4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2013 by Delphix. All rights reserved.
24 * Copyright 2013 Nexenta Systems, Inc. All rights reserved.
27 /* Portions Copyright 2007 Jeremy Teo */
28 /* Portions Copyright 2010 Robert Milkowski */
30 #include <sys/types.h>
31 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/sysmacros.h>
35 #include <sys/resource.h>
38 #include <sys/vnode.h>
42 #include <sys/taskq.h>
44 #include <sys/atomic.h>
45 #include <sys/namei.h>
47 #include <sys/cmn_err.h>
48 #include <sys/errno.h>
49 #include <sys/unistd.h>
50 #include <sys/zfs_dir.h>
51 #include <sys/zfs_ioctl.h>
52 #include <sys/fs/zfs.h>
54 #include <sys/dmu_objset.h>
60 #include <sys/dirent.h>
61 #include <sys/policy.h>
62 #include <sys/sunddi.h>
63 #include <sys/filio.h>
65 #include <sys/zfs_ctldir.h>
66 #include <sys/zfs_fuid.h>
67 #include <sys/zfs_sa.h>
69 #include <sys/zfs_rlock.h>
70 #include <sys/extdirent.h>
71 #include <sys/kidmap.h>
74 #include <sys/sf_buf.h>
75 #include <sys/sched.h>
77 #include <vm/vm_param.h>
78 #include <vm/vm_pageout.h>
83 * Each vnode op performs some logical unit of work. To do this, the ZPL must
84 * properly lock its in-core state, create a DMU transaction, do the work,
85 * record this work in the intent log (ZIL), commit the DMU transaction,
86 * and wait for the intent log to commit if it is a synchronous operation.
87 * Moreover, the vnode ops must work in both normal and log replay context.
88 * The ordering of events is important to avoid deadlocks and references
89 * to freed memory. The example below illustrates the following Big Rules:
91 * (1) A check must be made in each zfs thread for a mounted file system.
92 * This is done avoiding races using ZFS_ENTER(zfsvfs).
93 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
94 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
95 * can return EIO from the calling function.
97 * (2) VN_RELE() should always be the last thing except for zil_commit()
98 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
99 * First, if it's the last reference, the vnode/znode
100 * can be freed, so the zp may point to freed memory. Second, the last
101 * reference will call zfs_zinactive(), which may induce a lot of work --
102 * pushing cached pages (which acquires range locks) and syncing out
103 * cached atime changes. Third, zfs_zinactive() may require a new tx,
104 * which could deadlock the system if you were already holding one.
105 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
107 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
108 * as they can span dmu_tx_assign() calls.
110 * (4) Always pass TXG_NOWAIT as the second argument to dmu_tx_assign().
111 * This is critical because we don't want to block while holding locks.
112 * Note, in particular, that if a lock is sometimes acquired before
113 * the tx assigns, and sometimes after (e.g. z_lock), then failing to
114 * use a non-blocking assign can deadlock the system. The scenario:
116 * Thread A has grabbed a lock before calling dmu_tx_assign().
117 * Thread B is in an already-assigned tx, and blocks for this lock.
118 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
119 * forever, because the previous txg can't quiesce until B's tx commits.
121 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
122 * then drop all locks, call dmu_tx_wait(), and try again.
124 * (5) If the operation succeeded, generate the intent log entry for it
125 * before dropping locks. This ensures that the ordering of events
126 * in the intent log matches the order in which they actually occurred.
127 * During ZIL replay the zfs_log_* functions will update the sequence
128 * number to indicate the zil transaction has replayed.
130 * (6) At the end of each vnode op, the DMU tx must always commit,
131 * regardless of whether there were any errors.
133 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
134 * to ensure that synchronous semantics are provided when necessary.
136 * In general, this is how things should be ordered in each vnode op:
138 * ZFS_ENTER(zfsvfs); // exit if unmounted
140 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD())
141 * rw_enter(...); // grab any other locks you need
142 * tx = dmu_tx_create(...); // get DMU tx
143 * dmu_tx_hold_*(); // hold each object you might modify
144 * error = dmu_tx_assign(tx, TXG_NOWAIT); // try to assign
146 * rw_exit(...); // drop locks
147 * zfs_dirent_unlock(dl); // unlock directory entry
148 * VN_RELE(...); // release held vnodes
149 * if (error == ERESTART) {
154 * dmu_tx_abort(tx); // abort DMU tx
155 * ZFS_EXIT(zfsvfs); // finished in zfs
156 * return (error); // really out of space
158 * error = do_real_work(); // do whatever this VOP does
160 * zfs_log_*(...); // on success, make ZIL entry
161 * dmu_tx_commit(tx); // commit DMU tx -- error or not
162 * rw_exit(...); // drop locks
163 * zfs_dirent_unlock(dl); // unlock directory entry
164 * VN_RELE(...); // release held vnodes
165 * zil_commit(zilog, foid); // synchronous when necessary
166 * ZFS_EXIT(zfsvfs); // finished in zfs
167 * return (error); // done, report error
172 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
174 znode_t *zp = VTOZ(*vpp);
175 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
180 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
181 ((flag & FAPPEND) == 0)) {
183 return (SET_ERROR(EPERM));
186 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
187 ZTOV(zp)->v_type == VREG &&
188 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
189 if (fs_vscan(*vpp, cr, 0) != 0) {
191 return (SET_ERROR(EACCES));
195 /* Keep a count of the synchronous opens in the znode */
196 if (flag & (FSYNC | FDSYNC))
197 atomic_inc_32(&zp->z_sync_cnt);
205 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
206 caller_context_t *ct)
208 znode_t *zp = VTOZ(vp);
209 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
212 * Clean up any locks held by this process on the vp.
214 cleanlocks(vp, ddi_get_pid(), 0);
215 cleanshares(vp, ddi_get_pid());
220 /* Decrement the synchronous opens in the znode */
221 if ((flag & (FSYNC | FDSYNC)) && (count == 1))
222 atomic_dec_32(&zp->z_sync_cnt);
224 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
225 ZTOV(zp)->v_type == VREG &&
226 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
227 VERIFY(fs_vscan(vp, cr, 1) == 0);
234 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
235 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
238 zfs_holey(vnode_t *vp, u_long cmd, offset_t *off)
240 znode_t *zp = VTOZ(vp);
241 uint64_t noff = (uint64_t)*off; /* new offset */
246 file_sz = zp->z_size;
247 if (noff >= file_sz) {
248 return (SET_ERROR(ENXIO));
251 if (cmd == _FIO_SEEK_HOLE)
256 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
259 if ((error == ESRCH) || (noff > file_sz)) {
261 * Handle the virtual hole at the end of file.
267 return (SET_ERROR(ENXIO));
278 zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred,
279 int *rvalp, caller_context_t *ct)
291 * The following two ioctls are used by bfu. Faking out,
292 * necessary to avoid bfu errors.
301 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
302 return (SET_ERROR(EFAULT));
304 off = *(offset_t *)data;
307 zfsvfs = zp->z_zfsvfs;
311 /* offset parameter is in/out */
312 error = zfs_holey(vp, com, &off);
317 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
318 return (SET_ERROR(EFAULT));
320 *(offset_t *)data = off;
324 return (SET_ERROR(ENOTTY));
328 page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes)
335 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE
336 * aligned boundaries, if the range is not aligned. As a result a
337 * DEV_BSIZE subrange with partially dirty data may get marked as clean.
338 * It may happen that all DEV_BSIZE subranges are marked clean and thus
339 * the whole page would be considred clean despite have some dirty data.
340 * For this reason we should shrink the range to DEV_BSIZE aligned
341 * boundaries before calling vm_page_clear_dirty.
343 end = rounddown2(off + nbytes, DEV_BSIZE);
344 off = roundup2(off, DEV_BSIZE);
348 zfs_vmobject_assert_wlocked(obj);
351 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
353 if (vm_page_xbusied(pp)) {
355 * Reference the page before unlocking and
356 * sleeping so that the page daemon is less
357 * likely to reclaim it.
359 vm_page_reference(pp);
361 zfs_vmobject_wunlock(obj);
362 vm_page_busy_sleep(pp, "zfsmwb");
363 zfs_vmobject_wlock(obj);
367 } else if (pp == NULL) {
368 pp = vm_page_alloc(obj, OFF_TO_IDX(start),
369 VM_ALLOC_SYSTEM | VM_ALLOC_IFCACHED |
372 ASSERT(pp != NULL && !pp->valid);
377 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
378 vm_object_pip_add(obj, 1);
379 pmap_remove_write(pp);
381 vm_page_clear_dirty(pp, off, nbytes);
389 page_unbusy(vm_page_t pp)
393 vm_object_pip_subtract(pp->object, 1);
397 page_hold(vnode_t *vp, int64_t start)
403 zfs_vmobject_assert_wlocked(obj);
406 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
408 if (vm_page_xbusied(pp)) {
410 * Reference the page before unlocking and
411 * sleeping so that the page daemon is less
412 * likely to reclaim it.
414 vm_page_reference(pp);
416 zfs_vmobject_wunlock(obj);
417 vm_page_busy_sleep(pp, "zfsmwb");
418 zfs_vmobject_wlock(obj);
422 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
435 page_unhold(vm_page_t pp)
444 zfs_map_page(vm_page_t pp, struct sf_buf **sfp)
447 *sfp = sf_buf_alloc(pp, 0);
448 return ((caddr_t)sf_buf_kva(*sfp));
452 zfs_unmap_page(struct sf_buf *sf)
459 * When a file is memory mapped, we must keep the IO data synchronized
460 * between the DMU cache and the memory mapped pages. What this means:
462 * On Write: If we find a memory mapped page, we write to *both*
463 * the page and the dmu buffer.
466 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid,
467 int segflg, dmu_tx_t *tx)
474 ASSERT(vp->v_mount != NULL);
478 off = start & PAGEOFFSET;
479 zfs_vmobject_wlock(obj);
480 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
482 int nbytes = imin(PAGESIZE - off, len);
484 if (segflg == UIO_NOCOPY) {
485 pp = vm_page_lookup(obj, OFF_TO_IDX(start));
487 ("zfs update_pages: NULL page in putpages case"));
489 ("zfs update_pages: unaligned data in putpages case"));
490 KASSERT(pp->valid == VM_PAGE_BITS_ALL,
491 ("zfs update_pages: invalid page in putpages case"));
492 KASSERT(vm_page_sbusied(pp),
493 ("zfs update_pages: unbusy page in putpages case"));
494 KASSERT(!pmap_page_is_write_mapped(pp),
495 ("zfs update_pages: writable page in putpages case"));
496 zfs_vmobject_wunlock(obj);
498 va = zfs_map_page(pp, &sf);
499 (void) dmu_write(os, oid, start, nbytes, va, tx);
502 zfs_vmobject_wlock(obj);
504 } else if ((pp = page_busy(vp, start, off, nbytes)) != NULL) {
505 zfs_vmobject_wunlock(obj);
507 va = zfs_map_page(pp, &sf);
508 (void) dmu_read(os, oid, start+off, nbytes,
509 va+off, DMU_READ_PREFETCH);;
512 zfs_vmobject_wlock(obj);
518 if (segflg != UIO_NOCOPY)
519 vm_object_pip_wakeupn(obj, 0);
520 zfs_vmobject_wunlock(obj);
524 * Read with UIO_NOCOPY flag means that sendfile(2) requests
525 * ZFS to populate a range of page cache pages with data.
527 * NOTE: this function could be optimized to pre-allocate
528 * all pages in advance, drain exclusive busy on all of them,
529 * map them into contiguous KVA region and populate them
530 * in one single dmu_read() call.
533 mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio)
535 znode_t *zp = VTOZ(vp);
536 objset_t *os = zp->z_zfsvfs->z_os;
546 ASSERT(uio->uio_segflg == UIO_NOCOPY);
547 ASSERT(vp->v_mount != NULL);
550 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0);
552 zfs_vmobject_wlock(obj);
553 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) {
554 int bytes = MIN(PAGESIZE, len);
556 pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_SBUSY |
557 VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY);
558 if (pp->valid == 0) {
559 zfs_vmobject_wunlock(obj);
560 va = zfs_map_page(pp, &sf);
561 error = dmu_read(os, zp->z_id, start, bytes, va,
563 if (bytes != PAGESIZE && error == 0)
564 bzero(va + bytes, PAGESIZE - bytes);
566 zfs_vmobject_wlock(obj);
570 if (pp->wire_count == 0 && pp->valid == 0 &&
574 pp->valid = VM_PAGE_BITS_ALL;
575 vm_page_activate(pp);
582 uio->uio_resid -= bytes;
583 uio->uio_offset += bytes;
586 zfs_vmobject_wunlock(obj);
591 * When a file is memory mapped, we must keep the IO data synchronized
592 * between the DMU cache and the memory mapped pages. What this means:
594 * On Read: We "read" preferentially from memory mapped pages,
595 * else we default from the dmu buffer.
597 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
598 * the file is memory mapped.
601 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
603 znode_t *zp = VTOZ(vp);
604 objset_t *os = zp->z_zfsvfs->z_os;
612 ASSERT(vp->v_mount != NULL);
616 start = uio->uio_loffset;
617 off = start & PAGEOFFSET;
618 zfs_vmobject_wlock(obj);
619 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
621 uint64_t bytes = MIN(PAGESIZE - off, len);
623 if (pp = page_hold(vp, start)) {
627 zfs_vmobject_wunlock(obj);
628 va = zfs_map_page(pp, &sf);
629 error = uiomove(va + off, bytes, UIO_READ, uio);
631 zfs_vmobject_wlock(obj);
634 zfs_vmobject_wunlock(obj);
635 error = dmu_read_uio(os, zp->z_id, uio, bytes);
636 zfs_vmobject_wlock(obj);
643 zfs_vmobject_wunlock(obj);
647 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
650 * Read bytes from specified file into supplied buffer.
652 * IN: vp - vnode of file to be read from.
653 * uio - structure supplying read location, range info,
655 * ioflag - SYNC flags; used to provide FRSYNC semantics.
656 * cr - credentials of caller.
657 * ct - caller context
659 * OUT: uio - updated offset and range, buffer filled.
661 * RETURN: 0 on success, error code on failure.
664 * vp - atime updated if byte count > 0
668 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
670 znode_t *zp = VTOZ(vp);
671 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
682 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
684 return (SET_ERROR(EACCES));
688 * Validate file offset
690 if (uio->uio_loffset < (offset_t)0) {
692 return (SET_ERROR(EINVAL));
696 * Fasttrack empty reads
698 if (uio->uio_resid == 0) {
704 * Check for mandatory locks
706 if (MANDMODE(zp->z_mode)) {
707 if (error = chklock(vp, FREAD,
708 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
715 * If we're in FRSYNC mode, sync out this znode before reading it.
