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) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
111 * dmu_tx_assign(). This is critical because we don't want to block
112 * while holding locks.
114 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
115 * reduces lock contention and CPU usage when we must wait (note that if
116 * throughput is constrained by the storage, nearly every transaction
119 * Note, in particular, that if a lock is sometimes acquired before
120 * the tx assigns, and sometimes after (e.g. z_lock), then failing
121 * to use a non-blocking assign can deadlock the system. The scenario:
123 * Thread A has grabbed a lock before calling dmu_tx_assign().
124 * Thread B is in an already-assigned tx, and blocks for this lock.
125 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
126 * forever, because the previous txg can't quiesce until B's tx commits.
128 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
129 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
130 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
131 * to indicate that this operation has already called dmu_tx_wait().
132 * This will ensure that we don't retry forever, waiting a short bit
135 * (5) If the operation succeeded, generate the intent log entry for it
136 * before dropping locks. This ensures that the ordering of events
137 * in the intent log matches the order in which they actually occurred.
138 * During ZIL replay the zfs_log_* functions will update the sequence
139 * number to indicate the zil transaction has replayed.
141 * (6) At the end of each vnode op, the DMU tx must always commit,
142 * regardless of whether there were any errors.
144 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
145 * to ensure that synchronous semantics are provided when necessary.
147 * In general, this is how things should be ordered in each vnode op:
149 * ZFS_ENTER(zfsvfs); // exit if unmounted
151 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD())
152 * rw_enter(...); // grab any other locks you need
153 * tx = dmu_tx_create(...); // get DMU tx
154 * dmu_tx_hold_*(); // hold each object you might modify
155 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
157 * rw_exit(...); // drop locks
158 * zfs_dirent_unlock(dl); // unlock directory entry
159 * VN_RELE(...); // release held vnodes
160 * if (error == ERESTART) {
166 * dmu_tx_abort(tx); // abort DMU tx
167 * ZFS_EXIT(zfsvfs); // finished in zfs
168 * return (error); // really out of space
170 * error = do_real_work(); // do whatever this VOP does
172 * zfs_log_*(...); // on success, make ZIL entry
173 * dmu_tx_commit(tx); // commit DMU tx -- error or not
174 * rw_exit(...); // drop locks
175 * zfs_dirent_unlock(dl); // unlock directory entry
176 * VN_RELE(...); // release held vnodes
177 * zil_commit(zilog, foid); // synchronous when necessary
178 * ZFS_EXIT(zfsvfs); // finished in zfs
179 * return (error); // done, report error
184 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
186 znode_t *zp = VTOZ(*vpp);
187 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
192 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
193 ((flag & FAPPEND) == 0)) {
195 return (SET_ERROR(EPERM));
198 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
199 ZTOV(zp)->v_type == VREG &&
200 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
201 if (fs_vscan(*vpp, cr, 0) != 0) {
203 return (SET_ERROR(EACCES));
207 /* Keep a count of the synchronous opens in the znode */
208 if (flag & (FSYNC | FDSYNC))
209 atomic_inc_32(&zp->z_sync_cnt);
217 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
218 caller_context_t *ct)
220 znode_t *zp = VTOZ(vp);
221 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
224 * Clean up any locks held by this process on the vp.
226 cleanlocks(vp, ddi_get_pid(), 0);
227 cleanshares(vp, ddi_get_pid());
232 /* Decrement the synchronous opens in the znode */
233 if ((flag & (FSYNC | FDSYNC)) && (count == 1))
234 atomic_dec_32(&zp->z_sync_cnt);
236 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
237 ZTOV(zp)->v_type == VREG &&
238 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
239 VERIFY(fs_vscan(vp, cr, 1) == 0);
246 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
247 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
250 zfs_holey(vnode_t *vp, u_long cmd, offset_t *off)
252 znode_t *zp = VTOZ(vp);
253 uint64_t noff = (uint64_t)*off; /* new offset */
258 file_sz = zp->z_size;
259 if (noff >= file_sz) {
260 return (SET_ERROR(ENXIO));
263 if (cmd == _FIO_SEEK_HOLE)
268 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
271 if ((error == ESRCH) || (noff > file_sz)) {
273 * Handle the virtual hole at the end of file.
279 return (SET_ERROR(ENXIO));
290 zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred,
291 int *rvalp, caller_context_t *ct)
303 * The following two ioctls are used by bfu. Faking out,
304 * necessary to avoid bfu errors.
313 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
314 return (SET_ERROR(EFAULT));
316 off = *(offset_t *)data;
319 zfsvfs = zp->z_zfsvfs;
323 /* offset parameter is in/out */
324 error = zfs_holey(vp, com, &off);
329 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
330 return (SET_ERROR(EFAULT));
332 *(offset_t *)data = off;
336 return (SET_ERROR(ENOTTY));
340 page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes)
347 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE
348 * aligned boundaries, if the range is not aligned. As a result a
349 * DEV_BSIZE subrange with partially dirty data may get marked as clean.
350 * It may happen that all DEV_BSIZE subranges are marked clean and thus
351 * the whole page would be considred clean despite have some dirty data.
352 * For this reason we should shrink the range to DEV_BSIZE aligned
353 * boundaries before calling vm_page_clear_dirty.
355 end = rounddown2(off + nbytes, DEV_BSIZE);
356 off = roundup2(off, DEV_BSIZE);
360 zfs_vmobject_assert_wlocked(obj);
363 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
365 if (vm_page_xbusied(pp)) {
367 * Reference the page before unlocking and
368 * sleeping so that the page daemon is less
369 * likely to reclaim it.
371 vm_page_reference(pp);
373 zfs_vmobject_wunlock(obj);
374 vm_page_busy_sleep(pp, "zfsmwb");
375 zfs_vmobject_wlock(obj);
379 } else if (pp == NULL) {
380 pp = vm_page_alloc(obj, OFF_TO_IDX(start),
381 VM_ALLOC_SYSTEM | VM_ALLOC_IFCACHED |
384 ASSERT(pp != NULL && !pp->valid);
389 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
390 vm_object_pip_add(obj, 1);
391 pmap_remove_write(pp);
393 vm_page_clear_dirty(pp, off, nbytes);
401 page_unbusy(vm_page_t pp)
405 vm_object_pip_subtract(pp->object, 1);
409 page_hold(vnode_t *vp, int64_t start)
415 zfs_vmobject_assert_wlocked(obj);
418 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
420 if (vm_page_xbusied(pp)) {
422 * Reference the page before unlocking and
423 * sleeping so that the page daemon is less
424 * likely to reclaim it.
426 vm_page_reference(pp);
428 zfs_vmobject_wunlock(obj);
429 vm_page_busy_sleep(pp, "zfsmwb");
430 zfs_vmobject_wlock(obj);
434 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
447 page_unhold(vm_page_t pp)
456 zfs_map_page(vm_page_t pp, struct sf_buf **sfp)
459 *sfp = sf_buf_alloc(pp, 0);
460 return ((caddr_t)sf_buf_kva(*sfp));
464 zfs_unmap_page(struct sf_buf *sf)
471 * When a file is memory mapped, we must keep the IO data synchronized
472 * between the DMU cache and the memory mapped pages. What this means:
474 * On Write: If we find a memory mapped page, we write to *both*
475 * the page and the dmu buffer.
478 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid,
479 int segflg, dmu_tx_t *tx)
486 ASSERT(vp->v_mount != NULL);
490 off = start & PAGEOFFSET;
491 zfs_vmobject_wlock(obj);
492 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
494 int nbytes = imin(PAGESIZE - off, len);
496 if (segflg == UIO_NOCOPY) {
497 pp = vm_page_lookup(obj, OFF_TO_IDX(start));
499 ("zfs update_pages: NULL page in putpages case"));
501 ("zfs update_pages: unaligned data in putpages case"));
502 KASSERT(pp->valid == VM_PAGE_BITS_ALL,
503 ("zfs update_pages: invalid page in putpages case"));
504 KASSERT(vm_page_sbusied(pp),
505 ("zfs update_pages: unbusy page in putpages case"));
506 KASSERT(!pmap_page_is_write_mapped(pp),
507 ("zfs update_pages: writable page in putpages case"));
508 zfs_vmobject_wunlock(obj);
510 va = zfs_map_page(pp, &sf);
511 (void) dmu_write(os, oid, start, nbytes, va, tx);
514 zfs_vmobject_wlock(obj);
516 } else if ((pp = page_busy(vp, start, off, nbytes)) != NULL) {
517 zfs_vmobject_wunlock(obj);
519 va = zfs_map_page(pp, &sf);
520 (void) dmu_read(os, oid, start+off, nbytes,
521 va+off, DMU_READ_PREFETCH);;
524 zfs_vmobject_wlock(obj);
530 if (segflg != UIO_NOCOPY)
531 vm_object_pip_wakeupn(obj, 0);
532 zfs_vmobject_wunlock(obj);
536 * Read with UIO_NOCOPY flag means that sendfile(2) requests
537 * ZFS to populate a range of page cache pages with data.
539 * NOTE: this function could be optimized to pre-allocate
540 * all pages in advance, drain exclusive busy on all of them,
541 * map them into contiguous KVA region and populate them
542 * in one single dmu_read() call.
545 mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio)
547 znode_t *zp = VTOZ(vp);
548 objset_t *os = zp->z_zfsvfs->z_os;
558 ASSERT(uio->uio_segflg == UIO_NOCOPY);
559 ASSERT(vp->v_mount != NULL);
562 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0);
564 zfs_vmobject_wlock(obj);
565 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) {
566 int bytes = MIN(PAGESIZE, len);
568 pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_SBUSY |
569 VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY);
570 if (pp->valid == 0) {
571 zfs_vmobject_wunlock(obj);
572 va = zfs_map_page(pp, &sf);
573 error = dmu_read(os, zp->z_id, start, bytes, va,
575 if (bytes != PAGESIZE && error == 0)
576 bzero(va + bytes, PAGESIZE - bytes);
578 zfs_vmobject_wlock(obj);
582 if (pp->wire_count == 0 && pp->valid == 0 &&
586 pp->valid = VM_PAGE_BITS_ALL;
587 vm_page_activate(pp);
591 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
596 uio->uio_resid -= bytes;
597 uio->uio_offset += bytes;
600 zfs_vmobject_wunlock(obj);
605 * When a file is memory mapped, we must keep the IO data synchronized
606 * between the DMU cache and the memory mapped pages. What this means:
608 * On Read: We "read" preferentially from memory mapped pages,
609 * else we default from the dmu buffer.
611 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
612 * the file is memory mapped.
615 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
617 znode_t *zp = VTOZ(vp);
618 objset_t *os = zp->z_zfsvfs->z_os;
626 ASSERT(vp->v_mount != NULL);
630 start = uio->uio_loffset;
631 off = start & PAGEOFFSET;
632 zfs_vmobject_wlock(obj);
633 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
635 uint64_t bytes = MIN(PAGESIZE - off, len);
637 if (pp = page_hold(vp, start)) {
641 zfs_vmobject_wunlock(obj);
642 va = zfs_map_page(pp, &sf);
643 error = uiomove(va + off, bytes, UIO_READ, uio);
645 zfs_vmobject_wlock(obj);
648 zfs_vmobject_wunlock(obj);
649 error = dmu_read_uio(os, zp->z_id, uio, bytes);
650 zfs_vmobject_wlock(obj);
657 zfs_vmobject_wunlock(obj);
661 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
664 * Read bytes from specified file into supplied buffer.
666 * IN: vp - vnode of file to be read from.
667 * uio - structure supplying read location, range info,
669 * ioflag - SYNC flags; used to provide FRSYNC semantics.
670 * cr - credentials of caller.
671 * ct - caller context
673 * OUT: uio - updated offset and range, buffer filled.
675 * RETURN: 0 on success, error code on failure.
678 * vp - atime updated if byte count > 0
682 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
684 znode_t *zp = VTOZ(vp);
685 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
696 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
698 return (SET_ERROR(EACCES));
702 * Validate file offset
704 if (uio->uio_loffset < (offset_t)0) {
706 return (SET_ERROR(EINVAL));
710 * Fasttrack empty reads
712 if (uio->uio_resid == 0) {
718 * Check for mandatory locks
720 if (MANDMODE(zp->z_mode)) {
721 if (error = chklock(vp, FREAD,
722 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
729 * If we're in FRSYNC mode, sync out this znode before reading it.
732 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
733 zil_commit(zfsvfs->z_log, zp->z_id);
736 * Lock the range against changes.
738 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
741 * If we are reading past end-of-file we can skip
742 * to the end; but we might still need to set atime.
744 if (uio->uio_loffset >= zp->z_size) {
749 ASSERT(uio->uio_loffset < zp->z_size);
750 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
753 if ((uio->uio_extflg == UIO_XUIO) &&
754 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
756 int blksz = zp->z_blksz;
757 uint64_t offset = uio->uio_loffset;
759 xuio = (xuio_t *)uio;
761 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
764 ASSERT(offset + n <= blksz);
767 (void) dmu_xuio_init(xuio, nblk);
769 if (vn_has_cached_data(vp)) {
771 * For simplicity, we always allocate a full buffer
772 * even if we only expect to read a portion of a block.
774 while (--nblk >= 0) {
775 (void) dmu_xuio_add(xuio,
776 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
784 nbytes = MIN(n, zfs_read_chunk_size -
785 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
788 if (uio->uio_segflg == UIO_NOCOPY)
789 error = mappedread_sf(vp, nbytes, uio);
791 #endif /* __FreeBSD__ */
792 if (vn_has_cached_data(vp))
793 error = mappedread(vp, nbytes, uio);
795 error = dmu_read_uio(os, zp->z_id, uio, nbytes);
797 /* convert checksum errors into IO errors */
799 error = SET_ERROR(EIO);
806 zfs_range_unlock(rl);
808 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
814 * Write the bytes to a file.
816 * IN: vp - vnode of file to be written to.
817 * uio - structure supplying write location, range info,
819 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is
820 * set if in append mode.
821 * cr - credentials of caller.
822 * ct - caller context (NFS/CIFS fem monitor only)
824 * OUT: uio - updated offset and range.
826 * RETURN: 0 on success, error code on failure.
829 * vp - ctime|mtime updated if byte count > 0
834 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
836 znode_t *zp = VTOZ(vp);
837 rlim64_t limit = MAXOFFSET_T;
838 ssize_t start_resid = uio->uio_resid;
842 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
847 int max_blksz = zfsvfs->z_max_blksz;
850 iovec_t *aiov = NULL;
853 int iovcnt = uio->uio_iovcnt;
854 iovec_t *iovp = uio->uio_iov;
857 sa_bulk_attr_t bulk[4];
858 uint64_t mtime[2], ctime[2];
861 * Fasttrack empty write
867 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
873 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
874 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
875 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
877 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
881 * If immutable or not appending then return EPERM
883 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
884 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
885 (uio->uio_loffset < zp->z_size))) {
887 return (SET_ERROR(EPERM));
890 zilog = zfsvfs->z_log;
893 * Validate file offset
895 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
898 return (SET_ERROR(EINVAL));
902 * Check for mandatory locks before calling zfs_range_lock()
903 * in order to prevent a deadlock with locks set via fcntl().
905 if (MANDMODE((mode_t)zp->z_mode) &&
906 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
913 * Pre-fault the pages to ensure slow (eg NFS) pages
915 * Skip this if uio contains loaned arc_buf.
917 if ((uio->uio_extflg == UIO_XUIO) &&
918 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
919 xuio = (xuio_t *)uio;
921 uio_prefaultpages(MIN(n, max_blksz), uio);
925 * If in append mode, set the io offset pointer to eof.
927 if (ioflag & FAPPEND) {
929 * Obtain an appending range lock to guarantee file append
930 * semantics. We reset the write offset once we have the lock.
932 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
934 if (rl->r_len == UINT64_MAX) {
936 * We overlocked the file because this write will cause
937 * the file block size to increase.
938 * Note that zp_size cannot change with this lock held.
942 uio->uio_loffset = woff;
945 * Note that if the file block size will change as a result of
946 * this write, then this range lock will lock the entire file
947 * so that we can re-write the block safely.
949 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
952 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) {
953 zfs_range_unlock(rl);
959 zfs_range_unlock(rl);
961 return (SET_ERROR(EFBIG));
964 if ((woff + n) > limit || woff > (limit - n))
967 /* Will this write extend the file length? */
968 write_eof = (woff + n > zp->z_size);
970 end_size = MAX(zp->z_size, woff + n);
973 * Write the file in reasonable size chunks. Each chunk is written
974 * in a separate transaction; this keeps the intent log records small
975 * and allows us to do more fine-grained space accounting.
