diff --git a/module/os/freebsd/zfs/zfs_vnops_os.c b/module/os/freebsd/zfs/zfs_vnops_os.c index 45cf6fdfc409..513accf18a51 100644 --- a/module/os/freebsd/zfs/zfs_vnops_os.c +++ b/module/os/freebsd/zfs/zfs_vnops_os.c @@ -1,6434 +1,6434 @@ /* * 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 https://opensource.org/licenses/CDDL-1.0. * 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 #include #include #include #include #ifndef VN_OPEN_INVFS #define VN_OPEN_INVFS 0x0 #endif VFS_SMR_DECLARE; #if __FreeBSD_version < 1300103 #define NDFREE_PNBUF(ndp) NDFREE((ndp), NDF_ONLY_PNBUF) #endif #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 #if __FreeBSD_version >= 1400045 typedef uint64_t cookie_t; #else typedef ulong_t cookie_t; #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 */ static int zfs_open(vnode_t **vpp, int flag, cred_t *cr) { (void) cr; znode_t *zp = VTOZ(*vpp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; int error; if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) return (error); if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) && ((flag & FAPPEND) == 0)) { zfs_exit(zfsvfs, FTAG); return (SET_ERROR(EPERM)); } /* Keep a count of the synchronous opens in the znode */ if (flag & O_SYNC) atomic_inc_32(&zp->z_sync_cnt); zfs_exit(zfsvfs, FTAG); return (0); } static int zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr) { (void) offset, (void) cr; znode_t *zp = VTOZ(vp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; int error; if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) return (error); /* Decrement the synchronous opens in the znode */ if ((flag & O_SYNC) && (count == 1)) atomic_dec_32(&zp->z_sync_cnt); zfs_exit(zfsvfs, FTAG); return (0); } static int zfs_ioctl(vnode_t *vp, ulong_t com, intptr_t data, int flag, cred_t *cred, int *rvalp) { (void) flag, (void) cred, (void) 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) memset(va + bytes, 0, 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 */ static int zfs_lookup(vnode_t *dvp, const char *nm, vnode_t **vpp, struct componentname *cnp, int nameiop, cred_t *cr, 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); if ((error = zfs_enter_verify_zp(zfsvfs, zdp, FTAG)) != 0) return (error); #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, FTAG); return (SET_ERROR(EOPNOTSUPP)); } /* * We don't allow recursive attributes.. * Maybe someday we will. */ if (zdp->z_pflags & ZFS_XATTR) { zfs_exit(zfsvfs, FTAG); return (SET_ERROR(EINVAL)); } if ((error = zfs_get_xattrdir(VTOZ(dvp), &zp, cr, flags))) { zfs_exit(zfsvfs, FTAG); return (error); } *vpp = ZTOV(zp); /* * Do we have permission to get into attribute directory? */ error = zfs_zaccess(zp, ACE_EXECUTE, 0, B_FALSE, cr, NULL); if (error) { vrele(ZTOV(zp)); } zfs_exit(zfsvfs, FTAG); 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, NULL))) { zfs_exit(zfsvfs, FTAG); return (error); } } if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { zfs_exit(zfsvfs, FTAG); 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, FTAG); 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, FTAG); 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, FTAG); 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; if ((error = zfs_enter(zfsvfs, FTAG)) != 0) { vput(ZTOV(zp)); *vpp = NULL; return (error); } 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, FTAG); vput(ZTOV(zp)); break; } if (zp->z_id == parent) { zfs_exit(zfsvfs, FTAG); 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; #if __FreeBSD_version < 1400068 cnp->cn_flags |= SAVENAME; #endif break; } zfs_fallthrough; case DELETE: #if __FreeBSD_version < 1400068 if (error == 0) cnp->cn_flags |= SAVENAME; #endif 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 * mnt_ns - Unused on FreeBSD * * 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 */ 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, zidmap_t *mnt_ns) { (void) excl, (void) mode, (void) flag; znode_t *zp; zfsvfs_t *zfsvfs = dzp->z_zfsvfs; zilog_t *zilog; objset_t *os; dmu_tx_t *tx; int error; uid_t uid = crgetuid(cr); 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 */ if (zfsvfs->z_use_fuids == B_FALSE && (vsecp || (vap->va_mask & AT_XVATTR) || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) return (SET_ERROR(EINVAL)); if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0) return (error); 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, FTAG); 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, FTAG); 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, FTAG); 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, mnt_ns))) { 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, NULL)) != 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, FTAG); 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, FTAG); 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) */ 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; if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0) return (error); zp = VTOZ(vp); if ((error = zfs_verify_zp(zp)) != 0) { zfs_exit(zfsvfs, FTAG); return (error); } zilog = zfsvfs->z_log; xattr_obj = 0; xzp = NULL; if ((error = zfs_zaccess_delete(dzp, zp, cr, NULL))) { 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, FTAG); 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, FTAG); 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; #if __FreeBSD_version < 1400068 cnp->cn_flags |= SAVENAME; #endif cnp->cn_lkflags = LK_EXCLUSIVE | LK_RETRY; cnp->cn_cred = kcred; #if __FreeBSD_version < 1400037 cnp->cn_thread = curthread; #endif 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, 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 * mnt_ns - Unused on FreeBSD * * OUT: vpp - vnode of created directory. * * RETURN: 0 on success, error code on failure. * * Timestamps: * dvp - ctime|mtime updated * vp - ctime|mtime|atime updated */ int zfs_mkdir(znode_t *dzp, const char *dirname, vattr_t *vap, znode_t **zpp, cred_t *cr, int flags, vsecattr_t *vsecp, zidmap_t *mnt_ns) { (void) flags, (void) vsecp; znode_t *zp; zfsvfs_t *zfsvfs = dzp->z_zfsvfs; zilog_t *zilog; uint64_t txtype; dmu_tx_t *tx; int error; uid_t uid = crgetuid(cr); 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 */ if (zfsvfs->z_use_fuids == B_FALSE && ((vap->va_mask & AT_XVATTR) || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) return (SET_ERROR(EINVAL)); if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0) return (error); zilog = zfsvfs->z_log; if (dzp->z_pflags & ZFS_XATTR) { zfs_exit(zfsvfs, FTAG); return (SET_ERROR(EINVAL)); } if (zfsvfs->z_utf8 && u8_validate(dirname, strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { zfs_exit(zfsvfs, FTAG); 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, FTAG); return (error); } } if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, NULL, &acl_ids, NULL)) != 0) { zfs_exit(zfsvfs, FTAG); 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, FTAG); return (error); } ASSERT3P(zp, ==, NULL); if ((error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr, mnt_ns))) { zfs_acl_ids_free(&acl_ids); zfs_exit(zfsvfs, FTAG); return (error); } if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, zfs_inherit_projid(dzp))) { zfs_acl_ids_free(&acl_ids); zfs_exit(zfsvfs, FTAG); 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, FTAG); 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, FTAG); 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 */ 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; if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0) return (error); if ((error = zfs_verify_zp(zp)) != 0) { zfs_exit(zfsvfs, FTAG); return (error); } zilog = zfsvfs->z_log; if ((error = zfs_zaccess_delete(dzp, zp, cr, NULL))) { 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, FTAG); 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); if (zfsvfs->z_use_namecache) cache_vop_rmdir(dvp, vp); out: if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) zil_commit(zilog, 0); zfs_exit(zfsvfs, FTAG); 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 * * OUT: uio - updated offset and range, buffer filled. * eofp - set to true if end-of-file detected. * ncookies- number of entries in cookies * cookies - offsets to directory entries * * 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. */ static int zfs_readdir(vnode_t *vp, zfs_uio_t *uio, cred_t *cr, int *eofp, int *ncookies, cookie_t **cookies) { znode_t *zp = VTOZ(vp); iovec_t *iovp; 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; uint8_t type; int ncooks; cookie_t *cooks = NULL; if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) return (error); if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0) { zfs_exit(zfsvfs, FTAG); 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, FTAG); return (SET_ERROR(EINVAL)); } /* * Quit if directory has been removed (posix) */ if ((*eofp = zp->z_unlinked) != 0) { zfs_exit(zfsvfs, FTAG); 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; } 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 (*cooks), M_TEMP, M_WAITOK); *cookies = cooks; *ncookies = ncooks; } /* * 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); } 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; } /* * 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); /* * 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, FTAG); 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). */ 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; if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) return (error); 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, FTAG); 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, NULL))) { zfs_exit(zfsvfs, FTAG); 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_gen = zp->z_gen; 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, FTAG); 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. * mnt_ns - Unused on FreeBSD * * RETURN: 0 on success, error code on failure. * * Timestamps: * vp - ctime updated, mtime updated if size changed. */ int zfs_setattr(znode_t *zp, vattr_t *vap, int flags, cred_t *cr, zidmap_t *mnt_ns) { vnode_t *vp = ZTOV(zp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; objset_t *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)); if ((err = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) return (err); os = zfsvfs->z_os; 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, FTAG); return (SET_ERROR(EINVAL)); } if (mask & AT_SIZE && vp->v_type == VDIR) { zfs_exit(zfsvfs, FTAG); return (SET_ERROR(EISDIR)); } if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) { zfs_exit(zfsvfs, FTAG); 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, FTAG); 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, FTAG); 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, FTAG); 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, FTAG); return (SET_ERROR(EOPNOTSUPP)); } projid = xoap->xoa_projid; if (unlikely(projid == ZFS_INVALID_PROJID)) { zfs_exit(zfsvfs, FTAG); 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, FTAG); return (SET_ERROR(EOPNOTSUPP)); } } attrzp = NULL; aclp = NULL; if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { zfs_exit(zfsvfs, FTAG); 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, FTAG); 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, mnt_ns); } 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, mnt_ns) == 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, FTAG); 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, mnt_ns) == 0) { err = secpolicy_setid_setsticky_clear(vp, vap, &oldva, cr); if (err) { zfs_exit(zfsvfs, FTAG); 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, FTAG); 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, FTAG); return (err); } /* * Look up the directory entries corresponding to the source and target * directory/name pairs. */ static int zfs_rename_relock_lookup(znode_t *sdzp, const struct componentname *scnp, znode_t **szpp, znode_t *tdzp, const struct componentname *tcnp, znode_t **tzpp) { zfsvfs_t *zfsvfs; znode_t *szp, *tzp; int error; /* * 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); if ((error = zfs_enter_verify_zp(zfsvfs, sdzp, FTAG)) != 0) return (error); if ((error = zfs_verify_zp(tdzp)) != 0) { zfs_exit(zfsvfs, FTAG); return (error); } /* * Re-resolve svp to be certain it still exists and fetch the * correct vnode. */ error = zfs_dirent_lookup(sdzp, scnp->cn_nameptr, &szp, ZEXISTS); if (error != 0) { /* Source entry invalid or not there. */ if ((scnp->cn_flags & ISDOTDOT) != 0 || (scnp->cn_namelen == 1 && scnp->cn_nameptr[0] == '.')) error = SET_ERROR(EINVAL); goto out; } *szpp = szp; /* * Re-resolve tvp, if it disappeared we just carry on. */ error = zfs_dirent_lookup(tdzp, tcnp->cn_nameptr, &tzp, 0); if (error != 0) { vrele(ZTOV(szp)); if ((tcnp->cn_flags & ISDOTDOT) != 0) error = SET_ERROR(EINVAL); goto out; } *tzpp = tzp; out: zfs_exit(zfsvfs, FTAG); 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 zfs_rename_relock(struct vnode *sdvp, struct vnode **svpp, struct vnode *tdvp, struct vnode **tvpp, const struct componentname *scnp, const struct componentname *tcnp) { struct vnode *nvp, *svp, *tvp; znode_t *sdzp, *tdzp, *szp, *tzp; int error; VOP_UNLOCK1(tdvp); if (*tvpp != NULL && *tvpp != tdvp) VOP_UNLOCK1(*tvpp); relock: error = vn_lock(sdvp, LK_EXCLUSIVE); if (error) goto out; 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); sdzp = VTOZ(sdvp); error = zfs_rename_relock_lookup(sdzp, scnp, &szp, tdzp, tcnp, &tzp); if (error != 0) { VOP_UNLOCK1(sdvp); VOP_UNLOCK1(tdvp); goto out; } svp = ZTOV(szp); tvp = tzp != NULL ? ZTOV(tzp) : NULL; /* * 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 static int zfs_do_rename_impl(vnode_t *sdvp, vnode_t **svpp, struct componentname *scnp, vnode_t *tdvp, vnode_t **tvpp, struct componentname *tcnp, cred_t *cr); /* * 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". * scnp - Old entry name. * tdvp - Target directory to contain the "new entry". * tcnp - New entry name. * cr - credentials of caller. * INOUT: svpp - Source file * tvpp - Target file, may point to NULL initially * * RETURN: 0 on success, error code on failure. * * Timestamps: * sdvp,tdvp - ctime|mtime updated */ static int zfs_do_rename(vnode_t *sdvp, vnode_t **svpp, struct componentname *scnp, vnode_t *tdvp, vnode_t **tvpp, struct componentname *tcnp, cred_t *cr) { int error; ASSERT_VOP_ELOCKED(tdvp, __func__); if (*tvpp != NULL) ASSERT_VOP_ELOCKED(*tvpp, __func__); /* 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); } error = zfs_do_rename_impl(sdvp, svpp, scnp, tdvp, tvpp, tcnp, cr); VOP_UNLOCK1(sdvp); VOP_UNLOCK1(*svpp); out: if (*tvpp != NULL) VOP_UNLOCK1(*tvpp); if (tdvp != *tvpp) VOP_UNLOCK1(tdvp); return (error); } static int zfs_do_rename_impl(vnode_t *sdvp, vnode_t **svpp, struct componentname *scnp, vnode_t *tdvp, vnode_t **tvpp, struct componentname *tcnp, cred_t *cr) { dmu_tx_t *tx; zfsvfs_t *zfsvfs; zilog_t *zilog; znode_t *tdzp, *sdzp, *tzp, *szp; const char *snm = scnp->cn_nameptr; const char *tnm = tcnp->cn_nameptr; int error; tdzp = VTOZ(tdvp); sdzp = VTOZ(sdvp); zfsvfs = tdzp->z_zfsvfs; if ((error = zfs_enter_verify_zp(zfsvfs, tdzp, FTAG)) != 0) return (error); if ((error = zfs_verify_zp(sdzp)) != 0) { zfs_exit(zfsvfs, FTAG); return (error); } zilog = zfsvfs->z_log; if (zfsvfs->z_utf8 && u8_validate(tnm, strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { error = SET_ERROR(EILSEQ); goto out; } /* If source and target are the same file, there is nothing to do. */ if ((*svpp) == (*tvpp)) { error = 0; goto out; } if (((*svpp)->v_type == VDIR && (*svpp)->v_mountedhere != NULL) || ((*tvpp) != NULL && (*tvpp)->v_type == VDIR && (*tvpp)->v_mountedhere != NULL)) { error = SET_ERROR(EXDEV); goto out; } szp = VTOZ(*svpp); if ((error = zfs_verify_zp(szp)) != 0) { zfs_exit(zfsvfs, FTAG); return (error); } tzp = *tvpp == NULL ? NULL : VTOZ(*tvpp); if (tzp != NULL) { if ((error = zfs_verify_zp(tzp)) != 0) { zfs_exit(zfsvfs, FTAG); return (error); } } /* * 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 out; } /* * 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 