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.
24 * Copyright 2013 Nexenta Systems, Inc. All rights reserved.
27 /* Portions Copyright 2007 Jeremy Teo */
28 /* Portions Copyright 2010 Robert Milkowski */
30 #include <sys/types.h>
31 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/sysmacros.h>
35 #include <sys/resource.h>
38 #include <sys/vnode.h>
42 #include <sys/taskq.h>
44 #include <sys/atomic.h>
45 #include <sys/namei.h>
47 #include <sys/cmn_err.h>
48 #include <sys/errno.h>
49 #include <sys/unistd.h>
50 #include <sys/zfs_dir.h>
51 #include <sys/zfs_ioctl.h>
52 #include <sys/fs/zfs.h>
54 #include <sys/dmu_objset.h>
60 #include <sys/dirent.h>
61 #include <sys/policy.h>
62 #include <sys/sunddi.h>
63 #include <sys/filio.h>
65 #include <sys/zfs_ctldir.h>
66 #include <sys/zfs_fuid.h>
67 #include <sys/zfs_sa.h>
69 #include <sys/zfs_rlock.h>
70 #include <sys/extdirent.h>
71 #include <sys/kidmap.h>
74 #include <sys/sf_buf.h>
75 #include <sys/sched.h>
77 #include <vm/vm_param.h>
78 #include <vm/vm_pageout.h>
83 * Each vnode op performs some logical unit of work. To do this, the ZPL must
84 * properly lock its in-core state, create a DMU transaction, do the work,
85 * record this work in the intent log (ZIL), commit the DMU transaction,
86 * and wait for the intent log to commit if it is a synchronous operation.
87 * Moreover, the vnode ops must work in both normal and log replay context.
88 * The ordering of events is important to avoid deadlocks and references
89 * to freed memory. The example below illustrates the following Big Rules:
91 * (1) A check must be made in each zfs thread for a mounted file system.
92 * This is done avoiding races using ZFS_ENTER(zfsvfs).
93 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
94 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
95 * can return EIO from the calling function.
97 * (2) VN_RELE() should always be the last thing except for zil_commit()
98 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
99 * First, if it's the last reference, the vnode/znode
100 * can be freed, so the zp may point to freed memory. Second, the last
101 * reference will call zfs_zinactive(), which may induce a lot of work --
102 * pushing cached pages (which acquires range locks) and syncing out
103 * cached atime changes. Third, zfs_zinactive() may require a new tx,
104 * which could deadlock the system if you were already holding one.
105 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
107 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
108 * as they can span dmu_tx_assign() calls.
110 * (4) Always pass TXG_NOWAIT as the second argument to dmu_tx_assign().
111 * This is critical because we don't want to block while holding locks.
112 * Note, in particular, that if a lock is sometimes acquired before
113 * the tx assigns, and sometimes after (e.g. z_lock), then failing to
114 * use a non-blocking assign can deadlock the system. The scenario:
116 * Thread A has grabbed a lock before calling dmu_tx_assign().
117 * Thread B is in an already-assigned tx, and blocks for this lock.
118 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
119 * forever, because the previous txg can't quiesce until B's tx commits.
121 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
122 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
123 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
124 * to indicate that this operation has already called dmu_tx_wait().
125 * This will ensure that we don't retry forever, waiting a short bit
128 * (5) If the operation succeeded, generate the intent log entry for it
129 * before dropping locks. This ensures that the ordering of events
130 * in the intent log matches the order in which they actually occurred.
131 * During ZIL replay the zfs_log_* functions will update the sequence
132 * number to indicate the zil transaction has replayed.
134 * (6) At the end of each vnode op, the DMU tx must always commit,
135 * regardless of whether there were any errors.
137 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
138 * to ensure that synchronous semantics are provided when necessary.
140 * In general, this is how things should be ordered in each vnode op:
142 * ZFS_ENTER(zfsvfs); // exit if unmounted
144 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD())
145 * rw_enter(...); // grab any other locks you need
146 * tx = dmu_tx_create(...); // get DMU tx
147 * dmu_tx_hold_*(); // hold each object you might modify
148 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
150 * rw_exit(...); // drop locks
151 * zfs_dirent_unlock(dl); // unlock directory entry
152 * VN_RELE(...); // release held vnodes
153 * if (error == ERESTART) {
159 * dmu_tx_abort(tx); // abort DMU tx
160 * ZFS_EXIT(zfsvfs); // finished in zfs
161 * return (error); // really out of space
163 * error = do_real_work(); // do whatever this VOP does
165 * zfs_log_*(...); // on success, make ZIL entry
166 * dmu_tx_commit(tx); // commit DMU tx -- error or not
167 * rw_exit(...); // drop locks
168 * zfs_dirent_unlock(dl); // unlock directory entry
169 * VN_RELE(...); // release held vnodes
170 * zil_commit(zilog, foid); // synchronous when necessary
171 * ZFS_EXIT(zfsvfs); // finished in zfs
172 * return (error); // done, report error
177 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
179 znode_t *zp = VTOZ(*vpp);
180 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
185 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
186 ((flag & FAPPEND) == 0)) {
188 return (SET_ERROR(EPERM));
191 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
192 ZTOV(zp)->v_type == VREG &&
193 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
194 if (fs_vscan(*vpp, cr, 0) != 0) {
196 return (SET_ERROR(EACCES));
200 /* Keep a count of the synchronous opens in the znode */
201 if (flag & (FSYNC | FDSYNC))
202 atomic_inc_32(&zp->z_sync_cnt);
210 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
211 caller_context_t *ct)
213 znode_t *zp = VTOZ(vp);
214 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
217 * Clean up any locks held by this process on the vp.
219 cleanlocks(vp, ddi_get_pid(), 0);
220 cleanshares(vp, ddi_get_pid());
225 /* Decrement the synchronous opens in the znode */
226 if ((flag & (FSYNC | FDSYNC)) && (count == 1))
227 atomic_dec_32(&zp->z_sync_cnt);
229 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
230 ZTOV(zp)->v_type == VREG &&
231 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
232 VERIFY(fs_vscan(vp, cr, 1) == 0);
239 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
240 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
243 zfs_holey(vnode_t *vp, u_long cmd, offset_t *off)
245 znode_t *zp = VTOZ(vp);
246 uint64_t noff = (uint64_t)*off; /* new offset */
251 file_sz = zp->z_size;
252 if (noff >= file_sz) {
253 return (SET_ERROR(ENXIO));
256 if (cmd == _FIO_SEEK_HOLE)
261 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
264 if ((error == ESRCH) || (noff > file_sz)) {
266 * Handle the virtual hole at the end of file.
272 return (SET_ERROR(ENXIO));
283 zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred,
284 int *rvalp, caller_context_t *ct)
296 * The following two ioctls are used by bfu. Faking out,
297 * necessary to avoid bfu errors.
306 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
307 return (SET_ERROR(EFAULT));
309 off = *(offset_t *)data;
312 zfsvfs = zp->z_zfsvfs;
316 /* offset parameter is in/out */
317 error = zfs_holey(vp, com, &off);
322 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
323 return (SET_ERROR(EFAULT));
325 *(offset_t *)data = off;
329 return (SET_ERROR(ENOTTY));
333 page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes)
340 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE
341 * aligned boundaries, if the range is not aligned. As a result a
342 * DEV_BSIZE subrange with partially dirty data may get marked as clean.
343 * It may happen that all DEV_BSIZE subranges are marked clean and thus
344 * the whole page would be considred clean despite have some dirty data.
345 * For this reason we should shrink the range to DEV_BSIZE aligned
346 * boundaries before calling vm_page_clear_dirty.
348 end = rounddown2(off + nbytes, DEV_BSIZE);
349 off = roundup2(off, DEV_BSIZE);
353 zfs_vmobject_assert_wlocked(obj);
356 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
358 if (vm_page_xbusied(pp)) {
360 * Reference the page before unlocking and
361 * sleeping so that the page daemon is less
362 * likely to reclaim it.
364 vm_page_reference(pp);
366 zfs_vmobject_wunlock(obj);
367 vm_page_busy_sleep(pp, "zfsmwb");
368 zfs_vmobject_wlock(obj);
372 } else if (pp == NULL) {
373 pp = vm_page_alloc(obj, OFF_TO_IDX(start),
374 VM_ALLOC_SYSTEM | VM_ALLOC_IFCACHED |
377 ASSERT(pp != NULL && !pp->valid);
382 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
383 vm_object_pip_add(obj, 1);
384 pmap_remove_write(pp);
386 vm_page_clear_dirty(pp, off, nbytes);
394 page_unbusy(vm_page_t pp)
398 vm_object_pip_subtract(pp->object, 1);
402 page_hold(vnode_t *vp, int64_t start)
408 zfs_vmobject_assert_wlocked(obj);
411 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
413 if (vm_page_xbusied(pp)) {
415 * Reference the page before unlocking and
416 * sleeping so that the page daemon is less
417 * likely to reclaim it.
419 vm_page_reference(pp);
421 zfs_vmobject_wunlock(obj);
422 vm_page_busy_sleep(pp, "zfsmwb");
423 zfs_vmobject_wlock(obj);
427 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
440 page_unhold(vm_page_t pp)
449 zfs_map_page(vm_page_t pp, struct sf_buf **sfp)
452 *sfp = sf_buf_alloc(pp, 0);
453 return ((caddr_t)sf_buf_kva(*sfp));
457 zfs_unmap_page(struct sf_buf *sf)
464 * When a file is memory mapped, we must keep the IO data synchronized
465 * between the DMU cache and the memory mapped pages. What this means:
467 * On Write: If we find a memory mapped page, we write to *both*
468 * the page and the dmu buffer.
471 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid,
472 int segflg, dmu_tx_t *tx)
479 ASSERT(vp->v_mount != NULL);
483 off = start & PAGEOFFSET;
484 zfs_vmobject_wlock(obj);
485 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
487 int nbytes = imin(PAGESIZE - off, len);
489 if (segflg == UIO_NOCOPY) {
490 pp = vm_page_lookup(obj, OFF_TO_IDX(start));
492 ("zfs update_pages: NULL page in putpages case"));
494 ("zfs update_pages: unaligned data in putpages case"));
495 KASSERT(pp->valid == VM_PAGE_BITS_ALL,
496 ("zfs update_pages: invalid page in putpages case"));
497 KASSERT(vm_page_sbusied(pp),
498 ("zfs update_pages: unbusy page in putpages case"));
499 KASSERT(!pmap_page_is_write_mapped(pp),
500 ("zfs update_pages: writable page in putpages case"));
501 zfs_vmobject_wunlock(obj);
503 va = zfs_map_page(pp, &sf);
504 (void) dmu_write(os, oid, start, nbytes, va, tx);
507 zfs_vmobject_wlock(obj);
509 } else if ((pp = page_busy(vp, start, off, nbytes)) != NULL) {
510 zfs_vmobject_wunlock(obj);
512 va = zfs_map_page(pp, &sf);
513 (void) dmu_read(os, oid, start+off, nbytes,
514 va+off, DMU_READ_PREFETCH);;
517 zfs_vmobject_wlock(obj);
523 if (segflg != UIO_NOCOPY)
524 vm_object_pip_wakeupn(obj, 0);
525 zfs_vmobject_wunlock(obj);
529 * Read with UIO_NOCOPY flag means that sendfile(2) requests
530 * ZFS to populate a range of page cache pages with data.
532 * NOTE: this function could be optimized to pre-allocate
533 * all pages in advance, drain exclusive busy on all of them,
534 * map them into contiguous KVA region and populate them
535 * in one single dmu_read() call.
538 mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio)
540 znode_t *zp = VTOZ(vp);
541 objset_t *os = zp->z_zfsvfs->z_os;
551 ASSERT(uio->uio_segflg == UIO_NOCOPY);
552 ASSERT(vp->v_mount != NULL);
555 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0);
557 zfs_vmobject_wlock(obj);
558 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) {
559 int bytes = MIN(PAGESIZE, len);
561 pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_SBUSY |
562 VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY);
563 if (pp->valid == 0) {
564 zfs_vmobject_wunlock(obj);
565 va = zfs_map_page(pp, &sf);
566 error = dmu_read(os, zp->z_id, start, bytes, va,
568 if (bytes != PAGESIZE && error == 0)
569 bzero(va + bytes, PAGESIZE - bytes);
571 zfs_vmobject_wlock(obj);
575 if (pp->wire_count == 0 && pp->valid == 0 &&
579 pp->valid = VM_PAGE_BITS_ALL;
580 vm_page_activate(pp);
584 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
589 uio->uio_resid -= bytes;
590 uio->uio_offset += bytes;
593 zfs_vmobject_wunlock(obj);
598 * When a file is memory mapped, we must keep the IO data synchronized
599 * between the DMU cache and the memory mapped pages. What this means:
601 * On Read: We "read" preferentially from memory mapped pages,
602 * else we default from the dmu buffer.
604 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
605 * the file is memory mapped.
608 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
610 znode_t *zp = VTOZ(vp);
611 objset_t *os = zp->z_zfsvfs->z_os;
619 ASSERT(vp->v_mount != NULL);
623 start = uio->uio_loffset;
624 off = start & PAGEOFFSET;
625 zfs_vmobject_wlock(obj);
626 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
628 uint64_t bytes = MIN(PAGESIZE - off, len);
630 if (pp = page_hold(vp, start)) {
634 zfs_vmobject_wunlock(obj);
635 va = zfs_map_page(pp, &sf);
636 error = uiomove(va + off, bytes, UIO_READ, uio);
638 zfs_vmobject_wlock(obj);
641 zfs_vmobject_wunlock(obj);
642 error = dmu_read_uio(os, zp->z_id, uio, bytes);
643 zfs_vmobject_wlock(obj);
650 zfs_vmobject_wunlock(obj);
654 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
657 * Read bytes from specified file into supplied buffer.
659 * IN: vp - vnode of file to be read from.
660 * uio - structure supplying read location, range info,
662 * ioflag - SYNC flags; used to provide FRSYNC semantics.
663 * cr - credentials of caller.
664 * ct - caller context
666 * OUT: uio - updated offset and range, buffer filled.
668 * RETURN: 0 on success, error code on failure.
671 * vp - atime updated if byte count > 0
675 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
677 znode_t *zp = VTOZ(vp);
678 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
689 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
691 return (SET_ERROR(EACCES));
695 * Validate file offset
697 if (uio->uio_loffset < (offset_t)0) {
699 return (SET_ERROR(EINVAL));
703 * Fasttrack empty reads
705 if (uio->uio_resid == 0) {
711 * Check for mandatory locks
713 if (MANDMODE(zp->z_mode)) {
714 if (error = chklock(vp, FREAD,
715 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
722 * If we're in FRSYNC mode, sync out this znode before reading it.
725 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
726 zil_commit(zfsvfs->z_log, zp->z_id);
729 * Lock the range against changes.
731 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
734 * If we are reading past end-of-file we can skip
735 * to the end; but we might still need to set atime.
737 if (uio->uio_loffset >= zp->z_size) {
742 ASSERT(uio->uio_loffset < zp->z_size);
743 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
746 if ((uio->uio_extflg == UIO_XUIO) &&
747 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
749 int blksz = zp->z_blksz;
750 uint64_t offset = uio->uio_loffset;
752 xuio = (xuio_t *)uio;
754 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
757 ASSERT(offset + n <= blksz);
760 (void) dmu_xuio_init(xuio, nblk);
762 if (vn_has_cached_data(vp)) {
764 * For simplicity, we always allocate a full buffer
765 * even if we only expect to read a portion of a block.
767 while (--nblk >= 0) {
768 (void) dmu_xuio_add(xuio,
769 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
777 nbytes = MIN(n, zfs_read_chunk_size -
778 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
781 if (uio->uio_segflg == UIO_NOCOPY)
782 error = mappedread_sf(vp, nbytes, uio);
784 #endif /* __FreeBSD__ */
785 if (vn_has_cached_data(vp))
786 error = mappedread(vp, nbytes, uio);
788 error = dmu_read_uio(os, zp->z_id, uio, nbytes);
790 /* convert checksum errors into IO errors */
792 error = SET_ERROR(EIO);
799 zfs_range_unlock(rl);
801 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
807 * Write the bytes to a file.
809 * IN: vp - vnode of file to be written to.
810 * uio - structure supplying write location, range info,
812 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is
813 * set if in append mode.
814 * cr - credentials of caller.
815 * ct - caller context (NFS/CIFS fem monitor only)
817 * OUT: uio - updated offset and range.
819 * RETURN: 0 on success, error code on failure.
822 * vp - ctime|mtime updated if byte count > 0
827 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
829 znode_t *zp = VTOZ(vp);
830 rlim64_t limit = MAXOFFSET_T;
831 ssize_t start_resid = uio->uio_resid;
835 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
840 int max_blksz = zfsvfs->z_max_blksz;
843 iovec_t *aiov = NULL;
846 int iovcnt = uio->uio_iovcnt;
847 iovec_t *iovp = uio->uio_iov;
850 sa_bulk_attr_t bulk[4];
851 uint64_t mtime[2], ctime[2];
854 * Fasttrack empty write
860 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
866 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
867 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
868 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
870 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
874 * If immutable or not appending then return EPERM
876 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
877 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
878 (uio->uio_loffset < zp->z_size))) {
880 return (SET_ERROR(EPERM));
883 zilog = zfsvfs->z_log;
886 * Validate file offset
888 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
891 return (SET_ERROR(EINVAL));
895 * Check for mandatory locks before calling zfs_range_lock()
896 * in order to prevent a deadlock with locks set via fcntl().
