Index: head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_vnops.c =================================================================== --- head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_vnops.c (revision 351240) +++ head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_vnops.c (revision 351241) @@ -1,6069 +1,6070 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2012, 2015 by Delphix. All rights reserved. * Copyright (c) 2014 Integros [integros.com] * Copyright 2017 Nexenta Systems, Inc. */ /* Portions Copyright 2007 Jeremy Teo */ /* Portions Copyright 2010 Robert Milkowski */ #include #include #include #include #include #include #include #include #include #include #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 /* * Programming rules. * * Each vnode op performs some logical unit of work. To do this, the ZPL must * properly lock its in-core state, create a DMU transaction, do the work, * record this work in the intent log (ZIL), commit the DMU transaction, * and wait for the intent log to commit if it is a synchronous operation. * Moreover, the vnode ops must work in both normal and log replay context. * The ordering of events is important to avoid deadlocks and references * to freed memory. The example below illustrates the following Big Rules: * * (1) A check must be made in each zfs thread for a mounted file system. * This is done avoiding races using ZFS_ENTER(zfsvfs). * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros * can return EIO from the calling function. * * (2) VN_RELE() should always be the last thing except for zil_commit() * (if necessary) and ZFS_EXIT(). This is for 3 reasons: * First, if it's the last reference, the vnode/znode * can be freed, so the zp may point to freed memory. Second, the last * reference will call zfs_zinactive(), which may induce a lot of work -- * pushing cached pages (which acquires range locks) and syncing out * cached atime changes. Third, zfs_zinactive() may require a new tx, * which could deadlock the system if you were already holding one. * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC(). * * (3) All range locks must be grabbed before calling dmu_tx_assign(), * as they can span dmu_tx_assign() calls. * * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to * dmu_tx_assign(). This is critical because we don't want to block * while holding locks. * * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This * reduces lock contention and CPU usage when we must wait (note that if * throughput is constrained by the storage, nearly every transaction * must wait). * * Note, in particular, that if a lock is sometimes acquired before * the tx assigns, and sometimes after (e.g. z_lock), then failing * to use a non-blocking assign can deadlock the system. The scenario: * * Thread A has grabbed a lock before calling dmu_tx_assign(). * Thread B is in an already-assigned tx, and blocks for this lock. * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open() * forever, because the previous txg can't quiesce until B's tx commits. * * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT, * then drop all locks, call dmu_tx_wait(), and try again. On subsequent * calls to dmu_tx_assign(), pass TXG_NOTHROTTLE in addition to TXG_NOWAIT, * to indicate that this operation has already called dmu_tx_wait(). * This will ensure that we don't retry forever, waiting a short bit * each time. * * (5) If the operation succeeded, generate the intent log entry for it * before dropping locks. This ensures that the ordering of events * in the intent log matches the order in which they actually occurred. * During ZIL replay the zfs_log_* functions will update the sequence * number to indicate the zil transaction has replayed. * * (6) At the end of each vnode op, the DMU tx must always commit, * regardless of whether there were any errors. * * (7) After dropping all locks, invoke zil_commit(zilog, foid) * to ensure that synchronous semantics are provided when necessary. * * In general, this is how things should be ordered in each vnode op: * * ZFS_ENTER(zfsvfs); // exit if unmounted * top: * zfs_dirent_lookup(&dl, ...) // lock directory entry (may VN_HOLD()) * rw_enter(...); // grab any other locks you need * tx = dmu_tx_create(...); // get DMU tx * dmu_tx_hold_*(); // hold each object you might modify * error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT); * if (error) { * rw_exit(...); // drop locks * zfs_dirent_unlock(dl); // unlock directory entry * VN_RELE(...); // release held vnodes * if (error == ERESTART) { * waited = B_TRUE; * dmu_tx_wait(tx); * dmu_tx_abort(tx); * goto top; * } * dmu_tx_abort(tx); // abort DMU tx * ZFS_EXIT(zfsvfs); // finished in zfs * return (error); // really out of space * } * error = do_real_work(); // do whatever this VOP does * if (error == 0) * zfs_log_*(...); // on success, make ZIL entry * dmu_tx_commit(tx); // commit DMU tx -- error or not * rw_exit(...); // drop locks * zfs_dirent_unlock(dl); // unlock directory entry * VN_RELE(...); // release held vnodes * zil_commit(zilog, foid); // synchronous when necessary * ZFS_EXIT(zfsvfs); // finished in zfs * return (error); // done, report error */ /* ARGSUSED */ static int zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct) { znode_t *zp = VTOZ(*vpp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) && ((flag & FAPPEND) == 0)) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EPERM)); } if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && ZTOV(zp)->v_type == VREG && !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) { if (fs_vscan(*vpp, cr, 0) != 0) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EACCES)); } } /* Keep a count of the synchronous opens in the znode */ if (flag & (FSYNC | FDSYNC)) atomic_inc_32(&zp->z_sync_cnt); ZFS_EXIT(zfsvfs); return (0); } /* ARGSUSED */ static int zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr, caller_context_t *ct) { znode_t *zp = VTOZ(vp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; /* * Clean up any locks held by this process on the vp. */ cleanlocks(vp, ddi_get_pid(), 0); cleanshares(vp, ddi_get_pid()); ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); /* Decrement the synchronous opens in the znode */ if ((flag & (FSYNC | FDSYNC)) && (count == 1)) atomic_dec_32(&zp->z_sync_cnt); if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && ZTOV(zp)->v_type == VREG && !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) VERIFY(fs_vscan(vp, cr, 1) == 0); ZFS_EXIT(zfsvfs); return (0); } /* * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter. */ static int zfs_holey(vnode_t *vp, u_long cmd, offset_t *off) { znode_t *zp = VTOZ(vp); 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 == _FIO_SEEK_HOLE) hole = B_TRUE; else hole = B_FALSE; error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff); if (error == ESRCH) return (SET_ERROR(ENXIO)); /* * 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); } /* ARGSUSED */ static int zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred, int *rvalp, caller_context_t *ct) { offset_t off; offset_t ndata; dmu_object_info_t doi; int error; zfsvfs_t *zfsvfs; znode_t *zp; 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 _FIO_SEEK_DATA: case _FIO_SEEK_HOLE: { #ifdef illumos if (ddi_copyin((void *)data, &off, sizeof (off), flag)) return (SET_ERROR(EFAULT)); #else off = *(offset_t *)data; #endif zp = VTOZ(vp); zfsvfs = zp->z_zfsvfs; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); /* offset parameter is in/out */ error = zfs_holey(vp, com, &off); ZFS_EXIT(zfsvfs); if (error) return (error); #ifdef illumos if (ddi_copyout(&off, (void *)data, sizeof (off), flag)) return (SET_ERROR(EFAULT)); #else *(offset_t *)data = off; #endif return (0); } #ifdef illumos case _FIO_COUNT_FILLED: { /* * _FIO_COUNT_FILLED adds a new ioctl command which * exposes the number of filled blocks in a * ZFS object. */ zp = VTOZ(vp); zfsvfs = zp->z_zfsvfs; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); /* * Wait for all dirty blocks for this object * to get synced out to disk, and the DMU info * updated. */ error = dmu_object_wait_synced(zfsvfs->z_os, zp->z_id); if (error) { ZFS_EXIT(zfsvfs); return (error); } /* * Retrieve fill count from DMU object. */ error = dmu_object_info(zfsvfs->z_os, zp->z_id, &doi); if (error) { ZFS_EXIT(zfsvfs); return (error); } ndata = doi.doi_fill_count; ZFS_EXIT(zfsvfs); if (ddi_copyout(&ndata, (void *)data, sizeof (ndata), flag)) return (SET_ERROR(EFAULT)); return (0); } #endif } 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 considred 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(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; } 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; } return (pp); } static void page_unbusy(vm_page_t pp) { vm_page_sunbusy(pp); - vm_object_pip_subtract(pp->object, 1); + vm_object_pip_wakeup(pp->object); } static vm_page_t page_wire(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_lock(pp); vm_page_wire(pp); vm_page_unlock(pp); } else pp = NULL; break; } return (pp); } static void page_unwire(vm_page_t pp) { vm_page_lock(pp); vm_page_unwire(pp, PQ_ACTIVE); vm_page_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. */ static void update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid, int segflg, dmu_tx_t *tx) { vm_object_t obj; struct sf_buf *sf; caddr_t va; int off; ASSERT(segflg != UIO_NOCOPY); ASSERT(vp->v_mount != NULL); obj = vp->v_object; ASSERT(obj != NULL); off = start & PAGEOFFSET; zfs_vmobject_wlock(obj); + vm_object_pip_add(obj, 1); 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(obj); va = zfs_map_page(pp, &sf); (void) dmu_read(os, oid, start+off, nbytes, va+off, DMU_READ_PREFETCH);; zfs_unmap_page(sf); zfs_vmobject_wlock(obj); page_unbusy(pp); } len -= nbytes; off = 0; } - vm_object_pip_wakeupn(obj, 0); + vm_object_pip_wakeup(obj); zfs_vmobject_wunlock(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. */ static int mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio) { znode_t *zp = VTOZ(vp); 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 off; int error = 0; ASSERT(uio->uio_segflg == UIO_NOCOPY); ASSERT(vp->v_mount != NULL); obj = vp->v_object; ASSERT(obj != NULL); ASSERT((uio->uio_loffset & PAGEOFFSET) == 0); zfs_vmobject_wlock(obj); for (start = uio->uio_loffset; len > 0; start += PAGESIZE) { int bytes = MIN(PAGESIZE, len); pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_SBUSY | VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY); if (pp->valid == 0) { zfs_vmobject_wunlock(obj); va = zfs_map_page(pp, &sf); error = dmu_read(os, zp->z_id, start, bytes, va, DMU_READ_PREFETCH); if (bytes != PAGESIZE && error == 0) bzero(va + bytes, PAGESIZE - bytes); zfs_unmap_page(sf); zfs_vmobject_wlock(obj); vm_page_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); } else { ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); vm_page_sunbusy(pp); } if (error) break; uio->uio_resid -= bytes; uio->uio_offset += bytes; len -= bytes; } zfs_vmobject_wunlock(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. */ static int mappedread(vnode_t *vp, int nbytes, uio_t *uio) { znode_t *zp = VTOZ(vp); vm_object_t obj; int64_t start; caddr_t va; int len = nbytes; int off; int error = 0; ASSERT(vp->v_mount != NULL); obj = vp->v_object; ASSERT(obj != NULL); start = uio->uio_loffset; off = start & PAGEOFFSET; zfs_vmobject_wlock(obj); for (start &= PAGEMASK; len > 0; start += PAGESIZE) { vm_page_t pp; uint64_t bytes = MIN(PAGESIZE - off, len); if (pp = page_wire(vp, start)) { struct sf_buf *sf; caddr_t va; zfs_vmobject_wunlock(obj); va = zfs_map_page(pp, &sf); #ifdef illumos error = uiomove(va + off, bytes, UIO_READ, uio); #else error = vn_io_fault_uiomove(va + off, bytes, uio); #endif zfs_unmap_page(sf); zfs_vmobject_wlock(obj); page_unwire(pp); } else { zfs_vmobject_wunlock(obj); error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl), uio, bytes); zfs_vmobject_wlock(obj); } len -= bytes; off = 0; if (error) break; } zfs_vmobject_wunlock(obj); return (error); } offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */ /* * Read bytes from specified file into supplied buffer. * * IN: vp - vnode of file to be read from. * uio - structure supplying read location, range info, * and return buffer. * ioflag - SYNC flags; used to provide FRSYNC semantics. * cr - credentials of caller. * ct - caller context * * OUT: uio - updated offset and range, buffer filled. * * RETURN: 0 on success, error code on failure. * * Side Effects: * vp - atime updated if byte count > 0 */ /* ARGSUSED */ static int zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) { znode_t *zp = VTOZ(vp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; ssize_t n, nbytes; int error = 0; rl_t *rl; xuio_t *xuio = NULL; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); if (zp->z_pflags & ZFS_AV_QUARANTINED) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EACCES)); } /* * Validate file offset */ if (uio->uio_loffset < (offset_t)0) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EINVAL)); } /* * Fasttrack empty reads */ if (uio->uio_resid == 0) { ZFS_EXIT(zfsvfs); return (0); } /* * Check for mandatory locks */ if (MANDMODE(zp->z_mode)) { if (error = chklock(vp, FREAD, uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) { ZFS_EXIT(zfsvfs); return (error); } } /* * If we're in FRSYNC mode, sync out this znode before reading it. */ if (zfsvfs->z_log && (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)) zil_commit(zfsvfs->z_log, zp->z_id); /* * Lock the range against changes. */ rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, 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 (uio->uio_loffset >= zp->z_size) { error = 0; goto out; } ASSERT(uio->uio_loffset < zp->z_size); n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset); #ifdef illumos if ((uio->uio_extflg == UIO_XUIO) && (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) { int nblk; int blksz = zp->z_blksz; uint64_t offset = uio->uio_loffset; xuio = (xuio_t *)uio; if ((ISP2(blksz))) { nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset, blksz)) / blksz; } else { ASSERT(offset + n <= blksz); nblk = 1; } (void) dmu_xuio_init(xuio, nblk); if (vn_has_cached_data(vp)) { /* * For simplicity, we always allocate a full buffer * even if we only expect to read a portion of a block. */ while (--nblk >= 0) { (void) dmu_xuio_add(xuio, dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), blksz), 0, blksz); } } } #endif /* illumos */ while (n > 0) { nbytes = MIN(n, zfs_read_chunk_size - P2PHASE(uio->uio_loffset, zfs_read_chunk_size)); #ifdef __FreeBSD__ if (uio->uio_segflg == UIO_NOCOPY) error = mappedread_sf(vp, nbytes, uio); else #endif /* __FreeBSD__ */ if (vn_has_cached_data(vp)) { error = mappedread(vp, 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); break; } n -= nbytes; } out: zfs_range_unlock(rl); ZFS_ACCESSTIME_STAMP(zfsvfs, zp); ZFS_EXIT(zfsvfs); return (error); } /* * Write the bytes to a file. * * IN: vp - vnode of file to be written to. * uio - structure supplying write location, range info, * and data buffer. * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is * set if in append mode. * cr - credentials of caller. * ct - caller context (NFS/CIFS fem monitor only) * * OUT: uio - updated offset and range. * * RETURN: 0 on success, error code on failure. * * Timestamps: * vp - ctime|mtime updated if byte count > 0 */ /* ARGSUSED */ static int zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) { znode_t *zp = VTOZ(vp); rlim64_t limit = MAXOFFSET_T; ssize_t start_resid = uio->uio_resid; ssize_t tx_bytes; uint64_t end_size; dmu_tx_t *tx; zfsvfs_t *zfsvfs = zp->z_zfsvfs; zilog_t *zilog; offset_t woff; ssize_t n, nbytes; rl_t *rl; int max_blksz = zfsvfs->z_max_blksz; int error = 0; arc_buf_t *abuf; iovec_t *aiov = NULL; xuio_t *xuio = NULL; int i_iov = 0; int iovcnt = uio->uio_iovcnt; iovec_t *iovp = uio->uio_iov; int write_eof; int count = 0; sa_bulk_attr_t bulk[4]; uint64_t mtime[2], ctime[2]; /* * Fasttrack empty write */ n = start_resid; if (n == 0) return (0); if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) limit = MAXOFFSET_T; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); 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); /* * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our * callers might not be able to detect properly that we are read-only, * so check it explicitly here. */ if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { ZFS_EXIT(zfsvfs); 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 & FAPPEND) && (uio->uio_loffset < zp->z_size))) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EPERM)); } zilog = zfsvfs->z_log; /* * Validate file offset */ woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset; if (woff < 0) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EINVAL)); } /* * Check for mandatory locks before calling zfs_range_lock() * in order to prevent a deadlock with locks set via fcntl(). */ if (MANDMODE((mode_t)zp->z_mode) && (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) { ZFS_EXIT(zfsvfs); return (error); } #ifdef illumos /* * Pre-fault the pages to ensure slow (eg NFS) pages * don't hold up txg. * Skip this if uio contains loaned arc_buf. */ if ((uio->uio_extflg == UIO_XUIO) && (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) xuio = (xuio_t *)uio; else uio_prefaultpages(MIN(n, max_blksz), uio); #endif /* * If in append mode, set the io offset pointer to eof. */ if (ioflag & FAPPEND) { /* * Obtain an appending range lock to guarantee file append * semantics. We reset the write offset once we have the lock. */ rl = zfs_range_lock(zp, 0, n, RL_APPEND); woff = rl->r_off; if (rl->r_len == 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; } uio->uio_loffset = 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. */ rl = zfs_range_lock(zp, woff, n, RL_WRITER); } if (vn_rlimit_fsize(vp, uio, uio->uio_td)) { zfs_range_unlock(rl); ZFS_EXIT(zfsvfs); return (EFBIG); } if (woff >= limit) { zfs_range_unlock(rl); ZFS_EXIT(zfsvfs); return (SET_ERROR(EFBIG)); } if ((woff + n) > limit || woff > (limit - n)) n = limit - woff; /* Will this write extend the file length? */ write_eof = (woff + n > zp->z_size); end_size = MAX(zp->z_size, woff + n); /* * 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) { abuf = NULL; woff = uio->uio_loffset; if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { if (abuf != NULL) dmu_return_arcbuf(abuf); error = SET_ERROR(EDQUOT); break; } if (xuio && abuf == NULL) { ASSERT(i_iov < iovcnt); aiov = &iovp[i_iov]; abuf = dmu_xuio_arcbuf(xuio, i_iov); dmu_xuio_clear(xuio, i_iov); DTRACE_PROBE3(zfs_cp_write, int, i_iov, iovec_t *, aiov, arc_buf_t *, abuf); ASSERT((aiov->iov_base == abuf->b_data) || ((char *)aiov->iov_base - (char *)abuf->b_data + aiov->iov_len == arc_buf_size(abuf))); i_iov++; } else if (abuf == NULL && n >= max_blksz && woff >= zp->z_size && P2PHASE(woff, max_blksz) == 0 && zp->z_blksz == max_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). */ size_t cbytes; abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), max_blksz); ASSERT(abuf != NULL); ASSERT(arc_buf_size(abuf) == max_blksz); if (error = uiocopy(abuf->b_data, max_blksz, UIO_WRITE, uio, &cbytes)) { dmu_return_arcbuf(abuf); break; } ASSERT(cbytes == max_blksz); } /* * Start a transaction. */ tx = dmu_tx_create(zfsvfs->z_os); dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 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; } /* * If zfs_range_lock() over-locked we grow the blocksize * and then reduce the lock range. This will only happen * on the first iteration since zfs_range_reduce() will * shrink down r_len to the appropriate size. */ if (rl->r_len == UINT64_MAX) { uint64_t new_blksz; if (zp->z_blksz > max_blksz) { /* * File's blocksize is already larger than the * "recordsize" property. Only let it grow to * the next power of 2. */ ASSERT(!ISP2(zp->z_blksz)); new_blksz = MIN(end_size, 1 << highbit64(zp->z_blksz)); } else { new_blksz = MIN(end_size, max_blksz); } zfs_grow_blocksize(zp, new_blksz, tx); zfs_range_reduce(rl, woff, n); } /* * XXX - should we really limit each write to z_max_blksz? * Perhaps we should use SPA_MAXBLOCKSIZE chunks? */ nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz)); if (woff + nbytes > zp->z_size) vnode_pager_setsize(vp, woff + nbytes); if (abuf == NULL) { tx_bytes = uio->uio_resid; error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl), uio, nbytes, tx); tx_bytes -= uio->uio_resid; } else { tx_bytes = nbytes; ASSERT(xuio == NULL || tx_bytes == aiov->iov_len); /* * If this is not a full block write, but we are * extending the file past EOF and this data starts * block-aligned, use assign_arcbuf(). Otherwise, * write via dmu_write(). */ if (tx_bytes < max_blksz && (!write_eof || aiov->iov_base != abuf->b_data)) { ASSERT(xuio); dmu_write(zfsvfs->z_os, zp->z_id, woff, aiov->iov_len, aiov->iov_base, tx); dmu_return_arcbuf(abuf); xuio_stat_wbuf_copied(); } else { ASSERT(xuio || tx_bytes == max_blksz); dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl), woff, abuf, tx); } ASSERT(tx_bytes <= uio->uio_resid); uioskip(uio, tx_bytes); } if (tx_bytes && vn_has_cached_data(vp)) { update_pages(vp, woff, tx_bytes, zfsvfs->z_os, zp->z_id, uio->uio_segflg, tx); } /* * 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; } /* * Clear Set-UID/Set-GID bits on successful write if not * privileged and at least one of the excute bits is set. * * It would be nice to to 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(vp, cr, (zp->z_mode & S_ISUID) != 0 && zp->z_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); zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE); /* * Update the file size (zp_size) if it has changed; * account for possible concurrent updates. */ while ((end_size = zp->z_size) < uio->uio_loffset) { (void) atomic_cas_64(&zp->z_size, end_size, uio->uio_loffset); #ifdef illumos ASSERT(error == 0); #else ASSERT(error == 0 || error == EFAULT); #endif } /* * 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; if (error == 0) error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); else (void) sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag); dmu_tx_commit(tx); if (error != 0) break; ASSERT(tx_bytes == nbytes); n -= nbytes; #ifdef illumos if (!xuio && n > 0) uio_prefaultpages(MIN(n, max_blksz), uio); #endif } zfs_range_unlock(rl); /* * If we're in replay mode, or we made no progress, return error. * Otherwise, it's at least a partial write, so it's successful. */ if (zfsvfs->z_replay || uio->uio_resid == start_resid) { ZFS_EXIT(zfsvfs); return (error); } #ifdef __FreeBSD__ /* * EFAULT means that at least one page of the source buffer was not * available. VFS will re-try remaining I/O upon this error. */ if (error == EFAULT) { ZFS_EXIT(zfsvfs); return (error); } #endif if (ioflag & (FSYNC | FDSYNC) || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) zil_commit(zilog, zp->z_id); ZFS_EXIT(zfsvfs); return (0); } /* ARGSUSED */ void zfs_get_done(zgd_t *zgd, int error) { znode_t *zp = zgd->zgd_private; objset_t *os = zp->z_zfsvfs->z_os; if (zgd->zgd_db) dmu_buf_rele(zgd->zgd_db, zgd); zfs_range_unlock(zgd->zgd_rl); /* * Release the vnode asynchronously as we currently have the * txg stopped from syncing. */ VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os))); kmem_free(zgd, sizeof (zgd_t)); } #ifdef DEBUG static int zil_fault_io = 0; #endif /* * Get data to generate a TX_WRITE intent log record. */ int zfs_get_data(void *arg, 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; ASSERT3P(lwb, !=, NULL); ASSERT3P(zio, !=, 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. */ VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os))); return (SET_ERROR(ENOENT)); } zgd = (zgd_t *)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_rl = zfs_range_lock(zp, 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 */ /* * 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_rl = zfs_range_lock(zp, offset, size, RL_READER); if (zp->z_blksz == size) break; offset += blkoff; zfs_range_unlock(zgd->zgd_rl); } /* test for truncation needs to be done while range locked */ if (lr->lr_offset >= zp->z_size) error = SET_ERROR(ENOENT); #ifdef DEBUG if (zil_fault_io) { error = SET_ERROR(EIO); zil_fault_io = 0; } #endif if (error == 0) error = dmu_buf_hold(os, object, offset, zgd, &db, DMU_READ_NO_PREFETCH); 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); } /*ARGSUSED*/ static int zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr, caller_context_t *ct) { znode_t *zp = VTOZ(vp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; int error; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); if (flag & V_ACE_MASK) error = zfs_zaccess(zp, mode, flag, B_FALSE, cr); else error = zfs_zaccess_rwx(zp, mode, flag, cr); ZFS_EXIT(zfsvfs); return (error); } 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 = zdp->z_zfsvfs; int error; int ltype; ASSERT_VOP_LOCKED(dvp, __func__); #ifdef DIAGNOSTIC if ((zdp->z_pflags & ZFS_XATTR) == 0) VERIFY(!RRM_LOCK_HELD(&zfsvfs->z_teardown_lock)); #endif 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 (dvp->v_iflag & VI_DOOMED) { vrele(dvp); return (SET_ERROR(ENOENT)); } } return (0); } else if (name[0] == '.' && name[1] == '.' && name[2] == 0) { /* * Note that in this case, dvp is the child vnode, and we * are looking up the parent vnode - exactly reverse from * normal operation. Unlocking dvp requires some rather * tricky unlock/relock dance to prevent mp from being freed; * use vn_vget_ino_gen() which takes care of all that. * * XXX Note that there is a time window when both vnodes are * unlocked. It is possible, although highly unlikely, that * during that window the parent-child relationship between * the vnodes may change, for example, get reversed. * In that case we would have a wrong lock order for the vnodes. * All other filesystems seem to ignore this problem, so we * do the same here. * A potential solution could be implemented as follows: * - using LK_NOWAIT when locking the second vnode and retrying * if necessary * - checking that the parent-child relationship still holds * after locking both vnodes and retrying if it doesn't */ error = vn_vget_ino_gen(dvp, zfs_dd_callback, vp, lkflags, &vp); return (error); } else { error = vn_lock(vp, lkflags); if (error != 0) vrele(vp); return (error); } } /* * Lookup an entry in a directory, or an extended attribute directory. * If it exists, return a held vnode reference for it. * * IN: dvp - vnode of directory to search. * nm - name of entry to lookup. * pnp - full pathname to lookup [UNUSED]. * flags - LOOKUP_XATTR set if looking for an attribute. * rdir - root directory vnode [UNUSED]. * cr - credentials of caller. * ct - caller context * * OUT: vpp - vnode of located entry, NULL if not found. * * RETURN: 0 on success, error code on failure. * * Timestamps: * NA */ /* ARGSUSED */ static int zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp, int nameiop, cred_t *cr, kthread_t *td, int flags) { znode_t *zdp = VTOZ(dvp); znode_t *zp; zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 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, char *, nm); ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zdp); *vpp = NULL; if (flags & LOOKUP_XATTR) { #ifdef TODO /* * If the xattr property is off, refuse the lookup request. */ if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EINVAL)); } #endif /* * We don't allow recursive attributes.. * Maybe someday we will. */ if (zdp->z_pflags & ZFS_XATTR) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EINVAL)); } if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) { ZFS_EXIT(zfsvfs); return (error); } /* * Do we have permission to get into attribute directory? */ if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0, B_FALSE, cr)) { vrele(*vpp); *vpp = NULL; } ZFS_EXIT(zfsvfs); return (error); } /* * Check accessibility of directory. */ if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) { ZFS_EXIT(zfsvfs); return (error); } if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EILSEQ)); } /* * First handle the special cases. */ if ((cnp->cn_flags & ISDOTDOT) != 0) { /* * If we are a snapshot mounted under .zfs, return * the vp for the snapshot directory. */ if (zdp->z_id == zfsvfs->z_root && zfsvfs->z_parent != zfsvfs) { struct componentname cn; vnode_t *zfsctl_vp; int ltype; ZFS_EXIT(zfsvfs); ltype = VOP_ISLOCKED(dvp); VOP_UNLOCK(dvp, 0); error = zfsctl_root(zfsvfs->z_parent, LK_SHARED, &zfsctl_vp); if (error == 0) { cn.cn_nameptr = "snapshot"; cn.cn_namelen = strlen(cn.cn_nameptr); cn.cn_nameiop = cnp->cn_nameiop; cn.cn_flags = cnp->cn_flags & ~ISDOTDOT; cn.cn_lkflags = cnp->cn_lkflags; error = VOP_LOOKUP(zfsctl_vp, vpp, &cn); vput(zfsctl_vp); } vn_lock(dvp, ltype | LK_RETRY); return (error); } } if (zfs_has_ctldir(zdp) && strcmp(nm, ZFS_CTLDIR_NAME) == 0) { ZFS_EXIT(zfsvfs); if ((cnp->cn_flags & ISLASTCN) != 0 && nameiop != LOOKUP) return (SET_ERROR(ENOTSUP)); error = zfsctl_root(zfsvfs, cnp->cn_lkflags, vpp); return (error); } /* * The loop is retry the lookup if the parent-child relationship * changes during the dot-dot locking complexities. */ for (;;) { uint64_t parent; error = zfs_dirlook(zdp, nm, &zp); if (error == 0) *vpp = ZTOV(zp); ZFS_EXIT(zfsvfs); if (error != 0) break; error = zfs_lookup_lock(dvp, *vpp, nm, cnp->cn_lkflags); if (error != 0) { /* * If we've got a locking error, then the vnode * got reclaimed because of a force unmount. * We never enter doomed vnodes into the name cache. */ *vpp = NULL; return (error); } if ((cnp->cn_flags & ISDOTDOT) == 0) break; ZFS_ENTER(zfsvfs); if (zdp->z_sa_hdl == NULL) { error = SET_ERROR(EIO); } else { error = sa_lookup(zdp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent)); } if (error != 0) { ZFS_EXIT(zfsvfs); vput(ZTOV(zp)); break; } if (zp->z_id == parent) { ZFS_EXIT(zfsvfs); break; } vput(ZTOV(zp)); } out: if (error != 0) *vpp = NULL; /* Translate errors and add SAVENAME when needed. */ if (cnp->cn_flags & ISLASTCN) { switch (nameiop) { case CREATE: case RENAME: if (error == ENOENT) { error = EJUSTRETURN; cnp->cn_flags |= SAVENAME; break; } /* FALLTHROUGH */ case DELETE: if (error == 0) cnp->cn_flags |= SAVENAME; break; } } /* Insert name into cache (as non-existent) if appropriate. */ if (zfsvfs->z_use_namecache && error == ENOENT && (cnp->cn_flags & MAKEENTRY) != 0) cache_enter(dvp, NULL, cnp); /* Insert name into cache if appropriate. */ if (zfsvfs->z_use_namecache && error == 0 && (cnp->cn_flags & MAKEENTRY)) { if (!(cnp->cn_flags & ISLASTCN) || (nameiop != DELETE && nameiop != RENAME)) { cache_enter(dvp, *vpp, cnp); } } return (error); } /* * Attempt to create a new entry in a directory. If the entry * already exists, truncate the file if permissible, else return * an error. Return the vp of the created or trunc'd file. * * IN: dvp - vnode of directory to put new file entry in. * name - name of new file entry. * vap - attributes of new file. * excl - flag indicating exclusive or non-exclusive mode. * mode - mode to open file with. * cr - credentials of caller. * flag - large file flag [UNUSED]. * ct - caller context * vsecp - ACL to be set * * OUT: vpp - vnode of created or trunc'd entry. * * RETURN: 0 on success, error code on failure. * * Timestamps: * dvp - ctime|mtime updated if new entry created * vp - ctime|mtime always, atime if new */ /* ARGSUSED */ static int zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode, vnode_t **vpp, cred_t *cr, kthread_t *td) { znode_t *zp, *dzp = VTOZ(dvp); zfsvfs_t *zfsvfs = dzp->z_zfsvfs; zilog_t *zilog; objset_t *os; dmu_tx_t *tx; int error; ksid_t *ksid; uid_t uid; gid_t gid = crgetgid(cr); zfs_acl_ids_t acl_ids; boolean_t fuid_dirtied; void *vsecp = NULL; int flag = 0; uint64_t txtype; /* * If we have an ephemeral id, ACL, or XVATTR then * make sure file system is at proper version */ ksid = crgetsid(cr, KSID_OWNER); if (ksid) uid = ksid_getid(ksid); else uid = crgetuid(cr); if (zfsvfs->z_use_fuids == B_FALSE && (vsecp || (vap->va_mask & AT_XVATTR) || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) return (SET_ERROR(EINVAL)); ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(dzp); os = zfsvfs->z_os; zilog = zfsvfs->z_log; if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EILSEQ)); } if (vap->va_mask & AT_XVATTR) { if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap, crgetuid(cr), cr, vap->va_type)) != 0) { ZFS_EXIT(zfsvfs); return (error); } } *vpp = NULL; if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr)) vap->va_mode &= ~S_ISVTX; error = zfs_dirent_lookup(dzp, name, &zp, ZNEW); if (error) { ZFS_EXIT(zfsvfs); return (error); } ASSERT3P(zp, ==, NULL); /* * Create a new file object and update the directory * to reference it. */ if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { goto out; } /* * We only support the creation of regular files in * extended attribute directories. */ if ((dzp->z_pflags & ZFS_XATTR) && (vap->va_type != VREG)) { error = SET_ERROR(EINVAL); goto out; } if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, vsecp, &acl_ids)) != 0) goto out; if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { zfs_acl_ids_free(&acl_ids); error = SET_ERROR(EDQUOT); goto out; } getnewvnode_reserve(1); tx = dmu_tx_create(os); dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + ZFS_SA_BASE_ATTR_SIZE); fuid_dirtied = zfsvfs->z_fuid_dirty; if (fuid_dirtied) zfs_fuid_txhold(zfsvfs, tx); dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, acl_ids.z_aclp->z_acl_bytes); } error = dmu_tx_assign(tx, TXG_WAIT); if (error) { zfs_acl_ids_free(&acl_ids); dmu_tx_abort(tx); getnewvnode_drop_reserve(); ZFS_EXIT(zfsvfs); return (error); } zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); if (fuid_dirtied) zfs_fuid_sync(zfsvfs, tx); (void) zfs_link_create(dzp, name, zp, tx, ZNEW); txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap); zfs_log_create(zilog, tx, txtype, dzp, zp, name, vsecp, acl_ids.z_fuidp, vap); zfs_acl_ids_free(&acl_ids); dmu_tx_commit(tx); getnewvnode_drop_reserve(); out: if (error == 0) { *vpp = ZTOV(zp); } if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) zil_commit(zilog, 0); ZFS_EXIT(zfsvfs); return (error); } /* * Remove an entry from a directory. * * IN: dvp - vnode of directory to remove entry from. * name - name of entry to remove. * cr - credentials of caller. * ct - caller context * flags - case flags * * RETURN: 0 on success, error code on failure. * * Timestamps: * dvp - ctime|mtime * vp - ctime (if nlink > 0) */ /*ARGSUSED*/ static int zfs_remove(vnode_t *dvp, vnode_t *vp, char *name, cred_t *cr) { znode_t *dzp = VTOZ(dvp); znode_t *zp = VTOZ(vp); znode_t *xzp; zfsvfs_t *zfsvfs = dzp->z_zfsvfs; zilog_t *zilog; uint64_t acl_obj, xattr_obj; uint64_t obj = 0; dmu_tx_t *tx; boolean_t unlinked, toobig = FALSE; uint64_t txtype; int error; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(dzp); ZFS_VERIFY_ZP(zp); zilog = zfsvfs->z_log; zp = VTOZ(vp); xattr_obj = 0; xzp = NULL; if (error = zfs_zaccess_delete(dzp, zp, cr)) { goto out; } /* * Need to use rmdir for removing directories. */ if (vp->v_type == VDIR) { error = SET_ERROR(EPERM); goto out; } vnevent_remove(vp, dvp, name, ct); obj = zp->z_id; /* are there any extended attributes? */ error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &xattr_obj, sizeof (xattr_obj)); if (error == 0 && xattr_obj) { error = zfs_zget(zfsvfs, xattr_obj, &xzp); ASSERT0(error); } /* * We may delete the znode now, or we may put it in the unlinked set; * it depends on whether we're the last link, and on whether there are * other holds on the vnode. So we dmu_tx_hold() the right things to * allow for either case. */ tx = dmu_tx_create(zfsvfs->z_os); dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); zfs_sa_upgrade_txholds(tx, zp); zfs_sa_upgrade_txholds(tx, dzp); if (xzp) { dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE); } /* charge as an update -- would be nice not to charge at all */ dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); /* * Mark this transaction as typically resulting in a net free of space */ dmu_tx_mark_netfree(tx); error = dmu_tx_assign(tx, TXG_WAIT); if (error) { dmu_tx_abort(tx); ZFS_EXIT(zfsvfs); return (error); } /* * Remove the directory entry. */ error = zfs_link_destroy(dzp, name, zp, tx, ZEXISTS, &unlinked); if (error) { dmu_tx_commit(tx); goto out; } if (unlinked) { zfs_unlinked_add(zp, tx); vp->v_vflag |= VV_NOSYNC; } txtype = TX_REMOVE; zfs_log_remove(zilog, tx, txtype, dzp, name, obj); dmu_tx_commit(tx); out: if (xzp) vrele(ZTOV(xzp)); if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) zil_commit(zilog, 0); ZFS_EXIT(zfsvfs); return (error); } /* * Create a new directory and insert it into dvp using the name * provided. Return a pointer to the inserted directory. * * IN: dvp - vnode of directory to add subdir to. * dirname - name of new directory. * vap - attributes of new directory. * cr - credentials of caller. * ct - caller context * flags - case flags * vsecp - ACL to be set * * OUT: vpp - vnode of created directory. * * RETURN: 0 on success, error code on failure. * * Timestamps: * dvp - ctime|mtime updated * vp - ctime|mtime|atime updated */ /*ARGSUSED*/ static int zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr) { znode_t *zp, *dzp = VTOZ(dvp); zfsvfs_t *zfsvfs = dzp->z_zfsvfs; zilog_t *zilog; uint64_t txtype; dmu_tx_t *tx; int error; ksid_t *ksid; uid_t uid; gid_t gid = crgetgid(cr); zfs_acl_ids_t acl_ids; boolean_t fuid_dirtied; ASSERT(vap->va_type == VDIR); /* * If we have an ephemeral id, ACL, or XVATTR then * make sure file system is at proper version */ ksid = crgetsid(cr, KSID_OWNER); if (ksid) uid = ksid_getid(ksid); else uid = crgetuid(cr); if (zfsvfs->z_use_fuids == B_FALSE && ((vap->va_mask & AT_XVATTR) || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) return (SET_ERROR(EINVAL)); ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(dzp); zilog = zfsvfs->z_log; if (dzp->z_pflags & ZFS_XATTR) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EINVAL)); } if (zfsvfs->z_utf8 && u8_validate(dirname, strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EILSEQ)); } if (vap->va_mask & AT_XVATTR) { if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap, crgetuid(cr), cr, vap->va_type)) != 0) { ZFS_EXIT(zfsvfs); return (error); } } if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, NULL, &acl_ids)) != 0) { ZFS_EXIT(zfsvfs); return (error); } /* * First make sure the new directory doesn't exist. * * Existence is checked first to make sure we don't return * EACCES instead of EEXIST which can cause some applications * to fail. */ *vpp = NULL; if (error = zfs_dirent_lookup(dzp, dirname, &zp, ZNEW)) { zfs_acl_ids_free(&acl_ids); ZFS_EXIT(zfsvfs); return (error); } ASSERT3P(zp, ==, NULL); if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) { zfs_acl_ids_free(&acl_ids); ZFS_EXIT(zfsvfs); return (error); } if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { zfs_acl_ids_free(&acl_ids); ZFS_EXIT(zfsvfs); return (SET_ERROR(EDQUOT)); } /* * Add a new entry to the directory. */ getnewvnode_reserve(1); tx = dmu_tx_create(zfsvfs->z_os); dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); fuid_dirtied = zfsvfs->z_fuid_dirty; if (fuid_dirtied) zfs_fuid_txhold(zfsvfs, tx); if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, acl_ids.z_aclp->z_acl_bytes); } dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + ZFS_SA_BASE_ATTR_SIZE); error = dmu_tx_assign(tx, TXG_WAIT); if (error) { zfs_acl_ids_free(&acl_ids); dmu_tx_abort(tx); getnewvnode_drop_reserve(); ZFS_EXIT(zfsvfs); return (error); } /* * Create new node. */ zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); if (fuid_dirtied) zfs_fuid_sync(zfsvfs, tx); /* * Now put new name in parent dir. */ (void) zfs_link_create(dzp, dirname, zp, tx, ZNEW); *vpp = ZTOV(zp); txtype = zfs_log_create_txtype(Z_DIR, NULL, vap); zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, NULL, acl_ids.z_fuidp, vap); zfs_acl_ids_free(&acl_ids); dmu_tx_commit(tx); getnewvnode_drop_reserve(); if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) zil_commit(zilog, 0); ZFS_EXIT(zfsvfs); return (0); } /* * Remove a directory subdir entry. If the current working * directory is the same as the subdir to be removed, the * remove will fail. * * IN: dvp - vnode of directory to remove from. * name - name of directory to be removed. * cwd - vnode of current working directory. * cr - credentials of caller. * ct - caller context * flags - case flags * * RETURN: 0 on success, error code on failure. * * Timestamps: * dvp - ctime|mtime updated */ /*ARGSUSED*/ static int zfs_rmdir(vnode_t *dvp, vnode_t *vp, char *name, cred_t *cr) { znode_t *dzp = VTOZ(dvp); znode_t *zp = VTOZ(vp); zfsvfs_t *zfsvfs = dzp->z_zfsvfs; zilog_t *zilog; dmu_tx_t *tx; int error; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(dzp); ZFS_VERIFY_ZP(zp); zilog = zfsvfs->z_log; if (error = zfs_zaccess_delete(dzp, zp, cr)) { goto out; } if (vp->v_type != VDIR) { error = SET_ERROR(ENOTDIR); goto out; } vnevent_rmdir(vp, dvp, name, ct); tx = dmu_tx_create(zfsvfs->z_os); dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); zfs_sa_upgrade_txholds(tx, zp); zfs_sa_upgrade_txholds(tx, dzp); dmu_tx_mark_netfree(tx); error = dmu_tx_assign(tx, TXG_WAIT); if (error) { dmu_tx_abort(tx); ZFS_EXIT(zfsvfs); return (error); } cache_purge(dvp); 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); } dmu_tx_commit(tx); cache_purge(vp); out: if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) zil_commit(zilog, 0); ZFS_EXIT(zfsvfs); return (error); } /* * Read as many directory entries as will fit into the provided * buffer from the given directory cursor position (specified in * the uio structure). * * IN: vp - vnode of directory to read. * uio - structure supplying read location, range info, * and return buffer. * cr - credentials of caller. * ct - caller context * flags - case flags * * OUT: uio - updated offset and range, buffer filled. * eofp - set to true if end-of-file detected. * * RETURN: 0 on success, error code on failure. * * Timestamps: * vp - atime updated * * Note that the low 4 bits of the cookie returned by zap is always zero. * This allows us to use the low range for "special" directory entries: * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, * we use the offset 2 for the '.zfs' directory. */ /* ARGSUSED */ static int zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies) { znode_t *zp = VTOZ(vp); iovec_t *iovp; edirent_t *eodp; dirent64_t *odp; zfsvfs_t *zfsvfs = zp->z_zfsvfs; objset_t *os; caddr_t outbuf; size_t bufsize; zap_cursor_t zc; zap_attribute_t zap; uint_t bytes_wanted; uint64_t offset; /* must be unsigned; checks for < 1 */ uint64_t parent; int local_eof; int outcount; int error; uint8_t prefetch; boolean_t check_sysattrs; uint8_t type; int ncooks; u_long *cooks = NULL; int flags = 0; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0) { ZFS_EXIT(zfsvfs); return (error); } /* * If we are not given an eof variable, * use a local one. */ if (eofp == NULL) eofp = &local_eof; /* * Check for valid iov_len. */ if (uio->uio_iov->iov_len <= 0) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EINVAL)); } /* * Quit if directory has been removed (posix) */ if ((*eofp = zp->z_unlinked) != 0) { ZFS_EXIT(zfsvfs); return (0); } error = 0; os = zfsvfs->z_os; offset = uio->uio_loffset; 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 = uio->uio_iov; bytes_wanted = iovp->iov_len; if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { bufsize = bytes_wanted; outbuf = kmem_alloc(bufsize, KM_SLEEP); odp = (struct dirent64 *)outbuf; } else { bufsize = bytes_wanted; outbuf = NULL; odp = (struct dirent64 *)iovp->iov_base; } eodp = (struct edirent *)odp; if (ncookies != NULL) { /* * Minimum entry size is dirent size and 1 byte for a file name. */ ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1); cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK); *cookies = cooks; *ncookies = ncooks; } /* * If this VFS supports the system attribute view interface; and * we're looking at an extended attribute directory; and we care * about normalization conflicts on this vfs; then we must check * for normalization conflicts with the sysattr name space. */ #ifdef TODO check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm && (flags & V_RDDIR_ENTFLAGS); #else check_sysattrs = 0; #endif /* * Transform to file-system independent format */ outcount = 0; while (outcount < bytes_wanted) { ino64_t objnum; ushort_t reclen; off64_t *next = NULL; /* * Special case `.', `..', and `.zfs'. */ if (offset == 0) { (void) strcpy(zap.za_name, "."); zap.za_normalization_conflict = 0; objnum = zp->z_id; type = DT_DIR; } else if (offset == 1) { (void) strcpy(zap.za_name, ".."); zap.za_normalization_conflict = 0; objnum = parent; type = DT_DIR; } else if (offset == 2 && zfs_show_ctldir(zp)) { (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); zap.za_normalization_conflict = 0; objnum = ZFSCTL_INO_ROOT; type = DT_DIR; } else { /* * Grab next entry. */ if (error = zap_cursor_retrieve(&zc, &zap)) { if ((*eofp = (error == ENOENT)) != 0) break; else goto update; } if (zap.za_integer_length != 8 || zap.za_num_integers != 1) { cmn_err(CE_WARN, "zap_readdir: bad directory " "entry, obj = %lld, offset = %lld\n", (u_longlong_t)zp->z_id, (u_longlong_t)offset); error = SET_ERROR(ENXIO); goto update; } objnum = ZFS_DIRENT_OBJ(zap.za_first_integer); /* * MacOS X can extract the object type here such as: * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer); */ type = ZFS_DIRENT_TYPE(zap.za_first_integer); if (check_sysattrs && !zap.za_normalization_conflict) { #ifdef TODO zap.za_normalization_conflict = xattr_sysattr_casechk(zap.za_name); #else panic("%s:%u: TODO", __func__, __LINE__); #endif } } if (flags & V_RDDIR_ACCFILTER) { /* * If we have no access at all, don't include * this entry in the returned information */ znode_t *ezp; if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0) goto skip_entry; if (!zfs_has_access(ezp, cr)) { vrele(ZTOV(ezp)); goto skip_entry; } vrele(ZTOV(ezp)); } if (flags & V_RDDIR_ENTFLAGS) reclen = EDIRENT_RECLEN(strlen(zap.za_name)); else reclen = DIRENT64_RECLEN(strlen(zap.za_name)); /* * Will this entry fit in the buffer? */ if (outcount + reclen > bufsize) { /* * Did we manage to fit anything in the buffer? */ if (!outcount) { error = SET_ERROR(EINVAL); goto update; } break; } if (flags & V_RDDIR_ENTFLAGS) { /* * Add extended flag entry: */ eodp->ed_ino = objnum; eodp->ed_reclen = reclen; /* NOTE: ed_off is the offset for the *next* entry. */ next = &eodp->ed_off; eodp->ed_eflags = zap.za_normalization_conflict ? ED_CASE_CONFLICT : 0; (void) strncpy(eodp->ed_name, zap.za_name, EDIRENT_NAMELEN(reclen)); eodp = (edirent_t *)((intptr_t)eodp + reclen); } else { /* * Add normal entry: */ odp->d_ino = objnum; odp->d_reclen = reclen; odp->d_namlen = strlen(zap.za_name); /* NOTE: d_off is the offset for the *next* entry. */ next = &odp->d_off; (void) 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; ASSERT(outcount <= bufsize); /* Prefetch znode */ if (prefetch) dmu_prefetch(os, objnum, 0, 0, 0, ZIO_PRIORITY_SYNC_READ); skip_entry: /* * Move to the next entry, fill in the previous offset. */ if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { zap_cursor_advance(&zc); offset = zap_cursor_serialize(&zc); } else { offset += 1; } /* Fill the offset right after advancing the cursor. */ if (next != NULL) *next = offset; if (cooks != NULL) { *cooks++ = offset; ncooks--; KASSERT(ncooks >= 0, ("ncookies=%d", ncooks)); } } zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ /* Subtract unused cookies */ if (ncookies != NULL) *ncookies -= ncooks; if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { iovp->iov_base += outcount; iovp->iov_len -= outcount; uio->uio_resid -= outcount; } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { /* * Reset the pointer. */ offset = uio->uio_loffset; } update: zap_cursor_fini(&zc); if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) kmem_free(outbuf, bufsize); if (error == ENOENT) error = 0; ZFS_ACCESSTIME_STAMP(zfsvfs, zp); uio->uio_loffset = offset; ZFS_EXIT(zfsvfs); if (error != 0 && cookies != NULL) { free(*cookies, M_TEMP); *cookies = NULL; *ncookies = 0; } return (error); } ulong_t zfs_fsync_sync_cnt = 4; static int zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct) { znode_t *zp = VTOZ(vp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt); if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) { ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); zil_commit(zfsvfs->z_log, zp->z_id); ZFS_EXIT(zfsvfs); } return (0); } /* * 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. * ct - caller context * * OUT: vap - attribute values. * * RETURN: 0 (always succeeds). */ /* ARGSUSED */ static int zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, caller_context_t *ct) { znode_t *zp = VTOZ(vp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; int error = 0; uint32_t blksize; u_longlong_t nblocks; uint64_t mtime[2], ctime[2], crtime[2], rdev; xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ xoptattr_t *xoap = NULL; boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; sa_bulk_attr_t bulk[4]; int count = 0; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid); SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16); if (vp->v_type == VBLK || vp->v_type == VCHR) SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL, &rdev, 8); if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) { ZFS_EXIT(zfsvfs); return (error); } /* * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. * Also, if we are the owner don't bother, since owner should * always be allowed to read basic attributes of file. */ if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) && (vap->va_uid != crgetuid(cr))) { if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0, skipaclchk, cr)) { ZFS_EXIT(zfsvfs); return (error); } } /* * Return all attributes. It's cheaper to provide the answer * than to determine whether we were asked the question. */ vap->va_type = IFTOVT(zp->z_mode); vap->va_mode = zp->z_mode & ~S_IFMT; #ifdef illumos vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev; #else vn_fsid(vp, vap); #endif 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; #ifdef illumos vap->va_rdev = vp->v_rdev; #else if (vp->v_type == VBLK || vp->v_type == VCHR) vap->va_rdev = zfs_cmpldev(rdev); #endif vap->va_seq = zp->z_seq; vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */ vap->va_filerev = zp->z_seq; /* * Add in any requested optional attributes and the create time. * Also set the corresponding bits in the returned attribute bitmap. */ if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) { if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { xoap->xoa_archive = ((zp->z_pflags & ZFS_ARCHIVE) != 0); XVA_SET_RTN(xvap, XAT_ARCHIVE); } if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { xoap->xoa_readonly = ((zp->z_pflags & ZFS_READONLY) != 0); XVA_SET_RTN(xvap, XAT_READONLY); } if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { xoap->xoa_system = ((zp->z_pflags & ZFS_SYSTEM) != 0); XVA_SET_RTN(xvap, XAT_SYSTEM); } if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { xoap->xoa_hidden = ((zp->z_pflags & ZFS_HIDDEN) != 0); XVA_SET_RTN(xvap, XAT_HIDDEN); } if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { xoap->xoa_nounlink = ((zp->z_pflags & ZFS_NOUNLINK) != 0); XVA_SET_RTN(xvap, XAT_NOUNLINK); } if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { xoap->xoa_immutable = ((zp->z_pflags & ZFS_IMMUTABLE) != 0); XVA_SET_RTN(xvap, XAT_IMMUTABLE); } if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { xoap->xoa_appendonly = ((zp->z_pflags & ZFS_APPENDONLY) != 0); XVA_SET_RTN(xvap, XAT_APPENDONLY); } if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { xoap->xoa_nodump = ((zp->z_pflags & ZFS_NODUMP) != 0); XVA_SET_RTN(xvap, XAT_NODUMP); } if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { xoap->xoa_opaque = ((zp->z_pflags & ZFS_OPAQUE) != 0); XVA_SET_RTN(xvap, XAT_OPAQUE); } if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { xoap->xoa_av_quarantined = ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0); XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); } if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { xoap->xoa_av_modified = ((zp->z_pflags & ZFS_AV_MODIFIED) != 0); XVA_SET_RTN(xvap, XAT_AV_MODIFIED); } if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) && vp->v_type == VREG) { zfs_sa_get_scanstamp(zp, xvap); } if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0); XVA_SET_RTN(xvap, XAT_REPARSE); } if (XVA_ISSET_REQ(xvap, XAT_GEN)) { xoap->xoa_generation = zp->z_gen; XVA_SET_RTN(xvap, XAT_GEN); } if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { xoap->xoa_offline = ((zp->z_pflags & ZFS_OFFLINE) != 0); XVA_SET_RTN(xvap, XAT_OFFLINE); } if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { xoap->xoa_sparse = ((zp->z_pflags & ZFS_SPARSE) != 0); XVA_SET_RTN(xvap, XAT_SPARSE); } } ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime); ZFS_TIME_DECODE(&vap->va_mtime, mtime); ZFS_TIME_DECODE(&vap->va_ctime, ctime); ZFS_TIME_DECODE(&vap->va_birthtime, crtime); sa_object_size(zp->z_sa_hdl, &blksize, &nblocks); vap->va_blksize = blksize; vap->va_bytes = nblocks << 9; /* nblocks * 512 */ if (zp->z_blksz == 0) { /* * Block size hasn't been set; suggest maximal I/O transfers. */ vap->va_blksize = zfsvfs->z_max_blksz; } ZFS_EXIT(zfsvfs); return (0); } /* * Set the file attributes to the values contained in the * vattr structure. * * IN: vp - vnode of file to be modified. * vap - new attribute values. * If AT_XVATTR set, then optional attrs are being set * flags - ATTR_UTIME set if non-default time values provided. * - ATTR_NOACLCHECK (CIFS context only). * cr - credentials of caller. * ct - caller context * * RETURN: 0 on success, error code on failure. * * Timestamps: * vp - ctime updated, mtime updated if size changed. */ /* ARGSUSED */ static int zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, caller_context_t *ct) { znode_t *zp = VTOZ(vp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; 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]; znode_t *attrzp; int need_policy = FALSE; int err, err2; zfs_fuid_info_t *fuidp = NULL; xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ xoptattr_t *xoap; zfs_acl_t *aclp; boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; boolean_t fuid_dirtied = B_FALSE; sa_bulk_attr_t bulk[7], xattr_bulk[7]; int count = 0, xattr_count = 0; if (mask == 0) return (0); if (mask & AT_NOSET) return (SET_ERROR(EINVAL)); ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); zilog = zfsvfs->z_log; /* * Make sure that if we have ephemeral uid/gid or xvattr specified * that file system is at proper version level */ if (zfsvfs->z_use_fuids == B_FALSE && (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) || ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) || (mask & AT_XVATTR))) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EINVAL)); } if (mask & AT_SIZE && vp->v_type == VDIR) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EISDIR)); } if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EINVAL)); } /* * If this is an xvattr_t, then get a pointer to the structure of * optional attributes. If this is NULL, then we have a vattr_t. */ xoap = xva_getxoptattr(xvap); xva_init(&tmpxvattr); /* * Immutable files can only alter immutable bit and atime */ if ((zp->z_pflags & ZFS_IMMUTABLE) && ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) || ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EPERM)); } /* * Note: ZFS_READONLY is handled in zfs_zaccess_common. */ /* * Verify timestamps doesn't overflow 32 bits. * ZFS can handle large timestamps, but 32bit syscalls can't * handle times greater than 2039. This check should be removed * once large timestamps are fully supported. */ if (mask & (AT_ATIME | AT_MTIME)) { if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) || ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EOVERFLOW)); } } if (xoap && (mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME) && TIMESPEC_OVERFLOW(&vap->va_birthtime)) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EOVERFLOW)); } attrzp = NULL; aclp = NULL; /* Can this be moved to before the top label? */ if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EROFS)); } /* * First validate permissions */ if (mask & AT_SIZE) { /* * XXX - Note, we are not providing any open * mode flags here (like FNDELAY), so we may * block if there are locks present... this * should be addressed in openat(). */ /* XXX - would it be OK to generate a log record here? */ err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); if (err) { ZFS_EXIT(zfsvfs); return (err); } } if (mask & (AT_ATIME|AT_MTIME) || ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) || XVA_ISSET_REQ(xvap, XAT_READONLY) || XVA_ISSET_REQ(xvap, XAT_ARCHIVE) || XVA_ISSET_REQ(xvap, XAT_OFFLINE) || XVA_ISSET_REQ(xvap, XAT_SPARSE) || XVA_ISSET_REQ(xvap, XAT_CREATETIME) || XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) { need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0, skipaclchk, cr); } if (mask & (AT_UID|AT_GID)) { int idmask = (mask & (AT_UID|AT_GID)); int take_owner; int take_group; /* * NOTE: even if a new mode is being set, * we may clear S_ISUID/S_ISGID bits. */ if (!