718 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
719 zil_commit(zfsvfs->z_log, zp->z_id);
722 * Lock the range against changes.
724 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
727 * If we are reading past end-of-file we can skip
728 * to the end; but we might still need to set atime.
730 if (uio->uio_loffset >= zp->z_size) {
735 ASSERT(uio->uio_loffset < zp->z_size);
736 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
739 if ((uio->uio_extflg == UIO_XUIO) &&
740 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
742 int blksz = zp->z_blksz;
743 uint64_t offset = uio->uio_loffset;
745 xuio = (xuio_t *)uio;
747 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
750 ASSERT(offset + n <= blksz);
753 (void) dmu_xuio_init(xuio, nblk);
755 if (vn_has_cached_data(vp)) {
757 * For simplicity, we always allocate a full buffer
758 * even if we only expect to read a portion of a block.
760 while (--nblk >= 0) {
761 (void) dmu_xuio_add(xuio,
762 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
770 nbytes = MIN(n, zfs_read_chunk_size -
771 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
774 if (uio->uio_segflg == UIO_NOCOPY)
775 error = mappedread_sf(vp, nbytes, uio);
777 #endif /* __FreeBSD__ */
778 if (vn_has_cached_data(vp))
779 error = mappedread(vp, nbytes, uio);
781 error = dmu_read_uio(os, zp->z_id, uio, nbytes);
783 /* convert checksum errors into IO errors */
785 error = SET_ERROR(EIO);
792 zfs_range_unlock(rl);
794 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
800 * Write the bytes to a file.
802 * IN: vp - vnode of file to be written to.
803 * uio - structure supplying write location, range info,
805 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is
806 * set if in append mode.
807 * cr - credentials of caller.
808 * ct - caller context (NFS/CIFS fem monitor only)
810 * OUT: uio - updated offset and range.
812 * RETURN: 0 on success, error code on failure.
815 * vp - ctime|mtime updated if byte count > 0
820 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
822 znode_t *zp = VTOZ(vp);
823 rlim64_t limit = MAXOFFSET_T;
824 ssize_t start_resid = uio->uio_resid;
828 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
833 int max_blksz = zfsvfs->z_max_blksz;
836 iovec_t *aiov = NULL;
839 int iovcnt = uio->uio_iovcnt;
840 iovec_t *iovp = uio->uio_iov;
843 sa_bulk_attr_t bulk[4];
844 uint64_t mtime[2], ctime[2];
847 * Fasttrack empty write
853 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
859 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
860 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
861 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
863 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
867 * If immutable or not appending then return EPERM
869 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
870 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
871 (uio->uio_loffset < zp->z_size))) {
873 return (SET_ERROR(EPERM));
876 zilog = zfsvfs->z_log;
879 * Validate file offset
881 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
884 return (SET_ERROR(EINVAL));
888 * Check for mandatory locks before calling zfs_range_lock()
889 * in order to prevent a deadlock with locks set via fcntl().
891 if (MANDMODE((mode_t)zp->z_mode) &&
892 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
899 * Pre-fault the pages to ensure slow (eg NFS) pages
901 * Skip this if uio contains loaned arc_buf.
903 if ((uio->uio_extflg == UIO_XUIO) &&
904 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
905 xuio = (xuio_t *)uio;
907 uio_prefaultpages(MIN(n, max_blksz), uio);
911 * If in append mode, set the io offset pointer to eof.
913 if (ioflag & FAPPEND) {
915 * Obtain an appending range lock to guarantee file append
916 * semantics. We reset the write offset once we have the lock.
918 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
920 if (rl->r_len == UINT64_MAX) {
922 * We overlocked the file because this write will cause
923 * the file block size to increase.
924 * Note that zp_size cannot change with this lock held.
928 uio->uio_loffset = woff;
931 * Note that if the file block size will change as a result of
932 * this write, then this range lock will lock the entire file
933 * so that we can re-write the block safely.
935 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
938 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) {
939 zfs_range_unlock(rl);
945 zfs_range_unlock(rl);
947 return (SET_ERROR(EFBIG));
950 if ((woff + n) > limit || woff > (limit - n))
953 /* Will this write extend the file length? */
954 write_eof = (woff + n > zp->z_size);
956 end_size = MAX(zp->z_size, woff + n);
959 * Write the file in reasonable size chunks. Each chunk is written
960 * in a separate transaction; this keeps the intent log records small
961 * and allows us to do more fine-grained space accounting.
965 woff = uio->uio_loffset;
967 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
968 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
970 dmu_return_arcbuf(abuf);
971 error = SET_ERROR(EDQUOT);
975 if (xuio && abuf == NULL) {
976 ASSERT(i_iov < iovcnt);
978 abuf = dmu_xuio_arcbuf(xuio, i_iov);
979 dmu_xuio_clear(xuio, i_iov);
980 DTRACE_PROBE3(zfs_cp_write, int, i_iov,
981 iovec_t *, aiov, arc_buf_t *, abuf);
982 ASSERT((aiov->iov_base == abuf->b_data) ||
983 ((char *)aiov->iov_base - (char *)abuf->b_data +
984 aiov->iov_len == arc_buf_size(abuf)));
986 } else if (abuf == NULL && n >= max_blksz &&
987 woff >= zp->z_size &&
988 P2PHASE(woff, max_blksz) == 0 &&
989 zp->z_blksz == max_blksz) {
991 * This write covers a full block. "Borrow" a buffer
992 * from the dmu so that we can fill it before we enter
993 * a transaction. This avoids the possibility of
994 * holding up the transaction if the data copy hangs
995 * up on a pagefault (e.g., from an NFS server mapping).
999 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
1001 ASSERT(abuf != NULL);
1002 ASSERT(arc_buf_size(abuf) == max_blksz);
1003 if (error = uiocopy(abuf->b_data, max_blksz,
1004 UIO_WRITE, uio, &cbytes)) {
1005 dmu_return_arcbuf(abuf);
1008 ASSERT(cbytes == max_blksz);
1012 * Start a transaction.
1014 tx = dmu_tx_create(zfsvfs->z_os);
1015 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1016 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
1017 zfs_sa_upgrade_txholds(tx, zp);
1018 error = dmu_tx_assign(tx, TXG_NOWAIT);
1020 if (error == ERESTART) {
1027 dmu_return_arcbuf(abuf);
1032 * If zfs_range_lock() over-locked we grow the blocksize
1033 * and then reduce the lock range. This will only happen
1034 * on the first iteration since zfs_range_reduce() will
1035 * shrink down r_len to the appropriate size.
1037 if (rl->r_len == UINT64_MAX) {
1040 if (zp->z_blksz > max_blksz) {
1041 ASSERT(!ISP2(zp->z_blksz));
1042 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
1044 new_blksz = MIN(end_size, max_blksz);
1046 zfs_grow_blocksize(zp, new_blksz, tx);
1047 zfs_range_reduce(rl, woff, n);
1051 * XXX - should we really limit each write to z_max_blksz?
1052 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
1054 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
1056 if (woff + nbytes > zp->z_size)
1057 vnode_pager_setsize(vp, woff + nbytes);
1060 tx_bytes = uio->uio_resid;
1061 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
1063 tx_bytes -= uio->uio_resid;
1066 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
1068 * If this is not a full block write, but we are
1069 * extending the file past EOF and this data starts
1070 * block-aligned, use assign_arcbuf(). Otherwise,
1071 * write via dmu_write().
1073 if (tx_bytes < max_blksz && (!write_eof ||
1074 aiov->iov_base != abuf->b_data)) {
1076 dmu_write(zfsvfs->z_os, zp->z_id, woff,
1077 aiov->iov_len, aiov->iov_base, tx);
1078 dmu_return_arcbuf(abuf);
1079 xuio_stat_wbuf_copied();
1081 ASSERT(xuio || tx_bytes == max_blksz);
1082 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
1085 ASSERT(tx_bytes <= uio->uio_resid);
1086 uioskip(uio, tx_bytes);
1088 if (tx_bytes && vn_has_cached_data(vp)) {
1089 update_pages(vp, woff, tx_bytes, zfsvfs->z_os,
1090 zp->z_id, uio->uio_segflg, tx);
1094 * If we made no progress, we're done. If we made even
1095 * partial progress, update the znode and ZIL accordingly.
1097 if (tx_bytes == 0) {
1098 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
1099 (void *)&zp->z_size, sizeof (uint64_t), tx);
1106 * Clear Set-UID/Set-GID bits on successful write if not
1107 * privileged and at least one of the excute bits is set.
1109 * It would be nice to to this after all writes have
1110 * been done, but that would still expose the ISUID/ISGID
1111 * to another app after the partial write is committed.
1113 * Note: we don't call zfs_fuid_map_id() here because
1114 * user 0 is not an ephemeral uid.
1116 mutex_enter(&zp->z_acl_lock);
1117 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
1118 (S_IXUSR >> 6))) != 0 &&
1119 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
1120 secpolicy_vnode_setid_retain(vp, cr,
1121 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
1123 zp->z_mode &= ~(S_ISUID | S_ISGID);
1124 newmode = zp->z_mode;
1125 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
1126 (void *)&newmode, sizeof (uint64_t), tx);
1128 mutex_exit(&zp->z_acl_lock);
1130 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
1134 * Update the file size (zp_size) if it has changed;
1135 * account for possible concurrent updates.
1137 while ((end_size = zp->z_size) < uio->uio_loffset) {
1138 (void) atomic_cas_64(&zp->z_size, end_size,
1143 * If we are replaying and eof is non zero then force
1144 * the file size to the specified eof. Note, there's no
1145 * concurrency during replay.
1147 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
1148 zp->z_size = zfsvfs->z_replay_eof;
1150 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1152 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
1157 ASSERT(tx_bytes == nbytes);
1162 uio_prefaultpages(MIN(n, max_blksz), uio);
1166 zfs_range_unlock(rl);
1169 * If we're in replay mode, or we made no progress, return error.
1170 * Otherwise, it's at least a partial write, so it's successful.
1172 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1177 if (ioflag & (FSYNC | FDSYNC) ||
1178 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1179 zil_commit(zilog, zp->z_id);
1186 zfs_get_done(zgd_t *zgd, int error)
1188 znode_t *zp = zgd->zgd_private;
1189 objset_t *os = zp->z_zfsvfs->z_os;
1192 dmu_buf_rele(zgd->zgd_db, zgd);
1194 zfs_range_unlock(zgd->zgd_rl);
1197 * Release the vnode asynchronously as we currently have the
1198 * txg stopped from syncing.
1200 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1202 if (error == 0 && zgd->zgd_bp)
1203 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1205 kmem_free(zgd, sizeof (zgd_t));
1209 static int zil_fault_io = 0;
1213 * Get data to generate a TX_WRITE intent log record.
1216 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1218 zfsvfs_t *zfsvfs = arg;
1219 objset_t *os = zfsvfs->z_os;
1221 uint64_t object = lr->lr_foid;
1222 uint64_t offset = lr->lr_offset;
1223 uint64_t size = lr->lr_length;
1224 blkptr_t *bp = &lr->lr_blkptr;
1229 ASSERT(zio != NULL);
1233 * Nothing to do if the file has been removed
1235 if (zfs_zget(zfsvfs, object, &zp) != 0)
1236 return (SET_ERROR(ENOENT));
1237 if (zp->z_unlinked) {
1239 * Release the vnode asynchronously as we currently have the
1240 * txg stopped from syncing.
1242 VN_RELE_ASYNC(ZTOV(zp),
1243 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1244 return (SET_ERROR(ENOENT));
1247 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1248 zgd->zgd_zilog = zfsvfs->z_log;
1249 zgd->zgd_private = zp;
1252 * Write records come in two flavors: immediate and indirect.
1253 * For small writes it's cheaper to store the data with the
1254 * log record (immediate); for large writes it's cheaper to
1255 * sync the data and get a pointer to it (indirect) so that
1256 * we don't have to write the data twice.
1258 if (buf != NULL) { /* immediate write */
1259 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1260 /* test for truncation needs to be done while range locked */
1261 if (offset >= zp->z_size) {
1262 error = SET_ERROR(ENOENT);
1264 error = dmu_read(os, object, offset, size, buf,
1265 DMU_READ_NO_PREFETCH);
1267 ASSERT(error == 0 || error == ENOENT);
1268 } else { /* indirect write */
1270 * Have to lock the whole block to ensure when it's
1271 * written out and it's checksum is being calculated
1272 * that no one can change the data. We need to re-check
1273 * blocksize after we get the lock in case it's changed!
1278 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1280 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1282 if (zp->z_blksz == size)
1285 zfs_range_unlock(zgd->zgd_rl);
1287 /* test for truncation needs to be done while range locked */
1288 if (lr->lr_offset >= zp->z_size)
1289 error = SET_ERROR(ENOENT);
1292 error = SET_ERROR(EIO);
1297 error = dmu_buf_hold(os, object, offset, zgd, &db,
1298 DMU_READ_NO_PREFETCH);
1301 blkptr_t *obp = dmu_buf_get_blkptr(db);
1303 ASSERT(BP_IS_HOLE(bp));
1310 ASSERT(db->db_offset == offset);
1311 ASSERT(db->db_size == size);
1313 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1315 ASSERT(error || lr->lr_length <= zp->z_blksz);
1318 * On success, we need to wait for the write I/O
1319 * initiated by dmu_sync() to complete before we can
1320 * release this dbuf. We will finish everything up
1321 * in the zfs_get_done() callback.
1326 if (error == EALREADY) {
1327 lr->lr_common.lrc_txtype = TX_WRITE2;
1333 zfs_get_done(zgd, error);
1340 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1341 caller_context_t *ct)
1343 znode_t *zp = VTOZ(vp);
1344 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1350 if (flag & V_ACE_MASK)
1351 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1353 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1360 * If vnode is for a device return a specfs vnode instead.
1363 specvp_check(vnode_t **vpp, cred_t *cr)
1367 if (IS_DEVVP(*vpp)) {
1370 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1373 error = SET_ERROR(ENOSYS);
1381 * Lookup an entry in a directory, or an extended attribute directory.
1382 * If it exists, return a held vnode reference for it.
1384 * IN: dvp - vnode of directory to search.
1385 * nm - name of entry to lookup.
1386 * pnp - full pathname to lookup [UNUSED].
1387 * flags - LOOKUP_XATTR set if looking for an attribute.
1388 * rdir - root directory vnode [UNUSED].
1389 * cr - credentials of caller.
1390 * ct - caller context
1391 * direntflags - directory lookup flags
1392 * realpnp - returned pathname.
1394 * OUT: vpp - vnode of located entry, NULL if not found.
1396 * RETURN: 0 on success, error code on failure.