898 if (MANDMODE((mode_t)zp->z_mode) &&
899 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
906 * Pre-fault the pages to ensure slow (eg NFS) pages
908 * Skip this if uio contains loaned arc_buf.
910 if ((uio->uio_extflg == UIO_XUIO) &&
911 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
912 xuio = (xuio_t *)uio;
914 uio_prefaultpages(MIN(n, max_blksz), uio);
918 * If in append mode, set the io offset pointer to eof.
920 if (ioflag & FAPPEND) {
922 * Obtain an appending range lock to guarantee file append
923 * semantics. We reset the write offset once we have the lock.
925 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
927 if (rl->r_len == UINT64_MAX) {
929 * We overlocked the file because this write will cause
930 * the file block size to increase.
931 * Note that zp_size cannot change with this lock held.
935 uio->uio_loffset = woff;
938 * Note that if the file block size will change as a result of
939 * this write, then this range lock will lock the entire file
940 * so that we can re-write the block safely.
942 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
945 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) {
946 zfs_range_unlock(rl);
952 zfs_range_unlock(rl);
954 return (SET_ERROR(EFBIG));
957 if ((woff + n) > limit || woff > (limit - n))
960 /* Will this write extend the file length? */
961 write_eof = (woff + n > zp->z_size);
963 end_size = MAX(zp->z_size, woff + n);
966 * Write the file in reasonable size chunks. Each chunk is written
967 * in a separate transaction; this keeps the intent log records small
968 * and allows us to do more fine-grained space accounting.
972 woff = uio->uio_loffset;
974 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
975 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
977 dmu_return_arcbuf(abuf);
978 error = SET_ERROR(EDQUOT);
982 if (xuio && abuf == NULL) {
983 ASSERT(i_iov < iovcnt);
985 abuf = dmu_xuio_arcbuf(xuio, i_iov);
986 dmu_xuio_clear(xuio, i_iov);
987 DTRACE_PROBE3(zfs_cp_write, int, i_iov,
988 iovec_t *, aiov, arc_buf_t *, abuf);
989 ASSERT((aiov->iov_base == abuf->b_data) ||
990 ((char *)aiov->iov_base - (char *)abuf->b_data +
991 aiov->iov_len == arc_buf_size(abuf)));
993 } else if (abuf == NULL && n >= max_blksz &&
994 woff >= zp->z_size &&
995 P2PHASE(woff, max_blksz) == 0 &&
996 zp->z_blksz == max_blksz) {
998 * This write covers a full block. "Borrow" a buffer
999 * from the dmu so that we can fill it before we enter
1000 * a transaction. This avoids the possibility of
1001 * holding up the transaction if the data copy hangs
1002 * up on a pagefault (e.g., from an NFS server mapping).
1006 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
1008 ASSERT(abuf != NULL);
1009 ASSERT(arc_buf_size(abuf) == max_blksz);
1010 if (error = uiocopy(abuf->b_data, max_blksz,
1011 UIO_WRITE, uio, &cbytes)) {
1012 dmu_return_arcbuf(abuf);
1015 ASSERT(cbytes == max_blksz);
1019 * Start a transaction.
1021 tx = dmu_tx_create(zfsvfs->z_os);
1022 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1023 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
1024 zfs_sa_upgrade_txholds(tx, zp);
1025 error = dmu_tx_assign(tx, TXG_NOWAIT);
1027 if (error == ERESTART) {
1034 dmu_return_arcbuf(abuf);
1039 * If zfs_range_lock() over-locked we grow the blocksize
1040 * and then reduce the lock range. This will only happen
1041 * on the first iteration since zfs_range_reduce() will
1042 * shrink down r_len to the appropriate size.
1044 if (rl->r_len == UINT64_MAX) {
1047 if (zp->z_blksz > max_blksz) {
1048 ASSERT(!ISP2(zp->z_blksz));
1049 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
1051 new_blksz = MIN(end_size, max_blksz);
1053 zfs_grow_blocksize(zp, new_blksz, tx);
1054 zfs_range_reduce(rl, woff, n);
1058 * XXX - should we really limit each write to z_max_blksz?
1059 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
1061 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
1063 if (woff + nbytes > zp->z_size)
1064 vnode_pager_setsize(vp, woff + nbytes);
1067 tx_bytes = uio->uio_resid;
1068 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
1070 tx_bytes -= uio->uio_resid;
1073 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
1075 * If this is not a full block write, but we are
1076 * extending the file past EOF and this data starts
1077 * block-aligned, use assign_arcbuf(). Otherwise,
1078 * write via dmu_write().
1080 if (tx_bytes < max_blksz && (!write_eof ||
1081 aiov->iov_base != abuf->b_data)) {
1083 dmu_write(zfsvfs->z_os, zp->z_id, woff,
1084 aiov->iov_len, aiov->iov_base, tx);
1085 dmu_return_arcbuf(abuf);
1086 xuio_stat_wbuf_copied();
1088 ASSERT(xuio || tx_bytes == max_blksz);
1089 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
1092 ASSERT(tx_bytes <= uio->uio_resid);
1093 uioskip(uio, tx_bytes);
1095 if (tx_bytes && vn_has_cached_data(vp)) {
1096 update_pages(vp, woff, tx_bytes, zfsvfs->z_os,
1097 zp->z_id, uio->uio_segflg, tx);
1101 * If we made no progress, we're done. If we made even
1102 * partial progress, update the znode and ZIL accordingly.
1104 if (tx_bytes == 0) {
1105 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
1106 (void *)&zp->z_size, sizeof (uint64_t), tx);
1113 * Clear Set-UID/Set-GID bits on successful write if not
1114 * privileged and at least one of the excute bits is set.
1116 * It would be nice to to this after all writes have
1117 * been done, but that would still expose the ISUID/ISGID
1118 * to another app after the partial write is committed.
1120 * Note: we don't call zfs_fuid_map_id() here because
1121 * user 0 is not an ephemeral uid.
1123 mutex_enter(&zp->z_acl_lock);
1124 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
1125 (S_IXUSR >> 6))) != 0 &&
1126 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
1127 secpolicy_vnode_setid_retain(vp, cr,
1128 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
1130 zp->z_mode &= ~(S_ISUID | S_ISGID);
1131 newmode = zp->z_mode;
1132 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
1133 (void *)&newmode, sizeof (uint64_t), tx);
1135 mutex_exit(&zp->z_acl_lock);
1137 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
1141 * Update the file size (zp_size) if it has changed;
1142 * account for possible concurrent updates.
1144 while ((end_size = zp->z_size) < uio->uio_loffset) {
1145 (void) atomic_cas_64(&zp->z_size, end_size,
1150 * If we are replaying and eof is non zero then force
1151 * the file size to the specified eof. Note, there's no
1152 * concurrency during replay.
1154 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
1155 zp->z_size = zfsvfs->z_replay_eof;
1157 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1159 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
1164 ASSERT(tx_bytes == nbytes);
1169 uio_prefaultpages(MIN(n, max_blksz), uio);
1173 zfs_range_unlock(rl);
1176 * If we're in replay mode, or we made no progress, return error.
1177 * Otherwise, it's at least a partial write, so it's successful.
1179 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1184 if (ioflag & (FSYNC | FDSYNC) ||
1185 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1186 zil_commit(zilog, zp->z_id);
1193 zfs_get_done(zgd_t *zgd, int error)
1195 znode_t *zp = zgd->zgd_private;
1196 objset_t *os = zp->z_zfsvfs->z_os;
1199 dmu_buf_rele(zgd->zgd_db, zgd);
1201 zfs_range_unlock(zgd->zgd_rl);
1204 * Release the vnode asynchronously as we currently have the
1205 * txg stopped from syncing.
1207 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1209 if (error == 0 && zgd->zgd_bp)
1210 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1212 kmem_free(zgd, sizeof (zgd_t));
1216 static int zil_fault_io = 0;
1220 * Get data to generate a TX_WRITE intent log record.
1223 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1225 zfsvfs_t *zfsvfs = arg;
1226 objset_t *os = zfsvfs->z_os;
1228 uint64_t object = lr->lr_foid;
1229 uint64_t offset = lr->lr_offset;
1230 uint64_t size = lr->lr_length;
1231 blkptr_t *bp = &lr->lr_blkptr;
1236 ASSERT(zio != NULL);
1240 * Nothing to do if the file has been removed
1242 if (zfs_zget(zfsvfs, object, &zp) != 0)
1243 return (SET_ERROR(ENOENT));
1244 if (zp->z_unlinked) {
1246 * Release the vnode asynchronously as we currently have the
1247 * txg stopped from syncing.
1249 VN_RELE_ASYNC(ZTOV(zp),
1250 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1251 return (SET_ERROR(ENOENT));
1254 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1255 zgd->zgd_zilog = zfsvfs->z_log;
1256 zgd->zgd_private = zp;
1259 * Write records come in two flavors: immediate and indirect.
1260 * For small writes it's cheaper to store the data with the
1261 * log record (immediate); for large writes it's cheaper to
1262 * sync the data and get a pointer to it (indirect) so that
1263 * we don't have to write the data twice.
1265 if (buf != NULL) { /* immediate write */
1266 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1267 /* test for truncation needs to be done while range locked */
1268 if (offset >= zp->z_size) {
1269 error = SET_ERROR(ENOENT);
1271 error = dmu_read(os, object, offset, size, buf,
1272 DMU_READ_NO_PREFETCH);
1274 ASSERT(error == 0 || error == ENOENT);
1275 } else { /* indirect write */
1277 * Have to lock the whole block to ensure when it's
1278 * written out and it's checksum is being calculated
1279 * that no one can change the data. We need to re-check
1280 * blocksize after we get the lock in case it's changed!
1285 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1287 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1289 if (zp->z_blksz == size)
1292 zfs_range_unlock(zgd->zgd_rl);
1294 /* test for truncation needs to be done while range locked */
1295 if (lr->lr_offset >= zp->z_size)
1296 error = SET_ERROR(ENOENT);
1299 error = SET_ERROR(EIO);
1304 error = dmu_buf_hold(os, object, offset, zgd, &db,
1305 DMU_READ_NO_PREFETCH);
1308 blkptr_t *obp = dmu_buf_get_blkptr(db);
1310 ASSERT(BP_IS_HOLE(bp));
1317 ASSERT(db->db_offset == offset);
1318 ASSERT(db->db_size == size);
1320 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1322 ASSERT(error || lr->lr_length <= zp->z_blksz);
1325 * On success, we need to wait for the write I/O
1326 * initiated by dmu_sync() to complete before we can
1327 * release this dbuf. We will finish everything up
1328 * in the zfs_get_done() callback.
1333 if (error == EALREADY) {
1334 lr->lr_common.lrc_txtype = TX_WRITE2;
1340 zfs_get_done(zgd, error);
1347 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1348 caller_context_t *ct)
1350 znode_t *zp = VTOZ(vp);
1351 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1357 if (flag & V_ACE_MASK)
1358 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1360 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1367 * If vnode is for a device return a specfs vnode instead.
1370 specvp_check(vnode_t **vpp, cred_t *cr)
1374 if (IS_DEVVP(*vpp)) {
1377 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1380 error = SET_ERROR(ENOSYS);
1388 * Lookup an entry in a directory, or an extended attribute directory.
1389 * If it exists, return a held vnode reference for it.
1391 * IN: dvp - vnode of directory to search.
1392 * nm - name of entry to lookup.
1393 * pnp - full pathname to lookup [UNUSED].
1394 * flags - LOOKUP_XATTR set if looking for an attribute.
1395 * rdir - root directory vnode [UNUSED].
1396 * cr - credentials of caller.
1397 * ct - caller context
1398 * direntflags - directory lookup flags
1399 * realpnp - returned pathname.
1401 * OUT: vpp - vnode of located entry, NULL if not found.
1403 * RETURN: 0 on success, error code on failure.
1410 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp,
1411 int nameiop, cred_t *cr, kthread_t *td, int flags)
1413 znode_t *zdp = VTOZ(dvp);
1414 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1416 int *direntflags = NULL;
1417 void *realpnp = NULL;
1420 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1422 if (dvp->v_type != VDIR) {
1423 return (SET_ERROR(ENOTDIR));
1424 } else if (zdp->z_sa_hdl == NULL) {
1425 return (SET_ERROR(EIO));
1428 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1429 error = zfs_fastaccesschk_execute(zdp, cr);
1437 vnode_t *tvp = dnlc_lookup(dvp, nm);
1440 error = zfs_fastaccesschk_execute(zdp, cr);
1445 if (tvp == DNLC_NO_VNODE) {
1447 return (SET_ERROR(ENOENT));
1450 return (specvp_check(vpp, cr));
1456 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1463 if (flags & LOOKUP_XATTR) {
1466 * If the xattr property is off, refuse the lookup request.
1468 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1470 return (SET_ERROR(EINVAL));
1475 * We don't allow recursive attributes..
1476 * Maybe someday we will.
1478 if (zdp->z_pflags & ZFS_XATTR) {
1480 return (SET_ERROR(EINVAL));
1483 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1489 * Do we have permission to get into attribute directory?
1492 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1502 if (dvp->v_type != VDIR) {
1504 return (SET_ERROR(ENOTDIR));
1508 * Check accessibility of directory.
1511 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1516 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1517 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1519 return (SET_ERROR(EILSEQ));
1522 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1524 error = specvp_check(vpp, cr);
1526 /* Translate errors and add SAVENAME when needed. */
1527 if (cnp->cn_flags & ISLASTCN) {
1531 if (error == ENOENT) {
1532 error = EJUSTRETURN;
1533 cnp->cn_flags |= SAVENAME;
1539 cnp->cn_flags |= SAVENAME;
1543 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) {
1546 if (cnp->cn_flags & ISDOTDOT) {
1547 ltype = VOP_ISLOCKED(dvp);
1551 error = vn_lock(*vpp, cnp->cn_lkflags);
1552 if (cnp->cn_flags & ISDOTDOT)
1553 vn_lock(dvp, ltype | LK_RETRY);
1563 #ifdef FREEBSD_NAMECACHE
1565 * Insert name into cache (as non-existent) if appropriate.
1567 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
1568 cache_enter(dvp, *vpp, cnp);
1570 * Insert name into cache if appropriate.
1572 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) {
1573 if (!(cnp->cn_flags & ISLASTCN) ||
1574 (nameiop != DELETE && nameiop != RENAME)) {
1575 cache_enter(dvp, *vpp, cnp);
1584 * Attempt to create a new entry in a directory. If the entry
1585 * already exists, truncate the file if permissible, else return
1586 * an error. Return the vp of the created or trunc'd file.
1588 * IN: dvp - vnode of directory to put new file entry in.
1589 * name - name of new file entry.
1590 * vap - attributes of new file.
1591 * excl - flag indicating exclusive or non-exclusive mode.
1592 * mode - mode to open file with.
1593 * cr - credentials of caller.
1594 * flag - large file flag [UNUSED].
1595 * ct - caller context
1596 * vsecp - ACL to be set
1598 * OUT: vpp - vnode of created or trunc'd entry.
1600 * RETURN: 0 on success, error code on failure.
1603 * dvp - ctime|mtime updated if new entry created
1604 * vp - ctime|mtime always, atime if new
1609 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode,
1610 vnode_t **vpp, cred_t *cr, kthread_t *td)
1612 znode_t *zp, *dzp = VTOZ(dvp);
1613 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1621 gid_t gid = crgetgid(cr);
1622 zfs_acl_ids_t acl_ids;
1623 boolean_t fuid_dirtied;
1624 boolean_t have_acl = B_FALSE;
1625 boolean_t waited = B_FALSE;
1630 * If we have an ephemeral id, ACL, or XVATTR then
1631 * make sure file system is at proper version
1634 ksid = crgetsid(cr, KSID_OWNER);
1636 uid = ksid_getid(ksid);
1640 if (zfsvfs->z_use_fuids == B_FALSE &&
1641 (vsecp || (vap->va_mask & AT_XVATTR) ||
1642 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1643 return (SET_ERROR(EINVAL));
1648 zilog = zfsvfs->z_log;
1650 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1651 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1653 return (SET_ERROR(EILSEQ));
1656 if (vap->va_mask & AT_XVATTR) {
1657 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
1658 crgetuid(cr), cr, vap->va_type)) != 0) {
1666 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr))
1667 vap->va_mode &= ~S_ISVTX;
1669 if (*name == '\0') {
1671 * Null component name refers to the directory itself.