979 woff = uio->uio_loffset;
980 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
981 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
983 dmu_return_arcbuf(abuf);
984 error = SET_ERROR(EDQUOT);
988 if (xuio && abuf == NULL) {
989 ASSERT(i_iov < iovcnt);
991 abuf = dmu_xuio_arcbuf(xuio, i_iov);
992 dmu_xuio_clear(xuio, i_iov);
993 DTRACE_PROBE3(zfs_cp_write, int, i_iov,
994 iovec_t *, aiov, arc_buf_t *, abuf);
995 ASSERT((aiov->iov_base == abuf->b_data) ||
996 ((char *)aiov->iov_base - (char *)abuf->b_data +
997 aiov->iov_len == arc_buf_size(abuf)));
999 } else if (abuf == NULL && n >= max_blksz &&
1000 woff >= zp->z_size &&
1001 P2PHASE(woff, max_blksz) == 0 &&
1002 zp->z_blksz == max_blksz) {
1004 * This write covers a full block. "Borrow" a buffer
1005 * from the dmu so that we can fill it before we enter
1006 * a transaction. This avoids the possibility of
1007 * holding up the transaction if the data copy hangs
1008 * up on a pagefault (e.g., from an NFS server mapping).
1012 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
1014 ASSERT(abuf != NULL);
1015 ASSERT(arc_buf_size(abuf) == max_blksz);
1016 if (error = uiocopy(abuf->b_data, max_blksz,
1017 UIO_WRITE, uio, &cbytes)) {
1018 dmu_return_arcbuf(abuf);
1021 ASSERT(cbytes == max_blksz);
1025 * Start a transaction.
1027 tx = dmu_tx_create(zfsvfs->z_os);
1028 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1029 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
1030 zfs_sa_upgrade_txholds(tx, zp);
1031 error = dmu_tx_assign(tx, TXG_WAIT);
1035 dmu_return_arcbuf(abuf);
1040 * If zfs_range_lock() over-locked we grow the blocksize
1041 * and then reduce the lock range. This will only happen
1042 * on the first iteration since zfs_range_reduce() will
1043 * shrink down r_len to the appropriate size.
1045 if (rl->r_len == UINT64_MAX) {
1048 if (zp->z_blksz > max_blksz) {
1049 ASSERT(!ISP2(zp->z_blksz));
1050 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
1052 new_blksz = MIN(end_size, max_blksz);
1054 zfs_grow_blocksize(zp, new_blksz, tx);
1055 zfs_range_reduce(rl, woff, n);
1059 * XXX - should we really limit each write to z_max_blksz?
1060 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
1062 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
1064 if (woff + nbytes > zp->z_size)
1065 vnode_pager_setsize(vp, woff + nbytes);
1068 tx_bytes = uio->uio_resid;
1069 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
1071 tx_bytes -= uio->uio_resid;
1074 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
1076 * If this is not a full block write, but we are
1077 * extending the file past EOF and this data starts
1078 * block-aligned, use assign_arcbuf(). Otherwise,
1079 * write via dmu_write().
1081 if (tx_bytes < max_blksz && (!write_eof ||
1082 aiov->iov_base != abuf->b_data)) {
1084 dmu_write(zfsvfs->z_os, zp->z_id, woff,
1085 aiov->iov_len, aiov->iov_base, tx);
1086 dmu_return_arcbuf(abuf);
1087 xuio_stat_wbuf_copied();
1089 ASSERT(xuio || tx_bytes == max_blksz);
1090 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
1093 ASSERT(tx_bytes <= uio->uio_resid);
1094 uioskip(uio, tx_bytes);
1096 if (tx_bytes && vn_has_cached_data(vp)) {
1097 update_pages(vp, woff, tx_bytes, zfsvfs->z_os,
1098 zp->z_id, uio->uio_segflg, tx);
1102 * If we made no progress, we're done. If we made even
1103 * partial progress, update the znode and ZIL accordingly.
1105 if (tx_bytes == 0) {
1106 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
1107 (void *)&zp->z_size, sizeof (uint64_t), tx);
1114 * Clear Set-UID/Set-GID bits on successful write if not
1115 * privileged and at least one of the excute bits is set.
1117 * It would be nice to to this after all writes have
1118 * been done, but that would still expose the ISUID/ISGID
1119 * to another app after the partial write is committed.
1121 * Note: we don't call zfs_fuid_map_id() here because
1122 * user 0 is not an ephemeral uid.
1124 mutex_enter(&zp->z_acl_lock);
1125 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
1126 (S_IXUSR >> 6))) != 0 &&
1127 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
1128 secpolicy_vnode_setid_retain(vp, cr,
1129 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
1131 zp->z_mode &= ~(S_ISUID | S_ISGID);
1132 newmode = zp->z_mode;
1133 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
1134 (void *)&newmode, sizeof (uint64_t), tx);
1136 mutex_exit(&zp->z_acl_lock);
1138 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
1142 * Update the file size (zp_size) if it has changed;
1143 * account for possible concurrent updates.
1145 while ((end_size = zp->z_size) < uio->uio_loffset) {
1146 (void) atomic_cas_64(&zp->z_size, end_size,
1151 * If we are replaying and eof is non zero then force
1152 * the file size to the specified eof. Note, there's no
1153 * concurrency during replay.
1155 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
1156 zp->z_size = zfsvfs->z_replay_eof;
1158 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1160 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
1165 ASSERT(tx_bytes == nbytes);
1170 uio_prefaultpages(MIN(n, max_blksz), uio);
1174 zfs_range_unlock(rl);
1177 * If we're in replay mode, or we made no progress, return error.
1178 * Otherwise, it's at least a partial write, so it's successful.
1180 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1185 if (ioflag & (FSYNC | FDSYNC) ||
1186 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1187 zil_commit(zilog, zp->z_id);
1194 zfs_get_done(zgd_t *zgd, int error)
1196 znode_t *zp = zgd->zgd_private;
1197 objset_t *os = zp->z_zfsvfs->z_os;
1200 dmu_buf_rele(zgd->zgd_db, zgd);
1202 zfs_range_unlock(zgd->zgd_rl);
1205 * Release the vnode asynchronously as we currently have the
1206 * txg stopped from syncing.
1208 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1210 if (error == 0 && zgd->zgd_bp)
1211 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1213 kmem_free(zgd, sizeof (zgd_t));
1217 static int zil_fault_io = 0;
1221 * Get data to generate a TX_WRITE intent log record.
1224 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1226 zfsvfs_t *zfsvfs = arg;
1227 objset_t *os = zfsvfs->z_os;
1229 uint64_t object = lr->lr_foid;
1230 uint64_t offset = lr->lr_offset;
1231 uint64_t size = lr->lr_length;
1232 blkptr_t *bp = &lr->lr_blkptr;
1237 ASSERT(zio != NULL);
1241 * Nothing to do if the file has been removed
1243 if (zfs_zget(zfsvfs, object, &zp) != 0)
1244 return (SET_ERROR(ENOENT));
1245 if (zp->z_unlinked) {
1247 * Release the vnode asynchronously as we currently have the
1248 * txg stopped from syncing.
1250 VN_RELE_ASYNC(ZTOV(zp),
1251 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1252 return (SET_ERROR(ENOENT));
1255 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1256 zgd->zgd_zilog = zfsvfs->z_log;
1257 zgd->zgd_private = zp;
1260 * Write records come in two flavors: immediate and indirect.
1261 * For small writes it's cheaper to store the data with the
1262 * log record (immediate); for large writes it's cheaper to
1263 * sync the data and get a pointer to it (indirect) so that
1264 * we don't have to write the data twice.
1266 if (buf != NULL) { /* immediate write */
1267 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1268 /* test for truncation needs to be done while range locked */
1269 if (offset >= zp->z_size) {
1270 error = SET_ERROR(ENOENT);
1272 error = dmu_read(os, object, offset, size, buf,
1273 DMU_READ_NO_PREFETCH);
1275 ASSERT(error == 0 || error == ENOENT);
1276 } else { /* indirect write */
1278 * Have to lock the whole block to ensure when it's
1279 * written out and it's checksum is being calculated
1280 * that no one can change the data. We need to re-check
1281 * blocksize after we get the lock in case it's changed!
1286 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1288 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1290 if (zp->z_blksz == size)
1293 zfs_range_unlock(zgd->zgd_rl);
1295 /* test for truncation needs to be done while range locked */
1296 if (lr->lr_offset >= zp->z_size)
1297 error = SET_ERROR(ENOENT);
1300 error = SET_ERROR(EIO);
1305 error = dmu_buf_hold(os, object, offset, zgd, &db,
1306 DMU_READ_NO_PREFETCH);
1309 blkptr_t *obp = dmu_buf_get_blkptr(db);
1311 ASSERT(BP_IS_HOLE(bp));
1318 ASSERT(db->db_offset == offset);
1319 ASSERT(db->db_size == size);
1321 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1323 ASSERT(error || lr->lr_length <= zp->z_blksz);
1326 * On success, we need to wait for the write I/O
1327 * initiated by dmu_sync() to complete before we can
1328 * release this dbuf. We will finish everything up
1329 * in the zfs_get_done() callback.
1334 if (error == EALREADY) {
1335 lr->lr_common.lrc_txtype = TX_WRITE2;
1341 zfs_get_done(zgd, error);
1348 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1349 caller_context_t *ct)
1351 znode_t *zp = VTOZ(vp);
1352 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1358 if (flag & V_ACE_MASK)
1359 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1361 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1368 * If vnode is for a device return a specfs vnode instead.
1371 specvp_check(vnode_t **vpp, cred_t *cr)
1375 if (IS_DEVVP(*vpp)) {
1378 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1381 error = SET_ERROR(ENOSYS);
1389 * Lookup an entry in a directory, or an extended attribute directory.
1390 * If it exists, return a held vnode reference for it.
1392 * IN: dvp - vnode of directory to search.
1393 * nm - name of entry to lookup.
1394 * pnp - full pathname to lookup [UNUSED].
1395 * flags - LOOKUP_XATTR set if looking for an attribute.
1396 * rdir - root directory vnode [UNUSED].
1397 * cr - credentials of caller.
1398 * ct - caller context
1399 * direntflags - directory lookup flags
1400 * realpnp - returned pathname.
1402 * OUT: vpp - vnode of located entry, NULL if not found.
1404 * RETURN: 0 on success, error code on failure.
1411 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp,
1412 int nameiop, cred_t *cr, kthread_t *td, int flags)
1414 znode_t *zdp = VTOZ(dvp);
1415 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1417 int *direntflags = NULL;
1418 void *realpnp = NULL;
1421 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1423 if (dvp->v_type != VDIR) {
1424 return (SET_ERROR(ENOTDIR));
1425 } else if (zdp->z_sa_hdl == NULL) {
1426 return (SET_ERROR(EIO));
1429 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1430 error = zfs_fastaccesschk_execute(zdp, cr);
1438 vnode_t *tvp = dnlc_lookup(dvp, nm);
1441 error = zfs_fastaccesschk_execute(zdp, cr);
1446 if (tvp == DNLC_NO_VNODE) {
1448 return (SET_ERROR(ENOENT));
1451 return (specvp_check(vpp, cr));
1457 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1464 if (flags & LOOKUP_XATTR) {
1467 * If the xattr property is off, refuse the lookup request.
1469 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1471 return (SET_ERROR(EINVAL));
1476 * We don't allow recursive attributes..
1477 * Maybe someday we will.
1479 if (zdp->z_pflags & ZFS_XATTR) {
1481 return (SET_ERROR(EINVAL));
1484 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1490 * Do we have permission to get into attribute directory?
1493 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1503 if (dvp->v_type != VDIR) {
1505 return (SET_ERROR(ENOTDIR));
1509 * Check accessibility of directory.
1512 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1517 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1518 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1520 return (SET_ERROR(EILSEQ));
1523 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1525 error = specvp_check(vpp, cr);
1527 /* Translate errors and add SAVENAME when needed. */
1528 if (cnp->cn_flags & ISLASTCN) {
1532 if (error == ENOENT) {
1533 error = EJUSTRETURN;
1534 cnp->cn_flags |= SAVENAME;
1540 cnp->cn_flags |= SAVENAME;
1544 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) {
1547 if (cnp->cn_flags & ISDOTDOT) {
1548 ltype = VOP_ISLOCKED(dvp);
1552 error = vn_lock(*vpp, cnp->cn_lkflags);
1553 if (cnp->cn_flags & ISDOTDOT)
1554 vn_lock(dvp, ltype | LK_RETRY);
1564 #ifdef FREEBSD_NAMECACHE
1566 * Insert name into cache (as non-existent) if appropriate.
1568 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
1569 cache_enter(dvp, *vpp, cnp);
1571 * Insert name into cache if appropriate.
1573 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) {
1574 if (!(cnp->cn_flags & ISLASTCN) ||
1575 (nameiop != DELETE && nameiop != RENAME)) {
1576 cache_enter(dvp, *vpp, cnp);
1585 * Attempt to create a new entry in a directory. If the entry
1586 * already exists, truncate the file if permissible, else return
1587 * an error. Return the vp of the created or trunc'd file.
1589 * IN: dvp - vnode of directory to put new file entry in.
1590 * name - name of new file entry.
1591 * vap - attributes of new file.
1592 * excl - flag indicating exclusive or non-exclusive mode.
1593 * mode - mode to open file with.
1594 * cr - credentials of caller.
1595 * flag - large file flag [UNUSED].
1596 * ct - caller context
1597 * vsecp - ACL to be set
1599 * OUT: vpp - vnode of created or trunc'd entry.
1601 * RETURN: 0 on success, error code on failure.
1604 * dvp - ctime|mtime updated if new entry created
1605 * vp - ctime|mtime always, atime if new
1610 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode,
1611 vnode_t **vpp, cred_t *cr, kthread_t *td)
1613 znode_t *zp, *dzp = VTOZ(dvp);
1614 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1622 gid_t gid = crgetgid(cr);
1623 zfs_acl_ids_t acl_ids;
1624 boolean_t fuid_dirtied;
1625 boolean_t have_acl = B_FALSE;
1626 boolean_t waited = B_FALSE;
1631 * If we have an ephemeral id, ACL, or XVATTR then
1632 * make sure file system is at proper version
1635 ksid = crgetsid(cr, KSID_OWNER);
1637 uid = ksid_getid(ksid);
1641 if (zfsvfs->z_use_fuids == B_FALSE &&
1642 (vsecp || (vap->va_mask & AT_XVATTR) ||
1643 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1644 return (SET_ERROR(EINVAL));
1649 zilog = zfsvfs->z_log;
1651 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1652 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1654 return (SET_ERROR(EILSEQ));
1657 if (vap->va_mask & AT_XVATTR) {
1658 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
1659 crgetuid(cr), cr, vap->va_type)) != 0) {
1667 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr))
1668 vap->va_mode &= ~S_ISVTX;
1670 if (*name == '\0') {
1672 * Null component name refers to the directory itself.
1679 /* possible VN_HOLD(zp) */
1682 if (flag & FIGNORECASE)
1685 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1689 zfs_acl_ids_free(&acl_ids);
1690 if (strcmp(name, "..") == 0)
1691 error = SET_ERROR(EISDIR);
1701 * Create a new file object and update the directory
1704 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1706 zfs_acl_ids_free(&acl_ids);
1711 * We only support the creation of regular files in
1712 * extended attribute directories.
1715 if ((dzp->z_pflags & ZFS_XATTR) &&
1716 (vap->va_type != VREG)) {
1718 zfs_acl_ids_free(&acl_ids);
1719 error = SET_ERROR(EINVAL);
1723 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1724 cr, vsecp, &acl_ids)) != 0)
1728 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1729 zfs_acl_ids_free(&acl_ids);
1730 error = SET_ERROR(EDQUOT);
1734 tx = dmu_tx_create(os);
1736 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1737 ZFS_SA_BASE_ATTR_SIZE);
1739 fuid_dirtied = zfsvfs->z_fuid_dirty;
1741 zfs_fuid_txhold(zfsvfs, tx);
1742 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1743 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1744 if (!zfsvfs->z_use_sa &&
1745 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1746 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1747 0, acl_ids.z_aclp->z_acl_bytes);
1749 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1751 zfs_dirent_unlock(dl);
1752 if (error == ERESTART) {
1758 zfs_acl_ids_free(&acl_ids);
1763 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1766 zfs_fuid_sync(zfsvfs, tx);
1768 (void) zfs_link_create(dl, zp, tx, ZNEW);
1769 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1770 if (flag & FIGNORECASE)
1772 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1773 vsecp, acl_ids.z_fuidp, vap);
1774 zfs_acl_ids_free(&acl_ids);
1777 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1780 zfs_acl_ids_free(&acl_ids);
1784 * A directory entry already exists for this name.
1787 * Can't truncate an existing file if in exclusive mode.
1790 error = SET_ERROR(EEXIST);
1794 * Can't open a directory for writing.
