spurious slashes in a pathname (e.g., "foo///bar") and retry. * * Lockless lookup tries to elide checking for spurious slashes and should they * be present is guaranteed to fail to find an entry. In this case the caller * must check if the name starts with a slash and call this routine. It is * going to fast forward across the spurious slashes and set the state up for * retry. */ static int __noinline cache_fplookup_skip_slashes(struct cache_fpl *fpl) { struct nameidata *ndp; struct componentname *cnp; ndp = fpl->ndp; cnp = fpl->cnp; MPASS(*(cnp->cn_nameptr) == '/'); do { cnp->cn_nameptr++; cache_fpl_pathlen_dec(fpl); } while (*(cnp->cn_nameptr) == '/'); /* * Go back to one slash so that cache_fplookup_parse_advance has * something to skip. */ cnp->cn_nameptr--; cache_fpl_pathlen_inc(fpl); /* * cache_fplookup_parse_advance starts from ndp->ni_next */ ndp->ni_next = cnp->cn_nameptr; /* * See cache_fplookup_dot. */ fpl->tvp = fpl->dvp; fpl->tvp_seqc = fpl->dvp_seqc; return (0); } /* * Handle trailing slashes (e.g., "foo/"). * * If a trailing slash is found the terminal vnode must be a directory. * Regular lookup shortens the path by nulifying the first trailing slash and * sets the TRAILINGSLASH flag to denote this took place. There are several * checks on it performed later. * * Similarly to spurious slashes, lockless lookup handles this in a speculative * manner relying on an invariant that a non-directory vnode will get a miss. * In this case cn_nameptr[0] == '\0' and cn_namelen == 0. * * Thus for a path like "foo/bar/" the code unwinds the state back to "bar/" * and denotes this is the last path component, which avoids looping back. * * Only plain lookups are supported for now to restrict corner cases to handle. */ static int __noinline cache_fplookup_trailingslash(struct cache_fpl *fpl) { #ifdef INVARIANTS size_t ni_pathlen; #endif struct nameidata *ndp; struct componentname *cnp; struct namecache *ncp; struct vnode *tvp; char *cn_nameptr_orig, *cn_nameptr_slash; seqc_t tvp_seqc; u_char nc_flag; ndp = fpl->ndp; cnp = fpl->cnp; tvp = fpl->tvp; tvp_seqc = fpl->tvp_seqc; MPASS(fpl->dvp == fpl->tvp); KASSERT(cache_fpl_istrailingslash(fpl), ("%s: expected trailing slash at %p; string [%s]\n", __func__, fpl->nulchar - 1, cnp->cn_pnbuf)); KASSERT(cnp->cn_nameptr[0] == '\0', ("%s: expected nul char at %p; string [%s]\n", __func__, &cnp->cn_nameptr[0], cnp->cn_pnbuf)); KASSERT(cnp->cn_namelen == 0, ("%s: namelen 0 but got %ld; string [%s]\n", __func__, cnp->cn_namelen, cnp->cn_pnbuf)); MPASS(cnp->cn_nameptr > cnp->cn_pnbuf); if (cnp->cn_nameiop != LOOKUP) { return (cache_fpl_aborted(fpl)); } if (__predict_false(tvp->v_type != VDIR)) { if (!vn_seqc_consistent(tvp, tvp_seqc)) { return (cache_fpl_aborted(fpl)); } cache_fpl_smr_exit(fpl); return (cache_fpl_handled_error(fpl, ENOTDIR)); } /* * Denote the last component. */ ndp->ni_next = &cnp->cn_nameptr[0]; MPASS(cache_fpl_islastcn(ndp)); /* * Unwind trailing slashes. */ cn_nameptr_orig = cnp->cn_nameptr; while (cnp->cn_nameptr >= cnp->cn_pnbuf) { cnp->cn_nameptr--; if (cnp->cn_nameptr[0] != '/') { break; } } /* * Unwind to the beginning of the path component. * * Note the path may or may not have started with a slash. */ cn_nameptr_slash = cnp->cn_nameptr; while (cnp->cn_nameptr > cnp->cn_pnbuf) { cnp->cn_nameptr--; if (cnp->cn_nameptr[0] == '/') { break; } } if (cnp->cn_nameptr[0] == '/') { cnp->cn_nameptr++; } cnp->cn_namelen = cn_nameptr_slash - cnp->cn_nameptr + 1; cache_fpl_pathlen_add(fpl, cn_nameptr_orig - cnp->cn_nameptr); cache_fpl_checkpoint(fpl); #ifdef INVARIANTS ni_pathlen = fpl->nulchar - cnp->cn_nameptr + 1; if (ni_pathlen != fpl->debug.ni_pathlen) { panic("%s: mismatch (%zu != %zu) nulchar %p nameptr %p [%s] ; full string [%s]\n", __func__, ni_pathlen, fpl->debug.ni_pathlen, fpl->nulchar, cnp->cn_nameptr, cnp->cn_nameptr, cnp->cn_pnbuf); } #endif /* * If this was a "./" lookup the parent directory is already correct. */ if (cnp->cn_nameptr[0] == '.' && cnp->cn_namelen == 1) { return (0); } /* * Otherwise we need to look it up. */ tvp = fpl->tvp; ncp = atomic_load_consume_ptr(&tvp->v_cache_dd); if (__predict_false(ncp == NULL)) { return (cache_fpl_aborted(fpl)); } nc_flag = atomic_load_char(&ncp->nc_flag); if ((nc_flag & NCF_ISDOTDOT) != 0) { return (cache_fpl_aborted(fpl)); } fpl->dvp = ncp->nc_dvp; fpl->dvp_seqc = vn_seqc_read_any(fpl->dvp); if (seqc_in_modify(fpl->dvp_seqc)) { return (cache_fpl_aborted(fpl)); } return (0); } /* * See the API contract for VOP_FPLOOKUP_VEXEC. */ static int __noinline cache_fplookup_failed_vexec(struct cache_fpl *fpl, int error) { struct componentname *cnp; struct vnode *dvp; seqc_t dvp_seqc; cnp = fpl->cnp; dvp = fpl->dvp; dvp_seqc = fpl->dvp_seqc; /* * TODO: Due to ignoring trailing slashes lookup will perform a * permission check on the last dir when it should not be doing it. It * may fail, but said failure should be ignored. It is possible to fix * it up fully without resorting to regular lookup, but for now just * abort. */ if (cache_fpl_istrailingslash(fpl)) { return (cache_fpl_aborted(fpl)); } /* * Hack: delayed degenerate path checking. */ if (cnp->cn_nameptr[0] == '\0' && fpl->tvp == NULL) { return (cache_fplookup_degenerate(fpl)); } /* * Hack: delayed name len checking. */ if (__predict_false(cnp->cn_namelen > NAME_MAX)) { cache_fpl_smr_exit(fpl); return (cache_fpl_handled_error(fpl, ENAMETOOLONG)); } /* * Hack: they may be looking up foo/bar, where foo is not a directory. * In such a case we need to return ENOTDIR, but we may happen to get * here with a different error. */ if (dvp->v_type != VDIR) { error = ENOTDIR; } /* * Hack: handle O_SEARCH. * * Open Group Base Specifications Issue 7, 2018 edition states: * * If the access mode of the open file description associated with the * file descriptor is not O_SEARCH, the function shall check whether * directory searches are permitted using the current permissions of * the directory underlying the file descriptor. If the access mode is * O_SEARCH, the function shall not perform the check. * * * Regular lookup tests for the NOEXECCHECK flag for every path * component to decide whether to do the permission check. However, * since most lookups never have the flag (and when they do it is only * present for the first path component), lockless lookup only acts on * it if there is a permission problem. Here the flag is represented * with a boolean so that we don't have to clear it on the way out. * * For simplicity this always aborts. * TODO: check if this is the first lookup and ignore the permission * problem. Note the flag has to survive fallback (if it happens to be * performed). */ if (fpl->fsearch) { return (cache_fpl_aborted(fpl)); } switch (error) { case EAGAIN: if (!vn_seqc_consistent(dvp, dvp_seqc)) { error = cache_fpl_aborted(fpl); } else { cache_fpl_partial(fpl); } break; default: if (!vn_seqc_consistent(dvp, dvp_seqc)) { error = cache_fpl_aborted(fpl); } else { cache_fpl_smr_exit(fpl); cache_fpl_handled_error(fpl, error); } break; } return (error); } static int cache_fplookup_impl(struct vnode *dvp, struct cache_fpl *fpl) { struct nameidata *ndp; struct componentname *cnp; struct mount *mp; int error; ndp = fpl->ndp; cnp = fpl->cnp; cache_fpl_checkpoint(fpl); /* * The vnode at hand is almost always stable, skip checking for it. * Worst case this postpones the check towards the end of the iteration * of the main loop. */ fpl->dvp = dvp; fpl->dvp_seqc = vn_seqc_read_notmodify(fpl->dvp); mp = atomic_load_ptr(&dvp->v_mount); if (__predict_false(mp == NULL || !cache_fplookup_mp_supported(mp))) { return (cache_fpl_aborted(fpl)); } MPASS(fpl->tvp == NULL); for (;;) { cache_fplookup_parse(fpl); error = VOP_FPLOOKUP_VEXEC(fpl->dvp, cnp->cn_cred); if (__predict_false(error != 0)) { error = cache_fplookup_failed_vexec(fpl, error); break; } error = cache_fplookup_next(fpl); if (__predict_false(cache_fpl_terminated(fpl))) { break; } VNPASS(!seqc_in_modify(fpl->tvp_seqc), fpl->tvp); if (fpl->tvp->v_type == VLNK) { error = cache_fplookup_symlink(fpl); if (cache_fpl_terminated(fpl)) { break; } } else { if (cache_fpl_islastcn(ndp)) { error = cache_fplookup_final(fpl); break; } if (!vn_seqc_consistent(fpl->dvp, fpl->dvp_seqc)) { error = cache_fpl_aborted(fpl); break; } fpl->dvp = fpl->tvp; fpl->dvp_seqc = fpl->tvp_seqc; cache_fplookup_parse_advance(fpl); } cache_fpl_checkpoint(fpl); } return (error); } /* * Fast path lookup protected with SMR and sequence counters. * * Note: all VOP_FPLOOKUP_VEXEC routines have a comment referencing this one. * * Filesystems can opt in by setting the MNTK_FPLOOKUP flag and meeting criteria * outlined below. * * Traditional vnode lookup conceptually looks like this: * * vn_lock(current); * for (;;) { * next = find(); * vn_lock(next); * vn_unlock(current); * current = next; * if (last) * break; * } * return (current); * * Each jump to the next vnode is safe memory-wise and atomic with respect to * any modifications thanks to holding respective locks. * * The same guarantee can be provided with a combination of safe memory * reclamation and sequence counters instead. If all operations which affect * the relationship between the current vnode and the one we are looking for * also modify the counter, we can verify whether all the conditions held as * we made the jump. This includes things like permissions, mount points etc. * Counter modification is provided by enclosing relevant places in * vn_seqc_write_begin()/end() calls. * * Thus this translates to: * * vfs_smr_enter(); * dvp_seqc = seqc_read_any(dvp); * if (seqc_in_modify(dvp_seqc)) // someone is altering the vnode * abort(); * for (;;) { * tvp = find(); * tvp_seqc = seqc_read_any(tvp); * if (seqc_in_modify(tvp_seqc)) // someone is altering the target vnode * abort(); * if (!seqc_consistent(dvp, dvp_seqc) // someone is altering the vnode * abort(); * dvp = tvp; // we know nothing of importance has changed * dvp_seqc = tvp_seqc; // store the counter for the tvp iteration * if (last) * break; * } * vget(); // secure the vnode * if (!seqc_consistent(tvp, tvp_seqc) // final check * abort(); * // at this point we know nothing has changed for any parent<->child pair * // as they were crossed during the lookup, meaning we matched the guarantee * // of the locked variant * return (tvp); * * The API contract for VOP_FPLOOKUP_VEXEC routines is as follows: * - they are called while within vfs_smr protection which they must never exit * - EAGAIN can be returned to denote checking could not be performed, it is * always valid to return it * - if the sequence counter has not changed the result must be valid * - if the sequence counter has changed both false positives and false negatives * are permitted (since the result will be rejected later) * - for simple cases of unix permission checks vaccess_vexec_smr can be used * * Caveats to watch out for: * - vnodes are passed unlocked and unreferenced with nothing stopping * VOP_RECLAIM, in turn meaning that ->v_data can become NULL. It is advised * to use atomic_load_ptr to fetch it. * - the aforementioned object can also get freed, meaning absent other means it * should be protected with vfs_smr * - either safely checking permissions as they are modified or guaranteeing * their stability is left to the routine */ int cache_fplookup(struct nameidata *ndp, enum cache_fpl_status *status, struct pwd **pwdp) { struct cache_fpl fpl; struct pwd *pwd; struct vnode *dvp; struct componentname *cnp; int error; fpl.status = CACHE_FPL_STATUS_UNSET; fpl.in_smr = false; fpl.ndp = ndp; fpl.cnp = cnp = &ndp->ni_cnd; MPASS(ndp->ni_lcf == 0); - MPASS(curthread == cnp->cn_thread); KASSERT ((cnp->cn_flags & CACHE_FPL_INTERNAL_CN_FLAGS) == 0, ("%s: internal flags found in cn_flags %" PRIx64, __func__, cnp->cn_flags)); if ((cnp->cn_flags & SAVESTART) != 0) { MPASS(cnp->cn_nameiop != LOOKUP); } MPASS(cnp->cn_nameptr == cnp->cn_pnbuf); if (__predict_false(!cache_can_fplookup(&fpl))) { *status = fpl.status; SDT_PROBE3(vfs, fplookup, lookup, done, ndp, fpl.line, fpl.status); return (EOPNOTSUPP); } cache_fpl_checkpoint_outer(&fpl); cache_fpl_smr_enter_initial(&fpl); #ifdef INVARIANTS fpl.debug.ni_pathlen = ndp->ni_pathlen; #endif fpl.nulchar = &cnp->cn_nameptr[ndp->ni_pathlen - 1]; fpl.fsearch = false; fpl.savename = (cnp->cn_flags & SAVENAME) != 0; fpl.tvp = NULL; /* for degenerate path handling */ fpl.pwd = pwdp; pwd = pwd_get_smr(); *(fpl.pwd) = pwd; ndp->ni_rootdir = pwd->pwd_rdir; ndp->ni_topdir = pwd->pwd_jdir; if (cnp->cn_pnbuf[0] == '/') { dvp = cache_fpl_handle_root(&fpl); MPASS(ndp->ni_resflags == 0); ndp->ni_resflags = NIRES_ABS; } else { if (ndp->ni_dirfd == AT_FDCWD) { dvp = pwd->pwd_cdir; } else { error = cache_fplookup_dirfd(&fpl, &dvp); if (__predict_false(error != 0)) { goto out; } } } SDT_PROBE4(vfs, namei, lookup, entry, dvp, cnp->cn_pnbuf, cnp->cn_flags, true); error = cache_fplookup_impl(dvp, &fpl); out: cache_fpl_smr_assert_not_entered(&fpl); cache_fpl_assert_status(&fpl); *status = fpl.status; if (SDT_PROBES_ENABLED()) { SDT_PROBE3(vfs, fplookup, lookup, done, ndp, fpl.line, fpl.status); if (fpl.status == CACHE_FPL_STATUS_HANDLED) SDT_PROBE4(vfs, namei, lookup, return, error, ndp->ni_vp, true, ndp); } if (__predict_true(fpl.status == CACHE_FPL_STATUS_HANDLED)) { MPASS(error != CACHE_FPL_FAILED); if (error != 0) { MPASS(fpl.dvp == NULL); MPASS(fpl.tvp == NULL); MPASS(fpl.savename == false); } ndp->ni_dvp = fpl.dvp; ndp->ni_vp = fpl.tvp; if (fpl.savename) { cnp->cn_flags |= HASBUF; } else { cache_fpl_cleanup_cnp(cnp); } } return (error); } diff --git a/sys/kern/vfs_lookup.c b/sys/kern/vfs_lookup.c index e262ebde6ae2..c1cefae543da 100644 --- a/sys/kern/vfs_lookup.c +++ b/sys/kern/vfs_lookup.c @@ -1,1779 +1,1777 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vfs_lookup.c 8.4 (Berkeley) 2/16/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_capsicum.h" #include "opt_ktrace.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef KTRACE #include #endif #ifdef INVARIANTS #include #endif #include #include #include #define NAMEI_DIAGNOSTIC 1 #undef NAMEI_DIAGNOSTIC SDT_PROVIDER_DEFINE(vfs); SDT_PROBE_DEFINE4(vfs, namei, lookup, entry, "struct vnode *", "char *", "unsigned long", "bool"); SDT_PROBE_DEFINE4(vfs, namei, lookup, return, "int", "struct vnode *", "bool", "struct nameidata"); /* Allocation zone for namei. */ uma_zone_t namei_zone; /* Placeholder vnode for mp traversal. */ static struct vnode *vp_crossmp; static int crossmp_vop_islocked(struct vop_islocked_args *ap) { return (LK_SHARED); } static int crossmp_vop_lock1(struct vop_lock1_args *ap) { struct vnode *vp; struct lock *lk __diagused; const char *file __diagused; int flags, line __diagused; vp = ap->a_vp; lk = vp->v_vnlock; flags = ap->a_flags; file = ap->a_file; line = ap->a_line; if ((flags & LK_SHARED) == 0) panic("invalid lock request for crossmp"); WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER, file, line, flags & LK_INTERLOCK ? &VI_MTX(vp)->lock_object : NULL); WITNESS_LOCK(&lk->lock_object, 0, file, line); if ((flags & LK_INTERLOCK) != 0) VI_UNLOCK(vp); LOCK_LOG_LOCK("SLOCK", &lk->lock_object, 0, 0, ap->a_file, line); return (0); } static int crossmp_vop_unlock(struct vop_unlock_args *ap) { struct vnode *vp; struct lock *lk __diagused; vp = ap->a_vp; lk = vp->v_vnlock; WITNESS_UNLOCK(&lk->lock_object, 0, LOCK_FILE, LOCK_LINE); LOCK_LOG_LOCK("SUNLOCK", &lk->lock_object, 0, 0, LOCK_FILE, LOCK_LINE); return (0); } static struct vop_vector crossmp_vnodeops = { .vop_default = &default_vnodeops, .vop_islocked = crossmp_vop_islocked, .vop_lock1 = crossmp_vop_lock1, .vop_unlock = crossmp_vop_unlock, }; /* * VFS_VOP_VECTOR_REGISTER(crossmp_vnodeops) is not used here since the vnode * gets allocated early. See nameiinit for the direct call below. */ struct nameicap_tracker { struct vnode *dp; TAILQ_ENTRY(nameicap_tracker) nm_link; }; /* Zone for cap mode tracker elements used for dotdot capability checks. */ MALLOC_DEFINE(M_NAMEITRACKER, "namei_tracker", "namei tracking for dotdot"); static void nameiinit(void *dummy __unused) { namei_zone = uma_zcreate("NAMEI", MAXPATHLEN, NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); vfs_vector_op_register(&crossmp_vnodeops); getnewvnode("crossmp", NULL, &crossmp_vnodeops, &vp_crossmp); } SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nameiinit, NULL); static int lookup_cap_dotdot = 1; SYSCTL_INT(_vfs, OID_AUTO, lookup_cap_dotdot, CTLFLAG_RWTUN, &lookup_cap_dotdot, 0, "enables \"..\" components in path lookup in capability mode"); static int lookup_cap_dotdot_nonlocal = 1; SYSCTL_INT(_vfs, OID_AUTO, lookup_cap_dotdot_nonlocal, CTLFLAG_RWTUN, &lookup_cap_dotdot_nonlocal, 0, "enables \"..\" components in path lookup in capability mode " "on non-local mount"); static void nameicap_tracker_add(struct nameidata *ndp, struct vnode *dp) { struct nameicap_tracker *nt; if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0 || dp->v_type != VDIR) return; nt = TAILQ_LAST(&ndp->ni_cap_tracker, nameicap_tracker_head); if (nt != NULL && nt->dp == dp) return; nt = malloc(sizeof(*nt), M_NAMEITRACKER, M_WAITOK); vhold(dp); nt->dp = dp; TAILQ_INSERT_TAIL(&ndp->ni_cap_tracker, nt, nm_link); } static void nameicap_cleanup_from(struct nameidata *ndp, struct nameicap_tracker *first) { struct nameicap_tracker *nt, *nt1; nt = first; TAILQ_FOREACH_FROM_SAFE(nt, &ndp->ni_cap_tracker, nm_link, nt1) { TAILQ_REMOVE(&ndp->ni_cap_tracker, nt, nm_link); vdrop(nt->dp); free(nt, M_NAMEITRACKER); } } static void nameicap_cleanup(struct nameidata *ndp) { KASSERT(TAILQ_EMPTY(&ndp->ni_cap_tracker) || (ndp->ni_lcf & NI_LCF_CAP_DOTDOT) != 0, ("not strictrelative")); nameicap_cleanup_from(ndp, NULL); } /* * For dotdot lookups in capability mode, only allow the component * lookup to succeed if the resulting directory was already traversed * during the operation. This catches situations where already * traversed directory is moved to different parent, and then we walk * over it with dotdots. * * Also allow to force failure of dotdot lookups for non-local * filesystems, where external agents might assist local lookups to * escape the compartment. */ static int nameicap_check_dotdot(struct nameidata *ndp, struct vnode *dp) { struct nameicap_tracker *nt; struct mount *mp; if (dp == NULL || dp->v_type != VDIR || (ndp->ni_lcf & NI_LCF_STRICTRELATIVE) == 0) return (0); if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0) return (ENOTCAPABLE); mp = dp->v_mount; if (lookup_cap_dotdot_nonlocal == 0 && mp != NULL && (mp->mnt_flag & MNT_LOCAL) == 0) return (ENOTCAPABLE); TAILQ_FOREACH_REVERSE(nt, &ndp->ni_cap_tracker, nameicap_tracker_head, nm_link) { if (dp == nt->dp) { nt = TAILQ_NEXT(nt, nm_link); if (nt != NULL) nameicap_cleanup_from(ndp, nt); return (0); } } return (ENOTCAPABLE); } static void namei_cleanup_cnp(struct componentname *cnp) { uma_zfree(namei_zone, cnp->cn_pnbuf); #ifdef DIAGNOSTIC cnp->cn_pnbuf = NULL; cnp->cn_nameptr = NULL; #endif } static int namei_handle_root(struct nameidata *ndp, struct vnode **dpp) { struct componentname *cnp; cnp = &ndp->ni_cnd; if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0) { #ifdef KTRACE if (KTRPOINT(curthread, KTR_CAPFAIL)) ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL); #endif return (ENOTCAPABLE); } while (*(cnp->cn_nameptr) == '/') { cnp->cn_nameptr++; ndp->ni_pathlen--; } *dpp = ndp->ni_rootdir; vrefact(*dpp); return (0); } static int namei_setup(struct nameidata *ndp, struct vnode **dpp, struct pwd **pwdp) { struct componentname *cnp; struct file *dfp; struct thread *td; struct pwd *pwd; cap_rights_t rights; int error; bool startdir_used; cnp = &ndp->ni_cnd; - td = cnp->cn_thread; + td = curthread; startdir_used = false; *pwdp = NULL; *dpp = NULL; #ifdef CAPABILITY_MODE /* * In capability mode, lookups must be restricted to happen in * the subtree with the root specified by the file descriptor: * - The root must be real file descriptor, not the pseudo-descriptor * AT_FDCWD. * - The passed path must be relative and not absolute. * - If lookup_cap_dotdot is disabled, path must not contain the * '..' components. * - If lookup_cap_dotdot is enabled, we verify that all '..' * components lookups result in the directories which were * previously walked by us, which prevents an escape from * the relative root. */ if (IN_CAPABILITY_MODE(td) && (cnp->cn_flags & NOCAPCHECK) == 0) { ndp->ni_lcf |= NI_LCF_STRICTRELATIVE; ndp->ni_resflags |= NIRES_STRICTREL; if (ndp->ni_dirfd == AT_FDCWD) { #ifdef KTRACE if (KTRPOINT(td, KTR_CAPFAIL)) ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL); #endif return (ECAPMODE); } } #endif error = 0; /* * Get starting point for the translation. */ pwd = pwd_hold(td); /* * The reference on ni_rootdir is acquired in the block below to avoid * back-to-back atomics for absolute lookups. */ ndp->ni_rootdir = pwd->pwd_rdir; ndp->ni_topdir = pwd->pwd_jdir; if (cnp->cn_pnbuf[0] == '/') { ndp->ni_resflags |= NIRES_ABS; error = namei_handle_root(ndp, dpp); } else { if (ndp->ni_startdir != NULL) { *dpp = ndp->ni_startdir; startdir_used = true; } else if (ndp->ni_dirfd == AT_FDCWD) { *dpp = pwd->pwd_cdir; vrefact(*dpp); } else { rights = *ndp->ni_rightsneeded; cap_rights_set_one(&rights, CAP_LOOKUP); if (cnp->cn_flags & AUDITVNODE1) AUDIT_ARG_ATFD1(ndp->ni_dirfd); if (cnp->cn_flags & AUDITVNODE2) AUDIT_ARG_ATFD2(ndp->ni_dirfd); /* * Effectively inlined fgetvp_rights, because * we need to inspect the file as well as * grabbing the vnode. No check for O_PATH, * files to implement its semantic. */ error = fget_cap(td, ndp->ni_dirfd, &rights, &dfp, &ndp->ni_filecaps); if (error != 0) { /* * Preserve the error; it should either be EBADF * or capability-related, both of which can be * safely returned to the caller. */ } else { if (dfp->f_ops == &badfileops) { error = EBADF; } else if (dfp->f_vnode == NULL) { error = ENOTDIR; } else { *dpp = dfp->f_vnode; vref(*dpp); if ((dfp->f_flag & FSEARCH) != 0) cnp->cn_flags |= NOEXECCHECK; } fdrop(dfp, td); } #ifdef CAPABILITIES /* * If file descriptor doesn't have all rights, * all lookups relative to it must also be * strictly relative. */ CAP_ALL(&rights); if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) || ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL || ndp->ni_filecaps.fc_nioctls != -1) { ndp->ni_lcf |= NI_LCF_STRICTRELATIVE; ndp->ni_resflags |= NIRES_STRICTREL; } #endif } if (error == 0 && (*dpp)->v_type != VDIR && (cnp->cn_pnbuf[0] != '\0' || (cnp->cn_flags & EMPTYPATH) == 0)) error = ENOTDIR; } if (error == 0 && (cnp->cn_flags & RBENEATH) != 0) { if (cnp->cn_pnbuf[0] == '/') { error = ENOTCAPABLE; } else if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) == 0) { ndp->ni_lcf |= NI_LCF_STRICTRELATIVE | NI_LCF_CAP_DOTDOT; } } /* * If we are auditing the kernel pathname, save the user pathname. */ if (cnp->cn_flags & AUDITVNODE1) AUDIT_ARG_UPATH1_VP(td, ndp->ni_rootdir, *dpp, cnp->cn_pnbuf); if (cnp->cn_flags & AUDITVNODE2) AUDIT_ARG_UPATH2_VP(td, ndp->ni_rootdir, *dpp, cnp->cn_pnbuf); if (ndp->ni_startdir != NULL && !startdir_used) vrele(ndp->ni_startdir); if (error != 0) { if (*dpp != NULL) vrele(*dpp); pwd_drop(pwd); return (error); } if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0 && lookup_cap_dotdot != 0) ndp->ni_lcf |= NI_LCF_CAP_DOTDOT; SDT_PROBE4(vfs, namei, lookup, entry, *dpp, cnp->cn_pnbuf, cnp->cn_flags, false); *pwdp = pwd; return (0); } static int namei_getpath(struct nameidata *ndp) { struct componentname *cnp; int error; cnp = &ndp->ni_cnd; /* * Get a buffer for the name to be translated, and copy the * name into the buffer. */ cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK); if (ndp->ni_segflg == UIO_SYSSPACE) { error = copystr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN, &ndp->ni_pathlen); } else { error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN, &ndp->ni_pathlen); } if (__predict_false(error != 0)) return (error); /* * Don't allow empty pathnames unless EMPTYPATH is specified. * Caller checks for ENOENT as an indication for the empty path. */ if (__predict_false(*cnp->cn_pnbuf == '\0')) return (ENOENT); cnp->cn_nameptr = cnp->cn_pnbuf; return (0); } static int namei_emptypath(struct nameidata *ndp) { struct componentname *cnp; struct pwd *pwd; struct vnode *dp; int error; cnp = &ndp->ni_cnd; MPASS(*cnp->cn_pnbuf == '\0'); MPASS((cnp->cn_flags & EMPTYPATH) != 0); MPASS((cnp->cn_flags & (LOCKPARENT | WANTPARENT)) == 0); ndp->ni_resflags |= NIRES_EMPTYPATH; error = namei_setup(ndp, &dp, &pwd); if (error != 0) { namei_cleanup_cnp(cnp); goto errout; } /* * Usecount on dp already provided by namei_setup. */ ndp->ni_vp = dp; namei_cleanup_cnp(cnp); pwd_drop(pwd); NDVALIDATE(ndp); if ((cnp->cn_flags & LOCKLEAF) != 0) { VOP_LOCK(dp, (cnp->cn_flags & LOCKSHARED) != 0 ? LK_SHARED : LK_EXCLUSIVE); if (VN_IS_DOOMED(dp)) { vput(dp); error = ENOENT; goto errout; } } SDT_PROBE4(vfs, namei, lookup, return, 0, ndp->ni_vp, false, ndp); return (0); errout: SDT_PROBE4(vfs, namei, lookup, return, error, NULL, false, ndp); return (error); } /* * Convert a pathname into a pointer to a locked vnode. * * The FOLLOW flag is set when symbolic links are to be followed * when they occur at the end of the name translation process. * Symbolic links are always followed for all other pathname * components other than the last. * * The segflg defines whether the name is to be copied from user * space or kernel space. * * Overall outline of namei: * * copy in name * get starting directory * while (!done && !error) { * call lookup to search path. * if symbolic link, massage name in buffer and continue * } */ int namei(struct nameidata *ndp) { char *cp; /* pointer into pathname argument */ struct vnode *dp; /* the directory we are searching */ struct iovec aiov; /* uio for reading symbolic links */ struct componentname *cnp; struct thread *td; struct pwd *pwd; struct uio auio; int error, linklen; enum cache_fpl_status status; cnp = &ndp->ni_cnd; - td = cnp->cn_thread; + td = curthread; #ifdef INVARIANTS - KASSERT(cnp->cn_thread == curthread, - ("namei not using curthread")); KASSERT((ndp->ni_debugflags & NAMEI_DBG_CALLED) == 0, ("%s: repeated call to namei without NDREINIT", __func__)); KASSERT(ndp->ni_debugflags == NAMEI_DBG_INITED, ("%s: bad debugflags %d", __func__, ndp->ni_debugflags)); ndp->ni_debugflags |= NAMEI_DBG_CALLED; if (ndp->ni_startdir != NULL) ndp->ni_debugflags |= NAMEI_DBG_HADSTARTDIR; if (cnp->cn_flags & FAILIFEXISTS) { KASSERT(cnp->cn_nameiop == CREATE, ("%s: FAILIFEXISTS passed for op %d", __func__, cnp->cn_nameiop)); /* * The limitation below is to restrict hairy corner cases. */ KASSERT((cnp->cn_flags & (LOCKPARENT | LOCKLEAF)) == LOCKPARENT, ("%s: FAILIFEXISTS must be passed with LOCKPARENT and without LOCKLEAF", __func__)); } /* * For NDVALIDATE. * * While NDINIT may seem like a more natural place to do it, there are * callers which directly modify flags past invoking init. */ cnp->cn_origflags = cnp->cn_flags; #endif - ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred; + ndp->ni_cnd.cn_cred = td->td_ucred; KASSERT(ndp->ni_resflags == 0, ("%s: garbage in ni_resflags: %x\n", __func__, ndp->ni_resflags)); KASSERT(cnp->cn_cred && td->td_proc, ("namei: bad cred/proc")); KASSERT((cnp->cn_flags & NAMEI_INTERNAL_FLAGS) == 0, ("namei: unexpected flags: %" PRIx64 "\n", cnp->cn_flags & NAMEI_INTERNAL_FLAGS)); if (cnp->cn_flags & NOCACHE) KASSERT(cnp->cn_nameiop != LOOKUP, ("%s: NOCACHE passed with LOOKUP", __func__)); MPASS(ndp->ni_startdir == NULL || ndp->ni_startdir->v_type == VDIR || ndp->ni_startdir->v_type == VBAD); ndp->ni_lcf = 0; ndp->ni_loopcnt = 0; ndp->ni_vp = NULL; error = namei_getpath(ndp); if (__predict_false(error != 0)) { if (error == ENOENT && (cnp->cn_flags & EMPTYPATH) != 0) return (namei_emptypath(ndp)); namei_cleanup_cnp(cnp); SDT_PROBE4(vfs, namei, lookup, return, error, NULL, false, ndp); return (error); } #ifdef KTRACE if (KTRPOINT(td, KTR_NAMEI)) { ktrnamei(cnp->cn_pnbuf); } #endif /* * First try looking up the target without locking any vnodes. * * We may need to start from scratch or pick up where it left off. */ error = cache_fplookup(ndp, &status, &pwd); switch (status) { case CACHE_FPL_STATUS_UNSET: __assert_unreachable(); break; case CACHE_FPL_STATUS_HANDLED: if (error == 0) NDVALIDATE(ndp); return (error); case CACHE_FPL_STATUS_PARTIAL: TAILQ_INIT(&ndp->ni_cap_tracker); dp = ndp->ni_startdir; break; case CACHE_FPL_STATUS_DESTROYED: ndp->ni_loopcnt = 0; error = namei_getpath(ndp); if (__predict_false(error != 0)) { namei_cleanup_cnp(cnp); return (error); } /* FALLTHROUGH */ case CACHE_FPL_STATUS_ABORTED: TAILQ_INIT(&ndp->ni_cap_tracker); MPASS(ndp->ni_lcf == 0); error = namei_setup(ndp, &dp, &pwd); if (error != 0) { namei_cleanup_cnp(cnp); return (error); } break; } /* * Locked lookup. */ for (;;) { ndp->ni_startdir = dp; error = lookup(ndp); if (error != 0) goto out; /* * If not a symbolic link, we're done. */ if ((cnp->cn_flags & ISSYMLINK) == 0) { SDT_PROBE4(vfs, namei, lookup, return, error, (error == 0 ? ndp->ni_vp : NULL), false, ndp); if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) { namei_cleanup_cnp(cnp); } else cnp->cn_flags |= HASBUF; nameicap_cleanup(ndp); pwd_drop(pwd); if (error == 0) NDVALIDATE(ndp); return (error); } if (ndp->ni_loopcnt++ >= MAXSYMLINKS) { error = ELOOP; break; } #ifdef MAC if ((cnp->cn_flags & NOMACCHECK) == 0) { error = mac_vnode_check_readlink(td->td_ucred, ndp->ni_vp); if (error != 0) break; } #endif if (ndp->ni_pathlen > 1) cp = uma_zalloc(namei_zone, M_WAITOK); else cp = cnp->cn_pnbuf; aiov.iov_base = cp; aiov.iov_len = MAXPATHLEN; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_offset = 0; auio.uio_rw = UIO_READ; auio.uio_segflg = UIO_SYSSPACE; auio.uio_td = td; auio.uio_resid = MAXPATHLEN; error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred); if (error != 0) { if (ndp->ni_pathlen > 1) uma_zfree(namei_zone, cp); break; } linklen = MAXPATHLEN - auio.uio_resid; if (linklen == 0) { if (ndp->ni_pathlen > 1) uma_zfree(namei_zone, cp); error = ENOENT; break; } if (linklen + ndp->ni_pathlen > MAXPATHLEN) { if (ndp->ni_pathlen > 1) uma_zfree(namei_zone, cp); error = ENAMETOOLONG; break; } if (ndp->ni_pathlen > 1) { bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen); uma_zfree(namei_zone, cnp->cn_pnbuf); cnp->cn_pnbuf = cp; } else cnp->cn_pnbuf[linklen] = '\0'; ndp->ni_pathlen += linklen; vput(ndp->ni_vp); dp = ndp->ni_dvp; /* * Check if root directory should replace current directory. */ cnp->cn_nameptr = cnp->cn_pnbuf; if (*(cnp->cn_nameptr) == '/') { vrele(dp); error = namei_handle_root(ndp, &dp); if (error != 0) goto out; } } vput(ndp->ni_vp); ndp->ni_vp = NULL; vrele(ndp->ni_dvp); out: MPASS(error != 0); SDT_PROBE4(vfs, namei, lookup, return, error, NULL, false, ndp); namei_cleanup_cnp(cnp); nameicap_cleanup(ndp); pwd_drop(pwd); return (error); } static int compute_cn_lkflags(struct mount *mp, int lkflags, int cnflags) { if (mp == NULL || ((lkflags & LK_SHARED) && (!(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED) || ((cnflags & ISDOTDOT) && (mp->mnt_kern_flag & MNTK_LOOKUP_EXCL_DOTDOT))))) { lkflags &= ~LK_SHARED; lkflags |= LK_EXCLUSIVE; } lkflags |= LK_NODDLKTREAT; return (lkflags); } static __inline int needs_exclusive_leaf(struct mount *mp, int flags) { /* * Intermediate nodes can use shared locks, we only need to * force an exclusive lock for leaf nodes. */ if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF)) return (0); /* Always use exclusive locks if LOCKSHARED isn't set. */ if (!(flags & LOCKSHARED)) return (1); /* * For lookups during open(), if the mount point supports * extended shared operations, then use a shared lock for the * leaf node, otherwise use an exclusive lock. */ if ((flags & ISOPEN) != 0) return (!MNT_EXTENDED_SHARED(mp)); /* * Lookup requests outside of open() that specify LOCKSHARED * only need a shared lock on the leaf vnode. */ return (0); } /* * Various filesystems expect to be able to copy a name component with length * bounded by NAME_MAX into a directory entry buffer of size MAXNAMLEN. Make * sure that these are the same size. */ _Static_assert(MAXNAMLEN == NAME_MAX, "MAXNAMLEN and NAME_MAX have different values"); /* * Search a pathname. * This is a very central and rather complicated routine. * * The pathname is pointed to by ni_ptr and is of length ni_pathlen. * The starting directory is taken from ni_startdir. The pathname is * descended until done, or a symbolic link is encountered. The variable * ni_more is clear if the path is completed; it is set to one if a * symbolic link needing interpretation is encountered. * * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on * whether the name is to be looked up, created, renamed, or deleted. * When CREATE, RENAME, or DELETE is specified, information usable in * creating, renaming, or deleting a directory entry may be calculated. * If flag has LOCKPARENT or'ed into it, the parent directory is returned * locked. If flag has WANTPARENT or'ed into it, the parent directory is * returned unlocked. Otherwise the parent directory is not returned. If * the target of the pathname exists and LOCKLEAF is or'ed into the flag * the target is returned locked, otherwise it is returned unlocked. * When creating or renaming and LOCKPARENT is specified, the target may not * be ".". When deleting and LOCKPARENT is specified, the target may be ".". * * Overall outline of lookup: * * dirloop: * identify next component of name at ndp->ni_ptr * handle degenerate case where name is null string * if .. and crossing mount points and on mounted filesys, find parent * call VOP_LOOKUP routine for next component name * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set * component vnode returned in ni_vp (if it exists), locked. * if result vnode is mounted on and crossing mount points, * find mounted on vnode * if more components of name, do next level at dirloop * return the answer in ni_vp, locked if LOCKLEAF set * if LOCKPARENT set, return locked parent in ni_dvp * if WANTPARENT set, return unlocked parent in ni_dvp */ int lookup(struct nameidata *ndp) { char *cp; /* pointer into pathname argument */ char *prev_ni_next; /* saved ndp->ni_next */ char *nulchar; /* location of '\0' in cn_pnbuf */ struct vnode *dp = NULL; /* the directory we are searching */ struct vnode *tdp; /* saved dp */ struct mount *mp; /* mount table entry */ struct prison *pr; size_t prev_ni_pathlen; /* saved ndp->ni_pathlen */ int docache; /* == 0 do not cache last component */ int wantparent; /* 1 => wantparent or lockparent flag */ int rdonly; /* lookup read-only flag bit */ int error = 0; int dpunlocked = 0; /* dp has already been unlocked */ int relookup = 0; /* do not consume the path component */ struct componentname *cnp = &ndp->ni_cnd; int lkflags_save; int ni_dvp_unlocked; /* * Setup: break out flag bits into variables. */ ni_dvp_unlocked = 0; wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT); KASSERT(cnp->cn_nameiop == LOOKUP || wantparent, ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT.")); /* * When set to zero, docache causes the last component of the * pathname to be deleted from the cache and the full lookup * of the name to be done (via VOP_CACHEDLOOKUP()). Often * filesystems need some pre-computed values that are made * during the full lookup, for instance UFS sets dp->i_offset. * * The docache variable is set to zero when requested by the * NOCACHE flag and for all modifying operations except CREATE. */ docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE; if (cnp->cn_nameiop == DELETE || (wantparent && cnp->cn_nameiop != CREATE && cnp->cn_nameiop != LOOKUP)) docache = 0; rdonly = cnp->cn_flags & RDONLY; cnp->cn_flags &= ~ISSYMLINK; ndp->ni_dvp = NULL; /* * We use shared locks until we hit the parent of the last cn then * we adjust based on the requesting flags. */ cnp->cn_lkflags = LK_SHARED; dp = ndp->ni_startdir; ndp->ni_startdir = NULLVP; vn_lock(dp, compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY, cnp->cn_flags)); dirloop: /* * Search a new directory. * * The last component of the filename is left accessible via * cnp->cn_nameptr for callers that need the name. Callers needing * the name set the SAVENAME flag. When done, they assume * responsibility for freeing the pathname buffer. * * Store / as a temporary sentinel so that we only have one character * to test for. Pathnames tend to be short so this should not be * resulting in cache misses. */ nulchar = &cnp->cn_nameptr[ndp->ni_pathlen - 1]; KASSERT(*nulchar == '\0', ("%s: expected nul at %p; string [%s]\n", __func__, nulchar, cnp->cn_pnbuf)); *nulchar = '/'; for (cp = cnp->cn_nameptr; *cp != '/'; cp++) { KASSERT(*cp != '\0', ("%s: encountered unexpected nul; string [%s]\n", __func__, cnp->cn_nameptr)); continue; } *nulchar = '\0'; cnp->cn_namelen = cp - cnp->cn_nameptr; if (cnp->cn_namelen > NAME_MAX) { error = ENAMETOOLONG; goto bad; } #ifdef NAMEI_DIAGNOSTIC { char c = *cp; *cp = '\0'; printf("{%s}: ", cnp->cn_nameptr); *cp = c; } #endif prev_ni_pathlen = ndp->ni_pathlen; ndp->ni_pathlen -= cnp->cn_namelen; KASSERT(ndp->ni_pathlen <= PATH_MAX, ("%s: ni_pathlen underflow to %zd\n", __func__, ndp->ni_pathlen)); prev_ni_next = ndp->ni_next; ndp->ni_next = cp; /* * Replace multiple slashes by a single slash and trailing slashes * by a null. This must be done before VOP_LOOKUP() because some * fs's don't know about trailing slashes. Remember if there were * trailing slashes to handle symlinks, existing non-directories * and non-existing files that won't be directories specially later. */ while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) { cp++; ndp->ni_pathlen--; if (*cp == '\0') { *ndp->ni_next = '\0'; cnp->cn_flags |= TRAILINGSLASH; } } ndp->ni_next = cp; cnp->cn_flags |= MAKEENTRY; if (*cp == '\0' && docache == 0) cnp->cn_flags &= ~MAKEENTRY; if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.') cnp->cn_flags |= ISDOTDOT; else cnp->cn_flags &= ~ISDOTDOT; if (*ndp->ni_next == 0) cnp->cn_flags |= ISLASTCN; else cnp->cn_flags &= ~ISLASTCN; if ((cnp->cn_flags & ISLASTCN) != 0 && cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { error = EINVAL; goto bad; } nameicap_tracker_add(ndp, dp); /* * Check for degenerate name (e.g. / or "") * which is a way of talking about a directory, * e.g. like "/." or ".". */ if (cnp->cn_nameptr[0] == '\0') { if (dp->v_type != VDIR) { error = ENOTDIR; goto bad; } if (cnp->cn_nameiop != LOOKUP) { error = EISDIR; goto bad; } if (wantparent) { ndp->ni_dvp = dp; VREF(dp); } ndp->ni_vp = dp; if (cnp->cn_flags & AUDITVNODE1) AUDIT_ARG_VNODE1(dp); else if (cnp->cn_flags & AUDITVNODE2) AUDIT_ARG_VNODE2(dp); if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF))) VOP_UNLOCK(dp); /* XXX This should probably move to the top of function. */ if (cnp->cn_flags & SAVESTART) panic("lookup: SAVESTART"); goto success; } /* * Handle "..": five special cases. * 0. If doing a capability lookup and lookup_cap_dotdot is * disabled, return ENOTCAPABLE. * 1. Return an error if this is the last component of * the name and the operation is DELETE or RENAME. * 2. If at root directory (e.g. after chroot) * or at absolute root directory * then ignore it so can't get out. * 3. If this vnode is the root of a mounted * filesystem, then replace it with the * vnode which was mounted on so we take the * .. in the other filesystem. * 4. If the vnode is the top directory of * the jail or chroot, don't let them out. * 5. If doing a capability lookup and lookup_cap_dotdot is * enabled, return ENOTCAPABLE if the lookup would escape * from the initial file descriptor directory. Checks are * done by ensuring that namei() already traversed the * result of dotdot lookup. */ if (cnp->cn_flags & ISDOTDOT) { if ((ndp->ni_lcf & (NI_LCF_STRICTRELATIVE | NI_LCF_CAP_DOTDOT)) == NI_LCF_STRICTRELATIVE) { #ifdef KTRACE if (KTRPOINT(curthread, KTR_CAPFAIL)) ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL); #endif error = ENOTCAPABLE; goto bad; } if ((cnp->cn_flags & ISLASTCN) != 0 && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { error = EINVAL; goto bad; } for (;;) { for (pr = cnp->cn_cred->cr_prison; pr != NULL; pr = pr->pr_parent) if (dp == pr->pr_root) break; if (dp == ndp->ni_rootdir || dp == ndp->ni_topdir || dp == rootvnode || pr != NULL || ((dp->v_vflag & VV_ROOT) != 0 && (cnp->cn_flags & NOCROSSMOUNT) != 0)) { ndp->ni_dvp = dp; ndp->ni_vp = dp; VREF(dp); goto nextname; } if ((dp->v_vflag & VV_ROOT) == 0) break; if (VN_IS_DOOMED(dp)) { /* forced unmount */ error = ENOENT; goto bad; } tdp = dp; dp = dp->v_mount->mnt_vnodecovered; VREF(dp); vput(tdp); vn_lock(dp, compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY, ISDOTDOT)); error = nameicap_check_dotdot(ndp, dp); if (error != 0) { #ifdef KTRACE if (KTRPOINT(curthread, KTR_CAPFAIL)) ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL); #endif goto bad; } } } /* * We now have a segment name to search for, and a directory to search. */ unionlookup: #ifdef MAC - error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp, cnp); + error = mac_vnode_check_lookup(cnp->cn_cred, dp, cnp); if (error) goto bad; #endif ndp->ni_dvp = dp; ndp->ni_vp = NULL; ASSERT_VOP_LOCKED(dp, "lookup"); /* * If we have a shared lock we may need to upgrade the lock for the * last operation. */ if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN) && dp != vp_crossmp && VOP_ISLOCKED(dp) == LK_SHARED) vn_lock(dp, LK_UPGRADE|LK_RETRY); if (VN_IS_DOOMED(dp)) { error = ENOENT; goto bad; } /* * If we're looking up the last component and we need an exclusive * lock, adjust our lkflags. */ if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags)) cnp->cn_lkflags = LK_EXCLUSIVE; #ifdef NAMEI_DIAGNOSTIC vn_printf(dp, "lookup in "); #endif lkflags_save = cnp->cn_lkflags; cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags, cnp->cn_flags); error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp); cnp->cn_lkflags = lkflags_save; if (error != 0) { KASSERT(ndp->ni_vp == NULL, ("leaf should be empty")); #ifdef NAMEI_DIAGNOSTIC printf("not found\n"); #endif if ((error == ENOENT) && (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) && (dp->v_mount->mnt_flag & MNT_UNION)) { tdp = dp; dp = dp->v_mount->mnt_vnodecovered; VREF(dp); vput(tdp); vn_lock(dp, compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY, cnp->cn_flags)); nameicap_tracker_add(ndp, dp); goto unionlookup; } if (error == ERELOOKUP) { vref(dp); ndp->ni_vp = dp; error = 0; relookup = 1; goto good; } if (error != EJUSTRETURN) goto bad; /* * At this point, we know we're at the end of the * pathname. If creating / renaming, we can consider * allowing the file or directory to be created / renamed, * provided we're not on a read-only filesystem. */ if (rdonly) { error = EROFS; goto bad; } /* trailing slash only allowed for directories */ if ((cnp->cn_flags & TRAILINGSLASH) && !(cnp->cn_flags & WILLBEDIR)) { error = ENOENT; goto bad; } if ((cnp->cn_flags & LOCKPARENT) == 0) VOP_UNLOCK(dp); /* * We return with ni_vp NULL to indicate that the entry * doesn't currently exist, leaving a pointer to the * (possibly locked) directory vnode in ndp->ni_dvp. */ if (cnp->cn_flags & SAVESTART) { ndp->ni_startdir = ndp->ni_dvp; VREF(ndp->ni_startdir); } goto success; } good: #ifdef NAMEI_DIAGNOSTIC printf("found\n"); #endif dp = ndp->ni_vp; /* * Check to see if the vnode has been mounted on; * if so find the root of the mounted filesystem. */ while (dp->v_type == VDIR && (mp = dp->v_mountedhere) && (cnp->cn_flags & NOCROSSMOUNT) == 0) { if (vfs_busy(mp, 0)) continue; vput(dp); if (dp != ndp->ni_dvp) vput(ndp->ni_dvp); else vrele(ndp->ni_dvp); vrefact(vp_crossmp); ndp->ni_dvp = vp_crossmp; error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags, cnp->cn_flags), &tdp); vfs_unbusy(mp); if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT)) panic("vp_crossmp exclusively locked or reclaimed"); if (error) { dpunlocked = 1; goto bad2; } ndp->ni_vp = dp = tdp; } /* * Check for symbolic link */ if ((dp->v_type == VLNK) && ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) || *ndp->ni_next == '/')) { cnp->cn_flags |= ISSYMLINK; if (VN_IS_DOOMED(dp)) { /* * We can't know whether the directory was mounted with * NOSYMFOLLOW, so we can't follow safely. */ error = ENOENT; goto bad2; } if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) { error = EACCES; goto bad2; } /* * Symlink code always expects an unlocked dvp. */ if (ndp->ni_dvp != ndp->ni_vp) { VOP_UNLOCK(ndp->ni_dvp); ni_dvp_unlocked = 1; } goto success; } nextname: /* * Not a symbolic link that we will follow. Continue with the * next component if there is any; otherwise, we're done. */ KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/', ("lookup: invalid path state.")); if (relookup) { relookup = 0; ndp->ni_pathlen = prev_ni_pathlen; ndp->ni_next = prev_ni_next; if (ndp->ni_dvp != dp) vput(ndp->ni_dvp); else vrele(ndp->ni_dvp); goto dirloop; } if (cnp->cn_flags & ISDOTDOT) { error = nameicap_check_dotdot(ndp, ndp->ni_vp); if (error != 0) { #ifdef KTRACE if (KTRPOINT(curthread, KTR_CAPFAIL)) ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL); #endif goto bad2; } } if (*ndp->ni_next == '/') { cnp->cn_nameptr = ndp->ni_next; while (*cnp->cn_nameptr == '/') { cnp->cn_nameptr++; ndp->ni_pathlen--; } if (ndp->ni_dvp != dp) vput(ndp->ni_dvp); else vrele(ndp->ni_dvp); goto dirloop; } /* * If we're processing a path with a trailing slash, * check that the end result is a directory. */ if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) { error = ENOTDIR; goto bad2; } /* * Disallow directory write attempts on read-only filesystems. */ if (rdonly && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { error = EROFS; goto bad2; } if (cnp->cn_flags & SAVESTART) { ndp->ni_startdir = ndp->ni_dvp; VREF(ndp->ni_startdir); } if (!wantparent) { ni_dvp_unlocked = 2; if (ndp->ni_dvp != dp) vput(ndp->ni_dvp); else vrele(ndp->ni_dvp); } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) { VOP_UNLOCK(ndp->ni_dvp); ni_dvp_unlocked = 1; } if (cnp->cn_flags & AUDITVNODE1) AUDIT_ARG_VNODE1(dp); else if (cnp->cn_flags & AUDITVNODE2) AUDIT_ARG_VNODE2(dp); if ((cnp->cn_flags & LOCKLEAF) == 0) VOP_UNLOCK(dp); success: /* * FIXME: for lookups which only cross a mount point to fetch the * root vnode, ni_dvp will be set to vp_crossmp. This can be a problem * if either WANTPARENT or LOCKPARENT is set. */ /* * Because of shared lookup we may have the vnode shared locked, but * the caller may want it to be exclusively locked. */ if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) && VOP_ISLOCKED(dp) != LK_EXCLUSIVE) { vn_lock(dp, LK_UPGRADE | LK_RETRY); if (VN_IS_DOOMED(dp)) { error = ENOENT; goto bad2; } } if (ndp->ni_vp != NULL) { if ((cnp->cn_flags & ISDOTDOT) == 0) nameicap_tracker_add(ndp, ndp->ni_vp); if ((cnp->cn_flags & (FAILIFEXISTS | ISSYMLINK)) == FAILIFEXISTS) goto bad_eexist; } return (0); bad2: if (ni_dvp_unlocked != 2) { if (dp != ndp->ni_dvp && !ni_dvp_unlocked) vput(ndp->ni_dvp); else vrele(ndp->ni_dvp); } bad: if (!dpunlocked) vput(dp); ndp->ni_vp = NULL; return (error); bad_eexist: /* * FAILIFEXISTS handling. * * XXX namei called with LOCKPARENT but not LOCKLEAF has the strange * behaviour of leaving the vnode unlocked if the target is the same * vnode as the parent. */ MPASS((cnp->cn_flags & ISSYMLINK) == 0); if (ndp->ni_vp == ndp->ni_dvp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); vrele(ndp->ni_vp); ndp->ni_dvp = NULL; ndp->ni_vp = NULL; NDFREE(ndp, NDF_ONLY_PNBUF); return (EEXIST); } /* * relookup - lookup a path name component * Used by lookup to re-acquire things. */ int relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp) { struct vnode *dp = NULL; /* the directory we are searching */ int rdonly; /* lookup read-only flag bit */ int error = 0; KASSERT(cnp->cn_flags & ISLASTCN, ("relookup: Not given last component.")); /* * Setup: break out flag bits into variables. */ KASSERT((cnp->cn_flags & (LOCKPARENT | WANTPARENT)) != 0, ("relookup: parent not wanted")); rdonly = cnp->cn_flags & RDONLY; cnp->cn_flags &= ~ISSYMLINK; dp = dvp; cnp->cn_lkflags = LK_EXCLUSIVE; vn_lock(dp, LK_EXCLUSIVE | LK_RETRY); /* * Search a new directory. * * The last component of the filename is left accessible via * cnp->cn_nameptr for callers that need the name. Callers needing * the name set the SAVENAME flag. When done, they assume * responsibility for freeing the pathname buffer. */ #ifdef NAMEI_DIAGNOSTIC printf("{%s}: ", cnp->cn_nameptr); #endif /* * Check for "" which represents the root directory after slash * removal. */ if (cnp->cn_nameptr[0] == '\0') { /* * Support only LOOKUP for "/" because lookup() * can't succeed for CREATE, DELETE and RENAME. */ KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP")); KASSERT(dp->v_type == VDIR, ("dp is not a directory")); if (!(cnp->cn_flags & LOCKLEAF)) VOP_UNLOCK(dp); *vpp = dp; /* XXX This should probably move to the top of function. */ if (cnp->cn_flags & SAVESTART) panic("lookup: SAVESTART"); return (0); } if (cnp->cn_flags & ISDOTDOT) panic ("relookup: lookup on dot-dot"); /* * We now have a segment name to search for, and a directory to search. */ #ifdef NAMEI_DIAGNOSTIC vn_printf(dp, "search in "); #endif if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) { KASSERT(*vpp == NULL, ("leaf should be empty")); if (error != EJUSTRETURN) goto bad; /* * If creating and at end of pathname, then can consider * allowing file to be created. */ if (rdonly) { error = EROFS; goto bad; } /* ASSERT(dvp == ndp->ni_startdir) */ if (cnp->cn_flags & SAVESTART) VREF(dvp); if ((cnp->cn_flags & LOCKPARENT) == 0) VOP_UNLOCK(dp); /* * We return with ni_vp NULL to indicate that the entry * doesn't currently exist, leaving a pointer to the * (possibly locked) directory vnode in ndp->ni_dvp. */ return (0); } dp = *vpp; /* * Disallow directory write attempts on read-only filesystems. */ if (rdonly && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { if (dvp == dp) vrele(dvp); else vput(dvp); error = EROFS; goto bad; } /* * Set the parent lock/ref state to the requested state. */ if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) VOP_UNLOCK(dvp); /* * Check for symbolic link */ KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW), ("relookup: symlink found.\n")); /* ASSERT(dvp == ndp->ni_startdir) */ if (cnp->cn_flags & SAVESTART) VREF(dvp); if ((cnp->cn_flags & LOCKLEAF) == 0) VOP_UNLOCK(dp); return (0); bad: vput(dp); *vpp = NULL; return (error); } /* * Free data allocated by namei(); see namei(9) for details. */ void NDFREE_PNBUF(struct nameidata *ndp) { if ((ndp->ni_cnd.cn_flags & HASBUF) != 0) { MPASS((ndp->ni_cnd.cn_flags & (SAVENAME | SAVESTART)) != 0); uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf); ndp->ni_cnd.cn_flags &= ~HASBUF; } } /* * NDFREE_PNBUF replacement for callers that know there is no buffer. * * This is a hack. Preferably the VFS layer would not produce anything more * than it was asked to do. Unfortunately several non-LOOKUP cases can add the * HASBUF flag to the result. Even then an interface could be implemented where * the caller specifies what they expected to see in the result and what they * are going to take care of. * * In the meantime provide this kludge as a trivial replacement for NDFREE_PNBUF * calls scattered throughout the kernel where we know for a fact the flag must not * be seen. */ #ifdef INVARIANTS void NDFREE_NOTHING(struct nameidata *ndp) { struct componentname *cnp; cnp = &ndp->ni_cnd; KASSERT(cnp->cn_nameiop == LOOKUP, ("%s: got non-LOOKUP op %d\n", __func__, cnp->cn_nameiop)); KASSERT((cnp->cn_flags & (SAVENAME | HASBUF)) == 0, ("%s: bad flags \%" PRIx64 "\n", __func__, cnp->cn_flags)); } #endif void (NDFREE)(struct nameidata *ndp, const u_int flags) { int unlock_dvp; int unlock_vp; unlock_dvp = 0; unlock_vp = 0; if (!(flags & NDF_NO_FREE_PNBUF)) { NDFREE_PNBUF(ndp); } if (!(flags & NDF_NO_VP_UNLOCK) && (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp) unlock_vp = 1; if (!(flags & NDF_NO_DVP_UNLOCK) && (ndp->ni_cnd.cn_flags & LOCKPARENT) && ndp->ni_dvp != ndp->ni_vp) unlock_dvp = 1; if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) { if (unlock_vp) { vput(ndp->ni_vp); unlock_vp = 0; } else vrele(ndp->ni_vp); ndp->ni_vp = NULL; } if (unlock_vp) VOP_UNLOCK(ndp->ni_vp); if (!(flags & NDF_NO_DVP_RELE) && (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) { if (unlock_dvp) { vput(ndp->ni_dvp); unlock_dvp = 0; } else vrele(ndp->ni_dvp); ndp->ni_dvp = NULL; } if (unlock_dvp) VOP_UNLOCK(ndp->ni_dvp); if (!(flags & NDF_NO_STARTDIR_RELE) && (ndp->ni_cnd.cn_flags & SAVESTART)) { vrele(ndp->ni_startdir); ndp->ni_startdir = NULL; } } #ifdef INVARIANTS /* * Validate the final state of ndp after the lookup. * * Historically filesystems were allowed to modify cn_flags. Most notably they * can add SAVENAME to the request, resulting in HASBUF and pushing subsequent * clean up to the consumer. In practice this seems to only concern != LOOKUP * operations. * * As a step towards stricter API contract this routine validates the state to * clean up. Note validation is a work in progress with the intent of becoming * stricter over time. */ #define NDMODIFYINGFLAGS (LOCKLEAF | LOCKPARENT | WANTPARENT | SAVENAME | SAVESTART | HASBUF) void NDVALIDATE(struct nameidata *ndp) { struct componentname *cnp; u_int64_t used, orig; cnp = &ndp->ni_cnd; orig = cnp->cn_origflags; used = cnp->cn_flags; switch (cnp->cn_nameiop) { case LOOKUP: /* * For plain lookup we require strict conformance -- nothing * to clean up if it was not requested by the caller. */ orig &= NDMODIFYINGFLAGS; used &= NDMODIFYINGFLAGS; if ((orig & (SAVENAME | SAVESTART)) != 0) orig |= HASBUF; if (orig != used) { goto out_mismatch; } break; case CREATE: case DELETE: case RENAME: /* * Some filesystems set SAVENAME to provoke HASBUF, accomodate * for it until it gets fixed. */ orig &= NDMODIFYINGFLAGS; orig |= (SAVENAME | HASBUF); used &= NDMODIFYINGFLAGS; used |= (SAVENAME | HASBUF); if (orig != used) { goto out_mismatch; } break; } return; out_mismatch: panic("%s: mismatched flags for op %d: added %" PRIx64 ", " "removed %" PRIx64" (%" PRIx64" != %" PRIx64"; stored %" PRIx64" != %" PRIx64")", __func__, cnp->cn_nameiop, used & ~orig, orig &~ used, orig, used, cnp->cn_origflags, cnp->cn_flags); } #endif /* * Determine if there is a suitable alternate filename under the specified * prefix for the specified path. If the create flag is set, then the * alternate prefix will be used so long as the parent directory exists. * This is used by the various compatibility ABIs so that Linux binaries prefer * files under /compat/linux for example. The chosen path (whether under * the prefix or under /) is returned in a kernel malloc'd buffer pointed * to by pathbuf. The caller is responsible for free'ing the buffer from * the M_TEMP bucket if one is returned. */ int kern_alternate_path(struct thread *td, const char *prefix, const char *path, enum uio_seg pathseg, char **pathbuf, int create, int dirfd) { struct nameidata nd, ndroot; char *ptr, *buf, *cp; size_t len, sz; int error; buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK); *pathbuf = buf; /* Copy the prefix into the new pathname as a starting point. */ len = strlcpy(buf, prefix, MAXPATHLEN); if (len >= MAXPATHLEN) { *pathbuf = NULL; free(buf, M_TEMP); return (EINVAL); } sz = MAXPATHLEN - len; ptr = buf + len; /* Append the filename to the prefix. */ if (pathseg == UIO_SYSSPACE) error = copystr(path, ptr, sz, &len); else error = copyinstr(path, ptr, sz, &len); if (error) { *pathbuf = NULL; free(buf, M_TEMP); return (error); } /* Only use a prefix with absolute pathnames. */ if (*ptr != '/') { error = EINVAL; goto keeporig; } if (dirfd != AT_FDCWD) { /* * We want the original because the "prefix" is * included in the already opened dirfd. */ bcopy(ptr, buf, len); return (0); } /* * We know that there is a / somewhere in this pathname. * Search backwards for it, to find the file's parent dir * to see if it exists in the alternate tree. If it does, * and we want to create a file (cflag is set). We don't * need to worry about the root comparison in this case. */ if (create) { for (cp = &ptr[len] - 1; *cp != '/'; cp--); *cp = '\0'; NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td); error = namei(&nd); *cp = '/'; if (error != 0) goto keeporig; } else { NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td); error = namei(&nd); if (error != 0) goto keeporig; /* * We now compare the vnode of the prefix to the one * vnode asked. If they resolve to be the same, then we * ignore the match so that the real root gets used. * This avoids the problem of traversing "../.." to find the * root directory and never finding it, because "/" resolves * to the emulation root directory. This is expensive :-( */ NDINIT(&ndroot, LOOKUP, FOLLOW, UIO_SYSSPACE, prefix, td); /* We shouldn't ever get an error from this namei(). */ error = namei(&ndroot); if (error == 0) { if (nd.ni_vp == ndroot.ni_vp) error = ENOENT; NDFREE(&ndroot, NDF_ONLY_PNBUF); vrele(ndroot.ni_vp); } } NDFREE(&nd, NDF_ONLY_PNBUF); vrele(nd.ni_vp); keeporig: /* If there was an error, use the original path name. */ if (error) bcopy(ptr, buf, len); return (error); } diff --git a/sys/kern/vfs_subr.c b/sys/kern/vfs_subr.c index 33a556fbfa2b..69ac73bd3954 100644 --- a/sys/kern/vfs_subr.c +++ b/sys/kern/vfs_subr.c @@ -1,6955 +1,6955 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95 */ /* * External virtual filesystem routines */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include "opt_watchdog.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef DDB #include #endif static void delmntque(struct vnode *vp); static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag, int slptimeo); static void syncer_shutdown(void *arg, int howto); static int vtryrecycle(struct vnode *vp); static void v_init_counters(struct vnode *); static void vn_seqc_init(struct vnode *); static void vn_seqc_write_end_free(struct vnode *vp); static void vgonel(struct vnode *); static bool vhold_recycle_free(struct vnode *); static void vfs_knllock(void *arg); static void vfs_knlunlock(void *arg); static void vfs_knl_assert_lock(void *arg, int what); static void destroy_vpollinfo(struct vpollinfo *vi); static int v_inval_buf_range_locked(struct vnode *vp, struct bufobj *bo, daddr_t startlbn, daddr_t endlbn); static void vnlru_recalc(void); /* * Number of vnodes in existence. Increased whenever getnewvnode() * allocates a new vnode, decreased in vdropl() for VIRF_DOOMED vnode. */ static u_long __exclusive_cache_line numvnodes; SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "Number of vnodes in existence"); static counter_u64_t vnodes_created; SYSCTL_COUNTER_U64(_vfs, OID_AUTO, vnodes_created, CTLFLAG_RD, &vnodes_created, "Number of vnodes created by getnewvnode"); /* * Conversion tables for conversion from vnode types to inode formats * and back. */ enum vtype iftovt_tab[16] = { VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VNON }; int vttoif_tab[10] = { 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT }; /* * List of allocates vnodes in the system. */ static TAILQ_HEAD(freelst, vnode) vnode_list; static struct vnode *vnode_list_free_marker; static struct vnode *vnode_list_reclaim_marker; /* * "Free" vnode target. Free vnodes are rarely completely free, but are * just ones that are cheap to recycle. Usually they are for files which * have been stat'd but not read; these usually have inode and namecache * data attached to them. This target is the preferred minimum size of a * sub-cache consisting mostly of such files. The system balances the size * of this sub-cache with its complement to try to prevent either from * thrashing while the other is relatively inactive. The targets express * a preference for the best balance. * * "Above" this target there are 2 further targets (watermarks) related * to recyling of free vnodes. In the best-operating case, the cache is * exactly full, the free list has size between vlowat and vhiwat above the * free target, and recycling from it and normal use maintains this state. * Sometimes the free list is below vlowat or even empty, but this state * is even better for immediate use provided the cache is not full. * Otherwise, vnlru_proc() runs to reclaim enough vnodes (usually non-free * ones) to reach one of these states. The watermarks are currently hard- * coded as 4% and 9% of the available space higher. These and the default * of 25% for wantfreevnodes are too large if the memory size is large. * E.g., 9% of 75% of MAXVNODES is more than 566000 vnodes to reclaim * whenever vnlru_proc() becomes active. */ static long wantfreevnodes; static long __exclusive_cache_line freevnodes; SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "Number of \"free\" vnodes"); static long freevnodes_old; static counter_u64_t recycles_count; SYSCTL_COUNTER_U64(_vfs, OID_AUTO, recycles, CTLFLAG_RD, &recycles_count, "Number of vnodes recycled to meet vnode cache targets"); static counter_u64_t recycles_free_count; SYSCTL_COUNTER_U64(_vfs, OID_AUTO, recycles_free, CTLFLAG_RD, &recycles_free_count, "Number of free vnodes recycled to meet vnode cache targets"); static counter_u64_t deferred_inact; SYSCTL_COUNTER_U64(_vfs, OID_AUTO, deferred_inact, CTLFLAG_RD, &deferred_inact, "Number of times inactive processing was deferred"); /* To keep more than one thread at a time from running vfs_getnewfsid */ static struct mtx mntid_mtx; /* * Lock for any access to the following: * vnode_list * numvnodes * freevnodes */ static struct mtx __exclusive_cache_line vnode_list_mtx; /* Publicly exported FS */ struct nfs_public nfs_pub; static uma_zone_t buf_trie_zone; static smr_t buf_trie_smr; /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */ static uma_zone_t vnode_zone; MALLOC_DEFINE(M_VNODEPOLL, "VN POLL", "vnode poll"); __read_frequently smr_t vfs_smr; /* * The workitem queue. * * It is useful to delay writes of file data and filesystem metadata * for tens of seconds so that quickly created and deleted files need * not waste disk bandwidth being created and removed. To realize this, * we append vnodes to a "workitem" queue. When running with a soft * updates implementation, most pending metadata dependencies should * not wait for more than a few seconds. Thus, mounted on block devices * are delayed only about a half the time that file data is delayed. * Similarly, directory updates are more critical, so are only delayed * about a third the time that file data is delayed. Thus, there are * SYNCER_MAXDELAY queues that are processed round-robin at a rate of * one each second (driven off the filesystem syncer process). The * syncer_delayno variable indicates the next queue that is to be processed. * Items that need to be processed soon are placed in this queue: * * syncer_workitem_pending[syncer_delayno] * * A delay of fifteen seconds is done by placing the request fifteen * entries later in the queue: * * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask] * */ static int syncer_delayno; static long syncer_mask; LIST_HEAD(synclist, bufobj); static struct synclist *syncer_workitem_pending; /* * The sync_mtx protects: * bo->bo_synclist * sync_vnode_count * syncer_delayno * syncer_state * syncer_workitem_pending * syncer_worklist_len * rushjob */ static struct mtx sync_mtx; static struct cv sync_wakeup; #define SYNCER_MAXDELAY 32 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */ static int syncdelay = 30; /* max time to delay syncing data */ static int filedelay = 30; /* time to delay syncing files */ SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "Time to delay syncing files (in seconds)"); static int dirdelay = 29; /* time to delay syncing directories */ SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "Time to delay syncing directories (in seconds)"); static int metadelay = 28; /* time to delay syncing metadata */ SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "Time to delay syncing metadata (in seconds)"); static int rushjob; /* number of slots to run ASAP */ static int stat_rush_requests; /* number of times I/O speeded up */ SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "Number of times I/O speeded up (rush requests)"); #define VDBATCH_SIZE 8 struct vdbatch { u_int index; long freevnodes; struct mtx lock; struct vnode *tab[VDBATCH_SIZE]; }; DPCPU_DEFINE_STATIC(struct vdbatch, vd); static void vdbatch_dequeue(struct vnode *vp); /* * When shutting down the syncer, run it at four times normal speed. */ #define SYNCER_SHUTDOWN_SPEEDUP 4 static int sync_vnode_count; static int syncer_worklist_len; static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY } syncer_state; /* Target for maximum number of vnodes. */ u_long desiredvnodes; static u_long gapvnodes; /* gap between wanted and desired */ static u_long vhiwat; /* enough extras after expansion */ static u_long vlowat; /* minimal extras before expansion */ static u_long vstir; /* nonzero to stir non-free vnodes */ static volatile int vsmalltrigger = 8; /* pref to keep if > this many pages */ static u_long vnlru_read_freevnodes(void); /* * Note that no attempt is made to sanitize these parameters. */ static int sysctl_maxvnodes(SYSCTL_HANDLER_ARGS) { u_long val; int error; val = desiredvnodes; error = sysctl_handle_long(oidp, &val, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (val == desiredvnodes) return (0); mtx_lock(&vnode_list_mtx); desiredvnodes = val; wantfreevnodes = desiredvnodes / 4; vnlru_recalc(); mtx_unlock(&vnode_list_mtx); /* * XXX There is no protection against multiple threads changing * desiredvnodes at the same time. Locking above only helps vnlru and * getnewvnode. */ vfs_hash_changesize(desiredvnodes); cache_changesize(desiredvnodes); return (0); } SYSCTL_PROC(_kern, KERN_MAXVNODES, maxvnodes, CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0, sysctl_maxvnodes, "LU", "Target for maximum number of vnodes"); static int sysctl_wantfreevnodes(SYSCTL_HANDLER_ARGS) { u_long val; int error; val = wantfreevnodes; error = sysctl_handle_long(oidp, &val, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (val == wantfreevnodes) return (0); mtx_lock(&vnode_list_mtx); wantfreevnodes = val; vnlru_recalc(); mtx_unlock(&vnode_list_mtx); return (0); } SYSCTL_PROC(_vfs, OID_AUTO, wantfreevnodes, CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0, sysctl_wantfreevnodes, "LU", "Target for minimum number of \"free\" vnodes"); SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW, &wantfreevnodes, 0, "Old name for vfs.wantfreevnodes (legacy)"); static int vnlru_nowhere; SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0, "Number of times the vnlru process ran without success"); static int sysctl_try_reclaim_vnode(SYSCTL_HANDLER_ARGS) { struct vnode *vp; struct nameidata nd; char *buf; unsigned long ndflags; int error; if (req->newptr == NULL) return (EINVAL); if (req->newlen >= PATH_MAX) return (E2BIG); buf = malloc(PATH_MAX, M_TEMP, M_WAITOK); error = SYSCTL_IN(req, buf, req->newlen); if (error != 0) goto out; buf[req->newlen] = '\0'; ndflags = LOCKLEAF | NOFOLLOW | AUDITVNODE1 | SAVENAME; NDINIT(&nd, LOOKUP, ndflags, UIO_SYSSPACE, buf, curthread); if ((error = namei(&nd)) != 0) goto out; vp = nd.ni_vp; if (VN_IS_DOOMED(vp)) { /* * This vnode is being recycled. Return != 0 to let the caller * know that the sysctl had no effect. Return EAGAIN because a * subsequent call will likely succeed (since namei will create * a new vnode if necessary) */ error = EAGAIN; goto putvnode; } counter_u64_add(recycles_count, 1); vgone(vp); putvnode: NDFREE(&nd, 0); out: free(buf, M_TEMP); return (error); } static int sysctl_ftry_reclaim_vnode(SYSCTL_HANDLER_ARGS) { struct thread *td = curthread; struct vnode *vp; struct file *fp; int error; int fd; if (req->newptr == NULL) return (EBADF); error = sysctl_handle_int(oidp, &fd, 0, req); if (error != 0) return (error); error = getvnode(curthread, fd, &cap_fcntl_rights, &fp); if (error != 0) return (error); vp = fp->f_vnode; error = vn_lock(vp, LK_EXCLUSIVE); if (error != 0) goto drop; counter_u64_add(recycles_count, 1); vgone(vp); VOP_UNLOCK(vp); drop: fdrop(fp, td); return (error); } SYSCTL_PROC(_debug, OID_AUTO, try_reclaim_vnode, CTLTYPE_STRING | CTLFLAG_MPSAFE | CTLFLAG_WR, NULL, 0, sysctl_try_reclaim_vnode, "A", "Try to reclaim a vnode by its pathname"); SYSCTL_PROC(_debug, OID_AUTO, ftry_reclaim_vnode, CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_WR, NULL, 0, sysctl_ftry_reclaim_vnode, "I", "Try to reclaim a vnode by its file descriptor"); /* Shift count for (uintptr_t)vp to initialize vp->v_hash. */ static int vnsz2log; /* * Support for the bufobj clean & dirty pctrie. */ static void * buf_trie_alloc(struct pctrie *ptree) { return (uma_zalloc_smr(buf_trie_zone, M_NOWAIT)); } static void buf_trie_free(struct pctrie *ptree, void *node) { uma_zfree_smr(buf_trie_zone, node); } PCTRIE_DEFINE_SMR(BUF, buf, b_lblkno, buf_trie_alloc, buf_trie_free, buf_trie_smr); /* * Initialize the vnode management data structures. * * Reevaluate the following cap on the number of vnodes after the physical * memory size exceeds 512GB. In the limit, as the physical memory size * grows, the ratio of the memory size in KB to vnodes approaches 64:1. */ #ifndef MAXVNODES_MAX #define MAXVNODES_MAX (512UL * 1024 * 1024 / 64) /* 8M */ #endif static MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker"); static struct vnode * vn_alloc_marker(struct mount *mp) { struct vnode *vp; vp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO); vp->v_type = VMARKER; vp->v_mount = mp; return (vp); } static void vn_free_marker(struct vnode *vp) { MPASS(vp->v_type == VMARKER); free(vp, M_VNODE_MARKER); } #ifdef KASAN static int vnode_ctor(void *mem, int size, void *arg __unused, int flags __unused) { kasan_mark(mem, size, roundup2(size, UMA_ALIGN_PTR + 1), 0); return (0); } static void vnode_dtor(void *mem, int size, void *arg __unused) { size_t end1, end2, off1, off2; _Static_assert(offsetof(struct vnode, v_vnodelist) < offsetof(struct vnode, v_dbatchcpu), "KASAN marks require updating"); off1 = offsetof(struct vnode, v_vnodelist); off2 = offsetof(struct vnode, v_dbatchcpu); end1 = off1 + sizeof(((struct vnode *)NULL)->v_vnodelist); end2 = off2 + sizeof(((struct vnode *)NULL)->v_dbatchcpu); /* * Access to the v_vnodelist and v_dbatchcpu fields are permitted even * after the vnode has been freed. Try to get some KASAN coverage by * marking everything except those two fields as invalid. Because * KASAN's tracking is not byte-granular, any preceding fields sharing * the same 8-byte aligned word must also be marked valid. */ /* Handle the area from the start until v_vnodelist... */ off1 = rounddown2(off1, KASAN_SHADOW_SCALE); kasan_mark(mem, off1, off1, KASAN_UMA_FREED); /* ... then the area between v_vnodelist and v_dbatchcpu ... */ off1 = roundup2(end1, KASAN_SHADOW_SCALE); off2 = rounddown2(off2, KASAN_SHADOW_SCALE); if (off2 > off1) kasan_mark((void *)((char *)mem + off1), off2 - off1, off2 - off1, KASAN_UMA_FREED); /* ... and finally the area from v_dbatchcpu to the end. */ off2 = roundup2(end2, KASAN_SHADOW_SCALE); kasan_mark((void *)((char *)mem + off2), size - off2, size - off2, KASAN_UMA_FREED); } #endif /* KASAN */ /* * Initialize a vnode as it first enters the zone. */ static int vnode_init(void *mem, int size, int flags) { struct vnode *vp; vp = mem; bzero(vp, size); /* * Setup locks. */ vp->v_vnlock = &vp->v_lock; mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF); /* * By default, don't allow shared locks unless filesystems opt-in. */ lockinit(vp->v_vnlock, PVFS, "vnode", VLKTIMEOUT, LK_NOSHARE | LK_IS_VNODE); /* * Initialize bufobj. */ bufobj_init(&vp->v_bufobj, vp); /* * Initialize namecache. */ cache_vnode_init(vp); /* * Initialize rangelocks. */ rangelock_init(&vp->v_rl); vp->v_dbatchcpu = NOCPU; /* * Check vhold_recycle_free for an explanation. */ vp->v_holdcnt = VHOLD_NO_SMR; vp->v_type = VNON; mtx_lock(&vnode_list_mtx); TAILQ_INSERT_BEFORE(vnode_list_free_marker, vp, v_vnodelist); mtx_unlock(&vnode_list_mtx); return (0); } /* * Free a vnode when it is cleared from the zone. */ static void vnode_fini(void *mem, int size) { struct vnode *vp; struct bufobj *bo; vp = mem; vdbatch_dequeue(vp); mtx_lock(&vnode_list_mtx); TAILQ_REMOVE(&vnode_list, vp, v_vnodelist); mtx_unlock(&vnode_list_mtx); rangelock_destroy(&vp->v_rl); lockdestroy(vp->v_vnlock); mtx_destroy(&vp->v_interlock); bo = &vp->v_bufobj; rw_destroy(BO_LOCKPTR(bo)); kasan_mark(mem, size, size, 0); } /* * Provide the size of NFS nclnode and NFS fh for calculation of the * vnode memory consumption. The size is specified directly to * eliminate dependency on NFS-private header. * * Other filesystems may use bigger or smaller (like UFS and ZFS) * private inode data, but the NFS-based estimation is ample enough. * Still, we care about differences in the size between 64- and 32-bit * platforms. * * Namecache structure size is heuristically * sizeof(struct namecache_ts) + CACHE_PATH_CUTOFF + 1. */ #ifdef _LP64 #define NFS_NCLNODE_SZ (528 + 64) #define NC_SZ 148 #else #define NFS_NCLNODE_SZ (360 + 32) #define NC_SZ 92 #endif static void vntblinit(void *dummy __unused) { struct vdbatch *vd; uma_ctor ctor; uma_dtor dtor; int cpu, physvnodes, virtvnodes; u_int i; /* * Desiredvnodes is a function of the physical memory size and the * kernel's heap size. Generally speaking, it scales with the * physical memory size. The ratio of desiredvnodes to the physical * memory size is 1:16 until desiredvnodes exceeds 98,304. * Thereafter, the * marginal ratio of desiredvnodes to the physical memory size is * 1:64. However, desiredvnodes is limited by the kernel's heap * size. The memory required by desiredvnodes vnodes and vm objects * must not exceed 1/10th of the kernel's heap size. */ physvnodes = maxproc + pgtok(vm_cnt.v_page_count) / 64 + 3 * min(98304 * 16, pgtok(vm_cnt.v_page_count)) / 64; virtvnodes = vm_kmem_size / (10 * (sizeof(struct vm_object) + sizeof(struct vnode) + NC_SZ * ncsizefactor + NFS_NCLNODE_SZ)); desiredvnodes = min(physvnodes, virtvnodes); if (desiredvnodes > MAXVNODES_MAX) { if (bootverbose) printf("Reducing kern.maxvnodes %lu -> %lu\n", desiredvnodes, MAXVNODES_MAX); desiredvnodes = MAXVNODES_MAX; } wantfreevnodes = desiredvnodes / 4; mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF); TAILQ_INIT(&vnode_list); mtx_init(&vnode_list_mtx, "vnode_list", NULL, MTX_DEF); /* * The lock is taken to appease WITNESS. */ mtx_lock(&vnode_list_mtx); vnlru_recalc(); mtx_unlock(&vnode_list_mtx); vnode_list_free_marker = vn_alloc_marker(NULL); TAILQ_INSERT_HEAD(&vnode_list, vnode_list_free_marker, v_vnodelist); vnode_list_reclaim_marker = vn_alloc_marker(NULL); TAILQ_INSERT_HEAD(&vnode_list, vnode_list_reclaim_marker, v_vnodelist); #ifdef KASAN ctor = vnode_ctor; dtor = vnode_dtor; #else ctor = NULL; dtor = NULL; #endif vnode_zone = uma_zcreate("VNODE", sizeof(struct vnode), ctor, dtor, vnode_init, vnode_fini, UMA_ALIGN_PTR, UMA_ZONE_NOKASAN); uma_zone_set_smr(vnode_zone, vfs_smr); /* * Preallocate enough nodes to support one-per buf so that * we can not fail an insert. reassignbuf() callers can not * tolerate the insertion failure. */ buf_trie_zone = uma_zcreate("BUF TRIE", pctrie_node_size(), NULL, NULL, pctrie_zone_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE | UMA_ZONE_SMR); buf_trie_smr = uma_zone_get_smr(buf_trie_zone); uma_prealloc(buf_trie_zone, nbuf); vnodes_created = counter_u64_alloc(M_WAITOK); recycles_count = counter_u64_alloc(M_WAITOK); recycles_free_count = counter_u64_alloc(M_WAITOK); deferred_inact = counter_u64_alloc(M_WAITOK); /* * Initialize the filesystem syncer. */ syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE, &syncer_mask); syncer_maxdelay = syncer_mask + 1; mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF); cv_init(&sync_wakeup, "syncer"); for (i = 1; i <= sizeof(struct vnode); i <<= 1) vnsz2log++; vnsz2log--; CPU_FOREACH(cpu) { vd = DPCPU_ID_PTR((cpu), vd); bzero(vd, sizeof(*vd)); mtx_init(&vd->lock, "vdbatch", NULL, MTX_DEF); } } SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL); /* * Mark a mount point as busy. Used to synchronize access and to delay * unmounting. Eventually, mountlist_mtx is not released on failure. * * vfs_busy() is a custom lock, it can block the caller. * vfs_busy() only sleeps if the unmount is active on the mount point. * For a mountpoint mp, vfs_busy-enforced lock is before lock of any * vnode belonging to mp. * * Lookup uses vfs_busy() to traverse mount points. * root fs var fs * / vnode lock A / vnode lock (/var) D * /var vnode lock B /log vnode lock(/var/log) E * vfs_busy lock C vfs_busy lock F * * Within each file system, the lock order is C->A->B and F->D->E. * * When traversing across mounts, the system follows that lock order: * * C->A->B * | * +->F->D->E * * The lookup() process for namei("/var") illustrates the process: * VOP_LOOKUP() obtains B while A is held * vfs_busy() obtains a shared lock on F while A and B are held * vput() releases lock on B * vput() releases lock on A * VFS_ROOT() obtains lock on D while shared lock on F is held * vfs_unbusy() releases shared lock on F * vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A. * Attempt to lock A (instead of vp_crossmp) while D is held would * violate the global order, causing deadlocks. * * dounmount() locks B while F is drained. */ int vfs_busy(struct mount *mp, int flags) { struct mount_pcpu *mpcpu; MPASS((flags & ~MBF_MASK) == 0); CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags); if (vfs_op_thread_enter(mp, mpcpu)) { MPASS((mp->mnt_kern_flag & MNTK_DRAINING) == 0); MPASS((mp->mnt_kern_flag & MNTK_UNMOUNT) == 0); MPASS((mp->mnt_kern_flag & MNTK_REFEXPIRE) == 0); vfs_mp_count_add_pcpu(mpcpu, ref, 1); vfs_mp_count_add_pcpu(mpcpu, lockref, 1); vfs_op_thread_exit(mp, mpcpu); if (flags & MBF_MNTLSTLOCK) mtx_unlock(&mountlist_mtx); return (0); } MNT_ILOCK(mp); vfs_assert_mount_counters(mp); MNT_REF(mp); /* * If mount point is currently being unmounted, sleep until the * mount point fate is decided. If thread doing the unmounting fails, * it will clear MNTK_UNMOUNT flag before waking us up, indicating * that this mount point has survived the unmount attempt and vfs_busy * should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE * flag in addition to MNTK_UNMOUNT, indicating that mount point is * about to be really destroyed. vfs_busy needs to release its * reference on the mount point in this case and return with ENOENT, * telling the caller that mount mount it tried to busy is no longer * valid. */ while (mp->mnt_kern_flag & MNTK_UNMOUNT) { KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("%s: non-empty upper mount list with pending unmount", __func__)); if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) { MNT_REL(mp); MNT_IUNLOCK(mp); CTR1(KTR_VFS, "%s: failed busying before sleeping", __func__); return (ENOENT); } if (flags & MBF_MNTLSTLOCK) mtx_unlock(&mountlist_mtx); mp->mnt_kern_flag |= MNTK_MWAIT; msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0); if (flags & MBF_MNTLSTLOCK) mtx_lock(&mountlist_mtx); MNT_ILOCK(mp); } if (flags & MBF_MNTLSTLOCK) mtx_unlock(&mountlist_mtx); mp->mnt_lockref++; MNT_IUNLOCK(mp); return (0); } /* * Free a busy filesystem. */ void vfs_unbusy(struct mount *mp) { struct mount_pcpu *mpcpu; int c; CTR2(KTR_VFS, "%s: mp %p", __func__, mp); if (vfs_op_thread_enter(mp, mpcpu)) { MPASS((mp->mnt_kern_flag & MNTK_DRAINING) == 0); vfs_mp_count_sub_pcpu(mpcpu, lockref, 1); vfs_mp_count_sub_pcpu(mpcpu, ref, 1); vfs_op_thread_exit(mp, mpcpu); return; } MNT_ILOCK(mp); vfs_assert_mount_counters(mp); MNT_REL(mp); c = --mp->mnt_lockref; if (mp->mnt_vfs_ops == 0) { MPASS((mp->mnt_kern_flag & MNTK_DRAINING) == 0); MNT_IUNLOCK(mp); return; } if (c < 0) vfs_dump_mount_counters(mp); if (c == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) { MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT); CTR1(KTR_VFS, "%s: waking up waiters", __func__); mp->mnt_kern_flag &= ~MNTK_DRAINING; wakeup(&mp->mnt_lockref); } MNT_IUNLOCK(mp); } /* * Lookup a mount point by filesystem identifier. */ struct mount * vfs_getvfs(fsid_t *fsid) { struct mount *mp; CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid); mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { if (fsidcmp(&mp->mnt_stat.f_fsid, fsid) == 0) { vfs_ref(mp); mtx_unlock(&mountlist_mtx); return (mp); } } mtx_unlock(&mountlist_mtx); CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid); return ((struct mount *) 0); } /* * Lookup a mount point by filesystem identifier, busying it before * returning. * * To avoid congestion on mountlist_mtx, implement simple direct-mapped * cache for popular filesystem identifiers. The cache is lockess, using * the fact that struct mount's are never freed. In worst case we may * get pointer to unmounted or even different filesystem, so we have to * check what we got, and go slow way if so. */ struct mount * vfs_busyfs(fsid_t *fsid) { #define FSID_CACHE_SIZE 256 typedef struct mount * volatile vmp_t; static vmp_t cache[FSID_CACHE_SIZE]; struct mount *mp; int error; uint32_t hash; CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid); hash = fsid->val[0] ^ fsid->val[1]; hash = (hash >> 16 ^ hash) & (FSID_CACHE_SIZE - 1); mp = cache[hash]; if (mp == NULL || fsidcmp(&mp->mnt_stat.f_fsid, fsid) != 0) goto slow; if (vfs_busy(mp, 0) != 0) { cache[hash] = NULL; goto slow; } if (fsidcmp(&mp->mnt_stat.f_fsid, fsid) == 0) return (mp); else vfs_unbusy(mp); slow: mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { if (fsidcmp(&mp->mnt_stat.f_fsid, fsid) == 0) { error = vfs_busy(mp, MBF_MNTLSTLOCK); if (error) { cache[hash] = NULL; mtx_unlock(&mountlist_mtx); return (NULL); } cache[hash] = mp; return (mp); } } CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid); mtx_unlock(&mountlist_mtx); return ((struct mount *) 0); } /* * Check if a user can access privileged mount options. */ int vfs_suser(struct mount *mp, struct thread *td) { int error; if (jailed(td->td_ucred)) { /* * If the jail of the calling thread lacks permission for * this type of file system, deny immediately. */ if (!prison_allow(td->td_ucred, mp->mnt_vfc->vfc_prison_flag)) return (EPERM); /* * If the file system was mounted outside the jail of the * calling thread, deny immediately. */ if (prison_check(td->td_ucred, mp->mnt_cred) != 0) return (EPERM); } /* * If file system supports delegated administration, we don't check * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified * by the file system itself. * If this is not the user that did original mount, we check for * the PRIV_VFS_MOUNT_OWNER privilege. */ if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) && mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) { if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0) return (error); } return (0); } /* * Get a new unique fsid. Try to make its val[0] unique, since this value * will be used to create fake device numbers for stat(). Also try (but * not so hard) make its val[0] unique mod 2^16, since some emulators only * support 16-bit device numbers. We end up with unique val[0]'s for the * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls. * * Keep in mind that several mounts may be running in parallel. Starting * the search one past where the previous search terminated is both a * micro-optimization and a defense against returning the same fsid to * different mounts. */ void vfs_getnewfsid(struct mount *mp) { static uint16_t mntid_base; struct mount *nmp; fsid_t tfsid; int mtype; CTR2(KTR_VFS, "%s: mp %p", __func__, mp); mtx_lock(&mntid_mtx); mtype = mp->mnt_vfc->vfc_typenum; tfsid.val[1] = mtype; mtype = (mtype & 0xFF) << 24; for (;;) { tfsid.val[0] = makedev(255, mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF)); mntid_base++; if ((nmp = vfs_getvfs(&tfsid)) == NULL) break; vfs_rel(nmp); } mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; mp->mnt_stat.f_fsid.val[1] = tfsid.val[1]; mtx_unlock(&mntid_mtx); } /* * Knob to control the precision of file timestamps: * * 0 = seconds only; nanoseconds zeroed. * 1 = seconds and nanoseconds, accurate within 1/HZ. * 2 = seconds and nanoseconds, truncated to microseconds. * >=3 = seconds and nanoseconds, maximum precision. */ enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; static int timestamp_precision = TSP_USEC; SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, ×tamp_precision, 0, "File timestamp precision (0: seconds, " "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to us, " "3+: sec + ns (max. precision))"); /* * Get a current timestamp. */ void vfs_timestamp(struct timespec *tsp) { struct timeval tv; switch (timestamp_precision) { case TSP_SEC: tsp->tv_sec = time_second; tsp->tv_nsec = 0; break; case TSP_HZ: getnanotime(tsp); break; case TSP_USEC: microtime(&tv); TIMEVAL_TO_TIMESPEC(&tv, tsp); break; case TSP_NSEC: default: nanotime(tsp); break; } } /* * Set vnode attributes to VNOVAL */ void vattr_null(struct vattr *vap) { vap->va_type = VNON; vap->va_size = VNOVAL; vap->va_bytes = VNOVAL; vap->va_mode = VNOVAL; vap->va_nlink = VNOVAL; vap->va_uid = VNOVAL; vap->va_gid = VNOVAL; vap->va_fsid = VNOVAL; vap->va_fileid = VNOVAL; vap->va_blocksize = VNOVAL; vap->va_rdev = VNOVAL; vap->va_atime.tv_sec = VNOVAL; vap->va_atime.tv_nsec = VNOVAL; vap->va_mtime.tv_sec = VNOVAL; vap->va_mtime.tv_nsec = VNOVAL; vap->va_ctime.tv_sec = VNOVAL; vap->va_ctime.tv_nsec = VNOVAL; vap->va_birthtime.tv_sec = VNOVAL; vap->va_birthtime.tv_nsec = VNOVAL; vap->va_flags = VNOVAL; vap->va_gen = VNOVAL; vap->va_vaflags = 0; } /* * Try to reduce the total number of vnodes. * * This routine (and its user) are buggy in at least the following ways: * - all parameters were picked years ago when RAM sizes were significantly * smaller * - it can pick vnodes based on pages used by the vm object, but filesystems * like ZFS don't use it making the pick broken * - since ZFS has its own aging policy it gets partially combated by this one * - a dedicated method should be provided for filesystems to let them decide * whether the vnode should be recycled * * This routine is called when we have too many vnodes. It attempts * to free vnodes and will potentially free vnodes that still * have VM backing store (VM backing store is typically the cause * of a vnode blowout so we want to do this). Therefore, this operation * is not considered cheap. * * A number of conditions may prevent a vnode from being reclaimed. * the buffer cache may have references on the vnode, a directory * vnode may still have references due to the namei cache representing * underlying files, or the vnode may be in active use. It is not * desirable to reuse such vnodes. These conditions may cause the * number of vnodes to reach some minimum value regardless of what * you set kern.maxvnodes to. Do not set kern.maxvnodes too low. * * @param reclaim_nc_src Only reclaim directories with outgoing namecache * entries if this argument is strue * @param trigger Only reclaim vnodes with fewer than this many resident * pages. * @param target How many vnodes to reclaim. * @return The number of vnodes that were reclaimed. */ static int vlrureclaim(bool reclaim_nc_src, int trigger, u_long target) { struct vnode *vp, *mvp; struct mount *mp; struct vm_object *object; u_long done; bool retried; mtx_assert(&vnode_list_mtx, MA_OWNED); retried = false; done = 0; mvp = vnode_list_reclaim_marker; restart: vp = mvp; while (done < target) { vp = TAILQ_NEXT(vp, v_vnodelist); if (__predict_false(vp == NULL)) break; if (__predict_false(vp->v_type == VMARKER)) continue; /* * If it's been deconstructed already, it's still * referenced, or it exceeds the trigger, skip it. * Also skip free vnodes. We are trying to make space * to expand the free list, not reduce it. */ if (vp->v_usecount > 0 || vp->v_holdcnt == 0 || (!reclaim_nc_src && !LIST_EMPTY(&vp->v_cache_src))) goto next_iter; if (vp->v_type == VBAD || vp->v_type == VNON) goto next_iter; object = atomic_load_ptr(&vp->v_object); if (object == NULL || object->resident_page_count > trigger) { goto next_iter; } /* * Handle races against vnode allocation. Filesystems lock the * vnode some time after it gets returned from getnewvnode, * despite type and hold count being manipulated earlier. * Resorting to checking v_mount restores guarantees present * before the global list was reworked to contain all vnodes. */ if (!VI_TRYLOCK(vp)) goto next_iter; if (__predict_false(vp->v_type == VBAD || vp->v_type == VNON)) { VI_UNLOCK(vp); goto next_iter; } if (vp->v_mount == NULL) { VI_UNLOCK(vp); goto next_iter; } vholdl(vp); VI_UNLOCK(vp); TAILQ_REMOVE(&vnode_list, mvp, v_vnodelist); TAILQ_INSERT_AFTER(&vnode_list, vp, mvp, v_vnodelist); mtx_unlock(&vnode_list_mtx); if (vn_start_write(vp, &mp, V_NOWAIT) != 0) { vdrop(vp); goto next_iter_unlocked; } if (VOP_LOCK(vp, LK_EXCLUSIVE|LK_NOWAIT) != 0) { vdrop(vp); vn_finished_write(mp); goto next_iter_unlocked; } VI_LOCK(vp); if (vp->v_usecount > 0 || (!reclaim_nc_src && !LIST_EMPTY(&vp->v_cache_src)) || (vp->v_object != NULL && vp->v_object->handle == vp && vp->v_object->resident_page_count > trigger)) { VOP_UNLOCK(vp); vdropl(vp); vn_finished_write(mp); goto next_iter_unlocked; } counter_u64_add(recycles_count, 1); vgonel(vp); VOP_UNLOCK(vp); vdropl(vp); vn_finished_write(mp); done++; next_iter_unlocked: if (should_yield()) kern_yield(PRI_USER); mtx_lock(&vnode_list_mtx); goto restart; next_iter: MPASS(vp->v_type != VMARKER); if (!should_yield()) continue; TAILQ_REMOVE(&vnode_list, mvp, v_vnodelist); TAILQ_INSERT_AFTER(&vnode_list, vp, mvp, v_vnodelist); mtx_unlock(&vnode_list_mtx); kern_yield(PRI_USER); mtx_lock(&vnode_list_mtx); goto restart; } if (done == 0 && !retried) { TAILQ_REMOVE(&vnode_list, mvp, v_vnodelist); TAILQ_INSERT_HEAD(&vnode_list, mvp, v_vnodelist); retried = true; goto restart; } return (done); } static int max_vnlru_free = 10000; /* limit on vnode free requests per call */ SYSCTL_INT(_debug, OID_AUTO, max_vnlru_free, CTLFLAG_RW, &max_vnlru_free, 0, "limit on vnode free requests per call to the vnlru_free routine"); /* * Attempt to reduce the free list by the requested amount. */ static int vnlru_free_impl(int count, struct vfsops *mnt_op, struct vnode *mvp) { struct vnode *vp; struct mount *mp; int ocount; mtx_assert(&vnode_list_mtx, MA_OWNED); if (count > max_vnlru_free) count = max_vnlru_free; ocount = count; vp = mvp; for (;;) { if (count == 0) { break; } vp = TAILQ_NEXT(vp, v_vnodelist); if (__predict_false(vp == NULL)) { TAILQ_REMOVE(&vnode_list, mvp, v_vnodelist); TAILQ_INSERT_TAIL(&vnode_list, mvp, v_vnodelist); break; } if (__predict_false(vp->v_type == VMARKER)) continue; if (vp->v_holdcnt > 0) continue; /* * Don't recycle if our vnode is from different type * of mount point. Note that mp is type-safe, the * check does not reach unmapped address even if * vnode is reclaimed. */ if (mnt_op != NULL && (mp = vp->v_mount) != NULL && mp->mnt_op != mnt_op) { continue; } if (__predict_false(vp->v_type == VBAD || vp->v_type == VNON)) { continue; } if (!vhold_recycle_free(vp)) continue; TAILQ_REMOVE(&vnode_list, mvp, v_vnodelist); TAILQ_INSERT_AFTER(&vnode_list, vp, mvp, v_vnodelist); mtx_unlock(&vnode_list_mtx); if (vtryrecycle(vp) == 0) count--; mtx_lock(&vnode_list_mtx); vp = mvp; } return (ocount - count); } static int vnlru_free_locked(int count) { mtx_assert(&vnode_list_mtx, MA_OWNED); return (vnlru_free_impl(count, NULL, vnode_list_free_marker)); } void vnlru_free_vfsops(int count, struct vfsops *mnt_op, struct vnode *mvp) { MPASS(mnt_op != NULL); MPASS(mvp != NULL); VNPASS(mvp->v_type == VMARKER, mvp); mtx_lock(&vnode_list_mtx); vnlru_free_impl(count, mnt_op, mvp); mtx_unlock(&vnode_list_mtx); } struct vnode * vnlru_alloc_marker(void) { struct vnode *mvp; mvp = vn_alloc_marker(NULL); mtx_lock(&vnode_list_mtx); TAILQ_INSERT_BEFORE(vnode_list_free_marker, mvp, v_vnodelist); mtx_unlock(&vnode_list_mtx); return (mvp); } void vnlru_free_marker(struct vnode *mvp) { mtx_lock(&vnode_list_mtx); TAILQ_REMOVE(&vnode_list, mvp, v_vnodelist); mtx_unlock(&vnode_list_mtx); vn_free_marker(mvp); } static void vnlru_recalc(void) { mtx_assert(&vnode_list_mtx, MA_OWNED); gapvnodes = imax(desiredvnodes - wantfreevnodes, 100); vhiwat = gapvnodes / 11; /* 9% -- just under the 10% in vlrureclaim() */ vlowat = vhiwat / 2; } /* * Attempt to recycle vnodes in a context that is always safe to block. * Calling vlrurecycle() from the bowels of filesystem code has some * interesting deadlock problems. */ static struct proc *vnlruproc; static int vnlruproc_sig; /* * The main freevnodes counter is only updated when threads requeue their vnode * batches. CPUs are conditionally walked to compute a more accurate total. * * Limit how much of a slop are we willing to tolerate. Note: the actual value * at any given moment can still exceed slop, but it should not be by significant * margin in practice. */ #define VNLRU_FREEVNODES_SLOP 128 static __inline void vfs_freevnodes_inc(void) { struct vdbatch *vd; critical_enter(); vd = DPCPU_PTR(vd); vd->freevnodes++; critical_exit(); } static __inline void vfs_freevnodes_dec(void) { struct vdbatch *vd; critical_enter(); vd = DPCPU_PTR(vd); vd->freevnodes--; critical_exit(); } static u_long vnlru_read_freevnodes(void) { struct vdbatch *vd; long slop; int cpu; mtx_assert(&vnode_list_mtx, MA_OWNED); if (freevnodes > freevnodes_old) slop = freevnodes - freevnodes_old; else slop = freevnodes_old - freevnodes; if (slop < VNLRU_FREEVNODES_SLOP) return (freevnodes >= 0 ? freevnodes : 0); freevnodes_old = freevnodes; CPU_FOREACH(cpu) { vd = DPCPU_ID_PTR((cpu), vd); freevnodes_old += vd->freevnodes; } return (freevnodes_old >= 0 ? freevnodes_old : 0); } static bool vnlru_under(u_long rnumvnodes, u_long limit) { u_long rfreevnodes, space; if (__predict_false(rnumvnodes > desiredvnodes)) return (true); space = desiredvnodes - rnumvnodes; if (space < limit) { rfreevnodes = vnlru_read_freevnodes(); if (rfreevnodes > wantfreevnodes) space += rfreevnodes - wantfreevnodes; } return (space < limit); } static bool vnlru_under_unlocked(u_long rnumvnodes, u_long limit) { long rfreevnodes, space; if (__predict_false(rnumvnodes > desiredvnodes)) return (true); space = desiredvnodes - rnumvnodes; if (space < limit) { rfreevnodes = atomic_load_long(&freevnodes); if (rfreevnodes > wantfreevnodes) space += rfreevnodes - wantfreevnodes; } return (space < limit); } static void vnlru_kick(void) { mtx_assert(&vnode_list_mtx, MA_OWNED); if (vnlruproc_sig == 0) { vnlruproc_sig = 1; wakeup(vnlruproc); } } static void vnlru_proc(void) { u_long rnumvnodes, rfreevnodes, target; unsigned long onumvnodes; int done, force, trigger, usevnodes; bool reclaim_nc_src, want_reread; EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, vnlruproc, SHUTDOWN_PRI_FIRST); force = 0; want_reread = false; for (;;) { kproc_suspend_check(vnlruproc); mtx_lock(&vnode_list_mtx); rnumvnodes = atomic_load_long(&numvnodes); if (want_reread) { force = vnlru_under(numvnodes, vhiwat) ? 1 : 0; want_reread = false; } /* * If numvnodes is too large (due to desiredvnodes being * adjusted using its sysctl, or emergency growth), first * try to reduce it by discarding from the free list. */ if (rnumvnodes > desiredvnodes) { vnlru_free_locked(rnumvnodes - desiredvnodes); rnumvnodes = atomic_load_long(&numvnodes); } /* * Sleep if the vnode cache is in a good state. This is * when it is not over-full and has space for about a 4% * or 9% expansion (by growing its size or inexcessively * reducing its free list). Otherwise, try to reclaim * space for a 10% expansion. */ if (vstir && force == 0) { force = 1; vstir = 0; } if (force == 0 && !vnlru_under(rnumvnodes, vlowat)) { vnlruproc_sig = 0; wakeup(&vnlruproc_sig); msleep(vnlruproc, &vnode_list_mtx, PVFS|PDROP, "vlruwt", hz); continue; } rfreevnodes = vnlru_read_freevnodes(); onumvnodes = rnumvnodes; /* * Calculate parameters for recycling. These are the same * throughout the loop to give some semblance of fairness. * The trigger point is to avoid recycling vnodes with lots * of resident pages. We aren't trying to free memory; we * are trying to recycle or at least free vnodes. */ if (rnumvnodes <= desiredvnodes) usevnodes = rnumvnodes - rfreevnodes; else usevnodes = rnumvnodes; if (usevnodes <= 0) usevnodes = 1; /* * The trigger value is is chosen to give a conservatively * large value to ensure that it alone doesn't prevent * making progress. The value can easily be so large that * it is effectively infinite in some congested and * misconfigured cases, and this is necessary. Normally * it is about 8 to 100 (pages), which is quite large. */ trigger = vm_cnt.v_page_count * 2 / usevnodes; if (force < 2) trigger = vsmalltrigger; reclaim_nc_src = force >= 3; target = rnumvnodes * (int64_t)gapvnodes / imax(desiredvnodes, 1); target = target / 10 + 1; done = vlrureclaim(reclaim_nc_src, trigger, target); mtx_unlock(&vnode_list_mtx); if (onumvnodes > desiredvnodes && numvnodes <= desiredvnodes) uma_reclaim(UMA_RECLAIM_DRAIN); if (done == 0) { if (force == 0 || force == 1) { force = 2; continue; } if (force == 2) { force = 3; continue; } want_reread = true; force = 0; vnlru_nowhere++; tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3); } else { want_reread = true; kern_yield(PRI_USER); } } } static struct kproc_desc vnlru_kp = { "vnlru", vnlru_proc, &vnlruproc }; SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp); /* * Routines having to do with the management of the vnode table. */ /* * Try to recycle a freed vnode. We abort if anyone picks up a reference * before we actually vgone(). This function must be called with the vnode * held to prevent the vnode from being returned to the free list midway * through vgone(). */ static int vtryrecycle(struct vnode *vp) { struct mount *vnmp; CTR2(KTR_VFS, "%s: vp %p", __func__, vp); VNASSERT(vp->v_holdcnt, vp, ("vtryrecycle: Recycling vp %p without a reference.", vp)); /* * This vnode may found and locked via some other list, if so we * can't recycle it yet. */ if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { CTR2(KTR_VFS, "%s: impossible to recycle, vp %p lock is already held", __func__, vp); vdrop(vp); return (EWOULDBLOCK); } /* * Don't recycle if its filesystem is being suspended. */ if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) { VOP_UNLOCK(vp); CTR2(KTR_VFS, "%s: impossible to recycle, cannot start the write for %p", __func__, vp); vdrop(vp); return (EBUSY); } /* * If we got this far, we need to acquire the interlock and see if * anyone picked up this vnode from another list. If not, we will * mark it with DOOMED via vgonel() so that anyone who does find it * will skip over it. */ VI_LOCK(vp); if (vp->v_usecount) { VOP_UNLOCK(vp); vdropl(vp); vn_finished_write(vnmp); CTR2(KTR_VFS, "%s: impossible to recycle, %p is already referenced", __func__, vp); return (EBUSY); } if (!VN_IS_DOOMED(vp)) { counter_u64_add(recycles_free_count, 1); vgonel(vp); } VOP_UNLOCK(vp); vdropl(vp); vn_finished_write(vnmp); return (0); } /* * Allocate a new vnode. * * The operation never returns an error. Returning an error was disabled * in r145385 (dated 2005) with the following comment: * * XXX Not all VFS_VGET/ffs_vget callers check returns. * * Given the age of this commit (almost 15 years at the time of writing this * comment) restoring the ability to fail requires a significant audit of * all codepaths. * * The routine can try to free a vnode or stall for up to 1 second waiting for * vnlru to clear things up, but ultimately always performs a M_WAITOK allocation. */ static u_long vn_alloc_cyclecount; static struct vnode * __noinline vn_alloc_hard(struct mount *mp) { u_long rnumvnodes, rfreevnodes; mtx_lock(&vnode_list_mtx); rnumvnodes = atomic_load_long(&numvnodes); if (rnumvnodes + 1 < desiredvnodes) { vn_alloc_cyclecount = 0; goto alloc; } rfreevnodes = vnlru_read_freevnodes(); if (vn_alloc_cyclecount++ >= rfreevnodes) { vn_alloc_cyclecount = 0; vstir = 1; } /* * Grow the vnode cache if it will not be above its target max * after growing. Otherwise, if the free list is nonempty, try * to reclaim 1 item from it before growing the cache (possibly * above its target max if the reclamation failed or is delayed). * Otherwise, wait for some space. In all cases, schedule * vnlru_proc() if we are getting short of space. The watermarks * should be chosen so that we never wait or even reclaim from * the free list to below its target minimum. */ if (vnlru_free_locked(1) > 0) goto alloc; if (mp == NULL || (mp->mnt_kern_flag & MNTK_SUSPEND) == 0) { /* * Wait for space for a new vnode. */ vnlru_kick(); msleep(&vnlruproc_sig, &vnode_list_mtx, PVFS, "vlruwk", hz); if (atomic_load_long(&numvnodes) + 1 > desiredvnodes && vnlru_read_freevnodes() > 1) vnlru_free_locked(1); } alloc: rnumvnodes = atomic_fetchadd_long(&numvnodes, 1) + 1; if (vnlru_under(rnumvnodes, vlowat)) vnlru_kick(); mtx_unlock(&vnode_list_mtx); return (uma_zalloc_smr(vnode_zone, M_WAITOK)); } static struct vnode * vn_alloc(struct mount *mp) { u_long rnumvnodes; if (__predict_false(vn_alloc_cyclecount != 0)) return (vn_alloc_hard(mp)); rnumvnodes = atomic_fetchadd_long(&numvnodes, 1) + 1; if (__predict_false(vnlru_under_unlocked(rnumvnodes, vlowat))) { atomic_subtract_long(&numvnodes, 1); return (vn_alloc_hard(mp)); } return (uma_zalloc_smr(vnode_zone, M_WAITOK)); } static void vn_free(struct vnode *vp) { atomic_subtract_long(&numvnodes, 1); uma_zfree_smr(vnode_zone, vp); } /* * Return the next vnode from the free list. */ int getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops, struct vnode **vpp) { struct vnode *vp; struct thread *td; struct lock_object *lo; CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag); KASSERT(vops->registered, ("%s: not registered vector op %p\n", __func__, vops)); td = curthread; if (td->td_vp_reserved != NULL) { vp = td->td_vp_reserved; td->td_vp_reserved = NULL; } else { vp = vn_alloc(mp); } counter_u64_add(vnodes_created, 1); /* * Locks are given the generic name "vnode" when created. * Follow the historic practice of using the filesystem * name when they allocated, e.g., "zfs", "ufs", "nfs, etc. * * Locks live in a witness group keyed on their name. Thus, * when a lock is renamed, it must also move from the witness * group of its old name to the witness group of its new name. * * The change only needs to be made when the vnode moves * from one filesystem type to another. We ensure that each * filesystem use a single static name pointer for its tag so * that we can compare pointers rather than doing a strcmp(). */ lo = &vp->v_vnlock->lock_object; #ifdef WITNESS if (lo->lo_name != tag) { #endif lo->lo_name = tag; #ifdef WITNESS WITNESS_DESTROY(lo); WITNESS_INIT(lo, tag); } #endif /* * By default, don't allow shared locks unless filesystems opt-in. */ vp->v_vnlock->lock_object.lo_flags |= LK_NOSHARE; /* * Finalize various vnode identity bits. */ KASSERT(vp->v_object == NULL, ("stale v_object %p", vp)); KASSERT(vp->v_lockf == NULL, ("stale v_lockf %p", vp)); KASSERT(vp->v_pollinfo == NULL, ("stale v_pollinfo %p", vp)); vp->v_type = VNON; vp->v_op = vops; vp->v_irflag = 0; v_init_counters(vp); vn_seqc_init(vp); vp->v_bufobj.bo_ops = &buf_ops_bio; #ifdef DIAGNOSTIC if (mp == NULL && vops != &dead_vnodeops) printf("NULL mp in getnewvnode(9), tag %s\n", tag); #endif #ifdef MAC mac_vnode_init(vp); if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0) mac_vnode_associate_singlelabel(mp, vp); #endif if (mp != NULL) { vp->v_bufobj.bo_bsize = mp->mnt_stat.f_iosize; if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0) vp->v_vflag |= VV_NOKNOTE; } /* * For the filesystems which do not use vfs_hash_insert(), * still initialize v_hash to have vfs_hash_index() useful. * E.g., nullfs uses vfs_hash_index() on the lower vnode for * its own hashing. */ vp->v_hash = (uintptr_t)vp >> vnsz2log; *vpp = vp; return (0); } void getnewvnode_reserve(void) { struct thread *td; td = curthread; MPASS(td->td_vp_reserved == NULL); td->td_vp_reserved = vn_alloc(NULL); } void getnewvnode_drop_reserve(void) { struct thread *td; td = curthread; if (td->td_vp_reserved != NULL) { vn_free(td->td_vp_reserved); td->td_vp_reserved = NULL; } } static void __noinline freevnode(struct vnode *vp) { struct bufobj *bo; /* * The vnode has been marked for destruction, so free it. * * The vnode will be returned to the zone where it will * normally remain until it is needed for another vnode. We * need to cleanup (or verify that the cleanup has already * been done) any residual data left from its current use * so as not to contaminate the freshly allocated vnode. */ CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp); /* * Paired with vgone. */ vn_seqc_write_end_free(vp); bo = &vp->v_bufobj; VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't")); VNPASS(vp->v_holdcnt == VHOLD_NO_SMR, vp); VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count")); VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count")); VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's")); VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0")); VNASSERT(pctrie_is_empty(&bo->bo_clean.bv_root), vp, ("clean blk trie not empty")); VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0")); VNASSERT(pctrie_is_empty(&bo->bo_dirty.bv_root), vp, ("dirty blk trie not empty")); VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst")); VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src")); VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for ..")); VNASSERT(TAILQ_EMPTY(&vp->v_rl.rl_waiters), vp, ("Dangling rangelock waiters")); VNASSERT((vp->v_iflag & (VI_DOINGINACT | VI_OWEINACT)) == 0, vp, ("Leaked inactivation")); VI_UNLOCK(vp); #ifdef MAC mac_vnode_destroy(vp); #endif if (vp->v_pollinfo != NULL) { destroy_vpollinfo(vp->v_pollinfo); vp->v_pollinfo = NULL; } vp->v_mountedhere = NULL; vp->v_unpcb = NULL; vp->v_rdev = NULL; vp->v_fifoinfo = NULL; vp->v_iflag = 0; vp->v_vflag = 0; bo->bo_flag = 0; vn_free(vp); } /* * Delete from old mount point vnode list, if on one. */ static void delmntque(struct vnode *vp) { struct mount *mp; VNPASS((vp->v_mflag & VMP_LAZYLIST) == 0, vp); mp = vp->v_mount; if (mp == NULL) return; MNT_ILOCK(mp); VI_LOCK(vp); vp->v_mount = NULL; VI_UNLOCK(vp); VNASSERT(mp->mnt_nvnodelistsize > 0, vp, ("bad mount point vnode list size")); TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); mp->mnt_nvnodelistsize--; MNT_REL(mp); MNT_IUNLOCK(mp); } static void insmntque_stddtr(struct vnode *vp, void *dtr_arg) { vp->v_data = NULL; vp->v_op = &dead_vnodeops; vgone(vp); vput(vp); } /* * Insert into list of vnodes for the new mount point, if available. */ int insmntque1(struct vnode *vp, struct mount *mp, void (*dtr)(struct vnode *, void *), void *dtr_arg) { KASSERT(vp->v_mount == NULL, ("insmntque: vnode already on per mount vnode list")); VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)")); ASSERT_VOP_ELOCKED(vp, "insmntque: non-locked vp"); /* * We acquire the vnode interlock early to ensure that the * vnode cannot be recycled by another process releasing a * holdcnt on it before we get it on both the vnode list * and the active vnode list. The mount mutex protects only * manipulation of the vnode list and the vnode freelist * mutex protects only manipulation of the active vnode list. * Hence the need to hold the vnode interlock throughout. */ MNT_ILOCK(mp); VI_LOCK(vp); if (((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 && ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 || mp->mnt_nvnodelistsize == 0)) && (vp->v_vflag & VV_FORCEINSMQ) == 0) { VI_UNLOCK(vp); MNT_IUNLOCK(mp); if (dtr != NULL) dtr(vp, dtr_arg); return (EBUSY); } vp->v_mount = mp; MNT_REF(mp); TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); VNASSERT(mp->mnt_nvnodelistsize >= 0, vp, ("neg mount point vnode list size")); mp->mnt_nvnodelistsize++; VI_UNLOCK(vp); MNT_IUNLOCK(mp); return (0); } int insmntque(struct vnode *vp, struct mount *mp) { return (insmntque1(vp, mp, insmntque_stddtr, NULL)); } /* * Flush out and invalidate all buffers associated with a bufobj * Called with the underlying object locked. */ int bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo) { int error; BO_LOCK(bo); if (flags & V_SAVE) { error = bufobj_wwait(bo, slpflag, slptimeo); if (error) { BO_UNLOCK(bo); return (error); } if (bo->bo_dirty.bv_cnt > 0) { BO_UNLOCK(bo); do { error = BO_SYNC(bo, MNT_WAIT); } while (error == ERELOOKUP); if (error != 0) return (error); BO_LOCK(bo); if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) { BO_UNLOCK(bo); return (EBUSY); } } } /* * If you alter this loop please notice that interlock is dropped and * reacquired in flushbuflist. Special care is needed to ensure that * no race conditions occur from this. */ do { error = flushbuflist(&bo->bo_clean, flags, bo, slpflag, slptimeo); if (error == 0 && !