1678 /* possible VN_HOLD(zp) */
1681 if (flag & FIGNORECASE)
1684 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1688 zfs_acl_ids_free(&acl_ids);
1689 if (strcmp(name, "..") == 0)
1690 error = SET_ERROR(EISDIR);
1700 * Create a new file object and update the directory
1703 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1705 zfs_acl_ids_free(&acl_ids);
1710 * We only support the creation of regular files in
1711 * extended attribute directories.
1714 if ((dzp->z_pflags & ZFS_XATTR) &&
1715 (vap->va_type != VREG)) {
1717 zfs_acl_ids_free(&acl_ids);
1718 error = SET_ERROR(EINVAL);
1722 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1723 cr, vsecp, &acl_ids)) != 0)
1727 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1728 zfs_acl_ids_free(&acl_ids);
1729 error = SET_ERROR(EDQUOT);
1733 tx = dmu_tx_create(os);
1735 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1736 ZFS_SA_BASE_ATTR_SIZE);
1738 fuid_dirtied = zfsvfs->z_fuid_dirty;
1740 zfs_fuid_txhold(zfsvfs, tx);
1741 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1742 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1743 if (!zfsvfs->z_use_sa &&
1744 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1745 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1746 0, acl_ids.z_aclp->z_acl_bytes);
1748 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1750 zfs_dirent_unlock(dl);
1751 if (error == ERESTART) {
1757 zfs_acl_ids_free(&acl_ids);
1762 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1765 zfs_fuid_sync(zfsvfs, tx);
1767 (void) zfs_link_create(dl, zp, tx, ZNEW);
1768 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1769 if (flag & FIGNORECASE)
1771 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1772 vsecp, acl_ids.z_fuidp, vap);
1773 zfs_acl_ids_free(&acl_ids);
1776 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1779 zfs_acl_ids_free(&acl_ids);
1783 * A directory entry already exists for this name.
1786 * Can't truncate an existing file if in exclusive mode.
1789 error = SET_ERROR(EEXIST);
1793 * Can't open a directory for writing.
1795 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1796 error = SET_ERROR(EISDIR);
1800 * Verify requested access to file.
1802 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1806 mutex_enter(&dzp->z_lock);
1808 mutex_exit(&dzp->z_lock);
1811 * Truncate regular files if requested.
1813 if ((ZTOV(zp)->v_type == VREG) &&
1814 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1815 /* we can't hold any locks when calling zfs_freesp() */
1816 zfs_dirent_unlock(dl);
1818 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1820 vnevent_create(ZTOV(zp), ct);
1826 zfs_dirent_unlock(dl);
1833 error = specvp_check(vpp, cr);
1836 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1837 zil_commit(zilog, 0);
1844 * Remove an entry from a directory.
1846 * IN: dvp - vnode of directory to remove entry from.
1847 * name - name of entry to remove.
1848 * cr - credentials of caller.
1849 * ct - caller context
1850 * flags - case flags
1852 * RETURN: 0 on success, error code on failure.
1856 * vp - ctime (if nlink > 0)
1859 uint64_t null_xattr = 0;
1863 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1866 znode_t *zp, *dzp = VTOZ(dvp);
1869 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1871 uint64_t acl_obj, xattr_obj;
1872 uint64_t xattr_obj_unlinked = 0;
1876 boolean_t may_delete_now, delete_now = FALSE;
1877 boolean_t unlinked, toobig = FALSE;
1879 pathname_t *realnmp = NULL;
1883 boolean_t waited = B_FALSE;
1887 zilog = zfsvfs->z_log;
1889 if (flags & FIGNORECASE) {
1899 * Attempt to lock directory; fail if entry doesn't exist.
1901 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1911 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1916 * Need to use rmdir for removing directories.
1918 if (vp->v_type == VDIR) {
1919 error = SET_ERROR(EPERM);
1923 vnevent_remove(vp, dvp, name, ct);
1926 dnlc_remove(dvp, realnmp->pn_buf);
1928 dnlc_remove(dvp, name);
1931 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1935 * We may delete the znode now, or we may put it in the unlinked set;
1936 * it depends on whether we're the last link, and on whether there are
1937 * other holds on the vnode. So we dmu_tx_hold() the right things to
1938 * allow for either case.
1941 tx = dmu_tx_create(zfsvfs->z_os);
1942 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1943 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1944 zfs_sa_upgrade_txholds(tx, zp);
1945 zfs_sa_upgrade_txholds(tx, dzp);
1946 if (may_delete_now) {
1948 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1949 /* if the file is too big, only hold_free a token amount */
1950 dmu_tx_hold_free(tx, zp->z_id, 0,
1951 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1954 /* are there any extended attributes? */
1955 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1956 &xattr_obj, sizeof (xattr_obj));
1957 if (error == 0 && xattr_obj) {
1958 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1960 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1961 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1964 mutex_enter(&zp->z_lock);
1965 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1966 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1967 mutex_exit(&zp->z_lock);
1969 /* charge as an update -- would be nice not to charge at all */
1970 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1972 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1974 zfs_dirent_unlock(dl);
1978 if (error == ERESTART) {
1992 * Remove the directory entry.
1994 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
2004 * Hold z_lock so that we can make sure that the ACL obj
2005 * hasn't changed. Could have been deleted due to
2008 mutex_enter(&zp->z_lock);
2010 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
2011 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
2012 delete_now = may_delete_now && !toobig &&
2013 vp->v_count == 1 && !vn_has_cached_data(vp) &&
2014 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
2021 panic("zfs_remove: delete_now branch taken");
2023 if (xattr_obj_unlinked) {
2024 ASSERT3U(xzp->z_links, ==, 2);
2025 mutex_enter(&xzp->z_lock);
2026 xzp->z_unlinked = 1;
2028 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
2029 &xzp->z_links, sizeof (xzp->z_links), tx);
2030 ASSERT3U(error, ==, 0);
2031 mutex_exit(&xzp->z_lock);
2032 zfs_unlinked_add(xzp, tx);
2035 error = sa_remove(zp->z_sa_hdl,
2036 SA_ZPL_XATTR(zfsvfs), tx);
2038 error = sa_update(zp->z_sa_hdl,
2039 SA_ZPL_XATTR(zfsvfs), &null_xattr,
2040 sizeof (uint64_t), tx);
2045 ASSERT0(vp->v_count);
2047 mutex_exit(&zp->z_lock);
2048 zfs_znode_delete(zp, tx);
2049 } else if (unlinked) {
2050 mutex_exit(&zp->z_lock);
2051 zfs_unlinked_add(zp, tx);
2053 vp->v_vflag |= VV_NOSYNC;
2058 if (flags & FIGNORECASE)
2060 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
2067 zfs_dirent_unlock(dl);
2074 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2075 zil_commit(zilog, 0);
2082 * Create a new directory and insert it into dvp using the name
2083 * provided. Return a pointer to the inserted directory.
2085 * IN: dvp - vnode of directory to add subdir to.
2086 * dirname - name of new directory.
2087 * vap - attributes of new directory.
2088 * cr - credentials of caller.
2089 * ct - caller context
2090 * flags - case flags
2091 * vsecp - ACL to be set
2093 * OUT: vpp - vnode of created directory.
2095 * RETURN: 0 on success, error code on failure.
2098 * dvp - ctime|mtime updated
2099 * vp - ctime|mtime|atime updated
2103 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
2104 caller_context_t *ct, int flags, vsecattr_t *vsecp)
2106 znode_t *zp, *dzp = VTOZ(dvp);
2107 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2116 gid_t gid = crgetgid(cr);
2117 zfs_acl_ids_t acl_ids;
2118 boolean_t fuid_dirtied;
2119 boolean_t waited = B_FALSE;
2121 ASSERT(vap->va_type == VDIR);
2124 * If we have an ephemeral id, ACL, or XVATTR then
2125 * make sure file system is at proper version
2128 ksid = crgetsid(cr, KSID_OWNER);
2130 uid = ksid_getid(ksid);
2133 if (zfsvfs->z_use_fuids == B_FALSE &&
2134 (vsecp || (vap->va_mask & AT_XVATTR) ||
2135 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
2136 return (SET_ERROR(EINVAL));
2140 zilog = zfsvfs->z_log;
2142 if (dzp->z_pflags & ZFS_XATTR) {
2144 return (SET_ERROR(EINVAL));
2147 if (zfsvfs->z_utf8 && u8_validate(dirname,
2148 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2150 return (SET_ERROR(EILSEQ));
2152 if (flags & FIGNORECASE)
2155 if (vap->va_mask & AT_XVATTR) {
2156 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
2157 crgetuid(cr), cr, vap->va_type)) != 0) {
2163 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
2164 vsecp, &acl_ids)) != 0) {
2169 * First make sure the new directory doesn't exist.
2171 * Existence is checked first to make sure we don't return
2172 * EACCES instead of EEXIST which can cause some applications
2178 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
2180 zfs_acl_ids_free(&acl_ids);
2185 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
2186 zfs_acl_ids_free(&acl_ids);
2187 zfs_dirent_unlock(dl);
2192 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
2193 zfs_acl_ids_free(&acl_ids);
2194 zfs_dirent_unlock(dl);
2196 return (SET_ERROR(EDQUOT));
2200 * Add a new entry to the directory.
2202 tx = dmu_tx_create(zfsvfs->z_os);
2203 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2204 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2205 fuid_dirtied = zfsvfs->z_fuid_dirty;
2207 zfs_fuid_txhold(zfsvfs, tx);
2208 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2209 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2210 acl_ids.z_aclp->z_acl_bytes);
2213 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2214 ZFS_SA_BASE_ATTR_SIZE);
2216 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2218 zfs_dirent_unlock(dl);
2219 if (error == ERESTART) {
2225 zfs_acl_ids_free(&acl_ids);
2234 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2237 zfs_fuid_sync(zfsvfs, tx);
2240 * Now put new name in parent dir.
2242 (void) zfs_link_create(dl, zp, tx, ZNEW);
2246 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2247 if (flags & FIGNORECASE)
2249 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2250 acl_ids.z_fuidp, vap);
2252 zfs_acl_ids_free(&acl_ids);
2256 zfs_dirent_unlock(dl);
2258 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2259 zil_commit(zilog, 0);
2266 * Remove a directory subdir entry. If the current working
2267 * directory is the same as the subdir to be removed, the
2270 * IN: dvp - vnode of directory to remove from.
2271 * name - name of directory to be removed.
2272 * cwd - vnode of current working directory.
2273 * cr - credentials of caller.
2274 * ct - caller context
2275 * flags - case flags
2277 * RETURN: 0 on success, error code on failure.
2280 * dvp - ctime|mtime updated
2284 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2285 caller_context_t *ct, int flags)
2287 znode_t *dzp = VTOZ(dvp);
2290 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2296 boolean_t waited = B_FALSE;
2300 zilog = zfsvfs->z_log;
2302 if (flags & FIGNORECASE)
2308 * Attempt to lock directory; fail if entry doesn't exist.
2310 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2318 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2322 if (vp->v_type != VDIR) {
2323 error = SET_ERROR(ENOTDIR);
2328 error = SET_ERROR(EINVAL);
2332 vnevent_rmdir(vp, dvp, name, ct);
2335 * Grab a lock on the directory to make sure that noone is
2336 * trying to add (or lookup) entries while we are removing it.
2338 rw_enter(&zp->z_name_lock, RW_WRITER);
2341 * Grab a lock on the parent pointer to make sure we play well
2342 * with the treewalk and directory rename code.
2344 rw_enter(&zp->z_parent_lock, RW_WRITER);
2346 tx = dmu_tx_create(zfsvfs->z_os);
2347 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2348 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2349 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2350 zfs_sa_upgrade_txholds(tx, zp);
2351 zfs_sa_upgrade_txholds(tx, dzp);
2352 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2354 rw_exit(&zp->z_parent_lock);
2355 rw_exit(&zp->z_name_lock);
2356 zfs_dirent_unlock(dl);
2358 if (error == ERESTART) {
2369 #ifdef FREEBSD_NAMECACHE
2373 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2376 uint64_t txtype = TX_RMDIR;
2377 if (flags & FIGNORECASE)
2379 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2384 rw_exit(&zp->z_parent_lock);
2385 rw_exit(&zp->z_name_lock);
2386 #ifdef FREEBSD_NAMECACHE
2390 zfs_dirent_unlock(dl);
2394 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2395 zil_commit(zilog, 0);
2402 * Read as many directory entries as will fit into the provided
2403 * buffer from the given directory cursor position (specified in
2404 * the uio structure).
2406 * IN: vp - vnode of directory to read.
2407 * uio - structure supplying read location, range info,
2408 * and return buffer.
2409 * cr - credentials of caller.
2410 * ct - caller context
2411 * flags - case flags
2413 * OUT: uio - updated offset and range, buffer filled.
2414 * eofp - set to true if end-of-file detected.
2416 * RETURN: 0 on success, error code on failure.
2419 * vp - atime updated
2421 * Note that the low 4 bits of the cookie returned by zap is always zero.
2422 * This allows us to use the low range for "special" directory entries:
2423 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2424 * we use the offset 2 for the '.zfs' directory.
2428 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies)
2430 znode_t *zp = VTOZ(vp);
2434 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2439 zap_attribute_t zap;
2440 uint_t bytes_wanted;
2441 uint64_t offset; /* must be unsigned; checks for < 1 */
2447 boolean_t check_sysattrs;
2450 u_long *cooks = NULL;
2456 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2457 &parent, sizeof (parent))) != 0) {
2463 * If we are not given an eof variable,
2470 * Check for valid iov_len.
2472 if (uio->uio_iov->iov_len <= 0) {
2474 return (SET_ERROR(EINVAL));
2478 * Quit if directory has been removed (posix)
2480 if ((*eofp = zp->z_unlinked) != 0) {
2487 offset = uio->uio_loffset;
2488 prefetch = zp->z_zn_prefetch;
2491 * Initialize the iterator cursor.
2495 * Start iteration from the beginning of the directory.
2497 zap_cursor_init(&zc, os, zp->z_id);
2500 * The offset is a serialized cursor.
2502 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2506 * Get space to change directory entries into fs independent format.
2508 iovp = uio->uio_iov;
2509 bytes_wanted = iovp->iov_len;
2510 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2511 bufsize = bytes_wanted;
2512 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2513 odp = (struct dirent64 *)outbuf;
2515 bufsize = bytes_wanted;
2517 odp = (struct dirent64 *)iovp->iov_base;
2519 eodp = (struct edirent *)odp;
2521 if (ncookies != NULL) {
2523 * Minimum entry size is dirent size and 1 byte for a file name.
2525 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1);
2526 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK);
2531 * If this VFS supports the system attribute view interface; and
2532 * we're looking at an extended attribute directory; and we care
2533 * about normalization conflicts on this vfs; then we must check
2534 * for normalization conflicts with the sysattr name space.
2537 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2538 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2539 (flags & V_RDDIR_ENTFLAGS);
2545 * Transform to file-system independent format
2548 while (outcount < bytes_wanted) {
2551 off64_t *next = NULL;
2554 * Special case `.', `..', and `.zfs'.
2557 (void) strcpy(zap.za_name, ".");
2558 zap.za_normalization_conflict = 0;
2561 } else if (offset == 1) {
2562 (void) strcpy(zap.za_name, "..");
2563 zap.za_normalization_conflict = 0;
2566 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2567 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2568 zap.za_normalization_conflict = 0;
2569 objnum = ZFSCTL_INO_ROOT;
2575 if (error = zap_cursor_retrieve(&zc, &zap)) {
2576 if ((*eofp = (error == ENOENT)) != 0)
2582 if (zap.za_integer_length != 8 ||
2583 zap.za_num_integers != 1) {
2584 cmn_err(CE_WARN, "zap_readdir: bad directory "
2585 "entry, obj = %lld, offset = %lld\n",
2586 (u_longlong_t)zp->z_id,
2587 (u_longlong_t)offset);
2588 error = SET_ERROR(ENXIO);
2592 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2594 * MacOS X can extract the object type here such as:
2595 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2597 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2599 if (check_sysattrs && !zap.za_normalization_conflict) {
2601 zap.za_normalization_conflict =
2602 xattr_sysattr_casechk(zap.za_name);
2604 panic("%s:%u: TODO", __func__, __LINE__);
2609 if (flags & V_RDDIR_ACCFILTER) {
2611 * If we have no access at all, don't include
2612 * this entry in the returned information
2615 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2617 if (!zfs_has_access(ezp, cr)) {
2624 if (flags & V_RDDIR_ENTFLAGS)
2625 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2627 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2630 * Will this entry fit in the buffer?
2632 if (outcount + reclen > bufsize) {
2634 * Did we manage to fit anything in the buffer?
2637 error = SET_ERROR(EINVAL);
2642 if (flags & V_RDDIR_ENTFLAGS) {
2644 * Add extended flag entry:
2646 eodp->ed_ino = objnum;
2647 eodp->ed_reclen = reclen;
2648 /* NOTE: ed_off is the offset for the *next* entry */
2649 next = &(eodp->ed_off);
2650 eodp->ed_eflags = zap.za_normalization_conflict ?