out; } /* * 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, NULL))) goto out; 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 out; } /* * 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 out; } /* * 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 out; } else { cache_purge(tdvp); if (sdvp != tdvp) cache_purge(sdvp); } } else { if ((*tvpp)->v_type == VDIR) { error = SET_ERROR(EISDIR); goto out; } } } 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); 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 out_seq; } 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); } 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); out_seq: 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); out: if (error == 0 && zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) zil_commit(zilog, 0); zfs_exit(zfsvfs, FTAG); return (error); } int zfs_rename(znode_t *sdzp, const char *sname, znode_t *tdzp, const char *tname, cred_t *cr, int flags, uint64_t rflags, vattr_t *wo_vap, zidmap_t *mnt_ns) { struct componentname scn, tcn; vnode_t *sdvp, *tdvp; vnode_t *svp, *tvp; int error; svp = tvp = NULL; if (rflags != 0 || wo_vap != NULL) return (SET_ERROR(EINVAL)); 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_do_rename(sdvp, &svp, &scn, tdvp, &tvp, &tcn, cr); 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 * mnt_ns - Unused on FreeBSD * * RETURN: 0 on success, error code on failure. * * Timestamps: * dvp - ctime|mtime updated */ int zfs_symlink(znode_t *dzp, const char *name, vattr_t *vap, const char *link, znode_t **zpp, cred_t *cr, int flags, zidmap_t *mnt_ns) { (void) 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); if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0) return (error); zilog = zfsvfs->z_log; if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { zfs_exit(zfsvfs, FTAG); return (SET_ERROR(EILSEQ)); } if (len > MAXPATHLEN) { zfs_exit(zfsvfs, FTAG); return (SET_ERROR(ENAMETOOLONG)); } if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, NULL, &acl_ids, NULL)) != 0) { zfs_exit(zfsvfs, FTAG); 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, FTAG); return (error); } if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr, mnt_ns))) { zfs_acl_ids_free(&acl_ids); zfs_exit(zfsvfs, FTAG); return (error); } if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, 0 /* projid */)) { zfs_acl_ids_free(&acl_ids); zfs_exit(zfsvfs, FTAG); 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, FTAG); 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, FTAG); 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 */ static int zfs_readlink(vnode_t *vp, zfs_uio_t *uio, cred_t *cr, caller_context_t *ct) { (void) cr, (void) ct; znode_t *zp = VTOZ(vp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; int error; if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) return (error); 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, FTAG); 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 */ int zfs_link(znode_t *tdzp, znode_t *szp, const char *name, cred_t *cr, int flags) { (void) 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); if ((error = zfs_enter_verify_zp(zfsvfs, tdzp, FTAG)) != 0) return (error); 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, FTAG); return (SET_ERROR(EPERM)); } if ((error = zfs_verify_zp(szp)) != 0) { zfs_exit(zfsvfs, FTAG); return (error); } /* * 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, FTAG); return (SET_ERROR(EXDEV)); } if (szp->z_pflags & (ZFS_APPENDONLY | ZFS_IMMUTABLE | ZFS_READONLY)) { zfs_exit(zfsvfs, FTAG); 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, FTAG); return (error); } if (parent == zfsvfs->z_shares_dir) { zfs_exit(zfsvfs, FTAG); return (SET_ERROR(EPERM)); } if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { zfs_exit(zfsvfs, FTAG); 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, FTAG); 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, FTAG); return (SET_ERROR(EPERM)); } if ((error = zfs_zaccess(tdzp, ACE_ADD_FILE, 0, B_FALSE, cr, NULL))) { zfs_exit(zfsvfs, FTAG); return (error); } /* * Attempt to lock directory; fail if entry already exists. */ error = zfs_dirent_lookup(tdzp, name, &tzp, ZNEW); if (error) { zfs_exit(zfsvfs, FTAG); 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, FTAG); 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, FTAG); 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 */ int zfs_space(znode_t *zp, int cmd, flock64_t *bfp, int flag, offset_t offset, cred_t *cr) { (void) offset; zfsvfs_t *zfsvfs = ZTOZSB(zp); uint64_t off, len; int error; if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) return (error); if (cmd != F_FREESP) { zfs_exit(zfsvfs, FTAG); 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, FTAG); return (SET_ERROR(EROFS)); } if (bfp->l_len < 0) { zfs_exit(zfsvfs, FTAG); 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, NULL))) { zfs_exit(zfsvfs, FTAG); 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, FTAG); return (error); } static void zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct) { (void) cr, (void) 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); } _Static_assert(sizeof (struct zfid_short) <= sizeof (struct fid), "struct zfid_short bigger than struct fid"); _Static_assert(sizeof (struct zfid_long) <= sizeof (struct fid), "struct zfid_long bigger than struct fid"); static int zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct) { (void) 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; if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) return (error); if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &gen64, sizeof (uint64_t))) != 0) { zfs_exit(zfsvfs, FTAG); 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, FTAG); 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; int error; 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; if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) return (error); *valp = zfsvfs->z_acl_type == ZFSACLTYPE_POSIX ? 1 : 0; zfs_exit(zfsvfs, FTAG); #else *valp = 0; #endif return (0); case _PC_ACL_NFS4: zp = VTOZ(vp); zfsvfs = zp->z_zfsvfs; if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) return (error); *valp = zfsvfs->z_acl_type == ZFS_ACLTYPE_NFSV4 ? 1 : 0; zfs_exit(zfsvfs, FTAG); 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; 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; if (zfs_enter_verify_zp(zfsvfs, zp, FTAG) != 0) return (zfs_vm_pagerret_error); 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, FTAG); 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(zfsvfs->z_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); dataset_kstats_update_read_kstats(&zfsvfs->z_kstat, count*PAGE_SIZE); zfs_exit(zfsvfs, FTAG); 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; object = vp->v_object; KASSERT(ma[0]->object == object, ("mismatching object")); KASSERT(len > 0 && (len & PAGE_MASK) == 0, ("unexpected length")); pcount = btoc(len); ncount = pcount; for (i = 0; i < pcount; i++) rtvals[i] = zfs_vm_pagerret_error; if (zfs_enter_verify_zp(zfsvfs, zp, FTAG) != 0) return (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); dataset_kstats_update_write_kstats(&zfsvfs->z_kstat, len); zfs_exit(zfsvfs, FTAG); 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 |= O_APPEND; if (ioflags & IO_NDELAY) flags |= O_NONBLOCK; if (ioflags & IO_SYNC) flags |= O_SYNC; 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, 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; #if __FreeBSD_version < 1400068 ASSERT(cnp->cn_flags & SAVENAME); #endif 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, 0, mode, &zp, cnp->cn_cred, 0 /* flag */, NULL /* vsecattr */, NULL); 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) { #if __FreeBSD_version < 1400068 ASSERT(ap->a_cnp->cn_flags & SAVENAME); #endif 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; #if __FreeBSD_version < 1400068 ASSERT(ap->a_cnp->cn_flags & SAVENAME); #endif 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, 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; #if __FreeBSD_version < 1400068 ASSERT(cnp->cn_flags & SAVENAME); #endif 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; cookie_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) { 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, NULL)); } #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; #if __FreeBSD_version < 1400068 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART)); ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART)); #endif error = zfs_do_rename(fdvp, &fvp, ap->a_fcnp, tdvp, &tvp, ap->a_tcnp, ap->a_fcnp->cn_cred); 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; #if __FreeBSD_version < 1400068 ASSERT(cnp->cn_flags & SAVENAME); #endif 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 */, NULL); 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); #if __FreeBSD_version < 1400068 ASSERT(cnp->cn_flags & SAVENAME); #endif 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)); } } static int zfs_xattr_compat = 1; static int zfs_check_attrname(const char *name) { /* 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 a namespace prefix. */ if (ZFS_XA_NS_PREFIX_FORBIDDEN(name)) return (SET_ERROR(EINVAL)); return (0); } /* * FreeBSD's extended attributes namespace defines file name prefix for ZFS' * extended attribute name: * * NAMESPACE XATTR_COMPAT PREFIX * system * freebsd:system: * user 1 (none, can be used to access ZFS * fsattr(5) attributes created on Solaris) * user 0 user. */ static int zfs_create_attrname(int attrnamespace, const char *name, char *attrname, size_t size, boolean_t compat) { const char *namespace, *prefix, *suffix; memset(attrname, 0, size); switch (attrnamespace) { case EXTATTR_NAMESPACE_USER: if (compat) { /* * This is the default namespace by which we can access * all attributes created on Solaris. */ prefix = namespace = suffix = ""; } else { /* * This is compatible with the user namespace encoding * on Linux prior to xattr_compat, but nothing * else. */ prefix = ""; namespace = "user"; suffix = "."; } 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, LOOKUP_XATTR, B_FALSE); if (error != 0) return (error); flags = FREAD; #if __FreeBSD_version < 1400043 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, xvp, td); #else NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, xvp); #endif error = vn_open_cred(&nd, &flags, 0, VN_OPEN_INVFS, ap->a_cred, NULL); if (error != 0) return (SET_ERROR(error)); vp = nd.ni_vp; NDFREE_PNBUF(&nd); 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); } static int zfs_getextattr_impl(struct vop_getextattr_args *ap, boolean_t compat) { znode_t *zp = VTOZ(ap->a_vp); zfsvfs_t *zfsvfs = ZTOZSB(zp); char attrname[EXTATTR_MAXNAMELEN+1]; int error; error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, sizeof (attrname), compat); if (error != 0) return (error); error = ENOENT; if (zfsvfs->z_use_sa && zp->z_is_sa) error = zfs_getextattr_sa(ap, attrname); if (error == ENOENT) error = zfs_getextattr_dir(ap, attrname); return (error); } /* * 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); 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_check_attrname(ap->a_name); if (error != 0) return (error); if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) return (error); error = ENOENT; rw_enter(&zp->z_xattr_lock, RW_READER); error = zfs_getextattr_impl(ap, zfs_xattr_compat); if ((error == ENOENT || error == ENOATTR) && ap->a_attrnamespace == EXTATTR_NAMESPACE_USER) { /* * Fall back to the alternate namespace format if we failed to * find a user xattr. */ error = zfs_getextattr_impl(ap, !zfs_xattr_compat); } rw_exit(&zp->z_xattr_lock); zfs_exit(zfsvfs, FTAG); 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 nameidata nd; vnode_t *xvp = NULL, *vp; int error; error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, LOOKUP_XATTR, B_FALSE); if (error != 0) return (error); #if __FreeBSD_version < 1400043 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF, UIO_SYSSPACE, attrname, xvp, ap->a_td); #else NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF, UIO_SYSSPACE, attrname, xvp); #endif error = namei(&nd); if (error != 0) return (SET_ERROR(error)); vp = nd.ni_vp; error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd); NDFREE_PNBUF(&nd); 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, attrname, NULL, 0); if (error != 0) { zp->z_xattr_cached = NULL; nvlist_free(nvl); } return (error); } static int zfs_deleteextattr_impl(struct vop_deleteextattr_args *ap, boolean_t compat) { znode_t *zp = VTOZ(ap->a_vp); zfsvfs_t *zfsvfs = ZTOZSB(zp); char attrname[EXTATTR_MAXNAMELEN+1]; int error; error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, sizeof (attrname), compat); if (error != 0) return (error); error = ENOENT; if (zfsvfs->z_use_sa && zp->z_is_sa) error = zfs_deleteextattr_sa(ap, attrname); if (error == ENOENT) error = zfs_deleteextattr_dir(ap, attrname); 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); 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_check_attrname(ap->a_name); if (error != 0) return (error); if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) return (error); rw_enter(&zp->z_xattr_lock, RW_WRITER); error = zfs_deleteextattr_impl(ap, zfs_xattr_compat); if ((error == ENOENT || error == ENOATTR) && ap->a_attrnamespace == EXTATTR_NAMESPACE_USER) { /* * Fall back to the alternate namespace format if we failed to * find a user xattr. */ error = zfs_deleteextattr_impl(ap, !zfs_xattr_compat); } rw_exit(&zp->z_xattr_lock); zfs_exit(zfsvfs, FTAG); 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, LOOKUP_XATTR | CREATE_XATTR_DIR, B_FALSE); if (error != 0) return (error); flags = FFLAGS(O_WRONLY | O_CREAT); #if __FreeBSD_version < 1400043 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, xvp, td); #else NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, xvp); #endif error = vn_open_cred(&nd, &flags, 0600, VN_OPEN_INVFS, ap->a_cred, NULL); if (error != 0) return (SET_ERROR(error)); vp = nd.ni_vp; NDFREE_PNBUF(&nd); 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); } if (error == 0) error = zfs_sa_set_xattr(zp, attrname, buf, entry_size); kmem_free(buf, entry_size); if (error != 0) { zp->z_xattr_cached = NULL; nvlist_free(nvl); } return (error); } static int zfs_setextattr_impl(struct vop_setextattr_args *ap, boolean_t compat) { znode_t *zp = VTOZ(ap->a_vp); zfsvfs_t *zfsvfs = ZTOZSB(zp); char attrname[EXTATTR_MAXNAMELEN+1]; int error; error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, sizeof (attrname), compat); if (error != 0) return (error); struct vop_deleteextattr_args vda = { .a_vp = ap->a_vp, .a_attrnamespace = ap->a_attrnamespace, .a_name = ap->a_name, .a_cred = ap->a_cred, .a_td = ap->a_td, }; error = ENOENT; 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 != 0) { 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); } } if (error == 0 && ap->a_attrnamespace == EXTATTR_NAMESPACE_USER) { /* * Also clear all versions of the alternate compat name. */ zfs_deleteextattr_impl(&vda, !compat); } 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); 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_check_attrname(ap->a_name); if (error != 0) return (error); if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) return (error); rw_enter(&zp->z_xattr_lock, RW_WRITER); error = zfs_setextattr_impl(ap, zfs_xattr_compat); rw_exit(&zp->z_xattr_lock); zfs_exit(zfsvfs, FTAG); 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, 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); } #if __FreeBSD_version < 1400043 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED, UIO_SYSSPACE, ".", xvp, td); #else NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED, UIO_SYSSPACE, ".", xvp); #endif error = namei(&nd); if (error != 0) return (SET_ERROR(error)); vp = nd.ni_vp; NDFREE_PNBUF(&nd); 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 && ZFS_XA_NS_PREFIX_FORBIDDEN(dp->d_name)) 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 && ZFS_XA_NS_PREFIX_FORBIDDEN(name)) 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); } static int