1796 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1797 error = SET_ERROR(EISDIR);
1801 * Verify requested access to file.
1803 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1807 mutex_enter(&dzp->z_lock);
1809 mutex_exit(&dzp->z_lock);
1812 * Truncate regular files if requested.
1814 if ((ZTOV(zp)->v_type == VREG) &&
1815 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1816 /* we can't hold any locks when calling zfs_freesp() */
1817 zfs_dirent_unlock(dl);
1819 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1821 vnevent_create(ZTOV(zp), ct);
1827 zfs_dirent_unlock(dl);
1834 error = specvp_check(vpp, cr);
1837 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1838 zil_commit(zilog, 0);
1845 * Remove an entry from a directory.
1847 * IN: dvp - vnode of directory to remove entry from.
1848 * name - name of entry to remove.
1849 * cr - credentials of caller.
1850 * ct - caller context
1851 * flags - case flags
1853 * RETURN: 0 on success, error code on failure.
1857 * vp - ctime (if nlink > 0)
1860 uint64_t null_xattr = 0;
1864 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1867 znode_t *zp, *dzp = VTOZ(dvp);
1870 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1872 uint64_t acl_obj, xattr_obj;
1873 uint64_t xattr_obj_unlinked = 0;
1877 boolean_t may_delete_now, delete_now = FALSE;
1878 boolean_t unlinked, toobig = FALSE;
1880 pathname_t *realnmp = NULL;
1884 boolean_t waited = B_FALSE;
1888 zilog = zfsvfs->z_log;
1890 if (flags & FIGNORECASE) {
1900 * Attempt to lock directory; fail if entry doesn't exist.
1902 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1912 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1917 * Need to use rmdir for removing directories.
1919 if (vp->v_type == VDIR) {
1920 error = SET_ERROR(EPERM);
1924 vnevent_remove(vp, dvp, name, ct);
1927 dnlc_remove(dvp, realnmp->pn_buf);
1929 dnlc_remove(dvp, name);
1932 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1936 * We may delete the znode now, or we may put it in the unlinked set;
1937 * it depends on whether we're the last link, and on whether there are
1938 * other holds on the vnode. So we dmu_tx_hold() the right things to
1939 * allow for either case.
1942 tx = dmu_tx_create(zfsvfs->z_os);
1943 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1944 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1945 zfs_sa_upgrade_txholds(tx, zp);
1946 zfs_sa_upgrade_txholds(tx, dzp);
1947 if (may_delete_now) {
1949 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1950 /* if the file is too big, only hold_free a token amount */
1951 dmu_tx_hold_free(tx, zp->z_id, 0,
1952 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1955 /* are there any extended attributes? */
1956 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1957 &xattr_obj, sizeof (xattr_obj));
1958 if (error == 0 && xattr_obj) {
1959 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1961 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1962 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1965 mutex_enter(&zp->z_lock);
1966 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1967 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1968 mutex_exit(&zp->z_lock);
1970 /* charge as an update -- would be nice not to charge at all */
1971 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1973 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1975 zfs_dirent_unlock(dl);
1979 if (error == ERESTART) {
1993 * Remove the directory entry.
1995 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
2005 * Hold z_lock so that we can make sure that the ACL obj
2006 * hasn't changed. Could have been deleted due to
2009 mutex_enter(&zp->z_lock);
2011 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
2012 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
2013 delete_now = may_delete_now && !toobig &&
2014 vp->v_count == 1 && !vn_has_cached_data(vp) &&
2015 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
2022 panic("zfs_remove: delete_now branch taken");
2024 if (xattr_obj_unlinked) {
2025 ASSERT3U(xzp->z_links, ==, 2);
2026 mutex_enter(&xzp->z_lock);
2027 xzp->z_unlinked = 1;
2029 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
2030 &xzp->z_links, sizeof (xzp->z_links), tx);
2031 ASSERT3U(error, ==, 0);
2032 mutex_exit(&xzp->z_lock);
2033 zfs_unlinked_add(xzp, tx);
2036 error = sa_remove(zp->z_sa_hdl,
2037 SA_ZPL_XATTR(zfsvfs), tx);
2039 error = sa_update(zp->z_sa_hdl,
2040 SA_ZPL_XATTR(zfsvfs), &null_xattr,
2041 sizeof (uint64_t), tx);
2046 ASSERT0(vp->v_count);
2048 mutex_exit(&zp->z_lock);
2049 zfs_znode_delete(zp, tx);
2050 } else if (unlinked) {
2051 mutex_exit(&zp->z_lock);
2052 zfs_unlinked_add(zp, tx);
2054 vp->v_vflag |= VV_NOSYNC;
2059 if (flags & FIGNORECASE)
2061 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
2068 zfs_dirent_unlock(dl);
2075 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2076 zil_commit(zilog, 0);
2083 * Create a new directory and insert it into dvp using the name
2084 * provided. Return a pointer to the inserted directory.
2086 * IN: dvp - vnode of directory to add subdir to.
2087 * dirname - name of new directory.
2088 * vap - attributes of new directory.
2089 * cr - credentials of caller.
2090 * ct - caller context
2091 * flags - case flags
2092 * vsecp - ACL to be set
2094 * OUT: vpp - vnode of created directory.
2096 * RETURN: 0 on success, error code on failure.
2099 * dvp - ctime|mtime updated
2100 * vp - ctime|mtime|atime updated
2104 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
2105 caller_context_t *ct, int flags, vsecattr_t *vsecp)
2107 znode_t *zp, *dzp = VTOZ(dvp);
2108 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2117 gid_t gid = crgetgid(cr);
2118 zfs_acl_ids_t acl_ids;
2119 boolean_t fuid_dirtied;
2120 boolean_t waited = B_FALSE;
2122 ASSERT(vap->va_type == VDIR);
2125 * If we have an ephemeral id, ACL, or XVATTR then
2126 * make sure file system is at proper version
2129 ksid = crgetsid(cr, KSID_OWNER);
2131 uid = ksid_getid(ksid);
2134 if (zfsvfs->z_use_fuids == B_FALSE &&
2135 (vsecp || (vap->va_mask & AT_XVATTR) ||
2136 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
2137 return (SET_ERROR(EINVAL));
2141 zilog = zfsvfs->z_log;
2143 if (dzp->z_pflags & ZFS_XATTR) {
2145 return (SET_ERROR(EINVAL));
2148 if (zfsvfs->z_utf8 && u8_validate(dirname,
2149 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2151 return (SET_ERROR(EILSEQ));
2153 if (flags & FIGNORECASE)
2156 if (vap->va_mask & AT_XVATTR) {
2157 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
2158 crgetuid(cr), cr, vap->va_type)) != 0) {
2164 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
2165 vsecp, &acl_ids)) != 0) {
2170 * First make sure the new directory doesn't exist.
2172 * Existence is checked first to make sure we don't return
2173 * EACCES instead of EEXIST which can cause some applications
2179 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
2181 zfs_acl_ids_free(&acl_ids);
2186 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
2187 zfs_acl_ids_free(&acl_ids);
2188 zfs_dirent_unlock(dl);
2193 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
2194 zfs_acl_ids_free(&acl_ids);
2195 zfs_dirent_unlock(dl);
2197 return (SET_ERROR(EDQUOT));
2201 * Add a new entry to the directory.
2203 tx = dmu_tx_create(zfsvfs->z_os);
2204 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2205 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2206 fuid_dirtied = zfsvfs->z_fuid_dirty;
2208 zfs_fuid_txhold(zfsvfs, tx);
2209 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2210 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2211 acl_ids.z_aclp->z_acl_bytes);
2214 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2215 ZFS_SA_BASE_ATTR_SIZE);
2217 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2219 zfs_dirent_unlock(dl);
2220 if (error == ERESTART) {
2226 zfs_acl_ids_free(&acl_ids);
2235 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2238 zfs_fuid_sync(zfsvfs, tx);
2241 * Now put new name in parent dir.
2243 (void) zfs_link_create(dl, zp, tx, ZNEW);
2247 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2248 if (flags & FIGNORECASE)
2250 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2251 acl_ids.z_fuidp, vap);
2253 zfs_acl_ids_free(&acl_ids);
2257 zfs_dirent_unlock(dl);
2259 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2260 zil_commit(zilog, 0);
2267 * Remove a directory subdir entry. If the current working
2268 * directory is the same as the subdir to be removed, the
2271 * IN: dvp - vnode of directory to remove from.
2272 * name - name of directory to be removed.
2273 * cwd - vnode of current working directory.
2274 * cr - credentials of caller.
2275 * ct - caller context
2276 * flags - case flags
2278 * RETURN: 0 on success, error code on failure.
2281 * dvp - ctime|mtime updated
2285 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2286 caller_context_t *ct, int flags)
2288 znode_t *dzp = VTOZ(dvp);
2291 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2297 boolean_t waited = B_FALSE;
2301 zilog = zfsvfs->z_log;
2303 if (flags & FIGNORECASE)
2309 * Attempt to lock directory; fail if entry doesn't exist.
2311 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2319 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2323 if (vp->v_type != VDIR) {
2324 error = SET_ERROR(ENOTDIR);
2329 error = SET_ERROR(EINVAL);
2333 vnevent_rmdir(vp, dvp, name, ct);
2336 * Grab a lock on the directory to make sure that noone is
2337 * trying to add (or lookup) entries while we are removing it.
2339 rw_enter(&zp->z_name_lock, RW_WRITER);
2342 * Grab a lock on the parent pointer to make sure we play well
2343 * with the treewalk and directory rename code.
2345 rw_enter(&zp->z_parent_lock, RW_WRITER);
2347 tx = dmu_tx_create(zfsvfs->z_os);
2348 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2349 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2350 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2351 zfs_sa_upgrade_txholds(tx, zp);
2352 zfs_sa_upgrade_txholds(tx, dzp);
2353 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2355 rw_exit(&zp->z_parent_lock);
2356 rw_exit(&zp->z_name_lock);
2357 zfs_dirent_unlock(dl);
2359 if (error == ERESTART) {
2370 #ifdef FREEBSD_NAMECACHE
2374 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2377 uint64_t txtype = TX_RMDIR;
2378 if (flags & FIGNORECASE)
2380 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2385 rw_exit(&zp->z_parent_lock);
2386 rw_exit(&zp->z_name_lock);
2387 #ifdef FREEBSD_NAMECACHE
2391 zfs_dirent_unlock(dl);
2395 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2396 zil_commit(zilog, 0);
2403 * Read as many directory entries as will fit into the provided
2404 * buffer from the given directory cursor position (specified in
2405 * the uio structure).
2407 * IN: vp - vnode of directory to read.
2408 * uio - structure supplying read location, range info,
2409 * and return buffer.
2410 * cr - credentials of caller.
2411 * ct - caller context
2412 * flags - case flags
2414 * OUT: uio - updated offset and range, buffer filled.
2415 * eofp - set to true if end-of-file detected.
2417 * RETURN: 0 on success, error code on failure.
2420 * vp - atime updated
2422 * Note that the low 4 bits of the cookie returned by zap is always zero.
2423 * This allows us to use the low range for "special" directory entries:
2424 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2425 * we use the offset 2 for the '.zfs' directory.
2429 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies)
2431 znode_t *zp = VTOZ(vp);
2435 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2440 zap_attribute_t zap;
2441 uint_t bytes_wanted;
2442 uint64_t offset; /* must be unsigned; checks for < 1 */
2448 boolean_t check_sysattrs;
2451 u_long *cooks = NULL;
2457 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2458 &parent, sizeof (parent))) != 0) {
2464 * If we are not given an eof variable,
2471 * Check for valid iov_len.
2473 if (uio->uio_iov->iov_len <= 0) {
2475 return (SET_ERROR(EINVAL));
2479 * Quit if directory has been removed (posix)
2481 if ((*eofp = zp->z_unlinked) != 0) {
2488 offset = uio->uio_loffset;
2489 prefetch = zp->z_zn_prefetch;
2492 * Initialize the iterator cursor.
2496 * Start iteration from the beginning of the directory.
2498 zap_cursor_init(&zc, os, zp->z_id);
2501 * The offset is a serialized cursor.
2503 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2507 * Get space to change directory entries into fs independent format.
2509 iovp = uio->uio_iov;
2510 bytes_wanted = iovp->iov_len;
2511 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2512 bufsize = bytes_wanted;
2513 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2514 odp = (struct dirent64 *)outbuf;
2516 bufsize = bytes_wanted;
2518 odp = (struct dirent64 *)iovp->iov_base;
2520 eodp = (struct edirent *)odp;
2522 if (ncookies != NULL) {
2524 * Minimum entry size is dirent size and 1 byte for a file name.
2526 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1);
2527 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK);
2532 * If this VFS supports the system attribute view interface; and
2533 * we're looking at an extended attribute directory; and we care
2534 * about normalization conflicts on this vfs; then we must check
2535 * for normalization conflicts with the sysattr name space.
2538 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2539 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2540 (flags & V_RDDIR_ENTFLAGS);
2546 * Transform to file-system independent format
2549 while (outcount < bytes_wanted) {
2552 off64_t *next = NULL;
2555 * Special case `.', `..', and `.zfs'.
2558 (void) strcpy(zap.za_name, ".");
2559 zap.za_normalization_conflict = 0;
2562 } else if (offset == 1) {
2563 (void) strcpy(zap.za_name, "..");
2564 zap.za_normalization_conflict = 0;
2567 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2568 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2569 zap.za_normalization_conflict = 0;
2570 objnum = ZFSCTL_INO_ROOT;
2576 if (error = zap_cursor_retrieve(&zc, &zap)) {
2577 if ((*eofp = (error == ENOENT)) != 0)
2583 if (zap.za_integer_length != 8 ||
2584 zap.za_num_integers != 1) {
2585 cmn_err(CE_WARN, "zap_readdir: bad directory "
2586 "entry, obj = %lld, offset = %lld\n",
2587 (u_longlong_t)zp->z_id,
2588 (u_longlong_t)offset);
2589 error = SET_ERROR(ENXIO);
2593 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2595 * MacOS X can extract the object type here such as:
2596 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2598 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2600 if (check_sysattrs && !zap.za_normalization_conflict) {
2602 zap.za_normalization_conflict =
2603 xattr_sysattr_casechk(zap.za_name);
2605 panic("%s:%u: TODO", __func__, __LINE__);
2610 if (flags & V_RDDIR_ACCFILTER) {
2612 * If we have no access at all, don't include
2613 * this entry in the returned information
2616 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2618 if (!zfs_has_access(ezp, cr)) {
2625 if (flags & V_RDDIR_ENTFLAGS)
2626 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2628 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2631 * Will this entry fit in the buffer?
2633 if (outcount + reclen > bufsize) {
2635 * Did we manage to fit anything in the buffer?
2638 error = SET_ERROR(EINVAL);
2643 if (flags & V_RDDIR_ENTFLAGS) {
2645 * Add extended flag entry:
2647 eodp->ed_ino = objnum;
2648 eodp->ed_reclen = reclen;
2649 /* NOTE: ed_off is the offset for the *next* entry */
2650 next = &(eodp->ed_off);
2651 eodp->ed_eflags = zap.za_normalization_conflict ?
2652 ED_CASE_CONFLICT : 0;
2653 (void) strncpy(eodp->ed_name, zap.za_name,
2654 EDIRENT_NAMELEN(reclen));
2655 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2660 odp->d_ino = objnum;
2661 odp->d_reclen = reclen;
2662 odp->d_namlen = strlen(zap.za_name);
2663 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1);
2665 odp = (dirent64_t *)((intptr_t)odp + reclen);
2669 ASSERT(outcount <= bufsize);
2671 /* Prefetch znode */
2673 dmu_prefetch(os, objnum, 0, 0);
2677 * Move to the next entry, fill in the previous offset.
2679 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2680 zap_cursor_advance(&zc);
2681 offset = zap_cursor_serialize(&zc);
2686 if (cooks != NULL) {
2689 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks));
2692 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2694 /* Subtract unused cookies */
2695 if (ncookies != NULL)
2696 *ncookies -= ncooks;
2698 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2699 iovp->iov_base += outcount;
2700 iovp->iov_len -= outcount;
2701 uio->uio_resid -= outcount;
2702 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2704 * Reset the pointer.
2706 offset = uio->uio_loffset;
2710 zap_cursor_fini(&zc);
2711 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2712 kmem_free(outbuf, bufsize);
2714 if (error == ENOENT)
2717 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2719 uio->uio_loffset = offset;
2721 if (error != 0 && cookies != NULL) {
2722 free(*cookies, M_TEMP);
2729 ulong_t zfs_fsync_sync_cnt = 4;
2732 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2734 znode_t *zp = VTOZ(vp);
2735 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2737 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2739 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2742 zil_commit(zfsvfs->z_log, zp->z_id);
2750 * Get the requested file attributes and place them in the provided
2753 * IN: vp - vnode of file.