(mask & AT_MODE)) vap->va_mode = zp->z_mode; /* * Take ownership or chgrp to group we are a member of */ take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); take_group = (mask & AT_GID) && zfs_groupmember(zfsvfs, vap->va_gid, cr); /* * If both AT_UID and AT_GID are set then take_owner and * take_group must both be set in order to allow taking * ownership. * * Otherwise, send the check through secpolicy_vnode_setattr() * */ if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || ((idmask == AT_UID) && take_owner) || ((idmask == AT_GID) && take_group)) { if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0, skipaclchk, cr) == 0) { /* * Remove setuid/setgid for non-privileged users */ secpolicy_setid_clear(vap, vp, cr); trim_mask = (mask & (AT_UID|AT_GID)); } else { need_policy = TRUE; } } else { need_policy = TRUE; } } oldva.va_mode = zp->z_mode; zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid); if (mask & AT_XVATTR) { /* * Update xvattr mask to include only those attributes * that are actually changing. * * the bits will be restored prior to actually setting * the attributes so the caller thinks they were set. */ if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { if (xoap->xoa_appendonly != ((zp->z_pflags & ZFS_APPENDONLY) != 0)) { need_policy = TRUE; } else { XVA_CLR_REQ(xvap, XAT_APPENDONLY); XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY); } } if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { if (xoap->xoa_nounlink != ((zp->z_pflags & ZFS_NOUNLINK) != 0)) { need_policy = TRUE; } else { XVA_CLR_REQ(xvap, XAT_NOUNLINK); XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK); } } if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { if (xoap->xoa_immutable != ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) { need_policy = TRUE; } else { XVA_CLR_REQ(xvap, XAT_IMMUTABLE); XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE); } } if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { if (xoap->xoa_nodump != ((zp->z_pflags & ZFS_NODUMP) != 0)) { need_policy = TRUE; } else { XVA_CLR_REQ(xvap, XAT_NODUMP); XVA_SET_REQ(&tmpxvattr, XAT_NODUMP); } } if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { if (xoap->xoa_av_modified != ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) { need_policy = TRUE; } else { XVA_CLR_REQ(xvap, XAT_AV_MODIFIED); XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED); } } if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { if ((vp->v_type != VREG && xoap->xoa_av_quarantined) || xoap->xoa_av_quarantined != ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) { need_policy = TRUE; } else { XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED); XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED); } } if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EPERM)); } if (need_policy == FALSE && (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) || XVA_ISSET_REQ(xvap, XAT_OPAQUE))) { need_policy = TRUE; } } if (mask & AT_MODE) { if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) { err = secpolicy_setid_setsticky_clear(vp, vap, &oldva, cr); if (err) { ZFS_EXIT(zfsvfs); return (err); } trim_mask |= AT_MODE; } else { need_policy = TRUE; } } if (need_policy) { /* * If trim_mask is set then take ownership * has been granted or write_acl is present and user * has the ability to modify mode. In that case remove * UID|GID and or MODE from mask so that * secpolicy_vnode_setattr() doesn't revoke it. */ if (trim_mask) { saved_mask = vap->va_mask; vap->va_mask &= ~trim_mask; if (trim_mask & AT_MODE) { /* * Save the mode, as secpolicy_vnode_setattr() * will overwrite it with ova.va_mode. */ saved_mode = vap->va_mode; } } err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp); if (err) { ZFS_EXIT(zfsvfs); return (err); } if (trim_mask) { vap->va_mask |= saved_mask; if (trim_mask & AT_MODE) { /* * Recover the mode after * secpolicy_vnode_setattr(). */ vap->va_mode = saved_mode; } } } /* * secpolicy_vnode_setattr, or take ownership may have * changed va_mask */ mask = vap->va_mask; if ((mask & (AT_UID | AT_GID))) { 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_fuid_overquota(zfsvfs, B_FALSE, 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_fuid_overquota(zfsvfs, B_TRUE, new_gid)) { if (attrzp) vput(ZTOV(attrzp)); err = SET_ERROR(EDQUOT); goto out2; } } } tx = dmu_tx_create(zfsvfs->z_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)) 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 (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 (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); ASSERT(err == 0); if (attrzp) { err = zfs_acl_chown_setattr(attrzp); ASSERT(err == 0); } } 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; ASSERT3U((uintptr_t)aclp, !=, 0); 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)); } /* 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, B_TRUE); } 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, B_TRUE); 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, B_TRUE); } } /* * 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(xvap, XAT_AV_SCANSTAMP)) ASSERT(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); ASSERT(err2 == 0); } 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 (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) zil_commit(zilog, 0); ZFS_EXIT(zfsvfs); return (err); } /* * We acquire all but fdvp locks using non-blocking acquisitions. If we * fail to acquire any lock in the path we will drop all held locks, * acquire the new lock in a blocking fashion, and then release it and * restart the rename. This acquire/release step ensures that we do not * spin on a lock waiting for release. On error release all vnode locks * and decrement references the way tmpfs_rename() would do. */ static int zfs_rename_relock(struct vnode *sdvp, struct vnode **svpp, struct vnode *tdvp, struct vnode **tvpp, const struct componentname *scnp, const struct componentname *tcnp) { zfsvfs_t *zfsvfs; struct vnode *nvp, *svp, *tvp; znode_t *sdzp, *tdzp, *szp, *tzp; const char *snm = scnp->cn_nameptr; const char *tnm = tcnp->cn_nameptr; int error; VOP_UNLOCK(tdvp, 0); if (*tvpp != NULL && *tvpp != tdvp) VOP_UNLOCK(*tvpp, 0); relock: error = vn_lock(sdvp, LK_EXCLUSIVE); if (error) goto out; sdzp = VTOZ(sdvp); error = vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT); if (error != 0) { VOP_UNLOCK(sdvp, 0); if (error != EBUSY) goto out; error = vn_lock(tdvp, LK_EXCLUSIVE); if (error) goto out; VOP_UNLOCK(tdvp, 0); goto relock; } tdzp = VTOZ(tdvp); /* * Before using sdzp and tdzp we must ensure that they are live. * As a porting legacy from illumos we have two things to worry * about. One is typical for FreeBSD and it is that the vnode is * not reclaimed (doomed). The other is that the znode is live. * The current code can invalidate the znode without acquiring the * corresponding vnode lock if the object represented by the znode * and vnode is no longer valid after a rollback or receive operation. * z_teardown_lock hidden behind ZFS_ENTER and ZFS_EXIT is the lock * that protects the znodes from the invalidation. */ zfsvfs = sdzp->z_zfsvfs; ASSERT3P(zfsvfs, ==, tdzp->z_zfsvfs); ZFS_ENTER(zfsvfs); /* * We can not use ZFS_VERIFY_ZP() here because it could directly return * bypassing the cleanup code in the case of an error. */ if (tdzp->z_sa_hdl == NULL || sdzp->z_sa_hdl == NULL) { ZFS_EXIT(zfsvfs); VOP_UNLOCK(sdvp, 0); VOP_UNLOCK(tdvp, 0); error = SET_ERROR(EIO); goto out; } /* * Re-resolve svp to be certain it still exists and fetch the * correct vnode. */ error = zfs_dirent_lookup(sdzp, snm, &szp, ZEXISTS); if (error != 0) { /* Source entry invalid or not there. */ ZFS_EXIT(zfsvfs); VOP_UNLOCK(sdvp, 0); VOP_UNLOCK(tdvp, 0); if ((scnp->cn_flags & ISDOTDOT) != 0 || (scnp->cn_namelen == 1 && scnp->cn_nameptr[0] == '.')) error = SET_ERROR(EINVAL); goto out; } svp = ZTOV(szp); /* * Re-resolve tvp, if it disappeared we just carry on. */ error = zfs_dirent_lookup(tdzp, tnm, &tzp, 0); if (error != 0) { ZFS_EXIT(zfsvfs); VOP_UNLOCK(sdvp, 0); VOP_UNLOCK(tdvp, 0); vrele(svp); if ((tcnp->cn_flags & ISDOTDOT) != 0) error = SET_ERROR(EINVAL); goto out; } if (tzp != NULL) tvp = ZTOV(tzp); else tvp = NULL; /* * At present the vnode locks must be acquired before z_teardown_lock, * although it would be more logical to use the opposite order. */ ZFS_EXIT(zfsvfs); /* * Now try acquire locks on svp and tvp. */ nvp = svp; error = vn_lock(nvp, LK_EXCLUSIVE | LK_NOWAIT); if (error != 0) { VOP_UNLOCK(sdvp, 0); VOP_UNLOCK(tdvp, 0); 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_UNLOCK(nvp, 0); /* * 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_UNLOCK(sdvp, 0); VOP_UNLOCK(tdvp, 0); VOP_UNLOCK(*svpp, 0); 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_vnrele_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_vnrele_taskq(dmu_objset_pool(zfsvfs->z_os))); return (error); } /* * Move an entry from the provided source directory to the target * directory. Change the entry name as indicated. * * IN: sdvp - Source directory containing the "old entry". * snm - Old entry name. * tdvp - Target directory to contain the "new entry". * tnm - New entry name. * cr - credentials of caller. * ct - caller context * flags - case flags * * RETURN: 0 on success, error code on failure. * * Timestamps: * sdvp,tdvp - ctime|mtime updated */ /*ARGSUSED*/ static int zfs_rename(vnode_t *sdvp, vnode_t **svpp, struct componentname *scnp, vnode_t *tdvp, vnode_t **tvpp, struct componentname *tcnp, cred_t *cr) { zfsvfs_t *zfsvfs; znode_t *sdzp, *tdzp, *szp, *tzp; zilog_t *zilog = NULL; dmu_tx_t *tx; char *snm = scnp->cn_nameptr; char *tnm = tcnp->cn_nameptr; int error = 0; /* Reject renames across filesystems. */ if ((*svpp)->v_mount != tdvp->v_mount || ((*tvpp) != NULL && (*svpp)->v_mount != (*tvpp)->v_mount)) { error = SET_ERROR(EXDEV); goto out; } if (zfsctl_is_node(tdvp)) { error = SET_ERROR(EXDEV); goto out; } /* * Lock all four vnodes to ensure safety and semantics of renaming. */ error = zfs_rename_relock(sdvp, svpp, tdvp, tvpp, scnp, tcnp); if (error != 0) { /* no vnodes are locked in the case of error here */ return (error); } tdzp = VTOZ(tdvp); sdzp = VTOZ(sdvp); zfsvfs = tdzp->z_zfsvfs; zilog = zfsvfs->z_log; /* * After we re-enter ZFS_ENTER() we will have to revalidate all * znodes involved. */ ZFS_ENTER(zfsvfs); if (zfsvfs->z_utf8 && u8_validate(tnm, strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { error = SET_ERROR(EILSEQ); goto unlockout; } /* If source and target are the same file, there is nothing to do. */ if ((*svpp) == (*tvpp)) { error = 0; goto unlockout; } if (((*svpp)->v_type == VDIR && (*svpp)->v_mountedhere != NULL) || ((*tvpp) != NULL && (*tvpp)->v_type == VDIR && (*tvpp)->v_mountedhere != NULL)) { error = SET_ERROR(EXDEV); goto unlockout; } /* * We can not use ZFS_VERIFY_ZP() here because it could directly return * bypassing the cleanup code in the case of an error. */ if (tdzp->z_sa_hdl == NULL || sdzp->z_sa_hdl == NULL) { error = SET_ERROR(EIO); goto unlockout; } szp = VTOZ(*svpp); tzp = *tvpp == NULL ? NULL : VTOZ(*tvpp); if (szp->z_sa_hdl == NULL || (tzp != NULL && tzp->z_sa_hdl == NULL)) { error = SET_ERROR(EIO); goto unlockout; } /* * This is to prevent the creation of links into attribute space * by renaming a linked file into/outof an attribute directory. * See the comment in zfs_link() for why this is considered bad. */ if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) { error = SET_ERROR(EINVAL); goto unlockout; } /* * Must have write access at the source to remove the old entry * and write access at the target to create the new entry. * Note that if target and source are the same, this can be * done in a single check. */ if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) goto unlockout; if ((*svpp)->v_type == VDIR) { /* * Avoid ".", "..", and aliases of "." for obvious reasons. */ if ((scnp->cn_namelen == 1 && scnp->cn_nameptr[0] == '.') || sdzp == szp || (scnp->cn_flags | tcnp->cn_flags) & ISDOTDOT) { error = EINVAL; goto unlockout; } /* * Check to make sure rename is valid. * Can't do a move like this: /usr/a/b to /usr/a/b/c/d */ if (error = zfs_rename_check(szp, sdzp, tdzp)) goto unlockout; } /* * Does target exist? */ if (tzp) { /* * Source and target must be the same type. */ if ((*svpp)->v_type == VDIR) { if ((*tvpp)->v_type != VDIR) { error = SET_ERROR(ENOTDIR); goto unlockout; } else { cache_purge(tdvp); if (sdvp != tdvp) cache_purge(sdvp); } } else { if ((*tvpp)->v_type == VDIR) { error = SET_ERROR(EISDIR); goto unlockout; } } } vnevent_rename_src(*svpp, sdvp, scnp->cn_nameptr, ct); if (tzp) vnevent_rename_dest(*tvpp, tdvp, tnm, ct); /* * notify the target directory if it is not the same * as source directory. */ if (tdvp != sdvp) { vnevent_rename_dest_dir(tdvp, ct); } tx = dmu_tx_create(zfsvfs->z_os); dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE); dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); if (sdzp != tdzp) { dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE); zfs_sa_upgrade_txholds(tx, tdzp); } if (tzp) { dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE); zfs_sa_upgrade_txholds(tx, tzp); } zfs_sa_upgrade_txholds(tx, szp); dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); error = dmu_tx_assign(tx, TXG_WAIT); if (error) { dmu_tx_abort(tx); goto unlockout; } if (tzp) /* Attempt to remove the existing target */ error = zfs_link_destroy(tdzp, tnm, tzp, tx, 0, NULL); if (error == 0) { error = zfs_link_create(tdzp, tnm, szp, tx, ZRENAMING); if (error == 0) { szp->z_pflags |= ZFS_AV_MODIFIED; error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs), (void *)&szp->z_pflags, sizeof (uint64_t), tx); ASSERT0(error); error = zfs_link_destroy(sdzp, snm, szp, tx, ZRENAMING, NULL); if (error == 0) { zfs_log_rename(zilog, tx, TX_RENAME, sdzp, snm, tdzp, tnm, szp); /* * Update path information for the target vnode */ vn_renamepath(tdvp, *svpp, tnm, strlen(tnm)); } else { /* * At this point, we have successfully created * the target name, but have failed to remove * the source name. Since the create was done * with the ZRENAMING flag, there are * complications; for one, the link count is * wrong. The easiest way to deal with this * is to remove the newly created target, and * return the original error. This must * succeed; fortunately, it is very unlikely to * fail, since we just created it. */ VERIFY3U(zfs_link_destroy(tdzp, tnm, szp, tx, ZRENAMING, NULL), ==, 0); } } if (error == 0) { cache_purge(*svpp); if (*tvpp != NULL) cache_purge(*tvpp); cache_purge_negative(tdvp); } } dmu_tx_commit(tx); unlockout: /* all 4 vnodes are locked, ZFS_ENTER called */ ZFS_EXIT(zfsvfs); VOP_UNLOCK(*svpp, 0); VOP_UNLOCK(sdvp, 0); out: /* original two vnodes are locked */ if (error == 0 && zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) zil_commit(zilog, 0); if (*tvpp != NULL) VOP_UNLOCK(*tvpp, 0); if (tdvp != *tvpp) VOP_UNLOCK(tdvp, 0); return (error); } /* * Insert the indicated symbolic reference entry into the directory. * * IN: dvp - Directory to contain new symbolic link. * link - Name for new symlink entry. * vap - Attributes of new entry. * cr - credentials of caller. * ct - caller context * flags - case flags * * RETURN: 0 on success, error code on failure. * * Timestamps: * dvp - ctime|mtime updated */ /*ARGSUSED*/ static int zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link, cred_t *cr, kthread_t *td) { znode_t *zp, *dzp = VTOZ(dvp); 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; int flags = 0; ASSERT(vap->va_type == VLNK); ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(dzp); zilog = zfsvfs->z_log; if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EILSEQ)); } if (len > MAXPATHLEN) { ZFS_EXIT(zfsvfs); return (SET_ERROR(ENAMETOOLONG)); } if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, NULL, &acl_ids)) != 0) { ZFS_EXIT(zfsvfs); return (error); } /* * Attempt to lock directory; fail if entry already exists. */ error = zfs_dirent_lookup(dzp, name, &zp, ZNEW); if (error) { zfs_acl_ids_free(&acl_ids); ZFS_EXIT(zfsvfs); return (error); } if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { zfs_acl_ids_free(&acl_ids); ZFS_EXIT(zfsvfs); return (error); } if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { zfs_acl_ids_free(&acl_ids); ZFS_EXIT(zfsvfs); return (SET_ERROR(EDQUOT)); } getnewvnode_reserve(1); tx = dmu_tx_create(zfsvfs->z_os); fuid_dirtied = zfsvfs->z_fuid_dirty; dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + ZFS_SA_BASE_ATTR_SIZE + len); dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, acl_ids.z_aclp->z_acl_bytes); } if (fuid_dirtied) zfs_fuid_txhold(zfsvfs, tx); error = dmu_tx_assign(tx, TXG_WAIT); if (error) { zfs_acl_ids_free(&acl_ids); dmu_tx_abort(tx); getnewvnode_drop_reserve(); ZFS_EXIT(zfsvfs); return (error); } /* * Create a new object for the symlink. * for version 4 ZPL datsets 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), link, len, tx); else zfs_sa_symlink(zp, 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); *vpp = ZTOV(zp); zfs_acl_ids_free(&acl_ids); dmu_tx_commit(tx); getnewvnode_drop_reserve(); if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) zil_commit(zilog, 0); ZFS_EXIT(zfsvfs); return (error); } /* * Return, in the buffer contained in the provided uio structure, * the symbolic path referred to by vp. * * IN: vp - vnode of symbolic link. * uio - structure to contain the link path. * cr - credentials of caller. * ct - caller context * * OUT: uio - structure containing the link path. * * RETURN: 0 on success, error code on failure. * * Timestamps: * vp - atime updated */ /* ARGSUSED */ static int zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct) { znode_t *zp = VTOZ(vp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; int error; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); if (zp->z_is_sa) error = sa_lookup_uio(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs), uio); else error = zfs_sa_readlink(zp, uio); ZFS_ACCESSTIME_STAMP(zfsvfs, zp); ZFS_EXIT(zfsvfs); return (error); } /* * Insert a new entry into directory tdvp referencing svp. * * IN: tdvp - Directory to contain new entry. * svp - vnode of new entry. * name - name of new entry. * cr - credentials of caller. * ct - caller context * * RETURN: 0 on success, error code on failure. * * Timestamps: * tdvp - ctime|mtime updated * svp - ctime updated */ /* ARGSUSED */ static int zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr, caller_context_t *ct, int flags) { znode_t *dzp = VTOZ(tdvp); znode_t *tzp, *szp; zfsvfs_t *zfsvfs = dzp->z_zfsvfs; zilog_t *zilog; dmu_tx_t *tx; int error; uint64_t parent; uid_t owner; ASSERT(tdvp->v_type == VDIR); ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(dzp); zilog = zfsvfs->z_log; /* * POSIX dictates that we return EPERM here. * Better choices include ENOTSUP or EISDIR. */ if (svp->v_type == VDIR) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EPERM)); } szp = VTOZ(svp); ZFS_VERIFY_ZP(szp); if (szp->z_pflags & (ZFS_APPENDONLY | ZFS_IMMUTABLE | ZFS_READONLY)) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EPERM)); } /* Prevent links to .zfs/shares files */ if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), &parent, sizeof (uint64_t))) != 0) { ZFS_EXIT(zfsvfs); return (error); } if (parent == zfsvfs->z_shares_dir) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EPERM)); } if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EILSEQ)); } /* * We do not support links between attributes and non-attributes * because of the potential security risk of creating links * into "normal" file space in order to circumvent restrictions * imposed in attribute space. */ if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EINVAL)); } owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER); if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) { ZFS_EXIT(zfsvfs); return (SET_ERROR(EPERM)); } if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { ZFS_EXIT(zfsvfs); return (error); } /* * Attempt to lock directory; fail if entry already exists. */ error = zfs_dirent_lookup(dzp, name, &tzp, ZNEW); if (error) { ZFS_EXIT(zfsvfs); return (error); } tx = dmu_tx_create(zfsvfs->z_os); dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); zfs_sa_upgrade_txholds(tx, szp); zfs_sa_upgrade_txholds(tx, dzp); error = dmu_tx_assign(tx, TXG_WAIT); if (error) { dmu_tx_abort(tx); ZFS_EXIT(zfsvfs); return (error); } error = zfs_link_create(dzp, name, szp, tx, 0); if (error == 0) { uint64_t txtype = TX_LINK; zfs_log_link(zilog, tx, txtype, dzp, szp, name); } dmu_tx_commit(tx); if (error == 0) { vnevent_link(svp, ct); } if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) zil_commit(zilog, 0); ZFS_EXIT(zfsvfs); return (error); } /*ARGSUSED*/ void zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct) { znode_t *zp = VTOZ(vp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; int error; rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); if (zp->z_sa_hdl == NULL) { /* * The fs has been unmounted, or we did a * suspend/resume and this file no longer exists. */ rw_exit(&zfsvfs->z_teardown_inactive_lock); vrecycle(vp); return; } if (zp->z_unlinked) { /* * Fast path to recycle a vnode of a removed file. */ rw_exit(&zfsvfs->z_teardown_inactive_lock); 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); } } rw_exit(&zfsvfs->z_teardown_inactive_lock); } CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid)); CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid)); /*ARGSUSED*/ static int zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct) { znode_t *zp = VTOZ(vp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; uint32_t gen; uint64_t gen64; uint64_t object = zp->z_id; zfid_short_t *zfid; int size, i, error; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &gen64, sizeof (uint64_t))) != 0) { ZFS_EXIT(zfsvfs); return (error); } gen = (uint32_t)gen64; size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; #ifdef illumos if (fidp->fid_len < size) { fidp->fid_len = size; ZFS_EXIT(zfsvfs); return (SET_ERROR(ENOSPC)); } #else fidp->fid_len = size; #endif zfid = (zfid_short_t *)fidp; zfid->zf_len = size; for (i = 0; i < sizeof (zfid->zf_object); i++) zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); /* Must have a non-zero generation number to distinguish from .zfs */ if (gen == 0) gen = 1; for (i = 0; i < sizeof (zfid->zf_gen); i++) zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); if (size == LONG_FID_LEN) { uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); zfid_long_t *zlfid; zlfid = (zfid_long_t *)fidp; for (i = 0; i < sizeof (zlfid->zf_setid); i++) zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); /* XXX - this should be the generation number for the objset */ for (i = 0; i < sizeof (zlfid->zf_setgen); i++) zlfid->zf_setgen[i] = 0; } ZFS_EXIT(zfsvfs); return (0); } static int zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, caller_context_t *ct) { znode_t *zp, *xzp; 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); #ifdef illumos case _PC_XATTR_EXISTS: zp = VTOZ(vp); zfsvfs = zp->z_zfsvfs; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); *valp = 0; error = zfs_dirent_lookup(zp, "", &xzp, ZXATTR | ZEXISTS | ZSHARED); if (error == 0) { if (!zfs_dirempty(xzp)) *valp = 1; vrele(ZTOV(xzp)); } else if (error == ENOENT) { /* * If there aren't extended attributes, it's the * same as having zero of them. */ error = 0; } ZFS_EXIT(zfsvfs); return (error); case _PC_SATTR_ENABLED: case _PC_SATTR_EXISTS: *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && (vp->v_type == VREG || vp->v_type == VDIR); return (0); case _PC_ACCESS_FILTERING: *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) && vp->v_type == VDIR; return (0); case _PC_ACL_ENABLED: *valp = _ACL_ACE_ENABLED; return (0); #endif /* illumos */ case _PC_MIN_HOLE_SIZE: *valp = (int)SPA_MINBLOCKSIZE; return (0); #ifdef illumos case _PC_TIMESTAMP_RESOLUTION: /* nanosecond timestamp resolution */ *valp = 1L; return (0); #endif case _PC_ACL_EXTENDED: *valp = 0; return (0); case _PC_ACL_NFS4: *valp = 1; return (0); case _PC_ACL_PATH_MAX: *valp = ACL_MAX_ENTRIES; return (0); default: return (EOPNOTSUPP); } } /*ARGSUSED*/ static int zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, caller_context_t *ct) { znode_t *zp = VTOZ(vp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; int error; boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); error = zfs_getacl(zp, vsecp, skipaclchk, cr); ZFS_EXIT(zfsvfs); return (error); } /*ARGSUSED*/ int zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, caller_context_t *ct) { znode_t *zp = VTOZ(vp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; int error; boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; zilog_t *zilog = zfsvfs->z_log; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); error = zfs_setacl(zp, vsecp, skipaclchk, cr); if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) zil_commit(zilog, 0); ZFS_EXIT(zfsvfs); return (error); } static int zfs_getpages(struct vnode *vp, vm_page_t *ma, int count, int *rbehind, int *rahead) { znode_t *zp = VTOZ(vp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; objset_t *os = zp->z_zfsvfs->z_os; rl_t *rl; vm_object_t object; off_t start, end, obj_size; uint_t blksz; int pgsin_b, pgsin_a; int error; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); start = IDX_TO_OFF(ma[0]->pindex); end = IDX_TO_OFF(ma[count - 1]->pindex + 1); /* * Lock a range covering all required and optional pages. * Note that we need to handle the case of the block size growing. */ for (;;) { blksz = zp->z_blksz; rl = zfs_range_lock(zp, rounddown(start, blksz), roundup(end, blksz) - rounddown(start, blksz), RL_READER); if (blksz == zp->z_blksz) break; zfs_range_unlock(rl); } 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) { zfs_range_unlock(rl); ZFS_EXIT(zfsvfs); return (zfs_vm_pagerret_bad); } pgsin_b = 0; if (rbehind != NULL) { pgsin_b = OFF_TO_IDX(start - rounddown(start, blksz)); pgsin_b = MIN(*rbehind, pgsin_b); } pgsin_a = 0; if (rahead != NULL) { pgsin_a = OFF_TO_IDX(roundup(end, blksz) - end); if (end + IDX_TO_OFF(pgsin_a) >= obj_size) pgsin_a = OFF_TO_IDX(round_page(obj_size) - end); pgsin_a = MIN(*rahead, pgsin_a); } /* * NB: we need to pass the exact byte size of the data that we expect * to read after accounting for the file size. This is required because * ZFS will panic if we request DMU to read beyond the end of the last * allocated block. */ error = dmu_read_pages(os, zp->z_id, ma, count, &pgsin_b, &pgsin_a, MIN(end, obj_size) - (end - PAGE_SIZE)); zfs_range_unlock(rl); ZFS_ACCESSTIME_STAMP(zfsvfs, zp); ZFS_EXIT(zfsvfs); if (error != 0) return (zfs_vm_pagerret_error); VM_CNT_INC(v_vnodein); VM_CNT_ADD(v_vnodepgsin, count + pgsin_b + pgsin_a); if (rbehind != NULL) *rbehind = pgsin_b; if (rahead != NULL) *rahead = pgsin_a; return (zfs_vm_pagerret_ok); } static int zfs_freebsd_getpages(ap) struct vop_getpages_args /* { struct vnode *a_vp; vm_page_t *a_m; int a_count; int *a_rbehind; int *a_rahead; } */ *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; rl_t *rl; dmu_tx_t *tx; struct sf_buf *sf; vm_object_t object; vm_page_t m; caddr_t va; size_t tocopy; size_t lo_len; vm_ooffset_t lo_off; vm_ooffset_t off; uint_t blksz; int ncount; int pcount; int err; int i; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); object = vp->v_object; pcount = btoc(len); ncount = pcount; KASSERT(ma[0]->object == object, ("mismatching object")); KASSERT(len > 0 && (len & PAGE_MASK) == 0, ("unexpected length")); for (i = 0; i < pcount; i++) rtvals[i] = zfs_vm_pagerret_error; off = IDX_TO_OFF(ma[0]->pindex); blksz = zp->z_blksz; lo_off = rounddown(off, blksz); lo_len = roundup(len + (off - lo_off), blksz); rl = zfs_range_lock(zp, 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_owner_overquota(zfsvfs, zp, B_FALSE) || zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 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, B_TRUE); err = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); ASSERT0(err); zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0); 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_range_unlock(rl); if ((flags & (zfs_vm_pagerput_sync | zfs_vm_pagerput_inval)) != 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) zil_commit(zfsvfs->z_log, zp->z_id); ZFS_EXIT(zfsvfs); return (rtvals[0]); } int zfs_freebsd_putpages(ap) struct vop_putpages_args /* { struct vnode *a_vp; vm_page_t *a_m; int a_count; int a_sync; int *a_rtvals; } */ *ap; { return (zfs_putpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_sync, ap->a_rtvals)); } static int zfs_freebsd_bmap(ap) 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; } */ *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); } static int zfs_freebsd_open(ap) struct vop_open_args /* { struct vnode *a_vp; int a_mode; struct ucred *a_cred; struct thread *a_td; } */ *ap; { vnode_t *vp = ap->a_vp; znode_t *zp = VTOZ(vp); int error; error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL); if (error == 0) vnode_create_vobject(vp, zp->z_size, ap->a_td); return (error); } static int zfs_freebsd_close(ap) struct vop_close_args /* { struct vnode *a_vp; int a_fflag; struct ucred *a_cred; struct thread *a_td; } */ *ap; { return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL)); } static int zfs_freebsd_ioctl(ap) struct vop_ioctl_args /* { struct vnode *a_vp; u_long a_command; caddr_t a_data; int a_fflag; struct ucred *cred; struct thread *td; } */ *ap; { return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data, ap->a_fflag, ap->a_cred, NULL, NULL)); } static int ioflags(int ioflags) { int flags = 0; if (ioflags & IO_APPEND) flags |= FAPPEND; if (ioflags & IO_NDELAY) flags |= FNONBLOCK; if (ioflags & IO_SYNC) flags |= (FSYNC | FDSYNC | FRSYNC); return (flags); } static int zfs_freebsd_read(ap) struct vop_read_args /* { struct vnode *a_vp; struct uio *a_uio; int a_ioflag; struct ucred *a_cred; } */ *ap; { return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag), ap->a_cred, NULL)); } static int zfs_freebsd_write(ap) struct vop_write_args /* { struct vnode *a_vp; struct uio *a_uio; int a_ioflag; struct ucred *a_cred; } */ *ap; { return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag), ap->a_cred, NULL)); } static int zfs_freebsd_access(ap) struct vop_access_args /* { struct vnode *a_vp; accmode_t a_accmode; struct ucred *a_cred; struct thread *a_td; } */ *ap; { vnode_t *vp = ap->a_vp; znode_t *zp = VTOZ(vp); accmode_t accmode; int error = 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(ap->a_vp, accmode, 0, ap->a_cred, NULL); /* * VADMIN has to be handled by vaccess(). */ if (error == 0) { accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND); if (accmode != 0) { error = vaccess(vp->v_type, zp->z_mode, zp->z_uid, zp->z_gid, accmode, ap->a_cred, NULL); } } /* * 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); } static int zfs_freebsd_lookup(ap) struct vop_lookup_args /* { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; } */ *ap; { struct componentname *cnp = ap->a_cnp; char nm[NAME_MAX + 1]; ASSERT(cnp->cn_namelen < sizeof(nm)); strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm))); return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop, cnp->cn_cred, cnp->cn_thread, 0)); } static int zfs_cache_lookup(ap) struct vop_lookup_args /* { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; } */ *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)); } static int zfs_freebsd_create(ap) struct vop_create_args /* { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; struct vattr *a_vap; } */ *ap; { zfsvfs_t *zfsvfs; struct componentname *cnp = ap->a_cnp; vattr_t *vap = ap->a_vap; int error, mode; ASSERT(cnp->cn_flags & SAVENAME); vattr_init_mask(vap); mode = vap->va_mode & ALLPERMS; zfsvfs = ap->a_dvp->v_mount->mnt_data; error = zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode, ap->a_vpp, cnp->cn_cred, cnp->cn_thread); if (zfsvfs->z_use_namecache && error == 0 && (cnp->cn_flags & MAKEENTRY) != 0) cache_enter(ap->a_dvp, *ap->a_vpp, cnp); return (error); } static int zfs_freebsd_remove(ap) struct vop_remove_args /* { struct vnode *a_dvp; struct vnode *a_vp; struct componentname *a_cnp; } */ *ap; { ASSERT(ap->a_cnp->cn_flags & SAVENAME); return (zfs_remove(ap->a_dvp, ap->a_vp, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_cred)); } static int zfs_freebsd_mkdir(ap) struct vop_mkdir_args /* { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; struct vattr *a_vap; } */ *ap; { vattr_t *vap = ap->a_vap; ASSERT(ap->a_cnp->cn_flags & SAVENAME); vattr_init_mask(vap); return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp, ap->a_cnp->cn_cred)); } static int zfs_freebsd_rmdir(ap) struct vop_rmdir_args /* { struct vnode *a_dvp; struct vnode *a_vp; struct componentname *a_cnp; } */ *ap; { struct componentname *cnp = ap->a_cnp; ASSERT(cnp->cn_flags & SAVENAME); return (zfs_rmdir(ap->a_dvp, ap->a_vp, cnp->cn_nameptr, cnp->cn_cred)); } static int zfs_freebsd_readdir(ap) struct vop_readdir_args /* { struct vnode *a_vp; struct uio *a_uio; struct ucred *a_cred; int *a_eofflag; int *a_ncookies; u_long **a_cookies; } */ *ap; { return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag, ap->a_ncookies, ap->a_cookies)); } static int zfs_freebsd_fsync(ap) struct vop_fsync_args /* { struct vnode *a_vp; int a_waitfor; struct thread *a_td; } */ *ap; { vop_stdfsync(ap); return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL)); } static int zfs_freebsd_getattr(ap) struct vop_getattr_args /* { struct vnode *a_vp; struct vattr *a_vap; struct ucred *a_cred; } */ *ap; { vattr_t *vap = ap->a_vap; xvattr_t xvap; u_long 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, NULL); 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); } static int zfs_freebsd_setattr(ap) struct vop_setattr_args /* { struct vnode *a_vp; struct vattr *a_vap; struct ucred *a_cred; } */ *ap; { vnode_t *vp = ap->a_vp; vattr_t *vap = ap->a_vap; cred_t *cred = ap->a_cred; xvattr_t xvap; u_long 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 || 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_hidden); 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(vp, (vattr_t *)&xvap, 0, cred, NULL)); } static int zfs_freebsd_rename(ap) struct vop_rename_args /* { struct vnode *a_fdvp; struct vnode *a_fvp; struct componentname *a_fcnp; struct vnode *a_tdvp; struct vnode *a_tvp; struct componentname *a_tcnp; } */ *ap; { vnode_t *fdvp = ap->a_fdvp; vnode_t *fvp = ap->a_fvp; vnode_t *tdvp = ap->a_tdvp; vnode_t *tvp = ap->a_tvp; int error; ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART)); ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART)); error = zfs_rename(fdvp, &fvp, ap->a_fcnp, tdvp, &tvp, ap->a_tcnp, ap->a_fcnp->cn_cred); vrele(fdvp); vrele(fvp); vrele(tdvp); if (tvp != NULL) vrele(tvp); return (error); } static int zfs_freebsd_symlink(ap) struct vop_symlink_args /* { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; struct vattr *a_vap; char *a_target; } */ *ap; { struct componentname *cnp = ap->a_cnp; vattr_t *vap = ap->a_vap; ASSERT(cnp->cn_flags & SAVENAME); vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */ vattr_init_mask(vap); return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap, __DECONST(char *, ap->a_target), cnp->cn_cred, cnp->cn_thread)); } static int zfs_freebsd_readlink(ap) struct vop_readlink_args /* { struct vnode *a_vp; struct uio *a_uio; struct ucred *a_cred; } */ *ap; { return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL)); } static int zfs_freebsd_link(ap) struct vop_link_args /* { struct vnode *a_tdvp; struct vnode *a_vp; struct componentname *a_cnp; } */ *ap; { struct componentname *cnp = ap->a_cnp; vnode_t *vp = ap->a_vp; vnode_t *tdvp = ap->a_tdvp; if (tdvp->v_mount != vp->v_mount) return (EXDEV); ASSERT(cnp->cn_flags & SAVENAME); return (zfs_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0)); } static int zfs_freebsd_inactive(ap) struct vop_inactive_args /* { struct vnode *a_vp; struct thread *a_td; } */ *ap; { vnode_t *vp = ap->a_vp; zfs_inactive(vp, ap->a_td->td_ucred, NULL); return (0); } static int zfs_freebsd_reclaim(ap) struct vop_reclaim_args /* { struct vnode *a_vp; struct thread *a_td; } */ *ap; { vnode_t *vp = ap->a_vp; znode_t *zp = VTOZ(vp); zfsvfs_t *zfsvfs = zp->z_zfsvfs; ASSERT(zp != NULL); /* Destroy the vm object and flush associated pages. */ vnode_destroy_vobject(vp); /* * z_teardown_inactive_lock protects from a race with * zfs_znode_dmu_fini in zfsvfs_teardown during * force unmount. */ rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); if (zp->z_sa_hdl == NULL) zfs_znode_free(zp); else zfs_zinactive(zp); rw_exit(&zfsvfs->z_teardown_inactive_lock); vp->v_data = NULL; return (0); } static int zfs_freebsd_fid(ap) struct vop_fid_args /* { struct vnode *a_vp; struct fid *a_fid; } */ *ap; { return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL)); } static int zfs_freebsd_pathconf(ap) struct vop_pathconf_args /* { struct vnode *a_vp; int a_name; register_t *a_retval; } */ *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); case _PC_PIPE_BUF: if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO) { *ap->a_retval = PIPE_BUF; return (0); } return (EINVAL); default: return (vop_stdpathconf(ap)); } } /* * FreeBSD's extended attributes namespace defines file name prefix for ZFS' * extended attribute name: * * NAMESPACE PREFIX * system freebsd:system: * user (none, can be used to access ZFS fsattr(5) attributes * created on Solaris) */ static int zfs_create_attrname(int attrnamespace, const char *name, char *attrname, size_t size) { const char *namespace, *prefix, *suffix; /* We don't allow '/' character in attribute name. */ if (strchr(name, '/') != NULL) return (EINVAL); /* We don't allow attribute names that start with "freebsd:" string. */ if (strncmp(name, "freebsd:", 8) == 0) return (EINVAL); bzero(attrname, size); switch (attrnamespace) { case EXTATTR_NAMESPACE_USER: #if 0 prefix = "freebsd:"; namespace = EXTATTR_NAMESPACE_USER_STRING; suffix = ":"; #else /* * This is the default namespace by which we can access all * attributes created on Solaris. */ prefix = namespace = suffix = ""; #endif break; case EXTATTR_NAMESPACE_SYSTEM: prefix = "freebsd:"; namespace = EXTATTR_NAMESPACE_SYSTEM_STRING; suffix = ":"; break; case EXTATTR_NAMESPACE_EMPTY: default: return (EINVAL); } if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix, name) >= size) { return (ENAMETOOLONG); } return (0); } /* * Vnode operating to retrieve a named extended attribute. */ static int zfs_getextattr(struct vop_getextattr_args *ap) /* vop_getextattr { IN struct vnode *a_vp; IN int a_attrnamespace; IN const char *a_name; INOUT struct uio *a_uio; OUT size_t *a_size; IN struct ucred *a_cred; IN struct thread *a_td; }; */ { zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; struct thread *td = ap->a_td; struct nameidata nd; char attrname[255]; struct vattr va; vnode_t *xvp = NULL, *vp; int error, flags; error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, ap->a_cred, ap->a_td, VREAD); if (error != 0) return (error); error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, sizeof(attrname)); if (error != 0) return (error); ZFS_ENTER(zfsvfs); error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, LOOKUP_XATTR); if (error != 0) { ZFS_EXIT(zfsvfs); return (error); } flags = FREAD; NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, xvp, td); error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL); vp = nd.ni_vp; NDFREE(&nd, NDF_ONLY_PNBUF); if (error != 0) { ZFS_EXIT(zfsvfs); if (error == ENOENT) error = ENOATTR; return (error); } if (ap->a_size != NULL) { error = VOP_GETATTR(vp, &va, ap->a_cred); if (error == 0) *ap->a_size = (size_t)va.va_size; } else if (ap->a_uio != NULL) error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred); VOP_UNLOCK(vp, 0); vn_close(vp, flags, ap->a_cred, td); ZFS_EXIT(zfsvfs); return (error); } /* * Vnode operation to remove a named attribute. */ int zfs_deleteextattr(struct vop_deleteextattr_args *ap) /* vop_deleteextattr { IN struct vnode *a_vp; IN int a_attrnamespace; IN const char *a_name; IN struct ucred *a_cred; IN struct thread *a_td; }; */ { zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; struct thread *td = ap->a_td; struct nameidata nd; char attrname[255]; struct vattr va; vnode_t *xvp = NULL, *vp; int error, flags; error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, ap->a_cred, ap->a_td, VWRITE); if (error != 0) return (error); error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, sizeof(attrname)); if (error != 0) return (error); ZFS_ENTER(zfsvfs); error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, LOOKUP_XATTR); if (error != 0) { ZFS_EXIT(zfsvfs); return (error); } NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF, UIO_SYSSPACE, attrname, xvp, td); error = namei(&nd); vp = nd.ni_vp; if (error != 0) { ZFS_EXIT(zfsvfs); NDFREE(&nd, NDF_ONLY_PNBUF); if (error == ENOENT) error = ENOATTR; return (error); } error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd); NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if (vp == nd.ni_dvp) vrele(vp); else vput(vp); ZFS_EXIT(zfsvfs); return (error); } /* * Vnode operation to set a named attribute. */ static int zfs_setextattr(struct vop_setextattr_args *ap) /* vop_setextattr { IN struct vnode *a_vp; IN int a_attrnamespace; IN const char *a_name; INOUT struct uio *a_uio; IN struct ucred *a_cred; IN struct thread *a_td; }; */ { zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; struct thread *td = ap->a_td; struct nameidata nd; char attrname[255]; struct vattr va; vnode_t *xvp = NULL, *vp; int error, flags; error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, ap->a_cred, ap->a_td, VWRITE); if (error != 0) return (error); error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, sizeof(attrname)); if (error != 0) return (error); ZFS_ENTER(zfsvfs); error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, LOOKUP_XATTR | CREATE_XATTR_DIR); if (error != 0) { ZFS_EXIT(zfsvfs); return (error); } flags = FFLAGS(O_WRONLY | O_CREAT); NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, xvp, td); error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL); vp = nd.ni_vp; NDFREE(&nd, NDF_ONLY_PNBUF); if (error != 0) { ZFS_EXIT(zfsvfs); return (error); } VATTR_NULL(&va); va.va_size = 0; error = VOP_SETATTR(vp, &va, ap->a_cred); if (error == 0) VOP_WRITE(vp, ap->a_uio, IO_UNIT, ap->a_cred); VOP_UNLOCK(vp, 0); vn_close(vp, flags, ap->a_cred, td); ZFS_EXIT(zfsvfs); return (error); } /* * Vnode operation to retrieve extended attributes on a vnode. */ static int zfs_listextattr(struct vop_listextattr_args *ap) /* 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; }; */ { zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; struct thread *td = ap->a_td; struct nameidata nd; char attrprefix[16]; u_char dirbuf[sizeof(struct dirent)]; struct dirent *dp; struct iovec aiov; struct uio auio, *uio = ap->a_uio; size_t *sizep = ap->a_size; size_t plen; vnode_t *xvp = NULL, *vp; int done, error, eof, pos; error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, ap->a_cred, ap->a_td, VREAD); if (error != 0) return (error); error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix, sizeof(attrprefix)); if (error != 0) return (error); plen = strlen(attrprefix); ZFS_ENTER(zfsvfs); if (sizep != NULL) *sizep = 0; error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, LOOKUP_XATTR); if (error != 0) { ZFS_EXIT(zfsvfs); /* * ENOATTR means that the EA directory does not yet exist, * i.e. there are no extended attributes there. */ if (error == ENOATTR) error = 0; return (error); } NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED, UIO_SYSSPACE, ".", xvp, td); error = namei(&nd); vp = nd.ni_vp; NDFREE(&nd, NDF_ONLY_PNBUF); if (error != 0) { ZFS_EXIT(zfsvfs); return (error); } auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_segflg = UIO_SYSSPACE; auio.uio_td = td; auio.uio_rw = UIO_READ; auio.uio_offset = 0; do { u_char nlen; 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); done = sizeof(dirbuf) - auio.uio_resid; if (error != 0) break; for (pos = 0; pos < done;) { 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; if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0) continue; else if (strncmp(dp->d_name, attrprefix, plen) != 0) continue; nlen = dp->d_namlen - plen; if (sizep != NULL) *sizep += 1 + nlen; else if (uio != NULL) { /* * Format of extattr name entry is one byte for * length and the rest for name. */ error = uiomove(&nlen, 1, uio->uio_rw, uio); if (error == 0) { error = uiomove(dp->d_name + plen, nlen, uio->uio_rw, uio); } if (error != 0) break; } } } while (!eof && error == 0); vput(vp); ZFS_EXIT(zfsvfs); return (error); } int zfs_freebsd_getacl(ap) struct vop_getacl_args /* { struct vnode *vp; acl_type_t type; struct acl *aclp; struct ucred *cred; struct thread *td; } */ *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(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL)) 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); } int zfs_freebsd_setacl(ap) struct vop_setacl_args /* { struct vnode *vp; acl_type_t type; struct acl *aclp; struct ucred *cred; struct thread *td; } */ *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(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL); kmem_free(aaclp, aclbsize); return (error); } int zfs_freebsd_aclcheck(ap) struct vop_aclcheck_args /* { struct vnode *vp; acl_type_t type; struct acl *aclp; struct ucred *cred; struct thread *td; } */ *ap; { return (EOPNOTSUPP); } static int zfs_vptocnp(struct vop_vptocnp_args *ap) { vnode_t *covered_vp; vnode_t *vp = ap->a_vp;; zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; znode_t *zp = VTOZ(vp); int ltype; int error; ZFS_ENTER(zfsvfs); ZFS_VERIFY_ZP(zp); /* * If we are a snapshot mounted under .zfs, run the operation * on the covered vnode. */ if (zp->z_id != zfsvfs->z_root || zfsvfs->z_parent == zfsvfs) { char name[MAXNAMLEN + 1]; znode_t *dzp; size_t len; error = zfs_znode_parent_and_name(zp, &dzp, name); if (error == 0) { len = strlen(name); if (*ap->a_buflen < len) error = SET_ERROR(ENOMEM); } if (error == 0) { *ap->a_buflen -= len; bcopy(name, ap->a_buf + *ap->a_buflen, len); *ap->a_vpp = ZTOV(dzp); } ZFS_EXIT(zfsvfs); return (error); } ZFS_EXIT(zfsvfs); covered_vp = vp->v_mount->mnt_vnodecovered; vhold(covered_vp); ltype = VOP_ISLOCKED(vp); VOP_UNLOCK(vp, 0); error = vget(covered_vp, LK_SHARED | LK_VNHELD, curthread); if (error == 0) { error = VOP_VPTOCNP(covered_vp, ap->a_vpp, ap->a_cred, ap->a_buf, ap->a_buflen); vput(covered_vp); } vn_lock(vp, ltype | LK_RETRY); if ((vp->v_iflag & VI_DOOMED) != 0) error = SET_ERROR(ENOENT); return (error); } #ifdef DIAGNOSTIC static int zfs_lock(ap) struct vop_lock1_args /* { struct vnode *a_vp; int a_flags; char *file; int line; } */ *ap; { vnode_t *vp; znode_t *zp; int err; err = vop_stdlock(ap); if (err == 0 && (ap->a_flags & LK_NOWAIT) == 0) { vp = ap->a_vp; zp = vp->v_data; if (vp->v_mount != NULL && (vp->v_iflag & VI_DOOMED) == 0 && zp != NULL && (zp->z_pflags & ZFS_XATTR) == 0) VERIFY(!RRM_LOCK_HELD(&zp->z_zfsvfs->z_teardown_lock)); } return (err); } #endif struct vop_vector zfs_vnodeops; struct vop_vector zfs_fifoops; struct vop_vector zfs_shareops; struct vop_vector zfs_vnodeops = { .vop_default = &default_vnodeops, .vop_inactive = zfs_freebsd_inactive, .vop_reclaim = zfs_freebsd_reclaim, .vop_access = zfs_freebsd_access, .vop_allocate = VOP_EINVAL, .vop_lookup = zfs_cache_lookup, .vop_cachedlookup = zfs_freebsd_lookup, .vop_getattr = zfs_freebsd_getattr, .vop_setattr = zfs_freebsd_setattr, .vop_create = zfs_freebsd_create, .vop_mknod = 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, #ifdef DIAGNOSTIC .vop_lock1 = zfs_lock, #endif }; struct vop_vector zfs_fifoops = { .vop_default = &fifo_specops, .vop_fsync = zfs_freebsd_fsync, .