zfs_listextattr_impl(struct vop_listextattr_args *ap, boolean_t compat) { znode_t *zp = VTOZ(ap->a_vp); zfsvfs_t *zfsvfs = ZTOZSB(zp); char attrprefix[16]; int error; error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix, sizeof (attrprefix), compat); if (error != 0) return (error); if (zfsvfs->z_use_sa && zp->z_is_sa) error = zfs_listextattr_sa(ap, attrprefix); if (error == 0) error = zfs_listextattr_dir(ap, attrprefix); 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); 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)); if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) return (error); rw_enter(&zp->z_xattr_lock, RW_READER); error = zfs_listextattr_impl(ap, zfs_xattr_compat); if (error == 0 && ap->a_attrnamespace == EXTATTR_NAMESPACE_USER) { /* Also list user xattrs with the alternate format. */ error = zfs_listextattr_impl(ap, !zfs_xattr_compat); } rw_exit(&zp->z_xattr_lock); zfs_exit(zfsvfs, FTAG); 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; if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) return (error); /* * 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; memcpy(ap->a_buf + *ap->a_buflen, name, len); *ap->a_vpp = ZTOV(dzp); } zfs_exit(zfsvfs, FTAG); return (error); } zfs_exit(zfsvfs, FTAG); 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; if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) return (error); /* * 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, FTAG); 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, FTAG); 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, FTAG); return (error); } #endif #ifndef _SYS_SYSPROTO_H_ struct vop_copy_file_range_args { 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; } #endif /* * TODO: FreeBSD will only call file system-specific copy_file_range() if both * files resides under the same mountpoint. In case of ZFS we want to be called * even is files are in different datasets (but on the same pools, but we need * to check that ourselves). */ static int zfs_freebsd_copy_file_range(struct vop_copy_file_range_args *ap) { zfsvfs_t *outzfsvfs; struct vnode *invp = ap->a_invp; struct vnode *outvp = ap->a_outvp; struct mount *mp; struct uio io; int error; uint64_t len = *ap->a_lenp; /* * TODO: If offset/length is not aligned to recordsize, use * vn_generic_copy_file_range() on this fragment. * It would be better to do this after we lock the vnodes, but then we * need something else than vn_generic_copy_file_range(). */ vn_start_write(outvp, &mp, V_WAIT); if (__predict_true(mp == outvp->v_mount)) { outzfsvfs = (zfsvfs_t *)mp->mnt_data; if (!spa_feature_is_enabled(dmu_objset_spa(outzfsvfs->z_os), SPA_FEATURE_BLOCK_CLONING)) { goto bad_write_fallback; } } if (invp == outvp) { if (vn_lock(outvp, LK_EXCLUSIVE) != 0) { goto bad_write_fallback; } } else { #if (__FreeBSD_version >= 1302506 && __FreeBSD_version < 1400000) || \ __FreeBSD_version >= 1400086 vn_lock_pair(invp, false, LK_EXCLUSIVE, outvp, false, LK_EXCLUSIVE); #else vn_lock_pair(invp, false, outvp, false); #endif if (VN_IS_DOOMED(invp) || VN_IS_DOOMED(outvp)) { goto bad_locked_fallback; } } #ifdef MAC error = mac_vnode_check_write(curthread->td_ucred, ap->a_outcred, outvp); if (error != 0) goto out_locked; #endif io.uio_offset = *ap->a_outoffp; io.uio_resid = *ap->a_lenp; error = vn_rlimit_fsize(outvp, &io, ap->a_fsizetd); if (error != 0) goto out_locked; error = zfs_clone_range(VTOZ(invp), ap->a_inoffp, VTOZ(outvp), ap->a_outoffp, &len, ap->a_outcred); - if (error == EXDEV || error == EOPNOTSUPP) + if (error == EXDEV || error == EINVAL || error == EOPNOTSUPP) goto bad_locked_fallback; *ap->a_lenp = (size_t)len; out_locked: if (invp != outvp) VOP_UNLOCK(invp); VOP_UNLOCK(outvp); if (mp != NULL) vn_finished_write(mp); return (error); bad_locked_fallback: if (invp != outvp) VOP_UNLOCK(invp); VOP_UNLOCK(outvp); bad_write_fallback: if (mp != NULL) vn_finished_write(mp); 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); } 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 #if __FreeBSD_version >= 1400043 .vop_add_writecount = vop_stdadd_writecount_nomsync, #endif .vop_copy_file_range = zfs_freebsd_copy_file_range, }; 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, #if __FreeBSD_version >= 1400043 .vop_add_writecount = vop_stdadd_writecount_nomsync, #endif }; 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, #if __FreeBSD_version >= 1400043 .vop_add_writecount = vop_stdadd_writecount_nomsync, #endif }; VFS_VOP_VECTOR_REGISTER(zfs_shareops); ZFS_MODULE_PARAM(zfs, zfs_, xattr_compat, INT, ZMOD_RW, "Use legacy ZFS xattr naming for writing new user namespace xattrs"); diff --git a/module/os/linux/zfs/zpl_file_range.c b/module/os/linux/zfs/zpl_file_range.c index 72384b638be5..43ba9a498202 100644 --- a/module/os/linux/zfs/zpl_file_range.c +++ b/module/os/linux/zfs/zpl_file_range.c @@ -1,271 +1,271 @@ /* * 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 https://opensource.org/licenses/CDDL-1.0. * 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) 2023, Klara Inc. */ #ifdef CONFIG_COMPAT #include #endif #include #include #include #include #include /* * Clone part of a file via block cloning. * * Note that we are not required to update file offsets; the kernel will take * care of that depending on how it was called. */ static ssize_t __zpl_clone_file_range(struct file *src_file, loff_t src_off, struct file *dst_file, loff_t dst_off, size_t len) { struct inode *src_i = file_inode(src_file); struct inode *dst_i = file_inode(dst_file); uint64_t src_off_o = (uint64_t)src_off; uint64_t dst_off_o = (uint64_t)dst_off; uint64_t len_o = (uint64_t)len; cred_t *cr = CRED(); fstrans_cookie_t cookie; int err; if (!spa_feature_is_enabled( dmu_objset_spa(ITOZSB(dst_i)->z_os), SPA_FEATURE_BLOCK_CLONING)) return (-EOPNOTSUPP); if (src_i != dst_i) spl_inode_lock_shared(src_i); spl_inode_lock(dst_i); crhold(cr); cookie = spl_fstrans_mark(); err = -zfs_clone_range(ITOZ(src_i), &src_off_o, ITOZ(dst_i), &dst_off_o, &len_o, cr); spl_fstrans_unmark(cookie); crfree(cr); spl_inode_unlock(dst_i); if (src_i != dst_i) spl_inode_unlock_shared(src_i); if (err < 0) return (err); return ((ssize_t)len_o); } #if defined(HAVE_VFS_COPY_FILE_RANGE) || \ defined(HAVE_VFS_FILE_OPERATIONS_EXTEND) /* * Entry point for copy_file_range(). Copy len bytes from src_off in src_file * to dst_off in dst_file. We are permitted to do this however we like, so we * try to just clone the blocks, and if we can't support it, fall back to the * kernel's generic byte copy function. */ ssize_t zpl_copy_file_range(struct file *src_file, loff_t src_off, struct file *dst_file, loff_t dst_off, size_t len, unsigned int flags) { ssize_t ret; if (flags != 0) return (-EINVAL); /* Try to do it via zfs_clone_range() */ ret = __zpl_clone_file_range(src_file, src_off, dst_file, dst_off, len); #ifdef HAVE_VFS_GENERIC_COPY_FILE_RANGE /* * Since Linux 5.3 the filesystem driver is responsible for executing * an appropriate fallback, and a generic fallback function is provided. */ - if (ret == -EOPNOTSUPP || ret == -EXDEV) + if (ret == -EOPNOTSUPP || ret == -EINVAL || ret == -EXDEV) ret = generic_copy_file_range(src_file, src_off, dst_file, dst_off, len, flags); #else /* * Before Linux 5.3 the filesystem has to return -EOPNOTSUPP to signal * to the kernel that it should fallback to a content copy. */ - if (ret == -EXDEV) + if (ret == -EINVAL || ret == -EXDEV) ret = -EOPNOTSUPP; #endif /* HAVE_VFS_GENERIC_COPY_FILE_RANGE */ return (ret); } #endif /* HAVE_VFS_COPY_FILE_RANGE || HAVE_VFS_FILE_OPERATIONS_EXTEND */ #ifdef HAVE_VFS_REMAP_FILE_RANGE /* * Entry point for FICLONE/FICLONERANGE/FIDEDUPERANGE. * * FICLONE and FICLONERANGE are basically the same as copy_file_range(), except * that they must clone - they cannot fall back to copying. FICLONE is exactly * FICLONERANGE, for the entire file. We don't need to try to tell them apart; * the kernel will sort that out for us. * * FIDEDUPERANGE is for turning a non-clone into a clone, that is, compare the * range in both files and if they're the same, arrange for them to be backed * by the same storage. */ loff_t zpl_remap_file_range(struct file *src_file, loff_t src_off, struct file *dst_file, loff_t dst_off, loff_t len, unsigned int flags) { if (flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_CAN_SHORTEN)) return (-EINVAL); /* * REMAP_FILE_CAN_SHORTEN lets us know we can clone less than the given * range if we want. Its designed for filesystems that make data past * EOF available, and don't want it to be visible in both files. ZFS * doesn't do that, so we just turn the flag off. */ flags &= ~REMAP_FILE_CAN_SHORTEN; if (flags & REMAP_FILE_DEDUP) /* No support for dedup yet */ return (-EOPNOTSUPP); /* Zero length means to clone everything to the end of the file */ if (len == 0) len = i_size_read(file_inode(src_file)) - src_off; return (__zpl_clone_file_range(src_file, src_off, dst_file, dst_off, len)); } #endif /* HAVE_VFS_REMAP_FILE_RANGE */ #if defined(HAVE_VFS_CLONE_FILE_RANGE) || \ defined(HAVE_VFS_FILE_OPERATIONS_EXTEND) /* * Entry point for FICLONE and FICLONERANGE, before Linux 4.20. */ int zpl_clone_file_range(struct file *src_file, loff_t src_off, struct file *dst_file, loff_t dst_off, uint64_t len) { /* Zero length means to clone everything to the end of the file */ if (len == 0) len = i_size_read(file_inode(src_file)) - src_off; return (__zpl_clone_file_range(src_file, src_off, dst_file, dst_off, len)); } #endif /* HAVE_VFS_CLONE_FILE_RANGE || HAVE_VFS_FILE_OPERATIONS_EXTEND */ #ifdef HAVE_VFS_DEDUPE_FILE_RANGE /* * Entry point for FIDEDUPERANGE, before Linux 4.20. */ int zpl_dedupe_file_range(struct file *src_file, loff_t src_off, struct file *dst_file, loff_t dst_off, uint64_t len) { /* No support for dedup yet */ return (-EOPNOTSUPP); } #endif /* HAVE_VFS_DEDUPE_FILE_RANGE */ /* Entry point for FICLONE, before Linux 4.5. */ long zpl_ioctl_ficlone(struct file *dst_file, void *arg) { unsigned long sfd = (unsigned long)arg; struct file *src_file = fget(sfd); if (src_file == NULL) return (-EBADF); if (dst_file->f_op != src_file->f_op) return (-EXDEV); size_t len = i_size_read(file_inode(src_file)); ssize_t ret = __zpl_clone_file_range(src_file, 0, dst_file, 0, len); fput(src_file); if (ret < 0) { if (ret == -EOPNOTSUPP) return (-ENOTTY); return (ret); } if (ret != len) return (-EINVAL); return (0); } /* Entry point for FICLONERANGE, before Linux 4.5. */ long zpl_ioctl_ficlonerange(struct file *dst_file, void __user *arg) { zfs_ioc_compat_file_clone_range_t fcr; if (copy_from_user(&fcr, arg, sizeof (fcr))) return (-EFAULT); struct file *src_file = fget(fcr.fcr_src_fd); if (src_file == NULL) return (-EBADF); if (dst_file->f_op != src_file->f_op) return (-EXDEV); size_t len = fcr.fcr_src_length; if (len == 0) len = i_size_read(file_inode(src_file)) - fcr.fcr_src_offset; ssize_t ret = __zpl_clone_file_range(src_file, fcr.fcr_src_offset, dst_file, fcr.fcr_dest_offset, len); fput(src_file); if (ret < 0) { if (ret == -EOPNOTSUPP) return (-ENOTTY); return (ret); } if (ret != len) return (-EINVAL); return (0); } /* Entry point for FIDEDUPERANGE, before Linux 4.5. */ long zpl_ioctl_fideduperange(struct file *filp, void *arg) { (void) arg; /* No support for dedup yet */ return (-ENOTTY); } diff --git a/module/zfs/brt.c b/module/zfs/brt.c index e8218fb26888..ddd8eefe600b 100644 --- a/module/zfs/brt.c +++ b/module/zfs/brt.c @@ -1,1915 +1,1915 @@ /* * 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 https://opensource.org/licenses/CDDL-1.0. * 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) 2020, 2021, 2022 by Pawel Jakub Dawidek */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * Block Cloning design. * * Block Cloning allows to manually clone a file (or a subset of its blocks) * into another (or the same) file by just creating additional references to * the data blocks without copying the data itself. Those references are kept * in the Block Reference Tables (BRTs). * * In many ways this is similar to the existing deduplication, but there are * some important differences: * * - Deduplication is automatic and Block Cloning is not - one has to use a * dedicated system call(s) to clone the given file/blocks. * - Deduplication keeps all data blocks in its table, even those referenced * just once. Block Cloning creates an entry in its tables only when there * are at least two references to the given data block. If the block was * never explicitly cloned or the second to last reference was dropped, * there will be neither space nor performance overhead. * - Deduplication needs data to work - one needs to pass real data to the * write(2) syscall, so hash can be calculated. Block Cloning doesn't require * data, just block pointers to the data, so it is extremely fast, as we pay * neither the cost of reading the data, nor the cost of writing the data - * we operate exclusively on metadata. * - If the D (dedup) bit is not set in the block pointer, it means that * the block is not in the dedup table (DDT) and we won't consult the DDT * when we need to free the block. Block Cloning must be consulted on every * free, because we cannot modify the source BP (eg. by setting something * similar to the D bit), thus we have no hint if the block is in the * Block Reference Table (BRT), so we need to look into the BRT. There is * an optimization in place that allows us to eliminate the majority of BRT * lookups which is described below in the "Minimizing free penalty" section. * - The BRT entry is much smaller than the DDT entry - for BRT we only store * 64bit offset and 64bit reference counter. * - Dedup keys are cryptographic hashes, so two blocks that are close to each * other on disk are most likely in totally different parts of the DDT. * The BRT entry keys are offsets into a single top-level VDEV, so data blocks * from one file should have BRT entries close to each other. * - Scrub will only do a single pass over a block that is referenced multiple * times in the DDT. Unfortunately it is not currently (if at all) possible * with Block Cloning and block referenced multiple times will be scrubbed * multiple times. The new, sorted scrub should be able to eliminate * duplicated reads given enough memory. * - Deduplication requires cryptographically strong hash as a checksum or * additional data verification. Block Cloning works with any checksum * algorithm or even with checksumming disabled. * * As mentioned above, the BRT entries are much smaller than the DDT entries. * To uniquely identify a block we just need its vdev id and offset. We also * need to maintain a reference counter. The vdev id will often repeat, as there * is a small number of top-level VDEVs and a large number of blocks stored in * each VDEV. We take advantage of that to reduce the BRT entry size further by * maintaining one BRT for each top-level VDEV, so we can then have only offset * and counter as the BRT entry. * * Minimizing free penalty. * * Block Cloning allows creating additional references to any existing block. * When we free a block there is no hint in the block pointer whether the block * was cloned or not, so on each free we have to check if there is a * corresponding entry in the BRT or not. If there is, we need to decrease * the reference counter. Doing BRT lookup on every free can potentially be * expensive by requiring additional I/Os if the BRT doesn't fit into memory. * This is the main problem with deduplication, so we've learned our lesson and * try not to repeat the same mistake here. How do we do that? We divide each * top-level VDEV into 16MB regions. For each region we maintain a counter that * is a sum of all the BRT entries that have offsets within the region. This * creates the entries count array of 16bit numbers for each top-level VDEV. * The entries count array is always kept in memory and updated on disk in the * same transaction group as the BRT updates to keep everything in-sync. We can * keep the array in memory, because it is very small. With 16MB regions and * 1TB VDEV the array requires only 128kB of memory (we may decide to decrease * the region size even further in the future). Now, when we want to free * a block, we first consult the