2651 ED_CASE_CONFLICT : 0;
2652 (void) strncpy(eodp->ed_name, zap.za_name,
2653 EDIRENT_NAMELEN(reclen));
2654 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2659 odp->d_ino = objnum;
2660 odp->d_reclen = reclen;
2661 odp->d_namlen = strlen(zap.za_name);
2662 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1);
2664 odp = (dirent64_t *)((intptr_t)odp + reclen);
2668 ASSERT(outcount <= bufsize);
2670 /* Prefetch znode */
2672 dmu_prefetch(os, objnum, 0, 0);
2676 * Move to the next entry, fill in the previous offset.
2678 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2679 zap_cursor_advance(&zc);
2680 offset = zap_cursor_serialize(&zc);
2685 if (cooks != NULL) {
2688 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks));
2691 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2693 /* Subtract unused cookies */
2694 if (ncookies != NULL)
2695 *ncookies -= ncooks;
2697 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2698 iovp->iov_base += outcount;
2699 iovp->iov_len -= outcount;
2700 uio->uio_resid -= outcount;
2701 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2703 * Reset the pointer.
2705 offset = uio->uio_loffset;
2709 zap_cursor_fini(&zc);
2710 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2711 kmem_free(outbuf, bufsize);
2713 if (error == ENOENT)
2716 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2718 uio->uio_loffset = offset;
2720 if (error != 0 && cookies != NULL) {
2721 free(*cookies, M_TEMP);
2728 ulong_t zfs_fsync_sync_cnt = 4;
2731 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2733 znode_t *zp = VTOZ(vp);
2734 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2736 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2738 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2741 zil_commit(zfsvfs->z_log, zp->z_id);
2749 * Get the requested file attributes and place them in the provided
2752 * IN: vp - vnode of file.
2753 * vap - va_mask identifies requested attributes.
2754 * If AT_XVATTR set, then optional attrs are requested
2755 * flags - ATTR_NOACLCHECK (CIFS server context)
2756 * cr - credentials of caller.
2757 * ct - caller context
2759 * OUT: vap - attribute values.
2761 * RETURN: 0 (always succeeds).
2765 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2766 caller_context_t *ct)
2768 znode_t *zp = VTOZ(vp);
2769 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2772 u_longlong_t nblocks;
2774 uint64_t mtime[2], ctime[2], crtime[2], rdev;
2775 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2776 xoptattr_t *xoap = NULL;
2777 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2778 sa_bulk_attr_t bulk[4];
2784 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2786 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2787 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2788 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
2789 if (vp->v_type == VBLK || vp->v_type == VCHR)
2790 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
2793 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2799 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2800 * Also, if we are the owner don't bother, since owner should
2801 * always be allowed to read basic attributes of file.
2803 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2804 (vap->va_uid != crgetuid(cr))) {
2805 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2813 * Return all attributes. It's cheaper to provide the answer
2814 * than to determine whether we were asked the question.
2817 mutex_enter(&zp->z_lock);
2818 vap->va_type = IFTOVT(zp->z_mode);
2819 vap->va_mode = zp->z_mode & ~S_IFMT;
2821 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2823 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
2825 vap->va_nodeid = zp->z_id;
2826 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2827 links = zp->z_links + 1;
2829 links = zp->z_links;
2830 vap->va_nlink = MIN(links, LINK_MAX); /* nlink_t limit! */
2831 vap->va_size = zp->z_size;
2833 vap->va_rdev = vp->v_rdev;
2835 if (vp->v_type == VBLK || vp->v_type == VCHR)
2836 vap->va_rdev = zfs_cmpldev(rdev);
2838 vap->va_seq = zp->z_seq;
2839 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */
2842 * Add in any requested optional attributes and the create time.
2843 * Also set the corresponding bits in the returned attribute bitmap.
2845 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2846 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2848 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2849 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2852 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2853 xoap->xoa_readonly =
2854 ((zp->z_pflags & ZFS_READONLY) != 0);
2855 XVA_SET_RTN(xvap, XAT_READONLY);
2858 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2860 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2861 XVA_SET_RTN(xvap, XAT_SYSTEM);
2864 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2866 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2867 XVA_SET_RTN(xvap, XAT_HIDDEN);
2870 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2871 xoap->xoa_nounlink =
2872 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2873 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2876 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2877 xoap->xoa_immutable =
2878 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2879 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2882 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2883 xoap->xoa_appendonly =
2884 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2885 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2888 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2890 ((zp->z_pflags & ZFS_NODUMP) != 0);
2891 XVA_SET_RTN(xvap, XAT_NODUMP);
2894 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2896 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2897 XVA_SET_RTN(xvap, XAT_OPAQUE);
2900 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2901 xoap->xoa_av_quarantined =
2902 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2903 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2906 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2907 xoap->xoa_av_modified =
2908 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2909 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2912 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2913 vp->v_type == VREG) {
2914 zfs_sa_get_scanstamp(zp, xvap);
2917 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2920 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2921 times, sizeof (times));
2922 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2923 XVA_SET_RTN(xvap, XAT_CREATETIME);
2926 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2927 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2928 XVA_SET_RTN(xvap, XAT_REPARSE);
2930 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2931 xoap->xoa_generation = zp->z_gen;
2932 XVA_SET_RTN(xvap, XAT_GEN);
2935 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2937 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2938 XVA_SET_RTN(xvap, XAT_OFFLINE);
2941 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2943 ((zp->z_pflags & ZFS_SPARSE) != 0);
2944 XVA_SET_RTN(xvap, XAT_SPARSE);
2948 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2949 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2950 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2951 ZFS_TIME_DECODE(&vap->va_birthtime, crtime);
2953 mutex_exit(&zp->z_lock);
2955 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2956 vap->va_blksize = blksize;
2957 vap->va_bytes = nblocks << 9; /* nblocks * 512 */
2959 if (zp->z_blksz == 0) {
2961 * Block size hasn't been set; suggest maximal I/O transfers.
2963 vap->va_blksize = zfsvfs->z_max_blksz;
2971 * Set the file attributes to the values contained in the
2974 * IN: vp - vnode of file to be modified.
2975 * vap - new attribute values.
2976 * If AT_XVATTR set, then optional attrs are being set
2977 * flags - ATTR_UTIME set if non-default time values provided.
2978 * - ATTR_NOACLCHECK (CIFS context only).
2979 * cr - credentials of caller.
2980 * ct - caller context
2982 * RETURN: 0 on success, error code on failure.
2985 * vp - ctime updated, mtime updated if size changed.
2989 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2990 caller_context_t *ct)
2992 znode_t *zp = VTOZ(vp);
2993 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2998 uint_t mask = vap->va_mask;
2999 uint_t saved_mask = 0;
3000 uint64_t saved_mode;
3003 uint64_t new_uid, new_gid;
3005 uint64_t mtime[2], ctime[2];
3007 int need_policy = FALSE;
3009 zfs_fuid_info_t *fuidp = NULL;
3010 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
3013 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
3014 boolean_t fuid_dirtied = B_FALSE;
3015 sa_bulk_attr_t bulk[7], xattr_bulk[7];
3016 int count = 0, xattr_count = 0;
3021 if (mask & AT_NOSET)
3022 return (SET_ERROR(EINVAL));
3027 zilog = zfsvfs->z_log;
3030 * Make sure that if we have ephemeral uid/gid or xvattr specified
3031 * that file system is at proper version level
3034 if (zfsvfs->z_use_fuids == B_FALSE &&
3035 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
3036 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
3037 (mask & AT_XVATTR))) {
3039 return (SET_ERROR(EINVAL));
3042 if (mask & AT_SIZE && vp->v_type == VDIR) {
3044 return (SET_ERROR(EISDIR));
3047 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
3049 return (SET_ERROR(EINVAL));
3053 * If this is an xvattr_t, then get a pointer to the structure of
3054 * optional attributes. If this is NULL, then we have a vattr_t.
3056 xoap = xva_getxoptattr(xvap);
3058 xva_init(&tmpxvattr);
3061 * Immutable files can only alter immutable bit and atime
3063 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
3064 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
3065 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
3067 return (SET_ERROR(EPERM));
3070 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
3072 return (SET_ERROR(EPERM));
3076 * Verify timestamps doesn't overflow 32 bits.
3077 * ZFS can handle large timestamps, but 32bit syscalls can't
3078 * handle times greater than 2039. This check should be removed
3079 * once large timestamps are fully supported.
3081 if (mask & (AT_ATIME | AT_MTIME)) {
3082 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
3083 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
3085 return (SET_ERROR(EOVERFLOW));
3093 /* Can this be moved to before the top label? */
3094 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
3096 return (SET_ERROR(EROFS));
3100 * First validate permissions
3103 if (mask & AT_SIZE) {
3105 * XXX - Note, we are not providing any open
3106 * mode flags here (like FNDELAY), so we may
3107 * block if there are locks present... this
3108 * should be addressed in openat().
3110 /* XXX - would it be OK to generate a log record here? */
3111 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
3118 if (mask & (AT_ATIME|AT_MTIME) ||
3119 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
3120 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
3121 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
3122 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
3123 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
3124 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
3125 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
3126 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
3130 if (mask & (AT_UID|AT_GID)) {
3131 int idmask = (mask & (AT_UID|AT_GID));
3136 * NOTE: even if a new mode is being set,
3137 * we may clear S_ISUID/S_ISGID bits.
3140 if (!(mask & AT_MODE))
3141 vap->va_mode = zp->z_mode;
3144 * Take ownership or chgrp to group we are a member of
3147 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
3148 take_group = (mask & AT_GID) &&
3149 zfs_groupmember(zfsvfs, vap->va_gid, cr);
3152 * If both AT_UID and AT_GID are set then take_owner and
3153 * take_group must both be set in order to allow taking
3156 * Otherwise, send the check through secpolicy_vnode_setattr()
3160 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
3161 ((idmask == AT_UID) && take_owner) ||
3162 ((idmask == AT_GID) && take_group)) {
3163 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
3164 skipaclchk, cr) == 0) {
3166 * Remove setuid/setgid for non-privileged users
3168 secpolicy_setid_clear(vap, vp, cr);
3169 trim_mask = (mask & (AT_UID|AT_GID));
3178 mutex_enter(&zp->z_lock);
3179 oldva.va_mode = zp->z_mode;
3180 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
3181 if (mask & AT_XVATTR) {
3183 * Update xvattr mask to include only those attributes
3184 * that are actually changing.
3186 * the bits will be restored prior to actually setting
3187 * the attributes so the caller thinks they were set.
3189 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
3190 if (xoap->xoa_appendonly !=
3191 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
3194 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
3195 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
3199 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
3200 if (xoap->xoa_nounlink !=
3201 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
3204 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
3205 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
3209 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
3210 if (xoap->xoa_immutable !=
3211 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
3214 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
3215 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
3219 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
3220 if (xoap->xoa_nodump !=
3221 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
3224 XVA_CLR_REQ(xvap, XAT_NODUMP);
3225 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
3229 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3230 if (xoap->xoa_av_modified !=
3231 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3234 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3235 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
3239 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3240 if ((vp->v_type != VREG &&
3241 xoap->xoa_av_quarantined) ||
3242 xoap->xoa_av_quarantined !=
3243 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3246 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3247 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
3251 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3252 mutex_exit(&zp->z_lock);
3254 return (SET_ERROR(EPERM));
3257 if (need_policy == FALSE &&
3258 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3259 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3264 mutex_exit(&zp->z_lock);
3266 if (mask & AT_MODE) {
3267 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3268 err = secpolicy_setid_setsticky_clear(vp, vap,
3274 trim_mask |= AT_MODE;
3282 * If trim_mask is set then take ownership
3283 * has been granted or write_acl is present and user
3284 * has the ability to modify mode. In that case remove
3285 * UID|GID and or MODE from mask so that
3286 * secpolicy_vnode_setattr() doesn't revoke it.
3290 saved_mask = vap->va_mask;
3291 vap->va_mask &= ~trim_mask;
3292 if (trim_mask & AT_MODE) {
3294 * Save the mode, as secpolicy_vnode_setattr()
3295 * will overwrite it with ova.va_mode.
3297 saved_mode = vap->va_mode;
3300 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3301 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3308 vap->va_mask |= saved_mask;
3309 if (trim_mask & AT_MODE) {
3311 * Recover the mode after
3312 * secpolicy_vnode_setattr().
3314 vap->va_mode = saved_mode;
3320 * secpolicy_vnode_setattr, or take ownership may have
3323 mask = vap->va_mask;
3325 if ((mask & (AT_UID | AT_GID))) {
3326 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3327 &xattr_obj, sizeof (xattr_obj));
3329 if (err == 0 && xattr_obj) {
3330 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3334 if (mask & AT_UID) {
3335 new_uid = zfs_fuid_create(zfsvfs,
3336 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3337 if (new_uid != zp->z_uid &&
3338 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3340 VN_RELE(ZTOV(attrzp));
3341 err = SET_ERROR(EDQUOT);
3346 if (mask & AT_GID) {
3347 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3348 cr, ZFS_GROUP, &fuidp);
3349 if (new_gid != zp->z_gid &&
3350 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3352 VN_RELE(ZTOV(attrzp));
3353 err = SET_ERROR(EDQUOT);
3358 tx = dmu_tx_create(zfsvfs->z_os);
3360 if (mask & AT_MODE) {
3361 uint64_t pmode = zp->z_mode;
3363 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3365 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3366 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3367 err = SET_ERROR(EPERM);
3371 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3374 mutex_enter(&zp->z_lock);
3375 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3377 * Are we upgrading ACL from old V0 format
3380 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3381 zfs_znode_acl_version(zp) ==
3382 ZFS_ACL_VERSION_INITIAL) {
3383 dmu_tx_hold_free(tx, acl_obj, 0,
3385 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3386 0, aclp->z_acl_bytes);
3388 dmu_tx_hold_write(tx, acl_obj, 0,
3391 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3392 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3393 0, aclp->z_acl_bytes);
3395 mutex_exit(&zp->z_lock);
3396 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3398 if ((mask & AT_XVATTR) &&
3399 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3400 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3402 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3406 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3409 fuid_dirtied = zfsvfs->z_fuid_dirty;
3411 zfs_fuid_txhold(zfsvfs, tx);
3413 zfs_sa_upgrade_txholds(tx, zp);
3415 err = dmu_tx_assign(tx, TXG_NOWAIT);
3417 if (err == ERESTART)
3424 * Set each attribute requested.
3425 * We group settings according to the locks they need to acquire.
3427 * Note: you cannot set ctime directly, although it will be
3428 * updated as a side-effect of calling this function.
3432 if (mask & (AT_UID|AT_GID|AT_MODE))
3433 mutex_enter(&zp->z_acl_lock);
3434 mutex_enter(&zp->z_lock);
3436 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3437 &zp->z_pflags, sizeof (zp->z_pflags));
3440 if (mask & (AT_UID|AT_GID|AT_MODE))
3441 mutex_enter(&attrzp->z_acl_lock);
3442 mutex_enter(&attrzp->z_lock);
3443 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3444 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3445 sizeof (attrzp->z_pflags));
3448 if (mask & (AT_UID|AT_GID)) {
3450 if (mask & AT_UID) {
3451 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3452 &new_uid, sizeof (new_uid));
3453 zp->z_uid = new_uid;
3455 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3456 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3458 attrzp->z_uid = new_uid;
3462 if (mask & AT_GID) {
3463 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3464 NULL, &new_gid, sizeof (new_gid));
3465 zp->z_gid = new_gid;
3467 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3468 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3470 attrzp->z_gid = new_gid;
3473 if (!(mask & AT_MODE)) {
3474 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3475 NULL, &new_mode, sizeof (new_mode));
3476 new_mode = zp->z_mode;
3478 err = zfs_acl_chown_setattr(zp);
3481 err = zfs_acl_chown_setattr(attrzp);
3486 if (mask & AT_MODE) {
3487 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3488 &new_mode, sizeof (new_mode));
3489 zp->z_mode = new_mode;
3490 ASSERT3U((uintptr_t)aclp, !=, 0);
3491 err = zfs_aclset_common(zp, aclp, cr, tx);
3493 if (zp->z_acl_cached)
3494 zfs_acl_free(zp->z_acl_cached);
3495 zp->z_acl_cached = aclp;
3500 if (mask & AT_ATIME) {
3501 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3502 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3503 &zp->z_atime, sizeof (zp->z_atime));
3506 if (mask & AT_MTIME) {
3507 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3508 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3509 mtime, sizeof (mtime));
3512 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3513 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3514 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3515 NULL, mtime, sizeof (mtime));
3516 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3517 &ctime, sizeof (ctime));
3518 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3520 } else if (mask != 0) {
3521 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3522 &ctime, sizeof (ctime));
3523 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3526 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3527 SA_ZPL_CTIME(zfsvfs), NULL,
3528 &ctime, sizeof (ctime));
3529 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3530 mtime, ctime, B_TRUE);
3534 * Do this after setting timestamps to prevent timestamp
3535 * update from toggling bit
3538 if (xoap && (mask & AT_XVATTR)) {
3541 * restore trimmed off masks
3542 * so that return masks can be set for caller.