2754 * vap - va_mask identifies requested attributes.
2755 * If AT_XVATTR set, then optional attrs are requested
2756 * flags - ATTR_NOACLCHECK (CIFS server context)
2757 * cr - credentials of caller.
2758 * ct - caller context
2760 * OUT: vap - attribute values.
2762 * RETURN: 0 (always succeeds).
2766 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2767 caller_context_t *ct)
2769 znode_t *zp = VTOZ(vp);
2770 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2773 u_longlong_t nblocks;
2775 uint64_t mtime[2], ctime[2], crtime[2], rdev;
2776 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2777 xoptattr_t *xoap = NULL;
2778 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2779 sa_bulk_attr_t bulk[4];
2785 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2787 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2788 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2789 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
2790 if (vp->v_type == VBLK || vp->v_type == VCHR)
2791 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
2794 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2800 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2801 * Also, if we are the owner don't bother, since owner should
2802 * always be allowed to read basic attributes of file.
2804 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2805 (vap->va_uid != crgetuid(cr))) {
2806 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2814 * Return all attributes. It's cheaper to provide the answer
2815 * than to determine whether we were asked the question.
2818 mutex_enter(&zp->z_lock);
2819 vap->va_type = IFTOVT(zp->z_mode);
2820 vap->va_mode = zp->z_mode & ~S_IFMT;
2822 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2824 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
2826 vap->va_nodeid = zp->z_id;
2827 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2828 links = zp->z_links + 1;
2830 links = zp->z_links;
2831 vap->va_nlink = MIN(links, LINK_MAX); /* nlink_t limit! */
2832 vap->va_size = zp->z_size;
2834 vap->va_rdev = vp->v_rdev;
2836 if (vp->v_type == VBLK || vp->v_type == VCHR)
2837 vap->va_rdev = zfs_cmpldev(rdev);
2839 vap->va_seq = zp->z_seq;
2840 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */
2843 * Add in any requested optional attributes and the create time.
2844 * Also set the corresponding bits in the returned attribute bitmap.
2846 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2847 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2849 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2850 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2853 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2854 xoap->xoa_readonly =
2855 ((zp->z_pflags & ZFS_READONLY) != 0);
2856 XVA_SET_RTN(xvap, XAT_READONLY);
2859 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2861 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2862 XVA_SET_RTN(xvap, XAT_SYSTEM);
2865 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2867 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2868 XVA_SET_RTN(xvap, XAT_HIDDEN);
2871 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2872 xoap->xoa_nounlink =
2873 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2874 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2877 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2878 xoap->xoa_immutable =
2879 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2880 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2883 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2884 xoap->xoa_appendonly =
2885 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2886 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2889 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2891 ((zp->z_pflags & ZFS_NODUMP) != 0);
2892 XVA_SET_RTN(xvap, XAT_NODUMP);
2895 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2897 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2898 XVA_SET_RTN(xvap, XAT_OPAQUE);
2901 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2902 xoap->xoa_av_quarantined =
2903 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2904 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2907 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2908 xoap->xoa_av_modified =
2909 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2910 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2913 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2914 vp->v_type == VREG) {
2915 zfs_sa_get_scanstamp(zp, xvap);
2918 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2921 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2922 times, sizeof (times));
2923 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2924 XVA_SET_RTN(xvap, XAT_CREATETIME);
2927 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2928 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2929 XVA_SET_RTN(xvap, XAT_REPARSE);
2931 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2932 xoap->xoa_generation = zp->z_gen;
2933 XVA_SET_RTN(xvap, XAT_GEN);
2936 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2938 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2939 XVA_SET_RTN(xvap, XAT_OFFLINE);
2942 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2944 ((zp->z_pflags & ZFS_SPARSE) != 0);
2945 XVA_SET_RTN(xvap, XAT_SPARSE);
2949 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2950 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2951 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2952 ZFS_TIME_DECODE(&vap->va_birthtime, crtime);
2954 mutex_exit(&zp->z_lock);
2956 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2957 vap->va_blksize = blksize;
2958 vap->va_bytes = nblocks << 9; /* nblocks * 512 */
2960 if (zp->z_blksz == 0) {
2962 * Block size hasn't been set; suggest maximal I/O transfers.
2964 vap->va_blksize = zfsvfs->z_max_blksz;
2972 * Set the file attributes to the values contained in the
2975 * IN: vp - vnode of file to be modified.
2976 * vap - new attribute values.
2977 * If AT_XVATTR set, then optional attrs are being set
2978 * flags - ATTR_UTIME set if non-default time values provided.
2979 * - ATTR_NOACLCHECK (CIFS context only).
2980 * cr - credentials of caller.
2981 * ct - caller context
2983 * RETURN: 0 on success, error code on failure.
2986 * vp - ctime updated, mtime updated if size changed.
2990 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2991 caller_context_t *ct)
2993 znode_t *zp = VTOZ(vp);
2994 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2999 uint_t mask = vap->va_mask;
3000 uint_t saved_mask = 0;
3001 uint64_t saved_mode;
3004 uint64_t new_uid, new_gid;
3006 uint64_t mtime[2], ctime[2];
3008 int need_policy = FALSE;
3010 zfs_fuid_info_t *fuidp = NULL;
3011 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
3014 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
3015 boolean_t fuid_dirtied = B_FALSE;
3016 sa_bulk_attr_t bulk[7], xattr_bulk[7];
3017 int count = 0, xattr_count = 0;
3022 if (mask & AT_NOSET)
3023 return (SET_ERROR(EINVAL));
3028 zilog = zfsvfs->z_log;
3031 * Make sure that if we have ephemeral uid/gid or xvattr specified
3032 * that file system is at proper version level
3035 if (zfsvfs->z_use_fuids == B_FALSE &&
3036 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
3037 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
3038 (mask & AT_XVATTR))) {
3040 return (SET_ERROR(EINVAL));
3043 if (mask & AT_SIZE && vp->v_type == VDIR) {
3045 return (SET_ERROR(EISDIR));
3048 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
3050 return (SET_ERROR(EINVAL));
3054 * If this is an xvattr_t, then get a pointer to the structure of
3055 * optional attributes. If this is NULL, then we have a vattr_t.
3057 xoap = xva_getxoptattr(xvap);
3059 xva_init(&tmpxvattr);
3062 * Immutable files can only alter immutable bit and atime
3064 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
3065 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
3066 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
3068 return (SET_ERROR(EPERM));
3071 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
3073 return (SET_ERROR(EPERM));
3077 * Verify timestamps doesn't overflow 32 bits.
3078 * ZFS can handle large timestamps, but 32bit syscalls can't
3079 * handle times greater than 2039. This check should be removed
3080 * once large timestamps are fully supported.
3082 if (mask & (AT_ATIME | AT_MTIME)) {
3083 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
3084 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
3086 return (SET_ERROR(EOVERFLOW));
3094 /* Can this be moved to before the top label? */
3095 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
3097 return (SET_ERROR(EROFS));
3101 * First validate permissions
3104 if (mask & AT_SIZE) {
3106 * XXX - Note, we are not providing any open
3107 * mode flags here (like FNDELAY), so we may
3108 * block if there are locks present... this
3109 * should be addressed in openat().
3111 /* XXX - would it be OK to generate a log record here? */
3112 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
3119 if (mask & (AT_ATIME|AT_MTIME) ||
3120 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
3121 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
3122 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
3123 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
3124 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
3125 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
3126 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
3127 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
3131 if (mask & (AT_UID|AT_GID)) {
3132 int idmask = (mask & (AT_UID|AT_GID));
3137 * NOTE: even if a new mode is being set,
3138 * we may clear S_ISUID/S_ISGID bits.
3141 if (!(mask & AT_MODE))
3142 vap->va_mode = zp->z_mode;
3145 * Take ownership or chgrp to group we are a member of
3148 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
3149 take_group = (mask & AT_GID) &&
3150 zfs_groupmember(zfsvfs, vap->va_gid, cr);
3153 * If both AT_UID and AT_GID are set then take_owner and
3154 * take_group must both be set in order to allow taking
3157 * Otherwise, send the check through secpolicy_vnode_setattr()
3161 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
3162 ((idmask == AT_UID) && take_owner) ||
3163 ((idmask == AT_GID) && take_group)) {
3164 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
3165 skipaclchk, cr) == 0) {
3167 * Remove setuid/setgid for non-privileged users
3169 secpolicy_setid_clear(vap, vp, cr);
3170 trim_mask = (mask & (AT_UID|AT_GID));
3179 mutex_enter(&zp->z_lock);
3180 oldva.va_mode = zp->z_mode;
3181 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
3182 if (mask & AT_XVATTR) {
3184 * Update xvattr mask to include only those attributes
3185 * that are actually changing.
3187 * the bits will be restored prior to actually setting
3188 * the attributes so the caller thinks they were set.
3190 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
3191 if (xoap->xoa_appendonly !=
3192 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
3195 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
3196 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
3200 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
3201 if (xoap->xoa_nounlink !=
3202 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
3205 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
3206 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
3210 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
3211 if (xoap->xoa_immutable !=
3212 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
3215 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
3216 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
3220 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
3221 if (xoap->xoa_nodump !=
3222 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
3225 XVA_CLR_REQ(xvap, XAT_NODUMP);
3226 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
3230 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3231 if (xoap->xoa_av_modified !=
3232 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3235 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3236 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
3240 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3241 if ((vp->v_type != VREG &&
3242 xoap->xoa_av_quarantined) ||
3243 xoap->xoa_av_quarantined !=
3244 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3247 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3248 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
3252 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3253 mutex_exit(&zp->z_lock);
3255 return (SET_ERROR(EPERM));
3258 if (need_policy == FALSE &&
3259 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3260 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3265 mutex_exit(&zp->z_lock);
3267 if (mask & AT_MODE) {
3268 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3269 err = secpolicy_setid_setsticky_clear(vp, vap,
3275 trim_mask |= AT_MODE;
3283 * If trim_mask is set then take ownership
3284 * has been granted or write_acl is present and user
3285 * has the ability to modify mode. In that case remove
3286 * UID|GID and or MODE from mask so that
3287 * secpolicy_vnode_setattr() doesn't revoke it.
3291 saved_mask = vap->va_mask;
3292 vap->va_mask &= ~trim_mask;
3293 if (trim_mask & AT_MODE) {
3295 * Save the mode, as secpolicy_vnode_setattr()
3296 * will overwrite it with ova.va_mode.
3298 saved_mode = vap->va_mode;
3301 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3302 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3309 vap->va_mask |= saved_mask;
3310 if (trim_mask & AT_MODE) {
3312 * Recover the mode after
3313 * secpolicy_vnode_setattr().
3315 vap->va_mode = saved_mode;
3321 * secpolicy_vnode_setattr, or take ownership may have
3324 mask = vap->va_mask;
3326 if ((mask & (AT_UID | AT_GID))) {
3327 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3328 &xattr_obj, sizeof (xattr_obj));
3330 if (err == 0 && xattr_obj) {
3331 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3335 if (mask & AT_UID) {
3336 new_uid = zfs_fuid_create(zfsvfs,
3337 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3338 if (new_uid != zp->z_uid &&
3339 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3341 VN_RELE(ZTOV(attrzp));
3342 err = SET_ERROR(EDQUOT);
3347 if (mask & AT_GID) {
3348 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3349 cr, ZFS_GROUP, &fuidp);
3350 if (new_gid != zp->z_gid &&
3351 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3353 VN_RELE(ZTOV(attrzp));
3354 err = SET_ERROR(EDQUOT);
3359 tx = dmu_tx_create(zfsvfs->z_os);
3361 if (mask & AT_MODE) {
3362 uint64_t pmode = zp->z_mode;
3364 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3366 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3367 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3368 err = SET_ERROR(EPERM);
3372 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3375 mutex_enter(&zp->z_lock);
3376 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3378 * Are we upgrading ACL from old V0 format
3381 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3382 zfs_znode_acl_version(zp) ==
3383 ZFS_ACL_VERSION_INITIAL) {
3384 dmu_tx_hold_free(tx, acl_obj, 0,
3386 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3387 0, aclp->z_acl_bytes);
3389 dmu_tx_hold_write(tx, acl_obj, 0,
3392 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3393 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3394 0, aclp->z_acl_bytes);
3396 mutex_exit(&zp->z_lock);
3397 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3399 if ((mask & AT_XVATTR) &&
3400 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3401 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3403 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3407 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3410 fuid_dirtied = zfsvfs->z_fuid_dirty;
3412 zfs_fuid_txhold(zfsvfs, tx);
3414 zfs_sa_upgrade_txholds(tx, zp);
3416 err = dmu_tx_assign(tx, TXG_WAIT);
3422 * Set each attribute requested.
3423 * We group settings according to the locks they need to acquire.
3425 * Note: you cannot set ctime directly, although it will be
3426 * updated as a side-effect of calling this function.
3430 if (mask & (AT_UID|AT_GID|AT_MODE))
3431 mutex_enter(&zp->z_acl_lock);
3432 mutex_enter(&zp->z_lock);
3434 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3435 &zp->z_pflags, sizeof (zp->z_pflags));
3438 if (mask & (AT_UID|AT_GID|AT_MODE))
3439 mutex_enter(&attrzp->z_acl_lock);
3440 mutex_enter(&attrzp->z_lock);
3441 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3442 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3443 sizeof (attrzp->z_pflags));
3446 if (mask & (AT_UID|AT_GID)) {
3448 if (mask & AT_UID) {
3449 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3450 &new_uid, sizeof (new_uid));
3451 zp->z_uid = new_uid;
3453 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3454 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3456 attrzp->z_uid = new_uid;
3460 if (mask & AT_GID) {
3461 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3462 NULL, &new_gid, sizeof (new_gid));
3463 zp->z_gid = new_gid;
3465 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3466 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3468 attrzp->z_gid = new_gid;
3471 if (!(mask & AT_MODE)) {
3472 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3473 NULL, &new_mode, sizeof (new_mode));
3474 new_mode = zp->z_mode;
3476 err = zfs_acl_chown_setattr(zp);
3479 err = zfs_acl_chown_setattr(attrzp);
3484 if (mask & AT_MODE) {
3485 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3486 &new_mode, sizeof (new_mode));
3487 zp->z_mode = new_mode;
3488 ASSERT3U((uintptr_t)aclp, !=, 0);
3489 err = zfs_aclset_common(zp, aclp, cr, tx);
3491 if (zp->z_acl_cached)
3492 zfs_acl_free(zp->z_acl_cached);
3493 zp->z_acl_cached = aclp;
3498 if (mask & AT_ATIME) {
3499 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3500 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3501 &zp->z_atime, sizeof (zp->z_atime));
3504 if (mask & AT_MTIME) {
3505 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3506 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3507 mtime, sizeof (mtime));
3510 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3511 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3512 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3513 NULL, mtime, sizeof (mtime));
3514 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3515 &ctime, sizeof (ctime));
3516 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3518 } else if (mask != 0) {
3519 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3520 &ctime, sizeof (ctime));
3521 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3524 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3525 SA_ZPL_CTIME(zfsvfs), NULL,
3526 &ctime, sizeof (ctime));
3527 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3528 mtime, ctime, B_TRUE);
3532 * Do this after setting timestamps to prevent timestamp
3533 * update from toggling bit
3536 if (xoap && (mask & AT_XVATTR)) {
3539 * restore trimmed off masks
3540 * so that return masks can be set for caller.
3543 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3544 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3546 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3547 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3549 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3550 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3552 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3553 XVA_SET_REQ(xvap, XAT_NODUMP);
3555 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3556 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3558 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3559 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3562 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3563 ASSERT(vp->v_type == VREG);
3565 zfs_xvattr_set(zp, xvap, tx);
3569 zfs_fuid_sync(zfsvfs, tx);
3572 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3574 mutex_exit(&zp->z_lock);
3575 if (mask & (AT_UID|AT_GID|AT_MODE))
3576 mutex_exit(&zp->z_acl_lock);
3579 if (mask & (AT_UID|AT_GID|AT_MODE))
3580 mutex_exit(&attrzp->z_acl_lock);
3581 mutex_exit(&attrzp->z_lock);
3584 if (err == 0 && attrzp) {
3585 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3591 VN_RELE(ZTOV(attrzp));
3597 zfs_fuid_info_free(fuidp);
3603 if (err == ERESTART)
3606 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3611 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3612 zil_commit(zilog, 0);
3618 typedef struct zfs_zlock {
3619 krwlock_t *zl_rwlock; /* lock we acquired */
3620 znode_t *zl_znode; /* znode we held */
3621 struct zfs_zlock *zl_next; /* next in list */
3625 * Drop locks and release vnodes that were held by zfs_rename_lock().