array. If the counter for the whole region is * zero, there is no need to look for the BRT entry, as there isn't one for * sure. If the counter for the region is greater than zero, only then we will * do a BRT lookup and if an entry is found we will decrease the reference * counter in the BRT entry and in the entry counters array. * * The entry counters array is small, but can potentially be larger for very * large VDEVs or smaller regions. In this case we don't want to rewrite entire * array on every change. We then divide the array into 32kB block and keep * a bitmap of dirty blocks within a transaction group. When we sync the * transaction group we can only update the parts of the entry counters array * that were modified. Note: Keeping track of the dirty parts of the entry * counters array is implemented, but updating only parts of the array on disk * is not yet implemented - for now we will update entire array if there was * any change. * * The implementation tries to be economic: if BRT is not used, or no longer * used, there will be no entries in the MOS and no additional memory used (eg. * the entry counters array is only allocated if needed). * * Interaction between Deduplication and Block Cloning. * * If both functionalities are in use, we could end up with a block that is * referenced multiple times in both DDT and BRT. When we free one of the * references we couldn't tell where it belongs, so we would have to decide * what table takes the precedence: do we first clear DDT references or BRT * references? To avoid this dilemma BRT cooperates with DDT - if a given block * is being cloned using BRT and the BP has the D (dedup) bit set, BRT will * lookup DDT entry instead and increase the counter there. No BRT entry * will be created for a block which has the D (dedup) bit set. * BRT may be more efficient for manual deduplication, but if the block is * already in the DDT, then creating additional BRT entry would be less * efficient. This clever idea was proposed by Allan Jude. * * Block Cloning across datasets. * * Block Cloning is not limited to cloning blocks within the same dataset. * It is possible (and very useful) to clone blocks between different datasets. * One use case is recovering files from snapshots. By cloning the files into * dataset we need no additional storage. Without Block Cloning we would need * additional space for those files. * Another interesting use case is moving the files between datasets * (copying the file content to the new dataset and removing the source file). * In that case Block Cloning will only be used briefly, because the BRT entries * will be removed when the source is removed. * Note: currently it is not possible to clone blocks between encrypted * datasets, even if those datasets use the same encryption key (this includes * snapshots of encrypted datasets). Cloning blocks between datasets that use * the same keys should be possible and should be implemented in the future. * * Block Cloning flow through ZFS layers. * * Note: Block Cloning can be used both for cloning file system blocks and ZVOL * blocks. As of this writing no interface is implemented that allows for block * cloning within a ZVOL. * FreeBSD and Linux provides copy_file_range(2) system call and we will use it * for blocking cloning. * * ssize_t * copy_file_range(int infd, off_t *inoffp, int outfd, off_t *outoffp, * size_t len, unsigned int flags); * * Even though offsets and length represent bytes, they have to be - * block-aligned or we will return the EXDEV error so the upper layer can + * block-aligned or we will return an error so the upper layer can * fallback to the generic mechanism that will just copy the data. * Using copy_file_range(2) will call OS-independent zfs_clone_range() function. * This function was implemented based on zfs_write(), but instead of writing * the given data we first read block pointers using the new dmu_read_l0_bps() * function from the source file. Once we have BPs from the source file we call * the dmu_brt_clone() function on the destination file. This function * allocates BPs for us. We iterate over all source BPs. If the given BP is * a hole or an embedded block, we just copy BP as-is. If it points to a real * data we place this BP on a BRT pending list using the brt_pending_add() * function. * * We use this pending list to keep track of all BPs that got new references * within this transaction group. * * Some special cases to consider and how we address them: * - The block we want to clone may have been created within the same * transaction group that we are trying to clone. Such block has no BP - * allocated yet, so cannot be immediately cloned. We return EXDEV. + * allocated yet, so cannot be immediately cloned. We return EAGAIN. * - The block we want to clone may have been modified within the same - * transaction group. We return EXDEV. + * transaction group. We return EAGAIN. * - A block may be cloned multiple times during one transaction group (that's * why pending list is actually a tree and not an append-only list - this * way we can figure out faster if this block is cloned for the first time * in this txg or consecutive time). * - A block may be cloned and freed within the same transaction group * (see dbuf_undirty()). * - A block may be cloned and within the same transaction group the clone * can be cloned again (see dmu_read_l0_bps()). * - A file might have been deleted, but the caller still has a file descriptor * open to this file and clones it. * * When we free a block we have an additional step in the ZIO pipeline where we * call the zio_brt_free() function. We then call the brt_entry_decref() * that loads the corresponding BRT entry (if one exists) and decreases * reference counter. If this is not the last reference we will stop ZIO * pipeline here. If this is the last reference or the block is not in the * BRT, we continue the pipeline and free the block as usual. * * At the beginning of spa_sync() where there can be no more block cloning, * but before issuing frees we call brt_pending_apply(). This function applies * all the new clones to the BRT table - we load BRT entries and update * reference counters. To sync new BRT entries to disk, we use brt_sync() * function. This function will sync all dirty per-top-level-vdev BRTs, * the entry counters arrays, etc. * * Block Cloning and ZIL. * * Every clone operation is divided into chunks (similar to write) and each * chunk is cloned in a separate transaction. The chunk size is determined by * how many BPs we can fit into a single ZIL entry. * Replaying clone operation is different from the regular clone operation, * as when we log clone operations we cannot use the source object - it may * reside on a different dataset, so we log BPs we want to clone. * The ZIL is replayed when we mount the given dataset, not when the pool is * imported. Taking this into account it is possible that the pool is imported * without mounting datasets and the source dataset is destroyed before the * destination dataset is mounted and its ZIL replayed. * To address this situation we leverage zil_claim() mechanism where ZFS will * parse all the ZILs on pool import. When we come across TX_CLONE_RANGE * entries, we will bump reference counters for their BPs in the BRT and then * on mount and ZIL replay we will just attach BPs to the file without * bumping reference counters. * Note it is still possible that after zil_claim() we never mount the * destination, so we never replay its ZIL and we destroy it. This way we would * end up with leaked references in BRT. We address that too as ZFS gives us * a chance to clean this up on dataset destroy (see zil_free_clone_range()). */ /* * BRT - Block Reference Table. */ #define BRT_OBJECT_VDEV_PREFIX "com.fudosecurity:brt:vdev:" /* * We divide each VDEV into 16MB chunks. Each chunk is represented in memory * by a 16bit counter, thus 1TB VDEV requires 128kB of memory: (1TB / 16MB) * 2B * Each element in this array represents how many BRT entries do we have in this * chunk of storage. We always load this entire array into memory and update as * needed. By having it in memory we can quickly tell (during zio_free()) if * there are any BRT entries that we might need to update. * * This value cannot be larger than 16MB, at least as long as we support * 512 byte block sizes. With 512 byte block size we can have exactly * 32768 blocks in 16MB. In 32MB we could have 65536 blocks, which is one too * many for a 16bit counter. */ #define BRT_RANGESIZE (16 * 1024 * 1024) _Static_assert(BRT_RANGESIZE / SPA_MINBLOCKSIZE <= UINT16_MAX, "BRT_RANGESIZE is too large."); /* * We don't want to update the whole structure every time. Maintain bitmap * of dirty blocks within the regions, so that a single bit represents a * block size of entcounts. For example if we have a 1PB vdev then all * entcounts take 128MB of memory ((64TB / 16MB) * 2B). We can divide this * 128MB array of entcounts into 32kB disk blocks, as we don't want to update * the whole 128MB on disk when we have updated only a single entcount. * We maintain a bitmap where each 32kB disk block within 128MB entcounts array * is represented by a single bit. This gives us 4096 bits. A set bit in the * bitmap means that we had a change in at least one of the 16384 entcounts * that reside on a 32kB disk block (32kB / sizeof (uint16_t)). */ #define BRT_BLOCKSIZE (32 * 1024) #define BRT_RANGESIZE_TO_NBLOCKS(size) \ (((size) - 1) / BRT_BLOCKSIZE / sizeof (uint16_t) + 1) #define BRT_LITTLE_ENDIAN 0 #define BRT_BIG_ENDIAN 1 #ifdef _ZFS_LITTLE_ENDIAN #define BRT_NATIVE_BYTEORDER BRT_LITTLE_ENDIAN #define BRT_NON_NATIVE_BYTEORDER BRT_BIG_ENDIAN #else #define BRT_NATIVE_BYTEORDER BRT_BIG_ENDIAN #define BRT_NON_NATIVE_BYTEORDER BRT_LITTLE_ENDIAN #endif typedef struct brt_vdev_phys { uint64_t bvp_mos_entries; uint64_t bvp_size; uint64_t bvp_byteorder; uint64_t bvp_totalcount; uint64_t bvp_rangesize; uint64_t bvp_usedspace; uint64_t bvp_savedspace; } brt_vdev_phys_t; typedef struct brt_vdev { /* * VDEV id. */ uint64_t bv_vdevid; /* * Is the structure initiated? * (bv_entcount and bv_bitmap are allocated?) */ boolean_t bv_initiated; /* * Object number in the MOS for the entcount array and brt_vdev_phys. */ uint64_t bv_mos_brtvdev; /* * Object number in the MOS for the entries table. */ uint64_t bv_mos_entries; /* * Entries to sync. */ avl_tree_t bv_tree; /* * Does the bv_entcount[] array needs byte swapping? */ boolean_t bv_need_byteswap; /* * Number of entries in the bv_entcount[] array. */ uint64_t bv_size; /* * This is the array with BRT entry count per BRT_RANGESIZE. */ uint16_t *bv_entcount; /* * Sum of all bv_entcount[]s. */ uint64_t bv_totalcount; /* * Space on disk occupied by cloned blocks (without compression). */ uint64_t bv_usedspace; /* * How much additional space would be occupied without block cloning. */ uint64_t bv_savedspace; /* * brt_vdev_phys needs updating on disk. */ boolean_t bv_meta_dirty; /* * bv_entcount[] needs updating on disk. */ boolean_t bv_entcount_dirty; /* * bv_entcount[] potentially can be a bit too big to sychronize it all * when we just changed few entcounts. The fields below allow us to * track updates to bv_entcount[] array since the last sync. * A single bit in the bv_bitmap represents as many entcounts as can * fit into a single BRT_BLOCKSIZE. * For example we have 65536 entcounts in the bv_entcount array * (so the whole array is 128kB). We updated bv_entcount[2] and * bv_entcount[5]. In that case only first bit in the bv_bitmap will * be set and we will write only first BRT_BLOCKSIZE out of 128kB. */ ulong_t *bv_bitmap; uint64_t bv_nblocks; } brt_vdev_t; /* * In-core brt */ typedef struct brt { krwlock_t brt_lock; spa_t *brt_spa; #define brt_mos brt_spa->spa_meta_objset uint64_t brt_rangesize; uint64_t brt_usedspace; uint64_t brt_savedspace; avl_tree_t brt_pending_tree[TXG_SIZE]; kmutex_t brt_pending_lock[TXG_SIZE]; /* Sum of all entries across all bv_trees. */ uint64_t brt_nentries; brt_vdev_t *brt_vdevs; uint64_t brt_nvdevs; } brt_t; /* Size of bre_offset / sizeof (uint64_t). */ #define BRT_KEY_WORDS (1) /* * In-core brt entry. * On-disk we use bre_offset as the key and bre_refcount as the value. */ typedef struct brt_entry { uint64_t bre_offset; uint64_t bre_refcount; avl_node_t bre_node; } brt_entry_t; typedef struct brt_pending_entry { blkptr_t bpe_bp; int bpe_count; avl_node_t bpe_node; } brt_pending_entry_t; static kmem_cache_t *brt_entry_cache; static kmem_cache_t *brt_pending_entry_cache; /* * Enable/disable prefetching of BRT entries that we are going to modify. */ int zfs_brt_prefetch = 1; #ifdef ZFS_DEBUG #define BRT_DEBUG(...) do { \ if ((zfs_flags & ZFS_DEBUG_BRT) != 0) { \ __dprintf(B_TRUE, __FILE__, __func__, __LINE__, __VA_ARGS__); \ } \ } while (0) #else #define BRT_DEBUG(...) do { } while (0) #endif int brt_zap_leaf_blockshift = 12; int brt_zap_indirect_blockshift = 12; static kstat_t *brt_ksp; typedef struct brt_stats { kstat_named_t brt_addref_entry_in_memory; kstat_named_t brt_addref_entry_not_on_disk; kstat_named_t brt_addref_entry_on_disk; kstat_named_t brt_addref_entry_read_lost_race; kstat_named_t brt_decref_entry_in_memory; kstat_named_t brt_decref_entry_loaded_from_disk; kstat_named_t brt_decref_entry_not_in_memory; kstat_named_t brt_decref_entry_not_on_disk; kstat_named_t brt_decref_entry_read_lost_race; kstat_named_t brt_decref_entry_still_referenced; kstat_named_t brt_decref_free_data_later; kstat_named_t brt_decref_free_data_now; kstat_named_t brt_decref_no_entry; } brt_stats_t; static brt_stats_t brt_stats = { { "addref_entry_in_memory", KSTAT_DATA_UINT64 }, { "addref_entry_not_on_disk", KSTAT_DATA_UINT64 }, { "addref_entry_on_disk", KSTAT_DATA_UINT64 }, { "addref_entry_read_lost_race", KSTAT_DATA_UINT64 }, { "decref_entry_in_memory", KSTAT_DATA_UINT64 }, { "decref_entry_loaded_from_disk", KSTAT_DATA_UINT64 }, { "decref_entry_not_in_memory", KSTAT_DATA_UINT64 }, { "decref_entry_not_on_disk", KSTAT_DATA_UINT64 }, { "decref_entry_read_lost_race", KSTAT_DATA_UINT64 }, { "decref_entry_still_referenced", KSTAT_DATA_UINT64 }, { "decref_free_data_later", KSTAT_DATA_UINT64 }, { "decref_free_data_now", KSTAT_DATA_UINT64 }, { "decref_no_entry", KSTAT_DATA_UINT64 } }; struct { wmsum_t brt_addref_entry_in_memory; wmsum_t brt_addref_entry_not_on_disk; wmsum_t brt_addref_entry_on_disk; wmsum_t brt_addref_entry_read_lost_race; wmsum_t brt_decref_entry_in_memory; wmsum_t brt_decref_entry_loaded_from_disk; wmsum_t brt_decref_entry_not_in_memory; wmsum_t brt_decref_entry_not_on_disk; wmsum_t brt_decref_entry_read_lost_race; wmsum_t brt_decref_entry_still_referenced; wmsum_t brt_decref_free_data_later; wmsum_t brt_decref_free_data_now; wmsum_t brt_decref_no_entry; } brt_sums; #define BRTSTAT_BUMP(stat) wmsum_add(&brt_sums.stat, 1) static int brt_entry_compare(const void *x1, const void *x2); static int brt_pending_entry_compare(const void *x1, const void *x2); static void brt_rlock(brt_t *brt) { rw_enter(&brt->brt_lock, RW_READER); } static void brt_wlock(brt_t *brt) { rw_enter(&brt->brt_lock, RW_WRITER); } static void brt_unlock(brt_t *brt) { rw_exit(&brt->brt_lock); } static uint16_t brt_vdev_entcount_get(const brt_vdev_t *brtvd, uint64_t idx) { ASSERT3U(idx, <, brtvd->bv_size); if (brtvd->bv_need_byteswap) { return (BSWAP_16(brtvd->bv_entcount[idx])); } else { return (brtvd->bv_entcount[idx]); } } static void brt_vdev_entcount_set(brt_vdev_t *brtvd, uint64_t idx, uint16_t entcnt) { ASSERT3U(idx, <, brtvd->bv_size); if (brtvd->bv_need_byteswap) { brtvd->bv_entcount[idx] = BSWAP_16(entcnt); } else { brtvd->bv_entcount[idx] = entcnt; } } static void brt_vdev_entcount_inc(brt_vdev_t *brtvd, uint64_t idx) { uint16_t entcnt; ASSERT3U(idx, <, brtvd->bv_size); entcnt = brt_vdev_entcount_get(brtvd, idx); ASSERT(entcnt < UINT16_MAX); brt_vdev_entcount_set(brtvd, idx, entcnt + 1); } static void brt_vdev_entcount_dec(brt_vdev_t *brtvd, uint64_t idx) { uint16_t entcnt; ASSERT3U(idx, <, brtvd->bv_size); entcnt = brt_vdev_entcount_get(brtvd, idx); ASSERT(entcnt > 0); brt_vdev_entcount_set(brtvd, idx, entcnt - 1); } #ifdef ZFS_DEBUG static void brt_vdev_dump(brt_t *brt) { brt_vdev_t *brtvd; uint64_t vdevid; if ((zfs_flags & ZFS_DEBUG_BRT) == 0) { return; } if (brt->brt_nvdevs == 0) { zfs_dbgmsg("BRT empty"); return; } zfs_dbgmsg("BRT vdev dump:"); for (vdevid = 0; vdevid < brt->brt_nvdevs; vdevid++) { uint64_t idx; brtvd = &brt->brt_vdevs[vdevid]; zfs_dbgmsg(" vdevid=%llu/%llu meta_dirty=%d entcount_dirty=%d " "size=%llu totalcount=%llu nblocks=%llu bitmapsize=%zu\n", (u_longlong_t)vdevid, (u_longlong_t)brtvd->bv_vdevid, brtvd->bv_meta_dirty, brtvd->bv_entcount_dirty, (u_longlong_t)brtvd->bv_size, (u_longlong_t)brtvd->bv_totalcount, (u_longlong_t)brtvd->bv_nblocks, (size_t)BT_SIZEOFMAP(brtvd->bv_nblocks)); if (brtvd->bv_totalcount > 0) { zfs_dbgmsg(" entcounts:"); for (idx = 0; idx < brtvd->bv_size; idx++) { if (brt_vdev_entcount_get(brtvd, idx) > 0) { zfs_dbgmsg(" [%04llu] %hu", (u_longlong_t)idx, brt_vdev_entcount_get(brtvd, idx)); } } } if (brtvd->bv_entcount_dirty) { char *bitmap; bitmap = kmem_alloc(brtvd->bv_nblocks + 1, KM_SLEEP); for (idx = 0; idx < brtvd->bv_nblocks; idx++) { bitmap[idx] = BT_TEST(brtvd->bv_bitmap, idx) ? 