1403 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp,
1404 int nameiop, cred_t *cr, kthread_t *td, int flags)
1406 znode_t *zdp = VTOZ(dvp);
1407 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1409 int *direntflags = NULL;
1410 void *realpnp = NULL;
1413 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1415 if (dvp->v_type != VDIR) {
1416 return (SET_ERROR(ENOTDIR));
1417 } else if (zdp->z_sa_hdl == NULL) {
1418 return (SET_ERROR(EIO));
1421 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1422 error = zfs_fastaccesschk_execute(zdp, cr);
1430 vnode_t *tvp = dnlc_lookup(dvp, nm);
1433 error = zfs_fastaccesschk_execute(zdp, cr);
1438 if (tvp == DNLC_NO_VNODE) {
1440 return (SET_ERROR(ENOENT));
1443 return (specvp_check(vpp, cr));
1449 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1456 if (flags & LOOKUP_XATTR) {
1459 * If the xattr property is off, refuse the lookup request.
1461 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1463 return (SET_ERROR(EINVAL));
1468 * We don't allow recursive attributes..
1469 * Maybe someday we will.
1471 if (zdp->z_pflags & ZFS_XATTR) {
1473 return (SET_ERROR(EINVAL));
1476 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1482 * Do we have permission to get into attribute directory?
1485 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1495 if (dvp->v_type != VDIR) {
1497 return (SET_ERROR(ENOTDIR));
1501 * Check accessibility of directory.
1504 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1509 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1510 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1512 return (SET_ERROR(EILSEQ));
1515 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1517 error = specvp_check(vpp, cr);
1519 /* Translate errors and add SAVENAME when needed. */
1520 if (cnp->cn_flags & ISLASTCN) {
1524 if (error == ENOENT) {
1525 error = EJUSTRETURN;
1526 cnp->cn_flags |= SAVENAME;
1532 cnp->cn_flags |= SAVENAME;
1536 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) {
1539 if (cnp->cn_flags & ISDOTDOT) {
1540 ltype = VOP_ISLOCKED(dvp);
1544 error = vn_lock(*vpp, cnp->cn_lkflags);
1545 if (cnp->cn_flags & ISDOTDOT)
1546 vn_lock(dvp, ltype | LK_RETRY);
1556 #ifdef FREEBSD_NAMECACHE
1558 * Insert name into cache (as non-existent) if appropriate.
1560 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
1561 cache_enter(dvp, *vpp, cnp);
1563 * Insert name into cache if appropriate.
1565 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) {
1566 if (!(cnp->cn_flags & ISLASTCN) ||
1567 (nameiop != DELETE && nameiop != RENAME)) {
1568 cache_enter(dvp, *vpp, cnp);
1577 * Attempt to create a new entry in a directory. If the entry
1578 * already exists, truncate the file if permissible, else return
1579 * an error. Return the vp of the created or trunc'd file.
1581 * IN: dvp - vnode of directory to put new file entry in.
1582 * name - name of new file entry.
1583 * vap - attributes of new file.
1584 * excl - flag indicating exclusive or non-exclusive mode.
1585 * mode - mode to open file with.
1586 * cr - credentials of caller.
1587 * flag - large file flag [UNUSED].
1588 * ct - caller context
1589 * vsecp - ACL to be set
1591 * OUT: vpp - vnode of created or trunc'd entry.
1593 * RETURN: 0 on success, error code on failure.
1596 * dvp - ctime|mtime updated if new entry created
1597 * vp - ctime|mtime always, atime if new
1602 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode,
1603 vnode_t **vpp, cred_t *cr, kthread_t *td)
1605 znode_t *zp, *dzp = VTOZ(dvp);
1606 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1614 gid_t gid = crgetgid(cr);
1615 zfs_acl_ids_t acl_ids;
1616 boolean_t fuid_dirtied;
1617 boolean_t have_acl = B_FALSE;
1622 * If we have an ephemeral id, ACL, or XVATTR then
1623 * make sure file system is at proper version
1626 ksid = crgetsid(cr, KSID_OWNER);
1628 uid = ksid_getid(ksid);
1632 if (zfsvfs->z_use_fuids == B_FALSE &&
1633 (vsecp || (vap->va_mask & AT_XVATTR) ||
1634 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1635 return (SET_ERROR(EINVAL));
1640 zilog = zfsvfs->z_log;
1642 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1643 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1645 return (SET_ERROR(EILSEQ));
1648 if (vap->va_mask & AT_XVATTR) {
1649 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
1650 crgetuid(cr), cr, vap->va_type)) != 0) {
1658 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr))
1659 vap->va_mode &= ~S_ISVTX;
1661 if (*name == '\0') {
1663 * Null component name refers to the directory itself.
1670 /* possible VN_HOLD(zp) */
1673 if (flag & FIGNORECASE)
1676 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1680 zfs_acl_ids_free(&acl_ids);
1681 if (strcmp(name, "..") == 0)
1682 error = SET_ERROR(EISDIR);
1692 * Create a new file object and update the directory
1695 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1697 zfs_acl_ids_free(&acl_ids);
1702 * We only support the creation of regular files in
1703 * extended attribute directories.
1706 if ((dzp->z_pflags & ZFS_XATTR) &&
1707 (vap->va_type != VREG)) {
1709 zfs_acl_ids_free(&acl_ids);
1710 error = SET_ERROR(EINVAL);
1714 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1715 cr, vsecp, &acl_ids)) != 0)
1719 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1720 zfs_acl_ids_free(&acl_ids);
1721 error = SET_ERROR(EDQUOT);
1725 tx = dmu_tx_create(os);
1727 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1728 ZFS_SA_BASE_ATTR_SIZE);
1730 fuid_dirtied = zfsvfs->z_fuid_dirty;
1732 zfs_fuid_txhold(zfsvfs, tx);
1733 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1734 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1735 if (!zfsvfs->z_use_sa &&
1736 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1737 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1738 0, acl_ids.z_aclp->z_acl_bytes);
1740 error = dmu_tx_assign(tx, TXG_NOWAIT);
1742 zfs_dirent_unlock(dl);
1743 if (error == ERESTART) {
1748 zfs_acl_ids_free(&acl_ids);
1753 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1756 zfs_fuid_sync(zfsvfs, tx);
1758 (void) zfs_link_create(dl, zp, tx, ZNEW);
1759 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1760 if (flag & FIGNORECASE)
1762 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1763 vsecp, acl_ids.z_fuidp, vap);
1764 zfs_acl_ids_free(&acl_ids);
1767 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1770 zfs_acl_ids_free(&acl_ids);
1774 * A directory entry already exists for this name.
1777 * Can't truncate an existing file if in exclusive mode.
1780 error = SET_ERROR(EEXIST);
1784 * Can't open a directory for writing.
1786 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1787 error = SET_ERROR(EISDIR);
1791 * Verify requested access to file.
1793 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1797 mutex_enter(&dzp->z_lock);
1799 mutex_exit(&dzp->z_lock);
1802 * Truncate regular files if requested.
1804 if ((ZTOV(zp)->v_type == VREG) &&
1805 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1806 /* we can't hold any locks when calling zfs_freesp() */
1807 zfs_dirent_unlock(dl);
1809 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1811 vnevent_create(ZTOV(zp), ct);
1817 zfs_dirent_unlock(dl);
1824 error = specvp_check(vpp, cr);
1827 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1828 zil_commit(zilog, 0);
1835 * Remove an entry from a directory.
1837 * IN: dvp - vnode of directory to remove entry from.
1838 * name - name of entry to remove.
1839 * cr - credentials of caller.
1840 * ct - caller context
1841 * flags - case flags
1843 * RETURN: 0 on success, error code on failure.
1847 * vp - ctime (if nlink > 0)
1850 uint64_t null_xattr = 0;
1854 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1857 znode_t *zp, *dzp = VTOZ(dvp);
1860 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1862 uint64_t acl_obj, xattr_obj;
1863 uint64_t xattr_obj_unlinked = 0;
1867 boolean_t may_delete_now, delete_now = FALSE;
1868 boolean_t unlinked, toobig = FALSE;
1870 pathname_t *realnmp = NULL;
1877 zilog = zfsvfs->z_log;
1879 if (flags & FIGNORECASE) {
1889 * Attempt to lock directory; fail if entry doesn't exist.
1891 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1901 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1906 * Need to use rmdir for removing directories.
1908 if (vp->v_type == VDIR) {
1909 error = SET_ERROR(EPERM);
1913 vnevent_remove(vp, dvp, name, ct);
1916 dnlc_remove(dvp, realnmp->pn_buf);
1918 dnlc_remove(dvp, name);
1921 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1925 * We may delete the znode now, or we may put it in the unlinked set;
1926 * it depends on whether we're the last link, and on whether there are
1927 * other holds on the vnode. So we dmu_tx_hold() the right things to
1928 * allow for either case.
1931 tx = dmu_tx_create(zfsvfs->z_os);
1932 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1933 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1934 zfs_sa_upgrade_txholds(tx, zp);
1935 zfs_sa_upgrade_txholds(tx, dzp);
1936 if (may_delete_now) {
1938 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1939 /* if the file is too big, only hold_free a token amount */
1940 dmu_tx_hold_free(tx, zp->z_id, 0,
1941 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1944 /* are there any extended attributes? */
1945 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1946 &xattr_obj, sizeof (xattr_obj));
1947 if (error == 0 && xattr_obj) {
1948 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1950 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1951 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1954 mutex_enter(&zp->z_lock);
1955 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1956 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1957 mutex_exit(&zp->z_lock);
1959 /* charge as an update -- would be nice not to charge at all */
1960 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1962 error = dmu_tx_assign(tx, TXG_NOWAIT);
1964 zfs_dirent_unlock(dl);
1968 if (error == ERESTART) {
1981 * Remove the directory entry.
1983 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1993 * Hold z_lock so that we can make sure that the ACL obj
1994 * hasn't changed. Could have been deleted due to
1997 mutex_enter(&zp->z_lock);
1999 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
2000 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
2001 delete_now = may_delete_now && !toobig &&
2002 vp->v_count == 1 && !vn_has_cached_data(vp) &&
2003 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
2010 panic("zfs_remove: delete_now branch taken");
2012 if (xattr_obj_unlinked) {
2013 ASSERT3U(xzp->z_links, ==, 2);
2014 mutex_enter(&xzp->z_lock);
2015 xzp->z_unlinked = 1;
2017 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
2018 &xzp->z_links, sizeof (xzp->z_links), tx);
2019 ASSERT3U(error, ==, 0);
2020 mutex_exit(&xzp->z_lock);
2021 zfs_unlinked_add(xzp, tx);
2024 error = sa_remove(zp->z_sa_hdl,
2025 SA_ZPL_XATTR(zfsvfs), tx);
2027 error = sa_update(zp->z_sa_hdl,
2028 SA_ZPL_XATTR(zfsvfs), &null_xattr,
2029 sizeof (uint64_t), tx);
2034 ASSERT0(vp->v_count);
2036 mutex_exit(&zp->z_lock);
2037 zfs_znode_delete(zp, tx);
2038 } else if (unlinked) {
2039 mutex_exit(&zp->z_lock);
2040 zfs_unlinked_add(zp, tx);
2042 vp->v_vflag |= VV_NOSYNC;
2047 if (flags & FIGNORECASE)
2049 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
2056 zfs_dirent_unlock(dl);
2063 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2064 zil_commit(zilog, 0);
2071 * Create a new directory and insert it into dvp using the name
2072 * provided. Return a pointer to the inserted directory.
2074 * IN: dvp - vnode of directory to add subdir to.
2075 * dirname - name of new directory.
2076 * vap - attributes of new directory.
2077 * cr - credentials of caller.
2078 * ct - caller context
2079 * flags - case flags
2080 * vsecp - ACL to be set
2082 * OUT: vpp - vnode of created directory.
2084 * RETURN: 0 on success, error code on failure.
2087 * dvp - ctime|mtime updated
2088 * vp - ctime|mtime|atime updated
2092 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
2093 caller_context_t *ct, int flags, vsecattr_t *vsecp)
2095 znode_t *zp, *dzp = VTOZ(dvp);
2096 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2105 gid_t gid = crgetgid(cr);
2106 zfs_acl_ids_t acl_ids;
2107 boolean_t fuid_dirtied;
2109 ASSERT(vap->va_type == VDIR);
2112 * If we have an ephemeral id, ACL, or XVATTR then
2113 * make sure file system is at proper version
2116 ksid = crgetsid(cr, KSID_OWNER);
2118 uid = ksid_getid(ksid);
2121 if (zfsvfs->z_use_fuids == B_FALSE &&
2122 (vsecp || (vap->va_mask & AT_XVATTR) ||
2123 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
2124 return (SET_ERROR(EINVAL));
2128 zilog = zfsvfs->z_log;
2130 if (dzp->z_pflags & ZFS_XATTR) {
2132 return (SET_ERROR(EINVAL));
2135 if (zfsvfs->z_utf8 && u8_validate(dirname,
2136 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2138 return (SET_ERROR(EILSEQ));
2140 if (flags & FIGNORECASE)
2143 if (vap->va_mask & AT_XVATTR) {
2144 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
2145 crgetuid(cr), cr, vap->va_type)) != 0) {
2151 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
2152 vsecp, &acl_ids)) != 0) {
2157 * First make sure the new directory doesn't exist.
2159 * Existence is checked first to make sure we don't return
2160 * EACCES instead of EEXIST which can cause some applications
2166 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
2168 zfs_acl_ids_free(&acl_ids);
2173 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
2174 zfs_acl_ids_free(&acl_ids);
2175 zfs_dirent_unlock(dl);
2180 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
2181 zfs_acl_ids_free(&acl_ids);
2182 zfs_dirent_unlock(dl);
2184 return (SET_ERROR(EDQUOT));
2188 * Add a new entry to the directory.
2190 tx = dmu_tx_create(zfsvfs->z_os);
2191 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2192 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2193 fuid_dirtied = zfsvfs->z_fuid_dirty;
2195 zfs_fuid_txhold(zfsvfs, tx);
2196 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2197 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2198 acl_ids.z_aclp->z_acl_bytes);
2201 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2202 ZFS_SA_BASE_ATTR_SIZE);
2204 error = dmu_tx_assign(tx, TXG_NOWAIT);
2206 zfs_dirent_unlock(dl);
2207 if (error == ERESTART) {
2212 zfs_acl_ids_free(&acl_ids);
2221 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2224 zfs_fuid_sync(zfsvfs, tx);
2227 * Now put new name in parent dir.
2229 (void) zfs_link_create(dl, zp, tx, ZNEW);
2233 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2234 if (flags & FIGNORECASE)
2236 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2237 acl_ids.z_fuidp, vap);
2239 zfs_acl_ids_free(&acl_ids);
2243 zfs_dirent_unlock(dl);
2245 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2246 zil_commit(zilog, 0);
2253 * Remove a directory subdir entry. If the current working
2254 * directory is the same as the subdir to be removed, the
2257 * IN: dvp - vnode of directory to remove from.
