4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2013 by Delphix. All rights reserved.
26 /* Portions Copyright 2007 Jeremy Teo */
27 /* Portions Copyright 2010 Robert Milkowski */
29 #include <sys/types.h>
30 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/sysmacros.h>
34 #include <sys/resource.h>
36 #include <sys/vnode.h>
40 #include <sys/taskq.h>
42 #include <sys/atomic.h>
43 #include <sys/namei.h>
45 #include <sys/cmn_err.h>
46 #include <sys/errno.h>
47 #include <sys/unistd.h>
48 #include <sys/zfs_dir.h>
49 #include <sys/zfs_ioctl.h>
50 #include <sys/fs/zfs.h>
52 #include <sys/dmu_objset.h>
58 #include <sys/dirent.h>
59 #include <sys/policy.h>
60 #include <sys/sunddi.h>
61 #include <sys/filio.h>
63 #include <sys/zfs_ctldir.h>
64 #include <sys/zfs_fuid.h>
65 #include <sys/zfs_sa.h>
67 #include <sys/zfs_rlock.h>
68 #include <sys/extdirent.h>
69 #include <sys/kidmap.h>
72 #include <sys/sf_buf.h>
73 #include <sys/sched.h>
75 #include <vm/vm_pageout.h>
80 * Each vnode op performs some logical unit of work. To do this, the ZPL must
81 * properly lock its in-core state, create a DMU transaction, do the work,
82 * record this work in the intent log (ZIL), commit the DMU transaction,
83 * and wait for the intent log to commit if it is a synchronous operation.
84 * Moreover, the vnode ops must work in both normal and log replay context.
85 * The ordering of events is important to avoid deadlocks and references
86 * to freed memory. The example below illustrates the following Big Rules:
88 * (1) A check must be made in each zfs thread for a mounted file system.
89 * This is done avoiding races using ZFS_ENTER(zfsvfs).
90 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
91 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
92 * can return EIO from the calling function.
94 * (2) VN_RELE() should always be the last thing except for zil_commit()
95 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
96 * First, if it's the last reference, the vnode/znode
97 * can be freed, so the zp may point to freed memory. Second, the last
98 * reference will call zfs_zinactive(), which may induce a lot of work --
99 * pushing cached pages (which acquires range locks) and syncing out
100 * cached atime changes. Third, zfs_zinactive() may require a new tx,
101 * which could deadlock the system if you were already holding one.
102 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
104 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
105 * as they can span dmu_tx_assign() calls.
107 * (4) Always pass TXG_NOWAIT as the second argument to dmu_tx_assign().
108 * This is critical because we don't want to block while holding locks.
109 * Note, in particular, that if a lock is sometimes acquired before
110 * the tx assigns, and sometimes after (e.g. z_lock), then failing to
111 * use a non-blocking assign can deadlock the system. The scenario:
113 * Thread A has grabbed a lock before calling dmu_tx_assign().
114 * Thread B is in an already-assigned tx, and blocks for this lock.
115 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
116 * forever, because the previous txg can't quiesce until B's tx commits.
118 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
119 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
120 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
121 * to indicate that this operation has already called dmu_tx_wait().
122 * This will ensure that we don't retry forever, waiting a short bit
125 * (5) If the operation succeeded, generate the intent log entry for it
126 * before dropping locks. This ensures that the ordering of events
127 * in the intent log matches the order in which they actually occurred.
128 * During ZIL replay the zfs_log_* functions will update the sequence
129 * number to indicate the zil transaction has replayed.
131 * (6) At the end of each vnode op, the DMU tx must always commit,
132 * regardless of whether there were any errors.
134 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
135 * to ensure that synchronous semantics are provided when necessary.
137 * In general, this is how things should be ordered in each vnode op:
139 * ZFS_ENTER(zfsvfs); // exit if unmounted
141 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD())
142 * rw_enter(...); // grab any other locks you need
143 * tx = dmu_tx_create(...); // get DMU tx
144 * dmu_tx_hold_*(); // hold each object you might modify
145 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
147 * rw_exit(...); // drop locks
148 * zfs_dirent_unlock(dl); // unlock directory entry
149 * VN_RELE(...); // release held vnodes
150 * if (error == ERESTART) {
156 * dmu_tx_abort(tx); // abort DMU tx
157 * ZFS_EXIT(zfsvfs); // finished in zfs
158 * return (error); // really out of space
160 * error = do_real_work(); // do whatever this VOP does
162 * zfs_log_*(...); // on success, make ZIL entry
163 * dmu_tx_commit(tx); // commit DMU tx -- error or not
164 * rw_exit(...); // drop locks
165 * zfs_dirent_unlock(dl); // unlock directory entry
166 * VN_RELE(...); // release held vnodes
167 * zil_commit(zilog, foid); // synchronous when necessary
168 * ZFS_EXIT(zfsvfs); // finished in zfs
169 * return (error); // done, report error
174 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
176 znode_t *zp = VTOZ(*vpp);
177 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
182 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
183 ((flag & FAPPEND) == 0)) {
185 return (SET_ERROR(EPERM));
188 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
189 ZTOV(zp)->v_type == VREG &&
190 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
191 if (fs_vscan(*vpp, cr, 0) != 0) {
193 return (SET_ERROR(EACCES));
197 /* Keep a count of the synchronous opens in the znode */
198 if (flag & (FSYNC | FDSYNC))
199 atomic_inc_32(&zp->z_sync_cnt);
207 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
208 caller_context_t *ct)
210 znode_t *zp = VTOZ(vp);
211 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
214 * Clean up any locks held by this process on the vp.
216 cleanlocks(vp, ddi_get_pid(), 0);
217 cleanshares(vp, ddi_get_pid());
222 /* Decrement the synchronous opens in the znode */
223 if ((flag & (FSYNC | FDSYNC)) && (count == 1))
224 atomic_dec_32(&zp->z_sync_cnt);
226 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
227 ZTOV(zp)->v_type == VREG &&
228 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
229 VERIFY(fs_vscan(vp, cr, 1) == 0);
236 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
237 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
240 zfs_holey(vnode_t *vp, u_long cmd, offset_t *off)
242 znode_t *zp = VTOZ(vp);
243 uint64_t noff = (uint64_t)*off; /* new offset */
248 file_sz = zp->z_size;
249 if (noff >= file_sz) {
250 return (SET_ERROR(ENXIO));
253 if (cmd == _FIO_SEEK_HOLE)
258 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
261 if ((error == ESRCH) || (noff > file_sz)) {
263 * Handle the virtual hole at the end of file.
269 return (SET_ERROR(ENXIO));
280 zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred,
281 int *rvalp, caller_context_t *ct)
293 * The following two ioctls are used by bfu. Faking out,
294 * necessary to avoid bfu errors.
303 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
304 return (SET_ERROR(EFAULT));
306 off = *(offset_t *)data;
309 zfsvfs = zp->z_zfsvfs;
313 /* offset parameter is in/out */
314 error = zfs_holey(vp, com, &off);
319 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
320 return (SET_ERROR(EFAULT));
322 *(offset_t *)data = off;
326 return (SET_ERROR(ENOTTY));
330 page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes)
337 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE
338 * aligned boundaries, if the range is not aligned. As a result a
339 * DEV_BSIZE subrange with partially dirty data may get marked as clean.
340 * It may happen that all DEV_BSIZE subranges are marked clean and thus
341 * the whole page would be considred clean despite have some dirty data.
342 * For this reason we should shrink the range to DEV_BSIZE aligned
343 * boundaries before calling vm_page_clear_dirty.
345 end = rounddown2(off + nbytes, DEV_BSIZE);
346 off = roundup2(off, DEV_BSIZE);
350 VM_OBJECT_LOCK_ASSERT(obj, MA_OWNED);
353 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
355 if ((pp->oflags & VPO_BUSY) != 0) {
357 * Reference the page before unlocking and
358 * sleeping so that the page daemon is less
359 * likely to reclaim it.
361 vm_page_reference(pp);
362 vm_page_sleep(pp, "zfsmwb");
365 } else if (pp == NULL) {
366 pp = vm_page_alloc(obj, OFF_TO_IDX(start),
367 VM_ALLOC_SYSTEM | VM_ALLOC_IFCACHED |
370 ASSERT(pp != NULL && !pp->valid);
375 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
376 vm_object_pip_add(obj, 1);
377 vm_page_io_start(pp);
378 pmap_remove_write(pp);
380 vm_page_clear_dirty(pp, off, nbytes);
388 page_unbusy(vm_page_t pp)
391 vm_page_io_finish(pp);
392 vm_object_pip_subtract(pp->object, 1);
396 page_hold(vnode_t *vp, int64_t start)
402 VM_OBJECT_LOCK_ASSERT(obj, MA_OWNED);
405 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
407 if ((pp->oflags & VPO_BUSY) != 0) {
409 * Reference the page before unlocking and
410 * sleeping so that the page daemon is less
411 * likely to reclaim it.
413 vm_page_reference(pp);
414 vm_page_sleep(pp, "zfsmwb");
418 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
431 page_unhold(vm_page_t pp)
440 zfs_map_page(vm_page_t pp, struct sf_buf **sfp)
443 *sfp = sf_buf_alloc(pp, 0);
444 return ((caddr_t)sf_buf_kva(*sfp));
448 zfs_unmap_page(struct sf_buf *sf)
455 * When a file is memory mapped, we must keep the IO data synchronized
456 * between the DMU cache and the memory mapped pages. What this means:
458 * On Write: If we find a memory mapped page, we write to *both*
459 * the page and the dmu buffer.
462 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid,
463 int segflg, dmu_tx_t *tx)
470 ASSERT(vp->v_mount != NULL);
474 off = start & PAGEOFFSET;
476 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
478 int nbytes = imin(PAGESIZE - off, len);
480 if (segflg == UIO_NOCOPY) {
481 pp = vm_page_lookup(obj, OFF_TO_IDX(start));
483 ("zfs update_pages: NULL page in putpages case"));
485 ("zfs update_pages: unaligned data in putpages case"));
486 KASSERT(pp->valid == VM_PAGE_BITS_ALL,
487 ("zfs update_pages: invalid page in putpages case"));
488 KASSERT(pp->busy > 0,
489 ("zfs update_pages: unbusy page in putpages case"));
490 KASSERT(!pmap_page_is_write_mapped(pp),
491 ("zfs update_pages: writable page in putpages case"));
492 VM_OBJECT_UNLOCK(obj);
494 va = zfs_map_page(pp, &sf);
495 (void) dmu_write(os, oid, start, nbytes, va, tx);
500 } else if ((pp = page_busy(vp, start, off, nbytes)) != NULL) {
501 VM_OBJECT_UNLOCK(obj);
503 va = zfs_map_page(pp, &sf);
504 (void) dmu_read(os, oid, start+off, nbytes,
505 va+off, DMU_READ_PREFETCH);;
514 if (segflg != UIO_NOCOPY)
515 vm_object_pip_wakeupn(obj, 0);
516 VM_OBJECT_UNLOCK(obj);
520 * Read with UIO_NOCOPY flag means that sendfile(2) requests
521 * ZFS to populate a range of page cache pages with data.
523 * NOTE: this function could be optimized to pre-allocate
524 * all pages in advance, drain VPO_BUSY on all of them,
525 * map them into contiguous KVA region and populate them
526 * in one single dmu_read() call.
529 mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio)
531 znode_t *zp = VTOZ(vp);
532 objset_t *os = zp->z_zfsvfs->z_os;
542 ASSERT(uio->uio_segflg == UIO_NOCOPY);
543 ASSERT(vp->v_mount != NULL);
546 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0);
549 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) {
550 int bytes = MIN(PAGESIZE, len);
552 pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_NOBUSY |
553 VM_ALLOC_NORMAL | VM_ALLOC_RETRY | VM_ALLOC_IGN_SBUSY);
554 if (pp->valid == 0) {
555 vm_page_io_start(pp);
556 VM_OBJECT_UNLOCK(obj);
557 va = zfs_map_page(pp, &sf);
558 error = dmu_read(os, zp->z_id, start, bytes, va,
560 if (bytes != PAGESIZE && error == 0)
561 bzero(va + bytes, PAGESIZE - bytes);
564 vm_page_io_finish(pp);
567 if (pp->wire_count == 0 && pp->valid == 0 &&
568 pp->busy == 0 && !(pp->oflags & VPO_BUSY))
571 pp->valid = VM_PAGE_BITS_ALL;
572 vm_page_activate(pp);
576 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
580 uio->uio_resid -= bytes;
581 uio->uio_offset += bytes;
584 VM_OBJECT_UNLOCK(obj);
589 * When a file is memory mapped, we must keep the IO data synchronized
590 * between the DMU cache and the memory mapped pages. What this means:
592 * On Read: We "read" preferentially from memory mapped pages,
593 * else we default from the dmu buffer.
595 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
596 * the file is memory mapped.
599 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
601 znode_t *zp = VTOZ(vp);
602 objset_t *os = zp->z_zfsvfs->z_os;
610 ASSERT(vp->v_mount != NULL);
614 start = uio->uio_loffset;
615 off = start & PAGEOFFSET;
617 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
619 uint64_t bytes = MIN(PAGESIZE - off, len);
621 if (pp = page_hold(vp, start)) {
625 VM_OBJECT_UNLOCK(obj);
626 va = zfs_map_page(pp, &sf);
627 error = uiomove(va + off, bytes, UIO_READ, uio);
632 VM_OBJECT_UNLOCK(obj);
633 error = dmu_read_uio(os, zp->z_id, uio, bytes);
641 VM_OBJECT_UNLOCK(obj);
645 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
648 * Read bytes from specified file into supplied buffer.
650 * IN: vp - vnode of file to be read from.
651 * uio - structure supplying read location, range info,
653 * ioflag - SYNC flags; used to provide FRSYNC semantics.
654 * cr - credentials of caller.
655 * ct - caller context
657 * OUT: uio - updated offset and range, buffer filled.
659 * RETURN: 0 on success, error code on failure.
662 * vp - atime updated if byte count > 0
666 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
668 znode_t *zp = VTOZ(vp);
669 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
680 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
682 return (SET_ERROR(EACCES));
686 * Validate file offset
688 if (uio->uio_loffset < (offset_t)0) {
690 return (SET_ERROR(EINVAL));
694 * Fasttrack empty reads
696 if (uio->uio_resid == 0) {
702 * Check for mandatory locks
704 if (MANDMODE(zp->z_mode)) {
705 if (error = chklock(vp, FREAD,
706 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
713 * If we're in FRSYNC mode, sync out this znode before reading it.
716 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
717 zil_commit(zfsvfs->z_log, zp->z_id);
720 * Lock the range against changes.
722 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
725 * If we are reading past end-of-file we can skip
726 * to the end; but we might still need to set atime.
728 if (uio->uio_loffset >= zp->z_size) {
733 ASSERT(uio->uio_loffset < zp->z_size);
734 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
737 if ((uio->uio_extflg == UIO_XUIO) &&
738 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
740 int blksz = zp->z_blksz;
741 uint64_t offset = uio->uio_loffset;
743 xuio = (xuio_t *)uio;
745 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
748 ASSERT(offset + n <= blksz);
751 (void) dmu_xuio_init(xuio, nblk);
753 if (vn_has_cached_data(vp)) {
755 * For simplicity, we always allocate a full buffer
756 * even if we only expect to read a portion of a block.
758 while (--nblk >= 0) {
759 (void) dmu_xuio_add(xuio,
760 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
768 nbytes = MIN(n, zfs_read_chunk_size -
769 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
772 if (uio->uio_segflg == UIO_NOCOPY)
773 error = mappedread_sf(vp, nbytes, uio);
775 #endif /* __FreeBSD__ */
776 if (vn_has_cached_data(vp))
777 error = mappedread(vp, nbytes, uio);
779 error = dmu_read_uio(os, zp->z_id, uio, nbytes);
781 /* convert checksum errors into IO errors */
783 error = SET_ERROR(EIO);
790 zfs_range_unlock(rl);
792 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
798 * Write the bytes to a file.
800 * IN: vp - vnode of file to be written to.
801 * uio - structure supplying write location, range info,
803 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is
804 * set if in append mode.
805 * cr - credentials of caller.
806 * ct - caller context (NFS/CIFS fem monitor only)
808 * OUT: uio - updated offset and range.
810 * RETURN: 0 on success, error code on failure.