(flags & V_CLEANONLY)) error = flushbuflist(&bo->bo_dirty, flags, bo, slpflag, slptimeo); if (error != 0 && error != EAGAIN) { BO_UNLOCK(bo); return (error); } } while (error != 0); /* * Wait for I/O to complete. XXX needs cleaning up. The vnode can * have write I/O in-progress but if there is a VM object then the * VM object can also have read-I/O in-progress. */ do { bufobj_wwait(bo, 0, 0); if ((flags & V_VMIO) == 0 && bo->bo_object != NULL) { BO_UNLOCK(bo); vm_object_pip_wait_unlocked(bo->bo_object, "bovlbx"); BO_LOCK(bo); } } while (bo->bo_numoutput > 0); BO_UNLOCK(bo); /* * Destroy the copy in the VM cache, too. */ if (bo->bo_object != NULL && (flags & (V_ALT | V_NORMAL | V_CLEANONLY | V_VMIO)) == 0) { VM_OBJECT_WLOCK(bo->bo_object); vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ? OBJPR_CLEANONLY : 0); VM_OBJECT_WUNLOCK(bo->bo_object); } #ifdef INVARIANTS BO_LOCK(bo); if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY | V_VMIO | V_ALLOWCLEAN)) == 0 && (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0)) panic("vinvalbuf: flush failed"); if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY | V_VMIO)) == 0 && bo->bo_dirty.bv_cnt > 0) panic("vinvalbuf: flush dirty failed"); BO_UNLOCK(bo); #endif return (0); } /* * Flush out and invalidate all buffers associated with a vnode. * Called with the underlying object locked. */ int vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo) { CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags); ASSERT_VOP_LOCKED(vp, "vinvalbuf"); if (vp->v_object != NULL && vp->v_object->handle != vp) return (0); return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo)); } /* * Flush out buffers on the specified list. * */ static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag, int slptimeo) { struct buf *bp, *nbp; int retval, error; daddr_t lblkno; b_xflags_t xflags; ASSERT_BO_WLOCKED(bo); retval = 0; TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) { /* * If we are flushing both V_NORMAL and V_ALT buffers then * do not skip any buffers. If we are flushing only V_NORMAL * buffers then skip buffers marked as BX_ALTDATA. If we are * flushing only V_ALT buffers then skip buffers not marked * as BX_ALTDATA. */ if (((flags & (V_NORMAL | V_ALT)) != (V_NORMAL | V_ALT)) && (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA) != 0) || ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0))) { continue; } if (nbp != NULL) { lblkno = nbp->b_lblkno; xflags = nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN); } retval = EAGAIN; error = BUF_TIMELOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo), "flushbuf", slpflag, slptimeo); if (error) { BO_LOCK(bo); return (error != ENOLCK ? error : EAGAIN); } KASSERT(bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); /* * XXX Since there are no node locks for NFS, I * believe there is a slight chance that a delayed * write will occur while sleeping just above, so * check for it. */ if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && (flags & V_SAVE)) { bremfree(bp); bp->b_flags |= B_ASYNC; bwrite(bp); BO_LOCK(bo); return (EAGAIN); /* XXX: why not loop ? */ } bremfree(bp); bp->b_flags |= (B_INVAL | B_RELBUF); bp->b_flags &= ~B_ASYNC; brelse(bp); BO_LOCK(bo); if (nbp == NULL) break; nbp = gbincore(bo, lblkno); if (nbp == NULL || (nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) != xflags) break; /* nbp invalid */ } return (retval); } int bnoreuselist(struct bufv *bufv, struct bufobj *bo, daddr_t startn, daddr_t endn) { struct buf *bp; int error; daddr_t lblkno; ASSERT_BO_LOCKED(bo); for (lblkno = startn;;) { again: bp = BUF_PCTRIE_LOOKUP_GE(&bufv->bv_root, lblkno); if (bp == NULL || bp->b_lblkno >= endn || bp->b_lblkno < startn) break; error = BUF_TIMELOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo), "brlsfl", 0, 0); if (error != 0) { BO_RLOCK(bo); if (error == ENOLCK) goto again; return (error); } KASSERT(bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); lblkno = bp->b_lblkno + 1; if ((bp->b_flags & B_MANAGED) == 0) bremfree(bp); bp->b_flags |= B_RELBUF; /* * In the VMIO case, use the B_NOREUSE flag to hint that the * pages backing each buffer in the range are unlikely to be * reused. Dirty buffers will have the hint applied once * they've been written. */ if ((bp->b_flags & B_VMIO) != 0) bp->b_flags |= B_NOREUSE; brelse(bp); BO_RLOCK(bo); } return (0); } /* * Truncate a file's buffer and pages to a specified length. This * is in lieu of the old vinvalbuf mechanism, which performed unneeded * sync activity. */ int vtruncbuf(struct vnode *vp, off_t length, int blksize) { struct buf *bp, *nbp; struct bufobj *bo; daddr_t startlbn; CTR4(KTR_VFS, "%s: vp %p with block %d:%ju", __func__, vp, blksize, (uintmax_t)length); /* * Round up to the *next* lbn. */ startlbn = howmany(length, blksize); ASSERT_VOP_LOCKED(vp, "vtruncbuf"); bo = &vp->v_bufobj; restart_unlocked: BO_LOCK(bo); while (v_inval_buf_range_locked(vp, bo, startlbn, INT64_MAX) == EAGAIN) ; if (length > 0) { restartsync: TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { if (bp->b_lblkno > 0) continue; /* * Since we hold the vnode lock this should only * fail if we're racing with the buf daemon. */ if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) goto restart_unlocked; VNASSERT((bp->b_flags & B_DELWRI), vp, ("buf(%p) on dirty queue without DELWRI", bp)); bremfree(bp); bawrite(bp); BO_LOCK(bo); goto restartsync; } } bufobj_wwait(bo, 0, 0); BO_UNLOCK(bo); vnode_pager_setsize(vp, length); return (0); } /* * Invalidate the cached pages of a file's buffer within the range of block * numbers [startlbn, endlbn). */ void v_inval_buf_range(struct vnode *vp, daddr_t startlbn, daddr_t endlbn, int blksize) { struct bufobj *bo; off_t start, end; ASSERT_VOP_LOCKED(vp, "v_inval_buf_range"); start = blksize * startlbn; end = blksize * endlbn; bo = &vp->v_bufobj; BO_LOCK(bo); MPASS(blksize == bo->bo_bsize); while (v_inval_buf_range_locked(vp, bo, startlbn, endlbn) == EAGAIN) ; BO_UNLOCK(bo); vn_pages_remove(vp, OFF_TO_IDX(start), OFF_TO_IDX(end + PAGE_SIZE - 1)); } static int v_inval_buf_range_locked(struct vnode *vp, struct bufobj *bo, daddr_t startlbn, daddr_t endlbn) { struct buf *bp, *nbp; bool anyfreed; ASSERT_VOP_LOCKED(vp, "v_inval_buf_range_locked"); ASSERT_BO_LOCKED(bo); do { anyfreed = false; TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) { if (bp->b_lblkno < startlbn || bp->b_lblkno >= endlbn) continue; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) { BO_LOCK(bo); return (EAGAIN); } bremfree(bp); bp->b_flags |= B_INVAL | B_RELBUF; bp->b_flags &= ~B_ASYNC; brelse(bp); anyfreed = true; BO_LOCK(bo); if (nbp != NULL && (((nbp->b_xflags & BX_VNCLEAN) == 0) || nbp->b_vp != vp || (nbp->b_flags & B_DELWRI) != 0)) return (EAGAIN); } TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { if (bp->b_lblkno < startlbn || bp->b_lblkno >= endlbn) continue; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) { BO_LOCK(bo); return (EAGAIN); } bremfree(bp); bp->b_flags |= B_INVAL | B_RELBUF; bp->b_flags &= ~B_ASYNC; brelse(bp); anyfreed = true; BO_LOCK(bo); if (nbp != NULL && (((nbp->b_xflags & BX_VNDIRTY) == 0) || (nbp->b_vp != vp) || (nbp->b_flags & B_DELWRI) == 0)) return (EAGAIN); } } while (anyfreed); return (0); } static void buf_vlist_remove(struct buf *bp) { struct bufv *bv; b_xflags_t flags; flags = bp->b_xflags; KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); ASSERT_BO_WLOCKED(bp->b_bufobj); KASSERT((flags & (BX_VNDIRTY | BX_VNCLEAN)) != 0 && (flags & (BX_VNDIRTY | BX_VNCLEAN)) != (BX_VNDIRTY | BX_VNCLEAN), ("%s: buffer %p has invalid queue state", __func__, bp)); if ((flags & BX_VNDIRTY) != 0) bv = &bp->b_bufobj->bo_dirty; else bv = &bp->b_bufobj->bo_clean; BUF_PCTRIE_REMOVE(&bv->bv_root, bp->b_lblkno); TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs); bv->bv_cnt--; bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); } /* * Add the buffer to the sorted clean or dirty block list. * * NOTE: xflags is passed as a constant, optimizing this inline function! */ static void buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags) { struct bufv *bv; struct buf *n; int error; ASSERT_BO_WLOCKED(bo); KASSERT((bo->bo_flag & BO_NOBUFS) == 0, ("buf_vlist_add: bo %p does not allow bufs", bo)); KASSERT((xflags & BX_VNDIRTY) == 0 || (bo->bo_flag & BO_DEAD) == 0, ("dead bo %p", bo)); KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags)); bp->b_xflags |= xflags; if (xflags & BX_VNDIRTY) bv = &bo->bo_dirty; else bv = &bo->bo_clean; /* * Keep the list ordered. Optimize empty list insertion. Assume * we tend to grow at the tail so lookup_le should usually be cheaper * than _ge. */ if (bv->bv_cnt == 0 || bp->b_lblkno > TAILQ_LAST(&bv->bv_hd, buflists)->b_lblkno) TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs); else if ((n = BUF_PCTRIE_LOOKUP_LE(&bv->bv_root, bp->b_lblkno)) == NULL) TAILQ_INSERT_HEAD(&bv->bv_hd, bp, b_bobufs); else TAILQ_INSERT_AFTER(&bv->bv_hd, n, bp, b_bobufs); error = BUF_PCTRIE_INSERT(&bv->bv_root, bp); if (error) panic("buf_vlist_add: Preallocated nodes insufficient."); bv->bv_cnt++; } /* * Look up a buffer using the buffer tries. */ struct buf * gbincore(struct bufobj *bo, daddr_t lblkno) { struct buf *bp; ASSERT_BO_LOCKED(bo); bp = BUF_PCTRIE_LOOKUP(&bo->bo_clean.bv_root, lblkno); if (bp != NULL) return (bp); return (BUF_PCTRIE_LOOKUP(&bo->bo_dirty.bv_root, lblkno)); } /* * Look up a buf using the buffer tries, without the bufobj lock. This relies * on SMR for safe lookup, and bufs being in a no-free zone to provide type * stability of the result. Like other lockless lookups, the found buf may * already be invalid by the time this function returns. */ struct buf * gbincore_unlocked(struct bufobj *bo, daddr_t lblkno) { struct buf *bp; ASSERT_BO_UNLOCKED(bo); bp = BUF_PCTRIE_LOOKUP_UNLOCKED(&bo->bo_clean.bv_root, lblkno); if (bp != NULL) return (bp); return (BUF_PCTRIE_LOOKUP_UNLOCKED(&bo->bo_dirty.bv_root, lblkno)); } /* * Associate a buffer with a vnode. */ void bgetvp(struct vnode *vp, struct buf *bp) { struct bufobj *bo; bo = &vp->v_bufobj; ASSERT_BO_WLOCKED(bo); VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free")); CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags); VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp, ("bgetvp: bp already attached! %p", bp)); vhold(vp); bp->b_vp = vp; bp->b_bufobj = bo; /* * Insert onto list for new vnode. */ buf_vlist_add(bp, bo, BX_VNCLEAN); } /* * Disassociate a buffer from a vnode. */ void brelvp(struct buf *bp) { struct bufobj *bo; struct vnode *vp; CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); /* * Delete from old vnode list, if on one. */ vp = bp->b_vp; /* XXX */ bo = bp->b_bufobj; BO_LOCK(bo); buf_vlist_remove(bp); if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { bo->bo_flag &= ~BO_ONWORKLST; mtx_lock(&sync_mtx); LIST_REMOVE(bo, bo_synclist); syncer_worklist_len--; mtx_unlock(&sync_mtx); } bp->b_vp = NULL; bp->b_bufobj = NULL; BO_UNLOCK(bo); vdrop(vp); } /* * Add an item to the syncer work queue. */ static void vn_syncer_add_to_worklist(struct bufobj *bo, int delay) { int slot; ASSERT_BO_WLOCKED(bo); mtx_lock(&sync_mtx); if (bo->bo_flag & BO_ONWORKLST) LIST_REMOVE(bo, bo_synclist); else { bo->bo_flag |= BO_ONWORKLST; syncer_worklist_len++; } if (delay > syncer_maxdelay - 2) delay = syncer_maxdelay - 2; slot = (syncer_delayno + delay) & syncer_mask; LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist); mtx_unlock(&sync_mtx); } static int sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS) { int error, len; mtx_lock(&sync_mtx); len = syncer_worklist_len - sync_vnode_count; mtx_unlock(&sync_mtx); error = SYSCTL_OUT(req, &len, sizeof(len)); return (error); } SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_MPSAFE| CTLFLAG_RD, NULL, 0, sysctl_vfs_worklist_len, "I", "Syncer thread worklist length"); static struct proc *updateproc; static void sched_sync(void); static struct kproc_desc up_kp = { "syncer", sched_sync, &updateproc }; SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp); static int sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td) { struct vnode *vp; struct mount *mp; *bo = LIST_FIRST(slp); if (*bo == NULL) return (0); vp = bo2vnode(*bo); if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0) return (1); /* * We use vhold in case the vnode does not * successfully sync. vhold prevents the vnode from * going away when we unlock the sync_mtx so that * we can acquire the vnode interlock. */ vholdl(vp); mtx_unlock(&sync_mtx); VI_UNLOCK(vp); if (vn_start_write(vp, &mp, V_NOWAIT) != 0) { vdrop(vp); mtx_lock(&sync_mtx); return (*bo == LIST_FIRST(slp)); } vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); (void) VOP_FSYNC(vp, MNT_LAZY, td); VOP_UNLOCK(vp); vn_finished_write(mp); BO_LOCK(*bo); if (((*bo)->bo_flag & BO_ONWORKLST) != 0) { /* * Put us back on the worklist. The worklist * routine will remove us from our current * position and then add us back in at a later * position. */ vn_syncer_add_to_worklist(*bo, syncdelay); } BO_UNLOCK(*bo); vdrop(vp); mtx_lock(&sync_mtx); return (0); } static int first_printf = 1; /* * System filesystem synchronizer daemon. */ static void sched_sync(void) { struct synclist *next, *slp; struct bufobj *bo; long starttime; struct thread *td = curthread; int last_work_seen; int net_worklist_len; int syncer_final_iter; int error; last_work_seen = 0; syncer_final_iter = 0; syncer_state = SYNCER_RUNNING; starttime = time_uptime; td->td_pflags |= TDP_NORUNNINGBUF; EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc, SHUTDOWN_PRI_LAST); mtx_lock(&sync_mtx); for (;;) { if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter == 0) { mtx_unlock(&sync_mtx); kproc_suspend_check(td->td_proc); mtx_lock(&sync_mtx); } net_worklist_len = syncer_worklist_len - sync_vnode_count; if (syncer_state != SYNCER_RUNNING && starttime != time_uptime) { if (first_printf) { printf("\nSyncing disks, vnodes remaining... "); first_printf = 0; } printf("%d ", net_worklist_len); } starttime = time_uptime; /* * Push files whose dirty time has expired. Be careful * of interrupt race on slp queue. * * Skip over empty worklist slots when shutting down. */ do { slp = &syncer_workitem_pending[syncer_delayno]; syncer_delayno += 1; if (syncer_delayno == syncer_maxdelay) syncer_delayno = 0; next = &syncer_workitem_pending[syncer_delayno]; /* * If the worklist has wrapped since the * it was emptied of all but syncer vnodes, * switch to the FINAL_DELAY state and run * for one more second. */ if (syncer_state == SYNCER_SHUTTING_DOWN && net_worklist_len == 0 && last_work_seen == syncer_delayno) { syncer_state = SYNCER_FINAL_DELAY; syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP; } } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) && syncer_worklist_len > 0); /* * Keep track of the last time there was anything * on the worklist other than syncer vnodes. * Return to the SHUTTING_DOWN state if any * new work appears. */ if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING) last_work_seen = syncer_delayno; if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY) syncer_state = SYNCER_SHUTTING_DOWN; while (!LIST_EMPTY(slp)) { error = sync_vnode(slp, &bo, td); if (error == 1) { LIST_REMOVE(bo, bo_synclist); LIST_INSERT_HEAD(next, bo, bo_synclist); continue; } if (first_printf == 0) { /* * Drop the sync mutex, because some watchdog * drivers need to sleep while patting */ mtx_unlock(&sync_mtx); wdog_kern_pat(WD_LASTVAL); mtx_lock(&sync_mtx); } } if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0) syncer_final_iter--; /* * The variable rushjob allows the kernel to speed up the * processing of the filesystem syncer process. A rushjob * value of N tells the filesystem syncer to process the next * N seconds worth of work on its queue ASAP. Currently rushjob * is used by the soft update code to speed up the filesystem * syncer process when the incore state is getting so far * ahead of the disk that the kernel memory pool is being * threatened with exhaustion. */ if (rushjob > 0) { rushjob -= 1; continue; } /* * Just sleep for a short period of time between * iterations when shutting down to allow some I/O * to happen. * * If it has taken us less than a second to process the * current work, then wait. Otherwise start right over * again. We can still lose time if any single round * takes more than two seconds, but it does not really * matter as we are just trying to generally pace the * filesystem activity. */ if (syncer_state != SYNCER_RUNNING || time_uptime == starttime) { thread_lock(td); sched_prio(td, PPAUSE); thread_unlock(td); } if (syncer_state != SYNCER_RUNNING) cv_timedwait(&sync_wakeup, &sync_mtx, hz / SYNCER_SHUTDOWN_SPEEDUP); else if (time_uptime == starttime) cv_timedwait(&sync_wakeup, &sync_mtx, hz); } } /* * Request the syncer daemon to speed up its work. * We never push it to speed up more than half of its * normal turn time, otherwise it could take over the cpu. */ int speedup_syncer(void) { int ret = 0; mtx_lock(&sync_mtx); if (rushjob < syncdelay / 2) { rushjob += 1; stat_rush_requests += 1; ret = 1; } mtx_unlock(&sync_mtx); cv_broadcast(&sync_wakeup); return (ret); } /* * Tell the syncer to speed up its work and run though its work * list several times, then tell it to shut down. */ static void syncer_shutdown(void *arg, int howto) { if (howto & RB_NOSYNC) return; mtx_lock(&sync_mtx); syncer_state = SYNCER_SHUTTING_DOWN; rushjob = 0; mtx_unlock(&sync_mtx); cv_broadcast(&sync_wakeup); kproc_shutdown(arg, howto); } void syncer_suspend(void) { syncer_shutdown(updateproc, 0); } void syncer_resume(void) { mtx_lock(&sync_mtx); first_printf = 1; syncer_state = SYNCER_RUNNING; mtx_unlock(&sync_mtx); cv_broadcast(&sync_wakeup); kproc_resume(updateproc); } /* * Move the buffer between the clean and dirty lists of its vnode. */ void reassignbuf(struct buf *bp) { struct vnode *vp; struct bufobj *bo; int delay; #ifdef INVARIANTS struct bufv *bv; #endif vp = bp->b_vp; bo = bp->b_bufobj; KASSERT((bp->b_flags & B_PAGING) == 0, ("%s: cannot reassign paging buffer %p", __func__, bp)); CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); BO_LOCK(bo); buf_vlist_remove(bp); /* * If dirty, put on list of dirty buffers; otherwise insert onto list * of clean buffers. */ if (bp->b_flags & B_DELWRI) { if ((bo->bo_flag & BO_ONWORKLST) == 0) { switch (vp->v_type) { case VDIR: delay = dirdelay; break; case VCHR: delay = metadelay; break; default: delay = filedelay; } vn_syncer_add_to_worklist(bo, delay); } buf_vlist_add(bp, bo, BX_VNDIRTY); } else { buf_vlist_add(bp, bo, BX_VNCLEAN); if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { mtx_lock(&sync_mtx); LIST_REMOVE(bo, bo_synclist); syncer_worklist_len--; mtx_unlock(&sync_mtx); bo->bo_flag &= ~BO_ONWORKLST; } } #ifdef INVARIANTS bv = &bo->bo_clean; bp = TAILQ_FIRST(&bv->bv_hd); KASSERT(bp == NULL || bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); bp = TAILQ_LAST(&bv->bv_hd, buflists); KASSERT(bp == NULL || bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); bv = &bo->bo_dirty; bp = TAILQ_FIRST(&bv->bv_hd); KASSERT(bp == NULL || bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); bp = TAILQ_LAST(&bv->bv_hd, buflists); KASSERT(bp == NULL || bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); #endif BO_UNLOCK(bo); } static void v_init_counters(struct vnode *vp) { VNASSERT(vp->v_type == VNON && vp->v_data == NULL && vp->v_iflag == 0, vp, ("%s called for an initialized vnode", __FUNCTION__)); ASSERT_VI_UNLOCKED(vp, __FUNCTION__); refcount_init(&vp->v_holdcnt, 1); refcount_init(&vp->v_usecount, 1); } /* * Grab a particular vnode from the free list, increment its * reference count and lock it. VIRF_DOOMED is set if the vnode * is being destroyed. Only callers who specify LK_RETRY will * see doomed vnodes. If inactive processing was delayed in * vput try to do it here. * * usecount is manipulated using atomics without holding any locks. * * holdcnt can be manipulated using atomics without holding any locks, * except when transitioning 1<->0, in which case the interlock is held. * * Consumers which don't guarantee liveness of the vnode can use SMR to * try to get a reference. Note this operation can fail since the vnode * may be awaiting getting freed by the time they get to it. */ enum vgetstate vget_prep_smr(struct vnode *vp) { enum vgetstate vs; VFS_SMR_ASSERT_ENTERED(); if (refcount_acquire_if_not_zero(&vp->v_usecount)) { vs = VGET_USECOUNT; } else { if (vhold_smr(vp)) vs = VGET_HOLDCNT; else vs = VGET_NONE; } return (vs); } enum vgetstate vget_prep(struct vnode *vp) { enum vgetstate vs; if (refcount_acquire_if_not_zero(&vp->v_usecount)) { vs = VGET_USECOUNT; } else { vhold(vp); vs = VGET_HOLDCNT; } return (vs); } void vget_abort(struct vnode *vp, enum vgetstate vs) { switch (vs) { case VGET_USECOUNT: vrele(vp); break; case VGET_HOLDCNT: vdrop(vp); break; default: __assert_unreachable(); } } int vget(struct vnode *vp, int flags) { enum vgetstate vs; vs = vget_prep(vp); return (vget_finish(vp, flags, vs)); } int vget_finish(struct vnode *vp, int flags, enum vgetstate vs) { int error; if ((flags & LK_INTERLOCK) != 0) ASSERT_VI_LOCKED(vp, __func__); else ASSERT_VI_UNLOCKED(vp, __func__); VNPASS(vs == VGET_HOLDCNT || vs == VGET_USECOUNT, vp); VNPASS(vp->v_holdcnt > 0, vp); VNPASS(vs == VGET_HOLDCNT || vp->v_usecount > 0, vp); error = vn_lock(vp, flags); if (__predict_false(error != 0)) { vget_abort(vp, vs); CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__, vp); return (error); } vget_finish_ref(vp, vs); return (0); } void vget_finish_ref(struct vnode *vp, enum vgetstate vs) { int old; VNPASS(vs == VGET_HOLDCNT || vs == VGET_USECOUNT, vp); VNPASS(vp->v_holdcnt > 0, vp); VNPASS(vs == VGET_HOLDCNT || vp->v_usecount > 0, vp); if (vs == VGET_USECOUNT) return; /* * We hold the vnode. If the usecount is 0 it will be utilized to keep * the vnode around. Otherwise someone else lended their hold count and * we have to drop ours. */ old = atomic_fetchadd_int(&vp->v_usecount, 1); VNASSERT(old >= 0, vp, ("%s: wrong use count %d", __func__, old)); if (old != 0) { #ifdef INVARIANTS old = atomic_fetchadd_int(&vp->v_holdcnt, -1); VNASSERT(old > 1, vp, ("%s: wrong hold count %d", __func__, old)); #else refcount_release(&vp->v_holdcnt); #endif } } void vref(struct vnode *vp) { enum vgetstate vs; CTR2(KTR_VFS, "%s: vp %p", __func__, vp); vs = vget_prep(vp); vget_finish_ref(vp, vs); } void vrefact(struct vnode *vp) { CTR2(KTR_VFS, "%s: vp %p", __func__, vp); #ifdef INVARIANTS int old = atomic_fetchadd_int(&vp->v_usecount, 1); VNASSERT(old > 0, vp, ("%s: wrong use count %d", __func__, old)); #else refcount_acquire(&vp->v_usecount); #endif } void vlazy(struct vnode *vp) { struct mount *mp; VNASSERT(vp->v_holdcnt > 0, vp, ("%s: vnode not held", __func__)); if ((vp->v_mflag & VMP_LAZYLIST) != 0) return; /* * We may get here for inactive routines after the vnode got doomed. */ if (VN_IS_DOOMED(vp)) return; mp = vp->v_mount; mtx_lock(&mp->mnt_listmtx); if ((vp->v_mflag & VMP_LAZYLIST) == 0) { vp->v_mflag |= VMP_LAZYLIST; TAILQ_INSERT_TAIL(&mp->mnt_lazyvnodelist, vp, v_lazylist); mp->mnt_lazyvnodelistsize++; } mtx_unlock(&mp->mnt_listmtx); } static void vunlazy(struct vnode *vp) { struct mount *mp; ASSERT_VI_LOCKED(vp, __func__); VNPASS(!VN_IS_DOOMED(vp), vp); mp = vp->v_mount; mtx_lock(&mp->mnt_listmtx); VNPASS(vp->v_mflag & VMP_LAZYLIST, vp); /* * Don't remove the vnode from the lazy list if another thread * has increased the hold count. It may have re-enqueued the * vnode to the lazy list and is now responsible for its * removal. */ if (vp->v_holdcnt == 0) { vp->v_mflag &= ~VMP_LAZYLIST; TAILQ_REMOVE(&mp->mnt_lazyvnodelist, vp, v_lazylist); mp->mnt_lazyvnodelistsize--; } mtx_unlock(&mp->mnt_listmtx); } /* * This routine is only meant to be called from vgonel prior to dooming * the vnode. */ static void vunlazy_gone(struct vnode *vp) { struct mount *mp; ASSERT_VOP_ELOCKED(vp, __func__); ASSERT_VI_LOCKED(vp, __func__); VNPASS(!VN_IS_DOOMED(vp), vp); if (vp->v_mflag & VMP_LAZYLIST) { mp = vp->v_mount; mtx_lock(&mp->mnt_listmtx); VNPASS(vp->v_mflag & VMP_LAZYLIST, vp); vp->v_mflag &= ~VMP_LAZYLIST; TAILQ_REMOVE(&mp->mnt_lazyvnodelist, vp, v_lazylist); mp->mnt_lazyvnodelistsize--; mtx_unlock(&mp->mnt_listmtx); } } static void vdefer_inactive(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __func__); VNASSERT(vp->v_holdcnt > 0, vp, ("%s: vnode without hold count", __func__)); if (VN_IS_DOOMED(vp)) { vdropl(vp); return; } if (vp->v_iflag & VI_DEFINACT) { VNASSERT(vp->v_holdcnt > 1, vp, ("lost hold count")); vdropl(vp); return; } if (vp->v_usecount > 0) { vp->v_iflag &= ~VI_OWEINACT; vdropl(vp); return; } vlazy(vp); vp->v_iflag |= VI_DEFINACT; VI_UNLOCK(vp); counter_u64_add(deferred_inact, 1); } static void vdefer_inactive_unlocked(struct vnode *vp) { VI_LOCK(vp); if ((vp->v_iflag & VI_OWEINACT) == 0) { vdropl(vp); return; } vdefer_inactive(vp); } enum vput_op { VRELE, VPUT, VUNREF }; /* * Handle ->v_usecount transitioning to 0. * * By releasing the last usecount we take ownership of the hold count which * provides liveness of the vnode, meaning we have to vdrop. * * For all vnodes we may need to perform inactive processing. It requires an * exclusive lock on the vnode, while it is legal to call here with only a * shared lock (or no locks). If locking the vnode in an expected manner fails, * inactive processing gets deferred to the syncer. * * XXX Some filesystems pass in an exclusively locked vnode and strongly depend * on the lock being held all the way until VOP_INACTIVE. This in particular * happens with UFS which adds half-constructed vnodes to the hash, where they * can be found by other code. */ static void vput_final(struct vnode *vp, enum vput_op func) { int error; bool want_unlock; CTR2(KTR_VFS, "%s: vp %p", __func__, vp); VNPASS(vp->v_holdcnt > 0, vp); VI_LOCK(vp); /* * By the time we got here someone else might have transitioned * the count back to > 0. */ if (vp->v_usecount > 0) goto out; /* * If the vnode is doomed vgone already performed inactive processing * (if needed). */ if (VN_IS_DOOMED(vp)) goto out; if (__predict_true(VOP_NEED_INACTIVE(vp) == 0)) goto out; if (vp->v_iflag & VI_DOINGINACT) goto out; /* * Locking operations here will drop the interlock and possibly the * vnode lock, opening a window where the vnode can get doomed all the * while ->v_usecount is 0. Set VI_OWEINACT to let vgone know to * perform inactive. */ vp->v_iflag |= VI_OWEINACT; want_unlock = false; error = 0; switch (func) { case VRELE: switch (VOP_ISLOCKED(vp)) { case LK_EXCLUSIVE: break; case LK_EXCLOTHER: case 0: want_unlock = true; error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK); VI_LOCK(vp); break; default: /* * The lock has at least one sharer, but we have no way * to conclude whether this is us. Play it safe and * defer processing. */ error = EAGAIN; break; } break; case VPUT: want_unlock = true; if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK | LK_NOWAIT); VI_LOCK(vp); } break; case VUNREF: if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { error = VOP_LOCK(vp, LK_TRYUPGRADE | LK_INTERLOCK); VI_LOCK(vp); } break; } if (error == 0) { if (func == VUNREF) { VNASSERT((vp->v_vflag & VV_UNREF) == 0, vp, ("recursive vunref")); vp->v_vflag |= VV_UNREF; } for (;;) { error = vinactive(vp); if (want_unlock) VOP_UNLOCK(vp); if (error != ERELOOKUP || !want_unlock) break; VOP_LOCK(vp, LK_EXCLUSIVE); } if (func == VUNREF) vp->v_vflag &= ~VV_UNREF; vdropl(vp); } else { vdefer_inactive(vp); } return; out: if (func == VPUT) VOP_UNLOCK(vp); vdropl(vp); } /* * Decrement ->v_usecount for a vnode. * * Releasing the last use count requires additional processing, see vput_final * above for details. * * Comment above each variant denotes lock state on entry and exit. */ /* * in: any * out: same as passed in */ void vrele(struct vnode *vp) { ASSERT_VI_UNLOCKED(vp, __func__); if (!refcount_release(&vp->v_usecount)) return; vput_final(vp, VRELE); } /* * in: locked * out: unlocked */ void vput(struct vnode *vp) { ASSERT_VOP_LOCKED(vp, __func__); ASSERT_VI_UNLOCKED(vp, __func__); if (!refcount_release(&vp->v_usecount)) { VOP_UNLOCK(vp); return; } vput_final(vp, VPUT); } /* * in: locked * out: locked */ void vunref(struct vnode *vp) { ASSERT_VOP_LOCKED(vp, __func__); ASSERT_VI_UNLOCKED(vp, __func__); if (!refcount_release(&vp->v_usecount)) return; vput_final(vp, VUNREF); } void vhold(struct vnode *vp) { int old; CTR2(KTR_VFS, "%s: vp %p", __func__, vp); old = atomic_fetchadd_int(&vp->v_holdcnt, 1); VNASSERT(old >= 0 && (old & VHOLD_ALL_FLAGS) == 0, vp, ("%s: wrong hold count %d", __func__, old)); if (old == 0) vfs_freevnodes_dec(); } void vholdnz(struct vnode *vp) { CTR2(KTR_VFS, "%s: vp %p", __func__, vp); #ifdef INVARIANTS int old = atomic_fetchadd_int(&vp->v_holdcnt, 1); VNASSERT(old > 0 && (old & VHOLD_ALL_FLAGS) == 0, vp, ("%s: wrong hold count %d", __func__, old)); #else atomic_add_int(&vp->v_holdcnt, 1); #endif } /* * Grab a hold count unless the vnode is freed. * * Only use this routine if vfs smr is the only protection you have against * freeing the vnode. * * The code loops trying to add a hold count as long as the VHOLD_NO_SMR flag * is not set. After the flag is set the vnode becomes immutable to anyone but * the thread which managed to set the flag. * * It may be tempting to replace the loop with: * count = atomic_fetchadd_int(&vp->v_holdcnt, 1); * if (count & VHOLD_NO_SMR) { * backpedal and error out; * } * * However, while this is more performant, it hinders debugging by eliminating * the previously mentioned invariant. */ bool vhold_smr(struct vnode *vp) { int count; VFS_SMR_ASSERT_ENTERED(); count = atomic_load_int(&vp->v_holdcnt); for (;;) { if (count & VHOLD_NO_SMR) { VNASSERT((count & ~VHOLD_NO_SMR) == 0, vp, ("non-zero hold count with flags %d\n", count)); return (false); } VNASSERT(count >= 0, vp, ("invalid hold count %d\n", count)); if (atomic_fcmpset_int(&vp->v_holdcnt, &count, count + 1)) { if (count == 0) vfs_freevnodes_dec(); return (true); } } } /* * Hold a free vnode for recycling. * * Note: vnode_init references this comment. * * Attempts to recycle only need the global vnode list lock and have no use for * SMR. * * However, vnodes get inserted into the global list before they get fully * initialized and stay there until UMA decides to free the memory. This in * particular means the target can be found before it becomes usable and after * it becomes recycled. Picking up such vnodes is guarded with v_holdcnt set to * VHOLD_NO_SMR. * * Note: the vnode may gain more references after we transition the count 0->1. */ static bool vhold_recycle_free(struct vnode *vp) { int count; mtx_assert(&vnode_list_mtx, MA_OWNED); count = atomic_load_int(&vp->v_holdcnt); for (;;) { if (count & VHOLD_NO_SMR) { VNASSERT((count & ~VHOLD_NO_SMR) == 0, vp, ("non-zero hold count with flags %d\n", count)); return (false); } VNASSERT(count >= 0, vp, ("invalid hold count %d\n", count)); if (count > 0) { return (false); } if (atomic_fcmpset_int(&vp->v_holdcnt, &count, count + 1)) { vfs_freevnodes_dec(); return (true); } } } static void __noinline vdbatch_process(struct vdbatch *vd) { struct vnode *vp; int i; mtx_assert(&vd->lock, MA_OWNED); MPASS(curthread->td_pinned > 0); MPASS(vd->index == VDBATCH_SIZE); mtx_lock(&vnode_list_mtx); critical_enter(); freevnodes += vd->freevnodes; for (i = 0; i < VDBATCH_SIZE; i++) { vp = vd->tab[i]; TAILQ_REMOVE(&vnode_list, vp, v_vnodelist); TAILQ_INSERT_TAIL(&vnode_list, vp, v_vnodelist); MPASS(vp->v_dbatchcpu != NOCPU); vp->v_dbatchcpu = NOCPU; } mtx_unlock(&vnode_list_mtx); vd->freevnodes = 0; bzero(vd->tab, sizeof(vd->tab)); vd->index = 0; critical_exit(); } static void vdbatch_enqueue(struct vnode *vp) { struct vdbatch *vd; ASSERT_VI_LOCKED(vp, __func__); VNASSERT(!VN_IS_DOOMED(vp), vp, ("%s: deferring requeue of a doomed vnode", __func__)); if (vp->v_dbatchcpu != NOCPU) { VI_UNLOCK(vp); return; } sched_pin(); vd = DPCPU_PTR(vd); mtx_lock(&vd->lock); MPASS(vd->index < VDBATCH_SIZE); MPASS(vd->tab[vd->index] == NULL); /* * A hack: we depend on being pinned so that we know what to put in * ->v_dbatchcpu. */ vp->v_dbatchcpu = curcpu; vd->tab[vd->index] = vp; vd->index++; VI_UNLOCK(vp); if (vd->index == VDBATCH_SIZE) vdbatch_process(vd); mtx_unlock(&vd->lock); sched_unpin(); } /* * This routine must only be called for vnodes which are about to be * deallocated. Supporting dequeue for arbitrary vndoes would require * validating that the locked batch matches. */ static void vdbatch_dequeue(struct vnode *vp) { struct vdbatch *vd; int i; short cpu; VNASSERT(vp->v_type == VBAD || vp->v_type == VNON, vp, ("%s: called for a used vnode\n", __func__)); cpu = vp->v_dbatchcpu; if (cpu == NOCPU) return; vd = DPCPU_ID_PTR(cpu, vd); mtx_lock(&vd->lock); for (i = 0; i < vd->index; i++) { if (vd->tab[i] != vp) continue; vp->v_dbatchcpu = NOCPU; vd->index--; vd->tab[i] = vd->tab[vd->index]; vd->tab[vd->index] = NULL; break; } mtx_unlock(&vd->lock); /* * Either we dequeued the vnode above or the target CPU beat us to it. */ MPASS(vp->v_dbatchcpu == NOCPU); } /* * Drop the hold count of the vnode. If this is the last reference to * the vnode we place it on the free list unless it has been vgone'd * (marked VIRF_DOOMED) in which case we will free it. * * Because the vnode vm object keeps a hold reference on the vnode if * there is at least one resident non-cached page, the vnode cannot * leave the active list without the page cleanup done. */ static void __noinline vdropl_final(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __func__); VNPASS(VN_IS_DOOMED(vp), vp); /* * Set the VHOLD_NO_SMR flag. * * We may be racing against vhold_smr. If they win we can just pretend * we never got this far, they will vdrop later. */ if (__predict_false(!atomic_cmpset_int(&vp->v_holdcnt, 0, VHOLD_NO_SMR))) { vfs_freevnodes_inc(); VI_UNLOCK(vp); /* * We lost the aforementioned race. Any subsequent access is * invalid as they might have managed to vdropl on their own. */ return; } /* * Don't bump freevnodes as this one is going away. */ freevnode(vp); } void vdrop(struct vnode *vp) { ASSERT_VI_UNLOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); if (refcount_release_if_not_last(&vp->v_holdcnt)) return; VI_LOCK(vp); vdropl(vp); } void vdropl(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); if (!refcount_release(&vp->v_holdcnt)) { VI_UNLOCK(vp); return; } VNPASS((vp->v_iflag & VI_OWEINACT) == 0, vp); VNPASS((vp->v_iflag & VI_DEFINACT) == 0, vp); if (VN_IS_DOOMED(vp)) { vdropl_final(vp); return; } vfs_freevnodes_inc(); if (vp->v_mflag & VMP_LAZYLIST) { vunlazy(vp); } /* * Also unlocks the interlock. We can't assert on it as we * released our hold and by now the vnode might have been * freed. */ vdbatch_enqueue(vp); } /* * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT * flags. DOINGINACT prevents us from recursing in calls to vinactive. */ static int vinactivef(struct vnode *vp) { struct vm_object *obj; int error; ASSERT_VOP_ELOCKED(vp, "vinactive"); ASSERT_VI_LOCKED(vp, "vinactive"); VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp, ("vinactive: recursed on VI_DOINGINACT")); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); vp->v_iflag |= VI_DOINGINACT; vp->v_iflag &= ~VI_OWEINACT; VI_UNLOCK(vp); /* * Before moving off the active list, we must be sure that any * modified pages are converted into the vnode's dirty * buffers, since these will no longer be checked once the * vnode is on the inactive list. * * The write-out of the dirty pages is asynchronous. At the * point that VOP_INACTIVE() is called, there could still be * pending I/O and dirty pages in the object. */ if ((obj = vp->v_object) != NULL && (vp->v_vflag & VV_NOSYNC) == 0 && vm_object_mightbedirty(obj)) { VM_OBJECT_WLOCK(obj); vm_object_page_clean(obj, 0, 0, 0); VM_OBJECT_WUNLOCK(obj); } error = VOP_INACTIVE(vp); VI_LOCK(vp); VNASSERT(vp->v_iflag & VI_DOINGINACT, vp, ("vinactive: lost VI_DOINGINACT")); vp->v_iflag &= ~VI_DOINGINACT; return (error); } int vinactive(struct vnode *vp) { ASSERT_VOP_ELOCKED(vp, "vinactive"); ASSERT_VI_LOCKED(vp, "vinactive"); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); if ((vp->v_iflag & VI_OWEINACT) == 0) return (0); if (vp->v_iflag & VI_DOINGINACT) return (0); if (vp->v_usecount > 0) { vp->v_iflag &= ~VI_OWEINACT; return (0); } return (vinactivef(vp)); } /* * Remove any vnodes in the vnode table belonging to mount point mp. * * If FORCECLOSE is not specified, there should not be any active ones, * return error if any are found (nb: this is a user error, not a * system error). If FORCECLOSE is specified, detach any active vnodes * that are found. * * If WRITECLOSE is set, only flush out regular file vnodes open for * writing. * * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped. * * `rootrefs' specifies the base reference count for the root vnode * of this filesystem. The root vnode is considered busy if its * v_usecount exceeds this value. On a successful return, vflush(, td) * will call vrele() on the root vnode exactly rootrefs times. * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must * be zero. */ #ifdef DIAGNOSTIC static int busyprt = 0; /* print out busy vnodes */ SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes"); #endif int vflush(struct mount *mp, int rootrefs, int flags, struct thread *td) { struct vnode *vp, *mvp, *rootvp = NULL; struct vattr vattr; int busy = 0, error; CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp, rootrefs, flags); if (rootrefs > 0) { KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0, ("vflush: bad args")); /* * Get the filesystem root vnode. We can vput() it * immediately, since with rootrefs > 0, it won't go away. */ if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) { CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d", __func__, error); return (error); } vput(rootvp); } loop: MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { vholdl(vp); error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE); if (error) { vdrop(vp); MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); goto loop; } /* * Skip over a vnodes marked VV_SYSTEM. */ if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) { VOP_UNLOCK(vp); vdrop(vp); continue; } /* * If WRITECLOSE is set, flush out unlinked but still open * files (even if open only for reading) and regular file * vnodes open for writing. */ if (flags & WRITECLOSE) { if (vp->v_object != NULL) { VM_OBJECT_WLOCK(vp->v_object); vm_object_page_clean(vp->v_object, 0, 0, 0); VM_OBJECT_WUNLOCK(vp->v_object); } do { error = VOP_FSYNC(vp, MNT_WAIT, td); } while (error == ERELOOKUP); if (error != 0) { VOP_UNLOCK(vp); vdrop(vp); MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); return (error); } error = VOP_GETATTR(vp, &vattr, td->td_ucred); VI_LOCK(vp); if ((vp->v_type == VNON || (error == 0 && vattr.va_nlink > 0)) && (vp->v_writecount <= 0 || vp->v_type != VREG)) { VOP_UNLOCK(vp); vdropl(vp); continue; } } else VI_LOCK(vp); /* * With v_usecount == 0, all we need to do is clear out the * vnode data structures and we are done. * * If FORCECLOSE is set, forcibly close the vnode. */ if (vp->v_usecount == 0 || (flags & FORCECLOSE)) { vgonel(vp); } else { busy++; #ifdef DIAGNOSTIC if (busyprt) vn_printf(vp, "vflush: busy vnode "); #endif } VOP_UNLOCK(vp); vdropl(vp); } if (rootrefs > 0 && (flags & FORCECLOSE) == 0) { /* * If just the root vnode is busy, and if its refcount * is equal to `rootrefs', then go ahead and kill it. */ VI_LOCK(rootvp); KASSERT(busy > 0, ("vflush: not busy")); VNASSERT(rootvp->v_usecount >= rootrefs, rootvp, ("vflush: usecount %d < rootrefs %d", rootvp->v_usecount, rootrefs)); if (busy == 1 && rootvp->v_usecount == rootrefs) { VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK); vgone(rootvp); VOP_UNLOCK(rootvp); busy = 0; } else VI_UNLOCK(rootvp); } if (busy) { CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__, busy); return (EBUSY); } for (; rootrefs > 0; rootrefs--) vrele(rootvp); return (0); } /* * Recycle an unused vnode to the front of the free list. */ int vrecycle(struct vnode *vp) { int recycled; VI_LOCK(vp); recycled = vrecyclel(vp); VI_UNLOCK(vp); return (recycled); } /* * vrecycle, with the vp interlock held. */ int vrecyclel(struct vnode *vp) { int recycled; ASSERT_VOP_ELOCKED(vp, __func__); ASSERT_VI_LOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); recycled = 0; if (vp->v_usecount == 0) { recycled = 1; vgonel(vp); } return (recycled); } /* * Eliminate all activity associated with a vnode * in preparation for reuse. */ void vgone(struct vnode *vp) { VI_LOCK(vp); vgonel(vp); VI_UNLOCK(vp); } /* * Notify upper mounts about reclaimed or unlinked vnode. */ void vfs_notify_upper(struct vnode *vp, int event) { struct mount *mp; struct mount_upper_node *ump; mp = atomic_load_ptr(&vp->v_mount); if (mp == NULL) return; if (TAILQ_EMPTY(&mp->mnt_notify)) return; MNT_ILOCK(mp); mp->mnt_upper_pending++; KASSERT(mp->mnt_upper_pending > 0, ("%s: mnt_upper_pending %d", __func__, mp->mnt_upper_pending)); TAILQ_FOREACH(ump, &mp->mnt_notify, mnt_upper_link) { MNT_IUNLOCK(mp); switch (event) { case VFS_NOTIFY_UPPER_RECLAIM: VFS_RECLAIM_LOWERVP(ump->mp, vp); break; case VFS_NOTIFY_UPPER_UNLINK: VFS_UNLINK_LOWERVP(ump->mp, vp); break; default: KASSERT(0, ("invalid event %d", event)); break; } MNT_ILOCK(mp); } mp->mnt_upper_pending--; if ((mp->mnt_kern_flag & MNTK_UPPER_WAITER) != 0 && mp->mnt_upper_pending == 0) { mp->mnt_kern_flag &= ~MNTK_UPPER_WAITER; wakeup(&mp->mnt_uppers); } MNT_IUNLOCK(mp); } /* * vgone, with the vp interlock held. */ static void vgonel(struct vnode *vp) { struct thread *td; struct mount *mp; vm_object_t object; bool active, doinginact, oweinact; ASSERT_VOP_ELOCKED(vp, "vgonel"); ASSERT_VI_LOCKED(vp, "vgonel"); VNASSERT(vp->v_holdcnt, vp, ("vgonel: vp %p has no reference.", vp)); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); td = curthread; /* * Don't vgonel if we're already doomed. */ if (VN_IS_DOOMED(vp)) return; /* * Paired with freevnode. */ vn_seqc_write_begin_locked(vp); vunlazy_gone(vp); vn_irflag_set_locked(vp, VIRF_DOOMED); /* * Check to see if the vnode is in use. If so, we have to * call VOP_CLOSE() and VOP_INACTIVE(). * * It could be that VOP_INACTIVE() requested reclamation, in * which case we should avoid recursion, so check * VI_DOINGINACT. This is not precise but good enough. */ active = vp->v_usecount > 0; oweinact = (vp->v_iflag & VI_OWEINACT) != 0; doinginact = (vp->v_iflag & VI_DOINGINACT) != 0; /* * If we need to do inactive VI_OWEINACT will be set. */ if (vp->v_iflag & VI_DEFINACT) { VNASSERT(vp->v_holdcnt > 1, vp, ("lost hold count")); vp->v_iflag &= ~VI_DEFINACT; vdropl(vp); } else { VNASSERT(vp->v_holdcnt > 0, vp, ("vnode without hold count")); VI_UNLOCK(vp); } cache_purge_vgone(vp); vfs_notify_upper(vp, VFS_NOTIFY_UPPER_RECLAIM); /* * If purging an active vnode, it must be closed and * deactivated before being reclaimed. */ if (active) VOP_CLOSE(vp, FNONBLOCK, NOCRED, td); if (!doinginact) { do { if (oweinact || active) { VI_LOCK(vp); vinactivef(vp); oweinact = (vp->v_iflag & VI_OWEINACT) != 0; VI_UNLOCK(vp); } } while (oweinact); } if (vp->v_type == VSOCK) vfs_unp_reclaim(vp); /* * Clean out any buffers associated with the vnode. * If the flush fails, just toss the buffers. */ mp = NULL; if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd)) (void) vn_start_secondary_write(vp, &mp, V_WAIT); if (vinvalbuf(vp, V_SAVE, 0, 0) != 0) { while (vinvalbuf(vp, 0, 0, 0) != 0) ; } BO_LOCK(&vp->v_bufobj); KASSERT(TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd) && vp->v_bufobj.bo_dirty.bv_cnt == 0 && TAILQ_EMPTY(&vp->v_bufobj.bo_clean.bv_hd) && vp->v_bufobj.bo_clean.bv_cnt == 0, ("vp %p bufobj not invalidated", vp)); /* * For VMIO bufobj, BO_DEAD is set later, or in * vm_object_terminate() after the object's page queue is * flushed. */ object = vp->v_bufobj.bo_object; if (object == NULL) vp->v_bufobj.bo_flag |= BO_DEAD; BO_UNLOCK(&vp->v_bufobj); /* * Handle the VM part. Tmpfs handles v_object on its own (the * OBJT_VNODE check). Nullfs or other bypassing filesystems * should not touch the object borrowed from the lower vnode * (the handle check). */ if (object != NULL && object->type == OBJT_VNODE && object->handle == vp) vnode_destroy_vobject(vp); /* * Reclaim the vnode. */ if (VOP_RECLAIM(vp)) panic("vgone: cannot reclaim"); if (mp != NULL) vn_finished_secondary_write(mp); VNASSERT(vp->v_object == NULL, vp, ("vop_reclaim left v_object vp=%p", vp)); /* * Clear the advisory locks and wake up waiting threads. */ (void)VOP_ADVLOCKPURGE(vp); vp->v_lockf = NULL; /* * Delete from old mount point vnode list. */ delmntque(vp); /* * Done with purge, reset to the standard lock and invalidate * the vnode. */ VI_LOCK(vp); vp->v_vnlock = &vp->v_lock; vp->v_op = &dead_vnodeops; vp->v_type = VBAD; } /* * Print out a description of a vnode. */ static const char * const typename[] = {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD", "VMARKER"}; _Static_assert((VHOLD_ALL_FLAGS & ~VHOLD_NO_SMR) == 0, "new hold count flag not added to vn_printf"); void vn_printf(struct vnode *vp, const char *fmt, ...) { va_list ap; char buf[256], buf2[16]; u_long flags; u_int holdcnt; short irflag; va_start(ap, fmt); vprintf(fmt, ap); va_end(ap); printf("%p: ", (void *)vp); printf("type %s\n", typename[vp->v_type]); holdcnt = atomic_load_int(&vp->v_holdcnt); printf(" usecount %d, writecount %d, refcount %d seqc users %d", vp->v_usecount, vp->v_writecount, holdcnt & ~VHOLD_ALL_FLAGS, vp->v_seqc_users); switch (vp->v_type) { case VDIR: printf(" mountedhere %p\n", vp->v_mountedhere); break; case VCHR: printf(" rdev %p\n", vp->v_rdev); break; case VSOCK: printf(" socket %p\n", vp->v_unpcb); break; case VFIFO: printf(" fifoinfo %p\n", vp->v_fifoinfo); break; default: printf("\n"); break; } buf[0] = '\0'; buf[1] = '\0'; if (holdcnt & VHOLD_NO_SMR) strlcat(buf, "|VHOLD_NO_SMR", sizeof(buf)); printf(" hold count flags (%s)\n", buf + 1); buf[0] = '\0'; buf[1] = '\0'; irflag = vn_irflag_read(vp); if (irflag & VIRF_DOOMED) strlcat(buf, "|VIRF_DOOMED", sizeof(buf)); if (irflag & VIRF_PGREAD) strlcat(buf, "|VIRF_PGREAD", sizeof(buf)); if (irflag & VIRF_MOUNTPOINT) strlcat(buf, "|VIRF_MOUNTPOINT", sizeof(buf)); flags = irflag & ~(VIRF_DOOMED | VIRF_PGREAD | VIRF_MOUNTPOINT); if (flags != 0) { snprintf(buf2, sizeof(buf2), "|VIRF(0x%lx)", flags); strlcat(buf, buf2, sizeof(buf)); } if (vp->v_vflag & VV_ROOT) strlcat(buf, "|VV_ROOT", sizeof(buf)); if (vp->v_vflag & VV_ISTTY) strlcat(buf, "|VV_ISTTY", sizeof(buf)); if (vp->v_vflag & VV_NOSYNC) strlcat(buf, "|VV_NOSYNC", sizeof(buf)); if (vp->v_vflag & VV_ETERNALDEV) strlcat(buf, "|VV_ETERNALDEV", sizeof(buf)); if (vp->v_vflag & VV_CACHEDLABEL) strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf)); if (vp->v_vflag & VV_VMSIZEVNLOCK) strlcat(buf, "|VV_VMSIZEVNLOCK", sizeof(buf)); if (vp->v_vflag & VV_COPYONWRITE) strlcat(buf, "|VV_COPYONWRITE", sizeof(buf)); if (vp->v_vflag & VV_SYSTEM) strlcat(buf, "|VV_SYSTEM", sizeof(buf)); if (vp->v_vflag & VV_PROCDEP) strlcat(buf, "|VV_PROCDEP", sizeof(buf)); if (vp->v_vflag & VV_NOKNOTE) strlcat(buf, "|VV_NOKNOTE", sizeof(buf)); if (vp->v_vflag & VV_DELETED) strlcat(buf, "|VV_DELETED", sizeof(buf)); if (vp->v_vflag & VV_MD) strlcat(buf, "|VV_MD", sizeof(buf)); if (vp->v_vflag & VV_FORCEINSMQ) strlcat(buf, "|VV_FORCEINSMQ", sizeof(buf)); if (vp->v_vflag & VV_READLINK) strlcat(buf, "|VV_READLINK", sizeof(buf)); flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | VV_ETERNALDEV | VV_CACHEDLABEL | VV_VMSIZEVNLOCK | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP | VV_NOKNOTE | VV_DELETED | VV_MD | VV_FORCEINSMQ | VV_READLINK); if (flags != 0) { snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags); strlcat(buf, buf2, sizeof(buf)); } if (vp->v_iflag & VI_TEXT_REF) strlcat(buf, "|VI_TEXT_REF", sizeof(buf)); if (vp->v_iflag & VI_MOUNT) strlcat(buf, "|VI_MOUNT", sizeof(buf)); if (vp->v_iflag & VI_DOINGINACT) strlcat(buf, "|VI_DOINGINACT", sizeof(buf)); if (vp->v_iflag & VI_OWEINACT) strlcat(buf, "|VI_OWEINACT", sizeof(buf)); if (vp->v_iflag & VI_DEFINACT) strlcat(buf, "|VI_DEFINACT", sizeof(buf)); if (vp->v_iflag & VI_FOPENING) strlcat(buf, "|VI_FOPENING", sizeof(buf)); flags = vp->v_iflag & ~(VI_TEXT_REF | VI_MOUNT | VI_DOINGINACT | VI_OWEINACT | VI_DEFINACT | VI_FOPENING); if (flags != 0) { snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags); strlcat(buf, buf2, sizeof(buf)); } if (vp->v_mflag & VMP_LAZYLIST) strlcat(buf, "|VMP_LAZYLIST", sizeof(buf)); flags = vp->v_mflag & ~(VMP_LAZYLIST); if (flags != 0) { snprintf(buf2, sizeof(buf2), "|VMP(0x%lx)", flags); strlcat(buf, buf2, sizeof(buf)); } printf(" flags (%s)", buf + 1); if (mtx_owned(VI_MTX(vp))) printf(" VI_LOCKed"); printf("\n"); if (vp->v_object != NULL) printf(" v_object %p ref %d pages %d " "cleanbuf %d dirtybuf %d\n", vp->v_object, vp->v_object->ref_count, vp->v_object->resident_page_count, vp->v_bufobj.bo_clean.bv_cnt, vp->v_bufobj.bo_dirty.bv_cnt); printf(" "); lockmgr_printinfo(vp->v_vnlock); if (vp->v_data != NULL) VOP_PRINT(vp); } #ifdef DDB /* * List all of the locked vnodes in the system. * Called when debugging the kernel. */ DB_SHOW_COMMAND(lockedvnods, lockedvnodes) { struct mount *mp; struct vnode *vp; /* * Note: because this is DDB, we can't obey the locking semantics * for these structures, which means we could catch an inconsistent * state and dereference a nasty pointer. Not much to be done * about that. */ db_printf("Locked vnodes\n"); TAILQ_FOREACH(mp, &mountlist, mnt_list) { TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { if (vp->v_type != VMARKER && VOP_ISLOCKED(vp)) vn_printf(vp, "vnode "); } } } /* * Show details about the given vnode. */ DB_SHOW_COMMAND(vnode, db_show_vnode) { struct vnode *vp; if (!have_addr) return; vp = (struct vnode *)addr; vn_printf(vp, "vnode "); } /* * Show details about the given mount point. */ DB_SHOW_COMMAND(mount, db_show_mount) { struct mount *mp; struct vfsopt *opt; struct statfs *sp; struct vnode *vp; char buf[512]; uint64_t mflags; u_int flags; if (!have_addr) { /* No address given, print short info about all mount points. */ TAILQ_FOREACH(mp, &mountlist, mnt_list) { db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename); if (db_pager_quit) break; } db_printf("\nMore info: show mount \n"); return; } mp = (struct mount *)addr; db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename); buf[0] = '\0'; mflags = mp->mnt_flag; #define MNT_FLAG(flag) do { \ if (mflags & (flag)) { \ if (buf[0] != '\0') \ strlcat(buf, ", ", sizeof(buf)); \ strlcat(buf, (#flag) + 4, sizeof(buf)); \ mflags &= ~(flag); \ } \ } while (0) MNT_FLAG(MNT_RDONLY); MNT_FLAG(MNT_SYNCHRONOUS); MNT_FLAG(MNT_NOEXEC); MNT_FLAG(MNT_NOSUID); MNT_FLAG(MNT_NFS4ACLS); MNT_FLAG(MNT_UNION); MNT_FLAG(MNT_ASYNC); MNT_FLAG(MNT_SUIDDIR); MNT_FLAG(MNT_SOFTDEP); MNT_FLAG(MNT_NOSYMFOLLOW); MNT_FLAG(MNT_GJOURNAL); MNT_FLAG(MNT_MULTILABEL); MNT_FLAG(MNT_ACLS); MNT_FLAG(MNT_NOATIME); MNT_FLAG(MNT_NOCLUSTERR); MNT_FLAG(MNT_NOCLUSTERW); MNT_FLAG(MNT_SUJ); MNT_FLAG(MNT_EXRDONLY); MNT_FLAG(MNT_EXPORTED); MNT_FLAG(MNT_DEFEXPORTED); MNT_FLAG(MNT_EXPORTANON); MNT_FLAG(MNT_EXKERB); MNT_FLAG(MNT_EXPUBLIC); MNT_FLAG(MNT_LOCAL); MNT_FLAG(MNT_QUOTA); MNT_FLAG(MNT_ROOTFS); MNT_FLAG(MNT_USER); MNT_FLAG(MNT_IGNORE); MNT_FLAG(MNT_UPDATE); MNT_FLAG(MNT_DELEXPORT); MNT_FLAG(MNT_RELOAD); MNT_FLAG(MNT_FORCE); MNT_FLAG(MNT_SNAPSHOT); MNT_FLAG(MNT_BYFSID); #undef MNT_FLAG if (mflags != 0) { if (buf[0] != '\0') strlcat(buf, ", ", sizeof(buf)); snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), "0x%016jx", mflags); } db_printf(" mnt_flag = %s\n", buf); buf[0] = '\0'; flags = mp->mnt_kern_flag; #define MNT_KERN_FLAG(flag) do { \ if (flags & (flag)) { \ if (buf[0] != '\0') \ strlcat(buf, ", ", sizeof(buf)); \ strlcat(buf, (#flag) + 5, sizeof(buf)); \ flags &= ~(flag); \ } \ } while (0) MNT_KERN_FLAG(MNTK_UNMOUNTF); MNT_KERN_FLAG(MNTK_ASYNC); MNT_KERN_FLAG(MNTK_SOFTDEP); MNT_KERN_FLAG(MNTK_DRAINING); MNT_KERN_FLAG(MNTK_REFEXPIRE); MNT_KERN_FLAG(MNTK_EXTENDED_SHARED); MNT_KERN_FLAG(MNTK_SHARED_WRITES); MNT_KERN_FLAG(MNTK_NO_IOPF); MNT_KERN_FLAG(MNTK_RECURSE); MNT_KERN_FLAG(MNTK_UPPER_WAITER); MNT_KERN_FLAG(MNTK_LOOKUP_EXCL_DOTDOT); MNT_KERN_FLAG(MNTK_USES_BCACHE); MNT_KERN_FLAG(MNTK_FPLOOKUP); MNT_KERN_FLAG(MNTK_TASKQUEUE_WAITER); MNT_KERN_FLAG(MNTK_NOASYNC); MNT_KERN_FLAG(MNTK_UNMOUNT); MNT_KERN_FLAG(MNTK_MWAIT); MNT_KERN_FLAG(MNTK_SUSPEND); MNT_KERN_FLAG(MNTK_SUSPEND2); MNT_KERN_FLAG(MNTK_SUSPENDED); MNT_KERN_FLAG(MNTK_LOOKUP_SHARED); MNT_KERN_FLAG(MNTK_NOKNOTE); #undef MNT_KERN_FLAG if (flags != 0) { if (buf[0] != '\0') strlcat(buf, ", ", sizeof(buf)); snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), "0x%08x", flags); } db_printf(" mnt_kern_flag = %s\n", buf); db_printf(" mnt_opt = "); opt = TAILQ_FIRST(mp->mnt_opt); if (opt != NULL) { db_printf("%s", opt->name); opt = TAILQ_NEXT(opt, link); while (opt != NULL) { db_printf(", %s", opt->name); opt = TAILQ_NEXT(opt, link); } } db_printf("\n"); sp = &mp->mnt_stat; db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx " "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju " "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju " "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n", (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags, (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize, (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree, (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files, (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites, (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads, (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax, (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]); db_printf(" mnt_cred = { uid=%u ruid=%u", (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid); if (jailed(mp->mnt_cred)) db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id); db_printf(" }\n"); db_printf(" mnt_ref = %d (with %d in the struct)\n", vfs_mount_fetch_counter(mp, MNT_COUNT_REF), mp->mnt_ref); db_printf(" mnt_gen = %d\n", mp->mnt_gen); db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize); db_printf(" mnt_lazyvnodelistsize = %d\n", mp->mnt_lazyvnodelistsize); db_printf(" mnt_writeopcount = %d (with %d in the struct)\n", vfs_mount_fetch_counter(mp, MNT_COUNT_WRITEOPCOUNT), mp->mnt_writeopcount); db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max); db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed); db_printf(" mnt_lockref = %d (with %d in the struct)\n", vfs_mount_fetch_counter(mp, MNT_COUNT_LOCKREF), mp->mnt_lockref); db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes); db_printf(" mnt_secondary_accwrites = %d\n", mp->mnt_secondary_accwrites); db_printf(" mnt_gjprovider = %s\n", mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL"); db_printf(" mnt_vfs_ops = %d\n", mp->mnt_vfs_ops); db_printf("\n\nList of active vnodes\n"); TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { if (vp->v_type != VMARKER && vp->v_holdcnt > 0) { vn_printf(vp, "vnode "); if (db_pager_quit) break; } } db_printf("\n\nList of inactive vnodes\n"); TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { if (vp->v_type != VMARKER && vp->v_holdcnt == 0) { vn_printf(vp, "vnode "); if (db_pager_quit) break; } } } #endif /* DDB */ /* * Fill in a struct xvfsconf based on a struct vfsconf. */ static int vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp) { struct xvfsconf xvfsp; bzero(&xvfsp, sizeof(xvfsp)); strcpy(xvfsp.vfc_name, vfsp->vfc_name); xvfsp.vfc_typenum = vfsp->vfc_typenum; xvfsp.vfc_refcount = vfsp->vfc_refcount; xvfsp.vfc_flags = vfsp->vfc_flags; /* * These are unused in userland, we keep them * to not break binary compatibility. */ xvfsp.vfc_vfsops = NULL; xvfsp.vfc_next = NULL; return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp))); } #ifdef COMPAT_FREEBSD32 struct xvfsconf32 { uint32_t vfc_vfsops; char vfc_name[MFSNAMELEN]; int32_t vfc_typenum; int32_t vfc_refcount; int32_t vfc_flags; uint32_t vfc_next; }; static int vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp) { struct xvfsconf32 xvfsp; bzero(&xvfsp, sizeof(xvfsp)); strcpy(xvfsp.vfc_name, vfsp->vfc_name); xvfsp.vfc_typenum = vfsp->vfc_typenum; xvfsp.vfc_refcount = vfsp->vfc_refcount; xvfsp.vfc_flags = vfsp->vfc_flags; return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp))); } #endif /* * Top level filesystem related information gathering. */ static int sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS) { struct vfsconf *vfsp; int error; error = 0; vfsconf_slock(); TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { #ifdef COMPAT_FREEBSD32 if (req->flags & SCTL_MASK32) error = vfsconf2x32(req, vfsp); else #endif error = vfsconf2x(req, vfsp); if (error) break; } vfsconf_sunlock(); return (error); } SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, sysctl_vfs_conflist, "S,xvfsconf", "List of all configured filesystems"); #ifndef BURN_BRIDGES static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS); static int vfs_sysctl(SYSCTL_HANDLER_ARGS) { int *name = (int *)arg1 - 1; /* XXX */ u_int namelen = arg2 + 1; /* XXX */ struct vfsconf *vfsp; log(LOG_WARNING, "userland calling deprecated sysctl, " "please rebuild world\n"); #if 1 || defined(COMPAT_PRELITE2) /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ if (namelen == 1) return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); #endif switch (name[1]) { case VFS_MAXTYPENUM: if (namelen != 2) return (ENOTDIR); return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); case VFS_CONF: if (namelen != 3) return (ENOTDIR); /* overloaded */ vfsconf_slock(); TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { if (vfsp->vfc_typenum == name[2]) break; } vfsconf_sunlock(); if (vfsp == NULL) return (EOPNOTSUPP); #ifdef COMPAT_FREEBSD32 if (req->flags & SCTL_MASK32) return (vfsconf2x32(req, vfsp)); else #endif return (vfsconf2x(req, vfsp)); } return (EOPNOTSUPP); } static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP | CTLFLAG_MPSAFE, vfs_sysctl, "Generic filesystem"); #if 1 || defined(COMPAT_PRELITE2) static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) { int error; struct vfsconf *vfsp; struct ovfsconf ovfs; vfsconf_slock(); TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { bzero(&ovfs, sizeof(ovfs)); ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ strcpy(ovfs.vfc_name, vfsp->vfc_name); ovfs.vfc_index = vfsp->vfc_typenum; ovfs.vfc_refcount = vfsp->vfc_refcount; ovfs.vfc_flags = vfsp->vfc_flags; error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); if (error != 0) { vfsconf_sunlock(); return (error); } } vfsconf_sunlock(); return (0); } #endif /* 1 || COMPAT_PRELITE2 */ #endif /* !BURN_BRIDGES */ #define KINFO_VNODESLOP 10 #ifdef notyet /* * Dump vnode list (via sysctl). */ /* ARGSUSED */ static int sysctl_vnode(SYSCTL_HANDLER_ARGS) { struct xvnode *xvn; struct mount *mp; struct vnode *vp; int error, len, n; /* * Stale numvnodes access is not fatal here. */ req->lock = 0; len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn; if (!req->oldptr) /* Make an estimate */ return (SYSCTL_OUT(req, 0, len)); error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK); n = 0; mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) continue; MNT_ILOCK(mp); TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { if (n == len) break; vref(vp); xvn[n].xv_size = sizeof *xvn; xvn[n].xv_vnode = vp; xvn[n].xv_id = 0; /* XXX compat */ #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field XV_COPY(usecount); XV_COPY(writecount); XV_COPY(holdcnt); XV_COPY(mount); XV_COPY(numoutput); XV_COPY(type); #undef XV_COPY xvn[n].xv_flag = vp->v_vflag; switch (vp->v_type) { case VREG: case VDIR: case VLNK: break; case VBLK: case VCHR: if (vp->v_rdev == NULL) { vrele(vp); continue; } xvn[n].xv_dev = dev2udev(vp->v_rdev); break; case VSOCK: xvn[n].xv_socket = vp->v_socket; break; case VFIFO: xvn[n].xv_fifo = vp->v_fifoinfo; break; case VNON: case VBAD: default: /* shouldn't happen? */ vrele(vp); continue; } vrele(vp); ++n; } MNT_IUNLOCK(mp); mtx_lock(&mountlist_mtx); vfs_unbusy(mp); if (n == len) break; } mtx_unlock(&mountlist_mtx); error = SYSCTL_OUT(req, xvn, n * sizeof *xvn); free(xvn, M_TEMP); return (error); } SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 0, sysctl_vnode, "S,xvnode", ""); #endif static void unmount_or_warn(struct mount *mp) { int error; error = dounmount(mp, MNT_FORCE, curthread); if (error != 0) { printf("unmount of %s failed (", mp->mnt_stat.f_mntonname); if (error == EBUSY) printf("BUSY)\n"); else printf("%d)\n", error); } } /* * Unmount all filesystems. The list is traversed in reverse order * of mounting to avoid dependencies. */ void vfs_unmountall(void) { struct mount *mp, *tmp; CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__); /* * Since this only runs when rebooting, it is not interlocked. */ TAILQ_FOREACH_REVERSE_SAFE(mp, &mountlist, mntlist, mnt_list, tmp) { vfs_ref(mp); /* * Forcibly unmounting "/dev" before "/" would prevent clean * unmount of the latter. */ if (mp == rootdevmp) continue; unmount_or_warn(mp); } if (rootdevmp != NULL) unmount_or_warn(rootdevmp); } static void vfs_deferred_inactive(struct vnode *vp, int lkflags) { ASSERT_VI_LOCKED(vp, __func__); VNASSERT((vp->v_iflag & VI_DEFINACT) == 0, vp, ("VI_DEFINACT still set")); if ((vp->v_iflag & VI_OWEINACT) == 0) { vdropl(vp); return; } if (vn_lock(vp, lkflags) == 0) { VI_LOCK(vp); vinactive(vp); VOP_UNLOCK(vp); vdropl(vp); return; } vdefer_inactive_unlocked(vp); } static int vfs_periodic_inactive_filter(struct vnode *vp, void *arg) { return (vp->v_iflag & VI_DEFINACT); } static void __noinline vfs_periodic_inactive(struct mount *mp, int flags) { struct vnode *vp, *mvp; int lkflags; lkflags = LK_EXCLUSIVE | LK_INTERLOCK; if (flags != MNT_WAIT) lkflags |= LK_NOWAIT; MNT_VNODE_FOREACH_LAZY(vp, mp, mvp, vfs_periodic_inactive_filter, NULL) { if ((vp->v_iflag & VI_DEFINACT) == 0) { VI_UNLOCK(vp); continue; } vp->v_iflag &= ~VI_DEFINACT; vfs_deferred_inactive(vp, lkflags); } } static inline bool vfs_want_msync(struct vnode *vp) { struct vm_object *obj; /* * This test may be performed without any locks held. * We rely on vm_object's type stability. */ if (vp->v_vflag & VV_NOSYNC) return (false); obj = vp->v_object; return (obj != NULL && vm_object_mightbedirty(obj)); } static int vfs_periodic_msync_inactive_filter(struct vnode *vp, void *arg __unused) { if (vp->v_vflag & VV_NOSYNC) return (false); if (vp->v_iflag & VI_DEFINACT) return (true); return (vfs_want_msync(vp)); } static void __noinline vfs_periodic_msync_inactive(struct mount *mp, int flags) { struct vnode *vp, *mvp; struct vm_object *obj; int lkflags, objflags; bool seen_defer; lkflags = LK_EXCLUSIVE | LK_INTERLOCK; if (flags != MNT_WAIT) { lkflags |= LK_NOWAIT; objflags = OBJPC_NOSYNC; } else { objflags = OBJPC_SYNC; } MNT_VNODE_FOREACH_LAZY(vp, mp, mvp, vfs_periodic_msync_inactive_filter, NULL) { seen_defer = false; if (vp->v_iflag & VI_DEFINACT) { vp->v_iflag &= ~VI_DEFINACT; seen_defer = true; } if (!vfs_want_msync(vp)) { if (seen_defer) vfs_deferred_inactive(vp, lkflags); else VI_UNLOCK(vp); continue; } if (vget(vp, lkflags) == 0) { obj = vp->v_object; if (obj != NULL && (vp->v_vflag & VV_NOSYNC) == 0) { VM_OBJECT_WLOCK(obj); vm_object_page_clean(obj, 0, 0, objflags); VM_OBJECT_WUNLOCK(obj); } vput(vp); if (seen_defer) vdrop(vp); } else { if (seen_defer) vdefer_inactive_unlocked(vp); } } } void vfs_periodic(struct mount *mp, int flags) { CTR2(KTR_VFS, "%s: mp %p", __func__, mp); if ((mp->mnt_kern_flag & MNTK_NOMSYNC) != 0) vfs_periodic_inactive(mp, flags); else vfs_periodic_msync_inactive(mp, flags); } static void destroy_vpollinfo_free(struct vpollinfo *vi) { knlist_destroy(&vi->vpi_selinfo.si_note); mtx_destroy(&vi->vpi_lock); free(vi, M_VNODEPOLL); } static void destroy_vpollinfo(struct vpollinfo *vi) { knlist_clear(&vi->vpi_selinfo.si_note, 1); seldrain(&vi->vpi_selinfo); destroy_vpollinfo_free(vi); } /* * Initialize per-vnode helper structure to hold poll-related state. */ void v_addpollinfo(struct vnode *vp) { struct vpollinfo *vi; if (vp->v_pollinfo != NULL) return; vi = malloc(sizeof(*vi), M_VNODEPOLL, M_WAITOK | M_ZERO); mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF); knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock, vfs_knlunlock, vfs_knl_assert_lock); VI_LOCK(vp); if (vp->v_pollinfo != NULL) { VI_UNLOCK(vp); destroy_vpollinfo_free(vi); return; } vp->v_pollinfo = vi; VI_UNLOCK(vp); } /* * Record a process's interest in events which might happen to * a vnode. Because poll uses the historic select-style interface * internally, this routine serves as both the ``check for any * pending events'' and the ``record my interest in future events'' * functions. (These are done together, while the lock is held, * to avoid race conditions.) */ int vn_pollrecord(struct vnode *vp, struct thread *td, int events) { v_addpollinfo(vp); mtx_lock(&vp->v_pollinfo->vpi_lock); if (vp->v_pollinfo->vpi_revents & events) { /* * This leaves events we are not interested * in available for the other process which * which presumably had requested them * (otherwise they would never have been * recorded). */ events &= vp->v_pollinfo->vpi_revents; vp->v_pollinfo->vpi_revents &= ~events; mtx_unlock(&vp->v_pollinfo->vpi_lock); return (events); } vp->v_pollinfo->vpi_events |= events; selrecord(td, &vp->v_pollinfo->vpi_selinfo); mtx_unlock(&vp->v_pollinfo->vpi_lock); return (0); } /* * Routine to create and manage a filesystem syncer vnode. */ #define sync_close ((int (*)(struct vop_close_args *))nullop) static int sync_fsync(struct vop_fsync_args *); static int sync_inactive(struct vop_inactive_args *); static int sync_reclaim(struct vop_reclaim_args *); static struct vop_vector sync_vnodeops = { .vop_bypass = VOP_EOPNOTSUPP, .vop_close = sync_close, /* close */ .vop_fsync = sync_fsync, /* fsync */ .vop_inactive = sync_inactive, /* inactive */ .vop_need_inactive = vop_stdneed_inactive, /* need_inactive */ .vop_reclaim = sync_reclaim, /* reclaim */ .vop_lock1 = vop_stdlock, /* lock */ .vop_unlock = vop_stdunlock, /* unlock */ .vop_islocked = vop_stdislocked, /* islocked */ }; VFS_VOP_VECTOR_REGISTER(sync_vnodeops); /* * Create a new filesystem syncer vnode for the specified mount point. */ void vfs_allocate_syncvnode(struct mount *mp) { struct vnode *vp; struct bufobj *bo; static long start, incr, next; int error; /* Allocate a new vnode */ error = getnewvnode("syncer", mp, &sync_vnodeops, &vp); if (error != 0) panic("vfs_allocate_syncvnode: getnewvnode() failed"); vp->v_type = VNON; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); vp->v_vflag |= VV_FORCEINSMQ; error = insmntque(vp, mp); if (error != 0) panic("vfs_allocate_syncvnode: insmntque() failed"); vp->v_vflag &= ~VV_FORCEINSMQ; VOP_UNLOCK(vp); /* * Place the vnode onto the syncer worklist. We attempt to * scatter them about on the list so that they will go off * at evenly distributed times even if all the filesystems * are mounted at once. */ next += incr; if (next == 0 || next > syncer_maxdelay) { start /= 2; incr /= 2; if (start == 0) { start = syncer_maxdelay / 2; incr = syncer_maxdelay; } next = start; } bo = &vp->v_bufobj; BO_LOCK(bo); vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0); /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */ mtx_lock(&sync_mtx); sync_vnode_count++; if (mp->mnt_syncer == NULL) { mp->mnt_syncer = vp; vp = NULL; } mtx_unlock(&sync_mtx); BO_UNLOCK(bo); if (vp != NULL) { vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); vgone(vp); vput(vp); } } void vfs_deallocate_syncvnode(struct mount *mp) { struct vnode *vp; mtx_lock(&sync_mtx); vp = mp->mnt_syncer; if (vp != NULL) mp->mnt_syncer = NULL; mtx_unlock(&sync_mtx); if (vp != NULL) vrele(vp); } /* * Do a lazy sync of the filesystem. */ static int sync_fsync(struct vop_fsync_args *ap) { struct vnode *syncvp = ap->a_vp; struct mount *mp = syncvp->v_mount; int error, save; struct bufobj *bo; /* * We only need to do something if this is a lazy evaluation. */ if (ap->a_waitfor != MNT_LAZY) return (0); /* * Move ourselves to the back of the sync list. */ bo = &syncvp->v_bufobj; BO_LOCK(bo); vn_syncer_add_to_worklist(bo, syncdelay); BO_UNLOCK(bo); /* * Walk the list of vnodes pushing all that are dirty and * not already on the sync list. */ if (vfs_busy(mp, MBF_NOWAIT) != 0) return (0); if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) { vfs_unbusy(mp); return (0); } save = curthread_pflags_set(TDP_SYNCIO); /* * The filesystem at hand may be idle with free vnodes stored in the * batch. Return them instead of letting them stay there indefinitely. */ vfs_periodic(mp, MNT_NOWAIT); error = VFS_SYNC(mp, MNT_LAZY); curthread_pflags_restore(save); vn_finished_write(mp); vfs_unbusy(mp); return (error); } /* * The syncer vnode is no referenced. */ static int sync_inactive(struct vop_inactive_args *ap) { vgone(ap->a_vp); return (0); } /* * The syncer vnode is no longer needed and is being decommissioned. * * Modifications to the worklist must be protected by sync_mtx. */ static int sync_reclaim(struct vop_reclaim_args *ap) { struct vnode *vp = ap->a_vp; struct bufobj *bo; bo = &vp->v_bufobj; BO_LOCK(bo); mtx_lock(&sync_mtx); if (vp->v_mount->mnt_syncer == vp) vp->v_mount->mnt_syncer = NULL; if (bo->bo_flag & BO_ONWORKLST) { LIST_REMOVE(bo, bo_synclist); syncer_worklist_len--; sync_vnode_count--; bo->bo_flag &= ~BO_ONWORKLST; } mtx_unlock(&sync_mtx); BO_UNLOCK(bo); return (0); } int vn_need_pageq_flush(struct vnode *vp) { struct vm_object *obj; obj = vp->v_object; return (obj != NULL && (vp->v_vflag & VV_NOSYNC) == 0 && vm_object_mightbedirty(obj)); } /* * Check if vnode represents a disk device */ bool vn_isdisk_error(struct vnode *vp, int *errp) { int error; if (vp->v_type != VCHR) { error = ENOTBLK; goto out; } error = 0; dev_lock(); if (vp->v_rdev == NULL) error = ENXIO; else if (vp->v_rdev->si_devsw == NULL) error = ENXIO; else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK)) error = ENOTBLK; dev_unlock(); out: *errp = error; return (error == 0); } bool vn_isdisk(struct vnode *vp) { int error; return (vn_isdisk_error(vp, &error)); } /* * VOP_FPLOOKUP_VEXEC routines are subject to special circumstances, see * the comment above cache_fplookup for details. */ int vaccess_vexec_smr(mode_t file_mode, uid_t file_uid, gid_t file_gid, struct ucred *cred) { int error; VFS_SMR_ASSERT_ENTERED(); /* Check the owner. */ if (cred->cr_uid == file_uid) { if (file_mode & S_IXUSR) return (0); goto out_error; } /* Otherwise, check the groups (first match) */ if (groupmember(file_gid, cred)) { if (file_mode & S_IXGRP) return (0); goto out_error; } /* Otherwise, check everyone else. */ if (file_mode & S_IXOTH) return (0); out_error: /* * Permission check failed, but it is possible denial will get overwritten * (e.g., when root is traversing through a 700 directory owned by someone * else). * * vaccess() calls priv_check_cred which in turn can descent into MAC * modules overriding this result. It's quite unclear what semantics * are allowed for them to operate, thus for safety we don't call them * from within the SMR section. This also means if any such modules * are present, we have to let the regular lookup decide. */ error = priv_check_cred_vfs_lookup_nomac(cred); switch (error) { case 0: return (0); case EAGAIN: /* * MAC modules present. */ return (EAGAIN); case EPERM: return (EACCES); default: return (error); } } /* * Common filesystem object access control check routine. Accepts a * vnode's type, "mode", uid and gid, requested access mode, and credentials. * Returns 0 on success, or an errno on failure. */ int vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid, accmode_t accmode, struct ucred *cred) { accmode_t dac_granted; accmode_t priv_granted; KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0, ("invalid bit in accmode")); KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE), ("VAPPEND without VWRITE")); /* * Look for a normal, non-privileged way to access the file/directory * as requested. If it exists, go with that. */ dac_granted = 0; /* Check the owner. */ if (cred->cr_uid == file_uid) { dac_granted |= VADMIN; if (file_mode & S_IXUSR) dac_granted |= VEXEC; if (file_mode & S_IRUSR) dac_granted |= VREAD; if (file_mode & S_IWUSR) dac_granted |= (VWRITE | VAPPEND); if ((accmode & dac_granted) == accmode) return (0); goto privcheck; } /* Otherwise, check the groups (first match) */ if (groupmember(file_gid, cred)) { if (file_mode & S_IXGRP) dac_granted |= VEXEC; if (file_mode & S_IRGRP) dac_granted |= VREAD; if (file_mode & S_IWGRP) dac_granted |= (VWRITE | VAPPEND); if ((accmode & dac_granted) == accmode) return (0); goto privcheck; } /* Otherwise, check everyone else. */ if (file_mode & S_IXOTH) dac_granted |= VEXEC; if (file_mode & S_IROTH) dac_granted |= VREAD; if (file_mode & S_IWOTH) dac_granted |= (VWRITE | VAPPEND); if ((accmode & dac_granted) == accmode) return (0); privcheck: /* * Build a privilege mask to determine if the set of privileges * satisfies the requirements when combined with the granted mask * from above. For each privilege, if the privilege is required, * bitwise or the request type onto the priv_granted mask. */ priv_granted = 0; if (type == VDIR) { /* * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC * requests, instead of PRIV_VFS_EXEC. */ if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) && !priv_check_cred(cred, PRIV_VFS_LOOKUP)) priv_granted |= VEXEC; } else { /* * Ensure that at least one execute bit is on. Otherwise, * a privileged user will always succeed, and we don't want * this to happen unless the file really is executable. */ if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) && (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 && !priv_check_cred(cred, PRIV_VFS_EXEC)) priv_granted |= VEXEC; } if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) && !priv_check_cred(cred, PRIV_VFS_READ)) priv_granted |= VREAD; if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) && !priv_check_cred(cred, PRIV_VFS_WRITE)) priv_granted |= (VWRITE | VAPPEND); if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) && !priv_check_cred(cred, PRIV_VFS_ADMIN)) priv_granted |= VADMIN; if ((accmode & (priv_granted | dac_granted)) == accmode) { return (0); } return ((accmode & VADMIN) ? EPERM : EACCES); } /* * Credential check based on process requesting service, and per-attribute * permissions. */ int extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred, struct thread *td, accmode_t accmode) { /* * Kernel-invoked always succeeds. */ if (cred == NOCRED) return (0); /* * Do not allow privileged processes in jail to directly manipulate * system attributes. */ switch (attrnamespace) { case EXTATTR_NAMESPACE_SYSTEM: /* Potentially should be: return (EPERM); */ return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM)); case EXTATTR_NAMESPACE_USER: return (VOP_ACCESS(vp, accmode, cred, td)); default: return (EPERM); } } #ifdef DEBUG_VFS_LOCKS /* * This only exists to suppress warnings from unlocked specfs accesses. It is * no longer ok to have an unlocked VFS. */ #define IGNORE_LOCK(vp) (KERNEL_PANICKED() || (vp) == NULL || \ (vp)->v_type == VCHR || (vp)->v_type == VBAD) int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, "Drop into debugger on lock violation"); int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, "Check for interlock across VOPs"); int vfs_badlock_print = 1; /* Print lock violations. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, "Print lock violations"); int vfs_badlock_vnode = 1; /* Print vnode details on lock violations. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_vnode, CTLFLAG_RW, &vfs_badlock_vnode, 0, "Print vnode details on lock violations"); #ifdef KDB int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, "Print backtrace at lock violations"); #endif static void vfs_badlock(const char *msg, const char *str, struct vnode *vp) { #ifdef KDB if (vfs_badlock_backtrace) kdb_backtrace(); #endif if (vfs_badlock_vnode) vn_printf(vp, "vnode "); if (vfs_badlock_print) printf("%s: %p %s\n", str, (void *)vp, msg); if (vfs_badlock_ddb) kdb_enter(KDB_WHY_VFSLOCK, "lock violation"); } void assert_vi_locked(struct vnode *vp, const char *str) { if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp))) vfs_badlock("interlock is not locked but should be", str, vp); } void assert_vi_unlocked(struct vnode *vp, const char *str) { if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp))) vfs_badlock("interlock is locked but should not be", str, vp); } void assert_vop_locked(struct vnode *vp, const char *str) { int locked; if (!IGNORE_LOCK(vp)) { locked = VOP_ISLOCKED(vp); if (locked == 0 || locked == LK_EXCLOTHER) vfs_badlock("is not locked but should be", str, vp); } } void assert_vop_unlocked(struct vnode *vp, const char *str) { if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) vfs_badlock("is locked but should not be", str, vp); } void assert_vop_elocked(struct vnode *vp, const char *str) { if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) vfs_badlock("is not exclusive locked but should be", str, vp); } #endif /* DEBUG_VFS_LOCKS */ void vop_rename_fail(struct vop_rename_args *ap) { if (ap->a_tvp != NULL) vput(ap->a_tvp); if (ap->a_tdvp == ap->a_tvp) vrele(ap->a_tdvp); else vput(ap->a_tdvp); vrele(ap->a_fdvp); vrele(ap->a_fvp); } void vop_rename_pre(void *ap) { struct vop_rename_args *a = ap; #ifdef DEBUG_VFS_LOCKS if (a->a_tvp) ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME"); ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME"); ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME"); ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME"); /* Check the source (from). */ if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock && (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock)) ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked"); if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock) ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked"); /* Check the target. */ if (a->a_tvp) ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked"); ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked"); #endif /* * It may be tempting to add vn_seqc_write_begin/end calls here and * in vop_rename_post but that's not going to work out since some * filesystems relookup vnodes mid-rename. This is probably a bug. * * For now filesystems are expected to do the relevant calls after they * decide what vnodes to operate on. */ if (a->a_tdvp != a->a_fdvp) vhold(a->a_fdvp); if (a->a_tvp != a->a_fvp) vhold(a->a_fvp); vhold(a->a_tdvp); if (a->a_tvp) vhold(a->a_tvp); } #ifdef DEBUG_VFS_LOCKS void vop_fplookup_vexec_debugpre(void *ap __unused) { VFS_SMR_ASSERT_ENTERED(); } void vop_fplookup_vexec_debugpost(void *ap __unused, int rc __unused) { VFS_SMR_ASSERT_ENTERED(); } void vop_fplookup_symlink_debugpre(void *ap __unused) { VFS_SMR_ASSERT_ENTERED(); } void vop_fplookup_symlink_debugpost(void *ap __unused, int rc __unused) { VFS_SMR_ASSERT_ENTERED(); } void vop_strategy_debugpre(void *ap) { struct vop_strategy_args *a; struct buf *bp; a = ap; bp = a->a_bp; /* * Cluster ops lock their component buffers but not the IO container. */ if ((bp->b_flags & B_CLUSTER) != 0) return; if (!KERNEL_PANICKED() && !BUF_ISLOCKED(bp)) { if (vfs_badlock_print) printf( "VOP_STRATEGY: bp is not locked but should be\n"); if (vfs_badlock_ddb) kdb_enter(KDB_WHY_VFSLOCK, "lock violation"); } } void vop_lock_debugpre(void *ap) { struct vop_lock1_args *a = ap; if ((a->a_flags & LK_INTERLOCK) == 0) ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); else ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK"); } void vop_lock_debugpost(void *ap, int rc) { struct vop_lock1_args *a = ap; ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); if (rc == 0 && (a->a_flags & LK_EXCLOTHER) == 0) ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK"); } void vop_unlock_debugpre(void *ap) { struct vop_unlock_args *a = ap; ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK"); } void vop_need_inactive_debugpre(void *ap) { struct vop_need_inactive_args *a = ap; ASSERT_VI_LOCKED(a->a_vp, "VOP_NEED_INACTIVE"); } void vop_need_inactive_debugpost(void *ap, int rc) { struct vop_need_inactive_args *a = ap; ASSERT_VI_LOCKED(a->a_vp, "VOP_NEED_INACTIVE"); } #endif void vop_create_pre(void *ap) { struct vop_create_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_begin(dvp); } void vop_create_post(void *ap, int rc) { struct vop_create_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_end(dvp); if (!rc) VFS_KNOTE_LOCKED(dvp, NOTE_WRITE); } void vop_whiteout_pre(void *ap) { struct vop_whiteout_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_begin(dvp); } void vop_whiteout_post(void *ap, int rc) { struct vop_whiteout_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_end(dvp); } void vop_deleteextattr_pre(void *ap) { struct vop_deleteextattr_args *a; struct vnode *vp; a = ap; vp = a->a_vp; vn_seqc_write_begin(vp); } void vop_deleteextattr_post(void *ap, int rc) { struct vop_deleteextattr_args *a; struct vnode *vp; a = ap; vp = a->a_vp; vn_seqc_write_end(vp); if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB); } void vop_link_pre(void *ap) { struct vop_link_args *a; struct vnode *vp, *tdvp; a = ap; vp = a->a_vp; tdvp = a->a_tdvp; vn_seqc_write_begin(vp); vn_seqc_write_begin(tdvp); } void vop_link_post(void *ap, int rc) { struct vop_link_args *a; struct vnode *vp, *tdvp; a = ap; vp = a->a_vp; tdvp = a->a_tdvp; vn_seqc_write_end(vp); vn_seqc_write_end(tdvp); if (!rc) { VFS_KNOTE_LOCKED(vp, NOTE_LINK); VFS_KNOTE_LOCKED(tdvp, NOTE_WRITE); } } void vop_mkdir_pre(void *ap) { struct vop_mkdir_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_begin(dvp); } void vop_mkdir_post(void *ap, int rc) { struct vop_mkdir_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_end(dvp); if (!rc) VFS_KNOTE_LOCKED(dvp, NOTE_WRITE | NOTE_LINK); } #ifdef DEBUG_VFS_LOCKS void vop_mkdir_debugpost(void *ap, int rc) { struct vop_mkdir_args *a; a = ap; if (!rc) cache_validate(a->a_dvp, *a->a_vpp, a->a_cnp); } #endif void vop_mknod_pre(void *ap) { struct vop_mknod_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_begin(dvp); } void vop_mknod_post(void *ap, int rc) { struct vop_mknod_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_end(dvp); if (!rc) VFS_KNOTE_LOCKED(dvp, NOTE_WRITE); } void vop_reclaim_post(void *ap, int rc) { struct vop_reclaim_args *a; struct vnode *vp; a = ap; vp = a->a_vp; ASSERT_VOP_IN_SEQC(vp); if (!rc) VFS_KNOTE_LOCKED(vp, NOTE_REVOKE); } void vop_remove_pre(void *ap) { struct vop_remove_args *a; struct vnode *dvp, *vp; a = ap; dvp = a->a_dvp; vp = a->a_vp; vn_seqc_write_begin(dvp); vn_seqc_write_begin(vp); } void vop_remove_post(void *ap, int rc) { struct vop_remove_args *a; struct vnode *dvp, *vp; a = ap; dvp = a->a_dvp; vp = a->a_vp; vn_seqc_write_end(dvp); vn_seqc_write_end(vp); if (!rc) { VFS_KNOTE_LOCKED(dvp, NOTE_WRITE); VFS_KNOTE_LOCKED(vp, NOTE_DELETE); } } void vop_rename_post(void *ap, int rc) { struct vop_rename_args *a = ap; long hint; if (!rc) { hint = NOTE_WRITE; if (a->a_fdvp == a->a_tdvp) { if (a->a_tvp != NULL && a->a_tvp->v_type == VDIR) hint |= NOTE_LINK; VFS_KNOTE_UNLOCKED(a->a_fdvp, hint); VFS_KNOTE_UNLOCKED(a->a_tdvp, hint); } else { hint |= NOTE_EXTEND; if (a->a_fvp->v_type == VDIR) hint |= NOTE_LINK; VFS_KNOTE_UNLOCKED(a->a_fdvp, hint); if (a->a_fvp->v_type == VDIR && a->a_tvp != NULL && a->a_tvp->v_type == VDIR) hint &= ~NOTE_LINK; VFS_KNOTE_UNLOCKED(a->a_tdvp, hint); } VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME); if (a->a_tvp) VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE); } if (a->a_tdvp != a->a_fdvp) vdrop(a->a_fdvp); if (a->a_tvp != a->a_fvp) vdrop(a->a_fvp); vdrop(a->a_tdvp); if (a->a_tvp) vdrop(a->a_tvp); } void vop_rmdir_pre(void *ap) { struct vop_rmdir_args *a; struct vnode *dvp, *vp; a = ap; dvp = a->a_dvp; vp = a->a_vp; vn_seqc_write_begin(dvp); vn_seqc_write_begin(vp); } void vop_rmdir_post(void *ap, int rc) { struct vop_rmdir_args *a; struct vnode *dvp, *vp; a = ap; dvp = a->a_dvp; vp = a->a_vp; vn_seqc_write_end(dvp); vn_seqc_write_end(vp); if (!rc) { VFS_KNOTE_LOCKED(dvp, NOTE_WRITE | NOTE_LINK); VFS_KNOTE_LOCKED(vp, NOTE_DELETE); } } void vop_setattr_pre(void *ap) { struct vop_setattr_args *a; struct vnode *vp; a = ap; vp = a->a_vp; vn_seqc_write_begin(vp); } void vop_setattr_post(void *ap, int rc) { struct vop_setattr_args *a; struct vnode *vp; a = ap; vp = a->a_vp; vn_seqc_write_end(vp); if (!rc) VFS_KNOTE_LOCKED(vp, NOTE_ATTRIB); } void vop_setacl_pre(void *ap) { struct vop_setacl_args *a; struct vnode *vp; a = ap; vp = a->a_vp; vn_seqc_write_begin(vp); } void vop_setacl_post(void *ap, int rc __unused) { struct vop_setacl_args *a; struct vnode *vp; a = ap; vp = a->a_vp; vn_seqc_write_end(vp); } void vop_setextattr_pre(void *ap) { struct vop_setextattr_args *a; struct vnode *vp; a = ap; vp = a->a_vp; vn_seqc_write_begin(vp); } void vop_setextattr_post(void *ap, int rc) { struct vop_setextattr_args *a; struct vnode *vp; a = ap; vp = a->a_vp; vn_seqc_write_end(vp); if (!rc) VFS_KNOTE_LOCKED(vp, NOTE_ATTRIB); } void vop_symlink_pre(void *ap) { struct vop_symlink_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_begin(dvp); } void vop_symlink_post(void *ap, int rc) { struct vop_symlink_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_end(dvp); if (!rc) VFS_KNOTE_LOCKED(dvp, NOTE_WRITE); } void vop_open_post(void *ap, int rc) { struct vop_open_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_OPEN); } void vop_close_post(void *ap, int rc) { struct vop_close_args *a = ap; if (!rc && (a->a_cred != NOCRED || /* filter out revokes */ !VN_IS_DOOMED(a->a_vp))) { VFS_KNOTE_LOCKED(a->a_vp, (a->a_fflag & FWRITE) != 0 ? NOTE_CLOSE_WRITE : NOTE_CLOSE); } } void vop_read_post(void *ap, int rc) { struct vop_read_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_READ); } void vop_read_pgcache_post(void *ap, int rc) { struct vop_read_pgcache_args *a = ap; if (!rc) VFS_KNOTE_UNLOCKED(a->a_vp, NOTE_READ); } void vop_readdir_post(void *ap, int rc) { struct vop_readdir_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_READ); } static struct knlist fs_knlist; static void vfs_event_init(void *arg) { knlist_init_mtx(&fs_knlist, NULL); } /* XXX - correct order? */ SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL); void vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused) { KNOTE_UNLOCKED(&fs_knlist, event); } static int filt_fsattach(struct knote *kn); static void filt_fsdetach(struct knote *kn); static int filt_fsevent(struct knote *kn, long hint); struct filterops fs_filtops = { .f_isfd = 0, .f_attach = filt_fsattach, .f_detach = filt_fsdetach, .f_event = filt_fsevent }; static int filt_fsattach(struct knote *kn) { kn->kn_flags |= EV_CLEAR; knlist_add(&fs_knlist, kn, 0); return (0); } static void filt_fsdetach(struct knote *kn) { knlist_remove(&fs_knlist, kn, 0); } static int filt_fsevent(struct knote *kn, long hint) { kn->kn_fflags |= kn->kn_sfflags & hint; return (kn->kn_fflags != 0); } static int sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS) { struct vfsidctl vc; int error; struct mount *mp; error = SYSCTL_IN(req, &vc, sizeof(vc)); if (error) return (error); if (vc.vc_vers != VFS_CTL_VERS1) return (EINVAL); mp = vfs_getvfs(&vc.vc_fsid); if (mp == NULL) return (ENOENT); /* ensure that a specific sysctl goes to the right filesystem. */ if (strcmp(vc.vc_fstypename, "*") != 0 && strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) { vfs_rel(mp); return (EINVAL); } VCTLTOREQ(&vc, req); error = VFS_SYSCTL(mp, vc.vc_op, req); vfs_rel(mp); return (error); } SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_MPSAFE | CTLFLAG_WR, NULL, 0, sysctl_vfs_ctl, "", "Sysctl by fsid"); /* * Function to initialize a va_filerev field sensibly. * XXX: Wouldn't a random number make a lot more sense ?? */ u_quad_t init_va_filerev(void) { struct bintime bt; getbinuptime(&bt); return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL)); } static int filt_vfsread(struct knote *kn, long hint); static int filt_vfswrite(struct knote *kn, long hint); static int filt_vfsvnode(struct knote *kn, long hint); static void filt_vfsdetach(struct knote *kn); static struct filterops vfsread_filtops = { .f_isfd = 1, .f_detach = filt_vfsdetach, .f_event = filt_vfsread }; static struct filterops vfswrite_filtops = { .f_isfd = 1, .f_detach = filt_vfsdetach, .f_event = filt_vfswrite }; static struct filterops vfsvnode_filtops = { .f_isfd = 1, .f_detach = filt_vfsdetach, .f_event = filt_vfsvnode }; static void vfs_knllock(void *arg) { struct vnode *vp = arg; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); } static void vfs_knlunlock(void *arg) { struct vnode *vp = arg; VOP_UNLOCK(vp); } static void vfs_knl_assert_lock(void *arg, int what) { #ifdef DEBUG_VFS_LOCKS struct vnode *vp = arg; if (what == LA_LOCKED) ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked"); else ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked"); #endif } int vfs_kqfilter(struct vop_kqfilter_args *ap) { struct vnode *vp = ap->a_vp; struct knote *kn = ap->a_kn; struct knlist *knl; switch (kn->kn_filter) { case EVFILT_READ: kn->kn_fop = &vfsread_filtops; break; case EVFILT_WRITE: kn->kn_fop = &vfswrite_filtops; break; case EVFILT_VNODE: kn->kn_fop = &vfsvnode_filtops; break; default: return (EINVAL); } kn->kn_hook = (caddr_t)vp; v_addpollinfo(vp); if (vp->v_pollinfo == NULL) return (ENOMEM); knl = &vp->v_pollinfo->vpi_selinfo.si_note; vhold(vp); knlist_add(knl, kn, 0); return (0); } /* * Detach knote from vnode */ static void filt_vfsdetach(struct knote *kn) { struct vnode *vp = (struct vnode *)kn->kn_hook; KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo")); knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0); vdrop(vp); } /*ARGSUSED*/ static int filt_vfsread(struct knote *kn, long hint) { struct vnode *vp = (struct vnode *)kn->kn_hook; struct vattr va; int res; /* * filesystem is gone, so set the EOF flag and schedule * the knote for deletion. */ if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) { VI_LOCK(vp); kn->kn_flags |= (EV_EOF | EV_ONESHOT); VI_UNLOCK(vp); return (1); } if (VOP_GETATTR(vp, &va, curthread->td_ucred)) return (0); VI_LOCK(vp); kn->kn_data = va.va_size - kn->kn_fp->f_offset; res = (kn->kn_sfflags & NOTE_FILE_POLL) != 0 || kn->kn_data != 0; VI_UNLOCK(vp); return (res); } /*ARGSUSED*/ static int filt_vfswrite(struct knote *kn, long hint) { struct vnode *vp = (struct vnode *)kn->kn_hook; VI_LOCK(vp); /* * filesystem is gone, so set the EOF flag and schedule * the knote for deletion. */ if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) kn->kn_flags |= (EV_EOF | EV_ONESHOT); kn->kn_data = 0; VI_UNLOCK(vp); return (1); } static int filt_vfsvnode(struct knote *kn, long hint) { struct vnode *vp = (struct vnode *)kn->kn_hook; int res; VI_LOCK(vp); if (kn->kn_sfflags & hint) kn->kn_fflags |= hint; if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) { kn->kn_flags |= EV_EOF; VI_UNLOCK(vp); return (1); } res = (kn->kn_fflags != 0); VI_UNLOCK(vp); return (res); } /* * Returns whether the directory is empty or not. * If it is empty, the return value is 0; otherwise * the return value is an error value (which may * be ENOTEMPTY). */ int vfs_emptydir(struct vnode *vp) { struct uio uio; struct iovec iov; struct dirent *dirent, *dp, *endp; int error, eof; error = 0; eof = 0; ASSERT_VOP_LOCKED(vp, "vfs_emptydir"); dirent = malloc(sizeof(struct dirent), M_TEMP, M_WAITOK); iov.iov_base = dirent; iov.iov_len = sizeof(struct dirent); uio.uio_iov = &iov; uio.uio_iovcnt = 1; uio.uio_offset = 0; uio.uio_resid = sizeof(struct dirent); uio.uio_segflg = UIO_SYSSPACE; uio.uio_rw = UIO_READ; uio.uio_td = curthread; while (eof == 0 && error == 0) { error = VOP_READDIR(vp, &uio, curthread->td_ucred, &eof, NULL, NULL); if (error != 0) break; endp = (void *)((uint8_t *)dirent + sizeof(struct dirent) - uio.uio_resid); for (dp = dirent; dp < endp; dp = (void *)((uint8_t *)dp + GENERIC_DIRSIZ(dp))) { if (dp->d_type == DT_WHT) continue; if (dp->d_namlen == 0) continue; if (dp->d_type != DT_DIR && dp->d_type != DT_UNKNOWN) { error = ENOTEMPTY; break; } if (dp->d_namlen > 2) { error = ENOTEMPTY; break; } if (dp->d_namlen == 1 && dp->d_name[0] != '.') { error = ENOTEMPTY; break; } if (dp->d_namlen == 2 && dp->d_name[1] != '.') { error = ENOTEMPTY; break; } uio.uio_resid = sizeof(struct dirent); } } free(dirent, M_TEMP); return (error); } int vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off) { int error; if (dp->d_reclen > ap->a_uio->uio_resid) return (ENAMETOOLONG); error = uiomove(dp, dp->d_reclen, ap->a_uio); if (error) { if (ap->a_ncookies != NULL) { if (ap->a_cookies != NULL) free(ap->a_cookies, M_TEMP); ap->a_cookies = NULL; *ap->a_ncookies = 0; } return (error); } if (ap->a_ncookies == NULL) return (0); KASSERT(ap->a_cookies, ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!")); *ap->a_cookies = realloc(*ap->a_cookies, (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO); (*ap->a_cookies)[*ap->a_ncookies] = off; *ap->a_ncookies += 1; return (0); } /* * The purpose of this routine is to remove granularity from accmode_t, * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE, * VADMIN and VAPPEND. * * If it returns 0, the caller is supposed to continue with the usual * access checks using 'accmode' as modified by this routine. If it * returns nonzero value, the caller is supposed to return that value * as errno. * * Note that after this routine runs, accmode may be zero. */ int vfs_unixify_accmode(accmode_t *accmode) { /* * There is no way to specify explicit "deny" rule using * file mode or POSIX.1e ACLs. */ if (*accmode & VEXPLICIT_DENY) { *accmode = 0; return (0); } /* * None of these can be translated into usual access bits. * Also, the common case for NFSv4 ACLs is to not contain * either of these bits. Caller should check for VWRITE * on the containing directory instead. */ if (*accmode & (VDELETE_CHILD | VDELETE)) return (EPERM); if (*accmode & VADMIN_PERMS) { *accmode &= ~VADMIN_PERMS; *accmode |= VADMIN; } /* * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL * or VSYNCHRONIZE using file mode or POSIX.1e ACL. */ *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE); return (0); } /* * Clear out a doomed vnode (if any) and replace it with a new one as long * as the fs is not being unmounted. Return the root vnode to the caller. */ static int __noinline vfs_cache_root_fallback(struct mount *mp, int flags, struct vnode **vpp) { struct vnode *vp; int error; restart: if (mp->mnt_rootvnode != NULL) { MNT_ILOCK(mp); vp = mp->mnt_rootvnode; if (vp != NULL) { if (!VN_IS_DOOMED(vp)) { vrefact(vp); MNT_IUNLOCK(mp); error = vn_lock(vp, flags); if (error == 0) { *vpp = vp; return (0); } vrele(vp); goto restart; } /* * Clear the old one. */ mp->mnt_rootvnode = NULL; } MNT_IUNLOCK(mp); if (vp != NULL) { vfs_op_barrier_wait(mp); vrele(vp); } } error = VFS_CACHEDROOT(mp, flags, vpp); if (error != 0) return (error); if (mp->mnt_vfs_ops == 0) { MNT_ILOCK(mp); if (mp->mnt_vfs_ops != 0) { MNT_IUNLOCK(mp); return (0); } if (mp->mnt_rootvnode == NULL) { vrefact(*vpp); mp->mnt_rootvnode = *vpp; } else { if (mp->mnt_rootvnode != *vpp) { if (!VN_IS_DOOMED(mp->mnt_rootvnode)) { panic("%s: mismatch between vnode returned " " by VFS_CACHEDROOT and the one cached " " (%p != %p)", __func__, *vpp, mp->mnt_rootvnode); } } } MNT_IUNLOCK(mp); } return (0); } int vfs_cache_root(struct mount *mp, int flags, struct vnode **vpp) { struct mount_pcpu *mpcpu; struct vnode *vp; int error; if (!vfs_op_thread_enter(mp, mpcpu)) return (vfs_cache_root_fallback(mp, flags, vpp)); vp = atomic_load_ptr(&mp->mnt_rootvnode); if (vp == NULL || VN_IS_DOOMED(vp)) { vfs_op_thread_exit(mp, mpcpu); return (vfs_cache_root_fallback(mp, flags, vpp)); } vrefact(vp); vfs_op_thread_exit(mp, mpcpu); error = vn_lock(vp, flags); if (error != 0) { vrele(vp); return (vfs_cache_root_fallback(mp, flags, vpp)); } *vpp = vp; return (0); } struct vnode * vfs_cache_root_clear(struct mount *mp) { struct vnode *vp; /* * ops > 0 guarantees there is nobody who can see this vnode */ MPASS(mp->mnt_vfs_ops > 0); vp = mp->mnt_rootvnode; if (vp != NULL) vn_seqc_write_begin(vp); mp->mnt_rootvnode = NULL; return (vp); } void vfs_cache_root_set(struct mount *mp, struct vnode *vp) { MPASS(mp->mnt_vfs_ops > 0); vrefact(vp); mp->mnt_rootvnode = vp; } /* * These are helper functions for filesystems to traverse all * their vnodes. See MNT_VNODE_FOREACH_ALL() in sys/mount.h. * * This interface replaces MNT_VNODE_FOREACH. */ struct vnode * __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp) { struct vnode *vp; if (should_yield()) kern_yield(PRI_USER); MNT_ILOCK(mp); KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); for (vp = TAILQ_NEXT(*mvp, v_nmntvnodes); vp != NULL; vp = TAILQ_NEXT(vp, v_nmntvnodes)) { /* Allow a racy peek at VIRF_DOOMED to save a lock acquisition. */ if (vp->v_type == VMARKER || VN_IS_DOOMED(vp)) continue; VI_LOCK(vp); if (VN_IS_DOOMED(vp)) { VI_UNLOCK(vp); continue; } break; } if (vp == NULL) { __mnt_vnode_markerfree_all(mvp, mp); /* MNT_IUNLOCK(mp); -- done in above function */ mtx_assert(MNT_MTX(mp), MA_NOTOWNED); return (NULL); } TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); MNT_IUNLOCK(mp); return (vp); } struct vnode * __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp) { struct vnode *vp; *mvp = vn_alloc_marker(mp); MNT_ILOCK(mp); MNT_REF(mp); TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { /* Allow a racy peek at VIRF_DOOMED to save a lock acquisition. */ if (vp->v_type == VMARKER || VN_IS_DOOMED(vp)) continue; VI_LOCK(vp); if (VN_IS_DOOMED(vp)) { VI_UNLOCK(vp); continue; } break; } if (vp == NULL) { MNT_REL(mp); MNT_IUNLOCK(mp); vn_free_marker(*mvp); *mvp = NULL; return (NULL); } TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); MNT_IUNLOCK(mp); return (vp); } void __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp) { if (*mvp == NULL) { MNT_IUNLOCK(mp); return; } mtx_assert(MNT_MTX(mp), MA_OWNED); KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); MNT_REL(mp); MNT_IUNLOCK(mp); vn_free_marker(*mvp); *mvp = NULL; } /* * These are helper functions for filesystems to traverse their * lazy vnodes. See MNT_VNODE_FOREACH_LAZY() in sys/mount.h */ static void mnt_vnode_markerfree_lazy(struct vnode **mvp, struct mount *mp) { KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); MNT_ILOCK(mp); MNT_REL(mp); MNT_IUNLOCK(mp); vn_free_marker(*mvp); *mvp = NULL; } /* * Relock the mp mount vnode list lock with the vp vnode interlock in the * conventional lock order during mnt_vnode_next_lazy iteration. * * On entry, the mount vnode list lock is held and the vnode interlock is not. * The list lock is dropped and reacquired. On success, both locks are held. * On failure, the mount vnode list lock is held but the vnode interlock is * not, and the procedure may have yielded. */ static bool mnt_vnode_next_lazy_relock(struct vnode *mvp, struct mount *mp, struct vnode *vp) { VNASSERT(mvp->v_mount == mp && mvp->v_type == VMARKER && TAILQ_NEXT(mvp, v_lazylist) != NULL, mvp, ("%s: bad marker", __func__)); VNASSERT(vp->v_mount == mp && vp->v_type != VMARKER, vp, ("%s: inappropriate vnode", __func__)); ASSERT_VI_UNLOCKED(vp, __func__); mtx_assert(&mp->mnt_listmtx, MA_OWNED); TAILQ_REMOVE(&mp->mnt_lazyvnodelist, mvp, v_lazylist); TAILQ_INSERT_BEFORE(vp, mvp, v_lazylist); /* * Note we may be racing against vdrop which transitioned the hold * count to 0 and now waits for the ->mnt_listmtx lock. This is fine, * if we are the only user after we get the interlock we will just * vdrop. */ vhold(vp); mtx_unlock(&mp->mnt_listmtx); VI_LOCK(vp); if (VN_IS_DOOMED(vp)) { VNPASS((vp->v_mflag & VMP_LAZYLIST) == 0, vp); goto out_lost; } VNPASS(vp->v_mflag & VMP_LAZYLIST, vp); /* * There is nothing to do if we are the last user. */ if (!refcount_release_if_not_last(&vp->v_holdcnt)) goto out_lost; mtx_lock(&mp->mnt_listmtx); return (true); out_lost: vdropl(vp); maybe_yield(); mtx_lock(&mp->mnt_listmtx); return (false); } static struct vnode * mnt_vnode_next_lazy(struct vnode **mvp, struct mount *mp, mnt_lazy_cb_t *cb, void *cbarg) { struct vnode *vp; mtx_assert(&mp->mnt_listmtx, MA_OWNED); KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); restart: vp = TAILQ_NEXT(*mvp, v_lazylist); while (vp != NULL) { if (vp->v_type == VMARKER) { vp = TAILQ_NEXT(vp, v_lazylist); continue; } /* * See if we want to process the vnode. Note we may encounter a * long string of vnodes we don't care about and hog the list * as a result. Check for it and requeue the marker. */ VNPASS(!VN_IS_DOOMED(vp), vp); if (!cb(vp, cbarg)) { if (!should_yield()) { vp = TAILQ_NEXT(vp, v_lazylist); continue; } TAILQ_REMOVE(&mp->mnt_lazyvnodelist, *mvp, v_lazylist); TAILQ_INSERT_AFTER(&mp->mnt_lazyvnodelist, vp, *mvp, v_lazylist); mtx_unlock(&mp->mnt_listmtx); kern_yield(PRI_USER); mtx_lock(&mp->mnt_listmtx); goto restart; } /* * Try-lock because this is the wrong lock order. */ if (!VI_TRYLOCK(vp) && !mnt_vnode_next_lazy_relock(*mvp, mp, vp)) goto restart; KASSERT(vp->v_type != VMARKER, ("locked marker %p", vp)); KASSERT(vp->v_mount == mp || vp->v_mount == NULL, ("alien vnode on the lazy list %p %p", vp, mp)); VNPASS(vp->v_mount == mp, vp); VNPASS(!VN_IS_DOOMED(vp), vp); break; } TAILQ_REMOVE(&mp->mnt_lazyvnodelist, *mvp, v_lazylist); /* Check if we are done */ if (vp == NULL) { mtx_unlock(&mp->mnt_listmtx); mnt_vnode_markerfree_lazy(mvp, mp); return (NULL); } TAILQ_INSERT_AFTER(&mp->mnt_lazyvnodelist, vp, *mvp, v_lazylist); mtx_unlock(&mp->mnt_listmtx); ASSERT_VI_LOCKED(vp, "lazy iter"); return (vp); } struct vnode * __mnt_vnode_next_lazy(struct vnode **mvp, struct mount *mp, mnt_lazy_cb_t *cb, void *cbarg) { if (should_yield()) kern_yield(PRI_USER); mtx_lock(&mp->mnt_listmtx); return (mnt_vnode_next_lazy(mvp, mp, cb, cbarg)); } struct vnode * __mnt_vnode_first_lazy(struct vnode **mvp, struct mount *mp, mnt_lazy_cb_t *cb, void *cbarg) { struct vnode *vp; if (TAILQ_EMPTY(&mp->mnt_lazyvnodelist)) return (NULL); *mvp = vn_alloc_marker(mp); MNT_ILOCK(mp); MNT_REF(mp); MNT_IUNLOCK(mp); mtx_lock(&mp->mnt_listmtx); vp = TAILQ_FIRST(&mp->mnt_lazyvnodelist); if (vp == NULL) { mtx_unlock(&mp->mnt_listmtx); mnt_vnode_markerfree_lazy(mvp, mp); return (NULL); } TAILQ_INSERT_BEFORE(vp, *mvp, v_lazylist); return (mnt_vnode_next_lazy(mvp, mp, cb, cbarg)); } void __mnt_vnode_markerfree_lazy(struct vnode **mvp, struct mount *mp) { if (*mvp == NULL) return; mtx_lock(&mp->mnt_listmtx); TAILQ_REMOVE(&mp->mnt_lazyvnodelist, *mvp, v_lazylist); mtx_unlock(&mp->mnt_listmtx); mnt_vnode_markerfree_lazy(mvp, mp); } int vn_dir_check_exec(struct vnode *vp, struct componentname *cnp) { if ((cnp->cn_flags & NOEXECCHECK) != 0) { cnp->cn_flags &= ~NOEXECCHECK; return (0); } - return (VOP_ACCESS(vp, VEXEC, cnp->cn_cred, cnp->cn_thread)); + return (VOP_ACCESS(vp, VEXEC, cnp->cn_cred, curthread)); } /* * Do not use this variant unless you have means other than the hold count * to prevent the vnode from getting freed. */ void vn_seqc_write_begin_locked(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __func__); VNPASS(vp->v_holdcnt > 0, vp); VNPASS(vp->v_seqc_users >= 0, vp); vp->v_seqc_users++; if (vp->v_seqc_users == 1) seqc_sleepable_write_begin(&vp->v_seqc); } void vn_seqc_write_begin(struct vnode *vp) { VI_LOCK(vp); vn_seqc_write_begin_locked(vp); VI_UNLOCK(vp); } void vn_seqc_write_end_locked(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __func__); VNPASS(vp->v_seqc_users > 0, vp); vp->v_seqc_users--; if (vp->v_seqc_users == 0) seqc_sleepable_write_end(&vp->v_seqc); } void vn_seqc_write_end(struct vnode *vp) { VI_LOCK(vp); vn_seqc_write_end_locked(vp); VI_UNLOCK(vp); } /* * Special case handling for allocating and freeing vnodes. * * The counter remains unchanged on free so that a doomed vnode will * keep testing as in modify as long as it is accessible with SMR. */ static void vn_seqc_init(struct vnode *vp) { vp->v_seqc = 0; vp->v_seqc_users = 0; } static void vn_seqc_write_end_free(struct vnode *vp) { VNPASS(seqc_in_modify(vp->v_seqc), vp); VNPASS(vp->v_seqc_users == 1, vp); } void vn_irflag_set_locked(struct vnode *vp, short toset) { short flags; ASSERT_VI_LOCKED(vp, __func__); flags = vn_irflag_read(vp); VNASSERT((flags & toset) == 0, vp, ("%s: some of the passed flags already set (have %d, passed %d)\n", __func__, flags, toset)); atomic_store_short(&vp->v_irflag, flags | toset); } void vn_irflag_set(struct vnode *vp, short toset) { VI_LOCK(vp); vn_irflag_set_locked(vp, toset); VI_UNLOCK(vp); } void vn_irflag_set_cond_locked(struct vnode *vp, short toset) { short flags; ASSERT_VI_LOCKED(vp, __func__); flags = vn_irflag_read(vp); atomic_store_short(&vp->v_irflag, flags | toset); } void vn_irflag_set_cond(struct vnode *vp, short toset) { VI_LOCK(vp); vn_irflag_set_cond_locked(vp, toset); VI_UNLOCK(vp); } void vn_irflag_unset_locked(struct vnode *vp, short tounset) { short flags; ASSERT_VI_LOCKED(vp, __func__); flags = vn_irflag_read(vp); VNASSERT((flags & tounset) == tounset, vp, ("%s: some of the passed flags not set (have %d, passed %d)\n", __func__, flags, tounset)); atomic_store_short(&vp->v_irflag, flags & ~tounset); } void vn_irflag_unset(struct vnode *vp, short tounset) { VI_LOCK(vp); vn_irflag_unset_locked(vp, tounset); VI_UNLOCK(vp); } diff --git a/sys/kern/vfs_vnops.c b/sys/kern/vfs_vnops.c index aeeeee3e3a1d..2c25a25da6c6 100644 --- a/sys/kern/vfs_vnops.c +++ b/sys/kern/vfs_vnops.c @@ -1,3709 +1,3708 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Copyright (c) 2012 Konstantin Belousov * Copyright (c) 2013, 2014 The FreeBSD Foundation * * Portions of this software were developed by Konstantin Belousov * under sponsorship from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vfs_vnops.c 8.2 (Berkeley) 1/21/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_hwpmc_hooks.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef HWPMC_HOOKS #include #endif static fo_rdwr_t vn_read; static fo_rdwr_t vn_write; static fo_rdwr_t vn_io_fault; static fo_truncate_t vn_truncate; static fo_ioctl_t vn_ioctl; static fo_poll_t vn_poll; static fo_kqfilter_t vn_kqfilter; static fo_close_t vn_closefile; static fo_mmap_t vn_mmap; static fo_fallocate_t vn_fallocate; static fo_fspacectl_t vn_fspacectl; struct fileops vnops = { .fo_read = vn_io_fault, .fo_write = vn_io_fault, .fo_truncate = vn_truncate, .fo_ioctl = vn_ioctl, .fo_poll = vn_poll, .fo_kqfilter = vn_kqfilter, .fo_stat = vn_statfile, .fo_close = vn_closefile, .fo_chmod = vn_chmod, .fo_chown = vn_chown, .fo_sendfile = vn_sendfile, .fo_seek = vn_seek, .fo_fill_kinfo = vn_fill_kinfo, .fo_mmap = vn_mmap, .fo_fallocate = vn_fallocate, .fo_fspacectl = vn_fspacectl, .