3545 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3546 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3548 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3549 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3551 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3552 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3554 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3555 XVA_SET_REQ(xvap, XAT_NODUMP);
3557 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3558 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3560 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3561 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3564 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3565 ASSERT(vp->v_type == VREG);
3567 zfs_xvattr_set(zp, xvap, tx);
3571 zfs_fuid_sync(zfsvfs, tx);
3574 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3576 mutex_exit(&zp->z_lock);
3577 if (mask & (AT_UID|AT_GID|AT_MODE))
3578 mutex_exit(&zp->z_acl_lock);
3581 if (mask & (AT_UID|AT_GID|AT_MODE))
3582 mutex_exit(&attrzp->z_acl_lock);
3583 mutex_exit(&attrzp->z_lock);
3586 if (err == 0 && attrzp) {
3587 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3593 VN_RELE(ZTOV(attrzp));
3599 zfs_fuid_info_free(fuidp);
3605 if (err == ERESTART)
3608 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3613 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3614 zil_commit(zilog, 0);
3620 typedef struct zfs_zlock {
3621 krwlock_t *zl_rwlock; /* lock we acquired */
3622 znode_t *zl_znode; /* znode we held */
3623 struct zfs_zlock *zl_next; /* next in list */
3627 * Drop locks and release vnodes that were held by zfs_rename_lock().
3630 zfs_rename_unlock(zfs_zlock_t **zlpp)
3634 while ((zl = *zlpp) != NULL) {
3635 if (zl->zl_znode != NULL)
3636 VN_RELE(ZTOV(zl->zl_znode));
3637 rw_exit(zl->zl_rwlock);
3638 *zlpp = zl->zl_next;
3639 kmem_free(zl, sizeof (*zl));
3644 * Search back through the directory tree, using the ".." entries.
3645 * Lock each directory in the chain to prevent concurrent renames.
3646 * Fail any attempt to move a directory into one of its own descendants.
3647 * XXX - z_parent_lock can overlap with map or grow locks
3650 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3654 uint64_t rootid = zp->z_zfsvfs->z_root;
3655 uint64_t oidp = zp->z_id;
3656 krwlock_t *rwlp = &szp->z_parent_lock;
3657 krw_t rw = RW_WRITER;
3660 * First pass write-locks szp and compares to zp->z_id.
3661 * Later passes read-lock zp and compare to zp->z_parent.
3664 if (!rw_tryenter(rwlp, rw)) {
3666 * Another thread is renaming in this path.
3667 * Note that if we are a WRITER, we don't have any
3668 * parent_locks held yet.
3670 if (rw == RW_READER && zp->z_id > szp->z_id) {
3672 * Drop our locks and restart
3674 zfs_rename_unlock(&zl);
3678 rwlp = &szp->z_parent_lock;
3683 * Wait for other thread to drop its locks
3689 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3690 zl->zl_rwlock = rwlp;
3691 zl->zl_znode = NULL;
3692 zl->zl_next = *zlpp;
3695 if (oidp == szp->z_id) /* We're a descendant of szp */
3696 return (SET_ERROR(EINVAL));
3698 if (oidp == rootid) /* We've hit the top */
3701 if (rw == RW_READER) { /* i.e. not the first pass */
3702 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3707 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3708 &oidp, sizeof (oidp));
3709 rwlp = &zp->z_parent_lock;
3712 } while (zp->z_id != sdzp->z_id);
3718 * Move an entry from the provided source directory to the target
3719 * directory. Change the entry name as indicated.
3721 * IN: sdvp - Source directory containing the "old entry".
3722 * snm - Old entry name.
3723 * tdvp - Target directory to contain the "new entry".
3724 * tnm - New entry name.
3725 * cr - credentials of caller.
3726 * ct - caller context
3727 * flags - case flags
3729 * RETURN: 0 on success, error code on failure.
3732 * sdvp,tdvp - ctime|mtime updated
3736 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3737 caller_context_t *ct, int flags)
3739 znode_t *tdzp, *szp, *tzp;
3740 znode_t *sdzp = VTOZ(sdvp);
3741 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs;
3744 zfs_dirlock_t *sdl, *tdl;
3747 int cmp, serr, terr;
3750 boolean_t waited = B_FALSE;
3753 ZFS_VERIFY_ZP(sdzp);
3754 zilog = zfsvfs->z_log;
3757 * Make sure we have the real vp for the target directory.
3759 if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3763 ZFS_VERIFY_ZP(tdzp);
3766 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3767 * ctldir appear to have the same v_vfsp.
3769 if (tdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) {
3771 return (SET_ERROR(EXDEV));
3774 if (zfsvfs->z_utf8 && u8_validate(tnm,
3775 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3777 return (SET_ERROR(EILSEQ));
3780 if (flags & FIGNORECASE)
3789 * This is to prevent the creation of links into attribute space
3790 * by renaming a linked file into/outof an attribute directory.
3791 * See the comment in zfs_link() for why this is considered bad.
3793 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3795 return (SET_ERROR(EINVAL));
3799 * Lock source and target directory entries. To prevent deadlock,
3800 * a lock ordering must be defined. We lock the directory with
3801 * the smallest object id first, or if it's a tie, the one with
3802 * the lexically first name.
3804 if (sdzp->z_id < tdzp->z_id) {
3806 } else if (sdzp->z_id > tdzp->z_id) {
3810 * First compare the two name arguments without
3811 * considering any case folding.
3813 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3815 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3816 ASSERT(error == 0 || !zfsvfs->z_utf8);
3819 * POSIX: "If the old argument and the new argument
3820 * both refer to links to the same existing file,
3821 * the rename() function shall return successfully
3822 * and perform no other action."
3828 * If the file system is case-folding, then we may
3829 * have some more checking to do. A case-folding file
3830 * system is either supporting mixed case sensitivity
3831 * access or is completely case-insensitive. Note
3832 * that the file system is always case preserving.
3834 * In mixed sensitivity mode case sensitive behavior
3835 * is the default. FIGNORECASE must be used to
3836 * explicitly request case insensitive behavior.
3838 * If the source and target names provided differ only
3839 * by case (e.g., a request to rename 'tim' to 'Tim'),
3840 * we will treat this as a special case in the
3841 * case-insensitive mode: as long as the source name
3842 * is an exact match, we will allow this to proceed as
3843 * a name-change request.
3845 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3846 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3847 flags & FIGNORECASE)) &&
3848 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3851 * case preserving rename request, require exact
3860 * If the source and destination directories are the same, we should
3861 * grab the z_name_lock of that directory only once.
3865 rw_enter(&sdzp->z_name_lock, RW_READER);
3869 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3870 ZEXISTS | zflg, NULL, NULL);
3871 terr = zfs_dirent_lock(&tdl,
3872 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3874 terr = zfs_dirent_lock(&tdl,
3875 tdzp, tnm, &tzp, zflg, NULL, NULL);
3876 serr = zfs_dirent_lock(&sdl,
3877 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3883 * Source entry invalid or not there.
3886 zfs_dirent_unlock(tdl);
3892 rw_exit(&sdzp->z_name_lock);
3895 * FreeBSD: In OpenSolaris they only check if rename source is
3896 * ".." here, because "." is handled in their lookup. This is
3897 * not the case for FreeBSD, so we check for "." explicitly.
3899 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0)
3900 serr = SET_ERROR(EINVAL);
3905 zfs_dirent_unlock(sdl);
3909 rw_exit(&sdzp->z_name_lock);
3911 if (strcmp(tnm, "..") == 0)
3912 terr = SET_ERROR(EINVAL);
3918 * Must have write access at the source to remove the old entry
3919 * and write access at the target to create the new entry.
3920 * Note that if target and source are the same, this can be
3921 * done in a single check.
3924 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3927 if (ZTOV(szp)->v_type == VDIR) {
3929 * Check to make sure rename is valid.
3930 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3932 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3937 * Does target exist?
3941 * Source and target must be the same type.
3943 if (ZTOV(szp)->v_type == VDIR) {
3944 if (ZTOV(tzp)->v_type != VDIR) {
3945 error = SET_ERROR(ENOTDIR);
3949 if (ZTOV(tzp)->v_type == VDIR) {
3950 error = SET_ERROR(EISDIR);
3955 * POSIX dictates that when the source and target
3956 * entries refer to the same file object, rename
3957 * must do nothing and exit without error.
3959 if (szp->z_id == tzp->z_id) {
3965 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3967 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3970 * notify the target directory if it is not the same
3971 * as source directory.
3974 vnevent_rename_dest_dir(tdvp, ct);
3977 tx = dmu_tx_create(zfsvfs->z_os);
3978 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3979 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3980 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3981 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3983 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3984 zfs_sa_upgrade_txholds(tx, tdzp);
3987 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3988 zfs_sa_upgrade_txholds(tx, tzp);
3991 zfs_sa_upgrade_txholds(tx, szp);
3992 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3993 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3996 zfs_rename_unlock(&zl);
3997 zfs_dirent_unlock(sdl);
3998 zfs_dirent_unlock(tdl);
4001 rw_exit(&sdzp->z_name_lock);
4006 if (error == ERESTART) {
4017 if (tzp) /* Attempt to remove the existing target */
4018 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
4021 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
4023 szp->z_pflags |= ZFS_AV_MODIFIED;
4025 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
4026 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
4029 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
4031 zfs_log_rename(zilog, tx, TX_RENAME |
4032 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
4033 sdl->dl_name, tdzp, tdl->dl_name, szp);
4036 * Update path information for the target vnode
4038 vn_renamepath(tdvp, ZTOV(szp), tnm,
4042 * At this point, we have successfully created
4043 * the target name, but have failed to remove
4044 * the source name. Since the create was done
4045 * with the ZRENAMING flag, there are
4046 * complications; for one, the link count is
4047 * wrong. The easiest way to deal with this
4048 * is to remove the newly created target, and
4049 * return the original error. This must
4050 * succeed; fortunately, it is very unlikely to
4051 * fail, since we just created it.
4053 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
4054 ZRENAMING, NULL), ==, 0);
4057 #ifdef FREEBSD_NAMECACHE
4061 cache_purge(ZTOV(szp));
4063 cache_purge(ZTOV(tzp));
4071 zfs_rename_unlock(&zl);
4073 zfs_dirent_unlock(sdl);
4074 zfs_dirent_unlock(tdl);
4077 rw_exit(&sdzp->z_name_lock);
4084 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4085 zil_commit(zilog, 0);
4093 * Insert the indicated symbolic reference entry into the directory.
4095 * IN: dvp - Directory to contain new symbolic link.
4096 * link - Name for new symlink entry.
4097 * vap - Attributes of new entry.
4098 * cr - credentials of caller.
4099 * ct - caller context
4100 * flags - case flags
4102 * RETURN: 0 on success, error code on failure.
4105 * dvp - ctime|mtime updated
4109 zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link,
4110 cred_t *cr, kthread_t *td)
4112 znode_t *zp, *dzp = VTOZ(dvp);
4115 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4117 uint64_t len = strlen(link);
4120 zfs_acl_ids_t acl_ids;
4121 boolean_t fuid_dirtied;
4122 uint64_t txtype = TX_SYMLINK;
4123 boolean_t waited = B_FALSE;
4126 ASSERT(vap->va_type == VLNK);
4130 zilog = zfsvfs->z_log;
4132 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
4133 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4135 return (SET_ERROR(EILSEQ));
4137 if (flags & FIGNORECASE)
4140 if (len > MAXPATHLEN) {
4142 return (SET_ERROR(ENAMETOOLONG));
4145 if ((error = zfs_acl_ids_create(dzp, 0,
4146 vap, cr, NULL, &acl_ids)) != 0) {
4152 * Attempt to lock directory; fail if entry already exists.
4154 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
4156 zfs_acl_ids_free(&acl_ids);
4161 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4162 zfs_acl_ids_free(&acl_ids);
4163 zfs_dirent_unlock(dl);
4168 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
4169 zfs_acl_ids_free(&acl_ids);
4170 zfs_dirent_unlock(dl);
4172 return (SET_ERROR(EDQUOT));
4174 tx = dmu_tx_create(zfsvfs->z_os);
4175 fuid_dirtied = zfsvfs->z_fuid_dirty;
4176 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
4177 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4178 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
4179 ZFS_SA_BASE_ATTR_SIZE + len);
4180 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
4181 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
4182 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
4183 acl_ids.z_aclp->z_acl_bytes);
4186 zfs_fuid_txhold(zfsvfs, tx);
4187 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4189 zfs_dirent_unlock(dl);
4190 if (error == ERESTART) {
4196 zfs_acl_ids_free(&acl_ids);
4203 * Create a new object for the symlink.
4204 * for version 4 ZPL datsets the symlink will be an SA attribute
4206 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
4209 zfs_fuid_sync(zfsvfs, tx);
4211 mutex_enter(&zp->z_lock);
4213 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
4216 zfs_sa_symlink(zp, link, len, tx);
4217 mutex_exit(&zp->z_lock);
4220 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
4221 &zp->z_size, sizeof (zp->z_size), tx);
4223 * Insert the new object into the directory.
4225 (void) zfs_link_create(dl, zp, tx, ZNEW);
4227 if (flags & FIGNORECASE)
4229 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
4232 zfs_acl_ids_free(&acl_ids);
4236 zfs_dirent_unlock(dl);
4238 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4239 zil_commit(zilog, 0);
4246 * Return, in the buffer contained in the provided uio structure,
4247 * the symbolic path referred to by vp.
4249 * IN: vp - vnode of symbolic link.
4250 * uio - structure to contain the link path.
4251 * cr - credentials of caller.
4252 * ct - caller context
4254 * OUT: uio - structure containing the link path.
4256 * RETURN: 0 on success, error code on failure.
4259 * vp - atime updated
4263 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
4265 znode_t *zp = VTOZ(vp);
4266 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4272 mutex_enter(&zp->z_lock);
4274 error = sa_lookup_uio(zp->z_sa_hdl,
4275 SA_ZPL_SYMLINK(zfsvfs), uio);
4277 error = zfs_sa_readlink(zp, uio);
4278 mutex_exit(&zp->z_lock);
4280 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4287 * Insert a new entry into directory tdvp referencing svp.
4289 * IN: tdvp - Directory to contain new entry.
4290 * svp - vnode of new entry.
4291 * name - name of new entry.
4292 * cr - credentials of caller.
4293 * ct - caller context
4295 * RETURN: 0 on success, error code on failure.
4298 * tdvp - ctime|mtime updated
4299 * svp - ctime updated
4303 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
4304 caller_context_t *ct, int flags)
4306 znode_t *dzp = VTOZ(tdvp);
4308 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4317 boolean_t waited = B_FALSE;
4319 ASSERT(tdvp->v_type == VDIR);
4323 zilog = zfsvfs->z_log;
4325 if (VOP_REALVP(svp, &realvp, ct) == 0)
4329 * POSIX dictates that we return EPERM here.
4330 * Better choices include ENOTSUP or EISDIR.
4332 if (svp->v_type == VDIR) {
4334 return (SET_ERROR(EPERM));
4341 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
4342 * ctldir appear to have the same v_vfsp.
4344 if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) {
4346 return (SET_ERROR(EXDEV));
4349 /* Prevent links to .zfs/shares files */
4351 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4352 &parent, sizeof (uint64_t))) != 0) {
4356 if (parent == zfsvfs->z_shares_dir) {
4358 return (SET_ERROR(EPERM));
4361 if (zfsvfs->z_utf8 && u8_validate(name,
4362 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4364 return (SET_ERROR(EILSEQ));
4366 if (flags & FIGNORECASE)
4370 * We do not support links between attributes and non-attributes
4371 * because of the potential security risk of creating links
4372 * into "normal" file space in order to circumvent restrictions
4373 * imposed in attribute space.
4375 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4377 return (SET_ERROR(EINVAL));
4381 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4382 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) {
4384 return (SET_ERROR(EPERM));
4387 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4394 * Attempt to lock directory; fail if entry already exists.
4396 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4402 tx = dmu_tx_create(zfsvfs->z_os);
4403 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4404 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4405 zfs_sa_upgrade_txholds(tx, szp);
4406 zfs_sa_upgrade_txholds(tx, dzp);
4407 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4409 zfs_dirent_unlock(dl);
4410 if (error == ERESTART) {
4421 error = zfs_link_create(dl, szp, tx, 0);
4424 uint64_t txtype = TX_LINK;
4425 if (flags & FIGNORECASE)
4427 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4432 zfs_dirent_unlock(dl);
4435 vnevent_link(svp, ct);
4438 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4439 zil_commit(zilog, 0);
4447 * zfs_null_putapage() is used when the file system has been force
4448 * unmounted. It just drops the pages.
4452 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4453 size_t *lenp, int flags, cred_t *cr)
4455 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4460 * Push a page out to disk, klustering if possible.
4462 * IN: vp - file to push page to.
4463 * pp - page to push.
4464 * flags - additional flags.
4465 * cr - credentials of caller.
4467 * OUT: offp - start of range pushed.
4468 * lenp - len of range pushed.
4470 * RETURN: 0 on success, error code on failure.