813 * vp - ctime|mtime updated if byte count > 0
818 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
820 znode_t *zp = VTOZ(vp);
821 rlim64_t limit = MAXOFFSET_T;
822 ssize_t start_resid = uio->uio_resid;
826 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
831 int max_blksz = zfsvfs->z_max_blksz;
834 iovec_t *aiov = NULL;
837 int iovcnt = uio->uio_iovcnt;
838 iovec_t *iovp = uio->uio_iov;
841 sa_bulk_attr_t bulk[4];
842 uint64_t mtime[2], ctime[2];
845 * Fasttrack empty write
851 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
857 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
858 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
859 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
861 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
865 * If immutable or not appending then return EPERM
867 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
868 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
869 (uio->uio_loffset < zp->z_size))) {
871 return (SET_ERROR(EPERM));
874 zilog = zfsvfs->z_log;
877 * Validate file offset
879 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
882 return (SET_ERROR(EINVAL));
886 * Check for mandatory locks before calling zfs_range_lock()
887 * in order to prevent a deadlock with locks set via fcntl().
889 if (MANDMODE((mode_t)zp->z_mode) &&
890 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
897 * Pre-fault the pages to ensure slow (eg NFS) pages
899 * Skip this if uio contains loaned arc_buf.
901 if ((uio->uio_extflg == UIO_XUIO) &&
902 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
903 xuio = (xuio_t *)uio;
905 uio_prefaultpages(MIN(n, max_blksz), uio);
909 * If in append mode, set the io offset pointer to eof.
911 if (ioflag & FAPPEND) {
913 * Obtain an appending range lock to guarantee file append
914 * semantics. We reset the write offset once we have the lock.
916 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
918 if (rl->r_len == UINT64_MAX) {
920 * We overlocked the file because this write will cause
921 * the file block size to increase.
922 * Note that zp_size cannot change with this lock held.
926 uio->uio_loffset = woff;
929 * Note that if the file block size will change as a result of
930 * this write, then this range lock will lock the entire file
931 * so that we can re-write the block safely.
933 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
936 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) {
937 zfs_range_unlock(rl);
943 zfs_range_unlock(rl);
945 return (SET_ERROR(EFBIG));
948 if ((woff + n) > limit || woff > (limit - n))
951 /* Will this write extend the file length? */
952 write_eof = (woff + n > zp->z_size);
954 end_size = MAX(zp->z_size, woff + n);
957 * Write the file in reasonable size chunks. Each chunk is written
958 * in a separate transaction; this keeps the intent log records small
959 * and allows us to do more fine-grained space accounting.
963 woff = uio->uio_loffset;
965 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
966 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
968 dmu_return_arcbuf(abuf);
969 error = SET_ERROR(EDQUOT);
973 if (xuio && abuf == NULL) {
974 ASSERT(i_iov < iovcnt);
976 abuf = dmu_xuio_arcbuf(xuio, i_iov);
977 dmu_xuio_clear(xuio, i_iov);
978 DTRACE_PROBE3(zfs_cp_write, int, i_iov,
979 iovec_t *, aiov, arc_buf_t *, abuf);
980 ASSERT((aiov->iov_base == abuf->b_data) ||
981 ((char *)aiov->iov_base - (char *)abuf->b_data +
982 aiov->iov_len == arc_buf_size(abuf)));
984 } else if (abuf == NULL && n >= max_blksz &&
985 woff >= zp->z_size &&
986 P2PHASE(woff, max_blksz) == 0 &&
987 zp->z_blksz == max_blksz) {
989 * This write covers a full block. "Borrow" a buffer
990 * from the dmu so that we can fill it before we enter
991 * a transaction. This avoids the possibility of
992 * holding up the transaction if the data copy hangs
993 * up on a pagefault (e.g., from an NFS server mapping).
997 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
999 ASSERT(abuf != NULL);
1000 ASSERT(arc_buf_size(abuf) == max_blksz);
1001 if (error = uiocopy(abuf->b_data, max_blksz,
1002 UIO_WRITE, uio, &cbytes)) {
1003 dmu_return_arcbuf(abuf);
1006 ASSERT(cbytes == max_blksz);
1010 * Start a transaction.
1012 tx = dmu_tx_create(zfsvfs->z_os);
1013 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1014 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
1015 zfs_sa_upgrade_txholds(tx, zp);
1016 error = dmu_tx_assign(tx, TXG_NOWAIT);
1018 if (error == ERESTART) {
1025 dmu_return_arcbuf(abuf);
1030 * If zfs_range_lock() over-locked we grow the blocksize
1031 * and then reduce the lock range. This will only happen
1032 * on the first iteration since zfs_range_reduce() will
1033 * shrink down r_len to the appropriate size.
1035 if (rl->r_len == UINT64_MAX) {
1038 if (zp->z_blksz > max_blksz) {
1039 ASSERT(!ISP2(zp->z_blksz));
1040 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
1042 new_blksz = MIN(end_size, max_blksz);
1044 zfs_grow_blocksize(zp, new_blksz, tx);
1045 zfs_range_reduce(rl, woff, n);
1049 * XXX - should we really limit each write to z_max_blksz?
1050 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
1052 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
1054 if (woff + nbytes > zp->z_size)
1055 vnode_pager_setsize(vp, woff + nbytes);
1058 tx_bytes = uio->uio_resid;
1059 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
1061 tx_bytes -= uio->uio_resid;
1064 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
1066 * If this is not a full block write, but we are
1067 * extending the file past EOF and this data starts
1068 * block-aligned, use assign_arcbuf(). Otherwise,
1069 * write via dmu_write().
1071 if (tx_bytes < max_blksz && (!write_eof ||
1072 aiov->iov_base != abuf->b_data)) {
1074 dmu_write(zfsvfs->z_os, zp->z_id, woff,
1075 aiov->iov_len, aiov->iov_base, tx);
1076 dmu_return_arcbuf(abuf);
1077 xuio_stat_wbuf_copied();
1079 ASSERT(xuio || tx_bytes == max_blksz);
1080 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
1083 ASSERT(tx_bytes <= uio->uio_resid);
1084 uioskip(uio, tx_bytes);
1086 if (tx_bytes && vn_has_cached_data(vp)) {
1087 update_pages(vp, woff, tx_bytes, zfsvfs->z_os,
1088 zp->z_id, uio->uio_segflg, tx);
1092 * If we made no progress, we're done. If we made even
1093 * partial progress, update the znode and ZIL accordingly.
1095 if (tx_bytes == 0) {
1096 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
1097 (void *)&zp->z_size, sizeof (uint64_t), tx);
1104 * Clear Set-UID/Set-GID bits on successful write if not
1105 * privileged and at least one of the excute bits is set.
1107 * It would be nice to to this after all writes have
1108 * been done, but that would still expose the ISUID/ISGID
1109 * to another app after the partial write is committed.
1111 * Note: we don't call zfs_fuid_map_id() here because
1112 * user 0 is not an ephemeral uid.
1114 mutex_enter(&zp->z_acl_lock);
1115 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
1116 (S_IXUSR >> 6))) != 0 &&
1117 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
1118 secpolicy_vnode_setid_retain(vp, cr,
1119 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
1121 zp->z_mode &= ~(S_ISUID | S_ISGID);
1122 newmode = zp->z_mode;
1123 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
1124 (void *)&newmode, sizeof (uint64_t), tx);
1126 mutex_exit(&zp->z_acl_lock);
1128 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
1132 * Update the file size (zp_size) if it has changed;
1133 * account for possible concurrent updates.
1135 while ((end_size = zp->z_size) < uio->uio_loffset) {
1136 (void) atomic_cas_64(&zp->z_size, end_size,
1141 * If we are replaying and eof is non zero then force
1142 * the file size to the specified eof. Note, there's no
1143 * concurrency during replay.
1145 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
1146 zp->z_size = zfsvfs->z_replay_eof;
1148 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1150 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
1155 ASSERT(tx_bytes == nbytes);
1160 uio_prefaultpages(MIN(n, max_blksz), uio);
1164 zfs_range_unlock(rl);
1167 * If we're in replay mode, or we made no progress, return error.
1168 * Otherwise, it's at least a partial write, so it's successful.
1170 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1175 if (ioflag & (FSYNC | FDSYNC) ||
1176 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1177 zil_commit(zilog, zp->z_id);
1184 zfs_get_done(zgd_t *zgd, int error)
1186 znode_t *zp = zgd->zgd_private;
1187 objset_t *os = zp->z_zfsvfs->z_os;
1191 dmu_buf_rele(zgd->zgd_db, zgd);
1193 zfs_range_unlock(zgd->zgd_rl);
1195 vfslocked = VFS_LOCK_GIANT(zp->z_zfsvfs->z_vfs);
1197 * Release the vnode asynchronously as we currently have the
1198 * txg stopped from syncing.
1200 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1202 if (error == 0 && zgd->zgd_bp)
1203 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1205 kmem_free(zgd, sizeof (zgd_t));
1206 VFS_UNLOCK_GIANT(vfslocked);
1210 static int zil_fault_io = 0;
1214 * Get data to generate a TX_WRITE intent log record.
1217 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1219 zfsvfs_t *zfsvfs = arg;
1220 objset_t *os = zfsvfs->z_os;
1222 uint64_t object = lr->lr_foid;
1223 uint64_t offset = lr->lr_offset;
1224 uint64_t size = lr->lr_length;
1225 blkptr_t *bp = &lr->lr_blkptr;
1230 ASSERT(zio != NULL);
1234 * Nothing to do if the file has been removed
1236 if (zfs_zget(zfsvfs, object, &zp) != 0)
1237 return (SET_ERROR(ENOENT));
1238 if (zp->z_unlinked) {
1240 * Release the vnode asynchronously as we currently have the
1241 * txg stopped from syncing.
1243 VN_RELE_ASYNC(ZTOV(zp),
1244 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1245 return (SET_ERROR(ENOENT));
1248 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1249 zgd->zgd_zilog = zfsvfs->z_log;
1250 zgd->zgd_private = zp;
1253 * Write records come in two flavors: immediate and indirect.
1254 * For small writes it's cheaper to store the data with the
1255 * log record (immediate); for large writes it's cheaper to
1256 * sync the data and get a pointer to it (indirect) so that
1257 * we don't have to write the data twice.
1259 if (buf != NULL) { /* immediate write */
1260 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1261 /* test for truncation needs to be done while range locked */
1262 if (offset >= zp->z_size) {
1263 error = SET_ERROR(ENOENT);
1265 error = dmu_read(os, object, offset, size, buf,
1266 DMU_READ_NO_PREFETCH);
1268 ASSERT(error == 0 || error == ENOENT);
1269 } else { /* indirect write */
1271 * Have to lock the whole block to ensure when it's
1272 * written out and it's checksum is being calculated
1273 * that no one can change the data. We need to re-check
1274 * blocksize after we get the lock in case it's changed!
1279 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1281 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1283 if (zp->z_blksz == size)
1286 zfs_range_unlock(zgd->zgd_rl);
1288 /* test for truncation needs to be done while range locked */
1289 if (lr->lr_offset >= zp->z_size)
1290 error = SET_ERROR(ENOENT);
1293 error = SET_ERROR(EIO);
1298 error = dmu_buf_hold(os, object, offset, zgd, &db,
1299 DMU_READ_NO_PREFETCH);
1302 blkptr_t *obp = dmu_buf_get_blkptr(db);
1304 ASSERT(BP_IS_HOLE(bp));
1311 ASSERT(db->db_offset == offset);
1312 ASSERT(db->db_size == size);
1314 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1316 ASSERT(error || lr->lr_length <= zp->z_blksz);
1319 * On success, we need to wait for the write I/O
1320 * initiated by dmu_sync() to complete before we can
1321 * release this dbuf. We will finish everything up
1322 * in the zfs_get_done() callback.
1327 if (error == EALREADY) {
1328 lr->lr_common.lrc_txtype = TX_WRITE2;
1334 zfs_get_done(zgd, error);
1341 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1342 caller_context_t *ct)
1344 znode_t *zp = VTOZ(vp);
1345 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1351 if (flag & V_ACE_MASK)
1352 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1354 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1361 * If vnode is for a device return a specfs vnode instead.
1364 specvp_check(vnode_t **vpp, cred_t *cr)
1368 if (IS_DEVVP(*vpp)) {
1371 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1374 error = SET_ERROR(ENOSYS);
1382 * Lookup an entry in a directory, or an extended attribute directory.
1383 * If it exists, return a held vnode reference for it.
1385 * IN: dvp - vnode of directory to search.
1386 * nm - name of entry to lookup.
1387 * pnp - full pathname to lookup [UNUSED].
1388 * flags - LOOKUP_XATTR set if looking for an attribute.
1389 * rdir - root directory vnode [UNUSED].
1390 * cr - credentials of caller.
1391 * ct - caller context
1392 * direntflags - directory lookup flags
1393 * realpnp - returned pathname.
1395 * OUT: vpp - vnode of located entry, NULL if not found.
1397 * RETURN: 0 on success, error code on failure.
1404 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp,
1405 int nameiop, cred_t *cr, kthread_t *td, int flags)
1407 znode_t *zdp = VTOZ(dvp);
1408 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1410 int *direntflags = NULL;
1411 void *realpnp = NULL;
1414 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1416 if (dvp->v_type != VDIR) {
1417 return (SET_ERROR(ENOTDIR));
1418 } else if (zdp->z_sa_hdl == NULL) {
1419 return (SET_ERROR(EIO));
1422 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1423 error = zfs_fastaccesschk_execute(zdp, cr);
1431 vnode_t *tvp = dnlc_lookup(dvp, nm);
1434 error = zfs_fastaccesschk_execute(zdp, cr);
1439 if (tvp == DNLC_NO_VNODE) {
1441 return (SET_ERROR(ENOENT));
1444 return (specvp_check(vpp, cr));
1450 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1457 if (flags & LOOKUP_XATTR) {
1460 * If the xattr property is off, refuse the lookup request.
1462 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1464 return (SET_ERROR(EINVAL));
1469 * We don't allow recursive attributes..
1470 * Maybe someday we will.
1472 if (zdp->z_pflags & ZFS_XATTR) {
1474 return (SET_ERROR(EINVAL));
1477 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1483 * Do we have permission to get into attribute directory?
1486 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1496 if (dvp->v_type != VDIR) {
1498 return (SET_ERROR(ENOTDIR));
1502 * Check accessibility of directory.
1505 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1510 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1511 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1513 return (SET_ERROR(EILSEQ));
1516 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1518 error = specvp_check(vpp, cr);
1520 /* Translate errors and add SAVENAME when needed. */
1521 if (cnp->cn_flags & ISLASTCN) {
1525 if (error == ENOENT) {
1526 error = EJUSTRETURN;
1527 cnp->cn_flags |= SAVENAME;
1533 cnp->cn_flags |= SAVENAME;
1537 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) {
1540 if (cnp->cn_flags & ISDOTDOT) {
1541 ltype = VOP_ISLOCKED(dvp);
1545 error = vn_lock(*vpp, cnp->cn_lkflags);
1546 if (cnp->cn_flags & ISDOTDOT)
1547 vn_lock(dvp, ltype | LK_RETRY);
1557 #ifdef FREEBSD_NAMECACHE
1559 * Insert name into cache (as non-existent) if appropriate.
1561 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
1562 cache_enter(dvp, *vpp, cnp);
1564 * Insert name into cache if appropriate.
1566 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) {
1567 if (!(cnp->cn_flags & ISLASTCN) ||
1568 (nameiop != DELETE && nameiop != RENAME)) {
1569 cache_enter(dvp, *vpp, cnp);
1578 * Attempt to create a new entry in a directory. If the entry
1579 * already exists, truncate the file if permissible, else return
1580 * an error. Return the vp of the created or trunc'd file.
1582 * IN: dvp - vnode of directory to put new file entry in.
1583 * name - name of new file entry.
1584 * vap - attributes of new file.
1585 * excl - flag indicating exclusive or non-exclusive mode.
1586 * mode - mode to open file with.
1587 * cr - credentials of caller.
1588 * flag - large file flag [UNUSED].
1589 * ct - caller context
1590 * vsecp - ACL to be set
1592 * OUT: vpp - vnode of created or trunc'd entry.
1594 * RETURN: 0 on success, error code on failure.
1597 * dvp - ctime|mtime updated if new entry created
1598 * vp - ctime|mtime always, atime if new
1603 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode,
1604 vnode_t **vpp, cred_t *cr, kthread_t *td)
1606 znode_t *zp, *dzp = VTOZ(dvp);
1607 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1615 gid_t gid = crgetgid(cr);
1616 zfs_acl_ids_t acl_ids;
1617 boolean_t fuid_dirtied;
1618 boolean_t have_acl = B_FALSE;
1619 boolean_t waited = B_FALSE;
1624 * If we have an ephemeral id, ACL, or XVATTR then
1625 * make sure file system is at proper version
1628 ksid = crgetsid(cr, KSID_OWNER);
1630 uid = ksid_getid(ksid);
1634 if (zfsvfs->z_use_fuids == B_FALSE &&
1635 (vsecp || (vap->va_mask & AT_XVATTR) ||
1636 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1637 return (SET_ERROR(EINVAL));
1642 zilog = zfsvfs->z_log;
1644 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1645 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1647 return (SET_ERROR(EILSEQ));
1650 if (vap->va_mask & AT_XVATTR) {
1651 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
1652 crgetuid(cr), cr, vap->va_type)) != 0) {
1660 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr))
1661 vap->va_mode &= ~S_ISVTX;
1663 if (*name == '\0') {
1665 * Null component name refers to the directory itself.