2258 * name - name of directory to be removed.
2259 * cwd - vnode of current working directory.
2260 * cr - credentials of caller.
2261 * ct - caller context
2262 * flags - case flags
2264 * RETURN: 0 on success, error code on failure.
2267 * dvp - ctime|mtime updated
2271 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2272 caller_context_t *ct, int flags)
2274 znode_t *dzp = VTOZ(dvp);
2277 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2286 zilog = zfsvfs->z_log;
2288 if (flags & FIGNORECASE)
2294 * Attempt to lock directory; fail if entry doesn't exist.
2296 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2304 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2308 if (vp->v_type != VDIR) {
2309 error = SET_ERROR(ENOTDIR);
2314 error = SET_ERROR(EINVAL);
2318 vnevent_rmdir(vp, dvp, name, ct);
2321 * Grab a lock on the directory to make sure that noone is
2322 * trying to add (or lookup) entries while we are removing it.
2324 rw_enter(&zp->z_name_lock, RW_WRITER);
2327 * Grab a lock on the parent pointer to make sure we play well
2328 * with the treewalk and directory rename code.
2330 rw_enter(&zp->z_parent_lock, RW_WRITER);
2332 tx = dmu_tx_create(zfsvfs->z_os);
2333 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2334 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2335 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2336 zfs_sa_upgrade_txholds(tx, zp);
2337 zfs_sa_upgrade_txholds(tx, dzp);
2338 error = dmu_tx_assign(tx, TXG_NOWAIT);
2340 rw_exit(&zp->z_parent_lock);
2341 rw_exit(&zp->z_name_lock);
2342 zfs_dirent_unlock(dl);
2344 if (error == ERESTART) {
2354 #ifdef FREEBSD_NAMECACHE
2358 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2361 uint64_t txtype = TX_RMDIR;
2362 if (flags & FIGNORECASE)
2364 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2369 rw_exit(&zp->z_parent_lock);
2370 rw_exit(&zp->z_name_lock);
2371 #ifdef FREEBSD_NAMECACHE
2375 zfs_dirent_unlock(dl);
2379 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2380 zil_commit(zilog, 0);
2387 * Read as many directory entries as will fit into the provided
2388 * buffer from the given directory cursor position (specified in
2389 * the uio structure).
2391 * IN: vp - vnode of directory to read.
2392 * uio - structure supplying read location, range info,
2393 * and return buffer.
2394 * cr - credentials of caller.
2395 * ct - caller context
2396 * flags - case flags
2398 * OUT: uio - updated offset and range, buffer filled.
2399 * eofp - set to true if end-of-file detected.
2401 * RETURN: 0 on success, error code on failure.
2404 * vp - atime updated
2406 * Note that the low 4 bits of the cookie returned by zap is always zero.
2407 * This allows us to use the low range for "special" directory entries:
2408 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2409 * we use the offset 2 for the '.zfs' directory.
2413 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies)
2415 znode_t *zp = VTOZ(vp);
2419 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2424 zap_attribute_t zap;
2425 uint_t bytes_wanted;
2426 uint64_t offset; /* must be unsigned; checks for < 1 */
2432 boolean_t check_sysattrs;
2435 u_long *cooks = NULL;
2441 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2442 &parent, sizeof (parent))) != 0) {
2448 * If we are not given an eof variable,
2455 * Check for valid iov_len.
2457 if (uio->uio_iov->iov_len <= 0) {
2459 return (SET_ERROR(EINVAL));
2463 * Quit if directory has been removed (posix)
2465 if ((*eofp = zp->z_unlinked) != 0) {
2472 offset = uio->uio_loffset;
2473 prefetch = zp->z_zn_prefetch;
2476 * Initialize the iterator cursor.
2480 * Start iteration from the beginning of the directory.
2482 zap_cursor_init(&zc, os, zp->z_id);
2485 * The offset is a serialized cursor.
2487 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2491 * Get space to change directory entries into fs independent format.
2493 iovp = uio->uio_iov;
2494 bytes_wanted = iovp->iov_len;
2495 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2496 bufsize = bytes_wanted;
2497 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2498 odp = (struct dirent64 *)outbuf;
2500 bufsize = bytes_wanted;
2502 odp = (struct dirent64 *)iovp->iov_base;
2504 eodp = (struct edirent *)odp;
2506 if (ncookies != NULL) {
2508 * Minimum entry size is dirent size and 1 byte for a file name.
2510 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1);
2511 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK);
2516 * If this VFS supports the system attribute view interface; and
2517 * we're looking at an extended attribute directory; and we care
2518 * about normalization conflicts on this vfs; then we must check
2519 * for normalization conflicts with the sysattr name space.
2522 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2523 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2524 (flags & V_RDDIR_ENTFLAGS);
2530 * Transform to file-system independent format
2533 while (outcount < bytes_wanted) {
2536 off64_t *next = NULL;
2539 * Special case `.', `..', and `.zfs'.
2542 (void) strcpy(zap.za_name, ".");
2543 zap.za_normalization_conflict = 0;
2546 } else if (offset == 1) {
2547 (void) strcpy(zap.za_name, "..");
2548 zap.za_normalization_conflict = 0;
2551 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2552 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2553 zap.za_normalization_conflict = 0;
2554 objnum = ZFSCTL_INO_ROOT;
2560 if (error = zap_cursor_retrieve(&zc, &zap)) {
2561 if ((*eofp = (error == ENOENT)) != 0)
2567 if (zap.za_integer_length != 8 ||
2568 zap.za_num_integers != 1) {
2569 cmn_err(CE_WARN, "zap_readdir: bad directory "
2570 "entry, obj = %lld, offset = %lld\n",
2571 (u_longlong_t)zp->z_id,
2572 (u_longlong_t)offset);
2573 error = SET_ERROR(ENXIO);
2577 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2579 * MacOS X can extract the object type here such as:
2580 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2582 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2584 if (check_sysattrs && !zap.za_normalization_conflict) {
2586 zap.za_normalization_conflict =
2587 xattr_sysattr_casechk(zap.za_name);
2589 panic("%s:%u: TODO", __func__, __LINE__);
2594 if (flags & V_RDDIR_ACCFILTER) {
2596 * If we have no access at all, don't include
2597 * this entry in the returned information
2600 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2602 if (!zfs_has_access(ezp, cr)) {
2609 if (flags & V_RDDIR_ENTFLAGS)
2610 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2612 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2615 * Will this entry fit in the buffer?
2617 if (outcount + reclen > bufsize) {
2619 * Did we manage to fit anything in the buffer?
2622 error = SET_ERROR(EINVAL);
2627 if (flags & V_RDDIR_ENTFLAGS) {
2629 * Add extended flag entry:
2631 eodp->ed_ino = objnum;
2632 eodp->ed_reclen = reclen;
2633 /* NOTE: ed_off is the offset for the *next* entry */
2634 next = &(eodp->ed_off);
2635 eodp->ed_eflags = zap.za_normalization_conflict ?
2636 ED_CASE_CONFLICT : 0;
2637 (void) strncpy(eodp->ed_name, zap.za_name,
2638 EDIRENT_NAMELEN(reclen));
2639 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2644 odp->d_ino = objnum;
2645 odp->d_reclen = reclen;
2646 odp->d_namlen = strlen(zap.za_name);
2647 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1);
2649 odp = (dirent64_t *)((intptr_t)odp + reclen);
2653 ASSERT(outcount <= bufsize);
2655 /* Prefetch znode */
2657 dmu_prefetch(os, objnum, 0, 0);
2661 * Move to the next entry, fill in the previous offset.
2663 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2664 zap_cursor_advance(&zc);
2665 offset = zap_cursor_serialize(&zc);
2670 if (cooks != NULL) {
2673 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks));
2676 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2678 /* Subtract unused cookies */
2679 if (ncookies != NULL)
2680 *ncookies -= ncooks;
2682 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2683 iovp->iov_base += outcount;
2684 iovp->iov_len -= outcount;
2685 uio->uio_resid -= outcount;
2686 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2688 * Reset the pointer.
2690 offset = uio->uio_loffset;
2694 zap_cursor_fini(&zc);
2695 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2696 kmem_free(outbuf, bufsize);
2698 if (error == ENOENT)
2701 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2703 uio->uio_loffset = offset;
2705 if (error != 0 && cookies != NULL) {
2706 free(*cookies, M_TEMP);
2713 ulong_t zfs_fsync_sync_cnt = 4;
2716 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2718 znode_t *zp = VTOZ(vp);
2719 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2721 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2723 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2726 zil_commit(zfsvfs->z_log, zp->z_id);
2734 * Get the requested file attributes and place them in the provided
2737 * IN: vp - vnode of file.
2738 * vap - va_mask identifies requested attributes.
2739 * If AT_XVATTR set, then optional attrs are requested
2740 * flags - ATTR_NOACLCHECK (CIFS server context)
2741 * cr - credentials of caller.
2742 * ct - caller context
2744 * OUT: vap - attribute values.
2746 * RETURN: 0 (always succeeds).
2750 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2751 caller_context_t *ct)
2753 znode_t *zp = VTOZ(vp);
2754 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2757 u_longlong_t nblocks;
2759 uint64_t mtime[2], ctime[2], crtime[2], rdev;
2760 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2761 xoptattr_t *xoap = NULL;
2762 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2763 sa_bulk_attr_t bulk[4];
2769 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2771 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2772 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2773 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
2774 if (vp->v_type == VBLK || vp->v_type == VCHR)
2775 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
2778 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2784 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2785 * Also, if we are the owner don't bother, since owner should
2786 * always be allowed to read basic attributes of file.
2788 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2789 (vap->va_uid != crgetuid(cr))) {
2790 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2798 * Return all attributes. It's cheaper to provide the answer
2799 * than to determine whether we were asked the question.
2802 mutex_enter(&zp->z_lock);
2803 vap->va_type = IFTOVT(zp->z_mode);
2804 vap->va_mode = zp->z_mode & ~S_IFMT;
2806 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2808 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
2810 vap->va_nodeid = zp->z_id;
2811 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2812 links = zp->z_links + 1;
2814 links = zp->z_links;
2815 vap->va_nlink = MIN(links, LINK_MAX); /* nlink_t limit! */
2816 vap->va_size = zp->z_size;
2818 vap->va_rdev = vp->v_rdev;
2820 if (vp->v_type == VBLK || vp->v_type == VCHR)
2821 vap->va_rdev = zfs_cmpldev(rdev);
2823 vap->va_seq = zp->z_seq;
2824 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */
2825 vap->va_filerev = zp->z_seq;
2828 * Add in any requested optional attributes and the create time.
2829 * Also set the corresponding bits in the returned attribute bitmap.
2831 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2832 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2834 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2835 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2838 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2839 xoap->xoa_readonly =
2840 ((zp->z_pflags & ZFS_READONLY) != 0);
2841 XVA_SET_RTN(xvap, XAT_READONLY);
2844 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2846 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2847 XVA_SET_RTN(xvap, XAT_SYSTEM);
2850 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2852 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2853 XVA_SET_RTN(xvap, XAT_HIDDEN);
2856 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2857 xoap->xoa_nounlink =
2858 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2859 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2862 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2863 xoap->xoa_immutable =
2864 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2865 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2868 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2869 xoap->xoa_appendonly =
2870 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2871 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2874 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2876 ((zp->z_pflags & ZFS_NODUMP) != 0);
2877 XVA_SET_RTN(xvap, XAT_NODUMP);
2880 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2882 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2883 XVA_SET_RTN(xvap, XAT_OPAQUE);
2886 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2887 xoap->xoa_av_quarantined =
2888 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2889 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2892 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2893 xoap->xoa_av_modified =
2894 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2895 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2898 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2899 vp->v_type == VREG) {
2900 zfs_sa_get_scanstamp(zp, xvap);
2903 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2906 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2907 times, sizeof (times));
2908 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2909 XVA_SET_RTN(xvap, XAT_CREATETIME);
2912 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2913 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2914 XVA_SET_RTN(xvap, XAT_REPARSE);
2916 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2917 xoap->xoa_generation = zp->z_gen;
2918 XVA_SET_RTN(xvap, XAT_GEN);
2921 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2923 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2924 XVA_SET_RTN(xvap, XAT_OFFLINE);
2927 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2929 ((zp->z_pflags & ZFS_SPARSE) != 0);
2930 XVA_SET_RTN(xvap, XAT_SPARSE);
2934 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2935 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2936 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2937 ZFS_TIME_DECODE(&vap->va_birthtime, crtime);
2939 mutex_exit(&zp->z_lock);
2941 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2942 vap->va_blksize = blksize;
2943 vap->va_bytes = nblocks << 9; /* nblocks * 512 */
2945 if (zp->z_blksz == 0) {
2947 * Block size hasn't been set; suggest maximal I/O transfers.
2949 vap->va_blksize = zfsvfs->z_max_blksz;
2957 * Set the file attributes to the values contained in the
2960 * IN: vp - vnode of file to be modified.
2961 * vap - new attribute values.
2962 * If AT_XVATTR set, then optional attrs are being set
2963 * flags - ATTR_UTIME set if non-default time values provided.
2964 * - ATTR_NOACLCHECK (CIFS context only).
2965 * cr - credentials of caller.
2966 * ct - caller context
2968 * RETURN: 0 on success, error code on failure.
2971 * vp - ctime updated, mtime updated if size changed.
2975 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2976 caller_context_t *ct)
2978 znode_t *zp = VTOZ(vp);
2979 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2984 uint_t mask = vap->va_mask;
2985 uint_t saved_mask = 0;
2986 uint64_t saved_mode;
2989 uint64_t new_uid, new_gid;
2991 uint64_t mtime[2], ctime[2];
2993 int need_policy = FALSE;
2995 zfs_fuid_info_t *fuidp = NULL;
2996 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2999 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
3000 boolean_t fuid_dirtied = B_FALSE;
3001 sa_bulk_attr_t bulk[7], xattr_bulk[7];
3002 int count = 0, xattr_count = 0;
3007 if (mask & AT_NOSET)
3008 return (SET_ERROR(EINVAL));
3013 zilog = zfsvfs->z_log;
3016 * Make sure that if we have ephemeral uid/gid or xvattr specified
3017 * that file system is at proper version level
3020 if (zfsvfs->z_use_fuids == B_FALSE &&
3021 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
3022 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
3023 (mask & AT_XVATTR))) {
3025 return (SET_ERROR(EINVAL));
3028 if (mask & AT_SIZE && vp->v_type == VDIR) {
3030 return (SET_ERROR(EISDIR));
3033 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
3035 return (SET_ERROR(EINVAL));
3039 * If this is an xvattr_t, then get a pointer to the structure of
3040 * optional attributes. If this is NULL, then we have a vattr_t.