4472 * NOTE: callers must have locked the page to be pushed. On
4473 * exit, the page (and all other pages in the kluster) must be
4478 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4479 size_t *lenp, int flags, cred_t *cr)
4481 znode_t *zp = VTOZ(vp);
4482 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4484 u_offset_t off, koff;
4491 * If our blocksize is bigger than the page size, try to kluster
4492 * multiple pages so that we write a full block (thus avoiding
4493 * a read-modify-write).
4495 if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4496 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4497 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4498 ASSERT(koff <= zp->z_size);
4499 if (koff + klen > zp->z_size)
4500 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4501 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4503 ASSERT3U(btop(len), ==, btopr(len));
4506 * Can't push pages past end-of-file.
4508 if (off >= zp->z_size) {
4509 /* ignore all pages */
4512 } else if (off + len > zp->z_size) {
4513 int npages = btopr(zp->z_size - off);
4516 page_list_break(&pp, &trunc, npages);
4517 /* ignore pages past end of file */
4519 pvn_write_done(trunc, flags);
4520 len = zp->z_size - off;
4523 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4524 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4525 err = SET_ERROR(EDQUOT);
4529 tx = dmu_tx_create(zfsvfs->z_os);
4530 dmu_tx_hold_write(tx, zp->z_id, off, len);
4532 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4533 zfs_sa_upgrade_txholds(tx, zp);
4534 err = dmu_tx_assign(tx, TXG_NOWAIT);
4536 if (err == ERESTART) {
4545 if (zp->z_blksz <= PAGESIZE) {
4546 caddr_t va = zfs_map_page(pp, S_READ);
4547 ASSERT3U(len, <=, PAGESIZE);
4548 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4549 zfs_unmap_page(pp, va);
4551 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4555 uint64_t mtime[2], ctime[2];
4556 sa_bulk_attr_t bulk[3];
4559 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4561 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4563 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4565 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4567 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4572 pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4582 * Copy the portion of the file indicated from pages into the file.
4583 * The pages are stored in a page list attached to the files vnode.
4585 * IN: vp - vnode of file to push page data to.
4586 * off - position in file to put data.
4587 * len - amount of data to write.
4588 * flags - flags to control the operation.
4589 * cr - credentials of caller.
4590 * ct - caller context.
4592 * RETURN: 0 on success, error code on failure.
4595 * vp - ctime|mtime updated
4599 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4600 caller_context_t *ct)
4602 znode_t *zp = VTOZ(vp);
4603 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4615 * Align this request to the file block size in case we kluster.
4616 * XXX - this can result in pretty aggresive locking, which can
4617 * impact simultanious read/write access. One option might be
4618 * to break up long requests (len == 0) into block-by-block
4619 * operations to get narrower locking.
4621 blksz = zp->z_blksz;
4623 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4626 if (len > 0 && ISP2(blksz))
4627 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4633 * Search the entire vp list for pages >= io_off.
4635 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4636 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4639 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4641 if (off > zp->z_size) {
4642 /* past end of file */
4643 zfs_range_unlock(rl);
4648 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4650 for (off = io_off; io_off < off + len; io_off += io_len) {
4651 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4652 pp = page_lookup(vp, io_off,
4653 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4655 pp = page_lookup_nowait(vp, io_off,
4656 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4659 if (pp != NULL && pvn_getdirty(pp, flags)) {
4663 * Found a dirty page to push
4665 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4673 zfs_range_unlock(rl);
4674 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4675 zil_commit(zfsvfs->z_log, zp->z_id);
4683 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4685 znode_t *zp = VTOZ(vp);
4686 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4689 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4690 if (zp->z_sa_hdl == NULL) {
4692 * The fs has been unmounted, or we did a
4693 * suspend/resume and this file no longer exists.
4695 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4700 mutex_enter(&zp->z_lock);
4701 if (zp->z_unlinked) {
4703 * Fast path to recycle a vnode of a removed file.
4705 mutex_exit(&zp->z_lock);
4706 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4710 mutex_exit(&zp->z_lock);
4712 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4713 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4715 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4716 zfs_sa_upgrade_txholds(tx, zp);
4717 error = dmu_tx_assign(tx, TXG_WAIT);
4721 mutex_enter(&zp->z_lock);
4722 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4723 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4724 zp->z_atime_dirty = 0;
4725 mutex_exit(&zp->z_lock);
4729 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4734 * Bounds-check the seek operation.
4736 * IN: vp - vnode seeking within
4737 * ooff - old file offset
4738 * noffp - pointer to new file offset
4739 * ct - caller context
4741 * RETURN: 0 on success, EINVAL if new offset invalid.
4745 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4746 caller_context_t *ct)
4748 if (vp->v_type == VDIR)
4750 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4754 * Pre-filter the generic locking function to trap attempts to place
4755 * a mandatory lock on a memory mapped file.
4758 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4759 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4761 znode_t *zp = VTOZ(vp);
4762 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4768 * We are following the UFS semantics with respect to mapcnt
4769 * here: If we see that the file is mapped already, then we will
4770 * return an error, but we don't worry about races between this
4771 * function and zfs_map().
4773 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4775 return (SET_ERROR(EAGAIN));
4778 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4782 * If we can't find a page in the cache, we will create a new page
4783 * and fill it with file data. For efficiency, we may try to fill
4784 * multiple pages at once (klustering) to fill up the supplied page
4785 * list. Note that the pages to be filled are held with an exclusive
4786 * lock to prevent access by other threads while they are being filled.
4789 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4790 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4792 znode_t *zp = VTOZ(vp);
4793 page_t *pp, *cur_pp;
4794 objset_t *os = zp->z_zfsvfs->z_os;
4795 u_offset_t io_off, total;
4799 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4801 * We only have a single page, don't bother klustering
4805 pp = page_create_va(vp, io_off, io_len,
4806 PG_EXCL | PG_WAIT, seg, addr);
4809 * Try to find enough pages to fill the page list
4811 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4812 &io_len, off, plsz, 0);
4816 * The page already exists, nothing to do here.
4823 * Fill the pages in the kluster.
4826 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4829 ASSERT3U(io_off, ==, cur_pp->p_offset);
4830 va = zfs_map_page(cur_pp, S_WRITE);
4831 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4833 zfs_unmap_page(cur_pp, va);
4835 /* On error, toss the entire kluster */
4836 pvn_read_done(pp, B_ERROR);
4837 /* convert checksum errors into IO errors */
4839 err = SET_ERROR(EIO);
4842 cur_pp = cur_pp->p_next;
4846 * Fill in the page list array from the kluster starting
4847 * from the desired offset `off'.
4848 * NOTE: the page list will always be null terminated.
4850 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4851 ASSERT(pl == NULL || (*pl)->p_offset == off);
4857 * Return pointers to the pages for the file region [off, off + len]
4858 * in the pl array. If plsz is greater than len, this function may
4859 * also return page pointers from after the specified region
4860 * (i.e. the region [off, off + plsz]). These additional pages are
4861 * only returned if they are already in the cache, or were created as
4862 * part of a klustered read.
4864 * IN: vp - vnode of file to get data from.
4865 * off - position in file to get data from.
4866 * len - amount of data to retrieve.
4867 * plsz - length of provided page list.
4868 * seg - segment to obtain pages for.
4869 * addr - virtual address of fault.
4870 * rw - mode of created pages.
4871 * cr - credentials of caller.
4872 * ct - caller context.
4874 * OUT: protp - protection mode of created pages.
4875 * pl - list of pages created.
4877 * RETURN: 0 on success, error code on failure.
4880 * vp - atime updated
4884 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4885 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4886 enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4888 znode_t *zp = VTOZ(vp);
4889 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4893 /* we do our own caching, faultahead is unnecessary */
4896 else if (len > plsz)
4899 len = P2ROUNDUP(len, PAGESIZE);
4900 ASSERT(plsz >= len);
4909 * Loop through the requested range [off, off + len) looking
4910 * for pages. If we don't find a page, we will need to create
4911 * a new page and fill it with data from the file.
4914 if (*pl = page_lookup(vp, off, SE_SHARED))
4916 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4919 ASSERT3U((*pl)->p_offset, ==, off);
4923 ASSERT3U(len, >=, PAGESIZE);
4926 ASSERT3U(plsz, >=, PAGESIZE);
4933 * Fill out the page array with any pages already in the cache.
4936 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4943 * Release any pages we have previously locked.
4948 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4958 * Request a memory map for a section of a file. This code interacts
4959 * with common code and the VM system as follows:
4961 * - common code calls mmap(), which ends up in smmap_common()
4962 * - this calls VOP_MAP(), which takes you into (say) zfs
4963 * - zfs_map() calls as_map(), passing segvn_create() as the callback
4964 * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4965 * - zfs_addmap() updates z_mapcnt
4969 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4970 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4971 caller_context_t *ct)
4973 znode_t *zp = VTOZ(vp);
4974 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4975 segvn_crargs_t vn_a;
4981 if ((prot & PROT_WRITE) && (zp->z_pflags &
4982 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4984 return (SET_ERROR(EPERM));
4987 if ((prot & (PROT_READ | PROT_EXEC)) &&
4988 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4990 return (SET_ERROR(EACCES));
4993 if (vp->v_flag & VNOMAP) {
4995 return (SET_ERROR(ENOSYS));
4998 if (off < 0 || len > MAXOFFSET_T - off) {
5000 return (SET_ERROR(ENXIO));
5003 if (vp->v_type != VREG) {
5005 return (SET_ERROR(ENODEV));
5009 * If file is locked, disallow mapping.
5011 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
5013 return (SET_ERROR(EAGAIN));
5017 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
5025 vn_a.offset = (u_offset_t)off;
5026 vn_a.type = flags & MAP_TYPE;
5028 vn_a.maxprot = maxprot;
5031 vn_a.flags = flags & ~MAP_TYPE;
5033 vn_a.lgrp_mem_policy_flags = 0;
5035 error = as_map(as, *addrp, len, segvn_create, &vn_a);
5044 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5045 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
5046 caller_context_t *ct)
5048 uint64_t pages = btopr(len);
5050 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
5055 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
5056 * more accurate mtime for the associated file. Since we don't have a way of
5057 * detecting when the data was actually modified, we have to resort to
5058 * heuristics. If an explicit msync() is done, then we mark the mtime when the
5059 * last page is pushed. The problem occurs when the msync() call is omitted,
5060 * which by far the most common case:
5068 * putpage() via fsflush
5070 * If we wait until fsflush to come along, we can have a modification time that
5071 * is some arbitrary point in the future. In order to prevent this in the
5072 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
5077 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5078 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
5079 caller_context_t *ct)
5081 uint64_t pages = btopr(len);
5083 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
5084 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
5086 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
5087 vn_has_cached_data(vp))
5088 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
5094 * Free or allocate space in a file. Currently, this function only
5095 * supports the `F_FREESP' command. However, this command is somewhat
5096 * misnamed, as its functionality includes the ability to allocate as
5097 * well as free space.
5099 * IN: vp - vnode of file to free data in.
5100 * cmd - action to take (only F_FREESP supported).
5101 * bfp - section of file to free/alloc.
5102 * flag - current file open mode flags.
5103 * offset - current file offset.
5104 * cr - credentials of caller [UNUSED].
5105 * ct - caller context.
5107 * RETURN: 0 on success, error code on failure.
5110 * vp - ctime|mtime updated
5114 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
5115 offset_t offset, cred_t *cr, caller_context_t *ct)
5117 znode_t *zp = VTOZ(vp);
5118 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5125 if (cmd != F_FREESP) {
5127 return (SET_ERROR(EINVAL));
5130 if (error = convoff(vp, bfp, 0, offset)) {
5135 if (bfp->l_len < 0) {
5137 return (SET_ERROR(EINVAL));
5141 len = bfp->l_len; /* 0 means from off to end of file */
5143 error = zfs_freesp(zp, off, len, flag, TRUE);
5150 CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid));
5151 CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid));
5155 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
5157 znode_t *zp = VTOZ(vp);
5158 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5161 uint64_t object = zp->z_id;
5168 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
5169 &gen64, sizeof (uint64_t))) != 0) {
5174 gen = (uint32_t)gen64;
5176 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
5179 if (fidp->fid_len < size) {
5180 fidp->fid_len = size;
5182 return (SET_ERROR(ENOSPC));
5185 fidp->fid_len = size;
5188 zfid = (zfid_short_t *)fidp;
5190 zfid->zf_len = size;
5192 for (i = 0; i < sizeof (zfid->zf_object); i++)
5193 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
5195 /* Must have a non-zero generation number to distinguish from .zfs */
5198 for (i = 0; i < sizeof (zfid->zf_gen); i++)
5199 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
5201 if (size == LONG_FID_LEN) {
5202 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
5205 zlfid = (zfid_long_t *)fidp;
5207 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
5208 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
5210 /* XXX - this should be the generation number for the objset */
5211 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
5212 zlfid->zf_setgen[i] = 0;
5220 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
5221 caller_context_t *ct)
5233 case _PC_FILESIZEBITS:
5237 case _PC_XATTR_EXISTS:
5239 zfsvfs = zp->z_zfsvfs;
5243 error = zfs_dirent_lock(&dl, zp, "", &xzp,
5244 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
5246 zfs_dirent_unlock(dl);
5247 if (!zfs_dirempty(xzp))
5250 } else if (error == ENOENT) {
5252 * If there aren't extended attributes, it's the
5253 * same as having zero of them.
5260 case _PC_SATTR_ENABLED:
5261 case _PC_SATTR_EXISTS:
5262 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
5263 (vp->v_type == VREG || vp->v_type == VDIR);
5266 case _PC_ACCESS_FILTERING:
5267 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
5271 case _PC_ACL_ENABLED:
5272 *valp = _ACL_ACE_ENABLED;
5275 case _PC_MIN_HOLE_SIZE:
5276 *valp = (int)SPA_MINBLOCKSIZE;
5279 case _PC_TIMESTAMP_RESOLUTION:
5280 /* nanosecond timestamp resolution */
5284 case _PC_ACL_EXTENDED:
5292 case _PC_ACL_PATH_MAX:
5293 *valp = ACL_MAX_ENTRIES;
5297 return (EOPNOTSUPP);
5303 zfs_getsecattr(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;
5313 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5321 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5322 caller_context_t *ct)
5324 znode_t *zp = VTOZ(vp);
5325 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5327 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5328 zilog_t *zilog = zfsvfs->z_log;
5333 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5335 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5336 zil_commit(zilog, 0);
5344 * The smallest read we may consider to loan out an arcbuf.
5345 * This must be a power of 2.
5347 int zcr_blksz_min = (1 << 10); /* 1K */
5349 * If set to less than the file block size, allow loaning out of an
5350 * arcbuf for a partial block read. This must be a power of 2.
5352 int zcr_blksz_max = (1 << 17); /* 128K */
5356 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5357 caller_context_t *ct)
5359 znode_t *zp = VTOZ(vp);
5360 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5361 int max_blksz = zfsvfs->z_max_blksz;
5362 uio_t *uio = &xuio->xu_uio;
5363 ssize_t size = uio->uio_resid;
5364 offset_t offset = uio->uio_loffset;
5369 int preamble, postamble;
5371 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5372 return (SET_ERROR(EINVAL));
5379 * Loan out an arc_buf for write if write size is bigger than
5380 * max_blksz, and the file's block size is also max_blksz.
5383 if (size < blksz || zp->z_blksz != blksz) {
5385 return (SET_ERROR(EINVAL));
5388 * Caller requests buffers for write before knowing where the
5389 * write offset might be (e.g. NFS TCP write).
5394 preamble = P2PHASE(offset, blksz);
5396 preamble = blksz - preamble;
5401 postamble = P2PHASE(size, blksz);
5404 fullblk = size / blksz;
5405 (void) dmu_xuio_init(xuio,
5406 (preamble != 0) + fullblk + (postamble != 0));
5407 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5408 int, postamble, int,
5409 (preamble != 0) + fullblk + (postamble != 0));
5412 * Have to fix iov base/len for partial buffers. They
5413 * currently represent full arc_buf's.
5416 /* data begins in the middle of the arc_buf */
5417 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5420 (void) dmu_xuio_add(xuio, abuf,
5421 blksz - preamble, preamble);
5424 for (i = 0; i < fullblk; i++) {
5425 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5428 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5432 /* data ends in the middle of the arc_buf */
5433 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5436 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5441 * Loan out an arc_buf for read if the read size is larger than
5442 * the current file block size. Block alignment is not
5443 * considered. Partial arc_buf will be loaned out for read.
5445 blksz = zp->z_blksz;
5446 if (blksz < zcr_blksz_min)
5447 blksz = zcr_blksz_min;
5448 if (blksz > zcr_blksz_max)
5449 blksz = zcr_blksz_max;
5450 /* avoid potential complexity of dealing with it */
5451 if (blksz > max_blksz) {
5453 return (SET_ERROR(EINVAL));
5456 maxsize = zp->z_size - uio->uio_loffset;
5460 if (size < blksz || vn_has_cached_data(vp)) {
5462 return (SET_ERROR(EINVAL));
5467 return (SET_ERROR(EINVAL));
5470 uio->uio_extflg = UIO_XUIO;
5471 XUIO_XUZC_RW(xuio) = ioflag;
5478 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5482 int ioflag = XUIO_XUZC_RW(xuio);
5484 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5486 i = dmu_xuio_cnt(xuio);
5488 abuf = dmu_xuio_arcbuf(xuio, i);
5490 * if abuf == NULL, it must be a write buffer
5491 * that has been returned in zfs_write().