1672 /* possible VN_HOLD(zp) */
1675 if (flag & FIGNORECASE)
1678 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1682 zfs_acl_ids_free(&acl_ids);
1683 if (strcmp(name, "..") == 0)
1684 error = SET_ERROR(EISDIR);
1694 * Create a new file object and update the directory
1697 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1699 zfs_acl_ids_free(&acl_ids);
1704 * We only support the creation of regular files in
1705 * extended attribute directories.
1708 if ((dzp->z_pflags & ZFS_XATTR) &&
1709 (vap->va_type != VREG)) {
1711 zfs_acl_ids_free(&acl_ids);
1712 error = SET_ERROR(EINVAL);
1716 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1717 cr, vsecp, &acl_ids)) != 0)
1721 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1722 zfs_acl_ids_free(&acl_ids);
1723 error = SET_ERROR(EDQUOT);
1727 tx = dmu_tx_create(os);
1729 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1730 ZFS_SA_BASE_ATTR_SIZE);
1732 fuid_dirtied = zfsvfs->z_fuid_dirty;
1734 zfs_fuid_txhold(zfsvfs, tx);
1735 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1736 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1737 if (!zfsvfs->z_use_sa &&
1738 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1739 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1740 0, acl_ids.z_aclp->z_acl_bytes);
1742 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1744 zfs_dirent_unlock(dl);
1745 if (error == ERESTART) {
1751 zfs_acl_ids_free(&acl_ids);
1756 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1759 zfs_fuid_sync(zfsvfs, tx);
1761 (void) zfs_link_create(dl, zp, tx, ZNEW);
1762 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1763 if (flag & FIGNORECASE)
1765 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1766 vsecp, acl_ids.z_fuidp, vap);
1767 zfs_acl_ids_free(&acl_ids);
1770 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1773 zfs_acl_ids_free(&acl_ids);
1777 * A directory entry already exists for this name.
1780 * Can't truncate an existing file if in exclusive mode.
1783 error = SET_ERROR(EEXIST);
1787 * Can't open a directory for writing.
1789 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1790 error = SET_ERROR(EISDIR);
1794 * Verify requested access to file.
1796 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1800 mutex_enter(&dzp->z_lock);
1802 mutex_exit(&dzp->z_lock);
1805 * Truncate regular files if requested.
1807 if ((ZTOV(zp)->v_type == VREG) &&
1808 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1809 /* we can't hold any locks when calling zfs_freesp() */
1810 zfs_dirent_unlock(dl);
1812 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1814 vnevent_create(ZTOV(zp), ct);
1820 zfs_dirent_unlock(dl);
1827 error = specvp_check(vpp, cr);
1830 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1831 zil_commit(zilog, 0);
1838 * Remove an entry from a directory.
1840 * IN: dvp - vnode of directory to remove entry from.
1841 * name - name of entry to remove.
1842 * cr - credentials of caller.
1843 * ct - caller context
1844 * flags - case flags
1846 * RETURN: 0 on success, error code on failure.
1850 * vp - ctime (if nlink > 0)
1853 uint64_t null_xattr = 0;
1857 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1860 znode_t *zp, *dzp = VTOZ(dvp);
1863 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1865 uint64_t acl_obj, xattr_obj;
1866 uint64_t xattr_obj_unlinked = 0;
1870 boolean_t may_delete_now, delete_now = FALSE;
1871 boolean_t unlinked, toobig = FALSE;
1873 pathname_t *realnmp = NULL;
1877 boolean_t waited = B_FALSE;
1881 zilog = zfsvfs->z_log;
1883 if (flags & FIGNORECASE) {
1893 * Attempt to lock directory; fail if entry doesn't exist.
1895 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1905 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1910 * Need to use rmdir for removing directories.
1912 if (vp->v_type == VDIR) {
1913 error = SET_ERROR(EPERM);
1917 vnevent_remove(vp, dvp, name, ct);
1920 dnlc_remove(dvp, realnmp->pn_buf);
1922 dnlc_remove(dvp, name);
1925 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1929 * We may delete the znode now, or we may put it in the unlinked set;
1930 * it depends on whether we're the last link, and on whether there are
1931 * other holds on the vnode. So we dmu_tx_hold() the right things to
1932 * allow for either case.
1935 tx = dmu_tx_create(zfsvfs->z_os);
1936 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1937 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1938 zfs_sa_upgrade_txholds(tx, zp);
1939 zfs_sa_upgrade_txholds(tx, dzp);
1940 if (may_delete_now) {
1942 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1943 /* if the file is too big, only hold_free a token amount */
1944 dmu_tx_hold_free(tx, zp->z_id, 0,
1945 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1948 /* are there any extended attributes? */
1949 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1950 &xattr_obj, sizeof (xattr_obj));
1951 if (error == 0 && xattr_obj) {
1952 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1954 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1955 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1958 mutex_enter(&zp->z_lock);
1959 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1960 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1961 mutex_exit(&zp->z_lock);
1963 /* charge as an update -- would be nice not to charge at all */
1964 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1966 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1968 zfs_dirent_unlock(dl);
1972 if (error == ERESTART) {
1986 * Remove the directory entry.
1988 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1998 * Hold z_lock so that we can make sure that the ACL obj
1999 * hasn't changed. Could have been deleted due to
2002 mutex_enter(&zp->z_lock);
2004 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
2005 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
2006 delete_now = may_delete_now && !toobig &&
2007 vp->v_count == 1 && !vn_has_cached_data(vp) &&
2008 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
2015 panic("zfs_remove: delete_now branch taken");
2017 if (xattr_obj_unlinked) {
2018 ASSERT3U(xzp->z_links, ==, 2);
2019 mutex_enter(&xzp->z_lock);
2020 xzp->z_unlinked = 1;
2022 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
2023 &xzp->z_links, sizeof (xzp->z_links), tx);
2024 ASSERT3U(error, ==, 0);
2025 mutex_exit(&xzp->z_lock);
2026 zfs_unlinked_add(xzp, tx);
2029 error = sa_remove(zp->z_sa_hdl,
2030 SA_ZPL_XATTR(zfsvfs), tx);
2032 error = sa_update(zp->z_sa_hdl,
2033 SA_ZPL_XATTR(zfsvfs), &null_xattr,
2034 sizeof (uint64_t), tx);
2039 ASSERT0(vp->v_count);
2041 mutex_exit(&zp->z_lock);
2042 zfs_znode_delete(zp, tx);
2043 } else if (unlinked) {
2044 mutex_exit(&zp->z_lock);
2045 zfs_unlinked_add(zp, tx);
2047 vp->v_vflag |= VV_NOSYNC;
2052 if (flags & FIGNORECASE)
2054 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
2061 zfs_dirent_unlock(dl);
2068 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2069 zil_commit(zilog, 0);
2076 * Create a new directory and insert it into dvp using the name
2077 * provided. Return a pointer to the inserted directory.
2079 * IN: dvp - vnode of directory to add subdir to.
2080 * dirname - name of new directory.
2081 * vap - attributes of new directory.
2082 * cr - credentials of caller.
2083 * ct - caller context
2084 * flags - case flags
2085 * vsecp - ACL to be set
2087 * OUT: vpp - vnode of created directory.
2089 * RETURN: 0 on success, error code on failure.
2092 * dvp - ctime|mtime updated
2093 * vp - ctime|mtime|atime updated
2097 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
2098 caller_context_t *ct, int flags, vsecattr_t *vsecp)
2100 znode_t *zp, *dzp = VTOZ(dvp);
2101 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2110 gid_t gid = crgetgid(cr);
2111 zfs_acl_ids_t acl_ids;
2112 boolean_t fuid_dirtied;
2113 boolean_t waited = B_FALSE;
2115 ASSERT(vap->va_type == VDIR);
2118 * If we have an ephemeral id, ACL, or XVATTR then
2119 * make sure file system is at proper version
2122 ksid = crgetsid(cr, KSID_OWNER);
2124 uid = ksid_getid(ksid);
2127 if (zfsvfs->z_use_fuids == B_FALSE &&
2128 (vsecp || (vap->va_mask & AT_XVATTR) ||
2129 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
2130 return (SET_ERROR(EINVAL));
2134 zilog = zfsvfs->z_log;
2136 if (dzp->z_pflags & ZFS_XATTR) {
2138 return (SET_ERROR(EINVAL));
2141 if (zfsvfs->z_utf8 && u8_validate(dirname,
2142 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2144 return (SET_ERROR(EILSEQ));
2146 if (flags & FIGNORECASE)
2149 if (vap->va_mask & AT_XVATTR) {
2150 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
2151 crgetuid(cr), cr, vap->va_type)) != 0) {
2157 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
2158 vsecp, &acl_ids)) != 0) {
2163 * First make sure the new directory doesn't exist.
2165 * Existence is checked first to make sure we don't return
2166 * EACCES instead of EEXIST which can cause some applications
2172 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
2174 zfs_acl_ids_free(&acl_ids);
2179 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
2180 zfs_acl_ids_free(&acl_ids);
2181 zfs_dirent_unlock(dl);
2186 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
2187 zfs_acl_ids_free(&acl_ids);
2188 zfs_dirent_unlock(dl);
2190 return (SET_ERROR(EDQUOT));
2194 * Add a new entry to the directory.
2196 tx = dmu_tx_create(zfsvfs->z_os);
2197 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2198 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2199 fuid_dirtied = zfsvfs->z_fuid_dirty;
2201 zfs_fuid_txhold(zfsvfs, tx);
2202 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2203 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2204 acl_ids.z_aclp->z_acl_bytes);
2207 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2208 ZFS_SA_BASE_ATTR_SIZE);
2210 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2212 zfs_dirent_unlock(dl);
2213 if (error == ERESTART) {
2219 zfs_acl_ids_free(&acl_ids);
2228 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2231 zfs_fuid_sync(zfsvfs, tx);
2234 * Now put new name in parent dir.
2236 (void) zfs_link_create(dl, zp, tx, ZNEW);
2240 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2241 if (flags & FIGNORECASE)
2243 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2244 acl_ids.z_fuidp, vap);
2246 zfs_acl_ids_free(&acl_ids);
2250 zfs_dirent_unlock(dl);
2252 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2253 zil_commit(zilog, 0);
2260 * Remove a directory subdir entry. If the current working
2261 * directory is the same as the subdir to be removed, the
2264 * IN: dvp - vnode of directory to remove from.
2265 * name - name of directory to be removed.
2266 * cwd - vnode of current working directory.
2267 * cr - credentials of caller.
2268 * ct - caller context
2269 * flags - case flags
2271 * RETURN: 0 on success, error code on failure.
2274 * dvp - ctime|mtime updated
2278 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2279 caller_context_t *ct, int flags)
2281 znode_t *dzp = VTOZ(dvp);
2284 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2290 boolean_t waited = B_FALSE;
2294 zilog = zfsvfs->z_log;
2296 if (flags & FIGNORECASE)
2302 * Attempt to lock directory; fail if entry doesn't exist.
2304 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2312 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2316 if (vp->v_type != VDIR) {
2317 error = SET_ERROR(ENOTDIR);
2322 error = SET_ERROR(EINVAL);
2326 vnevent_rmdir(vp, dvp, name, ct);
2329 * Grab a lock on the directory to make sure that noone is
2330 * trying to add (or lookup) entries while we are removing it.
2332 rw_enter(&zp->z_name_lock, RW_WRITER);
2335 * Grab a lock on the parent pointer to make sure we play well
2336 * with the treewalk and directory rename code.
2338 rw_enter(&zp->z_parent_lock, RW_WRITER);
2340 tx = dmu_tx_create(zfsvfs->z_os);
2341 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2342 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2343 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2344 zfs_sa_upgrade_txholds(tx, zp);
2345 zfs_sa_upgrade_txholds(tx, dzp);
2346 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2348 rw_exit(&zp->z_parent_lock);
2349 rw_exit(&zp->z_name_lock);
2350 zfs_dirent_unlock(dl);
2352 if (error == ERESTART) {
2363 #ifdef FREEBSD_NAMECACHE
2367 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2370 uint64_t txtype = TX_RMDIR;
2371 if (flags & FIGNORECASE)
2373 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2378 rw_exit(&zp->z_parent_lock);
2379 rw_exit(&zp->z_name_lock);
2380 #ifdef FREEBSD_NAMECACHE
2384 zfs_dirent_unlock(dl);
2388 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2389 zil_commit(zilog, 0);
2396 * Read as many directory entries as will fit into the provided
2397 * buffer from the given directory cursor position (specified in
2398 * the uio structure).
2400 * IN: vp - vnode of directory to read.
2401 * uio - structure supplying read location, range info,
2402 * and return buffer.
2403 * cr - credentials of caller.
2404 * ct - caller context
2405 * flags - case flags
2407 * OUT: uio - updated offset and range, buffer filled.
2408 * eofp - set to true if end-of-file detected.
2410 * RETURN: 0 on success, error code on failure.
2413 * vp - atime updated
2415 * Note that the low 4 bits of the cookie returned by zap is always zero.
2416 * This allows us to use the low range for "special" directory entries:
2417 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2418 * we use the offset 2 for the '.zfs' directory.
2422 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies)
2424 znode_t *zp = VTOZ(vp);
2428 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2433 zap_attribute_t zap;
2434 uint_t bytes_wanted;
2435 uint64_t offset; /* must be unsigned; checks for < 1 */
2441 boolean_t check_sysattrs;
2444 u_long *cooks = NULL;
2450 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2451 &parent, sizeof (parent))) != 0) {
2457 * If we are not given an eof variable,
2464 * Check for valid iov_len.
2466 if (uio->uio_iov->iov_len <= 0) {
2468 return (SET_ERROR(EINVAL));
2472 * Quit if directory has been removed (posix)
2474 if ((*eofp = zp->z_unlinked) != 0) {
2481 offset = uio->uio_loffset;
2482 prefetch = zp->z_zn_prefetch;
2485 * Initialize the iterator cursor.
2489 * Start iteration from the beginning of the directory.
2491 zap_cursor_init(&zc, os, zp->z_id);
2494 * The offset is a serialized cursor.
2496 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2500 * Get space to change directory entries into fs independent format.
2502 iovp = uio->uio_iov;
2503 bytes_wanted = iovp->iov_len;
2504 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2505 bufsize = bytes_wanted;
2506 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2507 odp = (struct dirent64 *)outbuf;
2509 bufsize = bytes_wanted;
2511 odp = (struct dirent64 *)iovp->iov_base;
2513 eodp = (struct edirent *)odp;
2515 if (ncookies != NULL) {
2517 * Minimum entry size is dirent size and 1 byte for a file name.
2519 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1);
2520 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK);
2525 * If this VFS supports the system attribute view interface; and
2526 * we're looking at an extended attribute directory; and we care
2527 * about normalization conflicts on this vfs; then we must check
2528 * for normalization conflicts with the sysattr name space.
2531 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2532 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2533 (flags & V_RDDIR_ENTFLAGS);
2539 * Transform to file-system independent format
2542 while (outcount < bytes_wanted) {
2545 off64_t *next = NULL;
2548 * Special case `.', `..', and `.zfs'.
2551 (void) strcpy(zap.za_name, ".");
2552 zap.za_normalization_conflict = 0;
2555 } else if (offset == 1) {
2556 (void) strcpy(zap.za_name, "..");
2557 zap.za_normalization_conflict = 0;
2560 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2561 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2562 zap.za_normalization_conflict = 0;
2563 objnum = ZFSCTL_INO_ROOT;
2569 if (error = zap_cursor_retrieve(&zc, &zap)) {
2570 if ((*eofp = (error == ENOENT)) != 0)
2576 if (zap.za_integer_length != 8 ||
2577 zap.za_num_integers != 1) {
2578 cmn_err(CE_WARN, "zap_readdir: bad directory "
2579 "entry, obj = %lld, offset = %lld\n",
2580 (u_longlong_t)zp->z_id,
2581 (u_longlong_t)offset);
2582 error = SET_ERROR(ENXIO);
2586 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2588 * MacOS X can extract the object type here such as:
2589 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2591 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2593 if (check_sysattrs && !zap.za_normalization_conflict) {
2595 zap.za_normalization_conflict =
2596 xattr_sysattr_casechk(zap.za_name);
2598 panic("%s:%u: TODO", __func__, __LINE__);
2603 if (flags & V_RDDIR_ACCFILTER) {
2605 * If we have no access at all, don't include
2606 * this entry in the returned information
2609 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2611 if (!zfs_has_access(ezp, cr)) {
2618 if (flags & V_RDDIR_ENTFLAGS)
2619 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2621 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2624 * Will this entry fit in the buffer?