'x' : '.'; } bitmap[idx] = '\0'; zfs_dbgmsg(" bitmap: %s", bitmap); kmem_free(bitmap, brtvd->bv_nblocks + 1); } } } #endif static brt_vdev_t * brt_vdev(brt_t *brt, uint64_t vdevid) { brt_vdev_t *brtvd; ASSERT(RW_LOCK_HELD(&brt->brt_lock)); if (vdevid < brt->brt_nvdevs) { brtvd = &brt->brt_vdevs[vdevid]; } else { brtvd = NULL; } return (brtvd); } static void brt_vdev_create(brt_t *brt, brt_vdev_t *brtvd, dmu_tx_t *tx) { char name[64]; ASSERT(RW_WRITE_HELD(&brt->brt_lock)); ASSERT0(brtvd->bv_mos_brtvdev); ASSERT0(brtvd->bv_mos_entries); ASSERT(brtvd->bv_entcount != NULL); ASSERT(brtvd->bv_size > 0); ASSERT(brtvd->bv_bitmap != NULL); ASSERT(brtvd->bv_nblocks > 0); brtvd->bv_mos_entries = zap_create_flags(brt->brt_mos, 0, ZAP_FLAG_HASH64 | ZAP_FLAG_UINT64_KEY, DMU_OTN_ZAP_METADATA, brt_zap_leaf_blockshift, brt_zap_indirect_blockshift, DMU_OT_NONE, 0, tx); VERIFY(brtvd->bv_mos_entries != 0); BRT_DEBUG("MOS entries created, object=%llu", (u_longlong_t)brtvd->bv_mos_entries); /* * We allocate DMU buffer to store the bv_entcount[] array. * We will keep array size (bv_size) and cummulative count for all * bv_entcount[]s (bv_totalcount) in the bonus buffer. */ brtvd->bv_mos_brtvdev = dmu_object_alloc(brt->brt_mos, DMU_OTN_UINT64_METADATA, BRT_BLOCKSIZE, DMU_OTN_UINT64_METADATA, sizeof (brt_vdev_phys_t), tx); VERIFY(brtvd->bv_mos_brtvdev != 0); BRT_DEBUG("MOS BRT VDEV created, object=%llu", (u_longlong_t)brtvd->bv_mos_brtvdev); snprintf(name, sizeof (name), "%s%llu", BRT_OBJECT_VDEV_PREFIX, (u_longlong_t)brtvd->bv_vdevid); VERIFY0(zap_add(brt->brt_mos, DMU_POOL_DIRECTORY_OBJECT, name, sizeof (uint64_t), 1, &brtvd->bv_mos_brtvdev, tx)); BRT_DEBUG("Pool directory object created, object=%s", name); spa_feature_incr(brt->brt_spa, SPA_FEATURE_BLOCK_CLONING, tx); } static void brt_vdev_realloc(brt_t *brt, brt_vdev_t *brtvd) { vdev_t *vd; uint16_t *entcount; ulong_t *bitmap; uint64_t nblocks, size; ASSERT(RW_WRITE_HELD(&brt->brt_lock)); spa_config_enter(brt->brt_spa, SCL_VDEV, FTAG, RW_READER); vd = vdev_lookup_top(brt->brt_spa, brtvd->bv_vdevid); size = (vdev_get_min_asize(vd) - 1) / brt->brt_rangesize + 1; spa_config_exit(brt->brt_spa, SCL_VDEV, FTAG); entcount = vmem_zalloc(sizeof (entcount[0]) * size, KM_SLEEP); nblocks = BRT_RANGESIZE_TO_NBLOCKS(size); bitmap = kmem_zalloc(BT_SIZEOFMAP(nblocks), KM_SLEEP); if (!brtvd->bv_initiated) { ASSERT0(brtvd->bv_size); ASSERT(brtvd->bv_entcount == NULL); ASSERT(brtvd->bv_bitmap == NULL); ASSERT0(brtvd->bv_nblocks); avl_create(&brtvd->bv_tree, brt_entry_compare, sizeof (brt_entry_t), offsetof(brt_entry_t, bre_node)); } else { ASSERT(brtvd->bv_size > 0); ASSERT(brtvd->bv_entcount != NULL); ASSERT(brtvd->bv_bitmap != NULL); ASSERT(brtvd->bv_nblocks > 0); /* * TODO: Allow vdev shrinking. We only need to implement * shrinking the on-disk BRT VDEV object. * dmu_free_range(brt->brt_mos, brtvd->bv_mos_brtvdev, offset, * size, tx); */ ASSERT3U(brtvd->bv_size, <=, size); memcpy(entcount, brtvd->bv_entcount, sizeof (entcount[0]) * MIN(size, brtvd->bv_size)); memcpy(bitmap, brtvd->bv_bitmap, MIN(BT_SIZEOFMAP(nblocks), BT_SIZEOFMAP(brtvd->bv_nblocks))); vmem_free(brtvd->bv_entcount, sizeof (entcount[0]) * brtvd->bv_size); kmem_free(brtvd->bv_bitmap, BT_SIZEOFMAP(brtvd->bv_nblocks)); } brtvd->bv_size = size; brtvd->bv_entcount = entcount; brtvd->bv_bitmap = bitmap; brtvd->bv_nblocks = nblocks; if (!brtvd->bv_initiated) { brtvd->bv_need_byteswap = FALSE; brtvd->bv_initiated = TRUE; BRT_DEBUG("BRT VDEV %llu initiated.", (u_longlong_t)brtvd->bv_vdevid); } } static void brt_vdev_load(brt_t *brt, brt_vdev_t *brtvd) { char name[64]; dmu_buf_t *db; brt_vdev_phys_t *bvphys; int error; snprintf(name, sizeof (name), "%s%llu", BRT_OBJECT_VDEV_PREFIX, (u_longlong_t)brtvd->bv_vdevid); error = zap_lookup(brt->brt_mos, DMU_POOL_DIRECTORY_OBJECT, name, sizeof (uint64_t), 1, &brtvd->bv_mos_brtvdev); if (error != 0) return; ASSERT(brtvd->bv_mos_brtvdev != 0); error = dmu_bonus_hold(brt->brt_mos, brtvd->bv_mos_brtvdev, FTAG, &db); ASSERT0(error); if (error != 0) return; bvphys = db->db_data; if (brt->brt_rangesize == 0) { brt->brt_rangesize = bvphys->bvp_rangesize; } else { ASSERT3U(brt->brt_rangesize, ==, bvphys->bvp_rangesize); } ASSERT(!brtvd->bv_initiated); brt_vdev_realloc(brt, brtvd); /* TODO: We don't support VDEV shrinking. */ ASSERT3U(bvphys->bvp_size, <=, brtvd->bv_size); /* * If VDEV grew, we will leave new bv_entcount[] entries zeroed out. */ error = dmu_read(brt->brt_mos, brtvd->bv_mos_brtvdev, 0, MIN(brtvd->bv_size, bvphys->bvp_size) * sizeof (uint16_t), brtvd->bv_entcount, DMU_READ_NO_PREFETCH); ASSERT0(error); brtvd->bv_mos_entries = bvphys->bvp_mos_entries; ASSERT(brtvd->bv_mos_entries != 0); brtvd->bv_need_byteswap = (bvphys->bvp_byteorder != BRT_NATIVE_BYTEORDER); brtvd->bv_totalcount = bvphys->bvp_totalcount; brtvd->bv_usedspace = bvphys->bvp_usedspace; brtvd->bv_savedspace = bvphys->bvp_savedspace; brt->brt_usedspace += brtvd->bv_usedspace; brt->brt_savedspace += brtvd->bv_savedspace; dmu_buf_rele(db, FTAG); BRT_DEBUG("MOS BRT VDEV %s loaded: mos_brtvdev=%llu, mos_entries=%llu", name, (u_longlong_t)brtvd->bv_mos_brtvdev, (u_longlong_t)brtvd->bv_mos_entries); } static void brt_vdev_dealloc(brt_t *brt, brt_vdev_t *brtvd) { ASSERT(RW_WRITE_HELD(&brt->brt_lock)); ASSERT(brtvd->bv_initiated); vmem_free(brtvd->bv_entcount, sizeof (uint16_t) * brtvd->bv_size); brtvd->bv_entcount = NULL; kmem_free(brtvd->bv_bitmap, BT_SIZEOFMAP(brtvd->bv_nblocks)); brtvd->bv_bitmap = NULL; ASSERT0(avl_numnodes(&brtvd->bv_tree)); avl_destroy(&brtvd->bv_tree); brtvd->bv_size = 0; brtvd->bv_nblocks = 0; brtvd->bv_initiated = FALSE; BRT_DEBUG("BRT VDEV %llu deallocated.", (u_longlong_t)brtvd->bv_vdevid); } static void brt_vdev_destroy(brt_t *brt, brt_vdev_t *brtvd, dmu_tx_t *tx) { char name[64]; uint64_t count; dmu_buf_t *db; brt_vdev_phys_t *bvphys; ASSERT(RW_WRITE_HELD(&brt->brt_lock)); ASSERT(brtvd->bv_mos_brtvdev != 0); ASSERT(brtvd->bv_mos_entries != 0); VERIFY0(zap_count(brt->brt_mos, brtvd->bv_mos_entries, &count)); VERIFY0(count); VERIFY0(zap_destroy(brt->brt_mos, brtvd->bv_mos_entries, tx)); BRT_DEBUG("MOS entries destroyed, object=%llu", (u_longlong_t)brtvd->bv_mos_entries); brtvd->bv_mos_entries = 0; VERIFY0(dmu_bonus_hold(brt->brt_mos, brtvd->bv_mos_brtvdev, FTAG, &db)); bvphys = db->db_data; ASSERT0(bvphys->bvp_totalcount); ASSERT0(bvphys->bvp_usedspace); ASSERT0(bvphys->bvp_savedspace); dmu_buf_rele(db, FTAG); VERIFY0(dmu_object_free(brt->brt_mos, brtvd->bv_mos_brtvdev, tx)); BRT_DEBUG("MOS BRT VDEV destroyed, object=%llu", (u_longlong_t)brtvd->bv_mos_brtvdev); brtvd->bv_mos_brtvdev = 0; snprintf(name, sizeof (name), "%s%llu", BRT_OBJECT_VDEV_PREFIX, (u_longlong_t)brtvd->bv_vdevid); VERIFY0(zap_remove(brt->brt_mos, DMU_POOL_DIRECTORY_OBJECT, name, tx)); BRT_DEBUG("Pool directory object removed, object=%s", name); brt_vdev_dealloc(brt, brtvd); spa_feature_decr(brt->brt_spa, SPA_FEATURE_BLOCK_CLONING, tx); } static void brt_vdevs_expand(brt_t *brt, uint64_t nvdevs) { brt_vdev_t *brtvd, *vdevs; uint64_t vdevid; ASSERT(RW_WRITE_HELD(&brt->brt_lock)); ASSERT3U(nvdevs, >, brt->brt_nvdevs); vdevs = kmem_zalloc(sizeof (vdevs[0]) * nvdevs, KM_SLEEP); if (brt->brt_nvdevs > 0) { ASSERT(brt->brt_vdevs != NULL); memcpy(vdevs, brt->brt_vdevs, sizeof (brt_vdev_t) * brt->brt_nvdevs); kmem_free(brt->brt_vdevs, sizeof (brt_vdev_t) * brt->brt_nvdevs); } for (vdevid = brt->brt_nvdevs; vdevid < nvdevs; vdevid++) { brtvd = &vdevs[vdevid]; brtvd->bv_vdevid = vdevid; brtvd->bv_initiated = FALSE; } BRT_DEBUG("BRT VDEVs expanded from %llu to %llu.", (u_longlong_t)brt->brt_nvdevs, (u_longlong_t)nvdevs); brt->brt_vdevs = vdevs; brt->brt_nvdevs = nvdevs; } static boolean_t brt_vdev_lookup(brt_t *brt, brt_vdev_t *brtvd, const brt_entry_t *bre) { uint64_t idx; ASSERT(RW_LOCK_HELD(&brt->brt_lock)); idx = bre->bre_offset / brt->brt_rangesize; if (brtvd->bv_entcount != NULL && idx < brtvd->bv_size) { /* VDEV wasn't expanded. */ return (brt_vdev_entcount_get(brtvd, idx) > 0); } return (FALSE); } static void brt_vdev_addref(brt_t *brt, brt_vdev_t *brtvd, const brt_entry_t *bre, uint64_t dsize) { uint64_t idx; ASSERT(RW_LOCK_HELD(&brt->brt_lock)); ASSERT(brtvd != NULL); ASSERT(brtvd->bv_entcount != NULL); brt->brt_savedspace += dsize; brtvd->bv_savedspace += dsize; brtvd->bv_meta_dirty = TRUE; if (bre->bre_refcount > 1) { return; } brt->brt_usedspace += dsize; brtvd->bv_usedspace += dsize; idx = bre->bre_offset / brt->brt_rangesize; if (idx >= brtvd->bv_size) { /* VDEV has been expanded. */ brt_vdev_realloc(brt, brtvd); } ASSERT3U(idx, <, brtvd->bv_size); brtvd->bv_totalcount++; brt_vdev_entcount_inc(brtvd, idx); brtvd->bv_entcount_dirty = TRUE; idx = idx / BRT_BLOCKSIZE / 8; BT_SET(brtvd->bv_bitmap, idx); #ifdef ZFS_DEBUG brt_vdev_dump(brt); #endif } static void brt_vdev_decref(brt_t *brt, brt_vdev_t *brtvd, const brt_entry_t *bre, uint64_t dsize) { uint64_t idx; ASSERT(RW_WRITE_HELD(&brt->brt_lock)); ASSERT(brtvd != NULL); ASSERT(brtvd->bv_entcount != NULL); brt->brt_savedspace -= dsize; brtvd->bv_savedspace -= dsize; brtvd->bv_meta_dirty = TRUE; if (bre->bre_refcount > 0) { return; } brt->brt_usedspace -= dsize; brtvd->bv_usedspace -= dsize; idx = bre->bre_offset / brt->brt_rangesize; ASSERT3U(idx, <, brtvd->bv_size); ASSERT(brtvd->bv_totalcount > 0); brtvd->bv_totalcount--; brt_vdev_entcount_dec(brtvd, idx); brtvd->bv_entcount_dirty = TRUE; idx = idx / BRT_BLOCKSIZE / 8; BT_SET(brtvd->bv_bitmap, idx); #ifdef ZFS_DEBUG brt_vdev_dump(brt); #endif } static void brt_vdev_sync(brt_t *brt, brt_vdev_t *brtvd, dmu_tx_t *tx) { dmu_buf_t *db; brt_vdev_phys_t *bvphys; ASSERT(brtvd->bv_meta_dirty); ASSERT(brtvd->bv_mos_brtvdev != 0); ASSERT(dmu_tx_is_syncing(tx)); VERIFY0(dmu_bonus_hold(brt->brt_mos, brtvd->bv_mos_brtvdev, FTAG, &db)); if (brtvd->bv_entcount_dirty) { /* * TODO: Walk brtvd->bv_bitmap and write only the dirty blocks. */ dmu_write(brt->brt_mos, brtvd->bv_mos_brtvdev, 0, brtvd->bv_size * sizeof (brtvd->bv_entcount[0]), brtvd->bv_entcount, tx); memset(brtvd->bv_bitmap, 0, BT_SIZEOFMAP(brtvd->bv_nblocks)); brtvd->bv_entcount_dirty = FALSE; } dmu_buf_will_dirty(db, tx); bvphys = db->db_data; bvphys->bvp_mos_entries = brtvd->bv_mos_entries; bvphys->bvp_size = brtvd->bv_size; if (brtvd->bv_need_byteswap) { bvphys->bvp_byteorder = BRT_NON_NATIVE_BYTEORDER; } else { bvphys->bvp_byteorder = BRT_NATIVE_BYTEORDER; } bvphys->bvp_totalcount = brtvd->bv_totalcount; bvphys->bvp_rangesize = brt->brt_rangesize; bvphys->bvp_usedspace = brtvd->bv_usedspace; bvphys->bvp_savedspace = brtvd->bv_savedspace; dmu_buf_rele(db, FTAG); brtvd->bv_meta_dirty = FALSE; } static void brt_vdevs_alloc(brt_t *brt, boolean_t load) { brt_vdev_t *brtvd; uint64_t vdevid; brt_wlock(brt); brt_vdevs_expand(brt, brt->brt_spa->spa_root_vdev->vdev_children); if (load) { for (vdevid = 0; vdevid < brt->brt_nvdevs; vdevid++) { brtvd = &brt->brt_vdevs[vdevid]; ASSERT(brtvd->bv_entcount == NULL); brt_vdev_load(brt, brtvd); } } if (brt->brt_rangesize == 0) { brt->brt_rangesize = BRT_RANGESIZE; } brt_unlock(brt); } static void brt_vdevs_free(brt_t *brt) { brt_vdev_t *brtvd; uint64_t vdevid; brt_wlock(brt); for (vdevid = 0; vdevid < brt->brt_nvdevs; vdevid++) { brtvd = &brt->brt_vdevs[vdevid]; if (brtvd->bv_initiated) brt_vdev_dealloc(brt, brtvd); } kmem_free(brt->brt_vdevs, sizeof (brt_vdev_t) * brt->brt_nvdevs); brt_unlock(brt); } static void brt_entry_fill(const blkptr_t *bp, brt_entry_t *bre, uint64_t *vdevidp) { bre->bre_offset = DVA_GET_OFFSET(&bp->blk_dva[0]); bre->bre_refcount = 0; *vdevidp = DVA_GET_VDEV(&bp->blk_dva[0]); } static int brt_entry_compare(const void *x1, const void *x2) { const brt_entry_t *bre1 = x1; const brt_entry_t *bre2 = x2; return (TREE_CMP(bre1->bre_offset, bre2->bre_offset)); } static int brt_entry_lookup(brt_t *brt, brt_vdev_t *brtvd, brt_entry_t *bre) { uint64_t mos_entries; uint64_t one, physsize; int error; ASSERT(RW_LOCK_HELD(&brt->brt_lock)); if (!brt_vdev_lookup(brt, brtvd, bre)) return (SET_ERROR(ENOENT)); /* * Remember mos_entries object number. After we reacquire the BRT lock, * the brtvd pointer may be invalid. */ mos_entries = brtvd->bv_mos_entries; if (mos_entries == 0) return (SET_ERROR(ENOENT)); brt_unlock(brt); error = zap_length_uint64(brt->brt_mos, mos_entries, &bre->bre_offset, BRT_KEY_WORDS, &one, &physsize); if (error == 0) { ASSERT3U(one, ==, 1); ASSERT3U(physsize, ==, sizeof (bre->bre_refcount)); error = zap_lookup_uint64(brt->brt_mos, mos_entries, &bre->bre_offset, BRT_KEY_WORDS, 1, sizeof (bre->bre_refcount), &bre->bre_refcount); BRT_DEBUG("ZAP lookup: object=%llu vdev=%llu offset=%llu " "count=%llu error=%d", (u_longlong_t)mos_entries, (u_longlong_t)brtvd->bv_vdevid, (u_longlong_t)bre->bre_offset, error == 0 ? (u_longlong_t)bre->bre_refcount : 0, error); } brt_wlock(brt); return (error); } static void brt_entry_prefetch(brt_t *brt, uint64_t vdevid, brt_entry_t *bre) { brt_vdev_t *brtvd; uint64_t mos_entries = 0; brt_rlock(brt); brtvd = brt_vdev(brt, vdevid); if (brtvd != NULL) mos_entries = brtvd->bv_mos_entries; brt_unlock(brt); if (mos_entries == 0) return; BRT_DEBUG("ZAP prefetch: object=%llu vdev=%llu offset=%llu", (u_longlong_t)mos_entries, (u_longlong_t)vdevid, (u_longlong_t)bre->bre_offset); (void) zap_prefetch_uint64(brt->brt_mos, mos_entries, (uint64_t *)&bre->bre_offset, BRT_KEY_WORDS); } static int brt_entry_update(brt_t *brt, brt_vdev_t *brtvd, brt_entry_t *bre, dmu_tx_t *tx) { int error; ASSERT(RW_LOCK_HELD(&brt->brt_lock)); ASSERT(brtvd->bv_mos_entries != 0); ASSERT(bre->bre_refcount > 0); error = zap_update_uint64(brt->brt_mos, brtvd->bv_mos_entries, (uint64_t *)&bre->bre_offset, BRT_KEY_WORDS, 1, sizeof (bre->bre_refcount), &bre->bre_refcount, tx); BRT_DEBUG("ZAP update: object=%llu vdev=%llu offset=%llu count=%llu " "error=%d", (u_longlong_t)brtvd->bv_mos_entries, (u_longlong_t)brtvd->bv_vdevid, (u_longlong_t)bre->bre_offset, (u_longlong_t)bre->bre_refcount, error); return (error); } static int brt_entry_remove(brt_t *brt, brt_vdev_t *brtvd, brt_entry_t *bre, dmu_tx_t *tx) { int error; ASSERT(RW_LOCK_HELD(&brt->brt_lock)); ASSERT(brtvd->bv_mos_entries != 0); ASSERT0(bre->bre_refcount); error = zap_remove_uint64(brt->brt_mos, brtvd->bv_mos_entries, (uint64_t *)&bre->bre_offset, BRT_KEY_WORDS, tx); BRT_DEBUG("ZAP remove: object=%llu vdev=%llu offset=%llu count=%llu " "error=%d", (u_longlong_t)brtvd->bv_mos_entries, (u_longlong_t)brtvd->bv_vdevid, (u_longlong_t)bre->bre_offset, (u_longlong_t)bre->bre_refcount, error); return (error); } /* * Return TRUE if we _can_ have BRT entry for this bp. It might be false * positive, but gives us quick answer if we should look into BRT, which * may require reads and thus will be more expensive. */ boolean_t brt_maybe_exists(spa_t *spa, const blkptr_t *bp) { brt_t *brt = spa->spa_brt; brt_vdev_t *brtvd; brt_entry_t bre_search; boolean_t mayexists = FALSE; uint64_t vdevid; brt_entry_fill(bp, &bre_search, &vdevid); brt_rlock(brt); brtvd = brt_vdev(brt, vdevid); if (brtvd != NULL && brtvd->bv_initiated) { if (!avl_is_empty(&brtvd->bv_tree) || brt_vdev_lookup(brt, brtvd, &bre_search)) { mayexists = TRUE; } } brt_unlock(brt); return (mayexists); } uint64_t brt_get_dspace(spa_t *spa) { brt_t *brt = spa->spa_brt; if (brt == NULL) return (0); return (brt->brt_savedspace); } uint64_t brt_get_used(spa_t *spa) { brt_t *brt = spa->spa_brt; if (brt == NULL) return (0); return (brt->brt_usedspace); } uint64_t brt_get_saved(spa_t *spa) { brt_t *brt = spa->spa_brt; if (brt == NULL) return (0); return (brt->brt_savedspace); } uint64_t brt_get_ratio(spa_t *spa) { brt_t *brt = spa->spa_brt; if (brt->brt_usedspace == 0) return (100); return ((brt->brt_usedspace + brt->brt_savedspace) * 100 / brt->brt_usedspace); } static int brt_kstats_update(kstat_t *ksp, int rw) { brt_stats_t *bs = ksp->ks_data; if (rw == KSTAT_WRITE) return (EACCES); bs->brt_addref_entry_in_memory.value.ui64 = wmsum_value(&brt_sums.brt_addref_entry_in_memory); bs->brt_addref_entry_not_on_disk.value.ui64 = wmsum_value(&brt_sums.brt_addref_entry_not_on_disk); bs->brt_addref_entry_on_disk.value.ui64 = wmsum_value(&brt_sums.brt_addref_entry_on_disk); bs->brt_addref_entry_read_lost_race.value.ui64 = wmsum_value(&brt_sums.brt_addref_entry_read_lost_race); bs->brt_decref_entry_in_memory.value.ui64 = wmsum_value(&brt_sums.brt_decref_entry_in_memory); bs->brt_decref_entry_loaded_from_disk.value.ui64 = wmsum_value(&brt_sums.brt_decref_entry_loaded_from_disk); bs->brt_decref_entry_not_in_memory.value.ui64 = wmsum_value(&brt_sums.brt_decref_entry_not_in_memory); bs->brt_decref_entry_not_on_disk.value.ui64 = wmsum_value(&brt_sums.brt_decref_entry_not_on_disk); bs->brt_decref_entry_read_lost_race.value.ui64 = wmsum_value(&brt_sums.brt_decref_entry_read_lost_race); bs->brt_decref_entry_still_referenced.value.ui64 = wmsum_value(&brt_sums.brt_decref_entry_still_referenced); bs->brt_decref_free_data_later.value.ui64 = wmsum_value(&brt_sums.brt_decref_free_data_later); bs->brt_decref_free_data_now.value.ui64 = wmsum_value(&brt_sums.brt_decref_free_data_now); bs->brt_decref_no_entry.value.ui64 = wmsum_value(&brt_sums.brt_decref_no_entry); return (0); } static void brt_stat_init(void) { wmsum_init(&brt_sums.brt_addref_entry_in_memory, 0); wmsum_init(&brt_sums.brt_addref_entry_not_on_disk, 0); wmsum_init(&brt_sums.brt_addref_entry_on_disk, 0); wmsum_init(&brt_sums.brt_addref_entry_read_lost_race, 0); wmsum_init(&brt_sums.brt_decref_entry_in_memory, 0); wmsum_init(&brt_sums.brt_decref_entry_loaded_from_disk, 0); wmsum_init(&brt_sums.brt_decref_entry_not_in_memory, 0); wmsum_init(&brt_sums.brt_decref_entry_not_on_disk, 0); wmsum_init(&brt_sums.brt_decref_entry_read_lost_race, 0); wmsum_init(&brt_sums.brt_decref_entry_still_referenced, 0); wmsum_init(&brt_sums.brt_decref_free_data_later, 0); wmsum_init(&brt_sums.brt_decref_free_data_now, 0); wmsum_init(&brt_sums.brt_decref_no_entry, 0); brt_ksp = kstat_create("zfs", 0, "brtstats", "misc", KSTAT_TYPE_NAMED, sizeof (brt_stats) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL); if (brt_ksp != NULL) { brt_ksp->ks_data = &brt_stats; brt_ksp->ks_update = brt_kstats_update; kstat_install(brt_ksp); } } static void brt_stat_fini(void) { if (brt_ksp != NULL) { kstat_delete(brt_ksp); brt_ksp = NULL; } wmsum_fini(&brt_sums.brt_addref_entry_in_memory); wmsum_fini(&brt_sums.brt_addref_entry_not_on_disk); wmsum_fini(&brt_sums.brt_addref_entry_on_disk); wmsum_fini(&brt_sums.brt_addref_entry_read_lost_race); wmsum_fini(&brt_sums.brt_decref_entry_in_memory); wmsum_fini(&brt_sums.brt_decref_entry_loaded_from_disk); wmsum_fini(&brt_sums.brt_decref_entry_not_in_memory); wmsum_fini(&brt_sums.brt_decref_entry_not_on_disk); wmsum_fini(&brt_sums.brt_decref_entry_read_lost_race); wmsum_fini(&brt_sums.brt_decref_entry_still_referenced); wmsum_fini(&brt_sums.brt_decref_free_data_later); wmsum_fini(&brt_sums.brt_decref_free_data_now); wmsum_fini(&brt_sums.brt_decref_no_entry); } void brt_init(void) { brt_entry_cache = kmem_cache_create("brt_entry_cache", sizeof (brt_entry_t), 0, NULL, NULL, NULL, NULL, NULL, 0); brt_pending_entry_cache = kmem_cache_create("brt_pending_entry_cache", sizeof (brt_pending_entry_t), 0, NULL, NULL, NULL, NULL, NULL, 0); brt_stat_init(); } void brt_fini(void) { brt_stat_fini(); kmem_cache_destroy(brt_entry_cache); kmem_cache_destroy(brt_pending_entry_cache); } static brt_entry_t * brt_entry_alloc(const brt_entry_t *bre_init) { brt_entry_t *bre; bre = kmem_cache_alloc(brt_entry_cache, KM_SLEEP); bre->bre_offset = bre_init->bre_offset; bre->bre_refcount = bre_init->bre_refcount; return (bre); } static void brt_entry_free(brt_entry_t *bre) { kmem_cache_free(brt_entry_cache, bre); } static void brt_entry_addref(brt_t *brt, const blkptr_t *bp) { brt_vdev_t *brtvd; brt_entry_t *bre, *racebre; brt_entry_t bre_search; avl_index_t where; uint64_t vdevid; int error; ASSERT(!RW_WRITE_HELD(&brt->brt_lock)); brt_entry_fill(bp, &bre_search, &vdevid); brt_wlock(brt); brtvd = brt_vdev(brt, vdevid); if (brtvd == NULL) { ASSERT3U(vdevid, >=, brt->brt_nvdevs); /* New VDEV was added. */ brt_vdevs_expand(brt, vdevid + 1); brtvd = brt_vdev(brt, vdevid); } ASSERT(brtvd != NULL); if (!brtvd->bv_initiated) brt_vdev_realloc(brt, brtvd); bre = avl_find(&brtvd->bv_tree, &bre_search, NULL); if (bre != NULL) { BRTSTAT_BUMP(brt_addref_entry_in_memory); } else { /* * brt_entry_lookup() may drop the BRT (read) lock and * reacquire it (write). */ error = brt_entry_lookup(brt, brtvd, &bre_search); /* bre_search now contains correct bre_refcount */ ASSERT(error == 0 || error == ENOENT); if (error == 0) BRTSTAT_BUMP(brt_addref_entry_on_disk); else BRTSTAT_BUMP(brt_addref_entry_not_on_disk); /* * When the BRT lock was dropped, brt_vdevs[] may have been * expanded and reallocated, we need to update brtvd's pointer. */ brtvd = brt_vdev(brt, vdevid); ASSERT(brtvd != NULL); racebre = avl_find(&brtvd->bv_tree, &bre_search, &where); if (racebre == NULL) { bre = brt_entry_alloc(&bre_search); ASSERT(RW_WRITE_HELD(&brt->brt_lock)); avl_insert(&brtvd->bv_tree, bre, where); brt->brt_nentries++; } else { /* * The entry was added when the BRT lock was dropped in * brt_entry_lookup(). */ BRTSTAT_BUMP(brt_addref_entry_read_lost_race); bre = racebre; } } bre->bre_refcount++; brt_vdev_addref(brt, brtvd, bre, bp_get_dsize(brt->brt_spa, bp)); brt_unlock(brt); } /* Return TRUE if block should be freed immediately. */ boolean_t brt_entry_decref(spa_t *spa, const blkptr_t *bp) { brt_t *brt = spa->spa_brt; brt_vdev_t *brtvd; brt_entry_t *bre, *racebre; brt_entry_t bre_search; avl_index_t where; uint64_t vdevid; int error; brt_entry_fill(bp, &bre_search, &vdevid); brt_wlock(brt); brtvd = brt_vdev(brt, vdevid); ASSERT(brtvd != NULL); bre = avl_find(&brtvd->bv_tree, &bre_search, NULL); if (bre != NULL) { BRTSTAT_BUMP(brt_decref_entry_in_memory); goto out; } else { BRTSTAT_BUMP(brt_decref_entry_not_in_memory); } /* * brt_entry_lookup() may drop the BRT lock and reacquire it. */ error = brt_entry_lookup(brt, brtvd, &bre_search); /* bre_search now contains correct bre_refcount */ ASSERT(error == 0 || error == ENOENT); /* * When the BRT lock was dropped, brt_vdevs[] may have been expanded * and reallocated, we need to update brtvd's pointer. */ brtvd = brt_vdev(brt, vdevid); ASSERT(brtvd != NULL); if (error == ENOENT) { BRTSTAT_BUMP(brt_decref_entry_not_on_disk); bre = NULL; goto out; } racebre = avl_find(&brtvd->bv_tree, &bre_search, &where); if (racebre != NULL) { /* * The entry was added when the BRT lock was dropped in * brt_entry_lookup(). */ BRTSTAT_BUMP(brt_decref_entry_read_lost_race); bre = racebre; goto out; } BRTSTAT_BUMP(brt_decref_entry_loaded_from_disk); bre = brt_entry_alloc(&bre_search); ASSERT(RW_WRITE_HELD(&brt->brt_lock)); avl_insert(&brtvd->bv_tree, bre, where); brt->brt_nentries++; out: if (bre == NULL) { /* * This is a free of a regular (not cloned) block. */ brt_unlock(brt); BRTSTAT_BUMP(brt_decref_no_entry); return (B_TRUE); } if (bre->bre_refcount == 0) { brt_unlock(brt); BRTSTAT_BUMP(brt_decref_free_data_now); return (B_TRUE); } ASSERT(bre->bre_refcount > 0); bre->bre_refcount--; if (bre->bre_refcount == 0) BRTSTAT_BUMP(brt_decref_free_data_later); else BRTSTAT_BUMP(brt_decref_entry_still_referenced); brt_vdev_decref(brt, brtvd, bre, bp_get_dsize(brt->brt_spa, bp)); brt_unlock(brt); return (B_FALSE); } uint64_t brt_entry_get_refcount(spa_t *spa, const blkptr_t *bp) { brt_t *brt = spa->spa_brt; brt_vdev_t *brtvd; brt_entry_t bre_search, *bre; uint64_t vdevid, refcnt; int error; brt_entry_fill(bp, &bre_search, &vdevid); brt_rlock(brt); brtvd = brt_vdev(brt, vdevid); ASSERT(brtvd != NULL); bre = avl_find(&brtvd->bv_tree, &bre_search, NULL); if (bre == NULL) { error = brt_entry_lookup(brt, brtvd, &bre_search); ASSERT(error == 0 || error == ENOENT); if (error == ENOENT) refcnt = 0; else refcnt = bre_search.bre_refcount; } else refcnt = bre->bre_refcount; brt_unlock(brt); return (refcnt); } static void brt_prefetch(brt_t *brt, const blkptr_t *bp) { brt_entry_t bre; uint64_t vdevid; ASSERT(bp != NULL); if (!zfs_brt_prefetch) return; brt_entry_fill(bp, &bre, &vdevid); brt_entry_prefetch(brt, vdevid, &bre); } static int brt_pending_entry_compare(const void *x1, const void *x2) { const brt_pending_entry_t *bpe1 = x1, *bpe2 = x2; const blkptr_t *bp1 = &bpe1->bpe_bp, *bp2 = &bpe2->bpe_bp; int cmp; cmp = TREE_CMP(BP_PHYSICAL_BIRTH(bp1), BP_PHYSICAL_BIRTH(bp2)); if (cmp == 0) { cmp = TREE_CMP(DVA_GET_VDEV(&bp1->blk_dva[0]), DVA_GET_VDEV(&bp2->blk_dva[0])); if (cmp == 0) { cmp = TREE_CMP(DVA_GET_OFFSET(&bp1->blk_dva[0]), DVA_GET_OFFSET(&bp2->blk_dva[0])); } } return (cmp); } void brt_pending_add(spa_t *spa, const blkptr_t *bp, dmu_tx_t *tx) { brt_t *brt; avl_tree_t *pending_tree; kmutex_t *pending_lock; brt_pending_entry_t *bpe, *newbpe; avl_index_t where; uint64_t txg; brt = spa->spa_brt; txg = dmu_tx_get_txg(tx); ASSERT3U(txg, !=, 0); pending_tree = &brt->brt_pending_tree[txg & TXG_MASK]; pending_lock = &brt->brt_pending_lock[txg & TXG_MASK]; newbpe = kmem_cache_alloc(brt_pending_entry_cache, KM_SLEEP); newbpe->bpe_bp = *bp; newbpe->bpe_count = 1; mutex_enter(pending_lock); bpe = avl_find(pending_tree, newbpe, &where); if (bpe == NULL) { avl_insert(pending_tree, newbpe, where); newbpe = NULL; } else { bpe->bpe_count++; } mutex_exit(pending_lock); if (newbpe != NULL) { ASSERT(bpe != NULL); ASSERT(bpe != newbpe); kmem_cache_free(brt_pending_entry_cache, newbpe); } else { ASSERT(bpe == NULL); } /* Prefetch BRT entry, as we will need it in the syncing context. */ brt_prefetch(brt, bp); } void brt_pending_remove(spa_t *spa, const blkptr_t *bp, dmu_tx_t *tx) { brt_t *brt; avl_tree_t *pending_tree; kmutex_t *pending_lock; brt_pending_entry_t *bpe, bpe_search; uint64_t txg; brt = spa->spa_brt; txg = dmu_tx_get_txg(tx); ASSERT3U(txg, !=, 0); pending_tree = &brt->brt_pending_tree[txg & TXG_MASK]; pending_lock = &brt->brt_pending_lock[txg & TXG_MASK]; bpe_search.bpe_bp = *bp; mutex_enter(pending_lock); bpe = avl_find(pending_tree, &bpe_search, NULL); /* I believe we should always find bpe when this function is called. */ if (bpe != NULL) { ASSERT(bpe->bpe_count > 0); bpe->bpe_count--; if (bpe->bpe_count == 0) { avl_remove(pending_tree, bpe); kmem_cache_free(brt_pending_entry_cache, bpe); } } mutex_exit(pending_lock); } void brt_pending_apply(spa_t *spa, uint64_t txg) { brt_t *brt; brt_pending_entry_t *bpe; avl_tree_t *pending_tree; kmutex_t *pending_lock; void *c; ASSERT3U(txg, !=, 0); brt = spa->spa_brt; pending_tree = &brt->brt_pending_tree[txg & TXG_MASK]; pending_lock = &brt->brt_pending_lock[txg & TXG_MASK]; mutex_enter(pending_lock); c = NULL; while ((bpe = avl_destroy_nodes(pending_tree, &c)) != NULL) { boolean_t added_to_ddt; mutex_exit(pending_lock); for (int i = 0; i < bpe->bpe_count; i++) { /* * If the block has DEDUP bit set, it means that it * already exists in the DEDUP table, so we can just * use that instead of creating new entry in * the BRT table. */ if (BP_GET_DEDUP(&bpe->bpe_bp)) { added_to_ddt = ddt_addref(spa, &bpe->bpe_bp); } else { added_to_ddt = B_FALSE; } if (!added_to_ddt) brt_entry_addref(brt, &bpe->bpe_bp); } kmem_cache_free(brt_pending_entry_cache, bpe); mutex_enter(pending_lock); } mutex_exit(pending_lock); } static void brt_sync_entry(brt_t *brt, brt_vdev_t *brtvd, brt_entry_t *bre, dmu_tx_t *tx) { ASSERT(RW_WRITE_HELD(&brt->brt_lock)); ASSERT(brtvd->bv_mos_entries != 0); if (bre->bre_refcount == 0) { int error; error = brt_entry_remove(brt, brtvd, bre, tx); ASSERT(error == 0 || error == ENOENT); /* * If error == ENOENT then zfs_clone_range() was done from a * removed (but opened) file (open(), unlink()). */ ASSERT(brt_entry_lookup(brt, brtvd, bre) == ENOENT); } else { VERIFY0(brt_entry_update(brt, brtvd, bre, tx)); } } static void brt_sync_table(brt_t *brt, dmu_tx_t *tx) { brt_vdev_t *brtvd; brt_entry_t *bre; uint64_t vdevid; void *c; brt_wlock(brt); for (vdevid = 0; vdevid < brt->brt_nvdevs; vdevid++) { brtvd = &brt->brt_vdevs[vdevid]; if (!brtvd->bv_initiated) continue; if (!brtvd->bv_meta_dirty) { ASSERT(!brtvd->bv_entcount_dirty); ASSERT0(avl_numnodes(&brtvd->bv_tree)); continue; } ASSERT(!brtvd->bv_entcount_dirty || avl_numnodes(&brtvd->bv_tree) != 0); if (brtvd->bv_mos_brtvdev == 0) brt_vdev_create(brt, brtvd, tx); c = NULL; while ((bre = avl_destroy_nodes(&brtvd->bv_tree, &c)) != NULL) { brt_sync_entry(brt, brtvd, bre, tx); brt_entry_free(bre); ASSERT(brt->brt_nentries > 0); brt->brt_nentries--; } brt_vdev_sync(brt, brtvd, tx); if (brtvd->bv_totalcount == 0) brt_vdev_destroy(brt, brtvd, tx); } ASSERT0(brt->brt_nentries); brt_unlock(brt); } void brt_sync(spa_t *spa, uint64_t txg) { dmu_tx_t *tx; brt_t *brt; ASSERT(spa_syncing_txg(spa) == txg); brt = spa->spa_brt; brt_rlock(brt); if (brt->brt_nentries == 0) { /* No changes. */ brt_unlock(brt); return; } brt_unlock(brt); tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); brt_sync_table(brt, tx); dmu_tx_commit(tx); } static void brt_table_alloc(brt_t *brt) { for (int i = 0; i < TXG_SIZE; i++) { avl_create(&brt->brt_pending_tree[i], brt_pending_entry_compare, sizeof (brt_pending_entry_t), offsetof(brt_pending_entry_t, bpe_node)); mutex_init(&brt->brt_pending_lock[i], NULL, MUTEX_DEFAULT, NULL); } } static void brt_table_free(brt_t *brt) { for (int i = 0; i < TXG_SIZE; i++) { ASSERT(avl_is_empty(&brt->brt_pending_tree[i])); avl_destroy(&brt->brt_pending_tree[i]); mutex_destroy(&brt->brt_pending_lock[i]); } } static void brt_alloc(spa_t *spa) { brt_t *brt; ASSERT(spa->spa_brt == NULL); brt = kmem_zalloc(sizeof (*brt), KM_SLEEP); rw_init(&brt->brt_lock, NULL, RW_DEFAULT, NULL); brt->brt_spa = spa; brt->brt_rangesize = 0; brt->brt_nentries = 0; brt->brt_vdevs = NULL; brt->brt_nvdevs = 0; brt_table_alloc(brt); spa->spa_brt = brt; } void brt_create(spa_t *spa) { brt_alloc(spa); brt_vdevs_alloc(spa->spa_brt, B_FALSE); } int brt_load(spa_t *spa) { brt_alloc(spa); brt_vdevs_alloc(spa->spa_brt, B_TRUE); return (0); } void brt_unload(spa_t *spa) { brt_t *brt = spa->spa_brt; if (brt == NULL) return; brt_vdevs_free(brt); brt_table_free(brt); rw_destroy(&brt->brt_lock); kmem_free(brt, sizeof (*brt)); spa->spa_brt = NULL; } /* BEGIN CSTYLED */ ZFS_MODULE_PARAM(zfs_brt, zfs_brt_, prefetch, INT, ZMOD_RW, "Enable prefetching of BRT entries"); #ifdef ZFS_BRT_DEBUG ZFS_MODULE_PARAM(zfs_brt, zfs_brt_, debug, INT, ZMOD_RW, "BRT debug"); #endif /* END CSTYLED */ diff --git a/module/zfs/zfs_vnops.c b/module/zfs/zfs_vnops.c index 8ffcf371344e..a3d54c6e6792 100644 --- a/module/zfs/zfs_vnops.c +++ b/module/zfs/zfs_vnops.c @@ -1,1464 +1,1465 @@ /* * 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 https://opensource.org/licenses/CDDL-1.0. * 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, 2018 by Delphix. All rights reserved. * Copyright (c) 2015 by Chunwei Chen. All rights reserved. * Copyright 2017 Nexenta Systems, Inc. * Copyright (c) 2021, 2022 by Pawel Jakub Dawidek */ /* Portions Copyright 2007 Jeremy Teo */ /* Portions Copyright 2010 Robert Milkowski */ #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 ulong_t zfs_fsync_sync_cnt = 4; int zfs_fsync(znode_t *zp, int syncflag, cred_t *cr) { int error = 0; zfsvfs_t *zfsvfs = ZTOZSB(zp); (void) tsd_set(zfs_fsyncer_key, (void *)(uintptr_t)zfs_fsync_sync_cnt); if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) { if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) goto