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE }; const u_int io_hold_cnt = 16; static int vn_io_fault_enable = 1; SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RWTUN, &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance"); static int vn_io_fault_prefault = 0; SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_prefault, CTLFLAG_RWTUN, &vn_io_fault_prefault, 0, "Enable vn_io_fault prefaulting"); static int vn_io_pgcache_read_enable = 1; SYSCTL_INT(_debug, OID_AUTO, vn_io_pgcache_read_enable, CTLFLAG_RWTUN, &vn_io_pgcache_read_enable, 0, "Enable copying from page cache for reads, avoiding fs"); static u_long vn_io_faults_cnt; SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD, &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers"); static int vfs_allow_read_dir = 0; SYSCTL_INT(_security_bsd, OID_AUTO, allow_read_dir, CTLFLAG_RW, &vfs_allow_read_dir, 0, "Enable read(2) of directory by root for filesystems that support it"); /* * Returns true if vn_io_fault mode of handling the i/o request should * be used. */ static bool do_vn_io_fault(struct vnode *vp, struct uio *uio) { struct mount *mp; return (uio->uio_segflg == UIO_USERSPACE && vp->v_type == VREG && (mp = vp->v_mount) != NULL && (mp->mnt_kern_flag & MNTK_NO_IOPF) != 0 && vn_io_fault_enable); } /* * Structure used to pass arguments to vn_io_fault1(), to do either * file- or vnode-based I/O calls. */ struct vn_io_fault_args { enum { VN_IO_FAULT_FOP, VN_IO_FAULT_VOP } kind; struct ucred *cred; int flags; union { struct fop_args_tag { struct file *fp; fo_rdwr_t *doio; } fop_args; struct vop_args_tag { struct vnode *vp; } vop_args; } args; }; static int vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args, struct thread *td); int vn_open(struct nameidata *ndp, int *flagp, int cmode, struct file *fp) { - struct thread *td = ndp->ni_cnd.cn_thread; + struct thread *td = curthread; return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp)); } static uint64_t open2nameif(int fmode, u_int vn_open_flags) { uint64_t res; res = ISOPEN | LOCKLEAF; if ((fmode & O_RESOLVE_BENEATH) != 0) res |= RBENEATH; if ((fmode & O_EMPTY_PATH) != 0) res |= EMPTYPATH; if ((fmode & FREAD) != 0) res |= OPENREAD; if ((fmode & FWRITE) != 0) res |= OPENWRITE; if ((vn_open_flags & VN_OPEN_NOAUDIT) == 0) res |= AUDITVNODE1; if ((vn_open_flags & VN_OPEN_NOCAPCHECK) != 0) res |= NOCAPCHECK; return (res); } /* * Common code for vnode open operations via a name lookup. * Lookup the vnode and invoke VOP_CREATE if needed. * Check permissions, and call the VOP_OPEN or VOP_CREATE routine. * * Note that this does NOT free nameidata for the successful case, * due to the NDINIT being done elsewhere. */ int vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags, struct ucred *cred, struct file *fp) { struct vnode *vp; struct mount *mp; - struct thread *td = ndp->ni_cnd.cn_thread; struct vattr vat; struct vattr *vap = &vat; int fmode, error; bool first_open; restart: first_open = false; fmode = *flagp; if ((fmode & (O_CREAT | O_EXCL | O_DIRECTORY)) == (O_CREAT | O_EXCL | O_DIRECTORY) || (fmode & (O_CREAT | O_EMPTY_PATH)) == (O_CREAT | O_EMPTY_PATH)) return (EINVAL); else if ((fmode & (O_CREAT | O_DIRECTORY)) == O_CREAT) { ndp->ni_cnd.cn_nameiop = CREATE; ndp->ni_cnd.cn_flags = open2nameif(fmode, vn_open_flags); /* * Set NOCACHE to avoid flushing the cache when * rolling in many files at once. * * Set NC_KEEPPOSENTRY to keep positive entries if they already * exist despite NOCACHE. */ ndp->ni_cnd.cn_flags |= LOCKPARENT | NOCACHE | NC_KEEPPOSENTRY; if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0) ndp->ni_cnd.cn_flags |= FOLLOW; if ((vn_open_flags & VN_OPEN_INVFS) == 0) bwillwrite(); if ((error = namei(ndp)) != 0) return (error); if (ndp->ni_vp == NULL) { VATTR_NULL(vap); vap->va_type = VREG; vap->va_mode = cmode; if (fmode & O_EXCL) vap->va_vaflags |= VA_EXCLUSIVE; if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(ndp, NDF_ONLY_PNBUF); vput(ndp->ni_dvp); if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) return (error); NDREINIT(ndp); goto restart; } if ((vn_open_flags & VN_OPEN_NAMECACHE) != 0) ndp->ni_cnd.cn_flags |= MAKEENTRY; #ifdef MAC error = mac_vnode_check_create(cred, ndp->ni_dvp, &ndp->ni_cnd, vap); if (error == 0) #endif error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, vap); vp = ndp->ni_vp; if (error == 0 && (fmode & O_EXCL) != 0 && (fmode & (O_EXLOCK | O_SHLOCK)) != 0) { VI_LOCK(vp); vp->v_iflag |= VI_FOPENING; VI_UNLOCK(vp); first_open = true; } VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &vp : NULL, false); vn_finished_write(mp); if (error) { NDFREE(ndp, NDF_ONLY_PNBUF); if (error == ERELOOKUP) { NDREINIT(ndp); goto restart; } return (error); } fmode &= ~O_TRUNC; } else { if (ndp->ni_dvp == ndp->ni_vp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); ndp->ni_dvp = NULL; vp = ndp->ni_vp; if (fmode & O_EXCL) { error = EEXIST; goto bad; } if (vp->v_type == VDIR) { error = EISDIR; goto bad; } fmode &= ~O_CREAT; } } else { ndp->ni_cnd.cn_nameiop = LOOKUP; ndp->ni_cnd.cn_flags = open2nameif(fmode, vn_open_flags); ndp->ni_cnd.cn_flags |= (fmode & O_NOFOLLOW) != 0 ? NOFOLLOW : FOLLOW; if ((fmode & FWRITE) == 0) ndp->ni_cnd.cn_flags |= LOCKSHARED; if ((error = namei(ndp)) != 0) return (error); vp = ndp->ni_vp; } - error = vn_open_vnode(vp, fmode, cred, td, fp); + error = vn_open_vnode(vp, fmode, cred, curthread, fp); if (first_open) { VI_LOCK(vp); vp->v_iflag &= ~VI_FOPENING; wakeup(vp); VI_UNLOCK(vp); } if (error) goto bad; *flagp = fmode; return (0); bad: NDFREE(ndp, NDF_ONLY_PNBUF); vput(vp); *flagp = fmode; ndp->ni_vp = NULL; return (error); } static int vn_open_vnode_advlock(struct vnode *vp, int fmode, struct file *fp) { struct flock lf; int error, lock_flags, type; ASSERT_VOP_LOCKED(vp, "vn_open_vnode_advlock"); if ((fmode & (O_EXLOCK | O_SHLOCK)) == 0) return (0); KASSERT(fp != NULL, ("open with flock requires fp")); if (fp->f_type != DTYPE_NONE && fp->f_type != DTYPE_VNODE) return (EOPNOTSUPP); lock_flags = VOP_ISLOCKED(vp); VOP_UNLOCK(vp); lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; lf.l_type = (fmode & O_EXLOCK) != 0 ? F_WRLCK : F_RDLCK; type = F_FLOCK; if ((fmode & FNONBLOCK) == 0) type |= F_WAIT; if ((fmode & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) type |= F_FIRSTOPEN; error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type); if (error == 0) fp->f_flag |= FHASLOCK; vn_lock(vp, lock_flags | LK_RETRY); return (error); } /* * Common code for vnode open operations once a vnode is located. * Check permissions, and call the VOP_OPEN routine. */ int vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred, struct thread *td, struct file *fp) { accmode_t accmode; int error; if (vp->v_type == VLNK) { if ((fmode & O_PATH) == 0 || (fmode & FEXEC) != 0) return (EMLINK); } if (vp->v_type != VDIR && fmode & O_DIRECTORY) return (ENOTDIR); accmode = 0; if ((fmode & O_PATH) == 0) { if (vp->v_type == VSOCK) return (EOPNOTSUPP); if ((fmode & (FWRITE | O_TRUNC)) != 0) { if (vp->v_type == VDIR) return (EISDIR); accmode |= VWRITE; } if ((fmode & FREAD) != 0) accmode |= VREAD; if ((fmode & O_APPEND) && (fmode & FWRITE)) accmode |= VAPPEND; #ifdef MAC if ((fmode & O_CREAT) != 0) accmode |= VCREAT; #endif } if ((fmode & FEXEC) != 0) accmode |= VEXEC; #ifdef MAC if ((fmode & O_VERIFY) != 0) accmode |= VVERIFY; error = mac_vnode_check_open(cred, vp, accmode); if (error != 0) return (error); accmode &= ~(VCREAT | VVERIFY); #endif if ((fmode & O_CREAT) == 0 && accmode != 0) { error = VOP_ACCESS(vp, accmode, cred, td); if (error != 0) return (error); } if ((fmode & O_PATH) != 0) { if (vp->v_type != VFIFO && vp->v_type != VSOCK && VOP_ACCESS(vp, VREAD, cred, td) == 0) fp->f_flag |= FKQALLOWED; return (0); } if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) vn_lock(vp, LK_UPGRADE | LK_RETRY); error = VOP_OPEN(vp, fmode, cred, td, fp); if (error != 0) return (error); error = vn_open_vnode_advlock(vp, fmode, fp); if (error == 0 && (fmode & FWRITE) != 0) { error = VOP_ADD_WRITECOUNT(vp, 1); if (error == 0) { CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d", __func__, vp, vp->v_writecount); } } /* * Error from advlock or VOP_ADD_WRITECOUNT() still requires * calling VOP_CLOSE() to pair with earlier VOP_OPEN(). */ if (error != 0) { if (fp != NULL) { /* * Arrange the call by having fdrop() to use * vn_closefile(). This is to satisfy * filesystems like devfs or tmpfs, which * override fo_close(). */ fp->f_flag |= FOPENFAILED; fp->f_vnode = vp; if (fp->f_ops == &badfileops) { fp->f_type = DTYPE_VNODE; fp->f_ops = &vnops; } vref(vp); } else { /* * If there is no fp, due to kernel-mode open, * we can call VOP_CLOSE() now. */ if (vp->v_type != VFIFO && (fmode & FWRITE) != 0 && !MNT_EXTENDED_SHARED(vp->v_mount) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) vn_lock(vp, LK_UPGRADE | LK_RETRY); (void)VOP_CLOSE(vp, fmode & (FREAD | FWRITE | FEXEC), cred, td); } } ASSERT_VOP_LOCKED(vp, "vn_open_vnode"); return (error); } /* * Check for write permissions on the specified vnode. * Prototype text segments cannot be written. * It is racy. */ int vn_writechk(struct vnode *vp) { ASSERT_VOP_LOCKED(vp, "vn_writechk"); /* * If there's shared text associated with * the vnode, try to free it up once. If * we fail, we can't allow writing. */ if (VOP_IS_TEXT(vp)) return (ETXTBSY); return (0); } /* * Vnode close call */ static int vn_close1(struct vnode *vp, int flags, struct ucred *file_cred, struct thread *td, bool keep_ref) { struct mount *mp; int error, lock_flags; if (vp->v_type != VFIFO && (flags & FWRITE) == 0 && MNT_EXTENDED_SHARED(vp->v_mount)) lock_flags = LK_SHARED; else lock_flags = LK_EXCLUSIVE; vn_start_write(vp, &mp, V_WAIT); vn_lock(vp, lock_flags | LK_RETRY); AUDIT_ARG_VNODE1(vp); if ((flags & (FWRITE | FOPENFAILED)) == FWRITE) { VOP_ADD_WRITECOUNT_CHECKED(vp, -1); CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d", __func__, vp, vp->v_writecount); } error = VOP_CLOSE(vp, flags, file_cred, td); if (keep_ref) VOP_UNLOCK(vp); else vput(vp); vn_finished_write(mp); return (error); } int vn_close(struct vnode *vp, int flags, struct ucred *file_cred, struct thread *td) { return (vn_close1(vp, flags, file_cred, td, false)); } /* * Heuristic to detect sequential operation. */ static int sequential_heuristic(struct uio *uio, struct file *fp) { enum uio_rw rw; ASSERT_VOP_LOCKED(fp->f_vnode, __func__); rw = uio->uio_rw; if (fp->f_flag & FRDAHEAD) return (fp->f_seqcount[rw] << IO_SEQSHIFT); /* * Offset 0 is handled specially. open() sets f_seqcount to 1 so * that the first I/O is normally considered to be slightly * sequential. Seeking to offset 0 doesn't change sequentiality * unless previous seeks have reduced f_seqcount to 0, in which * case offset 0 is not special. */ if ((uio->uio_offset == 0 && fp->f_seqcount[rw] > 0) || uio->uio_offset == fp->f_nextoff[rw]) { /* * f_seqcount is in units of fixed-size blocks so that it * depends mainly on the amount of sequential I/O and not * much on the number of sequential I/O's. The fixed size * of 16384 is hard-coded here since it is (not quite) just * a magic size that works well here. This size is more * closely related to the best I/O size for real disks than * to any block size used by software. */ if (uio->uio_resid >= IO_SEQMAX * 16384) fp->f_seqcount[rw] = IO_SEQMAX; else { fp->f_seqcount[rw] += howmany(uio->uio_resid, 16384); if (fp->f_seqcount[rw] > IO_SEQMAX) fp->f_seqcount[rw] = IO_SEQMAX; } return (fp->f_seqcount[rw] << IO_SEQSHIFT); } /* Not sequential. Quickly draw-down sequentiality. */ if (fp->f_seqcount[rw] > 1) fp->f_seqcount[rw] = 1; else fp->f_seqcount[rw] = 0; return (0); } /* * Package up an I/O request on a vnode into a uio and do it. */ int vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset, enum uio_seg segflg, int ioflg, struct ucred *active_cred, struct ucred *file_cred, ssize_t *aresid, struct thread *td) { struct uio auio; struct iovec aiov; struct mount *mp; struct ucred *cred; void *rl_cookie; struct vn_io_fault_args args; int error, lock_flags; if (offset < 0 && vp->v_type != VCHR) return (EINVAL); auio.uio_iov = &aiov; auio.uio_iovcnt = 1; aiov.iov_base = base; aiov.iov_len = len; auio.uio_resid = len; auio.uio_offset = offset; auio.uio_segflg = segflg; auio.uio_rw = rw; auio.uio_td = td; error = 0; if ((ioflg & IO_NODELOCKED) == 0) { if ((ioflg & IO_RANGELOCKED) == 0) { if (rw == UIO_READ) { rl_cookie = vn_rangelock_rlock(vp, offset, offset + len); } else if ((ioflg & IO_APPEND) != 0) { rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); } else { rl_cookie = vn_rangelock_wlock(vp, offset, offset + len); } } else rl_cookie = NULL; mp = NULL; if (rw == UIO_WRITE) { if (vp->v_type != VCHR && (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) goto out; lock_flags = vn_lktype_write(mp, vp); } else lock_flags = LK_SHARED; vn_lock(vp, lock_flags | LK_RETRY); } else rl_cookie = NULL; ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); #ifdef MAC if ((ioflg & IO_NOMACCHECK) == 0) { if (rw == UIO_READ) error = mac_vnode_check_read(active_cred, file_cred, vp); else error = mac_vnode_check_write(active_cred, file_cred, vp); } #endif if (error == 0) { if (file_cred != NULL) cred = file_cred; else cred = active_cred; if (do_vn_io_fault(vp, &auio)) { args.kind = VN_IO_FAULT_VOP; args.cred = cred; args.flags = ioflg; args.args.vop_args.vp = vp; error = vn_io_fault1(vp, &auio, &args, td); } else if (rw == UIO_READ) { error = VOP_READ(vp, &auio, ioflg, cred); } else /* if (rw == UIO_WRITE) */ { error = VOP_WRITE(vp, &auio, ioflg, cred); } } if (aresid) *aresid = auio.uio_resid; else if (auio.uio_resid && error == 0) error = EIO; if ((ioflg & IO_NODELOCKED) == 0) { VOP_UNLOCK(vp); if (mp != NULL) vn_finished_write(mp); } out: if (rl_cookie != NULL) vn_rangelock_unlock(vp, rl_cookie); return (error); } /* * Package up an I/O request on a vnode into a uio and do it. The I/O * request is split up into smaller chunks and we try to avoid saturating * the buffer cache while potentially holding a vnode locked, so we * check bwillwrite() before calling vn_rdwr(). We also call kern_yield() * to give other processes a chance to lock the vnode (either other processes * core'ing the same binary, or unrelated processes scanning the directory). */ int vn_rdwr_inchunks(enum uio_rw rw, struct vnode *vp, void *base, size_t len, off_t offset, enum uio_seg segflg, int ioflg, struct ucred *active_cred, struct ucred *file_cred, size_t *aresid, struct thread *td) { int error = 0; ssize_t iaresid; do { int chunk; /* * Force `offset' to a multiple of MAXBSIZE except possibly * for the first chunk, so that filesystems only need to * write full blocks except possibly for the first and last * chunks. */ chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE; if (chunk > len) chunk = len; if (rw != UIO_READ && vp->v_type == VREG) bwillwrite(); iaresid = 0; error = vn_rdwr(rw, vp, base, chunk, offset, segflg, ioflg, active_cred, file_cred, &iaresid, td); len -= chunk; /* aresid calc already includes length */ if (error) break; offset += chunk; base = (char *)base + chunk; kern_yield(PRI_USER); } while (len); if (aresid) *aresid = len + iaresid; return (error); } #if OFF_MAX <= LONG_MAX off_t foffset_lock(struct file *fp, int flags) { volatile short *flagsp; off_t res; short state; KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); if ((flags & FOF_NOLOCK) != 0) return (atomic_load_long(&fp->f_offset)); /* * According to McKusick the vn lock was protecting f_offset here. * It is now protected by the FOFFSET_LOCKED flag. */ flagsp = &fp->f_vnread_flags; if (atomic_cmpset_acq_16(flagsp, 0, FOFFSET_LOCKED)) return (atomic_load_long(&fp->f_offset)); sleepq_lock(&fp->f_vnread_flags); state = atomic_load_16(flagsp); for (;;) { if ((state & FOFFSET_LOCKED) == 0) { if (!atomic_fcmpset_acq_16(flagsp, &state, FOFFSET_LOCKED)) continue; break; } if ((state & FOFFSET_LOCK_WAITING) == 0) { if (!atomic_fcmpset_acq_16(flagsp, &state, state | FOFFSET_LOCK_WAITING)) continue; } DROP_GIANT(); sleepq_add(&fp->f_vnread_flags, NULL, "vofflock", 0, 0); sleepq_wait(&fp->f_vnread_flags, PUSER -1); PICKUP_GIANT(); sleepq_lock(&fp->f_vnread_flags); state = atomic_load_16(flagsp); } res = atomic_load_long(&fp->f_offset); sleepq_release(&fp->f_vnread_flags); return (res); } void foffset_unlock(struct file *fp, off_t val, int flags) { volatile short *flagsp; short state; KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); if ((flags & FOF_NOUPDATE) == 0) atomic_store_long(&fp->f_offset, val); if ((flags & FOF_NEXTOFF_R) != 0) fp->f_nextoff[UIO_READ] = val; if ((flags & FOF_NEXTOFF_W) != 0) fp->f_nextoff[UIO_WRITE] = val; if ((flags & FOF_NOLOCK) != 0) return; flagsp = &fp->f_vnread_flags; state = atomic_load_16(flagsp); if ((state & FOFFSET_LOCK_WAITING) == 0 && atomic_cmpset_rel_16(flagsp, state, 0)) return; sleepq_lock(&fp->f_vnread_flags); MPASS((fp->f_vnread_flags & FOFFSET_LOCKED) != 0); MPASS((fp->f_vnread_flags & FOFFSET_LOCK_WAITING) != 0); fp->f_vnread_flags = 0; sleepq_broadcast(&fp->f_vnread_flags, SLEEPQ_SLEEP, 0, 0); sleepq_release(&fp->f_vnread_flags); } #else off_t foffset_lock(struct file *fp, int flags) { struct mtx *mtxp; off_t res; KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); mtxp = mtx_pool_find(mtxpool_sleep, fp); mtx_lock(mtxp); if ((flags & FOF_NOLOCK) == 0) { while (fp->f_vnread_flags & FOFFSET_LOCKED) { fp->f_vnread_flags |= FOFFSET_LOCK_WAITING; msleep(&fp->f_vnread_flags, mtxp, PUSER -1, "vofflock", 0); } fp->f_vnread_flags |= FOFFSET_LOCKED; } res = fp->f_offset; mtx_unlock(mtxp); return (res); } void foffset_unlock(struct file *fp, off_t val, int flags) { struct mtx *mtxp; KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); mtxp = mtx_pool_find(mtxpool_sleep, fp); mtx_lock(mtxp); if ((flags & FOF_NOUPDATE) == 0) fp->f_offset = val; if ((flags & FOF_NEXTOFF_R) != 0) fp->f_nextoff[UIO_READ] = val; if ((flags & FOF_NEXTOFF_W) != 0) fp->f_nextoff[UIO_WRITE] = val; if ((flags & FOF_NOLOCK) == 0) { KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0, ("Lost FOFFSET_LOCKED")); if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING) wakeup(&fp->f_vnread_flags); fp->f_vnread_flags = 0; } mtx_unlock(mtxp); } #endif void foffset_lock_uio(struct file *fp, struct uio *uio, int flags) { if ((flags & FOF_OFFSET) == 0) uio->uio_offset = foffset_lock(fp, flags); } void foffset_unlock_uio(struct file *fp, struct uio *uio, int flags) { if ((flags & FOF_OFFSET) == 0) foffset_unlock(fp, uio->uio_offset, flags); } static int get_advice(struct file *fp, struct uio *uio) { struct mtx *mtxp; int ret; ret = POSIX_FADV_NORMAL; if (fp->f_advice == NULL || fp->f_vnode->v_type != VREG) return (ret); mtxp = mtx_pool_find(mtxpool_sleep, fp); mtx_lock(mtxp); if (fp->f_advice != NULL && uio->uio_offset >= fp->f_advice->fa_start && uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end) ret = fp->f_advice->fa_advice; mtx_unlock(mtxp); return (ret); } static int get_write_ioflag(struct file *fp) { int ioflag; struct mount *mp; struct vnode *vp; ioflag = 0; vp = fp->f_vnode; mp = atomic_load_ptr(&vp->v_mount); if ((fp->f_flag & O_DIRECT) != 0) ioflag |= IO_DIRECT; if ((fp->f_flag & O_FSYNC) != 0 || (mp != NULL && (mp->mnt_flag & MNT_SYNCHRONOUS) != 0)) ioflag |= IO_SYNC; /* * For O_DSYNC we set both IO_SYNC and IO_DATASYNC, so that VOP_WRITE() * or VOP_DEALLOCATE() implementations that don't understand IO_DATASYNC * fall back to full O_SYNC behavior. */ if ((fp->f_flag & O_DSYNC) != 0) ioflag |= IO_SYNC | IO_DATASYNC; return (ioflag); } int vn_read_from_obj(struct vnode *vp, struct uio *uio) { vm_object_t obj; vm_page_t ma[io_hold_cnt + 2]; off_t off, vsz; ssize_t resid; int error, i, j; MPASS(uio->uio_resid <= ptoa(io_hold_cnt + 2)); obj = atomic_load_ptr(&vp->v_object); if (obj == NULL) return (EJUSTRETURN); /* * Depends on type stability of vm_objects. */ vm_object_pip_add(obj, 1); if ((obj->flags & OBJ_DEAD) != 0) { /* * Note that object might be already reused from the * vnode, and the OBJ_DEAD flag cleared. This is fine, * we recheck for DOOMED vnode state after all pages * are busied, and retract then. * * But we check for OBJ_DEAD to ensure that we do not * busy pages while vm_object_terminate_pages() * processes the queue. */ error = EJUSTRETURN; goto out_pip; } resid = uio->uio_resid; off = uio->uio_offset; for (i = 0; resid > 0; i++) { MPASS(i < io_hold_cnt + 2); ma[i] = vm_page_grab_unlocked(obj, atop(off), VM_ALLOC_NOCREAT | VM_ALLOC_SBUSY | VM_ALLOC_IGN_SBUSY | VM_ALLOC_NOWAIT); if (ma[i] == NULL) break; /* * Skip invalid pages. Valid mask can be partial only * at EOF, and we clip later. */ if (vm_page_none_valid(ma[i])) { vm_page_sunbusy(ma[i]); break; } resid -= PAGE_SIZE; off += PAGE_SIZE; } if (i == 0) { error = EJUSTRETURN; goto out_pip; } /* * Check VIRF_DOOMED after we busied our pages. Since * vgonel() terminates the vnode' vm_object, it cannot * process past pages busied by us. */ if (VN_IS_DOOMED(vp)) { error = EJUSTRETURN; goto out; } resid = PAGE_SIZE - (uio->uio_offset & PAGE_MASK) + ptoa(i - 1); if (resid > uio->uio_resid) resid = uio->uio_resid; /* * Unlocked read of vnp_size is safe because truncation cannot * pass busied page. But we load vnp_size into a local * variable so that possible concurrent extension does not * break calculation. */ #if defined(__powerpc__) && !defined(__powerpc64__) vsz = obj->un_pager.vnp.vnp_size; #else vsz = atomic_load_64(&obj->un_pager.vnp.vnp_size); #endif if (uio->uio_offset >= vsz) { error = EJUSTRETURN; goto out; } if (uio->uio_offset + resid > vsz) resid = vsz - uio->uio_offset; error = vn_io_fault_pgmove(ma, uio->uio_offset & PAGE_MASK, resid, uio); out: for (j = 0; j < i; j++) { if (error == 0) vm_page_reference(ma[j]); vm_page_sunbusy(ma[j]); } out_pip: vm_object_pip_wakeup(obj); if (error != 0) return (error); return (uio->uio_resid == 0 ? 0 : EJUSTRETURN); } /* * File table vnode read routine. */ static int vn_read(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { struct vnode *vp; off_t orig_offset; int error, ioflag; int advice; KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td)); KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET")); vp = fp->f_vnode; ioflag = 0; if (fp->f_flag & FNONBLOCK) ioflag |= IO_NDELAY; if (fp->f_flag & O_DIRECT) ioflag |= IO_DIRECT; /* * Try to read from page cache. VIRF_DOOMED check is racy but * allows us to avoid unneeded work outright. */ if (vn_io_pgcache_read_enable && !mac_vnode_check_read_enabled() && (vn_irflag_read(vp) & (VIRF_DOOMED | VIRF_PGREAD)) == VIRF_PGREAD) { error = VOP_READ_PGCACHE(vp, uio, ioflag, fp->f_cred); if (error == 0) { fp->f_nextoff[UIO_READ] = uio->uio_offset; return (0); } if (error != EJUSTRETURN) return (error); } advice = get_advice(fp, uio); vn_lock(vp, LK_SHARED | LK_RETRY); switch (advice) { case POSIX_FADV_NORMAL: case POSIX_FADV_SEQUENTIAL: case POSIX_FADV_NOREUSE: ioflag |= sequential_heuristic(uio, fp); break; case POSIX_FADV_RANDOM: /* Disable read-ahead for random I/O. */ break; } orig_offset = uio->uio_offset; #ifdef MAC error = mac_vnode_check_read(active_cred, fp->f_cred, vp); if (error == 0) #endif error = VOP_READ(vp, uio, ioflag, fp->f_cred); fp->f_nextoff[UIO_READ] = uio->uio_offset; VOP_UNLOCK(vp); if (error == 0 && advice == POSIX_FADV_NOREUSE && orig_offset != uio->uio_offset) /* * Use POSIX_FADV_DONTNEED to flush pages and buffers * for the backing file after a POSIX_FADV_NOREUSE * read(2). */ error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1, POSIX_FADV_DONTNEED); return (error); } /* * File table vnode write routine. */ static int vn_write(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { struct vnode *vp; struct mount *mp; off_t orig_offset; int error, ioflag; int advice; bool need_finished_write; KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td)); KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET")); vp = fp->f_vnode; if (vp->v_type == VREG) bwillwrite(); ioflag = IO_UNIT; if (vp->v_type == VREG && (fp->f_flag & O_APPEND) != 0) ioflag |= IO_APPEND; if ((fp->f_flag & FNONBLOCK) != 0) ioflag |= IO_NDELAY; ioflag |= get_write_ioflag(fp); mp = NULL; need_finished_write = false; if (vp->v_type != VCHR) { error = vn_start_write(vp, &mp, V_WAIT | PCATCH); if (error != 0) goto unlock; need_finished_write = true; } advice = get_advice(fp, uio); vn_lock(vp, vn_lktype_write(mp, vp) | LK_RETRY); switch (advice) { case POSIX_FADV_NORMAL: case POSIX_FADV_SEQUENTIAL: case POSIX_FADV_NOREUSE: ioflag |= sequential_heuristic(uio, fp); break; case POSIX_FADV_RANDOM: /* XXX: Is this correct? */ break; } orig_offset = uio->uio_offset; #ifdef MAC error = mac_vnode_check_write(active_cred, fp->f_cred, vp); if (error == 0) #endif error = VOP_WRITE(vp, uio, ioflag, fp->f_cred); fp->f_nextoff[UIO_WRITE] = uio->uio_offset; VOP_UNLOCK(vp); if (need_finished_write) vn_finished_write(mp); if (error == 0 && advice == POSIX_FADV_NOREUSE && orig_offset != uio->uio_offset) /* * Use POSIX_FADV_DONTNEED to flush pages and buffers * for the backing file after a POSIX_FADV_NOREUSE * write(2). */ error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1, POSIX_FADV_DONTNEED); unlock: return (error); } /* * The vn_io_fault() is a wrapper around vn_read() and vn_write() to * prevent the following deadlock: * * Assume that the thread A reads from the vnode vp1 into userspace * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is * currently not resident, then system ends up with the call chain * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] -> * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2) * which establishes lock order vp1->vn_lock, then vp2->vn_lock. * If, at the same time, thread B reads from vnode vp2 into buffer buf2 * backed by the pages of vnode vp1, and some page in buf2 is not * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock. * * To prevent the lock order reversal and deadlock, vn_io_fault() does * not allow page faults to happen during VOP_READ() or VOP_WRITE(). * Instead, it first tries to do the whole range i/o with pagefaults * disabled. If all pages in the i/o buffer are resident and mapped, * VOP will succeed (ignoring the genuine filesystem errors). * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do * i/o in chunks, with all pages in the chunk prefaulted and held * using vm_fault_quick_hold_pages(). * * Filesystems using this deadlock avoidance scheme should use the * array of the held pages from uio, saved in the curthread->td_ma, * instead of doing uiomove(). A helper function * vn_io_fault_uiomove() converts uiomove request into * uiomove_fromphys() over td_ma array. * * Since vnode locks do not cover the whole i/o anymore, rangelocks * make the current i/o request atomic with respect to other i/os and * truncations. */ /* * Decode vn_io_fault_args and perform the corresponding i/o. */ static int vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio, struct thread *td) { int error, save; error = 0; save = vm_fault_disable_pagefaults(); switch (args->kind) { case VN_IO_FAULT_FOP: error = (args->args.fop_args.doio)(args->args.fop_args.fp, uio, args->cred, args->flags, td); break; case VN_IO_FAULT_VOP: if (uio->uio_rw == UIO_READ) { error = VOP_READ(args->args.vop_args.vp, uio, args->flags, args->cred); } else if (uio->uio_rw == UIO_WRITE) { error = VOP_WRITE(args->args.vop_args.vp, uio, args->flags, args->cred); } break; default: panic("vn_io_fault_doio: unknown kind of io %d %d", args->kind, uio->uio_rw); } vm_fault_enable_pagefaults(save); return (error); } static int vn_io_fault_touch(char *base, const struct uio *uio) { int r; r = fubyte(base); if (r == -1 || (uio->uio_rw == UIO_READ && subyte(base, r) == -1)) return (EFAULT); return (0); } static int vn_io_fault_prefault_user(const struct uio *uio) { char *base; const struct iovec *iov; size_t len; ssize_t resid; int error, i; KASSERT(uio->uio_segflg == UIO_USERSPACE, ("vn_io_fault_prefault userspace")); error = i = 0; iov = uio->uio_iov; resid = uio->uio_resid; base = iov->iov_base; len = iov->iov_len; while (resid > 0) { error = vn_io_fault_touch(base, uio); if (error != 0) break; if (len < PAGE_SIZE) { if (len != 0) { error = vn_io_fault_touch(base + len - 1, uio); if (error != 0) break; resid -= len; } if (++i >= uio->uio_iovcnt) break; iov = uio->uio_iov + i; base = iov->iov_base; len = iov->iov_len; } else { len -= PAGE_SIZE; base += PAGE_SIZE; resid -= PAGE_SIZE; } } return (error); } /* * Common code for vn_io_fault(), agnostic to the kind of i/o request. * Uses vn_io_fault_doio() to make the call to an actual i/o function. * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request * into args and call vn_io_fault1() to handle faults during the user * mode buffer accesses. */ static int vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args, struct thread *td) { vm_page_t ma[io_hold_cnt + 2]; struct uio *uio_clone, short_uio; struct iovec short_iovec[1]; vm_page_t *prev_td_ma; vm_prot_t prot; vm_offset_t addr, end; size_t len, resid; ssize_t adv; int error, cnt, saveheld, prev_td_ma_cnt; if (vn_io_fault_prefault) { error = vn_io_fault_prefault_user(uio); if (error != 0) return (error); /* Or ignore ? */ } prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ; /* * The UFS follows IO_UNIT directive and replays back both * uio_offset and uio_resid if an error is encountered during the * operation. But, since the iovec may be already advanced, * uio is still in an inconsistent state. * * Cache a copy of the original uio, which is advanced to the redo * point using UIO_NOCOPY below. */ uio_clone = cloneuio(uio); resid = uio->uio_resid; short_uio.uio_segflg = UIO_USERSPACE; short_uio.uio_rw = uio->uio_rw; short_uio.uio_td = uio->uio_td; error = vn_io_fault_doio(args, uio, td); if (error != EFAULT) goto out; atomic_add_long(&vn_io_faults_cnt, 1); uio_clone->uio_segflg = UIO_NOCOPY; uiomove(NULL, resid - uio->uio_resid, uio_clone); uio_clone->uio_segflg = uio->uio_segflg; saveheld = curthread_pflags_set(TDP_UIOHELD); prev_td_ma = td->td_ma; prev_td_ma_cnt = td->td_ma_cnt; while (uio_clone->uio_resid != 0) { len = uio_clone->uio_iov->iov_len; if (len == 0) { KASSERT(uio_clone->uio_iovcnt >= 1, ("iovcnt underflow")); uio_clone->uio_iov++; uio_clone->uio_iovcnt--; continue; } if (len > ptoa(io_hold_cnt)) len = ptoa(io_hold_cnt); addr = (uintptr_t)uio_clone->uio_iov->iov_base; end = round_page(addr + len); if (end < addr) { error = EFAULT; break; } cnt = atop(end - trunc_page(addr)); /* * A perfectly misaligned address and length could cause * both the start and the end of the chunk to use partial * page. +2 accounts for such a situation. */ cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map, addr, len, prot, ma, io_hold_cnt + 2); if (cnt == -1) { error = EFAULT; break; } short_uio.uio_iov = &short_iovec[0]; short_iovec[0].iov_base = (void *)addr; short_uio.uio_iovcnt = 1; short_uio.uio_resid = short_iovec[0].iov_len = len; short_uio.uio_offset = uio_clone->uio_offset; td->td_ma = ma; td->td_ma_cnt = cnt; error = vn_io_fault_doio(args, &short_uio, td); vm_page_unhold_pages(ma, cnt); adv = len - short_uio.uio_resid; uio_clone->uio_iov->iov_base = (char *)uio_clone->uio_iov->iov_base + adv; uio_clone->uio_iov->iov_len -= adv; uio_clone->uio_resid -= adv; uio_clone->uio_offset += adv; uio->uio_resid -= adv; uio->uio_offset += adv; if (error != 0 || adv == 0) break; } td->td_ma = prev_td_ma; td->td_ma_cnt = prev_td_ma_cnt; curthread_pflags_restore(saveheld); out: free(uio_clone, M_IOV); return (error); } static int vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { fo_rdwr_t *doio; struct vnode *vp; void *rl_cookie; struct vn_io_fault_args args; int error; doio = uio->uio_rw == UIO_READ ? vn_read : vn_write; vp = fp->f_vnode; /* * The ability to read(2) on a directory has historically been * allowed for all users, but this can and has been the source of * at least one security issue in the past. As such, it is now hidden * away behind a sysctl for those that actually need it to use it, and * restricted to root when it's turned on to make it relatively safe to * leave on for longer sessions of need. */ if (vp->v_type == VDIR) { KASSERT(uio->uio_rw == UIO_READ, ("illegal write attempted on a directory")); if (!vfs_allow_read_dir) return (EISDIR); if ((error = priv_check(td, PRIV_VFS_READ_DIR)) != 0) return (EISDIR); } foffset_lock_uio(fp, uio, flags); if (do_vn_io_fault(vp, uio)) { args.kind = VN_IO_FAULT_FOP; args.args.fop_args.fp = fp; args.args.fop_args.doio = doio; args.cred = active_cred; args.flags = flags | FOF_OFFSET; if (uio->uio_rw == UIO_READ) { rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset, uio->uio_offset + uio->uio_resid); } else if ((fp->f_flag & O_APPEND) != 0 || (flags & FOF_OFFSET) == 0) { /* For appenders, punt and lock the whole range. */ rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); } else { rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset, uio->uio_offset + uio->uio_resid); } error = vn_io_fault1(vp, uio, &args, td); vn_rangelock_unlock(vp, rl_cookie); } else { error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td); } foffset_unlock_uio(fp, uio, flags); return (error); } /* * Helper function to perform the requested uiomove operation using * the held pages for io->uio_iov[0].iov_base buffer instead of * copyin/copyout. Access to the pages with uiomove_fromphys() * instead of iov_base prevents page faults that could occur due to * pmap_collect() invalidating the mapping created by * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or * object cleanup revoking the write access from page mappings. * * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove() * instead of plain uiomove(). */ int vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio) { struct uio transp_uio; struct iovec transp_iov[1]; struct thread *td; size_t adv; int error, pgadv; td = curthread; if ((td->td_pflags & TDP_UIOHELD) == 0 || uio->uio_segflg != UIO_USERSPACE) return (uiomove(data, xfersize, uio)); KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt)); transp_iov[0].iov_base = data; transp_uio.uio_iov = &transp_iov[0]; transp_uio.uio_iovcnt = 1; if (xfersize > uio->uio_resid) xfersize = uio->uio_resid; transp_uio.uio_resid = transp_iov[0].iov_len = xfersize; transp_uio.uio_offset = 0; transp_uio.uio_segflg = UIO_SYSSPACE; /* * Since transp_iov points to data, and td_ma page array * corresponds to original uio->uio_iov, we need to invert the * direction of the i/o operation as passed to * uiomove_fromphys(). */ switch (uio->uio_rw) { case UIO_WRITE: transp_uio.uio_rw = UIO_READ; break; case UIO_READ: transp_uio.uio_rw = UIO_WRITE; break; } transp_uio.uio_td = uio->uio_td; error = uiomove_fromphys(td->td_ma, ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK, xfersize, &transp_uio); adv = xfersize - transp_uio.uio_resid; pgadv = (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) - (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT); td->td_ma += pgadv; KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt, pgadv)); td->td_ma_cnt -= pgadv; uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv; uio->uio_iov->iov_len -= adv; uio->uio_resid -= adv; uio->uio_offset += adv; return (error); } int vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize, struct uio *uio) { struct thread *td; vm_offset_t iov_base; int cnt, pgadv; td = curthread; if ((td->td_pflags & TDP_UIOHELD) == 0 || uio->uio_segflg != UIO_USERSPACE) return (uiomove_fromphys(ma, offset, xfersize, uio)); KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt)); cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize; iov_base = (vm_offset_t)uio->uio_iov->iov_base; switch (uio->uio_rw) { case UIO_WRITE: pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma, offset, cnt); break; case UIO_READ: pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK, cnt); break; } pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT); td->td_ma += pgadv; KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt, pgadv)); td->td_ma_cnt -= pgadv; uio->uio_iov->iov_base = (char *)(iov_base + cnt); uio->uio_iov->iov_len -= cnt; uio->uio_resid -= cnt; uio->uio_offset += cnt; return (0); } /* * File table truncate routine. */ static int vn_truncate(struct file *fp, off_t length, struct ucred *active_cred, struct thread *td) { struct mount *mp; struct vnode *vp; void *rl_cookie; int error; vp = fp->f_vnode; retry: /* * Lock the whole range for truncation. Otherwise split i/o * might happen partly before and partly after the truncation. */ rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); error = vn_start_write(vp, &mp, V_WAIT | PCATCH); if (error) goto out1; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); AUDIT_ARG_VNODE1(vp); if (vp->v_type == VDIR) { error = EISDIR; goto out; } #ifdef MAC error = mac_vnode_check_write(active_cred, fp->f_cred, vp); if (error) goto out; #endif error = vn_truncate_locked(vp, length, (fp->f_flag & O_FSYNC) != 0, fp->f_cred); out: VOP_UNLOCK(vp); vn_finished_write(mp); out1: vn_rangelock_unlock(vp, rl_cookie); if (error == ERELOOKUP) goto retry; return (error); } /* * Truncate a file that is already locked. */ int vn_truncate_locked(struct vnode *vp, off_t length, bool sync, struct ucred *cred) { struct vattr vattr; int error; error = VOP_ADD_WRITECOUNT(vp, 1); if (error == 0) { VATTR_NULL(&vattr); vattr.va_size = length; if (sync) vattr.va_vaflags |= VA_SYNC; error = VOP_SETATTR(vp, &vattr, cred); VOP_ADD_WRITECOUNT_CHECKED(vp, -1); } return (error); } /* * File table vnode stat routine. */ int vn_statfile(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td) { struct vnode *vp = fp->f_vnode; int error; vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_STAT(vp, sb, active_cred, fp->f_cred, td); VOP_UNLOCK(vp); return (error); } /* * File table vnode ioctl routine. */ static int vn_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td) { struct vattr vattr; struct vnode *vp; struct fiobmap2_arg *bmarg; int error; vp = fp->f_vnode; switch (vp->v_type) { case VDIR: case VREG: switch (com) { case FIONREAD: vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_GETATTR(vp, &vattr, active_cred); VOP_UNLOCK(vp); if (error == 0) *(int *)data = vattr.va_size - fp->f_offset; return (error); case FIOBMAP2: bmarg = (struct fiobmap2_arg *)data; vn_lock(vp, LK_SHARED | LK_RETRY); #ifdef MAC error = mac_vnode_check_read(active_cred, fp->f_cred, vp); if (error == 0) #endif error = VOP_BMAP(vp, bmarg->bn, NULL, &bmarg->bn, &bmarg->runp, &bmarg->runb); VOP_UNLOCK(vp); return (error); case FIONBIO: case FIOASYNC: return (0); default: return (VOP_IOCTL(vp, com, data, fp->f_flag, active_cred, td)); } break; case VCHR: return (VOP_IOCTL(vp, com, data, fp->f_flag, active_cred, td)); default: return (ENOTTY); } } /* * File table vnode poll routine. */ static int vn_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td) { struct vnode *vp; int error; vp = fp->f_vnode; #if defined(MAC) || defined(AUDIT) if (AUDITING_TD(td) || mac_vnode_check_poll_enabled()) { vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); AUDIT_ARG_VNODE1(vp); error = mac_vnode_check_poll(active_cred, fp->f_cred, vp); VOP_UNLOCK(vp); if (error != 0) return (error); } #endif error = VOP_POLL(vp, events, fp->f_cred, td); return (error); } /* * Acquire the requested lock and then check for validity. LK_RETRY * permits vn_lock to return doomed vnodes. */ static int __noinline _vn_lock_fallback(struct vnode *vp, int flags, const char *file, int line, int error) { KASSERT((flags & LK_RETRY) == 0 || error == 0, ("vn_lock: error %d incompatible with flags %#x", error, flags)); if (error == 0) VNASSERT(VN_IS_DOOMED(vp), vp, ("vnode not doomed")); if ((flags & LK_RETRY) == 0) { if (error == 0) { VOP_UNLOCK(vp); error = ENOENT; } return (error); } /* * LK_RETRY case. * * Nothing to do if we got the lock. */ if (error == 0) return (0); /* * Interlock was dropped by the call in _vn_lock. */ flags &= ~LK_INTERLOCK; do { error = VOP_LOCK1(vp, flags, file, line); } while (error != 0); return (0); } int _vn_lock(struct vnode *vp, int flags, const char *file, int line) { int error; VNASSERT((flags & LK_TYPE_MASK) != 0, vp, ("vn_lock: no locktype (%d passed)", flags)); VNPASS(vp->v_holdcnt > 0, vp); error = VOP_LOCK1(vp, flags, file, line); if (__predict_false(error != 0 || VN_IS_DOOMED(vp))) return (_vn_lock_fallback(vp, flags, file, line, error)); return (0); } /* * File table vnode close routine. */ static int vn_closefile(struct file *fp, struct thread *td) { struct vnode *vp; struct flock lf; int error; bool ref; vp = fp->f_vnode; fp->f_ops = &badfileops; ref = (fp->f_flag & FHASLOCK) != 0; error = vn_close1(vp, fp->f_flag, fp->f_cred, td, ref); if (__predict_false(ref)) { lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; lf.l_type = F_UNLCK; (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK); vrele(vp); } return (error); } /* * Preparing to start a filesystem write operation. If the operation is * permitted, then we bump the count of operations in progress and * proceed. If a suspend request is in progress, we wait until the * suspension is over, and then proceed. */ static int vn_start_write_refed(struct mount *mp, int flags, bool mplocked) { struct mount_pcpu *mpcpu; int error, mflags; if (__predict_true(!mplocked) && (flags & V_XSLEEP) == 0 && vfs_op_thread_enter(mp, mpcpu)) { MPASS((mp->mnt_kern_flag & MNTK_SUSPEND) == 0); vfs_mp_count_add_pcpu(mpcpu, writeopcount, 1); vfs_op_thread_exit(mp, mpcpu); return (0); } if (mplocked) mtx_assert(MNT_MTX(mp), MA_OWNED); else MNT_ILOCK(mp); error = 0; /* * Check on status of suspension. */ if ((curthread->td_pflags & TDP_IGNSUSP) == 0 || mp->mnt_susp_owner != curthread) { mflags = ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0) | (PUSER - 1); while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { if (flags & V_NOWAIT) { error = EWOULDBLOCK; goto unlock; } error = msleep(&mp->mnt_flag, MNT_MTX(mp), mflags, "suspfs", 0); if (error) goto unlock; } } if (flags & V_XSLEEP) goto unlock; mp->mnt_writeopcount++; unlock: if (error != 0 || (flags & V_XSLEEP) != 0) MNT_REL(mp); MNT_IUNLOCK(mp); return (error); } int vn_start_write(struct vnode *vp, struct mount **mpp, int flags) { struct mount *mp; int error; KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL), ("V_MNTREF requires mp")); error = 0; /* * If a vnode is provided, get and return the mount point that * to which it will write. */ if (vp != NULL) { if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) { *mpp = NULL; if (error != EOPNOTSUPP) return (error); return (0); } } if ((mp = *mpp) == NULL) return (0); /* * VOP_GETWRITEMOUNT() returns with the mp refcount held through * a vfs_ref(). * As long as a vnode is not provided we need to acquire a * refcount for the provided mountpoint too, in order to * emulate a vfs_ref(). */ if (vp == NULL && (flags & V_MNTREF) == 0) vfs_ref(mp); return (vn_start_write_refed(mp, flags, false)); } /* * Secondary suspension. Used by operations such as vop_inactive * routines that are needed by the higher level functions. These * are allowed to proceed until all the higher level functions have * completed (indicated by mnt_writeopcount dropping to zero). At that * time, these operations are halted until the suspension is over. */ int vn_start_secondary_write(struct vnode *vp, struct mount **mpp, int flags) { struct mount *mp; int error; KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL), ("V_MNTREF requires mp")); retry: if (vp != NULL) { if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) { *mpp = NULL; if (error != EOPNOTSUPP) return (error); return (0); } } /* * If we are not suspended or have not yet reached suspended * mode, then let the operation proceed. */ if ((mp = *mpp) == NULL) return (0); /* * VOP_GETWRITEMOUNT() returns with the mp refcount held through * a vfs_ref(). * As long as a vnode is not provided we need to acquire a * refcount for the provided mountpoint too, in order to * emulate a vfs_ref(). */ MNT_ILOCK(mp); if (vp == NULL && (flags & V_MNTREF) == 0) MNT_REF(mp); if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) { mp->mnt_secondary_writes++; mp->mnt_secondary_accwrites++; MNT_IUNLOCK(mp); return (0); } if (flags & V_NOWAIT) { MNT_REL(mp); MNT_IUNLOCK(mp); return (EWOULDBLOCK); } /* * Wait for the suspension to finish. */ error = msleep(&mp->mnt_flag, MNT_MTX(mp), (PUSER - 1) | PDROP | ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0), "suspfs", 0); vfs_rel(mp); if (error == 0) goto retry; return (error); } /* * Filesystem write operation has completed. If we are suspending and this * operation is the last one, notify the suspender that the suspension is * now in effect. */ void vn_finished_write(struct mount *mp) { struct mount_pcpu *mpcpu; int c; if (mp == NULL) return; if (vfs_op_thread_enter(mp, mpcpu)) { vfs_mp_count_sub_pcpu(mpcpu, writeopcount, 1); vfs_mp_count_sub_pcpu(mpcpu, ref, 1); vfs_op_thread_exit(mp, mpcpu); return; } MNT_ILOCK(mp); vfs_assert_mount_counters(mp); MNT_REL(mp); c = --mp->mnt_writeopcount; if (mp->mnt_vfs_ops == 0) { MPASS((mp->mnt_kern_flag & MNTK_SUSPEND) == 0); MNT_IUNLOCK(mp); return; } if (c < 0) vfs_dump_mount_counters(mp); if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 && c == 0) wakeup(&mp->mnt_writeopcount); MNT_IUNLOCK(mp); } /* * Filesystem secondary write operation has completed. If we are * suspending and this operation is the last one, notify the suspender * that the suspension is now in effect. */ void vn_finished_secondary_write(struct mount *mp) { if (mp == NULL) return; MNT_ILOCK(mp); MNT_REL(mp); mp->mnt_secondary_writes--; if (mp->mnt_secondary_writes < 0) panic("vn_finished_secondary_write: neg cnt"); if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 && mp->mnt_secondary_writes <= 0) wakeup(&mp->mnt_secondary_writes); MNT_IUNLOCK(mp); } /* * Request a filesystem to suspend write operations. */ int vfs_write_suspend(struct mount *mp, int flags) { int error; vfs_op_enter(mp); MNT_ILOCK(mp); vfs_assert_mount_counters(mp); if (mp->mnt_susp_owner == curthread) { vfs_op_exit_locked(mp); MNT_IUNLOCK(mp); return (EALREADY); } while (mp->mnt_kern_flag & MNTK_SUSPEND) msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0); /* * Unmount holds a write reference on the mount point. If we * own busy reference and drain for writers, we deadlock with * the reference draining in the unmount path. Callers of * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if * vfs_busy() reference is owned and caller is not in the * unmount context. */ if ((flags & VS_SKIP_UNMOUNT) != 0 && (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) { vfs_op_exit_locked(mp); MNT_IUNLOCK(mp); return (EBUSY); } mp->mnt_kern_flag |= MNTK_SUSPEND; mp->mnt_susp_owner = curthread; if (mp->mnt_writeopcount > 0) (void) msleep(&mp->mnt_writeopcount, MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0); else MNT_IUNLOCK(mp); if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0) { vfs_write_resume(mp, 0); /* vfs_write_resume does vfs_op_exit() for us */ } return (error); } /* * Request a filesystem to resume write operations. */ void vfs_write_resume(struct mount *mp, int flags) { MNT_ILOCK(mp); if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner")); mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 | MNTK_SUSPENDED); mp->mnt_susp_owner = NULL; wakeup(&mp->mnt_writeopcount); wakeup(&mp->mnt_flag); curthread->td_pflags &= ~TDP_IGNSUSP; if ((flags & VR_START_WRITE) != 0) { MNT_REF(mp); mp->mnt_writeopcount++; } MNT_IUNLOCK(mp); if ((flags & VR_NO_SUSPCLR) == 0) VFS_SUSP_CLEAN(mp); vfs_op_exit(mp); } else if ((flags & VR_START_WRITE) != 0) { MNT_REF(mp); vn_start_write_refed(mp, 0, true); } else { MNT_IUNLOCK(mp); } } /* * Helper loop around vfs_write_suspend() for filesystem unmount VFS * methods. */ int vfs_write_suspend_umnt(struct mount *mp) { int error; KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0, ("vfs_write_suspend_umnt: recursed")); /* dounmount() already called vn_start_write(). */ for (;;) { vn_finished_write(mp); error = vfs_write_suspend(mp, 0); if (error != 0) { vn_start_write(NULL, &mp, V_WAIT); return (error); } MNT_ILOCK(mp); if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0) break; MNT_IUNLOCK(mp); vn_start_write(NULL, &mp, V_WAIT); } mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2); wakeup(&mp->mnt_flag); MNT_IUNLOCK(mp); curthread->td_pflags |= TDP_IGNSUSP; return (0); } /* * Implement kqueues for files by translating it to vnode operation. */ static int vn_kqfilter(struct file *fp, struct knote *kn) { return (VOP_KQFILTER(fp->f_vnode, kn)); } int vn_kqfilter_opath(struct file *fp, struct knote *kn) { if ((fp->f_flag & FKQALLOWED) == 0) return (EBADF); return (vn_kqfilter(fp, kn)); } /* * Simplified in-kernel wrapper calls for extended attribute access. * Both calls pass in a NULL credential, authorizing as "kernel" access. * Set IO_NODELOCKED in ioflg if the vnode is already locked. */ int vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, int *buflen, char *buf, struct thread *td) { struct uio auio; struct iovec iov; int error; iov.iov_len = *buflen; iov.iov_base = buf; auio.uio_iov = &iov; auio.uio_iovcnt = 1; auio.uio_rw = UIO_READ; auio.uio_segflg = UIO_SYSSPACE; auio.uio_td = td; auio.uio_offset = 0; auio.uio_resid = *buflen; if ((ioflg & IO_NODELOCKED) == 0) vn_lock(vp, LK_SHARED | LK_RETRY); ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); /* authorize attribute retrieval as kernel */ error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL, td); if ((ioflg & IO_NODELOCKED) == 0) VOP_UNLOCK(vp); if (error == 0) { *buflen = *buflen - auio.uio_resid; } return (error); } /* * XXX failure mode if partially written? */ int vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, int buflen, char *buf, struct thread *td) { struct uio auio; struct iovec iov; struct mount *mp; int error; iov.iov_len = buflen; iov.iov_base = buf; auio.uio_iov = &iov; auio.uio_iovcnt = 1; auio.uio_rw = UIO_WRITE; auio.uio_segflg = UIO_SYSSPACE; auio.uio_td = td; auio.uio_offset = 0; auio.uio_resid = buflen; if ((ioflg & IO_NODELOCKED) == 0) { if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0) return (error); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); } ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); /* authorize attribute setting as kernel */ error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td); if ((ioflg & IO_NODELOCKED) == 0) { vn_finished_write(mp); VOP_UNLOCK(vp); } return (error); } int vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, struct thread *td) { struct mount *mp; int error; if ((ioflg & IO_NODELOCKED) == 0) { if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0) return (error); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); } ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); /* authorize attribute removal as kernel */ error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td); if (error == EOPNOTSUPP) error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL, NULL, td); if ((ioflg & IO_NODELOCKED) == 0) { vn_finished_write(mp); VOP_UNLOCK(vp); } return (error); } static int vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags, struct vnode **rvp) { return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp)); } int vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp) { return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino, lkflags, rvp)); } int vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg, int lkflags, struct vnode **rvp) { struct mount *mp; int ltype, error; ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get"); mp = vp->v_mount; ltype = VOP_ISLOCKED(vp); KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED, ("vn_vget_ino: vp not locked")); error = vfs_busy(mp, MBF_NOWAIT); if (error != 0) { vfs_ref(mp); VOP_UNLOCK(vp); error = vfs_busy(mp, 0); vn_lock(vp, ltype | LK_RETRY); vfs_rel(mp); if (error != 0) return (ENOENT); if (VN_IS_DOOMED(vp)) { vfs_unbusy(mp); return (ENOENT); } } VOP_UNLOCK(vp); error = alloc(mp, alloc_arg, lkflags, rvp); vfs_unbusy(mp); if (error != 0 || *rvp != vp) vn_lock(vp, ltype | LK_RETRY); if (VN_IS_DOOMED(vp)) { if (error == 0) { if (*rvp == vp) vunref(vp); else vput(*rvp); } error = ENOENT; } return (error); } int vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio, struct thread *td) { off_t lim; bool ktr_write; if (td == NULL) return (0); /* * There are conditions where the limit is to be ignored. * However, since it is almost never reached, check it first. */ ktr_write = (td->td_pflags & TDP_INKTRACE) != 0; lim = lim_cur(td, RLIMIT_FSIZE); if (__predict_false(ktr_write)) lim = td->td_ktr_io_lim; if (__predict_true((uoff_t)uio->uio_offset + uio->uio_resid <= lim)) return (0); /* * The limit is reached. */ if (vp->v_type != VREG || (td->td_pflags2 & TDP2_ACCT) != 0) return (0); if (!ktr_write || ktr_filesize_limit_signal) { PROC_LOCK(td->td_proc); kern_psignal(td->td_proc, SIGXFSZ); PROC_UNLOCK(td->td_proc); } return (EFBIG); } int vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, struct thread *td) { struct vnode *vp; vp = fp->f_vnode; #ifdef AUDIT vn_lock(vp, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(vp); VOP_UNLOCK(vp); #endif return (setfmode(td, active_cred, vp, mode)); } int vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, struct thread *td) { struct vnode *vp; vp = fp->f_vnode; #ifdef AUDIT vn_lock(vp, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(vp); VOP_UNLOCK(vp); #endif return (setfown(td, active_cred, vp, uid, gid)); } void vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end) { vm_object_t object; if ((object = vp->v_object) == NULL) return; VM_OBJECT_WLOCK(object); vm_object_page_remove(object, start, end, 0); VM_OBJECT_WUNLOCK(object); } int vn_bmap_seekhole_locked(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred) { struct vattr va; daddr_t bn, bnp; uint64_t bsize; off_t noff; int error; KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA, ("%s: Wrong command %lu", __func__, cmd)); ASSERT_VOP_LOCKED(vp, "vn_bmap_seekhole_locked"); if (vp->v_type != VREG) { error = ENOTTY; goto out; } error = VOP_GETATTR(vp, &va, cred); if (error != 0) goto out; noff = *off; if (noff >= va.va_size) { error = ENXIO; goto out; } bsize = vp->v_mount->mnt_stat.f_iosize; for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize - noff % bsize) { error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL); if (error == EOPNOTSUPP) { error = ENOTTY; goto out; } if ((bnp == -1 && cmd == FIOSEEKHOLE) || (bnp != -1 && cmd == FIOSEEKDATA)) { noff = bn * bsize; if (noff < *off) noff = *off; goto out; } } if (noff > va.va_size) noff = va.va_size; /* noff == va.va_size. There is an implicit hole at the end of file. */ if (cmd == FIOSEEKDATA) error = ENXIO; out: if (error == 0) *off = noff; return (error); } int vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred) { int error; KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA, ("%s: Wrong command %lu", __func__, cmd)); if (vn_lock(vp, LK_SHARED) != 0) return (EBADF); error = vn_bmap_seekhole_locked(vp, cmd, off, cred); VOP_UNLOCK(vp); return (error); } int vn_seek(struct file *fp, off_t offset, int whence, struct thread *td) { struct ucred *cred; struct vnode *vp; struct vattr vattr; off_t foffset, size; int error, noneg; cred = td->td_ucred; vp = fp->f_vnode; foffset = foffset_lock(fp, 0); noneg = (vp->v_type != VCHR); error = 0; switch (whence) { case L_INCR: if (noneg && (foffset < 0 || (offset > 0 && foffset > OFF_MAX - offset))) { error = EOVERFLOW; break; } offset += foffset; break; case L_XTND: vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_GETATTR(vp, &vattr, cred); VOP_UNLOCK(vp); if (error) break; /* * If the file references a disk device, then fetch * the media size and use that to determine the ending * offset. */ if (vattr.va_size == 0 && vp->v_type == VCHR && fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0) vattr.va_size = size; if (noneg && (vattr.va_size > OFF_MAX || (offset > 0 && vattr.va_size > OFF_MAX - offset))) { error = EOVERFLOW; break; } offset += vattr.va_size; break; case L_SET: break; case SEEK_DATA: error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td); if (error == ENOTTY) error = EINVAL; break; case SEEK_HOLE: error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td); if (error == ENOTTY) error = EINVAL; break; default: error = EINVAL; } if (error == 0 && noneg && offset < 0) error = EINVAL; if (error != 0) goto drop; VFS_KNOTE_UNLOCKED(vp, 0); td->td_uretoff.tdu_off = offset; drop: foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0); return (error); } int vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred, struct thread *td) { int error; /* * Grant permission if the caller is the owner of the file, or * the super-user, or has ACL_WRITE_ATTRIBUTES permission on * on the file. If the time pointer is null, then write * permission on the file is also sufficient. * * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes: * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES * will be allowed to set the times [..] to the current * server time. */ error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td); if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0) error = VOP_ACCESS(vp, VWRITE, cred, td); return (error); } int vn_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) { struct vnode *vp; int error; if (fp->f_type == DTYPE_FIFO) kif->kf_type = KF_TYPE_FIFO; else kif->kf_type = KF_TYPE_VNODE; vp = fp->f_vnode; vref(vp); FILEDESC_SUNLOCK(fdp); error = vn_fill_kinfo_vnode(vp, kif); vrele(vp); FILEDESC_SLOCK(fdp); return (error); } static inline void vn_fill_junk(struct kinfo_file *kif) { size_t len, olen; /* * Simulate vn_fullpath returning changing values for a given * vp during e.g. coredump. */ len = (arc4random() % (sizeof(kif->kf_path) - 2)) + 1; olen = strlen(kif->kf_path); if (len < olen) strcpy(&kif->kf_path[len - 1], "$"); else for (; olen < len; olen++) strcpy(&kif->kf_path[olen], "A"); } int vn_fill_kinfo_vnode(struct vnode *vp, struct kinfo_file *kif) { struct vattr va; char *fullpath, *freepath; int error; kif->kf_un.kf_file.kf_file_type = vntype_to_kinfo(vp->v_type); freepath = NULL; fullpath = "-"; error = vn_fullpath(vp, &fullpath, &freepath); if (error == 0) { strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); } if (freepath != NULL) free(freepath, M_TEMP); KFAIL_POINT_CODE(DEBUG_FP, fill_kinfo_vnode__random_path, vn_fill_junk(kif); ); /* * Retrieve vnode attributes. */ va.va_fsid = VNOVAL; va.va_rdev = NODEV; vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_GETATTR(vp, &va, curthread->td_ucred); VOP_UNLOCK(vp); if (error != 0) return (error); if (va.va_fsid != VNOVAL) kif->kf_un.kf_file.kf_file_fsid = va.va_fsid; else kif->kf_un.kf_file.kf_file_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; kif->kf_un.kf_file.kf_file_fsid_freebsd11 = kif->kf_un.kf_file.kf_file_fsid; /* truncate */ kif->kf_un.kf_file.kf_file_fileid = va.va_fileid; kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode); kif->kf_un.kf_file.kf_file_size = va.va_size; kif->kf_un.kf_file.kf_file_rdev = va.va_rdev; kif->kf_un.kf_file.kf_file_rdev_freebsd11 = kif->kf_un.kf_file.kf_file_rdev; /* truncate */ return (0); } int vn_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff, struct thread *td) { #ifdef HWPMC_HOOKS struct pmckern_map_in pkm; #endif struct mount *mp; struct vnode *vp; vm_object_t object; vm_prot_t maxprot; boolean_t writecounted; int error; #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \ defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) /* * POSIX shared-memory objects are defined to have * kernel persistence, and are not defined to support * read(2)/write(2) -- or even open(2). Thus, we can * use MAP_ASYNC to trade on-disk coherence for speed. * The shm_open(3) library routine turns on the FPOSIXSHM * flag to request this behavior. */ if ((fp->f_flag & FPOSIXSHM) != 0) flags |= MAP_NOSYNC; #endif vp = fp->f_vnode; /* * Ensure that file and memory protections are * compatible. Note that we only worry about * writability if mapping is shared; in this case, * current and max prot are dictated by the open file. * XXX use the vnode instead? Problem is: what * credentials do we use for determination? What if * proc does a setuid? */ mp = vp->v_mount; if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) { maxprot = VM_PROT_NONE; if ((prot & VM_PROT_EXECUTE) != 0) return (EACCES); } else maxprot = VM_PROT_EXECUTE; if ((fp->f_flag & FREAD) != 0) maxprot |= VM_PROT_READ; else if ((prot & VM_PROT_READ) != 0) return (EACCES); /* * If we are sharing potential changes via MAP_SHARED and we * are trying to get write permission although we opened it * without asking for it, bail out. */ if ((flags & MAP_SHARED) != 0) { if ((fp->f_flag & FWRITE) != 0) maxprot |= VM_PROT_WRITE; else if ((prot & VM_PROT_WRITE) != 0) return (EACCES); } else { maxprot |= VM_PROT_WRITE; cap_maxprot |= VM_PROT_WRITE; } maxprot &= cap_maxprot; /* * For regular files and shared memory, POSIX requires that * the value of foff be a legitimate offset within the data * object. In particular, negative offsets are invalid. * Blocking negative offsets and overflows here avoids * possible wraparound or user-level access into reserved * ranges of the data object later. In contrast, POSIX does * not dictate how offsets are used by device drivers, so in * the case of a device mapping a negative offset is passed * on. */ if ( #ifdef _LP64 size > OFF_MAX || #endif foff > OFF_MAX - size) return (EINVAL); writecounted = FALSE; error = vm_mmap_vnode(td, size, prot, &maxprot, &flags, vp, &foff, &object, &writecounted); if (error != 0) return (error); error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object, foff, writecounted, td); if (error != 0) { /* * If this mapping was accounted for in the vnode's * writecount, then undo that now. */ if (writecounted) vm_pager_release_writecount(object, 0, size); vm_object_deallocate(object); } #ifdef HWPMC_HOOKS /* Inform hwpmc(4) if an executable is being mapped. */ if (PMC_HOOK_INSTALLED(PMC_FN_MMAP)) { if ((prot & VM_PROT_EXECUTE) != 0 && error == 0) { pkm.pm_file = vp; pkm.pm_address = (uintptr_t) *addr; PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_MMAP, (void *) &pkm); } } #endif return (error); } void vn_fsid(struct vnode *vp, struct vattr *va) { fsid_t *f; f = &vp->v_mount->mnt_stat.f_fsid; va->va_fsid = (uint32_t)f->val[1]; va->va_fsid <<= sizeof(f->val[1]) * NBBY; va->va_fsid += (uint32_t)f->val[0]; } int vn_fsync_buf(struct vnode *vp, int waitfor) { struct buf *bp, *nbp; struct bufobj *bo; struct mount *mp; int error, maxretry; error = 0; maxretry = 10000; /* large, arbitrarily chosen */ mp = NULL; if (vp->v_type == VCHR) { VI_LOCK(vp); mp = vp->v_rdev->si_mountpt; VI_UNLOCK(vp); } bo = &vp->v_bufobj; BO_LOCK(bo); loop1: /* * MARK/SCAN initialization to avoid infinite loops. */ TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) { bp->b_vflags &= ~BV_SCANNED; bp->b_error = 0; } /* * Flush all dirty buffers associated with a vnode. */ loop2: TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { if ((bp->b_vflags & BV_SCANNED) != 0) continue; bp->b_vflags |= BV_SCANNED; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { if (waitfor != MNT_WAIT) continue; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_INTERLOCK | LK_SLEEPFAIL, BO_LOCKPTR(bo)) != 0) { BO_LOCK(bo); goto loop1; } BO_LOCK(bo); } BO_UNLOCK(bo); KASSERT(bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); if ((bp->b_flags & B_DELWRI) == 0) panic("fsync: not dirty"); if ((vp->v_object != NULL) && (bp->b_flags & B_CLUSTEROK)) { vfs_bio_awrite(bp); } else { bremfree(bp); bawrite(bp); } if (maxretry < 1000) pause("dirty", hz < 1000 ? 1 : hz / 1000); BO_LOCK(bo); goto loop2; } /* * If synchronous the caller expects us to completely resolve all * dirty buffers in the system. Wait for in-progress I/O to * complete (which could include background bitmap writes), then * retry if dirty blocks still exist. */ if (waitfor == MNT_WAIT) { bufobj_wwait(bo, 0, 0); if (bo->bo_dirty.bv_cnt > 0) { /* * If we are unable to write any of these buffers * then we fail now rather than trying endlessly * to write them out. */ TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) if ((error = bp->b_error) != 0) break; if ((mp != NULL && mp->mnt_secondary_writes > 0) || (error == 0 && --maxretry >= 0)) goto loop1; if (error == 0) error = EAGAIN; } } BO_UNLOCK(bo); if (error != 0) vn_printf(vp, "fsync: giving up on dirty (error = %d) ", error); return (error); } /* * Copies a byte range from invp to outvp. Calls VOP_COPY_FILE_RANGE() * or vn_generic_copy_file_range() after rangelocking the byte ranges, * to do the actual copy. * vn_generic_copy_file_range() is factored out, so it can be called * from a VOP_COPY_FILE_RANGE() call as well, but handles vnodes from * different file systems. */ int vn_copy_file_range(struct vnode *invp, off_t *inoffp, struct vnode *outvp, off_t *outoffp, size_t *lenp, unsigned int flags, struct ucred *incred, struct ucred *outcred, struct thread *fsize_td) { int error; size_t len; uint64_t uval; len = *lenp; *lenp = 0; /* For error returns. */ error = 0; /* Do some sanity checks on the arguments. */ if (invp->v_type == VDIR || outvp->v_type == VDIR) error = EISDIR; else if (*inoffp < 0 || *outoffp < 0 || invp->v_type != VREG || outvp->v_type != VREG) error = EINVAL; if (error != 0) goto out; /* Ensure offset + len does not wrap around. */ uval = *inoffp; uval += len; if (uval > INT64_MAX) len = INT64_MAX - *inoffp; uval = *outoffp; uval += len; if (uval > INT64_MAX) len = INT64_MAX - *outoffp; if (len == 0) goto out; /* * If the two vnode are for the same file system, call * VOP_COPY_FILE_RANGE(), otherwise call vn_generic_copy_file_range() * which can handle copies across multiple file systems. */ *lenp = len; if (invp->v_mount == outvp->v_mount) error = VOP_COPY_FILE_RANGE(invp, inoffp, outvp, outoffp, lenp, flags, incred, outcred, fsize_td); else error = vn_generic_copy_file_range(invp, inoffp, outvp, outoffp, lenp, flags, incred, outcred, fsize_td); out: return (error); } /* * Test len bytes of data starting at dat for all bytes == 0. * Return true if all bytes are zero, false otherwise. * Expects dat to be well aligned. */ static bool mem_iszero(void *dat, int len) { int i; const u_int *p; const char *cp; for (p = dat; len > 0; len -= sizeof(*p), p++) { if (len >= sizeof(*p)) { if (*p != 0) return (false); } else { cp = (const char *)p; for (i = 0; i < len; i++, cp++) if (*cp != '\0') return (false); } } return (true); } /* * Look for a hole in the output file and, if found, adjust *outoffp * and *xferp to skip past the hole. * *xferp is the entire hole length to be written and xfer2 is how many bytes * to be written as 0's upon return. */ static off_t vn_skip_hole(struct vnode *outvp, off_t xfer2, off_t *outoffp, off_t *xferp, off_t *dataoffp, off_t *holeoffp, struct ucred *cred) { int error; off_t delta; if (*holeoffp == 0 || *holeoffp <= *outoffp) { *dataoffp = *outoffp; error = VOP_IOCTL(outvp, FIOSEEKDATA, dataoffp, 0, cred, curthread); if (error == 0) { *holeoffp = *dataoffp; error = VOP_IOCTL(outvp, FIOSEEKHOLE, holeoffp, 0, cred, curthread); } if (error != 0 || *holeoffp == *dataoffp) { /* * Since outvp is unlocked, it may be possible for * another thread to do a truncate(), lseek(), write() * creating a hole at startoff between the above * VOP_IOCTL() calls, if the other thread does not do * rangelocking. * If that happens, *holeoffp == *dataoffp and finding * the hole has failed, so disable vn_skip_hole(). */ *holeoffp = -1; /* Disable use of vn_skip_hole(). */ return (xfer2); } KASSERT(*dataoffp >= *outoffp, ("vn_skip_hole: dataoff=%jd < outoff=%jd", (intmax_t)*dataoffp, (intmax_t)*outoffp)); KASSERT(*holeoffp > *dataoffp, ("vn_skip_hole: holeoff=%jd <= dataoff=%jd", (intmax_t)*holeoffp, (intmax_t)*dataoffp)); } /* * If there is a hole before the data starts, advance *outoffp and * *xferp past the hole. */ if (*dataoffp > *outoffp) { delta = *dataoffp - *outoffp; if (delta >= *xferp) { /* Entire *xferp is a hole. */ *outoffp += *xferp; *xferp = 0; return (0); } *xferp -= delta; *outoffp += delta; xfer2 = MIN(xfer2, *xferp); } /* * If a hole starts before the end of this xfer2, reduce this xfer2 so * that the write ends at the start of the hole. * *holeoffp should always be greater than *outoffp, but for the * non-INVARIANTS case, check this to make sure xfer2 remains a sane * value. */ if (*holeoffp > *outoffp && *holeoffp < *outoffp + xfer2) xfer2 = *holeoffp - *outoffp; return (xfer2); } /* * Write an xfer sized chunk to outvp in blksize blocks from dat. * dat is a maximum of blksize in length and can be written repeatedly in * the chunk. * If growfile == true, just grow the file via vn_truncate_locked() instead * of doing actual writes. * If checkhole == true, a hole is being punched, so skip over any hole * already in the output file. */ static int vn_write_outvp(struct vnode *outvp, char *dat, off_t outoff, off_t xfer, u_long blksize, bool growfile, bool checkhole, struct ucred *cred) { struct mount *mp; off_t dataoff, holeoff, xfer2; int error; /* * Loop around doing writes of blksize until write has been completed. * Lock/unlock on each loop iteration so that a bwillwrite() can be * done for each iteration, since the xfer argument can be very * large if there is a large hole to punch in the output file. */ error = 0; holeoff = 0; do { xfer2 = MIN(xfer, blksize); if (checkhole) { /* * Punching a hole. Skip writing if there is * already a hole in the output file. */ xfer2 = vn_skip_hole(outvp, xfer2, &outoff, &xfer, &dataoff, &holeoff, cred); if (xfer == 0) break; if (holeoff < 0) checkhole = false; KASSERT(xfer2 > 0, ("vn_write_outvp: xfer2=%jd", (intmax_t)xfer2)); } bwillwrite(); mp = NULL; error = vn_start_write(outvp, &mp, V_WAIT); if (error != 0) break; if (growfile) { error = vn_lock(outvp, LK_EXCLUSIVE); if (error == 0) { error = vn_truncate_locked(outvp, outoff + xfer, false, cred); VOP_UNLOCK(outvp); } } else { error = vn_lock(outvp, vn_lktype_write(mp, outvp)); if (error == 0) { error = vn_rdwr(UIO_WRITE, outvp, dat, xfer2, outoff, UIO_SYSSPACE, IO_NODELOCKED, curthread->td_ucred, cred, NULL, curthread); outoff += xfer2; xfer -= xfer2; VOP_UNLOCK(outvp); } } if (mp != NULL) vn_finished_write(mp); } while (!growfile && xfer > 0 && error == 0); return (error); } /* * Copy a byte range of one file to another. This function can handle the * case where invp and outvp are on different file systems. * It can also be called by a VOP_COPY_FILE_RANGE() to do the work, if there * is no better file system specific way to do it. */ int vn_generic_copy_file_range(struct vnode *invp, off_t *inoffp, struct vnode *outvp, off_t *outoffp, size_t *lenp, unsigned int flags, struct ucred *incred, struct ucred *outcred, struct thread *fsize_td) { struct vattr va, inva; struct mount *mp; struct uio io; off_t startoff, endoff, xfer, xfer2; u_long blksize; int error, interrupted; bool cantseek, readzeros, eof, lastblock, holetoeof; ssize_t aresid; size_t copylen, len, rem, savlen; char *dat; long holein, holeout; struct timespec curts, endts; holein = holeout = 0; savlen = len = *lenp; error = 0; interrupted = 0; dat = NULL; error = vn_lock(invp, LK_SHARED); if (error != 0) goto out; if (VOP_PATHCONF(invp, _PC_MIN_HOLE_SIZE, &holein) != 0) holein = 0; if (holein > 0) error = VOP_GETATTR(invp, &inva, incred); VOP_UNLOCK(invp); if (error != 0) goto out; mp = NULL; error = vn_start_write(outvp, &mp, V_WAIT); if (error == 0) error = vn_lock(outvp, LK_EXCLUSIVE); if (error == 0) { /* * If fsize_td != NULL, do a vn_rlimit_fsize() call, * now that outvp is locked. */ if (fsize_td != NULL) { io.uio_offset = *outoffp; io.uio_resid = len; error = vn_rlimit_fsize(outvp, &io, fsize_td); if (error != 0) error = EFBIG; } if (VOP_PATHCONF(outvp, _PC_MIN_HOLE_SIZE, &holeout) != 0) holeout = 0; /* * Holes that are past EOF do not need to be written as a block * of zero bytes. So, truncate the output file as far as * possible and then use va.va_size to decide if writing 0 * bytes is necessary in the loop below. */ if (error == 0) error = VOP_GETATTR(outvp, &va, outcred); if (error == 0 && va.va_size > *outoffp && va.va_size <= *outoffp + len) { #ifdef MAC error = mac_vnode_check_write(curthread->td_ucred, outcred, outvp); if (error == 0) #endif error = vn_truncate_locked(outvp, *outoffp, false, outcred); if (error == 0) va.va_size = *outoffp; } VOP_UNLOCK(outvp); } if (mp != NULL) vn_finished_write(mp); if (error != 0) goto out; /* * Set the blksize to the larger of the hole sizes for invp and outvp. * If hole sizes aren't available, set the blksize to the larger * f_iosize of invp and outvp. * This code expects the hole sizes and f_iosizes to be powers of 2. * This value is clipped at 4Kbytes and 1Mbyte. */ blksize = MAX(holein, holeout); /* Clip len to end at an exact multiple of hole size. */ if (blksize > 1) { rem = *inoffp % blksize; if (rem > 0) rem = blksize - rem; if (len > rem && len - rem > blksize) len = savlen = rounddown(len - rem, blksize) + rem; } if (blksize <= 1) blksize = MAX(invp->v_mount->mnt_stat.f_iosize, outvp->v_mount->mnt_stat.f_iosize); if (blksize < 4096) blksize = 4096; else if (blksize > 1024 * 1024) blksize = 1024 * 1024; dat = malloc(blksize, M_TEMP, M_WAITOK); /* * If VOP_IOCTL(FIOSEEKHOLE) works for invp, use it and FIOSEEKDATA * to find holes. Otherwise, just scan the read block for all 0s * in the inner loop where the data copying is done. * Note that some file systems such as NFSv3, NFSv4.0 and NFSv4.1 may * support holes on the server, but do not support FIOSEEKHOLE. * The kernel flag COPY_FILE_RANGE_TIMEO1SEC is used to indicate * that this function should return after 1second with a partial * completion. */ if ((flags & COPY_FILE_RANGE_TIMEO1SEC) != 0) { getnanouptime(&endts); endts.tv_sec++; } else timespecclear(&endts); holetoeof = eof = false; while (len > 0 && error == 0 && !eof && interrupted == 0) { endoff = 0; /* To shut up compilers. */ cantseek = true; startoff = *inoffp; copylen = len; /* * Find the next data area. If there is just a hole to EOF, * FIOSEEKDATA should fail with ENXIO. * (I do not know if any file system will report a hole to * EOF via FIOSEEKHOLE, but I am pretty sure FIOSEEKDATA * will fail for those file systems.) * * For input files that don't support FIOSEEKDATA/FIOSEEKHOLE, * the code just falls through to the inner copy loop. */ error = EINVAL; if (holein > 0) { error = VOP_IOCTL(invp, FIOSEEKDATA, &startoff, 0, incred, curthread); if (error == ENXIO) { startoff = endoff = inva.va_size; eof = holetoeof = true; error = 0; } } if (error == 0 && !holetoeof) { endoff = startoff; error = VOP_IOCTL(invp, FIOSEEKHOLE, &endoff, 0, incred, curthread); /* * Since invp is unlocked, it may be possible for * another thread to do a truncate(), lseek(), write() * creating a hole at startoff between the above * VOP_IOCTL() calls, if the other thread does not do * rangelocking. * If that happens, startoff == endoff and finding * the hole has failed, so set an error. */ if (error == 0 && startoff == endoff) error = EINVAL; /* Any error. Reset to 0. */ } if (error == 0) { if (startoff > *inoffp) { /* Found hole before data block. */ xfer = MIN(startoff - *inoffp, len); if (*outoffp < va.va_size) { /* Must write 0s to punch hole. */ xfer2 = MIN(va.va_size - *outoffp, xfer); memset(dat, 0, MIN(xfer2, blksize)); error = vn_write_outvp(outvp, dat, *outoffp, xfer2, blksize, false, holeout > 0, outcred); } if (error == 0 && *outoffp + xfer > va.va_size && (xfer == len || holetoeof)) { /* Grow output file (hole at end). */ error = vn_write_outvp(outvp, dat, *outoffp, xfer, blksize, true, false, outcred); } if (error == 0) { *inoffp += xfer; *outoffp += xfer; len -= xfer; if (len < savlen) { interrupted = sig_intr(); if (timespecisset(&endts) && interrupted == 0) { getnanouptime(&curts); if (timespeccmp(&curts, &endts, >=)) interrupted = EINTR; } } } } copylen = MIN(len, endoff - startoff); cantseek = false; } else { cantseek = true; startoff = *inoffp; copylen = len; error = 0; } xfer = blksize; if (cantseek) { /* * Set first xfer to end at a block boundary, so that * holes are more likely detected in the loop below via * the for all bytes 0 method. */ xfer -= (*inoffp % blksize); } /* Loop copying the data block. */ while (copylen > 0 && error == 0 && !eof && interrupted == 0) { if (copylen < xfer) xfer = copylen; error = vn_lock(invp, LK_SHARED); if (error != 0) goto out; error = vn_rdwr(UIO_READ, invp, dat, xfer, startoff, UIO_SYSSPACE, IO_NODELOCKED, curthread->td_ucred, incred, &aresid, curthread); VOP_UNLOCK(invp); lastblock = false; if (error == 0 && aresid > 0) { /* Stop the copy at EOF on the input file. */ xfer -= aresid; eof = true; lastblock = true; } if (error == 0) { /* * Skip the write for holes past the initial EOF * of the output file, unless this is the last * write of the output file at EOF. */ readzeros = cantseek ? mem_iszero(dat, xfer) : false; if (xfer == len) lastblock = true; if (!cantseek || *outoffp < va.va_size || lastblock || !readzeros) error = vn_write_outvp(outvp, dat, *outoffp, xfer, blksize, readzeros && lastblock && *outoffp >= va.va_size, false, outcred); if (error == 0) { *inoffp += xfer; startoff += xfer; *outoffp += xfer; copylen -= xfer; len -= xfer; if (len < savlen) { interrupted = sig_intr(); if (timespecisset(&endts) && interrupted == 0) { getnanouptime(&curts); if (timespeccmp(&curts, &endts, >=)) interrupted = EINTR; } } } } xfer = blksize; } } out: *lenp = savlen - len; free(dat, M_TEMP); return (error); } static int vn_fallocate(struct file *fp, off_t offset, off_t len, struct thread *td) { struct mount *mp; struct vnode *vp; off_t olen, ooffset; int error; #ifdef AUDIT int audited_vnode1 = 0; #endif vp = fp->f_vnode; if (vp->v_type != VREG) return (ENODEV); /* Allocating blocks may take a long time, so iterate. */ for (;;) { olen = len; ooffset = offset; bwillwrite(); mp = NULL; error = vn_start_write(vp, &mp, V_WAIT | PCATCH); if (error != 0) break; error = vn_lock(vp, LK_EXCLUSIVE); if (error != 0) { vn_finished_write(mp); break; } #ifdef AUDIT if (!audited_vnode1) { AUDIT_ARG_VNODE1(vp); audited_vnode1 = 1; } #endif #ifdef MAC error = mac_vnode_check_write(td->td_ucred, fp->f_cred, vp); if (error == 0) #endif error = VOP_ALLOCATE(vp, &offset, &len); VOP_UNLOCK(vp); vn_finished_write(mp); if (olen + ooffset != offset + len) { panic("offset + len changed from %jx/%jx to %jx/%jx", ooffset, olen, offset, len); } if (error != 0 || len == 0) break; KASSERT(olen > len, ("Iteration did not make progress?")); maybe_yield(); } return (error); } static int vn_deallocate_impl(struct vnode *vp, off_t *offset, off_t *length, int flags, int ioflag, struct ucred *cred, struct ucred *active_cred, struct ucred *file_cred) { struct mount *mp; void *rl_cookie; off_t off, len; int error; #ifdef AUDIT bool audited_vnode1 = false; #endif rl_cookie = NULL; error = 0; mp = NULL; off = *offset; len = *length; if ((ioflag & (IO_NODELOCKED | IO_RANGELOCKED)) == 0) rl_cookie = vn_rangelock_wlock(vp, off, off + len); while (len > 0 && error == 0) { /* * Try to deallocate the longest range in one pass. * In case a pass takes too long to be executed, it returns * partial result. The residue will be proceeded in the next * pass. */ if ((ioflag & IO_NODELOCKED) == 0) { bwillwrite(); if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) goto out; vn_lock(vp, vn_lktype_write(mp, vp) | LK_RETRY); } #ifdef AUDIT if (!audited_vnode1) { AUDIT_ARG_VNODE1(vp); audited_vnode1 = true; } #endif #ifdef MAC if ((ioflag & IO_NOMACCHECK) == 0) error = mac_vnode_check_write(active_cred, file_cred, vp); #endif if (error == 0) error = VOP_DEALLOCATE(vp, &off, &len, flags, ioflag, cred); if ((ioflag & IO_NODELOCKED) == 0) { VOP_UNLOCK(vp); if (mp != NULL) { vn_finished_write(mp); mp = NULL; } } if (error == 0 && len != 0) maybe_yield(); } out: if (rl_cookie != NULL) vn_rangelock_unlock(vp, rl_cookie); *offset = off; *length = len; return (error); } /* * This function is supposed to be used in the situations where the deallocation * is not triggered by a user request. */ int vn_deallocate(struct vnode *vp, off_t *offset, off_t *length, int flags, int ioflag, struct ucred *active_cred, struct ucred *file_cred) { struct ucred *cred; if (*offset < 0 || *length <= 0 || *length > OFF_MAX - *offset || flags != 0) return (EINVAL); if (vp->v_type != VREG) return (ENODEV); cred = file_cred != NOCRED ? file_cred : active_cred; return (vn_deallocate_impl(vp, offset, length, flags, ioflag, cred, active_cred, file_cred)); } static int vn_fspacectl(struct file *fp, int cmd, off_t *offset, off_t *length, int flags, struct ucred *active_cred, struct thread *td) { int error; struct vnode *vp; int ioflag; vp = fp->f_vnode; if (cmd != SPACECTL_DEALLOC || *offset < 0 || *length <= 0 || *length > OFF_MAX - *offset || flags != 0) return (EINVAL); if (vp->v_type != VREG) return (ENODEV); ioflag = get_write_ioflag(fp); switch (cmd) { case SPACECTL_DEALLOC: error = vn_deallocate_impl(vp, offset, length, flags, ioflag, active_cred, active_cred, fp->f_cred); break; default: panic("vn_fspacectl: unknown cmd %d", cmd); } return (error); } static u_long vn_lock_pair_pause_cnt; SYSCTL_ULONG(_debug, OID_AUTO, vn_lock_pair_pause, CTLFLAG_RD, &vn_lock_pair_pause_cnt, 0, "Count of vn_lock_pair deadlocks"); u_int vn_lock_pair_pause_max; SYSCTL_UINT(_debug, OID_AUTO, vn_lock_pair_pause_max, CTLFLAG_RW, &vn_lock_pair_pause_max, 0, "Max ticks for vn_lock_pair deadlock avoidance sleep"); static void vn_lock_pair_pause(const char *wmesg) { atomic_add_long(&vn_lock_pair_pause_cnt, 1); pause(wmesg, prng32_bounded(vn_lock_pair_pause_max)); } /* * Lock pair of vnodes vp1, vp2, avoiding lock order reversal. * vp1_locked indicates whether vp1 is exclusively locked; if not, vp1 * must be unlocked. Same for vp2 and vp2_locked. One of the vnodes * can be NULL. * * The function returns with both vnodes exclusively locked, and * guarantees that it does not create lock order reversal with other * threads during its execution. Both vnodes could be unlocked * temporary (and reclaimed). */ void vn_lock_pair(struct vnode *vp1, bool vp1_locked, struct vnode *vp2, bool vp2_locked) { int error; if (vp1 == NULL && vp2 == NULL) return; if (vp1 != NULL) { if (vp1_locked) ASSERT_VOP_ELOCKED(vp1, "vp1"); else ASSERT_VOP_UNLOCKED(vp1, "vp1"); } else { vp1_locked = true; } if (vp2 != NULL) { if (vp2_locked) ASSERT_VOP_ELOCKED(vp2, "vp2"); else ASSERT_VOP_UNLOCKED(vp2, "vp2"); } else { vp2_locked = true; } if (!vp1_locked && !vp2_locked) { vn_lock(vp1, LK_EXCLUSIVE | LK_RETRY); vp1_locked = true; } for (;;) { if (vp1_locked && vp2_locked) break; if (vp1_locked && vp2 != NULL) { if (vp1 != NULL) { error = VOP_LOCK1(vp2, LK_EXCLUSIVE | LK_NOWAIT, __FILE__, __LINE__); if (error == 0) break; VOP_UNLOCK(vp1); vp1_locked = false; vn_lock_pair_pause("vlp1"); } vn_lock(vp2, LK_EXCLUSIVE | LK_RETRY); vp2_locked = true; } if (vp2_locked && vp1 != NULL) { if (vp2 != NULL) { error = VOP_LOCK1(vp1, LK_EXCLUSIVE | LK_NOWAIT, __FILE__, __LINE__); if (error == 0) break; VOP_UNLOCK(vp2); vp2_locked = false; vn_lock_pair_pause("vlp2"); } vn_lock(vp1, LK_EXCLUSIVE | LK_RETRY); vp1_locked = true; } } if (vp1 != NULL) ASSERT_VOP_ELOCKED(vp1, "vp1 ret"); if (vp2 != NULL) ASSERT_VOP_ELOCKED(vp2, "vp2 ret"); } int vn_lktype_write(struct mount *mp, struct vnode *vp) { if (MNT_SHARED_WRITES(mp) || (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) return (LK_SHARED); return (LK_EXCLUSIVE); } diff --git a/sys/sys/namei.h b/sys/sys/namei.h index 3f6e59289f0a..a467c6e1ebf3 100644 --- a/sys/sys/namei.h +++ b/sys/sys/namei.h @@ -1,331 +1,329 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1985, 1989, 1991, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)namei.h 8.5 (Berkeley) 1/9/95 * $FreeBSD$ */ #ifndef _SYS_NAMEI_H_ #define _SYS_NAMEI_H_ #include #include #include #include #include enum nameiop { LOOKUP, CREATE, DELETE, RENAME }; struct componentname { /* * Arguments to lookup. */ u_int64_t cn_origflags; /* flags to namei */ u_int64_t cn_flags; /* flags to namei */ - struct thread *cn_thread;/* thread requesting lookup */ struct ucred *cn_cred; /* credentials */ enum nameiop cn_nameiop; /* namei operation */ int cn_lkflags; /* Lock flags LK_EXCLUSIVE or LK_SHARED */ /* * Shared between lookup and commit routines. */ char *cn_pnbuf; /* pathname buffer */ char *cn_nameptr; /* pointer to looked up name */ long cn_namelen; /* length of looked up component */ }; struct nameicap_tracker; TAILQ_HEAD(nameicap_tracker_head, nameicap_tracker); /* * Encapsulation of namei parameters. */ struct nameidata { /* * Arguments to namei/lookup. */ const char *ni_dirp; /* pathname pointer */ enum uio_seg ni_segflg; /* location of pathname */ cap_rights_t *ni_rightsneeded; /* rights required to look up vnode */ /* * Arguments to lookup. */ struct vnode *ni_startdir; /* starting directory */ struct vnode *ni_rootdir; /* logical root directory */ struct vnode *ni_topdir; /* logical top directory */ int ni_dirfd; /* starting directory for *at functions */ int ni_lcf; /* local call flags */ /* * Results: returned from namei */ struct filecaps ni_filecaps; /* rights the *at base has */ /* * Results: returned from/manipulated by lookup */ struct vnode *ni_vp; /* vnode of result */ struct vnode *ni_dvp; /* vnode of intermediate directory */ /* * Results: flags returned from namei */ u_int ni_resflags; /* * Debug for validating API use by the callers. */ u_short ni_debugflags; /* * Shared between namei and lookup/commit routines. */ u_short ni_loopcnt; /* count of symlinks encountered */ size_t ni_pathlen; /* remaining chars in path */ char *ni_next; /* next location in pathname */ /* * Lookup parameters: this structure describes the subset of * information from the nameidata structure that is passed * through the VOP interface. */ struct componentname ni_cnd; struct nameicap_tracker_head ni_cap_tracker; /* * Private helper data for UFS, must be at the end. See * NDINIT_PREFILL(). */ seqc_t ni_dvp_seqc; seqc_t ni_vp_seqc; }; #ifdef _KERNEL enum cache_fpl_status { CACHE_FPL_STATUS_DESTROYED, CACHE_FPL_STATUS_ABORTED, CACHE_FPL_STATUS_PARTIAL, CACHE_FPL_STATUS_HANDLED, CACHE_FPL_STATUS_UNSET }; int cache_fplookup(struct nameidata *ndp, enum cache_fpl_status *status, struct pwd **pwdp); /* * Flags for namei. * * If modifying the list make sure to check whether NDVALIDATE needs updating. */ /* * Debug. */ #define NAMEI_DBG_INITED 0x0001 #define NAMEI_DBG_CALLED 0x0002 #define NAMEI_DBG_HADSTARTDIR 0x0004 /* * namei operational modifier flags, stored in ni_cnd.flags */ #define NC_NOMAKEENTRY 0x0001 /* name must not be added to cache */ #define NC_KEEPPOSENTRY 0x0002 /* don't evict a positive entry */ #define NOCACHE NC_NOMAKEENTRY /* for compatibility with older code */ #define LOCKLEAF 0x0004 /* lock vnode on return */ #define LOCKPARENT 0x0008 /* want parent vnode returned locked */ #define WANTPARENT 0x0010 /* want parent vnode returned unlocked */ #define FAILIFEXISTS 0x0020 /* return EEXIST if found */ #define FOLLOW 0x0040 /* follow symbolic links */ #define EMPTYPATH 0x0080 /* Allow empty path for *at */ #define LOCKSHARED 0x0100 /* Shared lock leaf */ #define NOFOLLOW 0x0000 /* do not follow symbolic links (pseudo) */ #define RBENEATH 0x100000000ULL /* No escape, even tmp, from start dir */ #define MODMASK 0xf000001ffULL /* mask of operational modifiers */ /* * Namei parameter descriptors. * * SAVENAME may be set by either the callers of namei or by VOP_LOOKUP. * If the caller of namei sets the flag (for example execve wants to * know the name of the program that is being executed), then it must * free the buffer. If VOP_LOOKUP sets the flag, then the buffer must * be freed by either the commit routine or the VOP_ABORT routine. * SAVESTART is set only by the callers of namei. It implies SAVENAME * plus the addition of saving the parent directory that contains the * name in ni_startdir. It allows repeated calls to lookup for the * name being sought. The caller is responsible for releasing the * buffer and for vrele'ing ni_startdir. */ #define RDONLY 0x00000200 /* lookup with read-only semantics */ #define SAVENAME 0x00000400 /* save pathname buffer */ #define SAVESTART 0x00000800 /* save starting directory */ #define ISWHITEOUT 0x00001000 /* found whiteout */ #define DOWHITEOUT 0x00002000 /* do whiteouts */ #define WILLBEDIR 0x00004000 /* new files will be dirs; allow trailing / */ #define ISOPEN 0x00008000 /* caller is opening; return a real vnode. */ #define NOCROSSMOUNT 0x00010000 /* do not cross mount points */ #define NOMACCHECK 0x00020000 /* do not perform MAC checks */ #define AUDITVNODE1 0x00040000 /* audit the looked up vnode information */ #define AUDITVNODE2 0x00080000 /* audit the looked up vnode information */ #define NOCAPCHECK 0x00100000 /* do not perform capability checks */ #define OPENREAD 0x00200000 /* open for reading */ #define OPENWRITE 0x00400000 /* open for writing */ /* UNUSED 0x00800000 */ #define HASBUF 0x01000000 /* has allocated pathname buffer */ #define NOEXECCHECK 0x02000000 /* do not perform exec check on dir */ #define MAKEENTRY 0x04000000 /* entry is to be added to name cache */ #define ISSYMLINK 0x08000000 /* symlink needs interpretation */ #define ISLASTCN 0x10000000 /* this is last component of pathname */ #define ISDOTDOT 0x20000000 /* current component name is .. */ #define TRAILINGSLASH 0x40000000 /* path ended in a slash */ #define PARAMASK 0x7ffffe00 /* mask of parameter descriptors */ /* * Flags which must not be passed in by callers. */ #define NAMEI_INTERNAL_FLAGS \ (HASBUF | NOEXECCHECK | MAKEENTRY | ISSYMLINK | ISLASTCN | ISDOTDOT | \ TRAILINGSLASH) /* * Namei results flags */ #define NIRES_ABS 0x00000001 /* Path was absolute */ #define NIRES_STRICTREL 0x00000002 /* Restricted lookup result */ #define NIRES_EMPTYPATH 0x00000004 /* EMPTYPATH used */ /* * Flags in ni_lcf, valid for the duration of the namei call. */ #define NI_LCF_STRICTRELATIVE 0x0001 /* relative lookup only */ #define NI_LCF_CAP_DOTDOT 0x0002 /* ".." in strictrelative case */ /* * Initialization of a nameidata structure. */ #define NDINIT(ndp, op, flags, segflg, namep, td) \ NDINIT_ALL(ndp, op, flags, segflg, namep, AT_FDCWD, NULL, &cap_no_rights, td) #define NDINIT_AT(ndp, op, flags, segflg, namep, dirfd, td) \ NDINIT_ALL(ndp, op, flags, segflg, namep, dirfd, NULL, &cap_no_rights, td) #define NDINIT_ATRIGHTS(ndp, op, flags, segflg, namep, dirfd, rightsp, td) \ NDINIT_ALL(ndp, op, flags, segflg, namep, dirfd, NULL, rightsp, td) #define NDINIT_ATVP(ndp, op, flags, segflg, namep, vp, td) \ NDINIT_ALL(ndp, op, flags, segflg, namep, AT_FDCWD, vp, &cap_no_rights, td) /* * Note the constant pattern may *hide* bugs. */ #ifdef INVARIANTS #define NDINIT_PREFILL(arg) memset(arg, 0xff, offsetof(struct nameidata, \ ni_dvp_seqc)) #define NDINIT_DBG(arg) { (arg)->ni_debugflags = NAMEI_DBG_INITED; } #define NDREINIT_DBG(arg) { \ if (((arg)->ni_debugflags & NAMEI_DBG_INITED) == 0) \ panic("namei data not inited"); \ if (((arg)->ni_debugflags & NAMEI_DBG_HADSTARTDIR) != 0) \ panic("NDREINIT on namei data with NAMEI_DBG_HADSTARTDIR"); \ (arg)->ni_debugflags = NAMEI_DBG_INITED; \ } #else #define NDINIT_PREFILL(arg) do { } while (0) #define NDINIT_DBG(arg) do { } while (0) #define NDREINIT_DBG(arg) do { } while (0) #endif #define NDINIT_ALL(ndp, op, flags, segflg, namep, dirfd, startdir, rightsp, td) \ do { \ struct nameidata *_ndp = (ndp); \ cap_rights_t *_rightsp = (rightsp); \ MPASS(_rightsp != NULL); \ NDINIT_PREFILL(_ndp); \ NDINIT_DBG(_ndp); \ _ndp->ni_cnd.cn_nameiop = op; \ _ndp->ni_cnd.cn_flags = flags; \ _ndp->ni_segflg = segflg; \ _ndp->ni_dirp = namep; \ _ndp->ni_dirfd = dirfd; \ _ndp->ni_startdir = startdir; \ _ndp->ni_resflags = 0; \ filecaps_init(&_ndp->ni_filecaps); \ - _ndp->ni_cnd.cn_thread = td; \ _ndp->ni_rightsneeded = _rightsp; \ } while (0) #define NDREINIT(ndp) do { \ struct nameidata *_ndp = (ndp); \ NDREINIT_DBG(_ndp); \ _ndp->ni_resflags = 0; \ _ndp->ni_startdir = NULL; \ } while (0) #define NDPREINIT(ndp) do { \ (ndp)->ni_dvp_seqc = SEQC_MOD; \ (ndp)->ni_vp_seqc = SEQC_MOD; \ } while (0) #define NDF_NO_DVP_RELE 0x00000001 #define NDF_NO_DVP_UNLOCK 0x00000002 #define NDF_NO_DVP_PUT 0x00000003 #define NDF_NO_VP_RELE 0x00000004 #define NDF_NO_VP_UNLOCK 0x00000008 #define NDF_NO_VP_PUT 0x0000000c #define NDF_NO_STARTDIR_RELE 0x00000010 #define NDF_NO_FREE_PNBUF 0x00000020 #define NDF_ONLY_PNBUF (~NDF_NO_FREE_PNBUF) void NDFREE_PNBUF(struct nameidata *); void NDFREE(struct nameidata *, const u_int); #define NDFREE(ndp, flags) do { \ struct nameidata *_ndp = (ndp); \ if (__builtin_constant_p(flags) && flags == NDF_ONLY_PNBUF) \ NDFREE_PNBUF(_ndp); \ else \ NDFREE(_ndp, flags); \ } while (0) #ifdef INVARIANTS void NDFREE_NOTHING(struct nameidata *); void NDVALIDATE(struct nameidata *); #else #define NDFREE_NOTHING(ndp) do { } while (0) #define NDVALIDATE(ndp) do { } while (0) #endif int namei(struct nameidata *ndp); int lookup(struct nameidata *ndp); int relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp); #endif /* * Stats on usefulness of namei caches. */ struct nchstats { long ncs_goodhits; /* hits that we can really use */ long ncs_neghits; /* negative hits that we can use */ long ncs_badhits; /* hits we must drop */ long ncs_falsehits; /* hits with id mismatch */ long ncs_miss; /* misses */ long ncs_long; /* long names that ignore cache */ long ncs_pass2; /* names found with passes == 2 */ long ncs_2passes; /* number of times we attempt it */ }; extern struct nchstats nchstats; #endif /* !_SYS_NAMEI_H_ */ diff --git a/sys/ufs/ffs/ffs_softdep.c b/sys/ufs/ffs/ffs_softdep.c index 6f1de407135e..ff55a918ade5 100644 --- a/sys/ufs/ffs/ffs_softdep.c +++ b/sys/ufs/ffs/ffs_softdep.c @@ -1,15259 +1,15258 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright 1998, 2000 Marshall Kirk McKusick. * Copyright 2009, 2010 Jeffrey W. Roberson * All rights reserved. * * The soft updates code is derived from the appendix of a University * of Michigan technical report (Gregory R. Ganger and Yale N. Patt, * "Soft Updates: A Solution to the Metadata Update Problem in File * Systems", CSE-TR-254-95, August 1995). * * Further information about soft updates can be obtained from: * * Marshall Kirk McKusick http://www.mckusick.com/softdep/ * 1614 Oxford Street mckusick@mckusick.com * Berkeley, CA 94709-1608 +1-510-843-9542 * USA * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR * TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE * USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. * * from: @(#)ffs_softdep.c 9.59 (McKusick) 6/21/00 */ #include __FBSDID("$FreeBSD$"); #include "opt_ffs.h" #include "opt_quota.h" #include "opt_ddb.