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, }; /* * special share hidden files vnode operations template */ struct vop_vector zfs_shareops = { .vop_default = &default_vnodeops, .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, }; Index: head/sys/kern/vfs_cluster.c =================================================================== --- head/sys/kern/vfs_cluster.c (revision 351240) +++ head/sys/kern/vfs_cluster.c (revision 351241) @@ -1,1078 +1,1079 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1993 * The Regents of the University of California. All rights reserved. * Modifications/enhancements: * Copyright (c) 1995 John S. Dyson. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vfs_cluster.c 8.7 (Berkeley) 2/13/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_debug_cluster.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #if defined(CLUSTERDEBUG) static int rcluster= 0; SYSCTL_INT(_debug, OID_AUTO, rcluster, CTLFLAG_RW, &rcluster, 0, "Debug VFS clustering code"); #endif static MALLOC_DEFINE(M_SEGMENT, "cl_savebuf", "cluster_save buffer"); static uma_zone_t cluster_pbuf_zone; static void cluster_init(void *); static struct cluster_save *cluster_collectbufs(struct vnode *vp, struct buf *last_bp, int gbflags); static struct buf *cluster_rbuild(struct vnode *vp, u_quad_t filesize, daddr_t lbn, daddr_t blkno, long size, int run, int gbflags, struct buf *fbp); static void cluster_callback(struct buf *); static int write_behind = 1; SYSCTL_INT(_vfs, OID_AUTO, write_behind, CTLFLAG_RW, &write_behind, 0, "Cluster write-behind; 0: disable, 1: enable, 2: backed off"); static int read_max = 64; SYSCTL_INT(_vfs, OID_AUTO, read_max, CTLFLAG_RW, &read_max, 0, "Cluster read-ahead max block count"); static int read_min = 1; SYSCTL_INT(_vfs, OID_AUTO, read_min, CTLFLAG_RW, &read_min, 0, "Cluster read min block count"); SYSINIT(cluster, SI_SUB_CPU, SI_ORDER_ANY, cluster_init, NULL); static void cluster_init(void *dummy) { cluster_pbuf_zone = pbuf_zsecond_create("clpbuf", nswbuf / 2); } /* * Read data to a buf, including read-ahead if we find this to be beneficial. * cluster_read replaces bread. */ int cluster_read(struct vnode *vp, u_quad_t filesize, daddr_t lblkno, long size, struct ucred *cred, long totread, int seqcount, int gbflags, struct buf **bpp) { struct buf *bp, *rbp, *reqbp; struct bufobj *bo; struct thread *td; daddr_t blkno, origblkno; int maxra, racluster; int error, ncontig; int i; error = 0; td = curthread; bo = &vp->v_bufobj; if (!unmapped_buf_allowed) gbflags &= ~GB_UNMAPPED; /* * Try to limit the amount of read-ahead by a few * ad-hoc parameters. This needs work!!! */ racluster = vp->v_mount->mnt_iosize_max / size; maxra = seqcount; maxra = min(read_max, maxra); maxra = min(nbuf/8, maxra); if (((u_quad_t)(lblkno + maxra + 1) * size) > filesize) maxra = (filesize / size) - lblkno; /* * get the requested block */ error = getblkx(vp, lblkno, size, 0, 0, gbflags, &bp); if (error != 0) { *bpp = NULL; return (error); } gbflags &= ~GB_NOSPARSE; origblkno = lblkno; *bpp = reqbp = bp; /* * if it is in the cache, then check to see if the reads have been * sequential. If they have, then try some read-ahead, otherwise * back-off on prospective read-aheads. */ if (bp->b_flags & B_CACHE) { if (!seqcount) { return 0; } else if ((bp->b_flags & B_RAM) == 0) { return 0; } else { bp->b_flags &= ~B_RAM; BO_RLOCK(bo); for (i = 1; i < maxra; i++) { /* * Stop if the buffer does not exist or it * is invalid (about to go away?) */ rbp = gbincore(&vp->v_bufobj, lblkno+i); if (rbp == NULL || (rbp->b_flags & B_INVAL)) break; /* * Set another read-ahead mark so we know * to check again. (If we can lock the * buffer without waiting) */ if ((((i % racluster) == (racluster - 1)) || (i == (maxra - 1))) && (0 == BUF_LOCK(rbp, LK_EXCLUSIVE | LK_NOWAIT, NULL))) { rbp->b_flags |= B_RAM; BUF_UNLOCK(rbp); } } BO_RUNLOCK(bo); if (i >= maxra) { return 0; } lblkno += i; } reqbp = bp = NULL; /* * If it isn't in the cache, then get a chunk from * disk if sequential, otherwise just get the block. */ } else { off_t firstread = bp->b_offset; int nblks; long minread; KASSERT(bp->b_offset != NOOFFSET, ("cluster_read: no buffer offset")); ncontig = 0; /* * Adjust totread if needed */ minread = read_min * size; if (minread > totread) totread = minread; /* * Compute the total number of blocks that we should read * synchronously. */ if (firstread + totread > filesize) totread = filesize - firstread; nblks = howmany(totread, size); if (nblks > racluster) nblks = racluster; /* * Now compute the number of contiguous blocks. */ if (nblks > 1) { error = VOP_BMAP(vp, lblkno, NULL, &blkno, &ncontig, NULL); /* * If this failed to map just do the original block. */ if (error || blkno == -1) ncontig = 0; } /* * If we have contiguous data available do a cluster * otherwise just read the requested block. */ if (ncontig) { /* Account for our first block. */ ncontig = min(ncontig + 1, nblks); if (ncontig < nblks) nblks = ncontig; bp = cluster_rbuild(vp, filesize, lblkno, blkno, size, nblks, gbflags, bp); lblkno += (bp->b_bufsize / size); } else { bp->b_flags |= B_RAM; bp->b_iocmd = BIO_READ; lblkno += 1; } } /* * handle the synchronous read so that it is available ASAP. */ if (bp) { if ((bp->b_flags & B_CLUSTER) == 0) { vfs_busy_pages(bp, 0); } bp->b_flags &= ~B_INVAL; bp->b_ioflags &= ~BIO_ERROR; if ((bp->b_flags & B_ASYNC) || bp->b_iodone != NULL) BUF_KERNPROC(bp); bp->b_iooffset = dbtob(bp->b_blkno); bstrategy(bp); #ifdef RACCT if (racct_enable) { PROC_LOCK(td->td_proc); racct_add_buf(td->td_proc, bp, 0); PROC_UNLOCK(td->td_proc); } #endif /* RACCT */ td->td_ru.ru_inblock++; } /* * If we have been doing sequential I/O, then do some read-ahead. */ while (lblkno < (origblkno + maxra)) { error = VOP_BMAP(vp, lblkno, NULL, &blkno, &ncontig, NULL); if (error) break; if (blkno == -1) break; /* * We could throttle ncontig here by maxra but we might as * well read the data if it is contiguous. We're throttled * by racluster anyway. */ if (ncontig) { ncontig = min(ncontig + 1, racluster); rbp = cluster_rbuild(vp, filesize, lblkno, blkno, size, ncontig, gbflags, NULL); lblkno += (rbp->b_bufsize / size); if (rbp->b_flags & B_DELWRI) { bqrelse(rbp); continue; } } else { rbp = getblk(vp, lblkno, size, 0, 0, gbflags); lblkno += 1; if (rbp->b_flags & B_DELWRI) { bqrelse(rbp); continue; } rbp->b_flags |= B_ASYNC | B_RAM; rbp->b_iocmd = BIO_READ; rbp->b_blkno = blkno; } if (rbp->b_flags & B_CACHE) { rbp->b_flags &= ~B_ASYNC; bqrelse(rbp); continue; } if ((rbp->b_flags & B_CLUSTER) == 0) { vfs_busy_pages(rbp, 0); } rbp->b_flags &= ~B_INVAL; rbp->b_ioflags &= ~BIO_ERROR; if ((rbp->b_flags & B_ASYNC) || rbp->b_iodone != NULL) BUF_KERNPROC(rbp); rbp->b_iooffset = dbtob(rbp->b_blkno); bstrategy(rbp); #ifdef RACCT if (racct_enable) { PROC_LOCK(td->td_proc); racct_add_buf(td->td_proc, rbp, 0); PROC_UNLOCK(td->td_proc); } #endif /* RACCT */ td->td_ru.ru_inblock++; } if (reqbp) { /* * Like bread, always brelse() the buffer when * returning an error. */ error = bufwait(reqbp); if (error != 0) { brelse(reqbp); *bpp = NULL; } } return (error); } /* * If blocks are contiguous on disk, use this to provide clustered * read ahead. We will read as many blocks as possible sequentially * and then parcel them up into logical blocks in the buffer hash table. */ static struct buf * cluster_rbuild(struct vnode *vp, u_quad_t filesize, daddr_t lbn, daddr_t blkno, long size, int run, int gbflags, struct buf *fbp) { struct buf *bp, *tbp; daddr_t bn; off_t off; long tinc, tsize; int i, inc, j, k, toff; KASSERT(size == vp->v_mount->mnt_stat.f_iosize, ("cluster_rbuild: size %ld != f_iosize %jd\n", size, (intmax_t)vp->v_mount->mnt_stat.f_iosize)); /* * avoid a division */ while ((u_quad_t) size * (lbn + run) > filesize) { --run; } if (fbp) { tbp = fbp; tbp->b_iocmd = BIO_READ; } else { tbp = getblk(vp, lbn, size, 0, 0, gbflags); if (tbp->b_flags & B_CACHE) return tbp; tbp->b_flags |= B_ASYNC | B_RAM; tbp->b_iocmd = BIO_READ; } tbp->b_blkno = blkno; if( (tbp->b_flags & B_MALLOC) || ((tbp->b_flags & B_VMIO) == 0) || (run <= 1) ) return tbp; bp = uma_zalloc(cluster_pbuf_zone, M_NOWAIT); if (bp == NULL) return tbp; /* * We are synthesizing a buffer out of vm_page_t's, but * if the block size is not page aligned then the starting * address may not be either. Inherit the b_data offset * from the original buffer. */ bp->b_flags = B_ASYNC | B_CLUSTER | B_VMIO; if ((gbflags & GB_UNMAPPED) != 0) { bp->b_data = unmapped_buf; } else { bp->b_data = (char *)((vm_offset_t)bp->b_data | ((vm_offset_t)tbp->b_data & PAGE_MASK)); } bp->b_iocmd = BIO_READ; bp->b_iodone = cluster_callback; bp->b_blkno = blkno; bp->b_lblkno = lbn; bp->b_offset = tbp->b_offset; KASSERT(bp->b_offset != NOOFFSET, ("cluster_rbuild: no buffer offset")); pbgetvp(vp, bp); TAILQ_INIT(&bp->b_cluster.cluster_head); bp->b_bcount = 0; bp->b_bufsize = 0; bp->b_npages = 0; inc = btodb(size); for (bn = blkno, i = 0; i < run; ++i, bn += inc) { if (i == 0) { VM_OBJECT_WLOCK(tbp->b_bufobj->bo_object); vfs_drain_busy_pages(tbp); vm_object_pip_add(tbp->b_bufobj->bo_object, tbp->b_npages); for (k = 0; k < tbp->b_npages; k++) vm_page_sbusy(tbp->b_pages[k]); VM_OBJECT_WUNLOCK(tbp->b_bufobj->bo_object); } else { if ((bp->b_npages * PAGE_SIZE) + round_page(size) > vp->v_mount->mnt_iosize_max) { break; } tbp = getblk(vp, lbn + i, size, 0, 0, GB_LOCK_NOWAIT | (gbflags & GB_UNMAPPED)); /* Don't wait around for locked bufs. */ if (tbp == NULL) break; /* * Stop scanning if the buffer is fully valid * (marked B_CACHE), or locked (may be doing a * background write), or if the buffer is not * VMIO backed. The clustering code can only deal * with VMIO-backed buffers. The bo lock is not * required for the BKGRDINPROG check since it * can not be set without the buf lock. */ if ((tbp->b_vflags & BV_BKGRDINPROG) || (tbp->b_flags & B_CACHE) || (tbp->b_flags & B_VMIO) == 0) { bqrelse(tbp); break; } /* * The buffer must be completely invalid in order to * take part in the cluster. If it is partially valid * then we stop. */ off = tbp->b_offset; tsize = size; VM_OBJECT_WLOCK(tbp->b_bufobj->bo_object); for (j = 0; tsize > 0; j++) { toff = off & PAGE_MASK; tinc = tsize; if (toff + tinc > PAGE_SIZE) tinc = PAGE_SIZE - toff; VM_OBJECT_ASSERT_WLOCKED(tbp->b_pages[j]->object); if ((tbp->b_pages[j]->valid & vm_page_bits(toff, tinc)) != 0) break; if (vm_page_xbusied(tbp->b_pages[j])) break; vm_object_pip_add(tbp->b_bufobj->bo_object, 1); vm_page_sbusy(tbp->b_pages[j]); off += tinc; tsize -= tinc; } if (tsize > 0) { clean_sbusy: - vm_object_pip_add(tbp->b_bufobj->bo_object, -j); + vm_object_pip_wakeupn(tbp->b_bufobj->bo_object, + j); for (k = 0; k < j; k++) vm_page_sunbusy(tbp->b_pages[k]); VM_OBJECT_WUNLOCK(tbp->b_bufobj->bo_object); bqrelse(tbp); break; } VM_OBJECT_WUNLOCK(tbp->b_bufobj->bo_object); /* * Set a read-ahead mark as appropriate */ if ((fbp && (i == 1)) || (i == (run - 1))) tbp->b_flags |= B_RAM; /* * Set the buffer up for an async read (XXX should * we do this only if we do not wind up brelse()ing?). * Set the block number if it isn't set, otherwise * if it is make sure it matches the block number we * expect. */ tbp->b_flags |= B_ASYNC; tbp->b_iocmd = BIO_READ; if (tbp->b_blkno == tbp->b_lblkno) { tbp->b_blkno = bn; } else if (tbp->b_blkno != bn) { VM_OBJECT_WLOCK(tbp->b_bufobj->bo_object); goto clean_sbusy; } } /* * XXX fbp from caller may not be B_ASYNC, but we are going * to biodone() it in cluster_callback() anyway */ BUF_KERNPROC(tbp); TAILQ_INSERT_TAIL(&bp->b_cluster.cluster_head, tbp, b_cluster.cluster_entry); VM_OBJECT_WLOCK(tbp->b_bufobj->bo_object); for (j = 0; j < tbp->b_npages; j += 1) { vm_page_t m; m = tbp->b_pages[j]; if ((bp->b_npages == 0) || (bp->b_pages[bp->b_npages-1] != m)) { bp->b_pages[bp->b_npages] = m; bp->b_npages++; } if (m->valid == VM_PAGE_BITS_ALL) tbp->b_pages[j] = bogus_page; } VM_OBJECT_WUNLOCK(tbp->b_bufobj->bo_object); /* * Don't inherit tbp->b_bufsize as it may be larger due to * a non-page-aligned size. Instead just aggregate using * 'size'. */ if (tbp->b_bcount != size) printf("warning: tbp->b_bcount wrong %ld vs %ld\n", tbp->b_bcount, size); if (tbp->b_bufsize != size) printf("warning: tbp->b_bufsize wrong %ld vs %ld\n", tbp->b_bufsize, size); bp->b_bcount += size; bp->b_bufsize += size; } /* * Fully valid pages in the cluster are already good and do not need * to be re-read from disk. Replace the page with bogus_page */ VM_OBJECT_WLOCK(bp->b_bufobj->bo_object); for (j = 0; j < bp->b_npages; j++) { VM_OBJECT_ASSERT_WLOCKED(bp->b_pages[j]->object); if (bp->b_pages[j]->valid == VM_PAGE_BITS_ALL) bp->b_pages[j] = bogus_page; } VM_OBJECT_WUNLOCK(bp->b_bufobj->bo_object); if (bp->b_bufsize > bp->b_kvasize) panic("cluster_rbuild: b_bufsize(%ld) > b_kvasize(%d)\n", bp->b_bufsize, bp->b_kvasize); if (buf_mapped(bp)) { pmap_qenter(trunc_page((vm_offset_t) bp->b_data), (vm_page_t *)bp->b_pages, bp->b_npages); } return (bp); } /* * Cleanup after a clustered read or write. * This is complicated by the fact that any of the buffers might have * extra memory (if there were no empty buffer headers at allocbuf time) * that we will need to shift around. */ static void cluster_callback(struct buf *bp) { struct buf *nbp, *tbp; int error = 0; /* * Must propagate errors to all the components. */ if (bp->b_ioflags & BIO_ERROR) error = bp->b_error; if (buf_mapped(bp)) { pmap_qremove(trunc_page((vm_offset_t) bp->b_data), bp->b_npages); } /* * Move memory from the large cluster buffer into the component * buffers and mark IO as done on these. */ for (tbp = TAILQ_FIRST(&bp->b_cluster.cluster_head); tbp; tbp = nbp) { nbp = TAILQ_NEXT(&tbp->b_cluster, cluster_entry); if (error) { tbp->b_ioflags |= BIO_ERROR; tbp->b_error = error; } else { tbp->b_dirtyoff = tbp->b_dirtyend = 0; tbp->b_flags &= ~B_INVAL; tbp->b_ioflags &= ~BIO_ERROR; /* * XXX the bdwrite()/bqrelse() issued during * cluster building clears B_RELBUF (see bqrelse() * comment). If direct I/O was specified, we have * to restore it here to allow the buffer and VM * to be freed. */ if (tbp->b_flags & B_DIRECT) tbp->b_flags |= B_RELBUF; } bufdone(tbp); } pbrelvp(bp); uma_zfree(cluster_pbuf_zone, bp); } /* * cluster_wbuild_wb: * * Implement modified write build for cluster. * * write_behind = 0 write behind disabled * write_behind = 1 write behind normal (default) * write_behind = 2 write behind backed-off */ static __inline int cluster_wbuild_wb(struct vnode *vp, long size, daddr_t start_lbn, int len, int gbflags) { int r = 0; switch (write_behind) { case 2: if (start_lbn < len) break; start_lbn -= len; /* FALLTHROUGH */ case 1: r = cluster_wbuild(vp, size, start_lbn, len, gbflags); /* FALLTHROUGH */ default: /* FALLTHROUGH */ break; } return(r); } /* * Do clustered write for FFS. * * Three cases: * 1. Write is not sequential (write asynchronously) * Write is sequential: * 2. beginning of cluster - begin cluster * 3. middle of a cluster - add to cluster * 4. end of a cluster - asynchronously write cluster */ void cluster_write(struct vnode *vp, struct buf *bp, u_quad_t filesize, int seqcount, int gbflags) { daddr_t lbn; int maxclen, cursize; int lblocksize; int async; if (!unmapped_buf_allowed) gbflags &= ~GB_UNMAPPED; if (vp->v_type == VREG) { async = DOINGASYNC(vp); lblocksize = vp->v_mount->mnt_stat.f_iosize; } else { async = 0; lblocksize = bp->b_bufsize; } lbn = bp->b_lblkno; KASSERT(bp->b_offset != NOOFFSET, ("cluster_write: no buffer offset")); /* Initialize vnode to beginning of file. */ if (lbn == 0) vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0; if (vp->v_clen == 0 || lbn != vp->v_lastw + 1 || (bp->b_blkno != vp->v_lasta + btodb(lblocksize))) { maxclen = vp->v_mount->mnt_iosize_max / lblocksize - 1; if (vp->v_clen != 0) { /* * Next block is not sequential. * * If we are not writing at end of file, the process * seeked to another point in the file since its last * write, or we have reached our maximum cluster size, * then push the previous cluster. Otherwise try * reallocating to make it sequential. * * Change to algorithm: only push previous cluster if * it was sequential from the point of view of the * seqcount heuristic, otherwise leave the buffer * intact so we can potentially optimize the I/O * later on in the buf_daemon or update daemon * flush. */ cursize = vp->v_lastw - vp->v_cstart + 1; if (((u_quad_t) bp->b_offset + lblocksize) != filesize || lbn != vp->v_lastw + 1 || vp->v_clen <= cursize) { if (!async && seqcount > 0) { cluster_wbuild_wb(vp, lblocksize, vp->v_cstart, cursize, gbflags); } } else { struct buf **bpp, **endbp; struct cluster_save *buflist; buflist = cluster_collectbufs(vp, bp, gbflags); endbp = &buflist->bs_children [buflist->bs_nchildren - 1]; if (VOP_REALLOCBLKS(vp, buflist)) { /* * Failed, push the previous cluster * if *really* writing sequentially * in the logical file (seqcount > 1), * otherwise delay it in the hopes that * the low level disk driver can * optimize the write ordering. */ for (bpp = buflist->bs_children; bpp < endbp; bpp++) brelse(*bpp); free(buflist, M_SEGMENT); if (seqcount > 1) { cluster_wbuild_wb(vp, lblocksize, vp->v_cstart, cursize, gbflags); } } else { /* * Succeeded, keep building cluster. */ for (bpp = buflist->bs_children; bpp <= endbp; bpp++) bdwrite(*bpp); free(buflist, M_SEGMENT); vp->v_lastw = lbn; vp->v_lasta = bp->b_blkno; return; } } } /* * Consider beginning a cluster. If at end of file, make * cluster as large as possible, otherwise find size of * existing cluster. */ if ((vp->v_type == VREG) && ((u_quad_t) bp->b_offset + lblocksize) != filesize && (bp->b_blkno == bp->b_lblkno) && (VOP_BMAP(vp, lbn, NULL, &bp->b_blkno, &maxclen, NULL) || bp->b_blkno == -1)) { bawrite(bp); vp->v_clen = 0; vp->v_lasta = bp->b_blkno; vp->v_cstart = lbn + 1; vp->v_lastw = lbn; return; } vp->v_clen = maxclen; if (!async && maxclen == 0) { /* I/O not contiguous */ vp->v_cstart = lbn + 1; bawrite(bp); } else { /* Wait for rest of cluster */ vp->v_cstart = lbn; bdwrite(bp); } } else if (lbn == vp->v_cstart + vp->v_clen) { /* * At end of cluster, write it out if seqcount tells us we * are operating sequentially, otherwise let the buf or * update daemon handle it. */ bdwrite(bp); if (seqcount > 1) { cluster_wbuild_wb(vp, lblocksize, vp->v_cstart, vp->v_clen + 1, gbflags); } vp->v_clen = 0; vp->v_cstart = lbn + 1; } else if (vm_page_count_severe()) { /* * We are low on memory, get it going NOW */ bawrite(bp); } else { /* * In the middle of a cluster, so just delay the I/O for now. */ bdwrite(bp); } vp->v_lastw = lbn; vp->v_lasta = bp->b_blkno; } /* * This is an awful lot like cluster_rbuild...wish they could be combined. * The last lbn argument is the current block on which I/O is being * performed. Check to see that it doesn't fall in the middle of * the current block (if last_bp == NULL). */ int cluster_wbuild(struct vnode *vp, long size, daddr_t start_lbn, int len, int gbflags) { struct buf *bp, *tbp; struct bufobj *bo; int i, j; int totalwritten = 0; int dbsize = btodb(size); if (!unmapped_buf_allowed) gbflags &= ~GB_UNMAPPED; bo = &vp->v_bufobj; while (len > 0) { /* * If the buffer is not delayed-write (i.e. dirty), or it * is delayed-write but either locked or inval, it cannot * partake in the clustered write. */ BO_LOCK(bo); if ((tbp = gbincore(&vp->v_bufobj, start_lbn)) == NULL || (tbp->b_vflags & BV_BKGRDINPROG)) { BO_UNLOCK(bo); ++start_lbn; --len; continue; } if (BUF_LOCK(tbp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK, BO_LOCKPTR(bo))) { ++start_lbn; --len; continue; } if ((tbp->b_flags & (B_INVAL | B_DELWRI)) != B_DELWRI) { BUF_UNLOCK(tbp); ++start_lbn; --len; continue; } bremfree(tbp); tbp->b_flags &= ~B_DONE; /* * Extra memory in the buffer, punt on this buffer. * XXX we could handle this in most cases, but we would * have to push the extra memory down to after our max * possible cluster size and then potentially pull it back * up if the cluster was terminated prematurely--too much * hassle. */ if (((tbp->b_flags & (B_CLUSTEROK | B_MALLOC | B_VMIO)) != (B_CLUSTEROK | B_VMIO)) || (tbp->b_bcount != tbp->b_bufsize) || (tbp->b_bcount != size) || (len == 1) || ((bp = uma_zalloc(cluster_pbuf_zone, (vp->v_vflag & VV_MD) != 0 ? M_NOWAIT : M_WAITOK)) == NULL)) { totalwritten += tbp->b_bufsize; bawrite(tbp); ++start_lbn; --len; continue; } /* * We got a pbuf to make the cluster in. * so initialise it. */ TAILQ_INIT(&bp->b_cluster.cluster_head); bp->b_bcount = 0; bp->b_bufsize = 0; bp->b_npages = 0; if (tbp->b_wcred != NOCRED) bp->b_wcred = crhold(tbp->b_wcred); bp->b_blkno = tbp->b_blkno; bp->b_lblkno = tbp->b_lblkno; bp->b_offset = tbp->b_offset; /* * We are synthesizing a buffer out of vm_page_t's, but * if the block size is not page aligned then the starting * address may not be either. Inherit the b_data offset * from the original buffer. */ if ((gbflags & GB_UNMAPPED) == 0 || (tbp->b_flags & B_VMIO) == 0) { bp->b_data = (char *)((vm_offset_t)bp->b_data | ((vm_offset_t)tbp->b_data & PAGE_MASK)); } else { bp->b_data = unmapped_buf; } bp->b_flags |= B_CLUSTER | (tbp->b_flags & (B_VMIO | B_NEEDCOMMIT)); bp->b_iodone = cluster_callback; pbgetvp(vp, bp); /* * From this location in the file, scan forward to see * if there are buffers with adjacent data that need to * be written as well. */ for (i = 0; i < len; ++i, ++start_lbn) { if (i != 0) { /* If not the first buffer */ /* * If the adjacent data is not even in core it * can't need to be written. */ BO_LOCK(bo); if ((tbp = gbincore(bo, start_lbn)) == NULL || (tbp->b_vflags & BV_BKGRDINPROG)) { BO_UNLOCK(bo); break; } /* * If it IS in core, but has different * characteristics, or is locked (which * means it could be undergoing a background * I/O or be in a weird state), then don't * cluster with it. */ if (BUF_LOCK(tbp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK, BO_LOCKPTR(bo))) break; if ((tbp->b_flags & (B_VMIO | B_CLUSTEROK | B_INVAL | B_DELWRI | B_NEEDCOMMIT)) != (B_DELWRI | B_CLUSTEROK | (bp->b_flags & (B_VMIO | B_NEEDCOMMIT))) || tbp->b_wcred != bp->b_wcred) { BUF_UNLOCK(tbp); break; } /* * Check that the combined cluster * would make sense with regard to pages * and would not be too large */ if ((tbp->b_bcount != size) || ((bp->b_blkno + (dbsize * i)) != tbp->b_blkno) || ((tbp->b_npages + bp->b_npages) > (vp->v_mount->mnt_iosize_max / PAGE_SIZE))) { BUF_UNLOCK(tbp); break; } /* * Ok, it's passed all the tests, * so remove it from the free list * and mark it busy. We will use it. */ bremfree(tbp); tbp->b_flags &= ~B_DONE; } /* end of code for non-first buffers only */ /* * If the IO is via the VM then we do some * special VM hackery (yuck). Since the buffer's * block size may not be page-aligned it is possible * for a page to be shared between two buffers. We * have to get rid of the duplication when building * the cluster. */ if (tbp->b_flags & B_VMIO) { vm_page_t m; VM_OBJECT_WLOCK(tbp->b_bufobj->bo_object); if (i == 0) { vfs_drain_busy_pages(tbp); } else { /* if not first buffer */ for (j = 0; j < tbp->b_npages; j += 1) { m = tbp->b_pages[j]; if (vm_page_xbusied(m)) { VM_OBJECT_WUNLOCK( tbp->b_object); bqrelse(tbp); goto finishcluster; } } } for (j = 0; j < tbp->b_npages; j += 1) { m = tbp->b_pages[j]; vm_page_sbusy(m); vm_object_pip_add(m->object, 1); if ((bp->b_npages == 0) || (bp->b_pages[bp->b_npages - 1] != m)) { bp->b_pages[bp->b_npages] = m; bp->b_npages++; } } VM_OBJECT_WUNLOCK(tbp->b_bufobj->bo_object); } bp->b_bcount += size; bp->b_bufsize += size; /* * If any of the clustered buffers have their * B_BARRIER flag set, transfer that request to * the cluster. */ bp->b_flags |= (tbp->b_flags & B_BARRIER); tbp->b_flags &= ~(B_DONE | B_BARRIER); tbp->b_flags |= B_ASYNC; tbp->b_ioflags &= ~BIO_ERROR; tbp->b_iocmd = BIO_WRITE; bundirty(tbp); reassignbuf(tbp); /* put on clean list */ bufobj_wref(tbp->b_bufobj); BUF_KERNPROC(tbp); buf_track(tbp, __func__); TAILQ_INSERT_TAIL(&bp->b_cluster.cluster_head, tbp, b_cluster.cluster_entry); } finishcluster: if (buf_mapped(bp)) { pmap_qenter(trunc_page((vm_offset_t) bp->b_data), (vm_page_t *)bp->b_pages, bp->b_npages); } if (bp->b_bufsize > bp->b_kvasize) panic( "cluster_wbuild: b_bufsize(%ld) > b_kvasize(%d)\n", bp->b_bufsize, bp->b_kvasize); totalwritten += bp->b_bufsize; bp->b_dirtyoff = 0; bp->b_dirtyend = bp->b_bufsize; bawrite(bp); len -= i; } return totalwritten; } /* * Collect together all the buffers in a cluster. * Plus add one additional buffer. */ static struct cluster_save * cluster_collectbufs(struct vnode *vp, struct buf *last_bp, int gbflags) { struct cluster_save *buflist; struct buf *bp; daddr_t lbn; int i, len; len = vp->v_lastw - vp->v_cstart + 1; buflist = malloc(sizeof(struct buf *) * (len + 1) + sizeof(*buflist), M_SEGMENT, M_WAITOK); buflist->bs_nchildren = 0; buflist->bs_children = (struct buf **) (buflist + 1); for (lbn = vp->v_cstart, i = 0; i < len; lbn++, i++) { (void)bread_gb(vp, lbn, last_bp->b_bcount, NOCRED, gbflags, &bp); buflist->bs_children[i] = bp; if (bp->b_blkno == bp->b_lblkno) VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, NULL, NULL); } buflist->bs_children[i] = bp = last_bp; if (bp->b_blkno == bp->b_lblkno) VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, NULL, NULL); buflist->bs_nchildren = i + 1; return (buflist); } Index: head/sys/kern/vfs_subr.c =================================================================== --- head/sys/kern/vfs_subr.c (revision 351240) +++ head/sys/kern/vfs_subr.c (revision 351241) @@ -1,5711 +1,5707 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95 */ /* * External virtual filesystem routines */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include "opt_watchdog.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef DDB #include #endif static void delmntque(struct vnode *vp); static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag, int slptimeo); static void syncer_shutdown(void *arg, int howto); static int vtryrecycle(struct vnode *vp); static void v_init_counters(struct vnode *); static void v_incr_usecount(struct vnode *); static void v_incr_usecount_locked(struct vnode *); static void v_incr_devcount(struct vnode *); static void v_decr_devcount(struct vnode *); static void vgonel(struct vnode *); static void vfs_knllock(void *arg); static void vfs_knlunlock(void *arg); static void vfs_knl_assert_locked(void *arg); static void vfs_knl_assert_unlocked(void *arg); static void vnlru_return_batches(struct vfsops *mnt_op); static void destroy_vpollinfo(struct vpollinfo *vi); static int v_inval_buf_range_locked(struct vnode *vp, struct bufobj *bo, daddr_t startlbn, daddr_t endlbn); /* * These fences are intended for cases where some synchronization is * needed between access of v_iflags and lockless vnode refcount (v_holdcnt * and v_usecount) updates. Access to v_iflags is generally synchronized * by the interlock, but we have some internal assertions that check vnode * flags without acquiring the lock. Thus, these fences are INVARIANTS-only * for now. */ #ifdef INVARIANTS #define VNODE_REFCOUNT_FENCE_ACQ() atomic_thread_fence_acq() #define VNODE_REFCOUNT_FENCE_REL() atomic_thread_fence_rel() #else #define VNODE_REFCOUNT_FENCE_ACQ() #define VNODE_REFCOUNT_FENCE_REL() #endif /* * Number of vnodes in existence. Increased whenever getnewvnode() * allocates a new vnode, decreased in vdropl() for VI_DOOMED vnode. */ static unsigned long numvnodes; SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "Number of vnodes in existence"); static counter_u64_t vnodes_created; SYSCTL_COUNTER_U64(_vfs, OID_AUTO, vnodes_created, CTLFLAG_RD, &vnodes_created, "Number of vnodes created by getnewvnode"); static u_long mnt_free_list_batch = 128; SYSCTL_ULONG(_vfs, OID_AUTO, mnt_free_list_batch, CTLFLAG_RW, &mnt_free_list_batch, 0, "Limit of vnodes held on mnt's free list"); /* * Conversion tables for conversion from vnode types to inode formats * and back. */ enum vtype iftovt_tab[16] = { VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VNON }; int vttoif_tab[10] = { 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT }; /* * List of vnodes that are ready for recycling. */ static TAILQ_HEAD(freelst, vnode) vnode_free_list; /* * "Free" vnode target. Free vnodes are rarely completely free, but are * just ones that are cheap to recycle. Usually they are for files which * have been stat'd but not read; these usually have inode and namecache * data attached to them. This target is the preferred minimum size of a * sub-cache consisting mostly of such files. The system balances the size * of this sub-cache with its complement to try to prevent either from * thrashing while the other is relatively inactive. The targets express * a preference for the best balance. * * "Above" this target there are 2 further targets (watermarks) related * to recyling of free vnodes. In the best-operating case, the cache is * exactly full, the free list has size between vlowat and vhiwat above the * free target, and recycling from it and normal use maintains this state. * Sometimes the free list is below vlowat or even empty, but this state * is even better for immediate use provided the cache is not full. * Otherwise, vnlru_proc() runs to reclaim enough vnodes (usually non-free * ones) to reach one of these states. The watermarks are currently hard- * coded as 4% and 9% of the available space higher. These and the default * of 25% for wantfreevnodes are too large if the memory size is large. * E.g., 9% of 75% of MAXVNODES is more than 566000 vnodes to reclaim * whenever vnlru_proc() becomes active. */ static u_long wantfreevnodes; SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "Target for minimum number of \"free\" vnodes"); static u_long freevnodes; SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "Number of \"free\" vnodes"); static counter_u64_t recycles_count; SYSCTL_COUNTER_U64(_vfs, OID_AUTO, recycles, CTLFLAG_RD, &recycles_count, "Number of vnodes recycled to meet vnode cache targets"); /* * Various variables used for debugging the new implementation of * reassignbuf(). * XXX these are probably of (very) limited utility now. */ static int reassignbufcalls; SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "Number of calls to reassignbuf"); static counter_u64_t free_owe_inact; SYSCTL_COUNTER_U64(_vfs, OID_AUTO, free_owe_inact, CTLFLAG_RD, &free_owe_inact, "Number of times free vnodes kept on active list due to VFS " "owing inactivation"); /* To keep more than one thread at a time from running vfs_getnewfsid */ static struct mtx mntid_mtx; /* * Lock for any access to the following: * vnode_free_list * numvnodes * freevnodes */ static struct mtx vnode_free_list_mtx; /* Publicly exported FS */ struct nfs_public nfs_pub; static uma_zone_t buf_trie_zone; /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */ static uma_zone_t vnode_zone; static uma_zone_t vnodepoll_zone; /* * The workitem queue. * * It is useful to delay writes of file data and filesystem metadata * for tens of seconds so that quickly created and deleted files need * not waste disk bandwidth being created and removed. To realize this, * we append vnodes to a "workitem" queue. When running with a soft * updates implementation, most pending metadata dependencies should * not wait for more than a few seconds. Thus, mounted on block devices * are delayed only about a half the time that file data is delayed. * Similarly, directory updates are more critical, so are only delayed * about a third the time that file data is delayed. Thus, there are * SYNCER_MAXDELAY queues that are processed round-robin at a rate of * one each second (driven off the filesystem syncer process). The * syncer_delayno variable indicates the next queue that is to be processed. * Items that need to be processed soon are placed in this queue: * * syncer_workitem_pending[syncer_delayno] * * A delay of fifteen seconds is done by placing the request fifteen * entries later in the queue: * * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask] * */ static int syncer_delayno; static long syncer_mask; LIST_HEAD(synclist, bufobj); static struct synclist *syncer_workitem_pending; /* * The sync_mtx protects: * bo->bo_synclist * sync_vnode_count * syncer_delayno * syncer_state * syncer_workitem_pending * syncer_worklist_len * rushjob */ static struct mtx sync_mtx; static struct cv sync_wakeup; #define SYNCER_MAXDELAY 32 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */ static int syncdelay = 30; /* max time to delay syncing data */ static int filedelay = 30; /* time to delay syncing files */ SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "Time to delay syncing files (in seconds)"); static int dirdelay = 29; /* time to delay syncing directories */ SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "Time to delay syncing directories (in seconds)"); static int metadelay = 28; /* time to delay syncing metadata */ SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "Time to delay syncing metadata (in seconds)"); static int rushjob; /* number of slots to run ASAP */ static int stat_rush_requests; /* number of times I/O speeded up */ SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "Number of times I/O speeded up (rush requests)"); /* * When shutting down the syncer, run it at four times normal speed. */ #define SYNCER_SHUTDOWN_SPEEDUP 4 static int sync_vnode_count; static int syncer_worklist_len; static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY } syncer_state; /* Target for maximum number of vnodes. */ int desiredvnodes; static int gapvnodes; /* gap between wanted and desired */ static int vhiwat; /* enough extras after expansion */ static int vlowat; /* minimal extras before expansion */ static int vstir; /* nonzero to stir non-free vnodes */ static volatile int vsmalltrigger = 8; /* pref to keep if > this many pages */ static int sysctl_update_desiredvnodes(SYSCTL_HANDLER_ARGS) { int error, old_desiredvnodes; old_desiredvnodes = desiredvnodes; if ((error = sysctl_handle_int(oidp, arg1, arg2, req)) != 0) return (error); if (old_desiredvnodes != desiredvnodes) { wantfreevnodes = desiredvnodes / 4; /* XXX locking seems to be incomplete. */ vfs_hash_changesize(desiredvnodes); cache_changesize(desiredvnodes); } return (0); } SYSCTL_PROC(_kern, KERN_MAXVNODES, maxvnodes, CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, &desiredvnodes, 0, sysctl_update_desiredvnodes, "I", "Target for maximum number of vnodes"); SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW, &wantfreevnodes, 0, "Old name for vfs.wantfreevnodes (legacy)"); static int vnlru_nowhere; SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0, "Number of times the vnlru process ran without success"); static int sysctl_try_reclaim_vnode(SYSCTL_HANDLER_ARGS) { struct vnode *vp; struct nameidata nd; char *buf; unsigned long ndflags; int error; if (req->newptr == NULL) return (EINVAL); if (req->newlen > PATH_MAX) return (E2BIG); buf = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); error = SYSCTL_IN(req, buf, req->newlen); if (error != 0) goto out; buf[req->newlen] = '\0'; ndflags = LOCKLEAF | NOFOLLOW | AUDITVNODE1 | NOCACHE | SAVENAME; NDINIT(&nd, LOOKUP, ndflags, UIO_SYSSPACE, buf, curthread); if ((error = namei(&nd)) != 0) goto out; vp = nd.ni_vp; if ((vp->v_iflag & VI_DOOMED) != 0) { /* * This vnode is being recycled. Return != 0 to let the caller * know that the sysctl had no effect. Return EAGAIN because a * subsequent call will likely succeed (since namei will create * a new vnode if necessary) */ error = EAGAIN; goto putvnode; } counter_u64_add(recycles_count, 1); vgone(vp); putvnode: NDFREE(&nd, 0); out: free(buf, M_TEMP); return (error); } static int sysctl_ftry_reclaim_vnode(SYSCTL_HANDLER_ARGS) { struct thread *td = curthread; struct vnode *vp; struct file *fp; int error; int fd; if (req->newptr == NULL) return (EBADF); error = sysctl_handle_int(oidp, &fd, 0, req); if (error != 0) return (error); error = getvnode(curthread, fd, &cap_fcntl_rights, &fp); if (error != 0) return (error); vp = fp->f_vnode; error = vn_lock(vp, LK_EXCLUSIVE); if (error != 0) goto drop; counter_u64_add(recycles_count, 1); vgone(vp); VOP_UNLOCK(vp, 0); drop: fdrop(fp, td); return (error); } SYSCTL_PROC(_debug, OID_AUTO, try_reclaim_vnode, CTLTYPE_STRING | CTLFLAG_MPSAFE | CTLFLAG_WR, NULL, 0, sysctl_try_reclaim_vnode, "A", "Try to reclaim a vnode by its pathname"); SYSCTL_PROC(_debug, OID_AUTO, ftry_reclaim_vnode, CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_WR, NULL, 0, sysctl_ftry_reclaim_vnode, "I", "Try to reclaim a vnode by its file descriptor"); /* Shift count for (uintptr_t)vp to initialize vp->v_hash. */ static int vnsz2log; /* * Support for the bufobj clean & dirty pctrie. */ static void * buf_trie_alloc(struct pctrie *ptree) { return uma_zalloc(buf_trie_zone, M_NOWAIT); } static void buf_trie_free(struct pctrie *ptree, void *node) { uma_zfree(buf_trie_zone, node); } PCTRIE_DEFINE(BUF, buf, b_lblkno, buf_trie_alloc, buf_trie_free); /* * Initialize the vnode management data structures. * * Reevaluate the following cap on the number of vnodes after the physical * memory size exceeds 512GB. In the limit, as the physical memory size * grows, the ratio of the memory size in KB to vnodes approaches 64:1. */ #ifndef MAXVNODES_MAX #define MAXVNODES_MAX (512 * 1024 * 1024 / 64) /* 8M */ #endif /* * Initialize a vnode as it first enters the zone. */ static int vnode_init(void *mem, int size, int flags) { struct vnode *vp; vp = mem; bzero(vp, size); /* * Setup locks. */ vp->v_vnlock = &vp->v_lock; mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF); /* * By default, don't allow shared locks unless filesystems opt-in. */ lockinit(vp->v_vnlock, PVFS, "vnode", VLKTIMEOUT, LK_NOSHARE | LK_IS_VNODE); /* * Initialize bufobj. */ bufobj_init(&vp->v_bufobj, vp); /* * Initialize namecache. */ LIST_INIT(&vp->v_cache_src); TAILQ_INIT(&vp->v_cache_dst); /* * Initialize rangelocks. */ rangelock_init(&vp->v_rl); return (0); } /* * Free a vnode when it is cleared from the zone. */ static void vnode_fini(void *mem, int size) { struct vnode *vp; struct bufobj *bo; vp = mem; rangelock_destroy(&vp->v_rl); lockdestroy(vp->v_vnlock); mtx_destroy(&vp->v_interlock); bo = &vp->v_bufobj; rw_destroy(BO_LOCKPTR(bo)); } /* * Provide the size of NFS nclnode and NFS fh for calculation of the * vnode memory consumption. The size is specified directly to * eliminate dependency on NFS-private header. * * Other filesystems may use bigger or smaller (like UFS and ZFS) * private inode data, but the NFS-based estimation is ample enough. * Still, we care about differences in the size between 64- and 32-bit * platforms. * * Namecache structure size is heuristically * sizeof(struct namecache_ts) + CACHE_PATH_CUTOFF + 1. */ #ifdef _LP64 #define NFS_NCLNODE_SZ (528 + 64) #define NC_SZ 148 #else #define NFS_NCLNODE_SZ (360 + 32) #define NC_SZ 92 #endif static void vntblinit(void *dummy __unused) { u_int i; int physvnodes, virtvnodes; /* * Desiredvnodes is a function of the physical memory size and the * kernel's heap size. Generally speaking, it scales with the * physical memory size. The ratio of desiredvnodes to the physical * memory size is 1:16 until desiredvnodes exceeds 98,304. * Thereafter, the * marginal ratio of desiredvnodes to the physical memory size is * 1:64. However, desiredvnodes is limited by the kernel's heap * size. The memory required by desiredvnodes vnodes and vm objects * must not exceed 1/10th of the kernel's heap size. */ physvnodes = maxproc + pgtok(vm_cnt.v_page_count) / 64 + 3 * min(98304 * 16, pgtok(vm_cnt.v_page_count)) / 64; virtvnodes = vm_kmem_size / (10 * (sizeof(struct vm_object) + sizeof(struct vnode) + NC_SZ * ncsizefactor + NFS_NCLNODE_SZ)); desiredvnodes = min(physvnodes, virtvnodes); if (desiredvnodes > MAXVNODES_MAX) { if (bootverbose) printf("Reducing kern.maxvnodes %d -> %d\n", desiredvnodes, MAXVNODES_MAX); desiredvnodes = MAXVNODES_MAX; } wantfreevnodes = desiredvnodes / 4; mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF); TAILQ_INIT(&vnode_free_list); mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF); vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL, vnode_init, vnode_fini, UMA_ALIGN_PTR, 0); vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); /* * Preallocate enough nodes to support one-per buf so that * we can not fail an insert. reassignbuf() callers can not * tolerate the insertion failure. */ buf_trie_zone = uma_zcreate("BUF TRIE", pctrie_node_size(), NULL, NULL, pctrie_zone_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE | UMA_ZONE_VM); uma_prealloc(buf_trie_zone, nbuf); vnodes_created = counter_u64_alloc(M_WAITOK); recycles_count = counter_u64_alloc(M_WAITOK); free_owe_inact = counter_u64_alloc(M_WAITOK); /* * Initialize the filesystem syncer. */ syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE, &syncer_mask); syncer_maxdelay = syncer_mask + 1; mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF); cv_init(&sync_wakeup, "syncer"); for (i = 1; i <= sizeof(struct vnode); i <<= 1) vnsz2log++; vnsz2log--; } SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL); /* * Mark a mount point as busy. Used to synchronize access and to delay * unmounting. Eventually, mountlist_mtx is not released on failure. * * vfs_busy() is a custom lock, it can block the caller. * vfs_busy() only sleeps if the unmount is active on the mount point. * For a mountpoint mp, vfs_busy-enforced lock is before lock of any * vnode belonging to mp. * * Lookup uses vfs_busy() to traverse mount points. * root fs var fs * / vnode lock A / vnode lock (/var) D * /var vnode lock B /log vnode lock(/var/log) E * vfs_busy lock C vfs_busy lock F * * Within each file system, the lock order is C->A->B and F->D->E. * * When traversing across mounts, the system follows that lock order: * * C->A->B * | * +->F->D->E * * The lookup() process for namei("/var") illustrates the process: * VOP_LOOKUP() obtains B while A is held * vfs_busy() obtains a shared lock on F while A and B are held * vput() releases lock on B * vput() releases lock on A * VFS_ROOT() obtains lock on D while shared lock on F is held * vfs_unbusy() releases shared lock on F * vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A. * Attempt to lock A (instead of vp_crossmp) while D is held would * violate the global order, causing deadlocks. * * dounmount() locks B while F is drained. */ int vfs_busy(struct mount *mp, int flags) { MPASS((flags & ~MBF_MASK) == 0); CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags); MNT_ILOCK(mp); MNT_REF(mp); /* * If mount point is currently being unmounted, sleep until the * mount point fate is decided. If thread doing the unmounting fails, * it will clear MNTK_UNMOUNT flag before waking us up, indicating * that this mount point has survived the unmount attempt and vfs_busy * should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE * flag in addition to MNTK_UNMOUNT, indicating that mount point is * about to be really destroyed. vfs_busy needs to release its * reference on the mount point in this case and return with ENOENT, * telling the caller that mount mount it tried to busy is no longer * valid. */ while (mp->mnt_kern_flag & MNTK_UNMOUNT) { if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) { MNT_REL(mp); MNT_IUNLOCK(mp); CTR1(KTR_VFS, "%s: failed busying before sleeping", __func__); return (ENOENT); } if (flags & MBF_MNTLSTLOCK) mtx_unlock(&mountlist_mtx); mp->mnt_kern_flag |= MNTK_MWAIT; msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0); if (flags & MBF_MNTLSTLOCK) mtx_lock(&mountlist_mtx); MNT_ILOCK(mp); } if (flags & MBF_MNTLSTLOCK) mtx_unlock(&mountlist_mtx); mp->mnt_lockref++; MNT_IUNLOCK(mp); return (0); } /* * Free a busy filesystem. */ void vfs_unbusy(struct mount *mp) { CTR2(KTR_VFS, "%s: mp %p", __func__, mp); MNT_ILOCK(mp); MNT_REL(mp); KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref")); mp->mnt_lockref--; if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) { MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT); CTR1(KTR_VFS, "%s: waking up waiters", __func__); mp->mnt_kern_flag &= ~MNTK_DRAINING; wakeup(&mp->mnt_lockref); } MNT_IUNLOCK(mp); } /* * Lookup a mount point by filesystem identifier. */ struct mount * vfs_getvfs(fsid_t *fsid) { struct mount *mp; CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid); mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { vfs_ref(mp); mtx_unlock(&mountlist_mtx); return (mp); } } mtx_unlock(&mountlist_mtx); CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid); return ((struct mount *) 0); } /* * Lookup a mount point by filesystem identifier, busying it before * returning. * * To avoid congestion on mountlist_mtx, implement simple direct-mapped * cache for popular filesystem identifiers. The cache is lockess, using * the fact that struct mount's are never freed. In worst case we may * get pointer to unmounted or even different filesystem, so we have to * check what we got, and go slow way if so. */ struct mount * vfs_busyfs(fsid_t *fsid) { #define FSID_CACHE_SIZE 256 typedef struct mount * volatile vmp_t; static vmp_t cache[FSID_CACHE_SIZE]; struct mount *mp; int error; uint32_t hash; CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid); hash = fsid->val[0] ^ fsid->val[1]; hash = (hash >> 16 ^ hash) & (FSID_CACHE_SIZE - 1); mp = cache[hash]; if (mp == NULL || mp->mnt_stat.f_fsid.val[0] != fsid->val[0] || mp->mnt_stat.f_fsid.val[1] != fsid->val[1]) goto slow; if (vfs_busy(mp, 0) != 0) { cache[hash] = NULL; goto slow; } if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) return (mp); else vfs_unbusy(mp); slow: mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { error = vfs_busy(mp, MBF_MNTLSTLOCK); if (error) { cache[hash] = NULL; mtx_unlock(&mountlist_mtx); return (NULL); } cache[hash] = mp; return (mp); } } CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid); mtx_unlock(&mountlist_mtx); return ((struct mount *) 0); } /* * Check if a user can access privileged mount options. */ int vfs_suser(struct mount *mp, struct thread *td) { int error; if (jailed(td->td_ucred)) { /* * If the jail of the calling thread lacks permission for * this type of file system, deny immediately. */ if (!prison_allow(td->td_ucred, mp->mnt_vfc->vfc_prison_flag)) return (EPERM); /* * If the file system was mounted outside the jail of the * calling thread, deny immediately. */ if (prison_check(td->td_ucred, mp->mnt_cred) != 0) return (EPERM); } /* * If file system supports delegated administration, we don't check * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified * by the file system itself. * If this is not the user that did original mount, we check for * the PRIV_VFS_MOUNT_OWNER privilege. */ if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) && mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) { if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0) return (error); } return (0); } /* * Get a new unique fsid. Try to make its val[0] unique, since this value * will be used to create fake device numbers for stat(). Also try (but * not so hard) make its val[0] unique mod 2^16, since some emulators only * support 16-bit device numbers. We end up with unique val[0]'s for the * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls. * * Keep in mind that several mounts may be running in parallel. Starting * the search one past where the previous search terminated is both a * micro-optimization and a defense against returning the same fsid to * different mounts. */ void vfs_getnewfsid(struct mount *mp) { static uint16_t mntid_base; struct mount *nmp; fsid_t tfsid; int mtype; CTR2(KTR_VFS, "%s: mp %p", __func__, mp); mtx_lock(&mntid_mtx); mtype = mp->mnt_vfc->vfc_typenum; tfsid.val[1] = mtype; mtype = (mtype & 0xFF) << 24; for (;;) { tfsid.val[0] = makedev(255, mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF)); mntid_base++; if ((nmp = vfs_getvfs(&tfsid)) == NULL) break; vfs_rel(nmp); } mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; mp->mnt_stat.f_fsid.val[1] = tfsid.val[1]; mtx_unlock(&mntid_mtx); } /* * Knob to control the precision of file timestamps: * * 0 = seconds only; nanoseconds zeroed. * 1 = seconds and nanoseconds, accurate within 1/HZ. * 2 = seconds and nanoseconds, truncated to microseconds. * >=3 = seconds and nanoseconds, maximum precision. */ enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; static int timestamp_precision = TSP_USEC; SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, ×tamp_precision, 0, "File timestamp precision (0: seconds, " "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to us, " "3+: sec + ns (max. precision))"); /* * Get a current timestamp. */ void vfs_timestamp(struct timespec *tsp) { struct timeval tv; switch (timestamp_precision) { case TSP_SEC: tsp->tv_sec = time_second; tsp->tv_nsec = 0; break; case TSP_HZ: getnanotime(tsp); break; case TSP_USEC: microtime(&tv); TIMEVAL_TO_TIMESPEC(&tv, tsp); break; case TSP_NSEC: default: nanotime(tsp); break; } } /* * Set vnode attributes to VNOVAL */ void vattr_null(struct vattr *vap) { vap->va_type = VNON; vap->va_size = VNOVAL; vap->va_bytes = VNOVAL; vap->va_mode = VNOVAL; vap->va_nlink = VNOVAL; vap->va_uid = VNOVAL; vap->va_gid = VNOVAL; vap->va_fsid = VNOVAL; vap->va_fileid = VNOVAL; vap->va_blocksize = VNOVAL; vap->va_rdev = VNOVAL; vap->va_atime.tv_sec = VNOVAL; vap->va_atime.tv_nsec = VNOVAL; vap->va_mtime.tv_sec = VNOVAL; vap->va_mtime.tv_nsec = VNOVAL; vap->va_ctime.tv_sec = VNOVAL; vap->va_ctime.tv_nsec = VNOVAL; vap->va_birthtime.tv_sec = VNOVAL; vap->va_birthtime.tv_nsec = VNOVAL; vap->va_flags = VNOVAL; vap->va_gen = VNOVAL; vap->va_vaflags = 0; } /* * This routine is called when we have too many vnodes. It attempts * to free vnodes and will potentially free vnodes that still * have VM backing store (VM backing store is typically the cause * of a vnode blowout so we want to do this). Therefore, this operation * is not considered cheap. * * A number of conditions may prevent a vnode from being reclaimed. * the buffer cache may have references on the vnode, a directory * vnode may still