3628 zfs_rename_unlock(zfs_zlock_t **zlpp)
3632 while ((zl = *zlpp) != NULL) {
3633 if (zl->zl_znode != NULL)
3634 VN_RELE(ZTOV(zl->zl_znode));
3635 rw_exit(zl->zl_rwlock);
3636 *zlpp = zl->zl_next;
3637 kmem_free(zl, sizeof (*zl));
3642 * Search back through the directory tree, using the ".." entries.
3643 * Lock each directory in the chain to prevent concurrent renames.
3644 * Fail any attempt to move a directory into one of its own descendants.
3645 * XXX - z_parent_lock can overlap with map or grow locks
3648 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3652 uint64_t rootid = zp->z_zfsvfs->z_root;
3653 uint64_t oidp = zp->z_id;
3654 krwlock_t *rwlp = &szp->z_parent_lock;
3655 krw_t rw = RW_WRITER;
3658 * First pass write-locks szp and compares to zp->z_id.
3659 * Later passes read-lock zp and compare to zp->z_parent.
3662 if (!rw_tryenter(rwlp, rw)) {
3664 * Another thread is renaming in this path.
3665 * Note that if we are a WRITER, we don't have any
3666 * parent_locks held yet.
3668 if (rw == RW_READER && zp->z_id > szp->z_id) {
3670 * Drop our locks and restart
3672 zfs_rename_unlock(&zl);
3676 rwlp = &szp->z_parent_lock;
3681 * Wait for other thread to drop its locks
3687 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3688 zl->zl_rwlock = rwlp;
3689 zl->zl_znode = NULL;
3690 zl->zl_next = *zlpp;
3693 if (oidp == szp->z_id) /* We're a descendant of szp */
3694 return (SET_ERROR(EINVAL));
3696 if (oidp == rootid) /* We've hit the top */
3699 if (rw == RW_READER) { /* i.e. not the first pass */
3700 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3705 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3706 &oidp, sizeof (oidp));
3707 rwlp = &zp->z_parent_lock;
3710 } while (zp->z_id != sdzp->z_id);
3716 * Move an entry from the provided source directory to the target
3717 * directory. Change the entry name as indicated.
3719 * IN: sdvp - Source directory containing the "old entry".
3720 * snm - Old entry name.
3721 * tdvp - Target directory to contain the "new entry".
3722 * tnm - New entry name.
3723 * cr - credentials of caller.
3724 * ct - caller context
3725 * flags - case flags
3727 * RETURN: 0 on success, error code on failure.
3730 * sdvp,tdvp - ctime|mtime updated
3734 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3735 caller_context_t *ct, int flags)
3737 znode_t *tdzp, *szp, *tzp;
3738 znode_t *sdzp = VTOZ(sdvp);
3739 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs;
3742 zfs_dirlock_t *sdl, *tdl;
3745 int cmp, serr, terr;
3748 boolean_t waited = B_FALSE;
3751 ZFS_VERIFY_ZP(sdzp);
3752 zilog = zfsvfs->z_log;
3755 * Make sure we have the real vp for the target directory.
3757 if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3761 ZFS_VERIFY_ZP(tdzp);
3764 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3765 * ctldir appear to have the same v_vfsp.
3767 if (tdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) {
3769 return (SET_ERROR(EXDEV));
3772 if (zfsvfs->z_utf8 && u8_validate(tnm,
3773 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3775 return (SET_ERROR(EILSEQ));
3778 if (flags & FIGNORECASE)
3787 * This is to prevent the creation of links into attribute space
3788 * by renaming a linked file into/outof an attribute directory.
3789 * See the comment in zfs_link() for why this is considered bad.
3791 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3793 return (SET_ERROR(EINVAL));
3797 * Lock source and target directory entries. To prevent deadlock,
3798 * a lock ordering must be defined. We lock the directory with
3799 * the smallest object id first, or if it's a tie, the one with
3800 * the lexically first name.
3802 if (sdzp->z_id < tdzp->z_id) {
3804 } else if (sdzp->z_id > tdzp->z_id) {
3808 * First compare the two name arguments without
3809 * considering any case folding.
3811 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3813 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3814 ASSERT(error == 0 || !zfsvfs->z_utf8);
3817 * POSIX: "If the old argument and the new argument
3818 * both refer to links to the same existing file,
3819 * the rename() function shall return successfully
3820 * and perform no other action."
3826 * If the file system is case-folding, then we may
3827 * have some more checking to do. A case-folding file
3828 * system is either supporting mixed case sensitivity
3829 * access or is completely case-insensitive. Note
3830 * that the file system is always case preserving.
3832 * In mixed sensitivity mode case sensitive behavior
3833 * is the default. FIGNORECASE must be used to
3834 * explicitly request case insensitive behavior.
3836 * If the source and target names provided differ only
3837 * by case (e.g., a request to rename 'tim' to 'Tim'),
3838 * we will treat this as a special case in the
3839 * case-insensitive mode: as long as the source name
3840 * is an exact match, we will allow this to proceed as
3841 * a name-change request.
3843 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3844 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3845 flags & FIGNORECASE)) &&
3846 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3849 * case preserving rename request, require exact
3858 * If the source and destination directories are the same, we should
3859 * grab the z_name_lock of that directory only once.
3863 rw_enter(&sdzp->z_name_lock, RW_READER);
3867 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3868 ZEXISTS | zflg, NULL, NULL);
3869 terr = zfs_dirent_lock(&tdl,
3870 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3872 terr = zfs_dirent_lock(&tdl,
3873 tdzp, tnm, &tzp, zflg, NULL, NULL);
3874 serr = zfs_dirent_lock(&sdl,
3875 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3881 * Source entry invalid or not there.
3884 zfs_dirent_unlock(tdl);
3890 rw_exit(&sdzp->z_name_lock);
3893 * FreeBSD: In OpenSolaris they only check if rename source is
3894 * ".." here, because "." is handled in their lookup. This is
3895 * not the case for FreeBSD, so we check for "." explicitly.
3897 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0)
3898 serr = SET_ERROR(EINVAL);
3903 zfs_dirent_unlock(sdl);
3907 rw_exit(&sdzp->z_name_lock);
3909 if (strcmp(tnm, "..") == 0)
3910 terr = SET_ERROR(EINVAL);
3916 * Must have write access at the source to remove the old entry
3917 * and write access at the target to create the new entry.
3918 * Note that if target and source are the same, this can be
3919 * done in a single check.
3922 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3925 if (ZTOV(szp)->v_type == VDIR) {
3927 * Check to make sure rename is valid.
3928 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3930 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3935 * Does target exist?
3939 * Source and target must be the same type.
3941 if (ZTOV(szp)->v_type == VDIR) {
3942 if (ZTOV(tzp)->v_type != VDIR) {
3943 error = SET_ERROR(ENOTDIR);
3947 if (ZTOV(tzp)->v_type == VDIR) {
3948 error = SET_ERROR(EISDIR);
3953 * POSIX dictates that when the source and target
3954 * entries refer to the same file object, rename
3955 * must do nothing and exit without error.
3957 if (szp->z_id == tzp->z_id) {
3963 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3965 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3968 * notify the target directory if it is not the same
3969 * as source directory.
3972 vnevent_rename_dest_dir(tdvp, ct);
3975 tx = dmu_tx_create(zfsvfs->z_os);
3976 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3977 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3978 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3979 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3981 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3982 zfs_sa_upgrade_txholds(tx, tdzp);
3985 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3986 zfs_sa_upgrade_txholds(tx, tzp);
3989 zfs_sa_upgrade_txholds(tx, szp);
3990 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3991 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3994 zfs_rename_unlock(&zl);
3995 zfs_dirent_unlock(sdl);
3996 zfs_dirent_unlock(tdl);
3999 rw_exit(&sdzp->z_name_lock);
4004 if (error == ERESTART) {
4015 if (tzp) /* Attempt to remove the existing target */
4016 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
4019 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
4021 szp->z_pflags |= ZFS_AV_MODIFIED;
4023 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
4024 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
4027 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
4029 zfs_log_rename(zilog, tx, TX_RENAME |
4030 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
4031 sdl->dl_name, tdzp, tdl->dl_name, szp);
4034 * Update path information for the target vnode
4036 vn_renamepath(tdvp, ZTOV(szp), tnm,
4040 * At this point, we have successfully created
4041 * the target name, but have failed to remove
4042 * the source name. Since the create was done
4043 * with the ZRENAMING flag, there are
4044 * complications; for one, the link count is
4045 * wrong. The easiest way to deal with this
4046 * is to remove the newly created target, and
4047 * return the original error. This must
4048 * succeed; fortunately, it is very unlikely to
4049 * fail, since we just created it.
4051 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
4052 ZRENAMING, NULL), ==, 0);
4055 #ifdef FREEBSD_NAMECACHE
4059 cache_purge(ZTOV(szp));
4061 cache_purge(ZTOV(tzp));
4069 zfs_rename_unlock(&zl);
4071 zfs_dirent_unlock(sdl);
4072 zfs_dirent_unlock(tdl);
4075 rw_exit(&sdzp->z_name_lock);
4082 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4083 zil_commit(zilog, 0);
4091 * Insert the indicated symbolic reference entry into the directory.
4093 * IN: dvp - Directory to contain new symbolic link.
4094 * link - Name for new symlink entry.
4095 * vap - Attributes of new entry.
4096 * cr - credentials of caller.
4097 * ct - caller context
4098 * flags - case flags
4100 * RETURN: 0 on success, error code on failure.
4103 * dvp - ctime|mtime updated
4107 zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link,
4108 cred_t *cr, kthread_t *td)
4110 znode_t *zp, *dzp = VTOZ(dvp);
4113 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4115 uint64_t len = strlen(link);
4118 zfs_acl_ids_t acl_ids;
4119 boolean_t fuid_dirtied;
4120 uint64_t txtype = TX_SYMLINK;
4121 boolean_t waited = B_FALSE;
4124 ASSERT(vap->va_type == VLNK);
4128 zilog = zfsvfs->z_log;
4130 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
4131 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4133 return (SET_ERROR(EILSEQ));
4135 if (flags & FIGNORECASE)
4138 if (len > MAXPATHLEN) {
4140 return (SET_ERROR(ENAMETOOLONG));
4143 if ((error = zfs_acl_ids_create(dzp, 0,
4144 vap, cr, NULL, &acl_ids)) != 0) {
4150 * Attempt to lock directory; fail if entry already exists.
4152 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
4154 zfs_acl_ids_free(&acl_ids);
4159 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4160 zfs_acl_ids_free(&acl_ids);
4161 zfs_dirent_unlock(dl);
4166 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
4167 zfs_acl_ids_free(&acl_ids);
4168 zfs_dirent_unlock(dl);
4170 return (SET_ERROR(EDQUOT));
4172 tx = dmu_tx_create(zfsvfs->z_os);
4173 fuid_dirtied = zfsvfs->z_fuid_dirty;
4174 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
4175 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4176 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
4177 ZFS_SA_BASE_ATTR_SIZE + len);
4178 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
4179 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
4180 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
4181 acl_ids.z_aclp->z_acl_bytes);
4184 zfs_fuid_txhold(zfsvfs, tx);
4185 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4187 zfs_dirent_unlock(dl);
4188 if (error == ERESTART) {
4194 zfs_acl_ids_free(&acl_ids);
4201 * Create a new object for the symlink.
4202 * for version 4 ZPL datsets the symlink will be an SA attribute
4204 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
4207 zfs_fuid_sync(zfsvfs, tx);
4209 mutex_enter(&zp->z_lock);
4211 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
4214 zfs_sa_symlink(zp, link, len, tx);
4215 mutex_exit(&zp->z_lock);
4218 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
4219 &zp->z_size, sizeof (zp->z_size), tx);
4221 * Insert the new object into the directory.
4223 (void) zfs_link_create(dl, zp, tx, ZNEW);
4225 if (flags & FIGNORECASE)
4227 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
4230 zfs_acl_ids_free(&acl_ids);
4234 zfs_dirent_unlock(dl);
4236 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4237 zil_commit(zilog, 0);
4244 * Return, in the buffer contained in the provided uio structure,
4245 * the symbolic path referred to by vp.
4247 * IN: vp - vnode of symbolic link.
4248 * uio - structure to contain the link path.
4249 * cr - credentials of caller.
4250 * ct - caller context
4252 * OUT: uio - structure containing the link path.
4254 * RETURN: 0 on success, error code on failure.
4257 * vp - atime updated
4261 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
4263 znode_t *zp = VTOZ(vp);
4264 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4270 mutex_enter(&zp->z_lock);
4272 error = sa_lookup_uio(zp->z_sa_hdl,
4273 SA_ZPL_SYMLINK(zfsvfs), uio);
4275 error = zfs_sa_readlink(zp, uio);
4276 mutex_exit(&zp->z_lock);
4278 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4285 * Insert a new entry into directory tdvp referencing svp.
4287 * IN: tdvp - Directory to contain new entry.
4288 * svp - vnode of new entry.
4289 * name - name of new entry.
4290 * cr - credentials of caller.
4291 * ct - caller context
4293 * RETURN: 0 on success, error code on failure.
4296 * tdvp - ctime|mtime updated
4297 * svp - ctime updated
4301 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
4302 caller_context_t *ct, int flags)
4304 znode_t *dzp = VTOZ(tdvp);
4306 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4315 boolean_t waited = B_FALSE;
4317 ASSERT(tdvp->v_type == VDIR);
4321 zilog = zfsvfs->z_log;
4323 if (VOP_REALVP(svp, &realvp, ct) == 0)
4327 * POSIX dictates that we return EPERM here.
4328 * Better choices include ENOTSUP or EISDIR.
4330 if (svp->v_type == VDIR) {
4332 return (SET_ERROR(EPERM));
4339 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
4340 * ctldir appear to have the same v_vfsp.
4342 if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) {
4344 return (SET_ERROR(EXDEV));
4347 /* Prevent links to .zfs/shares files */
4349 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4350 &parent, sizeof (uint64_t))) != 0) {
4354 if (parent == zfsvfs->z_shares_dir) {
4356 return (SET_ERROR(EPERM));
4359 if (zfsvfs->z_utf8 && u8_validate(name,
4360 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4362 return (SET_ERROR(EILSEQ));
4364 if (flags & FIGNORECASE)
4368 * We do not support links between attributes and non-attributes
4369 * because of the potential security risk of creating links
4370 * into "normal" file space in order to circumvent restrictions
4371 * imposed in attribute space.
4373 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4375 return (SET_ERROR(EINVAL));
4379 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4380 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) {
4382 return (SET_ERROR(EPERM));
4385 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4392 * Attempt to lock directory; fail if entry already exists.
4394 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4400 tx = dmu_tx_create(zfsvfs->z_os);
4401 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4402 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4403 zfs_sa_upgrade_txholds(tx, szp);
4404 zfs_sa_upgrade_txholds(tx, dzp);
4405 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4407 zfs_dirent_unlock(dl);
4408 if (error == ERESTART) {
4419 error = zfs_link_create(dl, szp, tx, 0);
4422 uint64_t txtype = TX_LINK;
4423 if (flags & FIGNORECASE)
4425 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4430 zfs_dirent_unlock(dl);
4433 vnevent_link(svp, ct);
4436 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4437 zil_commit(zilog, 0);
4445 * zfs_null_putapage() is used when the file system has been force
4446 * unmounted. It just drops the pages.
4450 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4451 size_t *lenp, int flags, cred_t *cr)
4453 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4458 * Push a page out to disk, klustering if possible.
4460 * IN: vp - file to push page to.
4461 * pp - page to push.
4462 * flags - additional flags.
4463 * cr - credentials of caller.
4465 * OUT: offp - start of range pushed.
4466 * lenp - len of range pushed.
4468 * RETURN: 0 on success, error code on failure.
4470 * NOTE: callers must have locked the page to be pushed. On
4471 * exit, the page (and all other pages in the kluster) must be
4476 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4477 size_t *lenp, int flags, cred_t *cr)
4479 znode_t *zp = VTOZ(vp);
4480 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4482 u_offset_t off, koff;
4489 * If our blocksize is bigger than the page size, try to kluster
4490 * multiple pages so that we write a full block (thus avoiding
4491 * a read-modify-write).
4493 if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4494 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4495 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4496 ASSERT(koff <= zp->z_size);
4497 if (koff + klen > zp->z_size)
4498 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4499 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4501 ASSERT3U(btop(len), ==, btopr(len));
4504 * Can't push pages past end-of-file.