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define KTR_SUJ 0 /* Define to KTR_SPARE. */ #ifndef SOFTUPDATES int softdep_flushfiles(oldmnt, flags, td) struct mount *oldmnt; int flags; struct thread *td; { panic("softdep_flushfiles called"); } int softdep_mount(devvp, mp, fs, cred) struct vnode *devvp; struct mount *mp; struct fs *fs; struct ucred *cred; { return (0); } void softdep_initialize() { return; } void softdep_uninitialize() { return; } void softdep_unmount(mp) struct mount *mp; { panic("softdep_unmount called"); } void softdep_setup_sbupdate(ump, fs, bp) struct ufsmount *ump; struct fs *fs; struct buf *bp; { panic("softdep_setup_sbupdate called"); } void softdep_setup_inomapdep(bp, ip, newinum, mode) struct buf *bp; struct inode *ip; ino_t newinum; int mode; { panic("softdep_setup_inomapdep called"); } void softdep_setup_blkmapdep(bp, mp, newblkno, frags, oldfrags) struct buf *bp; struct mount *mp; ufs2_daddr_t newblkno; int frags; int oldfrags; { panic("softdep_setup_blkmapdep called"); } void softdep_setup_allocdirect(ip, lbn, newblkno, oldblkno, newsize, oldsize, bp) struct inode *ip; ufs_lbn_t lbn; ufs2_daddr_t newblkno; ufs2_daddr_t oldblkno; long newsize; long oldsize; struct buf *bp; { panic("softdep_setup_allocdirect called"); } void softdep_setup_allocext(ip, lbn, newblkno, oldblkno, newsize, oldsize, bp) struct inode *ip; ufs_lbn_t lbn; ufs2_daddr_t newblkno; ufs2_daddr_t oldblkno; long newsize; long oldsize; struct buf *bp; { panic("softdep_setup_allocext called"); } void softdep_setup_allocindir_page(ip, lbn, bp, ptrno, newblkno, oldblkno, nbp) struct inode *ip; ufs_lbn_t lbn; struct buf *bp; int ptrno; ufs2_daddr_t newblkno; ufs2_daddr_t oldblkno; struct buf *nbp; { panic("softdep_setup_allocindir_page called"); } void softdep_setup_allocindir_meta(nbp, ip, bp, ptrno, newblkno) struct buf *nbp; struct inode *ip; struct buf *bp; int ptrno; ufs2_daddr_t newblkno; { panic("softdep_setup_allocindir_meta called"); } void softdep_journal_freeblocks(ip, cred, length, flags) struct inode *ip; struct ucred *cred; off_t length; int flags; { panic("softdep_journal_freeblocks called"); } void softdep_journal_fsync(ip) struct inode *ip; { panic("softdep_journal_fsync called"); } void softdep_setup_freeblocks(ip, length, flags) struct inode *ip; off_t length; int flags; { panic("softdep_setup_freeblocks called"); } void softdep_freefile(pvp, ino, mode) struct vnode *pvp; ino_t ino; int mode; { panic("softdep_freefile called"); } int softdep_setup_directory_add(bp, dp, diroffset, newinum, newdirbp, isnewblk) struct buf *bp; struct inode *dp; off_t diroffset; ino_t newinum; struct buf *newdirbp; int isnewblk; { panic("softdep_setup_directory_add called"); } void softdep_change_directoryentry_offset(bp, dp, base, oldloc, newloc, entrysize) struct buf *bp; struct inode *dp; caddr_t base; caddr_t oldloc; caddr_t newloc; int entrysize; { panic("softdep_change_directoryentry_offset called"); } void softdep_setup_remove(bp, dp, ip, isrmdir) struct buf *bp; struct inode *dp; struct inode *ip; int isrmdir; { panic("softdep_setup_remove called"); } void softdep_setup_directory_change(bp, dp, ip, newinum, isrmdir) struct buf *bp; struct inode *dp; struct inode *ip; ino_t newinum; int isrmdir; { panic("softdep_setup_directory_change called"); } void softdep_setup_blkfree(mp, bp, blkno, frags, wkhd) struct mount *mp; struct buf *bp; ufs2_daddr_t blkno; int frags; struct workhead *wkhd; { panic("%s called", __FUNCTION__); } void softdep_setup_inofree(mp, bp, ino, wkhd) struct mount *mp; struct buf *bp; ino_t ino; struct workhead *wkhd; { panic("%s called", __FUNCTION__); } void softdep_setup_unlink(dp, ip) struct inode *dp; struct inode *ip; { panic("%s called", __FUNCTION__); } void softdep_setup_link(dp, ip) struct inode *dp; struct inode *ip; { panic("%s called", __FUNCTION__); } void softdep_revert_link(dp, ip) struct inode *dp; struct inode *ip; { panic("%s called", __FUNCTION__); } void softdep_setup_rmdir(dp, ip) struct inode *dp; struct inode *ip; { panic("%s called", __FUNCTION__); } void softdep_revert_rmdir(dp, ip) struct inode *dp; struct inode *ip; { panic("%s called", __FUNCTION__); } void softdep_setup_create(dp, ip) struct inode *dp; struct inode *ip; { panic("%s called", __FUNCTION__); } void softdep_revert_create(dp, ip) struct inode *dp; struct inode *ip; { panic("%s called", __FUNCTION__); } void softdep_setup_mkdir(dp, ip) struct inode *dp; struct inode *ip; { panic("%s called", __FUNCTION__); } void softdep_revert_mkdir(dp, ip) struct inode *dp; struct inode *ip; { panic("%s called", __FUNCTION__); } void softdep_setup_dotdot_link(dp, ip) struct inode *dp; struct inode *ip; { panic("%s called", __FUNCTION__); } int softdep_prealloc(vp, waitok) struct vnode *vp; int waitok; { panic("%s called", __FUNCTION__); } int softdep_journal_lookup(mp, vpp) struct mount *mp; struct vnode **vpp; { return (ENOENT); } void softdep_change_linkcnt(ip) struct inode *ip; { panic("softdep_change_linkcnt called"); } void softdep_load_inodeblock(ip) struct inode *ip; { panic("softdep_load_inodeblock called"); } void softdep_update_inodeblock(ip, bp, waitfor) struct inode *ip; struct buf *bp; int waitfor; { panic("softdep_update_inodeblock called"); } int softdep_fsync(vp) struct vnode *vp; /* the "in_core" copy of the inode */ { return (0); } void softdep_fsync_mountdev(vp) struct vnode *vp; { return; } int softdep_flushworklist(oldmnt, countp, td) struct mount *oldmnt; int *countp; struct thread *td; { *countp = 0; return (0); } int softdep_sync_metadata(struct vnode *vp) { panic("softdep_sync_metadata called"); } int softdep_sync_buf(struct vnode *vp, struct buf *bp, int waitfor) { panic("softdep_sync_buf called"); } int softdep_slowdown(vp) struct vnode *vp; { panic("softdep_slowdown called"); } int softdep_request_cleanup(fs, vp, cred, resource) struct fs *fs; struct vnode *vp; struct ucred *cred; int resource; { return (0); } int softdep_check_suspend(struct mount *mp, struct vnode *devvp, int softdep_depcnt, int softdep_accdepcnt, int secondary_writes, int secondary_accwrites) { struct bufobj *bo; int error; (void) softdep_depcnt, (void) softdep_accdepcnt; bo = &devvp->v_bufobj; ASSERT_BO_WLOCKED(bo); MNT_ILOCK(mp); while (mp->mnt_secondary_writes != 0) { BO_UNLOCK(bo); msleep(&mp->mnt_secondary_writes, MNT_MTX(mp), (PUSER - 1) | PDROP, "secwr", 0); BO_LOCK(bo); MNT_ILOCK(mp); } /* * Reasons for needing more work before suspend: * - Dirty buffers on devvp. * - Secondary writes occurred after start of vnode sync loop */ error = 0; if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0 || secondary_writes != 0 || mp->mnt_secondary_writes != 0 || secondary_accwrites != mp->mnt_secondary_accwrites) error = EAGAIN; BO_UNLOCK(bo); return (error); } void softdep_get_depcounts(struct mount *mp, int *softdepactivep, int *softdepactiveaccp) { (void) mp; *softdepactivep = 0; *softdepactiveaccp = 0; } void softdep_buf_append(bp, wkhd) struct buf *bp; struct workhead *wkhd; { panic("softdep_buf_appendwork called"); } void softdep_inode_append(ip, cred, wkhd) struct inode *ip; struct ucred *cred; struct workhead *wkhd; { panic("softdep_inode_appendwork called"); } void softdep_freework(wkhd) struct workhead *wkhd; { panic("softdep_freework called"); } int softdep_prerename(fdvp, fvp, tdvp, tvp) struct vnode *fdvp; struct vnode *fvp; struct vnode *tdvp; struct vnode *tvp; { panic("softdep_prerename called"); } int softdep_prelink(dvp, vp, cnp) struct vnode *dvp; struct vnode *vp; struct componentname *cnp; { panic("softdep_prelink called"); } #else FEATURE(softupdates, "FFS soft-updates support"); static SYSCTL_NODE(_debug, OID_AUTO, softdep, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "soft updates stats"); static SYSCTL_NODE(_debug_softdep, OID_AUTO, total, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "total dependencies allocated"); static SYSCTL_NODE(_debug_softdep, OID_AUTO, highuse, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "high use dependencies allocated"); static SYSCTL_NODE(_debug_softdep, OID_AUTO, current, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "current dependencies allocated"); static SYSCTL_NODE(_debug_softdep, OID_AUTO, write, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "current dependencies written"); unsigned long dep_current[D_LAST + 1]; unsigned long dep_highuse[D_LAST + 1]; unsigned long dep_total[D_LAST + 1]; unsigned long dep_write[D_LAST + 1]; #define SOFTDEP_TYPE(type, str, long) \ static MALLOC_DEFINE(M_ ## type, #str, long); \ SYSCTL_ULONG(_debug_softdep_total, OID_AUTO, str, CTLFLAG_RD, \ &dep_total[D_ ## type], 0, ""); \ SYSCTL_ULONG(_debug_softdep_current, OID_AUTO, str, CTLFLAG_RD, \ &dep_current[D_ ## type], 0, ""); \ SYSCTL_ULONG(_debug_softdep_highuse, OID_AUTO, str, CTLFLAG_RD, \ &dep_highuse[D_ ## type], 0, ""); \ SYSCTL_ULONG(_debug_softdep_write, OID_AUTO, str, CTLFLAG_RD, \ &dep_write[D_ ## type], 0, ""); SOFTDEP_TYPE(PAGEDEP, pagedep, "File page dependencies"); SOFTDEP_TYPE(INODEDEP, inodedep, "Inode dependencies"); SOFTDEP_TYPE(BMSAFEMAP, bmsafemap, "Block or frag allocated from cyl group map"); SOFTDEP_TYPE(NEWBLK, newblk, "New block or frag allocation dependency"); SOFTDEP_TYPE(ALLOCDIRECT, allocdirect, "Block or frag dependency for an inode"); SOFTDEP_TYPE(INDIRDEP, indirdep, "Indirect block dependencies"); SOFTDEP_TYPE(ALLOCINDIR, allocindir, "Block dependency for an indirect block"); SOFTDEP_TYPE(FREEFRAG, freefrag, "Previously used frag for an inode"); SOFTDEP_TYPE(FREEBLKS, freeblks, "Blocks freed from an inode"); SOFTDEP_TYPE(FREEFILE, freefile, "Inode deallocated"); SOFTDEP_TYPE(DIRADD, diradd, "New directory entry"); SOFTDEP_TYPE(MKDIR, mkdir, "New directory"); SOFTDEP_TYPE(DIRREM, dirrem, "Directory entry deleted"); SOFTDEP_TYPE(NEWDIRBLK, newdirblk, "Unclaimed new directory block"); SOFTDEP_TYPE(FREEWORK, freework, "free an inode block"); SOFTDEP_TYPE(FREEDEP, freedep, "track a block free"); SOFTDEP_TYPE(JADDREF, jaddref, "Journal inode ref add"); SOFTDEP_TYPE(JREMREF, jremref, "Journal inode ref remove"); SOFTDEP_TYPE(JMVREF, jmvref, "Journal inode ref move"); SOFTDEP_TYPE(JNEWBLK, jnewblk, "Journal new block"); SOFTDEP_TYPE(JFREEBLK, jfreeblk, "Journal free block"); SOFTDEP_TYPE(JFREEFRAG, jfreefrag, "Journal free frag"); SOFTDEP_TYPE(JSEG, jseg, "Journal segment"); SOFTDEP_TYPE(JSEGDEP, jsegdep, "Journal segment complete"); SOFTDEP_TYPE(SBDEP, sbdep, "Superblock write dependency"); SOFTDEP_TYPE(JTRUNC, jtrunc, "Journal inode truncation"); SOFTDEP_TYPE(JFSYNC, jfsync, "Journal fsync complete"); static MALLOC_DEFINE(M_SENTINEL, "sentinel", "Worklist sentinel"); static MALLOC_DEFINE(M_SAVEDINO, "savedino", "Saved inodes"); static MALLOC_DEFINE(M_JBLOCKS, "jblocks", "Journal block locations"); static MALLOC_DEFINE(M_MOUNTDATA, "softdep", "Softdep per-mount data"); #define M_SOFTDEP_FLAGS (M_WAITOK) /* * translate from workitem type to memory type * MUST match the defines above, such that memtype[D_XXX] == M_XXX */ static struct malloc_type *memtype[] = { NULL, M_PAGEDEP, M_INODEDEP, M_BMSAFEMAP, M_NEWBLK, M_ALLOCDIRECT, M_INDIRDEP, M_ALLOCINDIR, M_FREEFRAG, M_FREEBLKS, M_FREEFILE, M_DIRADD, M_MKDIR, M_DIRREM, M_NEWDIRBLK, M_FREEWORK, M_FREEDEP, M_JADDREF, M_JREMREF, M_JMVREF, M_JNEWBLK, M_JFREEBLK, M_JFREEFRAG, M_JSEG, M_JSEGDEP, M_SBDEP, M_JTRUNC, M_JFSYNC, M_SENTINEL }; #define DtoM(type) (memtype[type]) /* * Names of malloc types. */ #define TYPENAME(type) \ ((unsigned)(type) <= D_LAST && (unsigned)(type) >= D_FIRST ? \ memtype[type]->ks_shortdesc : "???") /* * End system adaptation definitions. */ #define DOTDOT_OFFSET offsetof(struct dirtemplate, dotdot_ino) #define DOT_OFFSET offsetof(struct dirtemplate, dot_ino) /* * Internal function prototypes. */ static void check_clear_deps(struct mount *); static void softdep_error(char *, int); static int softdep_prerename_vnode(struct ufsmount *, struct vnode *); static int softdep_process_worklist(struct mount *, int); static int softdep_waitidle(struct mount *, int); static void drain_output(struct vnode *); static struct buf *getdirtybuf(struct buf *, struct rwlock *, int); static int check_inodedep_free(struct inodedep *); static void clear_remove(struct mount *); static void clear_inodedeps(struct mount *); static void unlinked_inodedep(struct mount *, struct inodedep *); static void clear_unlinked_inodedep(struct inodedep *); static struct inodedep *first_unlinked_inodedep(struct ufsmount *); static int flush_pagedep_deps(struct vnode *, struct mount *, struct diraddhd *, struct buf *); static int free_pagedep(struct pagedep *); static int flush_newblk_dep(struct vnode *, struct mount *, ufs_lbn_t); static int flush_inodedep_deps(struct vnode *, struct mount *, ino_t); static int flush_deplist(struct allocdirectlst *, int, int *); static int sync_cgs(struct mount *, int); static int handle_written_filepage(struct pagedep *, struct buf *, int); static int handle_written_sbdep(struct sbdep *, struct buf *); static void initiate_write_sbdep(struct sbdep *); static void diradd_inode_written(struct diradd *, struct inodedep *); static int handle_written_indirdep(struct indirdep *, struct buf *, struct buf**, int); static int handle_written_inodeblock(struct inodedep *, struct buf *, int); static int jnewblk_rollforward(struct jnewblk *, struct fs *, struct cg *, uint8_t *); static int handle_written_bmsafemap(struct bmsafemap *, struct buf *, int); static void handle_written_jaddref(struct jaddref *); static void handle_written_jremref(struct jremref *); static void handle_written_jseg(struct jseg *, struct buf *); static void handle_written_jnewblk(struct jnewblk *); static void handle_written_jblkdep(struct jblkdep *); static void handle_written_jfreefrag(struct jfreefrag *); static void complete_jseg(struct jseg *); static void complete_jsegs(struct jseg *); static void jseg_write(struct ufsmount *ump, struct jseg *, uint8_t *); static void jaddref_write(struct jaddref *, struct jseg *, uint8_t *); static void jremref_write(struct jremref *, struct jseg *, uint8_t *); static void jmvref_write(struct jmvref *, struct jseg *, uint8_t *); static void jtrunc_write(struct jtrunc *, struct jseg *, uint8_t *); static void jfsync_write(struct jfsync *, struct jseg *, uint8_t *data); static void jnewblk_write(struct jnewblk *, struct jseg *, uint8_t *); static void jfreeblk_write(struct jfreeblk *, struct jseg *, uint8_t *); static void jfreefrag_write(struct jfreefrag *, struct jseg *, uint8_t *); static inline void inoref_write(struct inoref *, struct jseg *, struct jrefrec *); static void handle_allocdirect_partdone(struct allocdirect *, struct workhead *); static struct jnewblk *cancel_newblk(struct newblk *, struct worklist *, struct workhead *); static void indirdep_complete(struct indirdep *); static int indirblk_lookup(struct mount *, ufs2_daddr_t); static void indirblk_insert(struct freework *); static void indirblk_remove(struct freework *); static void handle_allocindir_partdone(struct allocindir *); static void initiate_write_filepage(struct pagedep *, struct buf *); static void initiate_write_indirdep(struct indirdep*, struct buf *); static void handle_written_mkdir(struct mkdir *, int); static int jnewblk_rollback(struct jnewblk *, struct fs *, struct cg *, uint8_t *); static void initiate_write_bmsafemap(struct bmsafemap *, struct buf *); static void initiate_write_inodeblock_ufs1(struct inodedep *, struct buf *); static void initiate_write_inodeblock_ufs2(struct inodedep *, struct buf *); static void handle_workitem_freefile(struct freefile *); static int handle_workitem_remove(struct dirrem *, int); static struct dirrem *newdirrem(struct buf *, struct inode *, struct inode *, int, struct dirrem **); static struct indirdep *indirdep_lookup(struct mount *, struct inode *, struct buf *); static void cancel_indirdep(struct indirdep *, struct buf *, struct freeblks *); static void free_indirdep(struct indirdep *); static void free_diradd(struct diradd *, struct workhead *); static void merge_diradd(struct inodedep *, struct diradd *); static void complete_diradd(struct diradd *); static struct diradd *diradd_lookup(struct pagedep *, int); static struct jremref *cancel_diradd_dotdot(struct inode *, struct dirrem *, struct jremref *); static struct jremref *cancel_mkdir_dotdot(struct inode *, struct dirrem *, struct jremref *); static void cancel_diradd(struct diradd *, struct dirrem *, struct jremref *, struct jremref *, struct jremref *); static void dirrem_journal(struct dirrem *, struct jremref *, struct jremref *, struct jremref *); static void cancel_allocindir(struct allocindir *, struct buf *bp, struct freeblks *, int); static int setup_trunc_indir(struct freeblks *, struct inode *, ufs_lbn_t, ufs_lbn_t, ufs2_daddr_t); static void complete_trunc_indir(struct freework *); static void trunc_indirdep(struct indirdep *, struct freeblks *, struct buf *, int); static void complete_mkdir(struct mkdir *); static void free_newdirblk(struct newdirblk *); static void free_jremref(struct jremref *); static void free_jaddref(struct jaddref *); static void free_jsegdep(struct jsegdep *); static void free_jsegs(struct jblocks *); static void rele_jseg(struct jseg *); static void free_jseg(struct jseg *, struct jblocks *); static void free_jnewblk(struct jnewblk *); static void free_jblkdep(struct jblkdep *); static void free_jfreefrag(struct jfreefrag *); static void free_freedep(struct freedep *); static void journal_jremref(struct dirrem *, struct jremref *, struct inodedep *); static void cancel_jnewblk(struct jnewblk *, struct workhead *); static int cancel_jaddref(struct jaddref *, struct inodedep *, struct workhead *); static void cancel_jfreefrag(struct jfreefrag *); static inline void setup_freedirect(struct freeblks *, struct inode *, int, int); static inline void setup_freeext(struct freeblks *, struct inode *, int, int); static inline void setup_freeindir(struct freeblks *, struct inode *, int, ufs_lbn_t, int); static inline struct freeblks *newfreeblks(struct mount *, struct inode *); static void freeblks_free(struct ufsmount *, struct freeblks *, int); static void indir_trunc(struct freework *, ufs2_daddr_t, ufs_lbn_t); static ufs2_daddr_t blkcount(struct fs *, ufs2_daddr_t, off_t); static int trunc_check_buf(struct buf *, int *, ufs_lbn_t, int, int); static void trunc_dependencies(struct inode *, struct freeblks *, ufs_lbn_t, int, int); static void trunc_pages(struct inode *, off_t, ufs2_daddr_t, int); static int cancel_pagedep(struct pagedep *, struct freeblks *, int); static int deallocate_dependencies(struct buf *, struct freeblks *, int); static void newblk_freefrag(struct newblk*); static void free_newblk(struct newblk *); static void cancel_allocdirect(struct allocdirectlst *, struct allocdirect *, struct freeblks *); static int check_inode_unwritten(struct inodedep *); static int free_inodedep(struct inodedep *); static void freework_freeblock(struct freework *, u_long); static void freework_enqueue(struct freework *); static int handle_workitem_freeblocks(struct freeblks *, int); static int handle_complete_freeblocks(struct freeblks *, int); static void handle_workitem_indirblk(struct freework *); static void handle_written_freework(struct freework *); static void merge_inode_lists(struct allocdirectlst *,struct allocdirectlst *); static struct worklist *jnewblk_merge(struct worklist *, struct worklist *, struct workhead *); static struct freefrag *setup_allocindir_phase2(struct buf *, struct inode *, struct inodedep *, struct allocindir *, ufs_lbn_t); static struct allocindir *newallocindir(struct inode *, int, ufs2_daddr_t, ufs2_daddr_t, ufs_lbn_t); static void handle_workitem_freefrag(struct freefrag *); static struct freefrag *newfreefrag(struct inode *, ufs2_daddr_t, long, ufs_lbn_t, u_long); static void allocdirect_merge(struct allocdirectlst *, struct allocdirect *, struct allocdirect *); static struct freefrag *allocindir_merge(struct allocindir *, struct allocindir *); static int bmsafemap_find(struct bmsafemap_hashhead *, int, struct bmsafemap **); static struct bmsafemap *bmsafemap_lookup(struct mount *, struct buf *, int cg, struct bmsafemap *); static int newblk_find(struct newblk_hashhead *, ufs2_daddr_t, int, struct newblk **); static int newblk_lookup(struct mount *, ufs2_daddr_t, int, struct newblk **); static int inodedep_find(struct inodedep_hashhead *, ino_t, struct inodedep **); static int inodedep_lookup(struct mount *, ino_t, int, struct inodedep **); static int pagedep_lookup(struct mount *, struct buf *bp, ino_t, ufs_lbn_t, int, struct pagedep **); static int pagedep_find(struct pagedep_hashhead *, ino_t, ufs_lbn_t, struct pagedep **); static void pause_timer(void *); static int request_cleanup(struct mount *, int); static int softdep_request_cleanup_flush(struct mount *, struct ufsmount *); static void schedule_cleanup(struct mount *); static void softdep_ast_cleanup_proc(struct thread *); static struct ufsmount *softdep_bp_to_mp(struct buf *bp); static int process_worklist_item(struct mount *, int, int); static void process_removes(struct vnode *); static void process_truncates(struct vnode *); static void jwork_move(struct workhead *, struct workhead *); static void jwork_insert(struct workhead *, struct jsegdep *); static void add_to_worklist(struct worklist *, int); static void wake_worklist(struct worklist *); static void wait_worklist(struct worklist *, char *); static void remove_from_worklist(struct worklist *); static void softdep_flush(void *); static void softdep_flushjournal(struct mount *); static int softdep_speedup(struct ufsmount *); static void worklist_speedup(struct mount *); static int journal_mount(struct mount *, struct fs *, struct ucred *); static void journal_unmount(struct ufsmount *); static int journal_space(struct ufsmount *, int); static void journal_suspend(struct ufsmount *); static int journal_unsuspend(struct ufsmount *ump); static void add_to_journal(struct worklist *); static void remove_from_journal(struct worklist *); static bool softdep_excess_items(struct ufsmount *, int); static void softdep_process_journal(struct mount *, struct worklist *, int); static struct jremref *newjremref(struct dirrem *, struct inode *, struct inode *ip, off_t, nlink_t); static struct jaddref *newjaddref(struct inode *, ino_t, off_t, int16_t, uint16_t); static inline void newinoref(struct inoref *, ino_t, ino_t, off_t, nlink_t, uint16_t); static inline struct jsegdep *inoref_jseg(struct inoref *); static struct jmvref *newjmvref(struct inode *, ino_t, off_t, off_t); static struct jfreeblk *newjfreeblk(struct freeblks *, ufs_lbn_t, ufs2_daddr_t, int); static void adjust_newfreework(struct freeblks *, int); static struct jtrunc *newjtrunc(struct freeblks *, off_t, int); static void move_newblock_dep(struct jaddref *, struct inodedep *); static void cancel_jfreeblk(struct freeblks *, ufs2_daddr_t); static struct jfreefrag *newjfreefrag(struct freefrag *, struct inode *, ufs2_daddr_t, long, ufs_lbn_t); static struct freework *newfreework(struct ufsmount *, struct freeblks *, struct freework *, ufs_lbn_t, ufs2_daddr_t, int, int, int); static int jwait(struct worklist *, int); static struct inodedep *inodedep_lookup_ip(struct inode *); static int bmsafemap_backgroundwrite(struct bmsafemap *, struct buf *); static struct freefile *handle_bufwait(struct inodedep *, struct workhead *); static void handle_jwork(struct workhead *); static struct mkdir *setup_newdir(struct diradd *, ino_t, ino_t, struct buf *, struct mkdir **); static struct jblocks *jblocks_create(void); static ufs2_daddr_t jblocks_alloc(struct jblocks *, int, int *); static void jblocks_free(struct jblocks *, struct mount *, int); static void jblocks_destroy(struct jblocks *); static void jblocks_add(struct jblocks *, ufs2_daddr_t, int); /* * Exported softdep operations. */ static void softdep_disk_io_initiation(struct buf *); static void softdep_disk_write_complete(struct buf *); static void softdep_deallocate_dependencies(struct buf *); static int softdep_count_dependencies(struct buf *bp, int); /* * Global lock over all of soft updates. */ static struct mtx lk; MTX_SYSINIT(softdep_lock, &lk, "global softdep", MTX_DEF); #define ACQUIRE_GBLLOCK(lk) mtx_lock(lk) #define FREE_GBLLOCK(lk) mtx_unlock(lk) #define GBLLOCK_OWNED(lk) mtx_assert((lk), MA_OWNED) /* * Per-filesystem soft-updates locking. */ #define LOCK_PTR(ump) (&(ump)->um_softdep->sd_fslock) #define TRY_ACQUIRE_LOCK(ump) rw_try_wlock(&(ump)->um_softdep->sd_fslock) #define ACQUIRE_LOCK(ump) rw_wlock(&(ump)->um_softdep->sd_fslock) #define FREE_LOCK(ump) rw_wunlock(&(ump)->um_softdep->sd_fslock) #define LOCK_OWNED(ump) rw_assert(&(ump)->um_softdep->sd_fslock, \ RA_WLOCKED) #define BUF_AREC(bp) lockallowrecurse(&(bp)->b_lock) #define BUF_NOREC(bp) lockdisablerecurse(&(bp)->b_lock) /* * Worklist queue management. * These routines require that the lock be held. */ #ifndef /* NOT */ INVARIANTS #define WORKLIST_INSERT(head, item) do { \ (item)->wk_state |= ONWORKLIST; \ LIST_INSERT_HEAD(head, item, wk_list); \ } while (0) #define WORKLIST_REMOVE(item) do { \ (item)->wk_state &= ~ONWORKLIST; \ LIST_REMOVE(item, wk_list); \ } while (0) #define WORKLIST_INSERT_UNLOCKED WORKLIST_INSERT #define WORKLIST_REMOVE_UNLOCKED WORKLIST_REMOVE #else /* INVARIANTS */ static void worklist_insert(struct workhead *, struct worklist *, int, const char *, int); static void worklist_remove(struct worklist *, int, const char *, int); #define WORKLIST_INSERT(head, item) \ worklist_insert(head, item, 1, __func__, __LINE__) #define WORKLIST_INSERT_UNLOCKED(head, item)\ worklist_insert(head, item, 0, __func__, __LINE__) #define WORKLIST_REMOVE(item)\ worklist_remove(item, 1, __func__, __LINE__) #define WORKLIST_REMOVE_UNLOCKED(item)\ worklist_remove(item, 0, __func__, __LINE__) static void worklist_insert(head, item, locked, func, line) struct workhead *head; struct worklist *item; int locked; const char *func; int line; { if (locked) LOCK_OWNED(VFSTOUFS(item->wk_mp)); if (item->wk_state & ONWORKLIST) panic("worklist_insert: %p %s(0x%X) already on list, " "added in function %s at line %d", item, TYPENAME(item->wk_type), item->wk_state, item->wk_func, item->wk_line); item->wk_state |= ONWORKLIST; item->wk_func = func; item->wk_line = line; LIST_INSERT_HEAD(head, item, wk_list); } static void worklist_remove(item, locked, func, line) struct worklist *item; int locked; const char *func; int line; { if (locked) LOCK_OWNED(VFSTOUFS(item->wk_mp)); if ((item->wk_state & ONWORKLIST) == 0) panic("worklist_remove: %p %s(0x%X) not on list, " "removed in function %s at line %d", item, TYPENAME(item->wk_type), item->wk_state, item->wk_func, item->wk_line); item->wk_state &= ~ONWORKLIST; item->wk_func = func; item->wk_line = line; LIST_REMOVE(item, wk_list); } #endif /* INVARIANTS */ /* * Merge two jsegdeps keeping only the oldest one as newer references * can't be discarded until after older references. */ static inline struct jsegdep * jsegdep_merge(struct jsegdep *one, struct jsegdep *two) { struct jsegdep *swp; if (two == NULL) return (one); if (one->jd_seg->js_seq > two->jd_seg->js_seq) { swp = one; one = two; two = swp; } WORKLIST_REMOVE(&two->jd_list); free_jsegdep(two); return (one); } /* * If two freedeps are compatible free one to reduce list size. */ static inline struct freedep * freedep_merge(struct freedep *one, struct freedep *two) { if (two == NULL) return (one); if (one->fd_freework == two->fd_freework) { WORKLIST_REMOVE(&two->fd_list); free_freedep(two); } return (one); } /* * Move journal work from one list to another. Duplicate freedeps and * jsegdeps are coalesced to keep the lists as small as possible. */ static void jwork_move(dst, src) struct workhead *dst; struct workhead *src; { struct freedep *freedep; struct jsegdep *jsegdep; struct worklist *wkn; struct worklist *wk; KASSERT(dst != src, ("jwork_move: dst == src")); freedep = NULL; jsegdep = NULL; LIST_FOREACH_SAFE(wk, dst, wk_list, wkn) { if (wk->wk_type == D_JSEGDEP) jsegdep = jsegdep_merge(WK_JSEGDEP(wk), jsegdep); else if (wk->wk_type == D_FREEDEP) freedep = freedep_merge(WK_FREEDEP(wk), freedep); } while ((wk = LIST_FIRST(src)) != NULL) { WORKLIST_REMOVE(wk); WORKLIST_INSERT(dst, wk); if (wk->wk_type == D_JSEGDEP) { jsegdep = jsegdep_merge(WK_JSEGDEP(wk), jsegdep); continue; } if (wk->wk_type == D_FREEDEP) freedep = freedep_merge(WK_FREEDEP(wk), freedep); } } static void jwork_insert(dst, jsegdep) struct workhead *dst; struct jsegdep *jsegdep; { struct jsegdep *jsegdepn; struct worklist *wk; LIST_FOREACH(wk, dst, wk_list) if (wk->wk_type == D_JSEGDEP) break; if (wk == NULL) { WORKLIST_INSERT(dst, &jsegdep->jd_list); return; } jsegdepn = WK_JSEGDEP(wk); if (jsegdep->jd_seg->js_seq < jsegdepn->jd_seg->js_seq) { WORKLIST_REMOVE(wk); free_jsegdep(jsegdepn); WORKLIST_INSERT(dst, &jsegdep->jd_list); } else free_jsegdep(jsegdep); } /* * Routines for tracking and managing workitems. */ static void workitem_free(struct worklist *, int); static void workitem_alloc(struct worklist *, int, struct mount *); static void workitem_reassign(struct worklist *, int); #define WORKITEM_FREE(item, type) \ workitem_free((struct worklist *)(item), (type)) #define WORKITEM_REASSIGN(item, type) \ workitem_reassign((struct worklist *)(item), (type)) static void workitem_free(item, type) struct worklist *item; int type; { struct ufsmount *ump; #ifdef INVARIANTS if (item->wk_state & ONWORKLIST) panic("workitem_free: %s(0x%X) still on list, " "added in function %s at line %d", TYPENAME(item->wk_type), item->wk_state, item->wk_func, item->wk_line); if (item->wk_type != type && type != D_NEWBLK) panic("workitem_free: type mismatch %s != %s", TYPENAME(item->wk_type), TYPENAME(type)); #endif if (item->wk_state & IOWAITING) wakeup(item); ump = VFSTOUFS(item->wk_mp); LOCK_OWNED(ump); KASSERT(ump->softdep_deps > 0, ("workitem_free: %s: softdep_deps going negative", ump->um_fs->fs_fsmnt)); if (--ump->softdep_deps == 0 && ump->softdep_req) wakeup(&ump->softdep_deps); KASSERT(dep_current[item->wk_type] > 0, ("workitem_free: %s: dep_current[%s] going negative", ump->um_fs->fs_fsmnt, TYPENAME(item->wk_type))); KASSERT(ump->softdep_curdeps[item->wk_type] > 0, ("workitem_free: %s: softdep_curdeps[%s] going negative", ump->um_fs->fs_fsmnt, TYPENAME(item->wk_type))); atomic_subtract_long(&dep_current[item->wk_type], 1); ump->softdep_curdeps[item->wk_type] -= 1; LIST_REMOVE(item, wk_all); free(item, DtoM(type)); } static void workitem_alloc(item, type, mp) struct worklist *item; int type; struct mount *mp; { struct ufsmount *ump; item->wk_type = type; item->wk_mp = mp; item->wk_state = 0; ump = VFSTOUFS(mp); ACQUIRE_GBLLOCK(&lk); dep_current[type]++; if (dep_current[type] > dep_highuse[type]) dep_highuse[type] = dep_current[type]; dep_total[type]++; FREE_GBLLOCK(&lk); ACQUIRE_LOCK(ump); ump->softdep_curdeps[type] += 1; ump->softdep_deps++; ump->softdep_accdeps++; LIST_INSERT_HEAD(&ump->softdep_alldeps[type], item, wk_all); FREE_LOCK(ump); } static void workitem_reassign(item, newtype) struct worklist *item; int newtype; { struct ufsmount *ump; ump = VFSTOUFS(item->wk_mp); LOCK_OWNED(ump); KASSERT(ump->softdep_curdeps[item->wk_type] > 0, ("workitem_reassign: %s: softdep_curdeps[%s] going negative", VFSTOUFS(item->wk_mp)->um_fs->fs_fsmnt, TYPENAME(item->wk_type))); ump->softdep_curdeps[item->wk_type] -= 1; ump->softdep_curdeps[newtype] += 1; KASSERT(dep_current[item->wk_type] > 0, ("workitem_reassign: %s: dep_current[%s] going negative", VFSTOUFS(item->wk_mp)->um_fs->fs_fsmnt, TYPENAME(item->wk_type))); ACQUIRE_GBLLOCK(&lk); dep_current[newtype]++; dep_current[item->wk_type]--; if (dep_current[newtype] > dep_highuse[newtype]) dep_highuse[newtype] = dep_current[newtype]; dep_total[newtype]++; FREE_GBLLOCK(&lk); item->wk_type = newtype; LIST_REMOVE(item, wk_all); LIST_INSERT_HEAD(&ump->softdep_alldeps[newtype], item, wk_all); } /* * Workitem queue management */ static int max_softdeps; /* maximum number of structs before slowdown */ static int tickdelay = 2; /* number of ticks to pause during slowdown */ static int proc_waiting; /* tracks whether we have a timeout posted */ static int *stat_countp; /* statistic to count in proc_waiting timeout */ static struct callout softdep_callout; static int req_clear_inodedeps; /* syncer process flush some inodedeps */ static int req_clear_remove; /* syncer process flush some freeblks */ static int softdep_flushcache = 0; /* Should we do BIO_FLUSH? */ /* * runtime statistics */ static int stat_flush_threads; /* number of softdep flushing threads */ static int stat_worklist_push; /* number of worklist cleanups */ static int stat_delayed_inact; /* number of delayed inactivation cleanups */ static int stat_blk_limit_push; /* number of times block limit neared */ static int stat_ino_limit_push; /* number of times inode limit neared */ static int stat_blk_limit_hit; /* number of times block slowdown imposed */ static int stat_ino_limit_hit; /* number of times inode slowdown imposed */ static int stat_sync_limit_hit; /* number of synchronous slowdowns imposed */ static int stat_indir_blk_ptrs; /* bufs redirtied as indir ptrs not written */ static int stat_inode_bitmap; /* bufs redirtied as inode bitmap not written */ static int stat_direct_blk_ptrs;/* bufs redirtied as direct ptrs not written */ static int stat_dir_entry; /* bufs redirtied as dir entry cannot write */ static int stat_jaddref; /* bufs redirtied as ino bitmap can not write */ static int stat_jnewblk; /* bufs redirtied as blk bitmap can not write */ static int stat_journal_min; /* Times hit journal min threshold */ static int stat_journal_low; /* Times hit journal low threshold */ static int stat_journal_wait; /* Times blocked in jwait(). */ static int stat_jwait_filepage; /* Times blocked in jwait() for filepage. */ static int stat_jwait_freeblks; /* Times blocked in jwait() for freeblks. */ static int stat_jwait_inode; /* Times blocked in jwait() for inodes. */ static int stat_jwait_newblk; /* Times blocked in jwait() for newblks. */ static int stat_cleanup_high_delay; /* Maximum cleanup delay (in ticks) */ static int stat_cleanup_blkrequests; /* Number of block cleanup requests */ static int stat_cleanup_inorequests; /* Number of inode cleanup requests */ static int stat_cleanup_retries; /* Number of cleanups that needed to flush */ static int stat_cleanup_failures; /* Number of cleanup requests that failed */ static int stat_emptyjblocks; /* Number of potentially empty journal blocks */ SYSCTL_INT(_debug_softdep, OID_AUTO, max_softdeps, CTLFLAG_RW, &max_softdeps, 0, ""); SYSCTL_INT(_debug_softdep, OID_AUTO, tickdelay, CTLFLAG_RW, &tickdelay, 0, ""); SYSCTL_INT(_debug_softdep, OID_AUTO, flush_threads, CTLFLAG_RD, &stat_flush_threads, 0, ""); SYSCTL_INT(_debug_softdep, OID_AUTO, worklist_push, CTLFLAG_RW | CTLFLAG_STATS, &stat_worklist_push, 0,""); SYSCTL_INT(_debug_softdep, OID_AUTO, delayed_inactivations, CTLFLAG_RD, &stat_delayed_inact, 0, ""); SYSCTL_INT(_debug_softdep, OID_AUTO, blk_limit_push, CTLFLAG_RW | CTLFLAG_STATS, &stat_blk_limit_push, 0,""); SYSCTL_INT(_debug_softdep, OID_AUTO, ino_limit_push, CTLFLAG_RW | CTLFLAG_STATS, &stat_ino_limit_push, 0,""); SYSCTL_INT(_debug_softdep, OID_AUTO, blk_limit_hit, CTLFLAG_RW | CTLFLAG_STATS, &stat_blk_limit_hit, 0, ""); SYSCTL_INT(_debug_softdep, OID_AUTO, ino_limit_hit, CTLFLAG_RW | CTLFLAG_STATS, &stat_ino_limit_hit, 0, ""); SYSCTL_INT(_debug_softdep, OID_AUTO, sync_limit_hit, CTLFLAG_RW | CTLFLAG_STATS, &stat_sync_limit_hit, 0, ""); SYSCTL_INT(_debug_softdep, OID_AUTO, indir_blk_ptrs, CTLFLAG_RW | CTLFLAG_STATS, &stat_indir_blk_ptrs, 0, ""); SYSCTL_INT(_debug_softdep, OID_AUTO, inode_bitmap, CTLFLAG_RW | CTLFLAG_STATS, &stat_inode_bitmap, 0, ""); SYSCTL_INT(_debug_softdep, OID_AUTO, direct_blk_ptrs, CTLFLAG_RW | CTLFLAG_STATS, &stat_direct_blk_ptrs, 0, ""); SYSCTL_INT(_debug_softdep, OID_AUTO, dir_entry, CTLFLAG_RW | CTLFLAG_STATS, &stat_dir_entry, 0, ""); SYSCTL_INT(_debug_softdep, OID_AUTO, jaddref_rollback, CTLFLAG_RW | CTLFLAG_STATS, &stat_jaddref, 0, ""); SYSCTL_INT(_debug_softdep, OID_AUTO, jnewblk_rollback, CTLFLAG_RW | CTLFLAG_STATS, &stat_jnewblk, 0, ""); SYSCTL_INT(_debug_softdep, OID_AUTO, journal_low, CTLFLAG_RW | CTLFLAG_STATS, &stat_journal_low, 0, ""); SYSCTL_INT(_debug_softdep, OID_AUTO, journal_min, CTLFLAG_RW | CTLFLAG_STATS, &stat_journal_min, 0, ""); SYSCTL_INT(_debug_softdep, OID_AUTO, journal_wait, CTLFLAG_RW | CTLFLAG_STATS, &stat_journal_wait, 0, ""); SYSCTL_INT(_debug_softdep, OID_AUTO, jwait_filepage, CTLFLAG_RW | CTLFLAG_STATS, &stat_jwait_filepage, 0, ""); SYSCTL_INT(_debug_softdep, OID_AUTO, jwait_freeblks, CTLFLAG_RW | CTLFLAG_STATS, &stat_jwait_freeblks, 0, ""); SYSCTL_INT(_debug_softdep, OID_AUTO, jwait_inode, CTLFLAG_RW | CTLFLAG_STATS, &stat_jwait_inode, 0, ""); SYSCTL_INT(_debug_softdep, OID_AUTO, jwait_newblk, CTLFLAG_RW | CTLFLAG_STATS, &stat_jwait_newblk, 0, ""); SYSCTL_INT(_debug_softdep, OID_AUTO, cleanup_blkrequests, CTLFLAG_RW | CTLFLAG_STATS, &stat_cleanup_blkrequests, 0, ""); SYSCTL_INT(_debug_softdep, OID_AUTO, cleanup_inorequests, CTLFLAG_RW | CTLFLAG_STATS, &stat_cleanup_inorequests, 0, ""); SYSCTL_INT(_debug_softdep, OID_AUTO, cleanup_high_delay, CTLFLAG_RW | CTLFLAG_STATS, &stat_cleanup_high_delay, 0, ""); SYSCTL_INT(_debug_softdep, OID_AUTO, cleanup_retries, CTLFLAG_RW | CTLFLAG_STATS, &stat_cleanup_retries, 0, ""); SYSCTL_INT(_debug_softdep, OID_AUTO, cleanup_failures, CTLFLAG_RW | CTLFLAG_STATS, &stat_cleanup_failures, 0, ""); SYSCTL_INT(_debug_softdep, OID_AUTO, flushcache, CTLFLAG_RW, &softdep_flushcache, 0, ""); SYSCTL_INT(_debug_softdep, OID_AUTO, emptyjblocks, CTLFLAG_RD, &stat_emptyjblocks, 0, ""); SYSCTL_DECL(_vfs_ffs); /* Whether to recompute the summary at mount time */ static int compute_summary_at_mount = 0; SYSCTL_INT(_vfs_ffs, OID_AUTO, compute_summary_at_mount, CTLFLAG_RW, &compute_summary_at_mount, 0, "Recompute summary at mount"); static int print_threads = 0; SYSCTL_INT(_debug_softdep, OID_AUTO, print_threads, CTLFLAG_RW, &print_threads, 0, "Notify flusher thread start/stop"); /* List of all filesystems mounted with soft updates */ static TAILQ_HEAD(, mount_softdeps) softdepmounts; static void get_parent_vp_unlock_bp(struct mount *mp, struct buf *bp, struct diraddhd *diraddhdp, struct diraddhd *unfinishedp) { struct diradd *dap; /* * Requeue unfinished dependencies before * unlocking buffer, which could make * diraddhdp invalid. */ ACQUIRE_LOCK(VFSTOUFS(mp)); while ((dap = LIST_FIRST(unfinishedp)) != NULL) { LIST_REMOVE(dap, da_pdlist); LIST_INSERT_HEAD(diraddhdp, dap, da_pdlist); } FREE_LOCK(VFSTOUFS(mp)); bp->b_vflags &= ~BV_SCANNED; BUF_NOREC(bp); BUF_UNLOCK(bp); } /* * This function fetches inode inum on mount point mp. We already * hold a locked vnode vp, and might have a locked buffer bp belonging * to vp. * We must not block on acquiring the new inode lock as we will get * into a lock-order reversal with the buffer lock and possibly get a * deadlock. Thus if we cannot instantiate the requested vnode * without sleeping on its lock, we must unlock the vnode and the * buffer before doing a blocking on the vnode lock. We return * ERELOOKUP if we have had to unlock either the vnode or the buffer so * that the caller can reassess its state. * * Top-level VFS code (for syscalls and other consumers, e.g. callers * of VOP_FSYNC() in syncer) check for ERELOOKUP and restart at safe * point. * * Since callers expect to operate on fully constructed vnode, we also * recheck v_data after relock, and return ENOENT if NULL. * * If unlocking bp, we must unroll dequeueing its unfinished * dependencies, and clear scan flag, before unlocking. If unlocking * vp while it is under deactivation, we re-queue deactivation. */ static int get_parent_vp(struct vnode *vp, struct mount *mp, ino_t inum, struct buf *bp, struct diraddhd *diraddhdp, struct diraddhd *unfinishedp, struct vnode **rvp) { struct vnode *pvp; int error; bool bplocked; ASSERT_VOP_ELOCKED(vp, "child vnode must be locked"); for (bplocked = true, pvp = NULL;;) { error = ffs_vgetf(mp, inum, LK_EXCLUSIVE | LK_NOWAIT, &pvp, FFSV_FORCEINSMQ | FFSV_FORCEINODEDEP); if (error == 0) { /* * Since we could have unlocked vp, the inode * number could no longer indicate a * constructed node. In this case, we must * restart the syscall. */ if (VTOI(pvp)->i_mode == 0 || !bplocked) { if (bp != NULL && bplocked) get_parent_vp_unlock_bp(mp, bp, diraddhdp, unfinishedp); if (VTOI(pvp)->i_mode == 0) vgone(pvp); error = ERELOOKUP; goto out2; } goto out1; } if (bp != NULL && bplocked) { get_parent_vp_unlock_bp(mp, bp, diraddhdp, unfinishedp); bplocked = false; } /* * Do not drop vnode lock while inactivating during * vunref. This would result in leaks of the VI flags * and reclaiming of non-truncated vnode. Instead, * re-schedule inactivation hoping that we would be * able to sync inode later. */ if ((vp->v_iflag & VI_DOINGINACT) != 0 && (vp->v_vflag & VV_UNREF) != 0) { VI_LOCK(vp); vp->v_iflag |= VI_OWEINACT; VI_UNLOCK(vp); return (ERELOOKUP); } VOP_UNLOCK(vp); error = ffs_vgetf(mp, inum, LK_EXCLUSIVE, &pvp, FFSV_FORCEINSMQ | FFSV_FORCEINODEDEP); if (error != 0) { MPASS(error != ERELOOKUP); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); break; } if (VTOI(pvp)->i_mode == 0) { vgone(pvp); vput(pvp); pvp = NULL; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); error = ERELOOKUP; break; } error = vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT); if (error == 0) break; vput(pvp); pvp = NULL; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); if (vp->v_data == NULL) { error = ENOENT; break; } } if (bp != NULL) { MPASS(!bplocked); error = ERELOOKUP; } out2: if (error != 0 && pvp != NULL) { vput(pvp); pvp = NULL; } out1: *rvp = pvp; ASSERT_VOP_ELOCKED(vp, "child vnode must be locked on return"); return (error); } /* * This function cleans the worklist for a filesystem. * Each filesystem running with soft dependencies gets its own * thread to run in this function. The thread is started up in * softdep_mount and shutdown in softdep_unmount. They show up * as part of the kernel "bufdaemon" process whose process * entry is available in bufdaemonproc. */ static int searchfailed; extern struct proc *bufdaemonproc; static void softdep_flush(addr) void *addr; { struct mount *mp; struct thread *td; struct ufsmount *ump; int cleanups; td = curthread; td->td_pflags |= TDP_NORUNNINGBUF; mp = (struct mount *)addr; ump = VFSTOUFS(mp); atomic_add_int(&stat_flush_threads, 1); ACQUIRE_LOCK(ump); ump->softdep_flags &= ~FLUSH_STARTING; wakeup(&ump->softdep_flushtd); FREE_LOCK(ump); if (print_threads) { if (stat_flush_threads == 1) printf("Running %s at pid %d\n", bufdaemonproc->p_comm, bufdaemonproc->p_pid); printf("Start thread %s\n", td->td_name); } for (;;) { while (softdep_process_worklist(mp, 0) > 0 || (MOUNTEDSUJ(mp) && VFSTOUFS(mp)->softdep_jblocks->jb_suspended)) kthread_suspend_check(); ACQUIRE_LOCK(ump); if ((ump->softdep_flags & (FLUSH_CLEANUP | FLUSH_EXIT)) == 0) msleep(&ump->softdep_flushtd, LOCK_PTR(ump), PVM, "sdflush", hz / 2); ump->softdep_flags &= ~FLUSH_CLEANUP; /* * Check to see if we are done and need to exit. */ if ((ump->softdep_flags & FLUSH_EXIT) == 0) { FREE_LOCK(ump); continue; } ump->softdep_flags &= ~FLUSH_EXIT; cleanups = ump->um_softdep->sd_cleanups; FREE_LOCK(ump); wakeup(&ump->softdep_flags); if (print_threads) { printf("Stop thread %s: searchfailed %d, " "did cleanups %d\n", td->td_name, searchfailed, cleanups); } atomic_subtract_int(&stat_flush_threads, 1); kthread_exit(); panic("kthread_exit failed\n"); } } static void worklist_speedup(mp) struct mount *mp; { struct ufsmount *ump; ump = VFSTOUFS(mp); LOCK_OWNED(ump); if ((ump->softdep_flags & (FLUSH_CLEANUP | FLUSH_EXIT)) == 0) ump->softdep_flags |= FLUSH_CLEANUP; wakeup(&ump->softdep_flushtd); } static void softdep_send_speedup(struct ufsmount *ump, off_t shortage, u_int flags) { struct buf *bp; if ((ump->um_flags & UM_CANSPEEDUP) == 0) return; bp = malloc(sizeof(*bp), M_TRIM, M_WAITOK | M_ZERO); bp->b_iocmd = BIO_SPEEDUP; bp->b_ioflags = flags; bp->b_bcount = omin(shortage, LONG_MAX); g_vfs_strategy(ump->um_bo, bp); bufwait(bp); free(bp, M_TRIM); } static int softdep_speedup(ump) struct ufsmount *ump; { struct ufsmount *altump; struct mount_softdeps *sdp; LOCK_OWNED(ump); worklist_speedup(ump->um_mountp); bd_speedup(); /* * If we have global shortages, then we need other * filesystems to help with the cleanup. Here we wakeup a * flusher thread for a filesystem that is over its fair * share of resources. */ if (req_clear_inodedeps || req_clear_remove) { ACQUIRE_GBLLOCK(&lk); TAILQ_FOREACH(sdp, &softdepmounts, sd_next) { if ((altump = sdp->sd_ump) == ump) continue; if (((req_clear_inodedeps && altump->softdep_curdeps[D_INODEDEP] > max_softdeps / stat_flush_threads) || (req_clear_remove && altump->softdep_curdeps[D_DIRREM] > (max_softdeps / 2) / stat_flush_threads)) && TRY_ACQUIRE_LOCK(altump)) break; } if (sdp == NULL) { searchfailed++; FREE_GBLLOCK(&lk); } else { /* * Move to the end of the list so we pick a * different one on out next try. */ TAILQ_REMOVE(&softdepmounts, sdp, sd_next); TAILQ_INSERT_TAIL(&softdepmounts, sdp, sd_next); FREE_GBLLOCK(&lk); if ((altump->softdep_flags & (FLUSH_CLEANUP | FLUSH_EXIT)) == 0) altump->softdep_flags |= FLUSH_CLEANUP; altump->um_softdep->sd_cleanups++; wakeup(&altump->softdep_flushtd); FREE_LOCK(altump); } } return (speedup_syncer()); } /* * Add an item to the end of the work queue. * This routine requires that the lock be held. * This is the only routine that adds items to the list. * The following routine is the only one that removes items * and does so in order from first to last. */ #define WK_HEAD 0x0001 /* Add to HEAD. */ #define WK_NODELAY 0x0002 /* Process immediately. */ static void add_to_worklist(wk, flags) struct worklist *wk; int flags; { struct ufsmount *ump; ump = VFSTOUFS(wk->wk_mp); LOCK_OWNED(ump); if (wk->wk_state & ONWORKLIST) panic("add_to_worklist: %s(0x%X) already on list", TYPENAME(wk->wk_type), wk->wk_state); wk->wk_state |= ONWORKLIST; if (ump->softdep_on_worklist == 0) { LIST_INSERT_HEAD(&ump->softdep_workitem_pending, wk, wk_list); ump->softdep_worklist_tail = wk; } else if (flags & WK_HEAD) { LIST_INSERT_HEAD(&ump->softdep_workitem_pending, wk, wk_list); } else { LIST_INSERT_AFTER(ump->softdep_worklist_tail, wk, wk_list); ump->softdep_worklist_tail = wk; } ump->softdep_on_worklist += 1; if (flags & WK_NODELAY) worklist_speedup(wk->wk_mp); } /* * Remove the item to be processed. If we are removing the last * item on the list, we need to recalculate the tail pointer. */ static void remove_from_worklist(wk) struct worklist *wk; { struct ufsmount *ump; ump = VFSTOUFS(wk->wk_mp); if (ump->softdep_worklist_tail == wk) ump->softdep_worklist_tail = (struct worklist *)wk->wk_list.le_prev; WORKLIST_REMOVE(wk); ump->softdep_on_worklist -= 1; } static void wake_worklist(wk) struct worklist *wk; { if (wk->wk_state & IOWAITING) { wk->wk_state &= ~IOWAITING; wakeup(wk); } } static void wait_worklist(wk, wmesg) struct worklist *wk; char *wmesg; { struct ufsmount *ump; ump = VFSTOUFS(wk->wk_mp); wk->wk_state |= IOWAITING; msleep(wk, LOCK_PTR(ump), PVM, wmesg, 0); } /* * Process that runs once per second to handle items in the background queue. * * Note that we ensure that everything is done in the order in which they * appear in the queue. The code below depends on this property to ensure * that blocks of a file are freed before the inode itself is freed. This * ordering ensures that no new triples will be generated * until all the old ones have been purged from the dependency lists. */ static int softdep_process_worklist(mp, full) struct mount *mp; int full; { int cnt, matchcnt; struct ufsmount *ump; long starttime; KASSERT(mp != NULL, ("softdep_process_worklist: NULL mp")); ump = VFSTOUFS(mp); if (ump->um_softdep == NULL) return (0); matchcnt = 0; ACQUIRE_LOCK(ump); starttime = time_second; softdep_process_journal(mp, NULL, full ? MNT_WAIT : 0); check_clear_deps(mp); while (ump->softdep_on_worklist > 0) { if ((cnt = process_worklist_item(mp, 10, LK_NOWAIT)) == 0) break; else matchcnt += cnt; check_clear_deps(mp); /* * We do not generally want to stop for buffer space, but if * we are really being a buffer hog, we will stop and wait. */ if (should_yield()) { FREE_LOCK(ump); kern_yield(PRI_USER); bwillwrite(); ACQUIRE_LOCK(ump); } /* * Never allow processing to run for more than one * second. This gives the syncer thread the opportunity * to pause if appropriate. */ if (!full && starttime != time_second) break; } if (full == 0) journal_unsuspend(ump); FREE_LOCK(ump); return (matchcnt); } /* * Process all removes associated with a vnode if we are running out of * journal space. Any other process which attempts to flush these will * be unable as we have the vnodes locked. */ static void process_removes(vp) struct vnode *vp; { struct inodedep *inodedep; struct dirrem *dirrem; struct ufsmount *ump; struct mount *mp; ino_t inum; mp = vp->v_mount; ump = VFSTOUFS(mp); LOCK_OWNED(ump); inum = VTOI(vp)->i_number; for (;;) { top: if (inodedep_lookup(mp, inum, 0, &inodedep) == 0) return; LIST_FOREACH(dirrem, &inodedep->id_dirremhd, dm_inonext) { /* * If another thread is trying to lock this vnode * it will fail but we must wait for it to do so * before we can proceed. */ if (dirrem->dm_state & INPROGRESS) { wait_worklist(&dirrem->dm_list, "pwrwait"); goto top; } if ((dirrem->dm_state & (COMPLETE | ONWORKLIST)) == (COMPLETE | ONWORKLIST)) break; } if (dirrem == NULL) return; remove_from_worklist(&dirrem->dm_list); FREE_LOCK(ump); if (vn_start_secondary_write(NULL, &mp, V_NOWAIT)) panic("process_removes: suspended filesystem"); handle_workitem_remove(dirrem, 0); vn_finished_secondary_write(mp); ACQUIRE_LOCK(ump); } } /* * Process all truncations associated with a vnode if we are running out * of journal space. This is called when the vnode lock is already held * and no other process can clear the truncation. This function returns * a value greater than zero if it did any work. */ static void process_truncates(vp) struct vnode *vp; { struct inodedep *inodedep; struct freeblks *freeblks; struct ufsmount *ump; struct mount *mp; ino_t inum; int cgwait; mp = vp->v_mount; ump = VFSTOUFS(mp); LOCK_OWNED(ump); inum = VTOI(vp)->i_number; for (;;) { if (inodedep_lookup(mp, inum, 0, &inodedep) == 0) return; cgwait = 0; TAILQ_FOREACH(freeblks, &inodedep->id_freeblklst, fb_next) { /* Journal entries not yet written. */ if (!LIST_EMPTY(&freeblks->fb_jblkdephd)) { jwait(&LIST_FIRST( &freeblks->fb_jblkdephd)->jb_list, MNT_WAIT); break; } /* Another thread is executing this item. */ if (freeblks->fb_state & INPROGRESS) { wait_worklist(&freeblks->fb_list, "ptrwait"); break; } /* Freeblks is waiting on a inode write. */ if ((freeblks->fb_state & COMPLETE) == 0) { FREE_LOCK(ump); ffs_update(vp, 1); ACQUIRE_LOCK(ump); break; } if ((freeblks->fb_state & (ALLCOMPLETE | ONWORKLIST)) == (ALLCOMPLETE | ONWORKLIST)) { remove_from_worklist(&freeblks->fb_list); freeblks->fb_state |= INPROGRESS; FREE_LOCK(ump); if (vn_start_secondary_write(NULL, &mp, V_NOWAIT)) panic("process_truncates: " "suspended filesystem"); handle_workitem_freeblocks(freeblks, 0); vn_finished_secondary_write(mp); ACQUIRE_LOCK(ump); break; } if (freeblks->fb_cgwait) cgwait++; } if (cgwait) { FREE_LOCK(ump); sync_cgs(mp, MNT_WAIT); ffs_sync_snap(mp, MNT_WAIT); ACQUIRE_LOCK(ump); continue; } if (freeblks == NULL) break; } return; } /* * Process one item on the worklist. */ static int process_worklist_item(mp, target, flags) struct mount *mp; int target; int flags; { struct worklist sentinel; struct worklist *wk; struct ufsmount *ump; int matchcnt; int error; KASSERT(mp != NULL, ("process_worklist_item: NULL mp")); /* * If we are being called because of a process doing a * copy-on-write, then it is not safe to write as we may * recurse into the copy-on-write routine. */ if (curthread->td_pflags & TDP_COWINPROGRESS) return (-1); PHOLD(curproc); /* Don't let the stack go away. */ ump = VFSTOUFS(mp); LOCK_OWNED(ump); matchcnt = 0; sentinel.wk_mp = NULL; sentinel.wk_type = D_SENTINEL; LIST_INSERT_HEAD(&ump->softdep_workitem_pending, &sentinel, wk_list); for (wk = LIST_NEXT(&sentinel, wk_list); wk != NULL; wk = LIST_NEXT(&sentinel, wk_list)) { if (wk->wk_type == D_SENTINEL) { LIST_REMOVE(&sentinel, wk_list); LIST_INSERT_AFTER(wk, &sentinel, wk_list); continue; } if (wk->wk_state & INPROGRESS) panic("process_worklist_item: %p already in progress.", wk); wk->wk_state |= INPROGRESS; remove_from_worklist(wk); FREE_LOCK(ump); if (vn_start_secondary_write(NULL, &mp, V_NOWAIT)) panic("process_worklist_item: suspended filesystem"); switch (wk->wk_type) { case D_DIRREM: /* removal of a directory entry */ error = handle_workitem_remove(WK_DIRREM(wk), flags); break; case D_FREEBLKS: /* releasing blocks and/or fragments from a file */ error = handle_workitem_freeblocks(WK_FREEBLKS(wk), flags); break; case D_FREEFRAG: /* releasing a fragment when replaced as a file grows */ handle_workitem_freefrag(WK_FREEFRAG(wk)); error = 0; break; case D_FREEFILE: /* releasing an inode when its link count drops to 0 */ handle_workitem_freefile(WK_FREEFILE(wk)); error = 0; break; default: panic("%s_process_worklist: Unknown type %s", "softdep", TYPENAME(wk->wk_type)); /* NOTREACHED */ } vn_finished_secondary_write(mp); ACQUIRE_LOCK(ump); if (error == 0) { if (++matchcnt == target) break; continue; } /* * We have to retry the worklist item later. Wake up any * waiters who may be able to complete it immediately and * add the item back to the head so we don't try to execute * it again. */ wk->wk_state &= ~INPROGRESS; wake_worklist(wk); add_to_worklist(wk, WK_HEAD); } /* Sentinal could've become the tail from remove_from_worklist. */ if (ump->softdep_worklist_tail == &sentinel) ump->softdep_worklist_tail = (struct worklist *)sentinel.wk_list.le_prev; LIST_REMOVE(&sentinel, wk_list); PRELE(curproc); return (matchcnt); } /* * Move dependencies from one buffer to another. */ int softdep_move_dependencies(oldbp, newbp) struct buf *oldbp; struct buf *newbp; { struct worklist *wk, *wktail; struct ufsmount *ump; int dirty; if ((wk = LIST_FIRST(&oldbp->b_dep)) == NULL) return (0); KASSERT(MOUNTEDSOFTDEP(wk->wk_mp) != 0, ("softdep_move_dependencies called on non-softdep filesystem")); dirty = 0; wktail = NULL; ump = VFSTOUFS(wk->wk_mp); ACQUIRE_LOCK(ump); while ((wk = LIST_FIRST(&oldbp->b_dep)) != NULL) { LIST_REMOVE(wk, wk_list); if (wk->wk_type == D_BMSAFEMAP && bmsafemap_backgroundwrite(WK_BMSAFEMAP(wk), newbp)) dirty = 1; if (wktail == NULL) LIST_INSERT_HEAD(&newbp->b_dep, wk, wk_list); else LIST_INSERT_AFTER(wktail, wk, wk_list); wktail = wk; } FREE_LOCK(ump); return (dirty); } /* * Purge the work list of all items associated with a particular mount point. */ int softdep_flushworklist(oldmnt, countp, td) struct mount *oldmnt; int *countp; struct thread *td; { struct vnode *devvp; struct ufsmount *ump; int count, error; /* * Alternately flush the block device associated with the mount * point and process any dependencies that the flushing * creates. We continue until no more worklist dependencies * are found. */ *countp = 0; error = 0; ump = VFSTOUFS(oldmnt); devvp = ump->um_devvp; while ((count = softdep_process_worklist(oldmnt, 1)) > 0) { *countp += count; vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); error = VOP_FSYNC(devvp, MNT_WAIT, td); VOP_UNLOCK(devvp); if (error != 0) break; } return (error); } #define SU_WAITIDLE_RETRIES 20 static int softdep_waitidle(struct mount *mp, int flags __unused) { struct ufsmount *ump; struct vnode *devvp; struct thread *td; int error, i; ump = VFSTOUFS(mp); KASSERT(ump->um_softdep != NULL, ("softdep_waitidle called on non-softdep filesystem")); devvp = ump->um_devvp; td = curthread; error = 0; ACQUIRE_LOCK(ump); for (i = 0; i < SU_WAITIDLE_RETRIES && ump->softdep_deps != 0; i++) { ump->softdep_req = 1; KASSERT((flags & FORCECLOSE) == 0 || ump->softdep_on_worklist == 0, ("softdep_waitidle: work added after flush")); msleep(&ump->softdep_deps, LOCK_PTR(ump), PVM | PDROP, "softdeps", 10 * hz); vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); error = VOP_FSYNC(devvp, MNT_WAIT, td); VOP_UNLOCK(devvp); ACQUIRE_LOCK(ump); if (error != 0) break; } ump->softdep_req = 0; if (i == SU_WAITIDLE_RETRIES && error == 0 && ump->softdep_deps != 0) { error = EBUSY; printf("softdep_waitidle: Failed to flush worklist for %p\n", mp); } FREE_LOCK(ump); return (error); } /* * Flush all vnodes and worklist items associated with a specified mount point. */ int softdep_flushfiles(oldmnt, flags, td) struct mount *oldmnt; int flags; struct thread *td; { struct ufsmount *ump; #ifdef QUOTA int i; #endif int error, early, depcount, loopcnt, retry_flush_count, retry; int morework; ump = VFSTOUFS(oldmnt); KASSERT(ump->um_softdep != NULL, ("softdep_flushfiles called on non-softdep filesystem")); loopcnt = 10; retry_flush_count = 3; retry_flush: error = 0; /* * Alternately flush the vnodes associated with the mount * point and process any dependencies that the flushing * creates. In theory, this loop can happen at most twice, * but we give it a few extra just to be sure. */ for (; loopcnt > 0; loopcnt--) { /* * Do another flush in case any vnodes were brought in * as part of the cleanup operations. */ early = retry_flush_count == 1 || (oldmnt->mnt_kern_flag & MNTK_UNMOUNT) == 0 ? 0 : EARLYFLUSH; if ((error = ffs_flushfiles(oldmnt, flags | early, td)) != 0) break; if ((error = softdep_flushworklist(oldmnt, &depcount, td)) != 0 || depcount == 0) break; } /* * If we are unmounting then it is an error to fail. If we * are simply trying to downgrade to read-only, then filesystem * activity can keep us busy forever, so we just fail with EBUSY. */ if (loopcnt == 0) { if (oldmnt->mnt_kern_flag & MNTK_UNMOUNT) panic("softdep_flushfiles: looping"); error = EBUSY; } if (!error) error = softdep_waitidle(oldmnt, flags); if (!error) { if (oldmnt->mnt_kern_flag & MNTK_UNMOUNT) { retry = 0; MNT_ILOCK(oldmnt); morework = oldmnt->mnt_nvnodelistsize > 0; #ifdef QUOTA UFS_LOCK(ump); for (i = 0; i < MAXQUOTAS; i++) { if (ump->um_quotas[i] != NULLVP) morework = 1; } UFS_UNLOCK(ump); #endif if (morework) { if (--retry_flush_count > 0) { retry = 1; loopcnt = 3; } else error = EBUSY; } MNT_IUNLOCK(oldmnt); if (retry) goto retry_flush; } } return (error); } /* * Structure hashing. * * There are four types of structures that can be looked up: * 1) pagedep structures identified by mount point, inode number, * and logical block. * 2) inodedep structures identified by mount point and inode number. * 3) newblk structures identified by mount point and * physical block number. * 4) bmsafemap structures identified by mount point and * cylinder group number. * * The "pagedep" and "inodedep" dependency structures are hashed * separately from the file blocks and inodes to which they correspond. * This separation helps when the in-memory copy of an inode or * file block must be replaced. It also obviates the need to access * an inode or file page when simply updating (or de-allocating) * dependency structures. Lookup of newblk structures is needed to * find newly allocated blocks when trying to associate them with * their allocdirect or allocindir structure. * * The lookup routines optionally create and hash a new instance when * an existing entry is not found. The bmsafemap lookup routine always * allocates a new structure if an existing one is not found. */ #define DEPALLOC 0x0001 /* allocate structure if lookup fails */ /* * Structures and routines associated with pagedep caching. */ #define PAGEDEP_HASH(ump, inum, lbn) \ (&(ump)->pagedep_hashtbl[((inum) + (lbn)) & (ump)->pagedep_hash_size]) static int pagedep_find(pagedephd, ino, lbn, pagedeppp) struct pagedep_hashhead *pagedephd; ino_t ino; ufs_lbn_t lbn; struct pagedep **pagedeppp; { struct pagedep *pagedep; LIST_FOREACH(pagedep, pagedephd, pd_hash) { if (ino == pagedep->pd_ino && lbn == pagedep->pd_lbn) { *pagedeppp = pagedep; return (1); } } *pagedeppp = NULL; return (0); } /* * Look up a pagedep. Return 1 if found, 0 otherwise. * If not found, allocate if DEPALLOC flag is passed. * Found or allocated entry is returned in pagedeppp. */ static int pagedep_lookup(mp, bp, ino, lbn, flags, pagedeppp) struct mount *mp; struct buf *bp; ino_t ino; ufs_lbn_t lbn; int flags; struct pagedep **pagedeppp; { struct pagedep *pagedep; struct pagedep_hashhead *pagedephd; struct worklist *wk; struct ufsmount *ump; int ret; int i; ump = VFSTOUFS(mp); LOCK_OWNED(ump); if (bp) { LIST_FOREACH(wk, &bp->b_dep, wk_list) { if (wk->wk_type == D_PAGEDEP) { *pagedeppp = WK_PAGEDEP(wk); return (1); } } } pagedephd = PAGEDEP_HASH(ump, ino, lbn); ret = pagedep_find(pagedephd, ino, lbn, pagedeppp); if (ret) { if (((*pagedeppp)->pd_state & ONWORKLIST) == 0 && bp) WORKLIST_INSERT(&bp->b_dep, &(*pagedeppp)->pd_list); return (1); } if ((flags & DEPALLOC) == 0) return (0); FREE_LOCK(ump); pagedep = malloc(sizeof(struct pagedep), M_PAGEDEP, M_SOFTDEP_FLAGS|M_ZERO); workitem_alloc(&pagedep->pd_list, D_PAGEDEP, mp); ACQUIRE_LOCK(ump); ret = pagedep_find(pagedephd, ino, lbn, pagedeppp); if (*pagedeppp) { /* * This should never happen since we only create pagedeps * with the vnode lock held. Could be an assert. */ WORKITEM_FREE(pagedep, D_PAGEDEP); return (ret); } pagedep->pd_ino = ino; pagedep->pd_lbn = lbn; LIST_INIT(&pagedep->pd_dirremhd); LIST_INIT(&pagedep->pd_pendinghd); for (i = 0; i < DAHASHSZ; i++) LIST_INIT(&pagedep->pd_diraddhd[i]); LIST_INSERT_HEAD(pagedephd, pagedep, pd_hash); WORKLIST_INSERT(&bp->b_dep, &pagedep->pd_list); *pagedeppp = pagedep; return (0); } /* * Structures and routines associated with inodedep caching. */ #define INODEDEP_HASH(ump, inum) \ (&(ump)->inodedep_hashtbl[(inum) & (ump)->inodedep_hash_size]) static int inodedep_find(inodedephd, inum, inodedeppp) struct inodedep_hashhead *inodedephd; ino_t inum; struct inodedep **inodedeppp; { struct inodedep *inodedep; LIST_FOREACH(inodedep, inodedephd, id_hash) if (inum == inodedep->id_ino) break; if (inodedep) { *inodedeppp = inodedep; return (1); } *inodedeppp = NULL; return (0); } /* * Look up an inodedep. Return 1 if found, 0 if not found. * If not found, allocate if DEPALLOC flag is passed. * Found or allocated entry is returned in inodedeppp. */ static int inodedep_lookup(mp, inum, flags, inodedeppp) struct mount *mp; ino_t inum; int flags; struct inodedep **inodedeppp; { struct inodedep *inodedep; struct inodedep_hashhead *inodedephd; struct ufsmount *ump; struct fs *fs; ump = VFSTOUFS(mp); LOCK_OWNED(ump); fs = ump->um_fs; inodedephd = INODEDEP_HASH(ump, inum); if (inodedep_find(inodedephd, inum, inodedeppp)) return (1); if ((flags & DEPALLOC) == 0) return (0); /* * If the system is over its limit and our filesystem is * responsible for more than our share of that usage and * we are not in a rush, request some inodedep cleanup. */ if (softdep_excess_items(ump, D_INODEDEP)) schedule_cleanup(mp); else FREE_LOCK(ump); inodedep = malloc(sizeof(struct inodedep), M_INODEDEP, M_SOFTDEP_FLAGS); workitem_alloc(&inodedep->id_list, D_INODEDEP, mp); ACQUIRE_LOCK(ump); if (inodedep_find(inodedephd, inum, inodedeppp)) { WORKITEM_FREE(inodedep, D_INODEDEP); return (1); } inodedep->id_fs = fs; inodedep->id_ino = inum; inodedep->id_state = ALLCOMPLETE; inodedep->id_nlinkdelta = 0; inodedep->id_nlinkwrote = -1; inodedep->id_savedino1 = NULL; inodedep->id_savedsize = -1; inodedep->id_savedextsize = -1; inodedep->id_savednlink = -1; inodedep->id_bmsafemap = NULL; inodedep->id_mkdiradd = NULL; LIST_INIT(&inodedep->id_dirremhd); LIST_INIT(&inodedep->id_pendinghd); LIST_INIT(&inodedep->id_inowait); LIST_INIT(&inodedep->id_bufwait); TAILQ_INIT(&inodedep->id_inoreflst); TAILQ_INIT(&inodedep->id_inoupdt); TAILQ_INIT(&inodedep->id_newinoupdt); TAILQ_INIT(&inodedep->id_extupdt); TAILQ_INIT(&inodedep->id_newextupdt); TAILQ_INIT(&inodedep->id_freeblklst); LIST_INSERT_HEAD(inodedephd, inodedep, id_hash); *inodedeppp = inodedep; return (0); } /* * Structures and routines associated with newblk caching. */ #define NEWBLK_HASH(ump, inum) \ (&(ump)->newblk_hashtbl[(inum) & (ump)->newblk_hash_size]) static int newblk_find(newblkhd, newblkno, flags, newblkpp) struct newblk_hashhead *newblkhd; ufs2_daddr_t newblkno; int flags; struct newblk **newblkpp; { struct newblk *newblk; LIST_FOREACH(newblk, newblkhd, nb_hash) { if (newblkno != newblk->nb_newblkno) continue; /* * If we're creating a new dependency don't match those that * have already been converted to allocdirects. This is for * a frag extend. */ if ((flags & DEPALLOC) && newblk->nb_list.wk_type != D_NEWBLK) continue; break; } if (newblk) { *newblkpp = newblk; return (1); } *newblkpp = NULL; return (0); } /* * Look up a newblk. Return 1 if found, 0 if not found. * If not found, allocate if DEPALLOC flag is passed. * Found or allocated entry is returned in newblkpp. */ static int newblk_lookup(mp, newblkno, flags, newblkpp) struct mount *mp; ufs2_daddr_t newblkno; int flags; struct newblk **newblkpp; { struct newblk *newblk; struct newblk_hashhead *newblkhd; struct ufsmount *ump; ump = VFSTOUFS(mp); LOCK_OWNED(ump); newblkhd = NEWBLK_HASH(ump, newblkno); if (newblk_find(newblkhd, newblkno, flags, newblkpp)) return (1); if ((flags & DEPALLOC) == 0) return (0); if (softdep_excess_items(ump, D_NEWBLK) || softdep_excess_items(ump, D_ALLOCDIRECT) || softdep_excess_items(ump, D_ALLOCINDIR)) schedule_cleanup(mp); else FREE_LOCK(ump); newblk = malloc(sizeof(union allblk), M_NEWBLK, M_SOFTDEP_FLAGS | M_ZERO); workitem_alloc(&newblk->nb_list, D_NEWBLK, mp); ACQUIRE_LOCK(ump); if (newblk_find(newblkhd, newblkno, flags, newblkpp)) { WORKITEM_FREE(newblk, D_NEWBLK); return (1); } newblk->nb_freefrag = NULL; LIST_INIT(&newblk->nb_indirdeps); LIST_INIT(&newblk->nb_newdirblk); LIST_INIT(&newblk->nb_jwork); newblk->nb_state = ATTACHED; newblk->nb_newblkno = newblkno; LIST_INSERT_HEAD(newblkhd, newblk, nb_hash); *newblkpp = newblk; return (0); } /* * Structures and routines associated with freed indirect block caching. */ #define INDIR_HASH(ump, blkno) \ (&(ump)->indir_hashtbl[(blkno) & (ump)->indir_hash_size]) /* * Lookup an indirect block in the indir hash table. The freework is * removed and potentially freed. The caller must do a blocking journal * write before writing to the blkno. */ static int indirblk_lookup(mp, blkno) struct mount *mp; ufs2_daddr_t blkno; { struct freework *freework; struct indir_hashhead *wkhd; struct ufsmount *ump; ump = VFSTOUFS(mp); wkhd = INDIR_HASH(ump, blkno); TAILQ_FOREACH(freework, wkhd, fw_next) { if (freework->fw_blkno != blkno) continue; indirblk_remove(freework); return (1); } return (0); } /* * Insert an indirect block represented by freework into the indirblk * hash table so that it may prevent the block from being re-used prior * to the journal being written. */ static void indirblk_insert(freework) struct freework *freework; { struct jblocks *jblocks; struct jseg *jseg; struct ufsmount *ump; ump = VFSTOUFS(freework->fw_list.wk_mp); jblocks = ump->softdep_jblocks; jseg = TAILQ_LAST(&jblocks->jb_segs, jseglst); if (jseg == NULL) return; LIST_INSERT_HEAD(&jseg->js_indirs, freework, fw_segs); TAILQ_INSERT_HEAD(INDIR_HASH(ump, freework->fw_blkno), freework, fw_next); freework->fw_state &= ~DEPCOMPLETE; } static void indirblk_remove(freework) struct freework *freework; { struct ufsmount *ump; ump = VFSTOUFS(freework->fw_list.wk_mp); LIST_REMOVE(freework, fw_segs); TAILQ_REMOVE(INDIR_HASH(ump, freework->fw_blkno), freework, fw_next); freework->fw_state |= DEPCOMPLETE; if ((freework->fw_state & ALLCOMPLETE) == ALLCOMPLETE) WORKITEM_FREE(freework, D_FREEWORK); } /* * Executed during filesystem system initialization before * mounting any filesystems. */ void softdep_initialize() { TAILQ_INIT(&softdepmounts); #ifdef __LP64__ max_softdeps = desiredvnodes * 4; #else max_softdeps = desiredvnodes * 2; #endif /* initialise bioops hack */ bioops.io_start = softdep_disk_io_initiation; bioops.io_complete = softdep_disk_write_complete; bioops.io_deallocate = softdep_deallocate_dependencies; bioops.io_countdeps = softdep_count_dependencies; softdep_ast_cleanup = softdep_ast_cleanup_proc; /* Initialize the callout with an mtx. */ callout_init_mtx(&softdep_callout, &lk, 0); } /* * Executed after all filesystems have been unmounted during * filesystem module unload. */ void softdep_uninitialize() { /* clear bioops hack */ bioops.io_start = NULL; bioops.io_complete = NULL; bioops.io_deallocate = NULL; bioops.io_countdeps = NULL; softdep_ast_cleanup = NULL; callout_drain(&softdep_callout); } /* * Called at mount time to notify the dependency code that a * filesystem wishes to use it. */ int softdep_mount(devvp, mp, fs, cred) struct vnode *devvp; struct mount *mp; struct fs *fs; struct ucred *cred; { struct csum_total cstotal; struct mount_softdeps *sdp; struct ufsmount *ump; struct cg *cgp; struct buf *bp; u_int cyl, i; int error; ump = VFSTOUFS(mp); sdp = malloc(sizeof(struct mount_softdeps), M_MOUNTDATA, M_WAITOK | M_ZERO); rw_init(&sdp->sd_fslock, "SUrw"); sdp->sd_ump = ump; LIST_INIT(&sdp->sd_workitem_pending); LIST_INIT(&sdp->sd_journal_pending); TAILQ_INIT(&sdp->sd_unlinked); LIST_INIT(&sdp->sd_dirtycg); sdp->sd_worklist_tail = NULL; sdp->sd_on_worklist = 0; sdp->sd_deps = 0; LIST_INIT(&sdp->sd_mkdirlisthd); sdp->sd_pdhash = hashinit(desiredvnodes / 5, M_PAGEDEP, &sdp->sd_pdhashsize); sdp->sd_pdnextclean = 0; sdp->sd_idhash = hashinit(desiredvnodes, M_INODEDEP, &sdp->sd_idhashsize); sdp->sd_idnextclean = 0; sdp->sd_newblkhash = hashinit(max_softdeps / 2, M_NEWBLK, &sdp->sd_newblkhashsize); sdp->sd_bmhash = hashinit(1024, M_BMSAFEMAP, &sdp->sd_bmhashsize); i = 1 << (ffs(desiredvnodes / 10) - 1); sdp->sd_indirhash = malloc(i * sizeof(struct indir_hashhead), M_FREEWORK, M_WAITOK); sdp->sd_indirhashsize = i - 1; for (i = 0; i <= sdp->sd_indirhashsize; i++) TAILQ_INIT(&sdp->sd_indirhash[i]); for (i = 0; i <= D_LAST; i++) LIST_INIT(&sdp->sd_alldeps[i]); ACQUIRE_GBLLOCK(&lk); TAILQ_INSERT_TAIL(&softdepmounts, sdp, sd_next); FREE_GBLLOCK(&lk); ump->um_softdep = sdp; MNT_ILOCK(mp); mp->mnt_flag = (mp->mnt_flag & ~MNT_ASYNC) | MNT_SOFTDEP; if ((mp->mnt_kern_flag & MNTK_SOFTDEP) == 0) { mp->mnt_kern_flag = (mp->mnt_kern_flag & ~MNTK_ASYNC) | MNTK_SOFTDEP | MNTK_NOASYNC; } MNT_IUNLOCK(mp); if ((fs->fs_flags & FS_SUJ) && (error = journal_mount(mp, fs, cred)) != 0) { printf("Failed to start journal: %d\n", error); softdep_unmount(mp); return (error); } /* * Start our flushing thread in the bufdaemon process. */ ACQUIRE_LOCK(ump); ump->softdep_flags |= FLUSH_STARTING; FREE_LOCK(ump); kproc_kthread_add(&softdep_flush, mp, &bufdaemonproc, &ump->softdep_flushtd, 0, 0, "softdepflush", "%s worker", mp->mnt_stat.f_mntonname); ACQUIRE_LOCK(ump); while ((ump->softdep_flags & FLUSH_STARTING) != 0) { msleep(&ump->softdep_flushtd, LOCK_PTR(ump), PVM, "sdstart", hz / 2); } FREE_LOCK(ump); /* * When doing soft updates, the counters in the * superblock may have gotten out of sync. Recomputation * can take a long time and can be deferred for background * fsck. However, the old behavior of scanning the cylinder * groups and recalculating them at mount time is available * by setting vfs.ffs.compute_summary_at_mount to one. */ if (compute_summary_at_mount == 0 || fs->fs_clean != 0) return (0); bzero(&cstotal, sizeof cstotal); for (cyl = 0; cyl < fs->fs_ncg; cyl++) { if ((error = bread(devvp, fsbtodb(fs, cgtod(fs, cyl)), fs->fs_cgsize, cred, &bp)) != 0) { brelse(bp); softdep_unmount(mp); return (error); } cgp = (struct cg *)bp->b_data; cstotal.cs_nffree += cgp->cg_cs.cs_nffree; cstotal.cs_nbfree += cgp->cg_cs.cs_nbfree; cstotal.cs_nifree += cgp->cg_cs.cs_nifree; cstotal.cs_ndir += cgp->cg_cs.cs_ndir; fs->fs_cs(fs, cyl) = cgp->cg_cs; brelse(bp); } #ifdef INVARIANTS if (bcmp(&cstotal, &fs->fs_cstotal, sizeof cstotal)) printf("%s: superblock summary recomputed\n", fs->fs_fsmnt); #endif bcopy(&cstotal, &fs->fs_cstotal, sizeof cstotal); return (0); } void softdep_unmount(mp) struct mount *mp; { struct ufsmount *ump; struct mount_softdeps *ums; ump = VFSTOUFS(mp); KASSERT(ump->um_softdep != NULL, ("softdep_unmount called on non-softdep filesystem")); MNT_ILOCK(mp); mp->mnt_flag &= ~MNT_SOFTDEP; if ((mp->mnt_flag & MNT_SUJ) == 0) { MNT_IUNLOCK(mp); } else { mp->mnt_flag &= ~MNT_SUJ; MNT_IUNLOCK(mp); journal_unmount(ump); } /* * Shut down our flushing thread. Check for NULL is if * softdep_mount errors out before the thread has been created. */ if (ump->softdep_flushtd != NULL) { ACQUIRE_LOCK(ump); ump->softdep_flags |= FLUSH_EXIT; wakeup(&ump->softdep_flushtd); while ((ump->softdep_flags & FLUSH_EXIT) != 0) { msleep(&ump->softdep_flags, LOCK_PTR(ump), PVM, "sdwait", 0); } KASSERT((ump->softdep_flags & FLUSH_EXIT) == 0, ("Thread shutdown failed")); FREE_LOCK(ump); } /* * We are no longer have softdep structure attached to ump. */ ums = ump->um_softdep; ACQUIRE_GBLLOCK(&lk); TAILQ_REMOVE(&softdepmounts, ums, sd_next); FREE_GBLLOCK(&lk); ump->um_softdep = NULL; KASSERT(ums->sd_on_journal == 0, ("ump %p ums %p on_journal %d", ump, ums, ums->sd_on_journal)); KASSERT(ums->sd_on_worklist == 0, ("ump %p ums %p on_worklist %d", ump, ums, ums->sd_on_worklist)); KASSERT(ums->sd_deps == 0, ("ump %p ums %p deps %d", ump, ums, ums->sd_deps)); /* * Free up our resources. */ rw_destroy(&ums->sd_fslock); hashdestroy(ums->sd_pdhash, M_PAGEDEP, ums->sd_pdhashsize); hashdestroy(ums->sd_idhash, M_INODEDEP, ums->sd_idhashsize); hashdestroy(ums->sd_newblkhash, M_NEWBLK, ums->sd_newblkhashsize); hashdestroy(ums->sd_bmhash, M_BMSAFEMAP, ums->sd_bmhashsize); free(ums->sd_indirhash, M_FREEWORK); #ifdef INVARIANTS for (int i = 0; i <= D_LAST; i++) { KASSERT(ums->sd_curdeps[i] == 0, ("Unmount %s: Dep type %s != 0 (%ld)", ump->um_fs->fs_fsmnt, TYPENAME(i), ums->sd_curdeps[i])); KASSERT(LIST_EMPTY(&ums->sd_alldeps[i]), ("Unmount %s: Dep type %s not empty (%p)", ump->um_fs->fs_fsmnt, TYPENAME(i), LIST_FIRST(&ums->sd_alldeps[i]))); } #endif free(ums, M_MOUNTDATA); } static struct jblocks * jblocks_create(void) { struct jblocks *jblocks; jblocks = malloc(sizeof(*jblocks), M_JBLOCKS, M_WAITOK | M_ZERO); TAILQ_INIT(&jblocks->jb_segs); jblocks->jb_avail = 10; jblocks->jb_extent = malloc(sizeof(struct jextent) * jblocks->jb_avail, M_JBLOCKS, M_WAITOK | M_ZERO); return (jblocks); } static ufs2_daddr_t jblocks_alloc(jblocks, bytes, actual) struct jblocks *jblocks; int bytes; int *actual; { ufs2_daddr_t daddr; struct jextent *jext; int freecnt; int blocks; blocks = bytes / DEV_BSIZE; jext = &jblocks->jb_extent[jblocks->jb_head]; freecnt = jext->je_blocks - jblocks->jb_off; if (freecnt == 0) { jblocks->jb_off = 0; if (++jblocks->jb_head > jblocks->jb_used) jblocks->jb_head = 0; jext = &jblocks->jb_extent[jblocks->jb_head]; freecnt = jext->je_blocks; } if (freecnt > blocks) freecnt = blocks; *actual = freecnt * DEV_BSIZE; daddr = jext->je_daddr + jblocks->jb_off; jblocks->jb_off += freecnt; jblocks->jb_free -= freecnt; return (daddr); } static void jblocks_free(jblocks, mp, bytes) struct jblocks *jblocks; struct mount *mp; int bytes; { LOCK_OWNED(VFSTOUFS(mp)); jblocks->jb_free += bytes / DEV_BSIZE; if (jblocks->jb_suspended) worklist_speedup(mp); wakeup(jblocks); } static void jblocks_destroy(jblocks) struct jblocks *jblocks; { if (jblocks->jb_extent) free(jblocks->jb_extent, M_JBLOCKS); free(jblocks, M_JBLOCKS); } static void jblocks_add(jblocks, daddr, blocks) struct jblocks *jblocks; ufs2_daddr_t daddr; int blocks; { struct jextent *jext; jblocks->jb_blocks += blocks; jblocks->jb_free += blocks; jext = &jblocks->jb_extent[jblocks->jb_used]; /* Adding the first block. */ if (jext->je_daddr == 0) { jext->je_daddr = daddr; jext->je_blocks = blocks; return; } /* Extending the last extent. */ if (jext->je_daddr + jext->je_blocks == daddr) { jext->je_blocks += blocks; return; } /* Adding a new extent. */ if (++jblocks->jb_used == jblocks->jb_avail) { jblocks->jb_avail *= 2; jext = malloc(sizeof(struct jextent) * jblocks->jb_avail, M_JBLOCKS, M_WAITOK | M_ZERO); memcpy(jext, jblocks->jb_extent, sizeof(struct jextent) * jblocks->jb_used); free(jblocks->jb_extent, M_JBLOCKS); jblocks->jb_extent = jext; } jext = &jblocks->jb_extent[jblocks->jb_used]; jext->je_daddr = daddr; jext->je_blocks = blocks; return; } int softdep_journal_lookup(mp, vpp) struct mount *mp; struct vnode **vpp; { struct componentname cnp; struct vnode *dvp; ino_t sujournal; int error; error = VFS_VGET(mp, UFS_ROOTINO, LK_EXCLUSIVE, &dvp); if (error) return (error); bzero(&cnp, sizeof(cnp)); cnp.cn_nameiop = LOOKUP; cnp.cn_flags = ISLASTCN; - cnp.cn_thread = curthread; cnp.cn_cred = curthread->td_ucred; cnp.cn_pnbuf = SUJ_FILE; cnp.cn_nameptr = SUJ_FILE; cnp.cn_namelen = strlen(SUJ_FILE); error = ufs_lookup_ino(dvp, NULL, &cnp, &sujournal); vput(dvp); if (error != 0) return (error); error = VFS_VGET(mp, sujournal, LK_EXCLUSIVE, vpp); return (error); } /* * Open and verify the journal file. */ static int journal_mount(mp, fs, cred) struct mount *mp; struct fs *fs; struct ucred *cred; { struct jblocks *jblocks; struct ufsmount *ump; struct vnode *vp; struct inode *ip; ufs2_daddr_t blkno; int bcount; int error; int i; ump = VFSTOUFS(mp); ump->softdep_journal_tail = NULL; ump->softdep_on_journal = 0; ump->softdep_accdeps = 0; ump->softdep_req = 0; ump->softdep_jblocks = NULL; error = softdep_journal_lookup(mp, &vp); if (error != 0) { printf("Failed to find journal. Use tunefs to create one\n"); return (error); } ip = VTOI(vp); if (ip->i_size < SUJ_MIN) { error = ENOSPC; goto out; } bcount = lblkno(fs, ip->i_size); /* Only use whole blocks. */ jblocks = jblocks_create(); for (i = 0; i < bcount; i++) { error = ufs_bmaparray(vp, i, &blkno, NULL, NULL, NULL); if (error) break; jblocks_add(jblocks, blkno, fsbtodb(fs, fs->fs_frag)); } if (error) { jblocks_destroy(jblocks); goto out; } jblocks->jb_low = jblocks->jb_free / 3; /* Reserve 33%. */ jblocks->jb_min = jblocks->jb_free / 10; /* Suspend at 10%. */ ump->softdep_jblocks = jblocks; MNT_ILOCK(mp); mp->mnt_flag |= MNT_SUJ; MNT_IUNLOCK(mp); /* * Only validate the journal contents if the * filesystem is clean, otherwise we write the logs * but they'll never be used. If the filesystem was * still dirty when we mounted it the journal is * invalid and a new journal can only be valid if it * starts from a clean mount. */ if (fs->fs_clean) { DIP_SET(ip, i_modrev, fs->fs_mtime); ip->i_flags |= IN_MODIFIED; ffs_update(vp, 1); } out: vput(vp); return (error); } static void journal_unmount(ump) struct ufsmount *ump; { if (ump->softdep_jblocks) jblocks_destroy(ump->softdep_jblocks); ump->softdep_jblocks = NULL; } /* * Called when a journal record is ready to be written. Space is allocated * and the journal entry is created when the journal is flushed to stable * store. */ static void add_to_journal(wk) struct worklist *wk; { struct ufsmount *ump; ump = VFSTOUFS(wk->wk_mp); LOCK_OWNED(ump); if (wk->wk_state & ONWORKLIST) panic("add_to_journal: %s(0x%X) already on list", TYPENAME(wk->wk_type), wk->wk_state); wk->wk_state |= ONWORKLIST | DEPCOMPLETE; if (LIST_EMPTY(&ump->softdep_journal_pending)) { ump->softdep_jblocks->jb_age = ticks; LIST_INSERT_HEAD(&ump->softdep_journal_pending, wk, wk_list); } else LIST_INSERT_AFTER(ump->softdep_journal_tail, wk, wk_list); ump->softdep_journal_tail = wk; ump->softdep_on_journal += 1; } /* * Remove an arbitrary item for the journal worklist maintain the tail * pointer. This happens when a new operation obviates the need to * journal an old operation. */ static void remove_from_journal(wk) struct worklist *wk; { struct ufsmount *ump; ump = VFSTOUFS(wk->wk_mp); LOCK_OWNED(ump); #ifdef INVARIANTS { struct worklist *wkn; LIST_FOREACH(wkn, &ump->softdep_journal_pending, wk_list) if (wkn == wk) break; if (wkn == NULL) panic("remove_from_journal: %p is not in journal", wk); } #endif /* * We emulate a TAILQ to save space in most structures which do not * require TAILQ semantics. Here we must update the tail position * when removing the tail which is not the final entry. This works * only if the worklist linkage are at the beginning of the structure. */ if (ump->softdep_journal_tail == wk) ump->softdep_journal_tail = (struct worklist *)wk->wk_list.le_prev; WORKLIST_REMOVE(wk); ump->softdep_on_journal -= 1; } /* * Check for journal space as well as dependency limits so the prelink * code can throttle both journaled and non-journaled filesystems. * Threshold is 0 for low and 1 for min. */ static int journal_space(ump, thresh) struct ufsmount *ump; int thresh; { struct jblocks *jblocks; int limit, avail; jblocks = ump->softdep_jblocks; if (jblocks == NULL) return (1); /* * We use a tighter restriction here to prevent request_cleanup() * running in threads from running into locks we currently hold. * We have to be over the limit and our filesystem has to be * responsible for more than our share of that usage. */ limit = (max_softdeps / 10) * 9; if (dep_current[D_INODEDEP] > limit && ump->softdep_curdeps[D_INODEDEP] > limit / stat_flush_threads) return (0); if (thresh) thresh = jblocks->jb_min; else thresh = jblocks->jb_low; avail = (ump->softdep_on_journal * JREC_SIZE) / DEV_BSIZE; avail = jblocks->jb_free - avail; return (avail > thresh); } static void journal_suspend(ump) struct ufsmount *ump; { struct jblocks *jblocks; struct mount *mp; bool set; mp = UFSTOVFS(ump); if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) return; jblocks = ump->softdep_jblocks; vfs_op_enter(mp); set = false; MNT_ILOCK(mp); if ((mp->mnt_kern_flag & MNTK_SUSPEND) == 0) { stat_journal_min++; mp->mnt_kern_flag |= MNTK_SUSPEND; mp->mnt_susp_owner = ump->softdep_flushtd; set = true; } jblocks->jb_suspended = 1; MNT_IUNLOCK(mp); if (!set) vfs_op_exit(mp); } static int journal_unsuspend(struct ufsmount *ump) { struct jblocks *jblocks; struct mount *mp; mp = UFSTOVFS(ump); jblocks = ump->softdep_jblocks; if (jblocks != NULL && jblocks->jb_suspended && journal_space(ump, jblocks->jb_min)) { jblocks->jb_suspended = 0; FREE_LOCK(ump); mp->mnt_susp_owner = curthread; vfs_write_resume(mp, 0); ACQUIRE_LOCK(ump); return (1); } return (0); } static void journal_check_space(struct ufsmount *ump) { struct mount *mp; LOCK_OWNED(ump); if (journal_space(ump, 0) == 0) { softdep_speedup(ump); mp = UFSTOVFS(ump); FREE_LOCK(ump); VFS_SYNC(mp, MNT_NOWAIT); ffs_sbupdate(ump, MNT_WAIT, 0); ACQUIRE_LOCK(ump); if (journal_space(ump, 1) == 0) journal_suspend(ump); } } /* * Called before any allocation function to be certain that there is * sufficient space in the journal prior to creating any new records. * Since in the case of block allocation we may have multiple locked * buffers at the time of the actual allocation we can not block * when the journal records are created. Doing so would create a deadlock * if any of these buffers needed to be flushed to reclaim space. Instead * we require a sufficiently large amount of available space such that * each thread in the system could have passed this allocation check and * still have sufficient free space. With 20% of a minimum journal size * of 1MB we have 6553 records available. */ int softdep_prealloc(vp, waitok) struct vnode *vp; int waitok; { struct ufsmount *ump; KASSERT(MOUNTEDSOFTDEP(vp->v_mount) != 0, ("softdep_prealloc called on non-softdep filesystem")); /* * Nothing to do if we are not running journaled soft updates. * If we currently hold the snapshot lock, we must avoid * handling other resources that could cause deadlock. Do not * touch quotas vnode since it is typically recursed with * other vnode locks held. */ if (DOINGSUJ(vp) == 0 || IS_SNAPSHOT(VTOI(vp)) || (vp->v_vflag & VV_SYSTEM) != 0) return (0); ump = VFSTOUFS(vp->v_mount); ACQUIRE_LOCK(ump); if (journal_space(ump, 0)) { FREE_LOCK(ump); return (0); } stat_journal_low++; FREE_LOCK(ump); if (waitok == MNT_NOWAIT) return (ENOSPC); /* * Attempt to sync this vnode once to flush any journal * work attached to it. */ if ((curthread->td_pflags & TDP_COWINPROGRESS) == 0) ffs_syncvnode(vp, waitok, 0); ACQUIRE_LOCK(ump); process_removes(vp); process_truncates(vp); journal_check_space(ump); FREE_LOCK(ump); return (0); } /* * Try hard to sync all data and metadata for the vnode, and workitems * flushing which might conflict with the vnode lock. This is a * helper for softdep_prerename(). */ static int softdep_prerename_vnode(ump, vp) struct ufsmount *ump; struct vnode *vp; { int error; ASSERT_VOP_ELOCKED(vp, "prehandle"); if (vp->v_data == NULL) return (0); error = VOP_FSYNC(vp, MNT_WAIT, curthread); if (error != 0) return (error); ACQUIRE_LOCK(ump); process_removes(vp); process_truncates(vp); FREE_LOCK(ump); return (0); } /* * Must be called from VOP_RENAME() after all vnodes are locked. * Ensures that there is enough journal space for rename. It is * sufficiently different from softdep_prelink() by having to handle * four vnodes. */ int softdep_prerename(fdvp, fvp, tdvp, tvp) struct vnode *fdvp; struct vnode *fvp; struct vnode *tdvp; struct vnode *tvp; { struct ufsmount *ump; int error; ump = VFSTOUFS(fdvp->v_mount); if (journal_space(ump, 0)) return (0); VOP_UNLOCK(tdvp); VOP_UNLOCK(fvp); if (tvp != NULL && tvp != tdvp) VOP_UNLOCK(tvp); error = softdep_prerename_vnode(ump, fdvp); VOP_UNLOCK(fdvp); if (error != 0) return (error); VOP_LOCK(fvp, LK_EXCLUSIVE | LK_RETRY); error = softdep_prerename_vnode(ump, fvp); VOP_UNLOCK(fvp); if (error != 0) return (error); if (tdvp != fdvp) { VOP_LOCK(tdvp, LK_EXCLUSIVE | LK_RETRY); error = softdep_prerename_vnode(ump, tdvp); VOP_UNLOCK(tdvp); if (error != 0) return (error); } if (tvp != fvp && tvp != NULL) { VOP_LOCK(tvp, LK_EXCLUSIVE | LK_RETRY); error = softdep_prerename_vnode(ump, tvp); VOP_UNLOCK(tvp); if (error != 0) return (error); } ACQUIRE_LOCK(ump); softdep_speedup(ump); process_worklist_item(UFSTOVFS(ump), 2, LK_NOWAIT); journal_check_space(ump); FREE_LOCK(ump); return (ERELOOKUP); } /* * Before adjusting a link count on a vnode verify that we have sufficient * journal space. If not, process operations that depend on the currently * locked pair of vnodes to try to flush space as the syncer, buf daemon, * and softdep flush threads can not acquire these locks to reclaim space. * * Returns 0 if all owned locks are still valid and were not dropped * in the process, in other case it returns either an error from sync, * or ERELOOKUP if any of the locks were re-acquired. In the later * case, the state of the vnodes cannot be relied upon and our VFS * syscall must be restarted at top level from the lookup. */ int softdep_prelink(dvp, vp, cnp) struct vnode *dvp; struct vnode *vp; struct componentname *cnp; { struct ufsmount *ump; struct nameidata *ndp; ASSERT_VOP_ELOCKED(dvp, "prelink dvp"); if (vp != NULL) ASSERT_VOP_ELOCKED(vp, "prelink vp"); ump = VFSTOUFS(dvp->v_mount); /* * Nothing to do if we have sufficient journal space. We skip * flushing when vp is a snapshot to avoid deadlock where * another thread is trying to update the inodeblock for dvp * and is waiting on snaplk that vp holds. */ if (journal_space(ump, 0) || (vp != NULL && IS_SNAPSHOT(VTOI(vp)))) return (0); /* * Check if the journal space consumption can in theory be * accounted on dvp and vp. If the vnodes metadata was not * changed comparing with the previous round-trip into * softdep_prelink(), as indicated by the seqc generation * recorded in the nameidata, then there is no point in * starting the sync. */ ndp = __containerof(cnp, struct nameidata, ni_cnd); if (!seqc_in_modify(ndp->ni_dvp_seqc) && vn_seqc_consistent(dvp, ndp->ni_dvp_seqc) && (vp == NULL || (!seqc_in_modify(ndp->ni_vp_seqc) && vn_seqc_consistent(vp, ndp->ni_vp_seqc)))) return (0); stat_journal_low++; if (vp != NULL) { VOP_UNLOCK(dvp); ffs_syncvnode(vp, MNT_NOWAIT, 0); vn_lock_pair(dvp, false, vp, true); if (dvp->v_data == NULL) goto out; } if (vp != NULL) VOP_UNLOCK(vp); ffs_syncvnode(dvp, MNT_WAIT, 0); /* Process vp before dvp as it may create .. removes. */ if (vp != NULL) { VOP_UNLOCK(dvp); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); if (vp->v_data == NULL) { vn_lock_pair(dvp, false, vp, true); goto out; } ACQUIRE_LOCK(ump); process_removes(vp); process_truncates(vp); FREE_LOCK(ump); VOP_UNLOCK(vp); vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); if (dvp->v_data == NULL) { vn_lock_pair(dvp, true, vp, false); goto out; } } ACQUIRE_LOCK(ump); process_removes(dvp); process_truncates(dvp); VOP_UNLOCK(dvp); softdep_speedup(ump); process_worklist_item(UFSTOVFS(ump), 2, LK_NOWAIT); journal_check_space(ump); FREE_LOCK(ump); vn_lock_pair(dvp, false, vp, false); out: ndp->ni_dvp_seqc = vn_seqc_read_any(dvp); if (vp != NULL) ndp->ni_vp_seqc = vn_seqc_read_any(vp); return (ERELOOKUP); } static void jseg_write(ump, jseg, data) struct ufsmount *ump; struct jseg *jseg; uint8_t *data; { struct jsegrec *rec; rec = (struct jsegrec *)data; rec->jsr_seq = jseg->js_seq; rec->jsr_oldest = jseg->js_oldseq; rec->jsr_cnt = jseg->js_cnt; rec->jsr_blocks = jseg->js_size / ump->um_devvp->v_bufobj.bo_bsize; rec->jsr_crc = 0; rec->jsr_time = ump->um_fs->fs_mtime; } static inline void inoref_write(inoref, jseg, rec) struct inoref *inoref; struct jseg *jseg; struct jrefrec *rec; { inoref->if_jsegdep->jd_seg = jseg; rec->jr_ino = inoref->if_ino; rec->jr_parent = inoref->if_parent; rec->jr_nlink = inoref->if_nlink; rec->jr_mode = inoref->if_mode; rec->jr_diroff = inoref->if_diroff; } static void jaddref_write(jaddref, jseg, data) struct jaddref *jaddref; struct jseg *jseg; uint8_t *data; { struct jrefrec *rec; rec = (struct jrefrec *)data; rec->jr_op = JOP_ADDREF; inoref_write(&jaddref->ja_ref, jseg, rec); } static void jremref_write(jremref, jseg, data) struct jremref *jremref; struct jseg *jseg; uint8_t *data; { struct jrefrec *rec; rec = (struct jrefrec *)data; rec->jr_op = JOP_REMREF; inoref_write(&jremref->jr_ref, jseg, rec); } static void jmvref_write(jmvref, jseg, data) struct jmvref *jmvref; struct jseg *jseg; uint8_t *data; { struct jmvrec *rec; rec = (struct jmvrec *)data; rec->jm_op = JOP_MVREF; rec->jm_ino = jmvref->jm_ino; rec->jm_parent = jmvref->jm_parent; rec->jm_oldoff = jmvref->jm_oldoff; rec->jm_newoff = jmvref->jm_newoff; } static void jnewblk_write(jnewblk, jseg, data) struct jnewblk *jnewblk; struct jseg *jseg; uint8_t *data; { struct jblkrec *rec; jnewblk->jn_jsegdep->jd_seg = jseg; rec = (struct jblkrec *)data; rec->jb_op = JOP_NEWBLK; rec->jb_ino = jnewblk->jn_ino; rec->jb_blkno = jnewblk->jn_blkno; rec->jb_lbn = jnewblk->jn_lbn; rec->jb_frags = jnewblk->jn_frags; rec->jb_oldfrags = jnewblk->jn_oldfrags; } static void jfreeblk_write(jfreeblk, jseg, data) struct jfreeblk *jfreeblk; struct jseg *jseg; uint8_t *data; { struct jblkrec *rec; jfreeblk->jf_dep.jb_jsegdep->jd_seg = jseg; rec = (struct jblkrec *)data; rec->jb_op = JOP_FREEBLK; rec->jb_ino = jfreeblk->jf_ino; rec->jb_blkno = jfreeblk->jf_blkno; rec->jb_lbn = jfreeblk->jf_lbn; rec->jb_frags = jfreeblk->jf_frags; rec->jb_oldfrags = 0; } static void jfreefrag_write(jfreefrag, jseg, data) struct jfreefrag *jfreefrag; struct jseg *jseg; uint8_t *data; { struct jblkrec *rec; jfreefrag->fr_jsegdep->jd_seg = jseg; rec = (struct jblkrec *)data; rec->jb_op = JOP_FREEBLK; rec->jb_ino = jfreefrag->fr_ino; rec->jb_blkno = jfreefrag->fr_blkno; rec->jb_lbn = jfreefrag->fr_lbn; rec->jb_frags = jfreefrag->fr_frags; rec->jb_oldfrags = 0; } static void jtrunc_write(jtrunc, jseg, data) struct jtrunc *jtrunc; struct jseg *jseg; uint8_t *data; { struct jtrncrec *rec; jtrunc->jt_dep.jb_jsegdep->jd_seg = jseg; rec = (struct jtrncrec *)data; rec->jt_op = JOP_TRUNC; rec->jt_ino = jtrunc->jt_ino; rec->jt_size = jtrunc->jt_size; rec->jt_extsize = jtrunc->jt_extsize; } static void jfsync_write(jfsync, jseg, data) struct jfsync *jfsync; struct jseg *jseg; uint8_t *data; { struct jtrncrec *rec; rec = (struct jtrncrec *)data; rec->jt_op = JOP_SYNC; rec->jt_ino = jfsync->jfs_ino; rec->jt_size = jfsync->jfs_size; rec->jt_extsize = jfsync->jfs_extsize; } static void softdep_flushjournal(mp) struct mount *mp; { struct jblocks *jblocks; struct ufsmount *ump; if (MOUNTEDSUJ(mp) == 0) return; ump = VFSTOUFS(mp); jblocks = ump->softdep_jblocks; ACQUIRE_LOCK(ump); while (ump->softdep_on_journal) { jblocks->jb_needseg = 1; softdep_process_journal(mp, NULL, MNT_WAIT); } FREE_LOCK(ump); } static void softdep_synchronize_completed(struct bio *); static void softdep_synchronize(struct bio *, struct ufsmount *, void *); static void softdep_synchronize_completed(bp) struct bio *bp; { struct jseg *oldest; struct jseg *jseg; struct ufsmount *ump; /* * caller1 marks the last segment written before we issued the * synchronize cache. */ jseg = bp->bio_caller1; if (jseg == NULL) { g_destroy_bio(bp); return; } ump = VFSTOUFS(jseg->js_list.wk_mp); ACQUIRE_LOCK(ump); oldest = NULL; /* * Mark all the journal entries waiting on the synchronize cache * as completed so they may continue on. */ while (jseg != NULL && (jseg->js_state & COMPLETE) == 0) { jseg->js_state |= COMPLETE; oldest = jseg; jseg = TAILQ_PREV(jseg, jseglst, js_next); } /* * Restart deferred journal entry processing from the oldest * completed jseg. */ if (oldest) complete_jsegs(oldest); FREE_LOCK(ump); g_destroy_bio(bp); } /* * Send BIO_FLUSH/SYNCHRONIZE CACHE to the device to enforce write ordering * barriers. The journal must be written prior to any blocks that depend * on it and the journal can not be released until the blocks have be * written. This code handles both barriers simultaneously. */ static void softdep_synchronize(bp, ump, caller1) struct bio *bp; struct ufsmount *ump; void *caller1; { bp->bio_cmd = BIO_FLUSH; bp->bio_flags |= BIO_ORDERED; bp->bio_data = NULL; bp->bio_offset = ump->um_cp->provider->mediasize; bp->bio_length = 0; bp->bio_done = softdep_synchronize_completed; bp->bio_caller1 = caller1; g_io_request(bp, ump->um_cp); } /* * Flush some journal records to disk. */ static void softdep_process_journal(mp, needwk, flags) struct mount *mp; struct worklist *needwk; int flags; { struct jblocks *jblocks; struct ufsmount *ump; struct worklist *wk; struct jseg *jseg; struct buf *bp; struct bio *bio; uint8_t *data; struct fs *fs; int shouldflush; int segwritten; int jrecmin; /* Minimum records per block. */ int jrecmax; /* Maximum records per block. */ int size; int cnt; int off; int devbsize; ump = VFSTOUFS(mp); if (ump->um_softdep == NULL || ump->um_softdep->sd_jblocks == NULL) return; shouldflush = softdep_flushcache; bio = NULL; jseg = NULL; LOCK_OWNED(ump); fs = ump->um_fs; jblocks = ump->softdep_jblocks; devbsize = ump->um_devvp->v_bufobj.bo_bsize; /* * We write anywhere between a disk block and fs block. The upper * bound is picked to prevent buffer cache fragmentation and limit * processing time per I/O. */ jrecmin = (devbsize / JREC_SIZE) - 1; /* -1 for seg header */ jrecmax = (fs->fs_bsize / devbsize) * jrecmin; segwritten = 0; for (;;) { cnt = ump->softdep_on_journal; /* * Criteria for writing a segment: * 1) We have a full block. * 2) We're called from jwait() and haven't found the * journal item yet. * 3) Always write if needseg is set. * 4) If we are called from process_worklist and have * not yet written anything we write a partial block * to enforce a 1 second maximum latency on journal * entries. */ if (cnt < (jrecmax - 1) && needwk == NULL && jblocks->jb_needseg == 0 && (segwritten || cnt == 0)) break; cnt++; /* * Verify some free journal space. softdep_prealloc() should * guarantee that we don't run out so this is indicative of * a problem with the flow control. Try to recover * gracefully in any event. */ while (jblocks->jb_free == 0) { if (flags != MNT_WAIT) break; printf("softdep: Out of journal space!\n"); softdep_speedup(ump); msleep(jblocks, LOCK_PTR(ump), PRIBIO, "jblocks", hz); } FREE_LOCK(ump); jseg = malloc(sizeof(*jseg), M_JSEG, M_SOFTDEP_FLAGS); workitem_alloc(&jseg->js_list, D_JSEG, mp); LIST_INIT(&jseg->js_entries); LIST_INIT(&jseg->js_indirs); jseg->js_state = ATTACHED; if (shouldflush == 0) jseg->js_state |= COMPLETE; else if (bio == NULL) bio = g_alloc_bio(); jseg->js_jblocks = jblocks; bp = geteblk(fs->fs_bsize, 0); ACQUIRE_LOCK(ump); /* * If there was a race while we were allocating the block * and jseg the entry we care about was likely written. * We bail out in both the WAIT and NOWAIT case and assume * the caller will loop if the entry it cares about is * not written. */ cnt = ump->softdep_on_journal; if (cnt + jblocks->jb_needseg == 0 || jblocks->jb_free == 0) { bp->b_flags |= B_INVAL | B_NOCACHE; WORKITEM_FREE(jseg, D_JSEG); FREE_LOCK(ump); brelse(bp); ACQUIRE_LOCK(ump); break; } /* * Calculate the disk block size required for the available * records rounded to the min size. */ if (cnt == 0) size = devbsize; else if (cnt < jrecmax) size = howmany(cnt, jrecmin) * devbsize; else size = fs->fs_bsize; /* * Allocate a disk block for this journal data and account * for truncation of the requested size if enough contiguous * space was not available. */ bp->b_blkno = jblocks_alloc(jblocks, size, &size); bp->b_lblkno = bp->b_blkno; bp->b_offset = bp->b_blkno * DEV_BSIZE; bp->b_bcount = size; bp->b_flags &= ~B_INVAL; bp->b_flags |= B_VALIDSUSPWRT | B_NOCOPY; /* * Initialize our jseg with cnt records. Assign the next * sequence number to it and link it in-order. */ cnt = MIN(cnt, (size / devbsize) * jrecmin); jseg->js_buf = bp; jseg->js_cnt = cnt; jseg->js_refs = cnt + 1; /* Self ref. */ jseg->js_size = size; jseg->js_seq = jblocks->jb_nextseq++; if (jblocks->jb_oldestseg == NULL) jblocks->jb_oldestseg = jseg; jseg->js_oldseq = jblocks->jb_oldestseg->js_seq; TAILQ_INSERT_TAIL(&jblocks->jb_segs, jseg, js_next); if (jblocks->jb_writeseg == NULL) jblocks->jb_writeseg = jseg; /* * Start filling in records from the pending list. */ data = bp->b_data; off = 0; /* * Always put a header on the first block. * XXX As with below, there might not be a chance to get * into the loop. Ensure that something valid is written. */ jseg_write(ump, jseg, data); off += JREC_SIZE; data = bp->b_data + off; /* * XXX Something is wrong here. There's no work to do, * but we need to perform and I/O and allow it to complete * anyways. */ if (LIST_EMPTY(&ump->softdep_journal_pending)) stat_emptyjblocks++; while ((wk = LIST_FIRST(&ump->softdep_journal_pending)) != NULL) { if (cnt == 0) break; /* Place a segment header on every device block. */ if ((off % devbsize) == 0) { jseg_write(ump, jseg, data); off += JREC_SIZE; data = bp->b_data + off; } if (wk == needwk) needwk = NULL; remove_from_journal(wk); wk->wk_state |= INPROGRESS; WORKLIST_INSERT(&jseg->js_entries, wk); switch (wk->wk_type) { case D_JADDREF: jaddref_write(WK_JADDREF(wk), jseg, data); break; case D_JREMREF: jremref_write(WK_JREMREF(wk), jseg, data); break; case D_JMVREF: jmvref_write(WK_JMVREF(wk), jseg, data); break; case D_JNEWBLK: jnewblk_write(WK_JNEWBLK(wk), jseg, data); break; case D_JFREEBLK: jfreeblk_write(WK_JFREEBLK(wk), jseg, data); break; case D_JFREEFRAG: jfreefrag_write(WK_JFREEFRAG(wk), jseg, data); break; case D_JTRUNC: jtrunc_write(WK_JTRUNC(wk), jseg, data); break; case D_JFSYNC: jfsync_write(WK_JFSYNC(wk), jseg, data); break; default: panic("process_journal: Unknown type %s", TYPENAME(wk->wk_type)); /* NOTREACHED */ } off += JREC_SIZE; data = bp->b_data + off; cnt--; } /* Clear any remaining space so we don't leak kernel data */ if (size > off) bzero(data, size - off); /* * Write this one buffer and continue. */ segwritten = 1; jblocks->jb_needseg = 0; WORKLIST_INSERT(&bp->b_dep, &jseg->js_list); FREE_LOCK(ump); bp->b_xflags |= BX_CVTENXIO; pbgetvp(ump->um_devvp, bp); /* * We only do the blocking wait once we find the journal * entry we're looking for. */ if (needwk == NULL && flags == MNT_WAIT) bwrite(bp); else bawrite(bp); ACQUIRE_LOCK(ump); } /* * If we wrote a segment issue a synchronize cache so the journal * is reflected on disk before the data is written. Since reclaiming * journal space also requires writing a journal record this * process also enforces a barrier before reclamation. */ if (segwritten && shouldflush) { softdep_synchronize(bio, ump, TAILQ_LAST(&jblocks->jb_segs, jseglst)); } else if (bio) g_destroy_bio(bio); /* * If we've suspended the filesystem because we ran out of journal * space either try to sync it here to make some progress or * unsuspend it if we already have. */ if (flags == 0 && jblocks->jb_suspended) { if (journal_unsuspend(ump)) return; FREE_LOCK(ump); VFS_SYNC(mp, MNT_NOWAIT); ffs_sbupdate(ump, MNT_WAIT, 0); ACQUIRE_LOCK(ump); } } /* * Complete a jseg, allowing all dependencies awaiting journal writes * to proceed. Each journal dependency also attaches a jsegdep to dependent * structures so that the journal segment can be freed to reclaim space. */ static void complete_jseg(jseg) struct jseg *jseg; { struct worklist *wk; struct jmvref *jmvref; #ifdef INVARIANTS int i = 0; #endif while ((wk = LIST_FIRST(&jseg->js_entries)) != NULL) { WORKLIST_REMOVE(wk); wk->wk_state &= ~INPROGRESS; wk->wk_state |= COMPLETE; KASSERT(i++ < jseg->js_cnt, ("handle_written_jseg: overflow %d >= %d", i - 1, jseg->js_cnt)); switch (wk->wk_type) { case D_JADDREF: handle_written_jaddref(WK_JADDREF(wk)); break; case D_JREMREF: handle_written_jremref(WK_JREMREF(wk)); break; case D_JMVREF: rele_jseg(jseg); /* No jsegdep. */ jmvref = WK_JMVREF(wk); LIST_REMOVE(jmvref, jm_deps); if ((jmvref->jm_pagedep->pd_state & ONWORKLIST) == 0) free_pagedep(jmvref->jm_pagedep); WORKITEM_FREE(jmvref, D_JMVREF); break; case D_JNEWBLK: handle_written_jnewblk(WK_JNEWBLK(wk)); break; case D_JFREEBLK: handle_written_jblkdep(&WK_JFREEBLK(wk)->jf_dep); break; case D_JTRUNC: handle_written_jblkdep(&WK_JTRUNC(wk)->jt_dep); break; case D_JFSYNC: rele_jseg(jseg); /* No jsegdep. */ WORKITEM_FREE(wk, D_JFSYNC); break; case D_JFREEFRAG: handle_written_jfreefrag(WK_JFREEFRAG(wk)); break; default: panic("handle_written_jseg: Unknown type %s", TYPENAME(wk->wk_type)); /* NOTREACHED */ } } /* Release the self reference so the structure may be freed. */ rele_jseg(jseg); } /* * Determine which jsegs are ready for completion processing. Waits for * synchronize cache to complete as well as forcing in-order completion * of journal entries. */ static void complete_jsegs(jseg) struct jseg *jseg; { struct jblocks *jblocks; struct jseg *jsegn; jblocks = jseg->js_jblocks; /* * Don't allow out of order completions. If this isn't the first * block wait for it to write before we're done. */ if (jseg != jblocks->jb_writeseg) return; /* Iterate through available jsegs processing their entries. */ while (jseg && (jseg->js_state & ALLCOMPLETE) == ALLCOMPLETE) { jblocks->jb_oldestwrseq = jseg->js_oldseq; jsegn = TAILQ_NEXT(jseg, js_next); complete_jseg(jseg); jseg = jsegn; } jblocks->jb_writeseg = jseg; /* * Attempt to free jsegs now that oldestwrseq may have advanced. */ free_jsegs(jblocks); } /* * Mark a jseg as DEPCOMPLETE and throw away the buffer. Attempt to handle * the final completions. */ static void handle_written_jseg(jseg, bp) struct jseg *jseg; struct buf *bp; { if (jseg->js_refs == 0) panic("handle_written_jseg: No self-reference on %p", jseg); jseg->js_state |= DEPCOMPLETE; /* * We'll never need this buffer again, set flags so it will be * discarded. */ bp->b_flags |= B_INVAL | B_NOCACHE; pbrelvp(bp); complete_jsegs(jseg); } static inline struct jsegdep * inoref_jseg(inoref) struct inoref *inoref; { struct jsegdep *jsegdep; jsegdep = inoref->if_jsegdep; inoref->if_jsegdep = NULL; return (jsegdep); } /* * Called once a jremref has made it to stable store. The jremref is marked * complete and we attempt to free it. Any pagedeps writes sleeping waiting * for the jremref to complete will be awoken by free_jremref. */ static void handle_written_jremref(jremref) struct jremref *jremref; { struct inodedep *inodedep; struct jsegdep *jsegdep; struct dirrem *dirrem; /* Grab the jsegdep. */ jsegdep = inoref_jseg(&jremref->jr_ref); /* * Remove us from the inoref list. */ if (inodedep_lookup(jremref->jr_list.wk_mp, jremref->jr_ref.if_ino, 0, &inodedep) == 0) panic("handle_written_jremref: Lost inodedep"); TAILQ_REMOVE(&inodedep->id_inoreflst, &jremref->jr_ref, if_deps); /* * Complete the dirrem. */ dirrem = jremref->jr_dirrem; jremref->jr_dirrem = NULL; LIST_REMOVE(jremref, jr_deps); jsegdep->jd_state |= jremref->jr_state & MKDIR_PARENT; jwork_insert(&dirrem->dm_jwork, jsegdep); if (LIST_EMPTY(&dirrem->dm_jremrefhd) && (dirrem->dm_state & COMPLETE) != 0) add_to_worklist(&dirrem->dm_list, 0); free_jremref(jremref); } /* * Called once a jaddref has made it to stable store. The dependency is * marked complete and any dependent structures are added to the inode * bufwait list to be completed as soon as it is written. If a bitmap write * depends on this entry we move the inode into the inodedephd of the * bmsafemap dependency and attempt to remove the jaddref from the bmsafemap. */ static void handle_written_jaddref(jaddref) struct jaddref *jaddref; { struct jsegdep *jsegdep; struct inodedep *inodedep; struct diradd *diradd; struct mkdir *mkdir; /* Grab the jsegdep. */ jsegdep = inoref_jseg(&jaddref->ja_ref); mkdir = NULL; diradd = NULL; if (inodedep_lookup(jaddref->ja_list.wk_mp, jaddref->ja_ino, 0, &inodedep) == 0) panic("handle_written_jaddref: Lost inodedep."); if (jaddref->ja_diradd == NULL) panic("handle_written_jaddref: No dependency"); if (jaddref->ja_diradd->da_list.wk_type == D_DIRADD) { diradd = jaddref->ja_diradd; WORKLIST_INSERT(&inodedep->id_bufwait, &diradd->da_list); } else if (jaddref->ja_state & MKDIR_PARENT) { mkdir = jaddref->ja_mkdir; WORKLIST_INSERT(&inodedep->id_bufwait, &mkdir->md_list); } else if (jaddref->ja_state & MKDIR_BODY) mkdir = jaddref->ja_mkdir; else panic("handle_written_jaddref: Unknown dependency %p", jaddref->ja_diradd); jaddref->ja_diradd = NULL; /* also clears ja_mkdir */ /* * Remove us from the inode list. */ TAILQ_REMOVE(&inodedep->id_inoreflst, &jaddref->ja_ref, if_deps); /* * The mkdir may be waiting on the jaddref to clear before freeing. */ if (mkdir) { KASSERT(mkdir->md_list.wk_type == D_MKDIR, ("handle_written_jaddref: Incorrect type for mkdir %s", TYPENAME(mkdir->md_list.wk_type))); mkdir->md_jaddref = NULL; diradd = mkdir->md_diradd; mkdir->md_state |= DEPCOMPLETE; complete_mkdir(mkdir); } jwork_insert(&diradd->da_jwork, jsegdep); if (jaddref->ja_state & NEWBLOCK) { inodedep->id_state |= ONDEPLIST; LIST_INSERT_HEAD(&inodedep->id_bmsafemap->sm_inodedephd, inodedep, id_deps); } free_jaddref(jaddref); } /* * Called once a jnewblk journal is written. The allocdirect or allocindir * is placed in the bmsafemap to await notification of a written bitmap. If * the operation was canceled we add the segdep to the appropriate * dependency to free the journal space once the canceling operation * completes. */ static void handle_written_jnewblk(jnewblk) struct jnewblk *jnewblk; { struct bmsafemap *bmsafemap; struct freefrag *freefrag; struct freework *freework; struct jsegdep *jsegdep; struct newblk *newblk; /* Grab the jsegdep. */ jsegdep = jnewblk->jn_jsegdep; jnewblk->jn_jsegdep = NULL; if (jnewblk->jn_dep == NULL) panic("handle_written_jnewblk: No dependency for the segdep."); switch (jnewblk->jn_dep->wk_type) { case D_NEWBLK: case D_ALLOCDIRECT: case D_ALLOCINDIR: /* * Add the written block to the bmsafemap so it can * be notified when the bitmap is on disk. */ newblk = WK_NEWBLK(jnewblk->jn_dep); newblk->nb_jnewblk = NULL; if ((newblk->nb_state & GOINGAWAY) == 0) { bmsafemap = newblk->nb_bmsafemap; newblk->nb_state |= ONDEPLIST; LIST_INSERT_HEAD(&bmsafemap->sm_newblkhd, newblk, nb_deps); } jwork_insert(&newblk->nb_jwork, jsegdep); break; case D_FREEFRAG: /* * A newblock being removed by a freefrag when replaced by * frag extension. */ freefrag = WK_FREEFRAG(jnewblk->jn_dep); freefrag->ff_jdep = NULL; jwork_insert(&freefrag->ff_jwork, jsegdep); break; case D_FREEWORK: /* * A direct block was removed by truncate. */ freework = WK_FREEWORK(jnewblk->jn_dep); freework->fw_jnewblk = NULL; jwork_insert(&freework->fw_freeblks->fb_jwork, jsegdep); break; default: panic("handle_written_jnewblk: Unknown type %d.", jnewblk->jn_dep->wk_type); } jnewblk->jn_dep = NULL; free_jnewblk(jnewblk); } /* * Cancel a jfreefrag that won't be needed, probably due to colliding with * an in-flight allocation that has not yet been committed. Divorce us * from the freefrag and mark it DEPCOMPLETE so that it may be added * to the worklist. */ static void cancel_jfreefrag(jfreefrag) struct jfreefrag *jfreefrag; { struct freefrag *freefrag; if (jfreefrag->fr_jsegdep) { free_jsegdep(jfreefrag->fr_jsegdep); jfreefrag->fr_jsegdep = NULL; } freefrag = jfreefrag->fr_freefrag; jfreefrag->fr_freefrag = NULL; free_jfreefrag(jfreefrag); freefrag->ff_state |= DEPCOMPLETE; CTR1(KTR_SUJ, "cancel_jfreefrag: blkno %jd", freefrag->ff_blkno); } /* * Free a jfreefrag when the parent freefrag is rendered obsolete. */ static void free_jfreefrag(jfreefrag) struct jfreefrag *jfreefrag; { if (jfreefrag->fr_state & INPROGRESS) WORKLIST_REMOVE(&jfreefrag->fr_list); else if (jfreefrag->fr_state & ONWORKLIST) remove_from_journal(&jfreefrag->fr_list); if (jfreefrag->fr_freefrag != NULL) panic("free_jfreefrag: Still attached to a freefrag."); WORKITEM_FREE(jfreefrag, D_JFREEFRAG); } /* * Called when the journal write for a jfreefrag completes. The parent * freefrag is added to the worklist if this completes its dependencies. */ static void handle_written_jfreefrag(jfreefrag) struct jfreefrag *jfreefrag; { struct jsegdep *jsegdep; struct freefrag *freefrag; /* Grab the jsegdep. */ jsegdep = jfreefrag->fr_jsegdep; jfreefrag->fr_jsegdep = NULL; freefrag = jfreefrag->fr_freefrag; if (freefrag == NULL) panic("handle_written_jfreefrag: No freefrag."); freefrag->ff_state |= DEPCOMPLETE; freefrag->ff_jdep = NULL; jwork_insert(&freefrag->ff_jwork, jsegdep); if ((freefrag->ff_state & ALLCOMPLETE) == ALLCOMPLETE) add_to_worklist(&freefrag->ff_list, 0); jfreefrag->fr_freefrag = NULL; free_jfreefrag(jfreefrag); } /* * Called when the journal write for a jfreeblk completes. The jfreeblk * is removed from the freeblks list of pending journal writes and the * jsegdep is moved to the freeblks jwork to be completed when all blocks * have been reclaimed. */ static void handle_written_jblkdep(jblkdep) struct jblkdep *jblkdep; { struct freeblks *freeblks; struct jsegdep *jsegdep; /* Grab the jsegdep. */ jsegdep = jblkdep->jb_jsegdep; jblkdep->jb_jsegdep = NULL; freeblks = jblkdep->jb_freeblks; LIST_REMOVE(jblkdep, jb_deps); jwork_insert(&freeblks->fb_jwork, jsegdep); /* * If the freeblks is all journaled, we can add it to the worklist. */ if (LIST_EMPTY(&freeblks->fb_jblkdephd) && (freeblks->fb_state & ALLCOMPLETE) == ALLCOMPLETE) add_to_worklist(&freeblks->fb_list, WK_NODELAY); free_jblkdep(jblkdep); } static struct jsegdep * newjsegdep(struct worklist *wk) { struct jsegdep *jsegdep; jsegdep = malloc(sizeof(*jsegdep), M_JSEGDEP, M_SOFTDEP_FLAGS); workitem_alloc(&jsegdep->jd_list, D_JSEGDEP, wk->wk_mp); jsegdep->jd_seg = NULL; return (jsegdep); } static struct jmvref * newjmvref(dp, ino, oldoff, newoff) struct inode *dp; ino_t ino; off_t oldoff; off_t newoff; { struct jmvref *jmvref; jmvref = malloc(sizeof(*jmvref), M_JMVREF, M_SOFTDEP_FLAGS); workitem_alloc(&jmvref->jm_list, D_JMVREF, ITOVFS(dp)); jmvref->jm_list.wk_state = ATTACHED | DEPCOMPLETE; jmvref->jm_parent = dp->i_number; jmvref->jm_ino = ino; jmvref->jm_oldoff = oldoff; jmvref->jm_newoff = newoff; return (jmvref); } /* * Allocate a new jremref that tracks the removal of ip from dp with the * directory entry offset of diroff. Mark the entry as ATTACHED and * DEPCOMPLETE as we have all the information required for the journal write * and the directory has already been removed from the buffer. The caller * is responsible for linking the jremref into the pagedep and adding it * to the journal to write. The MKDIR_PARENT flag is set if we're doing * a DOTDOT addition so handle_workitem_remove() can properly assign * the jsegdep when we're done. */ static struct jremref * newjremref(struct dirrem *dirrem, struct inode *dp, struct inode *ip, off_t diroff, nlink_t nlink) { struct jremref *jremref; jremref = malloc(sizeof(*jremref), M_JREMREF, M_SOFTDEP_FLAGS); workitem_alloc(&jremref->jr_list, D_JREMREF, ITOVFS(dp)); jremref->jr_state = ATTACHED; newinoref(&jremref->jr_ref, ip->i_number, dp->i_number, diroff, nlink, ip->i_mode); jremref->jr_dirrem = dirrem; return (jremref); } static inline void newinoref(struct inoref *inoref, ino_t ino, ino_t parent, off_t diroff, nlink_t nlink, uint16_t mode) { inoref->if_jsegdep = newjsegdep(&inoref->if_list); inoref->if_diroff = diroff; inoref->if_ino = ino; inoref->if_parent = parent; inoref->if_nlink = nlink; inoref->if_mode = mode; } /* * Allocate a new jaddref to track the addition of ino to dp at diroff. The * directory offset may not be known until later. The caller is responsible * adding the entry to the journal when this information is available. nlink * should be the link count prior to the addition and mode is only required * to have the correct FMT. */ static struct jaddref * newjaddref(struct inode *dp, ino_t ino, off_t diroff, int16_t nlink, uint16_t mode) { struct jaddref *jaddref; jaddref = malloc(sizeof(*jaddref), M_JADDREF, M_SOFTDEP_FLAGS); workitem_alloc(&jaddref->ja_list, D_JADDREF, ITOVFS(dp)); jaddref->ja_state = ATTACHED; jaddref->ja_mkdir = NULL; newinoref(&jaddref->ja_ref, ino, dp->i_number, diroff, nlink, mode); return (jaddref); } /* * Create a new free dependency for a freework. The caller is responsible * for adjusting the reference count when it has the lock held. The freedep * will track an outstanding bitmap write that will ultimately clear the * freework to continue. */ static struct freedep * newfreedep(struct freework *freework) { struct freedep *freedep; freedep = malloc(sizeof(*freedep), M_FREEDEP, M_SOFTDEP_FLAGS); workitem_alloc(&freedep->fd_list, D_FREEDEP, freework->fw_list.wk_mp); freedep->fd_freework = freework; return (freedep); } /* * Free a freedep structure once the buffer it is linked to is written. If * this is the last reference to the freework schedule it for completion. */ static void free_freedep(freedep) struct freedep *freedep; { struct freework *freework; freework = freedep->fd_freework; freework->fw_freeblks->fb_cgwait--; if (--freework->fw_ref == 0) freework_enqueue(freework); WORKITEM_FREE(freedep, D_FREEDEP); } /* * Allocate a new freework structure that may be a level in an indirect * when parent is not NULL or a top level block when it is. The top level * freework structures are allocated without the per-filesystem lock held * and before the freeblks is visible outside of softdep_setup_freeblocks(). */ static struct freework * newfreework(ump, freeblks, parent, lbn, nb, frags, off, journal) struct ufsmount *ump; struct freeblks *freeblks; struct freework *parent; ufs_lbn_t lbn; ufs2_daddr_t nb; int frags; int off; int journal; { struct freework *freework; freework = malloc(sizeof(*freework), M_FREEWORK, M_SOFTDEP_FLAGS); workitem_alloc(&freework->fw_list, D_FREEWORK, freeblks->fb_list.wk_mp); freework->fw_state = ATTACHED; freework->fw_jnewblk = NULL; freework->fw_freeblks = freeblks; freework->fw_parent = parent; freework->fw_lbn = lbn; freework->fw_blkno = nb; freework->fw_frags = frags; freework->fw_indir = NULL; freework->fw_ref = (MOUNTEDSUJ(UFSTOVFS(ump)) == 0 || lbn >= -UFS_NXADDR) ? 