out; atomic_inc_32(&zp->z_sync_writes_cnt); zil_commit(zfsvfs->z_log, zp->z_id); atomic_dec_32(&zp->z_sync_writes_cnt); zfs_exit(zfsvfs, FTAG); } out: tsd_set(zfs_fsyncer_key, NULL); return (error); } #if defined(SEEK_HOLE) && defined(SEEK_DATA) /* * Lseek support for finding holes (cmd == SEEK_HOLE) and * data (cmd == SEEK_DATA). "off" is an in/out parameter. */ static int zfs_holey_common(znode_t *zp, ulong_t cmd, loff_t *off) { zfs_locked_range_t *lr; uint64_t noff = (uint64_t)*off; /* new offset */ uint64_t file_sz; int error; boolean_t hole; file_sz = zp->z_size; if (noff >= file_sz) { return (SET_ERROR(ENXIO)); } if (cmd == F_SEEK_HOLE) hole = B_TRUE; else hole = B_FALSE; /* Flush any mmap()'d data to disk */ if (zn_has_cached_data(zp, 0, file_sz - 1)) zn_flush_cached_data(zp, B_FALSE); lr = zfs_rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_READER); error = dmu_offset_next(ZTOZSB(zp)->z_os, zp->z_id, hole, &noff); zfs_rangelock_exit(lr); if (error == ESRCH) return (SET_ERROR(ENXIO)); /* File was dirty, so fall back to using generic logic */ if (error == EBUSY) { if (hole) *off = file_sz; return (0); } /* * We could find a hole that begins after the logical end-of-file, * because dmu_offset_next() only works on whole blocks. If the * EOF falls mid-block, then indicate that the "virtual hole" * at the end of the file begins at the logical EOF, rather than * at the end of the last block. */ if (noff > file_sz) { ASSERT(hole); noff = file_sz; } if (noff < *off) return (error); *off = noff; return (error); } int zfs_holey(znode_t *zp, ulong_t cmd, loff_t *off) { zfsvfs_t *zfsvfs = ZTOZSB(zp); int error; if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) return (error); error = zfs_holey_common(zp, cmd, off); zfs_exit(zfsvfs, FTAG); return (error); } #endif /* SEEK_HOLE && SEEK_DATA */ int zfs_access(znode_t *zp, int mode, int flag, cred_t *cr) { zfsvfs_t *zfsvfs = ZTOZSB(zp); int error; if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) return (error); if (flag & V_ACE_MASK) #if defined(__linux__) error = zfs_zaccess(zp, mode, flag, B_FALSE, cr, zfs_init_idmap); #else error = zfs_zaccess(zp, mode, flag, B_FALSE, cr, NULL); #endif else #if defined(__linux__) error = zfs_zaccess_rwx(zp, mode, flag, cr, zfs_init_idmap); #else error = zfs_zaccess_rwx(zp, mode, flag, cr, NULL); #endif zfs_exit(zfsvfs, FTAG); return (error); } static uint64_t zfs_vnops_read_chunk_size = 1024 * 1024; /* Tunable */ /* * Read bytes from specified file into supplied buffer. * * IN: zp - inode of file to be read from. * uio - structure supplying read location, range info, * and return buffer. * ioflag - O_SYNC flags; used to provide FRSYNC semantics. * O_DIRECT flag; used to bypass page cache. * cr - credentials of caller. * * OUT: uio - updated offset and range, buffer filled. * * RETURN: 0 on success, error code on failure. * * Side Effects: * inode - atime updated if byte count > 0 */ int zfs_read(struct znode *zp, zfs_uio_t *uio, int ioflag, cred_t *cr) { (void) cr; int error = 0; boolean_t frsync = B_FALSE; zfsvfs_t *zfsvfs = ZTOZSB(zp); if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) return (error); if (zp->z_pflags & ZFS_AV_QUARANTINED) { zfs_exit(zfsvfs, FTAG); return (SET_ERROR(EACCES)); } /* We don't copy out anything useful for directories. */ if (Z_ISDIR(ZTOTYPE(zp))) { zfs_exit(zfsvfs, FTAG); return (SET_ERROR(EISDIR)); } /* * Validate file offset */ if (zfs_uio_offset(uio) < (offset_t)0) { zfs_exit(zfsvfs, FTAG); return (SET_ERROR(EINVAL)); } /* * Fasttrack empty reads */ if (zfs_uio_resid(uio) == 0) { zfs_exit(zfsvfs, FTAG); return (0); } #ifdef FRSYNC /* * If we're in FRSYNC mode, sync out this znode before reading it. * Only do this for non-snapshots. * * Some platforms do not support FRSYNC and instead map it * to O_SYNC, which results in unnecessary calls to zil_commit. We * only honor FRSYNC requests on platforms which support it. */ frsync = !!(ioflag & FRSYNC); #endif if (zfsvfs->z_log && (frsync || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)) zil_commit(zfsvfs->z_log, zp->z_id); /* * Lock the range against changes. */ zfs_locked_range_t *lr = zfs_rangelock_enter(&zp->z_rangelock, zfs_uio_offset(uio), zfs_uio_resid(uio), RL_READER); /* * If we are reading past end-of-file we can skip * to the end; but we might still need to set atime. */ if (zfs_uio_offset(uio) >= zp->z_size) { error = 0; goto out; } ASSERT(zfs_uio_offset(uio) < zp->z_size); #if defined(__linux__) ssize_t start_offset = zfs_uio_offset(uio); #endif ssize_t n = MIN(zfs_uio_resid(uio), zp->z_size - zfs_uio_offset(uio)); ssize_t start_resid = n; while (n > 0) { ssize_t nbytes = MIN(n, zfs_vnops_read_chunk_size - P2PHASE(zfs_uio_offset(uio), zfs_vnops_read_chunk_size)); #ifdef UIO_NOCOPY if (zfs_uio_segflg(uio) == UIO_NOCOPY) error = mappedread_sf(zp, nbytes, uio); else #endif if (zn_has_cached_data(zp, zfs_uio_offset(uio), zfs_uio_offset(uio) + nbytes - 1) && !(ioflag & O_DIRECT)) { error = mappedread(zp, nbytes, uio); } else { error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl), uio, nbytes); } if (error) { /* convert checksum errors into IO errors */ if (error == ECKSUM) error = SET_ERROR(EIO); #if defined(__linux__) /* * if we actually read some bytes, bubbling EFAULT * up to become EAGAIN isn't what we want here... * * ...on Linux, at least. On FBSD, doing this breaks. */ if (error == EFAULT && (zfs_uio_offset(uio) - start_offset) != 0) error = 0; #endif break; } n -= nbytes; } int64_t nread = start_resid - n; dataset_kstats_update_read_kstats(&zfsvfs->z_kstat, nread); task_io_account_read(nread); out: zfs_rangelock_exit(lr); ZFS_ACCESSTIME_STAMP(zfsvfs, zp); zfs_exit(zfsvfs, FTAG); return (error); } static void zfs_clear_setid_bits_if_necessary(zfsvfs_t *zfsvfs, znode_t *zp, cred_t *cr, uint64_t *clear_setid_bits_txgp, dmu_tx_t *tx) { zilog_t *zilog = zfsvfs->z_log; const uint64_t uid = KUID_TO_SUID(ZTOUID(zp)); ASSERT(clear_setid_bits_txgp != NULL); ASSERT(tx != NULL); /* * Clear Set-UID/Set-GID bits on successful write if not * privileged and at least one of the execute bits is set. * * It would be nice to do this after all writes have * been done, but that would still expose the ISUID/ISGID * to another app after the partial write is committed. * * Note: we don't call zfs_fuid_map_id() here because * user 0 is not an ephemeral uid. */ mutex_enter(&zp->z_acl_lock); if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) | (S_IXUSR >> 6))) != 0 && (zp->z_mode & (S_ISUID | S_ISGID)) != 0 && secpolicy_vnode_setid_retain(zp, cr, ((zp->z_mode & S_ISUID) != 0 && uid == 0)) != 0) { uint64_t newmode; zp->z_mode &= ~(S_ISUID | S_ISGID); newmode = zp->z_mode; (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), (void *)&newmode, sizeof (uint64_t), tx); mutex_exit(&zp->z_acl_lock); /* * Make sure SUID/SGID bits will be removed when we replay the * log. If the setid bits are keep coming back, don't log more * than one TX_SETATTR per transaction group. */ if (*clear_setid_bits_txgp != dmu_tx_get_txg(tx)) { vattr_t va = {0}; va.va_mask = ATTR_MODE; va.va_nodeid = zp->z_id; va.va_mode = newmode; zfs_log_setattr(zilog, tx, TX_SETATTR, zp, &va, ATTR_MODE, NULL); *clear_setid_bits_txgp = dmu_tx_get_txg(tx); } } else { mutex_exit(&zp->z_acl_lock); } } /* * Write the bytes to a file. * * IN: zp - znode of file to be written to. * uio - structure supplying write location, range info, * and data buffer. * ioflag - O_APPEND flag set if in append mode. * O_DIRECT flag; used to bypass page cache. * cr - credentials of caller. * * OUT: uio - updated offset and range. * * RETURN: 0 if success * error code if failure * * Timestamps: * ip - ctime|mtime updated if byte count > 0 */ int zfs_write(znode_t *zp, zfs_uio_t *uio, int ioflag, cred_t *cr) { int error = 0, error1; ssize_t start_resid = zfs_uio_resid(uio); uint64_t clear_setid_bits_txg = 0; /* * Fasttrack empty write */ ssize_t n = start_resid; if (n == 0) return (0); zfsvfs_t *zfsvfs = ZTOZSB(zp); if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) return (error); sa_bulk_attr_t bulk[4]; int count = 0; uint64_t mtime[2], ctime[2]; 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_SIZE(zfsvfs), NULL, &zp->z_size, 8); SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, &zp->z_pflags, 8); /* * 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, FTAG); return (SET_ERROR(EROFS)); } /* * If immutable or not appending then return EPERM. * Intentionally allow ZFS_READONLY through here. * See zfs_zaccess_common() */ if ((zp->z_pflags & ZFS_IMMUTABLE) || ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & O_APPEND) && (zfs_uio_offset(uio) < zp->z_size))) { zfs_exit(zfsvfs, FTAG); return (SET_ERROR(EPERM)); } /* * Validate file offset */ offset_t woff = ioflag & O_APPEND ? zp->z_size : zfs_uio_offset(uio); if (woff < 0) { zfs_exit(zfsvfs, FTAG); return (SET_ERROR(EINVAL)); } /* * Pre-fault the pages to ensure slow (eg NFS) pages * don't hold up txg. */ ssize_t pfbytes = MIN(n, DMU_MAX_ACCESS >> 1); if (zfs_uio_prefaultpages(pfbytes, uio)) { zfs_exit(zfsvfs, FTAG); return (SET_ERROR(EFAULT)); } /* * If in append mode, set the io offset pointer to eof. */ zfs_locked_range_t *lr; if (ioflag & O_APPEND) { /* * Obtain an appending range lock to guarantee file append * semantics. We reset the write offset once we have the lock. */ lr = zfs_rangelock_enter(&zp->z_rangelock, 0, n, RL_APPEND); woff = lr->lr_offset; if (lr->lr_length == UINT64_MAX) { /* * We overlocked the file because this write will cause * the file block size to increase. * Note that zp_size cannot change with this lock held. */ woff = zp->z_size; } zfs_uio_setoffset(uio, woff); } else { /* * Note that if the file block size will change as a result of * this write, then this range lock will lock the entire file * so that we can re-write the block safely. */ lr = zfs_rangelock_enter(&zp->z_rangelock, woff, n, RL_WRITER); } if (zn_rlimit_fsize_uio(zp, uio)) { zfs_rangelock_exit(lr); zfs_exit(zfsvfs, FTAG); return (SET_ERROR(EFBIG)); } const rlim64_t limit = MAXOFFSET_T; if (woff >= limit) { zfs_rangelock_exit(lr); zfs_exit(zfsvfs, FTAG); return (SET_ERROR(EFBIG)); } if (n > limit - woff) n = limit - woff; uint64_t end_size = MAX(zp->z_size, woff + n); zilog_t *zilog = zfsvfs->z_log; const uint64_t uid = KUID_TO_SUID(ZTOUID(zp)); const uint64_t gid = KGID_TO_SGID(ZTOGID(zp)); const uint64_t projid = zp->z_projid; /* * Write the file in reasonable size chunks. Each chunk is written * in a separate transaction; this keeps the intent log records small * and allows us to do more fine-grained space accounting. */ while (n > 0) { woff = zfs_uio_offset(uio); if (zfs_id_overblockquota(zfsvfs, DMU_USERUSED_OBJECT, uid) || zfs_id_overblockquota(zfsvfs, DMU_GROUPUSED_OBJECT, gid) || (projid != ZFS_DEFAULT_PROJID && zfs_id_overblockquota(zfsvfs, DMU_PROJECTUSED_OBJECT, projid))) { error = SET_ERROR(EDQUOT); break; } uint64_t blksz; if (lr->lr_length == UINT64_MAX && zp->z_size <= zp->z_blksz) { if (zp->z_blksz > zfsvfs->z_max_blksz && !ISP2(zp->z_blksz)) { /* * File's blocksize is already larger than the * "recordsize" property. Only let it grow to * the next power of 2. */ blksz = 1 << highbit64(zp->z_blksz); } else { blksz = zfsvfs->z_max_blksz; } blksz = MIN(blksz, P2ROUNDUP(end_size, SPA_MINBLOCKSIZE)); blksz = MAX(blksz, zp->z_blksz); } else { blksz = zp->z_blksz; } arc_buf_t *abuf = NULL; ssize_t nbytes = n; if (n >= blksz && woff >= zp->z_size && P2PHASE(woff, blksz) == 0 && (blksz >= SPA_OLD_MAXBLOCKSIZE || n < 4 * blksz)) { /* * This write covers a full block. "Borrow" a buffer * from the dmu so that we can fill it before we enter * a transaction. This avoids the possibility of * holding up the transaction if the data copy hangs * up on a pagefault (e.g., from an NFS server mapping). */ abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), blksz); ASSERT(abuf != NULL); ASSERT(arc_buf_size(abuf) == blksz); if ((error = zfs_uiocopy(abuf->b_data, blksz, UIO_WRITE, uio, &nbytes))) { dmu_return_arcbuf(abuf); break; } ASSERT3S(nbytes, ==, blksz); } else { nbytes = MIN(n, (DMU_MAX_ACCESS >> 1) - P2PHASE(woff, blksz)); if (pfbytes < nbytes) { if (zfs_uio_prefaultpages(nbytes, uio)) { error = SET_ERROR(EFAULT); break; } pfbytes = nbytes; } } /* * Start a transaction. */ dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); dmu_buf_impl_t *db = (dmu_buf_impl_t *)sa_get_db(zp->z_sa_hdl); DB_DNODE_ENTER(db); dmu_tx_hold_write_by_dnode(tx, DB_DNODE(db), woff, nbytes); DB_DNODE_EXIT(db); zfs_sa_upgrade_txholds(tx, zp); error = dmu_tx_assign(tx, TXG_WAIT); if (error) { dmu_tx_abort(tx); if (abuf != NULL) dmu_return_arcbuf(abuf); break; } /* * NB: We must call zfs_clear_setid_bits_if_necessary before * committing the transaction! */ /* * If rangelock_enter() over-locked we grow the blocksize * and then reduce the lock range. This will only happen * on the first iteration since rangelock_reduce() will * shrink down lr_length to the appropriate size. */ if (lr->lr_length == UINT64_MAX) { zfs_grow_blocksize(zp, blksz, tx); zfs_rangelock_reduce(lr, woff, n); } ssize_t tx_bytes; if (abuf == NULL) { tx_bytes = zfs_uio_resid(uio); zfs_uio_fault_disable(uio, B_TRUE); error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl), uio, nbytes, tx); zfs_uio_fault_disable(uio, B_FALSE); #ifdef __linux__ if (error == EFAULT) { zfs_clear_setid_bits_if_necessary(zfsvfs, zp, cr, &clear_setid_bits_txg, tx); dmu_tx_commit(tx); /* * Account for partial writes before * continuing the loop. * Update needs to occur before the next * zfs_uio_prefaultpages, or prefaultpages may * error, and we may break the loop early. */ n -= tx_bytes - zfs_uio_resid(uio); pfbytes -= tx_bytes - zfs_uio_resid(uio); continue; } #endif /* * On FreeBSD, EFAULT should be propagated back to the * VFS, which will handle faulting and will retry. */ if (error != 0 && error != EFAULT) { zfs_clear_setid_bits_if_necessary(zfsvfs, zp, cr, &clear_setid_bits_txg, tx); dmu_tx_commit(tx); break; } tx_bytes -= zfs_uio_resid(uio); } else { /* * Thus, we're writing a full block at a block-aligned * offset and extending the file past EOF. * * dmu_assign_arcbuf_by_dbuf() will directly assign the * arc buffer to a dbuf. */ error = dmu_assign_arcbuf_by_dbuf( sa_get_db(zp->z_sa_hdl), woff, abuf, tx); if (error != 0) { /* * XXX This might not be necessary if * dmu_assign_arcbuf_by_dbuf is guaranteed * to be atomic. */ zfs_clear_setid_bits_if_necessary(zfsvfs, zp, cr, &clear_setid_bits_txg, tx); dmu_return_arcbuf(abuf); dmu_tx_commit(tx); break; } ASSERT3S(nbytes, <=, zfs_uio_resid(uio)); zfs_uioskip(uio, nbytes); tx_bytes = nbytes; } if (tx_bytes && zn_has_cached_data(zp, woff, woff + tx_bytes - 1) && !