5494 dmu_return_arcbuf(abuf);
5495 ASSERT(abuf || ioflag == UIO_WRITE);
5498 dmu_xuio_fini(xuio);
5503 * Predeclare these here so that the compiler assumes that
5504 * this is an "old style" function declaration that does
5505 * not include arguments => we won't get type mismatch errors
5506 * in the initializations that follow.
5508 static int zfs_inval();
5509 static int zfs_isdir();
5514 return (SET_ERROR(EINVAL));
5520 return (SET_ERROR(EISDIR));
5523 * Directory vnode operations template
5525 vnodeops_t *zfs_dvnodeops;
5526 const fs_operation_def_t zfs_dvnodeops_template[] = {
5527 VOPNAME_OPEN, { .vop_open = zfs_open },
5528 VOPNAME_CLOSE, { .vop_close = zfs_close },
5529 VOPNAME_READ, { .error = zfs_isdir },
5530 VOPNAME_WRITE, { .error = zfs_isdir },
5531 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5532 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5533 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5534 VOPNAME_ACCESS, { .vop_access = zfs_access },
5535 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5536 VOPNAME_CREATE, { .vop_create = zfs_create },
5537 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5538 VOPNAME_LINK, { .vop_link = zfs_link },
5539 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5540 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir },
5541 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5542 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5543 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink },
5544 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5545 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5546 VOPNAME_FID, { .vop_fid = zfs_fid },
5547 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5548 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5549 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5550 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5551 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5556 * Regular file vnode operations template
5558 vnodeops_t *zfs_fvnodeops;
5559 const fs_operation_def_t zfs_fvnodeops_template[] = {
5560 VOPNAME_OPEN, { .vop_open = zfs_open },
5561 VOPNAME_CLOSE, { .vop_close = zfs_close },
5562 VOPNAME_READ, { .vop_read = zfs_read },
5563 VOPNAME_WRITE, { .vop_write = zfs_write },
5564 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5565 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5566 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5567 VOPNAME_ACCESS, { .vop_access = zfs_access },
5568 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5569 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5570 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5571 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5572 VOPNAME_FID, { .vop_fid = zfs_fid },
5573 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5574 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock },
5575 VOPNAME_SPACE, { .vop_space = zfs_space },
5576 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage },
5577 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage },
5578 VOPNAME_MAP, { .vop_map = zfs_map },
5579 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap },
5580 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap },
5581 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5582 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5583 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5584 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5585 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf },
5586 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf },
5591 * Symbolic link vnode operations template
5593 vnodeops_t *zfs_symvnodeops;
5594 const fs_operation_def_t zfs_symvnodeops_template[] = {
5595 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5596 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5597 VOPNAME_ACCESS, { .vop_access = zfs_access },
5598 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5599 VOPNAME_READLINK, { .vop_readlink = zfs_readlink },
5600 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5601 VOPNAME_FID, { .vop_fid = zfs_fid },
5602 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5603 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5608 * special share hidden files vnode operations template
5610 vnodeops_t *zfs_sharevnodeops;
5611 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5612 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5613 VOPNAME_ACCESS, { .vop_access = zfs_access },
5614 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5615 VOPNAME_FID, { .vop_fid = zfs_fid },
5616 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5617 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5618 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5619 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5624 * Extended attribute directory vnode operations template
5626 * This template is identical to the directory vnodes
5627 * operation template except for restricted operations:
5631 * Note that there are other restrictions embedded in:
5632 * zfs_create() - restrict type to VREG
5633 * zfs_link() - no links into/out of attribute space
5634 * zfs_rename() - no moves into/out of attribute space
5636 vnodeops_t *zfs_xdvnodeops;
5637 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5638 VOPNAME_OPEN, { .vop_open = zfs_open },
5639 VOPNAME_CLOSE, { .vop_close = zfs_close },
5640 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5641 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5642 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5643 VOPNAME_ACCESS, { .vop_access = zfs_access },
5644 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5645 VOPNAME_CREATE, { .vop_create = zfs_create },
5646 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5647 VOPNAME_LINK, { .vop_link = zfs_link },
5648 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5649 VOPNAME_MKDIR, { .error = zfs_inval },
5650 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5651 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5652 VOPNAME_SYMLINK, { .error = zfs_inval },
5653 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5654 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5655 VOPNAME_FID, { .vop_fid = zfs_fid },
5656 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5657 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5658 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5659 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5660 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5665 * Error vnode operations template
5667 vnodeops_t *zfs_evnodeops;
5668 const fs_operation_def_t zfs_evnodeops_template[] = {
5669 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5670 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5676 ioflags(int ioflags)
5680 if (ioflags & IO_APPEND)
5682 if (ioflags & IO_NDELAY)
5684 if (ioflags & IO_SYNC)
5685 flags |= (FSYNC | FDSYNC | FRSYNC);
5691 zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage)
5693 znode_t *zp = VTOZ(vp);
5694 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5695 objset_t *os = zp->z_zfsvfs->z_os;
5696 vm_page_t mfirst, mlast, mreq;
5700 off_t startoff, endoff;
5702 vm_pindex_t reqstart, reqend;
5703 int pcount, lsize, reqsize, size;
5708 pcount = OFF_TO_IDX(round_page(count));
5710 object = mreq->object;
5713 KASSERT(vp->v_object == object, ("mismatching object"));
5715 if (pcount > 1 && zp->z_blksz > PAGESIZE) {
5716 startoff = rounddown(IDX_TO_OFF(mreq->pindex), zp->z_blksz);
5717 reqstart = OFF_TO_IDX(round_page(startoff));
5718 if (reqstart < m[0]->pindex)
5721 reqstart = reqstart - m[0]->pindex;
5722 endoff = roundup(IDX_TO_OFF(mreq->pindex) + PAGE_SIZE,
5724 reqend = OFF_TO_IDX(trunc_page(endoff)) - 1;
5725 if (reqend > m[pcount - 1]->pindex)
5726 reqend = m[pcount - 1]->pindex;
5727 reqsize = reqend - m[reqstart]->pindex + 1;
5728 KASSERT(reqstart <= reqpage && reqpage < reqstart + reqsize,
5729 ("reqpage beyond [reqstart, reqstart + reqsize[ bounds"));
5734 mfirst = m[reqstart];
5735 mlast = m[reqstart + reqsize - 1];
5737 zfs_vmobject_wlock(object);
5739 for (i = 0; i < reqstart; i++) {
5742 vm_page_unlock(m[i]);
5744 for (i = reqstart + reqsize; i < pcount; i++) {
5747 vm_page_unlock(m[i]);
5750 if (mreq->valid && reqsize == 1) {
5751 if (mreq->valid != VM_PAGE_BITS_ALL)
5752 vm_page_zero_invalid(mreq, TRUE);
5753 zfs_vmobject_wunlock(object);
5755 return (zfs_vm_pagerret_ok);
5758 PCPU_INC(cnt.v_vnodein);
5759 PCPU_ADD(cnt.v_vnodepgsin, reqsize);
5761 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) {
5762 for (i = reqstart; i < reqstart + reqsize; i++) {
5766 vm_page_unlock(m[i]);
5769 zfs_vmobject_wunlock(object);
5771 return (zfs_vm_pagerret_bad);
5775 if (IDX_TO_OFF(mlast->pindex) + lsize > object->un_pager.vnp.vnp_size)
5776 lsize = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mlast->pindex);
5778 zfs_vmobject_wunlock(object);
5780 for (i = reqstart; i < reqstart + reqsize; i++) {
5782 if (i == (reqstart + reqsize - 1))
5784 va = zfs_map_page(m[i], &sf);
5785 error = dmu_read(os, zp->z_id, IDX_TO_OFF(m[i]->pindex),
5786 size, va, DMU_READ_PREFETCH);
5787 if (size != PAGE_SIZE)
5788 bzero(va + size, PAGE_SIZE - size);
5794 zfs_vmobject_wlock(object);
5796 for (i = reqstart; i < reqstart + reqsize; i++) {
5798 m[i]->valid = VM_PAGE_BITS_ALL;
5799 KASSERT(m[i]->dirty == 0, ("zfs_getpages: page %p is dirty", m[i]));
5801 vm_page_readahead_finish(m[i]);
5804 zfs_vmobject_wunlock(object);
5806 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
5808 return (error ? zfs_vm_pagerret_error : zfs_vm_pagerret_ok);
5812 zfs_freebsd_getpages(ap)
5813 struct vop_getpages_args /* {
5818 vm_ooffset_t a_offset;
5822 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage));
5826 zfs_freebsd_bmap(ap)
5827 struct vop_bmap_args /* {
5830 struct bufobj **a_bop;
5837 if (ap->a_bop != NULL)
5838 *ap->a_bop = &ap->a_vp->v_bufobj;
5839 if (ap->a_bnp != NULL)
5840 *ap->a_bnp = ap->a_bn;
5841 if (ap->a_runp != NULL)
5843 if (ap->a_runb != NULL)
5850 zfs_freebsd_open(ap)
5851 struct vop_open_args /* {
5854 struct ucred *a_cred;
5855 struct thread *a_td;
5858 vnode_t *vp = ap->a_vp;
5859 znode_t *zp = VTOZ(vp);
5862 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL);
5864 vnode_create_vobject(vp, zp->z_size, ap->a_td);
5869 zfs_freebsd_close(ap)
5870 struct vop_close_args /* {
5873 struct ucred *a_cred;
5874 struct thread *a_td;
5878 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL));
5882 zfs_freebsd_ioctl(ap)
5883 struct vop_ioctl_args /* {
5893 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data,
5894 ap->a_fflag, ap->a_cred, NULL, NULL));
5898 zfs_freebsd_read(ap)
5899 struct vop_read_args /* {
5903 struct ucred *a_cred;
5907 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5912 zfs_freebsd_write(ap)
5913 struct vop_write_args /* {
5917 struct ucred *a_cred;
5921 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5926 zfs_freebsd_access(ap)
5927 struct vop_access_args /* {
5929 accmode_t a_accmode;
5930 struct ucred *a_cred;
5931 struct thread *a_td;
5934 vnode_t *vp = ap->a_vp;
5935 znode_t *zp = VTOZ(vp);
5940 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND,
5942 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND);
5944 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL);
5947 * VADMIN has to be handled by vaccess().
5950 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND);
5952 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
5953 zp->z_gid, accmode, ap->a_cred, NULL);
5958 * For VEXEC, ensure that at least one execute bit is set for
5961 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR &&
5962 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
5970 zfs_freebsd_lookup(ap)
5971 struct vop_lookup_args /* {
5972 struct vnode *a_dvp;
5973 struct vnode **a_vpp;
5974 struct componentname *a_cnp;
5977 struct componentname *cnp = ap->a_cnp;
5978 char nm[NAME_MAX + 1];
5980 ASSERT(cnp->cn_namelen < sizeof(nm));
5981 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm)));
5983 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop,
5984 cnp->cn_cred, cnp->cn_thread, 0));
5988 zfs_freebsd_create(ap)
5989 struct vop_create_args /* {
5990 struct vnode *a_dvp;
5991 struct vnode **a_vpp;
5992 struct componentname *a_cnp;
5993 struct vattr *a_vap;
5996 struct componentname *cnp = ap->a_cnp;
5997 vattr_t *vap = ap->a_vap;
6000 ASSERT(cnp->cn_flags & SAVENAME);
6002 vattr_init_mask(vap);
6003 mode = vap->va_mode & ALLPERMS;
6005 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode,
6006 ap->a_vpp, cnp->cn_cred, cnp->cn_thread));
6010 zfs_freebsd_remove(ap)
6011 struct vop_remove_args /* {
6012 struct vnode *a_dvp;
6014 struct componentname *a_cnp;
6018 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6020 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr,
6021 ap->a_cnp->cn_cred, NULL, 0));
6025 zfs_freebsd_mkdir(ap)
6026 struct vop_mkdir_args /* {
6027 struct vnode *a_dvp;
6028 struct vnode **a_vpp;
6029 struct componentname *a_cnp;
6030 struct vattr *a_vap;
6033 vattr_t *vap = ap->a_vap;
6035 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6037 vattr_init_mask(vap);
6039 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp,
6040 ap->a_cnp->cn_cred, NULL, 0, NULL));
6044 zfs_freebsd_rmdir(ap)
6045 struct vop_rmdir_args /* {
6046 struct vnode *a_dvp;
6048 struct componentname *a_cnp;
6051 struct componentname *cnp = ap->a_cnp;
6053 ASSERT(cnp->cn_flags & SAVENAME);
6055 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0));
6059 zfs_freebsd_readdir(ap)
6060 struct vop_readdir_args /* {
6063 struct ucred *a_cred;
6070 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag,
6071 ap->a_ncookies, ap->a_cookies));
6075 zfs_freebsd_fsync(ap)
6076 struct vop_fsync_args /* {
6079 struct thread *a_td;
6084 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL));
6088 zfs_freebsd_getattr(ap)
6089 struct vop_getattr_args /* {
6091 struct vattr *a_vap;
6092 struct ucred *a_cred;
6095 vattr_t *vap = ap->a_vap;
6101 xvap.xva_vattr = *vap;
6102 xvap.xva_vattr.va_mask |= AT_XVATTR;
6104 /* Convert chflags into ZFS-type flags. */
6105 /* XXX: what about SF_SETTABLE?. */
6106 XVA_SET_REQ(&xvap, XAT_IMMUTABLE);
6107 XVA_SET_REQ(&xvap, XAT_APPENDONLY);
6108 XVA_SET_REQ(&xvap, XAT_NOUNLINK);
6109 XVA_SET_REQ(&xvap, XAT_NODUMP);
6110 XVA_SET_REQ(&xvap, XAT_READONLY);
6111 XVA_SET_REQ(&xvap, XAT_ARCHIVE);
6112 XVA_SET_REQ(&xvap, XAT_SYSTEM);
6113 XVA_SET_REQ(&xvap, XAT_HIDDEN);
6114 XVA_SET_REQ(&xvap, XAT_REPARSE);
6115 XVA_SET_REQ(&xvap, XAT_OFFLINE);
6116 XVA_SET_REQ(&xvap, XAT_SPARSE);
6118 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL);
6122 /* Convert ZFS xattr into chflags. */
6123 #define FLAG_CHECK(fflag, xflag, xfield) do { \
6124 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \
6125 fflags |= (fflag); \
6127 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE,
6128 xvap.xva_xoptattrs.xoa_immutable);
6129 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY,
6130 xvap.xva_xoptattrs.xoa_appendonly);
6131 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK,
6132 xvap.xva_xoptattrs.xoa_nounlink);
6133 FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE,
6134 xvap.xva_xoptattrs.xoa_archive);
6135 FLAG_CHECK(UF_NODUMP, XAT_NODUMP,
6136 xvap.xva_xoptattrs.xoa_nodump);
6137 FLAG_CHECK(UF_READONLY, XAT_READONLY,
6138 xvap.xva_xoptattrs.xoa_readonly);
6139 FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM,
6140 xvap.xva_xoptattrs.xoa_system);
6141 FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN,
6142 xvap.xva_xoptattrs.xoa_hidden);
6143 FLAG_CHECK(UF_REPARSE, XAT_REPARSE,
6144 xvap.xva_xoptattrs.xoa_reparse);
6145 FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE,
6146 xvap.xva_xoptattrs.xoa_offline);
6147 FLAG_CHECK(UF_SPARSE, XAT_SPARSE,
6148 xvap.xva_xoptattrs.xoa_sparse);
6151 *vap = xvap.xva_vattr;
6152 vap->va_flags = fflags;
6157 zfs_freebsd_setattr(ap)
6158 struct vop_setattr_args /* {
6160 struct vattr *a_vap;
6161 struct ucred *a_cred;
6164 vnode_t *vp = ap->a_vp;
6165 vattr_t *vap = ap->a_vap;
6166 cred_t *cred = ap->a_cred;
6171 vattr_init_mask(vap);
6172 vap->va_mask &= ~AT_NOSET;
6175 xvap.xva_vattr = *vap;
6177 zflags = VTOZ(vp)->z_pflags;
6179 if (vap->va_flags != VNOVAL) {
6180 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs;
6183 if (zfsvfs->z_use_fuids == B_FALSE)
6184 return (EOPNOTSUPP);
6186 fflags = vap->va_flags;
6189 * We need to figure out whether it makes sense to allow
6190 * UF_REPARSE through, since we don't really have other
6191 * facilities to handle reparse points and zfs_setattr()
6192 * doesn't currently allow setting that attribute anyway.
6194 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE|
6195 UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE|
6196 UF_OFFLINE|UF_SPARSE)) != 0)
6197 return (EOPNOTSUPP);
6199 * Unprivileged processes are not permitted to unset system
6200 * flags, or modify flags if any system flags are set.