0 : NINDIR(ump->um_fs) + 1; freework->fw_start = freework->fw_off = off; if (journal) newjfreeblk(freeblks, lbn, nb, frags); if (parent == NULL) { ACQUIRE_LOCK(ump); WORKLIST_INSERT(&freeblks->fb_freeworkhd, &freework->fw_list); freeblks->fb_ref++; FREE_LOCK(ump); } return (freework); } /* * Eliminate a jfreeblk for a block that does not need journaling. */ static void cancel_jfreeblk(freeblks, blkno) struct freeblks *freeblks; ufs2_daddr_t blkno; { struct jfreeblk *jfreeblk; struct jblkdep *jblkdep; LIST_FOREACH(jblkdep, &freeblks->fb_jblkdephd, jb_deps) { if (jblkdep->jb_list.wk_type != D_JFREEBLK) continue; jfreeblk = WK_JFREEBLK(&jblkdep->jb_list); if (jfreeblk->jf_blkno == blkno) break; } if (jblkdep == NULL) return; CTR1(KTR_SUJ, "cancel_jfreeblk: blkno %jd", blkno); free_jsegdep(jblkdep->jb_jsegdep); LIST_REMOVE(jblkdep, jb_deps); WORKITEM_FREE(jfreeblk, D_JFREEBLK); } /* * Allocate a new jfreeblk to journal top level block pointer when truncating * a file. The caller must add this to the worklist when the per-filesystem * lock is held. */ static struct jfreeblk * newjfreeblk(freeblks, lbn, blkno, frags) struct freeblks *freeblks; ufs_lbn_t lbn; ufs2_daddr_t blkno; int frags; { struct jfreeblk *jfreeblk; jfreeblk = malloc(sizeof(*jfreeblk), M_JFREEBLK, M_SOFTDEP_FLAGS); workitem_alloc(&jfreeblk->jf_dep.jb_list, D_JFREEBLK, freeblks->fb_list.wk_mp); jfreeblk->jf_dep.jb_jsegdep = newjsegdep(&jfreeblk->jf_dep.jb_list); jfreeblk->jf_dep.jb_freeblks = freeblks; jfreeblk->jf_ino = freeblks->fb_inum; jfreeblk->jf_lbn = lbn; jfreeblk->jf_blkno = blkno; jfreeblk->jf_frags = frags; LIST_INSERT_HEAD(&freeblks->fb_jblkdephd, &jfreeblk->jf_dep, jb_deps); return (jfreeblk); } /* * The journal is only prepared to handle full-size block numbers, so we * have to adjust the record to reflect the change to a full-size block. * For example, suppose we have a block made up of fragments 8-15 and * want to free its last two fragments. We are given a request that says: * FREEBLK ino=5, blkno=14, lbn=0, frags=2, oldfrags=0 * where frags are the number of fragments to free and oldfrags are the * number of fragments to keep. To block align it, we have to change it to * have a valid full-size blkno, so it becomes: * FREEBLK ino=5, blkno=8, lbn=0, frags=2, oldfrags=6 */ static void adjust_newfreework(freeblks, frag_offset) struct freeblks *freeblks; int frag_offset; { struct jfreeblk *jfreeblk; KASSERT((LIST_FIRST(&freeblks->fb_jblkdephd) != NULL && LIST_FIRST(&freeblks->fb_jblkdephd)->jb_list.wk_type == D_JFREEBLK), ("adjust_newfreework: Missing freeblks dependency")); jfreeblk = WK_JFREEBLK(LIST_FIRST(&freeblks->fb_jblkdephd)); jfreeblk->jf_blkno -= frag_offset; jfreeblk->jf_frags += frag_offset; } /* * Allocate a new jtrunc to track a partial truncation. */ static struct jtrunc * newjtrunc(freeblks, size, extsize) struct freeblks *freeblks; off_t size; int extsize; { struct jtrunc *jtrunc; jtrunc = malloc(sizeof(*jtrunc), M_JTRUNC, M_SOFTDEP_FLAGS); workitem_alloc(&jtrunc->jt_dep.jb_list, D_JTRUNC, freeblks->fb_list.wk_mp); jtrunc->jt_dep.jb_jsegdep = newjsegdep(&jtrunc->jt_dep.jb_list); jtrunc->jt_dep.jb_freeblks = freeblks; jtrunc->jt_ino = freeblks->fb_inum; jtrunc->jt_size = size; jtrunc->jt_extsize = extsize; LIST_INSERT_HEAD(&freeblks->fb_jblkdephd, &jtrunc->jt_dep, jb_deps); return (jtrunc); } /* * If we're canceling a new bitmap we have to search for another ref * to move into the bmsafemap dep. This might be better expressed * with another structure. */ static void move_newblock_dep(jaddref, inodedep) struct jaddref *jaddref; struct inodedep *inodedep; { struct inoref *inoref; struct jaddref *jaddrefn; jaddrefn = NULL; for (inoref = TAILQ_NEXT(&jaddref->ja_ref, if_deps); inoref; inoref = TAILQ_NEXT(inoref, if_deps)) { if ((jaddref->ja_state & NEWBLOCK) && inoref->if_list.wk_type == D_JADDREF) { jaddrefn = (struct jaddref *)inoref; break; } } if (jaddrefn == NULL) return; jaddrefn->ja_state &= ~(ATTACHED | UNDONE); jaddrefn->ja_state |= jaddref->ja_state & (ATTACHED | UNDONE | NEWBLOCK); jaddref->ja_state &= ~(ATTACHED | UNDONE | NEWBLOCK); jaddref->ja_state |= ATTACHED; LIST_REMOVE(jaddref, ja_bmdeps); LIST_INSERT_HEAD(&inodedep->id_bmsafemap->sm_jaddrefhd, jaddrefn, ja_bmdeps); } /* * Cancel a jaddref either before it has been written or while it is being * written. This happens when a link is removed before the add reaches * the disk. The jaddref dependency is kept linked into the bmsafemap * and inode to prevent the link count or bitmap from reaching the disk * until handle_workitem_remove() re-adjusts the counts and bitmaps as * required. * * Returns 1 if the canceled addref requires journaling of the remove and * 0 otherwise. */ static int cancel_jaddref(jaddref, inodedep, wkhd) struct jaddref *jaddref; struct inodedep *inodedep; struct workhead *wkhd; { struct inoref *inoref; struct jsegdep *jsegdep; int needsj; KASSERT((jaddref->ja_state & COMPLETE) == 0, ("cancel_jaddref: Canceling complete jaddref")); if (jaddref->ja_state & (INPROGRESS | COMPLETE)) needsj = 1; else needsj = 0; if (inodedep == NULL) if (inodedep_lookup(jaddref->ja_list.wk_mp, jaddref->ja_ino, 0, &inodedep) == 0) panic("cancel_jaddref: Lost inodedep"); /* * We must adjust the nlink of any reference operation that follows * us so that it is consistent with the in-memory reference. This * ensures that inode nlink rollbacks always have the correct link. */ if (needsj == 0) { for (inoref = TAILQ_NEXT(&jaddref->ja_ref, if_deps); inoref; inoref = TAILQ_NEXT(inoref, if_deps)) { if (inoref->if_state & GOINGAWAY) break; inoref->if_nlink--; } } jsegdep = inoref_jseg(&jaddref->ja_ref); if (jaddref->ja_state & NEWBLOCK) move_newblock_dep(jaddref, inodedep); wake_worklist(&jaddref->ja_list); jaddref->ja_mkdir = NULL; if (jaddref->ja_state & INPROGRESS) { jaddref->ja_state &= ~INPROGRESS; WORKLIST_REMOVE(&jaddref->ja_list); jwork_insert(wkhd, jsegdep); } else { free_jsegdep(jsegdep); if (jaddref->ja_state & DEPCOMPLETE) remove_from_journal(&jaddref->ja_list); } jaddref->ja_state |= (GOINGAWAY | DEPCOMPLETE); /* * Leave NEWBLOCK jaddrefs on the inodedep so handle_workitem_remove * can arrange for them to be freed with the bitmap. Otherwise we * no longer need this addref attached to the inoreflst and it * will incorrectly adjust nlink if we leave it. */ if ((jaddref->ja_state & NEWBLOCK) == 0) { TAILQ_REMOVE(&inodedep->id_inoreflst, &jaddref->ja_ref, if_deps); jaddref->ja_state |= COMPLETE; free_jaddref(jaddref); return (needsj); } /* * Leave the head of the list for jsegdeps for fast merging. */ if (LIST_FIRST(wkhd) != NULL) { jaddref->ja_state |= ONWORKLIST; LIST_INSERT_AFTER(LIST_FIRST(wkhd), &jaddref->ja_list, wk_list); } else WORKLIST_INSERT(wkhd, &jaddref->ja_list); return (needsj); } /* * Attempt to free a jaddref structure when some work completes. This * should only succeed once the entry is written and all dependencies have * been notified. */ static void free_jaddref(jaddref) struct jaddref *jaddref; { if ((jaddref->ja_state & ALLCOMPLETE) != ALLCOMPLETE) return; if (jaddref->ja_ref.if_jsegdep) panic("free_jaddref: segdep attached to jaddref %p(0x%X)\n", jaddref, jaddref->ja_state); if (jaddref->ja_state & NEWBLOCK) LIST_REMOVE(jaddref, ja_bmdeps); if (jaddref->ja_state & (INPROGRESS | ONWORKLIST)) panic("free_jaddref: Bad state %p(0x%X)", jaddref, jaddref->ja_state); if (jaddref->ja_mkdir != NULL) panic("free_jaddref: Work pending, 0x%X\n", jaddref->ja_state); WORKITEM_FREE(jaddref, D_JADDREF); } /* * Free a jremref structure once it has been written or discarded. */ static void free_jremref(jremref) struct jremref *jremref; { if (jremref->jr_ref.if_jsegdep) free_jsegdep(jremref->jr_ref.if_jsegdep); if (jremref->jr_state & INPROGRESS) panic("free_jremref: IO still pending"); WORKITEM_FREE(jremref, D_JREMREF); } /* * Free a jnewblk structure. */ static void free_jnewblk(jnewblk) struct jnewblk *jnewblk; { if ((jnewblk->jn_state & ALLCOMPLETE) != ALLCOMPLETE) return; LIST_REMOVE(jnewblk, jn_deps); if (jnewblk->jn_dep != NULL) panic("free_jnewblk: Dependency still attached."); WORKITEM_FREE(jnewblk, D_JNEWBLK); } /* * Cancel a jnewblk which has been been made redundant by frag extension. */ static void cancel_jnewblk(jnewblk, wkhd) struct jnewblk *jnewblk; struct workhead *wkhd; { struct jsegdep *jsegdep; CTR1(KTR_SUJ, "cancel_jnewblk: blkno %jd", jnewblk->jn_blkno); jsegdep = jnewblk->jn_jsegdep; if (jnewblk->jn_jsegdep == NULL || jnewblk->jn_dep == NULL) panic("cancel_jnewblk: Invalid state"); jnewblk->jn_jsegdep = NULL; jnewblk->jn_dep = NULL; jnewblk->jn_state |= GOINGAWAY; if (jnewblk->jn_state & INPROGRESS) { jnewblk->jn_state &= ~INPROGRESS; WORKLIST_REMOVE(&jnewblk->jn_list); jwork_insert(wkhd, jsegdep); } else { free_jsegdep(jsegdep); remove_from_journal(&jnewblk->jn_list); } wake_worklist(&jnewblk->jn_list); WORKLIST_INSERT(wkhd, &jnewblk->jn_list); } static void free_jblkdep(jblkdep) struct jblkdep *jblkdep; { if (jblkdep->jb_list.wk_type == D_JFREEBLK) WORKITEM_FREE(jblkdep, D_JFREEBLK); else if (jblkdep->jb_list.wk_type == D_JTRUNC) WORKITEM_FREE(jblkdep, D_JTRUNC); else panic("free_jblkdep: Unexpected type %s", TYPENAME(jblkdep->jb_list.wk_type)); } /* * Free a single jseg once it is no longer referenced in memory or on * disk. Reclaim journal blocks and dependencies waiting for the segment * to disappear. */ static void free_jseg(jseg, jblocks) struct jseg *jseg; struct jblocks *jblocks; { struct freework *freework; /* * Free freework structures that were lingering to indicate freed * indirect blocks that forced journal write ordering on reallocate. */ while ((freework = LIST_FIRST(&jseg->js_indirs)) != NULL) indirblk_remove(freework); if (jblocks->jb_oldestseg == jseg) jblocks->jb_oldestseg = TAILQ_NEXT(jseg, js_next); TAILQ_REMOVE(&jblocks->jb_segs, jseg, js_next); jblocks_free(jblocks, jseg->js_list.wk_mp, jseg->js_size); KASSERT(LIST_EMPTY(&jseg->js_entries), ("free_jseg: Freed jseg has valid entries.")); WORKITEM_FREE(jseg, D_JSEG); } /* * Free all jsegs that meet the criteria for being reclaimed and update * oldestseg. */ static void free_jsegs(jblocks) struct jblocks *jblocks; { struct jseg *jseg; /* * Free only those jsegs which have none allocated before them to * preserve the journal space ordering. */ while ((jseg = TAILQ_FIRST(&jblocks->jb_segs)) != NULL) { /* * Only reclaim space when nothing depends on this journal * set and another set has written that it is no longer * valid. */ if (jseg->js_refs != 0) { jblocks->jb_oldestseg = jseg; return; } if ((jseg->js_state & ALLCOMPLETE) != ALLCOMPLETE) break; if (jseg->js_seq > jblocks->jb_oldestwrseq) break; /* * We can free jsegs that didn't write entries when * oldestwrseq == js_seq. */ if (jseg->js_seq == jblocks->jb_oldestwrseq && jseg->js_cnt != 0) break; free_jseg(jseg, jblocks); } /* * If we exited the loop above we still must discover the * oldest valid segment. */ if (jseg) for (jseg = jblocks->jb_oldestseg; jseg != NULL; jseg = TAILQ_NEXT(jseg, js_next)) if (jseg->js_refs != 0) break; jblocks->jb_oldestseg = jseg; /* * The journal has no valid records but some jsegs may still be * waiting on oldestwrseq to advance. We force a small record * out to permit these lingering records to be reclaimed. */ if (jblocks->jb_oldestseg == NULL && !TAILQ_EMPTY(&jblocks->jb_segs)) jblocks->jb_needseg = 1; } /* * Release one reference to a jseg and free it if the count reaches 0. This * should eventually reclaim journal space as well. */ static void rele_jseg(jseg) struct jseg *jseg; { KASSERT(jseg->js_refs > 0, ("free_jseg: Invalid refcnt %d", jseg->js_refs)); if (--jseg->js_refs != 0) return; free_jsegs(jseg->js_jblocks); } /* * Release a jsegdep and decrement the jseg count. */ static void free_jsegdep(jsegdep) struct jsegdep *jsegdep; { if (jsegdep->jd_seg) rele_jseg(jsegdep->jd_seg); WORKITEM_FREE(jsegdep, D_JSEGDEP); } /* * Wait for a journal item to make it to disk. Initiate journal processing * if required. */ static int jwait(wk, waitfor) struct worklist *wk; int waitfor; { LOCK_OWNED(VFSTOUFS(wk->wk_mp)); /* * Blocking journal waits cause slow synchronous behavior. Record * stats on the frequency of these blocking operations. */ if (waitfor == MNT_WAIT) { stat_journal_wait++; switch (wk->wk_type) { case D_JREMREF: case D_JMVREF: stat_jwait_filepage++; break; case D_JTRUNC: case D_JFREEBLK: stat_jwait_freeblks++; break; case D_JNEWBLK: stat_jwait_newblk++; break; case D_JADDREF: stat_jwait_inode++; break; default: break; } } /* * If IO has not started we process the journal. We can't mark the * worklist item as IOWAITING because we drop the lock while * processing the journal and the worklist entry may be freed after * this point. The caller may call back in and re-issue the request. */ if ((wk->wk_state & INPROGRESS) == 0) { softdep_process_journal(wk->wk_mp, wk, waitfor); if (waitfor != MNT_WAIT) return (EBUSY); return (0); } if (waitfor != MNT_WAIT) return (EBUSY); wait_worklist(wk, "jwait"); return (0); } /* * Lookup an inodedep based on an inode pointer and set the nlinkdelta as * appropriate. This is a convenience function to reduce duplicate code * for the setup and revert functions below. */ static struct inodedep * inodedep_lookup_ip(ip) struct inode *ip; { struct inodedep *inodedep; KASSERT(ip->i_nlink >= ip->i_effnlink, ("inodedep_lookup_ip: bad delta")); (void) inodedep_lookup(ITOVFS(ip), ip->i_number, DEPALLOC, &inodedep); inodedep->id_nlinkdelta = ip->i_nlink - ip->i_effnlink; KASSERT((inodedep->id_state & UNLINKED) == 0, ("inode unlinked")); return (inodedep); } /* * Called prior to creating a new inode and linking it to a directory. The * jaddref structure must already be allocated by softdep_setup_inomapdep * and it is discovered here so we can initialize the mode and update * nlinkdelta. */ void softdep_setup_create(dp, ip) struct inode *dp; struct inode *ip; { struct inodedep *inodedep; struct jaddref *jaddref; struct vnode *dvp; KASSERT(MOUNTEDSOFTDEP(ITOVFS(dp)) != 0, ("softdep_setup_create called on non-softdep filesystem")); KASSERT(ip->i_nlink == 1, ("softdep_setup_create: Invalid link count.")); dvp = ITOV(dp); ACQUIRE_LOCK(ITOUMP(dp)); inodedep = inodedep_lookup_ip(ip); if (DOINGSUJ(dvp)) { jaddref = (struct jaddref *)TAILQ_LAST(&inodedep->id_inoreflst, inoreflst); KASSERT(jaddref != NULL && jaddref->ja_parent == dp->i_number, ("softdep_setup_create: No addref structure present.")); } FREE_LOCK(ITOUMP(dp)); } /* * Create a jaddref structure to track the addition of a DOTDOT link when * we are reparenting an inode as part of a rename. This jaddref will be * found by softdep_setup_directory_change. Adjusts nlinkdelta for * non-journaling softdep. */ void softdep_setup_dotdot_link(dp, ip) struct inode *dp; struct inode *ip; { struct inodedep *inodedep; struct jaddref *jaddref; struct vnode *dvp; KASSERT(MOUNTEDSOFTDEP(ITOVFS(dp)) != 0, ("softdep_setup_dotdot_link called on non-softdep filesystem")); dvp = ITOV(dp); jaddref = NULL; /* * We don't set MKDIR_PARENT as this is not tied to a mkdir and * is used as a normal link would be. */ if (DOINGSUJ(dvp)) jaddref = newjaddref(ip, dp->i_number, DOTDOT_OFFSET, dp->i_effnlink - 1, dp->i_mode); ACQUIRE_LOCK(ITOUMP(dp)); inodedep = inodedep_lookup_ip(dp); if (jaddref) TAILQ_INSERT_TAIL(&inodedep->id_inoreflst, &jaddref->ja_ref, if_deps); FREE_LOCK(ITOUMP(dp)); } /* * Create a jaddref structure to track a new link to an inode. The directory * offset is not known until softdep_setup_directory_add or * softdep_setup_directory_change. Adjusts nlinkdelta for non-journaling * softdep. */ void softdep_setup_link(dp, ip) struct inode *dp; struct inode *ip; { struct inodedep *inodedep; struct jaddref *jaddref; struct vnode *dvp; KASSERT(MOUNTEDSOFTDEP(ITOVFS(dp)) != 0, ("softdep_setup_link called on non-softdep filesystem")); dvp = ITOV(dp); jaddref = NULL; if (DOINGSUJ(dvp)) jaddref = newjaddref(dp, ip->i_number, 0, ip->i_effnlink - 1, ip->i_mode); ACQUIRE_LOCK(ITOUMP(dp)); inodedep = inodedep_lookup_ip(ip); if (jaddref) TAILQ_INSERT_TAIL(&inodedep->id_inoreflst, &jaddref->ja_ref, if_deps); FREE_LOCK(ITOUMP(dp)); } /* * Called to create the jaddref structures to track . and .. references as * well as lookup and further initialize the incomplete jaddref created * by softdep_setup_inomapdep when the inode was allocated. Adjusts * nlinkdelta for non-journaling softdep. */ void softdep_setup_mkdir(dp, ip) struct inode *dp; struct inode *ip; { struct inodedep *inodedep; struct jaddref *dotdotaddref; struct jaddref *dotaddref; struct jaddref *jaddref; struct vnode *dvp; KASSERT(MOUNTEDSOFTDEP(ITOVFS(dp)) != 0, ("softdep_setup_mkdir called on non-softdep filesystem")); dvp = ITOV(dp); dotaddref = dotdotaddref = NULL; if (DOINGSUJ(dvp)) { dotaddref = newjaddref(ip, ip->i_number, DOT_OFFSET, 1, ip->i_mode); dotaddref->ja_state |= MKDIR_BODY; dotdotaddref = newjaddref(ip, dp->i_number, DOTDOT_OFFSET, dp->i_effnlink - 1, dp->i_mode); dotdotaddref->ja_state |= MKDIR_PARENT; } ACQUIRE_LOCK(ITOUMP(dp)); inodedep = inodedep_lookup_ip(ip); if (DOINGSUJ(dvp)) { jaddref = (struct jaddref *)TAILQ_LAST(&inodedep->id_inoreflst, inoreflst); KASSERT(jaddref != NULL, ("softdep_setup_mkdir: No addref structure present.")); KASSERT(jaddref->ja_parent == dp->i_number, ("softdep_setup_mkdir: bad parent %ju", (uintmax_t)jaddref->ja_parent)); TAILQ_INSERT_BEFORE(&jaddref->ja_ref, &dotaddref->ja_ref, if_deps); } inodedep = inodedep_lookup_ip(dp); if (DOINGSUJ(dvp)) TAILQ_INSERT_TAIL(&inodedep->id_inoreflst, &dotdotaddref->ja_ref, if_deps); FREE_LOCK(ITOUMP(dp)); } /* * Called to track nlinkdelta of the inode and parent directories prior to * unlinking a directory. */ void softdep_setup_rmdir(dp, ip) struct inode *dp; struct inode *ip; { struct vnode *dvp; KASSERT(MOUNTEDSOFTDEP(ITOVFS(dp)) != 0, ("softdep_setup_rmdir called on non-softdep filesystem")); dvp = ITOV(dp); ACQUIRE_LOCK(ITOUMP(dp)); (void) inodedep_lookup_ip(ip); (void) inodedep_lookup_ip(dp); FREE_LOCK(ITOUMP(dp)); } /* * Called to track nlinkdelta of the inode and parent directories prior to * unlink. */ void softdep_setup_unlink(dp, ip) struct inode *dp; struct inode *ip; { struct vnode *dvp; KASSERT(MOUNTEDSOFTDEP(ITOVFS(dp)) != 0, ("softdep_setup_unlink called on non-softdep filesystem")); dvp = ITOV(dp); ACQUIRE_LOCK(ITOUMP(dp)); (void) inodedep_lookup_ip(ip); (void) inodedep_lookup_ip(dp); FREE_LOCK(ITOUMP(dp)); } /* * Called to release the journal structures created by a failed non-directory * creation. Adjusts nlinkdelta for non-journaling softdep. */ void softdep_revert_create(dp, ip) struct inode *dp; struct inode *ip; { struct inodedep *inodedep; struct jaddref *jaddref; struct vnode *dvp; KASSERT(MOUNTEDSOFTDEP(ITOVFS((dp))) != 0, ("softdep_revert_create called on non-softdep filesystem")); dvp = ITOV(dp); ACQUIRE_LOCK(ITOUMP(dp)); inodedep = inodedep_lookup_ip(ip); if (DOINGSUJ(dvp)) { jaddref = (struct jaddref *)TAILQ_LAST(&inodedep->id_inoreflst, inoreflst); KASSERT(jaddref->ja_parent == dp->i_number, ("softdep_revert_create: addref parent mismatch")); cancel_jaddref(jaddref, inodedep, &inodedep->id_inowait); } FREE_LOCK(ITOUMP(dp)); } /* * Called to release the journal structures created by a failed link * addition. Adjusts nlinkdelta for non-journaling softdep. */ void softdep_revert_link(dp, ip) struct inode *dp; struct inode *ip; { struct inodedep *inodedep; struct jaddref *jaddref; struct vnode *dvp; KASSERT(MOUNTEDSOFTDEP(ITOVFS(dp)) != 0, ("softdep_revert_link called on non-softdep filesystem")); dvp = ITOV(dp); ACQUIRE_LOCK(ITOUMP(dp)); inodedep = inodedep_lookup_ip(ip); if (DOINGSUJ(dvp)) { jaddref = (struct jaddref *)TAILQ_LAST(&inodedep->id_inoreflst, inoreflst); KASSERT(jaddref->ja_parent == dp->i_number, ("softdep_revert_link: addref parent mismatch")); cancel_jaddref(jaddref, inodedep, &inodedep->id_inowait); } FREE_LOCK(ITOUMP(dp)); } /* * Called to release the journal structures created by a failed mkdir * attempt. Adjusts nlinkdelta for non-journaling softdep. */ void softdep_revert_mkdir(dp, ip) struct inode *dp; struct inode *ip; { struct inodedep *inodedep; struct jaddref *jaddref; struct jaddref *dotaddref; struct vnode *dvp; KASSERT(MOUNTEDSOFTDEP(ITOVFS(dp)) != 0, ("softdep_revert_mkdir called on non-softdep filesystem")); dvp = ITOV(dp); ACQUIRE_LOCK(ITOUMP(dp)); inodedep = inodedep_lookup_ip(dp); if (DOINGSUJ(dvp)) { jaddref = (struct jaddref *)TAILQ_LAST(&inodedep->id_inoreflst, inoreflst); KASSERT(jaddref->ja_parent == ip->i_number, ("softdep_revert_mkdir: dotdot addref parent mismatch")); cancel_jaddref(jaddref, inodedep, &inodedep->id_inowait); } inodedep = inodedep_lookup_ip(ip); if (DOINGSUJ(dvp)) { jaddref = (struct jaddref *)TAILQ_LAST(&inodedep->id_inoreflst, inoreflst); KASSERT(jaddref->ja_parent == dp->i_number, ("softdep_revert_mkdir: addref parent mismatch")); dotaddref = (struct jaddref *)TAILQ_PREV(&jaddref->ja_ref, inoreflst, if_deps); cancel_jaddref(jaddref, inodedep, &inodedep->id_inowait); KASSERT(dotaddref->ja_parent == ip->i_number, ("softdep_revert_mkdir: dot addref parent mismatch")); cancel_jaddref(dotaddref, inodedep, &inodedep->id_inowait); } FREE_LOCK(ITOUMP(dp)); } /* * Called to correct nlinkdelta after a failed rmdir. */ void softdep_revert_rmdir(dp, ip) struct inode *dp; struct inode *ip; { KASSERT(MOUNTEDSOFTDEP(ITOVFS(dp)) != 0, ("softdep_revert_rmdir called on non-softdep filesystem")); ACQUIRE_LOCK(ITOUMP(dp)); (void) inodedep_lookup_ip(ip); (void) inodedep_lookup_ip(dp); FREE_LOCK(ITOUMP(dp)); } /* * Protecting the freemaps (or bitmaps). * * To eliminate the need to execute fsck before mounting a filesystem * after a power failure, one must (conservatively) guarantee that the * on-disk copy of the bitmaps never indicate that a live inode or block is * free. So, when a block or inode is allocated, the bitmap should be * updated (on disk) before any new pointers. When a block or inode is * freed, the bitmap should not be updated until all pointers have been * reset. The latter dependency is handled by the delayed de-allocation * approach described below for block and inode de-allocation. The former * dependency is handled by calling the following procedure when a block or * inode is allocated. When an inode is allocated an "inodedep" is created * with its DEPCOMPLETE flag cleared until its bitmap is written to disk. * Each "inodedep" is also inserted into the hash indexing structure so * that any additional link additions can be made dependent on the inode * allocation. * * The ufs filesystem maintains a number of free block counts (e.g., per * cylinder group, per cylinder and per pair) * in addition to the bitmaps. These counts are used to improve efficiency * during allocation and therefore must be consistent with the bitmaps. * There is no convenient way to guarantee post-crash consistency of these * counts with simple update ordering, for two main reasons: (1) The counts * and bitmaps for a single cylinder group block are not in the same disk * sector. If a disk write is interrupted (e.g., by power failure), one may * be written and the other not. (2) Some of the counts are located in the * superblock rather than the cylinder group block. So, we focus our soft * updates implementation on protecting the bitmaps. When mounting a * filesystem, we recompute the auxiliary counts from the bitmaps. */ /* * Called just after updating the cylinder group block to allocate an inode. */ void softdep_setup_inomapdep(bp, ip, newinum, mode) struct buf *bp; /* buffer for cylgroup block with inode map */ struct inode *ip; /* inode related to allocation */ ino_t newinum; /* new inode number being allocated */ int mode; { struct inodedep *inodedep; struct bmsafemap *bmsafemap; struct jaddref *jaddref; struct mount *mp; struct fs *fs; mp = ITOVFS(ip); KASSERT(MOUNTEDSOFTDEP(mp) != 0, ("softdep_setup_inomapdep called on non-softdep filesystem")); fs = VFSTOUFS(mp)->um_fs; jaddref = NULL; /* * Allocate the journal reference add structure so that the bitmap * can be dependent on it. */ if (MOUNTEDSUJ(mp)) { jaddref = newjaddref(ip, newinum, 0, 0, mode); jaddref->ja_state |= NEWBLOCK; } /* * Create a dependency for the newly allocated inode. * Panic if it already exists as something is seriously wrong. * Otherwise add it to the dependency list for the buffer holding * the cylinder group map from which it was allocated. * * We have to preallocate a bmsafemap entry in case it is needed * in bmsafemap_lookup since once we allocate the inodedep, we * have to finish initializing it before we can FREE_LOCK(). * By preallocating, we avoid FREE_LOCK() while doing a malloc * in bmsafemap_lookup. We cannot call bmsafemap_lookup before * creating the inodedep as it can be freed during the time * that we FREE_LOCK() while allocating the inodedep. We must * call workitem_alloc() before entering the locked section as * it also acquires the lock and we must avoid trying doing so * recursively. */ bmsafemap = malloc(sizeof(struct bmsafemap), M_BMSAFEMAP, M_SOFTDEP_FLAGS); workitem_alloc(&bmsafemap->sm_list, D_BMSAFEMAP, mp); ACQUIRE_LOCK(ITOUMP(ip)); if ((inodedep_lookup(mp, newinum, DEPALLOC, &inodedep))) panic("softdep_setup_inomapdep: dependency %p for new" "inode already exists", inodedep); bmsafemap = bmsafemap_lookup(mp, bp, ino_to_cg(fs, newinum), bmsafemap); if (jaddref) { LIST_INSERT_HEAD(&bmsafemap->sm_jaddrefhd, jaddref, ja_bmdeps); TAILQ_INSERT_TAIL(&inodedep->id_inoreflst, &jaddref->ja_ref, if_deps); } else { inodedep->id_state |= ONDEPLIST; LIST_INSERT_HEAD(&bmsafemap->sm_inodedephd, inodedep, id_deps); } inodedep->id_bmsafemap = bmsafemap; inodedep->id_state &= ~DEPCOMPLETE; FREE_LOCK(ITOUMP(ip)); } /* * Called just after updating the cylinder group block to * allocate block or fragment. */ void softdep_setup_blkmapdep(bp, mp, newblkno, frags, oldfrags) struct buf *bp; /* buffer for cylgroup block with block map */ struct mount *mp; /* filesystem doing allocation */ ufs2_daddr_t newblkno; /* number of newly allocated block */ int frags; /* Number of fragments. */ int oldfrags; /* Previous number of fragments for extend. */ { struct newblk *newblk; struct bmsafemap *bmsafemap; struct jnewblk *jnewblk; struct ufsmount *ump; struct fs *fs; KASSERT(MOUNTEDSOFTDEP(mp) != 0, ("softdep_setup_blkmapdep called on non-softdep filesystem")); ump = VFSTOUFS(mp); fs = ump->um_fs; jnewblk = NULL; /* * Create a dependency for the newly allocated block. * Add it to the dependency list for the buffer holding * the cylinder group map from which it was allocated. */ if (MOUNTEDSUJ(mp)) { jnewblk = malloc(sizeof(*jnewblk), M_JNEWBLK, M_SOFTDEP_FLAGS); workitem_alloc(&jnewblk->jn_list, D_JNEWBLK, mp); jnewblk->jn_jsegdep = newjsegdep(&jnewblk->jn_list); jnewblk->jn_state = ATTACHED; jnewblk->jn_blkno = newblkno; jnewblk->jn_frags = frags; jnewblk->jn_oldfrags = oldfrags; #ifdef INVARIANTS { struct cg *cgp; uint8_t *blksfree; long bno; int i; cgp = (struct cg *)bp->b_data; blksfree = cg_blksfree(cgp); bno = dtogd(fs, jnewblk->jn_blkno); for (i = jnewblk->jn_oldfrags; i < jnewblk->jn_frags; i++) { if (isset(blksfree, bno + i)) panic("softdep_setup_blkmapdep: " "free fragment %d from %d-%d " "state 0x%X dep %p", i, jnewblk->jn_oldfrags, jnewblk->jn_frags, jnewblk->jn_state, jnewblk->jn_dep); } } #endif } CTR3(KTR_SUJ, "softdep_setup_blkmapdep: blkno %jd frags %d oldfrags %d", newblkno, frags, oldfrags); ACQUIRE_LOCK(ump); if (newblk_lookup(mp, newblkno, DEPALLOC, &newblk) != 0) panic("softdep_setup_blkmapdep: found block"); newblk->nb_bmsafemap = bmsafemap = bmsafemap_lookup(mp, bp, dtog(fs, newblkno), NULL); if (jnewblk) { jnewblk->jn_dep = (struct worklist *)newblk; LIST_INSERT_HEAD(&bmsafemap->sm_jnewblkhd, jnewblk, jn_deps); } else { newblk->nb_state |= ONDEPLIST; LIST_INSERT_HEAD(&bmsafemap->sm_newblkhd, newblk, nb_deps); } newblk->nb_bmsafemap = bmsafemap; newblk->nb_jnewblk = jnewblk; FREE_LOCK(ump); } #define BMSAFEMAP_HASH(ump, cg) \ (&(ump)->bmsafemap_hashtbl[(cg) & (ump)->bmsafemap_hash_size]) static int bmsafemap_find(bmsafemaphd, cg, bmsafemapp) struct bmsafemap_hashhead *bmsafemaphd; int cg; struct bmsafemap **bmsafemapp; { struct bmsafemap *bmsafemap; LIST_FOREACH(bmsafemap, bmsafemaphd, sm_hash) if (bmsafemap->sm_cg == cg) break; if (bmsafemap) { *bmsafemapp = bmsafemap; return (1); } *bmsafemapp = NULL; return (0); } /* * Find the bmsafemap associated with a cylinder group buffer. * If none exists, create one. The buffer must be locked when * this routine is called and this routine must be called with * the softdep lock held. To avoid giving up the lock while * allocating a new bmsafemap, a preallocated bmsafemap may be * provided. If it is provided but not needed, it is freed. */ static struct bmsafemap * bmsafemap_lookup(mp, bp, cg, newbmsafemap) struct mount *mp; struct buf *bp; int cg; struct bmsafemap *newbmsafemap; { struct bmsafemap_hashhead *bmsafemaphd; struct bmsafemap *bmsafemap, *collision; struct worklist *wk; struct ufsmount *ump; ump = VFSTOUFS(mp); LOCK_OWNED(ump); KASSERT(bp != NULL, ("bmsafemap_lookup: missing buffer")); LIST_FOREACH(wk, &bp->b_dep, wk_list) { if (wk->wk_type == D_BMSAFEMAP) { if (newbmsafemap) WORKITEM_FREE(newbmsafemap, D_BMSAFEMAP); return (WK_BMSAFEMAP(wk)); } } bmsafemaphd = BMSAFEMAP_HASH(ump, cg); if (bmsafemap_find(bmsafemaphd, cg, &bmsafemap) == 1) { if (newbmsafemap) WORKITEM_FREE(newbmsafemap, D_BMSAFEMAP); return (bmsafemap); } if (newbmsafemap) { bmsafemap = newbmsafemap; } else { FREE_LOCK(ump); bmsafemap = malloc(sizeof(struct bmsafemap), M_BMSAFEMAP, M_SOFTDEP_FLAGS); workitem_alloc(&bmsafemap->sm_list, D_BMSAFEMAP, mp); ACQUIRE_LOCK(ump); } bmsafemap->sm_buf = bp; LIST_INIT(&bmsafemap->sm_inodedephd); LIST_INIT(&bmsafemap->sm_inodedepwr); LIST_INIT(&bmsafemap->sm_newblkhd); LIST_INIT(&bmsafemap->sm_newblkwr); LIST_INIT(&bmsafemap->sm_jaddrefhd); LIST_INIT(&bmsafemap->sm_jnewblkhd); LIST_INIT(&bmsafemap->sm_freehd); LIST_INIT(&bmsafemap->sm_freewr); if (bmsafemap_find(bmsafemaphd, cg, &collision) == 1) { WORKITEM_FREE(bmsafemap, D_BMSAFEMAP); return (collision); } bmsafemap->sm_cg = cg; LIST_INSERT_HEAD(bmsafemaphd, bmsafemap, sm_hash); LIST_INSERT_HEAD(&ump->softdep_dirtycg, bmsafemap, sm_next); WORKLIST_INSERT(&bp->b_dep, &bmsafemap->sm_list); return (bmsafemap); } /* * Direct block allocation dependencies. * * When a new block is allocated, the corresponding disk locations must be * initialized (with zeros or new data) before the on-disk inode points to * them. Also, the freemap from which the block was allocated must be * updated (on disk) before the inode's pointer. These two dependencies are * independent of each other and are needed for all file blocks and indirect * blocks that are pointed to directly by the inode. Just before the * "in-core" version of the inode is updated with a newly allocated block * number, a procedure (below) is called to setup allocation dependency * structures. These structures are removed when the corresponding * dependencies are satisfied or when the block allocation becomes obsolete * (i.e., the file is deleted, the block is de-allocated, or the block is a * fragment that gets upgraded). All of these cases are handled in * procedures described later. * * When a file extension causes a fragment to be upgraded, either to a larger * fragment or to a full block, the on-disk location may change (if the * previous fragment could not simply be extended). In this case, the old * fragment must be de-allocated, but not until after the inode's pointer has * been updated. In most cases, this is handled by later procedures, which * will construct a "freefrag" structure to be added to the workitem queue * when the inode update is complete (or obsolete). The main exception to * this is when an allocation occurs while a pending allocation dependency * (for the same block pointer) remains. This case is handled in the main * allocation dependency setup procedure by immediately freeing the * unreferenced fragments. */ void softdep_setup_allocdirect(ip, off, newblkno, oldblkno, newsize, oldsize, bp) struct inode *ip; /* inode to which block is being added */ ufs_lbn_t off; /* block pointer within inode */ ufs2_daddr_t newblkno; /* disk block number being added */ ufs2_daddr_t oldblkno; /* previous block number, 0 unless frag */ long newsize; /* size of new block */ long oldsize; /* size of new block */ struct buf *bp; /* bp for allocated block */ { struct allocdirect *adp, *oldadp; struct allocdirectlst *adphead; struct freefrag *freefrag; struct inodedep *inodedep; struct pagedep *pagedep; struct jnewblk *jnewblk; struct newblk *newblk; struct mount *mp; ufs_lbn_t lbn; lbn = bp->b_lblkno; mp = ITOVFS(ip); KASSERT(MOUNTEDSOFTDEP(mp) != 0, ("softdep_setup_allocdirect called on non-softdep filesystem")); if (oldblkno && oldblkno != newblkno) /* * The usual case is that a smaller fragment that * was just allocated has been replaced with a bigger * fragment or a full-size block. If it is marked as * B_DELWRI, the current contents have not been written * to disk. It is possible that the block was written * earlier, but very uncommon. If the block has never * been written, there is no need to send a BIO_DELETE * for it when it is freed. The gain from avoiding the * TRIMs for the common case of unwritten blocks far * exceeds the cost of the write amplification for the * uncommon case of failing to send a TRIM for a block * that had been written. */ freefrag = newfreefrag(ip, oldblkno, oldsize, lbn, (bp->b_flags & B_DELWRI) != 0 ? NOTRIM_KEY : SINGLETON_KEY); else freefrag = NULL; CTR6(KTR_SUJ, "softdep_setup_allocdirect: ino %d blkno %jd oldblkno %jd " "off %jd newsize %ld oldsize %d", ip->i_number, newblkno, oldblkno, off, newsize, oldsize); ACQUIRE_LOCK(ITOUMP(ip)); if (off >= UFS_NDADDR) { if (lbn > 0) panic("softdep_setup_allocdirect: bad lbn %jd, off %jd", lbn, off); /* allocating an indirect block */ if (oldblkno != 0) panic("softdep_setup_allocdirect: non-zero indir"); } else { if (off != lbn) panic("softdep_setup_allocdirect: lbn %jd != off %jd", lbn, off); /* * Allocating a direct block. * * If we are allocating a directory block, then we must * allocate an associated pagedep to track additions and * deletions. */ if ((ip->i_mode & IFMT) == IFDIR) pagedep_lookup(mp, bp, ip->i_number, off, DEPALLOC, &pagedep); } if (newblk_lookup(mp, newblkno, 0, &newblk) == 0) panic("softdep_setup_allocdirect: lost block"); KASSERT(newblk->nb_list.wk_type == D_NEWBLK, ("softdep_setup_allocdirect: newblk already initialized")); /* * Convert the newblk to an allocdirect. */ WORKITEM_REASSIGN(newblk, D_ALLOCDIRECT); adp = (struct allocdirect *)newblk; newblk->nb_freefrag = freefrag; adp->ad_offset = off; adp->ad_oldblkno = oldblkno; adp->ad_newsize = newsize; adp->ad_oldsize = oldsize; /* * Finish initializing the journal. */ if ((jnewblk = newblk->nb_jnewblk) != NULL) { jnewblk->jn_ino = ip->i_number; jnewblk->jn_lbn = lbn; add_to_journal(&jnewblk->jn_list); } if (freefrag && freefrag->ff_jdep != NULL && freefrag->ff_jdep->wk_type == D_JFREEFRAG) add_to_journal(freefrag->ff_jdep); inodedep_lookup(mp, ip->i_number, DEPALLOC, &inodedep); adp->ad_inodedep = inodedep; WORKLIST_INSERT(&bp->b_dep, &newblk->nb_list); /* * The list of allocdirects must be kept in sorted and ascending * order so that the rollback routines can quickly determine the * first uncommitted block (the size of the file stored on disk * ends at the end of the lowest committed fragment, or if there * are no fragments, at the end of the highest committed block). * Since files generally grow, the typical case is that the new * block is to be added at the end of the list. We speed this * special case by checking against the last allocdirect in the * list before laboriously traversing the list looking for the * insertion point. */ adphead = &inodedep->id_newinoupdt; oldadp = TAILQ_LAST(adphead, allocdirectlst); if (oldadp == NULL || oldadp->ad_offset <= off) { /* insert at end of list */ TAILQ_INSERT_TAIL(adphead, adp, ad_next); if (oldadp != NULL && oldadp->ad_offset == off) allocdirect_merge(adphead, adp, oldadp); FREE_LOCK(ITOUMP(ip)); return; } TAILQ_FOREACH(oldadp, adphead, ad_next) { if (oldadp->ad_offset >= off) break; } if (oldadp == NULL) panic("softdep_setup_allocdirect: lost entry"); /* insert in middle of list */ TAILQ_INSERT_BEFORE(oldadp, adp, ad_next); if (oldadp->ad_offset == off) allocdirect_merge(adphead, adp, oldadp); FREE_LOCK(ITOUMP(ip)); } /* * Merge a newer and older journal record to be stored either in a * newblock or freefrag. This handles aggregating journal records for * fragment allocation into a second record as well as replacing a * journal free with an aborted journal allocation. A segment for the * oldest record will be placed on wkhd if it has been written. If not * the segment for the newer record will suffice. */ static struct worklist * jnewblk_merge(new, old, wkhd) struct worklist *new; struct worklist *old; struct workhead *wkhd; { struct jnewblk *njnewblk; struct jnewblk *jnewblk; /* Handle NULLs to simplify callers. */ if (new == NULL) return (old); if (old == NULL) return (new); /* Replace a jfreefrag with a jnewblk. */ if (new->wk_type == D_JFREEFRAG) { if (WK_JNEWBLK(old)->jn_blkno != WK_JFREEFRAG(new)->fr_blkno) panic("jnewblk_merge: blkno mismatch: %p, %p", old, new); cancel_jfreefrag(WK_JFREEFRAG(new)); return (old); } if (old->wk_type != D_JNEWBLK || new->wk_type != D_JNEWBLK) panic("jnewblk_merge: Bad type: old %d new %d\n", old->wk_type, new->wk_type); /* * Handle merging of two jnewblk records that describe * different sets of fragments in the same block. */ jnewblk = WK_JNEWBLK(old); njnewblk = WK_JNEWBLK(new); if (jnewblk->jn_blkno != njnewblk->jn_blkno) panic("jnewblk_merge: Merging disparate blocks."); /* * The record may be rolled back in the cg. */ if (jnewblk->jn_state & UNDONE) { jnewblk->jn_state &= ~UNDONE; njnewblk->jn_state |= UNDONE; njnewblk->jn_state &= ~ATTACHED; } /* * We modify the newer addref and free the older so that if neither * has been written the most up-to-date copy will be on disk. If * both have been written but rolled back we only temporarily need * one of them to fix the bits when the cg write completes. */ jnewblk->jn_state |= ATTACHED | COMPLETE; njnewblk->jn_oldfrags = jnewblk->jn_oldfrags; cancel_jnewblk(jnewblk, wkhd); WORKLIST_REMOVE(&jnewblk->jn_list); free_jnewblk(jnewblk); return (new); } /* * Replace an old allocdirect dependency with a newer one. */ static void allocdirect_merge(adphead, newadp, oldadp) struct allocdirectlst *adphead; /* head of list holding allocdirects */ struct allocdirect *newadp; /* allocdirect being added */ struct allocdirect *oldadp; /* existing allocdirect being checked */ { struct worklist *wk; struct freefrag *freefrag; freefrag = NULL; LOCK_OWNED(VFSTOUFS(newadp->ad_list.wk_mp)); if (newadp->ad_oldblkno != oldadp->ad_newblkno || newadp->ad_oldsize != oldadp->ad_newsize || newadp->ad_offset >= UFS_NDADDR) panic("%s %jd != new %jd || old size %ld != new %ld", "allocdirect_merge: old blkno", (intmax_t)newadp->ad_oldblkno, (intmax_t)oldadp->ad_newblkno, newadp->ad_oldsize, oldadp->ad_newsize); newadp->ad_oldblkno = oldadp->ad_oldblkno; newadp->ad_oldsize = oldadp->ad_oldsize; /* * If the old dependency had a fragment to free or had never * previously had a block allocated, then the new dependency * can immediately post its freefrag and adopt the old freefrag. * This action is done by swapping the freefrag dependencies. * The new dependency gains the old one's freefrag, and the * old one gets the new one and then immediately puts it on * the worklist when it is freed by free_newblk. It is * not possible to do this swap when the old dependency had a * non-zero size but no previous fragment to free. This condition * arises when the new block is an extension of the old block. * Here, the first part of the fragment allocated to the new * dependency is part of the block currently claimed on disk by * the old dependency, so cannot legitimately be freed until the * conditions for the new dependency are fulfilled. */ freefrag = newadp->ad_freefrag; if (oldadp->ad_freefrag != NULL || oldadp->ad_oldblkno == 0) { newadp->ad_freefrag = oldadp->ad_freefrag; oldadp->ad_freefrag = freefrag; } /* * If we are tracking a new directory-block allocation, * move it from the old allocdirect to the new allocdirect. */ if ((wk = LIST_FIRST(&oldadp->ad_newdirblk)) != NULL) { WORKLIST_REMOVE(wk); if (!LIST_EMPTY(&oldadp->ad_newdirblk)) panic("allocdirect_merge: extra newdirblk"); WORKLIST_INSERT(&newadp->ad_newdirblk, wk); } TAILQ_REMOVE(adphead, oldadp, ad_next); /* * We need to move any journal dependencies over to the freefrag * that releases this block if it exists. Otherwise we are * extending an existing block and we'll wait until that is * complete to release the journal space and extend the * new journal to cover this old space as well. */ if (freefrag == NULL) { if (oldadp->ad_newblkno != newadp->ad_newblkno) panic("allocdirect_merge: %jd != %jd", oldadp->ad_newblkno, newadp->ad_newblkno); newadp->ad_block.nb_jnewblk = (struct jnewblk *) jnewblk_merge(&newadp->ad_block.nb_jnewblk->jn_list, &oldadp->ad_block.nb_jnewblk->jn_list, &newadp->ad_block.nb_jwork); oldadp->ad_block.nb_jnewblk = NULL; cancel_newblk(&oldadp->ad_block, NULL, &newadp->ad_block.nb_jwork); } else { wk = (struct worklist *) cancel_newblk(&oldadp->ad_block, &freefrag->ff_list, &freefrag->ff_jwork); freefrag->ff_jdep = jnewblk_merge(freefrag->ff_jdep, wk, &freefrag->ff_jwork); } free_newblk(&oldadp->ad_block); } /* * Allocate a jfreefrag structure to journal a single block free. */ static struct jfreefrag * newjfreefrag(freefrag, ip, blkno, size, lbn) struct freefrag *freefrag; struct inode *ip; ufs2_daddr_t blkno; long size; ufs_lbn_t lbn; { struct jfreefrag *jfreefrag; struct fs *fs; fs = ITOFS(ip); jfreefrag = malloc(sizeof(struct jfreefrag), M_JFREEFRAG, M_SOFTDEP_FLAGS); workitem_alloc(&jfreefrag->fr_list, D_JFREEFRAG, ITOVFS(ip)); jfreefrag->fr_jsegdep = newjsegdep(&jfreefrag->fr_list); jfreefrag->fr_state = ATTACHED | DEPCOMPLETE; jfreefrag->fr_ino = ip->i_number; jfreefrag->fr_lbn = lbn; jfreefrag->fr_blkno = blkno; jfreefrag->fr_frags = numfrags(fs, size); jfreefrag->fr_freefrag = freefrag; return (jfreefrag); } /* * Allocate a new freefrag structure. */ static struct freefrag * newfreefrag(ip, blkno, size, lbn, key) struct inode *ip; ufs2_daddr_t blkno; long size; ufs_lbn_t lbn; u_long key; { struct freefrag *freefrag; struct ufsmount *ump; struct fs *fs; CTR4(KTR_SUJ, "newfreefrag: ino %d blkno %jd size %ld lbn %jd", ip->i_number, blkno, size, lbn); ump = ITOUMP(ip); fs = ump->um_fs; if (fragnum(fs, blkno) + numfrags(fs, size) > fs->fs_frag) panic("newfreefrag: frag size"); freefrag = malloc(sizeof(struct freefrag), M_FREEFRAG, M_SOFTDEP_FLAGS); workitem_alloc(&freefrag->ff_list, D_FREEFRAG, UFSTOVFS(ump)); freefrag->ff_state = ATTACHED; LIST_INIT(&freefrag->ff_jwork); freefrag->ff_inum = ip->i_number; freefrag->ff_vtype = ITOV(ip)->v_type; freefrag->ff_blkno = blkno; freefrag->ff_fragsize = size; freefrag->ff_key = key; if (MOUNTEDSUJ(UFSTOVFS(ump))) { freefrag->ff_jdep = (struct worklist *) newjfreefrag(freefrag, ip, blkno, size, lbn); } else { freefrag->ff_state |= DEPCOMPLETE; freefrag->ff_jdep = NULL; } return (freefrag); } /* * This workitem de-allocates fragments that were replaced during * file block allocation. */ static void handle_workitem_freefrag(freefrag) struct freefrag *freefrag; { struct ufsmount *ump = VFSTOUFS(freefrag->ff_list.wk_mp); struct workhead wkhd; CTR3(KTR_SUJ, "handle_workitem_freefrag: ino %d blkno %jd size %ld", freefrag->ff_inum, freefrag->ff_blkno, freefrag->ff_fragsize); /* * It would be illegal to add new completion items to the * freefrag after it was schedule to be done so it must be * safe to modify the list head here. */ LIST_INIT(&wkhd); ACQUIRE_LOCK(ump); LIST_SWAP(&freefrag->ff_jwork, &wkhd, worklist, wk_list); /* * If the journal has not been written we must cancel it here. */ if (freefrag->ff_jdep) { if (freefrag->ff_jdep->wk_type != D_JNEWBLK) panic("handle_workitem_freefrag: Unexpected type %d\n", freefrag->ff_jdep->wk_type); cancel_jnewblk(WK_JNEWBLK(freefrag->ff_jdep), &wkhd); } FREE_LOCK(ump); ffs_blkfree(ump, ump->um_fs, ump->um_devvp, freefrag->ff_blkno, freefrag->ff_fragsize, freefrag->ff_inum, freefrag->ff_vtype, &wkhd, freefrag->ff_key); ACQUIRE_LOCK(ump); WORKITEM_FREE(freefrag, D_FREEFRAG); FREE_LOCK(ump); } /* * Set up a dependency structure for an external attributes data block. * This routine follows much of the structure of softdep_setup_allocdirect. * See the description of softdep_setup_allocdirect above for details. */ void softdep_setup_allocext(ip, off, newblkno, oldblkno, newsize, oldsize, bp) struct inode *ip; ufs_lbn_t off; ufs2_daddr_t newblkno; ufs2_daddr_t oldblkno; long newsize; long oldsize; struct buf *bp; { struct allocdirect *adp, *oldadp; struct allocdirectlst *adphead; struct freefrag *freefrag; struct inodedep *inodedep; struct jnewblk *jnewblk; struct newblk *newblk; struct mount *mp; struct ufsmount *ump; ufs_lbn_t lbn; mp = ITOVFS(ip); ump = VFSTOUFS(mp); KASSERT(MOUNTEDSOFTDEP(mp) != 0, ("softdep_setup_allocext called on non-softdep filesystem")); KASSERT(off < UFS_NXADDR, ("softdep_setup_allocext: lbn %lld > UFS_NXADDR", (long long)off)); lbn = bp->b_lblkno; if (oldblkno && oldblkno != newblkno) /* * The usual case is that a smaller fragment that * was just allocated has been replaced with a bigger * fragment or a full-size block. If it is marked as * B_DELWRI, the current contents have not been written * to disk. It is possible that the block was written * earlier, but very uncommon. If the block has never * been written, there is no need to send a BIO_DELETE * for it when it is freed. The gain from avoiding the * TRIMs for the common case of unwritten blocks far * exceeds the cost of the write amplification for the * uncommon case of failing to send a TRIM for a block * that had been written. */ freefrag = newfreefrag(ip, oldblkno, oldsize, lbn, (bp->b_flags & B_DELWRI) != 0 ? NOTRIM_KEY : SINGLETON_KEY); else freefrag = NULL; ACQUIRE_LOCK(ump); if (newblk_lookup(mp, newblkno, 0, &newblk) == 0) panic("softdep_setup_allocext: lost block"); KASSERT(newblk->nb_list.wk_type == D_NEWBLK, ("softdep_setup_allocext: newblk already initialized")); /* * Convert the newblk to an allocdirect. */ WORKITEM_REASSIGN(newblk, D_ALLOCDIRECT); adp = (struct allocdirect *)newblk; newblk->nb_freefrag = freefrag; adp->ad_offset = off; adp->ad_oldblkno = oldblkno; adp->ad_newsize = newsize; adp->ad_oldsize = oldsize; adp->ad_state |= EXTDATA; /* * Finish initializing the journal. */ if ((jnewblk = newblk->nb_jnewblk) != NULL) { jnewblk->jn_ino = ip->i_number; jnewblk->jn_lbn = lbn; add_to_journal(&jnewblk->jn_list); } if (freefrag && freefrag->ff_jdep != NULL && freefrag->ff_jdep->wk_type == D_JFREEFRAG) add_to_journal(freefrag->ff_jdep); inodedep_lookup(mp, ip->i_number, DEPALLOC, &inodedep); adp->ad_inodedep = inodedep; WORKLIST_INSERT(&bp->b_dep, &newblk->nb_list); /* * The list of allocdirects must be kept in sorted and ascending * order so that the rollback routines can quickly determine the * first uncommitted block (the size of the file stored on disk * ends at the end of the lowest committed fragment, or if there * are no fragments, at the end of the highest committed block). * Since files generally grow, the typical case is that the new * block is to be added at the end of the list. We speed this * special case by checking against the last allocdirect in the * list before laboriously traversing the list looking for the * insertion point. */ adphead = &inodedep->id_newextupdt; oldadp = TAILQ_LAST(adphead, allocdirectlst); if (oldadp == NULL || oldadp->ad_offset <= off) { /* insert at end of list */ TAILQ_INSERT_TAIL(adphead, adp, ad_next); if (oldadp != NULL && oldadp->ad_offset == off) allocdirect_merge(adphead, adp, oldadp); FREE_LOCK(ump); return; } TAILQ_FOREACH(oldadp, adphead, ad_next) { if (oldadp->ad_offset >= off) break; } if (oldadp == NULL) panic("softdep_setup_allocext: lost entry"); /* insert in middle of list */ TAILQ_INSERT_BEFORE(oldadp, adp, ad_next); if (oldadp->ad_offset == off) allocdirect_merge(adphead, adp, oldadp); FREE_LOCK(ump); } /* * Indirect block allocation dependencies. * * The same dependencies that exist for a direct block also exist when * a new block is allocated and pointed to by an entry in a block of * indirect pointers. The undo/redo states described above are also * used here. Because an indirect block contains many pointers that * may have dependencies, a second copy of the entire in-memory indirect * block is kept. The buffer cache copy is always completely up-to-date. * The second copy, which is used only as a source for disk writes, * contains only the safe pointers (i.e., those that have no remaining * update dependencies). The second copy is freed when all pointers * are safe. The cache is not allowed to replace indirect blocks with * pending update dependencies. If a buffer containing an indirect * block with dependencies is written, these routines will mark it * dirty again. It can only be successfully written once all the * dependencies are removed. The ffs_fsync routine in conjunction with * softdep_sync_metadata work together to get all the dependencies * removed so that a file can be successfully written to disk. Three * procedures are used when setting up indirect block pointer * dependencies. The division is necessary because of the organization * of the "balloc" routine and because of the distinction between file * pages and file metadata blocks. */ /* * Allocate a new allocindir structure. */ static struct allocindir * newallocindir(ip, ptrno, newblkno, oldblkno, lbn) struct inode *ip; /* inode for file being extended */ int ptrno; /* offset of pointer in indirect block */ ufs2_daddr_t newblkno; /* disk block number being added */ ufs2_daddr_t oldblkno; /* previous block number, 0 if none */ ufs_lbn_t lbn; { struct newblk *newblk; struct allocindir *aip; struct freefrag *freefrag; struct jnewblk *jnewblk; if (oldblkno) freefrag = newfreefrag(ip, oldblkno, ITOFS(ip)->fs_bsize, lbn, SINGLETON_KEY); else freefrag = NULL; ACQUIRE_LOCK(ITOUMP(ip)); if (newblk_lookup(ITOVFS(ip), newblkno, 0, &newblk) == 0) panic("new_allocindir: lost block"); KASSERT(newblk->nb_list.wk_type == D_NEWBLK, ("newallocindir: newblk already initialized")); WORKITEM_REASSIGN(newblk, D_ALLOCINDIR); newblk->nb_freefrag = freefrag; aip = (struct allocindir *)newblk; aip->ai_offset = ptrno; aip->ai_oldblkno = oldblkno; aip->ai_lbn = lbn; if ((jnewblk = newblk->nb_jnewblk) != NULL) { jnewblk->jn_ino = ip->i_number; jnewblk->jn_lbn = lbn; add_to_journal(&jnewblk->jn_list); } if (freefrag && freefrag->ff_jdep != NULL && freefrag->ff_jdep->wk_type == D_JFREEFRAG) add_to_journal(freefrag->ff_jdep); return (aip); } /* * Called just before setting an indirect block pointer * to a newly allocated file page. */ void softdep_setup_allocindir_page(ip, lbn, bp, ptrno, newblkno, oldblkno, nbp) struct inode *ip; /* inode for file being extended */ ufs_lbn_t lbn; /* allocated block number within file */ struct buf *bp; /* buffer with indirect blk referencing page */ int ptrno; /* offset of pointer in indirect block */ ufs2_daddr_t newblkno; /* disk block number being added */ ufs2_daddr_t oldblkno; /* previous block number, 0 if none */ struct buf *nbp; /* buffer holding allocated page */ { struct inodedep *inodedep; struct freefrag *freefrag; struct allocindir *aip; struct pagedep *pagedep; struct mount *mp; struct ufsmount *ump; mp = ITOVFS(ip); ump = VFSTOUFS(mp); KASSERT(MOUNTEDSOFTDEP(mp) != 0, ("softdep_setup_allocindir_page called on non-softdep filesystem")); KASSERT(lbn == nbp->b_lblkno, ("softdep_setup_allocindir_page: lbn %jd != lblkno %jd", lbn, bp->b_lblkno)); CTR4(KTR_SUJ, "softdep_setup_allocindir_page: ino %d blkno %jd oldblkno %jd " "lbn %jd", ip->i_number, newblkno, oldblkno, lbn); ASSERT_VOP_LOCKED(ITOV(ip), "softdep_setup_allocindir_page"); aip = newallocindir(ip, ptrno, newblkno, oldblkno, lbn); (void) inodedep_lookup(mp, ip->i_number, DEPALLOC, &inodedep); /* * If we are allocating a directory page, then we must * allocate an associated pagedep to track additions and * deletions. */ if ((ip->i_mode & IFMT) == IFDIR) pagedep_lookup(mp, nbp, ip->i_number, lbn, DEPALLOC, &pagedep); WORKLIST_INSERT(&nbp->b_dep, &aip->ai_block.nb_list); freefrag = setup_allocindir_phase2(bp, ip, inodedep, aip, lbn); FREE_LOCK(ump); if (freefrag) handle_workitem_freefrag(freefrag); } /* * Called just before setting an indirect block pointer to a * newly allocated indirect block. */ void softdep_setup_allocindir_meta(nbp, ip, bp, ptrno, newblkno) struct buf *nbp; /* newly allocated indirect block */ struct inode *ip; /* inode for file being extended */ struct buf *bp; /* indirect block referencing allocated block */ int ptrno; /* offset of pointer in indirect block */ ufs2_daddr_t newblkno; /* disk block number being added */ { struct inodedep *inodedep; struct allocindir *aip; struct ufsmount *ump; ufs_lbn_t lbn; ump = ITOUMP(ip); KASSERT(MOUNTEDSOFTDEP(UFSTOVFS(ump)) != 0, ("softdep_setup_allocindir_meta called on non-softdep filesystem")); CTR3(KTR_SUJ, "softdep_setup_allocindir_meta: ino %d blkno %jd ptrno %d", ip->i_number, newblkno, ptrno); lbn = nbp->b_lblkno; ASSERT_VOP_LOCKED(ITOV(ip), "softdep_setup_allocindir_meta"); aip = newallocindir(ip, ptrno, newblkno, 0, lbn); inodedep_lookup(UFSTOVFS(ump), ip->i_number, DEPALLOC, &inodedep); WORKLIST_INSERT(&nbp->b_dep, &aip->ai_block.nb_list); if (setup_allocindir_phase2(bp, ip, inodedep, aip, lbn)) panic("softdep_setup_allocindir_meta: Block already existed"); FREE_LOCK(ump); } static void indirdep_complete(indirdep) struct indirdep *indirdep; { struct allocindir *aip; LIST_REMOVE(indirdep, ir_next); indirdep->ir_state |= DEPCOMPLETE; while ((aip = LIST_FIRST(&indirdep->ir_completehd)) != NULL) { LIST_REMOVE(aip, ai_next); free_newblk(&aip->ai_block); } /* * If this indirdep is not attached to a buf it was simply waiting * on completion to clear completehd. free_indirdep() asserts * that nothing is dangling. */ if ((indirdep->ir_state & ONWORKLIST) == 0) free_indirdep(indirdep); } static struct indirdep * indirdep_lookup(mp, ip, bp) struct mount *mp; struct inode *ip; struct buf *bp; { struct indirdep *indirdep, *newindirdep; struct newblk *newblk; struct ufsmount *ump; struct worklist *wk; struct fs *fs; ufs2_daddr_t blkno; ump = VFSTOUFS(mp); LOCK_OWNED(ump); indirdep = NULL; newindirdep = NULL; fs = ump->um_fs; for (;;) { LIST_FOREACH(wk, &bp->b_dep, wk_list) { if (wk->wk_type != D_INDIRDEP) continue; indirdep = WK_INDIRDEP(wk); break; } /* Found on the buffer worklist, no new structure to free. */ if (indirdep != NULL && newindirdep == NULL) return (indirdep); if (indirdep != NULL && newindirdep != NULL) panic("indirdep_lookup: simultaneous create"); /* None found on the buffer and a new structure is ready. */ if (indirdep == NULL && newindirdep != NULL) break; /* None found and no new structure available. */ FREE_LOCK(ump); newindirdep = malloc(sizeof(struct indirdep), M_INDIRDEP, M_SOFTDEP_FLAGS); workitem_alloc(&newindirdep->ir_list, D_INDIRDEP, mp); newindirdep->ir_state = ATTACHED; if (I_IS_UFS1(ip)) newindirdep->ir_state |= UFS1FMT; TAILQ_INIT(&newindirdep->ir_trunc); newindirdep->ir_saveddata = NULL; LIST_INIT(&newindirdep->ir_deplisthd); LIST_INIT(&newindirdep->ir_donehd); LIST_INIT(&newindirdep->ir_writehd); LIST_INIT(&newindirdep->ir_completehd); if (bp->b_blkno == bp->b_lblkno) { ufs_bmaparray(bp->b_vp, bp->b_lblkno, &blkno, bp, NULL, NULL); bp->b_blkno = blkno; } newindirdep->ir_freeblks = NULL; newindirdep->ir_savebp = getblk(ump->um_devvp, bp->b_blkno, bp->b_bcount, 0, 0, 0); newindirdep->ir_bp = bp; BUF_KERNPROC(newindirdep->ir_savebp); bcopy(bp->b_data, newindirdep->ir_savebp->b_data, bp->b_bcount); ACQUIRE_LOCK(ump); } indirdep = newindirdep; WORKLIST_INSERT(&bp->b_dep, &indirdep->ir_list); /* * If the block is not yet allocated we don't set DEPCOMPLETE so * that we don't free dependencies until the pointers are valid. * This could search b_dep for D_ALLOCDIRECT/D_ALLOCINDIR rather * than using the hash. */ if (newblk_lookup(mp, dbtofsb(fs, bp->b_blkno), 0, &newblk)) LIST_INSERT_HEAD(&newblk->nb_indirdeps, indirdep, ir_next); else indirdep->ir_state |= DEPCOMPLETE; return (indirdep); } /* * Called to finish the allocation of the "aip" allocated * by one of the two routines above. */ static struct freefrag * setup_allocindir_phase2(bp, ip, inodedep, aip, lbn) struct buf *bp; /* in-memory copy of the indirect block */ struct inode *ip; /* inode for file being extended */ struct inodedep *inodedep; /* Inodedep for ip */ struct allocindir *aip; /* allocindir allocated by the above routines */ ufs_lbn_t lbn; /* Logical block number for this block. */ { struct fs *fs; struct indirdep *indirdep; struct allocindir *oldaip; struct freefrag *freefrag; struct mount *mp; struct ufsmount *ump; mp = ITOVFS(ip); ump = VFSTOUFS(mp); LOCK_OWNED(ump); fs = ump->um_fs; if (bp->b_lblkno >= 0) panic("setup_allocindir_phase2: not indir blk"); KASSERT(aip->ai_offset >= 0 && aip->ai_offset < NINDIR(fs), ("setup_allocindir_phase2: Bad offset %d", aip->ai_offset)); indirdep = indirdep_lookup(mp, ip, bp); KASSERT(indirdep->ir_savebp != NULL, ("setup_allocindir_phase2 NULL ir_savebp")); aip->ai_indirdep = indirdep; /* * Check for an unwritten dependency for this indirect offset. If * there is, merge the old dependency into the new one. This happens * as a result of reallocblk only. */ freefrag = NULL; if (aip->ai_oldblkno != 0) { LIST_FOREACH(oldaip, &indirdep->ir_deplisthd, ai_next) { if (oldaip->ai_offset == aip->ai_offset) { freefrag = allocindir_merge(aip, oldaip); goto done; } } LIST_FOREACH(oldaip, &indirdep->ir_donehd, ai_next) { if (oldaip->ai_offset == aip->ai_offset) { freefrag = allocindir_merge(aip, oldaip); goto done; } } } done: LIST_INSERT_HEAD(&indirdep->ir_deplisthd, aip, ai_next); return (freefrag); } /* * Merge two allocindirs which refer to the same block. Move newblock * dependencies and setup the freefrags appropriately. */ static struct freefrag * allocindir_merge(aip, oldaip) struct allocindir *aip; struct allocindir *oldaip; { struct freefrag *freefrag; struct worklist *wk; if (oldaip->ai_newblkno != aip->ai_oldblkno) panic("allocindir_merge: blkno"); aip->ai_oldblkno = oldaip->ai_oldblkno; freefrag = aip->ai_freefrag; aip->ai_freefrag = oldaip->ai_freefrag; oldaip->ai_freefrag = NULL; KASSERT(freefrag != NULL, ("setup_allocindir_phase2: No freefrag")); /* * If we are tracking a new directory-block allocation, * move it from the old allocindir to the new allocindir. */ if ((wk = LIST_FIRST(&oldaip->ai_newdirblk)) != NULL) { WORKLIST_REMOVE(wk); if (!LIST_EMPTY(&oldaip->ai_newdirblk)) panic("allocindir_merge: extra newdirblk"); WORKLIST_INSERT(&aip->ai_newdirblk, wk); } /* * We can skip journaling for this freefrag and just complete * any pending journal work for the allocindir that is being * removed after the freefrag completes. */ if (freefrag->ff_jdep) cancel_jfreefrag(WK_JFREEFRAG(freefrag->ff_jdep)); LIST_REMOVE(oldaip, ai_next); freefrag->ff_jdep = (struct worklist *)cancel_newblk(&oldaip->ai_block, &freefrag->ff_list, &freefrag->ff_jwork); free_newblk(&oldaip->ai_block); return (freefrag); } static inline void setup_freedirect(freeblks, ip, i, needj) struct freeblks *freeblks; struct inode *ip; int i; int needj; { struct ufsmount *ump; ufs2_daddr_t blkno; int frags; blkno = DIP(ip, i_db[i]); if (blkno == 0) return; DIP_SET(ip, i_db[i], 0); ump = ITOUMP(ip); frags = sblksize(ump->um_fs, ip->i_size, i); frags = numfrags(ump->um_fs, frags); newfreework(ump, freeblks, NULL, i, blkno, frags, 0, needj); } static inline void setup_freeext(freeblks, ip, i, needj) struct freeblks *freeblks; struct inode *ip; int i; int needj; { struct ufsmount *ump; ufs2_daddr_t blkno; int frags; blkno = ip->i_din2->di_extb[i]; if (blkno == 0) return; ip->i_din2->di_extb[i] = 0; ump = ITOUMP(ip); frags = sblksize(ump->um_fs, ip->i_din2->di_extsize, i); frags = numfrags(ump->um_fs, frags); newfreework(ump, freeblks, NULL, -1 - i, blkno, frags, 0, needj); } static inline void setup_freeindir(freeblks, ip, i, lbn, needj) struct freeblks *freeblks; struct inode *ip; int i; ufs_lbn_t lbn; int needj; { struct ufsmount *ump; ufs2_daddr_t blkno; blkno = DIP(ip, i_ib[i]); if (blkno == 0) return; DIP_SET(ip, i_ib[i], 0); ump = ITOUMP(ip); newfreework(ump, freeblks, NULL, lbn, blkno, ump->um_fs->fs_frag, 0, needj); } static inline struct freeblks * newfreeblks(mp, ip) struct mount *mp; struct inode *ip; { struct freeblks *freeblks; freeblks = malloc(sizeof(struct freeblks), M_FREEBLKS, M_SOFTDEP_FLAGS|M_ZERO); workitem_alloc(&freeblks->fb_list, D_FREEBLKS, mp); LIST_INIT(&freeblks->fb_jblkdephd); LIST_INIT(&freeblks->fb_jwork); freeblks->fb_ref = 0; freeblks->fb_cgwait = 0; freeblks->fb_state = ATTACHED; freeblks->fb_uid = ip->i_uid; freeblks->fb_inum = ip->i_number; freeblks->fb_vtype = ITOV(ip)->v_type; freeblks->fb_modrev = DIP(ip, i_modrev); freeblks->fb_devvp = ITODEVVP(ip); freeblks->fb_chkcnt = 0; freeblks->fb_len = 0; return (freeblks); } static void trunc_indirdep(indirdep, freeblks, bp, off) struct indirdep *indirdep; struct freeblks *freeblks; struct buf *bp; int off; { struct allocindir *aip, *aipn; /* * The first set of allocindirs won't be in savedbp. */ LIST_FOREACH_SAFE(aip, &indirdep->ir_deplisthd, ai_next, aipn) if (aip->ai_offset > off) cancel_allocindir(aip, bp, freeblks, 1); LIST_FOREACH_SAFE(aip, &indirdep->ir_donehd, ai_next, aipn) if (aip->ai_offset > off) cancel_allocindir(aip, bp, freeblks, 1); /* * These will exist in savedbp. */ LIST_FOREACH_SAFE(aip, &indirdep->ir_writehd, ai_next, aipn) if (aip->ai_offset > off) cancel_allocindir(aip, NULL, freeblks, 0); LIST_FOREACH_SAFE(aip, &indirdep->ir_completehd, ai_next, aipn) if (aip->ai_offset > off) cancel_allocindir(aip, NULL, freeblks, 0); } /* * Follow the chain of indirects down to lastlbn creating a freework * structure for each. This will be used to start indir_trunc() at * the right offset and create the journal records for the parrtial * truncation. A second step will handle the truncated dependencies. */ static int setup_trunc_indir(freeblks, ip, lbn, lastlbn, blkno) struct freeblks *freeblks; struct inode *ip; ufs_lbn_t lbn; ufs_lbn_t lastlbn; ufs2_daddr_t blkno; { struct indirdep *indirdep; struct indirdep *indirn; struct freework *freework; struct newblk *newblk; struct mount *mp; struct ufsmount *ump; struct buf *bp; uint8_t *start; uint8_t *end; ufs_lbn_t lbnadd; int level; int error; int off; freework = NULL; if (blkno == 0) return (0); mp = freeblks->fb_list.wk_mp; ump = VFSTOUFS(mp); /* * Here, calls to VOP_BMAP() will fail. However, we already have * the on-disk address, so we just pass it to bread() instead of * having bread() attempt to calculate it using VOP_BMAP(). */ error = ffs_breadz(ump, ITOV(ip), lbn, blkptrtodb(ump, blkno), (int)mp->mnt_stat.f_iosize, NULL, NULL, 0, NOCRED, 0, NULL, &bp); if (error) return (error); level = lbn_level(lbn); lbnadd = lbn_offset(ump->um_fs, level); /* * Compute the offset of the last block we want to keep. Store * in the freework the first block we want to completely free. */ off = (lastlbn - -(lbn + level)) / lbnadd; if (off + 1 == NINDIR(ump->um_fs)) goto nowork; freework = newfreework(ump, freeblks, NULL, lbn, blkno, 0, off + 1, 0); /* * Link the freework into the indirdep. This will prevent any new * allocations from proceeding until we are finished with the * truncate and the block is written. */ ACQUIRE_LOCK(ump); indirdep = indirdep_lookup(mp, ip, bp); if (indirdep->ir_freeblks) panic("setup_trunc_indir: indirdep already truncated."); TAILQ_INSERT_TAIL(&indirdep->ir_trunc, freework, fw_next); freework->fw_indir = indirdep; /* * Cancel any allocindirs that will not make it to disk. * We have to do this for all copies of the indirdep that * live on this newblk. */ if ((indirdep->ir_state & DEPCOMPLETE) == 0) { if (newblk_lookup(mp, dbtofsb(ump->um_fs, bp->b_blkno), 0, &newblk) == 0) panic("setup_trunc_indir: lost block"); LIST_FOREACH(indirn, &newblk->nb_indirdeps, ir_next) trunc_indirdep(indirn, freeblks, bp, off); } else trunc_indirdep(indirdep, freeblks, bp, off); FREE_LOCK(ump); /* * Creation is protected by the buf lock. The saveddata is only * needed if a full truncation follows a partial truncation but it * is difficult to allocate in that case so we fetch it anyway. */ if (indirdep->ir_saveddata == NULL) indirdep->ir_saveddata = malloc(bp->b_bcount, M_INDIRDEP, M_SOFTDEP_FLAGS); nowork: /* Fetch the blkno of the child and the zero start offset. */ if (I_IS_UFS1(ip)) { blkno = ((ufs1_daddr_t *)bp->b_data)[off]; start = (uint8_t *)&((ufs1_daddr_t *)bp->b_data)[off+1]; } else { blkno = ((ufs2_daddr_t *)bp->b_data)[off]; start = (uint8_t *)&((ufs2_daddr_t *)bp->b_data)[off+1]; } if (freework) { /* Zero the truncated pointers. */ end = bp->b_data + bp->b_bcount; bzero(start, end - start); bdwrite(bp); } else bqrelse(bp); if (level == 0) return (0); lbn++; /* adjust level */ lbn -= (off * lbnadd); return setup_trunc_indir(freeblks, ip, lbn, lastlbn, blkno); } /* * Complete the partial truncation of an indirect block setup by * setup_trunc_indir(). This zeros the truncated pointers in the saved * copy and writes them to disk before the freeblks is allowed to complete. */ static void complete_trunc_indir(freework) struct freework *freework; { struct freework *fwn; struct indirdep *indirdep; struct ufsmount *ump; struct buf *bp; uintptr_t start; int count; ump = VFSTOUFS(freework->fw_list.wk_mp); LOCK_OWNED(ump); indirdep = freework->fw_indir; for (;;) { bp = indirdep->ir_bp; /* See if the block was discarded. */ if (bp == NULL) break; /* Inline part of getdirtybuf(). We dont want bremfree. */ if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL) == 0) break; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, LOCK_PTR(ump)) == 0) BUF_UNLOCK(bp); ACQUIRE_LOCK(ump); } freework->fw_state |= DEPCOMPLETE; TAILQ_REMOVE(&indirdep->ir_trunc, freework, fw_next); /* * Zero the pointers in the saved copy. */ if (indirdep->ir_state & UFS1FMT) start = sizeof(ufs1_daddr_t); else start = sizeof(ufs2_daddr_t); start *= freework->fw_start; count = indirdep->ir_savebp->b_bcount - start; start += (uintptr_t)indirdep->ir_savebp->b_data; bzero((char *)start, count); /* * We need to start the next truncation in the list if it has not * been started yet. */ fwn = TAILQ_FIRST(&indirdep->ir_trunc); if (fwn != NULL) { if (fwn->fw_freeblks == indirdep->ir_freeblks) TAILQ_REMOVE(&indirdep->ir_trunc, fwn, fw_next); if ((fwn->fw_state & ONWORKLIST) == 0) freework_enqueue(fwn); } /* * If bp is NULL the block was fully truncated, restore * the saved block list otherwise free it if it is no * longer needed. */ if (TAILQ_EMPTY(&indirdep->ir_trunc)) { if (bp == NULL) bcopy(indirdep->ir_saveddata, indirdep->ir_savebp->b_data, indirdep->ir_savebp->b_bcount); free(indirdep->ir_saveddata, M_INDIRDEP); indirdep->ir_saveddata = NULL; } /* * When bp is NULL there is a full truncation pending. We * must wait for this full truncation to be journaled before * we can release this freework because the disk pointers will * never be written as zero. */ if (bp == NULL) { if (LIST_EMPTY(&indirdep->ir_freeblks->fb_jblkdephd)) handle_written_freework(freework); else WORKLIST_INSERT(&indirdep->ir_freeblks->fb_freeworkhd, &freework->fw_list); if (fwn == NULL) { freework->fw_indir = (void *)0x0000deadbeef0000; bp = indirdep->ir_savebp; indirdep->ir_savebp = NULL; free_indirdep(indirdep); FREE_LOCK(ump); brelse(bp); ACQUIRE_LOCK(ump); } } else { /* Complete when the real copy is written. */ WORKLIST_INSERT(&bp->b_dep, &freework->fw_list); BUF_UNLOCK(bp); } } /* * Calculate the number of blocks we are going to release where datablocks * is the current total and length is the new file size. */ static ufs2_daddr_t blkcount(fs, datablocks, length) struct fs *fs; ufs2_daddr_t datablocks; off_t length; { off_t totblks, numblks; totblks = 0; numblks = howmany(length, fs->fs_bsize); if (numblks <= UFS_NDADDR) { totblks = howmany(length, fs->fs_fsize); goto out; } totblks = blkstofrags(fs, numblks); numblks -= UFS_NDADDR; /* * Count all single, then double, then triple indirects required. * Subtracting one indirects worth of blocks for each pass * acknowledges one of each pointed to by the inode. */ for (;;) { totblks += blkstofrags(fs, howmany(numblks, NINDIR(fs))); numblks -= NINDIR(fs); if (numblks <= 0) break; numblks = howmany(numblks, NINDIR(fs)); } out: totblks = fsbtodb(fs, totblks); /* * Handle sparse files. We can't reclaim more blocks than the inode * references. We will correct it later in handle_complete_freeblks() * when we know the real count. */ if (totblks > datablocks) return (0); return (datablocks - totblks); } /* * Handle freeblocks for journaled softupdate filesystems. * * Contrary to normal softupdates, we must preserve the block pointers in * indirects until their subordinates are free. This is to avoid journaling * every block that is freed which may consume more space than the journal * itself. The recovery program will see the free block journals at the * base of the truncated area and traverse them to reclaim space. The * pointers in the inode may be cleared immediately after the journal * records are written because each direct and indirect pointer in the * inode is recorded in a journal. This permits full truncation to proceed * asynchronously. The write order is journal -> inode -> cgs -> indirects. * * The algorithm is as follows: * 1) Traverse the in-memory state and create journal entries to release * the relevant blocks and full indirect trees. * 2) Traverse the indirect block chain adding partial truncation freework * records to indirects in the path to lastlbn. The freework will * prevent new allocation dependencies from being satisfied in this * indirect until the truncation completes. * 3) Read and lock the inode block, performing an update with the new size * and pointers. This prevents truncated data from becoming valid on * disk through step 4. * 4) Reap unsatisfied dependencies that are beyond the truncated area, * eliminate journal work for those records that do not require it. * 5) Schedule the journal records to be written followed by the inode block. * 6) Allocate any necessary frags for the end of file. * 7) Zero any partially truncated blocks. * * From this truncation proceeds asynchronously using the freework and * indir_trunc machinery. The file will not be extended again into a * partially truncated indirect block until all work is completed but * the normal dependency mechanism ensures that it is rolled back/forward * as appropriate. Further truncation may occur without delay and is * serialized in indir_trunc(). */ void softdep_journal_freeblocks(ip, cred, length, flags) struct inode *ip; /* The inode whose length is to be reduced */ struct ucred *cred; off_t length; /* The new length for the file */ int flags; /* IO_EXT and/or IO_NORMAL */ { struct freeblks *freeblks, *fbn; struct worklist *wk, *wkn; struct inodedep *inodedep; struct jblkdep *jblkdep; struct allocdirect *adp, *adpn; struct ufsmount *ump; struct fs *fs; struct buf *bp; struct vnode *vp; struct mount *mp; daddr_t dbn; ufs2_daddr_t extblocks, datablocks; ufs_lbn_t tmpval, lbn, lastlbn; int frags, lastoff, iboff, allocblock, needj, error, i; ump = ITOUMP(ip); mp = UFSTOVFS(ump); fs = ump->um_fs; KASSERT(MOUNTEDSOFTDEP(mp) != 0, ("softdep_journal_freeblocks called on non-softdep filesystem")); vp = ITOV(ip); needj = 1; iboff = -1; allocblock = 0; extblocks = 0; datablocks = 0; frags = 0; freeblks = newfreeblks(mp, ip); ACQUIRE_LOCK(ump); /* * If we're truncating a removed file that will never be written * we don't need to journal the block frees. The canceled journals * for the allocations will suffice. */ inodedep_lookup(mp, ip->i_number, DEPALLOC, &inodedep); if ((inodedep->id_state & (UNLINKED | DEPCOMPLETE)) == UNLINKED && length == 0) needj = 0; CTR3(KTR_SUJ, "softdep_journal_freeblks: ip %d length %ld needj %d", ip->i_number, length, needj); FREE_LOCK(ump); /* * Calculate the lbn that we are truncating to. This results in -1 * if we're truncating the 0 bytes. So it is the last lbn we want * to keep, not the first lbn we want to truncate. */ lastlbn = lblkno(fs, length + fs->fs_bsize - 1) - 1; lastoff = blkoff(fs, length); /* * Compute frags we are keeping in lastlbn. 0 means all. */ if (lastlbn >= 0 && lastlbn < UFS_NDADDR) { frags = fragroundup(fs, lastoff); /* adp offset of last valid allocdirect. */ iboff = lastlbn; } else if (lastlbn > 0) iboff = UFS_NDADDR; if (fs->fs_magic == FS_UFS2_MAGIC) extblocks = btodb(fragroundup(fs, ip->i_din2->di_extsize)); /* * Handle normal data blocks and indirects. This section saves * values used after the inode update to complete frag and indirect * truncation. */ if ((flags & IO_NORMAL) != 0) { /* * Handle truncation of whole direct and indirect blocks. */ for (i = iboff + 1; i < UFS_NDADDR; i++) setup_freedirect(freeblks, ip, i, needj); for (i = 0, tmpval = NINDIR(fs), lbn = UFS_NDADDR; i < UFS_NIADDR; i++, lbn += tmpval, tmpval *= NINDIR(fs)) { /* Release a whole indirect tree. */ if (lbn > lastlbn) { setup_freeindir(freeblks, ip, i, -lbn -i, needj); continue; } iboff = i + UFS_NDADDR; /* * Traverse partially truncated indirect tree. */ if (lbn <= lastlbn && lbn + tmpval - 1 > lastlbn) setup_trunc_indir(freeblks, ip, -lbn - i, lastlbn, DIP(ip, i_ib[i])); } /* * Handle partial truncation to a frag boundary. */ if (frags) { ufs2_daddr_t blkno; long oldfrags; oldfrags = blksize(fs, ip, lastlbn); blkno = DIP(ip, i_db[lastlbn]); if (blkno && oldfrags != frags) { oldfrags -= frags; oldfrags = numfrags(fs, oldfrags); blkno += numfrags(fs, frags); newfreework(ump, freeblks, NULL, lastlbn, blkno, oldfrags, 0, needj); if (needj) adjust_newfreework(freeblks, numfrags(fs, frags)); } else if (blkno == 0) allocblock = 1; } /* * Add a journal record for partial truncate if we are * handling indirect blocks. Non-indirects need no extra * journaling. */ if (length != 0 && lastlbn >= UFS_NDADDR) { UFS_INODE_SET_FLAG(ip, IN_TRUNCATED); newjtrunc(freeblks, length, 0); } ip->i_size = length; DIP_SET(ip, i_size, ip->i_size); UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE); datablocks = DIP(ip, i_blocks) - extblocks; if (length != 0) datablocks = blkcount(fs, datablocks, length); freeblks->fb_len = length; } if ((flags & IO_EXT) != 0) { for (i = 0; i < UFS_NXADDR; i++) setup_freeext(freeblks, ip, i, needj); ip->i_din2->di_extsize = 0; datablocks += extblocks; UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE); } #ifdef QUOTA /* Reference the quotas in case the block count is wrong in the end. */ quotaref(vp, freeblks->fb_quota); (void) chkdq(ip, -datablocks, NOCRED, FORCE); #endif freeblks->fb_chkcnt = -datablocks; UFS_LOCK(ump); fs->fs_pendingblocks += datablocks; UFS_UNLOCK(ump); DIP_SET(ip, i_blocks, DIP(ip, i_blocks) - datablocks); /* * Handle truncation of incomplete alloc direct dependencies. We * hold the inode block locked to prevent incomplete dependencies * from reaching the disk while we are eliminating those that * have been truncated. This is a partially inlined ffs_update(). */ ufs_itimes(vp); ip->i_flag &= ~(IN_LAZYACCESS | IN_LAZYMOD | IN_MODIFIED); dbn = fsbtodb(fs, ino_to_fsba(fs, ip->i_number)); error = ffs_breadz(ump, ump->um_devvp, dbn, dbn, (int)fs->fs_bsize, NULL, NULL, 0, cred, 0, NULL, &bp); if (error) { softdep_error("softdep_journal_freeblocks", error); return; } if (bp->b_bufsize == fs->fs_bsize) bp->b_flags |= B_CLUSTEROK; softdep_update_inodeblock(ip, bp, 0); if (ump->um_fstype == UFS1) { *((struct ufs1_dinode *)bp->b_data + ino_to_fsbo(fs, ip->i_number)) = *ip->i_din1; } else { ffs_update_dinode_ckhash(fs, ip->i_din2); *((struct ufs2_dinode *)bp->b_data + ino_to_fsbo(fs, ip->i_number)) = *ip->i_din2; } ACQUIRE_LOCK(ump); (void) inodedep_lookup(mp, ip->i_number, DEPALLOC, &inodedep); if ((inodedep->id_state & IOSTARTED) != 0) panic("softdep_setup_freeblocks: inode busy"); /* * Add the freeblks structure to the list of operations that * must await the zero'ed inode being written to disk. If we * still have a bitmap dependency (needj), then the inode * has never been written to disk, so we can process the * freeblks below once we have deleted the dependencies. */ if (needj) WORKLIST_INSERT(&bp->b_dep, &freeblks->fb_list); else freeblks->fb_state |= COMPLETE; if ((flags & IO_NORMAL) != 0) { TAILQ_FOREACH_SAFE(adp, &inodedep->id_inoupdt, ad_next, adpn) { if (adp->ad_offset > iboff) cancel_allocdirect(&inodedep->id_inoupdt, adp, freeblks); /* * Truncate the allocdirect. We could eliminate * or modify journal records as well. */ else if (adp->ad_offset == iboff && frags) adp->ad_newsize = frags; } } if ((flags & IO_EXT) != 0) while ((adp = TAILQ_FIRST(&inodedep->id_extupdt)) != NULL) cancel_allocdirect(&inodedep->id_extupdt, adp, freeblks); /* * Scan the bufwait list for newblock dependencies that will never * make it to disk. */ LIST_FOREACH_SAFE(wk, &inodedep->id_bufwait, wk_list, wkn) { if (wk->wk_type != D_ALLOCDIRECT) continue; adp = WK_ALLOCDIRECT(wk); if (((flags & IO_NORMAL) != 0 && (adp->ad_offset > iboff)) || ((flags & IO_EXT) != 0 && (adp->ad_state & EXTDATA))) { cancel_jfreeblk(freeblks, adp->ad_newblkno); cancel_newblk(WK_NEWBLK(wk), NULL, &freeblks->fb_jwork); WORKLIST_INSERT(&freeblks->fb_freeworkhd, wk); } } /* * Add journal work. */ LIST_FOREACH(jblkdep, &freeblks->fb_jblkdephd, jb_deps) add_to_journal(&jblkdep->jb_list); FREE_LOCK(ump); bdwrite(bp); /* * Truncate dependency structures beyond length. */ trunc_dependencies(ip, freeblks, lastlbn, frags, flags); /* * This is only set when we need to allocate a fragment because * none existed at the end of a frag-sized file. It handles only * allocating a new, zero filled block. */ if (allocblock) { ip->i_size = length - lastoff; DIP_SET(ip, i_size, ip->i_size); error = UFS_BALLOC(vp, length - 1, 1, cred, BA_CLRBUF, &bp); if (error != 0) { softdep_error("softdep_journal_freeblks", error); return; } ip->i_size = length; DIP_SET(ip, i_size, length); UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE); allocbuf(bp, frags); ffs_update(vp, 0); bawrite(bp); } else if (lastoff != 0 && vp->v_type != VDIR) { int size; /* * Zero the end of a truncated frag or block. */ size = sblksize(fs, length, lastlbn); error = bread(vp, lastlbn, size, cred, &bp); if (error == 0) { bzero((char *)bp->b_data + lastoff, size - lastoff); bawrite(bp); } else if (!ffs_fsfail_cleanup(ump, error)) { softdep_error("softdep_journal_freeblks", error); return; } } ACQUIRE_LOCK(ump); inodedep_lookup(mp, ip->i_number, DEPALLOC, &inodedep); TAILQ_INSERT_TAIL(&inodedep->id_freeblklst, freeblks, fb_next); freeblks->fb_state |= DEPCOMPLETE | ONDEPLIST; /* * We zero earlier truncations so they don't erroneously * update i_blocks. */ if (freeblks->fb_len == 0 && (flags & IO_NORMAL) != 0) TAILQ_FOREACH(fbn, &inodedep->id_freeblklst, fb_next) fbn->fb_len = 0; if ((freeblks->fb_state & ALLCOMPLETE) == ALLCOMPLETE && LIST_EMPTY(&freeblks->fb_jblkdephd)) freeblks->fb_state |= INPROGRESS; else freeblks = NULL; FREE_LOCK(ump); if (freeblks) handle_workitem_freeblocks(freeblks, 0); trunc_pages(ip, length, extblocks, flags); } /* * Flush a JOP_SYNC to the journal. */ void softdep_journal_fsync(ip) struct inode *ip; { struct jfsync *jfsync; struct ufsmount *ump; ump = ITOUMP(ip); KASSERT(MOUNTEDSOFTDEP(UFSTOVFS(ump)) != 0, ("softdep_journal_fsync called on non-softdep filesystem")); if ((ip->i_flag & IN_TRUNCATED) == 0) return; ip->i_flag &= ~IN_TRUNCATED; jfsync = malloc(sizeof(*jfsync), M_JFSYNC, M_SOFTDEP_FLAGS | M_ZERO); workitem_alloc(&jfsync->jfs_list, D_JFSYNC, UFSTOVFS(ump)); jfsync->jfs_size = ip->i_size; jfsync->jfs_ino = ip->i_number; ACQUIRE_LOCK(ump); add_to_journal(&jfsync->jfs_list); jwait(&jfsync->jfs_list, MNT_WAIT); FREE_LOCK(ump); } /* * Block de-allocation dependencies. * * When blocks are de-allocated, the on-disk pointers must be nullified before * the blocks are made available for use by other files. (The true * requirement is that old pointers must be nullified before new on-disk * pointers are set. We chose this slightly more stringent requirement to * reduce complexity.) Our implementation handles this dependency by updating * the inode (or indirect block) appropriately but delaying the actual block * de-allocation (i.e., freemap and free space count manipulation) until * after the updated versions reach stable storage. After the disk is * updated, the blocks can be safely de-allocated whenever it is convenient. * This implementation handles only the common case of reducing a file's * length to zero. Other cases are handled by the conventional synchronous * write approach. * * The ffs implementation with which we worked double-checks * the state of the block pointers and file size as it reduces * a file's length. Some of this code is replicated here in our * soft updates implementation. The freeblks->fb_chkcnt field is * used to transfer a part of this information to the procedure * that eventually de-allocates the blocks. * * This routine should be called from the routine that shortens * a file's length, before the inode's size or block pointers * are modified. It will save the block pointer information for * later release and zero the inode so that the calling routine * can release it. */ void softdep_setup_freeblocks(ip, length, flags) struct inode *ip; /* The inode whose length is to be reduced */ off_t length; /* The new length for the file */ int flags; /* IO_EXT and/or IO_NORMAL */ { struct ufs1_dinode *dp1; struct ufs2_dinode *dp2; struct freeblks *freeblks; struct inodedep *inodedep; struct allocdirect *adp; struct ufsmount *ump; struct buf *bp; struct fs *fs; ufs2_daddr_t extblocks, datablocks; struct mount *mp; int i, delay, error; ufs_lbn_t tmpval; ufs_lbn_t lbn; ump = ITOUMP(ip); mp = UFSTOVFS(ump); KASSERT(MOUNTEDSOFTDEP(mp) != 0, ("softdep_setup_freeblocks called on non-softdep filesystem")); CTR2(KTR_SUJ, "softdep_setup_freeblks: ip %d length %ld", ip->i_number, length); KASSERT(length == 0, ("softdep_setup_freeblocks: non-zero length")); fs = ump->um_fs; if ((error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)), (int)fs->fs_bsize, NOCRED, &bp)) != 0) { if (!ffs_fsfail_cleanup(ump, error)) softdep_error("softdep_setup_freeblocks", error); return; } freeblks = newfreeblks(mp, ip); extblocks = 0; datablocks = 0; if (fs->fs_magic == FS_UFS2_MAGIC) extblocks = btodb(fragroundup(fs, ip->i_din2->di_extsize)); if ((flags & IO_NORMAL) != 0) { for (i = 0; i < UFS_NDADDR; i++) setup_freedirect(freeblks, ip, i, 0); for (i = 0, tmpval = NINDIR(fs), lbn = UFS_NDADDR; i < UFS_NIADDR; i++, lbn += tmpval, tmpval *= NINDIR(fs)) setup_freeindir(freeblks, ip, i, -lbn -i, 0); ip->i_size = 0; DIP_SET(ip, i_size, 0); UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE); datablocks = DIP(ip, i_blocks) - extblocks; } if ((flags & IO_EXT) != 0) { for (i = 0; i < UFS_NXADDR; i++) setup_freeext(freeblks, ip, i, 0); ip->i_din2->di_extsize = 0; datablocks += extblocks; UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE); } #ifdef QUOTA /* Reference the quotas in case the block count is wrong in the end. */ quotaref(ITOV(ip), freeblks->fb_quota); (void) chkdq(ip, -datablocks, NOCRED, FORCE); #endif freeblks->fb_chkcnt = -datablocks; UFS_LOCK(ump); fs->fs_pendingblocks += datablocks; UFS_UNLOCK(ump); DIP_SET(ip, i_blocks, DIP(ip, i_blocks) - datablocks); /* * Push the zero'ed inode to its disk buffer so that we are free * to delete its dependencies below. Once the dependencies are gone * the buffer can be safely released. */ if (ump->um_fstype == UFS1) { dp1 = ((struct ufs1_dinode *)bp->b_data + ino_to_fsbo(fs, ip->i_number)); ip->i_din1->di_freelink = dp1->di_freelink; *dp1 = *ip->i_din1; } else { dp2 = ((struct ufs2_dinode *)bp->b_data + ino_to_fsbo(fs, ip->i_number)); ip->i_din2->di_freelink = dp2->di_freelink; ffs_update_dinode_ckhash(fs, ip->i_din2); *dp2 = *ip->i_din2; } /* * Find and eliminate any inode dependencies. */ ACQUIRE_LOCK(ump); (void) inodedep_lookup(mp, ip->i_number, DEPALLOC, &inodedep); if ((inodedep->id_state & IOSTARTED) != 0) panic("softdep_setup_freeblocks: inode busy"); /* * Add the freeblks structure to the list of operations that * must await the zero'ed inode being written to disk. If we * still have a bitmap dependency (delay == 0), then the inode * has never been written to disk, so we can process the * freeblks below once we have deleted the dependencies. */ delay = (inodedep->id_state & DEPCOMPLETE); if (delay) WORKLIST_INSERT(&bp->b_dep, &freeblks->fb_list); else freeblks->fb_state |= COMPLETE; /* * Because the file length has been truncated to zero, any * pending block allocation dependency structures associated * with this inode are obsolete and can simply be de-allocated. * We must first merge the two dependency lists to get rid of * any duplicate freefrag structures, then purge the merged list. * If we still have a bitmap dependency, then the inode has never * been written to disk, so we can free any fragments without delay. */ if (flags & IO_NORMAL) { merge_inode_lists(&inodedep->id_newinoupdt, &inodedep->id_inoupdt); while ((adp = TAILQ_FIRST(&inodedep->id_inoupdt)) != NULL) cancel_allocdirect(&inodedep->id_inoupdt, adp, freeblks); } if (flags & IO_EXT) { merge_inode_lists(&inodedep->id_newextupdt, &inodedep->id_extupdt); while ((adp = TAILQ_FIRST(&inodedep->id_extupdt)) != NULL) cancel_allocdirect(&inodedep->id_extupdt, adp, freeblks); } FREE_LOCK(ump); bdwrite(bp); trunc_dependencies(ip, freeblks, -1, 0, flags); ACQUIRE_LOCK(ump); if (inodedep_lookup(mp, ip->i_number, 0, &inodedep) != 0) (void) free_inodedep(inodedep); freeblks->fb_state |= DEPCOMPLETE; /* * If the inode with zeroed block pointers is now on disk * we can start freeing blocks. */ if ((freeblks->fb_state & ALLCOMPLETE) == ALLCOMPLETE) freeblks->fb_state |= INPROGRESS; else freeblks = NULL; FREE_LOCK(ump); if (freeblks) handle_workitem_freeblocks(freeblks, 0); trunc_pages(ip, length, extblocks, flags); } /* * Eliminate pages from the page cache that back parts of this inode and * adjust the vnode pager's idea of our size. This prevents stale data * from hanging around in the page cache. */ static void trunc_pages(ip, length, extblocks, flags) struct inode *ip; off_t length; ufs2_daddr_t extblocks; int flags; { struct vnode *vp; struct fs *fs; ufs_lbn_t lbn; off_t end, extend; vp = ITOV(ip); fs = ITOFS(ip); extend = OFF_TO_IDX(lblktosize(fs, -extblocks)); if ((flags & IO_EXT) != 0) vn_pages_remove(vp, extend, 0); if ((flags & IO_NORMAL) == 0) return; BO_LOCK(&vp->v_bufobj); drain_output(vp); BO_UNLOCK(&vp->v_bufobj); /* * The vnode pager eliminates file pages we eliminate indirects * below. */ vnode_pager_setsize(vp, length); /* * Calculate the end based on the last indirect we want to keep. If * the block extends into indirects we can just use the negative of * its lbn. Doubles and triples exist at lower numbers so we must * be careful not to remove those, if they exist. double and triple * indirect lbns do not overlap with others so it is not important * to verify how many levels are required. */ lbn = lblkno(fs, length); if (lbn >= UFS_NDADDR) { /* Calculate the virtual lbn of the triple indirect. */ lbn = -lbn - (UFS_NIADDR - 1); end = OFF_TO_IDX(lblktosize(fs, lbn)); } else end = extend; vn_pages_remove(vp, OFF_TO_IDX(OFF_MAX), end); } /* * See if the buf bp is in the range eliminated by truncation. */ static int trunc_check_buf(bp, blkoffp, lastlbn, lastoff, flags) struct buf *bp; int *blkoffp; ufs_lbn_t lastlbn; int lastoff; int flags; { ufs_lbn_t lbn; *blkoffp = 0; /* Only match ext/normal blocks as appropriate. */ if (((flags & IO_EXT) == 0 && (bp->b_xflags & BX_ALTDATA)) || ((flags & IO_NORMAL) == 0 && (bp->b_xflags & BX_ALTDATA) == 0)) return (0); /* ALTDATA is always a full truncation. */ if ((bp->b_xflags & BX_ALTDATA) != 0) return (1); /* -1 is full truncation. */ if (lastlbn == -1) return (1); /* * If this is a partial truncate we only want those * blocks and indirect blocks that cover the range * we're after. */ lbn = bp->b_lblkno; if (lbn < 0) lbn = -(lbn + lbn_level(lbn)); if (lbn < lastlbn) return (0); /* Here we only truncate lblkno if it's partial. */ if (lbn == lastlbn) { if (lastoff == 0) return (0); *blkoffp = lastoff; } return (1); } /* * Eliminate any dependencies that exist in memory beyond lblkno:off */ static void trunc_dependencies(ip, freeblks, lastlbn, lastoff, flags) struct inode *ip; struct freeblks *freeblks; ufs_lbn_t lastlbn; int lastoff; int flags; { struct bufobj *bo; struct vnode *vp; struct buf *bp; int blkoff; /* * We must wait for any I/O in progress to finish so that * all potential buffers on the dirty list will be visible. * Once they are all there, walk the list and get rid of * any dependencies. */ vp = ITOV(ip); bo = &vp->v_bufobj; BO_LOCK(bo); drain_output(vp); TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) bp->b_vflags &= ~BV_SCANNED; restart: TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) { if (bp->b_vflags & BV_SCANNED) continue; if (!trunc_check_buf(bp, &blkoff, lastlbn, lastoff, flags)) { bp->b_vflags |= BV_SCANNED; continue; } KASSERT(bp->b_bufobj == bo, ("Wrong object in buffer")); if ((bp = getdirtybuf(bp, BO_LOCKPTR(bo), MNT_WAIT)) == NULL) goto restart; BO_UNLOCK(bo); if (deallocate_dependencies(bp, freeblks, blkoff)) bqrelse(bp); else brelse(bp); BO_LOCK(bo); goto restart; } /* * Now do the work of vtruncbuf while also matching indirect blocks. */ TAILQ_FOREACH(bp, &bo->bo_clean.bv_hd, b_bobufs) bp->b_vflags &= ~BV_SCANNED; cleanrestart: TAILQ_FOREACH(bp, &bo->bo_clean.bv_hd, b_bobufs) { if (bp->b_vflags & BV_SCANNED) continue; if (!trunc_check_buf(bp, &blkoff, lastlbn, lastoff, flags)) { bp->b_vflags |= BV_SCANNED; continue; } if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) { BO_LOCK(bo); goto cleanrestart; } BO_LOCK(bo); bp->b_vflags |= BV_SCANNED; BO_UNLOCK(bo); bremfree(bp); if (blkoff != 0) { allocbuf(bp, blkoff); bqrelse(bp); } else { bp->b_flags |= B_INVAL | B_NOCACHE | B_RELBUF; brelse(bp); } BO_LOCK(bo); goto cleanrestart; } drain_output(vp); BO_UNLOCK(bo); } static int cancel_pagedep(pagedep, freeblks, blkoff) struct pagedep *pagedep; struct freeblks *freeblks; int blkoff; { struct jremref *jremref; struct jmvref *jmvref; struct dirrem *dirrem, *tmp; int i; /* * Copy any directory remove dependencies to the list * to be processed after the freeblks proceeds. If * directory entry never made it to disk they * can be dumped directly onto the work list. */ LIST_FOREACH_SAFE(dirrem, &pagedep->pd_dirremhd, dm_next, tmp) { /* Skip this directory removal if it is intended to remain. */ if (dirrem->dm_offset < blkoff) continue; /* * If there are any dirrems we wait for the journal write * to complete and then restart the buf scan as the lock * has been dropped. */ while ((jremref = LIST_FIRST(&dirrem->dm_jremrefhd)) != NULL) { jwait(&jremref->jr_list, MNT_WAIT); return (ERESTART); } LIST_REMOVE(dirrem, dm_next); dirrem->dm_dirinum = pagedep->pd_ino; WORKLIST_INSERT(&freeblks->fb_freeworkhd, &dirrem->dm_list); } while ((jmvref = LIST_FIRST(&pagedep->pd_jmvrefhd)) != NULL) { jwait(&jmvref->jm_list, MNT_WAIT); return (ERESTART); } /* * When we're partially truncating a pagedep we just want to flush * journal entries and return. There can not be any adds in the * truncated portion of the directory and newblk must remain if * part of the block remains. */ if (blkoff != 0) { struct diradd *dap; LIST_FOREACH(dap, &pagedep->pd_pendinghd, da_pdlist) if (dap->da_offset > blkoff) panic("cancel_pagedep: diradd %p off %d > %d", dap, dap->da_offset, blkoff); for (i = 0; i < DAHASHSZ; i++) LIST_FOREACH(dap, &pagedep->pd_diraddhd[i], da_pdlist) if (dap->da_offset > blkoff) panic("cancel_pagedep: diradd %p off %d > %d", dap, dap->da_offset, blkoff); return (0); } /* * There should be no directory add dependencies present * as the directory could not be truncated until all * children were removed. */ KASSERT(LIST_FIRST(&pagedep->pd_pendinghd) == NULL, ("deallocate_dependencies: pendinghd != NULL")); for (i = 0; i < DAHASHSZ; i++) KASSERT(LIST_FIRST(&pagedep->pd_diraddhd[i]) == NULL, ("deallocate_dependencies: diraddhd != NULL")); if ((pagedep->pd_state & NEWBLOCK) != 0) free_newdirblk(pagedep->pd_newdirblk); if (free_pagedep(pagedep) == 0) panic("Failed to free pagedep %p", pagedep); return (0); } /* * Reclaim any dependency structures from a buffer that is about to * be reallocated to a new vnode. The buffer must be locked, thus, * no I/O completion operations can occur while we are manipulating * its associated dependencies. The mutex is held so that other I/O's * associated with related dependencies do not occur. */ static int deallocate_dependencies(bp, freeblks, off) struct buf *bp; struct freeblks *freeblks; int off; { struct indirdep *indirdep; struct pagedep *pagedep; struct worklist *wk, *wkn; struct ufsmount *ump; ump = softdep_bp_to_mp(bp); if (ump == NULL) goto done; ACQUIRE_LOCK(ump); LIST_FOREACH_SAFE(wk, &bp->b_dep, wk_list, wkn) { switch (wk->wk_type) { case D_INDIRDEP: indirdep = WK_INDIRDEP(wk); if (bp->b_lblkno >= 0 || bp->b_blkno != indirdep->ir_savebp->b_lblkno) panic("deallocate_dependencies: not indir"); cancel_indirdep(indirdep, bp, freeblks); continue; case D_PAGEDEP: pagedep = WK_PAGEDEP(wk); if (cancel_pagedep(pagedep, freeblks, off)) { FREE_LOCK(ump); return (ERESTART); } continue; case D_ALLOCINDIR: /* * Simply remove the allocindir, we'll find it via * the indirdep where we can clear pointers if * needed. */ WORKLIST_REMOVE(wk); continue; case D_FREEWORK: /* * A truncation is waiting for the zero'd pointers * to be written. It can be freed when the freeblks * is journaled. */ WORKLIST_REMOVE(wk); wk->wk_state |= ONDEPLIST; WORKLIST_INSERT(&freeblks->fb_freeworkhd, wk); break; case D_ALLOCDIRECT: if (off != 0) continue; /* FALLTHROUGH */ default: panic("deallocate_dependencies: Unexpected type %s", TYPENAME(wk->wk_type)); /* NOTREACHED */ } } FREE_LOCK(ump); done: /* * Don't throw away this buf, we were partially truncating and * some deps may always remain. */ if (off) { allocbuf(bp, off); bp->b_vflags |= BV_SCANNED; return (EBUSY); } bp->b_flags |= B_INVAL | B_NOCACHE; return (0); } /* * An allocdirect is being canceled due to a truncate. We must make sure * the journal entry is released in concert with the blkfree that releases * the storage. Completed journal entries must not be released until the * space is no longer pointed to by the inode or in the bitmap. */ static void cancel_allocdirect(adphead, adp, freeblks) struct allocdirectlst *adphead; struct allocdirect *adp; struct freeblks *freeblks; { struct freework *freework; struct newblk *newblk; struct worklist *wk; TAILQ_REMOVE(adphead, adp, ad_next); newblk = (struct newblk *)adp; freework = NULL; /* * Find the correct freework structure. */ LIST_FOREACH(wk, &freeblks->fb_freeworkhd, wk_list) { if (wk->wk_type != D_FREEWORK) continue; freework = WK_FREEWORK(wk); if (freework->fw_blkno == newblk->nb_newblkno) break; } if (freework == NULL) panic("cancel_allocdirect: Freework not found"); /* * If a newblk exists at all we still have the journal entry that * initiated the allocation so we do not need to journal the free. */ cancel_jfreeblk(freeblks, freework->fw_blkno); /* * If the journal hasn't been written the jnewblk must be passed * to the call to ffs_blkfree that reclaims the space. We accomplish * this by linking the journal dependency into the freework to be * freed when freework_freeblock() is called. If the journal has * been written we can simply reclaim the journal space when the * freeblks work is complete. */ freework->fw_jnewblk = cancel_newblk(newblk, &freework->fw_list, &freeblks->fb_jwork); WORKLIST_INSERT(&freeblks->fb_freeworkhd, &newblk->nb_list); } /* * Cancel a new block allocation. May be an indirect or direct block. We * remove it from various lists and return any journal record that needs to * be resolved by the caller. * * A special consideration is made for indirects which were never pointed * at on disk and will never be found once this block is released. */ static struct jnewblk * cancel_newblk(newblk, wk, wkhd) struct newblk *newblk; struct worklist *wk; struct workhead *wkhd; { struct jnewblk *jnewblk; CTR1(KTR_SUJ, "cancel_newblk: blkno %jd", newblk->nb_newblkno); newblk->nb_state |= GOINGAWAY; /* * Previously we traversed the completedhd on each indirdep * attached to this newblk to cancel them and gather journal * work. Since we need only the oldest journal segment and * the lowest point on the tree will always have the oldest * journal segment we are free to release the segments * of any subordinates and may leave the indirdep list to * indirdep_complete() when this newblk is freed. */ if (newblk->nb_state & ONDEPLIST) { newblk->nb_state &= ~ONDEPLIST; LIST_REMOVE(newblk, nb_deps); } if (newblk->nb_state & ONWORKLIST) WORKLIST_REMOVE(&newblk->nb_list); /* * If the journal entry hasn't been written we save a pointer to * the dependency that frees it until it is written or the * superseding operation completes. */ jnewblk = newblk->nb_jnewblk; if (jnewblk != NULL && wk != NULL) { newblk->nb_jnewblk = NULL; jnewblk->jn_dep = wk; } if (!LIST_EMPTY(&newblk->nb_jwork)) jwork_move(wkhd, &newblk->nb_jwork); /* * When truncating we must free the newdirblk early to remove * the pagedep from the hash before returning. */ if ((wk = LIST_FIRST(&newblk->nb_newdirblk)) != NULL) free_newdirblk(WK_NEWDIRBLK(wk)); if (!LIST_EMPTY(&newblk->nb_newdirblk)) panic("cancel_newblk: extra newdirblk"); return (jnewblk); } /* * Schedule the freefrag associated with a newblk to be released once * the pointers are written and the previous block is no longer needed. */ static void newblk_freefrag(newblk) struct newblk *newblk; { struct freefrag *freefrag; if (newblk->nb_freefrag == NULL) return; freefrag = newblk->nb_freefrag; newblk->nb_freefrag = NULL; freefrag->ff_state |= COMPLETE; if ((freefrag->ff_state & ALLCOMPLETE) == ALLCOMPLETE) add_to_worklist(&freefrag->ff_list, 0); } /* * Free a newblk. Generate a new freefrag work request if appropriate. * This must be called after the inode pointer and any direct block pointers * are valid or fully removed via truncate or frag extension. */ static void free_newblk(newblk) struct newblk *newblk; { struct indirdep *indirdep; struct worklist *wk; KASSERT(newblk->nb_jnewblk == NULL, ("free_newblk: jnewblk %p still attached", newblk->nb_jnewblk)); KASSERT(newblk->nb_list.wk_type != D_NEWBLK, ("free_newblk: unclaimed newblk")); LOCK_OWNED(VFSTOUFS(newblk->nb_list.wk_mp)); newblk_freefrag(newblk); if (newblk->nb_state & ONDEPLIST) LIST_REMOVE(newblk, nb_deps); if (newblk->nb_state & ONWORKLIST) WORKLIST_REMOVE(&newblk->nb_list); LIST_REMOVE(newblk, nb_hash); if ((wk = LIST_FIRST(&newblk->nb_newdirblk)) != NULL) free_newdirblk(WK_NEWDIRBLK(wk)); if (!LIST_EMPTY(&newblk->nb_newdirblk)) panic("free_newblk: extra newdirblk"); while ((indirdep = LIST_FIRST(&newblk->nb_indirdeps)) != NULL) indirdep_complete(indirdep); handle_jwork(&newblk->nb_jwork); WORKITEM_FREE(newblk, D_NEWBLK); } /* * Free a newdirblk. Clear the NEWBLOCK flag on its associated pagedep. */ static void free_newdirblk(newdirblk) struct newdirblk *newdirblk; { struct pagedep *pagedep; struct diradd *dap; struct worklist *wk; LOCK_OWNED(VFSTOUFS(newdirblk->db_list.wk_mp)); WORKLIST_REMOVE(&newdirblk->db_list); /* * If the pagedep is still linked onto the directory buffer * dependency chain, then some of the entries on the * pd_pendinghd list may not be committed to disk yet. In * this case, we will simply clear the NEWBLOCK flag and * let the pd_pendinghd list be processed when the pagedep * is next written. If the pagedep is no longer on the buffer * dependency chain, then all the entries on the pd_pending * list are committed to disk and we can free them here. */ pagedep = newdirblk->db_pagedep; pagedep->pd_state &= ~NEWBLOCK; if ((pagedep->pd_state & ONWORKLIST) == 0) { while ((dap = LIST_FIRST(&pagedep->pd_pendinghd)) != NULL) free_diradd(dap, NULL); /* * If no dependencies remain, the pagedep will be freed. */ free_pagedep(pagedep); } /* Should only ever be one item in the list. */ while ((wk = LIST_FIRST(&newdirblk->db_mkdir)) != NULL) { WORKLIST_REMOVE(wk); handle_written_mkdir(WK_MKDIR(wk), MKDIR_BODY); } WORKITEM_FREE(newdirblk, D_NEWDIRBLK); } /* * Prepare an inode to be freed. The actual free operation is not * done until the zero'ed inode has been written to disk. */ void softdep_freefile(pvp, ino, mode) struct vnode *pvp; ino_t ino; int mode; { struct inode *ip = VTOI(pvp); struct inodedep *inodedep; struct freefile *freefile; struct freeblks *freeblks; struct ufsmount *ump; ump = ITOUMP(ip); KASSERT(MOUNTEDSOFTDEP(UFSTOVFS(ump)) != 0, ("softdep_freefile called on non-softdep filesystem")); /* * This sets up the inode de-allocation dependency. */ freefile = malloc(sizeof(struct freefile), M_FREEFILE, M_SOFTDEP_FLAGS); workitem_alloc(&freefile->fx_list, D_FREEFILE, pvp->v_mount); freefile->fx_mode = mode; freefile->fx_oldinum = ino; freefile->fx_devvp = ump->um_devvp; LIST_INIT(&freefile->fx_jwork); UFS_LOCK(ump); ump->um_fs->fs_pendinginodes += 1; UFS_UNLOCK(ump); /* * If the inodedep does not exist, then the zero'ed inode has * been written to disk. If the allocated inode has never been * written to disk, then the on-disk inode is zero'ed. In either * case we can free the file immediately. If the journal was * canceled before being written the inode will never make it to * disk and we must send the canceled journal entrys to * ffs_freefile() to be cleared in conjunction with the bitmap. * Any blocks waiting on the inode to write can be safely freed * here as it will never been written. */ ACQUIRE_LOCK(ump); inodedep_lookup(pvp->v_mount, ino, 0, &inodedep); if (inodedep) { /* * Clear out freeblks that no longer need to reference * this inode. */ while ((freeblks = TAILQ_FIRST(&inodedep->id_freeblklst)) != NULL) { TAILQ_REMOVE(&inodedep->id_freeblklst, freeblks, fb_next); freeblks->fb_state &= ~ONDEPLIST; } /* * Remove this inode from the unlinked list. */ if (inodedep->id_state & UNLINKED) { /* * Save the journal work to be freed with the bitmap * before we clear UNLINKED. Otherwise it can be lost * if the inode block is written. */ handle_bufwait(inodedep, &freefile->fx_jwork); clear_unlinked_inodedep(inodedep); /* * Re-acquire inodedep as we've dropped the * per-filesystem lock in clear_unlinked_inodedep(). */ inodedep_lookup(pvp->v_mount, ino, 0, &inodedep); } } if (inodedep == NULL || check_inode_unwritten(inodedep)) { FREE_LOCK(ump); handle_workitem_freefile(freefile); return; } if ((inodedep->id_state & DEPCOMPLETE) == 0) inodedep->id_state |= GOINGAWAY; WORKLIST_INSERT(&inodedep->id_inowait, &freefile->fx_list); FREE_LOCK(ump); if (ip->i_number == ino) UFS_INODE_SET_FLAG(ip, IN_MODIFIED); } /* * Check to see if an inode has never been written to disk. If * so free the inodedep and return success, otherwise return failure. * * If we still have a bitmap dependency, then the inode has never * been written to disk. Drop the dependency as it is no longer * necessary since the inode is being deallocated. We set the * ALLCOMPLETE flags since the bitmap now properly shows that the * inode is not allocated. Even if the inode is actively being * written, it has been rolled back to its zero'ed state, so we * are ensured that a zero inode is what is on the disk. For short * lived files, this change will usually result in removing all the * dependencies from the inode so that it can be freed immediately. */ static int check_inode_unwritten(inodedep) struct inodedep *inodedep; { LOCK_OWNED(VFSTOUFS(inodedep->id_list.wk_mp)); if ((inodedep->id_state & (DEPCOMPLETE | UNLINKED)) != 0 || !LIST_EMPTY(&inodedep->id_dirremhd) || !LIST_EMPTY(&inodedep->id_pendinghd) || !LIST_EMPTY(&inodedep->id_bufwait) || !LIST_EMPTY(&inodedep->id_inowait) || !TAILQ_EMPTY(&inodedep->id_inoreflst) || !TAILQ_EMPTY(&inodedep->id_inoupdt) || !TAILQ_EMPTY(&inodedep->id_newinoupdt) || !TAILQ_EMPTY(&inodedep->id_extupdt) || !TAILQ_EMPTY(&inodedep->id_newextupdt) || !TAILQ_EMPTY(&inodedep->id_freeblklst) || inodedep->id_mkdiradd != NULL || inodedep->id_nlinkdelta != 0) return (0); /* * Another process might be in initiate_write_inodeblock_ufs[12] * trying to allocate memory without holding "Softdep Lock". */ if ((inodedep->id_state & IOSTARTED) != 0 && inodedep->id_savedino1 == NULL) return (0); if (inodedep->id_state & ONDEPLIST) LIST_REMOVE(inodedep, id_deps); inodedep->id_state &= ~ONDEPLIST; inodedep->id_state |= ALLCOMPLETE; inodedep->id_bmsafemap = NULL; if (inodedep->id_state & ONWORKLIST) WORKLIST_REMOVE(&inodedep->id_list); if (inodedep->id_savedino1 != NULL) { free(inodedep->id_savedino1, M_SAVEDINO); inodedep->id_savedino1 = NULL; } if (free_inodedep(inodedep) == 0) panic("check_inode_unwritten: busy inode"); return (1); } static int check_inodedep_free(inodedep) struct inodedep *inodedep; { LOCK_OWNED(VFSTOUFS(inodedep->id_list.wk_mp)); if ((inodedep->id_state & ALLCOMPLETE) != ALLCOMPLETE || !LIST_EMPTY(&inodedep->id_dirremhd) || !LIST_EMPTY(&inodedep->id_pendinghd) || !LIST_EMPTY(&inodedep->id_bufwait) || !LIST_EMPTY(&inodedep->id_inowait) || !TAILQ_EMPTY(&inodedep->id_inoreflst) || !TAILQ_EMPTY(&inodedep->id_inoupdt) || !TAILQ_EMPTY(&inodedep->id_newinoupdt) || !TAILQ_EMPTY(&inodedep->id_extupdt) || !TAILQ_EMPTY(&inodedep->id_newextupdt) || !TAILQ_EMPTY(&inodedep->id_freeblklst) || inodedep->id_mkdiradd != NULL || inodedep->id_nlinkdelta != 0 || inodedep->id_savedino1 != NULL) return (0); return (1); } /* * Try to free an inodedep structure. Return 1 if it could be freed. */ static int free_inodedep(inodedep) struct inodedep *inodedep; { LOCK_OWNED(VFSTOUFS(inodedep->id_list.wk_mp)); if ((inodedep->id_state & (ONWORKLIST | UNLINKED)) != 0 || !check_inodedep_free(inodedep)) return (0); if (inodedep->id_state & ONDEPLIST) LIST_REMOVE(inodedep, id_deps); LIST_REMOVE(inodedep, id_hash); WORKITEM_FREE(inodedep, D_INODEDEP); return (1); } /* * Free the block referenced by a freework structure. The parent freeblks * structure is released and completed when the final cg bitmap reaches * the disk. This routine may be freeing a jnewblk which never made it to * disk in which case we do not have to wait as the operation is undone * in memory immediately. */ static void freework_freeblock(freework, key) struct freework *freework; u_long key; { struct freeblks *freeblks; struct jnewblk *jnewblk; struct ufsmount *ump; struct workhead wkhd; struct fs *fs; int bsize; int needj; ump = VFSTOUFS(freework->fw_list.wk_mp); LOCK_OWNED(ump); /* * Handle partial truncate separately. */ if (freework->fw_indir) { complete_trunc_indir(freework); return; } freeblks = freework->fw_freeblks; fs = ump->um_fs; needj = MOUNTEDSUJ(freeblks->fb_list.wk_mp) != 0; bsize = lfragtosize(fs, freework->fw_frags); LIST_INIT(&wkhd); /* * DEPCOMPLETE is cleared in indirblk_insert() if the block lives * on the indirblk hashtable and prevents premature freeing. */ freework->fw_state |= DEPCOMPLETE; /* * SUJ needs to wait for the segment referencing freed indirect * blocks to expire so that we know the checker will not confuse * a re-allocated indirect block with its old contents. */ if (needj && freework->fw_lbn <= -UFS_NDADDR) indirblk_insert(freework); /* * If we are canceling an existing jnewblk pass it to the free * routine, otherwise pass the freeblk which will ultimately * release the freeblks. If we're not journaling, we can just * free the freeblks immediately. */ jnewblk = freework->fw_jnewblk; if (jnewblk != NULL) { cancel_jnewblk(jnewblk, &wkhd); needj = 0; } else if (needj) { freework->fw_state |= DELAYEDFREE; freeblks->fb_cgwait++; WORKLIST_INSERT(&wkhd, &freework->fw_list); } FREE_LOCK(ump); freeblks_free(ump, freeblks, btodb(bsize)); CTR4(KTR_SUJ, "freework_freeblock: ino %jd blkno %jd lbn %jd size %d", freeblks->fb_inum, freework->fw_blkno, freework->fw_lbn, bsize); ffs_blkfree(ump, fs, freeblks->fb_devvp, freework->fw_blkno, bsize, freeblks->fb_inum, freeblks->fb_vtype, &wkhd, key); ACQUIRE_LOCK(ump); /* * The jnewblk will be discarded and the bits in the map never * made it to disk. We can immediately free the freeblk. */ if (needj == 0) handle_written_freework(freework); } /* * We enqueue freework items that need processing back on the freeblks and * add the freeblks to the worklist. This makes it easier to find all work * required to flush a truncation in process_truncates(). */ static void freework_enqueue(freework) struct freework *freework; { struct freeblks *freeblks; freeblks = freework->fw_freeblks; if ((freework->fw_state & INPROGRESS) == 0) WORKLIST_INSERT(&freeblks->fb_freeworkhd, &freework->fw_list); if ((freeblks->fb_state & (ONWORKLIST | INPROGRESS | ALLCOMPLETE)) == ALLCOMPLETE && LIST_EMPTY(&freeblks->fb_jblkdephd)) add_to_worklist(&freeblks->fb_list, WK_NODELAY); } /* * Start, continue, or finish the process of freeing an indirect block tree. * The free operation may be paused at any point with fw_off containing the * offset to restart from. This enables us to implement some flow control * for large truncates which may fan out and generate a huge number of * dependencies. */ static void handle_workitem_indirblk(freework) struct freework *freework; { struct freeblks *freeblks; struct ufsmount *ump; struct fs *fs; freeblks = freework->fw_freeblks; ump = VFSTOUFS(freeblks->fb_list.wk_mp); fs = ump->um_fs; if (freework->fw_state & DEPCOMPLETE) { handle_written_freework(freework); return; } if (freework->fw_off == NINDIR(fs)) { freework_freeblock(freework, SINGLETON_KEY); return; } freework->fw_state |= INPROGRESS; FREE_LOCK(ump); indir_trunc(freework, fsbtodb(fs, freework->fw_blkno), freework->fw_lbn); ACQUIRE_LOCK(ump); } /* * Called when a freework structure attached to a cg buf is written. The * ref on either the parent or the freeblks structure is released and * the freeblks is added back to the worklist if there is more work to do. */ static void handle_written_freework(freework) struct freework *freework; { struct freeblks *freeblks; struct freework *parent; freeblks = freework->fw_freeblks; parent = freework->fw_parent; if (freework->fw_state & DELAYEDFREE) freeblks->fb_cgwait--; freework->fw_state |= COMPLETE; if ((freework->fw_state & ALLCOMPLETE) == ALLCOMPLETE) WORKITEM_FREE(freework, D_FREEWORK); if (parent) { if (--parent->fw_ref == 0) freework_enqueue(parent); return; } if (--freeblks->fb_ref != 0) return; if ((freeblks->fb_state & (ALLCOMPLETE | ONWORKLIST | INPROGRESS)) == ALLCOMPLETE && LIST_EMPTY(&freeblks->fb_jblkdephd)) add_to_worklist(&freeblks->fb_list, WK_NODELAY); } /* * This workitem routine performs the block de-allocation. * The workitem is added to the pending list after the updated * inode block has been written to disk. As mentioned above, * checks regarding the number of blocks de-allocated (compared * to the number of blocks allocated for the file) are also * performed in this function. */ static int handle_workitem_freeblocks(freeblks, flags) struct freeblks *freeblks; int flags; { struct freework *freework; struct newblk *newblk; struct allocindir *aip; struct ufsmount *ump; struct worklist *wk; u_long key; KASSERT(LIST_EMPTY(&freeblks->fb_jblkdephd), ("handle_workitem_freeblocks: Journal entries not written.")); ump = VFSTOUFS(freeblks->fb_list.wk_mp); key = ffs_blkrelease_start(ump, freeblks->fb_devvp, freeblks->fb_inum); ACQUIRE_LOCK(ump); while ((wk = LIST_FIRST(&freeblks->fb_freeworkhd)) != NULL) { WORKLIST_REMOVE(wk); switch (wk->wk_type) { case D_DIRREM: wk->wk_state |= COMPLETE; add_to_worklist(wk, 0); continue; case D_ALLOCDIRECT: free_newblk(WK_NEWBLK(wk)); continue; case D_ALLOCINDIR: aip = WK_ALLOCINDIR(wk); freework = NULL; if (aip->ai_state & DELAYEDFREE) { FREE_LOCK(ump); freework = newfreework(ump, freeblks, NULL, aip->ai_lbn, aip->ai_newblkno, ump->um_fs->fs_frag, 0, 0); ACQUIRE_LOCK(ump); } newblk = WK_NEWBLK(wk); if (newblk->nb_jnewblk) { freework->fw_jnewblk = newblk->nb_jnewblk; newblk->nb_jnewblk->jn_dep = &freework->fw_list; newblk->nb_jnewblk = NULL; } free_newblk(newblk); continue; case D_FREEWORK: freework = WK_FREEWORK(wk); if (freework->fw_lbn <= -UFS_NDADDR) handle_workitem_indirblk(freework); else freework_freeblock(freework, key); continue; default: panic("handle_workitem_freeblocks: Unknown type %s", TYPENAME(wk->wk_type)); } } if (freeblks->fb_ref != 0) { freeblks->fb_state &= ~INPROGRESS; wake_worklist(&freeblks->fb_list); freeblks = NULL; } FREE_LOCK(ump); ffs_blkrelease_finish(ump, key); if (freeblks) return handle_complete_freeblocks(freeblks, flags); return (0); } /* * Handle completion of block free via truncate. This allows fs_pending * to track the actual free block count more closely than if we only updated * it at the end. We must be careful to handle cases where the block count * on free was incorrect. */ static void freeblks_free(ump, freeblks, blocks) struct ufsmount *ump; struct freeblks *freeblks; int blocks; { struct fs *fs; ufs2_daddr_t remain; UFS_LOCK(ump); remain = -freeblks->fb_chkcnt; freeblks->fb_chkcnt += blocks; if (remain > 0) { if (remain < blocks) blocks = remain; fs = ump->um_fs; fs->fs_pendingblocks -= blocks; } UFS_UNLOCK(ump); } /* * Once all of the freework workitems are complete we can retire the * freeblocks dependency and any journal work awaiting completion. This * can not be called until all other dependencies are stable on disk. */ static int handle_complete_freeblocks(freeblks, flags) struct freeblks *freeblks; int flags; { struct inodedep *inodedep; struct inode *ip; struct vnode *vp; struct fs *fs; struct ufsmount *ump; ufs2_daddr_t spare; ump = VFSTOUFS(freeblks->fb_list.wk_mp); fs = ump->um_fs; flags = LK_EXCLUSIVE | flags; spare = freeblks->fb_chkcnt; /* * If we did not release the expected number of blocks we may have * to adjust the inode block count here. Only do so if it wasn't * a truncation to zero and the modrev still matches. */ if (spare && freeblks->fb_len != 0) { if (ffs_vgetf(freeblks->fb_list.wk_mp, freeblks->fb_inum, flags, &vp, FFSV_FORCEINSMQ | FFSV_FORCEINODEDEP) != 0) return (EBUSY); ip = VTOI(vp); if (ip->i_mode == 0) { vgone(vp); } else if (DIP(ip, i_modrev) == freeblks->fb_modrev) { DIP_SET(ip, i_blocks, DIP(ip, i_blocks) - spare); UFS_INODE_SET_FLAG(ip, IN_CHANGE); /* * We must wait so this happens before the * journal is reclaimed. */ ffs_update(vp, 1); } vput(vp); } if (spare < 