2626 if (outcount + reclen > bufsize) {
2628 * Did we manage to fit anything in the buffer?
2631 error = SET_ERROR(EINVAL);
2636 if (flags & V_RDDIR_ENTFLAGS) {
2638 * Add extended flag entry:
2640 eodp->ed_ino = objnum;
2641 eodp->ed_reclen = reclen;
2642 /* NOTE: ed_off is the offset for the *next* entry */
2643 next = &(eodp->ed_off);
2644 eodp->ed_eflags = zap.za_normalization_conflict ?
2645 ED_CASE_CONFLICT : 0;
2646 (void) strncpy(eodp->ed_name, zap.za_name,
2647 EDIRENT_NAMELEN(reclen));
2648 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2653 odp->d_ino = objnum;
2654 odp->d_reclen = reclen;
2655 odp->d_namlen = strlen(zap.za_name);
2656 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1);
2658 odp = (dirent64_t *)((intptr_t)odp + reclen);
2662 ASSERT(outcount <= bufsize);
2664 /* Prefetch znode */
2666 dmu_prefetch(os, objnum, 0, 0);
2670 * Move to the next entry, fill in the previous offset.
2672 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2673 zap_cursor_advance(&zc);
2674 offset = zap_cursor_serialize(&zc);
2679 if (cooks != NULL) {
2682 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks));
2685 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2687 /* Subtract unused cookies */
2688 if (ncookies != NULL)
2689 *ncookies -= ncooks;
2691 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2692 iovp->iov_base += outcount;
2693 iovp->iov_len -= outcount;
2694 uio->uio_resid -= outcount;
2695 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2697 * Reset the pointer.
2699 offset = uio->uio_loffset;
2703 zap_cursor_fini(&zc);
2704 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2705 kmem_free(outbuf, bufsize);
2707 if (error == ENOENT)
2710 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2712 uio->uio_loffset = offset;
2714 if (error != 0 && cookies != NULL) {
2715 free(*cookies, M_TEMP);
2722 ulong_t zfs_fsync_sync_cnt = 4;
2725 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2727 znode_t *zp = VTOZ(vp);
2728 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2730 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2732 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2735 zil_commit(zfsvfs->z_log, zp->z_id);
2743 * Get the requested file attributes and place them in the provided
2746 * IN: vp - vnode of file.
2747 * vap - va_mask identifies requested attributes.
2748 * If AT_XVATTR set, then optional attrs are requested
2749 * flags - ATTR_NOACLCHECK (CIFS server context)
2750 * cr - credentials of caller.
2751 * ct - caller context
2753 * OUT: vap - attribute values.
2755 * RETURN: 0 (always succeeds).
2759 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2760 caller_context_t *ct)
2762 znode_t *zp = VTOZ(vp);
2763 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2766 u_longlong_t nblocks;
2768 uint64_t mtime[2], ctime[2], crtime[2], rdev;
2769 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2770 xoptattr_t *xoap = NULL;
2771 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2772 sa_bulk_attr_t bulk[4];
2778 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2780 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2781 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2782 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
2783 if (vp->v_type == VBLK || vp->v_type == VCHR)
2784 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
2787 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2793 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2794 * Also, if we are the owner don't bother, since owner should
2795 * always be allowed to read basic attributes of file.
2797 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2798 (vap->va_uid != crgetuid(cr))) {
2799 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2807 * Return all attributes. It's cheaper to provide the answer
2808 * than to determine whether we were asked the question.
2811 mutex_enter(&zp->z_lock);
2812 vap->va_type = IFTOVT(zp->z_mode);
2813 vap->va_mode = zp->z_mode & ~S_IFMT;
2815 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2817 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
2819 vap->va_nodeid = zp->z_id;
2820 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2821 links = zp->z_links + 1;
2823 links = zp->z_links;
2824 vap->va_nlink = MIN(links, LINK_MAX); /* nlink_t limit! */
2825 vap->va_size = zp->z_size;
2827 vap->va_rdev = vp->v_rdev;
2829 if (vp->v_type == VBLK || vp->v_type == VCHR)
2830 vap->va_rdev = zfs_cmpldev(rdev);
2832 vap->va_seq = zp->z_seq;
2833 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */
2836 * Add in any requested optional attributes and the create time.
2837 * Also set the corresponding bits in the returned attribute bitmap.
2839 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2840 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2842 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2843 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2846 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2847 xoap->xoa_readonly =
2848 ((zp->z_pflags & ZFS_READONLY) != 0);
2849 XVA_SET_RTN(xvap, XAT_READONLY);
2852 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2854 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2855 XVA_SET_RTN(xvap, XAT_SYSTEM);
2858 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2860 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2861 XVA_SET_RTN(xvap, XAT_HIDDEN);
2864 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2865 xoap->xoa_nounlink =
2866 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2867 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2870 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2871 xoap->xoa_immutable =
2872 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2873 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2876 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2877 xoap->xoa_appendonly =
2878 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2879 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2882 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2884 ((zp->z_pflags & ZFS_NODUMP) != 0);
2885 XVA_SET_RTN(xvap, XAT_NODUMP);
2888 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2890 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2891 XVA_SET_RTN(xvap, XAT_OPAQUE);
2894 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2895 xoap->xoa_av_quarantined =
2896 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2897 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2900 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2901 xoap->xoa_av_modified =
2902 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2903 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2906 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2907 vp->v_type == VREG) {
2908 zfs_sa_get_scanstamp(zp, xvap);
2911 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2914 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2915 times, sizeof (times));
2916 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2917 XVA_SET_RTN(xvap, XAT_CREATETIME);
2920 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2921 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2922 XVA_SET_RTN(xvap, XAT_REPARSE);
2924 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2925 xoap->xoa_generation = zp->z_gen;
2926 XVA_SET_RTN(xvap, XAT_GEN);
2929 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2931 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2932 XVA_SET_RTN(xvap, XAT_OFFLINE);
2935 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2937 ((zp->z_pflags & ZFS_SPARSE) != 0);
2938 XVA_SET_RTN(xvap, XAT_SPARSE);
2942 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2943 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2944 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2945 ZFS_TIME_DECODE(&vap->va_birthtime, crtime);
2947 mutex_exit(&zp->z_lock);
2949 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2950 vap->va_blksize = blksize;
2951 vap->va_bytes = nblocks << 9; /* nblocks * 512 */
2953 if (zp->z_blksz == 0) {
2955 * Block size hasn't been set; suggest maximal I/O transfers.
2957 vap->va_blksize = zfsvfs->z_max_blksz;
2965 * Set the file attributes to the values contained in the
2968 * IN: vp - vnode of file to be modified.
2969 * vap - new attribute values.
2970 * If AT_XVATTR set, then optional attrs are being set
2971 * flags - ATTR_UTIME set if non-default time values provided.
2972 * - ATTR_NOACLCHECK (CIFS context only).
2973 * cr - credentials of caller.
2974 * ct - caller context
2976 * RETURN: 0 on success, error code on failure.
2979 * vp - ctime updated, mtime updated if size changed.
2983 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2984 caller_context_t *ct)
2986 znode_t *zp = VTOZ(vp);
2987 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2992 uint_t mask = vap->va_mask;
2993 uint_t saved_mask = 0;
2994 uint64_t saved_mode;
2997 uint64_t new_uid, new_gid;
2999 uint64_t mtime[2], ctime[2];
3001 int need_policy = FALSE;
3003 zfs_fuid_info_t *fuidp = NULL;
3004 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
3007 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
3008 boolean_t fuid_dirtied = B_FALSE;
3009 sa_bulk_attr_t bulk[7], xattr_bulk[7];
3010 int count = 0, xattr_count = 0;
3015 if (mask & AT_NOSET)
3016 return (SET_ERROR(EINVAL));
3021 zilog = zfsvfs->z_log;
3024 * Make sure that if we have ephemeral uid/gid or xvattr specified
3025 * that file system is at proper version level
3028 if (zfsvfs->z_use_fuids == B_FALSE &&
3029 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
3030 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
3031 (mask & AT_XVATTR))) {
3033 return (SET_ERROR(EINVAL));
3036 if (mask & AT_SIZE && vp->v_type == VDIR) {
3038 return (SET_ERROR(EISDIR));
3041 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
3043 return (SET_ERROR(EINVAL));
3047 * If this is an xvattr_t, then get a pointer to the structure of
3048 * optional attributes. If this is NULL, then we have a vattr_t.
3050 xoap = xva_getxoptattr(xvap);
3052 xva_init(&tmpxvattr);
3055 * Immutable files can only alter immutable bit and atime
3057 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
3058 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
3059 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
3061 return (SET_ERROR(EPERM));
3064 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
3066 return (SET_ERROR(EPERM));
3070 * Verify timestamps doesn't overflow 32 bits.
3071 * ZFS can handle large timestamps, but 32bit syscalls can't
3072 * handle times greater than 2039. This check should be removed
3073 * once large timestamps are fully supported.
3075 if (mask & (AT_ATIME | AT_MTIME)) {
3076 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
3077 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
3079 return (SET_ERROR(EOVERFLOW));
3087 /* Can this be moved to before the top label? */
3088 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
3090 return (SET_ERROR(EROFS));
3094 * First validate permissions
3097 if (mask & AT_SIZE) {
3099 * XXX - Note, we are not providing any open
3100 * mode flags here (like FNDELAY), so we may
3101 * block if there are locks present... this
3102 * should be addressed in openat().
3104 /* XXX - would it be OK to generate a log record here? */
3105 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
3112 if (mask & (AT_ATIME|AT_MTIME) ||
3113 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
3114 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
3115 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
3116 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
3117 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
3118 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
3119 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
3120 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
3124 if (mask & (AT_UID|AT_GID)) {
3125 int idmask = (mask & (AT_UID|AT_GID));
3130 * NOTE: even if a new mode is being set,
3131 * we may clear S_ISUID/S_ISGID bits.
3134 if (!(mask & AT_MODE))
3135 vap->va_mode = zp->z_mode;
3138 * Take ownership or chgrp to group we are a member of
3141 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
3142 take_group = (mask & AT_GID) &&
3143 zfs_groupmember(zfsvfs, vap->va_gid, cr);
3146 * If both AT_UID and AT_GID are set then take_owner and
3147 * take_group must both be set in order to allow taking
3150 * Otherwise, send the check through secpolicy_vnode_setattr()
3154 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
3155 ((idmask == AT_UID) && take_owner) ||
3156 ((idmask == AT_GID) && take_group)) {
3157 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
3158 skipaclchk, cr) == 0) {
3160 * Remove setuid/setgid for non-privileged users
3162 secpolicy_setid_clear(vap, vp, cr);
3163 trim_mask = (mask & (AT_UID|AT_GID));
3172 mutex_enter(&zp->z_lock);
3173 oldva.va_mode = zp->z_mode;
3174 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
3175 if (mask & AT_XVATTR) {
3177 * Update xvattr mask to include only those attributes
3178 * that are actually changing.
3180 * the bits will be restored prior to actually setting
3181 * the attributes so the caller thinks they were set.
3183 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
3184 if (xoap->xoa_appendonly !=
3185 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
3188 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
3189 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
3193 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
3194 if (xoap->xoa_nounlink !=
3195 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
3198 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
3199 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
3203 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
3204 if (xoap->xoa_immutable !=
3205 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
3208 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
3209 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
3213 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
3214 if (xoap->xoa_nodump !=
3215 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
3218 XVA_CLR_REQ(xvap, XAT_NODUMP);
3219 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
3223 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3224 if (xoap->xoa_av_modified !=
3225 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3228 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3229 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
3233 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3234 if ((vp->v_type != VREG &&
3235 xoap->xoa_av_quarantined) ||
3236 xoap->xoa_av_quarantined !=
3237 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3240 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3241 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
3245 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3246 mutex_exit(&zp->z_lock);
3248 return (SET_ERROR(EPERM));
3251 if (need_policy == FALSE &&
3252 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3253 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3258 mutex_exit(&zp->z_lock);
3260 if (mask & AT_MODE) {
3261 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3262 err = secpolicy_setid_setsticky_clear(vp, vap,
3268 trim_mask |= AT_MODE;
3276 * If trim_mask is set then take ownership
3277 * has been granted or write_acl is present and user
3278 * has the ability to modify mode. In that case remove
3279 * UID|GID and or MODE from mask so that
3280 * secpolicy_vnode_setattr() doesn't revoke it.
3284 saved_mask = vap->va_mask;
3285 vap->va_mask &= ~trim_mask;
3286 if (trim_mask & AT_MODE) {
3288 * Save the mode, as secpolicy_vnode_setattr()
3289 * will overwrite it with ova.va_mode.
3291 saved_mode = vap->va_mode;
3294 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3295 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3302 vap->va_mask |= saved_mask;
3303 if (trim_mask & AT_MODE) {
3305 * Recover the mode after
3306 * secpolicy_vnode_setattr().
3308 vap->va_mode = saved_mode;
3314 * secpolicy_vnode_setattr, or take ownership may have
3317 mask = vap->va_mask;
3319 if ((mask & (AT_UID | AT_GID))) {
3320 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3321 &xattr_obj, sizeof (xattr_obj));
3323 if (err == 0 && xattr_obj) {
3324 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3328 if (mask & AT_UID) {
3329 new_uid = zfs_fuid_create(zfsvfs,
3330 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3331 if (new_uid != zp->z_uid &&
3332 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3334 VN_RELE(ZTOV(attrzp));
3335 err = SET_ERROR(EDQUOT);
3340 if (mask & AT_GID) {
3341 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3342 cr, ZFS_GROUP, &fuidp);
3343 if (new_gid != zp->z_gid &&
3344 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3346 VN_RELE(ZTOV(attrzp));
3347 err = SET_ERROR(EDQUOT);
3352 tx = dmu_tx_create(zfsvfs->z_os);
3354 if (mask & AT_MODE) {
3355 uint64_t pmode = zp->z_mode;
3357 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3359 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3360 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3361 err = SET_ERROR(EPERM);
3365 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3368 mutex_enter(&zp->z_lock);
3369 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3371 * Are we upgrading ACL from old V0 format
3374 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3375 zfs_znode_acl_version(zp) ==
3376 ZFS_ACL_VERSION_INITIAL) {
3377 dmu_tx_hold_free(tx, acl_obj, 0,
3379 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3380 0, aclp->z_acl_bytes);
3382 dmu_tx_hold_write(tx, acl_obj, 0,
3385 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3386 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3387 0, aclp->z_acl_bytes);
3389 mutex_exit(&zp->z_lock);
3390 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3392 if ((mask & AT_XVATTR) &&
3393 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3394 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3396 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3400 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3403 fuid_dirtied = zfsvfs->z_fuid_dirty;
3405 zfs_fuid_txhold(zfsvfs, tx);
3407 zfs_sa_upgrade_txholds(tx, zp);
3409 err = dmu_tx_assign(tx, TXG_NOWAIT);
3411 if (err == ERESTART)
3418 * Set each attribute requested.
3419 * We group settings according to the locks they need to acquire.
3421 * Note: you cannot set ctime directly, although it will be
3422 * updated as a side-effect of calling this function.
3426 if (mask & (AT_UID|AT_GID|AT_MODE))
3427 mutex_enter(&zp->z_acl_lock);
3428 mutex_enter(&zp->z_lock);
3430 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3431 &zp->z_pflags, sizeof (zp->z_pflags));
3434 if (mask & (AT_UID|AT_GID|AT_MODE))
3435 mutex_enter(&attrzp->z_acl_lock);
3436 mutex_enter(&attrzp->z_lock);
3437 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3438 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3439 sizeof (attrzp->z_pflags));
3442 if (mask & (AT_UID|AT_GID)) {
3444 if (mask & AT_UID) {
3445 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3446 &new_uid, sizeof (new_uid));
3447 zp->z_uid = new_uid;
3449 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3450 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3452 attrzp->z_uid = new_uid;
3456 if (mask & AT_GID) {
3457 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3458 NULL, &new_gid, sizeof (new_gid));
3459 zp->z_gid = new_gid;
3461 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3462 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3464 attrzp->z_gid = new_gid;
3467 if (!(mask & AT_MODE)) {
3468 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3469 NULL, &new_mode, sizeof (new_mode));
3470 new_mode = zp->z_mode;
3472 err = zfs_acl_chown_setattr(zp);
3475 err = zfs_acl_chown_setattr(attrzp);
3480 if (mask & AT_MODE) {
3481 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3482 &new_mode, sizeof (new_mode));
3483 zp->z_mode = new_mode;
3484 ASSERT3U((uintptr_t)aclp, !=, 0);
3485 err = zfs_aclset_common(zp, aclp, cr, tx);
3487 if (zp->z_acl_cached)
3488 zfs_acl_free(zp->z_acl_cached);
3489 zp->z_acl_cached = aclp;
3494 if (mask & AT_ATIME) {
3495 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3496 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3497 &zp->z_atime, sizeof (zp->z_atime));
3500 if (mask & AT_MTIME) {
3501 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3502 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3503 mtime, sizeof (mtime));
3506 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3507 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3508 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3509 NULL, mtime, sizeof (mtime));
3510 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3511 &ctime, sizeof (ctime));
3512 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3514 } else if (mask != 0) {
3515 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3516 &ctime, sizeof (ctime));
3517 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3520 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3521 SA_ZPL_CTIME(zfsvfs), NULL,
3522 &ctime, sizeof (ctime));
3523 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3524 mtime, ctime, B_TRUE);
3528 * Do this after setting timestamps to prevent timestamp
3529 * update from toggling bit
3532 if (xoap && (mask & AT_XVATTR)) {
3535 * restore trimmed off masks
3536 * so that return masks can be set for caller.