6201 * Privileged non-jail processes may not modify system flags
6202 * if securelevel > 0 and any existing system flags are set.
6203 * Privileged jail processes behave like privileged non-jail
6204 * processes if the security.jail.chflags_allowed sysctl is
6205 * is non-zero; otherwise, they behave like unprivileged
6208 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 ||
6209 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) {
6211 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6212 error = securelevel_gt(cred, 0);
6218 * Callers may only modify the file flags on objects they
6219 * have VADMIN rights for.
6221 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0)
6224 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6228 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) {
6233 #define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \
6234 if (((fflags & (fflag)) && !(zflags & (zflag))) || \
6235 ((zflags & (zflag)) && !(fflags & (fflag)))) { \
6236 XVA_SET_REQ(&xvap, (xflag)); \
6237 (xfield) = ((fflags & (fflag)) != 0); \
6240 /* Convert chflags into ZFS-type flags. */
6241 /* XXX: what about SF_SETTABLE?. */
6242 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE,
6243 xvap.xva_xoptattrs.xoa_immutable);
6244 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY,
6245 xvap.xva_xoptattrs.xoa_appendonly);
6246 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK,
6247 xvap.xva_xoptattrs.xoa_nounlink);
6248 FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE,
6249 xvap.xva_xoptattrs.xoa_archive);
6250 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP,
6251 xvap.xva_xoptattrs.xoa_nodump);
6252 FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY,
6253 xvap.xva_xoptattrs.xoa_readonly);
6254 FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM,
6255 xvap.xva_xoptattrs.xoa_system);
6256 FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN,
6257 xvap.xva_xoptattrs.xoa_hidden);
6258 FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE,
6259 xvap.xva_xoptattrs.xoa_hidden);
6260 FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE,
6261 xvap.xva_xoptattrs.xoa_offline);
6262 FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE,
6263 xvap.xva_xoptattrs.xoa_sparse);
6266 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL));
6270 zfs_freebsd_rename(ap)
6271 struct vop_rename_args /* {
6272 struct vnode *a_fdvp;
6273 struct vnode *a_fvp;
6274 struct componentname *a_fcnp;
6275 struct vnode *a_tdvp;
6276 struct vnode *a_tvp;
6277 struct componentname *a_tcnp;
6280 vnode_t *fdvp = ap->a_fdvp;
6281 vnode_t *fvp = ap->a_fvp;
6282 vnode_t *tdvp = ap->a_tdvp;
6283 vnode_t *tvp = ap->a_tvp;
6286 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART));
6287 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART));
6290 * Check for cross-device rename.
6292 if ((fdvp->v_mount != tdvp->v_mount) ||
6293 (tvp && (fdvp->v_mount != tvp->v_mount)))
6296 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp,
6297 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0);
6311 zfs_freebsd_symlink(ap)
6312 struct vop_symlink_args /* {
6313 struct vnode *a_dvp;
6314 struct vnode **a_vpp;
6315 struct componentname *a_cnp;
6316 struct vattr *a_vap;
6320 struct componentname *cnp = ap->a_cnp;
6321 vattr_t *vap = ap->a_vap;
6323 ASSERT(cnp->cn_flags & SAVENAME);
6325 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */
6326 vattr_init_mask(vap);
6328 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap,
6329 ap->a_target, cnp->cn_cred, cnp->cn_thread));
6333 zfs_freebsd_readlink(ap)
6334 struct vop_readlink_args /* {
6337 struct ucred *a_cred;
6341 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL));
6345 zfs_freebsd_link(ap)
6346 struct vop_link_args /* {
6347 struct vnode *a_tdvp;
6349 struct componentname *a_cnp;
6352 struct componentname *cnp = ap->a_cnp;
6353 vnode_t *vp = ap->a_vp;
6354 vnode_t *tdvp = ap->a_tdvp;
6356 if (tdvp->v_mount != vp->v_mount)
6359 ASSERT(cnp->cn_flags & SAVENAME);
6361 return (zfs_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0));
6365 zfs_freebsd_inactive(ap)
6366 struct vop_inactive_args /* {
6368 struct thread *a_td;
6371 vnode_t *vp = ap->a_vp;
6373 zfs_inactive(vp, ap->a_td->td_ucred, NULL);
6378 zfs_freebsd_reclaim(ap)
6379 struct vop_reclaim_args /* {
6381 struct thread *a_td;
6384 vnode_t *vp = ap->a_vp;
6385 znode_t *zp = VTOZ(vp);
6386 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
6390 /* Destroy the vm object and flush associated pages. */
6391 vnode_destroy_vobject(vp);
6394 * z_teardown_inactive_lock protects from a race with
6395 * zfs_znode_dmu_fini in zfsvfs_teardown during
6398 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
6399 if (zp->z_sa_hdl == NULL)
6403 rw_exit(&zfsvfs->z_teardown_inactive_lock);
6411 struct vop_fid_args /* {
6417 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL));
6421 zfs_freebsd_pathconf(ap)
6422 struct vop_pathconf_args /* {
6425 register_t *a_retval;
6431 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL);
6433 *ap->a_retval = val;
6434 else if (error == EOPNOTSUPP)
6435 error = vop_stdpathconf(ap);
6440 zfs_freebsd_fifo_pathconf(ap)
6441 struct vop_pathconf_args /* {
6444 register_t *a_retval;
6448 switch (ap->a_name) {
6449 case _PC_ACL_EXTENDED:
6451 case _PC_ACL_PATH_MAX:
6452 case _PC_MAC_PRESENT:
6453 return (zfs_freebsd_pathconf(ap));
6455 return (fifo_specops.vop_pathconf(ap));
6460 * FreeBSD's extended attributes namespace defines file name prefix for ZFS'
6461 * extended attribute name:
6464 * system freebsd:system:
6465 * user (none, can be used to access ZFS fsattr(5) attributes
6466 * created on Solaris)
6469 zfs_create_attrname(int attrnamespace, const char *name, char *attrname,
6472 const char *namespace, *prefix, *suffix;
6474 /* We don't allow '/' character in attribute name. */
6475 if (strchr(name, '/') != NULL)
6477 /* We don't allow attribute names that start with "freebsd:" string. */
6478 if (strncmp(name, "freebsd:", 8) == 0)
6481 bzero(attrname, size);
6483 switch (attrnamespace) {
6484 case EXTATTR_NAMESPACE_USER:
6486 prefix = "freebsd:";
6487 namespace = EXTATTR_NAMESPACE_USER_STRING;
6491 * This is the default namespace by which we can access all
6492 * attributes created on Solaris.
6494 prefix = namespace = suffix = "";
6497 case EXTATTR_NAMESPACE_SYSTEM:
6498 prefix = "freebsd:";
6499 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING;
6502 case EXTATTR_NAMESPACE_EMPTY:
6506 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix,
6508 return (ENAMETOOLONG);
6514 * Vnode operating to retrieve a named extended attribute.
6517 zfs_getextattr(struct vop_getextattr_args *ap)
6520 IN struct vnode *a_vp;
6521 IN int a_attrnamespace;
6522 IN const char *a_name;
6523 INOUT struct uio *a_uio;
6525 IN struct ucred *a_cred;
6526 IN struct thread *a_td;
6530 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6531 struct thread *td = ap->a_td;
6532 struct nameidata nd;
6535 vnode_t *xvp = NULL, *vp;
6538 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6539 ap->a_cred, ap->a_td, VREAD);
6543 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6550 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6558 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6560 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL);
6562 NDFREE(&nd, NDF_ONLY_PNBUF);
6565 if (error == ENOENT)
6570 if (ap->a_size != NULL) {
6571 error = VOP_GETATTR(vp, &va, ap->a_cred);
6573 *ap->a_size = (size_t)va.va_size;
6574 } else if (ap->a_uio != NULL)
6575 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6578 vn_close(vp, flags, ap->a_cred, td);
6585 * Vnode operation to remove a named attribute.
6588 zfs_deleteextattr(struct vop_deleteextattr_args *ap)
6591 IN struct vnode *a_vp;
6592 IN int a_attrnamespace;
6593 IN const char *a_name;
6594 IN struct ucred *a_cred;
6595 IN struct thread *a_td;
6599 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6600 struct thread *td = ap->a_td;
6601 struct nameidata nd;
6604 vnode_t *xvp = NULL, *vp;
6607 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6608 ap->a_cred, ap->a_td, VWRITE);
6612 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6619 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6626 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF,
6627 UIO_SYSSPACE, attrname, xvp, td);
6630 NDFREE(&nd, NDF_ONLY_PNBUF);
6633 if (error == ENOENT)
6637 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
6640 if (vp == nd.ni_dvp)
6650 * Vnode operation to set a named attribute.
6653 zfs_setextattr(struct vop_setextattr_args *ap)
6656 IN struct vnode *a_vp;
6657 IN int a_attrnamespace;
6658 IN const char *a_name;
6659 INOUT struct uio *a_uio;
6660 IN struct ucred *a_cred;
6661 IN struct thread *a_td;
6665 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6666 struct thread *td = ap->a_td;
6667 struct nameidata nd;
6670 vnode_t *xvp = NULL, *vp;
6673 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6674 ap->a_cred, ap->a_td, VWRITE);
6678 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6685 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6686 LOOKUP_XATTR | CREATE_XATTR_DIR);
6692 flags = FFLAGS(O_WRONLY | O_CREAT);
6693 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6695 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL);
6697 NDFREE(&nd, NDF_ONLY_PNBUF);
6705 error = VOP_SETATTR(vp, &va, ap->a_cred);
6707 VOP_WRITE(vp, ap->a_uio, IO_UNIT | IO_SYNC, ap->a_cred);
6710 vn_close(vp, flags, ap->a_cred, td);
6717 * Vnode operation to retrieve extended attributes on a vnode.
6720 zfs_listextattr(struct vop_listextattr_args *ap)
6723 IN struct vnode *a_vp;
6724 IN int a_attrnamespace;
6725 INOUT struct uio *a_uio;
6727 IN struct ucred *a_cred;
6728 IN struct thread *a_td;
6732 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6733 struct thread *td = ap->a_td;
6734 struct nameidata nd;
6735 char attrprefix[16];
6736 u_char dirbuf[sizeof(struct dirent)];
6739 struct uio auio, *uio = ap->a_uio;
6740 size_t *sizep = ap->a_size;
6742 vnode_t *xvp = NULL, *vp;
6743 int done, error, eof, pos;
6745 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6746 ap->a_cred, ap->a_td, VREAD);
6750 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix,
6751 sizeof(attrprefix));
6754 plen = strlen(attrprefix);
6761 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6766 * ENOATTR means that the EA directory does not yet exist,
6767 * i.e. there are no extended attributes there.
6769 if (error == ENOATTR)
6774 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED,
6775 UIO_SYSSPACE, ".", xvp, td);
6778 NDFREE(&nd, NDF_ONLY_PNBUF);
6784 auio.uio_iov = &aiov;
6785 auio.uio_iovcnt = 1;
6786 auio.uio_segflg = UIO_SYSSPACE;
6788 auio.uio_rw = UIO_READ;
6789 auio.uio_offset = 0;
6794 aiov.iov_base = (void *)dirbuf;
6795 aiov.iov_len = sizeof(dirbuf);
6796 auio.uio_resid = sizeof(dirbuf);
6797 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL);
6798 done = sizeof(dirbuf) - auio.uio_resid;
6801 for (pos = 0; pos < done;) {
6802 dp = (struct dirent *)(dirbuf + pos);
6803 pos += dp->d_reclen;
6805 * XXX: Temporarily we also accept DT_UNKNOWN, as this
6806 * is what we get when attribute was created on Solaris.
6808 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN)
6810 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0)
6812 else if (strncmp(dp->d_name, attrprefix, plen) != 0)
6814 nlen = dp->d_namlen - plen;
6817 else if (uio != NULL) {
6819 * Format of extattr name entry is one byte for
6820 * length and the rest for name.
6822 error = uiomove(&nlen, 1, uio->uio_rw, uio);
6824 error = uiomove(dp->d_name + plen, nlen,
6831 } while (!eof && error == 0);
6840 zfs_freebsd_getacl(ap)
6841 struct vop_getacl_args /* {
6850 vsecattr_t vsecattr;
6852 if (ap->a_type != ACL_TYPE_NFS4)
6855 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
6856 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL))
6859 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt);
6860 if (vsecattr.vsa_aclentp != NULL)
6861 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz);
6867 zfs_freebsd_setacl(ap)
6868 struct vop_setacl_args /* {
6877 vsecattr_t vsecattr;
6878 int aclbsize; /* size of acl list in bytes */
6881 if (ap->a_type != ACL_TYPE_NFS4)
6884 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES)
6888 * With NFSv4 ACLs, chmod(2) may need to add additional entries,
6889 * splitting every entry into two and appending "canonical six"
6890 * entries at the end. Don't allow for setting an ACL that would
6891 * cause chmod(2) to run out of ACL entries.
6893 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES)
6896 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR);
6900 vsecattr.vsa_mask = VSA_ACE;
6901 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t);
6902 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP);
6903 aaclp = vsecattr.vsa_aclentp;
6904 vsecattr.vsa_aclentsz = aclbsize;
6906 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp);
6907 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL);
6908 kmem_free(aaclp, aclbsize);
6914 zfs_freebsd_aclcheck(ap)
6915 struct vop_aclcheck_args /* {
6924 return (EOPNOTSUPP);
6927 struct vop_vector zfs_vnodeops;
6928 struct vop_vector zfs_fifoops;
6929 struct vop_vector zfs_shareops;
6931 struct vop_vector zfs_vnodeops = {
6932 .vop_default = &default_vnodeops,
6933 .vop_inactive = zfs_freebsd_inactive,
6934 .vop_reclaim = zfs_freebsd_reclaim,
6935 .vop_access = zfs_freebsd_access,
6936 #ifdef FREEBSD_NAMECACHE
6937 .vop_lookup = vfs_cache_lookup,
6938 .vop_cachedlookup = zfs_freebsd_lookup,
6940 .vop_lookup = zfs_freebsd_lookup,
6942 .vop_getattr = zfs_freebsd_getattr,
6943 .vop_setattr = zfs_freebsd_setattr,
6944 .vop_create = zfs_freebsd_create,
6945 .vop_mknod = zfs_freebsd_create,
6946 .vop_mkdir = zfs_freebsd_mkdir,
6947 .vop_readdir = zfs_freebsd_readdir,
6948 .vop_fsync = zfs_freebsd_fsync,
6949 .vop_open = zfs_freebsd_open,
6950 .vop_close = zfs_freebsd_close,
6951 .vop_rmdir = zfs_freebsd_rmdir,
6952 .vop_ioctl = zfs_freebsd_ioctl,
6953 .vop_link = zfs_freebsd_link,
6954 .vop_symlink = zfs_freebsd_symlink,
6955 .vop_readlink = zfs_freebsd_readlink,
6956 .vop_read = zfs_freebsd_read,
6957 .vop_write = zfs_freebsd_write,
6958 .vop_remove = zfs_freebsd_remove,
6959 .vop_rename = zfs_freebsd_rename,
6960 .vop_pathconf = zfs_freebsd_pathconf,
6961 .vop_bmap = zfs_freebsd_bmap,
6962 .vop_fid = zfs_freebsd_fid,
6963 .vop_getextattr = zfs_getextattr,
6964 .vop_deleteextattr = zfs_deleteextattr,
6965 .vop_setextattr = zfs_setextattr,
6966 .vop_listextattr = zfs_listextattr,
6967 .vop_getacl = zfs_freebsd_getacl,
6968 .vop_setacl = zfs_freebsd_setacl,
6969 .vop_aclcheck = zfs_freebsd_aclcheck,
6970 .vop_getpages = zfs_freebsd_getpages,
6973 struct vop_vector zfs_fifoops = {
6974 .vop_default = &fifo_specops,
6975 .vop_fsync = zfs_freebsd_fsync,
6976 .vop_access = zfs_freebsd_access,
6977 .vop_getattr = zfs_freebsd_getattr,
6978 .vop_inactive = zfs_freebsd_inactive,
6979 .vop_read = VOP_PANIC,
6980 .vop_reclaim = zfs_freebsd_reclaim,
6981 .vop_setattr = zfs_freebsd_setattr,
6982 .vop_write = VOP_PANIC,
6983 .vop_pathconf = zfs_freebsd_fifo_pathconf,
6984 .vop_fid = zfs_freebsd_fid,
6985 .vop_getacl = zfs_freebsd_getacl,
6986 .vop_setacl = zfs_freebsd_setacl,
6987 .vop_aclcheck = zfs_freebsd_aclcheck,
6991 * special share hidden files vnode operations template
6993 struct vop_vector zfs_shareops = {
6994 .vop_default = &default_vnodeops,
6995 .vop_access = zfs_freebsd_access,
6996 .vop_inactive = zfs_freebsd_inactive,
6997 .vop_reclaim = zfs_freebsd_reclaim,
6998 .vop_fid = zfs_freebsd_fid,
6999 .vop_pathconf = zfs_freebsd_pathconf,