0) { UFS_LOCK(ump); fs->fs_pendingblocks += spare; UFS_UNLOCK(ump); } #ifdef QUOTA /* Handle spare. */ if (spare) quotaadj(freeblks->fb_quota, ump, -spare); quotarele(freeblks->fb_quota); #endif ACQUIRE_LOCK(ump); if (freeblks->fb_state & ONDEPLIST) { inodedep_lookup(freeblks->fb_list.wk_mp, freeblks->fb_inum, 0, &inodedep); TAILQ_REMOVE(&inodedep->id_freeblklst, freeblks, fb_next); freeblks->fb_state &= ~ONDEPLIST; if (TAILQ_EMPTY(&inodedep->id_freeblklst)) free_inodedep(inodedep); } /* * All of the freeblock deps must be complete prior to this call * so it's now safe to complete earlier outstanding journal entries. */ handle_jwork(&freeblks->fb_jwork); WORKITEM_FREE(freeblks, D_FREEBLKS); FREE_LOCK(ump); return (0); } /* * Release blocks associated with the freeblks and stored in the indirect * block dbn. If level is greater than SINGLE, the block is an indirect block * and recursive calls to indirtrunc must be used to cleanse other indirect * blocks. * * This handles partial and complete truncation of blocks. Partial is noted * with goingaway == 0. In this case the freework is completed after the * zero'd indirects are written to disk. For full truncation the freework * is completed after the block is freed. */ static void indir_trunc(freework, dbn, lbn) struct freework *freework; ufs2_daddr_t dbn; ufs_lbn_t lbn; { struct freework *nfreework; struct workhead wkhd; struct freeblks *freeblks; struct buf *bp; struct fs *fs; struct indirdep *indirdep; struct mount *mp; struct ufsmount *ump; ufs1_daddr_t *bap1; ufs2_daddr_t nb, nnb, *bap2; ufs_lbn_t lbnadd, nlbn; u_long key; int nblocks, ufs1fmt, freedblocks; int goingaway, freedeps, needj, level, cnt, i, error; freeblks = freework->fw_freeblks; mp = freeblks->fb_list.wk_mp; ump = VFSTOUFS(mp); fs = ump->um_fs; /* * Get buffer of block pointers to be freed. There are three cases: * * 1) Partial truncate caches the indirdep pointer in the freework * which provides us a back copy to the save bp which holds the * pointers we want to clear. When this completes the zero * pointers are written to the real copy. * 2) The indirect is being completely truncated, cancel_indirdep() * eliminated the real copy and placed the indirdep on the saved * copy. The indirdep and buf are discarded when this completes. * 3) The indirect was not in memory, we read a copy off of the disk * using the devvp and drop and invalidate the buffer when we're * done. */ goingaway = 1; indirdep = NULL; if (freework->fw_indir != NULL) { goingaway = 0; indirdep = freework->fw_indir; bp = indirdep->ir_savebp; if (bp == NULL || bp->b_blkno != dbn) panic("indir_trunc: Bad saved buf %p blkno %jd", bp, (intmax_t)dbn); } else if ((bp = incore(&freeblks->fb_devvp->v_bufobj, dbn)) != NULL) { /* * The lock prevents the buf dep list from changing and * indirects on devvp should only ever have one dependency. */ indirdep = WK_INDIRDEP(LIST_FIRST(&bp->b_dep)); if (indirdep == NULL || (indirdep->ir_state & GOINGAWAY) == 0) panic("indir_trunc: Bad indirdep %p from buf %p", indirdep, bp); } else { error = ffs_breadz(ump, freeblks->fb_devvp, dbn, dbn, (int)fs->fs_bsize, NULL, NULL, 0, NOCRED, 0, NULL, &bp); if (error) return; } ACQUIRE_LOCK(ump); /* Protects against a race with complete_trunc_indir(). */ freework->fw_state &= ~INPROGRESS; /* * If we have an indirdep we need to enforce the truncation order * and discard it when it is complete. */ if (indirdep) { if (freework != TAILQ_FIRST(&indirdep->ir_trunc) && !TAILQ_EMPTY(&indirdep->ir_trunc)) { /* * Add the complete truncate to the list on the * indirdep to enforce in-order processing. */ if (freework->fw_indir == NULL) TAILQ_INSERT_TAIL(&indirdep->ir_trunc, freework, fw_next); FREE_LOCK(ump); return; } /* * If we're goingaway, free the indirdep. Otherwise it will * linger until the write completes. */ if (goingaway) { KASSERT(indirdep->ir_savebp == bp, ("indir_trunc: losing ir_savebp %p", indirdep->ir_savebp)); indirdep->ir_savebp = NULL; free_indirdep(indirdep); } } FREE_LOCK(ump); /* Initialize pointers depending on block size. */ if (ump->um_fstype == UFS1) { bap1 = (ufs1_daddr_t *)bp->b_data; nb = bap1[freework->fw_off]; ufs1fmt = 1; bap2 = NULL; } else { bap2 = (ufs2_daddr_t *)bp->b_data; nb = bap2[freework->fw_off]; ufs1fmt = 0; bap1 = NULL; } level = lbn_level(lbn); needj = MOUNTEDSUJ(UFSTOVFS(ump)) != 0; lbnadd = lbn_offset(fs, level); nblocks = btodb(fs->fs_bsize); nfreework = freework; freedeps = 0; cnt = 0; /* * Reclaim blocks. Traverses into nested indirect levels and * arranges for the current level to be freed when subordinates * are free when journaling. */ key = ffs_blkrelease_start(ump, freeblks->fb_devvp, freeblks->fb_inum); for (i = freework->fw_off; i < NINDIR(fs); i++, nb = nnb) { if (UFS_CHECK_BLKNO(mp, freeblks->fb_inum, nb, fs->fs_bsize) != 0) nb = 0; if (i != NINDIR(fs) - 1) { if (ufs1fmt) nnb = bap1[i+1]; else nnb = bap2[i+1]; } else nnb = 0; if (nb == 0) continue; cnt++; if (level != 0) { nlbn = (lbn + 1) - (i * lbnadd); if (needj != 0) { nfreework = newfreework(ump, freeblks, freework, nlbn, nb, fs->fs_frag, 0, 0); freedeps++; } indir_trunc(nfreework, fsbtodb(fs, nb), nlbn); } else { struct freedep *freedep; /* * Attempt to aggregate freedep dependencies for * all blocks being released to the same CG. */ LIST_INIT(&wkhd); if (needj != 0 && (nnb == 0 || (dtog(fs, nb) != dtog(fs, nnb)))) { freedep = newfreedep(freework); WORKLIST_INSERT_UNLOCKED(&wkhd, &freedep->fd_list); freedeps++; } CTR3(KTR_SUJ, "indir_trunc: ino %jd blkno %jd size %d", freeblks->fb_inum, nb, fs->fs_bsize); ffs_blkfree(ump, fs, freeblks->fb_devvp, nb, fs->fs_bsize, freeblks->fb_inum, freeblks->fb_vtype, &wkhd, key); } } ffs_blkrelease_finish(ump, key); if (goingaway) { bp->b_flags |= B_INVAL | B_NOCACHE; brelse(bp); } freedblocks = 0; if (level == 0) freedblocks = (nblocks * cnt); if (needj == 0) freedblocks += nblocks; freeblks_free(ump, freeblks, freedblocks); /* * If we are journaling set up the ref counts and offset so this * indirect can be completed when its children are free. */ if (needj) { ACQUIRE_LOCK(ump); freework->fw_off = i; freework->fw_ref += freedeps; freework->fw_ref -= NINDIR(fs) + 1; if (level == 0) freeblks->fb_cgwait += freedeps; if (freework->fw_ref == 0) freework_freeblock(freework, SINGLETON_KEY); FREE_LOCK(ump); return; } /* * If we're not journaling we can free the indirect now. */ dbn = dbtofsb(fs, dbn); CTR3(KTR_SUJ, "indir_trunc 2: ino %jd blkno %jd size %d", freeblks->fb_inum, dbn, fs->fs_bsize); ffs_blkfree(ump, fs, freeblks->fb_devvp, dbn, fs->fs_bsize, freeblks->fb_inum, freeblks->fb_vtype, NULL, SINGLETON_KEY); /* Non SUJ softdep does single-threaded truncations. */ if (freework->fw_blkno == dbn) { freework->fw_state |= ALLCOMPLETE; ACQUIRE_LOCK(ump); handle_written_freework(freework); FREE_LOCK(ump); } return; } /* * Cancel an allocindir when it is removed via truncation. When bp is not * NULL the indirect never appeared on disk and is scheduled to be freed * independently of the indir so we can more easily track journal work. */ static void cancel_allocindir(aip, bp, freeblks, trunc) struct allocindir *aip; struct buf *bp; struct freeblks *freeblks; int trunc; { struct indirdep *indirdep; struct freefrag *freefrag; struct newblk *newblk; newblk = (struct newblk *)aip; LIST_REMOVE(aip, ai_next); /* * We must eliminate the pointer in bp if it must be freed on its * own due to partial truncate or pending journal work. */ if (bp && (trunc || newblk->nb_jnewblk)) { /* * Clear the pointer and mark the aip to be freed * directly if it never existed on disk. */ aip->ai_state |= DELAYEDFREE; indirdep = aip->ai_indirdep; if (indirdep->ir_state & UFS1FMT) ((ufs1_daddr_t *)bp->b_data)[aip->ai_offset] = 0; else ((ufs2_daddr_t *)bp->b_data)[aip->ai_offset] = 0; } /* * When truncating the previous pointer will be freed via * savedbp. Eliminate the freefrag which would dup free. */ if (trunc && (freefrag = newblk->nb_freefrag) != NULL) { newblk->nb_freefrag = NULL; if (freefrag->ff_jdep) cancel_jfreefrag( WK_JFREEFRAG(freefrag->ff_jdep)); jwork_move(&freeblks->fb_jwork, &freefrag->ff_jwork); WORKITEM_FREE(freefrag, D_FREEFRAG); } /* * If the journal hasn't been written the jnewblk must be passed * to the call to ffs_blkfree that reclaims the space. We accomplish * this by leaving the journal dependency on the newblk to be freed * when a freework is created in handle_workitem_freeblocks(). */ cancel_newblk(newblk, NULL, &freeblks->fb_jwork); WORKLIST_INSERT(&freeblks->fb_freeworkhd, &newblk->nb_list); } /* * Create the mkdir dependencies for . and .. in a new directory. Link them * in to a newdirblk so any subsequent additions are tracked properly. The * caller is responsible for adding the mkdir1 dependency to the journal * and updating id_mkdiradd. This function returns with the per-filesystem * lock held. */ static struct mkdir * setup_newdir(dap, newinum, dinum, newdirbp, mkdirp) struct diradd *dap; ino_t newinum; ino_t dinum; struct buf *newdirbp; struct mkdir **mkdirp; { struct newblk *newblk; struct pagedep *pagedep; struct inodedep *inodedep; struct newdirblk *newdirblk; struct mkdir *mkdir1, *mkdir2; struct worklist *wk; struct jaddref *jaddref; struct ufsmount *ump; struct mount *mp; mp = dap->da_list.wk_mp; ump = VFSTOUFS(mp); newdirblk = malloc(sizeof(struct newdirblk), M_NEWDIRBLK, M_SOFTDEP_FLAGS); workitem_alloc(&newdirblk->db_list, D_NEWDIRBLK, mp); LIST_INIT(&newdirblk->db_mkdir); mkdir1 = malloc(sizeof(struct mkdir), M_MKDIR, M_SOFTDEP_FLAGS); workitem_alloc(&mkdir1->md_list, D_MKDIR, mp); mkdir1->md_state = ATTACHED | MKDIR_BODY; mkdir1->md_diradd = dap; mkdir1->md_jaddref = NULL; mkdir2 = malloc(sizeof(struct mkdir), M_MKDIR, M_SOFTDEP_FLAGS); workitem_alloc(&mkdir2->md_list, D_MKDIR, mp); mkdir2->md_state = ATTACHED | MKDIR_PARENT; mkdir2->md_diradd = dap; mkdir2->md_jaddref = NULL; if (MOUNTEDSUJ(mp) == 0) { mkdir1->md_state |= DEPCOMPLETE; mkdir2->md_state |= DEPCOMPLETE; } /* * Dependency on "." and ".." being written to disk. */ mkdir1->md_buf = newdirbp; ACQUIRE_LOCK(VFSTOUFS(mp)); LIST_INSERT_HEAD(&ump->softdep_mkdirlisthd, mkdir1, md_mkdirs); /* * We must link the pagedep, allocdirect, and newdirblk for * the initial file page so the pointer to the new directory * is not written until the directory contents are live and * any subsequent additions are not marked live until the * block is reachable via the inode. */ if (pagedep_lookup(mp, newdirbp, newinum, 0, 0, &pagedep) == 0) panic("setup_newdir: lost pagedep"); LIST_FOREACH(wk, &newdirbp->b_dep, wk_list) if (wk->wk_type == D_ALLOCDIRECT) break; if (wk == NULL) panic("setup_newdir: lost allocdirect"); if (pagedep->pd_state & NEWBLOCK) panic("setup_newdir: NEWBLOCK already set"); newblk = WK_NEWBLK(wk); pagedep->pd_state |= NEWBLOCK; pagedep->pd_newdirblk = newdirblk; newdirblk->db_pagedep = pagedep; WORKLIST_INSERT(&newblk->nb_newdirblk, &newdirblk->db_list); WORKLIST_INSERT(&newdirblk->db_mkdir, &mkdir1->md_list); /* * Look up the inodedep for the parent directory so that we * can link mkdir2 into the pending dotdot jaddref or * the inode write if there is none. If the inode is * ALLCOMPLETE and no jaddref is present all dependencies have * been satisfied and mkdir2 can be freed. */ inodedep_lookup(mp, dinum, 0, &inodedep); if (MOUNTEDSUJ(mp)) { if (inodedep == NULL) panic("setup_newdir: Lost parent."); jaddref = (struct jaddref *)TAILQ_LAST(&inodedep->id_inoreflst, inoreflst); KASSERT(jaddref != NULL && jaddref->ja_parent == newinum && (jaddref->ja_state & MKDIR_PARENT), ("setup_newdir: bad dotdot jaddref %p", jaddref)); LIST_INSERT_HEAD(&ump->softdep_mkdirlisthd, mkdir2, md_mkdirs); mkdir2->md_jaddref = jaddref; jaddref->ja_mkdir = mkdir2; } else if (inodedep == NULL || (inodedep->id_state & ALLCOMPLETE) == ALLCOMPLETE) { dap->da_state &= ~MKDIR_PARENT; WORKITEM_FREE(mkdir2, D_MKDIR); mkdir2 = NULL; } else { LIST_INSERT_HEAD(&ump->softdep_mkdirlisthd, mkdir2, md_mkdirs); WORKLIST_INSERT(&inodedep->id_bufwait, &mkdir2->md_list); } *mkdirp = mkdir2; return (mkdir1); } /* * Directory entry addition dependencies. * * When adding a new directory entry, the inode (with its incremented link * count) must be written to disk before the directory entry's pointer to it. * Also, if the inode is newly allocated, the corresponding freemap must be * updated (on disk) before the directory entry's pointer. These requirements * are met via undo/redo on the directory entry's pointer, which consists * simply of the inode number. * * As directory entries are added and deleted, the free space within a * directory block can become fragmented. The ufs filesystem will compact * a fragmented directory block to make space for a new entry. When this * occurs, the offsets of previously added entries change. Any "diradd" * dependency structures corresponding to these entries must be updated with * the new offsets. */ /* * This routine is called after the in-memory inode's link * count has been incremented, but before the directory entry's * pointer to the inode has been set. */ int softdep_setup_directory_add(bp, dp, diroffset, newinum, newdirbp, isnewblk) struct buf *bp; /* buffer containing directory block */ struct inode *dp; /* inode for directory */ off_t diroffset; /* offset of new entry in directory */ ino_t newinum; /* inode referenced by new directory entry */ struct buf *newdirbp; /* non-NULL => contents of new mkdir */ int isnewblk; /* entry is in a newly allocated block */ { int offset; /* offset of new entry within directory block */ ufs_lbn_t lbn; /* block in directory containing new entry */ struct fs *fs; struct diradd *dap; struct newblk *newblk; struct pagedep *pagedep; struct inodedep *inodedep; struct newdirblk *newdirblk; struct mkdir *mkdir1, *mkdir2; struct jaddref *jaddref; struct ufsmount *ump; struct mount *mp; int isindir; mp = ITOVFS(dp); ump = VFSTOUFS(mp); KASSERT(MOUNTEDSOFTDEP(mp) != 0, ("softdep_setup_directory_add called on non-softdep filesystem")); /* * Whiteouts have no dependencies. */ if (newinum == UFS_WINO) { if (newdirbp != NULL) bdwrite(newdirbp); return (0); } jaddref = NULL; mkdir1 = mkdir2 = NULL; fs = ump->um_fs; lbn = lblkno(fs, diroffset); offset = blkoff(fs, diroffset); dap = malloc(sizeof(struct diradd), M_DIRADD, M_SOFTDEP_FLAGS|M_ZERO); workitem_alloc(&dap->da_list, D_DIRADD, mp); dap->da_offset = offset; dap->da_newinum = newinum; dap->da_state = ATTACHED; LIST_INIT(&dap->da_jwork); isindir = bp->b_lblkno >= UFS_NDADDR; newdirblk = NULL; if (isnewblk && (isindir ? blkoff(fs, diroffset) : fragoff(fs, diroffset)) == 0) { newdirblk = malloc(sizeof(struct newdirblk), M_NEWDIRBLK, M_SOFTDEP_FLAGS); workitem_alloc(&newdirblk->db_list, D_NEWDIRBLK, mp); LIST_INIT(&newdirblk->db_mkdir); } /* * If we're creating a new directory setup the dependencies and set * the dap state to wait for them. Otherwise it's COMPLETE and * we can move on. */ if (newdirbp == NULL) { dap->da_state |= DEPCOMPLETE; ACQUIRE_LOCK(ump); } else { dap->da_state |= MKDIR_BODY | MKDIR_PARENT; mkdir1 = setup_newdir(dap, newinum, dp->i_number, newdirbp, &mkdir2); } /* * Link into parent directory pagedep to await its being written. */ pagedep_lookup(mp, bp, dp->i_number, lbn, DEPALLOC, &pagedep); #ifdef INVARIANTS if (diradd_lookup(pagedep, offset) != NULL) panic("softdep_setup_directory_add: %p already at off %d\n", diradd_lookup(pagedep, offset), offset); #endif dap->da_pagedep = pagedep; LIST_INSERT_HEAD(&pagedep->pd_diraddhd[DIRADDHASH(offset)], dap, da_pdlist); inodedep_lookup(mp, newinum, DEPALLOC, &inodedep); /* * If we're journaling, link the diradd into the jaddref so it * may be completed after the journal entry is written. Otherwise, * link the diradd into its inodedep. If the inode is not yet * written place it on the bufwait list, otherwise do the post-inode * write processing to put it on the id_pendinghd list. */ if (MOUNTEDSUJ(mp)) { jaddref = (struct jaddref *)TAILQ_LAST(&inodedep->id_inoreflst, inoreflst); KASSERT(jaddref != NULL && jaddref->ja_parent == dp->i_number, ("softdep_setup_directory_add: bad jaddref %p", jaddref)); jaddref->ja_diroff = diroffset; jaddref->ja_diradd = dap; add_to_journal(&jaddref->ja_list); } else if ((inodedep->id_state & ALLCOMPLETE) == ALLCOMPLETE) diradd_inode_written(dap, inodedep); else WORKLIST_INSERT(&inodedep->id_bufwait, &dap->da_list); /* * Add the journal entries for . and .. links now that the primary * link is written. */ if (mkdir1 != NULL && MOUNTEDSUJ(mp)) { jaddref = (struct jaddref *)TAILQ_PREV(&jaddref->ja_ref, inoreflst, if_deps); KASSERT(jaddref != NULL && jaddref->ja_ino == jaddref->ja_parent && (jaddref->ja_state & MKDIR_BODY), ("softdep_setup_directory_add: bad dot jaddref %p", jaddref)); mkdir1->md_jaddref = jaddref; jaddref->ja_mkdir = mkdir1; /* * It is important that the dotdot journal entry * is added prior to the dot entry since dot writes * both the dot and dotdot links. These both must * be added after the primary link for the journal * to remain consistent. */ add_to_journal(&mkdir2->md_jaddref->ja_list); add_to_journal(&jaddref->ja_list); } /* * If we are adding a new directory remember this diradd so that if * we rename it we can keep the dot and dotdot dependencies. If * we are adding a new name for an inode that has a mkdiradd we * must be in rename and we have to move the dot and dotdot * dependencies to this new name. The old name is being orphaned * soon. */ if (mkdir1 != NULL) { if (inodedep->id_mkdiradd != NULL) panic("softdep_setup_directory_add: Existing mkdir"); inodedep->id_mkdiradd = dap; } else if (inodedep->id_mkdiradd) merge_diradd(inodedep, dap); if (newdirblk != NULL) { /* * There is nothing to do if we are already tracking * this block. */ if ((pagedep->pd_state & NEWBLOCK) != 0) { WORKITEM_FREE(newdirblk, D_NEWDIRBLK); FREE_LOCK(ump); return (0); } if (newblk_lookup(mp, dbtofsb(fs, bp->b_blkno), 0, &newblk) == 0) panic("softdep_setup_directory_add: lost entry"); WORKLIST_INSERT(&newblk->nb_newdirblk, &newdirblk->db_list); pagedep->pd_state |= NEWBLOCK; pagedep->pd_newdirblk = newdirblk; newdirblk->db_pagedep = pagedep; FREE_LOCK(ump); /* * If we extended into an indirect signal direnter to sync. */ if (isindir) return (1); return (0); } FREE_LOCK(ump); return (0); } /* * This procedure is called to change the offset of a directory * entry when compacting a directory block which must be owned * exclusively by the caller. Note that the actual entry movement * must be done in this procedure to ensure that no I/O completions * occur while the move is in progress. */ void softdep_change_directoryentry_offset(bp, dp, base, oldloc, newloc, entrysize) struct buf *bp; /* Buffer holding directory block. */ struct inode *dp; /* inode for directory */ caddr_t base; /* address of dp->i_offset */ caddr_t oldloc; /* address of old directory location */ caddr_t newloc; /* address of new directory location */ int entrysize; /* size of directory entry */ { int offset, oldoffset, newoffset; struct pagedep *pagedep; struct jmvref *jmvref; struct diradd *dap; struct direct *de; struct mount *mp; struct ufsmount *ump; ufs_lbn_t lbn; int flags; mp = ITOVFS(dp); ump = VFSTOUFS(mp); KASSERT(MOUNTEDSOFTDEP(mp) != 0, ("softdep_change_directoryentry_offset called on " "non-softdep filesystem")); de = (struct direct *)oldloc; jmvref = NULL; flags = 0; /* * Moves are always journaled as it would be too complex to * determine if any affected adds or removes are present in the * journal. */ if (MOUNTEDSUJ(mp)) { flags = DEPALLOC; jmvref = newjmvref(dp, de->d_ino, I_OFFSET(dp) + (oldloc - base), I_OFFSET(dp) + (newloc - base)); } lbn = lblkno(ump->um_fs, I_OFFSET(dp)); offset = blkoff(ump->um_fs, I_OFFSET(dp)); oldoffset = offset + (oldloc - base); newoffset = offset + (newloc - base); ACQUIRE_LOCK(ump); if (pagedep_lookup(mp, bp, dp->i_number, lbn, flags, &pagedep) == 0) goto done; dap = diradd_lookup(pagedep, oldoffset); if (dap) { dap->da_offset = newoffset; newoffset = DIRADDHASH(newoffset); oldoffset = DIRADDHASH(oldoffset); if ((dap->da_state & ALLCOMPLETE) != ALLCOMPLETE && newoffset != oldoffset) { LIST_REMOVE(dap, da_pdlist); LIST_INSERT_HEAD(&pagedep->pd_diraddhd[newoffset], dap, da_pdlist); } } done: if (jmvref) { jmvref->jm_pagedep = pagedep; LIST_INSERT_HEAD(&pagedep->pd_jmvrefhd, jmvref, jm_deps); add_to_journal(&jmvref->jm_list); } bcopy(oldloc, newloc, entrysize); FREE_LOCK(ump); } /* * Move the mkdir dependencies and journal work from one diradd to another * when renaming a directory. The new name must depend on the mkdir deps * completing as the old name did. Directories can only have one valid link * at a time so one must be canonical. */ static void merge_diradd(inodedep, newdap) struct inodedep *inodedep; struct diradd *newdap; { struct diradd *olddap; struct mkdir *mkdir, *nextmd; struct ufsmount *ump; short state; olddap = inodedep->id_mkdiradd; inodedep->id_mkdiradd = newdap; if ((olddap->da_state & (MKDIR_PARENT | MKDIR_BODY)) != 0) { newdap->da_state &= ~DEPCOMPLETE; ump = VFSTOUFS(inodedep->id_list.wk_mp); for (mkdir = LIST_FIRST(&ump->softdep_mkdirlisthd); mkdir; mkdir = nextmd) { nextmd = LIST_NEXT(mkdir, md_mkdirs); if (mkdir->md_diradd != olddap) continue; mkdir->md_diradd = newdap; state = mkdir->md_state & (MKDIR_PARENT | MKDIR_BODY); newdap->da_state |= state; olddap->da_state &= ~state; if ((olddap->da_state & (MKDIR_PARENT | MKDIR_BODY)) == 0) break; } if ((olddap->da_state & (MKDIR_PARENT | MKDIR_BODY)) != 0) panic("merge_diradd: unfound ref"); } /* * Any mkdir related journal items are not safe to be freed until * the new name is stable. */ jwork_move(&newdap->da_jwork, &olddap->da_jwork); olddap->da_state |= DEPCOMPLETE; complete_diradd(olddap); } /* * Move the diradd to the pending list when all diradd dependencies are * complete. */ static void complete_diradd(dap) struct diradd *dap; { struct pagedep *pagedep; if ((dap->da_state & ALLCOMPLETE) == ALLCOMPLETE) { if (dap->da_state & DIRCHG) pagedep = dap->da_previous->dm_pagedep; else pagedep = dap->da_pagedep; LIST_REMOVE(dap, da_pdlist); LIST_INSERT_HEAD(&pagedep->pd_pendinghd, dap, da_pdlist); } } /* * Cancel a diradd when a dirrem overlaps with it. We must cancel the journal * add entries and conditonally journal the remove. */ static void cancel_diradd(dap, dirrem, jremref, dotremref, dotdotremref) struct diradd *dap; struct dirrem *dirrem; struct jremref *jremref; struct jremref *dotremref; struct jremref *dotdotremref; { struct inodedep *inodedep; struct jaddref *jaddref; struct inoref *inoref; struct ufsmount *ump; struct mkdir *mkdir; /* * If no remove references were allocated we're on a non-journaled * filesystem and can skip the cancel step. */ if (jremref == NULL) { free_diradd(dap, NULL); return; } /* * Cancel the primary name an free it if it does not require * journaling. */ if (inodedep_lookup(dap->da_list.wk_mp, dap->da_newinum, 0, &inodedep) != 0) { /* Abort the addref that reference this diradd. */ TAILQ_FOREACH(inoref, &inodedep->id_inoreflst, if_deps) { if (inoref->if_list.wk_type != D_JADDREF) continue; jaddref = (struct jaddref *)inoref; if (jaddref->ja_diradd != dap) continue; if (cancel_jaddref(jaddref, inodedep, &dirrem->dm_jwork) == 0) { free_jremref(jremref); jremref = NULL; } break; } } /* * Cancel subordinate names and free them if they do not require * journaling. */ if ((dap->da_state & (MKDIR_PARENT | MKDIR_BODY)) != 0) { ump = VFSTOUFS(dap->da_list.wk_mp); LIST_FOREACH(mkdir, &ump->softdep_mkdirlisthd, md_mkdirs) { if (mkdir->md_diradd != dap) continue; if ((jaddref = mkdir->md_jaddref) == NULL) continue; mkdir->md_jaddref = NULL; if (mkdir->md_state & MKDIR_PARENT) { if (cancel_jaddref(jaddref, NULL, &dirrem->dm_jwork) == 0) { free_jremref(dotdotremref); dotdotremref = NULL; } } else { if (cancel_jaddref(jaddref, inodedep, &dirrem->dm_jwork) == 0) { free_jremref(dotremref); dotremref = NULL; } } } } if (jremref) journal_jremref(dirrem, jremref, inodedep); if (dotremref) journal_jremref(dirrem, dotremref, inodedep); if (dotdotremref) journal_jremref(dirrem, dotdotremref, NULL); jwork_move(&dirrem->dm_jwork, &dap->da_jwork); free_diradd(dap, &dirrem->dm_jwork); } /* * Free a diradd dependency structure. */ static void free_diradd(dap, wkhd) struct diradd *dap; struct workhead *wkhd; { struct dirrem *dirrem; struct pagedep *pagedep; struct inodedep *inodedep; struct mkdir *mkdir, *nextmd; struct ufsmount *ump; ump = VFSTOUFS(dap->da_list.wk_mp); LOCK_OWNED(ump); LIST_REMOVE(dap, da_pdlist); if (dap->da_state & ONWORKLIST) WORKLIST_REMOVE(&dap->da_list); if ((dap->da_state & DIRCHG) == 0) { pagedep = dap->da_pagedep; } else { dirrem = dap->da_previous; pagedep = dirrem->dm_pagedep; dirrem->dm_dirinum = pagedep->pd_ino; dirrem->dm_state |= COMPLETE; if (LIST_EMPTY(&dirrem->dm_jremrefhd)) add_to_worklist(&dirrem->dm_list, 0); } if (inodedep_lookup(pagedep->pd_list.wk_mp, dap->da_newinum, 0, &inodedep) != 0) if (inodedep->id_mkdiradd == dap) inodedep->id_mkdiradd = NULL; if ((dap->da_state & (MKDIR_PARENT | MKDIR_BODY)) != 0) { for (mkdir = LIST_FIRST(&ump->softdep_mkdirlisthd); mkdir; mkdir = nextmd) { nextmd = LIST_NEXT(mkdir, md_mkdirs); if (mkdir->md_diradd != dap) continue; dap->da_state &= ~(mkdir->md_state & (MKDIR_PARENT | MKDIR_BODY)); LIST_REMOVE(mkdir, md_mkdirs); if (mkdir->md_state & ONWORKLIST) WORKLIST_REMOVE(&mkdir->md_list); if (mkdir->md_jaddref != NULL) panic("free_diradd: Unexpected jaddref"); WORKITEM_FREE(mkdir, D_MKDIR); if ((dap->da_state & (MKDIR_PARENT | MKDIR_BODY)) == 0) break; } if ((dap->da_state & (MKDIR_PARENT | MKDIR_BODY)) != 0) panic("free_diradd: unfound ref"); } if (inodedep) free_inodedep(inodedep); /* * Free any journal segments waiting for the directory write. */ handle_jwork(&dap->da_jwork); WORKITEM_FREE(dap, D_DIRADD); } /* * Directory entry removal dependencies. * * When removing a directory entry, the entry's inode pointer must be * zero'ed on disk before the corresponding inode's link count is decremented * (possibly freeing the inode for re-use). This dependency is handled by * updating the directory entry but delaying the inode count reduction until * after the directory block has been written to disk. After this point, the * inode count can be decremented whenever it is convenient. */ /* * This routine should be called immediately after removing * a directory entry. The inode's link count should not be * decremented by the calling procedure -- the soft updates * code will do this task when it is safe. */ void softdep_setup_remove(bp, dp, ip, isrmdir) struct buf *bp; /* buffer containing directory block */ struct inode *dp; /* inode for the directory being modified */ struct inode *ip; /* inode for directory entry being removed */ int isrmdir; /* indicates if doing RMDIR */ { struct dirrem *dirrem, *prevdirrem; struct inodedep *inodedep; struct ufsmount *ump; int direct; ump = ITOUMP(ip); KASSERT(MOUNTEDSOFTDEP(UFSTOVFS(ump)) != 0, ("softdep_setup_remove called on non-softdep filesystem")); /* * Allocate a new dirrem if appropriate and ACQUIRE_LOCK. We want * newdirrem() to setup the full directory remove which requires * isrmdir > 1. */ dirrem = newdirrem(bp, dp, ip, isrmdir, &prevdirrem); /* * Add the dirrem to the inodedep's pending remove list for quick * discovery later. */ if (inodedep_lookup(UFSTOVFS(ump), ip->i_number, 0, &inodedep) == 0) panic("softdep_setup_remove: Lost inodedep."); KASSERT((inodedep->id_state & UNLINKED) == 0, ("inode unlinked")); dirrem->dm_state |= ONDEPLIST; LIST_INSERT_HEAD(&inodedep->id_dirremhd, dirrem, dm_inonext); /* * If the COMPLETE flag is clear, then there were no active * entries and we want to roll back to a zeroed entry until * the new inode is committed to disk. If the COMPLETE flag is * set then we have deleted an entry that never made it to * disk. If the entry we deleted resulted from a name change, * then the old name still resides on disk. We cannot delete * its inode (returned to us in prevdirrem) until the zeroed * directory entry gets to disk. The new inode has never been * referenced on the disk, so can be deleted immediately. */ if ((dirrem->dm_state & COMPLETE) == 0) { LIST_INSERT_HEAD(&dirrem->dm_pagedep->pd_dirremhd, dirrem, dm_next); FREE_LOCK(ump); } else { if (prevdirrem != NULL) LIST_INSERT_HEAD(&dirrem->dm_pagedep->pd_dirremhd, prevdirrem, dm_next); dirrem->dm_dirinum = dirrem->dm_pagedep->pd_ino; direct = LIST_EMPTY(&dirrem->dm_jremrefhd); FREE_LOCK(ump); if (direct) handle_workitem_remove(dirrem, 0); } } /* * Check for an entry matching 'offset' on both the pd_dirraddhd list and the * pd_pendinghd list of a pagedep. */ static struct diradd * diradd_lookup(pagedep, offset) struct pagedep *pagedep; int offset; { struct diradd *dap; LIST_FOREACH(dap, &pagedep->pd_diraddhd[DIRADDHASH(offset)], da_pdlist) if (dap->da_offset == offset) return (dap); LIST_FOREACH(dap, &pagedep->pd_pendinghd, da_pdlist) if (dap->da_offset == offset) return (dap); return (NULL); } /* * Search for a .. diradd dependency in a directory that is being removed. * If the directory was renamed to a new parent we have a diradd rather * than a mkdir for the .. entry. We need to cancel it now before * it is found in truncate(). */ static struct jremref * cancel_diradd_dotdot(ip, dirrem, jremref) struct inode *ip; struct dirrem *dirrem; struct jremref *jremref; { struct pagedep *pagedep; struct diradd *dap; struct worklist *wk; if (pagedep_lookup(ITOVFS(ip), NULL, ip->i_number, 0, 0, &pagedep) == 0) return (jremref); dap = diradd_lookup(pagedep, DOTDOT_OFFSET); if (dap == NULL) return (jremref); cancel_diradd(dap, dirrem, jremref, NULL, NULL); /* * Mark any journal work as belonging to the parent so it is freed * with the .. reference. */ LIST_FOREACH(wk, &dirrem->dm_jwork, wk_list) wk->wk_state |= MKDIR_PARENT; return (NULL); } /* * Cancel the MKDIR_PARENT mkdir component of a diradd when we're going to * replace it with a dirrem/diradd pair as a result of re-parenting a * directory. This ensures that we don't simultaneously have a mkdir and * a diradd for the same .. entry. */ static struct jremref * cancel_mkdir_dotdot(ip, dirrem, jremref) struct inode *ip; struct dirrem *dirrem; struct jremref *jremref; { struct inodedep *inodedep; struct jaddref *jaddref; struct ufsmount *ump; struct mkdir *mkdir; struct diradd *dap; struct mount *mp; mp = ITOVFS(ip); if (inodedep_lookup(mp, ip->i_number, 0, &inodedep) == 0) return (jremref); dap = inodedep->id_mkdiradd; if (dap == NULL || (dap->da_state & MKDIR_PARENT) == 0) return (jremref); ump = VFSTOUFS(inodedep->id_list.wk_mp); for (mkdir = LIST_FIRST(&ump->softdep_mkdirlisthd); mkdir; mkdir = LIST_NEXT(mkdir, md_mkdirs)) if (mkdir->md_diradd == dap && mkdir->md_state & MKDIR_PARENT) break; if (mkdir == NULL) panic("cancel_mkdir_dotdot: Unable to find mkdir\n"); if ((jaddref = mkdir->md_jaddref) != NULL) { mkdir->md_jaddref = NULL; jaddref->ja_state &= ~MKDIR_PARENT; if (inodedep_lookup(mp, jaddref->ja_ino, 0, &inodedep) == 0) panic("cancel_mkdir_dotdot: Lost parent inodedep"); if (cancel_jaddref(jaddref, inodedep, &dirrem->dm_jwork)) { journal_jremref(dirrem, jremref, inodedep); jremref = NULL; } } if (mkdir->md_state & ONWORKLIST) WORKLIST_REMOVE(&mkdir->md_list); mkdir->md_state |= ALLCOMPLETE; complete_mkdir(mkdir); return (jremref); } static void journal_jremref(dirrem, jremref, inodedep) struct dirrem *dirrem; struct jremref *jremref; struct inodedep *inodedep; { if (inodedep == NULL) if (inodedep_lookup(jremref->jr_list.wk_mp, jremref->jr_ref.if_ino, 0, &inodedep) == 0) panic("journal_jremref: Lost inodedep"); LIST_INSERT_HEAD(&dirrem->dm_jremrefhd, jremref, jr_deps); TAILQ_INSERT_TAIL(&inodedep->id_inoreflst, &jremref->jr_ref, if_deps); add_to_journal(&jremref->jr_list); } static void dirrem_journal(dirrem, jremref, dotremref, dotdotremref) struct dirrem *dirrem; struct jremref *jremref; struct jremref *dotremref; struct jremref *dotdotremref; { struct inodedep *inodedep; if (inodedep_lookup(jremref->jr_list.wk_mp, jremref->jr_ref.if_ino, 0, &inodedep) == 0) panic("dirrem_journal: Lost inodedep"); journal_jremref(dirrem, jremref, inodedep); if (dotremref) journal_jremref(dirrem, dotremref, inodedep); if (dotdotremref) journal_jremref(dirrem, dotdotremref, NULL); } /* * Allocate a new dirrem if appropriate and return it along with * its associated pagedep. Called without a lock, returns with lock. */ static struct dirrem * newdirrem(bp, dp, ip, isrmdir, prevdirremp) struct buf *bp; /* buffer containing directory block */ struct inode *dp; /* inode for the directory being modified */ struct inode *ip; /* inode for directory entry being removed */ int isrmdir; /* indicates if doing RMDIR */ struct dirrem **prevdirremp; /* previously referenced inode, if any */ { int offset; ufs_lbn_t lbn; struct diradd *dap; struct dirrem *dirrem; struct pagedep *pagedep; struct jremref *jremref; struct jremref *dotremref; struct jremref *dotdotremref; struct vnode *dvp; struct ufsmount *ump; /* * Whiteouts have no deletion dependencies. */ if (ip == NULL) panic("newdirrem: whiteout"); dvp = ITOV(dp); ump = ITOUMP(dp); /* * If the system is over its limit and our filesystem is * responsible for more than our share of that usage and * we are not a snapshot, request some inodedep cleanup. * Limiting the number of dirrem structures will also limit * the number of freefile and freeblks structures. */ ACQUIRE_LOCK(ump); if (!IS_SNAPSHOT(ip) && softdep_excess_items(ump, D_DIRREM)) schedule_cleanup(UFSTOVFS(ump)); else FREE_LOCK(ump); dirrem = malloc(sizeof(struct dirrem), M_DIRREM, M_SOFTDEP_FLAGS | M_ZERO); workitem_alloc(&dirrem->dm_list, D_DIRREM, dvp->v_mount); LIST_INIT(&dirrem->dm_jremrefhd); LIST_INIT(&dirrem->dm_jwork); dirrem->dm_state = isrmdir ? RMDIR : 0; dirrem->dm_oldinum = ip->i_number; *prevdirremp = NULL; /* * Allocate remove reference structures to track journal write * dependencies. We will always have one for the link and * when doing directories we will always have one more for dot. * When renaming a directory we skip the dotdot link change so * this is not needed. */ jremref = dotremref = dotdotremref = NULL; if (DOINGSUJ(dvp)) { if (isrmdir) { jremref = newjremref(dirrem, dp, ip, I_OFFSET(dp), ip->i_effnlink + 2); dotremref = newjremref(dirrem, ip, ip, DOT_OFFSET, ip->i_effnlink + 1); dotdotremref = newjremref(dirrem, ip, dp, DOTDOT_OFFSET, dp->i_effnlink + 1); dotdotremref->jr_state |= MKDIR_PARENT; } else jremref = newjremref(dirrem, dp, ip, I_OFFSET(dp), ip->i_effnlink + 1); } ACQUIRE_LOCK(ump); lbn = lblkno(ump->um_fs, I_OFFSET(dp)); offset = blkoff(ump->um_fs, I_OFFSET(dp)); pagedep_lookup(UFSTOVFS(ump), bp, dp->i_number, lbn, DEPALLOC, &pagedep); dirrem->dm_pagedep = pagedep; dirrem->dm_offset = offset; /* * If we're renaming a .. link to a new directory, cancel any * existing MKDIR_PARENT mkdir. If it has already been canceled * the jremref is preserved for any potential diradd in this * location. This can not coincide with a rmdir. */ if (I_OFFSET(dp) == DOTDOT_OFFSET) { if (isrmdir) panic("newdirrem: .. directory change during remove?"); jremref = cancel_mkdir_dotdot(dp, dirrem, jremref); } /* * If we're removing a directory search for the .. dependency now and * cancel it. Any pending journal work will be added to the dirrem * to be completed when the workitem remove completes. */ if (isrmdir) dotdotremref = cancel_diradd_dotdot(ip, dirrem, dotdotremref); /* * Check for a diradd dependency for the same directory entry. * If present, then both dependencies become obsolete and can * be de-allocated. */ dap = diradd_lookup(pagedep, offset); if (dap == NULL) { /* * Link the jremref structures into the dirrem so they are * written prior to the pagedep. */ if (jremref) dirrem_journal(dirrem, jremref, dotremref, dotdotremref); return (dirrem); } /* * Must be ATTACHED at this point. */ if ((dap->da_state & ATTACHED) == 0) panic("newdirrem: not ATTACHED"); if (dap->da_newinum != ip->i_number) panic("newdirrem: inum %ju should be %ju", (uintmax_t)ip->i_number, (uintmax_t)dap->da_newinum); /* * If we are deleting a changed name that never made it to disk, * then return the dirrem describing the previous inode (which * represents the inode currently referenced from this entry on disk). */ if ((dap->da_state & DIRCHG) != 0) { *prevdirremp = dap->da_previous; dap->da_state &= ~DIRCHG; dap->da_pagedep = pagedep; } /* * We are deleting an entry that never made it to disk. * Mark it COMPLETE so we can delete its inode immediately. */ dirrem->dm_state |= COMPLETE; cancel_diradd(dap, dirrem, jremref, dotremref, dotdotremref); #ifdef INVARIANTS if (isrmdir == 0) { struct worklist *wk; LIST_FOREACH(wk, &dirrem->dm_jwork, wk_list) if (wk->wk_state & (MKDIR_BODY | MKDIR_PARENT)) panic("bad wk %p (0x%X)\n", wk, wk->wk_state); } #endif return (dirrem); } /* * Directory entry change dependencies. * * Changing an existing directory entry requires that an add operation * be completed first followed by a deletion. The semantics for the addition * are identical to the description of adding a new entry above except * that the rollback is to the old inode number rather than zero. Once * the addition dependency is completed, the removal is done as described * in the removal routine above. */ /* * This routine should be called immediately after changing * a directory entry. The inode's link count should not be * decremented by the calling procedure -- the soft updates * code will perform this task when it is safe. */ void softdep_setup_directory_change(bp, dp, ip, newinum, isrmdir) struct buf *bp; /* buffer containing directory block */ struct inode *dp; /* inode for the directory being modified */ struct inode *ip; /* inode for directory entry being removed */ ino_t newinum; /* new inode number for changed entry */ int isrmdir; /* indicates if doing RMDIR */ { int offset; struct diradd *dap = NULL; struct dirrem *dirrem, *prevdirrem; struct pagedep *pagedep; struct inodedep *inodedep; struct jaddref *jaddref; struct mount *mp; struct ufsmount *ump; mp = ITOVFS(dp); ump = VFSTOUFS(mp); offset = blkoff(ump->um_fs, I_OFFSET(dp)); KASSERT(MOUNTEDSOFTDEP(mp) != 0, ("softdep_setup_directory_change called on non-softdep filesystem")); /* * Whiteouts do not need diradd dependencies. */ if (newinum != UFS_WINO) { dap = malloc(sizeof(struct diradd), M_DIRADD, M_SOFTDEP_FLAGS|M_ZERO); workitem_alloc(&dap->da_list, D_DIRADD, mp); dap->da_state = DIRCHG | ATTACHED | DEPCOMPLETE; dap->da_offset = offset; dap->da_newinum = newinum; LIST_INIT(&dap->da_jwork); } /* * Allocate a new dirrem and ACQUIRE_LOCK. */ dirrem = newdirrem(bp, dp, ip, isrmdir, &prevdirrem); pagedep = dirrem->dm_pagedep; /* * The possible values for isrmdir: * 0 - non-directory file rename * 1 - directory rename within same directory * inum - directory rename to new directory of given inode number * When renaming to a new directory, we are both deleting and * creating a new directory entry, so the link count on the new * directory should not change. Thus we do not need the followup * dirrem which is usually done in handle_workitem_remove. We set * the DIRCHG flag to tell handle_workitem_remove to skip the * followup dirrem. */ if (isrmdir > 1) dirrem->dm_state |= DIRCHG; /* * Whiteouts have no additional dependencies, * so just put the dirrem on the correct list. */ if (newinum == UFS_WINO) { if ((dirrem->dm_state & COMPLETE) == 0) { LIST_INSERT_HEAD(&pagedep->pd_dirremhd, dirrem, dm_next); } else { dirrem->dm_dirinum = pagedep->pd_ino; if (LIST_EMPTY(&dirrem->dm_jremrefhd)) add_to_worklist(&dirrem->dm_list, 0); } FREE_LOCK(ump); return; } /* * Add the dirrem to the inodedep's pending remove list for quick * discovery later. A valid nlinkdelta ensures that this lookup * will not fail. */ if (inodedep_lookup(mp, ip->i_number, 0, &inodedep) == 0) panic("softdep_setup_directory_change: Lost inodedep."); dirrem->dm_state |= ONDEPLIST; LIST_INSERT_HEAD(&inodedep->id_dirremhd, dirrem, dm_inonext); /* * If the COMPLETE flag is clear, then there were no active * entries and we want to roll back to the previous inode until * the new inode is committed to disk. If the COMPLETE flag is * set, then we have deleted an entry that never made it to disk. * If the entry we deleted resulted from a name change, then the old * inode reference still resides on disk. Any rollback that we do * needs to be to that old inode (returned to us in prevdirrem). If * the entry we deleted resulted from a create, then there is * no entry on the disk, so we want to roll back to zero rather * than the uncommitted inode. In either of the COMPLETE cases we * want to immediately free the unwritten and unreferenced inode. */ if ((dirrem->dm_state & COMPLETE) == 0) { dap->da_previous = dirrem; } else { if (prevdirrem != NULL) { dap->da_previous = prevdirrem; } else { dap->da_state &= ~DIRCHG; dap->da_pagedep = pagedep; } dirrem->dm_dirinum = pagedep->pd_ino; if (LIST_EMPTY(&dirrem->dm_jremrefhd)) add_to_worklist(&dirrem->dm_list, 0); } /* * Lookup the jaddref for this journal entry. We must finish * initializing it and make the diradd write dependent on it. * If we're not journaling, put it on the id_bufwait list if the * inode is not yet written. If it is written, do the post-inode * write processing to put it on the id_pendinghd list. */ inodedep_lookup(mp, newinum, DEPALLOC, &inodedep); if (MOUNTEDSUJ(mp)) { jaddref = (struct jaddref *)TAILQ_LAST(&inodedep->id_inoreflst, inoreflst); KASSERT(jaddref != NULL && jaddref->ja_parent == dp->i_number, ("softdep_setup_directory_change: bad jaddref %p", jaddref)); jaddref->ja_diroff = I_OFFSET(dp); jaddref->ja_diradd = dap; LIST_INSERT_HEAD(&pagedep->pd_diraddhd[DIRADDHASH(offset)], dap, da_pdlist); add_to_journal(&jaddref->ja_list); } else if ((inodedep->id_state & ALLCOMPLETE) == ALLCOMPLETE) { dap->da_state |= COMPLETE; LIST_INSERT_HEAD(&pagedep->pd_pendinghd, dap, da_pdlist); WORKLIST_INSERT(&inodedep->id_pendinghd, &dap->da_list); } else { LIST_INSERT_HEAD(&pagedep->pd_diraddhd[DIRADDHASH(offset)], dap, da_pdlist); WORKLIST_INSERT(&inodedep->id_bufwait, &dap->da_list); } /* * If we're making a new name for a directory that has not been * committed when need to move the dot and dotdot references to * this new name. */ if (inodedep->id_mkdiradd && I_OFFSET(dp) != DOTDOT_OFFSET) merge_diradd(inodedep, dap); FREE_LOCK(ump); } /* * Called whenever the link count on an inode is changed. * It creates an inode dependency so that the new reference(s) * to the inode cannot be committed to disk until the updated * inode has been written. */ void softdep_change_linkcnt(ip) struct inode *ip; /* the inode with the increased link count */ { struct inodedep *inodedep; struct ufsmount *ump; ump = ITOUMP(ip); KASSERT(MOUNTEDSOFTDEP(UFSTOVFS(ump)) != 0, ("softdep_change_linkcnt called on non-softdep filesystem")); ACQUIRE_LOCK(ump); inodedep_lookup(UFSTOVFS(ump), ip->i_number, DEPALLOC, &inodedep); if (ip->i_nlink < ip->i_effnlink) panic("softdep_change_linkcnt: bad delta"); inodedep->id_nlinkdelta = ip->i_nlink - ip->i_effnlink; FREE_LOCK(ump); } /* * Attach a sbdep dependency to the superblock buf so that we can keep * track of the head of the linked list of referenced but unlinked inodes. */ void softdep_setup_sbupdate(ump, fs, bp) struct ufsmount *ump; struct fs *fs; struct buf *bp; { struct sbdep *sbdep; struct worklist *wk; KASSERT(MOUNTEDSOFTDEP(UFSTOVFS(ump)) != 0, ("softdep_setup_sbupdate called on non-softdep filesystem")); LIST_FOREACH(wk, &bp->b_dep, wk_list) if (wk->wk_type == D_SBDEP) break; if (wk != NULL) return; sbdep = malloc(sizeof(struct sbdep), M_SBDEP, M_SOFTDEP_FLAGS); workitem_alloc(&sbdep->sb_list, D_SBDEP, UFSTOVFS(ump)); sbdep->sb_fs = fs; sbdep->sb_ump = ump; ACQUIRE_LOCK(ump); WORKLIST_INSERT(&bp->b_dep, &sbdep->sb_list); FREE_LOCK(ump); } /* * Return the first unlinked inodedep which is ready to be the head of the * list. The inodedep and all those after it must have valid next pointers. */ static struct inodedep * first_unlinked_inodedep(ump) struct ufsmount *ump; { struct inodedep *inodedep; struct inodedep *idp; LOCK_OWNED(ump); for (inodedep = TAILQ_LAST(&ump->softdep_unlinked, inodedeplst); inodedep; inodedep = idp) { if ((inodedep->id_state & UNLINKNEXT) == 0) return (NULL); idp = TAILQ_PREV(inodedep, inodedeplst, id_unlinked); if (idp == NULL || (idp->id_state & UNLINKNEXT) == 0) break; if ((inodedep->id_state & UNLINKPREV) == 0) break; } return (inodedep); } /* * Set the sujfree unlinked head pointer prior to writing a superblock. */ static void initiate_write_sbdep(sbdep) struct sbdep *sbdep; { struct inodedep *inodedep; struct fs *bpfs; struct fs *fs; bpfs = sbdep->sb_fs; fs = sbdep->sb_ump->um_fs; inodedep = first_unlinked_inodedep(sbdep->sb_ump); if (inodedep) { fs->fs_sujfree = inodedep->id_ino; inodedep->id_state |= UNLINKPREV; } else fs->fs_sujfree = 0; bpfs->fs_sujfree = fs->fs_sujfree; /* * Because we have made changes to the superblock, we need to * recompute its check-hash. */ bpfs->fs_ckhash = ffs_calc_sbhash(bpfs); } /* * After a superblock is written determine whether it must be written again * due to a changing unlinked list head. */ static int handle_written_sbdep(sbdep, bp) struct sbdep *sbdep; struct buf *bp; { struct inodedep *inodedep; struct fs *fs; LOCK_OWNED(sbdep->sb_ump); fs = sbdep->sb_fs; /* * If the superblock doesn't match the in-memory list start over. */ inodedep = first_unlinked_inodedep(sbdep->sb_ump); if ((inodedep && fs->fs_sujfree != inodedep->id_ino) || (inodedep == NULL && fs->fs_sujfree != 0)) { bdirty(bp); return (1); } WORKITEM_FREE(sbdep, D_SBDEP); if (fs->fs_sujfree == 0) return (0); /* * Now that we have a record of this inode in stable store allow it * to be written to free up pending work. Inodes may see a lot of * write activity after they are unlinked which we must not hold up. */ for (; inodedep != NULL; inodedep = TAILQ_NEXT(inodedep, id_unlinked)) { if ((inodedep->id_state & UNLINKLINKS) != UNLINKLINKS) panic("handle_written_sbdep: Bad inodedep %p (0x%X)", inodedep, inodedep->id_state); if (inodedep->id_state & UNLINKONLIST) break; inodedep->id_state |= DEPCOMPLETE | UNLINKONLIST; } return (0); } /* * Mark an inodedep as unlinked and insert it into the in-memory unlinked list. */ static void unlinked_inodedep(mp, inodedep) struct mount *mp; struct inodedep *inodedep; { struct ufsmount *ump; ump = VFSTOUFS(mp); LOCK_OWNED(ump); if (MOUNTEDSUJ(mp) == 0) return; ump->um_fs->fs_fmod = 1; if (inodedep->id_state & UNLINKED) panic("unlinked_inodedep: %p already unlinked\n", inodedep); inodedep->id_state |= UNLINKED; TAILQ_INSERT_HEAD(&ump->softdep_unlinked, inodedep, id_unlinked); } /* * Remove an inodedep from the unlinked inodedep list. This may require * disk writes if the inode has made it that far. */ static void clear_unlinked_inodedep(inodedep) struct inodedep *inodedep; { struct ufs2_dinode *dip; struct ufsmount *ump; struct inodedep *idp; struct inodedep *idn; struct fs *fs, *bpfs; struct buf *bp; daddr_t dbn; ino_t ino; ino_t nino; ino_t pino; int error; ump = VFSTOUFS(inodedep->id_list.wk_mp); fs = ump->um_fs; ino = inodedep->id_ino; error = 0; for (;;) { LOCK_OWNED(ump); KASSERT((inodedep->id_state & UNLINKED) != 0, ("clear_unlinked_inodedep: inodedep %p not unlinked", inodedep)); /* * If nothing has yet been written simply remove us from * the in memory list and return. This is the most common * case where handle_workitem_remove() loses the final * reference. */ if ((inodedep->id_state & UNLINKLINKS) == 0) break; /* * If we have a NEXT pointer and no PREV pointer we can simply * clear NEXT's PREV and remove ourselves from the list. Be * careful not to clear PREV if the superblock points at * next as well. */ idn = TAILQ_NEXT(inodedep, id_unlinked); if ((inodedep->id_state & UNLINKLINKS) == UNLINKNEXT) { if (idn && fs->fs_sujfree != idn->id_ino) idn->id_state &= ~UNLINKPREV; break; } /* * Here we have an inodedep which is actually linked into * the list. We must remove it by forcing a write to the * link before us, whether it be the superblock or an inode. * Unfortunately the list may change while we're waiting * on the buf lock for either resource so we must loop until * we lock the right one. If both the superblock and an * inode point to this inode we must clear the inode first * followed by the superblock. */ idp = TAILQ_PREV(inodedep, inodedeplst, id_unlinked); pino = 0; if (idp && (idp->id_state & UNLINKNEXT)) pino = idp->id_ino; FREE_LOCK(ump); if (pino == 0) { bp = getblk(ump->um_devvp, btodb(fs->fs_sblockloc), (int)fs->fs_sbsize, 0, 0, 0); } else { dbn = fsbtodb(fs, ino_to_fsba(fs, pino)); error = ffs_breadz(ump, ump->um_devvp, dbn, dbn, (int)fs->fs_bsize, NULL, NULL, 0, NOCRED, 0, NULL, &bp); } ACQUIRE_LOCK(ump); if (error) break; /* If the list has changed restart the loop. */ idp = TAILQ_PREV(inodedep, inodedeplst, id_unlinked); nino = 0; if (idp && (idp->id_state & UNLINKNEXT)) nino = idp->id_ino; if (nino != pino || (inodedep->id_state & UNLINKPREV) != UNLINKPREV) { FREE_LOCK(ump); brelse(bp); ACQUIRE_LOCK(ump); continue; } nino = 0; idn = TAILQ_NEXT(inodedep, id_unlinked); if (idn) nino = idn->id_ino; /* * Remove us from the in memory list. After this we cannot * access the inodedep. */ KASSERT((inodedep->id_state & UNLINKED) != 0, ("clear_unlinked_inodedep: inodedep %p not unlinked", inodedep)); inodedep->id_state &= ~(UNLINKED | UNLINKLINKS | UNLINKONLIST); TAILQ_REMOVE(&ump->softdep_unlinked, inodedep, id_unlinked); FREE_LOCK(ump); /* * The predecessor's next pointer is manually updated here * so that the NEXT flag is never cleared for an element * that is in the list. */ if (pino == 0) { bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize); bpfs = (struct fs *)bp->b_data; ffs_oldfscompat_write(bpfs, ump); softdep_setup_sbupdate(ump, bpfs, bp); /* * Because we may have made changes to the superblock, * we need to recompute its check-hash. */ bpfs->fs_ckhash = ffs_calc_sbhash(bpfs); } else if (fs->fs_magic == FS_UFS1_MAGIC) { ((struct ufs1_dinode *)bp->b_data + ino_to_fsbo(fs, pino))->di_freelink = nino; } else { dip = (struct ufs2_dinode *)bp->b_data + ino_to_fsbo(fs, pino); dip->di_freelink = nino; ffs_update_dinode_ckhash(fs, dip); } /* * If the bwrite fails we have no recourse to recover. The * filesystem is corrupted already. */ bwrite(bp); ACQUIRE_LOCK(ump); /* * If the superblock pointer still needs to be cleared force * a write here. */ if (fs->fs_sujfree == ino) { FREE_LOCK(ump); bp = getblk(ump->um_devvp, btodb(fs->fs_sblockloc), (int)fs->fs_sbsize, 0, 0, 0); bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize); bpfs = (struct fs *)bp->b_data; ffs_oldfscompat_write(bpfs, ump); softdep_setup_sbupdate(ump, bpfs, bp); /* * Because we may have made changes to the superblock, * we need to recompute its check-hash. */ bpfs->fs_ckhash = ffs_calc_sbhash(bpfs); bwrite(bp); ACQUIRE_LOCK(ump); } if (fs->fs_sujfree != ino) return; panic("clear_unlinked_inodedep: Failed to clear free head"); } if (inodedep->id_ino == fs->fs_sujfree) panic("clear_unlinked_inodedep: Freeing head of free list"); inodedep->id_state &= ~(UNLINKED | UNLINKLINKS | UNLINKONLIST); TAILQ_REMOVE(&ump->softdep_unlinked, inodedep, id_unlinked); return; } /* * This workitem decrements the inode's link count. * If the link count reaches zero, the file is removed. */ static int handle_workitem_remove(dirrem, flags) struct dirrem *dirrem; int flags; { struct inodedep *inodedep; struct workhead dotdotwk; struct worklist *wk; struct ufsmount *ump; struct mount *mp; struct vnode *vp; struct inode *ip; ino_t oldinum; if (dirrem->dm_state & ONWORKLIST) panic("handle_workitem_remove: dirrem %p still on worklist", dirrem); oldinum = dirrem->dm_oldinum; mp = dirrem->dm_list.wk_mp; ump = VFSTOUFS(mp); flags |= LK_EXCLUSIVE; if (ffs_vgetf(mp, oldinum, flags, &vp, FFSV_FORCEINSMQ | FFSV_FORCEINODEDEP) != 0) return (EBUSY); ip = VTOI(vp); MPASS(ip->i_mode != 0); ACQUIRE_LOCK(ump); if ((inodedep_lookup(mp, oldinum, 0, &inodedep)) == 0) panic("handle_workitem_remove: lost inodedep"); if (dirrem->dm_state & ONDEPLIST) LIST_REMOVE(dirrem, dm_inonext); KASSERT(LIST_EMPTY(&dirrem->dm_jremrefhd), ("handle_workitem_remove: Journal entries not written.")); /* * Move all dependencies waiting on the remove to complete * from the dirrem to the inode inowait list to be completed * after the inode has been updated and written to disk. * * Any marked MKDIR_PARENT are saved to be completed when the * dotdot ref is removed unless DIRCHG is specified. For * directory change operations there will be no further * directory writes and the jsegdeps need to be moved along * with the rest to be completed when the inode is free or * stable in the inode free list. */ LIST_INIT(&dotdotwk); while ((wk = LIST_FIRST(&dirrem->dm_jwork)) != NULL) { WORKLIST_REMOVE(wk); if ((dirrem->dm_state & DIRCHG) == 0 && wk->wk_state & MKDIR_PARENT) { wk->wk_state &= ~MKDIR_PARENT; WORKLIST_INSERT(&dotdotwk, wk); continue; } WORKLIST_INSERT(&inodedep->id_inowait, wk); } LIST_SWAP(&dirrem->dm_jwork, &dotdotwk, worklist, wk_list); /* * Normal file deletion. */ if ((dirrem->dm_state & RMDIR) == 0) { ip->i_nlink--; KASSERT(ip->i_nlink >= 0, ("handle_workitem_remove: file ino " "%ju negative i_nlink %d", (intmax_t)ip->i_number, ip->i_nlink)); DIP_SET(ip, i_nlink, ip->i_nlink); UFS_INODE_SET_FLAG(ip, IN_CHANGE); if (ip->i_nlink < ip->i_effnlink) panic("handle_workitem_remove: bad file delta"); if (ip->i_nlink == 0) unlinked_inodedep(mp, inodedep); inodedep->id_nlinkdelta = ip->i_nlink - ip->i_effnlink; KASSERT(LIST_EMPTY(&dirrem->dm_jwork), ("handle_workitem_remove: worklist not empty. %s", TYPENAME(LIST_FIRST(&dirrem->dm_jwork)->wk_type))); WORKITEM_FREE(dirrem, D_DIRREM); FREE_LOCK(ump); goto out; } /* * Directory deletion. Decrement reference count for both the * just deleted parent directory entry and the reference for ".". * Arrange to have the reference count on the parent decremented * to account for the loss of "..". */ ip->i_nlink -= 2; KASSERT(ip->i_nlink >= 0, ("handle_workitem_remove: directory ino " "%ju negative i_nlink %d", (intmax_t)ip->i_number, ip->i_nlink)); DIP_SET(ip, i_nlink, ip->i_nlink); UFS_INODE_SET_FLAG(ip, IN_CHANGE); if (ip->i_nlink < ip->i_effnlink) panic("handle_workitem_remove: bad dir delta"); if (ip->i_nlink == 0) unlinked_inodedep(mp, inodedep); inodedep->id_nlinkdelta = ip->i_nlink - ip->i_effnlink; /* * Rename a directory to a new parent. Since, we are both deleting * and creating a new directory entry, the link count on the new * directory should not change. Thus we skip the followup dirrem. */ if (dirrem->dm_state & DIRCHG) { KASSERT(LIST_EMPTY(&dirrem->dm_jwork), ("handle_workitem_remove: DIRCHG and worklist not empty.")); WORKITEM_FREE(dirrem, D_DIRREM); FREE_LOCK(ump); goto out; } dirrem->dm_state = ONDEPLIST; dirrem->dm_oldinum = dirrem->dm_dirinum; /* * Place the dirrem on the parent's diremhd list. */ if (inodedep_lookup(mp, dirrem->dm_oldinum, 0, &inodedep) == 0) panic("handle_workitem_remove: lost dir inodedep"); LIST_INSERT_HEAD(&inodedep->id_dirremhd, dirrem, dm_inonext); /* * If the allocated inode has never been written to disk, then * the on-disk inode is zero'ed and we can remove the file * immediately. When journaling if the inode has been marked * unlinked and not DEPCOMPLETE we know it can never be written. */ inodedep_lookup(mp, oldinum, 0, &inodedep); if (inodedep == NULL || (inodedep->id_state & (DEPCOMPLETE | UNLINKED)) == UNLINKED || check_inode_unwritten(inodedep)) { FREE_LOCK(ump); vput(vp); return handle_workitem_remove(dirrem, flags); } WORKLIST_INSERT(&inodedep->id_inowait, &dirrem->dm_list); FREE_LOCK(ump); UFS_INODE_SET_FLAG(ip, IN_CHANGE); out: ffs_update(vp, 0); vput(vp); return (0); } /* * Inode de-allocation dependencies. * * When an inode's link count is reduced to zero, it can be de-allocated. We * found it convenient to postpone de-allocation until after the inode is * written to disk with its new link count (zero). At this point, all of the * on-disk inode's block pointers are nullified and, with careful dependency * list ordering, all dependencies related to the inode will be satisfied and * the corresponding dependency structures de-allocated. So, if/when the * inode is reused, there will be no mixing of old dependencies with new * ones. This artificial dependency is set up by the block de-allocation * procedure above (softdep_setup_freeblocks) and completed by the * following procedure. */ static void handle_workitem_freefile(freefile) struct freefile *freefile; { struct workhead wkhd; struct fs *fs; struct ufsmount *ump; int error; #ifdef INVARIANTS struct inodedep *idp; #endif ump = VFSTOUFS(freefile->fx_list.wk_mp); fs = ump->um_fs; #ifdef INVARIANTS ACQUIRE_LOCK(ump); error = inodedep_lookup(UFSTOVFS(ump), freefile->fx_oldinum, 0, &idp); FREE_LOCK(ump); if (error) panic("handle_workitem_freefile: inodedep %p survived", idp); #endif UFS_LOCK(ump); fs->fs_pendinginodes -= 1; UFS_UNLOCK(ump); LIST_INIT(&wkhd); LIST_SWAP(&freefile->fx_jwork, &wkhd, worklist, wk_list); if ((error = ffs_freefile(ump, fs, freefile->fx_devvp, freefile->fx_oldinum, freefile->fx_mode, &wkhd)) != 0) softdep_error("handle_workitem_freefile", error); ACQUIRE_LOCK(ump); WORKITEM_FREE(freefile, D_FREEFILE); FREE_LOCK(ump); } /* * Helper function which unlinks marker element from work list and returns * the next element on the list. */ static __inline struct worklist * markernext(struct worklist *marker) { struct worklist *next; next = LIST_NEXT(marker, wk_list); LIST_REMOVE(marker, wk_list); return next; } /* * Disk writes. * * The dependency structures constructed above are most actively used when file * system blocks are written to disk. No constraints are placed on when a * block can be written, but unsatisfied update dependencies are made safe by * modifying (or replacing) the source memory for the duration of the disk * write. When the disk write completes, the memory block is again brought * up-to-date. * * In-core inode structure reclamation. * * Because there are a finite number of "in-core" inode structures, they are * reused regularly. By transferring all inode-related dependencies to the * in-memory inode block and indexing them separately (via "inodedep"s), we * can allow "in-core" inode structures to be reused at any time and avoid * any increase in contention. * * Called just before entering the device driver to initiate a new disk I/O. * The buffer must be locked, thus, no I/O completion operations can occur * while we are manipulating its associated dependencies. */ static void softdep_disk_io_initiation(bp) struct buf *bp; /* structure describing disk write to occur */ { struct worklist *wk; struct worklist marker; struct inodedep *inodedep; struct freeblks *freeblks; struct jblkdep *jblkdep; struct newblk *newblk; struct ufsmount *ump; /* * We only care about write operations. There should never * be dependencies for reads. */ if (bp->b_iocmd != BIO_WRITE) panic("softdep_disk_io_initiation: not write"); if (bp->b_vflags & BV_BKGRDINPROG) panic("softdep_disk_io_initiation: Writing buffer with " "background write in progress: %p", bp); ump = softdep_bp_to_mp(bp); if (ump == NULL) return; marker.wk_type = D_LAST + 1; /* Not a normal workitem */ PHOLD(curproc); /* Don't swap out kernel stack */ ACQUIRE_LOCK(ump); /* * Do any necessary pre-I/O processing. */ for (wk = LIST_FIRST(&bp->b_dep); wk != NULL; wk = markernext(&marker)) { LIST_INSERT_AFTER(wk, &marker, wk_list); switch (wk->wk_type) { case D_PAGEDEP: initiate_write_filepage(WK_PAGEDEP(wk), bp); continue; case D_INODEDEP: inodedep = WK_INODEDEP(wk); if (inodedep->id_fs->fs_magic == FS_UFS1_MAGIC) initiate_write_inodeblock_ufs1(inodedep, bp); else initiate_write_inodeblock_ufs2(inodedep, bp); continue; case D_INDIRDEP: initiate_write_indirdep(WK_INDIRDEP(wk), bp); continue; case D_BMSAFEMAP: initiate_write_bmsafemap(WK_BMSAFEMAP(wk), bp); continue; case D_JSEG: WK_JSEG(wk)->js_buf = NULL; continue; case D_FREEBLKS: freeblks = WK_FREEBLKS(wk); jblkdep = LIST_FIRST(&freeblks->fb_jblkdephd); /* * We have to wait for the freeblks to be journaled * before we can write an inodeblock with updated * pointers. Be careful to arrange the marker so * we revisit the freeblks if it's not removed by * the first jwait(). */ if (jblkdep != NULL) { LIST_REMOVE(&marker, wk_list); LIST_INSERT_BEFORE(wk, &marker, wk_list); jwait(&jblkdep->jb_list, MNT_WAIT); } continue; case D_ALLOCDIRECT: case D_ALLOCINDIR: /* * We have to wait for the jnewblk to be journaled * before we can write to a block if the contents * may be confused with an earlier file's indirect * at recovery time. Handle the marker as described * above. */ newblk = WK_NEWBLK(wk); if (newblk->nb_jnewblk != NULL && indirblk_lookup(newblk->nb_list.wk_mp, newblk->nb_newblkno)) { LIST_REMOVE(&marker, wk_list); LIST_INSERT_BEFORE(wk, &marker, wk_list); jwait(&newblk->nb_jnewblk->jn_list, MNT_WAIT); } continue; case D_SBDEP: initiate_write_sbdep(WK_SBDEP(wk)); continue; case D_MKDIR: case D_FREEWORK: case D_FREEDEP: case D_JSEGDEP: continue; default: panic("handle_disk_io_initiation: Unexpected type %s", TYPENAME(wk->wk_type)); /* NOTREACHED */ } } FREE_LOCK(ump); PRELE(curproc); /* Allow swapout of kernel stack */ } /* * Called from within the procedure above to deal with unsatisfied * allocation dependencies in a directory. The buffer must be locked, * thus, no I/O completion operations can occur while we are * manipulating its associated dependencies. */ static void initiate_write_filepage(pagedep, bp) struct pagedep *pagedep; struct buf *bp; { struct jremref *jremref; struct jmvref *jmvref; struct dirrem *dirrem; struct diradd *dap; struct direct *ep; int i; if (pagedep->pd_state & IOSTARTED) { /* * This can only happen if there is a driver that does not * understand chaining. Here biodone will reissue the call * to strategy for the incomplete buffers. */ printf("initiate_write_filepage: already started\n"); return; } pagedep->pd_state |= IOSTARTED; /* * Wait for all journal remove dependencies to hit the disk. * We can not allow any potentially conflicting directory adds * to be visible before removes and rollback is too difficult. * The per-filesystem lock may be dropped and re-acquired, however * we hold the buf locked so the dependency can not go away. */ LIST_FOREACH(dirrem, &pagedep->pd_dirremhd, dm_next) while ((jremref = LIST_FIRST(&dirrem->dm_jremrefhd)) != NULL) jwait(&jremref->jr_list, MNT_WAIT); while ((jmvref = LIST_FIRST(&pagedep->pd_jmvrefhd)) != NULL) jwait(&jmvref->jm_list, MNT_WAIT); for (i = 0; i < DAHASHSZ; i++) { LIST_FOREACH(dap, &pagedep->pd_diraddhd[i], da_pdlist) { ep = (struct direct *) ((char *)bp->b_data + dap->da_offset); if (ep->d_ino != dap->da_newinum) panic("%s: dir inum %ju != new %ju", "initiate_write_filepage", (uintmax_t)ep->d_ino, (uintmax_t)dap->da_newinum); if (dap->da_state & DIRCHG) ep->d_ino = dap->da_previous->dm_oldinum; else ep->d_ino = 0; dap->da_state &= ~ATTACHED; dap->da_state |= UNDONE; } } } /* * Version of initiate_write_inodeblock that handles UFS1 dinodes. * Note that any bug fixes made to this routine must be done in the * version found below. * * Called from within the procedure above to deal with unsatisfied * allocation dependencies in an inodeblock. The buffer must be * locked, thus, no I/O completion operations can occur while we * are manipulating its associated dependencies. */ static void initiate_write_inodeblock_ufs1(inodedep, bp) struct inodedep *inodedep; struct buf *bp; /* The inode block */ { struct allocdirect *adp, *lastadp; struct ufs1_dinode *dp; struct ufs1_dinode *sip; struct inoref *inoref; struct ufsmount *ump; struct fs *fs; ufs_lbn_t i; #ifdef INVARIANTS ufs_lbn_t prevlbn = 0; #endif int deplist; if (inodedep->id_state & IOSTARTED) panic("initiate_write_inodeblock_ufs1: already started"); inodedep->id_state |= IOSTARTED; fs = inodedep->id_fs; ump = VFSTOUFS(inodedep->id_list.wk_mp); LOCK_OWNED(ump); dp = (struct ufs1_dinode *)bp->b_data + ino_to_fsbo(fs, inodedep->id_ino); /* * If we're on the unlinked list but have not yet written our * next pointer initialize it here. */ if ((inodedep->id_state & (UNLINKED | UNLINKNEXT)) == UNLINKED) { struct inodedep *inon; inon = TAILQ_NEXT(inodedep, id_unlinked); dp->di_freelink = inon ? inon->id_ino : 0; } /* * If the bitmap is not yet written, then the allocated * inode cannot be written to disk. */ if ((inodedep->id_state & DEPCOMPLETE) == 0) { if (inodedep->id_savedino1 != NULL) panic("initiate_write_inodeblock_ufs1: I/O underway"); FREE_LOCK(ump); sip = malloc(sizeof(struct ufs1_dinode), M_SAVEDINO, M_SOFTDEP_FLAGS); ACQUIRE_LOCK(ump); inodedep->id_savedino1 = sip; *inodedep->id_savedino1 = *dp; bzero((caddr_t)dp, sizeof(struct ufs1_dinode)); dp->di_gen = inodedep->id_savedino1->di_gen; dp->di_freelink = inodedep->id_savedino1->di_freelink; return; } /* * If no dependencies, then there is nothing to roll back. */ inodedep->id_savedsize = dp->di_size; inodedep->id_savedextsize = 0; inodedep->id_savednlink = dp->di_nlink; if (TAILQ_EMPTY(&inodedep->id_inoupdt) && TAILQ_EMPTY(&inodedep->id_inoreflst)) return; /* * Revert the link count to that of the first unwritten journal entry. */ inoref = TAILQ_FIRST(&inodedep->id_inoreflst); if (inoref) dp->di_nlink = inoref->if_nlink; /* * Set the dependencies to busy. */ for (deplist = 0, adp = TAILQ_FIRST(&inodedep->id_inoupdt); adp; adp = TAILQ_NEXT(adp, ad_next)) { #ifdef INVARIANTS if (deplist != 0 && prevlbn >= adp->ad_offset) panic("softdep_write_inodeblock: lbn order"); prevlbn = adp->ad_offset; if (adp->ad_offset < UFS_NDADDR && dp->di_db[adp->ad_offset] != adp->ad_newblkno) panic("initiate_write_inodeblock_ufs1: " "direct pointer #%jd mismatch %d != %jd", (intmax_t)adp->ad_offset, dp->di_db[adp->ad_offset], (intmax_t)adp->ad_newblkno); if (adp->ad_offset >= UFS_NDADDR && dp->di_ib[adp->ad_offset - UFS_NDADDR] != adp->ad_newblkno) panic("initiate_write_inodeblock_ufs1: " "indirect pointer #%jd mismatch %d != %jd", (intmax_t)adp->ad_offset - UFS_NDADDR, dp->di_ib[adp->ad_offset - UFS_NDADDR], (intmax_t)adp->ad_newblkno); deplist |= 1 << adp->ad_offset; if ((adp->ad_state & ATTACHED) == 0) panic("initiate_write_inodeblock_ufs1: " "Unknown state 0x%x", adp->ad_state); #endif /* INVARIANTS */ adp->ad_state &= ~ATTACHED; adp->ad_state |= UNDONE; } /* * The on-disk inode cannot claim to be any larger than the last * fragment that has been written. Otherwise, the on-disk inode * might have fragments that were not the last block in the file * which would corrupt the filesystem. */ for (lastadp = NULL, adp = TAILQ_FIRST(&inodedep->id_inoupdt); adp; lastadp = adp, adp = TAILQ_NEXT(adp, ad_next)) { if (adp->ad_offset >= UFS_NDADDR) break; dp->di_db[adp->ad_offset] = adp->ad_oldblkno; /* keep going until hitting a rollback to a frag */ if (adp->ad_oldsize == 0 || adp->ad_oldsize == fs->fs_bsize) continue; dp->di_size = fs->fs_bsize * adp->ad_offset + adp->ad_oldsize; for (i = adp->ad_offset + 1; i < UFS_NDADDR; i++) { #ifdef INVARIANTS if (dp->di_db[i] != 0 && (deplist & (1 << i)) == 0) panic("initiate_write_inodeblock_ufs1: " "lost dep1"); #endif /* INVARIANTS */ dp->di_db[i] = 0; } for (i = 0; i < UFS_NIADDR; i++) { #ifdef INVARIANTS if (dp->di_ib[i] != 0 && (deplist & ((1 << UFS_NDADDR) << i)) == 0) panic("initiate_write_inodeblock_ufs1: " "lost dep2"); #endif /* INVARIANTS */ dp->di_ib[i] = 0; } return; } /* * If we have zero'ed out the last allocated block of the file, * roll back the size to the last currently allocated block. * We know that this last allocated block is a full-sized as * we already checked for fragments in the loop above. */ if (lastadp != NULL && dp->di_size <= (lastadp->ad_offset + 1) * fs->fs_bsize) { for (i = lastadp->ad_offset; i >= 0; i--) if (dp->di_db[i] != 0) break; dp->di_size = (i + 1) * fs->fs_bsize; } /* * The only dependencies are for indirect blocks. * * The file size for indirect block additions is not guaranteed. * Such a guarantee would be non-trivial to achieve. The conventional * synchronous write implementation also does not make this guarantee. * Fsck should catch and fix discrepancies. Arguably, the file size * can be over-estimated without destroying integrity when the file * moves into the indirect blocks (i.e., is large). If we want to * postpone fsck, we are stuck with this argument. */ for (; adp; adp = TAILQ_NEXT(adp, ad_next)) dp->di_ib[adp->ad_offset - UFS_NDADDR] = 0; } /* * Version of initiate_write_inodeblock that handles UFS2 dinodes. * Note that any bug fixes made to this routine must be done in the * version found above. * * Called from within the procedure above to deal with unsatisfied * allocation dependencies in an inodeblock. The buffer must be * locked, thus, no I/O completion operations can occur while we * are manipulating its associated dependencies. */ static void initiate_write_inodeblock_ufs2(inodedep, bp) struct inodedep *inodedep; struct buf *bp; /* The inode block */ { struct allocdirect *adp, *lastadp; struct ufs2_dinode *dp; struct ufs2_dinode *sip; struct inoref *inoref; struct ufsmount *ump; struct fs *fs; ufs_lbn_t i; #ifdef INVARIANTS ufs_lbn_t prevlbn = 0; #endif int deplist; if (inodedep->id_state & IOSTARTED) panic("initiate_write_inodeblock_ufs2: already started"); inodedep->id_state |= IOSTARTED; fs = inodedep->id_fs; ump = VFSTOUFS(inodedep->id_list.wk_mp); LOCK_OWNED(ump); dp = (struct ufs2_dinode *)bp->b_data + ino_to_fsbo(fs, inodedep->id_ino); /* * If we're on the unlinked list but have not yet written our * next pointer initialize it here. */ if ((inodedep->id_state & (UNLINKED | UNLINKNEXT)) == UNLINKED) { struct inodedep *inon; inon = TAILQ_NEXT(inodedep, id_unlinked); dp->di_freelink = inon ? inon->id_ino : 0; ffs_update_dinode_ckhash(fs, dp); } /* * If the bitmap is not yet written, then the allocated * inode cannot be written to disk. */ if ((inodedep->id_state & DEPCOMPLETE) == 0) { if (inodedep->id_savedino2 != NULL) panic("initiate_write_inodeblock_ufs2: I/O underway"); FREE_LOCK(ump); sip = malloc(sizeof(struct ufs2_dinode), M_SAVEDINO, M_SOFTDEP_FLAGS); ACQUIRE_LOCK(ump); inodedep->id_savedino2 = sip; *inodedep->id_savedino2 = *dp; bzero((caddr_t)dp, sizeof(struct ufs2_dinode)); dp->di_gen = inodedep->id_savedino2->di_gen; dp->di_freelink = inodedep->id_savedino2->di_freelink; return; } /* * If no dependencies, then there is nothing to roll back. */ inodedep->id_savedsize = dp->di_size; inodedep->id_savedextsize = dp->di_extsize; inodedep->id_savednlink = dp->di_nlink; if (TAILQ_EMPTY(&inodedep->id_inoupdt) && TAILQ_EMPTY(&inodedep->id_extupdt) && TAILQ_EMPTY(&inodedep->id_inoreflst)) return; /* * Revert the link count to that of the first unwritten journal entry. */ inoref = TAILQ_FIRST(&inodedep->id_inoreflst); if (inoref) dp->di_nlink = inoref->if_nlink; /* * Set the ext data dependencies to busy. */ for (deplist = 0, adp = TAILQ_FIRST(&inodedep->id_extupdt); adp; adp = TAILQ_NEXT(adp, ad_next)) { #ifdef INVARIANTS if (deplist != 0 && prevlbn >= adp->ad_offset) panic("initiate_write_inodeblock_ufs2: lbn order"); prevlbn = adp->ad_offset; if (dp->di_extb[adp->ad_offset] != adp->ad_newblkno) panic("initiate_write_inodeblock_ufs2: " "ext pointer #%jd mismatch %jd != %jd", (intmax_t)adp->ad_offset, (intmax_t)dp->di_extb[adp->ad_offset], (intmax_t)adp->ad_newblkno); deplist |= 1 << adp->ad_offset; if ((adp->ad_state & ATTACHED) == 0) panic("initiate_write_inodeblock_ufs2: Unknown " "state 0x%x", adp->ad_state); #endif /* INVARIANTS */ adp->ad_state &= ~ATTACHED; adp->ad_state |= UNDONE; } /* * The on-disk inode cannot claim to be any larger than the last * fragment that has been written. Otherwise, the on-disk inode * might have fragments that were not the last block in the ext * data which would corrupt the filesystem. */ for (lastadp = NULL, adp = TAILQ_FIRST(&inodedep->id_extupdt); adp; lastadp = adp, adp = TAILQ_NEXT(adp, ad_next)) { dp->di_extb[adp->ad_offset] = adp->ad_oldblkno; /* keep going until hitting a rollback to a frag */ if (adp->ad_oldsize == 0 || adp->ad_oldsize == fs->fs_bsize) continue; dp->di_extsize = fs->fs_bsize * adp->ad_offset + adp->ad_oldsize; for (i = adp->ad_offset + 1; i < UFS_NXADDR; i++) { #ifdef INVARIANTS if (dp->di_extb[i] != 0 && (deplist & (1 << i)) == 0) panic("initiate_write_inodeblock_ufs2: " "lost dep1"); #endif /* INVARIANTS */ dp->di_extb[i] = 0; } lastadp = NULL; break; } /* * If we have zero'ed out the last allocated block of the ext * data, roll back the size to the last currently allocated block. * We know that this last allocated block is a full-sized as * we already checked for fragments in the loop above. */ if (lastadp != NULL && dp->di_extsize <= (lastadp->ad_offset + 1) * fs->fs_bsize) { for (i = lastadp->ad_offset; i >= 0; i--) if (dp->di_extb[i] != 0) break; dp->di_extsize = (i + 1) * fs->fs_bsize; } /* * Set the file data dependencies to busy. */ for (deplist = 0, adp = TAILQ_FIRST(&inodedep->id_inoupdt); adp; adp = TAILQ_NEXT(adp, ad_next)) { #ifdef INVARIANTS if (deplist != 0 && prevlbn >= adp->ad_offset) panic("softdep_write_inodeblock: lbn order"); if ((adp->ad_state & ATTACHED) == 0) panic("inodedep %p and adp %p not attached", inodedep, adp); prevlbn = adp->ad_offset; if (!ffs_fsfail_cleanup(ump, 0) && adp->ad_offset < UFS_NDADDR && dp->di_db[adp->ad_offset] != adp->ad_newblkno) panic("initiate_write_inodeblock_ufs2: " "direct pointer #%jd mismatch %jd != %jd", (intmax_t)adp->ad_offset, (intmax_t)dp->di_db[adp->ad_offset], (intmax_t)adp->ad_newblkno); if (!ffs_fsfail_cleanup(ump, 0) && adp->ad_offset >= UFS_NDADDR && dp->di_ib[adp->ad_offset - UFS_NDADDR] != adp->ad_newblkno) panic("initiate_write_inodeblock_ufs2: " "indirect pointer #%jd mismatch %jd != %jd", (intmax_t)adp->ad_offset - UFS_NDADDR, (intmax_t)dp->di_ib[adp->ad_offset - UFS_NDADDR], (intmax_t)adp->ad_newblkno); deplist |= 1 << adp->ad_offset; if ((adp->ad_state & ATTACHED) == 0) panic("initiate_write_inodeblock_ufs2: Unknown " "state 0x%x", adp->ad_state); #endif /* INVARIANTS */ adp->ad_state &= ~ATTACHED; adp->ad_state |= UNDONE; } /* * The on-disk inode cannot claim to be any larger than the last * fragment that has been written. Otherwise, the on-disk inode * might have fragments that were not the last block in the file * which would corrupt the filesystem. */ for (lastadp = NULL, adp = TAILQ_FIRST(&inodedep->id_inoupdt); adp; lastadp = adp, adp = TAILQ_NEXT(adp, ad_next)) { if (adp->ad_offset >= UFS_NDADDR) break; dp->di_db[adp->ad_offset] = adp->ad_oldblkno; /* keep going until hitting a rollback to a frag */ if (adp->ad_oldsize == 0 || adp->ad_oldsize == fs->fs_bsize) continue; dp->di_size = fs->fs_bsize * adp->ad_offset + adp->ad_oldsize; for (i = adp->ad_offset + 1; i < UFS_NDADDR; i++) { #ifdef INVARIANTS if (dp->di_db[i] != 0 && (deplist & (1 << i)) == 0) panic("initiate_write_inodeblock_ufs2: " "lost dep2"); #endif /* INVARIANTS */ dp->di_db[i] = 0; } for (i = 0; i < UFS_NIADDR; i++) { #ifdef INVARIANTS if (dp->di_ib[i] != 0 && (deplist & ((1 << UFS_NDADDR) << i)) == 0) panic("initiate_write_inodeblock_ufs2: " "lost dep3"); #endif /* INVARIANTS */ dp->di_ib[i] = 0; } ffs_update_dinode_ckhash(fs, dp); return; } /* * If we have zero'ed out the last allocated block of the file, * roll back the size to the last currently allocated block. * We know that this last allocated block is a full-sized as * we already checked for fragments in the loop above. */ if (lastadp != NULL && dp->di_size <= (lastadp->ad_offset + 1) * fs->fs_bsize) { for (i = lastadp->ad_offset; i >= 0; i--) if (dp->di_db[i] != 0) break; dp->di_size = (i + 1) * fs->fs_bsize; } /* * The only dependencies are for indirect blocks. * * The file size for indirect block additions is not guaranteed. * Such a guarantee would be non-trivial to achieve. The conventional * synchronous write implementation also does not make this guarantee. * Fsck should catch and fix discrepancies. Arguably, the file size * can be over-estimated without destroying integrity when the file * moves into the indirect blocks (i.e., is large). If we want to * postpone fsck, we are stuck with this argument. */ for (; adp; adp = TAILQ_NEXT(adp, ad_next)) dp->di_ib[adp->ad_offset - UFS_NDADDR] = 0; ffs_update_dinode_ckhash(fs, dp); } /* * Cancel an indirdep as a result of truncation. Release all of the * children allocindirs and place their journal work on the appropriate * list. */ static void cancel_indirdep(indirdep, bp, freeblks) struct indirdep *indirdep; struct buf *bp; struct freeblks *freeblks; { struct allocindir *aip; /* * None of the indirect pointers will ever be visible, * so they can simply be tossed. GOINGAWAY ensures * that allocated pointers will be saved in the buffer * cache until they are freed. Note that they will * only be able to be found by their physical address * since the inode mapping the logical address will * be gone. The save buffer used for the safe copy * was allocated in setup_allocindir_phase2 using * the physical address so it could be used for this * purpose. Hence we swap the safe copy with the real * copy, allowing the safe copy to be freed and holding * on to the real copy for later use in indir_trunc. */ if (indirdep->ir_state & GOINGAWAY) panic("cancel_indirdep: already gone"); if ((indirdep->ir_state & DEPCOMPLETE) == 0) { indirdep->ir_state |= DEPCOMPLETE; LIST_REMOVE(indirdep, ir_next); } indirdep->ir_state |= GOINGAWAY; /* * Pass in bp for blocks still have journal writes * pending so we can cancel them on their own. */ while ((aip = LIST_FIRST(&indirdep->ir_deplisthd)) != NULL) cancel_allocindir(aip, bp, freeblks, 0); while ((aip = LIST_FIRST(&indirdep->ir_donehd)) != NULL) cancel_allocindir(aip, NULL, freeblks, 0); while ((aip = LIST_FIRST(&indirdep->ir_writehd)) != NULL) cancel_allocindir(aip, NULL, freeblks, 0); while ((aip = LIST_FIRST(&indirdep->ir_completehd)) != NULL) cancel_allocindir(aip, NULL, freeblks, 0); /* * If there are pending partial truncations we need to keep the * old block copy around until they complete. This is because * the current b_data is not a perfect superset of the available * blocks. */ if (TAILQ_EMPTY(&indirdep->ir_trunc)) bcopy(bp->b_data, indirdep->ir_savebp->b_data, bp->b_bcount); else bcopy(bp->b_data, indirdep->ir_saveddata, bp->b_bcount); WORKLIST_REMOVE(&indirdep->ir_list); WORKLIST_INSERT(&indirdep->ir_savebp->b_dep, &indirdep->ir_list); indirdep->ir_bp = NULL; indirdep->ir_freeblks = freeblks; } /* * Free an indirdep once it no longer has new pointers to track. */ static void free_indirdep(indirdep) struct indirdep *indirdep; { KASSERT(TAILQ_EMPTY(&indirdep->ir_trunc), ("free_indirdep: Indir trunc list not empty.")); KASSERT(LIST_EMPTY(&indirdep->ir_completehd), ("free_indirdep: Complete head not empty.")); KASSERT(LIST_EMPTY(&indirdep->ir_writehd), ("free_indirdep: write head not empty.")); KASSERT(LIST_EMPTY(&indirdep->ir_donehd), ("free_indirdep: done head not empty.")); KASSERT(LIST_EMPTY(&indirdep->ir_deplisthd), ("free_indirdep: deplist head not empty.")); KASSERT((indirdep->ir_state & DEPCOMPLETE), ("free_indirdep: %p still on newblk list.", indirdep)); KASSERT(indirdep->ir_saveddata == NULL, ("free_indirdep: %p still has saved data.", indirdep)); KASSERT(indirdep->ir_savebp == NULL, ("free_indirdep: %p still has savebp buffer.", indirdep)); if (indirdep->ir_state & ONWORKLIST) WORKLIST_REMOVE(&indirdep->ir_list); WORKITEM_FREE(indirdep, D_INDIRDEP); } /* * Called before a write to an indirdep. This routine is responsible for * rolling back pointers to a safe state which includes only those * allocindirs which have been completed. */ static void initiate_write_indirdep(indirdep, bp) struct indirdep *indirdep; struct buf *bp; { struct ufsmount *ump; indirdep->ir_state |= IOSTARTED; if (indirdep->ir_state & GOINGAWAY) panic("disk_io_initiation: indirdep gone"); /* * If there are no remaining dependencies, this will be writing * the real pointers. */ if (LIST_EMPTY(&indirdep->ir_deplisthd) && TAILQ_EMPTY(&indirdep->ir_trunc)) return; /* * Replace up-to-date version with safe version. */ if (indirdep->ir_saveddata == NULL) { ump = VFSTOUFS(indirdep->ir_list.wk_mp); LOCK_OWNED(ump); FREE_LOCK(ump); indirdep->ir_saveddata = malloc(bp->b_bcount, M_INDIRDEP, M_SOFTDEP_FLAGS); ACQUIRE_LOCK(ump); } indirdep->ir_state &= ~ATTACHED; indirdep->ir_state |= UNDONE; bcopy(bp->b_data, indirdep->ir_saveddata, bp->b_bcount); bcopy(indirdep->ir_savebp->b_data, bp->b_data, bp->b_bcount); } /* * Called when an inode has been cleared in a cg bitmap. This finally * eliminates any canceled jaddrefs */ void softdep_setup_inofree(mp, bp, ino, wkhd) struct mount *mp; struct buf *bp; ino_t ino; struct workhead *wkhd; { struct worklist *wk, *wkn; struct inodedep *inodedep; struct ufsmount *ump; uint8_t *inosused; struct cg *cgp; struct fs *fs; KASSERT(MOUNTEDSOFTDEP(mp) != 0, ("softdep_setup_inofree called on non-softdep filesystem")); ump = VFSTOUFS(mp); ACQUIRE_LOCK(ump); if (!ffs_fsfail_cleanup(ump, 0)) { fs = ump->um_fs; cgp = (struct cg *)bp->b_data; inosused = cg_inosused(cgp); if (isset(inosused, ino % fs->fs_ipg)) panic("softdep_setup_inofree: inode %ju not freed.", (uintmax_t)ino); } if (inodedep_lookup(mp, ino, 0, &inodedep)) panic("softdep_setup_inofree: ino %ju has existing inodedep %p", (uintmax_t)ino, inodedep); if (wkhd) { LIST_FOREACH_SAFE(wk, wkhd, wk_list, wkn) { if (wk->wk_type != D_JADDREF) continue; WORKLIST_REMOVE(wk); /* * We can free immediately even if the jaddref * isn't attached in a background write as now * the bitmaps are reconciled. */ wk->wk_state |= COMPLETE | ATTACHED; free_jaddref(WK_JADDREF(wk)); } jwork_move(&bp->b_dep, wkhd); } FREE_LOCK(ump); } /* * Called via ffs_blkfree() after a set of frags has been cleared from a cg * map. Any dependencies waiting for the write to clear are added to the * buf's list and any jnewblks that are being canceled are discarded * immediately. */ void softdep_setup_blkfree(mp, bp, blkno, frags, wkhd) struct mount *mp; struct buf *bp; ufs2_daddr_t blkno; int frags; struct workhead *wkhd; { struct bmsafemap *bmsafemap; struct jnewblk *jnewblk; struct ufsmount *ump; struct worklist *wk; struct fs *fs; #ifdef INVARIANTS uint8_t *blksfree; struct cg *cgp; ufs2_daddr_t jstart; ufs2_daddr_t jend; ufs2_daddr_t end; long bno; int i; #endif CTR3(KTR_SUJ, "softdep_setup_blkfree: blkno %jd frags %d wk head %p", blkno, frags, wkhd); ump = VFSTOUFS(mp); KASSERT(MOUNTEDSOFTDEP(UFSTOVFS(ump)) != 0, ("softdep_setup_blkfree called on non-softdep filesystem")); ACQUIRE_LOCK(ump); /* Lookup the bmsafemap so we track when it is dirty. */ fs = ump->um_fs; bmsafemap = bmsafemap_lookup(mp, bp, dtog(fs, blkno), NULL); /* * Detach any jnewblks which have been canceled. They must linger * until the bitmap is cleared again by ffs_blkfree() to prevent * an unjournaled allocation from hitting the disk. */ if (wkhd) { while ((wk = LIST_FIRST(wkhd)) != NULL) { CTR2(KTR_SUJ, "softdep_setup_blkfree: blkno %jd wk type %d", blkno, wk->wk_type); WORKLIST_REMOVE(wk); if (wk->wk_type != D_JNEWBLK) { WORKLIST_INSERT(&bmsafemap->sm_freehd, wk); continue; } jnewblk = WK_JNEWBLK(wk); KASSERT(jnewblk->jn_state & GOINGAWAY, ("softdep_setup_blkfree: jnewblk not canceled.")); #ifdef INVARIANTS /* * Assert that this block is free in the bitmap * before we discard the jnewblk. */ cgp = (struct cg *)bp->b_data; blksfree = cg_blksfree(cgp); bno = dtogd(fs, jnewblk->jn_blkno); for (i = jnewblk->jn_oldfrags; i < jnewblk->jn_frags; i++) { if (isset(blksfree, bno + i)) continue; panic("softdep_setup_blkfree: not free"); } #endif /* * Even if it's not attached we can free immediately * as the new bitmap is correct. */ wk->wk_state |= COMPLETE | ATTACHED; free_jnewblk(jnewblk); } } #ifdef INVARIANTS /* * Assert that we are not freeing a block which has an outstanding * allocation dependency. */ fs = VFSTOUFS(mp)->um_fs; bmsafemap = bmsafemap_lookup(mp, bp, dtog(fs, blkno), NULL); end = blkno + frags; LIST_FOREACH(jnewblk, &bmsafemap->sm_jnewblkhd, jn_deps) { /* * Don't match against blocks that will be freed when the * background write is done. */ if ((jnewblk->jn_state & (ATTACHED | COMPLETE | DEPCOMPLETE)) == (COMPLETE | DEPCOMPLETE)) continue; jstart = jnewblk->jn_blkno + jnewblk->jn_oldfrags; jend = jnewblk->jn_blkno + jnewblk->jn_frags; if ((blkno >= jstart && blkno < jend) || (end > jstart && end <= jend)) { printf("state 0x%X %jd - %d %d dep %p\n", jnewblk->jn_state, jnewblk->jn_blkno, jnewblk->jn_oldfrags, jnewblk->jn_frags, jnewblk->jn_dep); panic("softdep_setup_blkfree: " "%jd-%jd(%d) overlaps with %jd-%jd", blkno, end, frags, jstart, jend); } } #endif FREE_LOCK(ump); } /* * Revert a block allocation when the journal record that describes it * is not yet written. */ static int jnewblk_rollback(jnewblk, fs, cgp, blksfree) struct jnewblk *jnewblk; struct fs *fs; struct cg *cgp; uint8_t *blksfree; { ufs1_daddr_t fragno; long cgbno, bbase; int frags, blk; int i; frags = 0; cgbno = dtogd(fs, jnewblk->jn_blkno); /* * We have to test which frags need to be rolled back. We may * be operating on a stale copy when doing background writes. */ for (i = jnewblk->jn_oldfrags; i < jnewblk->jn_frags; i++) if (isclr(blksfree, cgbno + i)) frags++; if (frags == 0) return (0); /* * This is mostly ffs_blkfree() sans some validation and * superblock updates. */ if (frags == fs->fs_frag) { fragno = fragstoblks(fs, cgbno); ffs_setblock(fs, blksfree, fragno); ffs_clusteracct(fs, cgp, fragno, 1); cgp->cg_cs.cs_nbfree++; } else { cgbno += jnewblk->jn_oldfrags; bbase = cgbno - fragnum(fs, cgbno); /* Decrement the old frags. */ blk = blkmap(fs, blksfree, bbase); ffs_fragacct(fs, blk, cgp->cg_frsum, -1); /* Deallocate the fragment */ for (i = 0; i < frags; i++) setbit(blksfree, cgbno + i); cgp->cg_cs.cs_nffree += frags; /* Add back in counts associated with the new frags */ blk = blkmap(fs, blksfree, bbase); ffs_fragacct(fs, blk, cgp->cg_frsum, 1); /* If a complete block has been reassembled, account for it. */ fragno = fragstoblks(fs, bbase); if (ffs_isblock(fs, blksfree, fragno)) { cgp->cg_cs.cs_nffree -= fs->fs_frag; ffs_clusteracct(fs, cgp, fragno, 1); cgp->cg_cs.cs_nbfree++; } } stat_jnewblk++; jnewblk->jn_state &= ~ATTACHED; jnewblk->jn_state |= UNDONE; return (frags); } static void initiate_write_bmsafemap(bmsafemap, bp) struct bmsafemap *bmsafemap; struct buf *bp; /* The cg block. */ { struct jaddref *jaddref; struct jnewblk *jnewblk; uint8_t *inosused; uint8_t *blksfree; struct cg *cgp; struct fs *fs; ino_t ino; /* * If this is a background write, we did this at the time that * the copy was made, so do not need to do it again. */ if (bmsafemap->sm_state & IOSTARTED) return; bmsafemap->sm_state |= IOSTARTED; /* * Clear any inode allocations which are pending journal writes. */ if (LIST_FIRST(&bmsafemap->sm_jaddrefhd) != NULL) { cgp = (struct cg *)bp->b_data; fs = VFSTOUFS(bmsafemap->sm_list.wk_mp)->um_fs; inosused = cg_inosused(cgp); LIST_FOREACH(jaddref, &bmsafemap->sm_jaddrefhd, ja_bmdeps) { ino = jaddref->ja_ino % fs->fs_ipg; if (isset(inosused, ino)) { if ((jaddref->ja_mode & IFMT) == IFDIR) cgp->cg_cs.cs_ndir--; cgp->cg_cs.cs_nifree++; clrbit(inosused, ino); jaddref->ja_state &= ~ATTACHED; jaddref->ja_state |= UNDONE; stat_jaddref++; } else panic("initiate_write_bmsafemap: inode %ju " "marked free", (uintmax_t)jaddref->ja_ino); } } /* * Clear any block allocations which are pending journal writes. */ if (LIST_FIRST(&bmsafemap->sm_jnewblkhd) != NULL) { cgp = (struct cg *)bp->b_data; fs = VFSTOUFS(bmsafemap->sm_list.wk_mp)->um_fs; blksfree = cg_blksfree(cgp); LIST_FOREACH(jnewblk, &bmsafemap->sm_jnewblkhd, jn_deps) { if (jnewblk_rollback(jnewblk, fs, cgp, blksfree)) continue; panic("initiate_write_bmsafemap: block %jd " "marked free", jnewblk->jn_blkno); } } /* * Move allocation lists to the written lists so they can be * cleared once the block write is complete. */ LIST_SWAP(&bmsafemap->sm_inodedephd, &bmsafemap->sm_inodedepwr, inodedep, id_deps); LIST_SWAP(&bmsafemap->sm_newblkhd, &bmsafemap->sm_newblkwr, newblk, nb_deps); LIST_SWAP(&bmsafemap->sm_freehd, &bmsafemap->sm_freewr, worklist, wk_list); } void softdep_handle_error(struct buf *bp) { struct ufsmount *ump; ump = softdep_bp_to_mp(bp); if (ump == NULL) return; if (ffs_fsfail_cleanup(ump, bp->b_error)) { /* * No future writes will succeed, so the on-disk image is safe. * Pretend that this write succeeded so that the softdep state * will be cleaned up naturally. */ bp->b_ioflags &= ~BIO_ERROR; bp->b_error = 0; } } /* * This routine is called during the completion interrupt * service routine for a disk write (from the procedure called * by the device driver to inform the filesystem caches of * a request completion). It should be called early in this * procedure, before the block is made available to other * processes or other routines are called. * */ static void softdep_disk_write_complete(bp) struct buf *bp; /* describes the completed disk write */ { struct worklist *wk; struct worklist *owk; struct ufsmount *ump; struct workhead reattach; struct freeblks *freeblks; struct buf *sbp; ump = softdep_bp_to_mp(bp); KASSERT(LIST_EMPTY(&bp->b_dep) || ump != NULL, ("softdep_disk_write_complete: softdep_bp_to_mp returned NULL " "with outstanding dependencies for buffer %p", bp)); if (ump == NULL) return; if ((bp->b_ioflags & BIO_ERROR) != 0) softdep_handle_error(bp); /* * If an error occurred while doing the write, then the data * has not hit the disk and the dependencies cannot be processed. * But we do have to go through and roll forward any dependencies * that were rolled back before the disk write. */ sbp = NULL; ACQUIRE_LOCK(ump); if ((bp->b_ioflags & BIO_ERROR) != 0 && (bp->b_flags & B_INVAL) == 0) { LIST_FOREACH(wk, &bp->b_dep, wk_list) { switch (wk->wk_type) { case D_PAGEDEP: handle_written_filepage(WK_PAGEDEP(wk), bp, 0); continue; case D_INODEDEP: handle_written_inodeblock(WK_INODEDEP(wk), bp, 0); continue; case D_BMSAFEMAP: handle_written_bmsafemap(WK_BMSAFEMAP(wk), bp, 0); continue; case D_INDIRDEP: handle_written_indirdep(WK_INDIRDEP(wk), bp, &sbp, 0); continue; default: /* nothing to roll forward */ continue; } } FREE_LOCK(ump); if (sbp) brelse(sbp); return; } LIST_INIT(&reattach); /* * Ump SU lock must not be released anywhere in this code segment. */ owk = NULL; while ((wk = LIST_FIRST(&bp->b_dep)) != NULL) { WORKLIST_REMOVE(wk); atomic_add_long(&dep_write[wk->wk_type], 1); if (wk == owk) panic("duplicate worklist: %p\n", wk); owk = wk; switch (wk->wk_type) { case D_PAGEDEP: if (handle_written_filepage(WK_PAGEDEP(wk), bp, WRITESUCCEEDED)) WORKLIST_INSERT(&reattach, wk); continue; case D_INODEDEP: if (handle_written_inodeblock(WK_INODEDEP(wk), bp, WRITESUCCEEDED)) WORKLIST_INSERT(&reattach, wk); continue; case D_BMSAFEMAP: if (handle_written_bmsafemap(WK_BMSAFEMAP(wk), bp, WRITESUCCEEDED)) WORKLIST_INSERT(&reattach, wk); continue; case D_MKDIR: handle_written_mkdir(WK_MKDIR(wk), MKDIR_BODY); continue; case D_ALLOCDIRECT: wk->wk_state |= COMPLETE; handle_allocdirect_partdone(WK_ALLOCDIRECT(wk), NULL); continue; case D_ALLOCINDIR: wk->wk_state |= COMPLETE; handle_allocindir_partdone(WK_ALLOCINDIR(wk)); continue; case D_INDIRDEP: if (handle_written_indirdep(WK_INDIRDEP(wk), bp, &sbp, WRITESUCCEEDED)) WORKLIST_INSERT(&reattach, wk); continue; case D_FREEBLKS: wk->wk_state |= COMPLETE; freeblks = WK_FREEBLKS(wk); if ((wk->wk_state & ALLCOMPLETE) == ALLCOMPLETE && LIST_EMPTY(&freeblks->fb_jblkdephd)) add_to_worklist(wk, WK_NODELAY); continue; case D_FREEWORK: handle_written_freework(WK_FREEWORK(wk)); break; case D_JSEGDEP: free_jsegdep(WK_JSEGDEP(wk)); continue; case D_JSEG: handle_written_jseg(WK_JSEG(wk), bp); continue; case D_SBDEP: if (handle_written_sbdep(WK_SBDEP(wk), bp)) WORKLIST_INSERT(&reattach, wk); continue; case D_FREEDEP: free_freedep(WK_FREEDEP(wk)); continue; default: panic("handle_disk_write_complete: Unknown type %s", TYPENAME(wk->wk_type)); /* NOTREACHED */ } } /* * Reattach any requests that must be redone. */ while ((wk = LIST_FIRST(&reattach)) != NULL) { WORKLIST_REMOVE(wk); WORKLIST_INSERT(&bp->b_dep, wk); } FREE_LOCK(ump); if (sbp) brelse(sbp); } /* * Called from within softdep_disk_write_complete above. */ static void handle_allocdirect_partdone(adp, wkhd) struct allocdirect *adp; /* the completed allocdirect */ struct workhead *wkhd; /* Work to do when inode is writtne. */ { struct allocdirectlst *listhead; struct allocdirect *listadp; struct inodedep *inodedep; long bsize; LOCK_OWNED(VFSTOUFS(adp->ad_block.nb_list.wk_mp)); if ((adp->ad_state & ALLCOMPLETE) != ALLCOMPLETE) return; /* * The on-disk inode cannot claim to be any larger than the last * fragment that has been written. Otherwise, the on-disk inode * might have fragments that were not the last block in the file * which would corrupt the filesystem. Thus, we cannot free any * allocdirects after one whose ad_oldblkno claims a fragment as * these blocks must be rolled back to zero before writing the inode. * We check the currently active set of allocdirects in id_inoupdt * or id_extupdt as appropriate. */ inodedep = adp->ad_inodedep; bsize = inodedep->id_fs->fs_bsize; if (adp->ad_state & EXTDATA) listhead = &inodedep->id_extupdt; else listhead = &inodedep->id_inoupdt; TAILQ_FOREACH(listadp, listhead, ad_next) { /* found our block */ if (listadp == adp) break; /* continue if ad_oldlbn is not a fragment */ if (listadp->ad_oldsize == 0 || listadp->ad_oldsize == bsize) continue; /* hit a fragment */ return; } /* * If we have reached the end of the current list without * finding the just finished dependency, then it must be * on the future dependency list. Future dependencies cannot * be freed until they are moved to the current list. */ if (listadp == NULL) { #ifdef INVARIANTS if (adp->ad_state & EXTDATA) listhead = &inodedep->id_newextupdt; else listhead = &inodedep->id_newinoupdt; TAILQ_FOREACH(listadp, listhead, ad_next) /* found our block */ if (listadp == adp) break; if (listadp == NULL) panic("handle_allocdirect_partdone: lost dep"); #endif /* INVARIANTS */ return; } /* * If we have found the just finished dependency, then queue * it along with anything that follows it that is complete. * Since the pointer has not yet been written in the inode * as the dependency prevents it, place the allocdirect on the * bufwait list where it will be freed once the pointer is * valid. */ if (wkhd == NULL) wkhd = &inodedep->id_bufwait; for (; adp; adp = listadp) { listadp = TAILQ_NEXT(adp, ad_next); if ((adp->ad_state & ALLCOMPLETE) != ALLCOMPLETE) return; TAILQ_REMOVE(listhead, adp, ad_next); WORKLIST_INSERT(wkhd, &adp->ad_block.nb_list); } } /* * Called from within softdep_disk_write_complete above. This routine * completes successfully written allocindirs. */ static void handle_allocindir_partdone(aip) struct allocindir *aip; /* the completed allocindir */ { struct indirdep *indirdep; if ((aip->ai_state & ALLCOMPLETE) != ALLCOMPLETE) return; indirdep = aip->ai_indirdep; LIST_REMOVE(aip, ai_next); /* * Don't set a pointer while the buffer is undergoing IO or while * we have active truncations. */ if (indirdep->ir_state & UNDONE || !TAILQ_EMPTY(&indirdep->ir_trunc)) { LIST_INSERT_HEAD(&indirdep->ir_donehd, aip, ai_next); return; } if (indirdep->ir_state & UFS1FMT) ((ufs1_daddr_t *)indirdep->ir_savebp->b_data)[aip->ai_offset] = aip->ai_newblkno; else ((ufs2_daddr_t *)indirdep->ir_savebp->b_data)[aip->ai_offset] = aip->ai_newblkno; /* * Await the pointer write before freeing the allocindir. */ LIST_INSERT_HEAD(&indirdep->ir_writehd, aip, ai_next); } /* * Release segments held on a jwork list. */ static void handle_jwork(wkhd) struct workhead *wkhd; { struct worklist *wk; while ((wk = LIST_FIRST(wkhd)) != NULL) { WORKLIST_REMOVE(wk); switch (wk->wk_type) { case D_JSEGDEP: free_jsegdep(WK_JSEGDEP(wk)); continue; case D_FREEDEP: free_freedep(WK_FREEDEP(wk)); continue; case D_FREEFRAG: rele_jseg(WK_JSEG(WK_FREEFRAG(wk)->ff_jdep)); WORKITEM_FREE(wk, D_FREEFRAG); continue; case D_FREEWORK: handle_written_freework(WK_FREEWORK(wk)); continue; default: panic("handle_jwork: Unknown type %s\n", TYPENAME(wk->wk_type)); } } } /* * Handle the bufwait list on an inode when it is safe to release items * held there. This normally happens after an inode block is written but * may be delayed and handled later if there are pending journal items that * are not yet safe to be released. */ static struct freefile * handle_bufwait(inodedep, refhd) struct inodedep *inodedep; struct workhead *refhd; { struct jaddref *jaddref; struct freefile *freefile; struct worklist *wk; freefile = NULL; while ((wk = LIST_FIRST(&inodedep->id_bufwait)) != NULL) { WORKLIST_REMOVE(wk); switch (wk->wk_type) { case D_FREEFILE: /* * We defer adding freefile to the worklist * until all other additions have been made to * ensure that it will be done after all the * old blocks have been freed. */ if (freefile != NULL) panic("handle_bufwait: freefile"); freefile = WK_FREEFILE(wk); continue; case D_MKDIR: handle_written_mkdir(WK_MKDIR(wk), MKDIR_PARENT); continue; case D_DIRADD: diradd_inode_written(WK_DIRADD(wk), inodedep); continue; case D_FREEFRAG: wk->wk_state |= COMPLETE; if ((wk->wk_state & ALLCOMPLETE) == ALLCOMPLETE) add_to_worklist(wk, 0); continue; case D_DIRREM: wk->wk_state |= COMPLETE; add_to_worklist(wk, 0); continue; case D_ALLOCDIRECT: case D_ALLOCINDIR: free_newblk(WK_NEWBLK(wk)); continue; case D_JNEWBLK: wk->wk_state |= COMPLETE; free_jnewblk(WK_JNEWBLK(wk)); continue; /* * Save freed journal segments and add references on * the supplied list which will delay their release * until the cg bitmap is cleared on disk. */ case D_JSEGDEP: if (refhd == NULL) free_jsegdep(WK_JSEGDEP(wk)); else WORKLIST_INSERT(refhd, wk); continue; case D_JADDREF: jaddref = WK_JADDREF(wk); TAILQ_REMOVE(&inodedep->id_inoreflst, &jaddref->ja_ref, if_deps); /* * Transfer any jaddrefs to the list to be freed with * the bitmap if we're handling a removed file. */ if (refhd == NULL) { wk->wk_state |= COMPLETE; free_jaddref(jaddref); } else WORKLIST_INSERT(refhd, wk); continue; default: panic("handle_bufwait: Unknown type %p(%s)", wk, TYPENAME(wk->wk_type)); /* NOTREACHED */ } } return (freefile); } /* * Called from within softdep_disk_write_complete above to restore * in-memory inode block contents to their most up-to-date state. Note * that this routine is always called from interrupt level with further * interrupts from this device blocked. * * If the write did not succeed, we will do all the roll-forward * operations, but we will not take the actions that will allow its * dependencies to be processed. */ static int handle_written_inodeblock(inodedep, bp, flags) struct inodedep *inodedep; struct buf *bp; /* buffer containing the inode block */ int flags; { struct freefile *freefile; struct allocdirect *adp, *nextadp; struct ufs1_dinode *dp1 = NULL; struct ufs2_dinode *dp2 = NULL; struct workhead wkhd; int hadchanges, fstype; ino_t freelink; LIST_INIT(&wkhd); hadchanges = 0; freefile = NULL; if ((inodedep->id_state & IOSTARTED) == 0) panic("handle_written_inodeblock: not started"); inodedep->id_state &= ~IOSTARTED; if (inodedep->id_fs->fs_magic == FS_UFS1_MAGIC) { fstype = UFS1; dp1 = (struct ufs1_dinode *)bp->b_data + ino_to_fsbo(inodedep->id_fs, inodedep->id_ino); freelink = dp1->di_freelink; } else { fstype = UFS2; dp2 = (struct ufs2_dinode *)bp->b_data + ino_to_fsbo(inodedep->id_fs, inodedep->id_ino); freelink = dp2->di_freelink; } /* * Leave this inodeblock dirty until it's in the list. */ if ((inodedep->id_state & (UNLINKED | UNLINKONLIST)) == UNLINKED && (flags & WRITESUCCEEDED)) { struct inodedep *inon; inon = TAILQ_NEXT(inodedep, id_unlinked); if ((inon == NULL && freelink == 0) || (inon && inon->id_ino == freelink)) { if (inon) inon->id_state |= UNLINKPREV; inodedep->id_state |= UNLINKNEXT; } hadchanges = 1; } /* * If we had to rollback the inode allocation because of * bitmaps being incomplete, then simply restore it. * Keep the block dirty so that it will not be reclaimed until * all associated dependencies have been cleared and the * corresponding updates written to disk. */ if (inodedep->id_savedino1 != NULL) { hadchanges = 1; if (fstype == UFS1) *dp1 = *inodedep->id_savedino1; else *dp2 = *inodedep->id_savedino2; free(inodedep->id_savedino1, M_SAVEDINO); inodedep->id_savedino1 = NULL; if ((bp->b_flags & B_DELWRI) == 0) stat_inode_bitmap++; bdirty(bp); /* * If the inode is clear here and GOINGAWAY it will never * be written. Process the bufwait and clear any pending * work which may include the freefile. */ if (inodedep->id_state & GOINGAWAY) goto bufwait; return (1); } if (flags & WRITESUCCEEDED) inodedep->id_state |= COMPLETE; /* * Roll forward anything that had to be rolled back before * the inode could be updated. */ for (adp = TAILQ_FIRST(&inodedep->id_inoupdt); adp; adp = nextadp) { nextadp = TAILQ_NEXT(adp, ad_next); if (adp->ad_state & ATTACHED) panic("handle_written_inodeblock: new entry"); if (fstype == UFS1) { if (adp->ad_offset < UFS_NDADDR) { if (dp1->di_db[adp->ad_offset]!=adp->ad_oldblkno) panic("%s %s #%jd mismatch %d != %jd", "handle_written_inodeblock:", "direct pointer", (intmax_t)adp->ad_offset, dp1->di_db[adp->ad_offset], (intmax_t)adp->ad_oldblkno); dp1->di_db[adp->ad_offset] = adp->ad_newblkno; } else { if (dp1->di_ib[adp->ad_offset - UFS_NDADDR] != 0) panic("%s: %s #%jd allocated as %d", "handle_written_inodeblock", "indirect pointer", (intmax_t)adp->ad_offset - UFS_NDADDR, dp1->di_ib[adp->ad_offset - UFS_NDADDR]); dp1->di_ib[adp->ad_offset - UFS_NDADDR] = adp->ad_newblkno; } } else { if (adp->ad_offset < UFS_NDADDR) { if (dp2->di_db[adp->ad_offset]!