4506 if (off >= zp->z_size) {
4507 /* ignore all pages */
4510 } else if (off + len > zp->z_size) {
4511 int npages = btopr(zp->z_size - off);
4514 page_list_break(&pp, &trunc, npages);
4515 /* ignore pages past end of file */
4517 pvn_write_done(trunc, flags);
4518 len = zp->z_size - off;
4521 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4522 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4523 err = SET_ERROR(EDQUOT);
4526 tx = dmu_tx_create(zfsvfs->z_os);
4527 dmu_tx_hold_write(tx, zp->z_id, off, len);
4529 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4530 zfs_sa_upgrade_txholds(tx, zp);
4531 err = dmu_tx_assign(tx, TXG_WAIT);
4537 if (zp->z_blksz <= PAGESIZE) {
4538 caddr_t va = zfs_map_page(pp, S_READ);
4539 ASSERT3U(len, <=, PAGESIZE);
4540 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4541 zfs_unmap_page(pp, va);
4543 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4547 uint64_t mtime[2], ctime[2];
4548 sa_bulk_attr_t bulk[3];
4551 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4553 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4555 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4557 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4559 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4564 pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4574 * Copy the portion of the file indicated from pages into the file.
4575 * The pages are stored in a page list attached to the files vnode.
4577 * IN: vp - vnode of file to push page data to.
4578 * off - position in file to put data.
4579 * len - amount of data to write.
4580 * flags - flags to control the operation.
4581 * cr - credentials of caller.
4582 * ct - caller context.
4584 * RETURN: 0 on success, error code on failure.
4587 * vp - ctime|mtime updated
4591 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4592 caller_context_t *ct)
4594 znode_t *zp = VTOZ(vp);
4595 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4607 * Align this request to the file block size in case we kluster.
4608 * XXX - this can result in pretty aggresive locking, which can
4609 * impact simultanious read/write access. One option might be
4610 * to break up long requests (len == 0) into block-by-block
4611 * operations to get narrower locking.
4613 blksz = zp->z_blksz;
4615 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4618 if (len > 0 && ISP2(blksz))
4619 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4625 * Search the entire vp list for pages >= io_off.
4627 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4628 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4631 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4633 if (off > zp->z_size) {
4634 /* past end of file */
4635 zfs_range_unlock(rl);
4640 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4642 for (off = io_off; io_off < off + len; io_off += io_len) {
4643 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4644 pp = page_lookup(vp, io_off,
4645 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4647 pp = page_lookup_nowait(vp, io_off,
4648 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4651 if (pp != NULL && pvn_getdirty(pp, flags)) {
4655 * Found a dirty page to push
4657 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4665 zfs_range_unlock(rl);
4666 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4667 zil_commit(zfsvfs->z_log, zp->z_id);
4675 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4677 znode_t *zp = VTOZ(vp);
4678 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4681 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4682 if (zp->z_sa_hdl == NULL) {
4684 * The fs has been unmounted, or we did a
4685 * suspend/resume and this file no longer exists.
4687 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4692 mutex_enter(&zp->z_lock);
4693 if (zp->z_unlinked) {
4695 * Fast path to recycle a vnode of a removed file.
4697 mutex_exit(&zp->z_lock);
4698 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4702 mutex_exit(&zp->z_lock);
4704 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4705 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4707 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4708 zfs_sa_upgrade_txholds(tx, zp);
4709 error = dmu_tx_assign(tx, TXG_WAIT);
4713 mutex_enter(&zp->z_lock);
4714 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4715 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4716 zp->z_atime_dirty = 0;
4717 mutex_exit(&zp->z_lock);
4721 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4726 * Bounds-check the seek operation.
4728 * IN: vp - vnode seeking within
4729 * ooff - old file offset
4730 * noffp - pointer to new file offset
4731 * ct - caller context
4733 * RETURN: 0 on success, EINVAL if new offset invalid.
4737 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4738 caller_context_t *ct)
4740 if (vp->v_type == VDIR)
4742 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4746 * Pre-filter the generic locking function to trap attempts to place
4747 * a mandatory lock on a memory mapped file.
4750 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4751 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4753 znode_t *zp = VTOZ(vp);
4754 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4760 * We are following the UFS semantics with respect to mapcnt
4761 * here: If we see that the file is mapped already, then we will
4762 * return an error, but we don't worry about races between this
4763 * function and zfs_map().
4765 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4767 return (SET_ERROR(EAGAIN));
4770 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4774 * If we can't find a page in the cache, we will create a new page
4775 * and fill it with file data. For efficiency, we may try to fill
4776 * multiple pages at once (klustering) to fill up the supplied page
4777 * list. Note that the pages to be filled are held with an exclusive
4778 * lock to prevent access by other threads while they are being filled.
4781 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4782 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4784 znode_t *zp = VTOZ(vp);
4785 page_t *pp, *cur_pp;
4786 objset_t *os = zp->z_zfsvfs->z_os;
4787 u_offset_t io_off, total;
4791 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4793 * We only have a single page, don't bother klustering
4797 pp = page_create_va(vp, io_off, io_len,
4798 PG_EXCL | PG_WAIT, seg, addr);
4801 * Try to find enough pages to fill the page list
4803 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4804 &io_len, off, plsz, 0);
4808 * The page already exists, nothing to do here.
4815 * Fill the pages in the kluster.
4818 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4821 ASSERT3U(io_off, ==, cur_pp->p_offset);
4822 va = zfs_map_page(cur_pp, S_WRITE);
4823 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4825 zfs_unmap_page(cur_pp, va);
4827 /* On error, toss the entire kluster */
4828 pvn_read_done(pp, B_ERROR);
4829 /* convert checksum errors into IO errors */
4831 err = SET_ERROR(EIO);
4834 cur_pp = cur_pp->p_next;
4838 * Fill in the page list array from the kluster starting
4839 * from the desired offset `off'.
4840 * NOTE: the page list will always be null terminated.
4842 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4843 ASSERT(pl == NULL || (*pl)->p_offset == off);
4849 * Return pointers to the pages for the file region [off, off + len]
4850 * in the pl array. If plsz is greater than len, this function may
4851 * also return page pointers from after the specified region
4852 * (i.e. the region [off, off + plsz]). These additional pages are
4853 * only returned if they are already in the cache, or were created as
4854 * part of a klustered read.
4856 * IN: vp - vnode of file to get data from.
4857 * off - position in file to get data from.
4858 * len - amount of data to retrieve.
4859 * plsz - length of provided page list.
4860 * seg - segment to obtain pages for.
4861 * addr - virtual address of fault.
4862 * rw - mode of created pages.
4863 * cr - credentials of caller.
4864 * ct - caller context.
4866 * OUT: protp - protection mode of created pages.
4867 * pl - list of pages created.
4869 * RETURN: 0 on success, error code on failure.
4872 * vp - atime updated
4876 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4877 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4878 enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4880 znode_t *zp = VTOZ(vp);
4881 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4885 /* we do our own caching, faultahead is unnecessary */
4888 else if (len > plsz)
4891 len = P2ROUNDUP(len, PAGESIZE);
4892 ASSERT(plsz >= len);
4901 * Loop through the requested range [off, off + len) looking
4902 * for pages. If we don't find a page, we will need to create
4903 * a new page and fill it with data from the file.
4906 if (*pl = page_lookup(vp, off, SE_SHARED))
4908 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4911 ASSERT3U((*pl)->p_offset, ==, off);
4915 ASSERT3U(len, >=, PAGESIZE);
4918 ASSERT3U(plsz, >=, PAGESIZE);
4925 * Fill out the page array with any pages already in the cache.
4928 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4935 * Release any pages we have previously locked.
4940 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4950 * Request a memory map for a section of a file. This code interacts
4951 * with common code and the VM system as follows:
4953 * - common code calls mmap(), which ends up in smmap_common()
4954 * - this calls VOP_MAP(), which takes you into (say) zfs
4955 * - zfs_map() calls as_map(), passing segvn_create() as the callback
4956 * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4957 * - zfs_addmap() updates z_mapcnt
4961 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4962 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4963 caller_context_t *ct)
4965 znode_t *zp = VTOZ(vp);
4966 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4967 segvn_crargs_t vn_a;
4973 if ((prot & PROT_WRITE) && (zp->z_pflags &
4974 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4976 return (SET_ERROR(EPERM));
4979 if ((prot & (PROT_READ | PROT_EXEC)) &&
4980 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4982 return (SET_ERROR(EACCES));
4985 if (vp->v_flag & VNOMAP) {
4987 return (SET_ERROR(ENOSYS));
4990 if (off < 0 || len > MAXOFFSET_T - off) {
4992 return (SET_ERROR(ENXIO));
4995 if (vp->v_type != VREG) {
4997 return (SET_ERROR(ENODEV));
5001 * If file is locked, disallow mapping.
5003 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
5005 return (SET_ERROR(EAGAIN));
5009 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
5017 vn_a.offset = (u_offset_t)off;
5018 vn_a.type = flags & MAP_TYPE;
5020 vn_a.maxprot = maxprot;
5023 vn_a.flags = flags & ~MAP_TYPE;
5025 vn_a.lgrp_mem_policy_flags = 0;
5027 error = as_map(as, *addrp, len, segvn_create, &vn_a);
5036 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5037 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
5038 caller_context_t *ct)
5040 uint64_t pages = btopr(len);
5042 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
5047 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
5048 * more accurate mtime for the associated file. Since we don't have a way of
5049 * detecting when the data was actually modified, we have to resort to
5050 * heuristics. If an explicit msync() is done, then we mark the mtime when the
5051 * last page is pushed. The problem occurs when the msync() call is omitted,
5052 * which by far the most common case:
5060 * putpage() via fsflush
5062 * If we wait until fsflush to come along, we can have a modification time that
5063 * is some arbitrary point in the future. In order to prevent this in the
5064 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
5069 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5070 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
5071 caller_context_t *ct)
5073 uint64_t pages = btopr(len);
5075 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
5076 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
5078 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
5079 vn_has_cached_data(vp))
5080 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
5086 * Free or allocate space in a file. Currently, this function only
5087 * supports the `F_FREESP' command. However, this command is somewhat
5088 * misnamed, as its functionality includes the ability to allocate as
5089 * well as free space.
5091 * IN: vp - vnode of file to free data in.
5092 * cmd - action to take (only F_FREESP supported).
5093 * bfp - section of file to free/alloc.
5094 * flag - current file open mode flags.
5095 * offset - current file offset.
5096 * cr - credentials of caller [UNUSED].
5097 * ct - caller context.
5099 * RETURN: 0 on success, error code on failure.
5102 * vp - ctime|mtime updated
5106 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
5107 offset_t offset, cred_t *cr, caller_context_t *ct)
5109 znode_t *zp = VTOZ(vp);
5110 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5117 if (cmd != F_FREESP) {
5119 return (SET_ERROR(EINVAL));
5122 if (error = convoff(vp, bfp, 0, offset)) {
5127 if (bfp->l_len < 0) {
5129 return (SET_ERROR(EINVAL));
5133 len = bfp->l_len; /* 0 means from off to end of file */
5135 error = zfs_freesp(zp, off, len, flag, TRUE);
5142 CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid));
5143 CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid));
5147 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
5149 znode_t *zp = VTOZ(vp);
5150 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5153 uint64_t object = zp->z_id;
5160 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
5161 &gen64, sizeof (uint64_t))) != 0) {
5166 gen = (uint32_t)gen64;
5168 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
5171 if (fidp->fid_len < size) {
5172 fidp->fid_len = size;
5174 return (SET_ERROR(ENOSPC));
5177 fidp->fid_len = size;
5180 zfid = (zfid_short_t *)fidp;
5182 zfid->zf_len = size;
5184 for (i = 0; i < sizeof (zfid->zf_object); i++)
5185 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
5187 /* Must have a non-zero generation number to distinguish from .zfs */
5190 for (i = 0; i < sizeof (zfid->zf_gen); i++)
5191 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
5193 if (size == LONG_FID_LEN) {
5194 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
5197 zlfid = (zfid_long_t *)fidp;
5199 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
5200 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
5202 /* XXX - this should be the generation number for the objset */
5203 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
5204 zlfid->zf_setgen[i] = 0;
5212 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
5213 caller_context_t *ct)
5225 case _PC_FILESIZEBITS:
5229 case _PC_XATTR_EXISTS:
5231 zfsvfs = zp->z_zfsvfs;
5235 error = zfs_dirent_lock(&dl, zp, "", &xzp,
5236 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
5238 zfs_dirent_unlock(dl);
5239 if (!zfs_dirempty(xzp))
5242 } else if (error == ENOENT) {
5244 * If there aren't extended attributes, it's the
5245 * same as having zero of them.
5252 case _PC_SATTR_ENABLED:
5253 case _PC_SATTR_EXISTS:
5254 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
5255 (vp->v_type == VREG || vp->v_type == VDIR);
5258 case _PC_ACCESS_FILTERING:
5259 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
5263 case _PC_ACL_ENABLED:
5264 *valp = _ACL_ACE_ENABLED;
5267 case _PC_MIN_HOLE_SIZE:
5268 *valp = (int)SPA_MINBLOCKSIZE;
5271 case _PC_TIMESTAMP_RESOLUTION:
5272 /* nanosecond timestamp resolution */
5276 case _PC_ACL_EXTENDED:
5284 case _PC_ACL_PATH_MAX:
5285 *valp = ACL_MAX_ENTRIES;
5289 return (EOPNOTSUPP);
5295 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5296 caller_context_t *ct)
5298 znode_t *zp = VTOZ(vp);
5299 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5301 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5305 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5313 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5314 caller_context_t *ct)
5316 znode_t *zp = VTOZ(vp);
5317 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5319 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5320 zilog_t *zilog = zfsvfs->z_log;
5325 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5327 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5328 zil_commit(zilog, 0);
5336 * The smallest read we may consider to loan out an arcbuf.
5337 * This must be a power of 2.
5339 int zcr_blksz_min = (1 << 10); /* 1K */
5341 * If set to less than the file block size, allow loaning out of an
5342 * arcbuf for a partial block read. This must be a power of 2.
5344 int zcr_blksz_max = (1 << 17); /* 128K */
5348 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5349 caller_context_t *ct)
5351 znode_t *zp = VTOZ(vp);
5352 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5353 int max_blksz = zfsvfs->z_max_blksz;
5354 uio_t *uio = &xuio->xu_uio;
5355 ssize_t size = uio->uio_resid;
5356 offset_t offset = uio->uio_loffset;
5361 int preamble, postamble;
5363 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5364 return (SET_ERROR(EINVAL));
5371 * Loan out an arc_buf for write if write size is bigger than
5372 * max_blksz, and the file's block size is also max_blksz.
5375 if (size < blksz || zp->z_blksz != blksz) {
5377 return (SET_ERROR(EINVAL));
5380 * Caller requests buffers for write before knowing where the
5381 * write offset might be (e.g. NFS TCP write).
5386 preamble = P2PHASE(offset, blksz);
5388 preamble = blksz - preamble;
5393 postamble = P2PHASE(size, blksz);
5396 fullblk = size / blksz;
5397 (void) dmu_xuio_init(xuio,
5398 (preamble != 0) + fullblk + (postamble != 0));
5399 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5400 int, postamble, int,
5401 (preamble != 0) + fullblk + (postamble != 0));
5404 * Have to fix iov base/len for partial buffers. They
5405 * currently represent full arc_buf's.
5408 /* data begins in the middle of the arc_buf */
5409 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5412 (void) dmu_xuio_add(xuio, abuf,
5413 blksz - preamble, preamble);
5416 for (i = 0; i < fullblk; i++) {
5417 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5420 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5424 /* data ends in the middle of the arc_buf */
5425 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5428 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5433 * Loan out an arc_buf for read if the read size is larger than
5434 * the current file block size. Block alignment is not
5435 * considered. Partial arc_buf will be loaned out for read.
5437 blksz = zp->z_blksz;
5438 if (blksz < zcr_blksz_min)
5439 blksz = zcr_blksz_min;
5440 if (blksz > zcr_blksz_max)
5441 blksz = zcr_blksz_max;
5442 /* avoid potential complexity of dealing with it */
5443 if (blksz > max_blksz) {
5445 return (SET_ERROR(EINVAL));
5448 maxsize = zp->z_size - uio->uio_loffset;
5452 if (size < blksz || vn_has_cached_data(vp)) {
5454 return (SET_ERROR(EINVAL));
5459 return (SET_ERROR(EINVAL));
5462 uio->uio_extflg = UIO_XUIO;
5463 XUIO_XUZC_RW(xuio) = ioflag;
5470 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5474 int ioflag = XUIO_XUZC_RW(xuio);
5476 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5478 i = dmu_xuio_cnt(xuio);
5480 abuf = dmu_xuio_arcbuf(xuio, i);
5482 * if abuf == NULL, it must be a write buffer
5483 * that has been returned in zfs_write().
5486 dmu_return_arcbuf(abuf);
5487 ASSERT(abuf || ioflag == UIO_WRITE);
5490 dmu_xuio_fini(xuio);
5495 * Predeclare these here so that the compiler assumes that
5496 * this is an "old style" function declaration that does
5497 * not include arguments => we won't get type mismatch errors
5498 * in the initializations that follow.