3042 xoap = xva_getxoptattr(xvap);
3044 xva_init(&tmpxvattr);
3047 * Immutable files can only alter immutable bit and atime
3049 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
3050 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
3051 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
3053 return (SET_ERROR(EPERM));
3056 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
3058 return (SET_ERROR(EPERM));
3062 * Verify timestamps doesn't overflow 32 bits.
3063 * ZFS can handle large timestamps, but 32bit syscalls can't
3064 * handle times greater than 2039. This check should be removed
3065 * once large timestamps are fully supported.
3067 if (mask & (AT_ATIME | AT_MTIME)) {
3068 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
3069 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
3071 return (SET_ERROR(EOVERFLOW));
3079 /* Can this be moved to before the top label? */
3080 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
3082 return (SET_ERROR(EROFS));
3086 * First validate permissions
3089 if (mask & AT_SIZE) {
3091 * XXX - Note, we are not providing any open
3092 * mode flags here (like FNDELAY), so we may
3093 * block if there are locks present... this
3094 * should be addressed in openat().
3096 /* XXX - would it be OK to generate a log record here? */
3097 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
3104 if (mask & (AT_ATIME|AT_MTIME) ||
3105 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
3106 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
3107 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
3108 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
3109 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
3110 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
3111 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
3112 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
3116 if (mask & (AT_UID|AT_GID)) {
3117 int idmask = (mask & (AT_UID|AT_GID));
3122 * NOTE: even if a new mode is being set,
3123 * we may clear S_ISUID/S_ISGID bits.
3126 if (!(mask & AT_MODE))
3127 vap->va_mode = zp->z_mode;
3130 * Take ownership or chgrp to group we are a member of
3133 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
3134 take_group = (mask & AT_GID) &&
3135 zfs_groupmember(zfsvfs, vap->va_gid, cr);
3138 * If both AT_UID and AT_GID are set then take_owner and
3139 * take_group must both be set in order to allow taking
3142 * Otherwise, send the check through secpolicy_vnode_setattr()
3146 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
3147 ((idmask == AT_UID) && take_owner) ||
3148 ((idmask == AT_GID) && take_group)) {
3149 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
3150 skipaclchk, cr) == 0) {
3152 * Remove setuid/setgid for non-privileged users
3154 secpolicy_setid_clear(vap, vp, cr);
3155 trim_mask = (mask & (AT_UID|AT_GID));
3164 mutex_enter(&zp->z_lock);
3165 oldva.va_mode = zp->z_mode;
3166 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
3167 if (mask & AT_XVATTR) {
3169 * Update xvattr mask to include only those attributes
3170 * that are actually changing.
3172 * the bits will be restored prior to actually setting
3173 * the attributes so the caller thinks they were set.
3175 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
3176 if (xoap->xoa_appendonly !=
3177 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
3180 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
3181 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
3185 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
3186 if (xoap->xoa_nounlink !=
3187 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
3190 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
3191 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
3195 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
3196 if (xoap->xoa_immutable !=
3197 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
3200 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
3201 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
3205 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
3206 if (xoap->xoa_nodump !=
3207 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
3210 XVA_CLR_REQ(xvap, XAT_NODUMP);
3211 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
3215 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3216 if (xoap->xoa_av_modified !=
3217 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3220 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3221 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
3225 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3226 if ((vp->v_type != VREG &&
3227 xoap->xoa_av_quarantined) ||
3228 xoap->xoa_av_quarantined !=
3229 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3232 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3233 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
3237 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3238 mutex_exit(&zp->z_lock);
3240 return (SET_ERROR(EPERM));
3243 if (need_policy == FALSE &&
3244 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3245 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3250 mutex_exit(&zp->z_lock);
3252 if (mask & AT_MODE) {
3253 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3254 err = secpolicy_setid_setsticky_clear(vp, vap,
3260 trim_mask |= AT_MODE;
3268 * If trim_mask is set then take ownership
3269 * has been granted or write_acl is present and user
3270 * has the ability to modify mode. In that case remove
3271 * UID|GID and or MODE from mask so that
3272 * secpolicy_vnode_setattr() doesn't revoke it.
3276 saved_mask = vap->va_mask;
3277 vap->va_mask &= ~trim_mask;
3278 if (trim_mask & AT_MODE) {
3280 * Save the mode, as secpolicy_vnode_setattr()
3281 * will overwrite it with ova.va_mode.
3283 saved_mode = vap->va_mode;
3286 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3287 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3294 vap->va_mask |= saved_mask;
3295 if (trim_mask & AT_MODE) {
3297 * Recover the mode after
3298 * secpolicy_vnode_setattr().
3300 vap->va_mode = saved_mode;
3306 * secpolicy_vnode_setattr, or take ownership may have
3309 mask = vap->va_mask;
3311 if ((mask & (AT_UID | AT_GID))) {
3312 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3313 &xattr_obj, sizeof (xattr_obj));
3315 if (err == 0 && xattr_obj) {
3316 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3320 if (mask & AT_UID) {
3321 new_uid = zfs_fuid_create(zfsvfs,
3322 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3323 if (new_uid != zp->z_uid &&
3324 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3326 VN_RELE(ZTOV(attrzp));
3327 err = SET_ERROR(EDQUOT);
3332 if (mask & AT_GID) {
3333 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3334 cr, ZFS_GROUP, &fuidp);
3335 if (new_gid != zp->z_gid &&
3336 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3338 VN_RELE(ZTOV(attrzp));
3339 err = SET_ERROR(EDQUOT);
3344 tx = dmu_tx_create(zfsvfs->z_os);
3346 if (mask & AT_MODE) {
3347 uint64_t pmode = zp->z_mode;
3349 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3351 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3352 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3353 err = SET_ERROR(EPERM);
3357 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3360 mutex_enter(&zp->z_lock);
3361 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3363 * Are we upgrading ACL from old V0 format
3366 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3367 zfs_znode_acl_version(zp) ==
3368 ZFS_ACL_VERSION_INITIAL) {
3369 dmu_tx_hold_free(tx, acl_obj, 0,
3371 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3372 0, aclp->z_acl_bytes);
3374 dmu_tx_hold_write(tx, acl_obj, 0,
3377 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3378 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3379 0, aclp->z_acl_bytes);
3381 mutex_exit(&zp->z_lock);
3382 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3384 if ((mask & AT_XVATTR) &&
3385 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3386 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3388 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3392 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3395 fuid_dirtied = zfsvfs->z_fuid_dirty;
3397 zfs_fuid_txhold(zfsvfs, tx);
3399 zfs_sa_upgrade_txholds(tx, zp);
3401 err = dmu_tx_assign(tx, TXG_NOWAIT);
3403 if (err == ERESTART)
3410 * Set each attribute requested.
3411 * We group settings according to the locks they need to acquire.
3413 * Note: you cannot set ctime directly, although it will be
3414 * updated as a side-effect of calling this function.
3418 if (mask & (AT_UID|AT_GID|AT_MODE))
3419 mutex_enter(&zp->z_acl_lock);
3420 mutex_enter(&zp->z_lock);
3422 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3423 &zp->z_pflags, sizeof (zp->z_pflags));
3426 if (mask & (AT_UID|AT_GID|AT_MODE))
3427 mutex_enter(&attrzp->z_acl_lock);
3428 mutex_enter(&attrzp->z_lock);
3429 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3430 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3431 sizeof (attrzp->z_pflags));
3434 if (mask & (AT_UID|AT_GID)) {
3436 if (mask & AT_UID) {
3437 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3438 &new_uid, sizeof (new_uid));
3439 zp->z_uid = new_uid;
3441 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3442 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3444 attrzp->z_uid = new_uid;
3448 if (mask & AT_GID) {
3449 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3450 NULL, &new_gid, sizeof (new_gid));
3451 zp->z_gid = new_gid;
3453 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3454 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3456 attrzp->z_gid = new_gid;
3459 if (!(mask & AT_MODE)) {
3460 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3461 NULL, &new_mode, sizeof (new_mode));
3462 new_mode = zp->z_mode;
3464 err = zfs_acl_chown_setattr(zp);
3467 err = zfs_acl_chown_setattr(attrzp);
3472 if (mask & AT_MODE) {
3473 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3474 &new_mode, sizeof (new_mode));
3475 zp->z_mode = new_mode;
3476 ASSERT3U((uintptr_t)aclp, !=, 0);
3477 err = zfs_aclset_common(zp, aclp, cr, tx);
3479 if (zp->z_acl_cached)
3480 zfs_acl_free(zp->z_acl_cached);
3481 zp->z_acl_cached = aclp;
3486 if (mask & AT_ATIME) {
3487 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3488 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3489 &zp->z_atime, sizeof (zp->z_atime));
3492 if (mask & AT_MTIME) {
3493 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3494 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3495 mtime, sizeof (mtime));
3498 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3499 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3500 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3501 NULL, mtime, sizeof (mtime));
3502 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3503 &ctime, sizeof (ctime));
3504 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3506 } else if (mask != 0) {
3507 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3508 &ctime, sizeof (ctime));
3509 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3512 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3513 SA_ZPL_CTIME(zfsvfs), NULL,
3514 &ctime, sizeof (ctime));
3515 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3516 mtime, ctime, B_TRUE);
3520 * Do this after setting timestamps to prevent timestamp
3521 * update from toggling bit
3524 if (xoap && (mask & AT_XVATTR)) {
3527 * restore trimmed off masks
3528 * so that return masks can be set for caller.
3531 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3532 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3534 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3535 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3537 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3538 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3540 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3541 XVA_SET_REQ(xvap, XAT_NODUMP);
3543 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3544 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3546 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3547 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3550 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3551 ASSERT(vp->v_type == VREG);
3553 zfs_xvattr_set(zp, xvap, tx);
3557 zfs_fuid_sync(zfsvfs, tx);
3560 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3562 mutex_exit(&zp->z_lock);
3563 if (mask & (AT_UID|AT_GID|AT_MODE))
3564 mutex_exit(&zp->z_acl_lock);
3567 if (mask & (AT_UID|AT_GID|AT_MODE))
3568 mutex_exit(&attrzp->z_acl_lock);
3569 mutex_exit(&attrzp->z_lock);
3572 if (err == 0 && attrzp) {
3573 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3579 VN_RELE(ZTOV(attrzp));
3585 zfs_fuid_info_free(fuidp);
3591 if (err == ERESTART)
3594 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3599 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3600 zil_commit(zilog, 0);
3606 typedef struct zfs_zlock {
3607 krwlock_t *zl_rwlock; /* lock we acquired */
3608 znode_t *zl_znode; /* znode we held */
3609 struct zfs_zlock *zl_next; /* next in list */
3613 * Drop locks and release vnodes that were held by zfs_rename_lock().
3616 zfs_rename_unlock(zfs_zlock_t **zlpp)
3620 while ((zl = *zlpp) != NULL) {
3621 if (zl->zl_znode != NULL)
3622 VN_RELE(ZTOV(zl->zl_znode));
3623 rw_exit(zl->zl_rwlock);
3624 *zlpp = zl->zl_next;
3625 kmem_free(zl, sizeof (*zl));
3630 * Search back through the directory tree, using the ".." entries.
3631 * Lock each directory in the chain to prevent concurrent renames.
3632 * Fail any attempt to move a directory into one of its own descendants.
3633 * XXX - z_parent_lock can overlap with map or grow locks
3636 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3640 uint64_t rootid = zp->z_zfsvfs->z_root;
3641 uint64_t oidp = zp->z_id;
3642 krwlock_t *rwlp = &szp->z_parent_lock;
3643 krw_t rw = RW_WRITER;
3646 * First pass write-locks szp and compares to zp->z_id.
3647 * Later passes read-lock zp and compare to zp->z_parent.
3650 if (!rw_tryenter(rwlp, rw)) {
3652 * Another thread is renaming in this path.
3653 * Note that if we are a WRITER, we don't have any
3654 * parent_locks held yet.
3656 if (rw == RW_READER && zp->z_id > szp->z_id) {
3658 * Drop our locks and restart
3660 zfs_rename_unlock(&zl);
3664 rwlp = &szp->z_parent_lock;
3669 * Wait for other thread to drop its locks
3675 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3676 zl->zl_rwlock = rwlp;
3677 zl->zl_znode = NULL;
3678 zl->zl_next = *zlpp;
3681 if (oidp == szp->z_id) /* We're a descendant of szp */
3682 return (SET_ERROR(EINVAL));
3684 if (oidp == rootid) /* We've hit the top */
3687 if (rw == RW_READER) { /* i.e. not the first pass */
3688 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3693 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3694 &oidp, sizeof (oidp));
3695 rwlp = &zp->z_parent_lock;
3698 } while (zp->z_id != sdzp->z_id);
3704 * Move an entry from the provided source directory to the target
3705 * directory. Change the entry name as indicated.
3707 * IN: sdvp - Source directory containing the "old entry".
3708 * snm - Old entry name.
3709 * tdvp - Target directory to contain the "new entry".
3710 * tnm - New entry name.
3711 * cr - credentials of caller.
3712 * ct - caller context
3713 * flags - case flags
3715 * RETURN: 0 on success, error code on failure.
3718 * sdvp,tdvp - ctime|mtime updated
3722 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3723 caller_context_t *ct, int flags)
3725 znode_t *tdzp, *szp, *tzp;
3726 znode_t *sdzp = VTOZ(sdvp);
3727 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs;
3730 zfs_dirlock_t *sdl, *tdl;
3733 int cmp, serr, terr;
3738 ZFS_VERIFY_ZP(sdzp);
3739 zilog = zfsvfs->z_log;
3742 * Make sure we have the real vp for the target directory.
3744 if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3748 ZFS_VERIFY_ZP(tdzp);
3751 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3752 * ctldir appear to have the same v_vfsp.
3754 if (tdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) {
3756 return (SET_ERROR(EXDEV));
3759 if (zfsvfs->z_utf8 && u8_validate(tnm,
3760 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3762 return (SET_ERROR(EILSEQ));
3765 if (flags & FIGNORECASE)
3774 * This is to prevent the creation of links into attribute space
3775 * by renaming a linked file into/outof an attribute directory.
3776 * See the comment in zfs_link() for why this is considered bad.
3778 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3780 return (SET_ERROR(EINVAL));
3784 * Lock source and target directory entries. To prevent deadlock,
3785 * a lock ordering must be defined. We lock the directory with
3786 * the smallest object id first, or if it's a tie, the one with
3787 * the lexically first name.
3789 if (sdzp->z_id < tdzp->z_id) {
3791 } else if (sdzp->z_id > tdzp->z_id) {
3795 * First compare the two name arguments without
3796 * considering any case folding.