(ioflag & O_DIRECT)) { update_pages(zp, woff, tx_bytes, zfsvfs->z_os); } /* * If we made no progress, we're done. If we made even * partial progress, update the znode and ZIL accordingly. */ if (tx_bytes == 0) { (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), (void *)&zp->z_size, sizeof (uint64_t), tx); dmu_tx_commit(tx); ASSERT(error != 0); break; } zfs_clear_setid_bits_if_necessary(zfsvfs, zp, cr, &clear_setid_bits_txg, tx); zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime); /* * Update the file size (zp_size) if it has changed; * account for possible concurrent updates. */ while ((end_size = zp->z_size) < zfs_uio_offset(uio)) { (void) atomic_cas_64(&zp->z_size, end_size, zfs_uio_offset(uio)); ASSERT(error == 0 || error == EFAULT); } /* * If we are replaying and eof is non zero then force * the file size to the specified eof. Note, there's no * concurrency during replay. */ if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0) zp->z_size = zfsvfs->z_replay_eof; error1 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); if (error1 != 0) /* Avoid clobbering EFAULT. */ error = error1; /* * NB: During replay, the TX_SETATTR record logged by * zfs_clear_setid_bits_if_necessary must precede any of * the TX_WRITE records logged here. */ zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag, NULL, NULL); dmu_tx_commit(tx); if (error != 0) break; ASSERT3S(tx_bytes, ==, nbytes); n -= nbytes; pfbytes -= nbytes; } zfs_znode_update_vfs(zp); zfs_rangelock_exit(lr); /* * If we're in replay mode, or we made no progress, or the * uio data is inaccessible return an error. Otherwise, it's * at least a partial write, so it's successful. */ if (zfsvfs->z_replay || zfs_uio_resid(uio) == start_resid || error == EFAULT) { zfs_exit(zfsvfs, FTAG); return (error); } if (ioflag & (O_SYNC | O_DSYNC) || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) zil_commit(zilog, zp->z_id); const int64_t nwritten = start_resid - zfs_uio_resid(uio); dataset_kstats_update_write_kstats(&zfsvfs->z_kstat, nwritten); task_io_account_write(nwritten); zfs_exit(zfsvfs, FTAG); return (0); } int zfs_getsecattr(znode_t *zp, vsecattr_t *vsecp, int flag, cred_t *cr) { zfsvfs_t *zfsvfs = ZTOZSB(zp); int error; boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) return (error); error = zfs_getacl(zp, vsecp, skipaclchk, cr); zfs_exit(zfsvfs, FTAG); return (error); } int zfs_setsecattr(znode_t *zp, vsecattr_t *vsecp, int flag, cred_t *cr) { zfsvfs_t *zfsvfs = ZTOZSB(zp); int error; boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; zilog_t *zilog = zfsvfs->z_log; if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) return (error); error = zfs_setacl(zp, vsecp, skipaclchk, cr); if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) zil_commit(zilog, 0); zfs_exit(zfsvfs, FTAG); return (error); } #ifdef ZFS_DEBUG static int zil_fault_io = 0; #endif static void zfs_get_done(zgd_t *zgd, int error); /* * Get data to generate a TX_WRITE intent log record. */ int zfs_get_data(void *arg, uint64_t gen, lr_write_t *lr, char *buf, struct lwb *lwb, zio_t *zio) { zfsvfs_t *zfsvfs = arg; objset_t *os = zfsvfs->z_os; znode_t *zp; uint64_t object = lr->lr_foid; uint64_t offset = lr->lr_offset; uint64_t size = lr->lr_length; dmu_buf_t *db; zgd_t *zgd; int error = 0; uint64_t zp_gen; ASSERT3P(lwb, !=, NULL); ASSERT3U(size, !=, 0); /* * Nothing to do if the file has been removed */ if (zfs_zget(zfsvfs, object, &zp) != 0) return (SET_ERROR(ENOENT)); if (zp->z_unlinked) { /* * Release the vnode asynchronously as we currently have the * txg stopped from syncing. */ zfs_zrele_async(zp); return (SET_ERROR(ENOENT)); } /* check if generation number matches */ if (sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen, sizeof (zp_gen)) != 0) { zfs_zrele_async(zp); return (SET_ERROR(EIO)); } if (zp_gen != gen) { zfs_zrele_async(zp); return (SET_ERROR(ENOENT)); } zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP); zgd->zgd_lwb = lwb; zgd->zgd_private = zp; /* * Write records come in two flavors: immediate and indirect. * For small writes it's cheaper to store the data with the * log record (immediate); for large writes it's cheaper to * sync the data and get a pointer to it (indirect) so that * we don't have to write the data twice. */ if (buf != NULL) { /* immediate write */ zgd->zgd_lr = zfs_rangelock_enter(&zp->z_rangelock, offset, size, RL_READER); /* test for truncation needs to be done while range locked */ if (offset >= zp->z_size) { error = SET_ERROR(ENOENT); } else { error = dmu_read(os, object, offset, size, buf, DMU_READ_NO_PREFETCH); } ASSERT(error == 0 || error == ENOENT); } else { /* indirect write */ ASSERT3P(zio, !=, NULL); /* * Have to lock the whole block to ensure when it's * written out and its checksum is being calculated * that no one can change the data. We need to re-check * blocksize after we get the lock in case it's changed! */ for (;;) { uint64_t blkoff; size = zp->z_blksz; blkoff = ISP2(size) ? P2PHASE(offset, size) : offset; offset -= blkoff; zgd->zgd_lr = zfs_rangelock_enter(&zp->z_rangelock, offset, size, RL_READER); if (zp->z_blksz == size) break; offset += blkoff; zfs_rangelock_exit(zgd->zgd_lr); } /* test for truncation needs to be done while range locked */ if (lr->lr_offset >= zp->z_size) error = SET_ERROR(ENOENT); #ifdef ZFS_DEBUG if (zil_fault_io) { error = SET_ERROR(EIO); zil_fault_io = 0; } #endif if (error == 0) error = dmu_buf_hold_noread(os, object, offset, zgd, &db); if (error == 0) { blkptr_t *bp = &lr->lr_blkptr; zgd->zgd_db = db; zgd->zgd_bp = bp; ASSERT(db->db_offset == offset); ASSERT(db->db_size == size); error = dmu_sync(zio, lr->lr_common.lrc_txg, zfs_get_done, zgd); ASSERT(error || lr->lr_length <= size); /* * On success, we need to wait for the write I/O * initiated by dmu_sync() to complete before we can * release this dbuf. We will finish everything up * in the zfs_get_done() callback. */ if (error == 0) return (0); if (error == EALREADY) { lr->lr_common.lrc_txtype = TX_WRITE2; /* * TX_WRITE2 relies on the data previously * written by the TX_WRITE that caused * EALREADY. We zero out the BP because * it is the old, currently-on-disk BP. */ zgd->zgd_bp = NULL; BP_ZERO(bp); error = 0; } } } zfs_get_done(zgd, error); return (error); } static void zfs_get_done(zgd_t *zgd, int error) { (void) error; znode_t *zp = zgd->zgd_private; if (zgd->zgd_db) dmu_buf_rele(zgd->zgd_db, zgd); zfs_rangelock_exit(zgd->zgd_lr); /* * Release the vnode asynchronously as we currently have the * txg stopped from syncing. */ zfs_zrele_async(zp); kmem_free(zgd, sizeof (zgd_t)); } static int zfs_enter_two(zfsvfs_t *zfsvfs1, zfsvfs_t *zfsvfs2, const char *tag) { int error; /* Swap. Not sure if the order of zfs_enter()s is important. */ if (zfsvfs1 > zfsvfs2) { zfsvfs_t *tmpzfsvfs; tmpzfsvfs = zfsvfs2; zfsvfs2 = zfsvfs1; zfsvfs1 = tmpzfsvfs; } error = zfs_enter(zfsvfs1, tag); if (error != 0) return (error); if (zfsvfs1 != zfsvfs2) { error = zfs_enter(zfsvfs2, tag); if (error != 0) { zfs_exit(zfsvfs1, tag); return (error); } } return (0); } static void zfs_exit_two(zfsvfs_t *zfsvfs1, zfsvfs_t *zfsvfs2, const char *tag) { zfs_exit(zfsvfs1, tag); if (zfsvfs1 != zfsvfs2) zfs_exit(zfsvfs2, tag); } /* * We split each clone request in chunks that can fit into a single ZIL * log entry. Each ZIL log entry can fit 130816 bytes for a block cloning * operation (see zil_max_log_data() and zfs_log_clone_range()). This gives * us room for storing 1022 block pointers. * * On success, the function return the number of bytes copied in *lenp. * Note, it doesn't return how much bytes are left to be copied. + * On errors which are caused by any file system limitations or + * brt limitations `EINVAL` is returned. In the most cases a user + * requested bad parameters, it could be possible to clone the file but + * some parameters don't match the requirements. */ int zfs_clone_range(znode_t *inzp, uint64_t *inoffp, znode_t *outzp, uint64_t *outoffp, uint64_t *lenp, cred_t *cr) { zfsvfs_t *inzfsvfs, *outzfsvfs; objset_t *inos, *outos; zfs_locked_range_t *inlr, *outlr; dmu_buf_impl_t *db; dmu_tx_t *tx; zilog_t *zilog; uint64_t inoff, outoff, len, done; uint64_t outsize, size; int error; int count = 0; sa_bulk_attr_t bulk[3]; uint64_t mtime[2], ctime[2]; uint64_t uid, gid, projid; blkptr_t *bps; size_t maxblocks, nbps; uint_t inblksz; uint64_t clear_setid_bits_txg = 0; inoff = *inoffp; outoff = *outoffp; len = *lenp; done = 0; inzfsvfs = ZTOZSB(inzp); outzfsvfs = ZTOZSB(outzp); /* * We need to call zfs_enter() potentially on two different datasets, * so we need a dedicated function for that. */ error = zfs_enter_two(inzfsvfs, outzfsvfs, FTAG); if (error != 0) return (error); inos = inzfsvfs->z_os; outos = outzfsvfs->z_os; /* * Both source and destination have to belong to the same storage pool. */ if (dmu_objset_spa(inos) != dmu_objset_spa(outos)) { zfs_exit_two(inzfsvfs, outzfsvfs, FTAG); return (SET_ERROR(EXDEV)); } /* * outos and inos belongs to the same storage pool. * see a few lines above, only one check. */ if (!spa_feature_is_enabled(dmu_objset_spa(outos), SPA_FEATURE_BLOCK_CLONING)) { zfs_exit_two(inzfsvfs, outzfsvfs, FTAG); return (SET_ERROR(EOPNOTSUPP)); } ASSERT(!outzfsvfs->z_replay); error = zfs_verify_zp(inzp); if (error == 0) error = zfs_verify_zp(outzp); if (error != 0) { zfs_exit_two(inzfsvfs, outzfsvfs, FTAG); return (error); } /* * We don't copy source file's flags that's why we don't allow to clone * files that are in quarantine. */ if (inzp->z_pflags & ZFS_AV_QUARANTINED) { zfs_exit_two(inzfsvfs, outzfsvfs, FTAG); return (SET_ERROR(EACCES)); } if (inoff >= inzp->z_size) { *lenp = 0; zfs_exit_two(inzfsvfs, outzfsvfs, FTAG); return (0); } if (len > inzp->z_size - inoff) { len = inzp->z_size - inoff; } if (len == 0) { *lenp = 0; zfs_exit_two(inzfsvfs, outzfsvfs, FTAG); return (0); } /* * Callers might not be able to detect properly that we are read-only, * so check it explicitly here. */ if (zfs_is_readonly(outzfsvfs)) { zfs_exit_two(inzfsvfs, outzfsvfs, FTAG); return (SET_ERROR(EROFS)); } /* * If immutable or not appending then return EPERM. * Intentionally allow ZFS_READONLY through here. * See zfs_zaccess_common() */ if ((outzp->z_pflags & ZFS_IMMUTABLE) != 0) { zfs_exit_two(inzfsvfs, outzfsvfs, FTAG); return (SET_ERROR(EPERM)); } /* * No overlapping if we are cloning within the same file. */ if (inzp == outzp) { if (inoff < outoff + len && outoff < inoff + len) { zfs_exit_two(inzfsvfs, outzfsvfs, FTAG); return (SET_ERROR(EINVAL)); } } /* * Maintain predictable lock order. */ if (inzp < outzp || (inzp == outzp && inoff < outoff)) { inlr = zfs_rangelock_enter(&inzp->z_rangelock, inoff, len, RL_READER); outlr = zfs_rangelock_enter(&outzp->z_rangelock, outoff, len, RL_WRITER); } else { outlr = zfs_rangelock_enter(&outzp->z_rangelock, outoff, len, RL_WRITER); inlr = zfs_rangelock_enter(&inzp->z_rangelock, inoff, len, RL_READER); } inblksz = inzp->z_blksz; /* * We cannot clone into files with different block size. */ if (inblksz != outzp->z_blksz && outzp->z_size > inblksz) { - error = SET_ERROR(EXDEV); + error = SET_ERROR(EINVAL); goto unlock; } /* * Offsets and len must be at block boundries. */ if ((inoff % inblksz) != 0 || (outoff % inblksz) != 0) { - error = SET_ERROR(EXDEV); + error = SET_ERROR(EINVAL); goto unlock; } /* * Length must be multipe of blksz, except for the end of the file. */ if ((len % inblksz) != 0 && (len < inzp->z_size - inoff || len < outzp->z_size - outoff)) { - error = SET_ERROR(EXDEV); + error = SET_ERROR(EINVAL); goto unlock; } error = zn_rlimit_fsize(outoff + len); if (error != 0) { goto unlock; } if (inoff >= MAXOFFSET_T || outoff >= MAXOFFSET_T) { error = SET_ERROR(EFBIG); goto unlock; } SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(outzfsvfs), NULL, &mtime, 16); SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(outzfsvfs), NULL, &ctime, 16); SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(outzfsvfs), NULL, &outzp->z_size, 8); zilog = outzfsvfs->z_log; maxblocks = zil_max_log_data(zilog, sizeof (lr_clone_range_t)) / sizeof (bps[0]); uid = KUID_TO_SUID(ZTOUID(outzp)); gid = KGID_TO_SGID(ZTOGID(outzp)); projid = outzp->z_projid; bps = vmem_alloc(sizeof (bps[0]) * maxblocks, KM_SLEEP); /* * Clone the file in reasonable size chunks. Each chunk is cloned * in a separate transaction; this keeps the intent log records small * and allows us to do more fine-grained space accounting. */ while (len > 0) { size = MIN(inblksz * maxblocks, len); if (zfs_id_overblockquota(outzfsvfs, DMU_USERUSED_OBJECT, uid) || zfs_id_overblockquota(outzfsvfs, DMU_GROUPUSED_OBJECT, gid) || (projid != ZFS_DEFAULT_PROJID && zfs_id_overblockquota(outzfsvfs, DMU_PROJECTUSED_OBJECT, projid))) { error = SET_ERROR(EDQUOT); break; } nbps = maxblocks; error = dmu_read_l0_bps(inos, inzp->z_id, inoff, size, bps, &nbps); if (error != 0) { /* * If we are tyring to clone a block that was created * in the current transaction group. Return an error, * so the caller can fallback to just copying the data. */ - if (error == EAGAIN) { - error = SET_ERROR(EXDEV); - } break; } /* * Encrypted data is fine as long as it comes from the same * dataset. * TODO: We want to extend it in the future to allow cloning to * datasets with the same keys, like clones or to be able to * clone a file from a snapshot of an encrypted dataset into the * dataset itself. */ if (BP_IS_PROTECTED(&bps[0])) { if (inzfsvfs != outzfsvfs) { error = SET_ERROR(EXDEV); break; } } /* * Start a transaction. */ tx = dmu_tx_create(outos); dmu_tx_hold_sa(tx, outzp->z_sa_hdl, B_FALSE); db = (dmu_buf_impl_t *)sa_get_db(outzp->z_sa_hdl); DB_DNODE_ENTER(db); dmu_tx_hold_clone_by_dnode(tx, DB_DNODE(db), outoff, size); DB_DNODE_EXIT(db); zfs_sa_upgrade_txholds(tx, outzp); error = dmu_tx_assign(tx, TXG_WAIT); if (error != 0) { dmu_tx_abort(tx); break; } /* * Copy source znode's block size. This only happens on the * first iteration since zfs_rangelock_reduce() will shrink down * lr_len to the appropriate size. */ if (outlr->lr_length == UINT64_MAX) { zfs_grow_blocksize(outzp, inblksz, tx); /* * Round range lock up to the block boundary, so we * prevent appends until we are done. */ zfs_rangelock_reduce(outlr, outoff, ((len - 1) / inblksz + 1) * inblksz); } error = dmu_brt_clone(outos, outzp->z_id, outoff, size, tx, bps, nbps, B_FALSE); if (error != 0) { dmu_tx_commit(tx); break; } zfs_clear_setid_bits_if_necessary(outzfsvfs, outzp, cr, &clear_setid_bits_txg, tx); zfs_tstamp_update_setup(outzp, CONTENT_MODIFIED, mtime, ctime); /* * Update the file size (zp_size) if it has changed; * account for possible concurrent updates. */ while ((outsize = outzp->z_size) < outoff + size) { (void) atomic_cas_64(&outzp->z_size, outsize, outoff + size); } error = sa_bulk_update(outzp->z_sa_hdl, bulk, count, tx); zfs_log_clone_range(zilog, tx, TX_CLONE_RANGE, outzp, outoff, size, inblksz, bps, nbps); dmu_tx_commit(tx); if (error != 0) break; inoff += size; outoff += size; len -= size; done += size; } vmem_free(bps, sizeof (bps[0]) * maxblocks); zfs_znode_update_vfs(outzp); unlock: zfs_rangelock_exit(outlr); zfs_rangelock_exit(inlr); if (done > 0) { /* * If we have made at least partial progress, reset the error. */ error = 0; ZFS_ACCESSTIME_STAMP(inzfsvfs, inzp); if (outos->os_sync == ZFS_SYNC_ALWAYS) { zil_commit(zilog, outzp->z_id); } *inoffp += done; *outoffp += done; *lenp = done; } zfs_exit_two(inzfsvfs, outzfsvfs, FTAG); return (error); } /* * Usual pattern would be to call zfs_clone_range() from zfs_replay_clone(), * but we cannot do that, because when replaying we don't have source znode * available. This is why we need a dedicated replay function. */ int zfs_clone_range_replay(znode_t *zp, uint64_t off, uint64_t len, uint64_t blksz, const blkptr_t *bps, size_t nbps) { zfsvfs_t *zfsvfs; dmu_buf_impl_t *db; dmu_tx_t *tx; int error; int count = 0; sa_bulk_attr_t bulk[3]; uint64_t mtime[2], ctime[2]; ASSERT3U(off, <, MAXOFFSET_T); ASSERT3U(len, >, 0); ASSERT3U(nbps, >, 0); zfsvfs = ZTOZSB(zp); ASSERT(spa_feature_is_enabled(dmu_objset_spa(zfsvfs->z_os), SPA_FEATURE_BLOCK_CLONING)); if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) return (error); ASSERT(zfsvfs->z_replay); ASSERT(!zfs_is_readonly(zfsvfs)); if ((off % blksz) != 0) { zfs_exit(zfsvfs, FTAG); return (SET_ERROR(EINVAL)); } 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_SIZE(zfsvfs), NULL, &zp->z_size, 8); /* * Start a transaction. */ tx = dmu_tx_create(zfsvfs->z_os); dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); db = (dmu_buf_impl_t *)sa_get_db(zp->z_sa_hdl); DB_DNODE_ENTER(db); dmu_tx_hold_clone_by_dnode(tx, DB_DNODE(db), off, len); DB_DNODE_EXIT(db); zfs_sa_upgrade_txholds(tx, zp); error = dmu_tx_assign(tx, TXG_WAIT); if (error != 0) { dmu_tx_abort(tx); zfs_exit(zfsvfs, FTAG); return (error); } if (zp->z_blksz < blksz) zfs_grow_blocksize(zp, blksz, tx); dmu_brt_clone(zfsvfs->z_os, zp->z_id, off, len, tx, bps, nbps, B_TRUE); zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime); if (zp->z_size < off + len) zp->z_size = off + len; error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); /* * zil_replaying() not only check if we are replaying ZIL, but also * updates the ZIL header to record replay progress. */ VERIFY(zil_replaying(zfsvfs->z_log, tx)); dmu_tx_commit(tx); zfs_znode_update_vfs(zp); zfs_exit(zfsvfs, FTAG); return (error); } EXPORT_SYMBOL(zfs_access); EXPORT_SYMBOL(zfs_fsync); EXPORT_SYMBOL(zfs_holey); EXPORT_SYMBOL(zfs_read); EXPORT_SYMBOL(zfs_write); EXPORT_SYMBOL(zfs_getsecattr); EXPORT_SYMBOL(zfs_setsecattr); EXPORT_SYMBOL(zfs_clone_range); EXPORT_SYMBOL(zfs_clone_range_replay); ZFS_MODULE_PARAM(zfs_vnops, zfs_vnops_, read_chunk_size, U64, ZMOD_RW, "Bytes to read per chunk");