have references due to the namei cache representing * underlying files, or the vnode may be in active use. It is not * desirable to reuse such vnodes. These conditions may cause the * number of vnodes to reach some minimum value regardless of what * you set kern.maxvnodes to. Do not set kern.maxvnodes too low. * * @param mp Try to reclaim vnodes from this mountpoint * @param reclaim_nc_src Only reclaim directories with outgoing namecache * entries if this argument is strue * @param trigger Only reclaim vnodes with fewer than this many resident * pages. * @return The number of vnodes that were reclaimed. */ static int vlrureclaim(struct mount *mp, bool reclaim_nc_src, int trigger) { struct vnode *vp; int count, done, target; done = 0; vn_start_write(NULL, &mp, V_WAIT); MNT_ILOCK(mp); count = mp->mnt_nvnodelistsize; target = count * (int64_t)gapvnodes / imax(desiredvnodes, 1); target = target / 10 + 1; while (count != 0 && done < target) { vp = TAILQ_FIRST(&mp->mnt_nvnodelist); while (vp != NULL && vp->v_type == VMARKER) vp = TAILQ_NEXT(vp, v_nmntvnodes); if (vp == NULL) break; /* * XXX LRU is completely broken for non-free vnodes. First * by calling here in mountpoint order, then by moving * unselected vnodes to the end here, and most grossly by * removing the vlruvp() function that was supposed to * maintain the order. (This function was born broken * since syncer problems prevented it doing anything.) The * order is closer to LRC (C = Created). * * LRU reclaiming of vnodes seems to have last worked in * FreeBSD-3 where LRU wasn't mentioned under any spelling. * Then there was no hold count, and inactive vnodes were * simply put on the free list in LRU order. The separate * lists also break LRU. We prefer to reclaim from the * free list for technical reasons. This tends to thrash * the free list to keep very unrecently used held vnodes. * The problem is mitigated by keeping the free list large. */ TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); --count; if (!VI_TRYLOCK(vp)) goto next_iter; /* * If it's been deconstructed already, it's still * referenced, or it exceeds the trigger, skip it. * Also skip free vnodes. We are trying to make space * to expand the free list, not reduce it. */ if (vp->v_usecount || (!reclaim_nc_src && !LIST_EMPTY(&vp->v_cache_src)) || ((vp->v_iflag & VI_FREE) != 0) || (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL && vp->v_object->resident_page_count > trigger)) { VI_UNLOCK(vp); goto next_iter; } MNT_IUNLOCK(mp); vholdl(vp); if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) { vdrop(vp); goto next_iter_mntunlocked; } VI_LOCK(vp); /* * v_usecount may have been bumped after VOP_LOCK() dropped * the vnode interlock and before it was locked again. * * It is not necessary to recheck VI_DOOMED because it can * only be set by another thread that holds both the vnode * lock and vnode interlock. If another thread has the * vnode lock before we get to VOP_LOCK() and obtains the * vnode interlock after VOP_LOCK() drops the vnode * interlock, the other thread will be unable to drop the * vnode lock before our VOP_LOCK() call fails. */ if (vp->v_usecount || (!reclaim_nc_src && !LIST_EMPTY(&vp->v_cache_src)) || (vp->v_iflag & VI_FREE) != 0 || (vp->v_object != NULL && vp->v_object->resident_page_count > trigger)) { VOP_UNLOCK(vp, LK_INTERLOCK); vdrop(vp); goto next_iter_mntunlocked; } KASSERT((vp->v_iflag & VI_DOOMED) == 0, ("VI_DOOMED unexpectedly detected in vlrureclaim()")); counter_u64_add(recycles_count, 1); vgonel(vp); VOP_UNLOCK(vp, 0); vdropl(vp); done++; next_iter_mntunlocked: if (!should_yield()) goto relock_mnt; goto yield; next_iter: if (!should_yield()) continue; MNT_IUNLOCK(mp); yield: kern_yield(PRI_USER); relock_mnt: MNT_ILOCK(mp); } MNT_IUNLOCK(mp); vn_finished_write(mp); return done; } static int max_vnlru_free = 10000; /* limit on vnode free requests per call */ SYSCTL_INT(_debug, OID_AUTO, max_vnlru_free, CTLFLAG_RW, &max_vnlru_free, 0, "limit on vnode free requests per call to the vnlru_free routine"); /* * Attempt to reduce the free list by the requested amount. */ static void vnlru_free_locked(int count, struct vfsops *mnt_op) { struct vnode *vp; struct mount *mp; bool tried_batches; tried_batches = false; mtx_assert(&vnode_free_list_mtx, MA_OWNED); if (count > max_vnlru_free) count = max_vnlru_free; for (; count > 0; count--) { vp = TAILQ_FIRST(&vnode_free_list); /* * The list can be modified while the free_list_mtx * has been dropped and vp could be NULL here. */ if (vp == NULL) { if (tried_batches) break; mtx_unlock(&vnode_free_list_mtx); vnlru_return_batches(mnt_op); tried_batches = true; mtx_lock(&vnode_free_list_mtx); continue; } VNASSERT(vp->v_op != NULL, vp, ("vnlru_free: vnode already reclaimed.")); KASSERT((vp->v_iflag & VI_FREE) != 0, ("Removing vnode not on freelist")); KASSERT((vp->v_iflag & VI_ACTIVE) == 0, ("Mangling active vnode")); TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist); /* * Don't recycle if our vnode is from different type * of mount point. Note that mp is type-safe, the * check does not reach unmapped address even if * vnode is reclaimed. * Don't recycle if we can't get the interlock without * blocking. */ if ((mnt_op != NULL && (mp = vp->v_mount) != NULL && mp->mnt_op != mnt_op) || !VI_TRYLOCK(vp)) { TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist); continue; } VNASSERT((vp->v_iflag & VI_FREE) != 0 && vp->v_holdcnt == 0, vp, ("vp inconsistent on freelist")); /* * The clear of VI_FREE prevents activation of the * vnode. There is no sense in putting the vnode on * the mount point active list, only to remove it * later during recycling. Inline the relevant part * of vholdl(), to avoid triggering assertions or * activating. */ freevnodes--; vp->v_iflag &= ~VI_FREE; VNODE_REFCOUNT_FENCE_REL(); refcount_acquire(&vp->v_holdcnt); mtx_unlock(&vnode_free_list_mtx); VI_UNLOCK(vp); vtryrecycle(vp); /* * If the recycled succeeded this vdrop will actually free * the vnode. If not it will simply place it back on * the free list. */ vdrop(vp); mtx_lock(&vnode_free_list_mtx); } } void vnlru_free(int count, struct vfsops *mnt_op) { mtx_lock(&vnode_free_list_mtx); vnlru_free_locked(count, mnt_op); mtx_unlock(&vnode_free_list_mtx); } /* XXX some names and initialization are bad for limits and watermarks. */ static int vspace(void) { int space; gapvnodes = imax(desiredvnodes - wantfreevnodes, 100); vhiwat = gapvnodes / 11; /* 9% -- just under the 10% in vlrureclaim() */ vlowat = vhiwat / 2; if (numvnodes > desiredvnodes) return (0); space = desiredvnodes - numvnodes; if (freevnodes > wantfreevnodes) space += freevnodes - wantfreevnodes; return (space); } static void vnlru_return_batch_locked(struct mount *mp) { struct vnode *vp; mtx_assert(&mp->mnt_listmtx, MA_OWNED); if (mp->mnt_tmpfreevnodelistsize == 0) return; TAILQ_FOREACH(vp, &mp->mnt_tmpfreevnodelist, v_actfreelist) { VNASSERT((vp->v_mflag & VMP_TMPMNTFREELIST) != 0, vp, ("vnode without VMP_TMPMNTFREELIST on mnt_tmpfreevnodelist")); vp->v_mflag &= ~VMP_TMPMNTFREELIST; } mtx_lock(&vnode_free_list_mtx); TAILQ_CONCAT(&vnode_free_list, &mp->mnt_tmpfreevnodelist, v_actfreelist); freevnodes += mp->mnt_tmpfreevnodelistsize; mtx_unlock(&vnode_free_list_mtx); mp->mnt_tmpfreevnodelistsize = 0; } static void vnlru_return_batch(struct mount *mp) { mtx_lock(&mp->mnt_listmtx); vnlru_return_batch_locked(mp); mtx_unlock(&mp->mnt_listmtx); } static void vnlru_return_batches(struct vfsops *mnt_op) { struct mount *mp, *nmp; bool need_unbusy; mtx_lock(&mountlist_mtx); for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { need_unbusy = false; if (mnt_op != NULL && mp->mnt_op != mnt_op) goto next; if (mp->mnt_tmpfreevnodelistsize == 0) goto next; if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) == 0) { vnlru_return_batch(mp); need_unbusy = true; mtx_lock(&mountlist_mtx); } next: nmp = TAILQ_NEXT(mp, mnt_list); if (need_unbusy) vfs_unbusy(mp); } mtx_unlock(&mountlist_mtx); } /* * Attempt to recycle vnodes in a context that is always safe to block. * Calling vlrurecycle() from the bowels of filesystem code has some * interesting deadlock problems. */ static struct proc *vnlruproc; static int vnlruproc_sig; static void vnlru_proc(void) { struct mount *mp, *nmp; unsigned long onumvnodes; int done, force, trigger, usevnodes; bool reclaim_nc_src; EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, vnlruproc, SHUTDOWN_PRI_FIRST); force = 0; for (;;) { kproc_suspend_check(vnlruproc); mtx_lock(&vnode_free_list_mtx); /* * If numvnodes is too large (due to desiredvnodes being * adjusted using its sysctl, or emergency growth), first * try to reduce it by discarding from the free list. */ if (numvnodes > desiredvnodes) vnlru_free_locked(numvnodes - desiredvnodes, NULL); /* * Sleep if the vnode cache is in a good state. This is * when it is not over-full and has space for about a 4% * or 9% expansion (by growing its size or inexcessively * reducing its free list). Otherwise, try to reclaim * space for a 10% expansion. */ if (vstir && force == 0) { force = 1; vstir = 0; } if (vspace() >= vlowat && force == 0) { vnlruproc_sig = 0; wakeup(&vnlruproc_sig); msleep(vnlruproc, &vnode_free_list_mtx, PVFS|PDROP, "vlruwt", hz); continue; } mtx_unlock(&vnode_free_list_mtx); done = 0; onumvnodes = numvnodes; /* * Calculate parameters for recycling. These are the same * throughout the loop to give some semblance of fairness. * The trigger point is to avoid recycling vnodes with lots * of resident pages. We aren't trying to free memory; we * are trying to recycle or at least free vnodes. */ if (numvnodes <= desiredvnodes) usevnodes = numvnodes - freevnodes; else usevnodes = numvnodes; if (usevnodes <= 0) usevnodes = 1; /* * The trigger value is is chosen to give a conservatively * large value to ensure that it alone doesn't prevent * making progress. The value can easily be so large that * it is effectively infinite in some congested and * misconfigured cases, and this is necessary. Normally * it is about 8 to 100 (pages), which is quite large. */ trigger = vm_cnt.v_page_count * 2 / usevnodes; if (force < 2) trigger = vsmalltrigger; reclaim_nc_src = force >= 3; mtx_lock(&mountlist_mtx); for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) { nmp = TAILQ_NEXT(mp, mnt_list); continue; } done += vlrureclaim(mp, reclaim_nc_src, trigger); mtx_lock(&mountlist_mtx); nmp = TAILQ_NEXT(mp, mnt_list); vfs_unbusy(mp); } mtx_unlock(&mountlist_mtx); if (onumvnodes > desiredvnodes && numvnodes <= desiredvnodes) uma_reclaim(); if (done == 0) { if (force == 0 || force == 1) { force = 2; continue; } if (force == 2) { force = 3; continue; } force = 0; vnlru_nowhere++; tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3); } else kern_yield(PRI_USER); /* * After becoming active to expand above low water, keep * active until above high water. */ force = vspace() < vhiwat; } } static struct kproc_desc vnlru_kp = { "vnlru", vnlru_proc, &vnlruproc }; SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp); /* * Routines having to do with the management of the vnode table. */ /* * Try to recycle a freed vnode. We abort if anyone picks up a reference * before we actually vgone(). This function must be called with the vnode * held to prevent the vnode from being returned to the free list midway * through vgone(). */ static int vtryrecycle(struct vnode *vp) { struct mount *vnmp; CTR2(KTR_VFS, "%s: vp %p", __func__, vp); VNASSERT(vp->v_holdcnt, vp, ("vtryrecycle: Recycling vp %p without a reference.", vp)); /* * This vnode may found and locked via some other list, if so we * can't recycle it yet. */ if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { CTR2(KTR_VFS, "%s: impossible to recycle, vp %p lock is already held", __func__, vp); return (EWOULDBLOCK); } /* * Don't recycle if its filesystem is being suspended. */ if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) { VOP_UNLOCK(vp, 0); CTR2(KTR_VFS, "%s: impossible to recycle, cannot start the write for %p", __func__, vp); return (EBUSY); } /* * If we got this far, we need to acquire the interlock and see if * anyone picked up this vnode from another list. If not, we will * mark it with DOOMED via vgonel() so that anyone who does find it * will skip over it. */ VI_LOCK(vp); if (vp->v_usecount) { VOP_UNLOCK(vp, LK_INTERLOCK); vn_finished_write(vnmp); CTR2(KTR_VFS, "%s: impossible to recycle, %p is already referenced", __func__, vp); return (EBUSY); } if ((vp->v_iflag & VI_DOOMED) == 0) { counter_u64_add(recycles_count, 1); vgonel(vp); } VOP_UNLOCK(vp, LK_INTERLOCK); vn_finished_write(vnmp); return (0); } static void vcheckspace(void) { if (vspace() < vlowat && vnlruproc_sig == 0) { vnlruproc_sig = 1; wakeup(vnlruproc); } } /* * Wait if necessary for space for a new vnode. */ static int getnewvnode_wait(int suspended) { mtx_assert(&vnode_free_list_mtx, MA_OWNED); if (numvnodes >= desiredvnodes) { if (suspended) { /* * The file system is being suspended. We cannot * risk a deadlock here, so allow allocation of * another vnode even if this would give too many. */ return (0); } if (vnlruproc_sig == 0) { vnlruproc_sig = 1; /* avoid unnecessary wakeups */ wakeup(vnlruproc); } msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS, "vlruwk", hz); } /* Post-adjust like the pre-adjust in getnewvnode(). */ if (numvnodes + 1 > desiredvnodes && freevnodes > 1) vnlru_free_locked(1, NULL); return (numvnodes >= desiredvnodes ? ENFILE : 0); } /* * This hack is fragile, and probably not needed any more now that the * watermark handling works. */ void getnewvnode_reserve(u_int count) { struct thread *td; /* Pre-adjust like the pre-adjust in getnewvnode(), with any count. */ /* XXX no longer so quick, but this part is not racy. */ mtx_lock(&vnode_free_list_mtx); if (numvnodes + count > desiredvnodes && freevnodes > wantfreevnodes) vnlru_free_locked(ulmin(numvnodes + count - desiredvnodes, freevnodes - wantfreevnodes), NULL); mtx_unlock(&vnode_free_list_mtx); td = curthread; /* First try to be quick and racy. */ if (atomic_fetchadd_long(&numvnodes, count) + count <= desiredvnodes) { td->td_vp_reserv += count; vcheckspace(); /* XXX no longer so quick, but more racy */ return; } else atomic_subtract_long(&numvnodes, count); mtx_lock(&vnode_free_list_mtx); while (count > 0) { if (getnewvnode_wait(0) == 0) { count--; td->td_vp_reserv++; atomic_add_long(&numvnodes, 1); } } vcheckspace(); mtx_unlock(&vnode_free_list_mtx); } /* * This hack is fragile, especially if desiredvnodes or wantvnodes are * misconfgured or changed significantly. Reducing desiredvnodes below * the reserved amount should cause bizarre behaviour like reducing it * below the number of active vnodes -- the system will try to reduce * numvnodes to match, but should fail, so the subtraction below should * not overflow. */ void getnewvnode_drop_reserve(void) { struct thread *td; td = curthread; atomic_subtract_long(&numvnodes, td->td_vp_reserv); td->td_vp_reserv = 0; } /* * Return the next vnode from the free list. */ int getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops, struct vnode **vpp) { struct vnode *vp; struct thread *td; struct lock_object *lo; static int cyclecount; int error __unused; CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag); vp = NULL; td = curthread; if (td->td_vp_reserv > 0) { td->td_vp_reserv -= 1; goto alloc; } mtx_lock(&vnode_free_list_mtx); if (numvnodes < desiredvnodes) cyclecount = 0; else if (cyclecount++ >= freevnodes) { cyclecount = 0; vstir = 1; } /* * Grow the vnode cache if it will not be above its target max * after growing. Otherwise, if the free list is nonempty, try * to reclaim 1 item from it before growing the cache (possibly * above its target max if the reclamation failed or is delayed). * Otherwise, wait for some space. In all cases, schedule * vnlru_proc() if we are getting short of space. The watermarks * should be chosen so that we never wait or even reclaim from * the free list to below its target minimum. */ if (numvnodes + 1 <= desiredvnodes) ; else if (freevnodes > 0) vnlru_free_locked(1, NULL); else { error = getnewvnode_wait(mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)); #if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */ if (error != 0) { mtx_unlock(&vnode_free_list_mtx); return (error); } #endif } vcheckspace(); atomic_add_long(&numvnodes, 1); mtx_unlock(&vnode_free_list_mtx); alloc: counter_u64_add(vnodes_created, 1); vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK); /* * Locks are given the generic name "vnode" when created. * Follow the historic practice of using the filesystem * name when they allocated, e.g., "zfs", "ufs", "nfs, etc. * * Locks live in a witness group keyed on their name. Thus, * when a lock is renamed, it must also move from the witness * group of its old name to the witness group of its new name. * * The change only needs to be made when the vnode moves * from one filesystem type to another. We ensure that each * filesystem use a single static name pointer for its tag so * that we can compare pointers rather than doing a strcmp(). */ lo = &vp->v_vnlock->lock_object; if (lo->lo_name != tag) { lo->lo_name = tag; WITNESS_DESTROY(lo); WITNESS_INIT(lo, tag); } /* * By default, don't allow shared locks unless filesystems opt-in. */ vp->v_vnlock->lock_object.lo_flags |= LK_NOSHARE; /* * Finalize various vnode identity bits. */ KASSERT(vp->v_object == NULL, ("stale v_object %p", vp)); KASSERT(vp->v_lockf == NULL, ("stale v_lockf %p", vp)); KASSERT(vp->v_pollinfo == NULL, ("stale v_pollinfo %p", vp)); vp->v_type = VNON; vp->v_tag = tag; vp->v_op = vops; v_init_counters(vp); vp->v_bufobj.bo_ops = &buf_ops_bio; #ifdef DIAGNOSTIC if (mp == NULL && vops != &dead_vnodeops) printf("NULL mp in getnewvnode(9), tag %s\n", tag); #endif #ifdef MAC mac_vnode_init(vp); if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0) mac_vnode_associate_singlelabel(mp, vp); #endif if (mp != NULL) { vp->v_bufobj.bo_bsize = mp->mnt_stat.f_iosize; if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0) vp->v_vflag |= VV_NOKNOTE; } /* * For the filesystems which do not use vfs_hash_insert(), * still initialize v_hash to have vfs_hash_index() useful. * E.g., nullfs uses vfs_hash_index() on the lower vnode for * its own hashing. */ vp->v_hash = (uintptr_t)vp >> vnsz2log; *vpp = vp; return (0); } /* * Delete from old mount point vnode list, if on one. */ static void delmntque(struct vnode *vp) { struct mount *mp; int active; mp = vp->v_mount; if (mp == NULL) return; MNT_ILOCK(mp); VI_LOCK(vp); KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize, ("Active vnode list size %d > Vnode list size %d", mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize)); active = vp->v_iflag & VI_ACTIVE; vp->v_iflag &= ~VI_ACTIVE; if (active) { mtx_lock(&mp->mnt_listmtx); TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist); mp->mnt_activevnodelistsize--; mtx_unlock(&mp->mnt_listmtx); } vp->v_mount = NULL; VI_UNLOCK(vp); VNASSERT(mp->mnt_nvnodelistsize > 0, vp, ("bad mount point vnode list size")); TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); mp->mnt_nvnodelistsize--; MNT_REL(mp); MNT_IUNLOCK(mp); } static void insmntque_stddtr(struct vnode *vp, void *dtr_arg) { vp->v_data = NULL; vp->v_op = &dead_vnodeops; vgone(vp); vput(vp); } /* * Insert into list of vnodes for the new mount point, if available. */ int insmntque1(struct vnode *vp, struct mount *mp, void (*dtr)(struct vnode *, void *), void *dtr_arg) { KASSERT(vp->v_mount == NULL, ("insmntque: vnode already on per mount vnode list")); VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)")); ASSERT_VOP_ELOCKED(vp, "insmntque: non-locked vp"); /* * We acquire the vnode interlock early to ensure that the * vnode cannot be recycled by another process releasing a * holdcnt on it before we get it on both the vnode list * and the active vnode list. The mount mutex protects only * manipulation of the vnode list and the vnode freelist * mutex protects only manipulation of the active vnode list. * Hence the need to hold the vnode interlock throughout. */ MNT_ILOCK(mp); VI_LOCK(vp); if (((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 && ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 || mp->mnt_nvnodelistsize == 0)) && (vp->v_vflag & VV_FORCEINSMQ) == 0) { VI_UNLOCK(vp); MNT_IUNLOCK(mp); if (dtr != NULL) dtr(vp, dtr_arg); return (EBUSY); } vp->v_mount = mp; MNT_REF(mp); TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); VNASSERT(mp->mnt_nvnodelistsize >= 0, vp, ("neg mount point vnode list size")); mp->mnt_nvnodelistsize++; KASSERT((vp->v_iflag & VI_ACTIVE) == 0, ("Activating already active vnode")); vp->v_iflag |= VI_ACTIVE; mtx_lock(&mp->mnt_listmtx); TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist); mp->mnt_activevnodelistsize++; mtx_unlock(&mp->mnt_listmtx); VI_UNLOCK(vp); MNT_IUNLOCK(mp); return (0); } int insmntque(struct vnode *vp, struct mount *mp) { return (insmntque1(vp, mp, insmntque_stddtr, NULL)); } /* * Flush out and invalidate all buffers associated with a bufobj * Called with the underlying object locked. */ int bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo) { int error; BO_LOCK(bo); if (flags & V_SAVE) { error = bufobj_wwait(bo, slpflag, slptimeo); if (error) { BO_UNLOCK(bo); return (error); } if (bo->bo_dirty.bv_cnt > 0) { BO_UNLOCK(bo); if ((error = BO_SYNC(bo, MNT_WAIT)) != 0) return (error); /* * XXX We could save a lock/unlock if this was only * enabled under INVARIANTS */ BO_LOCK(bo); if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) panic("vinvalbuf: dirty bufs"); } } /* * If you alter this loop please notice that interlock is dropped and * reacquired in flushbuflist. Special care is needed to ensure that * no race conditions occur from this. */ do { error = flushbuflist(&bo->bo_clean, flags, bo, slpflag, slptimeo); if (error == 0 && !(flags & V_CLEANONLY)) error = flushbuflist(&bo->bo_dirty, flags, bo, slpflag, slptimeo); if (error != 0 && error != EAGAIN) { BO_UNLOCK(bo); return (error); } } while (error != 0); /* * Wait for I/O to complete. XXX needs cleaning up. The vnode can * have write I/O in-progress but if there is a VM object then the * VM object can also have read-I/O in-progress. */ do { bufobj_wwait(bo, 0, 0); - if ((flags & V_VMIO) == 0) { + if ((flags & V_VMIO) == 0 && bo->bo_object != NULL) { BO_UNLOCK(bo); - if (bo->bo_object != NULL) { - VM_OBJECT_WLOCK(bo->bo_object); - vm_object_pip_wait(bo->bo_object, "bovlbx"); - VM_OBJECT_WUNLOCK(bo->bo_object); - } + vm_object_pip_wait_unlocked(bo->bo_object, "bovlbx"); BO_LOCK(bo); } } while (bo->bo_numoutput > 0); BO_UNLOCK(bo); /* * Destroy the copy in the VM cache, too. */ if (bo->bo_object != NULL && (flags & (V_ALT | V_NORMAL | V_CLEANONLY | V_VMIO)) == 0) { VM_OBJECT_WLOCK(bo->bo_object); vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ? OBJPR_CLEANONLY : 0); VM_OBJECT_WUNLOCK(bo->bo_object); } #ifdef INVARIANTS BO_LOCK(bo); if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY | V_VMIO | V_ALLOWCLEAN)) == 0 && (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0)) panic("vinvalbuf: flush failed"); if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY | V_VMIO)) == 0 && bo->bo_dirty.bv_cnt > 0) panic("vinvalbuf: flush dirty failed"); BO_UNLOCK(bo); #endif return (0); } /* * Flush out and invalidate all buffers associated with a vnode. * Called with the underlying object locked. */ int vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo) { CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags); ASSERT_VOP_LOCKED(vp, "vinvalbuf"); if (vp->v_object != NULL && vp->v_object->handle != vp) return (0); return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo)); } /* * Flush out buffers on the specified list. * */ static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag, int slptimeo) { struct buf *bp, *nbp; int retval, error; daddr_t lblkno; b_xflags_t xflags; ASSERT_BO_WLOCKED(bo); retval = 0; TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) { /* * If we are flushing both V_NORMAL and V_ALT buffers then * do not skip any buffers. If we are flushing only V_NORMAL * buffers then skip buffers marked as BX_ALTDATA. If we are * flushing only V_ALT buffers then skip buffers not marked * as BX_ALTDATA. */ if (((flags & (V_NORMAL | V_ALT)) != (V_NORMAL | V_ALT)) && (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA) != 0) || ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0))) { continue; } if (nbp != NULL) { lblkno = nbp->b_lblkno; xflags = nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN); } retval = EAGAIN; error = BUF_TIMELOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo), "flushbuf", slpflag, slptimeo); if (error) { BO_LOCK(bo); return (error != ENOLCK ? error : EAGAIN); } KASSERT(bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); /* * XXX Since there are no node locks for NFS, I * believe there is a slight chance that a delayed * write will occur while sleeping just above, so * check for it. */ if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && (flags & V_SAVE)) { bremfree(bp); bp->b_flags |= B_ASYNC; bwrite(bp); BO_LOCK(bo); return (EAGAIN); /* XXX: why not loop ? */ } bremfree(bp); bp->b_flags |= (B_INVAL | B_RELBUF); bp->b_flags &= ~B_ASYNC; brelse(bp); BO_LOCK(bo); if (nbp == NULL) break; nbp = gbincore(bo, lblkno); if (nbp == NULL || (nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) != xflags) break; /* nbp invalid */ } return (retval); } int bnoreuselist(struct bufv *bufv, struct bufobj *bo, daddr_t startn, daddr_t endn) { struct buf *bp; int error; daddr_t lblkno; ASSERT_BO_LOCKED(bo); for (lblkno = startn;;) { again: bp = BUF_PCTRIE_LOOKUP_GE(&bufv->bv_root, lblkno); if (bp == NULL || bp->b_lblkno >= endn || bp->b_lblkno < startn) break; error = BUF_TIMELOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo), "brlsfl", 0, 0); if (error != 0) { BO_RLOCK(bo); if (error == ENOLCK) goto again; return (error); } KASSERT(bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); lblkno = bp->b_lblkno + 1; if ((bp->b_flags & B_MANAGED) == 0) bremfree(bp); bp->b_flags |= B_RELBUF; /* * In the VMIO case, use the B_NOREUSE flag to hint that the * pages backing each buffer in the range are unlikely to be * reused. Dirty buffers will have the hint applied once * they've been written. */ if ((bp->b_flags & B_VMIO) != 0) bp->b_flags |= B_NOREUSE; brelse(bp); BO_RLOCK(bo); } return (0); } /* * Truncate a file's buffer and pages to a specified length. This * is in lieu of the old vinvalbuf mechanism, which performed unneeded * sync activity. */ int vtruncbuf(struct vnode *vp, off_t length, int blksize) { struct buf *bp, *nbp; struct bufobj *bo; daddr_t startlbn; CTR4(KTR_VFS, "%s: vp %p with block %d:%ju", __func__, vp, blksize, (uintmax_t)length); /* * Round up to the *next* lbn. */ startlbn = howmany(length, blksize); ASSERT_VOP_LOCKED(vp, "vtruncbuf"); bo = &vp->v_bufobj; restart_unlocked: BO_LOCK(bo); while (v_inval_buf_range_locked(vp, bo, startlbn, INT64_MAX) == EAGAIN) ; if (length > 0) { restartsync: TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { if (bp->b_lblkno > 0) continue; /* * Since we hold the vnode lock this should only * fail if we're racing with the buf daemon. */ if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) goto restart_unlocked; VNASSERT((bp->b_flags & B_DELWRI), vp, ("buf(%p) on dirty queue without DELWRI", bp)); bremfree(bp); bawrite(bp); BO_LOCK(bo); goto restartsync; } } bufobj_wwait(bo, 0, 0); BO_UNLOCK(bo); vnode_pager_setsize(vp, length); return (0); } /* * Invalidate the cached pages of a file's buffer within the range of block * numbers [startlbn, endlbn). */ void v_inval_buf_range(struct vnode *vp, daddr_t startlbn, daddr_t endlbn, int blksize) { struct bufobj *bo; off_t start, end; ASSERT_VOP_LOCKED(vp, "v_inval_buf_range"); start = blksize * startlbn; end = blksize * endlbn; bo = &vp->v_bufobj; BO_LOCK(bo); MPASS(blksize == bo->bo_bsize); while (v_inval_buf_range_locked(vp, bo, startlbn, endlbn) == EAGAIN) ; BO_UNLOCK(bo); vn_pages_remove(vp, OFF_TO_IDX(start), OFF_TO_IDX(end + PAGE_SIZE - 1)); } static int v_inval_buf_range_locked(struct vnode *vp, struct bufobj *bo, daddr_t startlbn, daddr_t endlbn) { struct buf *bp, *nbp; bool anyfreed; ASSERT_VOP_LOCKED(vp, "v_inval_buf_range_locked"); ASSERT_BO_LOCKED(bo); do { anyfreed = false; TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) { if (bp->b_lblkno < startlbn || bp->b_lblkno >= endlbn) continue; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) { BO_LOCK(bo); return (EAGAIN); } bremfree(bp); bp->b_flags |= B_INVAL | B_RELBUF; bp->b_flags &= ~B_ASYNC; brelse(bp); anyfreed = true; BO_LOCK(bo); if (nbp != NULL && (((nbp->b_xflags & BX_VNCLEAN) == 0) || nbp->b_vp != vp || (nbp->b_flags & B_DELWRI) != 0)) return (EAGAIN); } TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { if (bp->b_lblkno < startlbn || bp->b_lblkno >= endlbn) continue; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) { BO_LOCK(bo); return (EAGAIN); } bremfree(bp); bp->b_flags |= B_INVAL | B_RELBUF; bp->b_flags &= ~B_ASYNC; brelse(bp); anyfreed = true; BO_LOCK(bo); if (nbp != NULL && (((nbp->b_xflags & BX_VNDIRTY) == 0) || (nbp->b_vp != vp) || (nbp->b_flags & B_DELWRI) == 0)) return (EAGAIN); } } while (anyfreed); return (0); } static void buf_vlist_remove(struct buf *bp) { struct bufv *bv; KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); ASSERT_BO_WLOCKED(bp->b_bufobj); KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) != (BX_VNDIRTY|BX_VNCLEAN), ("buf_vlist_remove: Buf %p is on two lists", bp)); if (bp->b_xflags & BX_VNDIRTY) bv = &bp->b_bufobj->bo_dirty; else bv = &bp->b_bufobj->bo_clean; BUF_PCTRIE_REMOVE(&bv->bv_root, bp->b_lblkno); TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs); bv->bv_cnt--; bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); } /* * Add the buffer to the sorted clean or dirty block list. * * NOTE: xflags is passed as a constant, optimizing this inline function! */ static void buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags) { struct bufv *bv; struct buf *n; int error; ASSERT_BO_WLOCKED(bo); KASSERT((xflags & BX_VNDIRTY) == 0 || (bo->bo_flag & BO_DEAD) == 0, ("dead bo %p", bo)); KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags)); bp->b_xflags |= xflags; if (xflags & BX_VNDIRTY) bv = &bo->bo_dirty; else bv = &bo->bo_clean; /* * Keep the list ordered. Optimize empty list insertion. Assume * we tend to grow at the tail so lookup_le should usually be cheaper * than _ge. */ if (bv->bv_cnt == 0 || bp->b_lblkno > TAILQ_LAST(&bv->bv_hd, buflists)->b_lblkno) TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs); else if ((n = BUF_PCTRIE_LOOKUP_LE(&bv->bv_root, bp->b_lblkno)) == NULL) TAILQ_INSERT_HEAD(&bv->bv_hd, bp, b_bobufs); else TAILQ_INSERT_AFTER(&bv->bv_hd, n, bp, b_bobufs); error = BUF_PCTRIE_INSERT(&bv->bv_root, bp); if (error) panic("buf_vlist_add: Preallocated nodes insufficient."); bv->bv_cnt++; } /* * Look up a buffer using the buffer tries. */ struct buf * gbincore(struct bufobj *bo, daddr_t lblkno) { struct buf *bp; ASSERT_BO_LOCKED(bo); bp = BUF_PCTRIE_LOOKUP(&bo->bo_clean.bv_root, lblkno); if (bp != NULL) return (bp); return BUF_PCTRIE_LOOKUP(&bo->bo_dirty.bv_root, lblkno); } /* * Associate a buffer with a vnode. */ void bgetvp(struct vnode *vp, struct buf *bp) { struct bufobj *bo; bo = &vp->v_bufobj; ASSERT_BO_WLOCKED(bo); VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free")); CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags); VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp, ("bgetvp: bp already attached! %p", bp)); vhold(vp); bp->b_vp = vp; bp->b_bufobj = bo; /* * Insert onto list for new vnode. */ buf_vlist_add(bp, bo, BX_VNCLEAN); } /* * Disassociate a buffer from a vnode. */ void brelvp(struct buf *bp) { struct bufobj *bo; struct vnode *vp; CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); /* * Delete from old vnode list, if on one. */ vp = bp->b_vp; /* XXX */ bo = bp->b_bufobj; BO_LOCK(bo); if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) buf_vlist_remove(bp); else panic("brelvp: Buffer %p not on queue.", bp); if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { bo->bo_flag &= ~BO_ONWORKLST; mtx_lock(&sync_mtx); LIST_REMOVE(bo, bo_synclist); syncer_worklist_len--; mtx_unlock(&sync_mtx); } bp->b_vp = NULL; bp->b_bufobj = NULL; BO_UNLOCK(bo); vdrop(vp); } /* * Add an item to the syncer work queue. */ static void vn_syncer_add_to_worklist(struct bufobj *bo, int delay) { int slot; ASSERT_BO_WLOCKED(bo); mtx_lock(&sync_mtx); if (bo->bo_flag & BO_ONWORKLST) LIST_REMOVE(bo, bo_synclist); else { bo->bo_flag |= BO_ONWORKLST; syncer_worklist_len++; } if (delay > syncer_maxdelay - 2) delay = syncer_maxdelay - 2; slot = (syncer_delayno + delay) & syncer_mask; LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist); mtx_unlock(&sync_mtx); } static int sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS) { int error, len; mtx_lock(&sync_mtx); len = syncer_worklist_len - sync_vnode_count; mtx_unlock(&sync_mtx); error = SYSCTL_OUT(req, &len, sizeof(len)); return (error); } SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0, sysctl_vfs_worklist_len, "I", "Syncer thread worklist length"); static struct proc *updateproc; static void sched_sync(void); static struct kproc_desc up_kp = { "syncer", sched_sync, &updateproc }; SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp); static int sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td) { struct vnode *vp; struct mount *mp; *bo = LIST_FIRST(slp); if (*bo == NULL) return (0); vp = bo2vnode(*bo); if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0) return (1); /* * We use vhold in case the vnode does not * successfully sync. vhold prevents the vnode from * going away when we unlock the sync_mtx so that * we can acquire the vnode interlock. */ vholdl(vp); mtx_unlock(&sync_mtx); VI_UNLOCK(vp); if (vn_start_write(vp, &mp, V_NOWAIT) != 0) { vdrop(vp); mtx_lock(&sync_mtx); return (*bo == LIST_FIRST(slp)); } vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); (void) VOP_FSYNC(vp, MNT_LAZY, td); VOP_UNLOCK(vp, 0); vn_finished_write(mp); BO_LOCK(*bo); if (((*bo)->bo_flag & BO_ONWORKLST) != 0) { /* * Put us back on the worklist. The worklist * routine will remove us from our current * position and then add us back in at a later * position. */ vn_syncer_add_to_worklist(*bo, syncdelay); } BO_UNLOCK(*bo); vdrop(vp); mtx_lock(&sync_mtx); return (0); } static int first_printf = 1; /* * System filesystem synchronizer daemon. */ static void sched_sync(void) { struct synclist *next, *slp; struct bufobj *bo; long starttime; struct thread *td = curthread; int last_work_seen; int net_worklist_len; int syncer_final_iter; int error; last_work_seen = 0; syncer_final_iter = 0; syncer_state = SYNCER_RUNNING; starttime = time_uptime; td->td_pflags |= TDP_NORUNNINGBUF; EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc, SHUTDOWN_PRI_LAST); mtx_lock(&sync_mtx); for (;;) { if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter == 0) { mtx_unlock(&sync_mtx); kproc_suspend_check(td->td_proc); mtx_lock(&sync_mtx); } net_worklist_len = syncer_worklist_len - sync_vnode_count; if (syncer_state != SYNCER_RUNNING && starttime != time_uptime) { if (first_printf) { printf("\nSyncing disks, vnodes remaining... "); first_printf = 0; } printf("%d ", net_worklist_len); } starttime = time_uptime; /* * Push files whose dirty time has expired. Be careful * of interrupt race on slp queue. * * Skip over empty worklist slots when shutting down. */ do { slp = &syncer_workitem_pending[syncer_delayno]; syncer_delayno += 1; if (syncer_delayno == syncer_maxdelay) syncer_delayno = 0; next = &syncer_workitem_pending[syncer_delayno]; /* * If the worklist has wrapped since the * it was emptied of all but syncer vnodes, * switch to the FINAL_DELAY state and run * for one more second. */ if (syncer_state == SYNCER_SHUTTING_DOWN && net_worklist_len == 0 && last_work_seen == syncer_delayno) { syncer_state = SYNCER_FINAL_DELAY; syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP; } } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) && syncer_worklist_len > 0); /* * Keep track of the last time there was anything * on the worklist other than syncer vnodes. * Return to the SHUTTING_DOWN state if any * new work appears. */ if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING) last_work_seen = syncer_delayno; if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY) syncer_state = SYNCER_SHUTTING_DOWN; while (!LIST_EMPTY(slp)) { error = sync_vnode(slp, &bo, td); if (error == 1) { LIST_REMOVE(bo, bo_synclist); LIST_INSERT_HEAD(next, bo, bo_synclist); continue; } if (first_printf == 0) { /* * Drop the sync mutex, because some watchdog * drivers need to sleep while patting */ mtx_unlock(&sync_mtx); wdog_kern_pat(WD_LASTVAL); mtx_lock(&sync_mtx); } } if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0) syncer_final_iter--; /* * The variable rushjob allows the kernel to speed up the * processing of the filesystem syncer process. A rushjob * value of N tells the filesystem syncer to process the next * N seconds worth of work on its queue ASAP. Currently rushjob * is used by the soft update code to speed up the filesystem * syncer process when the incore state is getting so far * ahead of the disk that the kernel memory pool is being * threatened with exhaustion. */ if (rushjob > 0) { rushjob -= 1; continue; } /* * Just sleep for a short period of time between * iterations when shutting down to allow some I/O * to happen. * * If it has taken us less than a second to process the * current work, then wait. Otherwise start right over * again. We can still lose time if any single round * takes more than two seconds, but it does not really * matter as we are just trying to generally pace the * filesystem activity. */ if (syncer_state != SYNCER_RUNNING || time_uptime == starttime) { thread_lock(td); sched_prio(td, PPAUSE); thread_unlock(td); } if (syncer_state != SYNCER_RUNNING) cv_timedwait(&sync_wakeup, &sync_mtx, hz / SYNCER_SHUTDOWN_SPEEDUP); else if (time_uptime == starttime) cv_timedwait(&sync_wakeup, &sync_mtx, hz); } } /* * Request the syncer daemon to speed up its work. * We never push it to speed up more than half of its * normal turn time, otherwise it could take over the cpu. */ int speedup_syncer(void) { int ret = 0; mtx_lock(&sync_mtx); if (rushjob < syncdelay / 2) { rushjob += 1; stat_rush_requests += 1; ret = 1; } mtx_unlock(&sync_mtx); cv_broadcast(&sync_wakeup); return (ret); } /* * Tell the syncer to speed up its work and run though its work * list several times, then tell it to shut down. */ static void syncer_shutdown(void *arg, int howto) { if (howto & RB_NOSYNC) return; mtx_lock(&sync_mtx); syncer_state = SYNCER_SHUTTING_DOWN; rushjob = 0; mtx_unlock(&sync_mtx); cv_broadcast(&sync_wakeup); kproc_shutdown(arg, howto); } void syncer_suspend(void) { syncer_shutdown(updateproc, 0); } void syncer_resume(void) { mtx_lock(&sync_mtx); first_printf = 1; syncer_state = SYNCER_RUNNING; mtx_unlock(&sync_mtx); cv_broadcast(&sync_wakeup); kproc_resume(updateproc); } /* * Reassign a buffer from one vnode to another. * Used to assign file specific control information * (indirect blocks) to the vnode to which they belong. */ void reassignbuf(struct buf *bp) { struct vnode *vp; struct bufobj *bo; int delay; #ifdef INVARIANTS struct bufv *bv; #endif vp = bp->b_vp; bo = bp->b_bufobj; ++reassignbufcalls; CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); /* * B_PAGING flagged buffers cannot be reassigned because their vp * is not fully linked in. */ if (bp->b_flags & B_PAGING) panic("cannot reassign paging buffer"); /* * Delete from old vnode list, if on one. */ BO_LOCK(bo); if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) buf_vlist_remove(bp); else panic("reassignbuf: Buffer %p not on queue.", bp); /* * If dirty, put on list of dirty buffers; otherwise insert onto list * of clean buffers. */ if (bp->b_flags & B_DELWRI) { if ((bo->bo_flag & BO_ONWORKLST) == 0) { switch (vp->v_type) { case VDIR: delay = dirdelay; break; case VCHR: delay = metadelay; break; default: delay = filedelay; } vn_syncer_add_to_worklist(bo, delay); } buf_vlist_add(bp, bo, BX_VNDIRTY); } else { buf_vlist_add(bp, bo, BX_VNCLEAN); if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { mtx_lock(&sync_mtx); LIST_REMOVE(bo, bo_synclist); syncer_worklist_len--; mtx_unlock(&sync_mtx); bo->bo_flag &= ~BO_ONWORKLST; } } #ifdef INVARIANTS bv = &bo->bo_clean; bp = TAILQ_FIRST(&bv->bv_hd); KASSERT(bp == NULL || bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); bp = TAILQ_LAST(&bv->bv_hd, buflists); KASSERT(bp == NULL || bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); bv = &bo->bo_dirty; bp = TAILQ_FIRST(&bv->bv_hd); KASSERT(bp == NULL || bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); bp = TAILQ_LAST(&bv->bv_hd, buflists); KASSERT(bp == NULL || bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); #endif BO_UNLOCK(bo); } static void v_init_counters(struct vnode *vp) { VNASSERT(vp->v_type == VNON && vp->v_data == NULL && vp->v_iflag == 0, vp, ("%s called for an initialized vnode", __FUNCTION__)); ASSERT_VI_UNLOCKED(vp, __FUNCTION__); refcount_init(&vp->v_holdcnt, 1); refcount_init(&vp->v_usecount, 1); } static void v_incr_usecount_locked(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __func__); if ((vp->v_iflag & VI_OWEINACT) != 0) { VNASSERT(vp->v_usecount == 0, vp, ("vnode with usecount and VI_OWEINACT set")); vp->v_iflag &= ~VI_OWEINACT; } refcount_acquire(&vp->v_usecount); v_incr_devcount(vp); } /* * Increment the use count on the vnode, taking care to reference * the driver's usecount if this is a chardev. */ static void v_incr_usecount(struct vnode *vp) { ASSERT_VI_UNLOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); if (vp->v_type != VCHR && refcount_acquire_if_not_zero(&vp->v_usecount)) { VNODE_REFCOUNT_FENCE_ACQ(); VNASSERT((vp->v_iflag & VI_OWEINACT) == 0, vp, ("vnode with usecount and VI_OWEINACT set")); } else { VI_LOCK(vp); v_incr_usecount_locked(vp); VI_UNLOCK(vp); } } /* * Increment si_usecount of the associated device, if any. */ static void v_incr_devcount(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __FUNCTION__); if (vp->v_type == VCHR && vp->v_rdev != NULL) { dev_lock(); vp->v_rdev->si_usecount++; dev_unlock(); } } /* * Decrement si_usecount of the associated device, if any. */ static void v_decr_devcount(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __FUNCTION__); if (vp->v_type == VCHR && vp->v_rdev != NULL) { dev_lock(); vp->v_rdev->si_usecount--; dev_unlock(); } } /* * Grab a particular vnode from the free list, increment its * reference count and lock it. VI_DOOMED is set if the vnode * is being destroyed. Only callers who specify LK_RETRY will * see doomed vnodes. If inactive processing was delayed in * vput try to do it here. * * Notes on lockless counter manipulation: * _vhold, vputx and other routines make various decisions based * on either holdcnt or usecount being 0. As long as either counter * is not transitioning 0->1 nor 1->0, the manipulation can be done * with atomic operations. Otherwise the interlock is taken covering * both the atomic and additional actions. */ int vget(struct vnode *vp, int flags, struct thread *td) { int error, oweinact; VNASSERT((flags & LK_TYPE_MASK) != 0, vp, ("vget: invalid lock operation")); if ((flags & LK_INTERLOCK) != 0) ASSERT_VI_LOCKED(vp, __func__); else ASSERT_VI_UNLOCKED(vp, __func__); if ((flags & LK_VNHELD) != 0) VNASSERT((vp->v_holdcnt > 0), vp, ("vget: LK_VNHELD passed but vnode not held")); CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags); if ((flags & LK_VNHELD) == 0) _vhold(vp, (flags & LK_INTERLOCK) != 0); if ((error = vn_lock(vp, flags)) != 0) { vdrop(vp); CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__, vp); return (error); } if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0) panic("vget: vn_lock failed to return ENOENT\n"); /* * We don't guarantee that any particular close will * trigger inactive processing so just make a best effort * here at preventing a reference to a removed file. If * we don't succeed no harm is done. * * Upgrade our holdcnt to a usecount. */ if (vp->v_type == VCHR || !refcount_acquire_if_not_zero(&vp->v_usecount)) { VI_LOCK(vp); if ((vp->v_iflag & VI_OWEINACT) == 0) { oweinact = 0; } else { oweinact = 1; vp->v_iflag &= ~VI_OWEINACT; VNODE_REFCOUNT_FENCE_REL(); } refcount_acquire(&vp->v_usecount); v_incr_devcount(vp); if (oweinact && VOP_ISLOCKED(vp) == LK_EXCLUSIVE && (flags & LK_NOWAIT) == 0) vinactive(vp, td); VI_UNLOCK(vp); } return (0); } /* * Increase the reference (use) and hold count of a vnode. * This will also remove the vnode from the free list if it is presently free. */ void vref(struct vnode *vp) { CTR2(KTR_VFS, "%s: vp %p", __func__, vp); _vhold(vp, false); v_incr_usecount(vp); } void vrefl(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); _vhold(vp, true); v_incr_usecount_locked(vp); } void vrefact(struct vnode *vp) { CTR2(KTR_VFS, "%s: vp %p", __func__, vp); if (__predict_false(vp->v_type == VCHR)) { VNASSERT(vp->v_holdcnt > 0 && vp->v_usecount > 0, vp, ("%s: wrong ref counts", __func__)); vref(vp); return; } #ifdef INVARIANTS int old = atomic_fetchadd_int(&vp->v_holdcnt, 1); VNASSERT(old > 0, vp, ("%s: wrong hold count", __func__)); old = atomic_fetchadd_int(&vp->v_usecount, 1); VNASSERT(old > 0, vp, ("%s: wrong use count", __func__)); #else refcount_acquire(&vp->v_holdcnt); refcount_acquire(&vp->v_usecount); #endif } /* * Return reference count of a vnode. * * The results of this call are only guaranteed when some mechanism is used to * stop other processes from gaining references to the vnode. This may be the * case if the caller holds the only reference. This is also useful when stale * data is acceptable as race conditions may be accounted for by some other * means. */ int vrefcnt(struct vnode *vp) { return (vp->v_usecount); } #define VPUTX_VRELE 1 #define VPUTX_VPUT 2 #define VPUTX_VUNREF 3 /* * Decrement the use and hold counts for a vnode. * * See an explanation near vget() as to why atomic operation is safe. */ static void vputx(struct vnode *vp, int func) { int error; KASSERT(vp != NULL, ("vputx: null vp")); if (func == VPUTX_VUNREF) ASSERT_VOP_LOCKED(vp, "vunref"); else if (func == VPUTX_VPUT) ASSERT_VOP_LOCKED(vp, "vput"); else KASSERT(func == VPUTX_VRELE, ("vputx: wrong func")); ASSERT_VI_UNLOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); if (vp->v_type != VCHR && refcount_release_if_not_last(&vp->v_usecount)) { if (func == VPUTX_VPUT) VOP_UNLOCK(vp, 0); vdrop(vp); return; } VI_LOCK(vp); /* * We want to hold the vnode until the inactive finishes to * prevent vgone() races. We drop the use count here and the * hold count below when we're done. */ if (!refcount_release(&vp->v_usecount) || (vp->v_iflag & VI_DOINGINACT)) { if (func == VPUTX_VPUT) VOP_UNLOCK(vp, 0); v_decr_devcount(vp); vdropl(vp); return; } v_decr_devcount(vp); error = 0; if (vp->v_usecount != 0) { vn_printf(vp, "vputx: usecount not zero for vnode "); panic("vputx: usecount not zero"); } CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp); /* * We must call VOP_INACTIVE with the node locked. Mark * as VI_DOINGINACT to avoid recursion. */ vp->v_iflag |= VI_OWEINACT; switch (func) { case VPUTX_VRELE: error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK); VI_LOCK(vp); break; case VPUTX_VPUT: if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK | LK_NOWAIT); VI_LOCK(vp); } break; case VPUTX_VUNREF: if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { error = VOP_LOCK(vp, LK_TRYUPGRADE | LK_INTERLOCK); VI_LOCK(vp); } break; } VNASSERT(vp->v_usecount == 0 || (vp->v_iflag & VI_OWEINACT) == 0, vp, ("vnode with usecount and VI_OWEINACT set")); if (error == 0) { if (vp->v_iflag & VI_OWEINACT) vinactive(vp, curthread); if (func != VPUTX_VUNREF) VOP_UNLOCK(vp, 0); } vdropl(vp); } /* * Vnode put/release. * If count drops to zero, call inactive routine and return to freelist. */ void vrele(struct vnode *vp) { vputx(vp, VPUTX_VRELE); } /* * Release an already locked vnode. This give the same effects as * unlock+vrele(), but takes less time and avoids releasing and * re-aquiring the lock (as vrele() acquires the lock internally.) */ void vput(struct vnode *vp) { vputx(vp, VPUTX_VPUT); } /* * Release an exclusively locked vnode. Do not unlock the vnode lock. */ void vunref(struct vnode *vp) { vputx(vp, VPUTX_VUNREF); } /* * Increase the hold count and activate if this is the first reference. */ void _vhold(struct vnode *vp, bool locked) { struct mount *mp; if (locked) ASSERT_VI_LOCKED(vp, __func__); else ASSERT_VI_UNLOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); if (!locked) { if (refcount_acquire_if_not_zero(&vp->v_holdcnt)) { VNODE_REFCOUNT_FENCE_ACQ(); VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("_vhold: vnode with holdcnt is free")); return; } VI_LOCK(vp); } if ((vp->v_iflag & VI_FREE) == 0) { refcount_acquire(&vp->v_holdcnt); if (!locked) VI_UNLOCK(vp); return; } VNASSERT(vp->v_holdcnt == 0, vp, ("%s: wrong hold count", __func__)); VNASSERT(vp->v_op != NULL, vp, ("%s: vnode already reclaimed.", __func__)); /* * Remove a vnode from the free list, mark it as in use, * and put it on the active list. */ VNASSERT(vp->v_mount != NULL, vp, ("_vhold: vnode not on per mount vnode list")); mp = vp->v_mount; mtx_lock(&mp->mnt_listmtx); if ((vp->v_mflag & VMP_TMPMNTFREELIST) != 0) { TAILQ_REMOVE(&mp->mnt_tmpfreevnodelist, vp, v_actfreelist); mp->mnt_tmpfreevnodelistsize--; vp->v_mflag &= ~VMP_TMPMNTFREELIST; } else { mtx_lock(&vnode_free_list_mtx); TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist); freevnodes--; mtx_unlock(&vnode_free_list_mtx); } KASSERT((vp->v_iflag & VI_ACTIVE) == 0, ("Activating already active vnode")); vp->v_iflag &= ~VI_FREE; vp->v_iflag |= VI_ACTIVE; TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist); mp->mnt_activevnodelistsize++; mtx_unlock(&mp->mnt_listmtx); refcount_acquire(&vp->v_holdcnt); if (!locked) VI_UNLOCK(vp); } /* * Drop the hold count of the vnode. If this is the last reference to * the vnode we place it on the free list unless it has been vgone'd * (marked VI_DOOMED) in which case we will free it. * * Because the vnode vm object keeps a hold reference on the vnode if * there is at least one resident non-cached page, the vnode cannot * leave the active list without the page cleanup done. */ void _vdrop(struct vnode *vp, bool locked) { struct bufobj *bo; struct mount *mp; int active; if (locked) ASSERT_VI_LOCKED(vp, __func__); else ASSERT_VI_UNLOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); if ((int)vp->v_holdcnt <= 0) panic("vdrop: holdcnt %d", vp->v_holdcnt); if (!locked) { if (refcount_release_if_not_last(&vp->v_holdcnt)) return; VI_LOCK(vp); } if (refcount_release(&vp->v_holdcnt) == 0) { VI_UNLOCK(vp); return; } if ((vp->v_iflag & VI_DOOMED) == 0) { /* * Mark a vnode as free: remove it from its active list * and put it up for recycling on the freelist. */ VNASSERT(vp->v_op != NULL, vp, ("vdropl: vnode already reclaimed.")); VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free")); VNASSERT(vp->v_holdcnt == 0, vp, ("vdropl: freeing when we shouldn't")); active = vp->v_iflag & VI_ACTIVE; if ((vp->v_iflag & VI_OWEINACT) == 0) { vp->v_iflag &= ~VI_ACTIVE; mp = vp->v_mount; if (mp != NULL) { mtx_lock(&mp->mnt_listmtx); if (active) { TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist); mp->mnt_activevnodelistsize--; } TAILQ_INSERT_TAIL(&mp->mnt_tmpfreevnodelist, vp, v_actfreelist); mp->mnt_tmpfreevnodelistsize++; vp->v_iflag |= VI_FREE; vp->v_mflag |= VMP_TMPMNTFREELIST; VI_UNLOCK(vp); if (mp->mnt_tmpfreevnodelistsize >= mnt_free_list_batch) vnlru_return_batch_locked(mp); mtx_unlock(&mp->mnt_listmtx); } else { VNASSERT(active == 0, vp, ("vdropl: active