3798 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3800 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3801 ASSERT(error == 0 || !zfsvfs->z_utf8);
3804 * POSIX: "If the old argument and the new argument
3805 * both refer to links to the same existing file,
3806 * the rename() function shall return successfully
3807 * and perform no other action."
3813 * If the file system is case-folding, then we may
3814 * have some more checking to do. A case-folding file
3815 * system is either supporting mixed case sensitivity
3816 * access or is completely case-insensitive. Note
3817 * that the file system is always case preserving.
3819 * In mixed sensitivity mode case sensitive behavior
3820 * is the default. FIGNORECASE must be used to
3821 * explicitly request case insensitive behavior.
3823 * If the source and target names provided differ only
3824 * by case (e.g., a request to rename 'tim' to 'Tim'),
3825 * we will treat this as a special case in the
3826 * case-insensitive mode: as long as the source name
3827 * is an exact match, we will allow this to proceed as
3828 * a name-change request.
3830 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3831 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3832 flags & FIGNORECASE)) &&
3833 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3836 * case preserving rename request, require exact
3845 * If the source and destination directories are the same, we should
3846 * grab the z_name_lock of that directory only once.
3850 rw_enter(&sdzp->z_name_lock, RW_READER);
3854 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3855 ZEXISTS | zflg, NULL, NULL);
3856 terr = zfs_dirent_lock(&tdl,
3857 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3859 terr = zfs_dirent_lock(&tdl,
3860 tdzp, tnm, &tzp, zflg, NULL, NULL);
3861 serr = zfs_dirent_lock(&sdl,
3862 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3868 * Source entry invalid or not there.
3871 zfs_dirent_unlock(tdl);
3877 rw_exit(&sdzp->z_name_lock);
3880 * FreeBSD: In OpenSolaris they only check if rename source is
3881 * ".." here, because "." is handled in their lookup. This is
3882 * not the case for FreeBSD, so we check for "." explicitly.
3884 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0)
3885 serr = SET_ERROR(EINVAL);
3890 zfs_dirent_unlock(sdl);
3894 rw_exit(&sdzp->z_name_lock);
3896 if (strcmp(tnm, "..") == 0)
3897 terr = SET_ERROR(EINVAL);
3903 * Must have write access at the source to remove the old entry
3904 * and write access at the target to create the new entry.
3905 * Note that if target and source are the same, this can be
3906 * done in a single check.
3909 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3912 if (ZTOV(szp)->v_type == VDIR) {
3914 * Check to make sure rename is valid.
3915 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3917 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3922 * Does target exist?
3926 * Source and target must be the same type.
3928 if (ZTOV(szp)->v_type == VDIR) {
3929 if (ZTOV(tzp)->v_type != VDIR) {
3930 error = SET_ERROR(ENOTDIR);
3934 if (ZTOV(tzp)->v_type == VDIR) {
3935 error = SET_ERROR(EISDIR);
3940 * POSIX dictates that when the source and target
3941 * entries refer to the same file object, rename
3942 * must do nothing and exit without error.
3944 if (szp->z_id == tzp->z_id) {
3950 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3952 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3955 * notify the target directory if it is not the same
3956 * as source directory.
3959 vnevent_rename_dest_dir(tdvp, ct);
3962 tx = dmu_tx_create(zfsvfs->z_os);
3963 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3964 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3965 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3966 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3968 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3969 zfs_sa_upgrade_txholds(tx, tdzp);
3972 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3973 zfs_sa_upgrade_txholds(tx, tzp);
3976 zfs_sa_upgrade_txholds(tx, szp);
3977 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3978 error = dmu_tx_assign(tx, TXG_NOWAIT);
3981 zfs_rename_unlock(&zl);
3982 zfs_dirent_unlock(sdl);
3983 zfs_dirent_unlock(tdl);
3986 rw_exit(&sdzp->z_name_lock);
3991 if (error == ERESTART) {
4001 if (tzp) /* Attempt to remove the existing target */
4002 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
4005 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
4007 szp->z_pflags |= ZFS_AV_MODIFIED;
4009 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
4010 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
4013 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
4015 zfs_log_rename(zilog, tx, TX_RENAME |
4016 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
4017 sdl->dl_name, tdzp, tdl->dl_name, szp);
4020 * Update path information for the target vnode
4022 vn_renamepath(tdvp, ZTOV(szp), tnm,
4026 * At this point, we have successfully created
4027 * the target name, but have failed to remove
4028 * the source name. Since the create was done
4029 * with the ZRENAMING flag, there are
4030 * complications; for one, the link count is
4031 * wrong. The easiest way to deal with this
4032 * is to remove the newly created target, and
4033 * return the original error. This must
4034 * succeed; fortunately, it is very unlikely to
4035 * fail, since we just created it.
4037 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
4038 ZRENAMING, NULL), ==, 0);
4041 #ifdef FREEBSD_NAMECACHE
4045 cache_purge(ZTOV(szp));
4047 cache_purge(ZTOV(tzp));
4055 zfs_rename_unlock(&zl);
4057 zfs_dirent_unlock(sdl);
4058 zfs_dirent_unlock(tdl);
4061 rw_exit(&sdzp->z_name_lock);
4068 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4069 zil_commit(zilog, 0);
4077 * Insert the indicated symbolic reference entry into the directory.
4079 * IN: dvp - Directory to contain new symbolic link.
4080 * link - Name for new symlink entry.
4081 * vap - Attributes of new entry.
4082 * cr - credentials of caller.
4083 * ct - caller context
4084 * flags - case flags
4086 * RETURN: 0 on success, error code on failure.
4089 * dvp - ctime|mtime updated
4093 zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link,
4094 cred_t *cr, kthread_t *td)
4096 znode_t *zp, *dzp = VTOZ(dvp);
4099 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4101 uint64_t len = strlen(link);
4104 zfs_acl_ids_t acl_ids;
4105 boolean_t fuid_dirtied;
4106 uint64_t txtype = TX_SYMLINK;
4109 ASSERT(vap->va_type == VLNK);
4113 zilog = zfsvfs->z_log;
4115 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
4116 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4118 return (SET_ERROR(EILSEQ));
4120 if (flags & FIGNORECASE)
4123 if (len > MAXPATHLEN) {
4125 return (SET_ERROR(ENAMETOOLONG));
4128 if ((error = zfs_acl_ids_create(dzp, 0,
4129 vap, cr, NULL, &acl_ids)) != 0) {
4135 * Attempt to lock directory; fail if entry already exists.
4137 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
4139 zfs_acl_ids_free(&acl_ids);
4144 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4145 zfs_acl_ids_free(&acl_ids);
4146 zfs_dirent_unlock(dl);
4151 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
4152 zfs_acl_ids_free(&acl_ids);
4153 zfs_dirent_unlock(dl);
4155 return (SET_ERROR(EDQUOT));
4157 tx = dmu_tx_create(zfsvfs->z_os);
4158 fuid_dirtied = zfsvfs->z_fuid_dirty;
4159 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
4160 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4161 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
4162 ZFS_SA_BASE_ATTR_SIZE + len);
4163 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
4164 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
4165 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
4166 acl_ids.z_aclp->z_acl_bytes);
4169 zfs_fuid_txhold(zfsvfs, tx);
4170 error = dmu_tx_assign(tx, TXG_NOWAIT);
4172 zfs_dirent_unlock(dl);
4173 if (error == ERESTART) {
4178 zfs_acl_ids_free(&acl_ids);
4185 * Create a new object for the symlink.
4186 * for version 4 ZPL datsets the symlink will be an SA attribute
4188 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
4191 zfs_fuid_sync(zfsvfs, tx);
4193 mutex_enter(&zp->z_lock);
4195 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
4198 zfs_sa_symlink(zp, link, len, tx);
4199 mutex_exit(&zp->z_lock);
4202 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
4203 &zp->z_size, sizeof (zp->z_size), tx);
4205 * Insert the new object into the directory.
4207 (void) zfs_link_create(dl, zp, tx, ZNEW);
4209 if (flags & FIGNORECASE)
4211 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
4214 zfs_acl_ids_free(&acl_ids);
4218 zfs_dirent_unlock(dl);
4220 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4221 zil_commit(zilog, 0);
4228 * Return, in the buffer contained in the provided uio structure,
4229 * the symbolic path referred to by vp.
4231 * IN: vp - vnode of symbolic link.
4232 * uio - structure to contain the link path.
4233 * cr - credentials of caller.
4234 * ct - caller context
4236 * OUT: uio - structure containing the link path.
4238 * RETURN: 0 on success, error code on failure.
4241 * vp - atime updated
4245 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
4247 znode_t *zp = VTOZ(vp);
4248 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4254 mutex_enter(&zp->z_lock);
4256 error = sa_lookup_uio(zp->z_sa_hdl,
4257 SA_ZPL_SYMLINK(zfsvfs), uio);
4259 error = zfs_sa_readlink(zp, uio);
4260 mutex_exit(&zp->z_lock);
4262 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4269 * Insert a new entry into directory tdvp referencing svp.
4271 * IN: tdvp - Directory to contain new entry.
4272 * svp - vnode of new entry.
4273 * name - name of new entry.
4274 * cr - credentials of caller.
4275 * ct - caller context
4277 * RETURN: 0 on success, error code on failure.
4280 * tdvp - ctime|mtime updated
4281 * svp - ctime updated
4285 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
4286 caller_context_t *ct, int flags)
4288 znode_t *dzp = VTOZ(tdvp);
4290 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4300 ASSERT(tdvp->v_type == VDIR);
4304 zilog = zfsvfs->z_log;
4306 if (VOP_REALVP(svp, &realvp, ct) == 0)
4310 * POSIX dictates that we return EPERM here.
4311 * Better choices include ENOTSUP or EISDIR.
4313 if (svp->v_type == VDIR) {
4315 return (SET_ERROR(EPERM));
4322 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
4323 * ctldir appear to have the same v_vfsp.
4325 if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) {
4327 return (SET_ERROR(EXDEV));
4330 /* Prevent links to .zfs/shares files */
4332 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4333 &parent, sizeof (uint64_t))) != 0) {
4337 if (parent == zfsvfs->z_shares_dir) {
4339 return (SET_ERROR(EPERM));
4342 if (zfsvfs->z_utf8 && u8_validate(name,
4343 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4345 return (SET_ERROR(EILSEQ));
4347 if (flags & FIGNORECASE)
4351 * We do not support links between attributes and non-attributes
4352 * because of the potential security risk of creating links
4353 * into "normal" file space in order to circumvent restrictions
4354 * imposed in attribute space.
4356 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4358 return (SET_ERROR(EINVAL));
4362 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4363 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) {
4365 return (SET_ERROR(EPERM));
4368 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4375 * Attempt to lock directory; fail if entry already exists.
4377 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4383 tx = dmu_tx_create(zfsvfs->z_os);
4384 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4385 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4386 zfs_sa_upgrade_txholds(tx, szp);
4387 zfs_sa_upgrade_txholds(tx, dzp);
4388 error = dmu_tx_assign(tx, TXG_NOWAIT);
4390 zfs_dirent_unlock(dl);
4391 if (error == ERESTART) {
4401 error = zfs_link_create(dl, szp, tx, 0);
4404 uint64_t txtype = TX_LINK;
4405 if (flags & FIGNORECASE)
4407 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4412 zfs_dirent_unlock(dl);
4415 vnevent_link(svp, ct);
4418 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4419 zil_commit(zilog, 0);
4427 * zfs_null_putapage() is used when the file system has been force
4428 * unmounted. It just drops the pages.
4432 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4433 size_t *lenp, int flags, cred_t *cr)
4435 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4440 * Push a page out to disk, klustering if possible.
4442 * IN: vp - file to push page to.
4443 * pp - page to push.
4444 * flags - additional flags.
4445 * cr - credentials of caller.
4447 * OUT: offp - start of range pushed.
4448 * lenp - len of range pushed.
4450 * RETURN: 0 on success, error code on failure.
4452 * NOTE: callers must have locked the page to be pushed. On
4453 * exit, the page (and all other pages in the kluster) must be
4458 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4459 size_t *lenp, int flags, cred_t *cr)
4461 znode_t *zp = VTOZ(vp);
4462 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4464 u_offset_t off, koff;
4471 * If our blocksize is bigger than the page size, try to kluster
4472 * multiple pages so that we write a full block (thus avoiding
4473 * a read-modify-write).
4475 if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4476 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4477 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4478 ASSERT(koff <= zp->z_size);
4479 if (koff + klen > zp->z_size)
4480 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4481 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4483 ASSERT3U(btop(len), ==, btopr(len));
4486 * Can't push pages past end-of-file.
4488 if (off >= zp->z_size) {
4489 /* ignore all pages */
4492 } else if (off + len > zp->z_size) {
4493 int npages = btopr(zp->z_size - off);
4496 page_list_break(&pp, &trunc, npages);
4497 /* ignore pages past end of file */
4499 pvn_write_done(trunc, flags);
4500 len = zp->z_size - off;
4503 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4504 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4505 err = SET_ERROR(EDQUOT);
4509 tx = dmu_tx_create(zfsvfs->z_os);
4510 dmu_tx_hold_write(tx, zp->z_id, off, len);
4512 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4513 zfs_sa_upgrade_txholds(tx, zp);
4514 err = dmu_tx_assign(tx, TXG_NOWAIT);
4516 if (err == ERESTART) {
4525 if (zp->z_blksz <= PAGESIZE) {
4526 caddr_t va = zfs_map_page(pp, S_READ);
4527 ASSERT3U(len, <=, PAGESIZE);
4528 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4529 zfs_unmap_page(pp, va);
4531 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4535 uint64_t mtime[2], ctime[2];
4536 sa_bulk_attr_t bulk[3];
4539 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4541 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4543 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4545 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4547 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4552 pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4562 * Copy the portion of the file indicated from pages into the file.
4563 * The pages are stored in a page list attached to the files vnode.
4565 * IN: vp - vnode of file to push page data to.
4566 * off - position in file to put data.
4567 * len - amount of data to write.
4568 * flags - flags to control the operation.
4569 * cr - credentials of caller.
4570 * ct - caller context.
4572 * RETURN: 0 on success, error code on failure.
4575 * vp - ctime|mtime updated
4579 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4580 caller_context_t *ct)
4582 znode_t *zp = VTOZ(vp);
4583 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4595 * Align this request to the file block size in case we kluster.
4596 * XXX - this can result in pretty aggresive locking, which can
4597 * impact simultanious read/write access. One option might be
4598 * to break up long requests (len == 0) into block-by-block
4599 * operations to get narrower locking.
4601 blksz = zp->z_blksz;
4603 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4606 if (len > 0 && ISP2(blksz))
4607 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4613 * Search the entire vp list for pages >= io_off.