=adp->ad_oldblkno) panic("%s: %s #%jd %s %jd != %jd", "handle_written_inodeblock", "direct pointer", (intmax_t)adp->ad_offset, "mismatch", (intmax_t)dp2->di_db[adp->ad_offset], (intmax_t)adp->ad_oldblkno); dp2->di_db[adp->ad_offset] = adp->ad_newblkno; } else { if (dp2->di_ib[adp->ad_offset - UFS_NDADDR] != 0) panic("%s: %s #%jd allocated as %jd", "handle_written_inodeblock", "indirect pointer", (intmax_t)adp->ad_offset - UFS_NDADDR, (intmax_t) dp2->di_ib[adp->ad_offset - UFS_NDADDR]); dp2->di_ib[adp->ad_offset - UFS_NDADDR] = adp->ad_newblkno; } } adp->ad_state &= ~UNDONE; adp->ad_state |= ATTACHED; hadchanges = 1; } for (adp = TAILQ_FIRST(&inodedep->id_extupdt); adp; adp = nextadp) { nextadp = TAILQ_NEXT(adp, ad_next); if (adp->ad_state & ATTACHED) panic("handle_written_inodeblock: new entry"); if (dp2->di_extb[adp->ad_offset] != adp->ad_oldblkno) panic("%s: direct pointers #%jd %s %jd != %jd", "handle_written_inodeblock", (intmax_t)adp->ad_offset, "mismatch", (intmax_t)dp2->di_extb[adp->ad_offset], (intmax_t)adp->ad_oldblkno); dp2->di_extb[adp->ad_offset] = adp->ad_newblkno; adp->ad_state &= ~UNDONE; adp->ad_state |= ATTACHED; hadchanges = 1; } if (hadchanges && (bp->b_flags & B_DELWRI) == 0) stat_direct_blk_ptrs++; /* * Reset the file size to its most up-to-date value. */ if (inodedep->id_savedsize == -1 || inodedep->id_savedextsize == -1) panic("handle_written_inodeblock: bad size"); if (inodedep->id_savednlink > UFS_LINK_MAX) panic("handle_written_inodeblock: Invalid link count " "%jd for inodedep %p", (uintmax_t)inodedep->id_savednlink, inodedep); if (fstype == UFS1) { if (dp1->di_nlink != inodedep->id_savednlink) { dp1->di_nlink = inodedep->id_savednlink; hadchanges = 1; } if (dp1->di_size != inodedep->id_savedsize) { dp1->di_size = inodedep->id_savedsize; hadchanges = 1; } } else { if (dp2->di_nlink != inodedep->id_savednlink) { dp2->di_nlink = inodedep->id_savednlink; hadchanges = 1; } if (dp2->di_size != inodedep->id_savedsize) { dp2->di_size = inodedep->id_savedsize; hadchanges = 1; } if (dp2->di_extsize != inodedep->id_savedextsize) { dp2->di_extsize = inodedep->id_savedextsize; hadchanges = 1; } } inodedep->id_savedsize = -1; inodedep->id_savedextsize = -1; inodedep->id_savednlink = -1; /* * If there were any rollbacks in the inode block, then it must be * marked dirty so that its will eventually get written back in * its correct form. */ if (hadchanges) { if (fstype == UFS2) ffs_update_dinode_ckhash(inodedep->id_fs, dp2); bdirty(bp); } bufwait: /* * If the write did not succeed, we have done all the roll-forward * operations, but we cannot take the actions that will allow its * dependencies to be processed. */ if ((flags & WRITESUCCEEDED) == 0) return (hadchanges); /* * Process any allocdirects that completed during the update. */ if ((adp = TAILQ_FIRST(&inodedep->id_inoupdt)) != NULL) handle_allocdirect_partdone(adp, &wkhd); if ((adp = TAILQ_FIRST(&inodedep->id_extupdt)) != NULL) handle_allocdirect_partdone(adp, &wkhd); /* * Process deallocations that were held pending until the * inode had been written to disk. Freeing of the inode * is delayed until after all blocks have been freed to * avoid creation of new triples * before the old ones have been deleted. Completely * unlinked inodes are not processed until the unlinked * inode list is written or the last reference is removed. */ if ((inodedep->id_state & (UNLINKED | UNLINKONLIST)) != UNLINKED) { freefile = handle_bufwait(inodedep, NULL); if (freefile && !LIST_EMPTY(&wkhd)) { WORKLIST_INSERT(&wkhd, &freefile->fx_list); freefile = NULL; } } /* * Move rolled forward dependency completions to the bufwait list * now that those that were already written have been processed. */ if (!LIST_EMPTY(&wkhd) && hadchanges == 0) panic("handle_written_inodeblock: bufwait but no changes"); jwork_move(&inodedep->id_bufwait, &wkhd); if (freefile != NULL) { /* * If the inode is goingaway it was never written. Fake up * the state here so free_inodedep() can succeed. */ if (inodedep->id_state & GOINGAWAY) inodedep->id_state |= COMPLETE | DEPCOMPLETE; if (free_inodedep(inodedep) == 0) panic("handle_written_inodeblock: live inodedep %p", inodedep); add_to_worklist(&freefile->fx_list, 0); return (0); } /* * If no outstanding dependencies, free it. */ if (free_inodedep(inodedep) || (TAILQ_FIRST(&inodedep->id_inoreflst) == 0 && TAILQ_FIRST(&inodedep->id_inoupdt) == 0 && TAILQ_FIRST(&inodedep->id_extupdt) == 0 && LIST_FIRST(&inodedep->id_bufwait) == 0)) return (0); return (hadchanges); } /* * Perform needed roll-forwards and kick off any dependencies that * can now be processed. * * If the write did not succeed, we will do all the roll-forward * operations, but we will not take the actions that will allow its * dependencies to be processed. */ static int handle_written_indirdep(indirdep, bp, bpp, flags) struct indirdep *indirdep; struct buf *bp; struct buf **bpp; int flags; { struct allocindir *aip; struct buf *sbp; int chgs; if (indirdep->ir_state & GOINGAWAY) panic("handle_written_indirdep: indirdep gone"); if ((indirdep->ir_state & IOSTARTED) == 0) panic("handle_written_indirdep: IO not started"); chgs = 0; /* * If there were rollbacks revert them here. */ if (indirdep->ir_saveddata) { bcopy(indirdep->ir_saveddata, bp->b_data, bp->b_bcount); if (TAILQ_EMPTY(&indirdep->ir_trunc)) { free(indirdep->ir_saveddata, M_INDIRDEP); indirdep->ir_saveddata = NULL; } chgs = 1; } indirdep->ir_state &= ~(UNDONE | IOSTARTED); indirdep->ir_state |= ATTACHED; /* * If the write did not succeed, we have done all the roll-forward * operations, but we cannot take the actions that will allow its * dependencies to be processed. */ if ((flags & WRITESUCCEEDED) == 0) { stat_indir_blk_ptrs++; bdirty(bp); return (1); } /* * Move allocindirs with written pointers to the completehd if * the indirdep's pointer is not yet written. Otherwise * free them here. */ while ((aip = LIST_FIRST(&indirdep->ir_writehd)) != NULL) { LIST_REMOVE(aip, ai_next); if ((indirdep->ir_state & DEPCOMPLETE) == 0) { LIST_INSERT_HEAD(&indirdep->ir_completehd, aip, ai_next); newblk_freefrag(&aip->ai_block); continue; } free_newblk(&aip->ai_block); } /* * Move allocindirs that have finished dependency processing from * the done list to the write list after updating the pointers. */ if (TAILQ_EMPTY(&indirdep->ir_trunc)) { while ((aip = LIST_FIRST(&indirdep->ir_donehd)) != NULL) { handle_allocindir_partdone(aip); if (aip == LIST_FIRST(&indirdep->ir_donehd)) panic("disk_write_complete: not gone"); chgs = 1; } } /* * Preserve the indirdep if there were any changes or if it is not * yet valid on disk. */ if (chgs) { stat_indir_blk_ptrs++; bdirty(bp); return (1); } /* * If there were no changes we can discard the savedbp and detach * ourselves from the buf. We are only carrying completed pointers * in this case. */ sbp = indirdep->ir_savebp; sbp->b_flags |= B_INVAL | B_NOCACHE; indirdep->ir_savebp = NULL; indirdep->ir_bp = NULL; if (*bpp != NULL) panic("handle_written_indirdep: bp already exists."); *bpp = sbp; /* * The indirdep may not be freed until its parent points at it. */ if (indirdep->ir_state & DEPCOMPLETE) free_indirdep(indirdep); return (0); } /* * Process a diradd entry after its dependent inode has been written. */ static void diradd_inode_written(dap, inodedep) struct diradd *dap; struct inodedep *inodedep; { LOCK_OWNED(VFSTOUFS(dap->da_list.wk_mp)); dap->da_state |= COMPLETE; complete_diradd(dap); WORKLIST_INSERT(&inodedep->id_pendinghd, &dap->da_list); } /* * Returns true if the bmsafemap will have rollbacks when written. Must only * be called with the per-filesystem lock and the buf lock on the cg held. */ static int bmsafemap_backgroundwrite(bmsafemap, bp) struct bmsafemap *bmsafemap; struct buf *bp; { int dirty; LOCK_OWNED(VFSTOUFS(bmsafemap->sm_list.wk_mp)); dirty = !LIST_EMPTY(&bmsafemap->sm_jaddrefhd) | !LIST_EMPTY(&bmsafemap->sm_jnewblkhd); /* * If we're initiating a background write we need to process the * rollbacks as they exist now, not as they exist when IO starts. * No other consumers will look at the contents of the shadowed * buf so this is safe to do here. */ if (bp->b_xflags & BX_BKGRDMARKER) initiate_write_bmsafemap(bmsafemap, bp); return (dirty); } /* * Re-apply an allocation when a cg write is complete. */ static int jnewblk_rollforward(jnewblk, fs, cgp, blksfree) struct jnewblk *jnewblk; struct fs *fs; struct cg *cgp; uint8_t *blksfree; { ufs1_daddr_t fragno; ufs2_daddr_t blkno; long cgbno, bbase; int frags, blk; int i; frags = 0; cgbno = dtogd(fs, jnewblk->jn_blkno); for (i = jnewblk->jn_oldfrags; i < jnewblk->jn_frags; i++) { if (isclr(blksfree, cgbno + i)) panic("jnewblk_rollforward: re-allocated fragment"); frags++; } if (frags == fs->fs_frag) { blkno = fragstoblks(fs, cgbno); ffs_clrblock(fs, blksfree, (long)blkno); ffs_clusteracct(fs, cgp, blkno, -1); cgp->cg_cs.cs_nbfree--; } else { bbase = cgbno - fragnum(fs, cgbno); cgbno += jnewblk->jn_oldfrags; /* If a complete block had been reassembled, account for it. */ fragno = fragstoblks(fs, bbase); if (ffs_isblock(fs, blksfree, fragno)) { cgp->cg_cs.cs_nffree += fs->fs_frag; ffs_clusteracct(fs, cgp, fragno, -1); cgp->cg_cs.cs_nbfree--; } /* Decrement the old frags. */ blk = blkmap(fs, blksfree, bbase); ffs_fragacct(fs, blk, cgp->cg_frsum, -1); /* Allocate the fragment */ for (i = 0; i < frags; i++) clrbit(blksfree, cgbno + i); cgp->cg_cs.cs_nffree -= frags; /* Add back in counts associated with the new frags */ blk = blkmap(fs, blksfree, bbase); ffs_fragacct(fs, blk, cgp->cg_frsum, 1); } return (frags); } /* * Complete a write to a bmsafemap structure. Roll forward any bitmap * changes if it's not a background write. Set all written dependencies * to DEPCOMPLETE and free the structure if possible. * * If the write did not succeed, we will do all the roll-forward * operations, but we will not take the actions that will allow its * dependencies to be processed. */ static int handle_written_bmsafemap(bmsafemap, bp, flags) struct bmsafemap *bmsafemap; struct buf *bp; int flags; { struct newblk *newblk; struct inodedep *inodedep; struct jaddref *jaddref, *jatmp; struct jnewblk *jnewblk, *jntmp; struct ufsmount *ump; uint8_t *inosused; uint8_t *blksfree; struct cg *cgp; struct fs *fs; ino_t ino; int foreground; int chgs; if ((bmsafemap->sm_state & IOSTARTED) == 0) panic("handle_written_bmsafemap: Not started\n"); ump = VFSTOUFS(bmsafemap->sm_list.wk_mp); chgs = 0; bmsafemap->sm_state &= ~IOSTARTED; foreground = (bp->b_xflags & BX_BKGRDMARKER) == 0; /* * If write was successful, release journal work that was waiting * on the write. Otherwise move the work back. */ if (flags & WRITESUCCEEDED) handle_jwork(&bmsafemap->sm_freewr); else LIST_CONCAT(&bmsafemap->sm_freehd, &bmsafemap->sm_freewr, worklist, wk_list); /* * Restore unwritten inode allocation pending jaddref writes. */ if (!LIST_EMPTY(&bmsafemap->sm_jaddrefhd)) { cgp = (struct cg *)bp->b_data; fs = VFSTOUFS(bmsafemap->sm_list.wk_mp)->um_fs; inosused = cg_inosused(cgp); LIST_FOREACH_SAFE(jaddref, &bmsafemap->sm_jaddrefhd, ja_bmdeps, jatmp) { if ((jaddref->ja_state & UNDONE) == 0) continue; ino = jaddref->ja_ino % fs->fs_ipg; if (isset(inosused, ino)) panic("handle_written_bmsafemap: " "re-allocated inode"); /* Do the roll-forward only if it's a real copy. */ if (foreground) { if ((jaddref->ja_mode & IFMT) == IFDIR) cgp->cg_cs.cs_ndir++; cgp->cg_cs.cs_nifree--; setbit(inosused, ino); chgs = 1; } jaddref->ja_state &= ~UNDONE; jaddref->ja_state |= ATTACHED; free_jaddref(jaddref); } } /* * Restore any block allocations which are pending journal writes. */ if (LIST_FIRST(&bmsafemap->sm_jnewblkhd) != NULL) { cgp = (struct cg *)bp->b_data; fs = VFSTOUFS(bmsafemap->sm_list.wk_mp)->um_fs; blksfree = cg_blksfree(cgp); LIST_FOREACH_SAFE(jnewblk, &bmsafemap->sm_jnewblkhd, jn_deps, jntmp) { if ((jnewblk->jn_state & UNDONE) == 0) continue; /* Do the roll-forward only if it's a real copy. */ if (foreground && jnewblk_rollforward(jnewblk, fs, cgp, blksfree)) chgs = 1; jnewblk->jn_state &= ~(UNDONE | NEWBLOCK); jnewblk->jn_state |= ATTACHED; free_jnewblk(jnewblk); } } /* * If the write did not succeed, we have done all the roll-forward * operations, but we cannot take the actions that will allow its * dependencies to be processed. */ if ((flags & WRITESUCCEEDED) == 0) { LIST_CONCAT(&bmsafemap->sm_newblkhd, &bmsafemap->sm_newblkwr, newblk, nb_deps); LIST_CONCAT(&bmsafemap->sm_freehd, &bmsafemap->sm_freewr, worklist, wk_list); if (foreground) bdirty(bp); return (1); } while ((newblk = LIST_FIRST(&bmsafemap->sm_newblkwr))) { newblk->nb_state |= DEPCOMPLETE; newblk->nb_state &= ~ONDEPLIST; newblk->nb_bmsafemap = NULL; LIST_REMOVE(newblk, nb_deps); if (newblk->nb_list.wk_type == D_ALLOCDIRECT) handle_allocdirect_partdone( WK_ALLOCDIRECT(&newblk->nb_list), NULL); else if (newblk->nb_list.wk_type == D_ALLOCINDIR) handle_allocindir_partdone( WK_ALLOCINDIR(&newblk->nb_list)); else if (newblk->nb_list.wk_type != D_NEWBLK) panic("handle_written_bmsafemap: Unexpected type: %s", TYPENAME(newblk->nb_list.wk_type)); } while ((inodedep = LIST_FIRST(&bmsafemap->sm_inodedepwr)) != NULL) { inodedep->id_state |= DEPCOMPLETE; inodedep->id_state &= ~ONDEPLIST; LIST_REMOVE(inodedep, id_deps); inodedep->id_bmsafemap = NULL; } LIST_REMOVE(bmsafemap, sm_next); if (chgs == 0 && LIST_EMPTY(&bmsafemap->sm_jaddrefhd) && LIST_EMPTY(&bmsafemap->sm_jnewblkhd) && LIST_EMPTY(&bmsafemap->sm_newblkhd) && LIST_EMPTY(&bmsafemap->sm_inodedephd) && LIST_EMPTY(&bmsafemap->sm_freehd)) { LIST_REMOVE(bmsafemap, sm_hash); WORKITEM_FREE(bmsafemap, D_BMSAFEMAP); return (0); } LIST_INSERT_HEAD(&ump->softdep_dirtycg, bmsafemap, sm_next); if (foreground) bdirty(bp); return (1); } /* * Try to free a mkdir dependency. */ static void complete_mkdir(mkdir) struct mkdir *mkdir; { struct diradd *dap; if ((mkdir->md_state & ALLCOMPLETE) != ALLCOMPLETE) return; LIST_REMOVE(mkdir, md_mkdirs); dap = mkdir->md_diradd; dap->da_state &= ~(mkdir->md_state & (MKDIR_PARENT | MKDIR_BODY)); if ((dap->da_state & (MKDIR_PARENT | MKDIR_BODY)) == 0) { dap->da_state |= DEPCOMPLETE; complete_diradd(dap); } WORKITEM_FREE(mkdir, D_MKDIR); } /* * Handle the completion of a mkdir dependency. */ static void handle_written_mkdir(mkdir, type) struct mkdir *mkdir; int type; { if ((mkdir->md_state & (MKDIR_PARENT | MKDIR_BODY)) != type) panic("handle_written_mkdir: bad type"); mkdir->md_state |= COMPLETE; complete_mkdir(mkdir); } static int free_pagedep(pagedep) struct pagedep *pagedep; { int i; if (pagedep->pd_state & NEWBLOCK) return (0); if (!LIST_EMPTY(&pagedep->pd_dirremhd)) return (0); for (i = 0; i < DAHASHSZ; i++) if (!LIST_EMPTY(&pagedep->pd_diraddhd[i])) return (0); if (!LIST_EMPTY(&pagedep->pd_pendinghd)) return (0); if (!LIST_EMPTY(&pagedep->pd_jmvrefhd)) return (0); if (pagedep->pd_state & ONWORKLIST) WORKLIST_REMOVE(&pagedep->pd_list); LIST_REMOVE(pagedep, pd_hash); WORKITEM_FREE(pagedep, D_PAGEDEP); return (1); } /* * Called from within softdep_disk_write_complete above. * A write operation was just completed. Removed inodes can * now be freed and associated block pointers may be committed. * Note that this routine is always called from interrupt level * with further interrupts from this device blocked. * * If the write did not succeed, we will do all the roll-forward * operations, but we will not take the actions that will allow its * dependencies to be processed. */ static int handle_written_filepage(pagedep, bp, flags) struct pagedep *pagedep; struct buf *bp; /* buffer containing the written page */ int flags; { struct dirrem *dirrem; struct diradd *dap, *nextdap; struct direct *ep; int i, chgs; if ((pagedep->pd_state & IOSTARTED) == 0) panic("handle_written_filepage: not started"); pagedep->pd_state &= ~IOSTARTED; if ((flags & WRITESUCCEEDED) == 0) goto rollforward; /* * Process any directory removals that have been committed. */ while ((dirrem = LIST_FIRST(&pagedep->pd_dirremhd)) != NULL) { LIST_REMOVE(dirrem, dm_next); dirrem->dm_state |= COMPLETE; dirrem->dm_dirinum = pagedep->pd_ino; KASSERT(LIST_EMPTY(&dirrem->dm_jremrefhd), ("handle_written_filepage: Journal entries not written.")); add_to_worklist(&dirrem->dm_list, 0); } /* * Free any directory additions that have been committed. * If it is a newly allocated block, we have to wait until * the on-disk directory inode claims the new block. */ if ((pagedep->pd_state & NEWBLOCK) == 0) while ((dap = LIST_FIRST(&pagedep->pd_pendinghd)) != NULL) free_diradd(dap, NULL); rollforward: /* * Uncommitted directory entries must be restored. */ for (chgs = 0, i = 0; i < DAHASHSZ; i++) { for (dap = LIST_FIRST(&pagedep->pd_diraddhd[i]); dap; dap = nextdap) { nextdap = LIST_NEXT(dap, da_pdlist); if (dap->da_state & ATTACHED) panic("handle_written_filepage: attached"); ep = (struct direct *) ((char *)bp->b_data + dap->da_offset); ep->d_ino = dap->da_newinum; dap->da_state &= ~UNDONE; dap->da_state |= ATTACHED; chgs = 1; /* * If the inode referenced by the directory has * been written out, then the dependency can be * moved to the pending list. */ if ((dap->da_state & ALLCOMPLETE) == ALLCOMPLETE) { LIST_REMOVE(dap, da_pdlist); LIST_INSERT_HEAD(&pagedep->pd_pendinghd, dap, da_pdlist); } } } /* * If there were any rollbacks in the directory, then it must be * marked dirty so that its will eventually get written back in * its correct form. */ if (chgs || (flags & WRITESUCCEEDED) == 0) { if ((bp->b_flags & B_DELWRI) == 0) stat_dir_entry++; bdirty(bp); return (1); } /* * If we are not waiting for a new directory block to be * claimed by its inode, then the pagedep will be freed. * Otherwise it will remain to track any new entries on * the page in case they are fsync'ed. */ free_pagedep(pagedep); return (0); } /* * Writing back in-core inode structures. * * The filesystem only accesses an inode's contents when it occupies an * "in-core" inode structure. These "in-core" structures are separate from * the page frames used to cache inode blocks. Only the latter are * transferred to/from the disk. So, when the updated contents of the * "in-core" inode structure are copied to the corresponding in-memory inode * block, the dependencies are also transferred. The following procedure is * called when copying a dirty "in-core" inode to a cached inode block. */ /* * Called when an inode is loaded from disk. If the effective link count * differed from the actual link count when it was last flushed, then we * need to ensure that the correct effective link count is put back. */ void softdep_load_inodeblock(ip) struct inode *ip; /* the "in_core" copy of the inode */ { struct inodedep *inodedep; struct ufsmount *ump; ump = ITOUMP(ip); KASSERT(MOUNTEDSOFTDEP(UFSTOVFS(ump)) != 0, ("softdep_load_inodeblock called on non-softdep filesystem")); /* * Check for alternate nlink count. */ ip->i_effnlink = ip->i_nlink; ACQUIRE_LOCK(ump); if (inodedep_lookup(UFSTOVFS(ump), ip->i_number, 0, &inodedep) == 0) { FREE_LOCK(ump); return; } if (ip->i_nlink != inodedep->id_nlinkwrote && inodedep->id_nlinkwrote != -1) { KASSERT(ip->i_nlink == 0 && (ump->um_flags & UM_FSFAIL_CLEANUP) != 0, ("read bad i_nlink value")); ip->i_effnlink = ip->i_nlink = inodedep->id_nlinkwrote; } ip->i_effnlink -= inodedep->id_nlinkdelta; KASSERT(ip->i_effnlink >= 0, ("softdep_load_inodeblock: negative i_effnlink")); FREE_LOCK(ump); } /* * This routine is called just before the "in-core" inode * information is to be copied to the in-memory inode block. * Recall that an inode block contains several inodes. If * the force flag is set, then the dependencies will be * cleared so that the update can always be made. Note that * the buffer is locked when this routine is called, so we * will never be in the middle of writing the inode block * to disk. */ void softdep_update_inodeblock(ip, bp, waitfor) struct inode *ip; /* the "in_core" copy of the inode */ struct buf *bp; /* the buffer containing the inode block */ int waitfor; /* nonzero => update must be allowed */ { struct inodedep *inodedep; struct inoref *inoref; struct ufsmount *ump; struct worklist *wk; struct mount *mp; struct buf *ibp; struct fs *fs; int error; ump = ITOUMP(ip); mp = UFSTOVFS(ump); KASSERT(MOUNTEDSOFTDEP(mp) != 0, ("softdep_update_inodeblock called on non-softdep filesystem")); fs = ump->um_fs; /* * Preserve the freelink that is on disk. clear_unlinked_inodedep() * does not have access to the in-core ip so must write directly into * the inode block buffer when setting freelink. */ if (fs->fs_magic == FS_UFS1_MAGIC) DIP_SET(ip, i_freelink, ((struct ufs1_dinode *)bp->b_data + ino_to_fsbo(fs, ip->i_number))->di_freelink); else DIP_SET(ip, i_freelink, ((struct ufs2_dinode *)bp->b_data + ino_to_fsbo(fs, ip->i_number))->di_freelink); /* * If the effective link count is not equal to the actual link * count, then we must track the difference in an inodedep while * the inode is (potentially) tossed out of the cache. Otherwise, * if there is no existing inodedep, then there are no dependencies * to track. */ ACQUIRE_LOCK(ump); again: if (inodedep_lookup(mp, ip->i_number, 0, &inodedep) == 0) { FREE_LOCK(ump); if (ip->i_effnlink != ip->i_nlink) panic("softdep_update_inodeblock: bad link count"); return; } KASSERT(ip->i_nlink >= inodedep->id_nlinkdelta, ("softdep_update_inodeblock inconsistent ip %p i_nlink %d " "inodedep %p id_nlinkdelta %jd", ip, ip->i_nlink, inodedep, (intmax_t)inodedep->id_nlinkdelta)); inodedep->id_nlinkwrote = ip->i_nlink; if (inodedep->id_nlinkdelta != ip->i_nlink - ip->i_effnlink) panic("softdep_update_inodeblock: bad delta"); /* * If we're flushing all dependencies we must also move any waiting * for journal writes onto the bufwait list prior to I/O. */ if (waitfor) { TAILQ_FOREACH(inoref, &inodedep->id_inoreflst, if_deps) { if ((inoref->if_state & (DEPCOMPLETE | GOINGAWAY)) == DEPCOMPLETE) { jwait(&inoref->if_list, MNT_WAIT); goto again; } } } /* * Changes have been initiated. Anything depending on these * changes cannot occur until this inode has been written. */ inodedep->id_state &= ~COMPLETE; if ((inodedep->id_state & ONWORKLIST) == 0) WORKLIST_INSERT(&bp->b_dep, &inodedep->id_list); /* * Any new dependencies associated with the incore inode must * now be moved to the list associated with the buffer holding * the in-memory copy of the inode. Once merged process any * allocdirects that are completed by the merger. */ merge_inode_lists(&inodedep->id_newinoupdt, &inodedep->id_inoupdt); if (!TAILQ_EMPTY(&inodedep->id_inoupdt)) handle_allocdirect_partdone(TAILQ_FIRST(&inodedep->id_inoupdt), NULL); merge_inode_lists(&inodedep->id_newextupdt, &inodedep->id_extupdt); if (!TAILQ_EMPTY(&inodedep->id_extupdt)) handle_allocdirect_partdone(TAILQ_FIRST(&inodedep->id_extupdt), NULL); /* * Now that the inode has been pushed into the buffer, the * operations dependent on the inode being written to disk * can be moved to the id_bufwait so that they will be * processed when the buffer I/O completes. */ while ((wk = LIST_FIRST(&inodedep->id_inowait)) != NULL) { WORKLIST_REMOVE(wk); WORKLIST_INSERT(&inodedep->id_bufwait, wk); } /* * Newly allocated inodes cannot be written until the bitmap * that allocates them have been written (indicated by * DEPCOMPLETE being set in id_state). If we are doing a * forced sync (e.g., an fsync on a file), we force the bitmap * to be written so that the update can be done. */ if (waitfor == 0) { FREE_LOCK(ump); return; } retry: if ((inodedep->id_state & (DEPCOMPLETE | GOINGAWAY)) != 0) { FREE_LOCK(ump); return; } ibp = inodedep->id_bmsafemap->sm_buf; ibp = getdirtybuf(ibp, LOCK_PTR(ump), MNT_WAIT); if (ibp == NULL) { /* * If ibp came back as NULL, the dependency could have been * freed while we slept. Look it up again, and check to see * that it has completed. */ if (inodedep_lookup(mp, ip->i_number, 0, &inodedep) != 0) goto retry; FREE_LOCK(ump); return; } FREE_LOCK(ump); if ((error = bwrite(ibp)) != 0) softdep_error("softdep_update_inodeblock: bwrite", error); } /* * Merge the a new inode dependency list (such as id_newinoupdt) into an * old inode dependency list (such as id_inoupdt). */ static void merge_inode_lists(newlisthead, oldlisthead) struct allocdirectlst *newlisthead; struct allocdirectlst *oldlisthead; { struct allocdirect *listadp, *newadp; newadp = TAILQ_FIRST(newlisthead); if (newadp != NULL) LOCK_OWNED(VFSTOUFS(newadp->ad_block.nb_list.wk_mp)); for (listadp = TAILQ_FIRST(oldlisthead); listadp && newadp;) { if (listadp->ad_offset < newadp->ad_offset) { listadp = TAILQ_NEXT(listadp, ad_next); continue; } TAILQ_REMOVE(newlisthead, newadp, ad_next); TAILQ_INSERT_BEFORE(listadp, newadp, ad_next); if (listadp->ad_offset == newadp->ad_offset) { allocdirect_merge(oldlisthead, newadp, listadp); listadp = newadp; } newadp = TAILQ_FIRST(newlisthead); } while ((newadp = TAILQ_FIRST(newlisthead)) != NULL) { TAILQ_REMOVE(newlisthead, newadp, ad_next); TAILQ_INSERT_TAIL(oldlisthead, newadp, ad_next); } } /* * If we are doing an fsync, then we must ensure that any directory * entries for the inode have been written after the inode gets to disk. */ int softdep_fsync(vp) struct vnode *vp; /* the "in_core" copy of the inode */ { struct inodedep *inodedep; struct pagedep *pagedep; struct inoref *inoref; struct ufsmount *ump; struct worklist *wk; struct diradd *dap; struct mount *mp; struct vnode *pvp; struct inode *ip; struct buf *bp; struct fs *fs; struct thread *td = curthread; int error, flushparent, pagedep_new_block; ino_t parentino; ufs_lbn_t lbn; ip = VTOI(vp); mp = vp->v_mount; ump = VFSTOUFS(mp); fs = ump->um_fs; if (MOUNTEDSOFTDEP(mp) == 0) return (0); ACQUIRE_LOCK(ump); restart: if (inodedep_lookup(mp, ip->i_number, 0, &inodedep) == 0) { FREE_LOCK(ump); return (0); } TAILQ_FOREACH(inoref, &inodedep->id_inoreflst, if_deps) { if ((inoref->if_state & (DEPCOMPLETE | GOINGAWAY)) == DEPCOMPLETE) { jwait(&inoref->if_list, MNT_WAIT); goto restart; } } if (!LIST_EMPTY(&inodedep->id_inowait) || !TAILQ_EMPTY(&inodedep->id_extupdt) || !TAILQ_EMPTY(&inodedep->id_newextupdt) || !TAILQ_EMPTY(&inodedep->id_inoupdt) || !TAILQ_EMPTY(&inodedep->id_newinoupdt)) panic("softdep_fsync: pending ops %p", inodedep); for (error = 0, flushparent = 0; ; ) { if ((wk = LIST_FIRST(&inodedep->id_pendinghd)) == NULL) break; if (wk->wk_type != D_DIRADD) panic("softdep_fsync: Unexpected type %s", TYPENAME(wk->wk_type)); dap = WK_DIRADD(wk); /* * Flush our parent if this directory entry has a MKDIR_PARENT * dependency or is contained in a newly allocated block. */ if (dap->da_state & DIRCHG) pagedep = dap->da_previous->dm_pagedep; else pagedep = dap->da_pagedep; parentino = pagedep->pd_ino; lbn = pagedep->pd_lbn; if ((dap->da_state & (MKDIR_BODY | COMPLETE)) != COMPLETE) panic("softdep_fsync: dirty"); if ((dap->da_state & MKDIR_PARENT) || (pagedep->pd_state & NEWBLOCK)) flushparent = 1; else flushparent = 0; /* * If we are being fsync'ed as part of vgone'ing this vnode, * then we will not be able to release and recover the * vnode below, so we just have to give up on writing its * directory entry out. It will eventually be written, just * not now, but then the user was not asking to have it * written, so we are not breaking any promises. */ if (VN_IS_DOOMED(vp)) break; /* * We prevent deadlock by always fetching inodes from the * root, moving down the directory tree. Thus, when fetching * our parent directory, we first try to get the lock. If * that fails, we must unlock ourselves before requesting * the lock on our parent. See the comment in ufs_lookup * for details on possible races. */ FREE_LOCK(ump); error = get_parent_vp(vp, mp, parentino, NULL, NULL, NULL, &pvp); if (error == ERELOOKUP) error = 0; if (error != 0) return (error); /* * All MKDIR_PARENT dependencies and all the NEWBLOCK pagedeps * that are contained in direct blocks will be resolved by * doing a ffs_update. Pagedeps contained in indirect blocks * may require a complete sync'ing of the directory. So, we * try the cheap and fast ffs_update first, and if that fails, * then we do the slower ffs_syncvnode of the directory. */ if (flushparent) { int locked; if ((error = ffs_update(pvp, 1)) != 0) { vput(pvp); return (error); } ACQUIRE_LOCK(ump); locked = 1; if (inodedep_lookup(mp, ip->i_number, 0, &inodedep) != 0) { if ((wk = LIST_FIRST(&inodedep->id_pendinghd)) != NULL) { if (wk->wk_type != D_DIRADD) panic("softdep_fsync: Unexpected type %s", TYPENAME(wk->wk_type)); dap = WK_DIRADD(wk); if (dap->da_state & DIRCHG) pagedep = dap->da_previous->dm_pagedep; else pagedep = dap->da_pagedep; pagedep_new_block = pagedep->pd_state & NEWBLOCK; FREE_LOCK(ump); locked = 0; if (pagedep_new_block && (error = ffs_syncvnode(pvp, MNT_WAIT, 0))) { vput(pvp); return (error); } } } if (locked) FREE_LOCK(ump); } /* * Flush directory page containing the inode's name. */ error = bread(pvp, lbn, blksize(fs, VTOI(pvp), lbn), td->td_ucred, &bp); if (error == 0) error = bwrite(bp); else brelse(bp); vput(pvp); if (!ffs_fsfail_cleanup(ump, error)) return (error); ACQUIRE_LOCK(ump); if (inodedep_lookup(mp, ip->i_number, 0, &inodedep) == 0) break; } FREE_LOCK(ump); return (0); } /* * Flush all the dirty bitmaps associated with the block device * before flushing the rest of the dirty blocks so as to reduce * the number of dependencies that will have to be rolled back. * * XXX Unused? */ void softdep_fsync_mountdev(vp) struct vnode *vp; { struct buf *bp, *nbp; struct worklist *wk; struct bufobj *bo; if (!vn_isdisk(vp)) panic("softdep_fsync_mountdev: vnode not a disk"); bo = &vp->v_bufobj; restart: BO_LOCK(bo); TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { /* * If it is already scheduled, skip to the next buffer. */ if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) continue; if ((bp->b_flags & B_DELWRI) == 0) panic("softdep_fsync_mountdev: not dirty"); /* * We are only interested in bitmaps with outstanding * dependencies. */ if ((wk = LIST_FIRST(&bp->b_dep)) == NULL || wk->wk_type != D_BMSAFEMAP || (bp->b_vflags & BV_BKGRDINPROG)) { BUF_UNLOCK(bp); continue; } BO_UNLOCK(bo); bremfree(bp); (void) bawrite(bp); goto restart; } drain_output(vp); BO_UNLOCK(bo); } /* * Sync all cylinder groups that were dirty at the time this function is * called. Newly dirtied cgs will be inserted before the sentinel. This * is used to flush freedep activity that may be holding up writes to a * indirect block. */ static int sync_cgs(mp, waitfor) struct mount *mp; int waitfor; { struct bmsafemap *bmsafemap; struct bmsafemap *sentinel; struct ufsmount *ump; struct buf *bp; int error; sentinel = malloc(sizeof(*sentinel), M_BMSAFEMAP, M_ZERO | M_WAITOK); sentinel->sm_cg = -1; ump = VFSTOUFS(mp); error = 0; ACQUIRE_LOCK(ump); LIST_INSERT_HEAD(&ump->softdep_dirtycg, sentinel, sm_next); for (bmsafemap = LIST_NEXT(sentinel, sm_next); bmsafemap != NULL; bmsafemap = LIST_NEXT(sentinel, sm_next)) { /* Skip sentinels and cgs with no work to release. */ if (bmsafemap->sm_cg == -1 || (LIST_EMPTY(&bmsafemap->sm_freehd) && LIST_EMPTY(&bmsafemap->sm_freewr))) { LIST_REMOVE(sentinel, sm_next); LIST_INSERT_AFTER(bmsafemap, sentinel, sm_next); continue; } /* * If we don't get the lock and we're waiting try again, if * not move on to the next buf and try to sync it. */ bp = getdirtybuf(bmsafemap->sm_buf, LOCK_PTR(ump), waitfor); if (bp == NULL && waitfor == MNT_WAIT) continue; LIST_REMOVE(sentinel, sm_next); LIST_INSERT_AFTER(bmsafemap, sentinel, sm_next); if (bp == NULL) continue; FREE_LOCK(ump); if (waitfor == MNT_NOWAIT) bawrite(bp); else error = bwrite(bp); ACQUIRE_LOCK(ump); if (error) break; } LIST_REMOVE(sentinel, sm_next); FREE_LOCK(ump); free(sentinel, M_BMSAFEMAP); return (error); } /* * This routine is called when we are trying to synchronously flush a * file. This routine must eliminate any filesystem metadata dependencies * so that the syncing routine can succeed. */ int softdep_sync_metadata(struct vnode *vp) { struct inode *ip; int error; ip = VTOI(vp); KASSERT(MOUNTEDSOFTDEP(vp->v_mount) != 0, ("softdep_sync_metadata called on non-softdep filesystem")); /* * Ensure that any direct block dependencies have been cleared, * truncations are started, and inode references are journaled. */ ACQUIRE_LOCK(VFSTOUFS(vp->v_mount)); /* * Write all journal records to prevent rollbacks on devvp. */ if (vp->v_type == VCHR) softdep_flushjournal(vp->v_mount); error = flush_inodedep_deps(vp, vp->v_mount, ip->i_number); /* * Ensure that all truncates are written so we won't find deps on * indirect blocks. */ process_truncates(vp); FREE_LOCK(VFSTOUFS(vp->v_mount)); return (error); } /* * This routine is called when we are attempting to sync a buf with * dependencies. If waitfor is MNT_NOWAIT it attempts to schedule any * other IO it can but returns EBUSY if the buffer is not yet able to * be written. Dependencies which will not cause rollbacks will always * return 0. */ int softdep_sync_buf(struct vnode *vp, struct buf *bp, int waitfor) { struct indirdep *indirdep; struct pagedep *pagedep; struct allocindir *aip; struct newblk *newblk; struct ufsmount *ump; struct buf *nbp; struct worklist *wk; int i, error; KASSERT(MOUNTEDSOFTDEP(vp->v_mount) != 0, ("softdep_sync_buf called on non-softdep filesystem")); /* * For VCHR we just don't want to force flush any dependencies that * will cause rollbacks. */ if (vp->v_type == VCHR) { if (waitfor == MNT_NOWAIT && softdep_count_dependencies(bp, 0)) return (EBUSY); return (0); } ump = VFSTOUFS(vp->v_mount); ACQUIRE_LOCK(ump); /* * As we hold the buffer locked, none of its dependencies * will disappear. */ error = 0; top: LIST_FOREACH(wk, &bp->b_dep, wk_list) { switch (wk->wk_type) { case D_ALLOCDIRECT: case D_ALLOCINDIR: newblk = WK_NEWBLK(wk); if (newblk->nb_jnewblk != NULL) { if (waitfor == MNT_NOWAIT) { error = EBUSY; goto out_unlock; } jwait(&newblk->nb_jnewblk->jn_list, waitfor); goto top; } if (newblk->nb_state & DEPCOMPLETE || waitfor == MNT_NOWAIT) continue; nbp = newblk->nb_bmsafemap->sm_buf; nbp = getdirtybuf(nbp, LOCK_PTR(ump), waitfor); if (nbp == NULL) goto top; FREE_LOCK(ump); if ((error = bwrite(nbp)) != 0) goto out; ACQUIRE_LOCK(ump); continue; case D_INDIRDEP: indirdep = WK_INDIRDEP(wk); if (waitfor == MNT_NOWAIT) { if (!TAILQ_EMPTY(&indirdep->ir_trunc) || !LIST_EMPTY(&indirdep->ir_deplisthd)) { error = EBUSY; goto out_unlock; } } if (!TAILQ_EMPTY(&indirdep->ir_trunc)) panic("softdep_sync_buf: truncation pending."); restart: LIST_FOREACH(aip, &indirdep->ir_deplisthd, ai_next) { newblk = (struct newblk *)aip; if (newblk->nb_jnewblk != NULL) { jwait(&newblk->nb_jnewblk->jn_list, waitfor); goto restart; } if (newblk->nb_state & DEPCOMPLETE) continue; nbp = newblk->nb_bmsafemap->sm_buf; nbp = getdirtybuf(nbp, LOCK_PTR(ump), waitfor); if (nbp == NULL) goto restart; FREE_LOCK(ump); if ((error = bwrite(nbp)) != 0) goto out; ACQUIRE_LOCK(ump); goto restart; } continue; case D_PAGEDEP: /* * Only flush directory entries in synchronous passes. */ if (waitfor != MNT_WAIT) { error = EBUSY; goto out_unlock; } /* * While syncing snapshots, we must allow recursive * lookups. */ BUF_AREC(bp); /* * We are trying to sync a directory that may * have dependencies on both its own metadata * and/or dependencies on the inodes of any * recently allocated files. We walk its diradd * lists pushing out the associated inode. */ pagedep = WK_PAGEDEP(wk); for (i = 0; i < DAHASHSZ; i++) { if (LIST_FIRST(&pagedep->pd_diraddhd[i]) == 0) continue; error = flush_pagedep_deps(vp, wk->wk_mp, &pagedep->pd_diraddhd[i], bp); if (error != 0) { if (error != ERELOOKUP) BUF_NOREC(bp); goto out_unlock; } } BUF_NOREC(bp); continue; case D_FREEWORK: case D_FREEDEP: case D_JSEGDEP: case D_JNEWBLK: continue; default: panic("softdep_sync_buf: Unknown type %s", TYPENAME(wk->wk_type)); /* NOTREACHED */ } } out_unlock: FREE_LOCK(ump); out: return (error); } /* * Flush the dependencies associated with an inodedep. */ static int flush_inodedep_deps(vp, mp, ino) struct vnode *vp; struct mount *mp; ino_t ino; { struct inodedep *inodedep; struct inoref *inoref; struct ufsmount *ump; int error, waitfor; /* * This work is done in two passes. The first pass grabs most * of the buffers and begins asynchronously writing them. The * only way to wait for these asynchronous writes is to sleep * on the filesystem vnode which may stay busy for a long time * if the filesystem is active. So, instead, we make a second * pass over the dependencies blocking on each write. In the * usual case we will be blocking against a write that we * initiated, so when it is done the dependency will have been * resolved. Thus the second pass is expected to end quickly. * We give a brief window at the top of the loop to allow * any pending I/O to complete. */ ump = VFSTOUFS(mp); LOCK_OWNED(ump); for (error = 0, waitfor = MNT_NOWAIT; ; ) { if (error) return (error); FREE_LOCK(ump); ACQUIRE_LOCK(ump); restart: if (inodedep_lookup(mp, ino, 0, &inodedep) == 0) return (0); TAILQ_FOREACH(inoref, &inodedep->id_inoreflst, if_deps) { if ((inoref->if_state & (DEPCOMPLETE | GOINGAWAY)) == DEPCOMPLETE) { jwait(&inoref->if_list, MNT_WAIT); goto restart; } } if (flush_deplist(&inodedep->id_inoupdt, waitfor, &error) || flush_deplist(&inodedep->id_newinoupdt, waitfor, &error) || flush_deplist(&inodedep->id_extupdt, waitfor, &error) || flush_deplist(&inodedep->id_newextupdt, waitfor, &error)) continue; /* * If pass2, we are done, otherwise do pass 2. */ if (waitfor == MNT_WAIT) break; waitfor = MNT_WAIT; } /* * Try freeing inodedep in case all dependencies have been removed. */ if (inodedep_lookup(mp, ino, 0, &inodedep) != 0) (void) free_inodedep(inodedep); return (0); } /* * Flush an inode dependency list. */ static int flush_deplist(listhead, waitfor, errorp) struct allocdirectlst *listhead; int waitfor; int *errorp; { struct allocdirect *adp; struct newblk *newblk; struct ufsmount *ump; struct buf *bp; if ((adp = TAILQ_FIRST(listhead)) == NULL) return (0); ump = VFSTOUFS(adp->ad_list.wk_mp); LOCK_OWNED(ump); TAILQ_FOREACH(adp, listhead, ad_next) { newblk = (struct newblk *)adp; if (newblk->nb_jnewblk != NULL) { jwait(&newblk->nb_jnewblk->jn_list, MNT_WAIT); return (1); } if (newblk->nb_state & DEPCOMPLETE) continue; bp = newblk->nb_bmsafemap->sm_buf; bp = getdirtybuf(bp, LOCK_PTR(ump), waitfor); if (bp == NULL) { if (waitfor == MNT_NOWAIT) continue; return (1); } FREE_LOCK(ump); if (waitfor == MNT_NOWAIT) bawrite(bp); else *errorp = bwrite(bp); ACQUIRE_LOCK(ump); return (1); } return (0); } /* * Flush dependencies associated with an allocdirect block. */ static int flush_newblk_dep(vp, mp, lbn) struct vnode *vp; struct mount *mp; ufs_lbn_t lbn; { struct newblk *newblk; struct ufsmount *ump; struct bufobj *bo; struct inode *ip; struct buf *bp; ufs2_daddr_t blkno; int error; error = 0; bo = &vp->v_bufobj; ip = VTOI(vp); blkno = DIP(ip, i_db[lbn]); if (blkno == 0) panic("flush_newblk_dep: Missing block"); ump = VFSTOUFS(mp); ACQUIRE_LOCK(ump); /* * Loop until all dependencies related to this block are satisfied. * We must be careful to restart after each sleep in case a write * completes some part of this process for us. */ for (;;) { if (newblk_lookup(mp, blkno, 0, &newblk) == 0) { FREE_LOCK(ump); break; } if (newblk->nb_list.wk_type != D_ALLOCDIRECT) panic("flush_newblk_dep: Bad newblk %p", newblk); /* * Flush the journal. */ if (newblk->nb_jnewblk != NULL) { jwait(&newblk->nb_jnewblk->jn_list, MNT_WAIT); continue; } /* * Write the bitmap dependency. */ if ((newblk->nb_state & DEPCOMPLETE) == 0) { bp = newblk->nb_bmsafemap->sm_buf; bp = getdirtybuf(bp, LOCK_PTR(ump), MNT_WAIT); if (bp == NULL) continue; FREE_LOCK(ump); error = bwrite(bp); if (error) break; ACQUIRE_LOCK(ump); continue; } /* * Write the buffer. */ FREE_LOCK(ump); BO_LOCK(bo); bp = gbincore(bo, lbn); if (bp != NULL) { error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo)); if (error == ENOLCK) { ACQUIRE_LOCK(ump); error = 0; continue; /* Slept, retry */ } if (error != 0) break; /* Failed */ if (bp->b_flags & B_DELWRI) { bremfree(bp); error = bwrite(bp); if (error) break; } else BUF_UNLOCK(bp); } else BO_UNLOCK(bo); /* * We have to wait for the direct pointers to * point at the newdirblk before the dependency * will go away. */ error = ffs_update(vp, 1); if (error) break; ACQUIRE_LOCK(ump); } return (error); } /* * Eliminate a pagedep dependency by flushing out all its diradd dependencies. */ static int flush_pagedep_deps(pvp, mp, diraddhdp, locked_bp) struct vnode *pvp; struct mount *mp; struct diraddhd *diraddhdp; struct buf *locked_bp; { struct inodedep *inodedep; struct inoref *inoref; struct ufsmount *ump; struct diradd *dap; struct vnode *vp; int error = 0; struct buf *bp; ino_t inum; struct diraddhd unfinished; LIST_INIT(&unfinished); ump = VFSTOUFS(mp); LOCK_OWNED(ump); restart: while ((dap = LIST_FIRST(diraddhdp)) != NULL) { /* * Flush ourselves if this directory entry * has a MKDIR_PARENT dependency. */ if (dap->da_state & MKDIR_PARENT) { FREE_LOCK(ump); if ((error = ffs_update(pvp, 1)) != 0) break; ACQUIRE_LOCK(ump); /* * If that cleared dependencies, go on to next. */ if (dap != LIST_FIRST(diraddhdp)) continue; /* * All MKDIR_PARENT dependencies and all the * NEWBLOCK pagedeps that are contained in direct * blocks were resolved by doing above ffs_update. * Pagedeps contained in indirect blocks may * require a complete sync'ing of the directory. * We are in the midst of doing a complete sync, * so if they are not resolved in this pass we * defer them for now as they will be sync'ed by * our caller shortly. */ LIST_REMOVE(dap, da_pdlist); LIST_INSERT_HEAD(&unfinished, dap, da_pdlist); continue; } /* * A newly allocated directory must have its "." and * ".." entries written out before its name can be * committed in its parent. */ inum = dap->da_newinum; if (inodedep_lookup(UFSTOVFS(ump), inum, 0, &inodedep) == 0) panic("flush_pagedep_deps: lost inode1"); /* * Wait for any pending journal adds to complete so we don't * cause rollbacks while syncing. */ TAILQ_FOREACH(inoref, &inodedep->id_inoreflst, if_deps) { if ((inoref->if_state & (DEPCOMPLETE | GOINGAWAY)) == DEPCOMPLETE) { jwait(&inoref->if_list, MNT_WAIT); goto restart; } } if (dap->da_state & MKDIR_BODY) { FREE_LOCK(ump); error = get_parent_vp(pvp, mp, inum, locked_bp, diraddhdp, &unfinished, &vp); if (error != 0) break; error = flush_newblk_dep(vp, mp, 0); /* * If we still have the dependency we might need to * update the vnode to sync the new link count to * disk. */ if (error == 0 && dap == LIST_FIRST(diraddhdp)) error = ffs_update(vp, 1); vput(vp); if (error != 0) break; ACQUIRE_LOCK(ump); /* * If that cleared dependencies, go on to next. */ if (dap != LIST_FIRST(diraddhdp)) continue; if (dap->da_state & MKDIR_BODY) { inodedep_lookup(UFSTOVFS(ump), inum, 0, &inodedep); panic("flush_pagedep_deps: MKDIR_BODY " "inodedep %p dap %p vp %p", inodedep, dap, vp); } } /* * Flush the inode on which the directory entry depends. * Having accounted for MKDIR_PARENT and MKDIR_BODY above, * the only remaining dependency is that the updated inode * count must get pushed to disk. The inode has already * been pushed into its inode buffer (via VOP_UPDATE) at * the time of the reference count change. So we need only * locate that buffer, ensure that there will be no rollback * caused by a bitmap dependency, then write the inode buffer. */ retry: if (inodedep_lookup(UFSTOVFS(ump), inum, 0, &inodedep) == 0) panic("flush_pagedep_deps: lost inode"); /* * If the inode still has bitmap dependencies, * push them to disk. */ if ((inodedep->id_state & (DEPCOMPLETE | GOINGAWAY)) == 0) { bp = inodedep->id_bmsafemap->sm_buf; bp = getdirtybuf(bp, LOCK_PTR(ump), MNT_WAIT); if (bp == NULL) goto retry; FREE_LOCK(ump); if ((error = bwrite(bp)) != 0) break; ACQUIRE_LOCK(ump); if (dap != LIST_FIRST(diraddhdp)) continue; } /* * If the inode is still sitting in a buffer waiting * to be written or waiting for the link count to be * adjusted update it here to flush it to disk. */ if (dap == LIST_FIRST(diraddhdp)) { FREE_LOCK(ump); error = get_parent_vp(pvp, mp, inum, locked_bp, diraddhdp, &unfinished, &vp); if (error != 0) break; error = ffs_update(vp, 1); vput(vp); if (error) break; ACQUIRE_LOCK(ump); } /* * If we have failed to get rid of all the dependencies * then something is seriously wrong. */ if (dap == LIST_FIRST(diraddhdp)) { inodedep_lookup(UFSTOVFS(ump), inum, 0, &inodedep); panic("flush_pagedep_deps: failed to flush " "inodedep %p ino %ju dap %p", inodedep, (uintmax_t)inum, dap); } } if (error) ACQUIRE_LOCK(ump); while ((dap = LIST_FIRST(&unfinished)) != NULL) { LIST_REMOVE(dap, da_pdlist); LIST_INSERT_HEAD(diraddhdp, dap, da_pdlist); } return (error); } /* * A large burst of file addition or deletion activity can drive the * memory load excessively high. First attempt to slow things down * using the techniques below. If that fails, this routine requests * the offending operations to fall back to running synchronously * until the memory load returns to a reasonable level. */ int softdep_slowdown(vp) struct vnode *vp; { struct ufsmount *ump; int jlow; int max_softdeps_hard; KASSERT(MOUNTEDSOFTDEP(vp->v_mount) != 0, ("softdep_slowdown called on non-softdep filesystem")); ump = VFSTOUFS(vp->v_mount); ACQUIRE_LOCK(ump); jlow = 0; /* * Check for journal space if needed. */ if (DOINGSUJ(vp)) { if (journal_space(ump, 0) == 0) jlow = 1; } /* * If the system is under its limits and our filesystem is * not responsible for more than our share of the usage and * we are not low on journal space, then no need to slow down. */ max_softdeps_hard = max_softdeps * 11 / 10; if (dep_current[D_DIRREM] < max_softdeps_hard / 2 && dep_current[D_INODEDEP] < max_softdeps_hard && dep_current[D_INDIRDEP] < max_softdeps_hard / 1000 && dep_current[D_FREEBLKS] < max_softdeps_hard && jlow == 0 && ump->softdep_curdeps[D_DIRREM] < (max_softdeps_hard / 2) / stat_flush_threads && ump->softdep_curdeps[D_INODEDEP] < max_softdeps_hard / stat_flush_threads && ump->softdep_curdeps[D_INDIRDEP] < (max_softdeps_hard / 1000) / stat_flush_threads && ump->softdep_curdeps[D_FREEBLKS] < max_softdeps_hard / stat_flush_threads) { FREE_LOCK(ump); return (0); } /* * If the journal is low or our filesystem is over its limit * then speedup the cleanup. */ if (ump->softdep_curdeps[D_INDIRDEP] < (max_softdeps_hard / 1000) / stat_flush_threads || jlow) softdep_speedup(ump); stat_sync_limit_hit += 1; FREE_LOCK(ump); /* * We only slow down the rate at which new dependencies are * generated if we are not using journaling. With journaling, * the cleanup should always be sufficient to keep things * under control. */ if (DOINGSUJ(vp)) return (0); return (1); } static int softdep_request_cleanup_filter(struct vnode *vp, void *arg __unused) { return ((vp->v_iflag & VI_OWEINACT) != 0 && vp->v_usecount == 0 && ((vp->v_vflag & VV_NOSYNC) != 0 || VTOI(vp)->i_effnlink == 0)); } static void softdep_request_cleanup_inactivate(struct mount *mp) { struct vnode *vp, *mvp; int error; MNT_VNODE_FOREACH_LAZY(vp, mp, mvp, softdep_request_cleanup_filter, NULL) { vholdl(vp); vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY); VI_LOCK(vp); if (vp->v_data != NULL && vp->v_usecount == 0) { while ((vp->v_iflag & VI_OWEINACT) != 0) { error = vinactive(vp); if (error != 0 && error != ERELOOKUP) break; } atomic_add_int(&stat_delayed_inact, 1); } VOP_UNLOCK(vp); vdropl(vp); } } /* * Called by the allocation routines when they are about to fail * in the hope that we can free up the requested resource (inodes * or disk space). * * First check to see if the work list has anything on it. If it has, * clean up entries until we successfully free the requested resource. * Because this process holds inodes locked, we cannot handle any remove * requests that might block on a locked inode as that could lead to * deadlock. If the worklist yields none of the requested resource, * start syncing out vnodes to free up the needed space. */ int softdep_request_cleanup(fs, vp, cred, resource) struct fs *fs; struct vnode *vp; struct ucred *cred; int resource; { struct ufsmount *ump; struct mount *mp; long starttime; ufs2_daddr_t needed; int error, failed_vnode; /* * If we are being called because of a process doing a * copy-on-write, then it is not safe to process any * worklist items as we will recurse into the copyonwrite * routine. This will result in an incoherent snapshot. * If the vnode that we hold is a snapshot, we must avoid * handling other resources that could cause deadlock. */ if ((curthread->td_pflags & TDP_COWINPROGRESS) || IS_SNAPSHOT(VTOI(vp))) return (0); if (resource == FLUSH_BLOCKS_WAIT) stat_cleanup_blkrequests += 1; else stat_cleanup_inorequests += 1; mp = vp->v_mount; ump = VFSTOUFS(mp); mtx_assert(UFS_MTX(ump), MA_OWNED); UFS_UNLOCK(ump); error = ffs_update(vp, 1); if (error != 0 || MOUNTEDSOFTDEP(mp) == 0) { UFS_LOCK(ump); return (0); } /* * If we are in need of resources, start by cleaning up * any block removals associated with our inode. */ ACQUIRE_LOCK(ump); process_removes(vp); process_truncates(vp); FREE_LOCK(ump); /* * Now clean up at least as many resources as we will need. * * When requested to clean up inodes, the number that are needed * is set by the number of simultaneous writers (mnt_writeopcount) * plus a bit of slop (2) in case some more writers show up while * we are cleaning. * * When requested to free up space, the amount of space that * we need is enough blocks to allocate a full-sized segment * (fs_contigsumsize). The number of such segments that will * be needed is set by the number of simultaneous writers * (mnt_writeopcount) plus a bit of slop (2) in case some more * writers show up while we are cleaning. * * Additionally, if we are unpriviledged and allocating space, * we need to ensure that we clean up enough blocks to get the * needed number of blocks over the threshold of the minimum * number of blocks required to be kept free by the filesystem * (fs_minfree). */ if (resource == FLUSH_INODES_WAIT) { needed = vfs_mount_fetch_counter(vp->v_mount, MNT_COUNT_WRITEOPCOUNT) + 2; } else if (resource == FLUSH_BLOCKS_WAIT) { needed = (vfs_mount_fetch_counter(vp->v_mount, MNT_COUNT_WRITEOPCOUNT) + 2) * fs->fs_contigsumsize; if (priv_check_cred(cred, PRIV_VFS_BLOCKRESERVE)) needed += fragstoblks(fs, roundup((fs->fs_dsize * fs->fs_minfree / 100) - fs->fs_cstotal.cs_nffree, fs->fs_frag)); } else { printf("softdep_request_cleanup: Unknown resource type %d\n", resource); UFS_LOCK(ump); return (0); } starttime = time_second; retry: if (resource == FLUSH_BLOCKS_WAIT && fs->fs_cstotal.cs_nbfree <= needed) softdep_send_speedup(ump, needed * fs->fs_bsize, BIO_SPEEDUP_TRIM); if ((resource == FLUSH_BLOCKS_WAIT && ump->softdep_on_worklist > 0 && fs->fs_cstotal.cs_nbfree <= needed) || (resource == FLUSH_INODES_WAIT && fs->fs_pendinginodes > 0 && fs->fs_cstotal.cs_nifree <= needed)) { ACQUIRE_LOCK(ump); if (ump->softdep_on_worklist > 0 && process_worklist_item(UFSTOVFS(ump), ump->softdep_on_worklist, LK_NOWAIT) != 0) stat_worklist_push += 1; FREE_LOCK(ump); } /* * Check that there are vnodes pending inactivation. As they * have been unlinked, inactivating them will free up their * inodes. */ ACQUIRE_LOCK(ump); if (resource == FLUSH_INODES_WAIT && fs->fs_cstotal.cs_nifree <= needed && fs->fs_pendinginodes <= needed) { if ((ump->um_softdep->sd_flags & FLUSH_DI_ACTIVE) == 0) { ump->um_softdep->sd_flags |= FLUSH_DI_ACTIVE; FREE_LOCK(ump); softdep_request_cleanup_inactivate(mp); ACQUIRE_LOCK(ump); ump->um_softdep->sd_flags &= ~FLUSH_DI_ACTIVE; wakeup(&ump->um_softdep->sd_flags); } else { while ((ump->um_softdep->sd_flags & FLUSH_DI_ACTIVE) != 0) { msleep(&ump->um_softdep->sd_flags, LOCK_PTR(ump), PVM, "ffsvina", hz); } } } FREE_LOCK(ump); /* * If we still need resources and there are no more worklist * entries to process to obtain them, we have to start flushing * the dirty vnodes to force the release of additional requests * to the worklist that we can then process to reap addition * resources. We walk the vnodes associated with the mount point * until we get the needed worklist requests that we can reap. * * If there are several threads all needing to clean the same * mount point, only one is allowed to walk the mount list. * When several threads all try to walk the same mount list, * they end up competing with each other and often end up in * livelock. This approach ensures that forward progress is * made at the cost of occational ENOSPC errors being returned * that might otherwise have been avoided. */ error = 1; if ((resource == FLUSH_BLOCKS_WAIT && fs->fs_cstotal.cs_nbfree <= needed) || (resource == FLUSH_INODES_WAIT && fs->fs_pendinginodes > 0 && fs->fs_cstotal.cs_nifree <= needed)) { ACQUIRE_LOCK(ump); if ((ump->um_softdep->sd_flags & FLUSH_RC_ACTIVE) == 0) { ump->um_softdep->sd_flags |= FLUSH_RC_ACTIVE; FREE_LOCK(ump); failed_vnode = softdep_request_cleanup_flush(mp, ump); ACQUIRE_LOCK(ump); ump->um_softdep->sd_flags &= ~FLUSH_RC_ACTIVE; wakeup(&ump->um_softdep->sd_flags); FREE_LOCK(ump); if (ump->softdep_on_worklist > 0) { stat_cleanup_retries += 1; if (!failed_vnode) goto retry; } } else { while ((ump->um_softdep->sd_flags & FLUSH_RC_ACTIVE) != 0) { msleep(&ump->um_softdep->sd_flags, LOCK_PTR(ump), PVM, "ffsrca", hz); } FREE_LOCK(ump); error = 0; } stat_cleanup_failures += 1; } if (time_second - starttime > stat_cleanup_high_delay) stat_cleanup_high_delay = time_second - starttime; UFS_LOCK(ump); return (error); } /* * Scan the vnodes for the specified mount point flushing out any * vnodes that can be locked without waiting. Finally, try to flush * the device associated with the mount point if it can be locked * without waiting. * * We return 0 if we were able to lock every vnode in our scan. * If we had to skip one or more vnodes, we return 1. */ static int softdep_request_cleanup_flush(mp, ump) struct mount *mp; struct ufsmount *ump; { struct thread *td; struct vnode *lvp, *mvp; int failed_vnode; failed_vnode = 0; td = curthread; MNT_VNODE_FOREACH_ALL(lvp, mp, mvp) { if (TAILQ_FIRST(&lvp->v_bufobj.bo_dirty.bv_hd) == 0) { VI_UNLOCK(lvp); continue; } if (vget(lvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_NOWAIT) != 0) { failed_vnode = 1; continue; } if (lvp->v_vflag & VV_NOSYNC) { /* unlinked */ vput(lvp); continue; } (void) ffs_syncvnode(lvp, MNT_NOWAIT, 0); vput(lvp); } lvp = ump->um_devvp; if (vn_lock(lvp, LK_EXCLUSIVE | LK_NOWAIT) == 0) { VOP_FSYNC(lvp, MNT_NOWAIT, td); VOP_UNLOCK(lvp); } return (failed_vnode); } static bool softdep_excess_items(struct ufsmount *ump, int item) { KASSERT(item >= 0 && item < D_LAST, ("item %d", item)); return (dep_current[item] > max_softdeps && ump->softdep_curdeps[item] > max_softdeps / stat_flush_threads); } static void schedule_cleanup(struct mount *mp) { struct ufsmount *ump; struct thread *td; ump = VFSTOUFS(mp); LOCK_OWNED(ump); FREE_LOCK(ump); td = curthread; if ((td->td_pflags & TDP_KTHREAD) != 0 && (td->td_proc->p_flag2 & P2_AST_SU) == 0) { /* * No ast is delivered to kernel threads, so nobody * would deref the mp. Some kernel threads * explicitely check for AST, e.g. NFS daemon does * this in the serving loop. */ return; } if (td->td_su != NULL) vfs_rel(td->td_su); vfs_ref(mp); td->td_su = mp; thread_lock(td); td->td_flags |= TDF_ASTPENDING; thread_unlock(td); } static void softdep_ast_cleanup_proc(struct thread *td) { struct mount *mp; struct ufsmount *ump; int error; bool req; while ((mp = td->td_su) != NULL) { td->td_su = NULL; error = vfs_busy(mp, MBF_NOWAIT); vfs_rel(mp); if (error != 0) return; if (ffs_own_mount(mp) && MOUNTEDSOFTDEP(mp)) { ump = VFSTOUFS(mp); for (;;) { req = false; ACQUIRE_LOCK(ump); if (softdep_excess_items(ump, D_INODEDEP)) { req = true; request_cleanup(mp, FLUSH_INODES); } if (softdep_excess_items(ump, D_DIRREM)) { req = true; request_cleanup(mp, FLUSH_BLOCKS); } FREE_LOCK(ump); if (softdep_excess_items(ump, D_NEWBLK) || softdep_excess_items(ump, D_ALLOCDIRECT) || softdep_excess_items(ump, D_ALLOCINDIR)) { error = vn_start_write(NULL, &mp, V_WAIT); if (error == 0) { req = true; VFS_SYNC(mp, MNT_WAIT); vn_finished_write(mp); } } if ((td->td_pflags & TDP_KTHREAD) != 0 || !req) break; } } vfs_unbusy(mp); } if ((mp = td->td_su) != NULL) { td->td_su = NULL; vfs_rel(mp); } } /* * If memory utilization has gotten too high, deliberately slow things * down and speed up the I/O processing. */ static int request_cleanup(mp, resource) struct mount *mp; int resource; { struct thread *td = curthread; struct ufsmount *ump; ump = VFSTOUFS(mp); LOCK_OWNED(ump); /* * We never hold up the filesystem syncer or buf daemon. */ if (td->td_pflags & (TDP_SOFTDEP|TDP_NORUNNINGBUF)) return (0); /* * First check to see if the work list has gotten backlogged. * If it has, co-opt this process to help clean up two entries. * Because this process may hold inodes locked, we cannot * handle any remove requests that might block on a locked * inode as that could lead to deadlock. We set TDP_SOFTDEP * to avoid recursively processing the worklist. */ if (ump->softdep_on_worklist > max_softdeps / 10) { td->td_pflags |= TDP_SOFTDEP; process_worklist_item(mp, 2, LK_NOWAIT); td->td_pflags &= ~TDP_SOFTDEP; stat_worklist_push += 2; return(1); } /* * Next, we attempt to speed up the syncer process. If that * is successful, then we allow the process to continue. */ if (softdep_speedup(ump) && resource != FLUSH_BLOCKS_WAIT && resource != FLUSH_INODES_WAIT) return(0); /* * If we are resource constrained on inode dependencies, try * flushing some dirty inodes. Otherwise, we are constrained * by file deletions, so try accelerating flushes of directories * with removal dependencies. We would like to do the cleanup * here, but we probably hold an inode locked at this point and * that might deadlock against one that we try to clean. So, * the best that we can do is request the syncer daemon to do * the cleanup for us. */ switch (resource) { case FLUSH_INODES: case FLUSH_INODES_WAIT: ACQUIRE_GBLLOCK(&lk); stat_ino_limit_push += 1; req_clear_inodedeps += 1; FREE_GBLLOCK(&lk); stat_countp = &stat_ino_limit_hit; break; case FLUSH_BLOCKS: case FLUSH_BLOCKS_WAIT: ACQUIRE_GBLLOCK(&lk); stat_blk_limit_push += 1; req_clear_remove += 1; FREE_GBLLOCK(&lk); stat_countp = &stat_blk_limit_hit; break; default: panic("request_cleanup: unknown type"); } /* * Hopefully the syncer daemon will catch up and awaken us. * We wait at most tickdelay before proceeding in any case. */ ACQUIRE_GBLLOCK(&lk); FREE_LOCK(ump); proc_waiting += 1; if (callout_pending(&softdep_callout) == FALSE) callout_reset(&softdep_callout, tickdelay > 2 ? tickdelay : 2, pause_timer, 0); if ((td->td_pflags & TDP_KTHREAD) == 0) msleep((caddr_t)&proc_waiting, &lk, PPAUSE, "softupdate", 0); proc_waiting -= 1; FREE_GBLLOCK(&lk); ACQUIRE_LOCK(ump); return (1); } /* * Awaken processes pausing in request_cleanup and clear proc_waiting * to indicate that there is no longer a timer running. Pause_timer * will be called with the global softdep mutex (&lk) locked. */ static void pause_timer(arg) void *arg; { GBLLOCK_OWNED(&lk); /* * The callout_ API has acquired mtx and will hold it around this * function call. */ *stat_countp += proc_waiting; wakeup(&proc_waiting); } /* * If requested, try removing inode or removal dependencies. */ static void check_clear_deps(mp) struct mount *mp; { struct ufsmount *ump; bool suj_susp; /* * Tell the lower layers that any TRIM or WRITE transactions that have * been delayed for performance reasons should proceed to help alleviate * the shortage faster. The race between checking req_* and the softdep * mutex (lk) is fine since this is an advisory operation that at most * causes deferred work to be done sooner. */ ump = VFSTOUFS(mp); suj_susp = ump->um_softdep->sd_jblocks != NULL && ump->softdep_jblocks->jb_suspended; if (req_clear_remove || req_clear_inodedeps || suj_susp) { FREE_LOCK(ump); softdep_send_speedup(ump, 0, BIO_SPEEDUP_TRIM | BIO_SPEEDUP_WRITE); ACQUIRE_LOCK(ump); } /* * If we are suspended, it may be because of our using * too many inodedeps, so help clear them out. */ if (suj_susp) clear_inodedeps(mp); /* * General requests for cleanup of backed up dependencies */ ACQUIRE_GBLLOCK(&lk); if (req_clear_inodedeps) { req_clear_inodedeps -= 1; FREE_GBLLOCK(&lk); clear_inodedeps(mp); ACQUIRE_GBLLOCK(&lk); wakeup(&proc_waiting); } if (req_clear_remove) { req_clear_remove -= 1; FREE_GBLLOCK(&lk); clear_remove(mp); ACQUIRE_GBLLOCK(&lk); wakeup(&proc_waiting); } FREE_GBLLOCK(&lk); } /* * Flush out a directory with at least one removal dependency in an effort to * reduce the number of dirrem, freefile, and freeblks dependency structures. */ static void clear_remove(mp) struct mount *mp; { struct pagedep_hashhead *pagedephd; struct pagedep *pagedep; struct ufsmount *ump; struct vnode *vp; struct bufobj *bo; int error, cnt; ino_t ino; ump = VFSTOUFS(mp); LOCK_OWNED(ump); for (cnt = 0; cnt <= ump->pagedep_hash_size; cnt++) { pagedephd = &ump->pagedep_hashtbl[ump->pagedep_nextclean++]; if (ump->pagedep_nextclean > ump->pagedep_hash_size) ump->pagedep_nextclean = 0; LIST_FOREACH(pagedep, pagedephd, pd_hash) { if (LIST_EMPTY(&pagedep->pd_dirremhd)) continue; ino = pagedep->pd_ino; if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) continue; FREE_LOCK(ump); /* * Let unmount clear deps */ error = vfs_busy(mp, MBF_NOWAIT); if (error != 0) goto finish_write; error = ffs_vgetf(mp, ino, LK_EXCLUSIVE, &vp, FFSV_FORCEINSMQ | FFSV_FORCEINODEDEP); vfs_unbusy(mp); if (error != 0) { softdep_error("clear_remove: vget", error); goto finish_write; } MPASS(VTOI(vp)->i_mode != 0); if ((error = ffs_syncvnode(vp, MNT_NOWAIT, 0))) softdep_error("clear_remove: fsync", error); bo = &vp->v_bufobj; BO_LOCK(bo); drain_output(vp); BO_UNLOCK(bo); vput(vp); finish_write: vn_finished_write(mp); ACQUIRE_LOCK(ump); return; } } } /* * Clear out a block of dirty inodes in an effort to reduce * the number of inodedep dependency structures. */ static void clear_inodedeps(mp) struct mount *mp; { struct inodedep_hashhead *inodedephd; struct inodedep *inodedep; struct ufsmount *ump; struct vnode *vp; struct fs *fs; int error, cnt; ino_t firstino, lastino, ino; ump = VFSTOUFS(mp); fs = ump->um_fs; LOCK_OWNED(ump); /* * Pick a random inode dependency to be cleared. * We will then gather up all the inodes in its block * that have dependencies and flush them out. */ for (cnt = 0; cnt <= ump->inodedep_hash_size; cnt++) { inodedephd = &ump->inodedep_hashtbl[ump->inodedep_nextclean++]; if (ump->inodedep_nextclean > ump->inodedep_hash_size) ump->inodedep_nextclean = 0; if ((inodedep = LIST_FIRST(inodedephd)) != NULL) break; } if (inodedep == NULL) return; /* * Find the last inode in the block with dependencies. */ firstino = rounddown2(inodedep->id_ino, INOPB(fs)); for (lastino = firstino + INOPB(fs) - 1; lastino > firstino; lastino--) if (inodedep_lookup(mp, lastino, 0, &inodedep) != 0) break; /* * Asynchronously push all but the last inode with dependencies. * Synchronously push the last inode with dependencies to ensure * that the inode block gets written to free up the inodedeps. */ for (ino = firstino; ino <= lastino; ino++) { if (inodedep_lookup(mp, ino, 0, &inodedep) == 0) continue; if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) continue; FREE_LOCK(ump); error = vfs_busy(mp, MBF_NOWAIT); /* Let unmount clear deps */ if (error != 0) { vn_finished_write(mp); ACQUIRE_LOCK(ump); return; } if ((error = ffs_vgetf(mp, ino, LK_EXCLUSIVE, &vp, FFSV_FORCEINSMQ | FFSV_FORCEINODEDEP)) != 0) { softdep_error("clear_inodedeps: vget", error); vfs_unbusy(mp); vn_finished_write(mp); ACQUIRE_LOCK(ump); return; } vfs_unbusy(mp); if (VTOI(vp)->i_mode == 0) { vgone(vp); } else if (ino == lastino) { do { error = ffs_syncvnode(vp, MNT_WAIT, 0); } while (error == ERELOOKUP); if (error != 0) softdep_error("clear_inodedeps: fsync1", error); } else { if ((error = ffs_syncvnode(vp, MNT_NOWAIT, 0))) softdep_error("clear_inodedeps: fsync2", error); BO_LOCK(&vp->v_bufobj); drain_output(vp); BO_UNLOCK(&vp->v_bufobj); } vput(vp); vn_finished_write(mp); ACQUIRE_LOCK(ump); } } void softdep_buf_append(bp, wkhd) struct buf *bp; struct workhead *wkhd; { struct worklist *wk; struct ufsmount *ump; if ((wk = LIST_FIRST(wkhd)) == NULL) return; KASSERT(MOUNTEDSOFTDEP(wk->wk_mp) != 0, ("softdep_buf_append called on non-softdep filesystem")); ump = VFSTOUFS(wk->wk_mp); ACQUIRE_LOCK(ump); while ((wk = LIST_FIRST(wkhd)) != NULL) { WORKLIST_REMOVE(wk); WORKLIST_INSERT(&bp->b_dep, wk); } FREE_LOCK(ump); } void softdep_inode_append(ip, cred, wkhd) struct inode *ip; struct ucred *cred; struct workhead *wkhd; { struct buf *bp; struct fs *fs; struct ufsmount *ump; int error; ump = ITOUMP(ip); KASSERT(MOUNTEDSOFTDEP(UFSTOVFS(ump)) != 0, ("softdep_inode_append called on non-softdep filesystem")); fs = ump->um_fs; error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)), (int)fs->fs_bsize, cred, &bp); if (error) { bqrelse(bp); softdep_freework(wkhd); return; } softdep_buf_append(bp, wkhd); bqrelse(bp); } void softdep_freework(wkhd) struct workhead *wkhd; { struct worklist *wk; struct ufsmount *ump; if ((wk = LIST_FIRST(wkhd)) == NULL) return; KASSERT(MOUNTEDSOFTDEP(wk->wk_mp) != 0, ("softdep_freework called on non-softdep filesystem")); ump = VFSTOUFS(wk->wk_mp); ACQUIRE_LOCK(ump); handle_jwork(wkhd); FREE_LOCK(ump); } static struct ufsmount * softdep_bp_to_mp(bp) struct buf *bp; { struct mount *mp; struct vnode *vp; if (LIST_EMPTY(&bp->b_dep)) return (NULL); vp = bp->b_vp; KASSERT(vp != NULL, ("%s, buffer with dependencies lacks vnode", __func__)); /* * The ump mount point is stable after we get a correct * pointer, since bp is locked and this prevents unmount from * proceeding. But to get to it, we cannot dereference bp->b_dep * head wk_mp, because we do not yet own SU ump lock and * workitem might be freed while dereferenced. */ retry: switch (vp->v_type) { case VCHR: VI_LOCK(vp); mp = vp->v_type == VCHR ? vp->v_rdev->si_mountpt : NULL; VI_UNLOCK(vp); if (mp == NULL) goto retry; break; case VREG: case VDIR: case VLNK: case VFIFO: case VSOCK: mp = vp->v_mount; break; case VBLK: vn_printf(vp, "softdep_bp_to_mp: unexpected block device\n"); /* FALLTHROUGH */ case VNON: case VBAD: case VMARKER: mp = NULL; break; default: vn_printf(vp, "unknown vnode type"); mp = NULL; break; } return (VFSTOUFS(mp)); } /* * Function to determine if the buffer has outstanding dependencies * that will cause a roll-back if the buffer is written. If wantcount * is set, return number of dependencies, otherwise just yes or no. */ static int softdep_count_dependencies(bp, wantcount) struct buf *bp; int wantcount; { struct worklist *wk; struct ufsmount *ump; struct bmsafemap *bmsafemap; struct freework *freework; struct inodedep *inodedep; struct indirdep *indirdep; struct freeblks *freeblks; struct allocindir *aip; struct pagedep *pagedep; struct dirrem *dirrem; struct newblk *newblk; struct mkdir *mkdir; struct diradd *dap; int i, retval; ump = softdep_bp_to_mp(bp); if (ump == NULL) return (0); retval = 0; ACQUIRE_LOCK(ump); LIST_FOREACH(wk, &bp->b_dep, wk_list) { switch (wk->wk_type) { case D_INODEDEP: inodedep = WK_INODEDEP(wk); if ((inodedep->id_state & DEPCOMPLETE) == 0) { /* bitmap allocation dependency */ retval += 1; if (!wantcount) goto out; } if (TAILQ_FIRST(&inodedep->id_inoupdt)) { /* direct block pointer dependency */ retval += 1; if (!wantcount) goto out; } if (TAILQ_FIRST(&inodedep->id_extupdt)) { /* direct block pointer dependency */ retval += 1; if (!wantcount) goto out; } if (TAILQ_FIRST(&inodedep->id_inoreflst)) { /* Add reference dependency. */ retval += 1; if (!wantcount) goto out; } continue; case D_INDIRDEP: indirdep = WK_INDIRDEP(wk); TAILQ_FOREACH(freework, &indirdep->ir_trunc, fw_next) { /* indirect truncation dependency */ retval += 1; if (!wantcount) goto out; } LIST_FOREACH(aip, &indirdep->ir_deplisthd, ai_next) { /* indirect block pointer dependency */ retval += 1; if (!wantcount) goto out; } continue; case D_PAGEDEP: pagedep = WK_PAGEDEP(wk); LIST_FOREACH(dirrem, &pagedep->pd_dirremhd, dm_next) { if (LIST_FIRST(&dirrem->dm_jremrefhd)) { /* Journal remove ref dependency. */ retval += 1; if (!wantcount) goto out; } } for (i = 0; i < DAHASHSZ; i++) { LIST_FOREACH(dap, &pagedep->pd_diraddhd[i], da_pdlist) { /* directory entry dependency */ retval += 1; if (!wantcount) goto out; } } continue; case D_BMSAFEMAP: bmsafemap = WK_BMSAFEMAP(wk); if (LIST_FIRST(&bmsafemap->sm_jaddrefhd)) { /* Add reference dependency. */ retval += 1; if (!wantcount) goto out; } if (LIST_FIRST(&bmsafemap->sm_jnewblkhd)) { /* Allocate block dependency. */ retval += 1; if (!wantcount) goto out; } continue; case D_FREEBLKS: freeblks = WK_FREEBLKS(wk); if (LIST_FIRST(&freeblks->fb_jblkdephd)) { /* Freeblk journal dependency. */ retval += 1; if (!wantcount) goto out; } continue; case D_ALLOCDIRECT: case D_ALLOCINDIR: newblk = WK_NEWBLK(wk); if (newblk->nb_jnewblk) { /* Journal allocate dependency. */ retval += 1; if (!wantcount) goto out; } continue; case D_MKDIR: mkdir = WK_MKDIR(wk); if (mkdir->md_jaddref) { /* Journal reference dependency. */ retval += 1; if (!wantcount) goto out; } continue; case D_FREEWORK: case D_FREEDEP: case D_JSEGDEP: case D_JSEG: case D_SBDEP: /* never a dependency on these blocks */ continue; default: panic("softdep_count_dependencies: Unexpected type %s", TYPENAME(wk->wk_type)); /* NOTREACHED */ } } out: FREE_LOCK(ump); return (retval); } /* * Acquire exclusive access to a buffer. * Must be called with a locked mtx parameter. * Return acquired buffer or NULL on failure. */ static struct buf * getdirtybuf(bp, lock, waitfor) struct buf *bp; struct rwlock *lock; int waitfor; { int error; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL) != 0) { if (waitfor != MNT_WAIT) return (NULL); error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, lock); /* * Even if we successfully acquire bp here, we have dropped * lock, which may violates our guarantee. */ if (error == 0) BUF_UNLOCK(bp); else if (error != ENOLCK) panic("getdirtybuf: inconsistent lock: %d", error); rw_wlock(lock); return (NULL); } if ((bp->b_vflags & BV_BKGRDINPROG) != 0) { if (lock != BO_LOCKPTR(bp->b_bufobj) && waitfor == MNT_WAIT) { rw_wunlock(lock); BO_LOCK(bp->b_bufobj); BUF_UNLOCK(bp); if ((bp->b_vflags & BV_BKGRDINPROG) != 0) { bp->b_vflags |= BV_BKGRDWAIT; msleep(&bp->b_xflags, BO_LOCKPTR(bp->b_bufobj), PRIBIO | PDROP, "getbuf", 0); } else BO_UNLOCK(bp->b_bufobj); rw_wlock(lock); return (NULL); } BUF_UNLOCK(bp); if (waitfor != MNT_WAIT) return (NULL); #ifdef DEBUG_VFS_LOCKS if (bp->b_vp->v_type != VCHR) ASSERT_BO_WLOCKED(bp->b_bufobj); #endif bp->b_vflags |= BV_BKGRDWAIT; rw_sleep(&bp->b_xflags, lock, PRIBIO, "getbuf", 0); return (NULL); } if ((bp->b_flags & B_DELWRI) == 0) { BUF_UNLOCK(bp); return (NULL); } bremfree(bp); return (bp); } /* * Check if it is safe to suspend the file system now. On entry, * the vnode interlock for devvp should be held. Return 0 with * the mount interlock held if the file system can be suspended now, * otherwise return EAGAIN with the mount interlock held. */ int softdep_check_suspend(struct mount *mp, struct vnode *devvp, int softdep_depcnt, int softdep_accdepcnt, int secondary_writes, int secondary_accwrites) { struct buf *bp; struct bufobj *bo; struct ufsmount *ump; struct inodedep *inodedep; struct indirdep *indirdep; struct worklist *wk, *nextwk; int error, unlinked; bo = &devvp->v_bufobj; ASSERT_BO_WLOCKED(bo); /* * If we are not running with soft updates, then we need only * deal with secondary writes as we try to suspend. */ if (MOUNTEDSOFTDEP(mp) == 0) { MNT_ILOCK(mp); while (mp->mnt_secondary_writes != 0) { BO_UNLOCK(bo); msleep(&mp->mnt_secondary_writes, MNT_MTX(mp), (PUSER - 1) | PDROP, "secwr", 0); BO_LOCK(bo); MNT_ILOCK(mp); } /* * Reasons for needing more work before suspend: * - Dirty buffers on devvp. * - Secondary writes occurred after start of vnode sync loop */ error = 0; if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0 || secondary_writes != 0 || mp->mnt_secondary_writes != 0 || secondary_accwrites != mp->mnt_secondary_accwrites) error = EAGAIN; BO_UNLOCK(bo); return (error); } /* * If we are running with soft updates, then we need to coordinate * with them as we try to suspend. */ ump = VFSTOUFS(mp); for (;;) { if (!TRY_ACQUIRE_LOCK(ump)) { BO_UNLOCK(bo); ACQUIRE_LOCK(ump); FREE_LOCK(ump); BO_LOCK(bo); continue; } MNT_ILOCK(mp); if (mp->mnt_secondary_writes != 0) { FREE_LOCK(ump); BO_UNLOCK(bo); msleep(&mp->mnt_secondary_writes, MNT_MTX(mp), (PUSER - 1) | PDROP, "secwr", 0); BO_LOCK(bo); continue; } break; } unlinked = 0; if (MOUNTEDSUJ(mp)) { for (inodedep = TAILQ_FIRST(&ump->softdep_unlinked); inodedep != NULL; inodedep = TAILQ_NEXT(inodedep, id_unlinked)) { if ((inodedep->id_state & (UNLINKED | UNLINKLINKS | UNLINKONLIST)) != (UNLINKED | UNLINKLINKS | UNLINKONLIST) || !check_inodedep_free(inodedep)) continue; unlinked++; } } /* * XXX Check for orphaned indirdep dependency structures. * * During forcible unmount after a disk failure there is a * bug that causes one or more indirdep dependency structures * to fail to be deallocated. We check for them here and clean * them up so that the unmount can succeed. */ if ((ump->um_flags & UM_FSFAIL_CLEANUP) != 0 && ump->softdep_deps > 0 && ump->softdep_deps == ump->softdep_curdeps[D_INDIRDEP]) { LIST_FOREACH_SAFE(wk, &ump->softdep_alldeps[D_INDIRDEP], wk_all, nextwk) { indirdep = WK_INDIRDEP(wk); if ((indirdep->ir_state & (GOINGAWAY | DEPCOMPLETE)) != (GOINGAWAY | DEPCOMPLETE) || !TAILQ_EMPTY(&indirdep->ir_trunc) || !LIST_EMPTY(&indirdep->ir_completehd) || !LIST_EMPTY(&indirdep->ir_writehd) || !LIST_EMPTY(&indirdep->ir_donehd) || !LIST_EMPTY(&indirdep->ir_deplisthd) || indirdep->ir_saveddata != NULL || indirdep->ir_savebp == NULL) { printf("%s: skipping orphaned indirdep %p\n", __FUNCTION__, indirdep); continue; } printf("%s: freeing orphaned indirdep %p\n", __FUNCTION__, indirdep); bp = indirdep->ir_savebp; indirdep->ir_savebp = NULL; free_indirdep(indirdep); FREE_LOCK(ump); brelse(bp); while (!TRY_ACQUIRE_LOCK(ump)) { BO_UNLOCK(bo); ACQUIRE_LOCK(ump); FREE_LOCK(ump); BO_LOCK(bo); } } } /* * Reasons for needing more work before suspend: * - Dirty buffers on devvp. * - Dependency structures still exist * - Softdep activity occurred after start of vnode sync loop * - Secondary writes occurred after start of vnode sync loop */ error = 0; if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0 || softdep_depcnt != unlinked || ump->softdep_deps != unlinked || softdep_accdepcnt != ump->softdep_accdeps || secondary_writes != 0 || mp->mnt_secondary_writes != 0 || secondary_accwrites != mp->mnt_secondary_accwrites) error = EAGAIN; FREE_LOCK(ump); BO_UNLOCK(bo); return (error); } /* * Get the number of dependency structures for the file system, both * the current number and the total number allocated. These will * later be used to detect that softdep processing has occurred. */ void softdep_get_depcounts(struct mount *mp, int *softdep_depsp, int *softdep_accdepsp) { struct ufsmount *ump; if (MOUNTEDSOFTDEP(mp) == 0) { *softdep_depsp = 0; *softdep_accdepsp = 0; return; } ump = VFSTOUFS(mp); ACQUIRE_LOCK(ump); *softdep_depsp = ump->softdep_deps; *softdep_accdepsp = ump->softdep_accdeps; FREE_LOCK(ump); } /* * Wait for pending output on a vnode to complete. */ static void drain_output(vp) struct vnode *vp; { ASSERT_VOP_LOCKED(vp, "drain_output"); (void)bufobj_wwait(&vp->v_bufobj, 0, 0); } /* * Called whenever a buffer that is being invalidated or reallocated * contains dependencies. This should only happen if an I/O error has * occurred. The routine is called with the buffer locked. */ static void softdep_deallocate_dependencies(bp) struct buf *bp; { if ((bp->b_ioflags & BIO_ERROR) == 0) panic("softdep_deallocate_dependencies: dangling deps"); if (bp->b_vp != NULL && bp->b_vp->v_mount != NULL) softdep_error(bp->b_vp->v_mount->mnt_stat.f_mntonname, bp->b_error); else printf("softdep_deallocate_dependencies: " "got error %d while accessing filesystem\n", bp->b_error); if (bp->b_error != ENXIO) panic("softdep_deallocate_dependencies: unrecovered I/O error"); } /* * Function to handle asynchronous write errors in the filesystem. */ static void softdep_error(func, error) char *func; int error; { /* XXX should do something better! */ printf("%s: got error %d while accessing filesystem\n", func, error); } #ifdef DDB /* exported to ffs_vfsops.c */ extern void db_print_ffs(struct ufsmount *ump); void db_print_ffs(struct ufsmount *ump) { db_printf("mp %p (%s) devvp %p\n", ump->um_mountp, ump->um_mountp->mnt_stat.f_mntonname, ump->um_devvp); db_printf(" fs %p ", ump->um_fs); if (ump->um_softdep != NULL) { db_printf("su_wl %d su_deps %d su_req %d\n", ump->softdep_on_worklist, ump->softdep_deps, ump->softdep_req); } else { db_printf("su disabled\n"); } } static void worklist_print(struct worklist *wk, int verbose) { if (!verbose) { db_printf("%s: %p state 0x%b\n", TYPENAME(wk->wk_type), wk, (u_int)wk->wk_state, PRINT_SOFTDEP_FLAGS); return; } db_printf("worklist: %p type %s state 0x%b next %p\n ", wk, TYPENAME(wk->wk_type), (u_int)wk->wk_state, PRINT_SOFTDEP_FLAGS, LIST_NEXT(wk, wk_list)); db_print_ffs(VFSTOUFS(wk->wk_mp)); } static void inodedep_print(struct inodedep *inodedep, int verbose) { worklist_print(&inodedep->id_list, 0); db_printf(" fs %p ino %jd inoblk %jd delta %jd nlink %jd\n", inodedep->id_fs, (intmax_t)inodedep->id_ino, (intmax_t)fsbtodb(inodedep->id_fs, ino_to_fsba(inodedep->id_fs, inodedep->id_ino)), (intmax_t)inodedep->id_nlinkdelta, (intmax_t)inodedep->id_savednlink); if (verbose == 0) return; db_printf(" bmsafemap %p, mkdiradd %p, inoreflst %p\n", inodedep->id_bmsafemap, inodedep->id_mkdiradd, TAILQ_FIRST(&inodedep->id_inoreflst)); db_printf(" dirremhd %p, pendinghd %p, bufwait %p\n", LIST_FIRST(&inodedep->id_dirremhd), LIST_FIRST(&inodedep->id_pendinghd), LIST_FIRST(&inodedep->id_bufwait)); db_printf(" inowait %p, inoupdt %p, newinoupdt %p\n", LIST_FIRST(&inodedep->id_inowait), TAILQ_FIRST(&inodedep->id_inoupdt), TAILQ_FIRST(&inodedep->id_newinoupdt)); db_printf(" extupdt %p, newextupdt %p, freeblklst %p\n", TAILQ_FIRST(&inodedep->id_extupdt), TAILQ_FIRST(&inodedep->id_newextupdt), TAILQ_FIRST(&inodedep->id_freeblklst)); db_printf(" saveino %p, savedsize %jd, savedextsize %jd\n", inodedep->id_savedino1, (intmax_t)inodedep->id_savedsize, (intmax_t)inodedep->id_savedextsize); } static void newblk_print(struct newblk *nbp) { worklist_print(&nbp->nb_list, 0); db_printf(" newblkno %jd\n", (intmax_t)nbp->nb_newblkno); db_printf(" jnewblk %p, bmsafemap %p, freefrag %p\n", &nbp->nb_jnewblk, &nbp->nb_bmsafemap, &nbp->nb_freefrag); db_printf(" indirdeps %p, newdirblk %p, jwork %p\n", LIST_FIRST(&nbp->nb_indirdeps), LIST_FIRST(&nbp->nb_newdirblk), LIST_FIRST(&nbp->nb_jwork)); } static void allocdirect_print(struct allocdirect *adp) { newblk_print(&adp->ad_block); db_printf(" oldblkno %jd, oldsize %ld, newsize %ld\n", adp->ad_oldblkno, adp->ad_oldsize, adp->ad_newsize); db_printf(" offset %d, inodedep %p\n", adp->ad_offset, adp->ad_inodedep); } static void allocindir_print(struct allocindir *aip) { newblk_print(&aip->ai_block); db_printf(" oldblkno %jd, lbn %jd\n", (intmax_t)aip->ai_oldblkno, (intmax_t)aip->ai_lbn); db_printf(" offset %d, indirdep %p\n", aip->ai_offset, aip->ai_indirdep); } static void mkdir_print(struct mkdir *mkdir) { worklist_print(&mkdir->md_list, 0); db_printf(" diradd %p, jaddref %p, buf %p\n", mkdir->md_diradd, mkdir->md_jaddref, mkdir->md_buf); } DB_SHOW_COMMAND(sd_inodedep, db_show_sd_inodedep) { if (have_addr == 0) { db_printf("inodedep address required\n"); return; } inodedep_print((struct inodedep*)addr, 1); } DB_SHOW_COMMAND(sd_allinodedeps, db_show_sd_allinodedeps) { struct inodedep_hashhead *inodedephd; struct inodedep *inodedep; struct ufsmount *ump; int cnt; if (have_addr == 0) { db_printf("ufsmount address required\n"); return; } ump = (struct ufsmount *)addr; for (cnt = 0; cnt < ump->inodedep_hash_size; cnt++) { inodedephd = &ump->inodedep_hashtbl[cnt]; LIST_FOREACH(inodedep, inodedephd, id_hash) { inodedep_print(inodedep, 0); } } } DB_SHOW_COMMAND(sd_worklist, db_show_sd_worklist) { if (have_addr == 0) { db_printf("worklist address required\n"); return; } worklist_print((struct worklist *)addr, 1); } DB_SHOW_COMMAND(sd_workhead, db_show_sd_workhead) { struct worklist *wk; struct workhead *wkhd; if (have_addr == 0) { db_printf("worklist address required " "(for example value in bp->b_dep)\n"); return; } /* * We often do not have the address of the worklist head but * instead a pointer to its first entry (e.g., we have the * contents of bp->b_dep rather than &bp->b_dep). But the back * pointer of bp->b_dep will point at the head of the list, so * we cheat and use that instead. If we are in the middle of * a list we will still get the same result, so nothing * unexpected will result. */ wk = (struct worklist *)addr; if (wk == NULL) return; wkhd = (struct workhead *)wk->wk_list.le_prev; LIST_FOREACH(wk, wkhd, wk_list) { switch(wk->wk_type) { case D_INODEDEP: inodedep_print(WK_INODEDEP(wk), 0); continue; case D_ALLOCDIRECT: allocdirect_print(WK_ALLOCDIRECT(wk)); continue; case D_ALLOCINDIR: allocindir_print(WK_ALLOCINDIR(wk)); continue; case D_MKDIR: mkdir_print(WK_MKDIR(wk)); continue; default: worklist_print(wk, 0); continue; } } } DB_SHOW_COMMAND(sd_mkdir, db_show_sd_mkdir) { if (have_addr == 0) { db_printf("mkdir address required\n"); return; } mkdir_print((struct mkdir *)addr); } DB_SHOW_COMMAND(sd_mkdir_list, db_show_sd_mkdir_list) { struct mkdirlist *mkdirlisthd; struct mkdir *mkdir; if (have_addr == 0) { db_printf("mkdir listhead address required\n"); return; } mkdirlisthd = (struct mkdirlist *)addr; LIST_FOREACH(mkdir, mkdirlisthd, md_mkdirs) { mkdir_print(mkdir); if (mkdir->md_diradd != NULL) { db_printf(" "); worklist_print(&mkdir->md_diradd->da_list, 0); } if (mkdir->md_jaddref != NULL) { db_printf(" "); worklist_print(&mkdir->md_jaddref->ja_list, 0); } } } DB_SHOW_COMMAND(sd_allocdirect, db_show_sd_allocdirect) { if (have_addr == 0) { db_printf("allocdirect address required\n"); return; } allocdirect_print((struct allocdirect *)addr); } DB_SHOW_COMMAND(sd_allocindir, db_show_sd_allocindir) { if (have_addr == 0) { db_printf("allocindir address required\n"); return; } allocindir_print((struct allocindir *)addr); } #endif /* DDB */ #endif /* SOFTUPDATES */ diff --git a/sys/ufs/ufs/ufs_extattr.c b/sys/ufs/ufs/ufs_extattr.c index 140406d538dc..e9f0fb3ea762 100644 --- a/sys/ufs/ufs/ufs_extattr.c +++ b/sys/ufs/ufs/ufs_extattr.c @@ -1,1307 +1,1306 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 1999-2002 Robert N. M. Watson * Copyright (c) 2002-2003 Networks Associates Technology, Inc. * All rights reserved. * * This software was developed by Robert Watson for the TrustedBSD Project. * * This software was developed for the FreeBSD Project in part by Network * Associates Laboratories, the Security Research Division of Network * Associates, Inc. under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), * as part of the DARPA CHATS research program. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * */ /* * Support for filesystem extended attribute: UFS-specific support functions. */ #include __FBSDID("$FreeBSD$"); #include "opt_ufs.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef UFS_EXTATTR FEATURE(ufs_extattr, "ufs extended attribute support"); static MALLOC_DEFINE(M_UFS_EXTATTR, "ufs_extattr", "ufs extended attribute"); static int ufs_extattr_sync = 0; SYSCTL_INT(_debug, OID_AUTO, ufs_extattr_sync, CTLFLAG_RW, &ufs_extattr_sync, 0, ""); static int ufs_extattr_valid_attrname(int attrnamespace, const char *attrname); static int ufs_extattr_enable_with_open(struct ufsmount *ump, struct vnode *vp, int attrnamespace, const char *attrname, struct thread *td); static int ufs_extattr_enable(struct ufsmount *ump, int attrnamespace, const char *attrname, struct vnode *backing_vnode, struct thread *td); static int ufs_extattr_disable(struct ufsmount *ump, int attrnamespace, const char *attrname, struct thread *td); static int ufs_extattr_get(struct vnode *vp, int attrnamespace, const char *name, struct uio *uio, size_t *size, struct ucred *cred, struct thread *td); static int ufs_extattr_set(struct vnode *vp, int attrnamespace, const char *name, struct uio *uio, struct ucred *cred, struct thread *td); static int ufs_extattr_rm(struct vnode *vp, int attrnamespace, const char *name, struct ucred *cred, struct thread *td); #ifdef UFS_EXTATTR_AUTOSTART static int ufs_extattr_autostart_locked(struct mount *mp, struct thread *td); #endif static int ufs_extattr_start_locked(struct ufsmount *ump, struct thread *td); /* * Per-FS attribute lock protecting attribute operations. * * XXXRW: Perhaps something more fine-grained would be appropriate, but at * the end of the day we're going to contend on the vnode lock for the * backing file anyway. */ static void ufs_extattr_uepm_lock(struct ufsmount *ump) { sx_xlock(&ump->um_extattr.uepm_lock); } static void ufs_extattr_uepm_unlock(struct ufsmount *ump) { sx_xunlock(&ump->um_extattr.uepm_lock); } /*- * Determine whether the name passed is a valid name for an actual * attribute. * * Invalid currently consists of: * NULL pointer for attrname * zero-length attrname (used to retrieve application attribute list) */ static int ufs_extattr_valid_attrname(int attrnamespace, const char *attrname) { if (attrname == NULL) return (0); if (strlen(attrname) == 0) return (0); return (1); } /* * Locate an attribute given a name and mountpoint. * Must be holding uepm lock for the mount point. */ static struct ufs_extattr_list_entry * ufs_extattr_find_attr(struct ufsmount *ump, int attrnamespace, const char *attrname) { struct ufs_extattr_list_entry *search_attribute; sx_assert(&ump->um_extattr.uepm_lock, SA_XLOCKED); for (search_attribute = LIST_FIRST(&ump->um_extattr.uepm_list); search_attribute != NULL; search_attribute = LIST_NEXT(search_attribute, uele_entries)) { if (!(strncmp(attrname, search_attribute->uele_attrname, UFS_EXTATTR_MAXEXTATTRNAME)) && (attrnamespace == search_attribute->uele_attrnamespace)) { return (search_attribute); } } return (0); } /* * Initialize per-FS structures supporting extended attributes. Do not * start extended attributes yet. */ void ufs_extattr_uepm_init(struct ufs_extattr_per_mount *uepm) { uepm->uepm_flags = 0; LIST_INIT(&uepm->uepm_list); sx_init(&uepm->uepm_lock, "ufs_extattr_sx"); uepm->uepm_flags |= UFS_EXTATTR_UEPM_INITIALIZED; } /* * Destroy per-FS structures supporting extended attributes. Assumes * that EAs have already been stopped, and will panic if not. */ void ufs_extattr_uepm_destroy(struct ufs_extattr_per_mount *uepm) { if (!(uepm->uepm_flags & UFS_EXTATTR_UEPM_INITIALIZED)) panic("ufs_extattr_uepm_destroy: not initialized"); if ((uepm->uepm_flags & UFS_EXTATTR_UEPM_STARTED)) panic("ufs_extattr_uepm_destroy: called while still started"); /* * It's not clear that either order for the next two lines is * ideal, and it should never be a problem if this is only called * during unmount, and with vfs_busy(). */ uepm->uepm_flags &= ~UFS_EXTATTR_UEPM_INITIALIZED; sx_destroy(&uepm->uepm_lock); } /* * Start extended attribute support on an FS. */ int ufs_extattr_start(struct mount *mp, struct thread *td) { struct ufsmount *ump; int error = 0; ump = VFSTOUFS(mp); ufs_extattr_uepm_lock(ump); error = ufs_extattr_start_locked(ump, td); ufs_extattr_uepm_unlock(ump); return (error); } static int ufs_extattr_start_locked(struct ufsmount *ump, struct thread *td) { if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_INITIALIZED)) return (EOPNOTSUPP); if (ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED) return (EBUSY); ump->um_extattr.uepm_flags |= UFS_EXTATTR_UEPM_STARTED; ump->um_extattr.uepm_ucred = crhold(td->td_ucred); return (0); } #ifdef UFS_EXTATTR_AUTOSTART /* * Helper routine: given a locked parent directory and filename, return * the locked vnode of the inode associated with the name. Will not * follow symlinks, may return any type of vnode. Lock on parent will * be released even in the event of a failure. In the event that the * target is the parent (i.e., "."), there will be two references and * one lock, requiring the caller to possibly special-case. */ #define UE_GETDIR_LOCKPARENT 1 #define UE_GETDIR_LOCKPARENT_DONT 2 static int ufs_extattr_lookup(struct vnode *start_dvp, int lockparent, char *dirname, struct vnode **vp, struct thread *td) { struct vop_cachedlookup_args vargs; struct componentname cnp; struct vnode *target_vp; int error; bzero(&cnp, sizeof(cnp)); cnp.cn_nameiop = LOOKUP; cnp.cn_flags = ISLASTCN; if (lockparent == UE_GETDIR_LOCKPARENT) cnp.cn_flags |= LOCKPARENT; cnp.cn_lkflags = LK_EXCLUSIVE; - cnp.cn_thread = td; cnp.cn_cred = td->td_ucred; cnp.cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK); cnp.cn_nameptr = cnp.cn_pnbuf; error = copystr(dirname, cnp.cn_pnbuf, MAXPATHLEN, (size_t *) &cnp.cn_namelen); if (error) { if (lockparent == UE_GETDIR_LOCKPARENT_DONT) { VOP_UNLOCK(start_dvp); } uma_zfree(namei_zone, cnp.cn_pnbuf); printf("ufs_extattr_lookup: copystr failed\n"); return (error); } cnp.cn_namelen--; /* trim nul termination */ vargs.a_gen.a_desc = NULL; vargs.a_dvp = start_dvp; vargs.a_vpp = &target_vp; vargs.a_cnp = &cnp; error = ufs_lookup(&vargs); uma_zfree(namei_zone, cnp.cn_pnbuf); if (error) { /* * Error condition, may have to release the lock on the parent * if ufs_lookup() didn't. */ if (lockparent == UE_GETDIR_LOCKPARENT_DONT) VOP_UNLOCK(start_dvp); /* * Check that ufs_lookup() didn't release the lock when we * didn't want it to. */ if (lockparent == UE_GETDIR_LOCKPARENT) ASSERT_VOP_LOCKED(start_dvp, "ufs_extattr_lookup"); return (error); } /* if (target_vp == start_dvp) panic("ufs_extattr_lookup: target_vp == start_dvp"); */ if (target_vp != start_dvp && lockparent == UE_GETDIR_LOCKPARENT_DONT) VOP_UNLOCK(start_dvp); if (lockparent == UE_GETDIR_LOCKPARENT) ASSERT_VOP_LOCKED(start_dvp, "ufs_extattr_lookup"); /* printf("ufs_extattr_lookup: success\n"); */ *vp = target_vp; return (0); } #endif /* !UFS_EXTATTR_AUTOSTART */ /* * Enable an EA using the passed filesystem, backing vnode, attribute name, * namespace, and proc. Will perform a VOP_OPEN() on the vp, so expects vp * to be locked when passed in. The vnode will be returned unlocked, * regardless of success/failure of the function. As a result, the caller * will always need to vrele(), but not vput(). */ static int ufs_extattr_enable_with_open(struct ufsmount *ump, struct vnode *vp, int attrnamespace, const char *attrname, struct thread *td) { int error; error = VOP_OPEN(vp, FREAD|FWRITE, td->td_ucred, td, NULL); if (error) { printf("ufs_extattr_enable_with_open.VOP_OPEN(): failed " "with %d\n", error); VOP_UNLOCK(vp); return (error); } error = VOP_ADD_WRITECOUNT(vp, 1); if (error != 0) { VOP_CLOSE(vp, FREAD | FWRITE, td->td_ucred, td); VOP_UNLOCK(vp); return (error); } CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d", __func__, vp, vp->v_writecount); vref(vp); VOP_UNLOCK(vp); error = ufs_extattr_enable(ump, attrnamespace, attrname, vp, td); if (error != 0) vn_close(vp, FREAD|FWRITE, td->td_ucred, td); return (error); } #ifdef UFS_EXTATTR_AUTOSTART /* * Given a locked directory vnode, iterate over the names in the directory * and use ufs_extattr_lookup() to retrieve locked vnodes of potential * attribute files. Then invoke ufs_extattr_enable_with_open() on each * to attempt to start the attribute. Leaves the directory locked on * exit. */ static int ufs_extattr_iterate_directory(struct ufsmount *ump, struct vnode *dvp, int attrnamespace, struct thread *td) { struct vop_readdir_args vargs; struct dirent *dp, *edp; struct vnode *attr_vp; struct uio auio; struct iovec aiov; char *dirbuf; int error, eofflag = 0; if (dvp->v_type != VDIR) return (ENOTDIR); dirbuf = malloc(DIRBLKSIZ, M_TEMP, M_WAITOK); auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_rw = UIO_READ; auio.uio_segflg = UIO_SYSSPACE; auio.uio_td = td; auio.uio_offset = 0; vargs.a_gen.a_desc = NULL; vargs.a_vp = dvp; vargs.a_uio = &auio; vargs.a_cred = td->td_ucred; vargs.a_eofflag = &eofflag; vargs.a_ncookies = NULL; vargs.a_cookies = NULL; while (!eofflag) { auio.uio_resid = DIRBLKSIZ; aiov.iov_base = dirbuf; aiov.iov_len = DIRBLKSIZ; error = ufs_readdir(&vargs); if (error) { printf("ufs_extattr_iterate_directory: ufs_readdir " "%d\n", error); return (error); } edp = (struct dirent *)&dirbuf[DIRBLKSIZ - auio.uio_resid]; for (dp = (struct dirent *)dirbuf; dp < edp; ) { if (dp->d_reclen == 0) break; error = ufs_extattr_lookup(dvp, UE_GETDIR_LOCKPARENT, dp->d_name, &attr_vp, td); if (error) { printf("ufs_extattr_iterate_directory: lookup " "%s %d\n", dp->d_name, error); } else if (attr_vp == dvp) { vrele(attr_vp); } else if (attr_vp->v_type != VREG) { vput(attr_vp); } else { error = ufs_extattr_enable_with_open(ump, attr_vp, attrnamespace, dp->d_name, td); vrele(attr_vp); if (error) { printf("ufs_extattr_iterate_directory: " "enable %s %d\n", dp->d_name, error); } else if (bootverbose) { printf("UFS autostarted EA %s\n", dp->d_name); } } dp = (struct dirent *) ((char *)dp + dp->d_reclen); if (dp >= edp) break; } } free(dirbuf, M_TEMP); return (0); } /* * Auto-start of extended attributes, to be executed (optionally) at * mount-time. */ int ufs_extattr_autostart(struct mount *mp, struct thread *td) { struct ufsmount *ump; int error; ump = VFSTOUFS(mp); ufs_extattr_uepm_lock(ump); error = ufs_extattr_autostart_locked(mp, td); ufs_extattr_uepm_unlock(ump); return (error); } static int ufs_extattr_autostart_locked(struct mount *mp, struct thread *td) { struct vnode *rvp, *attr_dvp, *attr_system_dvp, *attr_user_dvp; struct ufsmount *ump = VFSTOUFS(mp); int error; /* * UFS_EXTATTR applies only to UFS1, as UFS2 uses native extended * attributes, so don't autostart. */ if (ump->um_fstype != UFS1) return (0); /* * Does UFS_EXTATTR_FSROOTSUBDIR exist off the filesystem root? * If so, automatically start EA's. */ error = VFS_ROOT(mp, LK_EXCLUSIVE, &rvp); if (error) { printf("ufs_extattr_autostart.VFS_ROOT() returned %d\n", error); return (error); } error = ufs_extattr_lookup(rvp, UE_GETDIR_LOCKPARENT_DONT, UFS_EXTATTR_FSROOTSUBDIR, &attr_dvp, td); if (error) { /* rvp ref'd but now unlocked */ vrele(rvp); return (error); } if (rvp == attr_dvp) { /* Should never happen. */ vput(rvp); vrele(attr_dvp); return (EINVAL); } vrele(rvp); if (attr_dvp->v_type != VDIR) { printf("ufs_extattr_autostart: %s != VDIR\n", UFS_EXTATTR_FSROOTSUBDIR); goto return_vput_attr_dvp; } error = ufs_extattr_start_locked(ump, td); if (error) { printf("ufs_extattr_autostart: ufs_extattr_start failed (%d)\n", error); goto return_vput_attr_dvp; } /* * Look for two subdirectories: UFS_EXTATTR_SUBDIR_SYSTEM, * UFS_EXTATTR_SUBDIR_USER. For each, iterate over the sub-directory, * and start with appropriate type. Failures in either don't * result in an over-all failure. attr_dvp is left locked to * be cleaned up on exit. */ error = ufs_extattr_lookup(attr_dvp, UE_GETDIR_LOCKPARENT, UFS_EXTATTR_SUBDIR_SYSTEM, &attr_system_dvp, td); if (!error) { error = ufs_extattr_iterate_directory(VFSTOUFS(mp), attr_system_dvp, EXTATTR_NAMESPACE_SYSTEM, td); if (error) printf("ufs_extattr_iterate_directory returned %d\n", error); vput(attr_system_dvp); } error = ufs_extattr_lookup(attr_dvp, UE_GETDIR_LOCKPARENT, UFS_EXTATTR_SUBDIR_USER, &attr_user_dvp, td); if (!error) { error = ufs_extattr_iterate_directory(VFSTOUFS(mp), attr_user_dvp, EXTATTR_NAMESPACE_USER, td); if (error) printf("ufs_extattr_iterate_directory returned %d\n", error); vput(attr_user_dvp); } /* Mask startup failures in sub-directories. */ error = 0; return_vput_attr_dvp: vput(attr_dvp); return (error); } #endif /* !UFS_EXTATTR_AUTOSTART */ /* * Stop extended attribute support on an FS. */ int ufs_extattr_stop(struct mount *mp, struct thread *td) { struct ufs_extattr_list_entry *uele; struct ufsmount *ump = VFSTOUFS(mp); int error = 0; ufs_extattr_uepm_lock(ump); if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED)) { error = EOPNOTSUPP; goto unlock; } while ((uele = LIST_FIRST(&ump->um_extattr.uepm_list)) != NULL) { ufs_extattr_disable(ump, uele->uele_attrnamespace, uele->uele_attrname, td); } ump->um_extattr.uepm_flags &= ~UFS_EXTATTR_UEPM_STARTED; crfree(ump->um_extattr.uepm_ucred); ump->um_extattr.uepm_ucred = NULL; unlock: ufs_extattr_uepm_unlock(ump); return (error); } /* * Enable a named attribute on the specified filesystem; provide an * unlocked backing vnode to hold the attribute data. */ static int ufs_extattr_enable(struct ufsmount *ump, int attrnamespace, const char *attrname, struct vnode *backing_vnode, struct thread *td) { struct ufs_extattr_list_entry *attribute; struct iovec aiov; struct uio auio; int error = 0; if (!ufs_extattr_valid_attrname(attrnamespace, attrname)) return (EINVAL); if (backing_vnode->v_type != VREG) return (EINVAL); attribute = malloc(sizeof(struct ufs_extattr_list_entry), M_UFS_EXTATTR, M_WAITOK); if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED)) { error = EOPNOTSUPP; goto free_exit; } if (ufs_extattr_find_attr(ump, attrnamespace, attrname)) { error = EEXIST; goto free_exit; } strncpy(attribute->uele_attrname, attrname, UFS_EXTATTR_MAXEXTATTRNAME); attribute->uele_attrnamespace = attrnamespace; bzero(&attribute->uele_fileheader, sizeof(struct ufs_extattr_fileheader)); attribute->uele_backing_vnode = backing_vnode; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; aiov.iov_base = (caddr_t) &attribute->uele_fileheader; aiov.iov_len = sizeof(struct ufs_extattr_fileheader); auio.uio_resid = sizeof(struct ufs_extattr_fileheader); auio.uio_offset = (off_t) 0; auio.uio_segflg = UIO_SYSSPACE; auio.uio_rw = UIO_READ; auio.uio_td = td; vn_lock(backing_vnode, LK_SHARED | LK_RETRY); error = VOP_READ(backing_vnode, &auio, IO_NODELOCKED, ump->um_extattr.uepm_ucred); if (error) goto unlock_free_exit; if (auio.uio_resid != 0) { printf("ufs_extattr_enable: malformed attribute header\n"); error = EINVAL; goto unlock_free_exit; } if (attribute->uele_fileheader.uef_magic != UFS_EXTATTR_MAGIC) { printf("ufs_extattr_enable: invalid attribute header magic\n"); error = EINVAL; goto unlock_free_exit; } if (attribute->uele_fileheader.uef_version != UFS_EXTATTR_VERSION) { printf("ufs_extattr_enable: incorrect attribute header " "version\n"); error = EINVAL; goto unlock_free_exit; } ASSERT_VOP_LOCKED(backing_vnode, "ufs_extattr_enable"); LIST_INSERT_HEAD(&ump->um_extattr.uepm_list, attribute, uele_entries); VOP_UNLOCK(backing_vnode); return (0); unlock_free_exit: VOP_UNLOCK(backing_vnode); free_exit: free(attribute, M_UFS_EXTATTR); return (error); } /* * Disable extended attribute support on an FS. */ static int ufs_extattr_disable(struct ufsmount *ump, int attrnamespace, const char *attrname, struct thread *td) { struct ufs_extattr_list_entry *uele; int error = 0; if (!ufs_extattr_valid_attrname(attrnamespace, attrname)) return (EINVAL); uele = ufs_extattr_find_attr(ump, attrnamespace, attrname); if (!uele) return (ENOATTR); LIST_REMOVE(uele, uele_entries); vn_lock(uele->uele_backing_vnode, LK_SHARED | LK_RETRY); ASSERT_VOP_LOCKED(uele->uele_backing_vnode, "ufs_extattr_disable"); VOP_UNLOCK(uele->uele_backing_vnode); error = vn_close(uele->uele_backing_vnode, FREAD|FWRITE, td->td_ucred, td); free(uele, M_UFS_EXTATTR); return (error); } /* * VFS call to manage extended attributes in UFS. If filename_vp is * non-NULL, it must be passed in locked, and regardless of errors in * processing, will be unlocked. */ int ufs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp, int attrnamespace, const char *attrname) { struct ufsmount *ump = VFSTOUFS(mp); struct thread *td = curthread; int error; /* * Processes with privilege, but in jail, are not allowed to * configure extended attributes. */ error = priv_check(td, PRIV_UFS_EXTATTRCTL); if (error) { if (filename_vp != NULL) VOP_UNLOCK(filename_vp); return (error); } /* * We only allow extattrctl(2) on UFS1 file systems, as UFS2 uses * native extended attributes. */ if (ump->um_fstype != UFS1) { if (filename_vp != NULL) VOP_UNLOCK(filename_vp); return (EOPNOTSUPP); } switch(cmd) { case UFS_EXTATTR_CMD_START: if (filename_vp != NULL) { VOP_UNLOCK(filename_vp); return (EINVAL); } if (attrname != NULL) return (EINVAL); error = ufs_extattr_start(mp, td); return (error); case UFS_EXTATTR_CMD_STOP: if (filename_vp != NULL) { VOP_UNLOCK(filename_vp); return (EINVAL); } if (attrname != NULL) return (EINVAL); error = ufs_extattr_stop(mp, td); return (error); case UFS_EXTATTR_CMD_ENABLE: if (filename_vp == NULL) return (EINVAL); if (attrname == NULL) { VOP_UNLOCK(filename_vp); return (EINVAL); } /* * ufs_extattr_enable_with_open() will always unlock the * vnode, regardless of failure. */ ufs_extattr_uepm_lock(ump); error = ufs_extattr_enable_with_open(ump, filename_vp, attrnamespace, attrname, td); ufs_extattr_uepm_unlock(ump); return (error); case UFS_EXTATTR_CMD_DISABLE: if (filename_vp != NULL) { VOP_UNLOCK(filename_vp); return (EINVAL); } if (attrname == NULL) return (EINVAL); ufs_extattr_uepm_lock(ump); error = ufs_extattr_disable(ump, attrnamespace, attrname, td); ufs_extattr_uepm_unlock(ump); return (error); default: return (EINVAL); } } /* * Vnode operating to retrieve a named extended attribute. */ int ufs_getextattr(struct vop_getextattr_args *ap) /* vop_getextattr { IN struct vnode *a_vp; IN int a_attrnamespace; IN const char *a_name; INOUT struct uio *a_uio; OUT size_t *a_size; IN struct ucred *a_cred; IN struct thread *a_td; }; */ { struct mount *mp = ap->a_vp->v_mount; struct ufsmount *ump = VFSTOUFS(mp); int error; ufs_extattr_uepm_lock(ump); error = ufs_extattr_get(ap->a_vp, ap->a_attrnamespace, ap->a_name, ap->a_uio, ap->a_size, ap->a_cred, ap->a_td); ufs_extattr_uepm_unlock(ump); return (error); } /* * Real work associated with retrieving a named attribute--assumes that * the attribute lock has already been grabbed. */ static int ufs_extattr_get(struct vnode *vp, int attrnamespace, const char *name, struct uio *uio, size_t *size, struct ucred *cred, struct thread *td) { struct ufs_extattr_list_entry *attribute; struct ufs_extattr_header ueh; struct iovec local_aiov; struct uio local_aio; struct mount *mp = vp->v_mount; struct ufsmount *ump = VFSTOUFS(mp); struct inode *ip = VTOI(vp); off_t base_offset; size_t len, old_len; int error = 0; if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED)) return (EOPNOTSUPP); if (strlen(name) == 0) return (EINVAL); error = extattr_check_cred(vp, attrnamespace, cred, td, VREAD); if (error) return (error); attribute = ufs_extattr_find_attr(ump, attrnamespace, name); if (!attribute) return (ENOATTR); /* * Allow only offsets of zero to encourage the read/replace * extended attribute semantic. Otherwise we can't guarantee * atomicity, as we don't provide locks for extended attributes. */ if (uio != NULL && uio->uio_offset != 0) return (ENXIO); /* * Find base offset of header in file based on file header size, and * data header size + maximum data size, indexed by inode number. */ base_offset = sizeof(struct ufs_extattr_fileheader) + ip->i_number * (sizeof(struct ufs_extattr_header) + attribute->uele_fileheader.uef_size); /* * Read in the data header to see if the data is defined, and if so * how much. */ bzero(&ueh, sizeof(struct ufs_extattr_header)); local_aiov.iov_base = (caddr_t) &ueh; local_aiov.iov_len = sizeof(struct ufs_extattr_header); local_aio.uio_iov = &local_aiov; local_aio.uio_iovcnt = 1; local_aio.uio_rw = UIO_READ; local_aio.uio_segflg = UIO_SYSSPACE; local_aio.uio_td = td; local_aio.uio_offset = base_offset; local_aio.uio_resid = sizeof(struct ufs_extattr_header); /* * Acquire locks. * * Don't need to get a lock on the backing file if the getattr is * being applied to the backing file, as the lock is already held. */ if (attribute->uele_backing_vnode != vp) vn_lock(attribute->uele_backing_vnode, LK_SHARED | LK_RETRY); error = VOP_READ(attribute->uele_backing_vnode, &local_aio, IO_NODELOCKED, ump->um_extattr.uepm_ucred); if (error) goto vopunlock_exit; /* Defined? */ if ((ueh.ueh_flags & UFS_EXTATTR_ATTR_FLAG_INUSE) == 0) { error = ENOATTR; goto vopunlock_exit; } /* Valid for the current inode generation? */ if (ueh.ueh_i_gen != ip->i_gen) { /* * The inode itself has a different generation number * than the attribute data. For now, the best solution * is to coerce this to undefined, and let it get cleaned * up by the next write or extattrctl clean. */ printf("ufs_extattr_get (%s): inode number inconsistency (%d, %ju)\n", mp->mnt_stat.f_mntonname, ueh.ueh_i_gen, (uintmax_t)ip->i_gen); error = ENOATTR; goto vopunlock_exit; } /* Local size consistency check. */ if (ueh.ueh_len > attribute->uele_fileheader.uef_size) { error = ENXIO; goto vopunlock_exit; } /* Return full data size if caller requested it. */ if (size != NULL) *size = ueh.ueh_len; /* Return data if the caller requested it. */ if (uio != NULL) { /* Allow for offset into the attribute data. */ uio->uio_offset = base_offset + sizeof(struct ufs_extattr_header); /* * Figure out maximum to transfer -- use buffer size and * local data limit. */ len = MIN(uio->uio_resid, ueh.ueh_len); old_len = uio->uio_resid; uio->uio_resid = len; error = VOP_READ(attribute->uele_backing_vnode, uio, IO_NODELOCKED, ump->um_extattr.uepm_ucred); if (error) goto vopunlock_exit; uio->uio_resid = old_len - (len - uio->uio_resid); } vopunlock_exit: if (uio != NULL) uio->uio_offset = 0; if (attribute->uele_backing_vnode != vp) VOP_UNLOCK(attribute->uele_backing_vnode); return (error); } /* * Vnode operation to remove a named attribute. */ int ufs_deleteextattr(struct vop_deleteextattr_args *ap) /* vop_deleteextattr { IN struct vnode *a_vp; IN int a_attrnamespace; IN const char *a_name; IN struct ucred *a_cred; IN struct thread *a_td; }; */ { struct mount *mp = ap->a_vp->v_mount; struct ufsmount *ump = VFSTOUFS(mp); int error; ufs_extattr_uepm_lock(ump); error = ufs_extattr_rm(ap->a_vp, ap->a_attrnamespace, ap->a_name, ap->a_cred, ap->a_td); ufs_extattr_uepm_unlock(ump); return (error); } /* * Vnode operation to set a named attribute. */ int ufs_setextattr(struct vop_setextattr_args *ap) /* vop_setextattr { IN struct vnode *a_vp; IN int a_attrnamespace; IN const char *a_name; INOUT struct uio *a_uio; IN struct ucred *a_cred; IN struct thread *a_td; }; */ { struct mount *mp = ap->a_vp->v_mount; struct ufsmount *ump = VFSTOUFS(mp); int error; /* * XXX: No longer a supported way to delete extended attributes. */ if (ap->a_uio == NULL) return (EINVAL); ufs_extattr_uepm_lock(ump); error = ufs_extattr_set(ap->a_vp, ap->a_attrnamespace, ap->a_name, ap->a_uio, ap->a_cred, ap->a_td); ufs_extattr_uepm_unlock(ump); return (error); } /* * Real work associated with setting a vnode's extended attributes; * assumes that the attribute lock has already been grabbed. */ static int ufs_extattr_set(struct vnode *vp, int attrnamespace, const char *name, struct uio *uio, struct ucred *cred, struct thread *td) { struct ufs_extattr_list_entry *attribute; struct ufs_extattr_header ueh; struct iovec local_aiov; struct uio local_aio; struct mount *mp = vp->v_mount; struct ufsmount *ump = VFSTOUFS(mp); struct inode *ip = VTOI(vp); off_t base_offset; int error = 0, ioflag; if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED)) return (EOPNOTSUPP); if (!ufs_extattr_valid_attrname(attrnamespace, name)) return (EINVAL); error = extattr_check_cred(vp, attrnamespace, cred, td, VWRITE); if (error) return (error); attribute = ufs_extattr_find_attr(ump, attrnamespace, name); if (!attribute) return (ENOATTR); /* * Early rejection of invalid offsets/length. * Reject: any offset but 0 (replace) * Any size greater than attribute size limit */ if (uio->uio_offset != 0 || uio->uio_resid > attribute->uele_fileheader.uef_size) return (ENXIO); /* * Find base offset of header in file based on file header size, and * data header size + maximum data size, indexed by inode number. */ base_offset = sizeof(struct ufs_extattr_fileheader) + ip->i_number * (sizeof(struct ufs_extattr_header) + attribute->uele_fileheader.uef_size); /* * Write out a data header for the data. */ ueh.ueh_len = uio->uio_resid; ueh.ueh_flags = UFS_EXTATTR_ATTR_FLAG_INUSE; ueh.ueh_i_gen = ip->i_gen; local_aiov.iov_base = (caddr_t) &ueh; local_aiov.iov_len = sizeof(struct ufs_extattr_header); local_aio.uio_iov = &local_aiov; local_aio.uio_iovcnt = 1; local_aio.uio_rw = UIO_WRITE; local_aio.uio_segflg = UIO_SYSSPACE; local_aio.uio_td = td; local_aio.uio_offset = base_offset; local_aio.uio_resid = sizeof(struct ufs_extattr_header); /* * Acquire locks. * * Don't need to get a lock on the backing file if the setattr is * being applied to the backing file, as the lock is already held. */ if (attribute->uele_backing_vnode != vp) vn_lock(attribute->uele_backing_vnode, LK_EXCLUSIVE | LK_RETRY); ioflag = IO_NODELOCKED; if (ufs_extattr_sync) ioflag |= IO_SYNC; error = VOP_WRITE(attribute->uele_backing_vnode, &local_aio, ioflag, ump->um_extattr.uepm_ucred); if (error) goto vopunlock_exit; if (local_aio.uio_resid != 0) { error = ENXIO; goto vopunlock_exit; } /* * Write out user data. */ uio->uio_offset = base_offset + sizeof(struct ufs_extattr_header); ioflag = IO_NODELOCKED; if (ufs_extattr_sync) ioflag |= IO_SYNC; error = VOP_WRITE(attribute->uele_backing_vnode, uio, ioflag, ump->um_extattr.uepm_ucred); vopunlock_exit: uio->uio_offset = 0; if (attribute->uele_backing_vnode != vp) VOP_UNLOCK(attribute->uele_backing_vnode); return (error); } /* * Real work associated with removing an extended attribute from a vnode. * Assumes the attribute lock has already been grabbed. */ static int ufs_extattr_rm(struct vnode *vp, int attrnamespace, const char *name, struct ucred *cred, struct thread *td) { struct ufs_extattr_list_entry *attribute; struct ufs_extattr_header ueh; struct iovec local_aiov; struct uio local_aio; struct mount *mp = vp->v_mount; struct ufsmount *ump = VFSTOUFS(mp); struct inode *ip = VTOI(vp); off_t base_offset; int error = 0, ioflag; if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED)) return (EOPNOTSUPP); if (!ufs_extattr_valid_attrname(attrnamespace, name)) return (EINVAL); error = extattr_check_cred(vp, attrnamespace, cred, td, VWRITE); if (error) return (error); attribute = ufs_extattr_find_attr(ump, attrnamespace, name); if (!attribute) return (ENOATTR); /* * Find base offset of header in file based on file header size, and * data header size + maximum data size, indexed by inode number. */ base_offset = sizeof(struct ufs_extattr_fileheader) + ip->i_number * (sizeof(struct ufs_extattr_header) + attribute->uele_fileheader.uef_size); /* * Check to see if currently defined. */ bzero(&ueh, sizeof(struct ufs_extattr_header)); local_aiov.iov_base = (caddr_t) &ueh; local_aiov.iov_len = sizeof(struct ufs_extattr_header); local_aio.uio_iov = &local_aiov; local_aio.uio_iovcnt = 1; local_aio.uio_rw = UIO_READ; local_aio.uio_segflg = UIO_SYSSPACE; local_aio.uio_td = td; local_aio.uio_offset = base_offset; local_aio.uio_resid = sizeof(struct ufs_extattr_header); /* * Don't need to get the lock on the backing vnode if the vnode we're * modifying is it, as we already hold the lock. */ if (attribute->uele_backing_vnode != vp) vn_lock(attribute->uele_backing_vnode, LK_EXCLUSIVE | LK_RETRY); error = VOP_READ(attribute->uele_backing_vnode, &local_aio, IO_NODELOCKED, ump->um_extattr.uepm_ucred); if (error) goto vopunlock_exit; /* Defined? */ if ((ueh.ueh_flags & UFS_EXTATTR_ATTR_FLAG_INUSE) == 0) { error = ENOATTR; goto vopunlock_exit; } /* Valid for the current inode generation? */ if (ueh.ueh_i_gen != ip->i_gen) { /* * The inode itself has a different generation number than * the attribute data. For now, the best solution is to * coerce this to undefined, and let it get cleaned up by * the next write or extattrctl clean. */ printf("ufs_extattr_rm (%s): inode number inconsistency (%d, %jd)\n", mp->mnt_stat.f_mntonname, ueh.ueh_i_gen, (intmax_t)ip->i_gen); error = ENOATTR; goto vopunlock_exit; } /* Flag it as not in use. */ ueh.ueh_flags = 0; ueh.ueh_len = 0; local_aiov.iov_base = (caddr_t) &ueh; local_aiov.iov_len = sizeof(struct ufs_extattr_header); local_aio.uio_iov = &local_aiov; local_aio.uio_iovcnt = 1; local_aio.uio_rw = UIO_WRITE; local_aio.uio_segflg = UIO_SYSSPACE; local_aio.uio_td = td; local_aio.uio_offset = base_offset; local_aio.uio_resid = sizeof(struct ufs_extattr_header); ioflag = IO_NODELOCKED; if (ufs_extattr_sync) ioflag |= IO_SYNC; error = VOP_WRITE(attribute->uele_backing_vnode, &local_aio, ioflag, ump->um_extattr.uepm_ucred); if (error) goto vopunlock_exit; if (local_aio.uio_resid != 0) error = ENXIO; vopunlock_exit: VOP_UNLOCK(attribute->uele_backing_vnode); return (error); } /* * Called by UFS when an inode is no longer active and should have its * attributes stripped. */ void ufs_extattr_vnode_inactive(struct vnode *vp) { struct ufs_extattr_list_entry *uele; struct mount *mp = vp->v_mount; struct ufsmount *ump = VFSTOUFS(mp); /* * In that case, we cannot lock. We should not have any active vnodes * on the fs if this is not yet initialized but is going to be, so * this can go unlocked. */ if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_INITIALIZED)) return; ufs_extattr_uepm_lock(ump); if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED)) { ufs_extattr_uepm_unlock(ump); return; } LIST_FOREACH(uele, &ump->um_extattr.uepm_list, uele_entries) ufs_extattr_rm(vp, uele->uele_attrnamespace, uele->uele_attrname, NULL, curthread); ufs_extattr_uepm_unlock(ump); } #endif /* !UFS_EXTATTR */ diff --git a/sys/ufs/ufs/ufs_lookup.c b/sys/ufs/ufs/ufs_lookup.c index fc78c017e2c6..3d24ad23f6e2 100644 --- a/sys/ufs/ufs/ufs_lookup.c +++ b/sys/ufs/ufs/ufs_lookup.c @@ -1,1608 +1,1608 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)ufs_lookup.c 8.15 (Berkeley) 6/16/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_ufs.h" #include "opt_quota.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef UFS_DIRHASH #include #endif #include #include #include #ifdef DIAGNOSTIC static int dirchk = 1; #else static int dirchk = 0; #endif SYSCTL_INT(_debug, OID_AUTO, dircheck, CTLFLAG_RW, &dirchk, 0, ""); static int ufs_delete_denied(struct vnode *vdp, struct vnode *tdp, struct ucred *cred, struct thread *td) { int error; #ifdef UFS_ACL /* * NFSv4 Minor Version 1, draft-ietf-nfsv4-minorversion1-03.txt * * 3.16.2.1. ACE4_DELETE vs. ACE4_DELETE_CHILD */ /* * XXX: Is this check required? */ error = VOP_ACCESS(vdp, VEXEC, cred, td); if (error) return (error); error = VOP_ACCESSX(tdp, VDELETE, cred, td); if (error == 0) return (0); error = VOP_ACCESSX(vdp, VDELETE_CHILD, cred, td); if (error == 0) return (0); error = VOP_ACCESSX(vdp, VEXPLICIT_DENY | VDELETE_CHILD, cred, td); if (error) return (error); #endif /* !UFS_ACL */ /* * Standard Unix access control - delete access requires VWRITE. */ error = VOP_ACCESS(vdp, VWRITE, cred, td); if (error) return (error); /* * If directory is "sticky", then user must own * the directory, or the file in it, else she * may not delete it (unless she's root). This * implements append-only directories. */ if ((VTOI(vdp)->i_mode & ISVTX) && VOP_ACCESS(vdp, VADMIN, cred, td) && VOP_ACCESS(tdp, VADMIN, cred, td)) return (EPERM); return (0); } /* * Convert a component of a pathname into a pointer to a locked inode. * This is a very central and rather complicated routine. * If the filesystem is not maintained in a strict tree hierarchy, * this can result in a deadlock situation (see comments in code below). * * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending * on whether the name is to be looked up, created, renamed, or deleted. * When CREATE, RENAME, or DELETE is specified, information usable in * creating, renaming, or deleting a directory entry may be calculated. * If flag has LOCKPARENT or'ed into it and the target of the pathname * exists, lookup returns both the target and its parent directory locked. * When creating or renaming and LOCKPARENT is specified, the target may * not be ".". When deleting and LOCKPARENT is specified, the target may * be "."