3539 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3540 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3542 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3543 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3545 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3546 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3548 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3549 XVA_SET_REQ(xvap, XAT_NODUMP);
3551 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3552 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3554 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3555 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3558 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3559 ASSERT(vp->v_type == VREG);
3561 zfs_xvattr_set(zp, xvap, tx);
3565 zfs_fuid_sync(zfsvfs, tx);
3568 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3570 mutex_exit(&zp->z_lock);
3571 if (mask & (AT_UID|AT_GID|AT_MODE))
3572 mutex_exit(&zp->z_acl_lock);
3575 if (mask & (AT_UID|AT_GID|AT_MODE))
3576 mutex_exit(&attrzp->z_acl_lock);
3577 mutex_exit(&attrzp->z_lock);
3580 if (err == 0 && attrzp) {
3581 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3587 VN_RELE(ZTOV(attrzp));
3593 zfs_fuid_info_free(fuidp);
3599 if (err == ERESTART)
3602 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3607 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3608 zil_commit(zilog, 0);
3614 typedef struct zfs_zlock {
3615 krwlock_t *zl_rwlock; /* lock we acquired */
3616 znode_t *zl_znode; /* znode we held */
3617 struct zfs_zlock *zl_next; /* next in list */
3621 * Drop locks and release vnodes that were held by zfs_rename_lock().
3624 zfs_rename_unlock(zfs_zlock_t **zlpp)
3628 while ((zl = *zlpp) != NULL) {
3629 if (zl->zl_znode != NULL)
3630 VN_RELE(ZTOV(zl->zl_znode));
3631 rw_exit(zl->zl_rwlock);
3632 *zlpp = zl->zl_next;
3633 kmem_free(zl, sizeof (*zl));
3638 * Search back through the directory tree, using the ".." entries.
3639 * Lock each directory in the chain to prevent concurrent renames.
3640 * Fail any attempt to move a directory into one of its own descendants.
3641 * XXX - z_parent_lock can overlap with map or grow locks
3644 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3648 uint64_t rootid = zp->z_zfsvfs->z_root;
3649 uint64_t oidp = zp->z_id;
3650 krwlock_t *rwlp = &szp->z_parent_lock;
3651 krw_t rw = RW_WRITER;
3654 * First pass write-locks szp and compares to zp->z_id.
3655 * Later passes read-lock zp and compare to zp->z_parent.
3658 if (!rw_tryenter(rwlp, rw)) {
3660 * Another thread is renaming in this path.
3661 * Note that if we are a WRITER, we don't have any
3662 * parent_locks held yet.
3664 if (rw == RW_READER && zp->z_id > szp->z_id) {
3666 * Drop our locks and restart
3668 zfs_rename_unlock(&zl);
3672 rwlp = &szp->z_parent_lock;
3677 * Wait for other thread to drop its locks
3683 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3684 zl->zl_rwlock = rwlp;
3685 zl->zl_znode = NULL;
3686 zl->zl_next = *zlpp;
3689 if (oidp == szp->z_id) /* We're a descendant of szp */
3690 return (SET_ERROR(EINVAL));
3692 if (oidp == rootid) /* We've hit the top */
3695 if (rw == RW_READER) { /* i.e. not the first pass */
3696 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3701 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3702 &oidp, sizeof (oidp));
3703 rwlp = &zp->z_parent_lock;
3706 } while (zp->z_id != sdzp->z_id);
3712 * Move an entry from the provided source directory to the target
3713 * directory. Change the entry name as indicated.
3715 * IN: sdvp - Source directory containing the "old entry".
3716 * snm - Old entry name.
3717 * tdvp - Target directory to contain the "new entry".
3718 * tnm - New entry name.
3719 * cr - credentials of caller.
3720 * ct - caller context
3721 * flags - case flags
3723 * RETURN: 0 on success, error code on failure.
3726 * sdvp,tdvp - ctime|mtime updated
3730 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3731 caller_context_t *ct, int flags)
3733 znode_t *tdzp, *szp, *tzp;
3734 znode_t *sdzp = VTOZ(sdvp);
3735 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs;
3738 zfs_dirlock_t *sdl, *tdl;
3741 int cmp, serr, terr;
3744 boolean_t waited = B_FALSE;
3747 ZFS_VERIFY_ZP(sdzp);
3748 zilog = zfsvfs->z_log;
3751 * Make sure we have the real vp for the target directory.
3753 if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3756 if (tdvp->v_vfsp != sdvp->v_vfsp || zfsctl_is_node(tdvp)) {
3758 return (SET_ERROR(EXDEV));
3762 ZFS_VERIFY_ZP(tdzp);
3763 if (zfsvfs->z_utf8 && u8_validate(tnm,
3764 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3766 return (SET_ERROR(EILSEQ));
3769 if (flags & FIGNORECASE)
3778 * This is to prevent the creation of links into attribute space
3779 * by renaming a linked file into/outof an attribute directory.
3780 * See the comment in zfs_link() for why this is considered bad.
3782 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3784 return (SET_ERROR(EINVAL));
3788 * Lock source and target directory entries. To prevent deadlock,
3789 * a lock ordering must be defined. We lock the directory with
3790 * the smallest object id first, or if it's a tie, the one with
3791 * the lexically first name.
3793 if (sdzp->z_id < tdzp->z_id) {
3795 } else if (sdzp->z_id > tdzp->z_id) {
3799 * First compare the two name arguments without
3800 * considering any case folding.
3802 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3804 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3805 ASSERT(error == 0 || !zfsvfs->z_utf8);
3808 * POSIX: "If the old argument and the new argument
3809 * both refer to links to the same existing file,
3810 * the rename() function shall return successfully
3811 * and perform no other action."
3817 * If the file system is case-folding, then we may
3818 * have some more checking to do. A case-folding file
3819 * system is either supporting mixed case sensitivity
3820 * access or is completely case-insensitive. Note
3821 * that the file system is always case preserving.
3823 * In mixed sensitivity mode case sensitive behavior
3824 * is the default. FIGNORECASE must be used to
3825 * explicitly request case insensitive behavior.
3827 * If the source and target names provided differ only
3828 * by case (e.g., a request to rename 'tim' to 'Tim'),
3829 * we will treat this as a special case in the
3830 * case-insensitive mode: as long as the source name
3831 * is an exact match, we will allow this to proceed as
3832 * a name-change request.
3834 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3835 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3836 flags & FIGNORECASE)) &&
3837 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3840 * case preserving rename request, require exact
3849 * If the source and destination directories are the same, we should
3850 * grab the z_name_lock of that directory only once.
3854 rw_enter(&sdzp->z_name_lock, RW_READER);
3858 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3859 ZEXISTS | zflg, NULL, NULL);
3860 terr = zfs_dirent_lock(&tdl,
3861 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3863 terr = zfs_dirent_lock(&tdl,
3864 tdzp, tnm, &tzp, zflg, NULL, NULL);
3865 serr = zfs_dirent_lock(&sdl,
3866 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3872 * Source entry invalid or not there.
3875 zfs_dirent_unlock(tdl);
3881 rw_exit(&sdzp->z_name_lock);
3884 * FreeBSD: In OpenSolaris they only check if rename source is
3885 * ".." here, because "." is handled in their lookup. This is
3886 * not the case for FreeBSD, so we check for "." explicitly.
3888 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0)
3889 serr = SET_ERROR(EINVAL);
3894 zfs_dirent_unlock(sdl);
3898 rw_exit(&sdzp->z_name_lock);
3900 if (strcmp(tnm, "..") == 0)
3901 terr = SET_ERROR(EINVAL);
3907 * Must have write access at the source to remove the old entry
3908 * and write access at the target to create the new entry.
3909 * Note that if target and source are the same, this can be
3910 * done in a single check.
3913 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3916 if (ZTOV(szp)->v_type == VDIR) {
3918 * Check to make sure rename is valid.
3919 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3921 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3926 * Does target exist?
3930 * Source and target must be the same type.
3932 if (ZTOV(szp)->v_type == VDIR) {
3933 if (ZTOV(tzp)->v_type != VDIR) {
3934 error = SET_ERROR(ENOTDIR);
3938 if (ZTOV(tzp)->v_type == VDIR) {
3939 error = SET_ERROR(EISDIR);
3944 * POSIX dictates that when the source and target
3945 * entries refer to the same file object, rename
3946 * must do nothing and exit without error.
3948 if (szp->z_id == tzp->z_id) {
3954 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3956 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3959 * notify the target directory if it is not the same
3960 * as source directory.
3963 vnevent_rename_dest_dir(tdvp, ct);
3966 tx = dmu_tx_create(zfsvfs->z_os);
3967 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3968 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3969 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3970 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3972 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3973 zfs_sa_upgrade_txholds(tx, tdzp);
3976 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3977 zfs_sa_upgrade_txholds(tx, tzp);
3980 zfs_sa_upgrade_txholds(tx, szp);
3981 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3982 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3985 zfs_rename_unlock(&zl);
3986 zfs_dirent_unlock(sdl);
3987 zfs_dirent_unlock(tdl);
3990 rw_exit(&sdzp->z_name_lock);
3995 if (error == ERESTART) {
4006 if (tzp) /* Attempt to remove the existing target */
4007 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
4010 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
4012 szp->z_pflags |= ZFS_AV_MODIFIED;
4014 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
4015 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
4018 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
4020 zfs_log_rename(zilog, tx, TX_RENAME |
4021 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
4022 sdl->dl_name, tdzp, tdl->dl_name, szp);
4025 * Update path information for the target vnode
4027 vn_renamepath(tdvp, ZTOV(szp), tnm,
4031 * At this point, we have successfully created
4032 * the target name, but have failed to remove
4033 * the source name. Since the create was done
4034 * with the ZRENAMING flag, there are
4035 * complications; for one, the link count is
4036 * wrong. The easiest way to deal with this
4037 * is to remove the newly created target, and
4038 * return the original error. This must
4039 * succeed; fortunately, it is very unlikely to
4040 * fail, since we just created it.
4042 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
4043 ZRENAMING, NULL), ==, 0);
4046 #ifdef FREEBSD_NAMECACHE
4057 zfs_rename_unlock(&zl);
4059 zfs_dirent_unlock(sdl);
4060 zfs_dirent_unlock(tdl);
4063 rw_exit(&sdzp->z_name_lock);
4070 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4071 zil_commit(zilog, 0);
4079 * Insert the indicated symbolic reference entry into the directory.
4081 * IN: dvp - Directory to contain new symbolic link.
4082 * link - Name for new symlink entry.
4083 * vap - Attributes of new entry.
4084 * cr - credentials of caller.
4085 * ct - caller context
4086 * flags - case flags
4088 * RETURN: 0 on success, error code on failure.
4091 * dvp - ctime|mtime updated
4095 zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link,
4096 cred_t *cr, kthread_t *td)
4098 znode_t *zp, *dzp = VTOZ(dvp);
4101 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4103 uint64_t len = strlen(link);
4106 zfs_acl_ids_t acl_ids;
4107 boolean_t fuid_dirtied;
4108 uint64_t txtype = TX_SYMLINK;
4109 boolean_t waited = B_FALSE;
4112 ASSERT(vap->va_type == VLNK);
4116 zilog = zfsvfs->z_log;
4118 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
4119 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4121 return (SET_ERROR(EILSEQ));
4123 if (flags & FIGNORECASE)
4126 if (len > MAXPATHLEN) {
4128 return (SET_ERROR(ENAMETOOLONG));
4131 if ((error = zfs_acl_ids_create(dzp, 0,
4132 vap, cr, NULL, &acl_ids)) != 0) {
4138 * Attempt to lock directory; fail if entry already exists.
4140 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
4142 zfs_acl_ids_free(&acl_ids);
4147 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4148 zfs_acl_ids_free(&acl_ids);
4149 zfs_dirent_unlock(dl);
4154 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
4155 zfs_acl_ids_free(&acl_ids);
4156 zfs_dirent_unlock(dl);
4158 return (SET_ERROR(EDQUOT));
4160 tx = dmu_tx_create(zfsvfs->z_os);
4161 fuid_dirtied = zfsvfs->z_fuid_dirty;
4162 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
4163 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4164 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
4165 ZFS_SA_BASE_ATTR_SIZE + len);
4166 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
4167 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
4168 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
4169 acl_ids.z_aclp->z_acl_bytes);
4172 zfs_fuid_txhold(zfsvfs, tx);
4173 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4175 zfs_dirent_unlock(dl);
4176 if (error == ERESTART) {
4182 zfs_acl_ids_free(&acl_ids);
4189 * Create a new object for the symlink.
4190 * for version 4 ZPL datsets the symlink will be an SA attribute
4192 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
4195 zfs_fuid_sync(zfsvfs, tx);
4197 mutex_enter(&zp->z_lock);
4199 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
4202 zfs_sa_symlink(zp, link, len, tx);
4203 mutex_exit(&zp->z_lock);
4206 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
4207 &zp->z_size, sizeof (zp->z_size), tx);
4209 * Insert the new object into the directory.
4211 (void) zfs_link_create(dl, zp, tx, ZNEW);
4213 if (flags & FIGNORECASE)
4215 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
4218 zfs_acl_ids_free(&acl_ids);
4222 zfs_dirent_unlock(dl);
4224 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4225 zil_commit(zilog, 0);
4232 * Return, in the buffer contained in the provided uio structure,
4233 * the symbolic path referred to by vp.
4235 * IN: vp - vnode of symbolic link.
4236 * uio - structure to contain the link path.
4237 * cr - credentials of caller.
4238 * ct - caller context
4240 * OUT: uio - structure containing the link path.
4242 * RETURN: 0 on success, error code on failure.
4245 * vp - atime updated
4249 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
4251 znode_t *zp = VTOZ(vp);
4252 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4258 mutex_enter(&zp->z_lock);
4260 error = sa_lookup_uio(zp->z_sa_hdl,
4261 SA_ZPL_SYMLINK(zfsvfs), uio);
4263 error = zfs_sa_readlink(zp, uio);
4264 mutex_exit(&zp->z_lock);
4266 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4273 * Insert a new entry into directory tdvp referencing svp.
4275 * IN: tdvp - Directory to contain new entry.
4276 * svp - vnode of new entry.
4277 * name - name of new entry.
4278 * cr - credentials of caller.
4279 * ct - caller context
4281 * RETURN: 0 on success, error code on failure.
4284 * tdvp - ctime|mtime updated
4285 * svp - ctime updated
4289 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
4290 caller_context_t *ct, int flags)
4292 znode_t *dzp = VTOZ(tdvp);
4294 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4303 boolean_t waited = B_FALSE;
4305 ASSERT(tdvp->v_type == VDIR);
4309 zilog = zfsvfs->z_log;
4311 if (VOP_REALVP(svp, &realvp, ct) == 0)
4315 * POSIX dictates that we return EPERM here.
4316 * Better choices include ENOTSUP or EISDIR.
4318 if (svp->v_type == VDIR) {
4320 return (SET_ERROR(EPERM));
4323 if (svp->v_vfsp != tdvp->v_vfsp || zfsctl_is_node(svp)) {
4325 return (SET_ERROR(EXDEV));
4331 /* Prevent links to .zfs/shares files */
4333 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4334 &parent, sizeof (uint64_t))) != 0) {
4338 if (parent == zfsvfs->z_shares_dir) {
4340 return (SET_ERROR(EPERM));
4343 if (zfsvfs->z_utf8 && u8_validate(name,
4344 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4346 return (SET_ERROR(EILSEQ));
4348 if (flags & FIGNORECASE)
4352 * We do not support links between attributes and non-attributes
4353 * because of the potential security risk of creating links
4354 * into "normal" file space in order to circumvent restrictions
4355 * imposed in attribute space.
4357 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4359 return (SET_ERROR(EINVAL));
4363 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4364 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) {
4366 return (SET_ERROR(EPERM));
4369 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4376 * Attempt to lock directory; fail if entry already exists.