4615 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4616 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4619 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4621 if (off > zp->z_size) {
4622 /* past end of file */
4623 zfs_range_unlock(rl);
4628 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4630 for (off = io_off; io_off < off + len; io_off += io_len) {
4631 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4632 pp = page_lookup(vp, io_off,
4633 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4635 pp = page_lookup_nowait(vp, io_off,
4636 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4639 if (pp != NULL && pvn_getdirty(pp, flags)) {
4643 * Found a dirty page to push
4645 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4653 zfs_range_unlock(rl);
4654 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4655 zil_commit(zfsvfs->z_log, zp->z_id);
4663 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4665 znode_t *zp = VTOZ(vp);
4666 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4669 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4670 if (zp->z_sa_hdl == NULL) {
4672 * The fs has been unmounted, or we did a
4673 * suspend/resume and this file no longer exists.
4675 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4680 mutex_enter(&zp->z_lock);
4681 if (zp->z_unlinked) {
4683 * Fast path to recycle a vnode of a removed file.
4685 mutex_exit(&zp->z_lock);
4686 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4690 mutex_exit(&zp->z_lock);
4692 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4693 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4695 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4696 zfs_sa_upgrade_txholds(tx, zp);
4697 error = dmu_tx_assign(tx, TXG_WAIT);
4701 mutex_enter(&zp->z_lock);
4702 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4703 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4704 zp->z_atime_dirty = 0;
4705 mutex_exit(&zp->z_lock);
4709 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4714 * Bounds-check the seek operation.
4716 * IN: vp - vnode seeking within
4717 * ooff - old file offset
4718 * noffp - pointer to new file offset
4719 * ct - caller context
4721 * RETURN: 0 on success, EINVAL if new offset invalid.
4725 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4726 caller_context_t *ct)
4728 if (vp->v_type == VDIR)
4730 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4734 * Pre-filter the generic locking function to trap attempts to place
4735 * a mandatory lock on a memory mapped file.
4738 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4739 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4741 znode_t *zp = VTOZ(vp);
4742 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4748 * We are following the UFS semantics with respect to mapcnt
4749 * here: If we see that the file is mapped already, then we will
4750 * return an error, but we don't worry about races between this
4751 * function and zfs_map().
4753 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4755 return (SET_ERROR(EAGAIN));
4758 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4762 * If we can't find a page in the cache, we will create a new page
4763 * and fill it with file data. For efficiency, we may try to fill
4764 * multiple pages at once (klustering) to fill up the supplied page
4765 * list. Note that the pages to be filled are held with an exclusive
4766 * lock to prevent access by other threads while they are being filled.
4769 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4770 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4772 znode_t *zp = VTOZ(vp);
4773 page_t *pp, *cur_pp;
4774 objset_t *os = zp->z_zfsvfs->z_os;
4775 u_offset_t io_off, total;
4779 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4781 * We only have a single page, don't bother klustering
4785 pp = page_create_va(vp, io_off, io_len,
4786 PG_EXCL | PG_WAIT, seg, addr);
4789 * Try to find enough pages to fill the page list
4791 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4792 &io_len, off, plsz, 0);
4796 * The page already exists, nothing to do here.
4803 * Fill the pages in the kluster.
4806 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4809 ASSERT3U(io_off, ==, cur_pp->p_offset);
4810 va = zfs_map_page(cur_pp, S_WRITE);
4811 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4813 zfs_unmap_page(cur_pp, va);
4815 /* On error, toss the entire kluster */
4816 pvn_read_done(pp, B_ERROR);
4817 /* convert checksum errors into IO errors */
4819 err = SET_ERROR(EIO);
4822 cur_pp = cur_pp->p_next;
4826 * Fill in the page list array from the kluster starting
4827 * from the desired offset `off'.
4828 * NOTE: the page list will always be null terminated.
4830 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4831 ASSERT(pl == NULL || (*pl)->p_offset == off);
4837 * Return pointers to the pages for the file region [off, off + len]
4838 * in the pl array. If plsz is greater than len, this function may
4839 * also return page pointers from after the specified region
4840 * (i.e. the region [off, off + plsz]). These additional pages are
4841 * only returned if they are already in the cache, or were created as
4842 * part of a klustered read.
4844 * IN: vp - vnode of file to get data from.
4845 * off - position in file to get data from.
4846 * len - amount of data to retrieve.
4847 * plsz - length of provided page list.
4848 * seg - segment to obtain pages for.
4849 * addr - virtual address of fault.
4850 * rw - mode of created pages.
4851 * cr - credentials of caller.
4852 * ct - caller context.
4854 * OUT: protp - protection mode of created pages.
4855 * pl - list of pages created.
4857 * RETURN: 0 on success, error code on failure.
4860 * vp - atime updated
4864 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4865 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4866 enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4868 znode_t *zp = VTOZ(vp);
4869 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4873 /* we do our own caching, faultahead is unnecessary */
4876 else if (len > plsz)
4879 len = P2ROUNDUP(len, PAGESIZE);
4880 ASSERT(plsz >= len);
4889 * Loop through the requested range [off, off + len) looking
4890 * for pages. If we don't find a page, we will need to create
4891 * a new page and fill it with data from the file.
4894 if (*pl = page_lookup(vp, off, SE_SHARED))
4896 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4899 ASSERT3U((*pl)->p_offset, ==, off);
4903 ASSERT3U(len, >=, PAGESIZE);
4906 ASSERT3U(plsz, >=, PAGESIZE);
4913 * Fill out the page array with any pages already in the cache.
4916 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4923 * Release any pages we have previously locked.
4928 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4938 * Request a memory map for a section of a file. This code interacts
4939 * with common code and the VM system as follows:
4941 * - common code calls mmap(), which ends up in smmap_common()
4942 * - this calls VOP_MAP(), which takes you into (say) zfs
4943 * - zfs_map() calls as_map(), passing segvn_create() as the callback
4944 * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4945 * - zfs_addmap() updates z_mapcnt
4949 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4950 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4951 caller_context_t *ct)
4953 znode_t *zp = VTOZ(vp);
4954 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4955 segvn_crargs_t vn_a;
4961 if ((prot & PROT_WRITE) && (zp->z_pflags &
4962 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4964 return (SET_ERROR(EPERM));
4967 if ((prot & (PROT_READ | PROT_EXEC)) &&
4968 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4970 return (SET_ERROR(EACCES));
4973 if (vp->v_flag & VNOMAP) {
4975 return (SET_ERROR(ENOSYS));
4978 if (off < 0 || len > MAXOFFSET_T - off) {
4980 return (SET_ERROR(ENXIO));
4983 if (vp->v_type != VREG) {
4985 return (SET_ERROR(ENODEV));
4989 * If file is locked, disallow mapping.
4991 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
4993 return (SET_ERROR(EAGAIN));
4997 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
5005 vn_a.offset = (u_offset_t)off;
5006 vn_a.type = flags & MAP_TYPE;
5008 vn_a.maxprot = maxprot;
5011 vn_a.flags = flags & ~MAP_TYPE;
5013 vn_a.lgrp_mem_policy_flags = 0;
5015 error = as_map(as, *addrp, len, segvn_create, &vn_a);
5024 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5025 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
5026 caller_context_t *ct)
5028 uint64_t pages = btopr(len);
5030 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
5035 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
5036 * more accurate mtime for the associated file. Since we don't have a way of
5037 * detecting when the data was actually modified, we have to resort to
5038 * heuristics. If an explicit msync() is done, then we mark the mtime when the
5039 * last page is pushed. The problem occurs when the msync() call is omitted,
5040 * which by far the most common case:
5048 * putpage() via fsflush
5050 * If we wait until fsflush to come along, we can have a modification time that
5051 * is some arbitrary point in the future. In order to prevent this in the
5052 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
5057 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5058 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
5059 caller_context_t *ct)
5061 uint64_t pages = btopr(len);
5063 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
5064 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
5066 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
5067 vn_has_cached_data(vp))
5068 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
5074 * Free or allocate space in a file. Currently, this function only
5075 * supports the `F_FREESP' command. However, this command is somewhat
5076 * misnamed, as its functionality includes the ability to allocate as
5077 * well as free space.
5079 * IN: vp - vnode of file to free data in.
5080 * cmd - action to take (only F_FREESP supported).
5081 * bfp - section of file to free/alloc.
5082 * flag - current file open mode flags.
5083 * offset - current file offset.
5084 * cr - credentials of caller [UNUSED].
5085 * ct - caller context.
5087 * RETURN: 0 on success, error code on failure.
5090 * vp - ctime|mtime updated
5094 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
5095 offset_t offset, cred_t *cr, caller_context_t *ct)
5097 znode_t *zp = VTOZ(vp);
5098 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5105 if (cmd != F_FREESP) {
5107 return (SET_ERROR(EINVAL));
5110 if (error = convoff(vp, bfp, 0, offset)) {
5115 if (bfp->l_len < 0) {
5117 return (SET_ERROR(EINVAL));
5121 len = bfp->l_len; /* 0 means from off to end of file */
5123 error = zfs_freesp(zp, off, len, flag, TRUE);
5130 CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid));
5131 CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid));
5135 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
5137 znode_t *zp = VTOZ(vp);
5138 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5141 uint64_t object = zp->z_id;
5148 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
5149 &gen64, sizeof (uint64_t))) != 0) {
5154 gen = (uint32_t)gen64;
5156 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
5159 if (fidp->fid_len < size) {
5160 fidp->fid_len = size;
5162 return (SET_ERROR(ENOSPC));
5165 fidp->fid_len = size;
5168 zfid = (zfid_short_t *)fidp;
5170 zfid->zf_len = size;
5172 for (i = 0; i < sizeof (zfid->zf_object); i++)
5173 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
5175 /* Must have a non-zero generation number to distinguish from .zfs */
5178 for (i = 0; i < sizeof (zfid->zf_gen); i++)
5179 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
5181 if (size == LONG_FID_LEN) {
5182 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
5185 zlfid = (zfid_long_t *)fidp;
5187 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
5188 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
5190 /* XXX - this should be the generation number for the objset */
5191 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
5192 zlfid->zf_setgen[i] = 0;
5200 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
5201 caller_context_t *ct)
5213 case _PC_FILESIZEBITS:
5217 case _PC_XATTR_EXISTS:
5219 zfsvfs = zp->z_zfsvfs;
5223 error = zfs_dirent_lock(&dl, zp, "", &xzp,
5224 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
5226 zfs_dirent_unlock(dl);
5227 if (!zfs_dirempty(xzp))
5230 } else if (error == ENOENT) {
5232 * If there aren't extended attributes, it's the
5233 * same as having zero of them.
5240 case _PC_SATTR_ENABLED:
5241 case _PC_SATTR_EXISTS:
5242 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
5243 (vp->v_type == VREG || vp->v_type == VDIR);
5246 case _PC_ACCESS_FILTERING:
5247 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
5251 case _PC_ACL_ENABLED:
5252 *valp = _ACL_ACE_ENABLED;
5255 case _PC_MIN_HOLE_SIZE:
5256 *valp = (int)SPA_MINBLOCKSIZE;
5259 case _PC_TIMESTAMP_RESOLUTION:
5260 /* nanosecond timestamp resolution */
5264 case _PC_ACL_EXTENDED:
5272 case _PC_ACL_PATH_MAX:
5273 *valp = ACL_MAX_ENTRIES;
5277 return (EOPNOTSUPP);
5283 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5284 caller_context_t *ct)
5286 znode_t *zp = VTOZ(vp);
5287 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5289 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5293 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5301 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5302 caller_context_t *ct)
5304 znode_t *zp = VTOZ(vp);
5305 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5307 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5308 zilog_t *zilog = zfsvfs->z_log;
5313 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5315 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5316 zil_commit(zilog, 0);
5324 * The smallest read we may consider to loan out an arcbuf.
5325 * This must be a power of 2.
5327 int zcr_blksz_min = (1 << 10); /* 1K */
5329 * If set to less than the file block size, allow loaning out of an
5330 * arcbuf for a partial block read. This must be a power of 2.
5332 int zcr_blksz_max = (1 << 17); /* 128K */
5336 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5337 caller_context_t *ct)
5339 znode_t *zp = VTOZ(vp);
5340 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5341 int max_blksz = zfsvfs->z_max_blksz;
5342 uio_t *uio = &xuio->xu_uio;
5343 ssize_t size = uio->uio_resid;
5344 offset_t offset = uio->uio_loffset;
5349 int preamble, postamble;
5351 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5352 return (SET_ERROR(EINVAL));
5359 * Loan out an arc_buf for write if write size is bigger than
5360 * max_blksz, and the file's block size is also max_blksz.
5363 if (size < blksz || zp->z_blksz != blksz) {
5365 return (SET_ERROR(EINVAL));
5368 * Caller requests buffers for write before knowing where the
5369 * write offset might be (e.g. NFS TCP write).
5374 preamble = P2PHASE(offset, blksz);
5376 preamble = blksz - preamble;
5381 postamble = P2PHASE(size, blksz);
5384 fullblk = size / blksz;
5385 (void) dmu_xuio_init(xuio,
5386 (preamble != 0) + fullblk + (postamble != 0));
5387 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5388 int, postamble, int,
5389 (preamble != 0) + fullblk + (postamble != 0));
5392 * Have to fix iov base/len for partial buffers. They
5393 * currently represent full arc_buf's.
5396 /* data begins in the middle of the arc_buf */
5397 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5400 (void) dmu_xuio_add(xuio, abuf,
5401 blksz - preamble, preamble);
5404 for (i = 0; i < fullblk; i++) {
5405 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5408 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5412 /* data ends in the middle of the arc_buf */
5413 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5416 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5421 * Loan out an arc_buf for read if the read size is larger than
5422 * the current file block size. Block alignment is not
5423 * considered. Partial arc_buf will be loaned out for read.
5425 blksz = zp->z_blksz;
5426 if (blksz < zcr_blksz_min)
5427 blksz = zcr_blksz_min;
5428 if (blksz > zcr_blksz_max)
5429 blksz = zcr_blksz_max;
5430 /* avoid potential complexity of dealing with it */
5431 if (blksz > max_blksz) {
5433 return (SET_ERROR(EINVAL));
5436 maxsize = zp->z_size - uio->uio_loffset;
5440 if (size < blksz || vn_has_cached_data(vp)) {
5442 return (SET_ERROR(EINVAL));
5447 return (SET_ERROR(EINVAL));
5450 uio->uio_extflg = UIO_XUIO;
5451 XUIO_XUZC_RW(xuio) = ioflag;
5458 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5462 int ioflag = XUIO_XUZC_RW(xuio);
5464 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5466 i = dmu_xuio_cnt(xuio);
5468 abuf = dmu_xuio_arcbuf(xuio, i);
5470 * if abuf == NULL, it must be a write buffer
5471 * that has been returned in zfs_write().