5500 static int zfs_inval();
5501 static int zfs_isdir();
5506 return (SET_ERROR(EINVAL));
5512 return (SET_ERROR(EISDIR));
5515 * Directory vnode operations template
5517 vnodeops_t *zfs_dvnodeops;
5518 const fs_operation_def_t zfs_dvnodeops_template[] = {
5519 VOPNAME_OPEN, { .vop_open = zfs_open },
5520 VOPNAME_CLOSE, { .vop_close = zfs_close },
5521 VOPNAME_READ, { .error = zfs_isdir },
5522 VOPNAME_WRITE, { .error = zfs_isdir },
5523 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5524 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5525 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5526 VOPNAME_ACCESS, { .vop_access = zfs_access },
5527 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5528 VOPNAME_CREATE, { .vop_create = zfs_create },
5529 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5530 VOPNAME_LINK, { .vop_link = zfs_link },
5531 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5532 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir },
5533 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5534 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5535 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink },
5536 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5537 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5538 VOPNAME_FID, { .vop_fid = zfs_fid },
5539 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5540 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5541 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5542 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5543 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5548 * Regular file vnode operations template
5550 vnodeops_t *zfs_fvnodeops;
5551 const fs_operation_def_t zfs_fvnodeops_template[] = {
5552 VOPNAME_OPEN, { .vop_open = zfs_open },
5553 VOPNAME_CLOSE, { .vop_close = zfs_close },
5554 VOPNAME_READ, { .vop_read = zfs_read },
5555 VOPNAME_WRITE, { .vop_write = zfs_write },
5556 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5557 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5558 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5559 VOPNAME_ACCESS, { .vop_access = zfs_access },
5560 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5561 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5562 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5563 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5564 VOPNAME_FID, { .vop_fid = zfs_fid },
5565 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5566 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock },
5567 VOPNAME_SPACE, { .vop_space = zfs_space },
5568 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage },
5569 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage },
5570 VOPNAME_MAP, { .vop_map = zfs_map },
5571 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap },
5572 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap },
5573 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5574 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5575 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5576 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5577 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf },
5578 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf },
5583 * Symbolic link vnode operations template
5585 vnodeops_t *zfs_symvnodeops;
5586 const fs_operation_def_t zfs_symvnodeops_template[] = {
5587 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5588 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5589 VOPNAME_ACCESS, { .vop_access = zfs_access },
5590 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5591 VOPNAME_READLINK, { .vop_readlink = zfs_readlink },
5592 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5593 VOPNAME_FID, { .vop_fid = zfs_fid },
5594 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5595 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5600 * special share hidden files vnode operations template
5602 vnodeops_t *zfs_sharevnodeops;
5603 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5604 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5605 VOPNAME_ACCESS, { .vop_access = zfs_access },
5606 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5607 VOPNAME_FID, { .vop_fid = zfs_fid },
5608 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5609 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5610 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5611 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5616 * Extended attribute directory vnode operations template
5618 * This template is identical to the directory vnodes
5619 * operation template except for restricted operations:
5623 * Note that there are other restrictions embedded in:
5624 * zfs_create() - restrict type to VREG
5625 * zfs_link() - no links into/out of attribute space
5626 * zfs_rename() - no moves into/out of attribute space
5628 vnodeops_t *zfs_xdvnodeops;
5629 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5630 VOPNAME_OPEN, { .vop_open = zfs_open },
5631 VOPNAME_CLOSE, { .vop_close = zfs_close },
5632 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5633 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5634 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5635 VOPNAME_ACCESS, { .vop_access = zfs_access },
5636 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5637 VOPNAME_CREATE, { .vop_create = zfs_create },
5638 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5639 VOPNAME_LINK, { .vop_link = zfs_link },
5640 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5641 VOPNAME_MKDIR, { .error = zfs_inval },
5642 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5643 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5644 VOPNAME_SYMLINK, { .error = zfs_inval },
5645 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5646 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5647 VOPNAME_FID, { .vop_fid = zfs_fid },
5648 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5649 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5650 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5651 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5652 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5657 * Error vnode operations template
5659 vnodeops_t *zfs_evnodeops;
5660 const fs_operation_def_t zfs_evnodeops_template[] = {
5661 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5662 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5668 ioflags(int ioflags)
5672 if (ioflags & IO_APPEND)
5674 if (ioflags & IO_NDELAY)
5676 if (ioflags & IO_SYNC)
5677 flags |= (FSYNC | FDSYNC | FRSYNC);
5683 zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage)
5685 znode_t *zp = VTOZ(vp);
5686 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5687 objset_t *os = zp->z_zfsvfs->z_os;
5688 vm_page_t mfirst, mlast, mreq;
5692 off_t startoff, endoff;
5694 vm_pindex_t reqstart, reqend;
5695 int pcount, lsize, reqsize, size;
5700 pcount = OFF_TO_IDX(round_page(count));
5702 object = mreq->object;
5705 KASSERT(vp->v_object == object, ("mismatching object"));
5707 if (pcount > 1 && zp->z_blksz > PAGESIZE) {
5708 startoff = rounddown(IDX_TO_OFF(mreq->pindex), zp->z_blksz);
5709 reqstart = OFF_TO_IDX(round_page(startoff));
5710 if (reqstart < m[0]->pindex)
5713 reqstart = reqstart - m[0]->pindex;
5714 endoff = roundup(IDX_TO_OFF(mreq->pindex) + PAGE_SIZE,
5716 reqend = OFF_TO_IDX(trunc_page(endoff)) - 1;
5717 if (reqend > m[pcount - 1]->pindex)
5718 reqend = m[pcount - 1]->pindex;
5719 reqsize = reqend - m[reqstart]->pindex + 1;
5720 KASSERT(reqstart <= reqpage && reqpage < reqstart + reqsize,
5721 ("reqpage beyond [reqstart, reqstart + reqsize[ bounds"));
5726 mfirst = m[reqstart];
5727 mlast = m[reqstart + reqsize - 1];
5729 zfs_vmobject_wlock(object);
5731 for (i = 0; i < reqstart; i++) {
5734 vm_page_unlock(m[i]);
5736 for (i = reqstart + reqsize; i < pcount; i++) {
5739 vm_page_unlock(m[i]);
5742 if (mreq->valid && reqsize == 1) {
5743 if (mreq->valid != VM_PAGE_BITS_ALL)
5744 vm_page_zero_invalid(mreq, TRUE);
5745 zfs_vmobject_wunlock(object);
5747 return (zfs_vm_pagerret_ok);
5750 PCPU_INC(cnt.v_vnodein);
5751 PCPU_ADD(cnt.v_vnodepgsin, reqsize);
5753 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) {
5754 for (i = reqstart; i < reqstart + reqsize; i++) {
5758 vm_page_unlock(m[i]);
5761 zfs_vmobject_wunlock(object);
5763 return (zfs_vm_pagerret_bad);
5767 if (IDX_TO_OFF(mlast->pindex) + lsize > object->un_pager.vnp.vnp_size)
5768 lsize = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mlast->pindex);
5770 zfs_vmobject_wunlock(object);
5772 for (i = reqstart; i < reqstart + reqsize; i++) {
5774 if (i == (reqstart + reqsize - 1))
5776 va = zfs_map_page(m[i], &sf);
5777 error = dmu_read(os, zp->z_id, IDX_TO_OFF(m[i]->pindex),
5778 size, va, DMU_READ_PREFETCH);
5779 if (size != PAGE_SIZE)
5780 bzero(va + size, PAGE_SIZE - size);
5786 zfs_vmobject_wlock(object);
5788 for (i = reqstart; i < reqstart + reqsize; i++) {
5790 m[i]->valid = VM_PAGE_BITS_ALL;
5791 KASSERT(m[i]->dirty == 0, ("zfs_getpages: page %p is dirty", m[i]));
5793 vm_page_readahead_finish(m[i]);
5796 zfs_vmobject_wunlock(object);
5798 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
5800 return (error ? zfs_vm_pagerret_error : zfs_vm_pagerret_ok);
5804 zfs_freebsd_getpages(ap)
5805 struct vop_getpages_args /* {
5810 vm_ooffset_t a_offset;
5814 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage));
5818 zfs_freebsd_bmap(ap)
5819 struct vop_bmap_args /* {
5822 struct bufobj **a_bop;
5829 if (ap->a_bop != NULL)
5830 *ap->a_bop = &ap->a_vp->v_bufobj;
5831 if (ap->a_bnp != NULL)
5832 *ap->a_bnp = ap->a_bn;
5833 if (ap->a_runp != NULL)
5835 if (ap->a_runb != NULL)
5842 zfs_freebsd_open(ap)
5843 struct vop_open_args /* {
5846 struct ucred *a_cred;
5847 struct thread *a_td;
5850 vnode_t *vp = ap->a_vp;
5851 znode_t *zp = VTOZ(vp);
5854 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL);
5856 vnode_create_vobject(vp, zp->z_size, ap->a_td);
5861 zfs_freebsd_close(ap)
5862 struct vop_close_args /* {
5865 struct ucred *a_cred;
5866 struct thread *a_td;
5870 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL));
5874 zfs_freebsd_ioctl(ap)
5875 struct vop_ioctl_args /* {
5885 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data,
5886 ap->a_fflag, ap->a_cred, NULL, NULL));
5890 zfs_freebsd_read(ap)
5891 struct vop_read_args /* {
5895 struct ucred *a_cred;
5899 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5904 zfs_freebsd_write(ap)
5905 struct vop_write_args /* {
5909 struct ucred *a_cred;
5913 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5918 zfs_freebsd_access(ap)
5919 struct vop_access_args /* {
5921 accmode_t a_accmode;
5922 struct ucred *a_cred;
5923 struct thread *a_td;
5926 vnode_t *vp = ap->a_vp;
5927 znode_t *zp = VTOZ(vp);
5932 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND,
5934 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND);
5936 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL);
5939 * VADMIN has to be handled by vaccess().
5942 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND);
5944 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
5945 zp->z_gid, accmode, ap->a_cred, NULL);
5950 * For VEXEC, ensure that at least one execute bit is set for
5953 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR &&
5954 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
5962 zfs_freebsd_lookup(ap)
5963 struct vop_lookup_args /* {
5964 struct vnode *a_dvp;
5965 struct vnode **a_vpp;
5966 struct componentname *a_cnp;
5969 struct componentname *cnp = ap->a_cnp;
5970 char nm[NAME_MAX + 1];
5972 ASSERT(cnp->cn_namelen < sizeof(nm));
5973 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm)));
5975 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop,
5976 cnp->cn_cred, cnp->cn_thread, 0));
5980 zfs_freebsd_create(ap)
5981 struct vop_create_args /* {
5982 struct vnode *a_dvp;
5983 struct vnode **a_vpp;
5984 struct componentname *a_cnp;
5985 struct vattr *a_vap;
5988 struct componentname *cnp = ap->a_cnp;
5989 vattr_t *vap = ap->a_vap;
5992 ASSERT(cnp->cn_flags & SAVENAME);
5994 vattr_init_mask(vap);
5995 mode = vap->va_mode & ALLPERMS;
5997 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode,
5998 ap->a_vpp, cnp->cn_cred, cnp->cn_thread));
6002 zfs_freebsd_remove(ap)
6003 struct vop_remove_args /* {
6004 struct vnode *a_dvp;
6006 struct componentname *a_cnp;
6010 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6012 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr,
6013 ap->a_cnp->cn_cred, NULL, 0));
6017 zfs_freebsd_mkdir(ap)
6018 struct vop_mkdir_args /* {
6019 struct vnode *a_dvp;
6020 struct vnode **a_vpp;
6021 struct componentname *a_cnp;
6022 struct vattr *a_vap;
6025 vattr_t *vap = ap->a_vap;
6027 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6029 vattr_init_mask(vap);
6031 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp,
6032 ap->a_cnp->cn_cred, NULL, 0, NULL));
6036 zfs_freebsd_rmdir(ap)
6037 struct vop_rmdir_args /* {
6038 struct vnode *a_dvp;
6040 struct componentname *a_cnp;
6043 struct componentname *cnp = ap->a_cnp;
6045 ASSERT(cnp->cn_flags & SAVENAME);
6047 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0));
6051 zfs_freebsd_readdir(ap)
6052 struct vop_readdir_args /* {
6055 struct ucred *a_cred;
6062 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag,
6063 ap->a_ncookies, ap->a_cookies));
6067 zfs_freebsd_fsync(ap)
6068 struct vop_fsync_args /* {
6071 struct thread *a_td;
6076 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL));
6080 zfs_freebsd_getattr(ap)
6081 struct vop_getattr_args /* {
6083 struct vattr *a_vap;
6084 struct ucred *a_cred;
6087 vattr_t *vap = ap->a_vap;
6093 xvap.xva_vattr = *vap;
6094 xvap.xva_vattr.va_mask |= AT_XVATTR;
6096 /* Convert chflags into ZFS-type flags. */
6097 /* XXX: what about SF_SETTABLE?. */
6098 XVA_SET_REQ(&xvap, XAT_IMMUTABLE);
6099 XVA_SET_REQ(&xvap, XAT_APPENDONLY);
6100 XVA_SET_REQ(&xvap, XAT_NOUNLINK);
6101 XVA_SET_REQ(&xvap, XAT_NODUMP);
6102 XVA_SET_REQ(&xvap, XAT_READONLY);
6103 XVA_SET_REQ(&xvap, XAT_ARCHIVE);
6104 XVA_SET_REQ(&xvap, XAT_SYSTEM);
6105 XVA_SET_REQ(&xvap, XAT_HIDDEN);
6106 XVA_SET_REQ(&xvap, XAT_REPARSE);
6107 XVA_SET_REQ(&xvap, XAT_OFFLINE);
6108 XVA_SET_REQ(&xvap, XAT_SPARSE);
6110 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL);
6114 /* Convert ZFS xattr into chflags. */
6115 #define FLAG_CHECK(fflag, xflag, xfield) do { \
6116 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \
6117 fflags |= (fflag); \
6119 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE,
6120 xvap.xva_xoptattrs.xoa_immutable);
6121 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY,
6122 xvap.xva_xoptattrs.xoa_appendonly);
6123 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK,
6124 xvap.xva_xoptattrs.xoa_nounlink);
6125 FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE,
6126 xvap.xva_xoptattrs.xoa_archive);
6127 FLAG_CHECK(UF_NODUMP, XAT_NODUMP,
6128 xvap.xva_xoptattrs.xoa_nodump);
6129 FLAG_CHECK(UF_READONLY, XAT_READONLY,
6130 xvap.xva_xoptattrs.xoa_readonly);
6131 FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM,
6132 xvap.xva_xoptattrs.xoa_system);
6133 FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN,
6134 xvap.xva_xoptattrs.xoa_hidden);
6135 FLAG_CHECK(UF_REPARSE, XAT_REPARSE,
6136 xvap.xva_xoptattrs.xoa_reparse);
6137 FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE,
6138 xvap.xva_xoptattrs.xoa_offline);
6139 FLAG_CHECK(UF_SPARSE, XAT_SPARSE,
6140 xvap.xva_xoptattrs.xoa_sparse);
6143 *vap = xvap.xva_vattr;
6144 vap->va_flags = fflags;
6149 zfs_freebsd_setattr(ap)
6150 struct vop_setattr_args /* {
6152 struct vattr *a_vap;
6153 struct ucred *a_cred;
6156 vnode_t *vp = ap->a_vp;
6157 vattr_t *vap = ap->a_vap;
6158 cred_t *cred = ap->a_cred;
6163 vattr_init_mask(vap);
6164 vap->va_mask &= ~AT_NOSET;
6167 xvap.xva_vattr = *vap;
6169 zflags = VTOZ(vp)->z_pflags;
6171 if (vap->va_flags != VNOVAL) {
6172 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs;
6175 if (zfsvfs->z_use_fuids == B_FALSE)
6176 return (EOPNOTSUPP);
6178 fflags = vap->va_flags;
6181 * We need to figure out whether it makes sense to allow
6182 * UF_REPARSE through, since we don't really have other
6183 * facilities to handle reparse points and zfs_setattr()
6184 * doesn't currently allow setting that attribute anyway.
6186 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE|
6187 UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE|
6188 UF_OFFLINE|UF_SPARSE)) != 0)
6189 return (EOPNOTSUPP);
6191 * Unprivileged processes are not permitted to unset system
6192 * flags, or modify flags if any system flags are set.
6193 * Privileged non-jail processes may not modify system flags
6194 * if securelevel > 0 and any existing system flags are set.