4378 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4384 tx = dmu_tx_create(zfsvfs->z_os);
4385 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4386 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4387 zfs_sa_upgrade_txholds(tx, szp);
4388 zfs_sa_upgrade_txholds(tx, dzp);
4389 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4391 zfs_dirent_unlock(dl);
4392 if (error == ERESTART) {
4403 error = zfs_link_create(dl, szp, tx, 0);
4406 uint64_t txtype = TX_LINK;
4407 if (flags & FIGNORECASE)
4409 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4414 zfs_dirent_unlock(dl);
4417 vnevent_link(svp, ct);
4420 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4421 zil_commit(zilog, 0);
4429 * zfs_null_putapage() is used when the file system has been force
4430 * unmounted. It just drops the pages.
4434 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4435 size_t *lenp, int flags, cred_t *cr)
4437 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4442 * Push a page out to disk, klustering if possible.
4444 * IN: vp - file to push page to.
4445 * pp - page to push.
4446 * flags - additional flags.
4447 * cr - credentials of caller.
4449 * OUT: offp - start of range pushed.
4450 * lenp - len of range pushed.
4452 * RETURN: 0 on success, error code on failure.
4454 * NOTE: callers must have locked the page to be pushed. On
4455 * exit, the page (and all other pages in the kluster) must be
4460 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4461 size_t *lenp, int flags, cred_t *cr)
4463 znode_t *zp = VTOZ(vp);
4464 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4466 u_offset_t off, koff;
4473 * If our blocksize is bigger than the page size, try to kluster
4474 * multiple pages so that we write a full block (thus avoiding
4475 * a read-modify-write).
4477 if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4478 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4479 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4480 ASSERT(koff <= zp->z_size);
4481 if (koff + klen > zp->z_size)
4482 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4483 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4485 ASSERT3U(btop(len), ==, btopr(len));
4488 * Can't push pages past end-of-file.
4490 if (off >= zp->z_size) {
4491 /* ignore all pages */
4494 } else if (off + len > zp->z_size) {
4495 int npages = btopr(zp->z_size - off);
4498 page_list_break(&pp, &trunc, npages);
4499 /* ignore pages past end of file */
4501 pvn_write_done(trunc, flags);
4502 len = zp->z_size - off;
4505 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4506 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4507 err = SET_ERROR(EDQUOT);
4511 tx = dmu_tx_create(zfsvfs->z_os);
4512 dmu_tx_hold_write(tx, zp->z_id, off, len);
4514 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4515 zfs_sa_upgrade_txholds(tx, zp);
4516 err = dmu_tx_assign(tx, TXG_NOWAIT);
4518 if (err == ERESTART) {
4527 if (zp->z_blksz <= PAGESIZE) {
4528 caddr_t va = zfs_map_page(pp, S_READ);
4529 ASSERT3U(len, <=, PAGESIZE);
4530 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4531 zfs_unmap_page(pp, va);
4533 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4537 uint64_t mtime[2], ctime[2];
4538 sa_bulk_attr_t bulk[3];
4541 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4543 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4545 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4547 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4549 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4554 pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4564 * Copy the portion of the file indicated from pages into the file.
4565 * The pages are stored in a page list attached to the files vnode.
4567 * IN: vp - vnode of file to push page data to.
4568 * off - position in file to put data.
4569 * len - amount of data to write.
4570 * flags - flags to control the operation.
4571 * cr - credentials of caller.
4572 * ct - caller context.
4574 * RETURN: 0 on success, error code on failure.
4577 * vp - ctime|mtime updated
4581 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4582 caller_context_t *ct)
4584 znode_t *zp = VTOZ(vp);
4585 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4597 * Align this request to the file block size in case we kluster.
4598 * XXX - this can result in pretty aggresive locking, which can
4599 * impact simultanious read/write access. One option might be
4600 * to break up long requests (len == 0) into block-by-block
4601 * operations to get narrower locking.
4603 blksz = zp->z_blksz;
4605 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4608 if (len > 0 && ISP2(blksz))
4609 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4615 * Search the entire vp list for pages >= io_off.
4617 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4618 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4621 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4623 if (off > zp->z_size) {
4624 /* past end of file */
4625 zfs_range_unlock(rl);
4630 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4632 for (off = io_off; io_off < off + len; io_off += io_len) {
4633 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4634 pp = page_lookup(vp, io_off,
4635 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4637 pp = page_lookup_nowait(vp, io_off,
4638 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4641 if (pp != NULL && pvn_getdirty(pp, flags)) {
4645 * Found a dirty page to push
4647 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4655 zfs_range_unlock(rl);
4656 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4657 zil_commit(zfsvfs->z_log, zp->z_id);
4665 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4667 znode_t *zp = VTOZ(vp);
4668 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4671 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4672 if (zp->z_sa_hdl == NULL) {
4674 * The fs has been unmounted, or we did a
4675 * suspend/resume and this file no longer exists.
4677 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4678 vrecycle(vp, curthread);
4682 mutex_enter(&zp->z_lock);
4683 if (zp->z_unlinked) {
4685 * Fast path to recycle a vnode of a removed file.
4687 mutex_exit(&zp->z_lock);
4688 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4689 vrecycle(vp, curthread);
4692 mutex_exit(&zp->z_lock);
4694 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4695 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4697 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4698 zfs_sa_upgrade_txholds(tx, zp);
4699 error = dmu_tx_assign(tx, TXG_WAIT);
4703 mutex_enter(&zp->z_lock);
4704 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4705 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4706 zp->z_atime_dirty = 0;
4707 mutex_exit(&zp->z_lock);
4711 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4716 * Bounds-check the seek operation.
4718 * IN: vp - vnode seeking within
4719 * ooff - old file offset
4720 * noffp - pointer to new file offset
4721 * ct - caller context
4723 * RETURN: 0 on success, EINVAL if new offset invalid.
4727 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4728 caller_context_t *ct)
4730 if (vp->v_type == VDIR)
4732 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4736 * Pre-filter the generic locking function to trap attempts to place
4737 * a mandatory lock on a memory mapped file.
4740 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4741 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4743 znode_t *zp = VTOZ(vp);
4744 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4750 * We are following the UFS semantics with respect to mapcnt
4751 * here: If we see that the file is mapped already, then we will
4752 * return an error, but we don't worry about races between this
4753 * function and zfs_map().
4755 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4757 return (SET_ERROR(EAGAIN));
4760 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4764 * If we can't find a page in the cache, we will create a new page
4765 * and fill it with file data. For efficiency, we may try to fill
4766 * multiple pages at once (klustering) to fill up the supplied page
4767 * list. Note that the pages to be filled are held with an exclusive
4768 * lock to prevent access by other threads while they are being filled.
4771 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4772 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4774 znode_t *zp = VTOZ(vp);
4775 page_t *pp, *cur_pp;
4776 objset_t *os = zp->z_zfsvfs->z_os;
4777 u_offset_t io_off, total;
4781 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4783 * We only have a single page, don't bother klustering
4787 pp = page_create_va(vp, io_off, io_len,
4788 PG_EXCL | PG_WAIT, seg, addr);
4791 * Try to find enough pages to fill the page list
4793 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4794 &io_len, off, plsz, 0);
4798 * The page already exists, nothing to do here.
4805 * Fill the pages in the kluster.
4808 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4811 ASSERT3U(io_off, ==, cur_pp->p_offset);
4812 va = zfs_map_page(cur_pp, S_WRITE);
4813 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4815 zfs_unmap_page(cur_pp, va);
4817 /* On error, toss the entire kluster */
4818 pvn_read_done(pp, B_ERROR);
4819 /* convert checksum errors into IO errors */
4821 err = SET_ERROR(EIO);
4824 cur_pp = cur_pp->p_next;
4828 * Fill in the page list array from the kluster starting
4829 * from the desired offset `off'.
4830 * NOTE: the page list will always be null terminated.
4832 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4833 ASSERT(pl == NULL || (*pl)->p_offset == off);
4839 * Return pointers to the pages for the file region [off, off + len]
4840 * in the pl array. If plsz is greater than len, this function may
4841 * also return page pointers from after the specified region
4842 * (i.e. the region [off, off + plsz]). These additional pages are
4843 * only returned if they are already in the cache, or were created as
4844 * part of a klustered read.
4846 * IN: vp - vnode of file to get data from.
4847 * off - position in file to get data from.
4848 * len - amount of data to retrieve.
4849 * plsz - length of provided page list.
4850 * seg - segment to obtain pages for.
4851 * addr - virtual address of fault.
4852 * rw - mode of created pages.
4853 * cr - credentials of caller.
4854 * ct - caller context.
4856 * OUT: protp - protection mode of created pages.
4857 * pl - list of pages created.
4859 * RETURN: 0 on success, error code on failure.
4862 * vp - atime updated
4866 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4867 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4868 enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4870 znode_t *zp = VTOZ(vp);
4871 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4875 /* we do our own caching, faultahead is unnecessary */
4878 else if (len > plsz)
4881 len = P2ROUNDUP(len, PAGESIZE);
4882 ASSERT(plsz >= len);
4891 * Loop through the requested range [off, off + len) looking
4892 * for pages. If we don't find a page, we will need to create
4893 * a new page and fill it with data from the file.
4896 if (*pl = page_lookup(vp, off, SE_SHARED))
4898 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4901 ASSERT3U((*pl)->p_offset, ==, off);
4905 ASSERT3U(len, >=, PAGESIZE);
4908 ASSERT3U(plsz, >=, PAGESIZE);
4915 * Fill out the page array with any pages already in the cache.
4918 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4925 * Release any pages we have previously locked.
4930 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4940 * Request a memory map for a section of a file. This code interacts
4941 * with common code and the VM system as follows:
4943 * - common code calls mmap(), which ends up in smmap_common()
4944 * - this calls VOP_MAP(), which takes you into (say) zfs
4945 * - zfs_map() calls as_map(), passing segvn_create() as the callback
4946 * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4947 * - zfs_addmap() updates z_mapcnt
4951 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4952 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4953 caller_context_t *ct)
4955 znode_t *zp = VTOZ(vp);
4956 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4957 segvn_crargs_t vn_a;
4963 if ((prot & PROT_WRITE) && (zp->z_pflags &
4964 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4966 return (SET_ERROR(EPERM));
4969 if ((prot & (PROT_READ | PROT_EXEC)) &&
4970 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4972 return (SET_ERROR(EACCES));
4975 if (vp->v_flag & VNOMAP) {
4977 return (SET_ERROR(ENOSYS));
4980 if (off < 0 || len > MAXOFFSET_T - off) {
4982 return (SET_ERROR(ENXIO));
4985 if (vp->v_type != VREG) {
4987 return (SET_ERROR(ENODEV));
4991 * If file is locked, disallow mapping.
4993 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
4995 return (SET_ERROR(EAGAIN));
4999 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
5007 vn_a.offset = (u_offset_t)off;
5008 vn_a.type = flags & MAP_TYPE;
5010 vn_a.maxprot = maxprot;
5013 vn_a.flags = flags & ~MAP_TYPE;
5015 vn_a.lgrp_mem_policy_flags = 0;
5017 error = as_map(as, *addrp, len, segvn_create, &vn_a);
5026 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5027 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
5028 caller_context_t *ct)
5030 uint64_t pages = btopr(len);
5032 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
5037 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
5038 * more accurate mtime for the associated file. Since we don't have a way of
5039 * detecting when the data was actually modified, we have to resort to
5040 * heuristics. If an explicit msync() is done, then we mark the mtime when the
5041 * last page is pushed. The problem occurs when the msync() call is omitted,
5042 * which by far the most common case:
5050 * putpage() via fsflush
5052 * If we wait until fsflush to come along, we can have a modification time that
5053 * is some arbitrary point in the future. In order to prevent this in the
5054 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
5059 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5060 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
5061 caller_context_t *ct)
5063 uint64_t pages = btopr(len);
5065 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
5066 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
5068 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
5069 vn_has_cached_data(vp))
5070 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
5076 * Free or allocate space in a file. Currently, this function only
5077 * supports the `F_FREESP' command. However, this command is somewhat
5078 * misnamed, as its functionality includes the ability to allocate as
5079 * well as free space.
5081 * IN: vp - vnode of file to free data in.
5082 * cmd - action to take (only F_FREESP supported).
5083 * bfp - section of file to free/alloc.
5084 * flag - current file open mode flags.
5085 * offset - current file offset.
5086 * cr - credentials of caller [UNUSED].
5087 * ct - caller context.
5089 * RETURN: 0 on success, error code on failure.
5092 * vp - ctime|mtime updated
5096 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
5097 offset_t offset, cred_t *cr, caller_context_t *ct)
5099 znode_t *zp = VTOZ(vp);
5100 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5107 if (cmd != F_FREESP) {
5109 return (SET_ERROR(EINVAL));
5112 if (error = convoff(vp, bfp, 0, offset)) {
5117 if (bfp->l_len < 0) {
5119 return (SET_ERROR(EINVAL));
5123 len = bfp->l_len; /* 0 means from off to end of file */
5125 error = zfs_freesp(zp, off, len, flag, TRUE);
5132 CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid));
5133 CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid));
5137 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
5139 znode_t *zp = VTOZ(vp);
5140 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5143 uint64_t object = zp->z_id;
5150 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
5151 &gen64, sizeof (uint64_t))) != 0) {
5156 gen = (uint32_t)gen64;
5158 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
5161 if (fidp->fid_len < size) {
5162 fidp->fid_len = size;
5164 return (SET_ERROR(ENOSPC));
5167 fidp->fid_len = size;
5170 zfid = (zfid_short_t *)fidp;
5172 zfid->zf_len = size;
5174 for (i = 0; i < sizeof (zfid->zf_object); i++)
5175 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
5177 /* Must have a non-zero generation number to distinguish from .zfs */
5180 for (i = 0; i < sizeof (zfid->zf_gen); i++)
5181 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
5183 if (size == LONG_FID_LEN) {
5184 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
5187 zlfid = (zfid_long_t *)fidp;
5189 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
5190 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
5192 /* XXX - this should be the generation number for the objset */
5193 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
5194 zlfid->zf_setgen[i] = 0;
5202 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
5203 caller_context_t *ct)
5215 case _PC_FILESIZEBITS:
5219 case _PC_XATTR_EXISTS:
5221 zfsvfs = zp->z_zfsvfs;
5225 error = zfs_dirent_lock(&dl, zp, "", &xzp,
5226 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
5228 zfs_dirent_unlock(dl);
5229 if (!zfs_dirempty(xzp))
5232 } else if (error == ENOENT) {
5234 * If there aren't extended attributes, it's the
5235 * same as having zero of them.
5242 case _PC_SATTR_ENABLED:
5243 case _PC_SATTR_EXISTS:
5244 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
5245 (vp->v_type == VREG || vp->v_type == VDIR);
5248 case _PC_ACCESS_FILTERING:
5249 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
5253 case _PC_ACL_ENABLED:
5254 *valp = _ACL_ACE_ENABLED;
5257 case _PC_MIN_HOLE_SIZE:
5258 *valp = (int)SPA_MINBLOCKSIZE;
5261 case _PC_TIMESTAMP_RESOLUTION:
5262 /* nanosecond timestamp resolution */
5266 case _PC_ACL_EXTENDED:
5274 case _PC_ACL_PATH_MAX:
5275 *valp = ACL_MAX_ENTRIES;
5279 return (EOPNOTSUPP);
5285 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5286 caller_context_t *ct)
5288 znode_t *zp = VTOZ(vp);
5289 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5291 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5295 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5303 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5304 caller_context_t *ct)
5306 znode_t *zp = VTOZ(vp);
5307 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5309 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5310 zilog_t *zilog = zfsvfs->z_log;
5315 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5317 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5318 zil_commit(zilog, 0);
5326 * The smallest read we may consider to loan out an arcbuf.
5327 * This must be a power of 2.
5329 int zcr_blksz_min = (1 << 10); /* 1K */
5331 * If set to less than the file block size, allow loaning out of an
5332 * arcbuf for a partial block read. This must be a power of 2.
5334 int zcr_blksz_max = (1 << 17); /* 128K */
5338 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5339 caller_context_t *ct)
5341 znode_t *zp = VTOZ(vp);
5342 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5343 int max_blksz = zfsvfs->z_max_blksz;
5344 uio_t *uio = &xuio->xu_uio;
5345 ssize_t size = uio->uio_resid;
5346 offset_t offset = uio->uio_loffset;
5351 int preamble, postamble;
5353 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5354 return (SET_ERROR(EINVAL));
5361 * Loan out an arc_buf for write if write size is bigger than
5362 * max_blksz, and the file's block size is also max_blksz.