vnode not on per mount " "vnode list")); mtx_lock(&vnode_free_list_mtx); TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist); freevnodes++; vp->v_iflag |= VI_FREE; VI_UNLOCK(vp); mtx_unlock(&vnode_free_list_mtx); } } else { VI_UNLOCK(vp); counter_u64_add(free_owe_inact, 1); } return; } /* * The vnode has been marked for destruction, so free it. * * The vnode will be returned to the zone where it will * normally remain until it is needed for another vnode. We * need to cleanup (or verify that the cleanup has already * been done) any residual data left from its current use * so as not to contaminate the freshly allocated vnode. */ CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp); atomic_subtract_long(&numvnodes, 1); bo = &vp->v_bufobj; VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("cleaned vnode still on the free list.")); VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't")); VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count")); VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count")); VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count")); VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's")); VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0")); VNASSERT(pctrie_is_empty(&bo->bo_clean.bv_root), vp, ("clean blk trie not empty")); VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0")); VNASSERT(pctrie_is_empty(&bo->bo_dirty.bv_root), vp, ("dirty blk trie not empty")); VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst")); VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src")); VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for ..")); VNASSERT(TAILQ_EMPTY(&vp->v_rl.rl_waiters), vp, ("Dangling rangelock waiters")); VI_UNLOCK(vp); #ifdef MAC mac_vnode_destroy(vp); #endif if (vp->v_pollinfo != NULL) { destroy_vpollinfo(vp->v_pollinfo); vp->v_pollinfo = NULL; } #ifdef INVARIANTS /* XXX Elsewhere we detect an already freed vnode via NULL v_op. */ vp->v_op = NULL; #endif vp->v_mountedhere = NULL; vp->v_unpcb = NULL; vp->v_rdev = NULL; vp->v_fifoinfo = NULL; vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0; vp->v_iflag = 0; vp->v_vflag = 0; bo->bo_flag = 0; uma_zfree(vnode_zone, vp); } /* * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT * flags. DOINGINACT prevents us from recursing in calls to vinactive. * OWEINACT tracks whether a vnode missed a call to inactive due to a * failed lock upgrade. */ void vinactive(struct vnode *vp, struct thread *td) { struct vm_object *obj; ASSERT_VOP_ELOCKED(vp, "vinactive"); ASSERT_VI_LOCKED(vp, "vinactive"); VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp, ("vinactive: recursed on VI_DOINGINACT")); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); vp->v_iflag |= VI_DOINGINACT; vp->v_iflag &= ~VI_OWEINACT; VI_UNLOCK(vp); /* * Before moving off the active list, we must be sure that any * modified pages are converted into the vnode's dirty * buffers, since these will no longer be checked once the * vnode is on the inactive list. * * The write-out of the dirty pages is asynchronous. At the * point that VOP_INACTIVE() is called, there could still be * pending I/O and dirty pages in the object. */ if ((obj = vp->v_object) != NULL && (vp->v_vflag & VV_NOSYNC) == 0 && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) { VM_OBJECT_WLOCK(obj); vm_object_page_clean(obj, 0, 0, 0); VM_OBJECT_WUNLOCK(obj); } VOP_INACTIVE(vp, td); VI_LOCK(vp); VNASSERT(vp->v_iflag & VI_DOINGINACT, vp, ("vinactive: lost VI_DOINGINACT")); vp->v_iflag &= ~VI_DOINGINACT; } /* * Remove any vnodes in the vnode table belonging to mount point mp. * * If FORCECLOSE is not specified, there should not be any active ones, * return error if any are found (nb: this is a user error, not a * system error). If FORCECLOSE is specified, detach any active vnodes * that are found. * * If WRITECLOSE is set, only flush out regular file vnodes open for * writing. * * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped. * * `rootrefs' specifies the base reference count for the root vnode * of this filesystem. The root vnode is considered busy if its * v_usecount exceeds this value. On a successful return, vflush(, td) * will call vrele() on the root vnode exactly rootrefs times. * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must * be zero. */ #ifdef DIAGNOSTIC static int busyprt = 0; /* print out busy vnodes */ SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes"); #endif int vflush(struct mount *mp, int rootrefs, int flags, struct thread *td) { struct vnode *vp, *mvp, *rootvp = NULL; struct vattr vattr; int busy = 0, error; CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp, rootrefs, flags); if (rootrefs > 0) { KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0, ("vflush: bad args")); /* * Get the filesystem root vnode. We can vput() it * immediately, since with rootrefs > 0, it won't go away. */ if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) { CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d", __func__, error); return (error); } vput(rootvp); } loop: MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { vholdl(vp); error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE); if (error) { vdrop(vp); MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); goto loop; } /* * Skip over a vnodes marked VV_SYSTEM. */ if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) { VOP_UNLOCK(vp, 0); vdrop(vp); continue; } /* * If WRITECLOSE is set, flush out unlinked but still open * files (even if open only for reading) and regular file * vnodes open for writing. */ if (flags & WRITECLOSE) { if (vp->v_object != NULL) { VM_OBJECT_WLOCK(vp->v_object); vm_object_page_clean(vp->v_object, 0, 0, 0); VM_OBJECT_WUNLOCK(vp->v_object); } error = VOP_FSYNC(vp, MNT_WAIT, td); if (error != 0) { VOP_UNLOCK(vp, 0); vdrop(vp); MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); return (error); } error = VOP_GETATTR(vp, &vattr, td->td_ucred); VI_LOCK(vp); if ((vp->v_type == VNON || (error == 0 && vattr.va_nlink > 0)) && (vp->v_writecount <= 0 || vp->v_type != VREG)) { VOP_UNLOCK(vp, 0); vdropl(vp); continue; } } else VI_LOCK(vp); /* * With v_usecount == 0, all we need to do is clear out the * vnode data structures and we are done. * * If FORCECLOSE is set, forcibly close the vnode. */ if (vp->v_usecount == 0 || (flags & FORCECLOSE)) { vgonel(vp); } else { busy++; #ifdef DIAGNOSTIC if (busyprt) vn_printf(vp, "vflush: busy vnode "); #endif } VOP_UNLOCK(vp, 0); vdropl(vp); } if (rootrefs > 0 && (flags & FORCECLOSE) == 0) { /* * If just the root vnode is busy, and if its refcount * is equal to `rootrefs', then go ahead and kill it. */ VI_LOCK(rootvp); KASSERT(busy > 0, ("vflush: not busy")); VNASSERT(rootvp->v_usecount >= rootrefs, rootvp, ("vflush: usecount %d < rootrefs %d", rootvp->v_usecount, rootrefs)); if (busy == 1 && rootvp->v_usecount == rootrefs) { VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK); vgone(rootvp); VOP_UNLOCK(rootvp, 0); busy = 0; } else VI_UNLOCK(rootvp); } if (busy) { CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__, busy); return (EBUSY); } for (; rootrefs > 0; rootrefs--) vrele(rootvp); return (0); } /* * Recycle an unused vnode to the front of the free list. */ int vrecycle(struct vnode *vp) { int recycled; VI_LOCK(vp); recycled = vrecyclel(vp); VI_UNLOCK(vp); return (recycled); } /* * vrecycle, with the vp interlock held. */ int vrecyclel(struct vnode *vp) { int recycled; ASSERT_VOP_ELOCKED(vp, __func__); ASSERT_VI_LOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); recycled = 0; if (vp->v_usecount == 0) { recycled = 1; vgonel(vp); } return (recycled); } /* * Eliminate all activity associated with a vnode * in preparation for reuse. */ void vgone(struct vnode *vp) { VI_LOCK(vp); vgonel(vp); VI_UNLOCK(vp); } static void notify_lowervp_vfs_dummy(struct mount *mp __unused, struct vnode *lowervp __unused) { } /* * Notify upper mounts about reclaimed or unlinked vnode. */ void vfs_notify_upper(struct vnode *vp, int event) { static struct vfsops vgonel_vfsops = { .vfs_reclaim_lowervp = notify_lowervp_vfs_dummy, .vfs_unlink_lowervp = notify_lowervp_vfs_dummy, }; struct mount *mp, *ump, *mmp; mp = vp->v_mount; if (mp == NULL) return; MNT_ILOCK(mp); if (TAILQ_EMPTY(&mp->mnt_uppers)) goto unlock; MNT_IUNLOCK(mp); mmp = malloc(sizeof(struct mount), M_TEMP, M_WAITOK | M_ZERO); mmp->mnt_op = &vgonel_vfsops; mmp->mnt_kern_flag |= MNTK_MARKER; MNT_ILOCK(mp); mp->mnt_kern_flag |= MNTK_VGONE_UPPER; for (ump = TAILQ_FIRST(&mp->mnt_uppers); ump != NULL;) { if ((ump->mnt_kern_flag & MNTK_MARKER) != 0) { ump = TAILQ_NEXT(ump, mnt_upper_link); continue; } TAILQ_INSERT_AFTER(&mp->mnt_uppers, ump, mmp, mnt_upper_link); MNT_IUNLOCK(mp); switch (event) { case VFS_NOTIFY_UPPER_RECLAIM: VFS_RECLAIM_LOWERVP(ump, vp); break; case VFS_NOTIFY_UPPER_UNLINK: VFS_UNLINK_LOWERVP(ump, vp); break; default: KASSERT(0, ("invalid event %d", event)); break; } MNT_ILOCK(mp); ump = TAILQ_NEXT(mmp, mnt_upper_link); TAILQ_REMOVE(&mp->mnt_uppers, mmp, mnt_upper_link); } free(mmp, M_TEMP); mp->mnt_kern_flag &= ~MNTK_VGONE_UPPER; if ((mp->mnt_kern_flag & MNTK_VGONE_WAITER) != 0) { mp->mnt_kern_flag &= ~MNTK_VGONE_WAITER; wakeup(&mp->mnt_uppers); } unlock: MNT_IUNLOCK(mp); } /* * vgone, with the vp interlock held. */ static void vgonel(struct vnode *vp) { struct thread *td; int oweinact; int active; struct mount *mp; ASSERT_VOP_ELOCKED(vp, "vgonel"); ASSERT_VI_LOCKED(vp, "vgonel"); VNASSERT(vp->v_holdcnt, vp, ("vgonel: vp %p has no reference.", vp)); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); td = curthread; /* * Don't vgonel if we're already doomed. */ if (vp->v_iflag & VI_DOOMED) return; vp->v_iflag |= VI_DOOMED; /* * Check to see if the vnode is in use. If so, we have to call * VOP_CLOSE() and VOP_INACTIVE(). */ active = vp->v_usecount; oweinact = (vp->v_iflag & VI_OWEINACT); VI_UNLOCK(vp); vfs_notify_upper(vp, VFS_NOTIFY_UPPER_RECLAIM); /* * If purging an active vnode, it must be closed and * deactivated before being reclaimed. */ if (active) VOP_CLOSE(vp, FNONBLOCK, NOCRED, td); if (oweinact || active) { VI_LOCK(vp); if ((vp->v_iflag & VI_DOINGINACT) == 0) vinactive(vp, td); VI_UNLOCK(vp); } if (vp->v_type == VSOCK) vfs_unp_reclaim(vp); /* * Clean out any buffers associated with the vnode. * If the flush fails, just toss the buffers. */ mp = NULL; if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd)) (void) vn_start_secondary_write(vp, &mp, V_WAIT); if (vinvalbuf(vp, V_SAVE, 0, 0) != 0) { while (vinvalbuf(vp, 0, 0, 0) != 0) ; } BO_LOCK(&vp->v_bufobj); KASSERT(TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd) && vp->v_bufobj.bo_dirty.bv_cnt == 0 && TAILQ_EMPTY(&vp->v_bufobj.bo_clean.bv_hd) && vp->v_bufobj.bo_clean.bv_cnt == 0, ("vp %p bufobj not invalidated", vp)); /* * For VMIO bufobj, BO_DEAD is set in vm_object_terminate() * after the object's page queue is flushed. */ if (vp->v_bufobj.bo_object == NULL) vp->v_bufobj.bo_flag |= BO_DEAD; BO_UNLOCK(&vp->v_bufobj); /* * Reclaim the vnode. */ if (VOP_RECLAIM(vp, td)) panic("vgone: cannot reclaim"); if (mp != NULL) vn_finished_secondary_write(mp); VNASSERT(vp->v_object == NULL, vp, ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag)); /* * Clear the advisory locks and wake up waiting threads. */ (void)VOP_ADVLOCKPURGE(vp); vp->v_lockf = NULL; /* * Delete from old mount point vnode list. */ delmntque(vp); cache_purge(vp); /* * Done with purge, reset to the standard lock and invalidate * the vnode. */ VI_LOCK(vp); vp->v_vnlock = &vp->v_lock; vp->v_op = &dead_vnodeops; vp->v_tag = "none"; vp->v_type = VBAD; } /* * Calculate the total number of references to a special device. */ int vcount(struct vnode *vp) { int count; dev_lock(); count = vp->v_rdev->si_usecount; dev_unlock(); return (count); } /* * Same as above, but using the struct cdev *as argument */ int count_dev(struct cdev *dev) { int count; dev_lock(); count = dev->si_usecount; dev_unlock(); return(count); } /* * Print out a description of a vnode. */ static char *typename[] = {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD", "VMARKER"}; void vn_printf(struct vnode *vp, const char *fmt, ...) { va_list ap; char buf[256], buf2[16]; u_long flags; va_start(ap, fmt); vprintf(fmt, ap); va_end(ap); printf("%p: ", (void *)vp); printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]); printf(" usecount %d, writecount %d, refcount %d", vp->v_usecount, vp->v_writecount, vp->v_holdcnt); switch (vp->v_type) { case VDIR: printf(" mountedhere %p\n", vp->v_mountedhere); break; case VCHR: printf(" rdev %p\n", vp->v_rdev); break; case VSOCK: printf(" socket %p\n", vp->v_unpcb); break; case VFIFO: printf(" fifoinfo %p\n", vp->v_fifoinfo); break; default: printf("\n"); break; } buf[0] = '\0'; buf[1] = '\0'; if (vp->v_vflag & VV_ROOT) strlcat(buf, "|VV_ROOT", sizeof(buf)); if (vp->v_vflag & VV_ISTTY) strlcat(buf, "|VV_ISTTY", sizeof(buf)); if (vp->v_vflag & VV_NOSYNC) strlcat(buf, "|VV_NOSYNC", sizeof(buf)); if (vp->v_vflag & VV_ETERNALDEV) strlcat(buf, "|VV_ETERNALDEV", sizeof(buf)); if (vp->v_vflag & VV_CACHEDLABEL) strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf)); if (vp->v_vflag & VV_COPYONWRITE) strlcat(buf, "|VV_COPYONWRITE", sizeof(buf)); if (vp->v_vflag & VV_SYSTEM) strlcat(buf, "|VV_SYSTEM", sizeof(buf)); if (vp->v_vflag & VV_PROCDEP) strlcat(buf, "|VV_PROCDEP", sizeof(buf)); if (vp->v_vflag & VV_NOKNOTE) strlcat(buf, "|VV_NOKNOTE", sizeof(buf)); if (vp->v_vflag & VV_DELETED) strlcat(buf, "|VV_DELETED", sizeof(buf)); if (vp->v_vflag & VV_MD) strlcat(buf, "|VV_MD", sizeof(buf)); if (vp->v_vflag & VV_FORCEINSMQ) strlcat(buf, "|VV_FORCEINSMQ", sizeof(buf)); flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | VV_ETERNALDEV | VV_CACHEDLABEL | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP | VV_NOKNOTE | VV_DELETED | VV_MD | VV_FORCEINSMQ); if (flags != 0) { snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags); strlcat(buf, buf2, sizeof(buf)); } if (vp->v_iflag & VI_MOUNT) strlcat(buf, "|VI_MOUNT", sizeof(buf)); if (vp->v_iflag & VI_DOOMED) strlcat(buf, "|VI_DOOMED", sizeof(buf)); if (vp->v_iflag & VI_FREE) strlcat(buf, "|VI_FREE", sizeof(buf)); if (vp->v_iflag & VI_ACTIVE) strlcat(buf, "|VI_ACTIVE", sizeof(buf)); if (vp->v_iflag & VI_DOINGINACT) strlcat(buf, "|VI_DOINGINACT", sizeof(buf)); if (vp->v_iflag & VI_OWEINACT) strlcat(buf, "|VI_OWEINACT", sizeof(buf)); flags = vp->v_iflag & ~(VI_MOUNT | VI_DOOMED | VI_FREE | VI_ACTIVE | VI_DOINGINACT | VI_OWEINACT); if (flags != 0) { snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags); strlcat(buf, buf2, sizeof(buf)); } printf(" flags (%s)\n", buf + 1); if (mtx_owned(VI_MTX(vp))) printf(" VI_LOCKed"); if (vp->v_object != NULL) printf(" v_object %p ref %d pages %d " "cleanbuf %d dirtybuf %d\n", vp->v_object, vp->v_object->ref_count, vp->v_object->resident_page_count, vp->v_bufobj.bo_clean.bv_cnt, vp->v_bufobj.bo_dirty.bv_cnt); printf(" "); lockmgr_printinfo(vp->v_vnlock); if (vp->v_data != NULL) VOP_PRINT(vp); } #ifdef DDB /* * List all of the locked vnodes in the system. * Called when debugging the kernel. */ DB_SHOW_COMMAND(lockedvnods, lockedvnodes) { struct mount *mp; struct vnode *vp; /* * Note: because this is DDB, we can't obey the locking semantics * for these structures, which means we could catch an inconsistent * state and dereference a nasty pointer. Not much to be done * about that. */ db_printf("Locked vnodes\n"); TAILQ_FOREACH(mp, &mountlist, mnt_list) { TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { if (vp->v_type != VMARKER && VOP_ISLOCKED(vp)) vn_printf(vp, "vnode "); } } } /* * Show details about the given vnode. */ DB_SHOW_COMMAND(vnode, db_show_vnode) { struct vnode *vp; if (!have_addr) return; vp = (struct vnode *)addr; vn_printf(vp, "vnode "); } /* * Show details about the given mount point. */ DB_SHOW_COMMAND(mount, db_show_mount) { struct mount *mp; struct vfsopt *opt; struct statfs *sp; struct vnode *vp; char buf[512]; uint64_t mflags; u_int flags; if (!have_addr) { /* No address given, print short info about all mount points. */ TAILQ_FOREACH(mp, &mountlist, mnt_list) { db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename); if (db_pager_quit) break; } db_printf("\nMore info: show mount \n"); return; } mp = (struct mount *)addr; db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename); buf[0] = '\0'; mflags = mp->mnt_flag; #define MNT_FLAG(flag) do { \ if (mflags & (flag)) { \ if (buf[0] != '\0') \ strlcat(buf, ", ", sizeof(buf)); \ strlcat(buf, (#flag) + 4, sizeof(buf)); \ mflags &= ~(flag); \ } \ } while (0) MNT_FLAG(MNT_RDONLY); MNT_FLAG(MNT_SYNCHRONOUS); MNT_FLAG(MNT_NOEXEC); MNT_FLAG(MNT_NOSUID); MNT_FLAG(MNT_NFS4ACLS); MNT_FLAG(MNT_UNION); MNT_FLAG(MNT_ASYNC); MNT_FLAG(MNT_SUIDDIR); MNT_FLAG(MNT_SOFTDEP); MNT_FLAG(MNT_NOSYMFOLLOW); MNT_FLAG(MNT_GJOURNAL); MNT_FLAG(MNT_MULTILABEL); MNT_FLAG(MNT_ACLS); MNT_FLAG(MNT_NOATIME); MNT_FLAG(MNT_NOCLUSTERR); MNT_FLAG(MNT_NOCLUSTERW); MNT_FLAG(MNT_SUJ); MNT_FLAG(MNT_EXRDONLY); MNT_FLAG(MNT_EXPORTED); MNT_FLAG(MNT_DEFEXPORTED); MNT_FLAG(MNT_EXPORTANON); MNT_FLAG(MNT_EXKERB); MNT_FLAG(MNT_EXPUBLIC); MNT_FLAG(MNT_LOCAL); MNT_FLAG(MNT_QUOTA); MNT_FLAG(MNT_ROOTFS); MNT_FLAG(MNT_USER); MNT_FLAG(MNT_IGNORE); MNT_FLAG(MNT_UPDATE); MNT_FLAG(MNT_DELEXPORT); MNT_FLAG(MNT_RELOAD); MNT_FLAG(MNT_FORCE); MNT_FLAG(MNT_SNAPSHOT); MNT_FLAG(MNT_BYFSID); #undef MNT_FLAG if (mflags != 0) { if (buf[0] != '\0') strlcat(buf, ", ", sizeof(buf)); snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), "0x%016jx", mflags); } db_printf(" mnt_flag = %s\n", buf); buf[0] = '\0'; flags = mp->mnt_kern_flag; #define MNT_KERN_FLAG(flag) do { \ if (flags & (flag)) { \ if (buf[0] != '\0') \ strlcat(buf, ", ", sizeof(buf)); \ strlcat(buf, (#flag) + 5, sizeof(buf)); \ flags &= ~(flag); \ } \ } while (0) MNT_KERN_FLAG(MNTK_UNMOUNTF); MNT_KERN_FLAG(MNTK_ASYNC); MNT_KERN_FLAG(MNTK_SOFTDEP); MNT_KERN_FLAG(MNTK_NOINSMNTQ); MNT_KERN_FLAG(MNTK_DRAINING); MNT_KERN_FLAG(MNTK_REFEXPIRE); MNT_KERN_FLAG(MNTK_EXTENDED_SHARED); MNT_KERN_FLAG(MNTK_SHARED_WRITES); MNT_KERN_FLAG(MNTK_NO_IOPF); MNT_KERN_FLAG(MNTK_VGONE_UPPER); MNT_KERN_FLAG(MNTK_VGONE_WAITER); MNT_KERN_FLAG(MNTK_LOOKUP_EXCL_DOTDOT); MNT_KERN_FLAG(MNTK_MARKER); MNT_KERN_FLAG(MNTK_USES_BCACHE); MNT_KERN_FLAG(MNTK_NOASYNC); MNT_KERN_FLAG(MNTK_UNMOUNT); MNT_KERN_FLAG(MNTK_MWAIT); MNT_KERN_FLAG(MNTK_SUSPEND); MNT_KERN_FLAG(MNTK_SUSPEND2); MNT_KERN_FLAG(MNTK_SUSPENDED); MNT_KERN_FLAG(MNTK_LOOKUP_SHARED); MNT_KERN_FLAG(MNTK_NOKNOTE); #undef MNT_KERN_FLAG if (flags != 0) { if (buf[0] != '\0') strlcat(buf, ", ", sizeof(buf)); snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), "0x%08x", flags); } db_printf(" mnt_kern_flag = %s\n", buf); db_printf(" mnt_opt = "); opt = TAILQ_FIRST(mp->mnt_opt); if (opt != NULL) { db_printf("%s", opt->name); opt = TAILQ_NEXT(opt, link); while (opt != NULL) { db_printf(", %s", opt->name); opt = TAILQ_NEXT(opt, link); } } db_printf("\n"); sp = &mp->mnt_stat; db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx " "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju " "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju " "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n", (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags, (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize, (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree, (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files, (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites, (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads, (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax, (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]); db_printf(" mnt_cred = { uid=%u ruid=%u", (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid); if (jailed(mp->mnt_cred)) db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id); db_printf(" }\n"); db_printf(" mnt_ref = %d\n", mp->mnt_ref); db_printf(" mnt_gen = %d\n", mp->mnt_gen); db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize); db_printf(" mnt_activevnodelistsize = %d\n", mp->mnt_activevnodelistsize); db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount); db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen); db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max); db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed); db_printf(" mnt_lockref = %d\n", mp->mnt_lockref); db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes); db_printf(" mnt_secondary_accwrites = %d\n", mp->mnt_secondary_accwrites); db_printf(" mnt_gjprovider = %s\n", mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL"); db_printf("\n\nList of active vnodes\n"); TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) { if (vp->v_type != VMARKER) { vn_printf(vp, "vnode "); if (db_pager_quit) break; } } db_printf("\n\nList of inactive vnodes\n"); TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) { vn_printf(vp, "vnode "); if (db_pager_quit) break; } } } #endif /* DDB */ /* * Fill in a struct xvfsconf based on a struct vfsconf. */ static int vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp) { struct xvfsconf xvfsp; bzero(&xvfsp, sizeof(xvfsp)); strcpy(xvfsp.vfc_name, vfsp->vfc_name); xvfsp.vfc_typenum = vfsp->vfc_typenum; xvfsp.vfc_refcount = vfsp->vfc_refcount; xvfsp.vfc_flags = vfsp->vfc_flags; /* * These are unused in userland, we keep them * to not break binary compatibility. */ xvfsp.vfc_vfsops = NULL; xvfsp.vfc_next = NULL; return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp))); } #ifdef COMPAT_FREEBSD32 struct xvfsconf32 { uint32_t vfc_vfsops; char vfc_name[MFSNAMELEN]; int32_t vfc_typenum; int32_t vfc_refcount; int32_t vfc_flags; uint32_t vfc_next; }; static int vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp) { struct xvfsconf32 xvfsp; bzero(&xvfsp, sizeof(xvfsp)); strcpy(xvfsp.vfc_name, vfsp->vfc_name); xvfsp.vfc_typenum = vfsp->vfc_typenum; xvfsp.vfc_refcount = vfsp->vfc_refcount; xvfsp.vfc_flags = vfsp->vfc_flags; return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp))); } #endif /* * Top level filesystem related information gathering. */ static int sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS) { struct vfsconf *vfsp; int error; error = 0; vfsconf_slock(); TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { #ifdef COMPAT_FREEBSD32 if (req->flags & SCTL_MASK32) error = vfsconf2x32(req, vfsp); else #endif error = vfsconf2x(req, vfsp); if (error) break; } vfsconf_sunlock(); return (error); } SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, sysctl_vfs_conflist, "S,xvfsconf", "List of all configured filesystems"); #ifndef BURN_BRIDGES static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS); static int vfs_sysctl(SYSCTL_HANDLER_ARGS) { int *name = (int *)arg1 - 1; /* XXX */ u_int namelen = arg2 + 1; /* XXX */ struct vfsconf *vfsp; log(LOG_WARNING, "userland calling deprecated sysctl, " "please rebuild world\n"); #if 1 || defined(COMPAT_PRELITE2) /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ if (namelen == 1) return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); #endif switch (name[1]) { case VFS_MAXTYPENUM: if (namelen != 2) return (ENOTDIR); return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); case VFS_CONF: if (namelen != 3) return (ENOTDIR); /* overloaded */ vfsconf_slock(); TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { if (vfsp->vfc_typenum == name[2]) break; } vfsconf_sunlock(); if (vfsp == NULL) return (EOPNOTSUPP); #ifdef COMPAT_FREEBSD32 if (req->flags & SCTL_MASK32) return (vfsconf2x32(req, vfsp)); else #endif return (vfsconf2x(req, vfsp)); } return (EOPNOTSUPP); } static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP | CTLFLAG_MPSAFE, vfs_sysctl, "Generic filesystem"); #if 1 || defined(COMPAT_PRELITE2) static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) { int error; struct vfsconf *vfsp; struct ovfsconf ovfs; vfsconf_slock(); TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { bzero(&ovfs, sizeof(ovfs)); ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ strcpy(ovfs.vfc_name, vfsp->vfc_name); ovfs.vfc_index = vfsp->vfc_typenum; ovfs.vfc_refcount = vfsp->vfc_refcount; ovfs.vfc_flags = vfsp->vfc_flags; error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); if (error != 0) { vfsconf_sunlock(); return (error); } } vfsconf_sunlock(); return (0); } #endif /* 1 || COMPAT_PRELITE2 */ #endif /* !BURN_BRIDGES */ #define KINFO_VNODESLOP 10 #ifdef notyet /* * Dump vnode list (via sysctl). */ /* ARGSUSED */ static int sysctl_vnode(SYSCTL_HANDLER_ARGS) { struct xvnode *xvn; struct mount *mp; struct vnode *vp; int error, len, n; /* * Stale numvnodes access is not fatal here. */ req->lock = 0; len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn; if (!req->oldptr) /* Make an estimate */ return (SYSCTL_OUT(req, 0, len)); error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK); n = 0; mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) continue; MNT_ILOCK(mp); TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { if (n == len) break; vref(vp); xvn[n].xv_size = sizeof *xvn; xvn[n].xv_vnode = vp; xvn[n].xv_id = 0; /* XXX compat */ #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field XV_COPY(usecount); XV_COPY(writecount); XV_COPY(holdcnt); XV_COPY(mount); XV_COPY(numoutput); XV_COPY(type); #undef XV_COPY xvn[n].xv_flag = vp->v_vflag; switch (vp->v_type) { case VREG: case VDIR: case VLNK: break; case VBLK: case VCHR: if (vp->v_rdev == NULL) { vrele(vp); continue; } xvn[n].xv_dev = dev2udev(vp->v_rdev); break; case VSOCK: xvn[n].xv_socket = vp->v_socket; break; case VFIFO: xvn[n].xv_fifo = vp->v_fifoinfo; break; case VNON: case VBAD: default: /* shouldn't happen? */ vrele(vp); continue; } vrele(vp); ++n; } MNT_IUNLOCK(mp); mtx_lock(&mountlist_mtx); vfs_unbusy(mp); if (n == len) break; } mtx_unlock(&mountlist_mtx); error = SYSCTL_OUT(req, xvn, n * sizeof *xvn); free(xvn, M_TEMP); return (error); } SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 0, sysctl_vnode, "S,xvnode", ""); #endif static void unmount_or_warn(struct mount *mp) { int error; error = dounmount(mp, MNT_FORCE, curthread); if (error != 0) { printf("unmount of %s failed (", mp->mnt_stat.f_mntonname); if (error == EBUSY) printf("BUSY)\n"); else printf("%d)\n", error); } } /* * Unmount all filesystems. The list is traversed in reverse order * of mounting to avoid dependencies. */ void vfs_unmountall(void) { struct mount *mp, *tmp; CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__); /* * Since this only runs when rebooting, it is not interlocked. */ TAILQ_FOREACH_REVERSE_SAFE(mp, &mountlist, mntlist, mnt_list, tmp) { vfs_ref(mp); /* * Forcibly unmounting "/dev" before "/" would prevent clean * unmount of the latter. */ if (mp == rootdevmp) continue; unmount_or_warn(mp); } if (rootdevmp != NULL) unmount_or_warn(rootdevmp); } /* * perform msync on all vnodes under a mount point * the mount point must be locked. */ void vfs_msync(struct mount *mp, int flags) { struct vnode *vp, *mvp; struct vm_object *obj; CTR2(KTR_VFS, "%s: mp %p", __func__, mp); vnlru_return_batch(mp); MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) { obj = vp->v_object; if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 && (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) { if (!vget(vp, LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK, curthread)) { if (vp->v_vflag & VV_NOSYNC) { /* unlinked */ vput(vp); continue; } obj = vp->v_object; if (obj != NULL) { VM_OBJECT_WLOCK(obj); vm_object_page_clean(obj, 0, 0, flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC); VM_OBJECT_WUNLOCK(obj); } vput(vp); } } else VI_UNLOCK(vp); } } static void destroy_vpollinfo_free(struct vpollinfo *vi) { knlist_destroy(&vi->vpi_selinfo.si_note); mtx_destroy(&vi->vpi_lock); uma_zfree(vnodepoll_zone, vi); } static void destroy_vpollinfo(struct vpollinfo *vi) { knlist_clear(&vi->vpi_selinfo.si_note, 1); seldrain(&vi->vpi_selinfo); destroy_vpollinfo_free(vi); } /* * Initialize per-vnode helper structure to hold poll-related state. */ void v_addpollinfo(struct vnode *vp) { struct vpollinfo *vi; if (vp->v_pollinfo != NULL) return; vi = uma_zalloc(vnodepoll_zone, M_WAITOK | M_ZERO); mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF); knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock, vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked); VI_LOCK(vp); if (vp->v_pollinfo != NULL) { VI_UNLOCK(vp); destroy_vpollinfo_free(vi); return; } vp->v_pollinfo = vi; VI_UNLOCK(vp); } /* * Record a process's interest in events which might happen to * a vnode. Because poll uses the historic select-style interface * internally, this routine serves as both the ``check for any * pending events'' and the ``record my interest in future events'' * functions. (These are done together, while the lock is held, * to avoid race conditions.) */ int vn_pollrecord(struct vnode *vp, struct thread *td, int events) { v_addpollinfo(vp); mtx_lock(&vp->v_pollinfo->vpi_lock); if (vp->v_pollinfo->vpi_revents & events) { /* * This leaves events we are not interested * in available for the other process which * which presumably had requested them * (otherwise they would never have been * recorded). */ events &= vp->v_pollinfo->vpi_revents; vp->v_pollinfo->vpi_revents &= ~events; mtx_unlock(&vp->v_pollinfo->vpi_lock); return (events); } vp->v_pollinfo->vpi_events |= events; selrecord(td, &vp->v_pollinfo->vpi_selinfo); mtx_unlock(&vp->v_pollinfo->vpi_lock); return (0); } /* * Routine to create and manage a filesystem syncer vnode. */ #define sync_close ((int (*)(struct vop_close_args *))nullop) static int sync_fsync(struct vop_fsync_args *); static int sync_inactive(struct vop_inactive_args *); static int sync_reclaim(struct vop_reclaim_args *); static struct vop_vector sync_vnodeops = { .vop_bypass = VOP_EOPNOTSUPP, .vop_close = sync_close, /* close */ .vop_fsync = sync_fsync, /* fsync */ .vop_inactive = sync_inactive, /* inactive */ .vop_reclaim = sync_reclaim, /* reclaim */ .vop_lock1 = vop_stdlock, /* lock */ .vop_unlock = vop_stdunlock, /* unlock */ .vop_islocked = vop_stdislocked, /* islocked */ }; /* * Create a new filesystem syncer vnode for the specified mount point. */ void vfs_allocate_syncvnode(struct mount *mp) { struct vnode *vp; struct bufobj *bo; static long start, incr, next; int error; /* Allocate a new vnode */ error = getnewvnode("syncer", mp, &sync_vnodeops, &vp); if (error != 0) panic("vfs_allocate_syncvnode: getnewvnode() failed"); vp->v_type = VNON; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); vp->v_vflag |= VV_FORCEINSMQ; error = insmntque(vp, mp); if (error != 0) panic("vfs_allocate_syncvnode: insmntque() failed"); vp->v_vflag &= ~VV_FORCEINSMQ; VOP_UNLOCK(vp, 0); /* * Place the vnode onto the syncer worklist. We attempt to * scatter them about on the list so that they will go off * at evenly distributed times even if all the filesystems * are mounted at once. */ next += incr; if (next == 0 || next > syncer_maxdelay) { start /= 2; incr /= 2; if (start == 0) { start = syncer_maxdelay / 2; incr = syncer_maxdelay; } next = start; } bo = &vp->v_bufobj; BO_LOCK(bo); vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0); /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */ mtx_lock(&sync_mtx); sync_vnode_count++; if (mp->mnt_syncer == NULL) { mp->mnt_syncer = vp; vp = NULL; } mtx_unlock(&sync_mtx); BO_UNLOCK(bo); if (vp != NULL) { vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); vgone(vp); vput(vp); } } void vfs_deallocate_syncvnode(struct mount *mp) { struct vnode *vp; mtx_lock(&sync_mtx); vp = mp->mnt_syncer; if (vp != NULL) mp->mnt_syncer = NULL; mtx_unlock(&sync_mtx); if (vp != NULL) vrele(vp); } /* * Do a lazy sync of the filesystem. */ static int sync_fsync(struct vop_fsync_args *ap) { struct vnode *syncvp = ap->a_vp; struct mount *mp = syncvp->v_mount; int error, save; struct bufobj *bo; /* * We only need to do something if this is a lazy evaluation. */ if (ap->a_waitfor != MNT_LAZY) return (0); /* * Move ourselves to the back of the sync list. */ bo = &syncvp->v_bufobj; BO_LOCK(bo); vn_syncer_add_to_worklist(bo, syncdelay); BO_UNLOCK(bo); /* * Walk the list of vnodes pushing all that are dirty and * not already on the sync list. */ if (vfs_busy(mp, MBF_NOWAIT) != 0) return (0); if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) { vfs_unbusy(mp); return (0); } save = curthread_pflags_set(TDP_SYNCIO); vfs_msync(mp, MNT_NOWAIT); error = VFS_SYNC(mp, MNT_LAZY); curthread_pflags_restore(save); vn_finished_write(mp); vfs_unbusy(mp); return (error); } /* * The syncer vnode is no referenced. */ static int sync_inactive(struct vop_inactive_args *ap) { vgone(ap->a_vp); return (0); } /* * The syncer vnode is no longer needed and is being decommissioned. * * Modifications to the worklist must be protected by sync_mtx. */ static int sync_reclaim(struct vop_reclaim_args *ap) { struct vnode *vp = ap->a_vp; struct bufobj *bo; bo = &vp->v_bufobj; BO_LOCK(bo); mtx_lock(&sync_mtx); if (vp->v_mount->mnt_syncer == vp) vp->v_mount->mnt_syncer = NULL; if (bo->bo_flag & BO_ONWORKLST) { LIST_REMOVE(bo, bo_synclist); syncer_worklist_len--; sync_vnode_count--; bo->bo_flag &= ~BO_ONWORKLST; } mtx_unlock(&sync_mtx); BO_UNLOCK(bo); return (0); } /* * Check if vnode represents a disk device */ int vn_isdisk(struct vnode *vp, int *errp) { int error; if (vp->v_type != VCHR) { error = ENOTBLK; goto out; } error = 0; dev_lock(); if (vp->v_rdev == NULL) error = ENXIO; else if (vp->v_rdev->si_devsw == NULL) error = ENXIO; else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK)) error = ENOTBLK; dev_unlock(); out: if (errp != NULL) *errp = error; return (error == 0); } /* * Common filesystem object access control check routine. Accepts a * vnode's type, "mode", uid and gid, requested access mode, credentials, * and optional call-by-reference privused argument allowing vaccess() * to indicate to the caller whether privilege was used to satisfy the * request (obsoleted). Returns 0 on success, or an errno on failure. */ int vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid, accmode_t accmode, struct ucred *cred, int *privused) { accmode_t dac_granted; accmode_t priv_granted; KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0, ("invalid bit in accmode")); KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE), ("VAPPEND without VWRITE")); /* * Look for a normal, non-privileged way to access the file/directory * as requested. If it exists, go with that. */ if (privused != NULL) *privused = 0; dac_granted = 0; /* Check the owner. */ if (cred->cr_uid == file_uid) { dac_granted |= VADMIN; if (file_mode & S_IXUSR) dac_granted |= VEXEC; if (file_mode & S_IRUSR) dac_granted |= VREAD; if (file_mode & S_IWUSR) dac_granted |= (VWRITE | VAPPEND); if ((accmode & dac_granted) == accmode) return (0); goto privcheck; } /* Otherwise, check the groups (first match) */ if (groupmember(file_gid, cred)) { if (file_mode & S_IXGRP) dac_granted |= VEXEC; if (file_mode & S_IRGRP) dac_granted |= VREAD; if (file_mode & S_IWGRP) dac_granted |= (VWRITE | VAPPEND); if ((accmode & dac_granted) == accmode) return (0); goto privcheck; } /* Otherwise, check everyone else. */ if (file_mode & S_IXOTH) dac_granted |= VEXEC; if (file_mode & S_IROTH) dac_granted |= VREAD; if (file_mode & S_IWOTH) dac_granted |= (VWRITE | VAPPEND); if ((accmode & dac_granted) == accmode) return (0); privcheck: /* * Build a privilege mask to determine if the set of privileges * satisfies the requirements when combined with the granted mask * from above. For each privilege, if the privilege is required, * bitwise or the request type onto the priv_granted mask. */ priv_granted = 0; if (type == VDIR) { /* * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC * requests, instead of PRIV_VFS_EXEC. */ if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) && !priv_check_cred(cred, PRIV_VFS_LOOKUP)) priv_granted |= VEXEC; } else { /* * Ensure that at least one execute bit is on. Otherwise, * a privileged user will always succeed, and we don't want * this to happen unless the file really is executable. */ if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) && (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 && !priv_check_cred(cred, PRIV_VFS_EXEC)) priv_granted |= VEXEC; } if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) && !priv_check_cred(cred, PRIV_VFS_READ)) priv_granted |= VREAD; if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) && !priv_check_cred(cred, PRIV_VFS_WRITE)) priv_granted |= (VWRITE | VAPPEND); if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) && !priv_check_cred(cred, PRIV_VFS_ADMIN)) priv_granted |= VADMIN; if ((accmode & (priv_granted | dac_granted)) == accmode) { /* XXX audit: privilege used */ if (privused != NULL) *privused = 1; return (0); } return ((accmode & VADMIN) ? EPERM : EACCES); } /* * Credential check based on process requesting service, and per-attribute * permissions. */ int extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred, struct thread *td, accmode_t accmode) { /* * Kernel-invoked always succeeds. */ if (cred == NOCRED) return (0); /* * Do not allow privileged processes in jail to directly manipulate * system attributes. */ switch (attrnamespace) { case EXTATTR_NAMESPACE_SYSTEM: /* Potentially should be: return (EPERM); */ return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM)); case EXTATTR_NAMESPACE_USER: return (VOP_ACCESS(vp, accmode, cred, td)); default: return (EPERM); } } #ifdef DEBUG_VFS_LOCKS /* * This only exists to suppress warnings from unlocked specfs accesses. It is * no longer ok to have an unlocked VFS. */ #define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL || \ (vp)->v_type == VCHR || (vp)->v_type == VBAD) int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, "Drop into debugger on lock violation"); int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, "Check for interlock across VOPs"); int vfs_badlock_print = 1; /* Print lock violations. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, "Print lock violations"); int vfs_badlock_vnode = 1; /* Print vnode details on lock violations. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_vnode, CTLFLAG_RW, &vfs_badlock_vnode, 0, "Print vnode details on lock violations"); #ifdef KDB int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, "Print backtrace at lock violations"); #endif static void vfs_badlock(const char *msg, const char *str, struct vnode *vp) { #ifdef KDB if (vfs_badlock_backtrace) kdb_backtrace(); #endif if (vfs_badlock_vnode) vn_printf(vp, "vnode "); if (vfs_badlock_print) printf("%s: %p %s\n", str, (void *)vp, msg); if (vfs_badlock_ddb) kdb_enter(KDB_WHY_VFSLOCK, "lock violation"); } void assert_vi_locked(struct vnode *vp, const char *str) { if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp))) vfs_badlock("interlock is not locked but should be", str, vp); } void assert_vi_unlocked(struct vnode *vp, const char *str) { if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp))) vfs_badlock("interlock is locked but should not be", str, vp); } void assert_vop_locked(struct vnode *vp, const char *str) { int locked; if (!IGNORE_LOCK(vp)) { locked = VOP_ISLOCKED(vp); if (locked == 0 || locked == LK_EXCLOTHER) vfs_badlock("is not locked but should be", str, vp); } } void assert_vop_unlocked(struct vnode *vp, const char *str) { if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) vfs_badlock("is locked but should not be", str, vp); } void assert_vop_elocked(struct vnode *vp, const char *str) { if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) vfs_badlock("is not exclusive locked but should be", str, vp); } #endif /* DEBUG_VFS_LOCKS */ void vop_rename_fail(struct vop_rename_args *ap) { if (ap->a_tvp != NULL) vput(ap->a_tvp); if (ap->a_tdvp == ap->a_tvp) vrele(ap->a_tdvp); else vput(ap->a_tdvp); vrele(ap->a_fdvp); vrele(ap->a_fvp); } void vop_rename_pre(void *ap) { struct vop_rename_args *a = ap; #ifdef DEBUG_VFS_LOCKS if (a->a_tvp) ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME"); ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME"); ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME"); ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME"); /* Check the source (from). */ if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock && (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock)) ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked"); if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock) ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked"); /* Check the target. */ if (a->a_tvp) ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked"); ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked"); #endif if (a->a_tdvp != a->a_fdvp) vhold(a->a_fdvp); if (a->a_tvp != a->a_fvp) vhold(a->a_fvp); vhold(a->a_tdvp); if (a->a_tvp) vhold(a->a_tvp); } #ifdef DEBUG_VFS_LOCKS void vop_strategy_pre(void *ap) { struct vop_strategy_args *a; struct buf *bp; a = ap; bp = a->a_bp; /* * Cluster ops lock their component buffers but not the IO container. */ if ((bp->b_flags & B_CLUSTER) != 0) return; if (panicstr == NULL && !BUF_ISLOCKED(bp)) { if (vfs_badlock_print) printf( "VOP_STRATEGY: bp is not locked but should be\n"); if (vfs_badlock_ddb) kdb_enter(KDB_WHY_VFSLOCK, "lock violation"); } } void vop_lock_pre(void *ap) { struct vop_lock1_args *a = ap; if ((a->a_flags & LK_INTERLOCK) == 0) ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); else ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK"); } void vop_lock_post(void *ap, int rc) { struct vop_lock1_args *a = ap; ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); if (rc == 0 && (a->a_flags & LK_EXCLOTHER) == 0) ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK"); } void vop_unlock_pre(void *ap) { struct vop_unlock_args *a = ap; if (a->a_flags & LK_INTERLOCK) ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK"); ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK"); } void vop_unlock_post(void *ap, int rc) { struct vop_unlock_args *a = ap; if (a->a_flags & LK_INTERLOCK) ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK"); } #endif void vop_create_post(void *ap, int rc) { struct vop_create_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); } void vop_deleteextattr_post(void *ap, int rc) { struct vop_deleteextattr_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB); } void vop_link_post(void *ap, int rc) { struct vop_link_args *a = ap; if (!rc) { VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK); VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE); } } void vop_mkdir_post(void *ap, int rc) { struct vop_mkdir_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK); } void vop_mknod_post(void *ap, int rc) { struct vop_mknod_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); } void vop_reclaim_post(void *ap, int rc) { struct vop_reclaim_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_REVOKE); } void vop_remove_post(void *ap, int rc) { struct vop_remove_args *a = ap; if (!rc) { VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE); } } void vop_rename_post(void *ap, int rc) { struct vop_rename_args *a = ap; long hint; if (!rc) { hint = NOTE_WRITE; if (a->a_fdvp == a->a_tdvp) { if (a->a_tvp != NULL && a->a_tvp->v_type == VDIR) hint |= NOTE_LINK; VFS_KNOTE_UNLOCKED(a->a_fdvp, hint); VFS_KNOTE_UNLOCKED(a->a_tdvp, hint); } else { hint |= NOTE_EXTEND; if (a->a_fvp->v_type == VDIR) hint |= NOTE_LINK; VFS_KNOTE_UNLOCKED(a->a_fdvp, hint); if (a->a_fvp->v_type == VDIR && a->a_tvp != NULL && a->a_tvp->v_type == VDIR) hint &= ~NOTE_LINK; VFS_KNOTE_UNLOCKED(a->a_tdvp, hint); } VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME); if (a->a_tvp) VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE); } if (a->a_tdvp != a->a_fdvp) vdrop(a->a_fdvp); if (a->a_tvp != a->a_fvp) vdrop(a->a_fvp); vdrop(a->a_tdvp); if (a->a_tvp) vdrop(a->a_tvp); } void vop_rmdir_post(void *ap, int rc) { struct vop_rmdir_args *a = ap; if (!rc) { VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK); VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE); } } void vop_setattr_post(void *ap, int rc) { struct vop_setattr_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB); } void vop_setextattr_post(void *ap, int rc) { struct vop_setextattr_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB); } void vop_symlink_post(void *ap, int rc) { struct vop_symlink_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); } void vop_open_post(void *ap, int rc) { struct vop_open_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_OPEN); } void vop_close_post(void *ap, int rc) { struct vop_close_args *a = ap; if (!rc && (a->a_cred != NOCRED || /* filter out revokes */ (a->a_vp->v_iflag & VI_DOOMED) == 0)) { VFS_KNOTE_LOCKED(a->a_vp, (a->a_fflag & FWRITE) != 0 ? NOTE_CLOSE_WRITE : NOTE_CLOSE); } } void vop_read_post(void *ap, int rc) { struct vop_read_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_READ); } void vop_readdir_post(void *ap, int rc) { struct vop_readdir_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_READ); } static struct knlist fs_knlist; static void vfs_event_init(void *arg) { knlist_init_mtx(&fs_knlist, NULL); } /* XXX - correct order? */ SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL); void vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused) { KNOTE_UNLOCKED(&fs_knlist, event); } static int filt_fsattach(struct knote *kn); static void filt_fsdetach(struct knote *kn); static int filt_fsevent(struct knote *kn, long hint); struct filterops fs_filtops = { .f_isfd = 0, .f_attach = filt_fsattach, .f_detach = filt_fsdetach, .f_event = filt_fsevent }; static int filt_fsattach(struct knote *kn) { kn->kn_flags |= EV_CLEAR; knlist_add(&fs_knlist, kn, 0); return (0); } static void filt_fsdetach(struct knote *kn) { knlist_remove(&fs_knlist, kn, 0); } static int filt_fsevent(struct knote *kn, long hint) { kn->kn_fflags |= hint; return (kn->kn_fflags != 0); } static int sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS) { struct vfsidctl vc; int error; struct mount *mp; error = SYSCTL_IN(req, &vc, sizeof(vc)); if (error) return (error); if (vc.vc_vers != VFS_CTL_VERS1) return (EINVAL); mp = vfs_getvfs(&vc.vc_fsid); if (mp == NULL) return (ENOENT); /* ensure that a specific sysctl goes to the right filesystem. */ if (strcmp(vc.vc_fstypename, "*") != 0 && strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) { vfs_rel(mp); return (EINVAL); } VCTLTOREQ(&vc, req); error = VFS_SYSCTL(mp, vc.vc_op, req); vfs_rel(mp); return (error); } SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR, NULL, 0, sysctl_vfs_ctl, "", "Sysctl by fsid"); /* * Function to initialize a va_filerev field sensibly. * XXX: Wouldn't a random number make a lot more sense ?? */ u_quad_t init_va_filerev(void) { struct bintime bt; getbinuptime(&bt); return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL)); } static int filt_vfsread(struct knote *kn, long hint); static int filt_vfswrite(struct knote *kn, long hint); static int filt_vfsvnode(struct knote *kn, long hint); static void filt_vfsdetach(struct knote *kn); static struct filterops vfsread_filtops = { .f_isfd = 1, .f_detach = filt_vfsdetach, .f_event = filt_vfsread }; static struct filterops vfswrite_filtops = { .f_isfd = 1, .f_detach = filt_vfsdetach, .f_event = filt_vfswrite }; static struct filterops vfsvnode_filtops = { .f_isfd = 1, .f_detach = filt_vfsdetach, .f_event = filt_vfsvnode }; static void vfs_knllock(void *arg) { struct vnode *vp = arg; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); } static void vfs_knlunlock(void *arg) { struct vnode *vp = arg; VOP_UNLOCK(vp, 0); } static void vfs_knl_assert_locked(void *arg) { #ifdef DEBUG_VFS_LOCKS struct vnode *vp = arg; ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked"); #endif } static void vfs_knl_assert_unlocked(void *arg) { #ifdef DEBUG_VFS_LOCKS struct vnode *vp = arg; ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked"); #endif } int vfs_kqfilter(struct vop_kqfilter_args *ap) { struct vnode *vp = ap->a_vp; struct knote *kn = ap->a_kn; struct knlist *knl; switch (kn->kn_filter) { case EVFILT_READ: kn->kn_fop = &vfsread_filtops; break; case EVFILT_WRITE: kn->kn_fop = &vfswrite_filtops; break; case EVFILT_VNODE: kn->kn_fop = &vfsvnode_filtops; break; default: return (EINVAL); } kn->kn_hook = (caddr_t)vp; v_addpollinfo(vp); if (vp->v_pollinfo == NULL) return (ENOMEM); knl = &vp->v_pollinfo->vpi_selinfo.si_note; vhold(vp); knlist_add(knl, kn, 0); return (0); } /* * Detach knote from vnode */ static void filt_vfsdetach(struct knote *kn) { struct vnode *vp = (struct vnode *)kn->kn_hook; KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo")); knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0); vdrop(vp); } /*ARGSUSED*/ static int filt_vfsread(struct knote *kn, long hint) { struct vnode *vp = (struct vnode *)kn->kn_hook; struct vattr va; int res; /* * filesystem is gone, so set the EOF flag and schedule * the knote for deletion. */ if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) { VI_LOCK(vp); kn->kn_flags |= (EV_EOF | EV_ONESHOT); VI_UNLOCK(vp); return (1); } if (VOP_GETATTR(vp, &va, curthread->td_ucred)) return (0); VI_LOCK(vp); kn->kn_data = va.va_size - kn->kn_fp->f_offset; res = (kn->kn_sfflags & NOTE_FILE_POLL) != 0 || kn->kn_data != 0; VI_UNLOCK(vp); return (res); } /*ARGSUSED*/ static int filt_vfswrite(struct knote *kn, long hint) { struct vnode *vp = (struct vnode *)kn->kn_hook; VI_LOCK(vp); /* * filesystem is gone, so set the EOF flag and schedule * the knote for deletion. */ if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) kn->kn_flags |= (EV_EOF | EV_ONESHOT); kn->kn_data = 0; VI_UNLOCK(vp); return (1); } static int filt_vfsvnode(struct knote *kn, long hint) { struct vnode *vp = (struct vnode *)kn->kn_hook; int res; VI_LOCK(vp); if (kn->kn_sfflags & hint) kn->kn_fflags |= hint; if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) { kn->kn_flags |= EV_EOF; VI_UNLOCK(vp); return (1); } res = (kn->kn_fflags != 0); VI_UNLOCK(vp); return (res); } int vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off) { int error; if (dp->d_reclen > ap->a_uio->uio_resid) return (ENAMETOOLONG); error = uiomove(dp, dp->d_reclen, ap->a_uio); if (error) { if (ap->a_ncookies != NULL) { if (ap->a_cookies != NULL) free(ap->a_cookies, M_TEMP); ap->a_cookies = NULL; *ap->a_ncookies = 0; } return (error); } if (ap->a_ncookies == NULL) return (0); KASSERT(ap->a_cookies, ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!")); *ap->a_cookies = realloc(*ap->a_cookies, (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO); (*ap->a_cookies)[*ap->a_ncookies] = off; *ap->a_ncookies += 1; return (0); } /* * Mark for update the access time of the file if the filesystem * supports VOP_MARKATIME. This functionality is used by execve and * mmap, so we want to avoid the I/O implied by directly setting * va_atime for the sake of efficiency. */ void vfs_mark_atime(struct vnode *vp, struct ucred *cred) { struct mount *mp; mp = vp->v_mount; ASSERT_VOP_LOCKED(vp, "vfs_mark_atime"); if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) (void)VOP_MARKATIME(vp); } /* * The purpose of this routine is to remove granularity from accmode_t, * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE, * VADMIN and VAPPEND. * * If it returns 0, the caller is supposed to continue with the usual * access checks using 'accmode' as modified by this routine. If it * returns nonzero value, the caller is supposed to return that value * as errno. * * Note that after this routine runs, accmode may be zero. */ int vfs_unixify_accmode(accmode_t *accmode) { /* * There is no way to specify explicit "deny" rule using * file mode or POSIX.1e ACLs. */ if (*accmode & VEXPLICIT_DENY) { *accmode = 0; return (0); } /* * None of these can be translated into usual access bits. * Also, the common case