6195 * Privileged jail processes behave like privileged non-jail
6196 * processes if the security.jail.chflags_allowed sysctl is
6197 * is non-zero; otherwise, they behave like unprivileged
6200 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 ||
6201 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) {
6203 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6204 error = securelevel_gt(cred, 0);
6210 * Callers may only modify the file flags on objects they
6211 * have VADMIN rights for.
6213 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0)
6216 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6220 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) {
6225 #define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \
6226 if (((fflags & (fflag)) && !(zflags & (zflag))) || \
6227 ((zflags & (zflag)) && !(fflags & (fflag)))) { \
6228 XVA_SET_REQ(&xvap, (xflag)); \
6229 (xfield) = ((fflags & (fflag)) != 0); \
6232 /* Convert chflags into ZFS-type flags. */
6233 /* XXX: what about SF_SETTABLE?. */
6234 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE,
6235 xvap.xva_xoptattrs.xoa_immutable);
6236 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY,
6237 xvap.xva_xoptattrs.xoa_appendonly);
6238 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK,
6239 xvap.xva_xoptattrs.xoa_nounlink);
6240 FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE,
6241 xvap.xva_xoptattrs.xoa_archive);
6242 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP,
6243 xvap.xva_xoptattrs.xoa_nodump);
6244 FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY,
6245 xvap.xva_xoptattrs.xoa_readonly);
6246 FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM,
6247 xvap.xva_xoptattrs.xoa_system);
6248 FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN,
6249 xvap.xva_xoptattrs.xoa_hidden);
6250 FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE,
6251 xvap.xva_xoptattrs.xoa_hidden);
6252 FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE,
6253 xvap.xva_xoptattrs.xoa_offline);
6254 FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE,
6255 xvap.xva_xoptattrs.xoa_sparse);
6258 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL));
6262 zfs_freebsd_rename(ap)
6263 struct vop_rename_args /* {
6264 struct vnode *a_fdvp;
6265 struct vnode *a_fvp;
6266 struct componentname *a_fcnp;
6267 struct vnode *a_tdvp;
6268 struct vnode *a_tvp;
6269 struct componentname *a_tcnp;
6272 vnode_t *fdvp = ap->a_fdvp;
6273 vnode_t *fvp = ap->a_fvp;
6274 vnode_t *tdvp = ap->a_tdvp;
6275 vnode_t *tvp = ap->a_tvp;
6278 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART));
6279 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART));
6282 * Check for cross-device rename.
6284 if ((fdvp->v_mount != tdvp->v_mount) ||
6285 (tvp && (fdvp->v_mount != tvp->v_mount)))
6288 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp,
6289 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0);
6303 zfs_freebsd_symlink(ap)
6304 struct vop_symlink_args /* {
6305 struct vnode *a_dvp;
6306 struct vnode **a_vpp;
6307 struct componentname *a_cnp;
6308 struct vattr *a_vap;
6312 struct componentname *cnp = ap->a_cnp;
6313 vattr_t *vap = ap->a_vap;
6315 ASSERT(cnp->cn_flags & SAVENAME);
6317 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */
6318 vattr_init_mask(vap);
6320 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap,
6321 ap->a_target, cnp->cn_cred, cnp->cn_thread));
6325 zfs_freebsd_readlink(ap)
6326 struct vop_readlink_args /* {
6329 struct ucred *a_cred;
6333 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL));
6337 zfs_freebsd_link(ap)
6338 struct vop_link_args /* {
6339 struct vnode *a_tdvp;
6341 struct componentname *a_cnp;
6344 struct componentname *cnp = ap->a_cnp;
6345 vnode_t *vp = ap->a_vp;
6346 vnode_t *tdvp = ap->a_tdvp;
6348 if (tdvp->v_mount != vp->v_mount)
6351 ASSERT(cnp->cn_flags & SAVENAME);
6353 return (zfs_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0));
6357 zfs_freebsd_inactive(ap)
6358 struct vop_inactive_args /* {
6360 struct thread *a_td;
6363 vnode_t *vp = ap->a_vp;
6365 zfs_inactive(vp, ap->a_td->td_ucred, NULL);
6370 zfs_freebsd_reclaim(ap)
6371 struct vop_reclaim_args /* {
6373 struct thread *a_td;
6376 vnode_t *vp = ap->a_vp;
6377 znode_t *zp = VTOZ(vp);
6378 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
6382 /* Destroy the vm object and flush associated pages. */
6383 vnode_destroy_vobject(vp);
6386 * z_teardown_inactive_lock protects from a race with
6387 * zfs_znode_dmu_fini in zfsvfs_teardown during
6390 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
6391 if (zp->z_sa_hdl == NULL)
6395 rw_exit(&zfsvfs->z_teardown_inactive_lock);
6403 struct vop_fid_args /* {
6409 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL));
6413 zfs_freebsd_pathconf(ap)
6414 struct vop_pathconf_args /* {
6417 register_t *a_retval;
6423 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL);
6425 *ap->a_retval = val;
6426 else if (error == EOPNOTSUPP)
6427 error = vop_stdpathconf(ap);
6432 zfs_freebsd_fifo_pathconf(ap)
6433 struct vop_pathconf_args /* {
6436 register_t *a_retval;
6440 switch (ap->a_name) {
6441 case _PC_ACL_EXTENDED:
6443 case _PC_ACL_PATH_MAX:
6444 case _PC_MAC_PRESENT:
6445 return (zfs_freebsd_pathconf(ap));
6447 return (fifo_specops.vop_pathconf(ap));
6452 * FreeBSD's extended attributes namespace defines file name prefix for ZFS'
6453 * extended attribute name:
6456 * system freebsd:system:
6457 * user (none, can be used to access ZFS fsattr(5) attributes
6458 * created on Solaris)
6461 zfs_create_attrname(int attrnamespace, const char *name, char *attrname,
6464 const char *namespace, *prefix, *suffix;
6466 /* We don't allow '/' character in attribute name. */
6467 if (strchr(name, '/') != NULL)
6469 /* We don't allow attribute names that start with "freebsd:" string. */
6470 if (strncmp(name, "freebsd:", 8) == 0)
6473 bzero(attrname, size);
6475 switch (attrnamespace) {
6476 case EXTATTR_NAMESPACE_USER:
6478 prefix = "freebsd:";
6479 namespace = EXTATTR_NAMESPACE_USER_STRING;
6483 * This is the default namespace by which we can access all
6484 * attributes created on Solaris.
6486 prefix = namespace = suffix = "";
6489 case EXTATTR_NAMESPACE_SYSTEM:
6490 prefix = "freebsd:";
6491 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING;
6494 case EXTATTR_NAMESPACE_EMPTY:
6498 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix,
6500 return (ENAMETOOLONG);
6506 * Vnode operating to retrieve a named extended attribute.
6509 zfs_getextattr(struct vop_getextattr_args *ap)
6512 IN struct vnode *a_vp;
6513 IN int a_attrnamespace;
6514 IN const char *a_name;
6515 INOUT struct uio *a_uio;
6517 IN struct ucred *a_cred;
6518 IN struct thread *a_td;
6522 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6523 struct thread *td = ap->a_td;
6524 struct nameidata nd;
6527 vnode_t *xvp = NULL, *vp;
6530 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6531 ap->a_cred, ap->a_td, VREAD);
6535 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6542 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6550 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6552 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL);
6554 NDFREE(&nd, NDF_ONLY_PNBUF);
6557 if (error == ENOENT)
6562 if (ap->a_size != NULL) {
6563 error = VOP_GETATTR(vp, &va, ap->a_cred);
6565 *ap->a_size = (size_t)va.va_size;
6566 } else if (ap->a_uio != NULL)
6567 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6570 vn_close(vp, flags, ap->a_cred, td);
6577 * Vnode operation to remove a named attribute.
6580 zfs_deleteextattr(struct vop_deleteextattr_args *ap)
6583 IN struct vnode *a_vp;
6584 IN int a_attrnamespace;
6585 IN const char *a_name;
6586 IN struct ucred *a_cred;
6587 IN struct thread *a_td;
6591 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6592 struct thread *td = ap->a_td;
6593 struct nameidata nd;
6596 vnode_t *xvp = NULL, *vp;
6599 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6600 ap->a_cred, ap->a_td, VWRITE);
6604 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6611 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6618 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF,
6619 UIO_SYSSPACE, attrname, xvp, td);
6622 NDFREE(&nd, NDF_ONLY_PNBUF);
6625 if (error == ENOENT)
6629 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
6632 if (vp == nd.ni_dvp)
6642 * Vnode operation to set a named attribute.
6645 zfs_setextattr(struct vop_setextattr_args *ap)
6648 IN struct vnode *a_vp;
6649 IN int a_attrnamespace;
6650 IN const char *a_name;
6651 INOUT struct uio *a_uio;
6652 IN struct ucred *a_cred;
6653 IN struct thread *a_td;
6657 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6658 struct thread *td = ap->a_td;
6659 struct nameidata nd;
6662 vnode_t *xvp = NULL, *vp;
6665 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6666 ap->a_cred, ap->a_td, VWRITE);
6670 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6677 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6678 LOOKUP_XATTR | CREATE_XATTR_DIR);
6684 flags = FFLAGS(O_WRONLY | O_CREAT);
6685 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6687 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL);
6689 NDFREE(&nd, NDF_ONLY_PNBUF);
6697 error = VOP_SETATTR(vp, &va, ap->a_cred);
6699 VOP_WRITE(vp, ap->a_uio, IO_UNIT | IO_SYNC, ap->a_cred);
6702 vn_close(vp, flags, ap->a_cred, td);
6709 * Vnode operation to retrieve extended attributes on a vnode.
6712 zfs_listextattr(struct vop_listextattr_args *ap)
6715 IN struct vnode *a_vp;
6716 IN int a_attrnamespace;
6717 INOUT struct uio *a_uio;
6719 IN struct ucred *a_cred;
6720 IN struct thread *a_td;
6724 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6725 struct thread *td = ap->a_td;
6726 struct nameidata nd;
6727 char attrprefix[16];
6728 u_char dirbuf[sizeof(struct dirent)];
6731 struct uio auio, *uio = ap->a_uio;
6732 size_t *sizep = ap->a_size;
6734 vnode_t *xvp = NULL, *vp;
6735 int done, error, eof, pos;
6737 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6738 ap->a_cred, ap->a_td, VREAD);
6742 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix,
6743 sizeof(attrprefix));
6746 plen = strlen(attrprefix);
6753 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6758 * ENOATTR means that the EA directory does not yet exist,
6759 * i.e. there are no extended attributes there.
6761 if (error == ENOATTR)
6766 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED,
6767 UIO_SYSSPACE, ".", xvp, td);
6770 NDFREE(&nd, NDF_ONLY_PNBUF);
6776 auio.uio_iov = &aiov;
6777 auio.uio_iovcnt = 1;
6778 auio.uio_segflg = UIO_SYSSPACE;
6780 auio.uio_rw = UIO_READ;
6781 auio.uio_offset = 0;
6786 aiov.iov_base = (void *)dirbuf;
6787 aiov.iov_len = sizeof(dirbuf);
6788 auio.uio_resid = sizeof(dirbuf);
6789 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL);
6790 done = sizeof(dirbuf) - auio.uio_resid;
6793 for (pos = 0; pos < done;) {
6794 dp = (struct dirent *)(dirbuf + pos);
6795 pos += dp->d_reclen;
6797 * XXX: Temporarily we also accept DT_UNKNOWN, as this
6798 * is what we get when attribute was created on Solaris.
6800 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN)
6802 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0)
6804 else if (strncmp(dp->d_name, attrprefix, plen) != 0)
6806 nlen = dp->d_namlen - plen;
6809 else if (uio != NULL) {
6811 * Format of extattr name entry is one byte for
6812 * length and the rest for name.
6814 error = uiomove(&nlen, 1, uio->uio_rw, uio);
6816 error = uiomove(dp->d_name + plen, nlen,
6823 } while (!eof && error == 0);
6832 zfs_freebsd_getacl(ap)
6833 struct vop_getacl_args /* {
6842 vsecattr_t vsecattr;
6844 if (ap->a_type != ACL_TYPE_NFS4)
6847 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
6848 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL))
6851 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt);
6852 if (vsecattr.vsa_aclentp != NULL)
6853 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz);
6859 zfs_freebsd_setacl(ap)
6860 struct vop_setacl_args /* {
6869 vsecattr_t vsecattr;
6870 int aclbsize; /* size of acl list in bytes */
6873 if (ap->a_type != ACL_TYPE_NFS4)
6876 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES)
6880 * With NFSv4 ACLs, chmod(2) may need to add additional entries,
6881 * splitting every entry into two and appending "canonical six"
6882 * entries at the end. Don't allow for setting an ACL that would
6883 * cause chmod(2) to run out of ACL entries.
6885 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES)
6888 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR);
6892 vsecattr.vsa_mask = VSA_ACE;
6893 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t);
6894 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP);
6895 aaclp = vsecattr.vsa_aclentp;
6896 vsecattr.vsa_aclentsz = aclbsize;
6898 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp);
6899 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL);
6900 kmem_free(aaclp, aclbsize);
6906 zfs_freebsd_aclcheck(ap)
6907 struct vop_aclcheck_args /* {
6916 return (EOPNOTSUPP);
6919 struct vop_vector zfs_vnodeops;
6920 struct vop_vector zfs_fifoops;
6921 struct vop_vector zfs_shareops;
6923 struct vop_vector zfs_vnodeops = {
6924 .vop_default = &default_vnodeops,
6925 .vop_inactive = zfs_freebsd_inactive,
6926 .vop_reclaim = zfs_freebsd_reclaim,
6927 .vop_access = zfs_freebsd_access,
6928 #ifdef FREEBSD_NAMECACHE
6929 .vop_lookup = vfs_cache_lookup,
6930 .vop_cachedlookup = zfs_freebsd_lookup,
6932 .vop_lookup = zfs_freebsd_lookup,
6934 .vop_getattr = zfs_freebsd_getattr,
6935 .vop_setattr = zfs_freebsd_setattr,
6936 .vop_create = zfs_freebsd_create,
6937 .vop_mknod = zfs_freebsd_create,
6938 .vop_mkdir = zfs_freebsd_mkdir,
6939 .vop_readdir = zfs_freebsd_readdir,
6940 .vop_fsync = zfs_freebsd_fsync,
6941 .vop_open = zfs_freebsd_open,
6942 .vop_close = zfs_freebsd_close,
6943 .vop_rmdir = zfs_freebsd_rmdir,
6944 .vop_ioctl = zfs_freebsd_ioctl,
6945 .vop_link = zfs_freebsd_link,
6946 .vop_symlink = zfs_freebsd_symlink,
6947 .vop_readlink = zfs_freebsd_readlink,
6948 .vop_read = zfs_freebsd_read,
6949 .vop_write = zfs_freebsd_write,
6950 .vop_remove = zfs_freebsd_remove,
6951 .vop_rename = zfs_freebsd_rename,
6952 .vop_pathconf = zfs_freebsd_pathconf,
6953 .vop_bmap = zfs_freebsd_bmap,
6954 .vop_fid = zfs_freebsd_fid,
6955 .vop_getextattr = zfs_getextattr,
6956 .vop_deleteextattr = zfs_deleteextattr,
6957 .vop_setextattr = zfs_setextattr,
6958 .vop_listextattr = zfs_listextattr,
6959 .vop_getacl = zfs_freebsd_getacl,
6960 .vop_setacl = zfs_freebsd_setacl,
6961 .vop_aclcheck = zfs_freebsd_aclcheck,
6962 .vop_getpages = zfs_freebsd_getpages,
6965 struct vop_vector zfs_fifoops = {
6966 .vop_default = &fifo_specops,
6967 .vop_fsync = zfs_freebsd_fsync,
6968 .vop_access = zfs_freebsd_access,
6969 .vop_getattr = zfs_freebsd_getattr,
6970 .vop_inactive = zfs_freebsd_inactive,
6971 .vop_read = VOP_PANIC,
6972 .vop_reclaim = zfs_freebsd_reclaim,
6973 .vop_setattr = zfs_freebsd_setattr,
6974 .vop_write = VOP_PANIC,
6975 .vop_pathconf = zfs_freebsd_fifo_pathconf,
6976 .vop_fid = zfs_freebsd_fid,
6977 .vop_getacl = zfs_freebsd_getacl,
6978 .vop_setacl = zfs_freebsd_setacl,
6979 .vop_aclcheck = zfs_freebsd_aclcheck,
6983 * special share hidden files vnode operations template
6985 struct vop_vector zfs_shareops = {
6986 .vop_default = &default_vnodeops,
6987 .vop_access = zfs_freebsd_access,
6988 .vop_inactive = zfs_freebsd_inactive,
6989 .vop_reclaim = zfs_freebsd_reclaim,
6990 .vop_fid = zfs_freebsd_fid,
6991 .vop_pathconf = zfs_freebsd_pathconf,