5365 if (size < blksz || zp->z_blksz != blksz) {
5367 return (SET_ERROR(EINVAL));
5370 * Caller requests buffers for write before knowing where the
5371 * write offset might be (e.g. NFS TCP write).
5376 preamble = P2PHASE(offset, blksz);
5378 preamble = blksz - preamble;
5383 postamble = P2PHASE(size, blksz);
5386 fullblk = size / blksz;
5387 (void) dmu_xuio_init(xuio,
5388 (preamble != 0) + fullblk + (postamble != 0));
5389 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5390 int, postamble, int,
5391 (preamble != 0) + fullblk + (postamble != 0));
5394 * Have to fix iov base/len for partial buffers. They
5395 * currently represent full arc_buf's.
5398 /* data begins in the middle of the arc_buf */
5399 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5402 (void) dmu_xuio_add(xuio, abuf,
5403 blksz - preamble, preamble);
5406 for (i = 0; i < fullblk; i++) {
5407 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5410 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5414 /* data ends in the middle of the arc_buf */
5415 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5418 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5423 * Loan out an arc_buf for read if the read size is larger than
5424 * the current file block size. Block alignment is not
5425 * considered. Partial arc_buf will be loaned out for read.
5427 blksz = zp->z_blksz;
5428 if (blksz < zcr_blksz_min)
5429 blksz = zcr_blksz_min;
5430 if (blksz > zcr_blksz_max)
5431 blksz = zcr_blksz_max;
5432 /* avoid potential complexity of dealing with it */
5433 if (blksz > max_blksz) {
5435 return (SET_ERROR(EINVAL));
5438 maxsize = zp->z_size - uio->uio_loffset;
5442 if (size < blksz || vn_has_cached_data(vp)) {
5444 return (SET_ERROR(EINVAL));
5449 return (SET_ERROR(EINVAL));
5452 uio->uio_extflg = UIO_XUIO;
5453 XUIO_XUZC_RW(xuio) = ioflag;
5460 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5464 int ioflag = XUIO_XUZC_RW(xuio);
5466 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5468 i = dmu_xuio_cnt(xuio);
5470 abuf = dmu_xuio_arcbuf(xuio, i);
5472 * if abuf == NULL, it must be a write buffer
5473 * that has been returned in zfs_write().
5476 dmu_return_arcbuf(abuf);
5477 ASSERT(abuf || ioflag == UIO_WRITE);
5480 dmu_xuio_fini(xuio);
5485 * Predeclare these here so that the compiler assumes that
5486 * this is an "old style" function declaration that does
5487 * not include arguments => we won't get type mismatch errors
5488 * in the initializations that follow.
5490 static int zfs_inval();
5491 static int zfs_isdir();
5496 return (SET_ERROR(EINVAL));
5502 return (SET_ERROR(EISDIR));
5505 * Directory vnode operations template
5507 vnodeops_t *zfs_dvnodeops;
5508 const fs_operation_def_t zfs_dvnodeops_template[] = {
5509 VOPNAME_OPEN, { .vop_open = zfs_open },
5510 VOPNAME_CLOSE, { .vop_close = zfs_close },
5511 VOPNAME_READ, { .error = zfs_isdir },
5512 VOPNAME_WRITE, { .error = zfs_isdir },
5513 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5514 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5515 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5516 VOPNAME_ACCESS, { .vop_access = zfs_access },
5517 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5518 VOPNAME_CREATE, { .vop_create = zfs_create },
5519 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5520 VOPNAME_LINK, { .vop_link = zfs_link },
5521 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5522 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir },
5523 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5524 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5525 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink },
5526 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5527 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5528 VOPNAME_FID, { .vop_fid = zfs_fid },
5529 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5530 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5531 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5532 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5533 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5538 * Regular file vnode operations template
5540 vnodeops_t *zfs_fvnodeops;
5541 const fs_operation_def_t zfs_fvnodeops_template[] = {
5542 VOPNAME_OPEN, { .vop_open = zfs_open },
5543 VOPNAME_CLOSE, { .vop_close = zfs_close },
5544 VOPNAME_READ, { .vop_read = zfs_read },
5545 VOPNAME_WRITE, { .vop_write = zfs_write },
5546 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5547 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5548 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5549 VOPNAME_ACCESS, { .vop_access = zfs_access },
5550 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5551 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5552 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5553 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5554 VOPNAME_FID, { .vop_fid = zfs_fid },
5555 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5556 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock },
5557 VOPNAME_SPACE, { .vop_space = zfs_space },
5558 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage },
5559 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage },
5560 VOPNAME_MAP, { .vop_map = zfs_map },
5561 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap },
5562 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap },
5563 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5564 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5565 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5566 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5567 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf },
5568 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf },
5573 * Symbolic link vnode operations template
5575 vnodeops_t *zfs_symvnodeops;
5576 const fs_operation_def_t zfs_symvnodeops_template[] = {
5577 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5578 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5579 VOPNAME_ACCESS, { .vop_access = zfs_access },
5580 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5581 VOPNAME_READLINK, { .vop_readlink = zfs_readlink },
5582 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5583 VOPNAME_FID, { .vop_fid = zfs_fid },
5584 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5585 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5590 * special share hidden files vnode operations template
5592 vnodeops_t *zfs_sharevnodeops;
5593 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5594 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5595 VOPNAME_ACCESS, { .vop_access = zfs_access },
5596 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5597 VOPNAME_FID, { .vop_fid = zfs_fid },
5598 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5599 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5600 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5601 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5606 * Extended attribute directory vnode operations template
5608 * This template is identical to the directory vnodes
5609 * operation template except for restricted operations:
5613 * Note that there are other restrictions embedded in:
5614 * zfs_create() - restrict type to VREG
5615 * zfs_link() - no links into/out of attribute space
5616 * zfs_rename() - no moves into/out of attribute space
5618 vnodeops_t *zfs_xdvnodeops;
5619 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5620 VOPNAME_OPEN, { .vop_open = zfs_open },
5621 VOPNAME_CLOSE, { .vop_close = zfs_close },
5622 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5623 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5624 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5625 VOPNAME_ACCESS, { .vop_access = zfs_access },
5626 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5627 VOPNAME_CREATE, { .vop_create = zfs_create },
5628 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5629 VOPNAME_LINK, { .vop_link = zfs_link },
5630 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5631 VOPNAME_MKDIR, { .error = zfs_inval },
5632 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5633 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5634 VOPNAME_SYMLINK, { .error = zfs_inval },
5635 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5636 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5637 VOPNAME_FID, { .vop_fid = zfs_fid },
5638 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5639 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5640 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5641 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5642 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5647 * Error vnode operations template
5649 vnodeops_t *zfs_evnodeops;
5650 const fs_operation_def_t zfs_evnodeops_template[] = {
5651 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5652 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5658 ioflags(int ioflags)
5662 if (ioflags & IO_APPEND)
5664 if (ioflags & IO_NDELAY)
5666 if (ioflags & IO_SYNC)
5667 flags |= (FSYNC | FDSYNC | FRSYNC);
5673 zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage)
5675 znode_t *zp = VTOZ(vp);
5676 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5677 objset_t *os = zp->z_zfsvfs->z_os;
5678 vm_page_t mfirst, mlast, mreq;
5682 off_t startoff, endoff;
5684 vm_pindex_t reqstart, reqend;
5685 int pcount, lsize, reqsize, size;
5690 pcount = OFF_TO_IDX(round_page(count));
5692 object = mreq->object;
5695 KASSERT(vp->v_object == object, ("mismatching object"));
5697 if (pcount > 1 && zp->z_blksz > PAGESIZE) {
5698 startoff = rounddown(IDX_TO_OFF(mreq->pindex), zp->z_blksz);
5699 reqstart = OFF_TO_IDX(round_page(startoff));
5700 if (reqstart < m[0]->pindex)
5703 reqstart = reqstart - m[0]->pindex;
5704 endoff = roundup(IDX_TO_OFF(mreq->pindex) + PAGE_SIZE,
5706 reqend = OFF_TO_IDX(trunc_page(endoff)) - 1;
5707 if (reqend > m[pcount - 1]->pindex)
5708 reqend = m[pcount - 1]->pindex;
5709 reqsize = reqend - m[reqstart]->pindex + 1;
5710 KASSERT(reqstart <= reqpage && reqpage < reqstart + reqsize,
5711 ("reqpage beyond [reqstart, reqstart + reqsize[ bounds"));
5716 mfirst = m[reqstart];
5717 mlast = m[reqstart + reqsize - 1];
5719 VM_OBJECT_LOCK(object);
5721 for (i = 0; i < reqstart; i++) {
5724 vm_page_unlock(m[i]);
5726 for (i = reqstart + reqsize; i < pcount; i++) {
5729 vm_page_unlock(m[i]);
5732 if (mreq->valid && reqsize == 1) {
5733 if (mreq->valid != VM_PAGE_BITS_ALL)
5734 vm_page_zero_invalid(mreq, TRUE);
5735 VM_OBJECT_UNLOCK(object);
5737 return (VM_PAGER_OK);
5740 PCPU_INC(cnt.v_vnodein);
5741 PCPU_ADD(cnt.v_vnodepgsin, reqsize);
5743 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) {
5744 for (i = reqstart; i < reqstart + reqsize; i++) {
5748 vm_page_unlock(m[i]);
5751 VM_OBJECT_UNLOCK(object);
5753 return (VM_PAGER_BAD);
5757 if (IDX_TO_OFF(mlast->pindex) + lsize > object->un_pager.vnp.vnp_size)
5758 lsize = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mlast->pindex);
5760 VM_OBJECT_UNLOCK(object);
5762 for (i = reqstart; i < reqstart + reqsize; i++) {
5764 if (i == (reqstart + reqsize - 1))
5766 va = zfs_map_page(m[i], &sf);
5767 error = dmu_read(os, zp->z_id, IDX_TO_OFF(m[i]->pindex),
5768 size, va, DMU_READ_PREFETCH);
5769 if (size != PAGE_SIZE)
5770 bzero(va + size, PAGE_SIZE - size);
5776 VM_OBJECT_LOCK(object);
5778 for (i = reqstart; i < reqstart + reqsize; i++) {
5780 m[i]->valid = VM_PAGE_BITS_ALL;
5781 KASSERT(m[i]->dirty == 0, ("zfs_getpages: page %p is dirty", m[i]));
5783 vm_page_readahead_finish(m[i]);
5786 VM_OBJECT_UNLOCK(object);
5788 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
5790 return (error ? VM_PAGER_ERROR : VM_PAGER_OK);
5794 zfs_freebsd_getpages(ap)
5795 struct vop_getpages_args /* {
5800 vm_ooffset_t a_offset;
5804 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage));
5808 zfs_freebsd_bmap(ap)
5809 struct vop_bmap_args /* {
5812 struct bufobj **a_bop;
5819 if (ap->a_bop != NULL)
5820 *ap->a_bop = &ap->a_vp->v_bufobj;
5821 if (ap->a_bnp != NULL)
5822 *ap->a_bnp = ap->a_bn;
5823 if (ap->a_runp != NULL)
5825 if (ap->a_runb != NULL)
5832 zfs_freebsd_open(ap)
5833 struct vop_open_args /* {
5836 struct ucred *a_cred;
5837 struct thread *a_td;
5840 vnode_t *vp = ap->a_vp;
5841 znode_t *zp = VTOZ(vp);
5844 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL);
5846 vnode_create_vobject(vp, zp->z_size, ap->a_td);
5851 zfs_freebsd_close(ap)
5852 struct vop_close_args /* {
5855 struct ucred *a_cred;
5856 struct thread *a_td;
5860 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL));
5864 zfs_freebsd_ioctl(ap)
5865 struct vop_ioctl_args /* {
5875 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data,
5876 ap->a_fflag, ap->a_cred, NULL, NULL));
5880 zfs_freebsd_read(ap)
5881 struct vop_read_args /* {
5885 struct ucred *a_cred;
5889 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5894 zfs_freebsd_write(ap)
5895 struct vop_write_args /* {
5899 struct ucred *a_cred;
5903 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5908 zfs_freebsd_access(ap)
5909 struct vop_access_args /* {
5911 accmode_t a_accmode;
5912 struct ucred *a_cred;
5913 struct thread *a_td;
5916 vnode_t *vp = ap->a_vp;
5917 znode_t *zp = VTOZ(vp);
5922 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND,
5924 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND);
5926 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL);
5929 * VADMIN has to be handled by vaccess().
5932 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND);
5934 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
5935 zp->z_gid, accmode, ap->a_cred, NULL);
5940 * For VEXEC, ensure that at least one execute bit is set for
5943 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR &&
5944 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
5952 zfs_freebsd_lookup(ap)
5953 struct vop_lookup_args /* {
5954 struct vnode *a_dvp;
5955 struct vnode **a_vpp;
5956 struct componentname *a_cnp;
5959 struct componentname *cnp = ap->a_cnp;
5960 char nm[NAME_MAX + 1];
5962 ASSERT(cnp->cn_namelen < sizeof(nm));
5963 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm)));
5965 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop,
5966 cnp->cn_cred, cnp->cn_thread, 0));
5970 zfs_freebsd_create(ap)
5971 struct vop_create_args /* {
5972 struct vnode *a_dvp;
5973 struct vnode **a_vpp;
5974 struct componentname *a_cnp;
5975 struct vattr *a_vap;
5978 struct componentname *cnp = ap->a_cnp;
5979 vattr_t *vap = ap->a_vap;
5982 ASSERT(cnp->cn_flags & SAVENAME);
5984 vattr_init_mask(vap);
5985 mode = vap->va_mode & ALLPERMS;
5987 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode,
5988 ap->a_vpp, cnp->cn_cred, cnp->cn_thread));
5992 zfs_freebsd_remove(ap)
5993 struct vop_remove_args /* {
5994 struct vnode *a_dvp;
5996 struct componentname *a_cnp;
6000 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6002 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr,
6003 ap->a_cnp->cn_cred, NULL, 0));
6007 zfs_freebsd_mkdir(ap)
6008 struct vop_mkdir_args /* {
6009 struct vnode *a_dvp;
6010 struct vnode **a_vpp;
6011 struct componentname *a_cnp;
6012 struct vattr *a_vap;
6015 vattr_t *vap = ap->a_vap;
6017 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6019 vattr_init_mask(vap);
6021 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp,
6022 ap->a_cnp->cn_cred, NULL, 0, NULL));
6026 zfs_freebsd_rmdir(ap)
6027 struct vop_rmdir_args /* {
6028 struct vnode *a_dvp;
6030 struct componentname *a_cnp;
6033 struct componentname *cnp = ap->a_cnp;
6035 ASSERT(cnp->cn_flags & SAVENAME);
6037 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0));
6041 zfs_freebsd_readdir(ap)
6042 struct vop_readdir_args /* {
6045 struct ucred *a_cred;
6052 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag,
6053 ap->a_ncookies, ap->a_cookies));
6057 zfs_freebsd_fsync(ap)
6058 struct vop_fsync_args /* {
6061 struct thread *a_td;
6066 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL));
6070 zfs_freebsd_getattr(ap)
6071 struct vop_getattr_args /* {
6073 struct vattr *a_vap;
6074 struct ucred *a_cred;
6077 vattr_t *vap = ap->a_vap;
6083 xvap.xva_vattr = *vap;
6084 xvap.xva_vattr.va_mask |= AT_XVATTR;
6086 /* Convert chflags into ZFS-type flags. */
6087 /* XXX: what about SF_SETTABLE?. */
6088 XVA_SET_REQ(&xvap, XAT_IMMUTABLE);
6089 XVA_SET_REQ(&xvap, XAT_APPENDONLY);
6090 XVA_SET_REQ(&xvap, XAT_NOUNLINK);
6091 XVA_SET_REQ(&xvap, XAT_NODUMP);
6092 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL);
6096 /* Convert ZFS xattr into chflags. */
6097 #define FLAG_CHECK(fflag, xflag, xfield) do { \
6098 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \
6099 fflags |= (fflag); \
6101 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE,
6102 xvap.xva_xoptattrs.xoa_immutable);
6103 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY,
6104 xvap.xva_xoptattrs.xoa_appendonly);
6105 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK,
6106 xvap.xva_xoptattrs.xoa_nounlink);
6107 FLAG_CHECK(UF_NODUMP, XAT_NODUMP,
6108 xvap.xva_xoptattrs.xoa_nodump);
6110 *vap = xvap.xva_vattr;
6111 vap->va_flags = fflags;
6116 zfs_freebsd_setattr(ap)
6117 struct vop_setattr_args /* {
6119 struct vattr *a_vap;
6120 struct ucred *a_cred;
6123 vnode_t *vp = ap->a_vp;
6124 vattr_t *vap = ap->a_vap;
6125 cred_t *cred = ap->a_cred;
6130 vattr_init_mask(vap);
6131 vap->va_mask &= ~AT_NOSET;
6134 xvap.xva_vattr = *vap;
6136 zflags = VTOZ(vp)->z_pflags;
6138 if (vap->va_flags != VNOVAL) {
6139 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs;
6142 if (zfsvfs->z_use_fuids == B_FALSE)
6143 return (EOPNOTSUPP);
6145 fflags = vap->va_flags;
6146 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_NODUMP)) != 0)
6147 return (EOPNOTSUPP);
6149 * Unprivileged processes are not permitted to unset system
6150 * flags, or modify flags if any system flags are set.