for NFSv4 ACLs is to not contain * either of these bits. Caller should check for VWRITE * on the containing directory instead. */ if (*accmode & (VDELETE_CHILD | VDELETE)) return (EPERM); if (*accmode & VADMIN_PERMS) { *accmode &= ~VADMIN_PERMS; *accmode |= VADMIN; } /* * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL * or VSYNCHRONIZE using file mode or POSIX.1e ACL. */ *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE); return (0); } /* * These are helper functions for filesystems to traverse all * their vnodes. See MNT_VNODE_FOREACH_ALL() in sys/mount.h. * * This interface replaces MNT_VNODE_FOREACH. */ MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker"); struct vnode * __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp) { struct vnode *vp; if (should_yield()) kern_yield(PRI_USER); MNT_ILOCK(mp); KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); for (vp = TAILQ_NEXT(*mvp, v_nmntvnodes); vp != NULL; vp = TAILQ_NEXT(vp, v_nmntvnodes)) { /* Allow a racy peek at VI_DOOMED to save a lock acquisition. */ if (vp->v_type == VMARKER || (vp->v_iflag & VI_DOOMED) != 0) continue; VI_LOCK(vp); if ((vp->v_iflag & VI_DOOMED) != 0) { VI_UNLOCK(vp); continue; } break; } if (vp == NULL) { __mnt_vnode_markerfree_all(mvp, mp); /* MNT_IUNLOCK(mp); -- done in above function */ mtx_assert(MNT_MTX(mp), MA_NOTOWNED); return (NULL); } TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); MNT_IUNLOCK(mp); return (vp); } struct vnode * __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp) { struct vnode *vp; *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO); MNT_ILOCK(mp); MNT_REF(mp); (*mvp)->v_mount = mp; (*mvp)->v_type = VMARKER; TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { /* Allow a racy peek at VI_DOOMED to save a lock acquisition. */ if (vp->v_type == VMARKER || (vp->v_iflag & VI_DOOMED) != 0) continue; VI_LOCK(vp); if ((vp->v_iflag & VI_DOOMED) != 0) { VI_UNLOCK(vp); continue; } break; } if (vp == NULL) { MNT_REL(mp); MNT_IUNLOCK(mp); free(*mvp, M_VNODE_MARKER); *mvp = NULL; return (NULL); } TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); MNT_IUNLOCK(mp); return (vp); } void __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp) { if (*mvp == NULL) { MNT_IUNLOCK(mp); return; } mtx_assert(MNT_MTX(mp), MA_OWNED); KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); MNT_REL(mp); MNT_IUNLOCK(mp); free(*mvp, M_VNODE_MARKER); *mvp = NULL; } /* * These are helper functions for filesystems to traverse their * active vnodes. See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h */ static void mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp) { KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); MNT_ILOCK(mp); MNT_REL(mp); MNT_IUNLOCK(mp); free(*mvp, M_VNODE_MARKER); *mvp = NULL; } /* * Relock the mp mount vnode list lock with the vp vnode interlock in the * conventional lock order during mnt_vnode_next_active iteration. * * On entry, the mount vnode list lock is held and the vnode interlock is not. * The list lock is dropped and reacquired. On success, both locks are held. * On failure, the mount vnode list lock is held but the vnode interlock is * not, and the procedure may have yielded. */ static bool mnt_vnode_next_active_relock(struct vnode *mvp, struct mount *mp, struct vnode *vp) { const struct vnode *tmp; bool held, ret; VNASSERT(mvp->v_mount == mp && mvp->v_type == VMARKER && TAILQ_NEXT(mvp, v_actfreelist) != NULL, mvp, ("%s: bad marker", __func__)); VNASSERT(vp->v_mount == mp && vp->v_type != VMARKER, vp, ("%s: inappropriate vnode", __func__)); ASSERT_VI_UNLOCKED(vp, __func__); mtx_assert(&mp->mnt_listmtx, MA_OWNED); ret = false; TAILQ_REMOVE(&mp->mnt_activevnodelist, mvp, v_actfreelist); TAILQ_INSERT_BEFORE(vp, mvp, v_actfreelist); /* * Use a hold to prevent vp from disappearing while the mount vnode * list lock is dropped and reacquired. Normally a hold would be * acquired with vhold(), but that might try to acquire the vnode * interlock, which would be a LOR with the mount vnode list lock. */ held = refcount_acquire_if_not_zero(&vp->v_holdcnt); mtx_unlock(&mp->mnt_listmtx); if (!held) goto abort; VI_LOCK(vp); if (!refcount_release_if_not_last(&vp->v_holdcnt)) { vdropl(vp); goto abort; } mtx_lock(&mp->mnt_listmtx); /* * Determine whether the vnode is still the next one after the marker, * excepting any other markers. If the vnode has not been doomed by * vgone() then the hold should have ensured that it remained on the * active list. If it has been doomed but is still on the active list, * don't abort, but rather skip over it (avoid spinning on doomed * vnodes). */ tmp = mvp; do { tmp = TAILQ_NEXT(tmp, v_actfreelist); } while (tmp != NULL && tmp->v_type == VMARKER); if (tmp != vp) { mtx_unlock(&mp->mnt_listmtx); VI_UNLOCK(vp); goto abort; } ret = true; goto out; abort: maybe_yield(); mtx_lock(&mp->mnt_listmtx); out: if (ret) ASSERT_VI_LOCKED(vp, __func__); else ASSERT_VI_UNLOCKED(vp, __func__); mtx_assert(&mp->mnt_listmtx, MA_OWNED); return (ret); } static struct vnode * mnt_vnode_next_active(struct vnode **mvp, struct mount *mp) { struct vnode *vp, *nvp; mtx_assert(&mp->mnt_listmtx, MA_OWNED); KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); restart: vp = TAILQ_NEXT(*mvp, v_actfreelist); while (vp != NULL) { if (vp->v_type == VMARKER) { vp = TAILQ_NEXT(vp, v_actfreelist); continue; } /* * Try-lock because this is the wrong lock order. If that does * not succeed, drop the mount vnode list lock and try to * reacquire it and the vnode interlock in the right order. */ if (!VI_TRYLOCK(vp) && !mnt_vnode_next_active_relock(*mvp, mp, vp)) goto restart; KASSERT(vp->v_type != VMARKER, ("locked marker %p", vp)); KASSERT(vp->v_mount == mp || vp->v_mount == NULL, ("alien vnode on the active list %p %p", vp, mp)); if (vp->v_mount == mp && (vp->v_iflag & VI_DOOMED) == 0) break; nvp = TAILQ_NEXT(vp, v_actfreelist); VI_UNLOCK(vp); vp = nvp; } TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist); /* Check if we are done */ if (vp == NULL) { mtx_unlock(&mp->mnt_listmtx); mnt_vnode_markerfree_active(mvp, mp); return (NULL); } TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist); mtx_unlock(&mp->mnt_listmtx); ASSERT_VI_LOCKED(vp, "active iter"); KASSERT((vp->v_iflag & VI_ACTIVE) != 0, ("Non-active vp %p", vp)); return (vp); } struct vnode * __mnt_vnode_next_active(struct vnode **mvp, struct mount *mp) { if (should_yield()) kern_yield(PRI_USER); mtx_lock(&mp->mnt_listmtx); return (mnt_vnode_next_active(mvp, mp)); } struct vnode * __mnt_vnode_first_active(struct vnode **mvp, struct mount *mp) { struct vnode *vp; *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO); MNT_ILOCK(mp); MNT_REF(mp); MNT_IUNLOCK(mp); (*mvp)->v_type = VMARKER; (*mvp)->v_mount = mp; mtx_lock(&mp->mnt_listmtx); vp = TAILQ_FIRST(&mp->mnt_activevnodelist); if (vp == NULL) { mtx_unlock(&mp->mnt_listmtx); mnt_vnode_markerfree_active(mvp, mp); return (NULL); } TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist); return (mnt_vnode_next_active(mvp, mp)); } void __mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp) { if (*mvp == NULL) return; mtx_lock(&mp->mnt_listmtx); TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist); mtx_unlock(&mp->mnt_listmtx); mnt_vnode_markerfree_active(mvp, mp); } Index: head/sys/vm/swap_pager.c =================================================================== --- head/sys/vm/swap_pager.c (revision 351240) +++ head/sys/vm/swap_pager.c (revision 351241) @@ -1,2993 +1,2993 @@ /*- * SPDX-License-Identifier: BSD-4-Clause * * Copyright (c) 1998 Matthew Dillon, * Copyright (c) 1994 John S. Dyson * Copyright (c) 1990 University of Utah. * Copyright (c) 1982, 1986, 1989, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * the Systems Programming Group of the University of Utah Computer * Science Department. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by the University of * California, Berkeley and its contributors. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * New Swap System * Matthew Dillon * * Radix Bitmap 'blists'. * * - The new swapper uses the new radix bitmap code. This should scale * to arbitrarily small or arbitrarily large swap spaces and an almost * arbitrary degree of fragmentation. * * Features: * * - on the fly reallocation of swap during putpages. The new system * does not try to keep previously allocated swap blocks for dirty * pages. * * - on the fly deallocation of swap * * - No more garbage collection required. Unnecessarily allocated swap * blocks only exist for dirty vm_page_t's now and these are already * cycled (in a high-load system) by the pager. We also do on-the-fly * removal of invalidated swap blocks when a page is destroyed * or renamed. * * from: Utah $Hdr: swap_pager.c 1.4 91/04/30$ * * @(#)swap_pager.c 8.9 (Berkeley) 3/21/94 * @(#)vm_swap.c 8.5 (Berkeley) 2/17/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_vm.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * MAX_PAGEOUT_CLUSTER must be a power of 2 between 1 and 64. * The 64-page limit is due to the radix code (kern/subr_blist.c). */ #ifndef MAX_PAGEOUT_CLUSTER #define MAX_PAGEOUT_CLUSTER 32 #endif #if !defined(SWB_NPAGES) #define SWB_NPAGES MAX_PAGEOUT_CLUSTER #endif #define SWAP_META_PAGES PCTRIE_COUNT /* * A swblk structure maps each page index within a * SWAP_META_PAGES-aligned and sized range to the address of an * on-disk swap block (or SWAPBLK_NONE). The collection of these * mappings for an entire vm object is implemented as a pc-trie. */ struct swblk { vm_pindex_t p; daddr_t d[SWAP_META_PAGES]; }; static MALLOC_DEFINE(M_VMPGDATA, "vm_pgdata", "swap pager private data"); static struct mtx sw_dev_mtx; static TAILQ_HEAD(, swdevt) swtailq = TAILQ_HEAD_INITIALIZER(swtailq); static struct swdevt *swdevhd; /* Allocate from here next */ static int nswapdev; /* Number of swap devices */ int swap_pager_avail; static struct sx swdev_syscall_lock; /* serialize swap(on|off) */ static u_long swap_reserved; static u_long swap_total; static int sysctl_page_shift(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_vm, OID_AUTO, swap_reserved, CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_MPSAFE, &swap_reserved, 0, sysctl_page_shift, "A", "Amount of swap storage needed to back all allocated anonymous memory."); SYSCTL_PROC(_vm, OID_AUTO, swap_total, CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_MPSAFE, &swap_total, 0, sysctl_page_shift, "A", "Total amount of available swap storage."); static int overcommit = 0; SYSCTL_INT(_vm, VM_OVERCOMMIT, overcommit, CTLFLAG_RW, &overcommit, 0, "Configure virtual memory overcommit behavior. See tuning(7) " "for details."); static unsigned long swzone; SYSCTL_ULONG(_vm, OID_AUTO, swzone, CTLFLAG_RD, &swzone, 0, "Actual size of swap metadata zone"); static unsigned long swap_maxpages; SYSCTL_ULONG(_vm, OID_AUTO, swap_maxpages, CTLFLAG_RD, &swap_maxpages, 0, "Maximum amount of swap supported"); /* bits from overcommit */ #define SWAP_RESERVE_FORCE_ON (1 << 0) #define SWAP_RESERVE_RLIMIT_ON (1 << 1) #define SWAP_RESERVE_ALLOW_NONWIRED (1 << 2) static int sysctl_page_shift(SYSCTL_HANDLER_ARGS) { uint64_t newval; u_long value = *(u_long *)arg1; newval = ((uint64_t)value) << PAGE_SHIFT; return (sysctl_handle_64(oidp, &newval, 0, req)); } int swap_reserve(vm_ooffset_t incr) { return (swap_reserve_by_cred(incr, curthread->td_ucred)); } int swap_reserve_by_cred(vm_ooffset_t incr, struct ucred *cred) { u_long r, s, prev, pincr; int res, error; static int curfail; static struct timeval lastfail; struct uidinfo *uip; uip = cred->cr_ruidinfo; KASSERT((incr & PAGE_MASK) == 0, ("%s: incr: %ju & PAGE_MASK", __func__, (uintmax_t)incr)); #ifdef RACCT if (racct_enable) { PROC_LOCK(curproc); error = racct_add(curproc, RACCT_SWAP, incr); PROC_UNLOCK(curproc); if (error != 0) return (0); } #endif pincr = atop(incr); res = 0; prev = atomic_fetchadd_long(&swap_reserved, pincr); r = prev + pincr; if (overcommit & SWAP_RESERVE_ALLOW_NONWIRED) { s = vm_cnt.v_page_count - vm_cnt.v_free_reserved - vm_wire_count(); } else s = 0; s += swap_total; if ((overcommit & SWAP_RESERVE_FORCE_ON) == 0 || r <= s || (error = priv_check(curthread, PRIV_VM_SWAP_NOQUOTA)) == 0) { res = 1; } else { prev = atomic_fetchadd_long(&swap_reserved, -pincr); if (prev < pincr) panic("swap_reserved < incr on overcommit fail"); } if (res) { prev = atomic_fetchadd_long(&uip->ui_vmsize, pincr); if ((overcommit & SWAP_RESERVE_RLIMIT_ON) != 0 && prev + pincr > lim_cur(curthread, RLIMIT_SWAP) && priv_check(curthread, PRIV_VM_SWAP_NORLIMIT)) { res = 0; prev = atomic_fetchadd_long(&uip->ui_vmsize, -pincr); if (prev < pincr) panic("uip->ui_vmsize < incr on overcommit fail"); } } if (!res && ppsratecheck(&lastfail, &curfail, 1)) { printf("uid %d, pid %d: swap reservation for %jd bytes failed\n", uip->ui_uid, curproc->p_pid, incr); } #ifdef RACCT if (racct_enable && !res) { PROC_LOCK(curproc); racct_sub(curproc, RACCT_SWAP, incr); PROC_UNLOCK(curproc); } #endif return (res); } void swap_reserve_force(vm_ooffset_t incr) { struct uidinfo *uip; u_long pincr; KASSERT((incr & PAGE_MASK) == 0, ("%s: incr: %ju & PAGE_MASK", __func__, (uintmax_t)incr)); PROC_LOCK(curproc); #ifdef RACCT if (racct_enable) racct_add_force(curproc, RACCT_SWAP, incr); #endif pincr = atop(incr); atomic_add_long(&swap_reserved, pincr); uip = curproc->p_ucred->cr_ruidinfo; atomic_add_long(&uip->ui_vmsize, pincr); PROC_UNLOCK(curproc); } void swap_release(vm_ooffset_t decr) { struct ucred *cred; PROC_LOCK(curproc); cred = curproc->p_ucred; swap_release_by_cred(decr, cred); PROC_UNLOCK(curproc); } void swap_release_by_cred(vm_ooffset_t decr, struct ucred *cred) { u_long prev, pdecr; struct uidinfo *uip; uip = cred->cr_ruidinfo; KASSERT((decr & PAGE_MASK) == 0, ("%s: decr: %ju & PAGE_MASK", __func__, (uintmax_t)decr)); pdecr = atop(decr); prev = atomic_fetchadd_long(&swap_reserved, -pdecr); if (prev < pdecr) panic("swap_reserved < decr"); prev = atomic_fetchadd_long(&uip->ui_vmsize, -pdecr); if (prev < pdecr) printf("negative vmsize for uid = %d\n", uip->ui_uid); #ifdef RACCT if (racct_enable) racct_sub_cred(cred, RACCT_SWAP, decr); #endif } #define SWM_POP 0x01 /* pop out */ static int swap_pager_full = 2; /* swap space exhaustion (task killing) */ static int swap_pager_almost_full = 1; /* swap space exhaustion (w/hysteresis)*/ static struct mtx swbuf_mtx; /* to sync nsw_wcount_async */ static int nsw_wcount_async; /* limit async write buffers */ static int nsw_wcount_async_max;/* assigned maximum */ static int nsw_cluster_max; /* maximum VOP I/O allowed */ static int sysctl_swap_async_max(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_vm, OID_AUTO, swap_async_max, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 0, sysctl_swap_async_max, "I", "Maximum running async swap ops"); static int sysctl_swap_fragmentation(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_vm, OID_AUTO, swap_fragmentation, CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, sysctl_swap_fragmentation, "A", "Swap Fragmentation Info"); static struct sx sw_alloc_sx; /* * "named" and "unnamed" anon region objects. Try to reduce the overhead * of searching a named list by hashing it just a little. */ #define NOBJLISTS 8 #define NOBJLIST(handle) \ (&swap_pager_object_list[((int)(intptr_t)handle >> 4) & (NOBJLISTS-1)]) static struct pagerlst swap_pager_object_list[NOBJLISTS]; static uma_zone_t swwbuf_zone; static uma_zone_t swrbuf_zone; static uma_zone_t swblk_zone; static uma_zone_t swpctrie_zone; /* * pagerops for OBJT_SWAP - "swap pager". Some ops are also global procedure * calls hooked from other parts of the VM system and do not appear here. * (see vm/swap_pager.h). */ static vm_object_t swap_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot, vm_ooffset_t offset, struct ucred *); static void swap_pager_dealloc(vm_object_t object); static int swap_pager_getpages(vm_object_t, vm_page_t *, int, int *, int *); static int swap_pager_getpages_async(vm_object_t, vm_page_t *, int, int *, int *, pgo_getpages_iodone_t, void *); static void swap_pager_putpages(vm_object_t, vm_page_t *, int, boolean_t, int *); static boolean_t swap_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *before, int *after); static void swap_pager_init(void); static void swap_pager_unswapped(vm_page_t); static void swap_pager_swapoff(struct swdevt *sp); struct pagerops swappagerops = { .pgo_init = swap_pager_init, /* early system initialization of pager */ .pgo_alloc = swap_pager_alloc, /* allocate an OBJT_SWAP object */ .pgo_dealloc = swap_pager_dealloc, /* deallocate an OBJT_SWAP object */ .pgo_getpages = swap_pager_getpages, /* pagein */ .pgo_getpages_async = swap_pager_getpages_async, /* pagein (async) */ .pgo_putpages = swap_pager_putpages, /* pageout */ .pgo_haspage = swap_pager_haspage, /* get backing store status for page */ .pgo_pageunswapped = swap_pager_unswapped, /* remove swap related to page */ }; /* * swap_*() routines are externally accessible. swp_*() routines are * internal. */ static int nswap_lowat = 128; /* in pages, swap_pager_almost_full warn */ static int nswap_hiwat = 512; /* in pages, swap_pager_almost_full warn */ SYSCTL_INT(_vm, OID_AUTO, dmmax, CTLFLAG_RD, &nsw_cluster_max, 0, "Maximum size of a swap block in pages"); static void swp_sizecheck(void); static void swp_pager_async_iodone(struct buf *bp); static bool swp_pager_swblk_empty(struct swblk *sb, int start, int limit); static int swapongeom(struct vnode *); static int swaponvp(struct thread *, struct vnode *, u_long); static int swapoff_one(struct swdevt *sp, struct ucred *cred); /* * Swap bitmap functions */ static void swp_pager_freeswapspace(daddr_t blk, daddr_t npages); static daddr_t swp_pager_getswapspace(int *npages, int limit); /* * Metadata functions */ static daddr_t swp_pager_meta_build(vm_object_t, vm_pindex_t, daddr_t); static void swp_pager_meta_free(vm_object_t, vm_pindex_t, vm_pindex_t); static void swp_pager_meta_free_all(vm_object_t); static daddr_t swp_pager_meta_ctl(vm_object_t, vm_pindex_t, int); static void swp_pager_init_freerange(daddr_t *start, daddr_t *num) { *start = SWAPBLK_NONE; *num = 0; } static void swp_pager_update_freerange(daddr_t *start, daddr_t *num, daddr_t addr) { if (*start + *num == addr) { (*num)++; } else { swp_pager_freeswapspace(*start, *num); *start = addr; *num = 1; } } static void * swblk_trie_alloc(struct pctrie *ptree) { return (uma_zalloc(swpctrie_zone, M_NOWAIT | (curproc == pageproc ? M_USE_RESERVE : 0))); } static void swblk_trie_free(struct pctrie *ptree, void *node) { uma_zfree(swpctrie_zone, node); } PCTRIE_DEFINE(SWAP, swblk, p, swblk_trie_alloc, swblk_trie_free); /* * SWP_SIZECHECK() - update swap_pager_full indication * * update the swap_pager_almost_full indication and warn when we are * about to run out of swap space, using lowat/hiwat hysteresis. * * Clear swap_pager_full ( task killing ) indication when lowat is met. * * No restrictions on call * This routine may not block. */ static void swp_sizecheck(void) { if (swap_pager_avail < nswap_lowat) { if (swap_pager_almost_full == 0) { printf("swap_pager: out of swap space\n"); swap_pager_almost_full = 1; } } else { swap_pager_full = 0; if (swap_pager_avail > nswap_hiwat) swap_pager_almost_full = 0; } } /* * SWAP_PAGER_INIT() - initialize the swap pager! * * Expected to be started from system init. NOTE: This code is run * before much else so be careful what you depend on. Most of the VM * system has yet to be initialized at this point. */ static void swap_pager_init(void) { /* * Initialize object lists */ int i; for (i = 0; i < NOBJLISTS; ++i) TAILQ_INIT(&swap_pager_object_list[i]); mtx_init(&sw_dev_mtx, "swapdev", NULL, MTX_DEF); sx_init(&sw_alloc_sx, "swspsx"); sx_init(&swdev_syscall_lock, "swsysc"); } /* * SWAP_PAGER_SWAP_INIT() - swap pager initialization from pageout process * * Expected to be started from pageout process once, prior to entering * its main loop. */ void swap_pager_swap_init(void) { unsigned long n, n2; /* * Number of in-transit swap bp operations. Don't * exhaust the pbufs completely. Make sure we * initialize workable values (0 will work for hysteresis * but it isn't very efficient). * * The nsw_cluster_max is constrained by the bp->b_pages[] * array, which has MAXPHYS / PAGE_SIZE entries, and our locally * defined MAX_PAGEOUT_CLUSTER. Also be aware that swap ops are * constrained by the swap device interleave stripe size. * * Currently we hardwire nsw_wcount_async to 4. This limit is * designed to prevent other I/O from having high latencies due to * our pageout I/O. The value 4 works well for one or two active swap * devices but is probably a little low if you have more. Even so, * a higher value would probably generate only a limited improvement * with three or four active swap devices since the system does not * typically have to pageout at extreme bandwidths. We will want * at least 2 per swap devices, and 4 is a pretty good value if you * have one NFS swap device due to the command/ack latency over NFS. * So it all works out pretty well. */ nsw_cluster_max = min(MAXPHYS / PAGE_SIZE, MAX_PAGEOUT_CLUSTER); nsw_wcount_async = 4; nsw_wcount_async_max = nsw_wcount_async; mtx_init(&swbuf_mtx, "async swbuf mutex", NULL, MTX_DEF); swwbuf_zone = pbuf_zsecond_create("swwbuf", nswbuf / 4); swrbuf_zone = pbuf_zsecond_create("swrbuf", nswbuf / 2); /* * Initialize our zone, taking the user's requested size or * estimating the number we need based on the number of pages * in the system. */ n = maxswzone != 0 ? maxswzone / sizeof(struct swblk) : vm_cnt.v_page_count / 2; swpctrie_zone = uma_zcreate("swpctrie", pctrie_node_size(), NULL, NULL, pctrie_zone_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM); if (swpctrie_zone == NULL) panic("failed to create swap pctrie zone."); swblk_zone = uma_zcreate("swblk", sizeof(struct swblk), NULL, NULL, NULL, NULL, _Alignof(struct swblk) - 1, UMA_ZONE_VM); if (swblk_zone == NULL) panic("failed to create swap blk zone."); n2 = n; do { if (uma_zone_reserve_kva(swblk_zone, n)) break; /* * if the allocation failed, try a zone two thirds the * size of the previous attempt. */ n -= ((n + 2) / 3); } while (n > 0); /* * Often uma_zone_reserve_kva() cannot reserve exactly the * requested size. Account for the difference when * calculating swap_maxpages. */ n = uma_zone_get_max(swblk_zone); if (n < n2) printf("Swap blk zone entries changed from %lu to %lu.\n", n2, n); swap_maxpages = n * SWAP_META_PAGES; swzone = n * sizeof(struct swblk); if (!uma_zone_reserve_kva(swpctrie_zone, n)) printf("Cannot reserve swap pctrie zone, " "reduce kern.maxswzone.\n"); } static vm_object_t swap_pager_alloc_init(void *handle, struct ucred *cred, vm_ooffset_t size, vm_ooffset_t offset) { vm_object_t object; if (cred != NULL) { if (!swap_reserve_by_cred(size, cred)) return (NULL); crhold(cred); } /* * The un_pager.swp.swp_blks trie is initialized by * vm_object_allocate() to ensure the correct order of * visibility to other threads. */ object = vm_object_allocate(OBJT_SWAP, OFF_TO_IDX(offset + PAGE_MASK + size)); object->handle = handle; if (cred != NULL) { object->cred = cred; object->charge = size; } return (object); } /* * SWAP_PAGER_ALLOC() - allocate a new OBJT_SWAP VM object and instantiate * its metadata structures. * * This routine is called from the mmap and fork code to create a new * OBJT_SWAP object. * * This routine must ensure that no live duplicate is created for * the named object request, which is protected against by * holding the sw_alloc_sx lock in case handle != NULL. */ static vm_object_t swap_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot, vm_ooffset_t offset, struct ucred *cred) { vm_object_t object; if (handle != NULL) { /* * Reference existing named region or allocate new one. There * should not be a race here against swp_pager_meta_build() * as called from vm_page_remove() in regards to the lookup * of the handle. */ sx_xlock(&sw_alloc_sx); object = vm_pager_object_lookup(NOBJLIST(handle), handle); if (object == NULL) { object = swap_pager_alloc_init(handle, cred, size, offset); if (object != NULL) { TAILQ_INSERT_TAIL(NOBJLIST(object->handle), object, pager_object_list); } } sx_xunlock(&sw_alloc_sx); } else { object = swap_pager_alloc_init(handle, cred, size, offset); } return (object); } /* * SWAP_PAGER_DEALLOC() - remove swap metadata from object * * The swap backing for the object is destroyed. The code is * designed such that we can reinstantiate it later, but this * routine is typically called only when the entire object is * about to be destroyed. * * The object must be locked. */ static void swap_pager_dealloc(vm_object_t object) { VM_OBJECT_ASSERT_WLOCKED(object); KASSERT((object->flags & OBJ_DEAD) != 0, ("dealloc of reachable obj")); /* * Remove from list right away so lookups will fail if we block for * pageout completion. */ if (object->handle != NULL) { VM_OBJECT_WUNLOCK(object); sx_xlock(&sw_alloc_sx); TAILQ_REMOVE(NOBJLIST(object->handle), object, pager_object_list); sx_xunlock(&sw_alloc_sx); VM_OBJECT_WLOCK(object); } vm_object_pip_wait(object, "swpdea"); /* * Free all remaining metadata. We only bother to free it from * the swap meta data. We do not attempt to free swapblk's still * associated with vm_page_t's for this object. We do not care * if paging is still in progress on some objects. */ swp_pager_meta_free_all(object); object->handle = NULL; object->type = OBJT_DEAD; } /************************************************************************ * SWAP PAGER BITMAP ROUTINES * ************************************************************************/ /* * SWP_PAGER_GETSWAPSPACE() - allocate raw swap space * * Allocate swap for up to the requested number of pages, and at * least a minimum number of pages. The starting swap block number * (a page index) is returned or SWAPBLK_NONE if the allocation * failed. * * Also has the side effect of advising that somebody made a mistake * when they configured swap and didn't configure enough. * * This routine may not sleep. * * We allocate in round-robin fashion from the configured devices. */ static daddr_t swp_pager_getswapspace(int *io_npages, int limit) { daddr_t blk; struct swdevt *sp; int mpages, npages; blk = SWAPBLK_NONE; mpages = *io_npages; npages = imin(BLIST_MAX_ALLOC, mpages); mtx_lock(&sw_dev_mtx); sp = swdevhd; while (!TAILQ_EMPTY(&swtailq)) { if (sp == NULL) sp = TAILQ_FIRST(&swtailq); if ((sp->sw_flags & SW_CLOSING) == 0) blk = blist_alloc(sp->sw_blist, &npages, mpages); if (blk != SWAPBLK_NONE) break; sp = TAILQ_NEXT(sp, sw_list); if (swdevhd == sp) { if (npages <= limit) break; mpages = npages - 1; npages >>= 1; } } if (blk != SWAPBLK_NONE) { *io_npages = npages; blk += sp->sw_first; sp->sw_used += npages; swap_pager_avail -= npages; swp_sizecheck(); swdevhd = TAILQ_NEXT(sp, sw_list); } else { if (swap_pager_full != 2) { printf("swp_pager_getswapspace(%d): failed\n", *io_npages); swap_pager_full = 2; swap_pager_almost_full = 1; } swdevhd = NULL; } mtx_unlock(&sw_dev_mtx); return (blk); } static bool swp_pager_isondev(daddr_t blk, struct swdevt *sp) { return (blk >= sp->sw_first && blk < sp->sw_end); } static void swp_pager_strategy(struct buf *bp) { struct swdevt *sp; mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(sp, &swtailq, sw_list) { if (swp_pager_isondev(bp->b_blkno, sp)) { mtx_unlock(&sw_dev_mtx); if ((sp->sw_flags & SW_UNMAPPED) != 0 && unmapped_buf_allowed) { bp->b_data = unmapped_buf; bp->b_offset = 0; } else { pmap_qenter((vm_offset_t)bp->b_data, &bp->b_pages[0], bp->b_bcount / PAGE_SIZE); } sp->sw_strategy(bp, sp); return; } } panic("Swapdev not found"); } /* * SWP_PAGER_FREESWAPSPACE() - free raw swap space * * This routine returns the specified swap blocks back to the bitmap. * * This routine may not sleep. */ static void swp_pager_freeswapspace(daddr_t blk, daddr_t npages) { struct swdevt *sp; if (npages == 0) return; mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(sp, &swtailq, sw_list) { if (swp_pager_isondev(blk, sp)) { sp->sw_used -= npages; /* * If we are attempting to stop swapping on * this device, we don't want to mark any * blocks free lest they be reused. */ if ((sp->sw_flags & SW_CLOSING) == 0) { blist_free(sp->sw_blist, blk - sp->sw_first, npages); swap_pager_avail += npages; swp_sizecheck(); } mtx_unlock(&sw_dev_mtx); return; } } panic("Swapdev not found"); } /* * SYSCTL_SWAP_FRAGMENTATION() - produce raw swap space stats */ static int sysctl_swap_fragmentation(SYSCTL_HANDLER_ARGS) { struct sbuf sbuf; struct swdevt *sp; const char *devname; int error; error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); sbuf_new_for_sysctl(&sbuf, NULL, 128, req); mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(sp, &swtailq, sw_list) { if (vn_isdisk(sp->sw_vp, NULL)) devname = devtoname(sp->sw_vp->v_rdev); else devname = "[file]"; sbuf_printf(&sbuf, "\nFree space on device %s:\n", devname); blist_stats(sp->sw_blist, &sbuf); } mtx_unlock(&sw_dev_mtx); error = sbuf_finish(&sbuf); sbuf_delete(&sbuf); return (error); } /* * SWAP_PAGER_FREESPACE() - frees swap blocks associated with a page * range within an object. * * This is a globally accessible routine. * * This routine removes swapblk assignments from swap metadata. * * The external callers of this routine typically have already destroyed * or renamed vm_page_t's associated with this range in the object so * we should be ok. * * The object must be locked. */ void swap_pager_freespace(vm_object_t object, vm_pindex_t start, vm_size_t size) { swp_pager_meta_free(object, start, size); } /* * SWAP_PAGER_RESERVE() - reserve swap blocks in object * * Assigns swap blocks to the specified range within the object. The * swap blocks are not zeroed. Any previous swap assignment is destroyed. * * Returns 0 on success, -1 on failure. */ int swap_pager_reserve(vm_object_t object, vm_pindex_t start, vm_size_t size) { daddr_t addr, blk, n_free, s_free; int i, j, n; swp_pager_init_freerange(&s_free, &n_free); VM_OBJECT_WLOCK(object); for (i = 0; i < size; i += n) { n = size - i; blk = swp_pager_getswapspace(&n, 1); if (blk == SWAPBLK_NONE) { swp_pager_meta_free(object, start, i); VM_OBJECT_WUNLOCK(object); return (-1); } for (j = 0; j < n; ++j) { addr = swp_pager_meta_build(object, start + i + j, blk + j); if (addr != SWAPBLK_NONE) swp_pager_update_freerange(&s_free, &n_free, addr); } } swp_pager_freeswapspace(s_free, n_free); VM_OBJECT_WUNLOCK(object); return (0); } /* * SWAP_PAGER_COPY() - copy blocks from source pager to destination pager * and destroy the source. * * Copy any valid swapblks from the source to the destination. In * cases where both the source and destination have a valid swapblk, * we keep the destination's. * * This routine is allowed to sleep. It may sleep allocating metadata * indirectly through swp_pager_meta_build() or if paging is still in * progress on the source. * * The source object contains no vm_page_t's (which is just as well) * * The source object is of type OBJT_SWAP. * * The source and destination objects must be locked. * Both object locks may temporarily be released. */ void swap_pager_copy(vm_object_t srcobject, vm_object_t dstobject, vm_pindex_t offset, int destroysource) { vm_pindex_t i; daddr_t dstaddr, n_free, s_free, srcaddr; VM_OBJECT_ASSERT_WLOCKED(srcobject); VM_OBJECT_ASSERT_WLOCKED(dstobject); /* * If destroysource is set, we remove the source object from the * swap_pager internal queue now. */ if (destroysource && srcobject->handle != NULL) { vm_object_pip_add(srcobject, 1); VM_OBJECT_WUNLOCK(srcobject); vm_object_pip_add(dstobject, 1); VM_OBJECT_WUNLOCK(dstobject); sx_xlock(&sw_alloc_sx); TAILQ_REMOVE(NOBJLIST(srcobject->handle), srcobject, pager_object_list); sx_xunlock(&sw_alloc_sx); VM_OBJECT_WLOCK(dstobject); vm_object_pip_wakeup(dstobject); VM_OBJECT_WLOCK(srcobject); vm_object_pip_wakeup(srcobject); } /* * Transfer source to destination. */ swp_pager_init_freerange(&s_free, &n_free); for (i = 0; i < dstobject->size; ++i) { srcaddr = swp_pager_meta_ctl(srcobject, i + offset, SWM_POP); if (srcaddr == SWAPBLK_NONE) continue; dstaddr = swp_pager_meta_ctl(dstobject, i, 0); if (dstaddr != SWAPBLK_NONE) { /* * Destination has valid swapblk or it is represented * by a resident page. We destroy the source block. */ swp_pager_update_freerange(&s_free, &n_free, srcaddr); continue; } /* * Destination has no swapblk and is not resident, * copy source. * * swp_pager_meta_build() can sleep. */ vm_object_pip_add(srcobject, 1); VM_OBJECT_WUNLOCK(srcobject); vm_object_pip_add(dstobject, 1); dstaddr = swp_pager_meta_build(dstobject, i, srcaddr); KASSERT(dstaddr == SWAPBLK_NONE, ("Unexpected destination swapblk")); vm_object_pip_wakeup(dstobject); VM_OBJECT_WLOCK(srcobject); vm_object_pip_wakeup(srcobject); } swp_pager_freeswapspace(s_free, n_free); /* * Free left over swap blocks in source. * * We have to revert the type to OBJT_DEFAULT so we do not accidentally * double-remove the object from the swap queues. */ if (destroysource) { swp_pager_meta_free_all(srcobject); /* * Reverting the type is not necessary, the caller is going * to destroy srcobject directly, but I'm doing it here * for consistency since we've removed the object from its * queues. */ srcobject->type = OBJT_DEFAULT; } } /* * SWAP_PAGER_HASPAGE() - determine if we have good backing store for * the requested page. * * We determine whether good backing store exists for the requested * page and return TRUE if it does, FALSE if it doesn't. * * If TRUE, we also try to determine how much valid, contiguous backing * store exists before and after the requested page. */ static boolean_t swap_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *before, int *after) { daddr_t blk, blk0; int i; VM_OBJECT_ASSERT_LOCKED(object); /* * do we have good backing store at the requested index ? */ blk0 = swp_pager_meta_ctl(object, pindex, 0); if (blk0 == SWAPBLK_NONE) { if (before) *before = 0; if (after) *after = 0; return (FALSE); } /* * find backwards-looking contiguous good backing store */ if (before != NULL) { for (i = 1; i < SWB_NPAGES; i++) { if (i > pindex) break; blk = swp_pager_meta_ctl(object, pindex - i, 0); if (blk != blk0 - i) break; } *before = i - 1; } /* * find forward-looking contiguous good backing store */ if (after != NULL) { for (i = 1; i < SWB_NPAGES; i++) { blk = swp_pager_meta_ctl(object, pindex + i, 0); if (blk != blk0 + i) break; } *after = i - 1; } return (TRUE); } /* * SWAP_PAGER_PAGE_UNSWAPPED() - remove swap backing store related to page * * This removes any associated swap backing store, whether valid or * not, from the page. * * This routine is typically called when a page is made dirty, at * which point any associated swap can be freed. MADV_FREE also * calls us in a special-case situation * * NOTE!!! If the page is clean and the swap was valid, the caller * should make the page dirty before calling this routine. This routine * does NOT change the m->dirty status of the page. Also: MADV_FREE * depends on it. * * This routine may not sleep. * * The object containing the page must be locked. */ static void swap_pager_unswapped(vm_page_t m) { daddr_t srcaddr; srcaddr = swp_pager_meta_ctl(m->object, m->pindex, SWM_POP); if (srcaddr != SWAPBLK_NONE) swp_pager_freeswapspace(srcaddr, 1); } /* * swap_pager_getpages() - bring pages in from swap * * Attempt to page in the pages in array "ma" of length "count". The * caller may optionally specify that additional pages preceding and * succeeding the specified range be paged in. The number of such pages * is returned in the "rbehind" and "rahead" parameters, and they will * be in the inactive queue upon return. * * The pages in "ma" must be busied and will remain busied upon return. */ static int swap_pager_getpages(vm_object_t object, vm_page_t *ma, int count, int *rbehind, int *rahead) { struct buf *bp; vm_page_t bm, mpred, msucc, p; vm_pindex_t pindex; daddr_t blk; int i, maxahead, maxbehind, reqcount; reqcount = count; /* * Determine the final number of read-behind pages and * allocate them BEFORE releasing the object lock. Otherwise, * there can be a problematic race with vm_object_split(). * Specifically, vm_object_split() might first transfer pages * that precede ma[0] in the current object to a new object, * and then this function incorrectly recreates those pages as * read-behind pages in the current object. */ if (!swap_pager_haspage(object, ma[0]->pindex, &maxbehind, &maxahead)) return (VM_PAGER_FAIL); /* * Clip the readahead and readbehind ranges to exclude resident pages. */ if (rahead != NULL) { KASSERT(reqcount - 1 <= maxahead, ("page count %d extends beyond swap block", reqcount)); *rahead = imin(*rahead, maxahead - (reqcount - 1)); pindex = ma[reqcount - 1]->pindex; msucc = TAILQ_NEXT(ma[reqcount - 1], listq); if (msucc != NULL && msucc->pindex - pindex - 1 < *rahead) *rahead = msucc->pindex - pindex - 1; } if (rbehind != NULL) { *rbehind = imin(*rbehind, maxbehind); pindex = ma[0]->pindex; mpred = TAILQ_PREV(ma[0], pglist, listq); if (mpred != NULL && pindex - mpred->pindex - 1 < *rbehind) *rbehind = pindex - mpred->pindex - 1; } bm = ma[0]; for (i = 0; i < count; i++) ma[i]->oflags |= VPO_SWAPINPROG; /* * Allocate readahead and readbehind pages. */ if (rbehind != NULL) { for (i = 1; i <= *rbehind; i++) { p = vm_page_alloc(object, ma[0]->pindex - i, VM_ALLOC_NORMAL); if (p == NULL) break; p->oflags |= VPO_SWAPINPROG; bm = p; } *rbehind = i - 1; } if (rahead != NULL) { for (i = 0; i < *rahead; i++) { p = vm_page_alloc(object, ma[reqcount - 1]->pindex + i + 1, VM_ALLOC_NORMAL); if (p == NULL) break; p->oflags |= VPO_SWAPINPROG; } *rahead = i; } if (rbehind != NULL) count += *rbehind; if (rahead != NULL) count += *rahead; vm_object_pip_add(object, count); pindex = bm->pindex; blk = swp_pager_meta_ctl(object, pindex, 0); KASSERT(blk != SWAPBLK_NONE, ("no swap blocking containing %p(%jx)", object, (uintmax_t)pindex)); VM_OBJECT_WUNLOCK(object); bp = uma_zalloc(swrbuf_zone, M_WAITOK); /* Pages cannot leave the object while busy. */ for (i = 0, p = bm; i < count; i++, p = TAILQ_NEXT(p, listq)) { MPASS(p->pindex == bm->pindex + i); bp->b_pages[i] = p; } bp->b_flags |= B_PAGING; bp->b_iocmd = BIO_READ; bp->b_iodone = swp_pager_async_iodone; bp->b_rcred = crhold(thread0.td_ucred); bp->b_wcred = crhold(thread0.td_ucred); bp->b_blkno = blk; bp->b_bcount = PAGE_SIZE * count; bp->b_bufsize = PAGE_SIZE * count; bp->b_npages = count; bp->b_pgbefore = rbehind != NULL ? *rbehind : 0; bp->b_pgafter = rahead != NULL ? *rahead : 0; VM_CNT_INC(v_swapin); VM_CNT_ADD(v_swappgsin, count); /* * perform the I/O. NOTE!!! bp cannot be considered valid after * this point because we automatically release it on completion. * Instead, we look at the one page we are interested in which we * still hold a lock on even through the I/O completion. * * The other pages in our ma[] array are also released on completion, * so we cannot assume they are valid anymore either. * * NOTE: b_blkno is destroyed by the call to swapdev_strategy */ BUF_KERNPROC(bp); swp_pager_strategy(bp); /* * Wait for the pages we want to complete. VPO_SWAPINPROG is always * cleared on completion. If an I/O error occurs, SWAPBLK_NONE * is set in the metadata for each page in the request. */ VM_OBJECT_WLOCK(object); while ((ma[0]->oflags & VPO_SWAPINPROG) != 0) { ma[0]->oflags |= VPO_SWAPSLEEP; VM_CNT_INC(v_intrans); - if (VM_OBJECT_SLEEP(object, &object->paging_in_progress, PSWP, + if (VM_OBJECT_SLEEP(object, &object->handle, PSWP, "swread", hz * 20)) { printf( "swap_pager: indefinite wait buffer: bufobj: %p, blkno: %jd, size: %ld\n", bp->b_bufobj, (intmax_t)bp->b_blkno, bp->b_bcount); } } /* * If we had an unrecoverable read error pages will not be valid. */ for (i = 0; i < reqcount; i++) if (ma[i]->valid != VM_PAGE_BITS_ALL) return (VM_PAGER_ERROR); return (VM_PAGER_OK); /* * A final note: in a low swap situation, we cannot deallocate swap * and mark a page dirty here because the caller is likely to mark * the page clean when we return, causing the page to possibly revert * to all-zero's later. */ } /* * swap_pager_getpages_async(): * * Right now this is emulation of asynchronous operation on top of * swap_pager_getpages(). */ static int swap_pager_getpages_async(vm_object_t object, vm_page_t *ma, int count, int *rbehind, int *rahead, pgo_getpages_iodone_t iodone, void *arg) { int r, error; r = swap_pager_getpages(object, ma, count, rbehind, rahead); VM_OBJECT_WUNLOCK(object); switch (r) { case VM_PAGER_OK: error = 0; break; case VM_PAGER_ERROR: error = EIO; break; case VM_PAGER_FAIL: error = EINVAL; break; default: panic("unhandled swap_pager_getpages() error %d", r); } (iodone)(arg, ma, count, error); VM_OBJECT_WLOCK(object); return (r); } /* * swap_pager_putpages: * * Assign swap (if necessary) and initiate I/O on the specified pages. * * We support both OBJT_DEFAULT and OBJT_SWAP objects. DEFAULT objects * are automatically converted to SWAP objects. * * In a low memory situation we may block in VOP_STRATEGY(), but the new * vm_page reservation system coupled with properly written VFS devices * should ensure that no low-memory deadlock occurs. This is an area * which needs work. * * The parent has N vm_object_pip_add() references prior to * calling us and will remove references for rtvals[] that are * not set to VM_PAGER_PEND. We need to remove the rest on I/O * completion. * * The parent has soft-busy'd the pages it passes us and will unbusy * those whose rtvals[] entry is not set to VM_PAGER_PEND on return. * We need to unbusy the rest on I/O completion. */ static void swap_pager_putpages(vm_object_t object, vm_page_t *ma, int count, int flags, int *rtvals) { struct buf *bp; daddr_t addr, blk, n_free, s_free; vm_page_t mreq; int i, j, n; bool async; KASSERT(count == 0 || ma[0]->object == object, ("%s: object mismatch %p/%p", __func__, object, ma[0]->object)); /* * Step 1 * * Turn object into OBJT_SWAP. Force sync if not a pageout process. */ if (object->type != OBJT_SWAP) { addr = swp_pager_meta_build(object, 0, SWAPBLK_NONE); KASSERT(addr == SWAPBLK_NONE, ("unexpected object swap block")); } VM_OBJECT_WUNLOCK(object); async = curproc == pageproc && (flags & VM_PAGER_PUT_SYNC) == 0; swp_pager_init_freerange(&s_free, &n_free); /* * Step 2 * * Assign swap blocks and issue I/O. We reallocate swap on the fly. * The page is left dirty until the pageout operation completes * successfully. */ for (i = 0; i < count; i += n) { /* Maximum I/O size is limited by maximum swap block size. */ n = min(count - i, nsw_cluster_max); /* Get a block of swap of size up to size n. */ blk = swp_pager_getswapspace(&n, 4); if (blk == SWAPBLK_NONE) { for (j = 0; j < n; ++j) rtvals[i + j] = VM_PAGER_FAIL; continue; } /* * All I/O parameters have been satisfied. Build the I/O * request and assign the swap space. */ if (async) { mtx_lock(&swbuf_mtx); while (nsw_wcount_async == 0) msleep(&nsw_wcount_async, &swbuf_mtx, PVM, "swbufa", 0); nsw_wcount_async--; mtx_unlock(&swbuf_mtx); } bp = uma_zalloc(swwbuf_zone, M_WAITOK); if (async) bp->b_flags = B_ASYNC; bp->b_flags |= B_PAGING; bp->b_iocmd = BIO_WRITE; bp->b_rcred = crhold(thread0.td_ucred); bp->b_wcred = crhold(thread0.td_ucred); bp->b_bcount = PAGE_SIZE * n; bp->b_bufsize = PAGE_SIZE * n; bp->b_blkno = blk; VM_OBJECT_WLOCK(object); for (j = 0; j < n; ++j) { mreq = ma[i + j]; addr = swp_pager_meta_build(mreq->object, mreq->pindex, blk + j); if (addr != SWAPBLK_NONE) swp_pager_update_freerange(&s_free, &n_free, addr); MPASS(mreq->dirty == VM_PAGE_BITS_ALL); mreq->oflags |= VPO_SWAPINPROG; bp->b_pages[j] = mreq; } VM_OBJECT_WUNLOCK(object); bp->b_npages = n; /* * Must set dirty range for NFS to work. */ bp->b_dirtyoff = 0; bp->b_dirtyend = bp->b_bcount; VM_CNT_INC(v_swapout); VM_CNT_ADD(v_swappgsout, bp->b_npages); /* * We unconditionally set rtvals[] to VM_PAGER_PEND so that we * can call the async completion routine at the end of a * synchronous I/O operation. Otherwise, our caller would * perform duplicate unbusy and wakeup operations on the page * and object, respectively. */ for (j = 0; j < n; j++) rtvals[i + j] = VM_PAGER_PEND; /* * asynchronous * * NOTE: b_blkno is destroyed by the call to swapdev_strategy. */ if (async) { bp->b_iodone = swp_pager_async_iodone; BUF_KERNPROC(bp); swp_pager_strategy(bp); continue; } /* * synchronous * * NOTE: b_blkno is destroyed by the call to swapdev_strategy. */ bp->b_iodone = bdone; swp_pager_strategy(bp); /* * Wait for the sync I/O to complete. */ bwait(bp, PVM, "swwrt"); /* * Now that we are through with the bp, we can call the * normal async completion, which frees everything up. */ swp_pager_async_iodone(bp); } swp_pager_freeswapspace(s_free, n_free); VM_OBJECT_WLOCK(object); } /* * swp_pager_async_iodone: * * Completion routine for asynchronous reads and writes from/to swap. * Also called manually by synchronous code to finish up a bp. * * This routine may not sleep. */ static void swp_pager_async_iodone(struct buf *bp) { int i; vm_object_t object = NULL; /* * Report error - unless we ran out of memory, in which case * we've already logged it in swapgeom_strategy(). */ if (bp->b_ioflags & BIO_ERROR && bp->b_error != ENOMEM) { printf( "swap_pager: I/O error - %s failed; blkno %ld," "size %ld, error %d\n", ((bp->b_iocmd == BIO_READ) ? "pagein" : "pageout"), (long)bp->b_blkno, (long)bp->b_bcount, bp->b_error ); } /* * remove the mapping for kernel virtual */ if (buf_mapped(bp)) pmap_qremove((vm_offset_t)bp->b_data, bp->b_npages); else bp->b_data = bp->b_kvabase; if (bp->b_npages) { object = bp->b_pages[0]->object; VM_OBJECT_WLOCK(object); } /* * cleanup pages. If an error occurs writing to swap, we are in * very serious trouble. If it happens to be a disk error, though, * we may be able to recover by reassigning the swap later on. So * in this case we remove the m->swapblk assignment for the page * but do not free it in the rlist. The errornous block(s) are thus * never reallocated as swap. Redirty the page and continue. */ for (i = 0; i < bp->b_npages; ++i) { vm_page_t m = bp->b_pages[i]; m->oflags &= ~VPO_SWAPINPROG; if (m->oflags & VPO_SWAPSLEEP) { m->oflags &= ~VPO_SWAPSLEEP; - wakeup(&object->paging_in_progress); + wakeup(&object->handle); } if (bp->b_ioflags & BIO_ERROR) { /* * If an error occurs I'd love to throw the swapblk * away without freeing it back to swapspace, so it * can never be used again. But I can't from an * interrupt. */ if (bp->b_iocmd == BIO_READ) { /* * NOTE: for reads, m->dirty will probably * be overridden by the original caller of * getpages so don't play cute tricks here. */ m->valid = 0; } else { /* * If a write error occurs, reactivate page * so it doesn't clog the inactive list, * then finish the I/O. */ MPASS(m->dirty == VM_PAGE_BITS_ALL); vm_page_lock(m); vm_page_activate(m); vm_page_unlock(m); vm_page_sunbusy(m); } } else if (bp->b_iocmd == BIO_READ) { /* * NOTE: for reads, m->dirty will probably be * overridden by the original caller of getpages so * we cannot set them in order to free the underlying * swap in a low-swap situation. I don't think we'd * want to do that anyway, but it was an optimization * that existed in the old swapper for a time before * it got ripped out due to precisely this problem. */ KASSERT(!pmap_page_is_mapped(m), ("swp_pager_async_iodone: page %p is mapped", m)); KASSERT(m->dirty == 0, ("swp_pager_async_iodone: page %p is dirty", m)); m->valid = VM_PAGE_BITS_ALL; if (i < bp->b_pgbefore || i >= bp->b_npages - bp->b_pgafter) vm_page_readahead_finish(m); } else { /* * For write success, clear the dirty * status, then finish the I/O ( which decrements the * busy count and possibly wakes waiter's up ). * A page is only written to swap after a period of * inactivity. Therefore, we do not expect it to be * reused. */ KASSERT(!pmap_page_is_write_mapped(m), ("swp_pager_async_iodone: page %p is not write" " protected", m)); vm_page_undirty(m); vm_page_lock(m); vm_page_deactivate_noreuse(m); vm_page_unlock(m); vm_page_sunbusy(m); } } /* * adjust pip. NOTE: the original parent may still have its own * pip refs on the object. */ if (object != NULL) { vm_object_pip_wakeupn(object, bp->b_npages); VM_OBJECT_WUNLOCK(object); } /* * swapdev_strategy() manually sets b_vp and b_bufobj before calling * bstrategy(). Set them back to NULL now we're done with it, or we'll * trigger a KASSERT in relpbuf(). */ if (bp->b_vp) { bp->b_vp = NULL; bp->b_bufobj = NULL; } /* * release the physical I/O buffer */ if (bp->b_flags & B_ASYNC) { mtx_lock(&swbuf_mtx); if (++nsw_wcount_async == 1) wakeup(&nsw_wcount_async); mtx_unlock(&swbuf_mtx); } uma_zfree((bp->b_iocmd == BIO_READ) ? swrbuf_zone : swwbuf_zone, bp); } int swap_pager_nswapdev(void) { return (nswapdev); } static void swp_pager_force_dirty(vm_page_t m) { vm_page_dirty(m); #ifdef INVARIANTS vm_page_lock(m); if (!vm_page_wired(m) && m->queue == PQ_NONE) panic("page %p is neither wired nor queued", m); vm_page_unlock(m); #endif vm_page_xunbusy(m); swap_pager_unswapped(m); } static void swp_pager_force_launder(vm_page_t m) { vm_page_dirty(m); vm_page_lock(m); vm_page_launder(m); vm_page_unlock(m); vm_page_xunbusy(m); swap_pager_unswapped(m); } /* * SWP_PAGER_FORCE_PAGEIN() - force swap blocks to be paged in * * This routine dissociates pages starting at the given index within an * object from their backing store, paging them in if they do not reside * in memory. Pages that are paged in are marked dirty and placed in the * laundry queue. Pages are marked dirty because they no longer have * backing store. They are placed in the laundry queue because they have * not been accessed recently. Otherwise, they would already reside in * memory. */ static void swp_pager_force_pagein(vm_object_t object, vm_pindex_t pindex, int npages) { vm_page_t ma[npages]; int i, j; KASSERT(npages > 0, ("%s: No pages", __func__)); KASSERT(npages <= MAXPHYS / PAGE_SIZE, ("%s: Too many pages: %d", __func__, npages)); vm_object_pip_add(object, npages); vm_page_grab_pages(object, pindex, VM_ALLOC_NORMAL, ma, npages); for (i = j = 0;; i++) { /* Count nonresident pages, to page-in all at once. */ if (i < npages && ma[i]->valid != VM_PAGE_BITS_ALL) continue; if (j < i) { /* Page-in nonresident pages. Mark for laundering. */ if (swap_pager_getpages(object, &ma[j], i - j, NULL, NULL) != VM_PAGER_OK) panic("%s: read from swap failed", __func__); do { swp_pager_force_launder(ma[j]); } while (++j < i); } if (i == npages) break; /* Mark dirty a resident page. */ swp_pager_force_dirty(ma[j++]); } vm_object_pip_wakeupn(object, npages); } /* * swap_pager_swapoff_object: * * Page in all of the pages that have been paged out for an object * to a swap device. */ static void swap_pager_swapoff_object(struct swdevt *sp, vm_object_t object) { struct swblk *sb; vm_pindex_t pi, s_pindex; daddr_t blk, n_blks, s_blk; int i; n_blks = 0; for (pi = 0; (sb = SWAP_PCTRIE_LOOKUP_GE( &object->un_pager.swp.swp_blks, pi)) != NULL; ) { for (i = 0; i < SWAP_META_PAGES; i++) { blk = sb->d[i]; if (!swp_pager_isondev(blk, sp)) blk = SWAPBLK_NONE; /* * If there are no blocks/pages accumulated, start a new * accumulation here. */ if (n_blks == 0) { if (blk != SWAPBLK_NONE) { s_blk = blk; s_pindex = sb->p + i; n_blks = 1; } continue; } /* * If the accumulation can be extended without breaking * the sequence of consecutive blocks and pages that * swp_pager_force_pagein() depends on, do so. */ if (n_blks < MAXPHYS / PAGE_SIZE && s_blk + n_blks == blk && s_pindex + n_blks == sb->p + i) { ++n_blks; continue; } /* * The sequence of consecutive blocks and pages cannot * be extended, so page them all in here. Then, * because doing so involves releasing and reacquiring * a lock that protects the swap block pctrie, do not * rely on the current swap block. Break this loop and * re-fetch the same pindex from the pctrie again. */ swp_pager_force_pagein(object, s_pindex, n_blks); n_blks = 0; break; } if (i == SWAP_META_PAGES) pi = sb->p + SWAP_META_PAGES; } if (n_blks > 