., but the caller must check to ensure it does an vrele and vput * instead of two vputs. * * This routine is actually used as VOP_CACHEDLOOKUP method, and the * filesystem employs the generic vfs_cache_lookup() as VOP_LOOKUP * method. * * vfs_cache_lookup() performs the following for us: * check that it is a directory * check accessibility of directory * check for modification attempts on read-only mounts * if name found in cache * if at end of path and deleting or creating * drop it * else * return name. * return VOP_CACHEDLOOKUP() * * Overall outline of ufs_lookup: * * search for name in directory, to found or notfound * notfound: * if creating, return locked directory, leaving info on available slots * else return error * found: * if at end of path and deleting, return information to allow delete * if at end of path and rewriting (RENAME and LOCKPARENT), lock target * inode and return info to allow rewrite * if not at end, add name to cache; if at end and neither creating * nor deleting, add name to cache */ int ufs_lookup(ap) struct vop_cachedlookup_args /* { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; } */ *ap; { return (ufs_lookup_ino(ap->a_dvp, ap->a_vpp, ap->a_cnp, NULL)); } int ufs_lookup_ino(struct vnode *vdp, struct vnode **vpp, struct componentname *cnp, ino_t *dd_ino) { struct inode *dp; /* inode for directory being searched */ struct buf *bp; /* a buffer of directory entries */ struct direct *ep; /* the current directory entry */ int entryoffsetinblock; /* offset of ep in bp's buffer */ enum {NONE, COMPACT, FOUND} slotstatus; doff_t slotoffset; /* offset of area with free space */ doff_t i_diroff; /* cached i_diroff value. */ doff_t i_offset; /* cached i_offset value. */ int slotsize; /* size of area at slotoffset */ int slotfreespace; /* amount of space free in slot */ int slotneeded; /* size of the entry we're seeking */ int numdirpasses; /* strategy for directory search */ doff_t endsearch; /* offset to end directory search */ doff_t prevoff; /* prev entry dp->i_offset */ struct vnode *pdp; /* saved dp during symlink work */ struct vnode *tdp; /* returned by VFS_VGET */ doff_t enduseful; /* pointer past last used dir slot */ u_long bmask; /* block offset mask */ int namlen, error; struct ucred *cred = cnp->cn_cred; int flags = cnp->cn_flags; int nameiop = cnp->cn_nameiop; ino_t ino, ino1; int ltype; if (vpp != NULL) *vpp = NULL; dp = VTOI(vdp); if (dp->i_effnlink == 0) return (ENOENT); /* * Create a vm object if vmiodirenable is enabled. * Alternatively we could call vnode_create_vobject * in VFS_VGET but we could end up creating objects * that are never used. */ - vnode_create_vobject(vdp, DIP(dp, i_size), cnp->cn_thread); + vnode_create_vobject(vdp, DIP(dp, i_size), curthread); bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1; #ifdef DEBUG_VFS_LOCKS /* * Assert that the directory vnode is locked, and locked * exclusively for the last component lookup for modifying * operations. * * The directory-modifying operations need to save * intermediate state in the inode between namei() call and * actual directory manipulations. See fields in the struct * inode marked as 'used during directory lookup'. We must * ensure that upgrade in namei() does not happen, since * upgrade might need to unlock vdp. If quotas are enabled, * getinoquota() also requires exclusive lock to modify inode. */ ASSERT_VOP_LOCKED(vdp, "ufs_lookup1"); if ((nameiop == CREATE || nameiop == DELETE || nameiop == RENAME) && (flags & (LOCKPARENT | ISLASTCN)) == (LOCKPARENT | ISLASTCN)) ASSERT_VOP_ELOCKED(vdp, "ufs_lookup2"); #endif restart: bp = NULL; slotoffset = -1; /* * We now have a segment name to search for, and a directory to search. * * Suppress search for slots unless creating * file and at end of pathname, in which case * we watch for a place to put the new file in * case it doesn't already exist. */ ino = 0; i_diroff = dp->i_diroff; slotstatus = FOUND; slotfreespace = slotsize = slotneeded = 0; if ((nameiop == CREATE || nameiop == RENAME) && (flags & ISLASTCN)) { slotstatus = NONE; slotneeded = DIRECTSIZ(cnp->cn_namelen); } #ifdef UFS_DIRHASH /* * Use dirhash for fast operations on large directories. The logic * to determine whether to hash the directory is contained within * ufsdirhash_build(); a zero return means that it decided to hash * this directory and it successfully built up the hash table. */ if (ufsdirhash_build(dp) == 0) { /* Look for a free slot if needed. */ enduseful = dp->i_size; if (slotstatus != FOUND) { slotoffset = ufsdirhash_findfree(dp, slotneeded, &slotsize); if (slotoffset >= 0) { slotstatus = COMPACT; enduseful = ufsdirhash_enduseful(dp); if (enduseful < 0) enduseful = dp->i_size; } } /* Look up the component. */ numdirpasses = 1; entryoffsetinblock = 0; /* silence compiler warning */ switch (ufsdirhash_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen, &i_offset, &bp, nameiop == DELETE ? &prevoff : NULL)) { case 0: ep = (struct direct *)((char *)bp->b_data + (i_offset & bmask)); goto foundentry; case ENOENT: i_offset = roundup2(dp->i_size, DIRBLKSIZ); goto notfound; default: /* Something failed; just do a linear search. */ break; } } #endif /* UFS_DIRHASH */ /* * If there is cached information on a previous search of * this directory, pick up where we last left off. * We cache only lookups as these are the most common * and have the greatest payoff. Caching CREATE has little * benefit as it usually must search the entire directory * to determine that the entry does not exist. Caching the * location of the last DELETE or RENAME has not reduced * profiling time and hence has been removed in the interest * of simplicity. */ if (nameiop != LOOKUP || i_diroff == 0 || i_diroff >= dp->i_size) { entryoffsetinblock = 0; i_offset = 0; numdirpasses = 1; } else { i_offset = i_diroff; if ((entryoffsetinblock = i_offset & bmask) && (error = UFS_BLKATOFF(vdp, (off_t)i_offset, NULL, &bp))) return (error); numdirpasses = 2; nchstats.ncs_2passes++; } prevoff = i_offset; endsearch = roundup2(dp->i_size, DIRBLKSIZ); enduseful = 0; searchloop: while (i_offset < endsearch) { /* * If necessary, get the next directory block. */ if ((i_offset & bmask) == 0) { if (bp != NULL) brelse(bp); error = UFS_BLKATOFF(vdp, (off_t)i_offset, NULL, &bp); if (error) return (error); entryoffsetinblock = 0; } /* * If still looking for a slot, and at a DIRBLKSIZE * boundary, have to start looking for free space again. */ if (slotstatus == NONE && (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) { slotoffset = -1; slotfreespace = 0; } /* * Get pointer to next entry. * Full validation checks are slow, so we only check * enough to insure forward progress through the * directory. Complete checks can be run by patching * "dirchk" to be true. */ ep = (struct direct *)((char *)bp->b_data + entryoffsetinblock); if (ep->d_reclen == 0 || ep->d_reclen > DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) || (dirchk && ufs_dirbadentry(vdp, ep, entryoffsetinblock))) { int i; ufs_dirbad(dp, i_offset, "mangled entry"); i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)); i_offset += i; entryoffsetinblock += i; continue; } /* * If an appropriate sized slot has not yet been found, * check to see if one is available. Also accumulate space * in the current block so that we can determine if * compaction is viable. */ if (slotstatus != FOUND) { int size = ep->d_reclen; if (ep->d_ino != 0) size -= DIRSIZ(OFSFMT(vdp), ep); if (size > 0) { if (size >= slotneeded) { slotstatus = FOUND; slotoffset = i_offset; slotsize = ep->d_reclen; } else if (slotstatus == NONE) { slotfreespace += size; if (slotoffset == -1) slotoffset = i_offset; if (slotfreespace >= slotneeded) { slotstatus = COMPACT; slotsize = i_offset + ep->d_reclen - slotoffset; } } } } /* * Check for a name match. */ if (ep->d_ino) { # if (BYTE_ORDER == LITTLE_ENDIAN) if (OFSFMT(vdp)) namlen = ep->d_type; else namlen = ep->d_namlen; # else namlen = ep->d_namlen; # endif if (namlen == cnp->cn_namelen && (cnp->cn_nameptr[0] == ep->d_name[0]) && !bcmp(cnp->cn_nameptr, ep->d_name, (unsigned)namlen)) { #ifdef UFS_DIRHASH foundentry: #endif /* * Save directory entry's inode number and * reclen in ndp->ni_ufs area, and release * directory buffer. */ if (!OFSFMT(vdp) && ep->d_type == DT_WHT) { slotstatus = FOUND; slotoffset = i_offset; slotsize = ep->d_reclen; enduseful = dp->i_size; cnp->cn_flags |= ISWHITEOUT; numdirpasses--; goto notfound; } ino = ep->d_ino; goto found; } } prevoff = i_offset; i_offset += ep->d_reclen; entryoffsetinblock += ep->d_reclen; if (ep->d_ino) enduseful = i_offset; } notfound: /* * If we started in the middle of the directory and failed * to find our target, we must check the beginning as well. */ if (numdirpasses == 2) { numdirpasses--; i_offset = 0; endsearch = i_diroff; goto searchloop; } if (bp != NULL) brelse(bp); /* * If creating, and at end of pathname and current * directory has not been removed, then can consider * allowing file to be created. */ if ((nameiop == CREATE || nameiop == RENAME || (nameiop == DELETE && (cnp->cn_flags & DOWHITEOUT) && (cnp->cn_flags & ISWHITEOUT))) && (flags & ISLASTCN) && dp->i_effnlink != 0) { /* * Access for write is interpreted as allowing * creation of files in the directory. * * XXX: Fix the comment above. */ if (flags & WILLBEDIR) - error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred, cnp->cn_thread); + error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred, curthread); else - error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread); + error = VOP_ACCESS(vdp, VWRITE, cred, curthread); if (error) return (error); /* * Return an indication of where the new directory * entry should be put. If we didn't find a slot, * then set dp->i_count to 0 indicating * that the new slot belongs at the end of the * directory. If we found a slot, then the new entry * can be put in the range from dp->i_offset to * dp->i_offset + dp->i_count. */ if (slotstatus == NONE) { SET_I_OFFSET(dp, roundup2(dp->i_size, DIRBLKSIZ)); SET_I_COUNT(dp, 0); enduseful = I_OFFSET(dp); } else if (nameiop == DELETE) { SET_I_OFFSET(dp, slotoffset); if ((I_OFFSET(dp) & (DIRBLKSIZ - 1)) == 0) SET_I_COUNT(dp, 0); else SET_I_COUNT(dp, I_OFFSET(dp) - prevoff); } else { SET_I_OFFSET(dp, slotoffset); SET_I_COUNT(dp, slotsize); if (enduseful < slotoffset + slotsize) enduseful = slotoffset + slotsize; } SET_I_ENDOFF(dp, roundup2(enduseful, DIRBLKSIZ)); /* * We return with the directory locked, so that * the parameters we set up above will still be * valid if we actually decide to do a direnter(). * We return ni_vp == NULL to indicate that the entry * does not currently exist; we leave a pointer to * the (locked) directory inode in ndp->ni_dvp. * The pathname buffer is saved so that the name * can be obtained later. * * NB - if the directory is unlocked, then this * information cannot be used. */ cnp->cn_flags |= SAVENAME; return (EJUSTRETURN); } /* * Insert name into cache (as non-existent) if appropriate. */ if ((cnp->cn_flags & MAKEENTRY) != 0) cache_enter(vdp, NULL, cnp); return (ENOENT); found: if (dd_ino != NULL) *dd_ino = ino; if (numdirpasses == 2) nchstats.ncs_pass2++; /* * Check that directory length properly reflects presence * of this entry. */ if (i_offset + DIRSIZ(OFSFMT(vdp), ep) > dp->i_size) { ufs_dirbad(dp, i_offset, "i_size too small"); dp->i_size = i_offset + DIRSIZ(OFSFMT(vdp), ep); DIP_SET(dp, i_size, dp->i_size); UFS_INODE_SET_FLAG(dp, IN_SIZEMOD | IN_CHANGE | IN_UPDATE); } brelse(bp); /* * Found component in pathname. * If the final component of path name, save information * in the cache as to where the entry was found. */ if ((flags & ISLASTCN) && nameiop == LOOKUP) dp->i_diroff = rounddown2(i_offset, DIRBLKSIZ); /* * If deleting, and at end of pathname, return * parameters which can be used to remove file. */ if (nameiop == DELETE && (flags & ISLASTCN)) { if (flags & LOCKPARENT) ASSERT_VOP_ELOCKED(vdp, __FUNCTION__); if (VOP_ISLOCKED(vdp) == LK_EXCLUSIVE) { /* * Return pointer to current entry in * dp->i_offset, and distance past previous * entry (if there is a previous entry in this * block) in dp->i_count. * * We shouldn't be setting these in the * WANTPARENT case (first lookup in rename()), but any * lookups that will result in directory changes will * overwrite these. */ SET_I_OFFSET(dp, i_offset); if ((I_OFFSET(dp) & (DIRBLKSIZ - 1)) == 0) SET_I_COUNT(dp, 0); else SET_I_COUNT(dp, I_OFFSET(dp) - prevoff); } if (dd_ino != NULL) return (0); /* * Save directory inode pointer in ndp->ni_dvp for * dirremove(). */ if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE, &tdp)) != 0) return (error); - error = ufs_delete_denied(vdp, tdp, cred, cnp->cn_thread); + error = ufs_delete_denied(vdp, tdp, cred, curthread); if (error) { vput(tdp); return (error); } if (dp->i_number == ino) { VREF(vdp); *vpp = vdp; vput(tdp); return (0); } *vpp = tdp; return (0); } /* * If rewriting (RENAME), return the inode and the * information required to rewrite the present directory * Must get inode of directory entry to verify it's a * regular file, or empty directory. */ if (nameiop == RENAME && (flags & ISLASTCN)) { if (flags & WILLBEDIR) - error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred, cnp->cn_thread); + error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred, curthread); else - error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread); + error = VOP_ACCESS(vdp, VWRITE, cred, curthread); if (error) return (error); /* * Careful about locking second inode. * This can only occur if the target is ".". */ SET_I_OFFSET(dp, i_offset); if (dp->i_number == ino) return (EISDIR); if (dd_ino != NULL) return (0); if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE, &tdp)) != 0) return (error); - error = ufs_delete_denied(vdp, tdp, cred, cnp->cn_thread); + error = ufs_delete_denied(vdp, tdp, cred, curthread); if (error) { vput(tdp); return (error); } #ifdef SunOS_doesnt_do_that /* * The only purpose of this check is to return the correct * error. Assume that we want to rename directory "a" * to a file "b", and that we have no ACL_WRITE_DATA on * a containing directory, but we _do_ have ACL_APPEND_DATA. * In that case, the VOP_ACCESS check above will return 0, * and the operation will fail with ENOTDIR instead * of EACCESS. */ if (tdp->v_type == VDIR) - error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred, cnp->cn_thread); + error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred, curthread); else - error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread); + error = VOP_ACCESS(vdp, VWRITE, cred, curthread); if (error) { vput(tdp); return (error); } #endif *vpp = tdp; cnp->cn_flags |= SAVENAME; return (0); } if (dd_ino != NULL) return (0); /* * Step through the translation in the name. We do not `vput' the * directory because we may need it again if a symbolic link * is relative to the current directory. Instead we save it * unlocked as "pdp". We must get the target inode before unlocking * the directory to insure that the inode will not be removed * before we get it. We prevent deadlock by always fetching * inodes from the root, moving down the directory tree. Thus * when following backward pointers ".." we must unlock the * parent directory before getting the requested directory. * There is a potential race condition here if both the current * and parent directories are removed before the VFS_VGET for the * inode associated with ".." returns. We hope that this occurs * infrequently since we cannot avoid this race condition without * implementing a sophisticated deadlock detection algorithm. * Note also that this simple deadlock detection scheme will not * work if the filesystem has any hard links other than ".." * that point backwards in the directory structure. */ pdp = vdp; if (flags & ISDOTDOT) { error = vn_vget_ino(pdp, ino, cnp->cn_lkflags, &tdp); if (error) return (error); /* * Recheck that ".." entry in the vdp directory points * to the inode we looked up before vdp lock was * dropped. */ error = ufs_lookup_ino(pdp, NULL, cnp, &ino1); if (error) { vput(tdp); return (error); } if (ino1 != ino) { vput(tdp); goto restart; } *vpp = tdp; } else if (dp->i_number == ino) { VREF(vdp); /* we want ourself, ie "." */ /* * When we lookup "." we still can be asked to lock it * differently. */ ltype = cnp->cn_lkflags & LK_TYPE_MASK; if (ltype != VOP_ISLOCKED(vdp)) { if (ltype == LK_EXCLUSIVE) vn_lock(vdp, LK_UPGRADE | LK_RETRY); else /* if (ltype == LK_SHARED) */ vn_lock(vdp, LK_DOWNGRADE | LK_RETRY); /* * Relock for the "." case may left us with * reclaimed vnode. */ if (VN_IS_DOOMED(vdp)) { vrele(vdp); return (ENOENT); } } *vpp = vdp; } else { error = VFS_VGET(pdp->v_mount, ino, cnp->cn_lkflags, &tdp); if (error == 0 && VTOI(tdp)->i_mode == 0) { vgone(tdp); vput(tdp); error = ENOENT; } if (error) return (error); *vpp = tdp; } /* * Insert name into cache if appropriate. */ if (cnp->cn_flags & MAKEENTRY) cache_enter(vdp, *vpp, cnp); return (0); } void ufs_dirbad(ip, offset, how) struct inode *ip; doff_t offset; char *how; { struct mount *mp; mp = ITOV(ip)->v_mount; if ((mp->mnt_flag & MNT_RDONLY) == 0) panic("ufs_dirbad: %s: bad dir ino %ju at offset %ld: %s", mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number, (long)offset, how); else (void)printf("%s: bad dir ino %ju at offset %ld: %s\n", mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number, (long)offset, how); } /* * Do consistency checking on a directory entry: * record length must be multiple of 4 * entry must fit in rest of its DIRBLKSIZ block * record must be large enough to contain entry * name is not longer than UFS_MAXNAMLEN * name must be as long as advertised, and null terminated */ int ufs_dirbadentry(dp, ep, entryoffsetinblock) struct vnode *dp; struct direct *ep; int entryoffsetinblock; { int i, namlen; # if (BYTE_ORDER == LITTLE_ENDIAN) if (OFSFMT(dp)) namlen = ep->d_type; else namlen = ep->d_namlen; # else namlen = ep->d_namlen; # endif if ((ep->d_reclen & 0x3) != 0 || ep->d_reclen > DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) || ep->d_reclen < DIRSIZ(OFSFMT(dp), ep) || namlen > UFS_MAXNAMLEN) { /*return (1); */ printf("First bad\n"); goto bad; } if (ep->d_ino == 0) return (0); for (i = 0; i < namlen; i++) if (ep->d_name[i] == '\0') { /*return (1); */ printf("Second bad\n"); goto bad; } if (ep->d_name[i]) goto bad; return (0); bad: return (1); } /* * Construct a new directory entry after a call to namei, using the * parameters that it left in the componentname argument cnp. The * argument ip is the inode to which the new directory entry will refer. */ void ufs_makedirentry(ip, cnp, newdirp) struct inode *ip; struct componentname *cnp; struct direct *newdirp; { u_int namelen; namelen = (unsigned)cnp->cn_namelen; KASSERT((cnp->cn_flags & SAVENAME) != 0, ("ufs_makedirentry: missing name")); KASSERT(namelen <= UFS_MAXNAMLEN, ("ufs_makedirentry: name too long")); newdirp->d_ino = ip->i_number; newdirp->d_namlen = namelen; /* Zero out after-name padding */ *(u_int32_t *)(&newdirp->d_name[namelen & ~(DIR_ROUNDUP - 1)]) = 0; bcopy(cnp->cn_nameptr, newdirp->d_name, namelen); if (!OFSFMT(ITOV(ip))) newdirp->d_type = IFTODT(ip->i_mode); else { newdirp->d_type = 0; # if (BYTE_ORDER == LITTLE_ENDIAN) { u_char tmp = newdirp->d_namlen; newdirp->d_namlen = newdirp->d_type; newdirp->d_type = tmp; } # endif } } /* * Write a directory entry after a call to namei, using the parameters * that it left in nameidata. The argument dirp is the new directory * entry contents. Dvp is a pointer to the directory to be written, * which was left locked by namei. Remaining parameters (dp->i_offset, * dp->i_count) indicate how the space for the new entry is to be obtained. * Non-null bp indicates that a directory is being created (for the * soft dependency code). */ int ufs_direnter(dvp, tvp, dirp, cnp, newdirbp) struct vnode *dvp; struct vnode *tvp; struct direct *dirp; struct componentname *cnp; struct buf *newdirbp; { struct ucred *cr; struct thread *td; int newentrysize; struct inode *dp; struct buf *bp; u_int dsize; struct direct *ep, *nep; u_int64_t old_isize; int error, ret, blkoff, loc, spacefree, flags, namlen; char *dirbuf; td = curthread; /* XXX */ cr = td->td_ucred; dp = VTOI(dvp); newentrysize = DIRSIZ(OFSFMT(dvp), dirp); if (I_COUNT(dp) == 0) { /* * If dp->i_count is 0, then namei could find no * space in the directory. Here, dp->i_offset will * be on a directory block boundary and we will write the * new entry into a fresh block. */ if (I_OFFSET(dp) & (DIRBLKSIZ - 1)) panic("ufs_direnter: newblk"); flags = BA_CLRBUF; if (!DOINGSOFTDEP(dvp) && !DOINGASYNC(dvp)) flags |= IO_SYNC; #ifdef QUOTA if ((error = getinoquota(dp)) != 0) { if (DOINGSOFTDEP(dvp) && newdirbp != NULL) bdwrite(newdirbp); return (error); } #endif old_isize = dp->i_size; vnode_pager_setsize(dvp, (u_long)I_OFFSET(dp) + DIRBLKSIZ); if ((error = UFS_BALLOC(dvp, (off_t)I_OFFSET(dp), DIRBLKSIZ, cr, flags, &bp)) != 0) { if (DOINGSOFTDEP(dvp) && newdirbp != NULL) bdwrite(newdirbp); vnode_pager_setsize(dvp, (u_long)old_isize); return (error); } dp->i_size = I_OFFSET(dp) + DIRBLKSIZ; DIP_SET(dp, i_size, dp->i_size); SET_I_ENDOFF(dp, dp->i_size); UFS_INODE_SET_FLAG(dp, IN_SIZEMOD | IN_CHANGE | IN_UPDATE); dirp->d_reclen = DIRBLKSIZ; blkoff = I_OFFSET(dp) & (VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_iosize - 1); bcopy((caddr_t)dirp, (caddr_t)bp->b_data + blkoff,newentrysize); #ifdef UFS_DIRHASH if (dp->i_dirhash != NULL) { ufsdirhash_newblk(dp, I_OFFSET(dp)); ufsdirhash_add(dp, dirp, I_OFFSET(dp)); ufsdirhash_checkblock(dp, (char *)bp->b_data + blkoff, I_OFFSET(dp)); } #endif if (DOINGSOFTDEP(dvp)) { /* * Ensure that the entire newly allocated block is a * valid directory so that future growth within the * block does not have to ensure that the block is * written before the inode. */ blkoff += DIRBLKSIZ; while (blkoff < bp->b_bcount) { ((struct direct *) (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ; blkoff += DIRBLKSIZ; } if (softdep_setup_directory_add(bp, dp, I_OFFSET(dp), dirp->d_ino, newdirbp, 1)) UFS_INODE_SET_FLAG(dp, IN_NEEDSYNC); if (newdirbp) bdwrite(newdirbp); bdwrite(bp); return (UFS_UPDATE(dvp, 0)); } if (DOINGASYNC(dvp)) { bdwrite(bp); return (UFS_UPDATE(dvp, 0)); } error = bwrite(bp); ret = UFS_UPDATE(dvp, 1); if (error == 0) return (ret); return (error); } /* * If dp->i_count is non-zero, then namei found space for the new * entry in the range dp->i_offset to dp->i_offset + dp->i_count * in the directory. To use this space, we may have to compact * the entries located there, by copying them together towards the * beginning of the block, leaving the free space in one usable * chunk at the end. */ /* * Increase size of directory if entry eats into new space. * This should never push the size past a new multiple of * DIRBLKSIZE. * * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN. */ if (I_OFFSET(dp) + I_COUNT(dp) > dp->i_size) { dp->i_size = I_OFFSET(dp) + I_COUNT(dp); DIP_SET(dp, i_size, dp->i_size); UFS_INODE_SET_FLAG(dp, IN_SIZEMOD | IN_MODIFIED); } /* * Get the block containing the space for the new directory entry. */ error = UFS_BLKATOFF(dvp, (off_t)I_OFFSET(dp), &dirbuf, &bp); if (error) { if (DOINGSOFTDEP(dvp) && newdirbp != NULL) bdwrite(newdirbp); return (error); } /* * Find space for the new entry. In the simple case, the entry at * offset base will have the space. If it does not, then namei * arranged that compacting the region dp->i_offset to * dp->i_offset + dp->i_count would yield the space. */ ep = (struct direct *)dirbuf; dsize = ep->d_ino ? DIRSIZ(OFSFMT(dvp), ep) : 0; spacefree = ep->d_reclen - dsize; for (loc = ep->d_reclen; loc < I_COUNT(dp); ) { nep = (struct direct *)(dirbuf + loc); /* Trim the existing slot (NB: dsize may be zero). */ ep->d_reclen = dsize; ep = (struct direct *)((char *)ep + dsize); /* Read nep->d_reclen now as the bcopy() may clobber it. */ loc += nep->d_reclen; if (nep->d_ino == 0) { /* * A mid-block unused entry. Such entries are * never created by the kernel, but fsck_ffs * can create them (and it doesn't fix them). * * Add up the free space, and initialise the * relocated entry since we don't bcopy it. */ spacefree += nep->d_reclen; ep->d_ino = 0; dsize = 0; continue; } dsize = DIRSIZ(OFSFMT(dvp), nep); spacefree += nep->d_reclen - dsize; #ifdef UFS_DIRHASH if (dp->i_dirhash != NULL) ufsdirhash_move(dp, nep, I_OFFSET(dp) + ((char *)nep - dirbuf), I_OFFSET(dp) + ((char *)ep - dirbuf)); #endif if (DOINGSOFTDEP(dvp)) softdep_change_directoryentry_offset(bp, dp, dirbuf, (caddr_t)nep, (caddr_t)ep, dsize); else bcopy((caddr_t)nep, (caddr_t)ep, dsize); } /* * Here, `ep' points to a directory entry containing `dsize' in-use * bytes followed by `spacefree' unused bytes. If ep->d_ino == 0, * then the entry is completely unused (dsize == 0). The value * of ep->d_reclen is always indeterminate. * * Update the pointer fields in the previous entry (if any), * copy in the new entry, and write out the block. */ # if (BYTE_ORDER == LITTLE_ENDIAN) if (OFSFMT(dvp)) namlen = ep->d_type; else namlen = ep->d_namlen; # else namlen = ep->d_namlen; # endif if (ep->d_ino == 0 || (ep->d_ino == UFS_WINO && namlen == dirp->d_namlen && bcmp(ep->d_name, dirp->d_name, dirp->d_namlen) == 0)) { if (spacefree + dsize < newentrysize) panic("ufs_direnter: compact1"); dirp->d_reclen = spacefree + dsize; } else { if (spacefree < newentrysize) panic("ufs_direnter: compact2"); dirp->d_reclen = spacefree; ep->d_reclen = dsize; ep = (struct direct *)((char *)ep + dsize); } #ifdef UFS_DIRHASH if (dp->i_dirhash != NULL && (ep->d_ino == 0 || dirp->d_reclen == spacefree)) ufsdirhash_add(dp, dirp, I_OFFSET(dp) + ((char *)ep - dirbuf)); #endif bcopy((caddr_t)dirp, (caddr_t)ep, (u_int)newentrysize); #ifdef UFS_DIRHASH if (dp->i_dirhash != NULL) ufsdirhash_checkblock(dp, dirbuf - (I_OFFSET(dp) & (DIRBLKSIZ - 1)), rounddown2(I_OFFSET(dp), DIRBLKSIZ)); #endif if (DOINGSOFTDEP(dvp)) { (void) softdep_setup_directory_add(bp, dp, I_OFFSET(dp) + (caddr_t)ep - dirbuf, dirp->d_ino, newdirbp, 0); if (newdirbp != NULL) bdwrite(newdirbp); bdwrite(bp); } else { if (DOINGASYNC(dvp)) { bdwrite(bp); error = 0; } else { error = bwrite(bp); } } /* * If all went well, and the directory can be shortened, * mark directory inode with the truncation request. */ UFS_INODE_SET_FLAG(dp, IN_CHANGE | IN_UPDATE | (error == 0 && I_ENDOFF(dp) != 0 && I_ENDOFF(dp) < dp->i_size ? IN_ENDOFF : 0)); return (error); } /* * Remove a directory entry after a call to namei, using * the parameters which it left in nameidata. The entry * dp->i_offset contains the offset into the directory of the * entry to be eliminated. The dp->i_count field contains the * size of the previous record in the directory. If this * is 0, the first entry is being deleted, so we need only * zero the inode number to mark the entry as free. If the * entry is not the first in the directory, we must reclaim * the space of the now empty record by adding the record size * to the size of the previous entry. */ int ufs_dirremove(dvp, ip, flags, isrmdir) struct vnode *dvp; struct inode *ip; int flags; int isrmdir; { struct inode *dp; struct direct *ep, *rep; struct buf *bp; off_t offset; int error; dp = VTOI(dvp); /* * Adjust the link count early so softdep can block if necessary. */ if (ip) { ip->i_effnlink--; UFS_INODE_SET_FLAG(ip, IN_CHANGE); if (DOINGSOFTDEP(dvp)) { softdep_setup_unlink(dp, ip); } else { ip->i_nlink--; DIP_SET(ip, i_nlink, ip->i_nlink); UFS_INODE_SET_FLAG(ip, IN_CHANGE); } } if (flags & DOWHITEOUT) offset = I_OFFSET(dp); else offset = I_OFFSET(dp) - I_COUNT(dp); if ((error = UFS_BLKATOFF(dvp, offset, (char **)&ep, &bp)) != 0) { if (ip) { ip->i_effnlink++; UFS_INODE_SET_FLAG(ip, IN_CHANGE); if (DOINGSOFTDEP(dvp)) { softdep_change_linkcnt(ip); } else { ip->i_nlink++; DIP_SET(ip, i_nlink, ip->i_nlink); UFS_INODE_SET_FLAG(ip, IN_CHANGE); } } return (error); } if (flags & DOWHITEOUT) { /* * Whiteout entry: set d_ino to UFS_WINO. */ ep->d_ino = UFS_WINO; ep->d_type = DT_WHT; goto out; } /* Set 'rep' to the entry being removed. */ if (I_COUNT(dp) == 0) rep = ep; else rep = (struct direct *)((char *)ep + ep->d_reclen); #ifdef UFS_DIRHASH /* * Remove the dirhash entry. This is complicated by the fact * that `ep' is the previous entry when dp->i_count != 0. */ if (dp->i_dirhash != NULL) ufsdirhash_remove(dp, rep, I_OFFSET(dp)); #endif if (ip && rep->d_ino != ip->i_number) panic("ufs_dirremove: ip %ju does not match dirent ino %ju\n", (uintmax_t)ip->i_number, (uintmax_t)rep->d_ino); /* * Zero out the file directory entry metadata to reduce disk * scavenging disclosure. */ bzero(&rep->d_name[0], rep->d_namlen); rep->d_namlen = 0; rep->d_type = 0; rep->d_ino = 0; if (I_COUNT(dp) != 0) { /* * Collapse new free space into previous entry. */ ep->d_reclen += rep->d_reclen; rep->d_reclen = 0; } #ifdef UFS_DIRHASH if (dp->i_dirhash != NULL) ufsdirhash_checkblock(dp, (char *)ep - ((I_OFFSET(dp) - I_COUNT(dp)) & (DIRBLKSIZ - 1)), rounddown2(I_OFFSET(dp), DIRBLKSIZ)); #endif out: error = 0; if (DOINGSOFTDEP(dvp)) { if (ip) softdep_setup_remove(bp, dp, ip, isrmdir); if (softdep_slowdown(dvp)) error = bwrite(bp); else bdwrite(bp); } else { if (flags & DOWHITEOUT) error = bwrite(bp); else if (DOINGASYNC(dvp)) bdwrite(bp); else error = bwrite(bp); } UFS_INODE_SET_FLAG(dp, IN_CHANGE | IN_UPDATE); /* * If the last named reference to a snapshot goes away, * drop its snapshot reference so that it will be reclaimed * when last open reference goes away. */ if (ip != NULL && (ip->i_flags & SF_SNAPSHOT) != 0 && ip->i_effnlink == 0) UFS_SNAPGONE(ip); return (error); } /* * Rewrite an existing directory entry to point at the inode * supplied. The parameters describing the directory entry are * set up by a call to namei. */ int ufs_dirrewrite(dp, oip, newinum, newtype, isrmdir) struct inode *dp, *oip; ino_t newinum; int newtype; int isrmdir; { struct buf *bp; struct direct *ep; struct vnode *vdp = ITOV(dp); int error; /* * Drop the link before we lock the buf so softdep can block if * necessary. */ oip->i_effnlink--; UFS_INODE_SET_FLAG(oip, IN_CHANGE); if (DOINGSOFTDEP(vdp)) { softdep_setup_unlink(dp, oip); } else { oip->i_nlink--; DIP_SET(oip, i_nlink, oip->i_nlink); UFS_INODE_SET_FLAG(oip, IN_CHANGE); } error = UFS_BLKATOFF(vdp, (off_t)I_OFFSET(dp), (char **)&ep, &bp); if (error == 0 && ep->d_namlen == 2 && ep->d_name[1] == '.' && ep->d_name[0] == '.' && ep->d_ino != oip->i_number) { brelse(bp); error = EIDRM; } if (error) { oip->i_effnlink++; UFS_INODE_SET_FLAG(oip, IN_CHANGE); if (DOINGSOFTDEP(vdp)) { softdep_change_linkcnt(oip); } else { oip->i_nlink++; DIP_SET(oip, i_nlink, oip->i_nlink); UFS_INODE_SET_FLAG(oip, IN_CHANGE); } return (error); } ep->d_ino = newinum; if (!OFSFMT(vdp)) ep->d_type = newtype; if (DOINGSOFTDEP(vdp)) { softdep_setup_directory_change(bp, dp, oip, newinum, isrmdir); bdwrite(bp); } else { if (DOINGASYNC(vdp)) { bdwrite(bp); error = 0; } else { error = bwrite(bp); } } UFS_INODE_SET_FLAG(dp, IN_CHANGE | IN_UPDATE); /* * If the last named reference to a snapshot goes away, * drop its snapshot reference so that it will be reclaimed * when last open reference goes away. */ if ((oip->i_flags & SF_SNAPSHOT) != 0 && oip->i_effnlink == 0) UFS_SNAPGONE(oip); return (error); } /* * Check if a directory is empty or not. * Inode supplied must be locked. * * Using a struct dirtemplate here is not precisely * what we want, but better than using a struct direct. * * NB: does not handle corrupted directories. */ int ufs_dirempty(ip, parentino, cred) struct inode *ip; ino_t parentino; struct ucred *cred; { doff_t off; struct dirtemplate dbuf; struct direct *dp = (struct direct *)&dbuf; int error, namlen; ssize_t count; #define MINDIRSIZ (sizeof (struct dirtemplate) / 2) for (off = 0; off < ip->i_size; off += dp->d_reclen) { error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, off, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, &count, (struct thread *)0); /* * Since we read MINDIRSIZ, residual must * be 0 unless we're at end of file. */ if (error || count != 0) return (0); /* avoid infinite loops */ if (dp->d_reclen == 0) return (0); /* skip empty entries */ if (dp->d_ino == 0 || dp->d_ino == UFS_WINO) continue; /* accept only "." and ".." */ # if (BYTE_ORDER == LITTLE_ENDIAN) if (OFSFMT(ITOV(ip))) namlen = dp->d_type; else namlen = dp->d_namlen; # else namlen = dp->d_namlen; # endif if (namlen > 2) return (0); if (dp->d_name[0] != '.') return (0); /* * At this point namlen must be 1 or 2. * 1 implies ".", 2 implies ".." if second * char is also "." */ if (namlen == 1 && dp->d_ino == ip->i_number) continue; if (dp->d_name[1] == '.' && dp->d_ino == parentino) continue; return (0); } return (1); } static int ufs_dir_dd_ino(struct vnode *vp, struct ucred *cred, ino_t *dd_ino, struct vnode **dd_vp) { struct dirtemplate dirbuf; struct vnode *ddvp; int error, namlen; ASSERT_VOP_LOCKED(vp, "ufs_dir_dd_ino"); *dd_vp = NULL; if (vp->v_type != VDIR) return (ENOTDIR); /* * First check to see if we have it in the name cache. */ if ((ddvp = vn_dir_dd_ino(vp)) != NULL) { KASSERT(ddvp->v_mount == vp->v_mount, ("ufs_dir_dd_ino: Unexpected mount point crossing")); *dd_ino = VTOI(ddvp)->i_number; *dd_vp = ddvp; return (0); } /* * Have to read the directory. */ error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf, sizeof (struct dirtemplate), (off_t)0, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, NULL, NULL); if (error != 0) return (error); #if (BYTE_ORDER == LITTLE_ENDIAN) if (OFSFMT(vp)) namlen = dirbuf.dotdot_type; else namlen = dirbuf.dotdot_namlen; #else namlen = dirbuf.dotdot_namlen; #endif if (namlen != 2 || dirbuf.dotdot_name[0] != '.' || dirbuf.dotdot_name[1] != '.') return (ENOTDIR); *dd_ino = dirbuf.dotdot_ino; return (0); } /* * Check if source directory is in the path of the target directory. */ int ufs_checkpath(ino_t source_ino, ino_t parent_ino, struct inode *target, struct ucred *cred, ino_t *wait_ino) { struct mount *mp; struct vnode *tvp, *vp, *vp1; int error; ino_t dd_ino; vp = tvp = ITOV(target); mp = vp->v_mount; *wait_ino = 0; sx_assert(&VFSTOUFS(mp)->um_checkpath_lock, SA_XLOCKED); if (target->i_number == source_ino) return (EEXIST); if (target->i_number == parent_ino) return (0); if (target->i_number == UFS_ROOTINO) return (0); for (;;) { error = ufs_dir_dd_ino(vp, cred, &dd_ino, &vp1); if (error != 0) break; if (dd_ino == source_ino) { error = EINVAL; break; } if (dd_ino == UFS_ROOTINO) break; if (dd_ino == parent_ino) break; if (vp1 == NULL) { error = VFS_VGET(mp, dd_ino, LK_SHARED | LK_NOWAIT, &vp1); if (error != 0) { *wait_ino = dd_ino; break; } } KASSERT(dd_ino == VTOI(vp1)->i_number, ("directory %ju reparented\n", (uintmax_t)VTOI(vp1)->i_number)); if (vp != tvp) vput(vp); vp = vp1; } if (error == ENOTDIR) panic("checkpath: .. not a directory\n"); if (vp1 != NULL) vput(vp1); if (vp != tvp) vput(vp); return (error); } #ifdef DIAGNOSTIC static void ufs_assert_inode_offset_owner(struct inode *ip, struct iown_tracker *tr, const char *name, const char *file, int line) { char msg[128]; snprintf(msg, sizeof(msg), "at %s@%d", file, line); ASSERT_VOP_ELOCKED(ITOV(ip), msg); MPASS((ip->i_mode & IFMT) == IFDIR); if (curthread == tr->tr_owner && ip->i_lock_gen == tr->tr_gen) return; printf("locked at\n"); stack_print(&tr->tr_st); printf("unlocked at\n"); stack_print(&tr->tr_unlock); panic("%s ip %p %jd offset owner %p %d gen %d " "curthread %p %d gen %d at %s@%d\n", name, ip, (uintmax_t)ip->i_number, tr->tr_owner, tr->tr_owner->td_tid, tr->tr_gen, curthread, curthread->td_tid, ip->i_lock_gen, file, line); } static void ufs_set_inode_offset_owner(struct inode *ip, struct iown_tracker *tr, const char *file, int line) { char msg[128]; snprintf(msg, sizeof(msg), "at %s@%d", file, line); ASSERT_VOP_ELOCKED(ITOV(ip), msg); MPASS((ip->i_mode & IFMT) == IFDIR); tr->tr_owner = curthread; tr->tr_gen = ip->i_lock_gen; stack_save(&tr->tr_st); } static void ufs_init_one_tracker(struct iown_tracker *tr) { tr->tr_owner = NULL; stack_zero(&tr->tr_st); } void ufs_init_trackers(struct inode *ip) { ufs_init_one_tracker(&ip->i_offset_tracker); ufs_init_one_tracker(&ip->i_count_tracker); ufs_init_one_tracker(&ip->i_endoff_tracker); } void ufs_unlock_tracker(struct inode *ip) { if (ip->i_count_tracker.tr_gen == ip->i_lock_gen) stack_save(&ip->i_count_tracker.tr_unlock); if (ip->i_offset_tracker.tr_gen == ip->i_lock_gen) stack_save(&ip->i_offset_tracker.tr_unlock); if (ip->i_endoff_tracker.tr_gen == ip->i_lock_gen) stack_save(&ip->i_endoff_tracker.tr_unlock); ip->i_lock_gen++; } doff_t ufs_get_i_offset(struct inode *ip, const char *file, int line) { ufs_assert_inode_offset_owner(ip, &ip->i_offset_tracker, "i_offset", file, line); return (ip->i_offset); } void ufs_set_i_offset(struct inode *ip, doff_t off, const char *file, int line) { ufs_set_inode_offset_owner(ip, &ip->i_offset_tracker, file, line); ip->i_offset = off; } int32_t ufs_get_i_count(struct inode *ip, const char *file, int line) { ufs_assert_inode_offset_owner(ip, &ip->i_count_tracker, "i_count", file, line); return (ip->i_count); } void ufs_set_i_count(struct inode *ip, int32_t cnt, const char *file, int line) { ufs_set_inode_offset_owner(ip, &ip->i_count_tracker, file, line); ip->i_count = cnt; } doff_t ufs_get_i_endoff(struct inode *ip, const char *file, int line) { ufs_assert_inode_offset_owner(ip, &ip->i_endoff_tracker, "i_endoff", file, line); return (ip->i_endoff); } void ufs_set_i_endoff(struct inode *ip, doff_t off, const char *file, int line) { ufs_set_inode_offset_owner(ip, &ip->i_endoff_tracker, file, line); ip->i_endoff = off; } #endif diff --git a/sys/ufs/ufs/ufs_vnops.c b/sys/ufs/ufs/ufs_vnops.c index e962ef8ca48c..02dada0c63a2 100644 --- a/sys/ufs/ufs/ufs_vnops.c +++ b/sys/ufs/ufs/ufs_vnops.c @@ -1,3018 +1,3018 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1989, 1993, 1995 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)ufs_vnops.c 8.27 (Berkeley) 5/27/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_quota.h" #include "opt_suiddir.h" #include "opt_ufs.h" #include "opt_ffs.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* XXX */ #include #include #include #include #include #include #include #include #include #ifdef UFS_DIRHASH #include #endif #ifdef UFS_GJOURNAL #include FEATURE(ufs_gjournal, "Journaling support through GEOM for UFS"); #endif #ifdef QUOTA FEATURE(ufs_quota, "UFS disk quotas support"); FEATURE(ufs_quota64, "64bit UFS disk quotas support"); #endif #ifdef SUIDDIR FEATURE(suiddir, "Give all new files in directory the same ownership as the directory"); #endif VFS_SMR_DECLARE; #include static vop_accessx_t ufs_accessx; static vop_fplookup_vexec_t ufs_fplookup_vexec; static int ufs_chmod(struct vnode *, int, struct ucred *, struct thread *); static int ufs_chown(struct vnode *, uid_t, gid_t, struct ucred *, struct thread *); static vop_close_t ufs_close; static vop_create_t ufs_create; static vop_stat_t ufs_stat; static vop_getattr_t ufs_getattr; static vop_ioctl_t ufs_ioctl; static vop_link_t ufs_link; static int ufs_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *, const char *); static vop_mmapped_t ufs_mmapped; static vop_mkdir_t ufs_mkdir; static vop_mknod_t ufs_mknod; static vop_open_t ufs_open; static vop_pathconf_t ufs_pathconf; static vop_print_t ufs_print; static vop_readlink_t ufs_readlink; static vop_remove_t ufs_remove; static vop_rename_t ufs_rename; static vop_rmdir_t ufs_rmdir; static vop_setattr_t ufs_setattr; static vop_strategy_t ufs_strategy; static vop_symlink_t ufs_symlink; static vop_whiteout_t ufs_whiteout; static vop_close_t ufsfifo_close; SYSCTL_NODE(_vfs, OID_AUTO, ufs, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "UFS filesystem"); /* * A virgin directory (no blushing please). */ static struct dirtemplate mastertemplate = { 0, 12, DT_DIR, 1, ".", 0, DIRBLKSIZ - 12, DT_DIR, 2, ".." }; static struct odirtemplate omastertemplate = { 0, 12, 1, ".", 0, DIRBLKSIZ - 12, 2, ".." }; static void ufs_itimes_locked(struct vnode *vp) { struct inode *ip; struct timespec ts; ASSERT_VI_LOCKED(vp, __func__); ip = VTOI(vp); if (UFS_RDONLY(ip)) goto out; if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0) return; if ((vp->v_type == VBLK || vp->v_type == VCHR) && !DOINGSOFTDEP(vp)) UFS_INODE_SET_FLAG(ip, IN_LAZYMOD); else if (((vp->v_mount->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND)) == 0) || (ip->i_flag & (IN_CHANGE | IN_UPDATE))) UFS_INODE_SET_FLAG(ip, IN_MODIFIED); else if (ip->i_flag & IN_ACCESS) UFS_INODE_SET_FLAG(ip, IN_LAZYACCESS); vfs_timestamp(&ts); if (ip->i_flag & IN_ACCESS) { DIP_SET(ip, i_atime, ts.tv_sec); DIP_SET(ip, i_atimensec, ts.tv_nsec); } if (ip->i_flag & IN_UPDATE) { DIP_SET(ip, i_mtime, ts.tv_sec); DIP_SET(ip, i_mtimensec, ts.tv_nsec); } if (ip->i_flag & IN_CHANGE) { DIP_SET(ip, i_ctime, ts.tv_sec); DIP_SET(ip, i_ctimensec, ts.tv_nsec); DIP_SET(ip, i_modrev, DIP(ip, i_modrev) + 1); } out: ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE); } void ufs_itimes(struct vnode *vp) { VI_LOCK(vp); ufs_itimes_locked(vp); VI_UNLOCK(vp); } /* * Create a regular file */ static int ufs_create(ap) struct vop_create_args /* { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; struct vattr *a_vap; } */ *ap; { int error; error = ufs_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode), ap->a_dvp, ap->a_vpp, ap->a_cnp, "ufs_create"); if (error != 0) return (error); if ((ap->a_cnp->cn_flags & MAKEENTRY) != 0) cache_enter(ap->a_dvp, *ap->a_vpp, ap->a_cnp); return (0); } /* * Mknod vnode call */ /* ARGSUSED */ static int ufs_mknod(ap) struct vop_mknod_args /* { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; struct vattr *a_vap; } */ *ap; { struct vattr *vap = ap->a_vap; struct vnode **vpp = ap->a_vpp; struct inode *ip; ino_t ino; int error; error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), ap->a_dvp, vpp, ap->a_cnp, "ufs_mknod"); if (error) return (error); ip = VTOI(*vpp); UFS_INODE_SET_FLAG(ip, IN_ACCESS | IN_CHANGE | IN_UPDATE); if (vap->va_rdev != VNOVAL) { /* * Want to be able to use this to make badblock * inodes, so don't truncate the dev number. */ DIP_SET(ip, i_rdev, vap->va_rdev); } /* * Remove inode, then reload it through VFS_VGET(). This is * needed to do further inode initialization, for instance * fifo, which was too early for VFS_VGET() done as part of * UFS_VALLOC(). */ (*vpp)->v_type = VNON; ino = ip->i_number; /* Save this before vgone() invalidates ip. */ vgone(*vpp); vput(*vpp); error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp); if (error) { *vpp = NULL; return (error); } return (0); } /* * Open called. */ /* ARGSUSED */ static int ufs_open(struct vop_open_args *ap) { struct vnode *vp = ap->a_vp; struct inode *ip; if (vp->v_type == VCHR || vp->v_type == VBLK) return (EOPNOTSUPP); ip = VTOI(vp); vnode_create_vobject(vp, DIP(ip, i_size), ap->a_td); if (vp->v_type == VREG && (vn_irflag_read(vp) & VIRF_PGREAD) == 0) { vn_irflag_set_cond(vp, VIRF_PGREAD); } /* * Files marked append-only must be opened for appending. */ if ((ip->i_flags & APPEND) && (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE) return (EPERM); return (0); } /* * Close called. * * Update the times on the inode. */ /* ARGSUSED */ static int ufs_close(ap) struct vop_close_args /* { struct vnode *a_vp; int a_fflag; struct ucred *a_cred; struct thread *a_td; } */ *ap; { struct vnode *vp = ap->a_vp; int usecount; VI_LOCK(vp); usecount = vp->v_usecount; if (usecount > 1) ufs_itimes_locked(vp); VI_UNLOCK(vp); return (0); } static int ufs_accessx(ap) struct vop_accessx_args /* { struct vnode *a_vp; accmode_t a_accmode; struct ucred *a_cred; struct thread *a_td; } */ *ap; { struct vnode *vp = ap->a_vp; struct inode *ip = VTOI(vp); accmode_t accmode = ap->a_accmode; int error; #ifdef UFS_ACL struct acl *acl; acl_type_t type; #endif /* * Disallow write attempts on read-only filesystems; * unless the file is a socket, fifo, or a block or * character device resident on the filesystem. */ if (accmode & VMODIFY_PERMS) { switch (vp->v_type) { case VDIR: case VLNK: case VREG: if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); #ifdef QUOTA /* * Inode is accounted in the quotas only if struct * dquot is attached to it. VOP_ACCESS() is called * from vn_open_cred() and provides a convenient * point to call getinoquota(). The lock mode is * exclusive when the file is opening for write. */ if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE) { error = getinoquota(ip); if (error != 0) return (error); } #endif break; default: break; } } /* * If immutable bit set, nobody gets to write it. "& ~VADMIN_PERMS" * permits the owner of the file to remove the IMMUTABLE flag. */ if ((accmode & (VMODIFY_PERMS & ~VADMIN_PERMS)) && (ip->i_flags & (IMMUTABLE | SF_SNAPSHOT))) return (EPERM); #ifdef UFS_ACL if ((vp->v_mount->mnt_flag & (MNT_ACLS | MNT_NFS4ACLS)) != 0) { if (vp->v_mount->mnt_flag & MNT_NFS4ACLS) type = ACL_TYPE_NFS4; else type = ACL_TYPE_ACCESS; acl = acl_alloc(M_WAITOK); if (type == ACL_TYPE_NFS4) error = ufs_getacl_nfs4_internal(vp, acl, ap->a_td); else error = VOP_GETACL(vp, type, acl, ap->a_cred, ap->a_td); switch (error) { case 0: if (type == ACL_TYPE_NFS4) { error = vaccess_acl_nfs4(vp->v_type, ip->i_uid, ip->i_gid, acl, accmode, ap->a_cred); } else { error = vfs_unixify_accmode(&accmode); if (error == 0) error = vaccess_acl_posix1e(vp->v_type, ip->i_uid, ip->i_gid, acl, accmode, ap->a_cred); } break; default: if (error != EOPNOTSUPP) printf( "ufs_accessx(): Error retrieving ACL on object (%d).\n", error); /* * XXX: Fall back until debugged. Should * eventually possibly log an error, and return * EPERM for safety. */ error = vfs_unixify_accmode(&accmode); if (error == 0) error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid, accmode, ap->a_cred); } acl_free(acl); return (error); } #endif /* !UFS_ACL */ error = vfs_unixify_accmode(&accmode); if (error == 0) error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid, accmode, ap->a_cred); return (error); } /* * VOP_FPLOOKUP_VEXEC routines are subject to special circumstances, see * the comment above cache_fplookup for details. */ static int ufs_fplookup_vexec(ap) struct vop_fplookup_vexec_args /* { struct vnode *a_vp; struct ucred *a_cred; struct thread *a_td; } */ *ap; { struct vnode *vp; struct inode *ip; struct ucred *cred; mode_t all_x, mode; vp = ap->a_vp; ip = VTOI_SMR(vp); if (__predict_false(ip == NULL)) return (EAGAIN); /* * XXX ACL race * * ACLs are not supported and UFS clears/sets this flag on mount and * remount. However, we may still be racing with seeing them and there * is no provision to make sure they were accounted for. This matches * the behavior of the locked case, since the lookup there is also * racy: mount takes no measures to block anyone from progressing. */ all_x = S_IXUSR | S_IXGRP | S_IXOTH; mode = atomic_load_short(&ip->i_mode); if (__predict_true((mode & all_x) == all_x)) return (0); cred = ap->a_cred; return (vaccess_vexec_smr(mode, ip->i_uid, ip->i_gid, cred)); } /* ARGSUSED */ static int ufs_stat(struct vop_stat_args *ap) { struct vnode *vp = ap->a_vp; struct inode *ip = VTOI(vp); struct stat *sb = ap->a_sb; int error; error = vop_stat_helper_pre(ap); if (__predict_false(error)) return (error); VI_LOCK(vp); ufs_itimes_locked(vp); if (I_IS_UFS1(ip)) { sb->st_atim.tv_sec = ip->i_din1->di_atime; sb->st_atim.tv_nsec = ip->i_din1->di_atimensec; } else { sb->st_atim.tv_sec = ip->i_din2->di_atime; sb->st_atim.tv_nsec = ip->i_din2->di_atimensec; } VI_UNLOCK(vp); sb->st_dev = dev2udev(ITOUMP(ip)->um_dev); sb->st_ino = ip->i_number; sb->st_mode = (ip->i_mode & ~IFMT) | VTTOIF(vp->v_type); sb->st_nlink = ip->i_effnlink; sb->st_uid = ip->i_uid; sb->st_gid = ip->i_gid; if (I_IS_UFS1(ip)) { sb->st_rdev = ip->i_din1->di_rdev; sb->st_size = ip->i_din1->di_size; sb->st_mtim.tv_sec = ip->i_din1->di_mtime; sb->st_mtim.tv_nsec = ip->i_din1->di_mtimensec; sb->st_ctim.tv_sec = ip->i_din1->di_ctime; sb->st_ctim.tv_nsec = ip->i_din1->di_ctimensec; sb->st_birthtim.tv_sec = -1; sb->st_birthtim.tv_nsec = 0; sb->st_blocks = dbtob((u_quad_t)ip->i_din1->di_blocks) / S_BLKSIZE; } else { sb->st_rdev = ip->i_din2->di_rdev; sb->st_size = ip->i_din2->di_size; sb->st_mtim.tv_sec = ip->i_din2->di_mtime; sb->st_mtim.tv_nsec = ip->i_din2->di_mtimensec; sb->st_ctim.tv_sec = ip->i_din2->di_ctime; sb->st_ctim.tv_nsec = ip->i_din2->di_ctimensec; sb->st_birthtim.tv_sec = ip->i_din2->di_birthtime; sb->st_birthtim.tv_nsec = ip->i_din2->di_birthnsec; sb->st_blocks = dbtob((u_quad_t)ip->i_din2->di_blocks) / S_BLKSIZE; } sb->st_blksize = max(PAGE_SIZE, vp->v_mount->mnt_stat.f_iosize); sb->st_flags = ip->i_flags; sb->st_gen = ip->i_gen; return (vop_stat_helper_post(ap, error)); } /* ARGSUSED */ static int ufs_getattr(ap) struct vop_getattr_args /* { struct vnode *a_vp; struct vattr *a_vap; struct ucred *a_cred; } */ *ap; { struct vnode *vp = ap->a_vp; struct inode *ip = VTOI(vp); struct vattr *vap = ap->a_vap; VI_LOCK(vp); ufs_itimes_locked(vp); if (I_IS_UFS1(ip)) { vap->va_atime.tv_sec = ip->i_din1->di_atime; vap->va_atime.tv_nsec = ip->i_din1->di_atimensec; } else { vap->va_atime.tv_sec = ip->i_din2->di_atime; vap->va_atime.tv_nsec = ip->i_din2->di_atimensec; } VI_UNLOCK(vp); /* * Copy from inode table */ vap->va_fsid = dev2udev(ITOUMP(ip)->um_dev); vap->va_fileid = ip->i_number; vap->va_mode = ip->i_mode & ~IFMT; vap->va_nlink = ip->i_effnlink; vap->va_uid = ip->i_uid; vap->va_gid = ip->i_gid; if (I_IS_UFS1(ip)) { vap->va_rdev = ip->i_din1->di_rdev; vap->va_size = ip->i_din1->di_size; vap->va_mtime.tv_sec = ip->i_din1->di_mtime; vap->va_mtime.tv_nsec = ip->i_din1->di_mtimensec; vap->va_ctime.tv_sec = ip->i_din1->di_ctime; vap->va_ctime.tv_nsec = ip->i_din1->di_ctimensec; vap->va_bytes = dbtob((u_quad_t)ip->i_din1->di_blocks); vap->va_filerev = ip->i_din1->di_modrev; } else { vap->va_rdev = ip->i_din2->di_rdev; vap->va_size = ip->i_din2->di_size; vap->va_mtime.tv_sec = ip->i_din2->di_mtime; vap->va_mtime.tv_nsec = ip->i_din2->di_mtimensec; vap->va_ctime.tv_sec = ip->i_din2->di_ctime; vap->va_ctime.tv_nsec = ip->i_din2->di_ctimensec; vap->va_birthtime.tv_sec = ip->i_din2->di_birthtime; vap->va_birthtime.tv_nsec = ip->i_din2->di_birthnsec; vap->va_bytes = dbtob((u_quad_t)ip->i_din2->di_blocks); vap->va_filerev = ip->i_din2->di_modrev; } vap->va_flags = ip->i_flags; vap->va_gen = ip->i_gen; vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; vap->va_type = IFTOVT(ip->i_mode); return (0); } /* * Set attribute vnode op. called from several syscalls */ static int ufs_setattr(ap) struct vop_setattr_args /* { struct vnode *a_vp; struct vattr *a_vap; struct ucred *a_cred; } */ *ap; { struct vattr *vap = ap->a_vap; struct vnode *vp = ap->a_vp; struct inode *ip = VTOI(vp); struct ucred *cred = ap->a_cred; struct thread *td = curthread; int error; /* * Check for unsettable attributes. */ if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) || (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) || (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) || ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) { return (EINVAL); } if (vap->va_flags != VNOVAL) { if ((vap->va_flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE | SF_NOUNLINK | SF_SNAPSHOT | UF_APPEND | UF_ARCHIVE | UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP | UF_NOUNLINK | UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE | UF_SPARSE | UF_SYSTEM)) != 0) return (EOPNOTSUPP); if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); /* * Callers may only modify the file flags on objects they * have VADMIN rights for. */ if ((error = VOP_ACCESS(vp, VADMIN, cred, td))) return (error); /* * Unprivileged processes are not permitted to unset system * flags, or modify flags if any system flags are set. * Privileged non-jail processes may not modify system flags * if securelevel > 0 and any existing system flags are set. * Privileged jail processes behave like privileged non-jail * processes if the PR_ALLOW_CHFLAGS permission bit is set; * otherwise, they behave like unprivileged processes. */ if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS)) { if (ip->i_flags & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) { error = securelevel_gt(cred, 0); if (error) return (error); } /* The snapshot flag cannot be toggled. */ if ((vap->va_flags ^ ip->i_flags) & SF_SNAPSHOT) return (EPERM); } else { if (ip->i_flags & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) || ((vap->va_flags ^ ip->i_flags) & SF_SETTABLE)) return (EPERM); } ip->i_flags = vap->va_flags; DIP_SET(ip, i_flags, vap->va_flags); UFS_INODE_SET_FLAG(ip, IN_CHANGE); error = UFS_UPDATE(vp, 0); if (ip->i_flags & (IMMUTABLE | APPEND)) return (error); } /* * If immutable or append, no one can change any of its attributes * except the ones already handled (in some cases, file flags * including the immutability flags themselves for the superuser). */ if (ip->i_flags & (IMMUTABLE | APPEND)) return (EPERM); /* * Go through the fields and update iff not VNOVAL. */ if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) { if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); if ((error = ufs_chown(vp, vap->va_uid, vap->va_gid, cred, td)) != 0) return (error); } if (vap->va_size != VNOVAL) { /* * XXX most of the following special cases should be in * callers instead of in N filesystems. The VDIR check * mostly already is. */ switch (vp->v_type) { case VDIR: return (EISDIR); case VLNK: case VREG: /* * Truncation should have an effect in these cases. * Disallow it if the filesystem is read-only or * the file is being snapshotted. */ if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); if ((ip->i_flags & SF_SNAPSHOT) != 0) return (EPERM); break; default: /* * According to POSIX, the result is unspecified * for file types other than regular files, * directories and shared memory objects. We * don't support shared memory objects in the file * system, and have dubious support for truncating * symlinks. Just ignore the request in other cases. */ return (0); } if ((error = UFS_TRUNCATE(vp, vap->va_size, IO_NORMAL | ((vap->va_vaflags & VA_SYNC) != 0 ? IO_SYNC : 0), cred)) != 0) return (error); } if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL || vap->va_birthtime.tv_sec != VNOVAL) { if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); if ((ip->i_flags & SF_SNAPSHOT) != 0) return (EPERM); error = vn_utimes_perm(vp, vap, cred, td); if (error != 0) return (error); UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_MODIFIED); if (vap->va_atime.tv_sec != VNOVAL) { ip->i_flag &= ~IN_ACCESS; DIP_SET(ip, i_atime, vap->va_atime.tv_sec); DIP_SET(ip, i_atimensec, vap->va_atime.tv_nsec); } if (vap->va_mtime.tv_sec != VNOVAL) { ip->i_flag &= ~IN_UPDATE; DIP_SET(ip, i_mtime, vap->va_mtime.tv_sec); DIP_SET(ip, i_mtimensec, vap->va_mtime.tv_nsec); } if (vap->va_birthtime.tv_sec != VNOVAL && I_IS_UFS2(ip)) { ip->i_din2->di_birthtime = vap->va_birthtime.tv_sec; ip->i_din2->di_birthnsec = vap->va_birthtime.tv_nsec; } error = UFS_UPDATE(vp, 0); if (error) return (error); } error = 0; if (vap->va_mode != (mode_t)VNOVAL) { if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); if ((ip->i_flags & SF_SNAPSHOT) != 0 && (vap->va_mode & (S_IXUSR | S_IWUSR | S_IXGRP | S_IWGRP | S_IXOTH | S_IWOTH))) return (EPERM); error = ufs_chmod(vp, (int)vap->va_mode, cred, td); } return (error); } #ifdef UFS_ACL static int ufs_update_nfs4_acl_after_mode_change(struct vnode *vp, int mode, int file_owner_id, struct ucred *cred, struct thread *td) { int error; struct acl *aclp; aclp = acl_alloc(M_WAITOK); error = ufs_getacl_nfs4_internal(vp, aclp, td); /* * We don't have to handle EOPNOTSUPP here, as the filesystem claims * it supports ACLs. */ if (error) goto out; acl_nfs4_sync_acl_from_mode(aclp, mode, file_owner_id); error = ufs_setacl_nfs4_internal(vp, aclp, td); out: acl_free(aclp); return (error); } #endif /* UFS_ACL */ static int ufs_mmapped(ap) struct vop_mmapped_args /* { struct vnode *a_vp; } */ *ap; { struct vnode *vp; struct inode *ip; struct mount *mp; vp = ap->a_vp; ip = VTOI(vp); mp = vp->v_mount; if ((mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) UFS_INODE_SET_FLAG_SHARED(ip, IN_ACCESS); /* * XXXKIB No UFS_UPDATE(ap->a_vp, 0) there. */ return (0); } /* * Change the mode on a file. * Inode must be locked before calling. */ static int ufs_chmod(vp, mode, cred, td) struct vnode *vp; int mode; struct ucred *cred; struct thread *td; { struct inode *ip = VTOI(vp); int newmode, error; /* * To modify the permissions on a file, must possess VADMIN * for that file. */ if ((error = VOP_ACCESSX(vp, VWRITE_ACL, cred, td))) return (error); /* * Privileged processes may set the sticky bit on non-directories, * as well as set the setgid bit on a file with a group that the * process is not a member of. Both of these are allowed in * jail(8). */ if (vp->v_type != VDIR && (mode & S_ISTXT)) { if (priv_check_cred(cred, PRIV_VFS_STICKYFILE)) return (EFTYPE); } if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) { error = priv_check_cred(cred, PRIV_VFS_SETGID); if (error) return (error); } /* * Deny setting setuid if we are not the file owner. */ if ((mode & ISUID) && ip->i_uid != cred->cr_uid) { error = priv_check_cred(cred, PRIV_VFS_ADMIN); if (error) return (error); } newmode = ip->i_mode & ~ALLPERMS; newmode |= (mode & ALLPERMS); UFS_INODE_SET_MODE(ip, newmode); DIP_SET(ip, i_mode, ip->i_mode); UFS_INODE_SET_FLAG(ip, IN_CHANGE); #ifdef UFS_ACL if ((vp->v_mount->mnt_flag & MNT_NFS4ACLS) != 0) error = ufs_update_nfs4_acl_after_mode_change(vp, mode, ip->i_uid, cred, td); #endif if (error == 0 && (ip->i_flag & IN_CHANGE) != 0) error = UFS_UPDATE(vp, 0); return (error); } /* * Perform chown operation on inode ip; * inode must be locked prior to call. */ static int ufs_chown(vp, uid, gid, cred, td) struct vnode *vp; uid_t uid; gid_t gid; struct ucred *cred; struct thread *td; { struct inode *ip = VTOI(vp); uid_t ouid; gid_t ogid; int error = 0; #ifdef QUOTA int i; ufs2_daddr_t change; #endif if (uid == (uid_t)VNOVAL) uid = ip->i_uid; if (gid == (gid_t)VNOVAL) gid = ip->i_gid; /* * To modify the ownership of a file, must possess VADMIN for that * file. */ if ((error = VOP_ACCESSX(vp, VWRITE_OWNER, cred, td))) return (error); /* * To change the owner of a file, or change the group of a file to a * group of which we are not a member, the caller must have * privilege. */ if (((uid != ip->i_uid && uid != cred->cr_uid) || (gid != ip->i_gid && !groupmember(gid, cred))) && (error = priv_check_cred(cred, PRIV_VFS_CHOWN))) return (error); ogid = ip->i_gid; ouid = ip->i_uid; #ifdef QUOTA if ((error = getinoquota(ip)) != 0) return (error); if (ouid == uid) { dqrele(vp, ip->i_dquot[USRQUOTA]); ip->i_dquot[USRQUOTA] = NODQUOT; } if (ogid == gid) { dqrele(vp, ip->i_dquot[GRPQUOTA]); ip->i_dquot[GRPQUOTA] = NODQUOT; } change = DIP(ip, i_blocks); (void) chkdq(ip, -change, cred, CHOWN|FORCE); (void) chkiq(ip, -1, cred, CHOWN|FORCE); for (i = 0; i < MAXQUOTAS; i++) { dqrele(vp, ip->i_dquot[i]); ip->i_dquot[i] = NODQUOT; } #endif ip->i_gid = gid; DIP_SET(ip, i_gid, gid); ip->i_uid = uid; DIP_SET(ip, i_uid, uid); #ifdef QUOTA if ((error = getinoquota(ip)) == 0) { if (ouid == uid) { dqrele(vp, ip->i_dquot[USRQUOTA]); ip->i_dquot[USRQUOTA] = NODQUOT; } if (ogid == gid) { dqrele(vp, ip->i_dquot[GRPQUOTA]); ip->i_dquot[GRPQUOTA] = NODQUOT; } if ((error = chkdq(ip, change, cred, CHOWN)) == 0) { if ((error = chkiq(ip, 1, cred, CHOWN)) == 0) goto good; else (void) chkdq(ip, -change, cred, CHOWN|FORCE); } for (i = 0; i < MAXQUOTAS; i++) { dqrele(vp, ip->i_dquot[i]); ip->i_dquot[i] = NODQUOT; } } ip->i_gid = ogid; DIP_SET(ip, i_gid, ogid); ip->i_uid = ouid; DIP_SET(ip, i_uid, ouid); if (getinoquota(ip) == 0) { if (ouid == uid) { dqrele(vp, ip->i_dquot[USRQUOTA]); ip->i_dquot[USRQUOTA] = NODQUOT; } if (ogid == gid) { dqrele(vp, ip->i_dquot[GRPQUOTA]); ip->i_dquot[GRPQUOTA] = NODQUOT; } (void) chkdq(ip, change, cred, FORCE|CHOWN); (void) chkiq(ip, 1, cred, FORCE|CHOWN); (void) getinoquota(ip); } return (error); good: if (getinoquota(ip)) panic("ufs_chown: lost quota"); #endif /* QUOTA */ UFS_INODE_SET_FLAG(ip, IN_CHANGE); if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) { if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID)) { UFS_INODE_SET_MODE(ip, ip->i_mode & ~(ISUID | ISGID)); DIP_SET(ip, i_mode, ip->i_mode); } } error = UFS_UPDATE(vp, 0); return (error); } static int ufs_remove(ap) struct vop_remove_args /* { struct vnode *a_dvp; struct vnode *a_vp; struct componentname *a_cnp; } */ *ap; { struct inode *ip; struct vnode *vp = ap->a_vp; struct vnode *dvp = ap->a_dvp; int error; struct thread *td; td = curthread; ip = VTOI(vp); if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || (VTOI(dvp)->i_flags & APPEND)) return (EPERM); if (DOINGSUJ(dvp)) { error = softdep_prelink(dvp, vp, ap->a_cnp); if (error != 0) { MPASS(error == ERELOOKUP); return (error); } } #ifdef UFS_GJOURNAL ufs_gjournal_orphan(vp); #endif error = ufs_dirremove(dvp, ip, ap->a_cnp->cn_flags, 0); if (ip->i_nlink <= 0) vp->v_vflag |= VV_NOSYNC; if ((ip->i_flags & SF_SNAPSHOT) != 0) { /* * Avoid deadlock where another thread is trying to * update the inodeblock for dvp and is waiting on * snaplk. Temporary unlock the vnode lock for the * unlinked file and sync the directory. This should * allow vput() of the directory to not block later on * while holding the snapshot vnode locked, assuming * that the directory hasn't been unlinked too. */ VOP_UNLOCK(vp); (void) VOP_FSYNC(dvp, MNT_WAIT, td); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); } return (error); } static void print_bad_link_count(const char *funcname, struct vnode *dvp) { struct inode *dip; dip = VTOI(dvp); uprintf("%s: Bad link count %d on parent inode %jd in file system %s\n", funcname, dip->i_effnlink, (intmax_t)dip->i_number, dvp->v_mount->mnt_stat.f_mntonname); } /* * link vnode call */ static int ufs_link(ap) struct vop_link_args /* { struct vnode *a_tdvp; struct vnode *a_vp; struct componentname *a_cnp; } */ *ap; { struct vnode *vp = ap->a_vp; struct vnode *tdvp = ap->a_tdvp; struct componentname *cnp = ap->a_cnp; struct inode *ip; struct direct newdir; int error; #ifdef INVARIANTS if ((cnp->cn_flags & HASBUF) == 0) panic("ufs_link: no name"); #endif if (DOINGSUJ(tdvp)) { error = softdep_prelink(tdvp, vp, cnp); if (error != 0) { MPASS(error == ERELOOKUP); return (error); } } if (VTOI(tdvp)->i_effnlink < 2) { print_bad_link_count("ufs_link", tdvp); error = EINVAL; goto out; } ip = VTOI(vp); if (ip->i_nlink >= UFS_LINK_MAX) { error = EMLINK; goto out; } /* * The file may have been removed after namei droped the original * lock. */ if (ip->i_effnlink == 0) { error = ENOENT; goto out; } if (ip->i_flags & (IMMUTABLE | APPEND)) { error = EPERM; goto out; } ip->i_effnlink++; ip->i_nlink++; DIP_SET(ip, i_nlink, ip->i_nlink); UFS_INODE_SET_FLAG(ip, IN_CHANGE); if (DOINGSOFTDEP(vp)) softdep_setup_link(VTOI(tdvp), ip); error = UFS_UPDATE(vp, !DOINGSOFTDEP(vp) && !DOINGASYNC(vp)); if (!error) { ufs_makedirentry(ip, cnp, &newdir); error = ufs_direnter(tdvp, vp, &newdir, cnp, NULL); } if (error) { ip->i_effnlink--; ip->i_nlink--; DIP_SET(ip, i_nlink, ip->i_nlink); UFS_INODE_SET_FLAG(ip, IN_CHANGE); if (DOINGSOFTDEP(vp)) softdep_revert_link(VTOI(tdvp), ip); } out: return (error); } /* * whiteout vnode call */ static int ufs_whiteout(ap) struct vop_whiteout_args /* { struct vnode *a_dvp; struct componentname *a_cnp; int a_flags; } */ *ap; { struct vnode *dvp = ap->a_dvp; struct componentname *cnp = ap->a_cnp; struct direct newdir; int error = 0; if (DOINGSUJ(dvp) && (ap->a_flags == CREATE || ap->a_flags == DELETE)) { error = softdep_prelink(dvp, NULL, cnp); if (error != 0) { MPASS(error == ERELOOKUP); return (error); } } switch (ap->a_flags) { case LOOKUP: /* 4.4 format directories support whiteout operations */ if (!OFSFMT(dvp)) return (0); return (EOPNOTSUPP); case CREATE: /* create a new directory whiteout */ #ifdef INVARIANTS if ((cnp->cn_flags & SAVENAME) == 0) panic("ufs_whiteout: missing name"); if (OFSFMT(dvp)) panic("ufs_whiteout: old format filesystem"); #endif newdir.d_ino = UFS_WINO; newdir.d_namlen = cnp->cn_namelen; bcopy(cnp->cn_nameptr, newdir.d_name, (unsigned)cnp->cn_namelen + 1); newdir.d_type = DT_WHT; error = ufs_direnter(dvp, NULL, &newdir, cnp, NULL); break; case DELETE: /* remove an existing directory whiteout */ #ifdef INVARIANTS if (OFSFMT(dvp)) panic("ufs_whiteout: old format filesystem"); #endif cnp->cn_flags &= ~DOWHITEOUT; error = ufs_dirremove(dvp, NULL, cnp->cn_flags, 0); break; default: panic("ufs_whiteout: unknown op"); } return (error); } static volatile int rename_restarts; SYSCTL_INT(_vfs_ufs, OID_AUTO, rename_restarts, CTLFLAG_RD, __DEVOLATILE(int *, &rename_restarts), 0, "Times rename had to restart due to lock contention"); /* * Rename system call. * rename("foo", "bar"); * is essentially * unlink("bar"); * link("foo", "bar"); * unlink("foo"); * but ``atomically''. Can't do full commit without saving state in the * inode on disk which isn't feasible at this time. Best we can do is * always guarantee the target exists. * * Basic algorithm is: * * 1) Bump link count on source while we're linking it to the * target. This also ensure the inode won't be deleted out * from underneath us while we work (it may be truncated by * a concurrent `trunc' or `open' for creation). * 2) Link source to destination. If destination already exists, * delete it first. * 3) Unlink source reference to inode if still around. If a * directory was moved and the parent of the destination * is different from the source, patch the ".." entry in the * directory. */ static int ufs_rename(ap) struct vop_rename_args /* { struct vnode *a_fdvp; struct vnode *a_fvp; struct componentname *a_fcnp; struct vnode *a_tdvp; struct vnode *a_tvp; struct componentname *a_tcnp; } */ *ap; { struct vnode *tvp = ap->a_tvp; struct vnode *tdvp = ap->a_tdvp; struct vnode *fvp = ap->a_fvp; struct vnode *fdvp = ap->a_fdvp; struct vnode *nvp; struct componentname *tcnp = ap->a_tcnp; struct componentname *fcnp = ap->a_fcnp; - struct thread *td = fcnp->cn_thread; + struct thread *td = curthread; struct inode *fip, *tip, *tdp, *fdp; struct direct newdir; off_t endoff; int doingdirectory, newparent; int error = 0; struct mount *mp; ino_t ino; seqc_t fdvp_s, fvp_s, tdvp_s, tvp_s; bool checkpath_locked, want_seqc_end; checkpath_locked = want_seqc_end = false; #ifdef INVARIANTS if ((tcnp->cn_flags & HASBUF) == 0 || (fcnp->cn_flags & HASBUF) == 0) panic("ufs_rename: no name"); #endif endoff = 0; mp = tdvp->v_mount; VOP_UNLOCK(tdvp); if (tvp && tvp != tdvp) VOP_UNLOCK(tvp); /* * Check for cross-device rename. */ if ((fvp->v_mount != tdvp->v_mount) || (tvp && (fvp->v_mount != tvp->v_mount))) { error = EXDEV; mp = NULL; goto releout; } fdvp_s = fvp_s = tdvp_s = tvp_s = SEQC_MOD; relock: /* * We need to acquire 2 to 4 locks depending on whether tvp is NULL * and fdvp and tdvp are the same directory. Subsequently we need * to double-check all paths and in the directory rename case we * need to verify that we are not creating a directory loop. To * handle this we acquire all but fdvp using non-blocking * acquisitions. If we fail to acquire any lock in the path we will * drop all held locks, acquire the new lock in a blocking fashion, * and then release it and restart the rename. This acquire/release * step ensures that we do not spin on a lock waiting for release. */ error = vn_lock(fdvp, LK_EXCLUSIVE); if (error) goto releout; if (vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { VOP_UNLOCK(fdvp); error = vn_lock(tdvp, LK_EXCLUSIVE); if (error) goto releout; VOP_UNLOCK(tdvp); atomic_add_int(&rename_restarts, 1); goto relock; } /* * Re-resolve fvp to be certain it still exists and fetch the * correct vnode. */ error = ufs_lookup_ino(fdvp, NULL, fcnp, &ino); if (error) { VOP_UNLOCK(fdvp); VOP_UNLOCK(tdvp); goto releout; } error = VFS_VGET(mp, ino, LK_EXCLUSIVE | LK_NOWAIT, &nvp); if (error) { VOP_UNLOCK(fdvp); VOP_UNLOCK(tdvp); if (error != EBUSY) goto releout; error = VFS_VGET(mp, ino, LK_EXCLUSIVE, &nvp); if (error != 0) goto releout; VOP_UNLOCK(nvp); vrele(fvp); fvp = nvp; atomic_add_int(&rename_restarts, 1); goto relock; } vrele(fvp); fvp = nvp; /* * Re-resolve tvp and acquire the vnode lock if present. */ error = ufs_lookup_ino(tdvp, NULL, tcnp, &ino); if (error != 0 && error != EJUSTRETURN) { VOP_UNLOCK(fdvp); VOP_UNLOCK(tdvp); VOP_UNLOCK(fvp); goto releout; } /* * If tvp disappeared we just carry on. */ if (error == EJUSTRETURN && tvp != NULL) { vrele(tvp); tvp = NULL; } /* * Get the tvp ino if the lookup succeeded. We may have to restart * if the non-blocking acquire fails. */ if (error == 0) { nvp = NULL; error = VFS_VGET(mp, ino, LK_EXCLUSIVE | LK_NOWAIT, &nvp); if (tvp) vrele(tvp); tvp = nvp; if (error) { VOP_UNLOCK(fdvp); VOP_UNLOCK(tdvp); VOP_UNLOCK(fvp); if (error != EBUSY) goto releout; error = VFS_VGET(mp, ino, LK_EXCLUSIVE, &nvp); if (error != 0) goto releout; vput(nvp); atomic_add_int(&rename_restarts, 1); goto relock; } } if (DOINGSUJ(fdvp) && (seqc_in_modify(fdvp_s) || !vn_seqc_consistent(fdvp, fdvp_s) || seqc_in_modify(fvp_s) || !vn_seqc_consistent(fvp, fvp_s) || seqc_in_modify(tdvp_s) || !vn_seqc_consistent(tdvp, tdvp_s) || (tvp != NULL && (seqc_in_modify(tvp_s) || !vn_seqc_consistent(tvp, tvp_s))))) { error = softdep_prerename(fdvp, fvp, tdvp, tvp); if (error != 0) { if (error == ERELOOKUP) { fdvp_s = vn_seqc_read_any(fdvp); fvp_s = vn_seqc_read_any(fvp); tdvp_s = vn_seqc_read_any(tdvp); if (tvp != NULL) tvp_s = vn_seqc_read_any(tvp); atomic_add_int(&rename_restarts, 1); goto relock; } goto releout; } } fdp = VTOI(fdvp); fip = VTOI(fvp); tdp = VTOI(tdvp); tip = NULL; if (tvp) tip = VTOI(tvp); if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || (VTOI(tdvp)->i_flags & APPEND))) { error = EPERM; goto unlockout; } /* * Renaming a file to itself has no effect. The upper layers should * not call us in that case. However, things could change after * we drop the locks above. */ if (fvp == tvp) { error = 0; goto unlockout; } doingdirectory = 0; newparent = 0; ino = fip->i_number; if (fip->i_nlink >= UFS_LINK_MAX) { error = EMLINK; goto unlockout; } if ((fip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || (fdp->i_flags & APPEND)) { error = EPERM; goto unlockout; } if ((fip->i_mode & IFMT) == IFDIR) { /* * Avoid ".", "..", and aliases of "." for obvious reasons. */ if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') || fdp == fip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT) { error = EINVAL; goto unlockout; } if (fdp->i_number != tdp->i_number) newparent = tdp->i_number; doingdirectory = 1; } if ((fvp->v_type == VDIR && fvp->v_mountedhere != NULL) || (tvp != NULL && tvp->v_type == VDIR && tvp->v_mountedhere != NULL)) { error = EXDEV; goto unlockout; } /* * If ".." must be changed (ie the directory gets a new * parent) then the source directory must not be in the * directory hierarchy above the target, as this would * orphan everything below the source directory. Also * the user must have write permission in the source so * as to be able to change "..". */ if (doingdirectory && newparent) { - error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread); + error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, curthread); if (error) goto unlockout; sx_xlock(&VFSTOUFS(mp)->um_checkpath_lock); checkpath_locked = true; error = ufs_checkpath(ino, fdp->i_number, tdp, tcnp->cn_cred, &ino); /* * We encountered a lock that we have to wait for. Unlock * everything else and VGET before restarting. */ if (ino) { sx_xunlock(&VFSTOUFS(mp)->um_checkpath_lock); checkpath_locked = false; VOP_UNLOCK(fdvp); VOP_UNLOCK(fvp); VOP_UNLOCK(tdvp); if (tvp) VOP_UNLOCK(tvp); error = VFS_VGET(mp, ino, LK_SHARED, &nvp); if (error == 0) vput(nvp); atomic_add_int(&rename_restarts, 1); goto relock; } if (error) goto unlockout; if ((tcnp->cn_flags & SAVESTART) == 0) panic("ufs_rename: lost to startdir"); } if (fip->i_effnlink == 0 || fdp->i_effnlink == 0 || tdp->i_effnlink == 0) panic("Bad effnlink fip %p, fdp %p, tdp %p", fip, fdp, tdp); if (tvp != NULL) vn_seqc_write_begin(tvp); vn_seqc_write_begin(tdvp); vn_seqc_write_begin(fvp); vn_seqc_write_begin(fdvp); want_seqc_end = true; /* * 1) Bump link count while we're moving stuff * around. If we crash somewhere before * completing our work, the link count * may be wrong, but correctable. */ fip->i_effnlink++; fip->i_nlink++; DIP_SET(fip, i_nlink, fip->i_nlink); UFS_INODE_SET_FLAG(fip, IN_CHANGE); if (DOINGSOFTDEP(fvp)) softdep_setup_link(tdp, fip); error = UFS_UPDATE(fvp, !DOINGSOFTDEP(fvp) && !DOINGASYNC(fvp)); if (error) goto bad; /* * 2) If target doesn't exist, link the target * to the source and unlink the source. * Otherwise, rewrite the target directory * entry to reference the source inode and * expunge the original entry's existence. */ if (tip == NULL) { if (ITODEV(tdp) != ITODEV(fip)) panic("ufs_rename: EXDEV"); if (doingdirectory && newparent) { /* * Account for ".." in new directory. * When source and destination have the same * parent we don't adjust the link count. The * actual link modification is completed when * .. is rewritten below. */ if (tdp->i_nlink >= UFS_LINK_MAX) { error = EMLINK; goto bad; } } ufs_makedirentry(fip, tcnp, &newdir); error = ufs_direnter(tdvp, NULL, &newdir, tcnp, NULL); if (error) goto bad; /* Setup tdvp for directory compaction if needed. */ if (I_COUNT(tdp) != 0 && I_ENDOFF(tdp) != 0 && I_ENDOFF(tdp) < tdp->i_size) endoff = I_ENDOFF(tdp); } else { if (ITODEV(tip) != ITODEV(tdp) || ITODEV(tip) != ITODEV(fip)) panic("ufs_rename: EXDEV"); /* * Short circuit rename(foo, foo). */ if (tip->i_number == fip->i_number) panic("ufs_rename: same file"); /* * If the parent directory is "sticky", then the caller * must possess VADMIN for the parent directory, or the * destination of the rename. This implements append-only * directories. */ if ((tdp->i_mode & S_ISTXT) && VOP_ACCESS(tdvp, VADMIN, tcnp->cn_cred, td) && VOP_ACCESS(tvp, VADMIN, tcnp->cn_cred, td)) { error = EPERM; goto bad; } /* * Target must be empty if a directory and have no links * to it. Also, ensure source and target are compatible * (both directories, or both not directories). */ if ((tip->i_mode & IFMT) == IFDIR) { if ((tip->i_effnlink > 2) || !ufs_dirempty(tip, tdp->i_number, tcnp->cn_cred)) { error = ENOTEMPTY; goto bad; } if (!doingdirectory) { error = ENOTDIR; goto bad; } cache_purge(tdvp); } else if (doingdirectory) { error = EISDIR; goto bad; } if (doingdirectory) { if (!newparent) { tdp->i_effnlink--; if (DOINGSOFTDEP(tdvp)) softdep_change_linkcnt(tdp); } tip->i_effnlink--; if (DOINGSOFTDEP(tvp)) softdep_change_linkcnt(tip); } error = ufs_dirrewrite(tdp, tip, fip->i_number, IFTODT(fip->i_mode), (doingdirectory && newparent) ? newparent : doingdirectory); if (error) { if (doingdirectory) { if (!newparent) { tdp->i_effnlink++; if (DOINGSOFTDEP(tdvp)) softdep_change_linkcnt(tdp); } tip->i_effnlink++; if (DOINGSOFTDEP(tvp)) softdep_change_linkcnt(tip); } goto bad; } if (doingdirectory && !DOINGSOFTDEP(tvp)) { /* * The only stuff left in the directory is "." * and "..". The "." reference is inconsequential * since we are quashing it. We have removed the "." * reference and the reference in the parent directory, * but there may be other hard links. The soft * dependency code will arrange to do these operations * after the parent directory entry has been deleted on * disk, so when running with that code we avoid doing * them now. */ if (!newparent) { tdp->i_nlink--; DIP_SET(tdp, i_nlink, tdp->i_nlink); UFS_INODE_SET_FLAG(tdp, IN_CHANGE); } tip->i_nlink--; DIP_SET(tip, i_nlink, tip->i_nlink); UFS_INODE_SET_FLAG(tip, IN_CHANGE); } } /* * 3) Unlink the source. We have to resolve the path again to * fixup the directory offset and count for ufs_dirremove. */ if (fdvp == tdvp) { error = ufs_lookup_ino(fdvp, NULL, fcnp, &ino); if (error) panic("ufs_rename: from entry went away!"); if (ino != fip->i_number) panic("ufs_rename: ino mismatch %ju != %ju\n", (uintmax_t)ino, (uintmax_t)fip->i_number); } /* * If the source is a directory with a * new parent, the link count of the old * parent directory must be decremented * and ".." set to point to the new parent. */ if (doingdirectory && newparent) { /* * If tip exists we simply use its link, otherwise we must * add a new one. */ if (tip == NULL) { tdp->i_effnlink++; tdp->i_nlink++; DIP_SET(tdp, i_nlink, tdp->i_nlink); UFS_INODE_SET_FLAG(tdp, IN_CHANGE); if (DOINGSOFTDEP(tdvp)) softdep_setup_dotdot_link(tdp, fip); error = UFS_UPDATE(tdvp, !DOINGSOFTDEP(tdvp) && !DOINGASYNC(tdvp)); /* Don't go to bad here as the new link exists. */ if (error) goto unlockout; } else if (DOINGSUJ(tdvp)) /* Journal must account for each new link. */ softdep_setup_dotdot_link(tdp, fip); SET_I_OFFSET(fip, mastertemplate.dot_reclen); ufs_dirrewrite(fip, fdp, newparent, DT_DIR, 0); cache_purge(fdvp); } error = ufs_dirremove(fdvp, fip, fcnp->cn_flags, 0); /* * The kern_renameat() looks up the fvp using the DELETE flag, which * causes the removal of the name cache entry for fvp. * As the relookup of the fvp is done in two steps: * ufs_lookup_ino() and then VFS_VGET(), another thread might do a * normal lookup of the from name just before the VFS_VGET() call, * causing the cache entry to be re-instantiated. * * The same issue also applies to tvp if it exists as * otherwise we may have a stale name cache entry for the new * name that references the old i-node if it has other links * or open file descriptors. */ cache_vop_rename(fdvp, fvp, tdvp, tvp, fcnp, tcnp); unlockout: if (want_seqc_end) { if (tvp != NULL) vn_seqc_write_end(tvp); vn_seqc_write_end(tdvp); vn_seqc_write_end(fvp); vn_seqc_write_end(fdvp); } if (checkpath_locked) sx_xunlock(&VFSTOUFS(mp)->um_checkpath_lock); vput(fdvp); vput(fvp); /* * If compaction or fsync was requested do it in * ffs_vput_pair() now that other locks are no longer needed. */ if (error == 0 && endoff != 0) { UFS_INODE_SET_FLAG(tdp, IN_ENDOFF); SET_I_ENDOFF(tdp, endoff); } VOP_VPUT_PAIR(tdvp, &tvp, true); return (error); bad: fip->i_effnlink--; fip->i_nlink--; DIP_SET(fip, i_nlink, fip->i_nlink); UFS_INODE_SET_FLAG(fip, IN_CHANGE); if (DOINGSOFTDEP(fvp)) softdep_revert_link(tdp, fip); goto unlockout; releout: if (want_seqc_end) { if (tvp != NULL) vn_seqc_write_end(tvp); vn_seqc_write_end(tdvp); vn_seqc_write_end(fvp); vn_seqc_write_end(fdvp); } vrele(fdvp); vrele(fvp); vrele(tdvp); if (tvp) vrele(tvp); return (error); } #ifdef UFS_ACL static int ufs_do_posix1e_acl_inheritance_dir(struct vnode *dvp, struct vnode *tvp, mode_t dmode, struct ucred *cred, struct thread *td) { int error; struct inode *ip = VTOI(tvp); struct acl *dacl, *acl; acl = acl_alloc(M_WAITOK); dacl = acl_alloc(M_WAITOK); /* * Retrieve default ACL from parent, if any. */ error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td); switch (error) { case 0: /* * Retrieved a default ACL, so merge mode and ACL if * necessary. If the ACL is empty, fall through to * the "not defined or available" case. */ if (acl->acl_cnt != 0) { dmode = acl_posix1e_newfilemode(dmode, acl); UFS_INODE_SET_MODE(ip, dmode); DIP_SET(ip, i_mode, dmode); *dacl = *acl; ufs_sync_acl_from_inode(ip, acl); break; } /* FALLTHROUGH */ case EOPNOTSUPP: /* * Just use the mode as-is. */ UFS_INODE_SET_MODE(ip, dmode); DIP_SET(ip, i_mode, dmode); error = 0; goto out; default: goto out; } /* * XXX: If we abort now, will Soft Updates notify the extattr * code that the EAs for the file need to be released? */ error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td); if (error == 0) error = VOP_SETACL(tvp, ACL_TYPE_DEFAULT, dacl, cred, td); switch (error) { case 0: break; case EOPNOTSUPP: /* * XXX: This should not happen, as EOPNOTSUPP above * was supposed to free acl. */ printf("ufs_mkdir: VOP_GETACL() but no VOP_SETACL()\n"); /* panic("ufs_mkdir: VOP_GETACL() but no VOP_SETACL()"); */ break; default: goto out; } out: acl_free(acl); acl_free(dacl); return (error); } static int ufs_do_posix1e_acl_inheritance_file(struct vnode *dvp, struct vnode *tvp, mode_t mode, struct ucred *cred, struct thread *td) { int error; struct inode *ip = VTOI(tvp); struct acl *acl; acl = acl_alloc(M_WAITOK); /* * Retrieve default ACL for parent, if any. */ error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td); switch (error) { case 0: /* * Retrieved a default ACL, so merge mode and ACL if * necessary. */ if (acl->acl_cnt != 0) { /* * Two possible ways for default ACL to not * be present. First, the EA can be * undefined, or second, the default ACL can * be blank. If it's blank, fall through to * the it's not defined case. */ mode = acl_posix1e_newfilemode(mode, acl); UFS_INODE_SET_MODE(ip, mode); DIP_SET(ip, i_mode, mode); ufs_sync_acl_from_inode(ip, acl); break; } /* FALLTHROUGH */ case EOPNOTSUPP: /* * Just use the mode as-is. */ UFS_INODE_SET_MODE(ip, mode); DIP_SET(ip, i_mode, mode); error = 0; goto out; default: goto out; } /* * XXX: If we abort now, will Soft Updates notify the extattr * code that the EAs for the file need to be released? */ error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td); switch (error) { case 0: break; case EOPNOTSUPP: /* * XXX: This should not happen, as EOPNOTSUPP above was * supposed to free acl. */ printf("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() " "but no VOP_SETACL()\n"); /* panic("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() " "but no VOP_SETACL()"); */ break; default: goto out; } out: acl_free(acl); return (error); } static int ufs_do_nfs4_acl_inheritance(struct vnode *dvp, struct vnode *tvp, mode_t child_mode, struct ucred *cred, struct thread *td) { int error; struct acl *parent_aclp, *child_aclp; parent_aclp = acl_alloc(M_WAITOK); child_aclp = acl_alloc(M_WAITOK | M_ZERO); error = ufs_getacl_nfs4_internal(dvp, parent_aclp, td); if (error) goto out; acl_nfs4_compute_inherited_acl(parent_aclp, child_aclp, child_mode, VTOI(tvp)->i_uid, tvp->v_type == VDIR); error = ufs_setacl_nfs4_internal(tvp, child_aclp, td); if (error) goto out; out: acl_free(parent_aclp); acl_free(child_aclp); return (error); } #endif /* * Mkdir system call */ static int ufs_mkdir(ap) struct vop_mkdir_args /* { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; struct vattr *a_vap; } */ *ap; { struct vnode *dvp = ap->a_dvp; struct vattr *vap = ap->a_vap; struct componentname *cnp = ap->a_cnp; struct inode *ip, *dp; struct vnode *tvp; struct buf *bp; struct dirtemplate dirtemplate, *dtp; struct direct newdir; int error, dmode; long blkoff; #ifdef INVARIANTS if ((cnp->cn_flags & HASBUF) == 0) panic("ufs_mkdir: no name"); #endif dp = VTOI(dvp); if (dp->i_nlink >= UFS_LINK_MAX) { error = EMLINK; goto out; } dmode = vap->va_mode & 0777; dmode |= IFDIR; /* * Must simulate part of ufs_makeinode here to acquire the inode, * but not have it entered in the parent directory. The entry is * made later after writing "." and ".." entries. */ if (dp->i_effnlink < 2) { print_bad_link_count("ufs_mkdir", dvp); error = EINVAL; goto out; } if (DOINGSUJ(dvp)) { error = softdep_prelink(dvp, NULL, cnp); if (error != 0) { MPASS(error == ERELOOKUP); return (error); } } error = UFS_VALLOC(dvp, dmode, cnp->cn_cred, &tvp); if (error) goto out; vn_seqc_write_begin(tvp); ip = VTOI(tvp); ip->i_gid = dp->i_gid; DIP_SET(ip, i_gid, dp->i_gid); #ifdef SUIDDIR { #ifdef QUOTA struct ucred ucred, *ucp; gid_t ucred_group; ucp = cnp->cn_cred; #endif /* * If we are hacking owners here, (only do this where told to) * and we are not giving it TO root, (would subvert quotas) * then go ahead and give it to the other user. * The new directory also inherits the SUID bit. * If user's UID and dir UID are the same, * 'give it away' so that the SUID is still forced on. */ if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) && (dp->i_mode & ISUID) && dp->i_uid) { dmode |= ISUID; ip->i_uid = dp->i_uid; DIP_SET(ip, i_uid, dp->i_uid); #ifdef QUOTA if (dp->i_uid != cnp->cn_cred->cr_uid) { /* * Make sure the correct user gets charged * for the space. * Make a dummy credential for the victim. * XXX This seems to never be accessed out of * our context so a stack variable is ok. */ refcount_init(&ucred.cr_ref, 1); ucred.cr_uid = ip->i_uid; ucred.cr_ngroups = 1; ucred.cr_groups = &ucred_group; ucred.cr_groups[0] = dp->i_gid; ucp = &ucred; } #endif } else { ip->i_uid = cnp->cn_cred->cr_uid; DIP_SET(ip, i_uid, ip->i_uid); } #ifdef QUOTA if ((error = getinoquota(ip)) || (error = chkiq(ip, 1, ucp, 0))) { if (DOINGSOFTDEP(tvp)) softdep_revert_link(dp, ip); UFS_VFREE(tvp, ip->i_number, dmode); vn_seqc_write_end(tvp); vgone(tvp); vput(tvp); return (error); } #endif } #else /* !SUIDDIR */ ip->i_uid = cnp->cn_cred->cr_uid; DIP_SET(ip, i_uid, ip->i_uid); #ifdef QUOTA if ((error = getinoquota(ip)) || (error = chkiq(ip, 1, cnp->cn_cred, 0))) { if (DOINGSOFTDEP(tvp)) softdep_revert_link(dp, ip); UFS_VFREE(tvp, ip->i_number, dmode); vn_seqc_write_end(tvp); vgone(tvp); vput(tvp); return (error); } #endif #endif /* !SUIDDIR */ UFS_INODE_SET_FLAG(ip, IN_ACCESS | IN_CHANGE | IN_UPDATE); UFS_INODE_SET_MODE(ip, dmode); DIP_SET(ip, i_mode, dmode); tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */ ip->i_effnlink = 2; ip->i_nlink = 2; DIP_SET(ip, i_nlink, 2); if (cnp->cn_flags & ISWHITEOUT) { ip->i_flags |= UF_OPAQUE; DIP_SET(ip, i_flags, ip->i_flags); } /* * Bump link count in parent directory to reflect work done below. * Should be done before reference is created so cleanup is * possible if we crash. */ dp->i_effnlink++; dp->i_nlink++; DIP_SET(dp, i_nlink, dp->i_nlink); UFS_INODE_SET_FLAG(dp, IN_CHANGE); if (DOINGSOFTDEP(dvp)) softdep_setup_mkdir(dp, ip); error = UFS_UPDATE(dvp, !DOINGSOFTDEP(dvp) && !DOINGASYNC(dvp)); if (error) goto bad; #ifdef MAC if (dvp->v_mount->mnt_flag & MNT_MULTILABEL) { error = mac_vnode_create_extattr(cnp->cn_cred, dvp->v_mount, dvp, tvp, cnp); if (error) goto bad; } #endif #ifdef UFS_ACL if (dvp->v_mount->mnt_flag & MNT_ACLS) { error = ufs_do_posix1e_acl_inheritance_dir(dvp, tvp, dmode, - cnp->cn_cred, cnp->cn_thread); + cnp->cn_cred, curthread); if (error) goto bad; } else if (dvp->v_mount->mnt_flag & MNT_NFS4ACLS) { error = ufs_do_nfs4_acl_inheritance(dvp, tvp, dmode, - cnp->cn_cred, cnp->cn_thread); + cnp->cn_cred, curthread); if (error) goto bad; } #endif /* !UFS_ACL */ /* * Initialize directory with "." and ".." from static template. */ if (!OFSFMT(dvp)) dtp = &mastertemplate; else dtp = (struct dirtemplate *)&omastertemplate; dirtemplate = *dtp; dirtemplate.dot_ino = ip->i_number; dirtemplate.dotdot_ino = dp->i_number; vnode_pager_setsize(tvp, DIRBLKSIZ); if ((error = UFS_BALLOC(tvp, (off_t)0, DIRBLKSIZ, cnp->cn_cred, BA_CLRBUF, &bp)) != 0) goto bad; ip->i_size = DIRBLKSIZ; DIP_SET(ip, i_size, DIRBLKSIZ); UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE); bcopy((caddr_t)&dirtemplate, (caddr_t)bp->b_data, sizeof dirtemplate); if (DOINGSOFTDEP(tvp)) { /* * Ensure that the entire newly allocated block is a * valid directory so that future growth within the * block does not have to ensure that the block is * written before the inode. */ blkoff = DIRBLKSIZ; while (blkoff < bp->b_bcount) { ((struct direct *) (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ; blkoff += DIRBLKSIZ; } } if ((error = UFS_UPDATE(tvp, !DOINGSOFTDEP(tvp) && !DOINGASYNC(tvp))) != 0) { (void)bwrite(bp); goto bad; } /* * Directory set up, now install its entry in the parent directory. * * If we are not doing soft dependencies, then we must write out the * buffer containing the new directory body before entering the new * name in the parent. If we are doing soft dependencies, then the * buffer containing the new directory body will be passed to and * released in the soft dependency code after the code has attached * an appropriate ordering dependency to the buffer which ensures that * the buffer is written before the new name is written in the parent. */ if (DOINGASYNC(dvp)) bdwrite(bp); else if (!DOINGSOFTDEP(dvp) && ((error = bwrite(bp)))) goto bad; ufs_makedirentry(ip, cnp, &newdir); error = ufs_direnter(dvp, tvp, &newdir, cnp, bp); bad: if (error == 0) { *ap->a_vpp = tvp; vn_seqc_write_end(tvp); } else { dp->i_effnlink--; dp->i_nlink--; DIP_SET(dp, i_nlink, dp->i_nlink); UFS_INODE_SET_FLAG(dp, IN_CHANGE); /* * No need to do an explicit VOP_TRUNCATE here, vrele will * do this for us because we set the link count to 0. */ ip->i_effnlink = 0; ip->i_nlink = 0; DIP_SET(ip, i_nlink, 0); UFS_INODE_SET_FLAG(ip, IN_CHANGE); if (DOINGSOFTDEP(tvp)) softdep_revert_mkdir(dp, ip); vn_seqc_write_end(tvp); vgone(tvp); vput(tvp); } out: return (error); } /* * Rmdir system call. */ static int ufs_rmdir(ap) struct vop_rmdir_args /* { struct vnode *a_dvp; struct vnode *a_vp; struct componentname *a_cnp; } */ *ap; { struct vnode *vp = ap->a_vp; struct vnode *dvp = ap->a_dvp; struct componentname *cnp = ap->a_cnp; struct inode *ip, *dp; int error; ip = VTOI(vp); dp = VTOI(dvp); /* * Do not remove a directory that is in the process of being renamed. * Verify the directory is empty (and valid). Rmdir ".." will not be * valid since ".." will contain a reference to the current directory * and thus be non-empty. Do not allow the removal of mounted on * directories (this can happen when an NFS exported filesystem * tries to remove a locally mounted on directory). */ error = 0; if (dp->i_effnlink <= 2) { if (dp->i_effnlink == 2) print_bad_link_count("ufs_rmdir", dvp); error = EINVAL; goto out; } if (!ufs_dirempty(ip, dp->i_number, cnp->cn_cred)) { error = ENOTEMPTY; goto out; } if ((dp->i_flags & APPEND) || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) { error = EPERM; goto out; } if (vp->v_mountedhere != 0) { error = EINVAL; goto out; } if (DOINGSUJ(dvp)) { error = softdep_prelink(dvp, vp, cnp); if (error != 0) { MPASS(error == ERELOOKUP); return (error); } } #ifdef UFS_GJOURNAL ufs_gjournal_orphan(vp); #endif /* * Delete reference to directory before purging * inode. If we crash in between, the directory * will be reattached to lost+found, */ dp->i_effnlink--; ip->i_effnlink--; if (DOINGSOFTDEP(vp)) softdep_setup_rmdir(dp, ip); error = ufs_dirremove(dvp, ip, cnp->cn_flags, 1); if (error) { dp->i_effnlink++; ip->i_effnlink++; if (DOINGSOFTDEP(vp)) softdep_revert_rmdir(dp, ip); goto out; } /* * The only stuff left in the directory is "." and "..". The "." * reference is inconsequential since we are quashing it. The soft * dependency code will arrange to do these operations after * the parent directory entry has been deleted on disk, so * when running with that code we avoid doing them now. */ if (!DOINGSOFTDEP(vp)) { dp->i_nlink--; DIP_SET(dp, i_nlink, dp->i_nlink); UFS_INODE_SET_FLAG(dp, IN_CHANGE); error = UFS_UPDATE(dvp, 0); ip->i_nlink--; DIP_SET(ip, i_nlink, ip->i_nlink); UFS_INODE_SET_FLAG(ip, IN_CHANGE); } cache_vop_rmdir(dvp, vp); #ifdef UFS_DIRHASH /* Kill any active hash; i_effnlink == 0, so it will not come back. */ if (ip->i_dirhash != NULL) ufsdirhash_free(ip); #endif out: return (error); } /* * symlink -- make a symbolic link */ static int ufs_symlink(ap) struct vop_symlink_args /* { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; struct vattr *a_vap; const char *a_target; } */ *ap; { struct vnode *vp, **vpp = ap->a_vpp; struct inode *ip; int len, error; error = ufs_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp, vpp, ap->a_cnp, "ufs_symlink"); if (error) return (error); vp = *vpp; len = strlen(ap->a_target); if (len < VFSTOUFS(vp->v_mount)->um_maxsymlinklen) { ip = VTOI(vp); bcopy(ap->a_target, SHORTLINK(ip), len); ip->i_size = len; DIP_SET(ip, i_size, len); UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE); error = UFS_UPDATE(vp, 0); } else error = vn_rdwr(UIO_WRITE, vp, __DECONST(void *, ap->a_target), len, (off_t)0, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, ap->a_cnp->cn_cred, NOCRED, NULL, NULL); if (error) vput(vp); return (error); } /* * Vnode op for reading directories. */ int ufs_readdir(ap) struct vop_readdir_args /* { struct vnode *a_vp; struct uio *a_uio; struct ucred *a_cred; int *a_eofflag; int *a_ncookies; u_long **a_cookies; } */ *ap; { struct vnode *vp = ap->a_vp; struct uio *uio = ap->a_uio; struct buf *bp; struct inode *ip; struct direct *dp, *edp; u_long *cookies; struct dirent dstdp; off_t offset, startoffset; size_t readcnt, skipcnt; ssize_t startresid; u_int ncookies; int error; if (uio->uio_offset < 0) return (EINVAL); ip = VTOI(vp); if (ip->i_effnlink == 0) return (0); if (ap->a_ncookies != NULL) { if (uio->uio_resid < 0) ncookies = 0; else ncookies = uio->uio_resid; if (uio->uio_offset >= ip->i_size) ncookies = 0; else if (ip->i_size - uio->uio_offset < ncookies) ncookies = ip->i_size - uio->uio_offset; ncookies = ncookies / (offsetof(struct direct, d_name) + 4) + 1; cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK); *ap->a_ncookies = ncookies; *ap->a_cookies = cookies; } else { ncookies = 0; cookies = NULL; } offset = startoffset = uio->uio_offset; startresid = uio->uio_resid; error = 0; while (error == 0 && uio->uio_resid > 0 && uio->uio_offset < ip->i_size) { error = UFS_BLKATOFF(vp, uio->uio_offset, NULL, &bp); if (error) break; if (bp->b_offset + bp->b_bcount > ip->i_size) readcnt = ip->i_size - bp->b_offset; else readcnt = bp->b_bcount; skipcnt = (size_t)(uio->uio_offset - bp->b_offset) & ~(size_t)(DIRBLKSIZ - 1); offset = bp->b_offset + skipcnt; dp = (struct direct *)&bp->b_data[skipcnt]; edp = (struct direct *)&bp->b_data[readcnt]; while (error == 0 && uio->uio_resid > 0 && dp < edp) { if (dp->d_reclen <= offsetof(struct direct, d_name) || (caddr_t)dp + dp->d_reclen > (caddr_t)edp) { error = EIO; break; } #if BYTE_ORDER == LITTLE_ENDIAN /* Old filesystem format. */ if (OFSFMT(vp)) { dstdp.d_namlen = dp->d_type; dstdp.d_type = dp->d_namlen; } else #endif { dstdp.d_namlen = dp->d_namlen; dstdp.d_type = dp->d_type; } if (offsetof(struct direct, d_name) + dstdp.d_namlen > dp->d_reclen) { error = EIO; break; } if (offset < startoffset || dp->d_ino == 0) goto nextentry; dstdp.d_fileno = dp->d_ino; dstdp.d_reclen = GENERIC_DIRSIZ(&dstdp); bcopy(dp->d_name, dstdp.d_name, dstdp.d_namlen); /* NOTE: d_off is the offset of the *next* entry. */ dstdp.d_off = offset + dp->d_reclen; dirent_terminate(&dstdp); if (dstdp.d_reclen > uio->uio_resid) { if (uio->uio_resid == startresid) error = EINVAL; else error = EJUSTRETURN; break; } /* Advance dp. */ error = uiomove((caddr_t)&dstdp, dstdp.d_reclen, uio); if (error) break; if (cookies != NULL) { KASSERT(ncookies > 0, ("ufs_readdir: cookies buffer too small")); *cookies = offset + dp->d_reclen; cookies++; ncookies--; } nextentry: offset += dp->d_reclen; dp = (struct direct *)((caddr_t)dp + dp->d_reclen); } bqrelse(bp); uio->uio_offset = offset; } /* We need to correct uio_offset. */ uio->uio_offset = offset; if (error == EJUSTRETURN) error = 0; if (ap->a_ncookies != NULL) { if (error == 0) { ap->a_ncookies -= ncookies; } else { free(*ap->a_cookies, M_TEMP); *ap->a_ncookies = 0; *ap->a_cookies = NULL; } } if (error == 0 && ap->a_eofflag) *ap->a_eofflag = ip->i_size <= uio->uio_offset; return (error); } /* * Return target name of a symbolic link */ static int ufs_readlink(ap) struct vop_readlink_args /* { struct vnode *a_vp; struct uio *a_uio; struct ucred *a_cred; } */ *ap; { struct vnode *vp = ap->a_vp; struct inode *ip = VTOI(vp); doff_t isize; isize = ip->i_size; if (isize < VFSTOUFS(vp->v_mount)->um_maxsymlinklen) return (uiomove(SHORTLINK(ip), isize, ap->a_uio)); return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred)); } /* * Calculate the logical to physical mapping if not done already, * then call the device strategy routine. * * In order to be able to swap to a file, the ufs_bmaparray() operation may not * deadlock on memory. See ufs_bmap() for details. */ static int ufs_strategy(ap) struct vop_strategy_args /* { struct vnode *a_vp; struct buf *a_bp; } */ *ap; { struct buf *bp = ap->a_bp; struct vnode *vp = ap->a_vp; ufs2_daddr_t blkno; int error; if (bp->b_blkno == bp->b_lblkno) { error = ufs_bmaparray(vp, bp->b_lblkno, &blkno, bp, NULL, NULL); bp->b_blkno = blkno; if (error) { bp->b_error = error; bp->b_ioflags |= BIO_ERROR; bufdone(bp); return (0); } if ((long)bp->b_blkno == -1) vfs_bio_clrbuf(bp); } if ((long)bp->b_blkno == -1) { bufdone(bp); return (0); } bp->b_iooffset = dbtob(bp->b_blkno); BO_STRATEGY(VFSTOUFS(vp->v_mount)->um_bo, bp); return (0); } /* * Print out the contents of an inode. */ static int ufs_print(ap) struct vop_print_args /* { struct vnode *a_vp; } */ *ap; { struct vnode *vp = ap->a_vp; struct inode *ip = VTOI(vp); printf("\tnlink=%d, effnlink=%d, size=%jd", ip->i_nlink, ip->i_effnlink, (intmax_t)ip->i_size); if (I_IS_UFS2(ip)) printf(", extsize %d", ip->i_din2->di_extsize); printf("\n\tgeneration=%jx, uid=%d, gid=%d, flags=0x%b\n", (uintmax_t)ip->i_gen, ip->i_uid, ip->i_gid, (u_int)ip->i_flags, PRINT_INODE_FLAGS); printf("\tino %lu, on dev %s", (u_long)ip->i_number, devtoname(ITODEV(ip))); if (vp->v_type == VFIFO) fifo_printinfo(vp); printf("\n"); return (0); } /* * Close wrapper for fifos. * * Update the times on the inode then do device close. */ static int ufsfifo_close(ap) struct vop_close_args /* { struct vnode *a_vp; int a_fflag; struct ucred *a_cred; struct thread *a_td; } */ *ap; { struct vnode *vp = ap->a_vp; int usecount; VI_LOCK(vp); usecount = vp->v_usecount; if (usecount > 1) ufs_itimes_locked(vp); VI_UNLOCK(vp); return (fifo_specops.vop_close(ap)); } /* * Return POSIX pathconf information applicable to ufs filesystems. */ static int ufs_pathconf(ap) struct vop_pathconf_args /* { struct vnode *a_vp; int a_name; int *a_retval; } */ *ap; { int error; error = 0; switch (ap->a_name) { case _PC_LINK_MAX: *ap->a_retval = UFS_LINK_MAX; break; case _PC_NAME_MAX: *ap->a_retval = UFS_MAXNAMLEN; break; case _PC_PIPE_BUF: if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO) *ap->a_retval = PIPE_BUF; else error = EINVAL; break; case _PC_CHOWN_RESTRICTED: *ap->a_retval = 1; break; case _PC_NO_TRUNC: *ap->a_retval = 1; break; #ifdef UFS_ACL case _PC_ACL_EXTENDED: if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS) *ap->a_retval = 1; else *ap->a_retval = 0; break; case _PC_ACL_NFS4: if (ap->a_vp->v_mount->mnt_flag & MNT_NFS4ACLS) *ap->a_retval = 1; else *ap->a_retval = 0; break; #endif case _PC_ACL_PATH_MAX: #ifdef UFS_ACL if (ap->a_vp->v_mount->mnt_flag & (MNT_ACLS | MNT_NFS4ACLS)) *ap->a_retval = ACL_MAX_ENTRIES; else *ap->a_retval = 3; #else *ap->a_retval = 3; #endif break; #ifdef MAC case _PC_MAC_PRESENT: if (ap->a_vp->v_mount->mnt_flag & MNT_MULTILABEL) *ap->a_retval = 1; else *ap->a_retval = 0; break; #endif case _PC_MIN_HOLE_SIZE: *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize; break; case _PC_PRIO_IO: *ap->a_retval = 0; break; case _PC_SYNC_IO: *ap->a_retval = 0; break; case _PC_ALLOC_SIZE_MIN: *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize; break; case _PC_FILESIZEBITS: *ap->a_retval = 64; break; case _PC_REC_INCR_XFER_SIZE: *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize; break; case _PC_REC_MAX_XFER_SIZE: *ap->a_retval = -1; /* means ``unlimited'' */ break; case _PC_REC_MIN_XFER_SIZE: *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize; break; case _PC_REC_XFER_ALIGN: *ap->a_retval = PAGE_SIZE; break; case _PC_SYMLINK_MAX: *ap->a_retval = MAXPATHLEN; break; default: error = vop_stdpathconf(ap); break; } return (error); } /* * Initialize the vnode associated with a new inode, handle aliased * vnodes. */ int ufs_vinit(mntp, fifoops, vpp) struct mount *mntp; struct vop_vector *fifoops; struct vnode **vpp; { struct inode *ip; struct vnode *vp; vp = *vpp; ASSERT_VOP_LOCKED(vp, "ufs_vinit"); ip = VTOI(vp); vp->v_type = IFTOVT(ip->i_mode); /* * Only unallocated inodes should be of type VNON. */ if (ip->i_mode != 0 && vp->v_type == VNON) return (EINVAL); if (vp->v_type == VFIFO) vp->v_op = fifoops; if (ip->i_number == UFS_ROOTINO) vp->v_vflag |= VV_ROOT; *vpp = vp; return (0); } /* * Allocate a new inode. * Vnode dvp must be locked. */ static int ufs_makeinode(mode, dvp, vpp, cnp, callfunc) int mode; struct vnode *dvp; struct vnode **vpp; struct componentname *cnp; const char *callfunc; { struct inode *ip, *pdir; struct direct newdir; struct vnode *tvp; int error; pdir = VTOI(dvp); #ifdef INVARIANTS if ((cnp->cn_flags & HASBUF) == 0) panic("%s: no name", callfunc); #endif *vpp = NULL; if ((mode & IFMT) == 0) mode |= IFREG; if (pdir->i_effnlink < 2) { print_bad_link_count(callfunc, dvp); return (EINVAL); } if (DOINGSUJ(dvp)) { error = softdep_prelink(dvp, NULL, cnp); if (error != 0) { MPASS(error == ERELOOKUP); return (error); } } error = UFS_VALLOC(dvp, mode, cnp->cn_cred, &tvp); if (error) return (error); ip = VTOI(tvp); ip->i_gid = pdir->i_gid; DIP_SET(ip, i_gid, pdir->i_gid); #ifdef SUIDDIR { #ifdef QUOTA struct ucred ucred, *ucp; gid_t ucred_group; ucp = cnp->cn_cred; #endif /* * If we are not the owner of the directory, * and we are hacking owners here, (only do this where told to) * and we are not giving it TO root, (would subvert quotas) * then go ahead and give it to the other user. * Note that this drops off the execute bits for security. */ if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) && (pdir->i_mode & ISUID) && (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) { ip->i_uid = pdir->i_uid; DIP_SET(ip, i_uid, ip->i_uid); mode &= ~07111; #ifdef QUOTA /* * Make sure the correct user gets charged * for the space. * Quickly knock up a dummy credential for the victim. * XXX This seems to never be accessed out of our * context so a stack variable is ok. */ refcount_init(&ucred.cr_ref, 1); ucred.cr_uid = ip->i_uid; ucred.cr_ngroups = 1; ucred.cr_groups = &ucred_group; ucred.cr_groups[0] = pdir->i_gid; ucp = &ucred; #endif } else { ip->i_uid = cnp->cn_cred->cr_uid; DIP_SET(ip, i_uid, ip->i_uid); } #ifdef QUOTA if ((error = getinoquota(ip)) || (error = chkiq(ip, 1, ucp, 0))) { if (DOINGSOFTDEP(tvp)) softdep_revert_link(pdir, ip); UFS_VFREE(tvp, ip->i_number, mode); vgone(tvp); vput(tvp); return (error); } #endif } #else /* !SUIDDIR */ ip->i_uid = cnp->cn_cred->cr_uid; DIP_SET(ip, i_uid, ip->i_uid); #ifdef QUOTA if ((error = getinoquota(ip)) || (error = chkiq(ip, 1, cnp->cn_cred, 0))) { if (DOINGSOFTDEP(tvp)) softdep_revert_link(pdir, ip); UFS_VFREE(tvp, ip->i_number, mode); vgone(tvp); vput(tvp); return (error); } #endif #endif /* !SUIDDIR */ vn_seqc_write_begin(tvp); /* Mostly to cover asserts */ UFS_INODE_SET_FLAG(ip, IN_ACCESS | IN_CHANGE | IN_UPDATE); UFS_INODE_SET_MODE(ip, mode); DIP_SET(ip, i_mode, mode); tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */ ip->i_effnlink = 1; ip->i_nlink = 1; DIP_SET(ip, i_nlink, 1); if (DOINGSOFTDEP(tvp)) softdep_setup_create(VTOI(dvp), ip); if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred) && priv_check_cred(cnp->cn_cred, PRIV_VFS_SETGID)) { UFS_INODE_SET_MODE(ip, ip->i_mode & ~ISGID); DIP_SET(ip, i_mode, ip->i_mode); } if (cnp->cn_flags & ISWHITEOUT) { ip->i_flags |= UF_OPAQUE; DIP_SET(ip, i_flags, ip->i_flags); } /* * Make sure inode goes to disk before directory entry. */ error = UFS_UPDATE(tvp, !DOINGSOFTDEP(tvp) && !DOINGASYNC(tvp)); if (error) goto bad; #ifdef MAC if (dvp->v_mount->mnt_flag & MNT_MULTILABEL) { error = mac_vnode_create_extattr(cnp->cn_cred, dvp->v_mount, dvp, tvp, cnp); if (error) goto bad; } #endif #ifdef UFS_ACL if (dvp->v_mount->mnt_flag & MNT_ACLS) { error = ufs_do_posix1e_acl_inheritance_file(dvp, tvp, mode, - cnp->cn_cred, cnp->cn_thread); + cnp->cn_cred, curthread); if (error) goto bad; } else if (dvp->v_mount->mnt_flag & MNT_NFS4ACLS) { error = ufs_do_nfs4_acl_inheritance(dvp, tvp, mode, - cnp->cn_cred, cnp->cn_thread); + cnp->cn_cred, curthread); if (error) goto bad; } #endif /* !UFS_ACL */ ufs_makedirentry(ip, cnp, &newdir); error = ufs_direnter(dvp, tvp, &newdir, cnp, NULL); if (error) goto bad; vn_seqc_write_end(tvp); *vpp = tvp; return (0); bad: /* * Write error occurred trying to update the inode * or the directory so must deallocate the inode. */ ip->i_effnlink = 0; ip->i_nlink = 0; DIP_SET(ip, i_nlink, 0); UFS_INODE_SET_FLAG(ip, IN_CHANGE); if (DOINGSOFTDEP(tvp)) softdep_revert_create(VTOI(dvp), ip); vn_seqc_write_end(tvp); vgone(tvp); vput(tvp); return (error); } static int ufs_ioctl(struct vop_ioctl_args *ap) { struct vnode *vp; int error; vp = ap->a_vp; switch (ap->a_command) { case FIOSEEKDATA: error = vn_lock(vp, LK_SHARED); if (error == 0) { error = ufs_bmap_seekdata(vp, (off_t *)ap->a_data); VOP_UNLOCK(vp); } else error = EBADF; return (error); case FIOSEEKHOLE: return (vn_bmap_seekhole(vp, ap->a_command, (off_t *)ap->a_data, ap->a_cred)); default: return (ENOTTY); } } static int ufs_read_pgcache(struct vop_read_pgcache_args *ap) { struct uio *uio; struct vnode *vp; uio = ap->a_uio; vp = ap->a_vp; VNPASS((vn_irflag_read(vp) & VIRF_PGREAD) != 0, vp); if (uio->uio_resid > ptoa(io_hold_cnt) || uio->uio_offset < 0 || (ap->a_ioflag & IO_DIRECT) != 0) return (EJUSTRETURN); return (vn_read_from_obj(vp, uio)); } /* Global vfs data structures for ufs. */ struct vop_vector ufs_vnodeops = { .vop_default = &default_vnodeops, .vop_fsync = VOP_PANIC, .vop_read = VOP_PANIC, .vop_reallocblks = VOP_PANIC, .vop_write = VOP_PANIC, .vop_accessx = ufs_accessx, .vop_bmap = ufs_bmap, .vop_fplookup_vexec = ufs_fplookup_vexec, .vop_fplookup_symlink = VOP_EAGAIN, .vop_cachedlookup = ufs_lookup, .vop_close = ufs_close, .vop_create = ufs_create, .vop_stat = ufs_stat, .vop_getattr = ufs_getattr, .vop_inactive = ufs_inactive, .vop_ioctl = ufs_ioctl, .vop_link = ufs_link, .vop_lookup = vfs_cache_lookup, .vop_mmapped = ufs_mmapped, .vop_mkdir = ufs_mkdir, .vop_mknod = ufs_mknod, .vop_need_inactive = ufs_need_inactive, .vop_open = ufs_open, .vop_pathconf = ufs_pathconf, .vop_poll = vop_stdpoll, .vop_print = ufs_print, .vop_read_pgcache = ufs_read_pgcache, .vop_readdir = ufs_readdir, .vop_readlink = ufs_readlink, .vop_reclaim = ufs_reclaim, .vop_remove = ufs_remove, .vop_rename = ufs_rename, .vop_rmdir = ufs_rmdir, .vop_setattr = ufs_setattr, #ifdef MAC .vop_setlabel = vop_stdsetlabel_ea, #endif .vop_strategy = ufs_strategy, .vop_symlink = ufs_symlink, .vop_whiteout = ufs_whiteout, #ifdef UFS_EXTATTR .vop_getextattr = ufs_getextattr, .vop_deleteextattr = ufs_deleteextattr, .vop_setextattr = ufs_setextattr, #endif #ifdef UFS_ACL .vop_getacl = ufs_getacl, .vop_setacl = ufs_setacl, .vop_aclcheck = ufs_aclcheck, #endif }; VFS_VOP_VECTOR_REGISTER(ufs_vnodeops); struct vop_vector ufs_fifoops = { .vop_default = &fifo_specops, .vop_fsync = VOP_PANIC, .vop_accessx = ufs_accessx, .vop_close = ufsfifo_close, .vop_getattr = ufs_getattr, .vop_inactive = ufs_inactive, .vop_pathconf = ufs_pathconf, .vop_print = ufs_print, .vop_read = VOP_PANIC, .vop_reclaim = ufs_reclaim, .vop_setattr = ufs_setattr, #ifdef MAC .vop_setlabel = vop_stdsetlabel_ea, #endif .vop_write = VOP_PANIC, #ifdef UFS_EXTATTR .vop_getextattr = ufs_getextattr, .vop_deleteextattr = ufs_deleteextattr, .vop_setextattr = ufs_setextattr, #endif #ifdef UFS_ACL .vop_getacl = ufs_getacl, .vop_setacl = ufs_setacl, .vop_aclcheck = ufs_aclcheck, #endif }; VFS_VOP_VECTOR_REGISTER(ufs_fifoops);