6151 * Privileged non-jail processes may not modify system flags
6152 * if securelevel > 0 and any existing system flags are set.
6153 * Privileged jail processes behave like privileged non-jail
6154 * processes if the security.jail.chflags_allowed sysctl is
6155 * is non-zero; otherwise, they behave like unprivileged
6158 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 ||
6159 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) {
6161 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6162 error = securelevel_gt(cred, 0);
6168 * Callers may only modify the file flags on objects they
6169 * have VADMIN rights for.
6171 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0)
6174 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6178 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) {
6183 #define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \
6184 if (((fflags & (fflag)) && !(zflags & (zflag))) || \
6185 ((zflags & (zflag)) && !(fflags & (fflag)))) { \
6186 XVA_SET_REQ(&xvap, (xflag)); \
6187 (xfield) = ((fflags & (fflag)) != 0); \
6190 /* Convert chflags into ZFS-type flags. */
6191 /* XXX: what about SF_SETTABLE?. */
6192 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE,
6193 xvap.xva_xoptattrs.xoa_immutable);
6194 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY,
6195 xvap.xva_xoptattrs.xoa_appendonly);
6196 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK,
6197 xvap.xva_xoptattrs.xoa_nounlink);
6198 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP,
6199 xvap.xva_xoptattrs.xoa_nodump);
6202 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL));
6206 zfs_freebsd_rename(ap)
6207 struct vop_rename_args /* {
6208 struct vnode *a_fdvp;
6209 struct vnode *a_fvp;
6210 struct componentname *a_fcnp;
6211 struct vnode *a_tdvp;
6212 struct vnode *a_tvp;
6213 struct componentname *a_tcnp;
6216 vnode_t *fdvp = ap->a_fdvp;
6217 vnode_t *fvp = ap->a_fvp;
6218 vnode_t *tdvp = ap->a_tdvp;
6219 vnode_t *tvp = ap->a_tvp;
6222 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART));
6223 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART));
6225 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp,
6226 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0);
6241 zfs_freebsd_symlink(ap)
6242 struct vop_symlink_args /* {
6243 struct vnode *a_dvp;
6244 struct vnode **a_vpp;
6245 struct componentname *a_cnp;
6246 struct vattr *a_vap;
6250 struct componentname *cnp = ap->a_cnp;
6251 vattr_t *vap = ap->a_vap;
6253 ASSERT(cnp->cn_flags & SAVENAME);
6255 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */
6256 vattr_init_mask(vap);
6258 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap,
6259 ap->a_target, cnp->cn_cred, cnp->cn_thread));
6263 zfs_freebsd_readlink(ap)
6264 struct vop_readlink_args /* {
6267 struct ucred *a_cred;
6271 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL));
6275 zfs_freebsd_link(ap)
6276 struct vop_link_args /* {
6277 struct vnode *a_tdvp;
6279 struct componentname *a_cnp;
6282 struct componentname *cnp = ap->a_cnp;
6284 ASSERT(cnp->cn_flags & SAVENAME);
6286 return (zfs_link(ap->a_tdvp, ap->a_vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0));
6290 zfs_freebsd_inactive(ap)
6291 struct vop_inactive_args /* {
6293 struct thread *a_td;
6296 vnode_t *vp = ap->a_vp;
6298 zfs_inactive(vp, ap->a_td->td_ucred, NULL);
6303 zfs_freebsd_reclaim(ap)
6304 struct vop_reclaim_args /* {
6306 struct thread *a_td;
6309 vnode_t *vp = ap->a_vp;
6310 znode_t *zp = VTOZ(vp);
6311 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
6315 /* Destroy the vm object and flush associated pages. */
6316 vnode_destroy_vobject(vp);
6319 * z_teardown_inactive_lock protects from a race with
6320 * zfs_znode_dmu_fini in zfsvfs_teardown during
6323 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
6324 if (zp->z_sa_hdl == NULL)
6328 rw_exit(&zfsvfs->z_teardown_inactive_lock);
6336 struct vop_fid_args /* {
6342 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL));
6346 zfs_freebsd_pathconf(ap)
6347 struct vop_pathconf_args /* {
6350 register_t *a_retval;
6356 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL);
6358 *ap->a_retval = val;
6359 else if (error == EOPNOTSUPP)
6360 error = vop_stdpathconf(ap);
6365 zfs_freebsd_fifo_pathconf(ap)
6366 struct vop_pathconf_args /* {
6369 register_t *a_retval;
6373 switch (ap->a_name) {
6374 case _PC_ACL_EXTENDED:
6376 case _PC_ACL_PATH_MAX:
6377 case _PC_MAC_PRESENT:
6378 return (zfs_freebsd_pathconf(ap));
6380 return (fifo_specops.vop_pathconf(ap));
6385 * FreeBSD's extended attributes namespace defines file name prefix for ZFS'
6386 * extended attribute name:
6389 * system freebsd:system:
6390 * user (none, can be used to access ZFS fsattr(5) attributes
6391 * created on Solaris)
6394 zfs_create_attrname(int attrnamespace, const char *name, char *attrname,
6397 const char *namespace, *prefix, *suffix;
6399 /* We don't allow '/' character in attribute name. */
6400 if (strchr(name, '/') != NULL)
6402 /* We don't allow attribute names that start with "freebsd:" string. */
6403 if (strncmp(name, "freebsd:", 8) == 0)
6406 bzero(attrname, size);
6408 switch (attrnamespace) {
6409 case EXTATTR_NAMESPACE_USER:
6411 prefix = "freebsd:";
6412 namespace = EXTATTR_NAMESPACE_USER_STRING;
6416 * This is the default namespace by which we can access all
6417 * attributes created on Solaris.
6419 prefix = namespace = suffix = "";
6422 case EXTATTR_NAMESPACE_SYSTEM:
6423 prefix = "freebsd:";
6424 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING;
6427 case EXTATTR_NAMESPACE_EMPTY:
6431 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix,
6433 return (ENAMETOOLONG);
6439 * Vnode operating to retrieve a named extended attribute.
6442 zfs_getextattr(struct vop_getextattr_args *ap)
6445 IN struct vnode *a_vp;
6446 IN int a_attrnamespace;
6447 IN const char *a_name;
6448 INOUT struct uio *a_uio;
6450 IN struct ucred *a_cred;
6451 IN struct thread *a_td;
6455 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6456 struct thread *td = ap->a_td;
6457 struct nameidata nd;
6460 vnode_t *xvp = NULL, *vp;
6463 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6464 ap->a_cred, ap->a_td, VREAD);
6468 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6475 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6483 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_SYSSPACE, attrname,
6485 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL);
6487 NDFREE(&nd, NDF_ONLY_PNBUF);
6490 if (error == ENOENT)
6495 if (ap->a_size != NULL) {
6496 error = VOP_GETATTR(vp, &va, ap->a_cred);
6498 *ap->a_size = (size_t)va.va_size;
6499 } else if (ap->a_uio != NULL)
6500 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6503 vn_close(vp, flags, ap->a_cred, td);
6510 * Vnode operation to remove a named attribute.
6513 zfs_deleteextattr(struct vop_deleteextattr_args *ap)
6516 IN struct vnode *a_vp;
6517 IN int a_attrnamespace;
6518 IN const char *a_name;
6519 IN struct ucred *a_cred;
6520 IN struct thread *a_td;
6524 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6525 struct thread *td = ap->a_td;
6526 struct nameidata nd;
6529 vnode_t *xvp = NULL, *vp;
6532 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6533 ap->a_cred, ap->a_td, VWRITE);
6537 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6544 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6551 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF | MPSAFE,
6552 UIO_SYSSPACE, attrname, xvp, td);
6555 NDFREE(&nd, NDF_ONLY_PNBUF);
6558 if (error == ENOENT)
6562 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
6565 if (vp == nd.ni_dvp)
6575 * Vnode operation to set a named attribute.
6578 zfs_setextattr(struct vop_setextattr_args *ap)
6581 IN struct vnode *a_vp;
6582 IN int a_attrnamespace;
6583 IN const char *a_name;
6584 INOUT struct uio *a_uio;
6585 IN struct ucred *a_cred;
6586 IN struct thread *a_td;
6590 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6591 struct thread *td = ap->a_td;
6592 struct nameidata nd;
6595 vnode_t *xvp = NULL, *vp;
6598 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6599 ap->a_cred, ap->a_td, VWRITE);
6603 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6610 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6611 LOOKUP_XATTR | CREATE_XATTR_DIR);
6617 flags = FFLAGS(O_WRONLY | O_CREAT);
6618 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_SYSSPACE, attrname,
6620 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL);
6622 NDFREE(&nd, NDF_ONLY_PNBUF);
6630 error = VOP_SETATTR(vp, &va, ap->a_cred);
6632 VOP_WRITE(vp, ap->a_uio, IO_UNIT | IO_SYNC, ap->a_cred);
6635 vn_close(vp, flags, ap->a_cred, td);
6642 * Vnode operation to retrieve extended attributes on a vnode.
6645 zfs_listextattr(struct vop_listextattr_args *ap)
6648 IN struct vnode *a_vp;
6649 IN int a_attrnamespace;
6650 INOUT struct uio *a_uio;
6652 IN struct ucred *a_cred;
6653 IN struct thread *a_td;
6657 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6658 struct thread *td = ap->a_td;
6659 struct nameidata nd;
6660 char attrprefix[16];
6661 u_char dirbuf[sizeof(struct dirent)];
6664 struct uio auio, *uio = ap->a_uio;
6665 size_t *sizep = ap->a_size;
6667 vnode_t *xvp = NULL, *vp;
6668 int done, error, eof, pos;
6670 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6671 ap->a_cred, ap->a_td, VREAD);
6675 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix,
6676 sizeof(attrprefix));
6679 plen = strlen(attrprefix);
6686 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6691 * ENOATTR means that the EA directory does not yet exist,
6692 * i.e. there are no extended attributes there.
6694 if (error == ENOATTR)
6699 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED | MPSAFE,
6700 UIO_SYSSPACE, ".", xvp, td);
6703 NDFREE(&nd, NDF_ONLY_PNBUF);
6709 auio.uio_iov = &aiov;
6710 auio.uio_iovcnt = 1;
6711 auio.uio_segflg = UIO_SYSSPACE;
6713 auio.uio_rw = UIO_READ;
6714 auio.uio_offset = 0;
6719 aiov.iov_base = (void *)dirbuf;
6720 aiov.iov_len = sizeof(dirbuf);
6721 auio.uio_resid = sizeof(dirbuf);
6722 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL);
6723 done = sizeof(dirbuf) - auio.uio_resid;
6726 for (pos = 0; pos < done;) {
6727 dp = (struct dirent *)(dirbuf + pos);
6728 pos += dp->d_reclen;
6730 * XXX: Temporarily we also accept DT_UNKNOWN, as this
6731 * is what we get when attribute was created on Solaris.
6733 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN)
6735 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0)
6737 else if (strncmp(dp->d_name, attrprefix, plen) != 0)
6739 nlen = dp->d_namlen - plen;
6742 else if (uio != NULL) {
6744 * Format of extattr name entry is one byte for
6745 * length and the rest for name.
6747 error = uiomove(&nlen, 1, uio->uio_rw, uio);
6749 error = uiomove(dp->d_name + plen, nlen,
6756 } while (!eof && error == 0);
6765 zfs_freebsd_getacl(ap)
6766 struct vop_getacl_args /* {
6775 vsecattr_t vsecattr;
6777 if (ap->a_type != ACL_TYPE_NFS4)
6780 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
6781 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL))
6784 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt);
6785 if (vsecattr.vsa_aclentp != NULL)
6786 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz);
6792 zfs_freebsd_setacl(ap)
6793 struct vop_setacl_args /* {
6802 vsecattr_t vsecattr;
6803 int aclbsize; /* size of acl list in bytes */
6806 if (ap->a_type != ACL_TYPE_NFS4)
6809 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES)
6813 * With NFSv4 ACLs, chmod(2) may need to add additional entries,
6814 * splitting every entry into two and appending "canonical six"
6815 * entries at the end. Don't allow for setting an ACL that would
6816 * cause chmod(2) to run out of ACL entries.
6818 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES)
6821 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR);
6825 vsecattr.vsa_mask = VSA_ACE;
6826 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t);
6827 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP);
6828 aaclp = vsecattr.vsa_aclentp;
6829 vsecattr.vsa_aclentsz = aclbsize;
6831 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp);
6832 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL);
6833 kmem_free(aaclp, aclbsize);
6839 zfs_freebsd_aclcheck(ap)
6840 struct vop_aclcheck_args /* {
6849 return (EOPNOTSUPP);
6852 struct vop_vector zfs_vnodeops;
6853 struct vop_vector zfs_fifoops;
6854 struct vop_vector zfs_shareops;
6856 struct vop_vector zfs_vnodeops = {
6857 .vop_default = &default_vnodeops,
6858 .vop_inactive = zfs_freebsd_inactive,
6859 .vop_reclaim = zfs_freebsd_reclaim,
6860 .vop_access = zfs_freebsd_access,
6861 #ifdef FREEBSD_NAMECACHE
6862 .vop_lookup = vfs_cache_lookup,
6863 .vop_cachedlookup = zfs_freebsd_lookup,
6865 .vop_lookup = zfs_freebsd_lookup,
6867 .vop_getattr = zfs_freebsd_getattr,
6868 .vop_setattr = zfs_freebsd_setattr,
6869 .vop_create = zfs_freebsd_create,
6870 .vop_mknod = zfs_freebsd_create,
6871 .vop_mkdir = zfs_freebsd_mkdir,
6872 .vop_readdir = zfs_freebsd_readdir,
6873 .vop_fsync = zfs_freebsd_fsync,
6874 .vop_open = zfs_freebsd_open,
6875 .vop_close = zfs_freebsd_close,
6876 .vop_rmdir = zfs_freebsd_rmdir,
6877 .vop_ioctl = zfs_freebsd_ioctl,
6878 .vop_link = zfs_freebsd_link,
6879 .vop_symlink = zfs_freebsd_symlink,
6880 .vop_readlink = zfs_freebsd_readlink,
6881 .vop_read = zfs_freebsd_read,
6882 .vop_write = zfs_freebsd_write,
6883 .vop_remove = zfs_freebsd_remove,
6884 .vop_rename = zfs_freebsd_rename,
6885 .vop_pathconf = zfs_freebsd_pathconf,
6886 .vop_bmap = zfs_freebsd_bmap,
6887 .vop_fid = zfs_freebsd_fid,
6888 .vop_getextattr = zfs_getextattr,
6889 .vop_deleteextattr = zfs_deleteextattr,
6890 .vop_setextattr = zfs_setextattr,
6891 .vop_listextattr = zfs_listextattr,
6892 .vop_getacl = zfs_freebsd_getacl,
6893 .vop_setacl = zfs_freebsd_setacl,
6894 .vop_aclcheck = zfs_freebsd_aclcheck,
6895 .vop_getpages = zfs_freebsd_getpages,
6898 struct vop_vector zfs_fifoops = {
6899 .vop_default = &fifo_specops,
6900 .vop_fsync = zfs_freebsd_fsync,
6901 .vop_access = zfs_freebsd_access,
6902 .vop_getattr = zfs_freebsd_getattr,
6903 .vop_inactive = zfs_freebsd_inactive,
6904 .vop_read = VOP_PANIC,
6905 .vop_reclaim = zfs_freebsd_reclaim,
6906 .vop_setattr = zfs_freebsd_setattr,
6907 .vop_write = VOP_PANIC,
6908 .vop_pathconf = zfs_freebsd_fifo_pathconf,
6909 .vop_fid = zfs_freebsd_fid,
6910 .vop_getacl = zfs_freebsd_getacl,
6911 .vop_setacl = zfs_freebsd_setacl,
6912 .vop_aclcheck = zfs_freebsd_aclcheck,
6916 * special share hidden files vnode operations template
6918 struct vop_vector zfs_shareops = {
6919 .vop_default = &default_vnodeops,
6920 .vop_access = zfs_freebsd_access,
6921 .vop_inactive = zfs_freebsd_inactive,
6922 .vop_reclaim = zfs_freebsd_reclaim,
6923 .vop_fid = zfs_freebsd_fid,
6924 .vop_pathconf = zfs_freebsd_pathconf,