0) swp_pager_force_pagein(object, s_pindex, n_blks); } /* * swap_pager_swapoff: * * Page in all of the pages that have been paged out to the * given device. The corresponding blocks in the bitmap must be * marked as allocated and the device must be flagged SW_CLOSING. * There may be no processes swapped out to the device. * * This routine may block. */ static void swap_pager_swapoff(struct swdevt *sp) { vm_object_t object; int retries; sx_assert(&swdev_syscall_lock, SA_XLOCKED); retries = 0; full_rescan: mtx_lock(&vm_object_list_mtx); TAILQ_FOREACH(object, &vm_object_list, object_list) { if (object->type != OBJT_SWAP) continue; mtx_unlock(&vm_object_list_mtx); /* Depends on type-stability. */ VM_OBJECT_WLOCK(object); /* * Dead objects are eventually terminated on their own. */ if ((object->flags & OBJ_DEAD) != 0) goto next_obj; /* * Sync with fences placed after pctrie * initialization. We must not access pctrie below * unless we checked that our object is swap and not * dead. */ atomic_thread_fence_acq(); if (object->type != OBJT_SWAP) goto next_obj; swap_pager_swapoff_object(sp, object); next_obj: VM_OBJECT_WUNLOCK(object); mtx_lock(&vm_object_list_mtx); } mtx_unlock(&vm_object_list_mtx); if (sp->sw_used) { /* * Objects may be locked or paging to the device being * removed, so we will miss their pages and need to * make another pass. We have marked this device as * SW_CLOSING, so the activity should finish soon. */ retries++; if (retries > 100) { panic("swapoff: failed to locate %d swap blocks", sp->sw_used); } pause("swpoff", hz / 20); goto full_rescan; } EVENTHANDLER_INVOKE(swapoff, sp); } /************************************************************************ * SWAP META DATA * ************************************************************************ * * These routines manipulate the swap metadata stored in the * OBJT_SWAP object. * * Swap metadata is implemented with a global hash and not directly * linked into the object. Instead the object simply contains * appropriate tracking counters. */ /* * SWP_PAGER_SWBLK_EMPTY() - is a range of blocks free? */ static bool swp_pager_swblk_empty(struct swblk *sb, int start, int limit) { int i; MPASS(0 <= start && start <= limit && limit <= SWAP_META_PAGES); for (i = start; i < limit; i++) { if (sb->d[i] != SWAPBLK_NONE) return (false); } return (true); } /* * SWP_PAGER_META_BUILD() - add swap block to swap meta data for object * * We first convert the object to a swap object if it is a default * object. * * The specified swapblk is added to the object's swap metadata. If * the swapblk is not valid, it is freed instead. Any previously * assigned swapblk is returned. */ static daddr_t swp_pager_meta_build(vm_object_t object, vm_pindex_t pindex, daddr_t swapblk) { static volatile int swblk_zone_exhausted, swpctrie_zone_exhausted; struct swblk *sb, *sb1; vm_pindex_t modpi, rdpi; daddr_t prev_swapblk; int error, i; VM_OBJECT_ASSERT_WLOCKED(object); /* * Convert default object to swap object if necessary */ if (object->type != OBJT_SWAP) { pctrie_init(&object->un_pager.swp.swp_blks); /* * Ensure that swap_pager_swapoff()'s iteration over * object_list does not see a garbage pctrie. */ atomic_thread_fence_rel(); object->type = OBJT_SWAP; KASSERT(object->handle == NULL, ("default pager with handle")); } rdpi = rounddown(pindex, SWAP_META_PAGES); sb = SWAP_PCTRIE_LOOKUP(&object->un_pager.swp.swp_blks, rdpi); if (sb == NULL) { if (swapblk == SWAPBLK_NONE) return (SWAPBLK_NONE); for (;;) { sb = uma_zalloc(swblk_zone, M_NOWAIT | (curproc == pageproc ? M_USE_RESERVE : 0)); if (sb != NULL) { sb->p = rdpi; for (i = 0; i < SWAP_META_PAGES; i++) sb->d[i] = SWAPBLK_NONE; if (atomic_cmpset_int(&swblk_zone_exhausted, 1, 0)) printf("swblk zone ok\n"); break; } VM_OBJECT_WUNLOCK(object); if (uma_zone_exhausted(swblk_zone)) { if (atomic_cmpset_int(&swblk_zone_exhausted, 0, 1)) printf("swap blk zone exhausted, " "increase kern.maxswzone\n"); vm_pageout_oom(VM_OOM_SWAPZ); pause("swzonxb", 10); } else uma_zwait(swblk_zone); VM_OBJECT_WLOCK(object); sb = SWAP_PCTRIE_LOOKUP(&object->un_pager.swp.swp_blks, rdpi); if (sb != NULL) /* * Somebody swapped out a nearby page, * allocating swblk at the rdpi index, * while we dropped the object lock. */ goto allocated; } for (;;) { error = SWAP_PCTRIE_INSERT( &object->un_pager.swp.swp_blks, sb); if (error == 0) { if (atomic_cmpset_int(&swpctrie_zone_exhausted, 1, 0)) printf("swpctrie zone ok\n"); break; } VM_OBJECT_WUNLOCK(object); if (uma_zone_exhausted(swpctrie_zone)) { if (atomic_cmpset_int(&swpctrie_zone_exhausted, 0, 1)) printf("swap pctrie zone exhausted, " "increase kern.maxswzone\n"); vm_pageout_oom(VM_OOM_SWAPZ); pause("swzonxp", 10); } else uma_zwait(swpctrie_zone); VM_OBJECT_WLOCK(object); sb1 = SWAP_PCTRIE_LOOKUP(&object->un_pager.swp.swp_blks, rdpi); if (sb1 != NULL) { uma_zfree(swblk_zone, sb); sb = sb1; goto allocated; } } } allocated: MPASS(sb->p == rdpi); modpi = pindex % SWAP_META_PAGES; /* Return prior contents of metadata. */ prev_swapblk = sb->d[modpi]; /* Enter block into metadata. */ sb->d[modpi] = swapblk; /* * Free the swblk if we end up with the empty page run. */ if (swapblk == SWAPBLK_NONE && swp_pager_swblk_empty(sb, 0, SWAP_META_PAGES)) { SWAP_PCTRIE_REMOVE(&object->un_pager.swp.swp_blks, rdpi); uma_zfree(swblk_zone, sb); } return (prev_swapblk); } /* * SWP_PAGER_META_FREE() - free a range of blocks in the object's swap metadata * * The requested range of blocks is freed, with any associated swap * returned to the swap bitmap. * * This routine will free swap metadata structures as they are cleaned * out. This routine does *NOT* operate on swap metadata associated * with resident pages. */ static void swp_pager_meta_free(vm_object_t object, vm_pindex_t pindex, vm_pindex_t count) { struct swblk *sb; daddr_t n_free, s_free; vm_pindex_t last; int i, limit, start; VM_OBJECT_ASSERT_WLOCKED(object); if (object->type != OBJT_SWAP || count == 0) return; swp_pager_init_freerange(&s_free, &n_free); last = pindex + count; for (;;) { sb = SWAP_PCTRIE_LOOKUP_GE(&object->un_pager.swp.swp_blks, rounddown(pindex, SWAP_META_PAGES)); if (sb == NULL || sb->p >= last) break; start = pindex > sb->p ? pindex - sb->p : 0; limit = last - sb->p < SWAP_META_PAGES ? last - sb->p : SWAP_META_PAGES; for (i = start; i < limit; i++) { if (sb->d[i] == SWAPBLK_NONE) continue; swp_pager_update_freerange(&s_free, &n_free, sb->d[i]); sb->d[i] = SWAPBLK_NONE; } pindex = sb->p + SWAP_META_PAGES; if (swp_pager_swblk_empty(sb, 0, start) && swp_pager_swblk_empty(sb, limit, SWAP_META_PAGES)) { SWAP_PCTRIE_REMOVE(&object->un_pager.swp.swp_blks, sb->p); uma_zfree(swblk_zone, sb); } } swp_pager_freeswapspace(s_free, n_free); } /* * SWP_PAGER_META_FREE_ALL() - destroy all swap metadata associated with object * * This routine locates and destroys all swap metadata associated with * an object. */ static void swp_pager_meta_free_all(vm_object_t object) { struct swblk *sb; daddr_t n_free, s_free; vm_pindex_t pindex; int i; VM_OBJECT_ASSERT_WLOCKED(object); if (object->type != OBJT_SWAP) return; swp_pager_init_freerange(&s_free, &n_free); for (pindex = 0; (sb = SWAP_PCTRIE_LOOKUP_GE( &object->un_pager.swp.swp_blks, pindex)) != NULL;) { pindex = sb->p + SWAP_META_PAGES; for (i = 0; i < SWAP_META_PAGES; i++) { if (sb->d[i] == SWAPBLK_NONE) continue; swp_pager_update_freerange(&s_free, &n_free, sb->d[i]); } SWAP_PCTRIE_REMOVE(&object->un_pager.swp.swp_blks, sb->p); uma_zfree(swblk_zone, sb); } swp_pager_freeswapspace(s_free, n_free); } /* * SWP_PAGER_METACTL() - misc control of swap meta data. * * This routine is capable of looking up, or removing swapblk * assignments in the swap meta data. It returns the swapblk being * looked-up, popped, or SWAPBLK_NONE if the block was invalid. * * When acting on a busy resident page and paging is in progress, we * have to wait until paging is complete but otherwise can act on the * busy page. * * SWM_POP remove from meta data but do not free it */ static daddr_t swp_pager_meta_ctl(vm_object_t object, vm_pindex_t pindex, int flags) { struct swblk *sb; daddr_t r1; if ((flags & SWM_POP) != 0) VM_OBJECT_ASSERT_WLOCKED(object); else VM_OBJECT_ASSERT_LOCKED(object); /* * The meta data only exists if the object is OBJT_SWAP * and even then might not be allocated yet. */ if (object->type != OBJT_SWAP) return (SWAPBLK_NONE); sb = SWAP_PCTRIE_LOOKUP(&object->un_pager.swp.swp_blks, rounddown(pindex, SWAP_META_PAGES)); if (sb == NULL) return (SWAPBLK_NONE); r1 = sb->d[pindex % SWAP_META_PAGES]; if (r1 == SWAPBLK_NONE) return (SWAPBLK_NONE); if ((flags & SWM_POP) != 0) { sb->d[pindex % SWAP_META_PAGES] = SWAPBLK_NONE; if (swp_pager_swblk_empty(sb, 0, SWAP_META_PAGES)) { SWAP_PCTRIE_REMOVE(&object->un_pager.swp.swp_blks, rounddown(pindex, SWAP_META_PAGES)); uma_zfree(swblk_zone, sb); } } return (r1); } /* * Returns the least page index which is greater than or equal to the * parameter pindex and for which there is a swap block allocated. * Returns object's size if the object's type is not swap or if there * are no allocated swap blocks for the object after the requested * pindex. */ vm_pindex_t swap_pager_find_least(vm_object_t object, vm_pindex_t pindex) { struct swblk *sb; int i; VM_OBJECT_ASSERT_LOCKED(object); if (object->type != OBJT_SWAP) return (object->size); sb = SWAP_PCTRIE_LOOKUP_GE(&object->un_pager.swp.swp_blks, rounddown(pindex, SWAP_META_PAGES)); if (sb == NULL) return (object->size); if (sb->p < pindex) { for (i = pindex % SWAP_META_PAGES; i < SWAP_META_PAGES; i++) { if (sb->d[i] != SWAPBLK_NONE) return (sb->p + i); } sb = SWAP_PCTRIE_LOOKUP_GE(&object->un_pager.swp.swp_blks, roundup(pindex, SWAP_META_PAGES)); if (sb == NULL) return (object->size); } for (i = 0; i < SWAP_META_PAGES; i++) { if (sb->d[i] != SWAPBLK_NONE) return (sb->p + i); } /* * We get here if a swblk is present in the trie but it * doesn't map any blocks. */ MPASS(0); return (object->size); } /* * System call swapon(name) enables swapping on device name, * which must be in the swdevsw. Return EBUSY * if already swapping on this device. */ #ifndef _SYS_SYSPROTO_H_ struct swapon_args { char *name; }; #endif /* * MPSAFE */ /* ARGSUSED */ int sys_swapon(struct thread *td, struct swapon_args *uap) { struct vattr attr; struct vnode *vp; struct nameidata nd; int error; error = priv_check(td, PRIV_SWAPON); if (error) return (error); sx_xlock(&swdev_syscall_lock); /* * Swap metadata may not fit in the KVM if we have physical * memory of >1GB. */ if (swblk_zone == NULL) { error = ENOMEM; goto done; } NDINIT(&nd, LOOKUP, ISOPEN | FOLLOW | AUDITVNODE1, UIO_USERSPACE, uap->name, td); error = namei(&nd); if (error) goto done; NDFREE(&nd, NDF_ONLY_PNBUF); vp = nd.ni_vp; if (vn_isdisk(vp, &error)) { error = swapongeom(vp); } else if (vp->v_type == VREG && (vp->v_mount->mnt_vfc->vfc_flags & VFCF_NETWORK) != 0 && (error = VOP_GETATTR(vp, &attr, td->td_ucred)) == 0) { /* * Allow direct swapping to NFS regular files in the same * way that nfs_mountroot() sets up diskless swapping. */ error = swaponvp(td, vp, attr.va_size / DEV_BSIZE); } if (error) vrele(vp); done: sx_xunlock(&swdev_syscall_lock); return (error); } /* * Check that the total amount of swap currently configured does not * exceed half the theoretical maximum. If it does, print a warning * message. */ static void swapon_check_swzone(void) { unsigned long maxpages, npages; npages = swap_total; /* absolute maximum we can handle assuming 100% efficiency */ maxpages = uma_zone_get_max(swblk_zone) * SWAP_META_PAGES; /* recommend using no more than half that amount */ if (npages > maxpages / 2) { printf("warning: total configured swap (%lu pages) " "exceeds maximum recommended amount (%lu pages).\n", npages, maxpages / 2); printf("warning: increase kern.maxswzone " "or reduce amount of swap.\n"); } } static void swaponsomething(struct vnode *vp, void *id, u_long nblks, sw_strategy_t *strategy, sw_close_t *close, dev_t dev, int flags) { struct swdevt *sp, *tsp; swblk_t dvbase; u_long mblocks; /* * nblks is in DEV_BSIZE'd chunks, convert to PAGE_SIZE'd chunks. * First chop nblks off to page-align it, then convert. * * sw->sw_nblks is in page-sized chunks now too. */ nblks &= ~(ctodb(1) - 1); nblks = dbtoc(nblks); /* * If we go beyond this, we get overflows in the radix * tree bitmap code. */ mblocks = 0x40000000 / BLIST_META_RADIX; if (nblks > mblocks) { printf( "WARNING: reducing swap size to maximum of %luMB per unit\n", mblocks / 1024 / 1024 * PAGE_SIZE); nblks = mblocks; } sp = malloc(sizeof *sp, M_VMPGDATA, M_WAITOK | M_ZERO); sp->sw_vp = vp; sp->sw_id = id; sp->sw_dev = dev; sp->sw_nblks = nblks; sp->sw_used = 0; sp->sw_strategy = strategy; sp->sw_close = close; sp->sw_flags = flags; sp->sw_blist = blist_create(nblks, M_WAITOK); /* * Do not free the first blocks in order to avoid overwriting * any bsd label at the front of the partition */ blist_free(sp->sw_blist, howmany(BBSIZE, PAGE_SIZE), nblks - howmany(BBSIZE, PAGE_SIZE)); dvbase = 0; mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(tsp, &swtailq, sw_list) { if (tsp->sw_end >= dvbase) { /* * We put one uncovered page between the devices * in order to definitively prevent any cross-device * I/O requests */ dvbase = tsp->sw_end + 1; } } sp->sw_first = dvbase; sp->sw_end = dvbase + nblks; TAILQ_INSERT_TAIL(&swtailq, sp, sw_list); nswapdev++; swap_pager_avail += nblks - howmany(BBSIZE, PAGE_SIZE); swap_total += nblks; swapon_check_swzone(); swp_sizecheck(); mtx_unlock(&sw_dev_mtx); EVENTHANDLER_INVOKE(swapon, sp); } /* * SYSCALL: swapoff(devname) * * Disable swapping on the given device. * * XXX: Badly designed system call: it should use a device index * rather than filename as specification. We keep sw_vp around * only to make this work. */ #ifndef _SYS_SYSPROTO_H_ struct swapoff_args { char *name; }; #endif /* * MPSAFE */ /* ARGSUSED */ int sys_swapoff(struct thread *td, struct swapoff_args *uap) { struct vnode *vp; struct nameidata nd; struct swdevt *sp; int error; error = priv_check(td, PRIV_SWAPOFF); if (error) return (error); sx_xlock(&swdev_syscall_lock); NDINIT(&nd, LOOKUP, FOLLOW | AUDITVNODE1, UIO_USERSPACE, uap->name, td); error = namei(&nd); if (error) goto done; NDFREE(&nd, NDF_ONLY_PNBUF); vp = nd.ni_vp; mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(sp, &swtailq, sw_list) { if (sp->sw_vp == vp) break; } mtx_unlock(&sw_dev_mtx); if (sp == NULL) { error = EINVAL; goto done; } error = swapoff_one(sp, td->td_ucred); done: sx_xunlock(&swdev_syscall_lock); return (error); } static int swapoff_one(struct swdevt *sp, struct ucred *cred) { u_long nblks; #ifdef MAC int error; #endif sx_assert(&swdev_syscall_lock, SA_XLOCKED); #ifdef MAC (void) vn_lock(sp->sw_vp, LK_EXCLUSIVE | LK_RETRY); error = mac_system_check_swapoff(cred, sp->sw_vp); (void) VOP_UNLOCK(sp->sw_vp, 0); if (error != 0) return (error); #endif nblks = sp->sw_nblks; /* * We can turn off this swap device safely only if the * available virtual memory in the system will fit the amount * of data we will have to page back in, plus an epsilon so * the system doesn't become critically low on swap space. */ if (vm_free_count() + swap_pager_avail < nblks + nswap_lowat) return (ENOMEM); /* * Prevent further allocations on this device. */ mtx_lock(&sw_dev_mtx); sp->sw_flags |= SW_CLOSING; swap_pager_avail -= blist_fill(sp->sw_blist, 0, nblks); swap_total -= nblks; mtx_unlock(&sw_dev_mtx); /* * Page in the contents of the device and close it. */ swap_pager_swapoff(sp); sp->sw_close(curthread, sp); mtx_lock(&sw_dev_mtx); sp->sw_id = NULL; TAILQ_REMOVE(&swtailq, sp, sw_list); nswapdev--; if (nswapdev == 0) { swap_pager_full = 2; swap_pager_almost_full = 1; } if (swdevhd == sp) swdevhd = NULL; mtx_unlock(&sw_dev_mtx); blist_destroy(sp->sw_blist); free(sp, M_VMPGDATA); return (0); } void swapoff_all(void) { struct swdevt *sp, *spt; const char *devname; int error; sx_xlock(&swdev_syscall_lock); mtx_lock(&sw_dev_mtx); TAILQ_FOREACH_SAFE(sp, &swtailq, sw_list, spt) { mtx_unlock(&sw_dev_mtx); if (vn_isdisk(sp->sw_vp, NULL)) devname = devtoname(sp->sw_vp->v_rdev); else devname = "[file]"; error = swapoff_one(sp, thread0.td_ucred); if (error != 0) { printf("Cannot remove swap device %s (error=%d), " "skipping.\n", devname, error); } else if (bootverbose) { printf("Swap device %s removed.\n", devname); } mtx_lock(&sw_dev_mtx); } mtx_unlock(&sw_dev_mtx); sx_xunlock(&swdev_syscall_lock); } void swap_pager_status(int *total, int *used) { struct swdevt *sp; *total = 0; *used = 0; mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(sp, &swtailq, sw_list) { *total += sp->sw_nblks; *used += sp->sw_used; } mtx_unlock(&sw_dev_mtx); } int swap_dev_info(int name, struct xswdev *xs, char *devname, size_t len) { struct swdevt *sp; const char *tmp_devname; int error, n; n = 0; error = ENOENT; mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(sp, &swtailq, sw_list) { if (n != name) { n++; continue; } xs->xsw_version = XSWDEV_VERSION; xs->xsw_dev = sp->sw_dev; xs->xsw_flags = sp->sw_flags; xs->xsw_nblks = sp->sw_nblks; xs->xsw_used = sp->sw_used; if (devname != NULL) { if (vn_isdisk(sp->sw_vp, NULL)) tmp_devname = devtoname(sp->sw_vp->v_rdev); else tmp_devname = "[file]"; strncpy(devname, tmp_devname, len); } error = 0; break; } mtx_unlock(&sw_dev_mtx); return (error); } #if defined(COMPAT_FREEBSD11) #define XSWDEV_VERSION_11 1 struct xswdev11 { u_int xsw_version; uint32_t xsw_dev; int xsw_flags; int xsw_nblks; int xsw_used; }; #endif #if defined(__amd64__) && defined(COMPAT_FREEBSD32) struct xswdev32 { u_int xsw_version; u_int xsw_dev1, xsw_dev2; int xsw_flags; int xsw_nblks; int xsw_used; }; #endif static int sysctl_vm_swap_info(SYSCTL_HANDLER_ARGS) { struct xswdev xs; #if defined(__amd64__) && defined(COMPAT_FREEBSD32) struct xswdev32 xs32; #endif #if defined(COMPAT_FREEBSD11) struct xswdev11 xs11; #endif int error; if (arg2 != 1) /* name length */ return (EINVAL); error = swap_dev_info(*(int *)arg1, &xs, NULL, 0); if (error != 0) return (error); #if defined(__amd64__) && defined(COMPAT_FREEBSD32) if (req->oldlen == sizeof(xs32)) { xs32.xsw_version = XSWDEV_VERSION; xs32.xsw_dev1 = xs.xsw_dev; xs32.xsw_dev2 = xs.xsw_dev >> 32; xs32.xsw_flags = xs.xsw_flags; xs32.xsw_nblks = xs.xsw_nblks; xs32.xsw_used = xs.xsw_used; error = SYSCTL_OUT(req, &xs32, sizeof(xs32)); return (error); } #endif #if defined(COMPAT_FREEBSD11) if (req->oldlen == sizeof(xs11)) { xs11.xsw_version = XSWDEV_VERSION_11; xs11.xsw_dev = xs.xsw_dev; /* truncation */ xs11.xsw_flags = xs.xsw_flags; xs11.xsw_nblks = xs.xsw_nblks; xs11.xsw_used = xs.xsw_used; error = SYSCTL_OUT(req, &xs11, sizeof(xs11)); return (error); } #endif error = SYSCTL_OUT(req, &xs, sizeof(xs)); return (error); } SYSCTL_INT(_vm, OID_AUTO, nswapdev, CTLFLAG_RD, &nswapdev, 0, "Number of swap devices"); SYSCTL_NODE(_vm, OID_AUTO, swap_info, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_vm_swap_info, "Swap statistics by device"); /* * Count the approximate swap usage in pages for a vmspace. The * shadowed or not yet copied on write swap blocks are not accounted. * The map must be locked. */ long vmspace_swap_count(struct vmspace *vmspace) { vm_map_t map; vm_map_entry_t cur; vm_object_t object; struct swblk *sb; vm_pindex_t e, pi; long count; int i; map = &vmspace->vm_map; count = 0; for (cur = map->header.next; cur != &map->header; cur = cur->next) { if ((cur->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) continue; object = cur->object.vm_object; if (object == NULL || object->type != OBJT_SWAP) continue; VM_OBJECT_RLOCK(object); if (object->type != OBJT_SWAP) goto unlock; pi = OFF_TO_IDX(cur->offset); e = pi + OFF_TO_IDX(cur->end - cur->start); for (;; pi = sb->p + SWAP_META_PAGES) { sb = SWAP_PCTRIE_LOOKUP_GE( &object->un_pager.swp.swp_blks, pi); if (sb == NULL || sb->p >= e) break; for (i = 0; i < SWAP_META_PAGES; i++) { if (sb->p + i < e && sb->d[i] != SWAPBLK_NONE) count++; } } unlock: VM_OBJECT_RUNLOCK(object); } return (count); } /* * GEOM backend * * Swapping onto disk devices. * */ static g_orphan_t swapgeom_orphan; static struct g_class g_swap_class = { .name = "SWAP", .version = G_VERSION, .orphan = swapgeom_orphan, }; DECLARE_GEOM_CLASS(g_swap_class, g_class); static void swapgeom_close_ev(void *arg, int flags) { struct g_consumer *cp; cp = arg; g_access(cp, -1, -1, 0); g_detach(cp); g_destroy_consumer(cp); } /* * Add a reference to the g_consumer for an inflight transaction. */ static void swapgeom_acquire(struct g_consumer *cp) { mtx_assert(&sw_dev_mtx, MA_OWNED); cp->index++; } /* * Remove a reference from the g_consumer. Post a close event if all * references go away, since the function might be called from the * biodone context. */ static void swapgeom_release(struct g_consumer *cp, struct swdevt *sp) { mtx_assert(&sw_dev_mtx, MA_OWNED); cp->index--; if (cp->index == 0) { if (g_post_event(swapgeom_close_ev, cp, M_NOWAIT, NULL) == 0) sp->sw_id = NULL; } } static void swapgeom_done(struct bio *bp2) { struct swdevt *sp; struct buf *bp; struct g_consumer *cp; bp = bp2->bio_caller2; cp = bp2->bio_from; bp->b_ioflags = bp2->bio_flags; if (bp2->bio_error) bp->b_ioflags |= BIO_ERROR; bp->b_resid = bp->b_bcount - bp2->bio_completed; bp->b_error = bp2->bio_error; bp->b_caller1 = NULL; bufdone(bp); sp = bp2->bio_caller1; mtx_lock(&sw_dev_mtx); swapgeom_release(cp, sp); mtx_unlock(&sw_dev_mtx); g_destroy_bio(bp2); } static void swapgeom_strategy(struct buf *bp, struct swdevt *sp) { struct bio *bio; struct g_consumer *cp; mtx_lock(&sw_dev_mtx); cp = sp->sw_id; if (cp == NULL) { mtx_unlock(&sw_dev_mtx); bp->b_error = ENXIO; bp->b_ioflags |= BIO_ERROR; bufdone(bp); return; } swapgeom_acquire(cp); mtx_unlock(&sw_dev_mtx); if (bp->b_iocmd == BIO_WRITE) bio = g_new_bio(); else bio = g_alloc_bio(); if (bio == NULL) { mtx_lock(&sw_dev_mtx); swapgeom_release(cp, sp); mtx_unlock(&sw_dev_mtx); bp->b_error = ENOMEM; bp->b_ioflags |= BIO_ERROR; printf("swap_pager: cannot allocate bio\n"); bufdone(bp); return; } bp->b_caller1 = bio; bio->bio_caller1 = sp; bio->bio_caller2 = bp; bio->bio_cmd = bp->b_iocmd; bio->bio_offset = (bp->b_blkno - sp->sw_first) * PAGE_SIZE; bio->bio_length = bp->b_bcount; bio->bio_done = swapgeom_done; if (!buf_mapped(bp)) { bio->bio_ma = bp->b_pages; bio->bio_data = unmapped_buf; bio->bio_ma_offset = (vm_offset_t)bp->b_offset & PAGE_MASK; bio->bio_ma_n = bp->b_npages; bio->bio_flags |= BIO_UNMAPPED; } else { bio->bio_data = bp->b_data; bio->bio_ma = NULL; } g_io_request(bio, cp); return; } static void swapgeom_orphan(struct g_consumer *cp) { struct swdevt *sp; int destroy; mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(sp, &swtailq, sw_list) { if (sp->sw_id == cp) { sp->sw_flags |= SW_CLOSING; break; } } /* * Drop reference we were created with. Do directly since we're in a * special context where we don't have to queue the call to * swapgeom_close_ev(). */ cp->index--; destroy = ((sp != NULL) && (cp->index == 0)); if (destroy) sp->sw_id = NULL; mtx_unlock(&sw_dev_mtx); if (destroy) swapgeom_close_ev(cp, 0); } static void swapgeom_close(struct thread *td, struct swdevt *sw) { struct g_consumer *cp; mtx_lock(&sw_dev_mtx); cp = sw->sw_id; sw->sw_id = NULL; mtx_unlock(&sw_dev_mtx); /* * swapgeom_close() may be called from the biodone context, * where we cannot perform topology changes. Delegate the * work to the events thread. */ if (cp != NULL) g_waitfor_event(swapgeom_close_ev, cp, M_WAITOK, NULL); } static int swapongeom_locked(struct cdev *dev, struct vnode *vp) { struct g_provider *pp; struct g_consumer *cp; static struct g_geom *gp; struct swdevt *sp; u_long nblks; int error; pp = g_dev_getprovider(dev); if (pp == NULL) return (ENODEV); mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(sp, &swtailq, sw_list) { cp = sp->sw_id; if (cp != NULL && cp->provider == pp) { mtx_unlock(&sw_dev_mtx); return (EBUSY); } } mtx_unlock(&sw_dev_mtx); if (gp == NULL) gp = g_new_geomf(&g_swap_class, "swap"); cp = g_new_consumer(gp); cp->index = 1; /* Number of active I/Os, plus one for being active. */ cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE; g_attach(cp, pp); /* * XXX: Every time you think you can improve the margin for * footshooting, somebody depends on the ability to do so: * savecore(8) wants to write to our swapdev so we cannot * set an exclusive count :-( */ error = g_access(cp, 1, 1, 0); if (error != 0) { g_detach(cp); g_destroy_consumer(cp); return (error); } nblks = pp->mediasize / DEV_BSIZE; swaponsomething(vp, cp, nblks, swapgeom_strategy, swapgeom_close, dev2udev(dev), (pp->flags & G_PF_ACCEPT_UNMAPPED) != 0 ? SW_UNMAPPED : 0); return (0); } static int swapongeom(struct vnode *vp) { int error; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); if (vp->v_type != VCHR || (vp->v_iflag & VI_DOOMED) != 0) { error = ENOENT; } else { g_topology_lock(); error = swapongeom_locked(vp->v_rdev, vp); g_topology_unlock(); } VOP_UNLOCK(vp, 0); return (error); } /* * VNODE backend * * This is used mainly for network filesystem (read: probably only tested * with NFS) swapfiles. * */ static void swapdev_strategy(struct buf *bp, struct swdevt *sp) { struct vnode *vp2; bp->b_blkno = ctodb(bp->b_blkno - sp->sw_first); vp2 = sp->sw_id; vhold(vp2); if (bp->b_iocmd == BIO_WRITE) { if (bp->b_bufobj) bufobj_wdrop(bp->b_bufobj); bufobj_wref(&vp2->v_bufobj); } if (bp->b_bufobj != &vp2->v_bufobj) bp->b_bufobj = &vp2->v_bufobj; bp->b_vp = vp2; bp->b_iooffset = dbtob(bp->b_blkno); bstrategy(bp); return; } static void swapdev_close(struct thread *td, struct swdevt *sp) { VOP_CLOSE(sp->sw_vp, FREAD | FWRITE, td->td_ucred, td); vrele(sp->sw_vp); } static int swaponvp(struct thread *td, struct vnode *vp, u_long nblks) { struct swdevt *sp; int error; if (nblks == 0) return (ENXIO); mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(sp, &swtailq, sw_list) { if (sp->sw_id == vp) { mtx_unlock(&sw_dev_mtx); return (EBUSY); } } mtx_unlock(&sw_dev_mtx); (void) vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); #ifdef MAC error = mac_system_check_swapon(td->td_ucred, vp); if (error == 0) #endif error = VOP_OPEN(vp, FREAD | FWRITE, td->td_ucred, td, NULL); (void) VOP_UNLOCK(vp, 0); if (error) return (error); swaponsomething(vp, vp, nblks, swapdev_strategy, swapdev_close, NODEV, 0); return (0); } static int sysctl_swap_async_max(SYSCTL_HANDLER_ARGS) { int error, new, n; new = nsw_wcount_async_max; error = sysctl_handle_int(oidp, &new, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (new > nswbuf / 2 || new < 1) return (EINVAL); mtx_lock(&swbuf_mtx); while (nsw_wcount_async_max != new) { /* * Adjust difference. If the current async count is too low, * we will need to sqeeze our update slowly in. Sleep with a * higher priority than getpbuf() to finish faster. */ n = new - nsw_wcount_async_max; if (nsw_wcount_async + n >= 0) { nsw_wcount_async += n; nsw_wcount_async_max += n; wakeup(&nsw_wcount_async); } else { nsw_wcount_async_max -= nsw_wcount_async; nsw_wcount_async = 0; msleep(&nsw_wcount_async, &swbuf_mtx, PSWP, "swpsysctl", 0); } } mtx_unlock(&swbuf_mtx); return (0); } Index: head/sys/vm/vm_object.c =================================================================== --- head/sys/vm/vm_object.c (revision 351240) +++ head/sys/vm/vm_object.c (revision 351241) @@ -1,2687 +1,2681 @@ /*- * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU) * * Copyright (c) 1991, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * The Mach Operating System project at Carnegie-Mellon University. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from: @(#)vm_object.c 8.5 (Berkeley) 3/22/94 * * * Copyright (c) 1987, 1990 Carnegie-Mellon University. * All rights reserved. * * Authors: Avadis Tevanian, Jr., Michael Wayne Young * * Permission to use, copy, modify and distribute this software and * its documentation is hereby granted, provided that both the copyright * notice and this permission notice appear in all copies of the * software, derivative works or modified versions, and any portions * thereof, and that both notices appear in supporting documentation. * * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. * * Carnegie Mellon requests users of this software to return to * * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU * School of Computer Science * Carnegie Mellon University * Pittsburgh PA 15213-3890 * * any improvements or extensions that they make and grant Carnegie the * rights to redistribute these changes. */ /* * Virtual memory object module. */ #include __FBSDID("$FreeBSD$"); #include "opt_vm.h" #include #include #include #include #include #include #include #include #include #include #include /* for curproc, pageproc */ +#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 int old_msync; SYSCTL_INT(_vm, OID_AUTO, old_msync, CTLFLAG_RW, &old_msync, 0, "Use old (insecure) msync behavior"); static int vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int pagerflags, int flags, boolean_t *clearobjflags, boolean_t *eio); static boolean_t vm_object_page_remove_write(vm_page_t p, int flags, boolean_t *clearobjflags); static void vm_object_qcollapse(vm_object_t object); static void vm_object_vndeallocate(vm_object_t object); /* * Virtual memory objects maintain the actual data * associated with allocated virtual memory. A given * page of memory exists within exactly one object. * * An object is only deallocated when all "references" * are given up. Only one "reference" to a given * region of an object should be writeable. * * Associated with each object is a list of all resident * memory pages belonging to that object; this list is * maintained by the "vm_page" module, and locked by the object's * lock. * * Each object also records a "pager" routine which is * used to retrieve (and store) pages to the proper backing * storage. In addition, objects may be backed by other * objects from which they were virtual-copied. * * The only items within the object structure which are * modified after time of creation are: * reference count locked by object's lock * pager routine locked by object's lock * */ struct object_q vm_object_list; struct mtx vm_object_list_mtx; /* lock for object list and count */ struct vm_object kernel_object_store; static SYSCTL_NODE(_vm_stats, OID_AUTO, object, CTLFLAG_RD, 0, "VM object stats"); static counter_u64_t object_collapses = EARLY_COUNTER; SYSCTL_COUNTER_U64(_vm_stats_object, OID_AUTO, collapses, CTLFLAG_RD, &object_collapses, "VM object collapses"); static counter_u64_t object_bypasses = EARLY_COUNTER; SYSCTL_COUNTER_U64(_vm_stats_object, OID_AUTO, bypasses, CTLFLAG_RD, &object_bypasses, "VM object bypasses"); static void counter_startup(void) { object_collapses = counter_u64_alloc(M_WAITOK); object_bypasses = counter_u64_alloc(M_WAITOK); } SYSINIT(object_counters, SI_SUB_CPU, SI_ORDER_ANY, counter_startup, NULL); static uma_zone_t obj_zone; static int vm_object_zinit(void *mem, int size, int flags); #ifdef INVARIANTS static void vm_object_zdtor(void *mem, int size, void *arg); static void vm_object_zdtor(void *mem, int size, void *arg) { vm_object_t object; object = (vm_object_t)mem; KASSERT(object->ref_count == 0, ("object %p ref_count = %d", object, object->ref_count)); KASSERT(TAILQ_EMPTY(&object->memq), ("object %p has resident pages in its memq", object)); KASSERT(vm_radix_is_empty(&object->rtree), ("object %p has resident pages in its trie", object)); #if VM_NRESERVLEVEL > 0 KASSERT(LIST_EMPTY(&object->rvq), ("object %p has reservations", object)); #endif KASSERT(object->paging_in_progress == 0, ("object %p paging_in_progress = %d", object, object->paging_in_progress)); KASSERT(object->resident_page_count == 0, ("object %p resident_page_count = %d", object, object->resident_page_count)); KASSERT(object->shadow_count == 0, ("object %p shadow_count = %d", object, object->shadow_count)); KASSERT(object->type == OBJT_DEAD, ("object %p has non-dead type %d", object, object->type)); } #endif static int vm_object_zinit(void *mem, int size, int flags) { vm_object_t object; object = (vm_object_t)mem; rw_init_flags(&object->lock, "vm object", RW_DUPOK | RW_NEW); /* These are true for any object that has been freed */ object->type = OBJT_DEAD; object->ref_count = 0; vm_radix_init(&object->rtree); - object->paging_in_progress = 0; + refcount_init(&object->paging_in_progress, 0); object->resident_page_count = 0; object->shadow_count = 0; object->flags = OBJ_DEAD; mtx_lock(&vm_object_list_mtx); TAILQ_INSERT_TAIL(&vm_object_list, object, object_list); mtx_unlock(&vm_object_list_mtx); return (0); } static void _vm_object_allocate(objtype_t type, vm_pindex_t size, vm_object_t object) { TAILQ_INIT(&object->memq); LIST_INIT(&object->shadow_head); object->type = type; if (type == OBJT_SWAP) pctrie_init(&object->un_pager.swp.swp_blks); /* * Ensure that swap_pager_swapoff() iteration over object_list * sees up to date type and pctrie head if it observed * non-dead object. */ atomic_thread_fence_rel(); switch (type) { case OBJT_DEAD: panic("_vm_object_allocate: can't create OBJT_DEAD"); case OBJT_DEFAULT: case OBJT_SWAP: object->flags = OBJ_ONEMAPPING; break; case OBJT_DEVICE: case OBJT_SG: object->flags = OBJ_FICTITIOUS | OBJ_UNMANAGED; break; case OBJT_MGTDEVICE: object->flags = OBJ_FICTITIOUS; break; case OBJT_PHYS: object->flags = OBJ_UNMANAGED; break; case OBJT_VNODE: object->flags = 0; break; default: panic("_vm_object_allocate: type %d is undefined", type); } object->size = size; object->domain.dr_policy = NULL; object->generation = 1; object->ref_count = 1; object->memattr = VM_MEMATTR_DEFAULT; object->cred = NULL; object->charge = 0; object->handle = NULL; object->backing_object = NULL; object->backing_object_offset = (vm_ooffset_t) 0; #if VM_NRESERVLEVEL > 0 LIST_INIT(&object->rvq); #endif umtx_shm_object_init(object); } /* * vm_object_init: * * Initialize the VM objects module. */ void vm_object_init(void) { TAILQ_INIT(&vm_object_list); mtx_init(&vm_object_list_mtx, "vm object_list", NULL, MTX_DEF); rw_init(&kernel_object->lock, "kernel vm object"); _vm_object_allocate(OBJT_PHYS, atop(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS), kernel_object); #if VM_NRESERVLEVEL > 0 kernel_object->flags |= OBJ_COLORED; kernel_object->pg_color = (u_short)atop(VM_MIN_KERNEL_ADDRESS); #endif /* * The lock portion of struct vm_object must be type stable due * to vm_pageout_fallback_object_lock locking a vm object * without holding any references to it. */ obj_zone = uma_zcreate("VM OBJECT", sizeof (struct vm_object), NULL, #ifdef INVARIANTS vm_object_zdtor, #else NULL, #endif vm_object_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); vm_radix_zinit(); } void vm_object_clear_flag(vm_object_t object, u_short bits) { VM_OBJECT_ASSERT_WLOCKED(object); object->flags &= ~bits; } /* * Sets the default memory attribute for the specified object. Pages * that are allocated to this object are by default assigned this memory * attribute. * * Presently, this function must be called before any pages are allocated * to the object. In the future, this requirement may be relaxed for * "default" and "swap" objects. */ int vm_object_set_memattr(vm_object_t object, vm_memattr_t memattr) { VM_OBJECT_ASSERT_WLOCKED(object); switch (object->type) { case OBJT_DEFAULT: case OBJT_DEVICE: case OBJT_MGTDEVICE: case OBJT_PHYS: case OBJT_SG: case OBJT_SWAP: case OBJT_VNODE: if (!TAILQ_EMPTY(&object->memq)) return (KERN_FAILURE); break; case OBJT_DEAD: return (KERN_INVALID_ARGUMENT); default: panic("vm_object_set_memattr: object %p is of undefined type", object); } object->memattr = memattr; return (KERN_SUCCESS); } void vm_object_pip_add(vm_object_t object, short i) { - VM_OBJECT_ASSERT_WLOCKED(object); - object->paging_in_progress += i; + refcount_acquiren(&object->paging_in_progress, i); } void -vm_object_pip_subtract(vm_object_t object, short i) +vm_object_pip_wakeup(vm_object_t object) { - VM_OBJECT_ASSERT_WLOCKED(object); - object->paging_in_progress -= i; + refcount_release(&object->paging_in_progress); } void -vm_object_pip_wakeup(vm_object_t object) +vm_object_pip_wakeupn(vm_object_t object, short i) { - VM_OBJECT_ASSERT_WLOCKED(object); - object->paging_in_progress--; - if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) { - vm_object_clear_flag(object, OBJ_PIPWNT); - wakeup(object); - } + refcount_releasen(&object->paging_in_progress, i); } void -vm_object_pip_wakeupn(vm_object_t object, short i) +vm_object_pip_wait(vm_object_t object, char *waitid) { VM_OBJECT_ASSERT_WLOCKED(object); - if (i) - object->paging_in_progress -= i; - if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) { - vm_object_clear_flag(object, OBJ_PIPWNT); - wakeup(object); + + while (object->paging_in_progress) { + VM_OBJECT_WUNLOCK(object); + refcount_wait(&object->paging_in_progress, waitid, PVM); + VM_OBJECT_WLOCK(object); } } void -vm_object_pip_wait(vm_object_t object, char *waitid) +vm_object_pip_wait_unlocked(vm_object_t object, char *waitid) { - VM_OBJECT_ASSERT_WLOCKED(object); - while (object->paging_in_progress) { - object->flags |= OBJ_PIPWNT; - VM_OBJECT_SLEEP(object, object, PVM, waitid, 0); - } + VM_OBJECT_ASSERT_UNLOCKED(object); + + while (object->paging_in_progress) + refcount_wait(&object->paging_in_progress, waitid, PVM); } /* * vm_object_allocate: * * Returns a new object with the given size. */ vm_object_t vm_object_allocate(objtype_t type, vm_pindex_t size) { vm_object_t object; object = (vm_object_t)uma_zalloc(obj_zone, M_WAITOK); _vm_object_allocate(type, size, object); return (object); } /* * vm_object_reference: * * Gets another reference to the given object. Note: OBJ_DEAD * objects can be referenced during final cleaning. */ void vm_object_reference(vm_object_t object) { if (object == NULL) return; VM_OBJECT_WLOCK(object); vm_object_reference_locked(object); VM_OBJECT_WUNLOCK(object); } /* * vm_object_reference_locked: * * Gets another reference to the given object. * * The object must be locked. */ void vm_object_reference_locked(vm_object_t object) { struct vnode *vp; VM_OBJECT_ASSERT_WLOCKED(object); object->ref_count++; if (object->type == OBJT_VNODE) { vp = object->handle; vref(vp); } } /* * Handle deallocating an object of type OBJT_VNODE. */ static void vm_object_vndeallocate(vm_object_t object) { struct vnode *vp = (struct vnode *) object->handle; VM_OBJECT_ASSERT_WLOCKED(object); KASSERT(object->type == OBJT_VNODE, ("vm_object_vndeallocate: not a vnode object")); KASSERT(vp != NULL, ("vm_object_vndeallocate: missing vp")); #ifdef INVARIANTS if (object->ref_count == 0) { vn_printf(vp, "vm_object_vndeallocate "); panic("vm_object_vndeallocate: bad object reference count"); } #endif if (!umtx_shm_vnobj_persistent && object->ref_count == 1) umtx_shm_object_terminated(object); object->ref_count--; /* vrele may need the vnode lock. */ VM_OBJECT_WUNLOCK(object); vrele(vp); } /* * vm_object_deallocate: * * Release a reference to the specified object, * gained either through a vm_object_allocate * or a vm_object_reference call. When all references * are gone, storage associated with this object * may be relinquished. * * No object may be locked. */ void vm_object_deallocate(vm_object_t object) { vm_object_t temp; struct vnode *vp; while (object != NULL) { VM_OBJECT_WLOCK(object); if (object->type == OBJT_VNODE) { vm_object_vndeallocate(object); return; } KASSERT(object->ref_count != 0, ("vm_object_deallocate: object deallocated too many times: %d", object->type)); /* * If the reference count goes to 0 we start calling * vm_object_terminate() on the object chain. * A ref count of 1 may be a special case depending on the * shadow count being 0 or 1. */ object->ref_count--; if (object->ref_count > 1) { VM_OBJECT_WUNLOCK(object); return; } else if (object->ref_count == 1) { if (object->type == OBJT_SWAP && (object->flags & OBJ_TMPFS) != 0) { vp = object->un_pager.swp.swp_tmpfs; vhold(vp); VM_OBJECT_WUNLOCK(object); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); VM_OBJECT_WLOCK(object); if (object->type == OBJT_DEAD || object->ref_count != 1) { VM_OBJECT_WUNLOCK(object); VOP_UNLOCK(vp, 0); vdrop(vp); return; } if ((object->flags & OBJ_TMPFS) != 0) VOP_UNSET_TEXT(vp); VOP_UNLOCK(vp, 0); vdrop(vp); } if (object->shadow_count == 0 && object->handle == NULL && (object->type == OBJT_DEFAULT || (object->type == OBJT_SWAP && (object->flags & OBJ_TMPFS_NODE) == 0))) { vm_object_set_flag(object, OBJ_ONEMAPPING); } else if ((object->shadow_count == 1) && (object->handle == NULL) && (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) { vm_object_t robject; robject = LIST_FIRST(&object->shadow_head); KASSERT(robject != NULL, ("vm_object_deallocate: ref_count: %d, shadow_count: %d", object->ref_count, object->shadow_count)); KASSERT((robject->flags & OBJ_TMPFS_NODE) == 0, ("shadowed tmpfs v_object %p", object)); if (!VM_OBJECT_TRYWLOCK(robject)) { /* * Avoid a potential deadlock. */ object->ref_count++; VM_OBJECT_WUNLOCK(object); /* * More likely than not the thread * holding robject's lock has lower * priority than the current thread. * Let the lower priority thread run. */ pause("vmo_de", 1); continue; } /* * Collapse object into its shadow unless its * shadow is dead. In that case, object will * be deallocated by the thread that is * deallocating its shadow. */ if ((robject->flags & OBJ_DEAD) == 0 && (robject->handle == NULL) && (robject->type == OBJT_DEFAULT || robject->type == OBJT_SWAP)) { robject->ref_count++; retry: if (robject->paging_in_progress) { VM_OBJECT_WUNLOCK(object); vm_object_pip_wait(robject, "objde1"); temp = robject->backing_object; if (object == temp) { VM_OBJECT_WLOCK(object); goto retry; } } else if (object->paging_in_progress) { VM_OBJECT_WUNLOCK(robject); - object->flags |= OBJ_PIPWNT; - VM_OBJECT_SLEEP(object, object, - PDROP | PVM, "objde2", 0); + VM_OBJECT_WUNLOCK(object); + refcount_wait( + &object->paging_in_progress, + "objde2", PVM); VM_OBJECT_WLOCK(robject); temp = robject->backing_object; if (object == temp) { VM_OBJECT_WLOCK(object); goto retry; } } else VM_OBJECT_WUNLOCK(object); if (robject->ref_count == 1) { robject->ref_count--; object = robject; goto doterm; } object = robject; vm_object_collapse(object); VM_OBJECT_WUNLOCK(object); continue; } VM_OBJECT_WUNLOCK(robject); } VM_OBJECT_WUNLOCK(object); return; } doterm: umtx_shm_object_terminated(object); temp = object->backing_object; if (temp != NULL) { KASSERT((object->flags & OBJ_TMPFS_NODE) == 0, ("shadowed tmpfs v_object 2 %p", object)); VM_OBJECT_WLOCK(temp); LIST_REMOVE(object, shadow_list); temp->shadow_count--; VM_OBJECT_WUNLOCK(temp); object->backing_object = NULL; } /* * Don't double-terminate, we could be in a termination * recursion due to the terminate having to sync data * to disk. */ if ((object->flags & OBJ_DEAD) == 0) vm_object_terminate(object); else VM_OBJECT_WUNLOCK(object); object = temp; } } /* * vm_object_destroy removes the object from the global object list * and frees the space for the object. */ void vm_object_destroy(vm_object_t object) { /* * Release the allocation charge. */ if (object->cred != NULL) { swap_release_by_cred(object->charge, object->cred); object->charge = 0; crfree(object->cred); object->cred = NULL; } /* * Free the space for the object. */ uma_zfree(obj_zone, object); } /* * vm_object_terminate_pages removes any remaining pageable pages * from the object and resets the object to an empty state. */ static void vm_object_terminate_pages(vm_object_t object) { vm_page_t p, p_next; struct mtx *mtx; VM_OBJECT_ASSERT_WLOCKED(object); mtx = NULL; /* * Free any remaining pageable pages. This also removes them from the * paging queues. However, don't free wired pages, just remove them * from the object. Rather than incrementally removing each page from * the object, the page and object are reset to any empty state. */ TAILQ_FOREACH_SAFE(p, &object->memq, listq, p_next) { vm_page_assert_unbusied(p); if ((object->flags & OBJ_UNMANAGED) == 0) /* * vm_page_free_prep() only needs the page * lock for managed pages. */ vm_page_change_lock(p, &mtx); p->object = NULL; if (vm_page_wired(p)) continue; VM_CNT_INC(v_pfree); vm_page_free(p); } if (mtx != NULL) mtx_unlock(mtx); /* * If the object contained any pages, then reset it to an empty state. * None of the object's fields, including "resident_page_count", were * modified by the preceding loop. */ if (object->resident_page_count != 0) { vm_radix_reclaim_allnodes(&object->rtree); TAILQ_INIT(&object->memq); object->resident_page_count = 0; if (object->type == OBJT_VNODE) vdrop(object->handle); } } /* * vm_object_terminate actually destroys the specified object, freeing * up all previously used resources. * * The object must be locked. * This routine may block. */ void vm_object_terminate(vm_object_t object) { VM_OBJECT_ASSERT_WLOCKED(object); /* * Make sure no one uses us. */ vm_object_set_flag(object, OBJ_DEAD); /* - * wait for the pageout daemon to be done with the object - */ - vm_object_pip_wait(object, "objtrm"); - - KASSERT(!object->paging_in_progress, - ("vm_object_terminate: pageout in progress")); - - /* * Clean and free the pages, as appropriate. All references to the * object are gone, so we don't need to lock it. */ if (object->type == OBJT_VNODE) { struct vnode *vp = (struct vnode *)object->handle; /* * Clean pages and flush buffers. */ vm_object_page_clean(object, 0, 0, OBJPC_SYNC); VM_OBJECT_WUNLOCK(object); vinvalbuf(vp, V_SAVE, 0, 0); BO_LOCK(&vp->v_bufobj); vp->v_bufobj.bo_flag |= BO_DEAD; BO_UNLOCK(&vp->v_bufobj); VM_OBJECT_WLOCK(object); } + + /* + * wait for the pageout daemon to be done with the object + */ + vm_object_pip_wait(object, "objtrm"); + + KASSERT(!object->paging_in_progress, + ("vm_object_terminate: pageout in progress")); KASSERT(object->ref_count == 0, ("vm_object_terminate: object with references, ref_count=%d", object->ref_count)); if ((object->flags & OBJ_PG_DTOR) == 0) vm_object_terminate_pages(object); #if VM_NRESERVLEVEL > 0 if (__predict_false(!LIST_EMPTY(&object->rvq))) vm_reserv_break_all(object); #endif KASSERT(object->cred == NULL || object->type == OBJT_DEFAULT || object->type == OBJT_SWAP, ("%s: non-swap obj %p has cred", __func__, object)); /* * Let the pager know object is dead. */ vm_pager_deallocate(object); VM_OBJECT_WUNLOCK(object); vm_object_destroy(object); } /* * Make the page read-only so that we can clear the object flags. However, if * this is a nosync mmap then the object is likely to stay dirty so do not * mess with the page and do not clear the object flags. Returns TRUE if the * page should be flushed, and FALSE otherwise. */ static boolean_t vm_object_page_remove_write(vm_page_t p, int flags, boolean_t *clearobjflags) { /* * If we have been asked to skip nosync pages and this is a * nosync page, skip it. Note that the object flags were not * cleared in this case so we do not have to set them. */ if ((flags & OBJPC_NOSYNC) != 0 && (p->oflags & VPO_NOSYNC) != 0) { *clearobjflags = FALSE; return (FALSE); } else { pmap_remove_write(p); return (p->dirty != 0); } } /* * vm_object_page_clean * * Clean all dirty pages in the specified range of object. Leaves page * on whatever queue it is currently on. If NOSYNC is set then do not * write out pages with VPO_NOSYNC set (originally comes from MAP_NOSYNC), * leaving the object dirty. * * When stuffing pages asynchronously, allow clustering. XXX we need a * synchronous clustering mode implementation. * * Odd semantics: if start == end, we clean everything. * * The object must be locked. * * Returns FALSE if some page from the range was not written, as * reported by the pager, and TRUE otherwise. */ boolean_t vm_object_page_clean(vm_object_t object, vm_ooffset_t start, vm_ooffset_t end, int flags) { vm_page_t np, p; vm_pindex_t pi, tend, tstart; int curgeneration, n, pagerflags; boolean_t clearobjflags, eio, res; VM_OBJECT_ASSERT_WLOCKED(object); /* * The OBJ_MIGHTBEDIRTY flag is only set for OBJT_VNODE * objects. The check below prevents the function from * operating on non-vnode objects. */ if ((object->flags & OBJ_MIGHTBEDIRTY) == 0 || object->resident_page_count == 0) return (TRUE); pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) != 0 ? VM_PAGER_PUT_SYNC : VM_PAGER_CLUSTER_OK; pagerflags |= (flags & OBJPC_INVAL) != 0 ? VM_PAGER_PUT_INVAL : 0; tstart = OFF_TO_IDX(start); tend = (end == 0) ? object->size : OFF_TO_IDX(end + PAGE_MASK); clearobjflags = tstart == 0 && tend >= object->size; res = TRUE; rescan: curgeneration = object->generation; for (p = vm_page_find_least(object, tstart); p != NULL; p = np) { pi = p->pindex; if (pi >= tend) break; np = TAILQ_NEXT(p, listq); if (p->valid == 0) continue; if (vm_page_sleep_if_busy(p, "vpcwai")) { if (object->generation != curgeneration) { if ((flags & OBJPC_SYNC) != 0) goto rescan; else clearobjflags = FALSE; } np = vm_page_find_least(object, pi); continue; } if (!vm_object_page_remove_write(p, flags, &clearobjflags)) continue; n = vm_object_page_collect_flush(object, p, pagerflags, flags, &clearobjflags, &eio); if (eio) { res = FALSE; clearobjflags = FALSE; } if (object->generation != curgeneration) { if ((flags & OBJPC_SYNC) != 0) goto rescan; else clearobjflags = FALSE; } /* * If the VOP_PUTPAGES() did a truncated write, so * that even the first page of the run is not fully * written, vm_pageout_flush() returns 0 as the run * length. Since the condition that caused truncated * write may be permanent, e.g. exhausted free space, * accepting n == 0 would cause an infinite loop. * * Forwarding the iterator leaves the unwritten page * behind, but there is not much we can do there if * filesystem refuses to write it. */ if (n == 0) { n = 1; clearobjflags = FALSE; } np = vm_page_find_least(object, pi + n); } #if 0 VOP_FSYNC(vp, (pagerflags & VM_PAGER_PUT_SYNC) ? MNT_WAIT : 0); #endif if (clearobjflags) vm_object_clear_flag(object, OBJ_MIGHTBEDIRTY); return (res); } static int vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int pagerflags, int flags, boolean_t *clearobjflags, boolean_t *eio) { vm_page_t ma[vm_pageout_page_count], p_first, tp; int count, i, mreq, runlen; vm_page_lock_assert(p, MA_NOTOWNED); VM_OBJECT_ASSERT_WLOCKED(object); count = 1; mreq = 0; for (tp = p; count < vm_pageout_page_count; count++) { tp = vm_page_next(tp); if (tp == NULL || vm_page_busied(tp)) break; if (!vm_object_page_remove_write(tp, flags, clearobjflags)) break; } for (p_first = p; count < vm_pageout_page_count; count++) { tp = vm_page_prev(p_first); if (tp == NULL || vm_page_busied(tp)) break; if (!vm_object_page_remove_write(tp, flags, clearobjflags)) break; p_first = tp; mreq++; } for (tp = p_first, i = 0; i < count; tp = TAILQ_NEXT(tp, listq), i++) ma[i] = tp; vm_pageout_flush(ma, count, pagerflags, mreq, &runlen, eio); return (runlen); } /* * Note that there is absolutely no sense in writing out * anonymous objects, so we track down the vnode object * to write out. * We invalidate (remove) all pages from the address space * for semantic correctness. * * If the backing object is a device object with unmanaged pages, then any * mappings to the specified range of pages must be removed before this * function is called. * * Note: certain anonymous maps, such as MAP_NOSYNC maps, * may start out with a NULL object. */ boolean_t vm_object_sync(vm_object_t object, vm_ooffset_t offset, vm_size_t size, boolean_t syncio, boolean_t invalidate) { vm_object_t backing_object; struct vnode *vp; struct mount *mp; int error, flags, fsync_after; boolean_t res; if (object == NULL) return (TRUE); res = TRUE; error = 0; VM_OBJECT_WLOCK(object); while ((backing_object = object->backing_object) != NULL) { VM_OBJECT_WLOCK(backing_object); offset += object->backing_object_offset; VM_OBJECT_WUNLOCK(object); object = backing_object; if (object->size < OFF_TO_IDX(offset + size)) size = IDX_TO_OFF(object->size) - offset; } /* * Flush pages if writing is allowed, invalidate them * if invalidation requested. Pages undergoing I/O * will be ignored by vm_object_page_remove(). * * We cannot lock the vnode and then wait for paging * to complete without deadlocking against vm_fault. * Instead we simply call vm_object_page_remove() and * allow it to block internally on a page-by-page * basis when it encounters pages undergoing async * I/O. */ if (object->type == OBJT_VNODE && (object->flags & OBJ_MIGHTBEDIRTY) != 0 && ((vp = object->handle)->v_vflag & VV_NOSYNC) == 0) { VM_OBJECT_WUNLOCK(object); (void) vn_start_write(vp, &mp, V_WAIT); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); if (syncio && !invalidate && offset == 0 && atop(size) == object->size) { /* * If syncing the whole mapping of the file, * it is faster to schedule all the writes in * async mode, also allowing the clustering, * and then wait for i/o to complete. */ flags = 0; fsync_after = TRUE; } else { flags = (syncio || invalidate) ? OBJPC_SYNC : 0; flags |= invalidate ? (OBJPC_SYNC | OBJPC_INVAL) : 0; fsync_after = FALSE; } VM_OBJECT_WLOCK(object); res = vm_object_page_clean(object, offset, offset + size, flags); VM_OBJECT_WUNLOCK(object); if (fsync_after) error = VOP_FSYNC(vp, MNT_WAIT, curthread); VOP_UNLOCK(vp, 0); vn_finished_write(mp); if (error != 0) res = FALSE; VM_OBJECT_WLOCK(object); } if ((object->type == OBJT_VNODE || object->type == OBJT_DEVICE) && invalidate) { if (object->type == OBJT_DEVICE) /* * The option OBJPR_NOTMAPPED must be passed here * because vm_object_page_remove() cannot remove * unmanaged mappings. */ flags = OBJPR_NOTMAPPED; else if (old_msync) flags = 0; else flags = OBJPR_CLEANONLY; vm_object_page_remove(object, OFF_TO_IDX(offset), OFF_TO_IDX(offset + size + PAGE_MASK), flags); } VM_OBJECT_WUNLOCK(object); return (res); } /* * Determine whether the given advice can be applied to the object. Advice is * not applied to unmanaged pages since they never belong to page queues, and * since MADV_FREE is destructive, it can apply only to anonymous pages that * have been mapped at most once. */ static bool vm_object_advice_applies(vm_object_t object, int advice) { if ((object->flags & OBJ_UNMANAGED) != 0) return (false); if (advice != MADV_FREE) return (true); return ((object->type == OBJT_DEFAULT || object->type == OBJT_SWAP) && (object->flags & OBJ_ONEMAPPING) != 0); } static void vm_object_madvise_freespace(vm_object_t object, int advice, vm_pindex_t pindex, vm_size_t size) { if (advice == MADV_FREE && object->type == OBJT_SWAP) swap_pager_freespace(object, pindex, size); } /* * vm_object_madvise: * * Implements the madvise function at the object/page level. * * MADV_WILLNEED (any object) * * Activate the specified pages if they are resident. * * MADV_DONTNEED (any object) * * Deactivate the specified pages if they are resident. * * MADV_FREE (OBJT_DEFAULT/OBJT_SWAP objects, * OBJ_ONEMAPPING only) * * Deactivate and clean the specified pages if they are * resident. This permits the process to reuse the pages * without faulting or the kernel to reclaim the pages * without I/O. */ void vm_object_madvise(vm_object_t object, vm_pindex_t pindex, vm_pindex_t end, int advice) { vm_pindex_t tpindex; vm_object_t backing_object, tobject; vm_page_t m, tm; if (object == NULL) return; relookup: VM_OBJECT_WLOCK(object); if (!vm_object_advice_applies(object, advice)) { VM_OBJECT_WUNLOCK(object); return; } for (m = vm_page_find_least(object, pindex); pindex < end; pindex++) { tobject = object; /* * If the next page isn't resident in the top-level object, we * need to search the shadow chain. When applying MADV_FREE, we * take care to release any swap space used to store * non-resident pages. */ if (m == NULL || pindex < m->pindex) { /* * Optimize a common case: if the top-level object has * no backing object, we can skip over the non-resident * range in constant time. */ if (object->backing_object == NULL) { tpindex = (m != NULL && m->pindex < end) ? m->pindex : end; vm_object_madvise_freespace(object, advice, pindex, tpindex - pindex); if ((pindex = tpindex) == end) break; goto next_page; } tpindex = pindex; do { vm_object_madvise_freespace(tobject, advice, tpindex, 1); /* * Prepare to search the next object in the * chain. */ backing_object = tobject->backing_object; if (backing_object == NULL) goto next_pindex; VM_OBJECT_WLOCK(backing_object); tpindex += OFF_TO_IDX(tobject->backing_object_offset); if (tobject != object) VM_OBJECT_WUNLOCK(tobject); tobject = backing_object; if (!vm_object_advice_applies(tobject, advice)) goto next_pindex; } while ((tm = vm_page_lookup(tobject, tpindex)) == NULL); } else { next_page: tm = m; m = TAILQ_NEXT(m, listq); } /* * If the page is not in a normal state, skip it. */ if (tm->valid != VM_PAGE_BITS_ALL) goto next_pindex; vm_page_lock(tm); if (vm_page_wired(tm)) { vm_page_unlock(tm); goto next_pindex; } KASSERT((tm->flags & PG_FICTITIOUS) == 0, ("vm_object_madvise: page %p is fictitious", tm)); KASSERT((tm->oflags & VPO_UNMANAGED) == 0, ("vm_object_madvise: page %p is not managed", tm)); if (vm_page_busied(tm)) { if (object != tobject) VM_OBJECT_WUNLOCK(tobject); VM_OBJECT_WUNLOCK(object); if (advice == MADV_WILLNEED) { /* * Reference the page before unlocking and * sleeping so that the page daemon is less * likely to reclaim it. */ vm_page_aflag_set(tm, PGA_REFERENCED); } vm_page_busy_sleep(tm, "madvpo", false); goto relookup; } vm_page_advise(tm, advice); vm_page_unlock(tm); vm_object_madvise_freespace(tobject, advice, tm->pindex, 1); next_pindex: if (tobject != object) VM_OBJECT_WUNLOCK(tobject); } VM_OBJECT_WUNLOCK(object); } /* * vm_object_shadow: * * Create a new object which is backed by the * specified existing object range. The source * object reference is deallocated. * * The new object and offset into that object * are returned in the source parameters. */ void vm_object_shadow( vm_object_t *object, /* IN/OUT */ vm_ooffset_t *offset, /* IN/OUT */ vm_size_t length) { vm_object_t source; vm_object_t result; source = *object; /* * Don't create the new object if the old object isn't shared. */ if (source != NULL) { VM_OBJECT_WLOCK(source); if (source->ref_count == 1 && source->handle == NULL && (source->type == OBJT_DEFAULT || source->type == OBJT_SWAP)) { VM_OBJECT_WUNLOCK(source); return; } VM_OBJECT_WUNLOCK(source); } /* * Allocate a new object with the given length. */ result = vm_object_allocate(OBJT_DEFAULT, atop(length)); /* * The new object shadows the source object, adding a reference to it. * Our caller changes his reference to point to the new object, * removing a reference to the source object. Net result: no change * of reference count. * * Try to optimize the result object's page color when shadowing * in order to maintain page coloring consistency in the combined * shadowed object. */ result->backing_object = source; /* * Store the offset into the source object, and fix up the offset into * the new object. */ result->backing_object_offset = *offset; if (source != NULL) { VM_OBJECT_WLOCK(source); result->domain = source->domain; LIST_INSERT_HEAD(&source->shadow_head, result, shadow_list); source->shadow_count++; #if VM_NRESERVLEVEL > 0 result->flags |= source->flags & OBJ_COLORED; result->pg_color = (source->pg_color + OFF_TO_IDX(*offset)) & ((1 << (VM_NFREEORDER - 1)) - 1); #endif VM_OBJECT_WUNLOCK(source); } /* * Return the new things */ *offset = 0; *object = result; } /* * vm_object_split: * * Split the pages in a map entry into a new object. This affords * easier removal of unused pages, and keeps object inheritance from * being a negative impact on memory usage. */ void vm_object_split(vm_map_entry_t entry) { vm_page_t m, m_next; vm_object_t orig_object, new_object, source; vm_pindex_t idx, offidxstart; vm_size_t size; orig_object = entry->object.vm_object; if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP) return; if (orig_object->ref_count <= 1) return; VM_OBJECT_WUNLOCK(orig_object); offidxstart = OFF_TO_IDX(entry->offset); size = atop(entry->end - entry->start); /* * If swap_pager_copy() is later called, it will convert new_object * into a swap object. */ new_object = vm_object_allocate(OBJT_DEFAULT, size); /* * At this point, the new object is still private, so the order in * which the original and new objects are locked does not matter. */ VM_OBJECT_WLOCK(new_object); VM_OBJECT_WLOCK(orig_object); new_object->domain = orig_object->domain; source = orig_object->backing_object; if (source != NULL) { VM_OBJECT_WLOCK(source); if ((source->flags & OBJ_DEAD) != 0) { VM_OBJECT_WUNLOCK(source); VM_OBJECT_WUNLOCK(orig_object); VM_OBJECT_WUNLOCK(new_object); vm_object_deallocate(new_object); VM_OBJECT_WLOCK(orig_object); return; } LIST_INSERT_HEAD(&source->shadow_head, new_object, shadow_list); source->shadow_count++; vm_object_reference_locked(source); /* for new_object */ vm_object_clear_flag(source, OBJ_ONEMAPPING); VM_OBJECT_WUNLOCK(source); new_object->backing_object_offset = orig_object->backing_object_offset + entry->offset; new_object->backing_object = source; } if (orig_object->cred != NULL) { new_object->cred = orig_object->cred; crhold(orig_object->cred); new_object->charge = ptoa(size); KASSERT(orig_object->charge >= ptoa(size), ("orig_object->charge < 0")); orig_object->charge -= ptoa(size); } retry: m = vm_page_find_least(orig_object, offidxstart); for (; m != NULL && (idx = m->pindex - offidxstart) < size; m = m_next) { m_next = TAILQ_NEXT(m, listq); /* * We must wait for pending I/O to complete before we can * rename the page. * * We do not have to VM_PROT_NONE the page as mappings should * not be changed by this operation. */ if (vm_page_busied(m)) { VM_OBJECT_WUNLOCK(new_object); vm_page_lock(m); VM_OBJECT_WUNLOCK(orig_object); vm_page_busy_sleep(m, "spltwt", false); VM_OBJECT_WLOCK(orig_object); VM_OBJECT_WLOCK(new_object); goto retry; } /* vm_page_rename() will dirty the page. */ if (vm_page_rename(m, new_object, idx)) { VM_OBJECT_WUNLOCK(new_object); VM_OBJECT_WUNLOCK(orig_object); vm_radix_wait(); VM_OBJECT_WLOCK(orig_object); VM_OBJECT_WLOCK(new_object); goto retry; } #if VM_NRESERVLEVEL > 0 /* * If some of the reservation's allocated pages remain with * the original object, then transferring the reservation to * the new object is neither particularly beneficial nor * particularly harmful as compared to leaving the reservation * with the original object. If, however, all of the * reservation's allocated pages are transferred to the new * object, then transferring the reservation is typically * beneficial. Determining which of these two cases applies * would be more costly than unconditionally renaming the * reservation. */ vm_reserv_rename(m, new_object, orig_object, offidxstart); #endif if (orig_object->type == OBJT_SWAP) vm_page_xbusy(m); } if (orig_object->type == OBJT_SWAP) { /* * swap_pager_copy() can sleep, in which case the orig_object's * and new_object's locks are released and reacquired. */ swap_pager_copy(orig_object, new_object, offidxstart, 0); TAILQ_FOREACH(m, &new_object->memq, listq) vm_page_xunbusy(m); } VM_OBJECT_WUNLOCK(orig_object); VM_OBJECT_WUNLOCK(new_object); entry->object.vm_object = new_object; entry->offset = 0LL; vm_object_deallocate(orig_object); VM_OBJECT_WLOCK(new_object); } #define OBSC_COLLAPSE_NOWAIT 0x0002 #define OBSC_COLLAPSE_WAIT 0x0004 static vm_page_t vm_object_collapse_scan_wait(vm_object_t object, vm_page_t p, vm_page_t next, int op) { vm_object_t backing_object; VM_OBJECT_ASSERT_WLOCKED(object); backing_object = object->backing_object; VM_OBJECT_ASSERT_WLOCKED(backing_object); KASSERT(p == NULL || vm_page_busied(p), ("unbusy page %p", p)); KASSERT(p == NULL || p->object == object || p->object == backing_object, ("invalid ownership %p %p %p", p, object, backing_object)); if ((op & OBSC_COLLAPSE_NOWAIT) != 0) return (next); if (p != NULL) vm_page_lock(p); VM_OBJECT_WUNLOCK(object); VM_OBJECT_WUNLOCK(backing_object); /* The page is only NULL when rename fails. */ if (p == NULL) vm_radix_wait(); else vm_page_busy_sleep(p, "vmocol", false); VM_OBJECT_WLOCK(object); VM_OBJECT_WLOCK(backing_object); return (TAILQ_FIRST(&backing_object->memq)); } static bool vm_object_scan_all_shadowed(vm_object_t object) { vm_object_t backing_object; vm_page_t p, pp; vm_pindex_t backing_offset_index, new_pindex, pi, ps; VM_OBJECT_ASSERT_WLOCKED(object); VM_OBJECT_ASSERT_WLOCKED(object->backing_object); backing_object = object->backing_object; if (backing_object->type != OBJT_DEFAULT && backing_object->type != OBJT_SWAP) return (false); pi = backing_offset_index = OFF_TO_IDX(object->backing_object_offset); p = vm_page_find_least(backing_object, pi); ps = swap_pager_find_least(backing_object, pi); /* * Only check pages inside the parent object's range and * inside the parent object's mapping of the backing object. */ for (;; pi++) { if (p != NULL && p->pindex < pi) p = TAILQ_NEXT(p, listq); if (ps < pi) ps = swap_pager_find_least(backing_object, pi); if (p == NULL && ps >= backing_object->size) break; else if (p == NULL) pi = ps; else pi = MIN(p->pindex, ps); new_pindex = pi - backing_offset_index; if (new_pindex >= object->size) break; /* * See if the parent has the page or if the parent's object * pager has the page. If the parent has the page but the page * is not valid, the parent's object pager must have the page. * * If this fails, the parent does not completely shadow the * object and we might as well give up now. */ pp = vm_page_lookup(object, new_pindex); if ((pp == NULL || pp->valid == 0) && !vm_pager_has_page(object, new_pindex, NULL, NULL)) return (false); } return (true); } static bool vm_object_collapse_scan(vm_object_t object, int op) { vm_object_t backing_object; vm_page_t next, p, pp; vm_pindex_t backing_offset_index, new_pindex; VM_OBJECT_ASSERT_WLOCKED(object); VM_OBJECT_ASSERT_WLOCKED(object->backing_object); backing_object = object->backing_object; backing_offset_index = OFF_TO_IDX(object->backing_object_offset); /* * Initial conditions */ if ((op & OBSC_COLLAPSE_WAIT) != 0) vm_object_set_flag(backing_object, OBJ_DEAD); /* * Our scan */ for (p = TAILQ_FIRST(&backing_object->memq); p != NULL; p = next) { next = TAILQ_NEXT(p, listq); new_pindex = p->pindex - backing_offset_index; /* * Check for busy page */ if (vm_page_busied(p)) { next = vm_object_collapse_scan_wait(object, p, next, op); continue; } KASSERT(p->object == backing_object, ("vm_object_collapse_scan: object mismatch")); if (p->pindex < backing_offset_index || new_pindex >= object->size) { if (backing_object->type == OBJT_SWAP) swap_pager_freespace(backing_object, p->pindex, 1); /* * Page is out of the parent object's range, we can * simply destroy it. */ vm_page_lock(p); KASSERT(!pmap_page_is_mapped(p), ("freeing mapped page %p", p)); if (vm_page_remove(p)) vm_page_free(p); vm_page_unlock(p); continue; } pp = vm_page_lookup(object, new_pindex); if (pp != NULL && vm_page_busied(pp)) { /* * The page in the parent is busy and possibly not * (yet) valid. Until its state is finalized by the * busy bit owner, we can't tell whether it shadows the * original page. Therefore, we must either skip it * and the original (backing_object) page or wait for * its state to be finalized. * * This is due to a race with vm_fault() where we must * unbusy the original (backing_obj) page before we can * (re)lock the parent. Hence we can get here. */ next = vm_object_collapse_scan_wait(object, pp, next, op); continue; } KASSERT(pp == NULL || pp->valid != 0, ("unbusy invalid page %p", pp)); if (pp != NULL || vm_pager_has_page(object, new_pindex, NULL, NULL)) { /* * The page already exists in the parent OR swap exists * for this location in the parent. Leave the parent's * page alone. Destroy the original page from the * backing object. */ if (backing_object->type == OBJT_SWAP) swap_pager_freespace(backing_object, p->pindex, 1); vm_page_lock(p); KASSERT(!pmap_page_is_mapped(p), ("freeing mapped page %p", p)); if (vm_page_remove(p)) vm_page_free(p); vm_page_unlock(p); continue; } /* * Page does not exist in parent, rename the page from the * backing object to the main object. * * If the page was mapped to a process, it can remain mapped * through the rename. vm_page_rename() will dirty the page. */ if (vm_page_rename(p, object, new_pindex)) { next = vm_object_collapse_scan_wait(object, NULL, next, op); continue; } /* Use the old pindex to free the right page. */ if (backing_object->type == OBJT_SWAP) swap_pager_freespace(backing_object, new_pindex + backing_offset_index, 1); #if VM_NRESERVLEVEL > 0 /* * Rename the reservation. */ vm_reserv_rename(p, object, backing_object, backing_offset_index); #endif } return (true); } /* * this version of collapse allows the operation to occur earlier and * when paging_in_progress is true for an object... This is not a complete * operation, but should plug 99.9% of the rest of the leaks. */ static void vm_object_qcollapse(vm_object_t object) { vm_object_t backing_object = object->backing_object; VM_OBJECT_ASSERT_WLOCKED(object); VM_OBJECT_ASSERT_WLOCKED(backing_object); if (backing_object->ref_count != 1) return; vm_object_collapse_scan(object, OBSC_COLLAPSE_NOWAIT); } /* * vm_object_collapse: * * Collapse an object with the object backing it. * Pages in the backing object are moved into the * parent, and the backing object is deallocated. */ void vm_object_collapse(vm_object_t object) { vm_object_t backing_object, new_backing_object; VM_OBJECT_ASSERT_WLOCKED(object); while (TRUE) { /* * Verify that the conditions are right for collapse: * * The object exists and the backing object exists. */ if ((backing_object = object->backing_object) == NULL) break; /* * we check the backing object first, because it is most likely * not collapsable. */ VM_OBJECT_WLOCK(backing_object); if (backing_object->handle != NULL || (backing_object->type != OBJT_DEFAULT && backing_object->type != OBJT_SWAP) || (backing_object->flags & (OBJ_DEAD | OBJ_NOSPLIT)) != 0 || object->handle != NULL || (object->type != OBJT_DEFAULT && object->type != OBJT_SWAP) || (object->flags & OBJ_DEAD)) { VM_OBJECT_WUNLOCK(backing_object); break; } if (object->paging_in_progress != 0 || backing_object->paging_in_progress != 0) { vm_object_qcollapse(object); VM_OBJECT_WUNLOCK(backing_object); break; } /* * We know that we can either collapse the backing object (if * the parent is the only reference to it) or (perhaps) have * the parent bypass the object if the parent happens to shadow * all the resident pages in the entire backing object. * * This is ignoring pager-backed pages such as swap pages. * vm_object_collapse_scan fails the shadowing test in this * case. */ if (backing_object->ref_count == 1) { vm_object_pip_add(object, 1); vm_object_pip_add(backing_object, 1); /* * If there is exactly one reference to the backing * object, we can collapse it into the parent. */ vm_object_collapse_scan(object, OBSC_COLLAPSE_WAIT); #if VM_NRESERVLEVEL > 0 /* * Break any reservations from backing_object. */ if (__predict_false(!LIST_EMPTY(&backing_object->rvq))) vm_reserv_break_all(backing_object); #endif /* * Move the pager from backing_object to object. */ if (backing_object->type == OBJT_SWAP) { /* * swap_pager_copy() can sleep, in which case * the backing_object's and object's locks are * released and reacquired. * Since swap_pager_copy() is being asked to * destroy the source, it will change the * backing_object's type to OBJT_DEFAULT. */ swap_pager_copy( backing_object, object, OFF_TO_IDX(object->backing_object_offset), TRUE); } /* * Object now shadows whatever backing_object did. * Note that the reference to * backing_object->backing_object moves from within * backing_object to within object. */ LIST_REMOVE(object, shadow_list); backing_object->shadow_count--; if (backing_object->backing_object) { VM_OBJECT_WLOCK(backing_object->backing_object); LIST_REMOVE(backing_object, shadow_list); LIST_INSERT_HEAD( &backing_object->backing_object->shadow_head, object, shadow_list); /* * The shadow_count has not changed. */ VM_OBJECT_WUNLOCK(backing_object->backing_object); } object->backing_object = backing_object->backing_object; object->backing_object_offset += backing_object->backing_object_offset; /* * Discard backing_object. * * Since the backing object has no pages, no pager left, * and no object references within it, all that is * necessary is to dispose of it. */ KASSERT(backing_object->ref_count == 1, ( "backing_object %p was somehow re-referenced during collapse!", backing_object)); vm_object_pip_wakeup(backing_object); backing_object->type = OBJT_DEAD; backing_object->ref_count = 0; VM_OBJECT_WUNLOCK(backing_object); vm_object_destroy(backing_object); vm_object_pip_wakeup(object); counter_u64_add(object_collapses, 1); } else { /* * If we do not entirely shadow the backing object, * there is nothing we can do so we give up. */ if (object->resident_page_count != object->size && !vm_object_scan_all_shadowed(object)) { VM_OBJECT_WUNLOCK(backing_object); break; } /* * Make the parent shadow the next object in the * chain. Deallocating backing_object will not remove * it, since its reference count is at least 2. */ LIST_REMOVE(object, shadow_list); backing_object->shadow_count--; new_backing_object = backing_object->backing_object; if ((object->backing_object = new_backing_object) != NULL) { VM_OBJECT_WLOCK(new_backing_object); LIST_INSERT_HEAD( &new_backing_object->shadow_head, object, shadow_list ); new_backing_object->shadow_count++; vm_object_reference_locked(new_backing_object); VM_OBJECT_WUNLOCK(new_backing_object); object->backing_object_offset += backing_object->backing_object_offset; } /* * Drop the reference count on backing_object. Since * its ref_count was at least 2, it will not vanish. */ backing_object->ref_count--; VM_OBJECT_WUNLOCK(backing_object); counter_u64_add(object_bypasses, 1); } /* * Try again with this object's new backing object. */ } } /* * vm_object_page_remove: * * For the given object, either frees or invalidates each of the * specified pages. In general, a page is freed. However, if a page is * wired for any reason other than the existence of a managed, wired * mapping, then it may be invalidated but not removed from the object. * Pages are specified by the given range ["start", "end") and the option * OBJPR_CLEANONLY. As a special case, if "end" is zero, then the range * extends from "start" to the end of the object. If the option * OBJPR_CLEANONLY is specified, then only the non-dirty pages within the * specified range are affected. If the option OBJPR_NOTMAPPED is * specified, then the pages within the specified range must have no * mappings. Otherwise, if this option is not specified, any mappings to * the specified pages are removed before the pages are freed or * invalidated. * * In general, this operation should only be performed on objects that * contain managed pages. There are, however, two exceptions. First, it * is performed on the kernel and kmem objects by vm_map_entry_delete(). * Second, it is used by msync(..., MS_INVALIDATE) to invalidate device- * backed pages. In both of these cases, the option OBJPR_CLEANONLY must * not be specified and the option OBJPR_NOTMAPPED must be specified. * * The object must be locked. */ void vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end, int options) { vm_page_t p, next; struct mtx *mtx; VM_OBJECT_ASSERT_WLOCKED(object); KASSERT((object->flags & OBJ_UNMANAGED) == 0 || (options & (OBJPR_CLEANONLY | OBJPR_NOTMAPPED)) == OBJPR_NOTMAPPED, ("vm_object_page_remove: illegal options for object %p", object)); if (object->resident_page_count == 0) return; vm_object_pip_add(object, 1); again: p = vm_page_find_least(object, start); mtx = NULL; /* * Here, the variable "p" is either (1) the page with the least pindex * greater than or equal to the parameter "start" or (2) NULL. */ for (; p != NULL && (p->pindex < end || end == 0); p = next) { next = TAILQ_NEXT(p, listq); /* * If the page is wired for any reason besides the existence * of managed, wired mappings, then it cannot be freed. For * example, fictitious pages, which represent device memory, * are inherently wired and cannot be freed. They can, * however, be invalidated if the option OBJPR_CLEANONLY is * not specified. */ vm_page_change_lock(p, &mtx); if (vm_page_xbusied(p)) { VM_OBJECT_WUNLOCK(object); vm_page_busy_sleep(p, "vmopax", true); VM_OBJECT_WLOCK(object); goto again; } if (vm_page_wired(p)) { if ((options & OBJPR_NOTMAPPED) == 0 && object->ref_count != 0) pmap_remove_all(p); if ((options & OBJPR_CLEANONLY) == 0) { p->valid = 0; vm_page_undirty(p); } continue; } if (vm_page_busied(p)) { VM_OBJECT_WUNLOCK(object); vm_page_busy_sleep(p, "vmopar", false); VM_OBJECT_WLOCK(object); goto again; } KASSERT((p->flags & PG_FICTITIOUS) == 0, ("vm_object_page_remove: page %p is fictitious", p)); if ((options & OBJPR_CLEANONLY) != 0 && p->valid != 0) { if ((options & OBJPR_NOTMAPPED) == 0 && object->ref_count != 0) pmap_remove_write(p); if (p->dirty != 0) continue; } if ((options & OBJPR_NOTMAPPED) == 0 && object->ref_count != 0) pmap_remove_all(p); vm_page_free(p); } if (mtx != NULL) mtx_unlock(mtx); vm_object_pip_wakeup(object); } /* * vm_object_page_noreuse: * * For the given object, attempt to move the specified pages to * the head of the inactive queue. This bypasses regular LRU * operation and allows the pages to be reused quickly under memory * pressure. If a page is wired for any reason, then it will not * be queued. Pages are specified by the range ["start", "end"). * As a special case, if "end" is zero, then the range extends from * "start" to the end of the object. * * This operation should only be performed on objects that * contain non-fictitious, managed pages. * * The object must be locked. */ void vm_object_page_noreuse(vm_object_t object, vm_pindex_t start, vm_pindex_t end) { struct mtx *mtx; vm_page_t p, next; VM_OBJECT_ASSERT_LOCKED(object); KASSERT((object->flags & (OBJ_FICTITIOUS | OBJ_UNMANAGED)) == 0, ("vm_object_page_noreuse: illegal object %p", object)); if (object->resident_page_count == 0) return; p = vm_page_find_least(object, start); /* * Here, the variable "p" is either (1) the page with the least pindex * greater than or equal to the parameter "start" or (2) NULL. */ mtx = NULL; for (; p != NULL && (p->pindex < end || end == 0); p = next) { next = TAILQ_NEXT(p, listq); vm_page_change_lock(p, &mtx); vm_page_deactivate_noreuse(p); } if (mtx != NULL) mtx_unlock(mtx); } /* * Populate the specified range of the object with valid pages. Returns * TRUE if the range is successfully populated and FALSE otherwise. * * Note: This function should be optimized to pass a larger array of * pages to vm_pager_get_pages() before it is applied to a non- * OBJT_DEVICE object. * * The object must be locked. */ boolean_t vm_object_populate(vm_object_t object, vm_pindex_t start, vm_pindex_t end) { vm_page_t m; vm_pindex_t pindex; int rv; VM_OBJECT_ASSERT_WLOCKED(object); for (pindex = start; pindex < end; pindex++) { m = vm_page_grab(object, pindex, VM_ALLOC_NORMAL); if (m->valid != VM_PAGE_BITS_ALL) { rv = vm_pager_get_pages(object, &m, 1, NULL, NULL); if (rv != VM_PAGER_OK) { vm_page_lock(m); vm_page_free(m); vm_page_unlock(m); break; } } /* * Keep "m" busy because a subsequent iteration may unlock * the object. */ } if (pindex > start) { m = vm_page_lookup(object, start); while (m != NULL && m->pindex < pindex) { vm_page_xunbusy(m); m = TAILQ_NEXT(m, listq); } } return (pindex == end); } /* * Routine: vm_object_coalesce * Function: Coalesces two objects backing up adjoining * regions of memory into a single object. * * returns TRUE if objects were combined. * * NOTE: Only works at the moment if the second object is NULL - * if it's not, which object do we lock first? * * Parameters: * prev_object First object to coalesce * prev_offset Offset into prev_object * prev_size Size of reference to prev_object * next_size Size of reference to the second object * reserved Indicator that extension region has * swap accounted for * * Conditions: * The object must *not* be locked. */ boolean_t vm_object_coalesce(vm_object_t prev_object, vm_ooffset_t prev_offset, vm_size_t prev_size, vm_size_t next_size, boolean_t reserved) { vm_pindex_t next_pindex; if (prev_object == NULL) return (TRUE); VM_OBJECT_WLOCK(prev_object); if ((prev_object->type != OBJT_DEFAULT && prev_object->type != OBJT_SWAP) || (prev_object->flags & OBJ_TMPFS_NODE) != 0) { VM_OBJECT_WUNLOCK(prev_object); return (FALSE); } /* * Try to collapse the object first */ vm_object_collapse(prev_object); /* * Can't coalesce if: . more than one reference . paged out . shadows * another object . has a copy elsewhere (any of which mean that the * pages not mapped to prev_entry may be in use anyway) */ if (prev_object->backing_object != NULL) { VM_OBJECT_WUNLOCK(prev_object); return (FALSE); } prev_size >>= PAGE_SHIFT; next_size >>= PAGE_SHIFT; next_pindex = OFF_TO_IDX(prev_offset) + prev_size; if (prev_object->ref_count > 1 && prev_object->size != next_pindex && (prev_object->flags & OBJ_ONEMAPPING) == 0) { VM_OBJECT_WUNLOCK(prev_object); return (FALSE); } /* * Account for the charge. */ if (prev_object->cred != NULL) { /* * If prev_object was charged, then this mapping, * although not charged now, may become writable * later. Non-NULL cred in the object would prevent * swap reservation during enabling of the write * access, so reserve swap now. Failed reservation * cause allocation of the separate object for the map * entry, and swap reservation for this entry is * managed in appropriate time. */ if (!reserved && !swap_reserve_by_cred(ptoa(next_size), prev_object->cred)) { VM_OBJECT_WUNLOCK(prev_object); return (FALSE); } prev_object->charge += ptoa(next_size); } /* * Remove any pages that may still be in the object from a previous * deallocation. */ if (next_pindex < prev_object->size) { vm_object_page_remove(prev_object, next_pindex, next_pindex + next_size, 0); if (prev_object->type == OBJT_SWAP) swap_pager_freespace(prev_object, next_pindex, next_size); #if 0 if (prev_object->cred != NULL) { KASSERT(prev_object->charge >= ptoa(prev_object->size - next_pindex), ("object %p overcharged 1 %jx %jx", prev_object, (uintmax_t)next_pindex, (uintmax_t)next_size)); prev_object->charge -= ptoa(prev_object->size - next_pindex); } #endif } /* * Extend the object if necessary. */ if (next_pindex + next_size > prev_object->size) prev_object->size = next_pindex + next_size; VM_OBJECT_WUNLOCK(prev_object); return (TRUE); } void vm_object_set_writeable_dirty(vm_object_t object) { VM_OBJECT_ASSERT_WLOCKED(object); if (object->type != OBJT_VNODE) { if ((object->flags & OBJ_TMPFS_NODE) != 0) { KASSERT(object->type == OBJT_SWAP, ("non-swap tmpfs")); vm_object_set_flag(object, OBJ_TMPFS_DIRTY); } return; } object->generation++; if ((object->flags & OBJ_MIGHTBEDIRTY) != 0) return; vm_object_set_flag(object, OBJ_MIGHTBEDIRTY); } /* * vm_object_unwire: * * For each page offset within the specified range of the given object, * find the highest-level page in the shadow chain and unwire it. A page * must exist at every page offset, and the highest-level page must be * wired. */ void vm_object_unwire(vm_object_t object, vm_ooffset_t offset, vm_size_t length, uint8_t queue) { vm_object_t tobject, t1object; vm_page_t m, tm; vm_pindex_t end_pindex, pindex, tpindex; int depth, locked_depth; KASSERT((offset & PAGE_MASK) == 0, ("vm_object_unwire: offset is not page aligned")); KASSERT((length & PAGE_MASK) == 0, ("vm_object_unwire: length is not a multiple of PAGE_SIZE")); /* The wired count of a fictitious page never changes. */ if ((object->flags & OBJ_FICTITIOUS) != 0) return; pindex = OFF_TO_IDX(offset); end_pindex = pindex + atop(length); again: locked_depth = 1; VM_OBJECT_RLOCK(object); m = vm_page_find_least(object, pindex); while (pindex < end_pindex) { if (m == NULL || pindex < m->pindex) { /* * The first object in the shadow chain doesn't * contain a page at the current index. Therefore, * the page must exist in a backing object. */ tobject = object; tpindex = pindex; depth = 0; do { tpindex += OFF_TO_IDX(tobject->backing_object_offset); tobject = tobject->backing_object; KASSERT(tobject != NULL, ("vm_object_unwire: missing page")); if ((tobject->flags & OBJ_FICTITIOUS) != 0) goto next_page; depth++; if (depth == locked_depth) { locked_depth++; VM_OBJECT_RLOCK(tobject); } } while ((tm = vm_page_lookup(tobject, tpindex)) == NULL); } else { tm = m; m = TAILQ_NEXT(m, listq); } vm_page_lock(tm); if (vm_page_xbusied(tm)) { for (tobject = object; locked_depth >= 1; locked_depth--) { t1object = tobject->backing_object; VM_OBJECT_RUNLOCK(tobject); tobject = t1object; } vm_page_busy_sleep(tm, "unwbo", true); goto again; } vm_page_unwire(tm, queue); vm_page_unlock(tm); next_page: pindex++; } /* Release the accumulated object locks. */ for (tobject = object; locked_depth >= 1; locked_depth--) { t1object = tobject->backing_object; VM_OBJECT_RUNLOCK(tobject); tobject = t1object; } } /* * Return the vnode for the given object, or NULL if none exists. * For tmpfs objects, the function may return NULL if there is * no vnode allocated at the time of the call. */ struct vnode * vm_object_vnode(vm_object_t object) { struct vnode *vp; VM_OBJECT_ASSERT_LOCKED(object); if (object->type == OBJT_VNODE) { vp = object->handle; KASSERT(vp != NULL, ("%s: OBJT_VNODE has no vnode", __func__)); } else if (object->type == OBJT_SWAP && (object->flags & OBJ_TMPFS) != 0) { vp = object->un_pager.swp.swp_tmpfs; KASSERT(vp != NULL, ("%s: OBJT_TMPFS has no vnode", __func__)); } else { vp = NULL; } return (vp); } /* * Return the kvme type of the given object. * If vpp is not NULL, set it to the object's vm_object_vnode() or NULL. */ int vm_object_kvme_type(vm_object_t object, struct vnode **vpp) { VM_OBJECT_ASSERT_LOCKED(object); if (vpp != NULL) *vpp = vm_object_vnode(object); switch (object->type) { case OBJT_DEFAULT: return (KVME_TYPE_DEFAULT); case OBJT_VNODE: return (KVME_TYPE_VNODE); case OBJT_SWAP: if ((object->flags & OBJ_TMPFS_NODE) != 0) return (KVME_TYPE_VNODE); return (KVME_TYPE_SWAP); case OBJT_DEVICE: return (KVME_TYPE_DEVICE); case OBJT_PHYS: return (KVME_TYPE_PHYS); case OBJT_DEAD: return (KVME_TYPE_DEAD); case OBJT_SG: return (KVME_TYPE_SG); case OBJT_MGTDEVICE: return (KVME_TYPE_MGTDEVICE); default: return (KVME_TYPE_UNKNOWN); } } static int sysctl_vm_object_list(SYSCTL_HANDLER_ARGS) { struct kinfo_vmobject *kvo; char *fullpath, *freepath; struct vnode *vp; struct vattr va; vm_object_t obj; vm_page_t m; int count, error; if (req->oldptr == NULL) { /* * If an old buffer has not been provided, generate an * estimate of the space needed for a subsequent call. */ mtx_lock(&vm_object_list_mtx); count = 0; TAILQ_FOREACH(obj, &vm_object_list, object_list) { if (obj->type == OBJT_DEAD) continue; count++; } mtx_unlock(&vm_object_list_mtx); return (SYSCTL_OUT(req, NULL, sizeof(struct kinfo_vmobject) * count * 11 / 10)); } kvo = malloc(sizeof(*kvo), M_TEMP, M_WAITOK); error = 0; /* * VM objects are type stable and are never removed from the * list once added. This allows us to safely read obj->object_list * after reacquiring the VM object lock. */ mtx_lock(&vm_object_list_mtx); TAILQ_FOREACH(obj, &vm_object_list, object_list) { if (obj->type == OBJT_DEAD) continue; VM_OBJECT_RLOCK(obj); if (obj->type == OBJT_DEAD) { VM_OBJECT_RUNLOCK(obj); continue; } mtx_unlock(&vm_object_list_mtx); kvo->kvo_size = ptoa(obj->size); kvo->kvo_resident = obj->resident_page_count; kvo->kvo_ref_count = obj->ref_count; kvo->kvo_shadow_count = obj->shadow_count; kvo->kvo_memattr = obj->memattr; kvo->kvo_active = 0; kvo->kvo_inactive = 0; TAILQ_FOREACH(m, &obj->memq, listq) { /* * A page may belong to the object but be * dequeued and set to PQ_NONE while the * object lock is not held. This makes the * reads of m->queue below racy, and we do not * count pages set to PQ_NONE. However, this * sysctl is only meant to give an * approximation of the system anyway. */ if (m->queue == PQ_ACTIVE) kvo->kvo_active++; else if (m->queue == PQ_INACTIVE) kvo->kvo_inactive++; } kvo->kvo_vn_fileid = 0; kvo->kvo_vn_fsid = 0; kvo->kvo_vn_fsid_freebsd11 = 0; freepath = NULL; fullpath = ""; kvo->kvo_type = vm_object_kvme_type(obj, &vp); if (vp != NULL) vref(vp); VM_OBJECT_RUNLOCK(obj); if (vp != NULL) { vn_fullpath(curthread, vp, &fullpath, &freepath); vn_lock(vp, LK_SHARED | LK_RETRY); if (VOP_GETATTR(vp, &va, curthread->td_ucred) == 0) { kvo->kvo_vn_fileid = va.va_fileid; kvo->kvo_vn_fsid = va.va_fsid; kvo->kvo_vn_fsid_freebsd11 = va.va_fsid; /* truncate */ } vput(vp); } strlcpy(kvo->kvo_path, fullpath, sizeof(kvo->kvo_path)); if (freepath != NULL) free(freepath, M_TEMP); /* Pack record size down */ kvo->kvo_structsize = offsetof(struct kinfo_vmobject, kvo_path) + strlen(kvo->kvo_path) + 1; kvo->kvo_structsize = roundup(kvo->kvo_structsize, sizeof(uint64_t)); error = SYSCTL_OUT(req, kvo, kvo->kvo_structsize); mtx_lock(&vm_object_list_mtx); if (error) break; } mtx_unlock(&vm_object_list_mtx); free(kvo, M_TEMP); return (error); } SYSCTL_PROC(_vm, OID_AUTO, objects, CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_object_list, "S,kinfo_vmobject", "List of VM objects"); #include "opt_ddb.h" #ifdef DDB #include #include #include static int _vm_object_in_map(vm_map_t map, vm_object_t object, vm_map_entry_t entry) { vm_map_t tmpm; vm_map_entry_t tmpe; vm_object_t obj; int entcount; if (map == 0) return 0; if (entry == 0) { tmpe = map->header.next; entcount = map->nentries; while (entcount-- && (tmpe != &map->header)) { if (_vm_object_in_map(map, object, tmpe)) { return 1; } tmpe = tmpe->next; } } else if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { tmpm = entry->object.sub_map; tmpe = tmpm->header.next; entcount = tmpm->nentries; while (entcount-- && tmpe != &tmpm->header) { if (_vm_object_in_map(tmpm, object, tmpe)) { return 1; } tmpe = tmpe->next; } } else if ((obj = entry->object.vm_object) != NULL) { for (; obj; obj = obj->backing_object) if (obj == object) { return 1; } } return 0; } static int vm_object_in_map(vm_object_t object) { struct proc *p; /* sx_slock(&allproc_lock); */ FOREACH_PROC_IN_SYSTEM(p) { if (!p->p_vmspace /* || (p->p_flag & (P_SYSTEM|P_WEXIT)) */) continue; if (_vm_object_in_map(&p->p_vmspace->vm_map, object, 0)) { /* sx_sunlock(&allproc_lock); */ return 1; } } /* sx_sunlock(&allproc_lock); */ if (_vm_object_in_map(kernel_map, object, 0)) return 1; return 0; } DB_SHOW_COMMAND(vmochk, vm_object_check) { vm_object_t object; /* * make sure that internal objs are in a map somewhere * and none have zero ref counts. */ TAILQ_FOREACH(object, &vm_object_list, object_list) { if (object->handle == NULL && (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) { if (object->ref_count == 0) { db_printf("vmochk: internal obj has zero ref count: %ld\n", (long)object->size); } if (!vm_object_in_map(object)) { db_printf( "vmochk: internal obj is not in a map: " "ref: %d, size: %lu: 0x%lx, backing_object: %p\n", object->ref_count, (u_long)object->size, (u_long)object->size, (void *)object->backing_object); } } } } /* * vm_object_print: [ debug ] */ DB_SHOW_COMMAND(object, vm_object_print_static) { /* XXX convert args. */ vm_object_t object = (vm_object_t)addr; boolean_t full = have_addr; vm_page_t p; /* XXX count is an (unused) arg. Avoid shadowing it. */ #define count was_count int count; if (object == NULL) return; db_iprintf( "Object %p: type=%d, size=0x%jx, res=%d, ref=%d, flags=0x%x ruid %d charge %jx\n", object, (int)object->type, (uintmax_t)object->size, object->resident_page_count, object->ref_count, object->flags, object->cred ? object->cred->cr_ruid : -1, (uintmax_t)object->charge); db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%jx\n", object->shadow_count, object->backing_object ? object->backing_object->ref_count : 0, object->backing_object, (uintmax_t)object->backing_object_offset); if (!full) return; db_indent += 2; count = 0; TAILQ_FOREACH(p, &object->memq, listq) { if (count == 0) db_iprintf("memory:="); else if (count == 6) { db_printf("\n"); db_iprintf(" ..."); count = 0; } else db_printf(","); count++; db_printf("(off=0x%jx,page=0x%jx)", (uintmax_t)p->pindex, (uintmax_t)VM_PAGE_TO_PHYS(p)); } if (count != 0) db_printf("\n"); db_indent -= 2; } /* XXX. */ #undef count /* XXX need this non-static entry for calling from vm_map_print. */ void vm_object_print( /* db_expr_t */ long addr, boolean_t have_addr, /* db_expr_t */ long count, char *modif) { vm_object_print_static(addr, have_addr, count, modif); } DB_SHOW_COMMAND(vmopag, vm_object_print_pages) { vm_object_t object; vm_pindex_t fidx; vm_paddr_t pa; vm_page_t m, prev_m; int rcount, nl, c; nl = 0; TAILQ_FOREACH(object, &vm_object_list, object_list) { db_printf("new object: %p\n", (void *)object); if (nl > 18) { c = cngetc(); if (c != ' ') return; nl = 0; } nl++; rcount = 0; fidx = 0; pa = -1; TAILQ_FOREACH(m, &object->memq, listq) { if (m->pindex > 128) break; if ((prev_m = TAILQ_PREV(m, pglist, listq)) != NULL && prev_m->pindex + 1 != m->pindex) { if (rcount) { db_printf(" index(%ld)run(%d)pa(0x%lx)\n", (long)fidx, rcount, (long)pa); if (nl > 18) { c = cngetc(); if (c != ' ') return; nl = 0; } nl++; rcount = 0; } } if (rcount && (VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) { ++rcount; continue; } if (rcount) { db_printf(" index(%ld)run(%d)pa(0x%lx)\n", (long)fidx, rcount, (long)pa); if (nl > 18) { c = cngetc(); if (c != ' ') return; nl = 0; } nl++; } fidx = m->pindex; pa = VM_PAGE_TO_PHYS(m); rcount = 1; } if (rcount) { db_printf(" index(%ld)run(%d)pa(0x%lx)\n", (long)fidx, rcount, (long)pa); if (nl > 18) { c = cngetc(); if (c != ' ') return; nl = 0; } nl++; } } } #endif /* DDB */ Index: head/sys/vm/vm_object.h =================================================================== --- head/sys/vm/vm_object.h (revision 351240) +++ head/sys/vm/vm_object.h (revision 351241) @@ -1,352 +1,351 @@ /*- * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU) * * Copyright (c) 1991, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * The Mach Operating System project at Carnegie-Mellon University. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from: @(#)vm_object.h 8.3 (Berkeley) 1/12/94 * * * Copyright (c) 1987, 1990 Carnegie-Mellon University. * All rights reserved. * * Authors: Avadis Tevanian, Jr., Michael Wayne Young * * Permission to use, copy, modify and distribute this software and * its documentation is hereby granted, provided that both the copyright * notice and this permission notice appear in all copies of the * software, derivative works or modified versions, and any portions * thereof, and that both notices appear in supporting documentation. * * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. * * Carnegie Mellon requests users of this software to return to * * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU * School of Computer Science * Carnegie Mellon University * Pittsburgh PA 15213-3890 * * any improvements or extensions that they make and grant Carnegie the * rights to redistribute these changes. * * $FreeBSD$ */ /* * Virtual memory object module definitions. */ #ifndef _VM_OBJECT_ #define _VM_OBJECT_ #include #include #include #include #include #include #include /* * Types defined: * * vm_object_t Virtual memory object. * * List of locks * (c) const until freed * (o) per-object lock * (f) free pages queue mutex * */ #ifndef VM_PAGE_HAVE_PGLIST TAILQ_HEAD(pglist, vm_page); #define VM_PAGE_HAVE_PGLIST #endif struct vm_object { struct rwlock lock; TAILQ_ENTRY(vm_object) object_list; /* list of all objects */ LIST_HEAD(, vm_object) shadow_head; /* objects that this is a shadow for */ LIST_ENTRY(vm_object) shadow_list; /* chain of shadow objects */ struct pglist memq; /* list of resident pages */ struct vm_radix rtree; /* root of the resident page radix trie*/ vm_pindex_t size; /* Object size */ struct domainset_ref domain; /* NUMA policy. */ int generation; /* generation ID */ int ref_count; /* How many refs?? */ int shadow_count; /* how many objects that this is a shadow for */ vm_memattr_t memattr; /* default memory attribute for pages */ objtype_t type; /* type of pager */ u_short flags; /* see below */ u_short pg_color; /* (c) color of first page in obj */ - u_int paging_in_progress; /* Paging (in or out) so don't collapse or destroy */ + volatile u_int paging_in_progress; /* Paging (in or out) so don't collapse or destroy */ int resident_page_count; /* number of resident pages */ struct vm_object *backing_object; /* object that I'm a shadow of */ vm_ooffset_t backing_object_offset;/* Offset in backing object */ TAILQ_ENTRY(vm_object) pager_object_list; /* list of all objects of this pager type */ LIST_HEAD(, vm_reserv) rvq; /* list of reservations */ void *handle; union { /* * VNode pager * * vnp_size - current size of file */ struct { off_t vnp_size; vm_ooffset_t writemappings; } vnp; /* * Device pager * * devp_pglist - list of allocated pages */ struct { TAILQ_HEAD(, vm_page) devp_pglist; struct cdev_pager_ops *ops; struct cdev *dev; } devp; /* * SG pager * * sgp_pglist - list of allocated pages */ struct { TAILQ_HEAD(, vm_page) sgp_pglist; } sgp; /* * Swap pager * * swp_tmpfs - back-pointer to the tmpfs vnode, * if any, which uses the vm object * as backing store. The handle * cannot be reused for linking, * because the vnode can be * reclaimed and recreated, making * the handle changed and hash-chain * invalid. * * swp_blks - pc-trie of the allocated swap blocks. * */ struct { void *swp_tmpfs; struct pctrie swp_blks; } swp; } un_pager; struct ucred *cred; vm_ooffset_t charge; void *umtx_data; }; /* * Flags */ #define OBJ_FICTITIOUS 0x0001 /* (c) contains fictitious pages */ #define OBJ_UNMANAGED 0x0002 /* (c) contains unmanaged pages */ #define OBJ_POPULATE 0x0004 /* pager implements populate() */ #define OBJ_DEAD 0x0008 /* dead objects (during rundown) */ #define OBJ_NOSPLIT 0x0010 /* dont split this object */ #define OBJ_UMTXDEAD 0x0020 /* umtx pshared was terminated */ -#define OBJ_PIPWNT 0x0040 /* paging in progress wanted */ #define OBJ_PG_DTOR 0x0080 /* dont reset object, leave that for dtor */ #define OBJ_MIGHTBEDIRTY 0x0100 /* object might be dirty, only for vnode */ #define OBJ_TMPFS_NODE 0x0200 /* object belongs to tmpfs VREG node */ #define OBJ_TMPFS_DIRTY 0x0400 /* dirty tmpfs obj */ #define OBJ_COLORED 0x1000 /* pg_color is defined */ #define OBJ_ONEMAPPING 0x2000 /* One USE (a single, non-forked) mapping flag */ #define OBJ_DISCONNECTWNT 0x4000 /* disconnect from vnode wanted */ #define OBJ_TMPFS 0x8000 /* has tmpfs vnode allocated */ /* * Helpers to perform conversion between vm_object page indexes and offsets. * IDX_TO_OFF() converts an index into an offset. * OFF_TO_IDX() converts an offset into an index. * OBJ_MAX_SIZE specifies the maximum page index corresponding to the * maximum unsigned offset. */ #define IDX_TO_OFF(idx) (((vm_ooffset_t)(idx)) << PAGE_SHIFT) #define OFF_TO_IDX(off) ((vm_pindex_t)(((vm_ooffset_t)(off)) >> PAGE_SHIFT)) #define OBJ_MAX_SIZE (OFF_TO_IDX(UINT64_MAX) + 1) #ifdef _KERNEL #define OBJPC_SYNC 0x1 /* sync I/O */ #define OBJPC_INVAL 0x2 /* invalidate */ #define OBJPC_NOSYNC 0x4 /* skip if VPO_NOSYNC */ /* * The following options are supported by vm_object_page_remove(). */ #define OBJPR_CLEANONLY 0x1 /* Don't remove dirty pages. */ #define OBJPR_NOTMAPPED 0x2 /* Don't unmap pages. */ TAILQ_HEAD(object_q, vm_object); extern struct object_q vm_object_list; /* list of allocated objects */ extern struct mtx vm_object_list_mtx; /* lock for object list and count */ extern struct vm_object kernel_object_store; /* kernel and kmem are aliased for backwards KPI compat. */ #define kernel_object (&kernel_object_store) #define kmem_object (&kernel_object_store) #define VM_OBJECT_ASSERT_LOCKED(object) \ rw_assert(&(object)->lock, RA_LOCKED) #define VM_OBJECT_ASSERT_RLOCKED(object) \ rw_assert(&(object)->lock, RA_RLOCKED) #define VM_OBJECT_ASSERT_WLOCKED(object) \ rw_assert(&(object)->lock, RA_WLOCKED) #define VM_OBJECT_ASSERT_UNLOCKED(object) \ rw_assert(&(object)->lock, RA_UNLOCKED) #define VM_OBJECT_LOCK_DOWNGRADE(object) \ rw_downgrade(&(object)->lock) #define VM_OBJECT_RLOCK(object) \ rw_rlock(&(object)->lock) #define VM_OBJECT_RUNLOCK(object) \ rw_runlock(&(object)->lock) #define VM_OBJECT_SLEEP(object, wchan, pri, wmesg, timo) \ rw_sleep((wchan), &(object)->lock, (pri), (wmesg), (timo)) #define VM_OBJECT_TRYRLOCK(object) \ rw_try_rlock(&(object)->lock) #define VM_OBJECT_TRYWLOCK(object) \ rw_try_wlock(&(object)->lock) #define VM_OBJECT_TRYUPGRADE(object) \ rw_try_upgrade(&(object)->lock) #define VM_OBJECT_WLOCK(object) \ rw_wlock(&(object)->lock) #define VM_OBJECT_WOWNED(object) \ rw_wowned(&(object)->lock) #define VM_OBJECT_WUNLOCK(object) \ rw_wunlock(&(object)->lock) #define VM_OBJECT_DROP(object) \ lock_class_rw.lc_unlock(&(object)->lock.lock_object) #define VM_OBJECT_PICKUP(object, state) \ lock_class_rw.lc_lock(&(object)->lock.lock_object, (state)) struct vnode; /* * The object must be locked or thread private. */ static __inline void vm_object_set_flag(vm_object_t object, u_short bits) { object->flags |= bits; } /* * Conditionally set the object's color, which (1) enables the allocation * of physical memory reservations for anonymous objects and larger-than- * superpage-sized named objects and (2) determines the first page offset * within the object at which a reservation may be allocated. In other * words, the color determines the alignment of the object with respect * to the largest superpage boundary. When mapping named objects, like * files or POSIX shared memory objects, the color should be set to zero * before a virtual address is selected for the mapping. In contrast, * for anonymous objects, the color may be set after the virtual address * is selected. * * The object must be locked. */ static __inline void vm_object_color(vm_object_t object, u_short color) { if ((object->flags & OBJ_COLORED) == 0) { object->pg_color = color; object->flags |= OBJ_COLORED; } } static __inline bool vm_object_reserv(vm_object_t object) { if (object != NULL && (object->flags & (OBJ_COLORED | OBJ_FICTITIOUS)) == OBJ_COLORED) { return (true); } return (false); } void vm_object_clear_flag(vm_object_t object, u_short bits); void vm_object_pip_add(vm_object_t object, short i); -void vm_object_pip_subtract(vm_object_t object, short i); void vm_object_pip_wakeup(vm_object_t object); void vm_object_pip_wakeupn(vm_object_t object, short i); void vm_object_pip_wait(vm_object_t object, char *waitid); +void vm_object_pip_wait_unlocked(vm_object_t object, char *waitid); void umtx_shm_object_init(vm_object_t object); void umtx_shm_object_terminated(vm_object_t object); extern int umtx_shm_vnobj_persistent; vm_object_t vm_object_allocate (objtype_t, vm_pindex_t); boolean_t vm_object_coalesce(vm_object_t, vm_ooffset_t, vm_size_t, vm_size_t, boolean_t); void vm_object_collapse (vm_object_t); void vm_object_deallocate (vm_object_t); void vm_object_destroy (vm_object_t); void vm_object_terminate (vm_object_t); void vm_object_set_writeable_dirty (vm_object_t); void vm_object_init (void); int vm_object_kvme_type(vm_object_t object, struct vnode **vpp); void vm_object_madvise(vm_object_t, vm_pindex_t, vm_pindex_t, int); boolean_t vm_object_page_clean(vm_object_t object, vm_ooffset_t start, vm_ooffset_t end, int flags); void vm_object_page_noreuse(vm_object_t object, vm_pindex_t start, vm_pindex_t end); void vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end, int options); boolean_t vm_object_populate(vm_object_t, vm_pindex_t, vm_pindex_t); void vm_object_print(long addr, boolean_t have_addr, long count, char *modif); void vm_object_reference (vm_object_t); void vm_object_reference_locked(vm_object_t); int vm_object_set_memattr(vm_object_t object, vm_memattr_t memattr); void vm_object_shadow (vm_object_t *, vm_ooffset_t *, vm_size_t); void vm_object_split(vm_map_entry_t); boolean_t vm_object_sync(vm_object_t, vm_ooffset_t, vm_size_t, boolean_t, boolean_t); void vm_object_unwire(vm_object_t object, vm_ooffset_t offset, vm_size_t length, uint8_t queue); struct vnode *vm_object_vnode(vm_object_t object); #endif /* _KERNEL */ #endif /* _VM_OBJECT_ */