5474 dmu_return_arcbuf(abuf);
5475 ASSERT(abuf || ioflag == UIO_WRITE);
5478 dmu_xuio_fini(xuio);
5483 * Predeclare these here so that the compiler assumes that
5484 * this is an "old style" function declaration that does
5485 * not include arguments => we won't get type mismatch errors
5486 * in the initializations that follow.
5488 static int zfs_inval();
5489 static int zfs_isdir();
5494 return (SET_ERROR(EINVAL));
5500 return (SET_ERROR(EISDIR));
5503 * Directory vnode operations template
5505 vnodeops_t *zfs_dvnodeops;
5506 const fs_operation_def_t zfs_dvnodeops_template[] = {
5507 VOPNAME_OPEN, { .vop_open = zfs_open },
5508 VOPNAME_CLOSE, { .vop_close = zfs_close },
5509 VOPNAME_READ, { .error = zfs_isdir },
5510 VOPNAME_WRITE, { .error = zfs_isdir },
5511 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5512 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5513 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5514 VOPNAME_ACCESS, { .vop_access = zfs_access },
5515 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5516 VOPNAME_CREATE, { .vop_create = zfs_create },
5517 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5518 VOPNAME_LINK, { .vop_link = zfs_link },
5519 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5520 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir },
5521 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5522 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5523 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink },
5524 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5525 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5526 VOPNAME_FID, { .vop_fid = zfs_fid },
5527 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5528 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5529 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5530 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5531 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5536 * Regular file vnode operations template
5538 vnodeops_t *zfs_fvnodeops;
5539 const fs_operation_def_t zfs_fvnodeops_template[] = {
5540 VOPNAME_OPEN, { .vop_open = zfs_open },
5541 VOPNAME_CLOSE, { .vop_close = zfs_close },
5542 VOPNAME_READ, { .vop_read = zfs_read },
5543 VOPNAME_WRITE, { .vop_write = zfs_write },
5544 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5545 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5546 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5547 VOPNAME_ACCESS, { .vop_access = zfs_access },
5548 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5549 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5550 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5551 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5552 VOPNAME_FID, { .vop_fid = zfs_fid },
5553 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5554 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock },
5555 VOPNAME_SPACE, { .vop_space = zfs_space },
5556 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage },
5557 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage },
5558 VOPNAME_MAP, { .vop_map = zfs_map },
5559 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap },
5560 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap },
5561 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5562 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5563 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5564 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5565 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf },
5566 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf },
5571 * Symbolic link vnode operations template
5573 vnodeops_t *zfs_symvnodeops;
5574 const fs_operation_def_t zfs_symvnodeops_template[] = {
5575 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5576 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5577 VOPNAME_ACCESS, { .vop_access = zfs_access },
5578 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5579 VOPNAME_READLINK, { .vop_readlink = zfs_readlink },
5580 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5581 VOPNAME_FID, { .vop_fid = zfs_fid },
5582 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5583 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5588 * special share hidden files vnode operations template
5590 vnodeops_t *zfs_sharevnodeops;
5591 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5592 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5593 VOPNAME_ACCESS, { .vop_access = zfs_access },
5594 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5595 VOPNAME_FID, { .vop_fid = zfs_fid },
5596 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5597 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5598 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5599 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5604 * Extended attribute directory vnode operations template
5606 * This template is identical to the directory vnodes
5607 * operation template except for restricted operations:
5611 * Note that there are other restrictions embedded in:
5612 * zfs_create() - restrict type to VREG
5613 * zfs_link() - no links into/out of attribute space
5614 * zfs_rename() - no moves into/out of attribute space
5616 vnodeops_t *zfs_xdvnodeops;
5617 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5618 VOPNAME_OPEN, { .vop_open = zfs_open },
5619 VOPNAME_CLOSE, { .vop_close = zfs_close },
5620 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5621 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5622 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5623 VOPNAME_ACCESS, { .vop_access = zfs_access },
5624 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5625 VOPNAME_CREATE, { .vop_create = zfs_create },
5626 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5627 VOPNAME_LINK, { .vop_link = zfs_link },
5628 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5629 VOPNAME_MKDIR, { .error = zfs_inval },
5630 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5631 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5632 VOPNAME_SYMLINK, { .error = zfs_inval },
5633 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5634 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5635 VOPNAME_FID, { .vop_fid = zfs_fid },
5636 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5637 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5638 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5639 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5640 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5645 * Error vnode operations template
5647 vnodeops_t *zfs_evnodeops;
5648 const fs_operation_def_t zfs_evnodeops_template[] = {
5649 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5650 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5656 ioflags(int ioflags)
5660 if (ioflags & IO_APPEND)
5662 if (ioflags & IO_NDELAY)
5664 if (ioflags & IO_SYNC)
5665 flags |= (FSYNC | FDSYNC | FRSYNC);
5671 zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage)
5673 znode_t *zp = VTOZ(vp);
5674 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5675 objset_t *os = zp->z_zfsvfs->z_os;
5676 vm_page_t mfirst, mlast, mreq;
5680 off_t startoff, endoff;
5682 vm_pindex_t reqstart, reqend;
5683 int pcount, lsize, reqsize, size;
5688 pcount = OFF_TO_IDX(round_page(count));
5690 object = mreq->object;
5693 KASSERT(vp->v_object == object, ("mismatching object"));
5695 if (pcount > 1 && zp->z_blksz > PAGESIZE) {
5696 startoff = rounddown(IDX_TO_OFF(mreq->pindex), zp->z_blksz);
5697 reqstart = OFF_TO_IDX(round_page(startoff));
5698 if (reqstart < m[0]->pindex)
5701 reqstart = reqstart - m[0]->pindex;
5702 endoff = roundup(IDX_TO_OFF(mreq->pindex) + PAGE_SIZE,
5704 reqend = OFF_TO_IDX(trunc_page(endoff)) - 1;
5705 if (reqend > m[pcount - 1]->pindex)
5706 reqend = m[pcount - 1]->pindex;
5707 reqsize = reqend - m[reqstart]->pindex + 1;
5708 KASSERT(reqstart <= reqpage && reqpage < reqstart + reqsize,
5709 ("reqpage beyond [reqstart, reqstart + reqsize[ bounds"));
5714 mfirst = m[reqstart];
5715 mlast = m[reqstart + reqsize - 1];
5717 zfs_vmobject_wlock(object);
5719 for (i = 0; i < reqstart; i++) {
5722 vm_page_unlock(m[i]);
5724 for (i = reqstart + reqsize; i < pcount; i++) {
5727 vm_page_unlock(m[i]);
5730 if (mreq->valid && reqsize == 1) {
5731 if (mreq->valid != VM_PAGE_BITS_ALL)
5732 vm_page_zero_invalid(mreq, TRUE);
5733 zfs_vmobject_wunlock(object);
5735 return (zfs_vm_pagerret_ok);
5738 PCPU_INC(cnt.v_vnodein);
5739 PCPU_ADD(cnt.v_vnodepgsin, reqsize);
5741 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) {
5742 for (i = reqstart; i < reqstart + reqsize; i++) {
5746 vm_page_unlock(m[i]);
5749 zfs_vmobject_wunlock(object);
5751 return (zfs_vm_pagerret_bad);
5755 if (IDX_TO_OFF(mlast->pindex) + lsize > object->un_pager.vnp.vnp_size)
5756 lsize = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mlast->pindex);
5758 zfs_vmobject_wunlock(object);
5760 for (i = reqstart; i < reqstart + reqsize; i++) {
5762 if (i == (reqstart + reqsize - 1))
5764 va = zfs_map_page(m[i], &sf);
5765 error = dmu_read(os, zp->z_id, IDX_TO_OFF(m[i]->pindex),
5766 size, va, DMU_READ_PREFETCH);
5767 if (size != PAGE_SIZE)
5768 bzero(va + size, PAGE_SIZE - size);
5774 zfs_vmobject_wlock(object);
5776 for (i = reqstart; i < reqstart + reqsize; i++) {
5778 m[i]->valid = VM_PAGE_BITS_ALL;
5779 KASSERT(m[i]->dirty == 0, ("zfs_getpages: page %p is dirty", m[i]));
5781 vm_page_readahead_finish(m[i]);
5784 zfs_vmobject_wunlock(object);
5786 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
5788 return (error ? zfs_vm_pagerret_error : zfs_vm_pagerret_ok);
5792 zfs_freebsd_getpages(ap)
5793 struct vop_getpages_args /* {
5798 vm_ooffset_t a_offset;
5802 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage));
5806 zfs_freebsd_bmap(ap)
5807 struct vop_bmap_args /* {
5810 struct bufobj **a_bop;
5817 if (ap->a_bop != NULL)
5818 *ap->a_bop = &ap->a_vp->v_bufobj;
5819 if (ap->a_bnp != NULL)
5820 *ap->a_bnp = ap->a_bn;
5821 if (ap->a_runp != NULL)
5823 if (ap->a_runb != NULL)
5830 zfs_freebsd_open(ap)
5831 struct vop_open_args /* {
5834 struct ucred *a_cred;
5835 struct thread *a_td;
5838 vnode_t *vp = ap->a_vp;
5839 znode_t *zp = VTOZ(vp);
5842 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL);
5844 vnode_create_vobject(vp, zp->z_size, ap->a_td);
5849 zfs_freebsd_close(ap)
5850 struct vop_close_args /* {
5853 struct ucred *a_cred;
5854 struct thread *a_td;
5858 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL));
5862 zfs_freebsd_ioctl(ap)
5863 struct vop_ioctl_args /* {
5873 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data,
5874 ap->a_fflag, ap->a_cred, NULL, NULL));
5878 zfs_freebsd_read(ap)
5879 struct vop_read_args /* {
5883 struct ucred *a_cred;
5887 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5892 zfs_freebsd_write(ap)
5893 struct vop_write_args /* {
5897 struct ucred *a_cred;
5901 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5906 zfs_freebsd_access(ap)
5907 struct vop_access_args /* {
5909 accmode_t a_accmode;
5910 struct ucred *a_cred;
5911 struct thread *a_td;
5914 vnode_t *vp = ap->a_vp;
5915 znode_t *zp = VTOZ(vp);
5920 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND,
5922 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND);
5924 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL);
5927 * VADMIN has to be handled by vaccess().
5930 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND);
5932 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
5933 zp->z_gid, accmode, ap->a_cred, NULL);
5938 * For VEXEC, ensure that at least one execute bit is set for
5941 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR &&
5942 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
5950 zfs_freebsd_lookup(ap)
5951 struct vop_lookup_args /* {
5952 struct vnode *a_dvp;
5953 struct vnode **a_vpp;
5954 struct componentname *a_cnp;
5957 struct componentname *cnp = ap->a_cnp;
5958 char nm[NAME_MAX + 1];
5960 ASSERT(cnp->cn_namelen < sizeof(nm));
5961 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm)));
5963 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop,
5964 cnp->cn_cred, cnp->cn_thread, 0));
5968 zfs_freebsd_create(ap)
5969 struct vop_create_args /* {
5970 struct vnode *a_dvp;
5971 struct vnode **a_vpp;
5972 struct componentname *a_cnp;
5973 struct vattr *a_vap;
5976 struct componentname *cnp = ap->a_cnp;
5977 vattr_t *vap = ap->a_vap;
5980 ASSERT(cnp->cn_flags & SAVENAME);
5982 vattr_init_mask(vap);
5983 mode = vap->va_mode & ALLPERMS;
5985 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode,
5986 ap->a_vpp, cnp->cn_cred, cnp->cn_thread));
5990 zfs_freebsd_remove(ap)
5991 struct vop_remove_args /* {
5992 struct vnode *a_dvp;
5994 struct componentname *a_cnp;
5998 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6000 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr,
6001 ap->a_cnp->cn_cred, NULL, 0));
6005 zfs_freebsd_mkdir(ap)
6006 struct vop_mkdir_args /* {
6007 struct vnode *a_dvp;
6008 struct vnode **a_vpp;
6009 struct componentname *a_cnp;
6010 struct vattr *a_vap;
6013 vattr_t *vap = ap->a_vap;
6015 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6017 vattr_init_mask(vap);
6019 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp,
6020 ap->a_cnp->cn_cred, NULL, 0, NULL));
6024 zfs_freebsd_rmdir(ap)
6025 struct vop_rmdir_args /* {
6026 struct vnode *a_dvp;
6028 struct componentname *a_cnp;
6031 struct componentname *cnp = ap->a_cnp;
6033 ASSERT(cnp->cn_flags & SAVENAME);
6035 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0));
6039 zfs_freebsd_readdir(ap)
6040 struct vop_readdir_args /* {
6043 struct ucred *a_cred;
6050 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag,
6051 ap->a_ncookies, ap->a_cookies));
6055 zfs_freebsd_fsync(ap)
6056 struct vop_fsync_args /* {
6059 struct thread *a_td;
6064 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL));
6068 zfs_freebsd_getattr(ap)
6069 struct vop_getattr_args /* {
6071 struct vattr *a_vap;
6072 struct ucred *a_cred;
6075 vattr_t *vap = ap->a_vap;
6081 xvap.xva_vattr = *vap;
6082 xvap.xva_vattr.va_mask |= AT_XVATTR;
6084 /* Convert chflags into ZFS-type flags. */
6085 /* XXX: what about SF_SETTABLE?. */
6086 XVA_SET_REQ(&xvap, XAT_IMMUTABLE);
6087 XVA_SET_REQ(&xvap, XAT_APPENDONLY);
6088 XVA_SET_REQ(&xvap, XAT_NOUNLINK);
6089 XVA_SET_REQ(&xvap, XAT_NODUMP);
6090 XVA_SET_REQ(&xvap, XAT_READONLY);
6091 XVA_SET_REQ(&xvap, XAT_ARCHIVE);
6092 XVA_SET_REQ(&xvap, XAT_SYSTEM);
6093 XVA_SET_REQ(&xvap, XAT_HIDDEN);
6094 XVA_SET_REQ(&xvap, XAT_REPARSE);
6095 XVA_SET_REQ(&xvap, XAT_OFFLINE);
6096 XVA_SET_REQ(&xvap, XAT_SPARSE);
6098 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL);
6102 /* Convert ZFS xattr into chflags. */
6103 #define FLAG_CHECK(fflag, xflag, xfield) do { \
6104 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \
6105 fflags |= (fflag); \
6107 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE,
6108 xvap.xva_xoptattrs.xoa_immutable);
6109 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY,
6110 xvap.xva_xoptattrs.xoa_appendonly);
6111 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK,
6112 xvap.xva_xoptattrs.xoa_nounlink);
6113 FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE,
6114 xvap.xva_xoptattrs.xoa_archive);
6115 FLAG_CHECK(UF_NODUMP, XAT_NODUMP,
6116 xvap.xva_xoptattrs.xoa_nodump);
6117 FLAG_CHECK(UF_READONLY, XAT_READONLY,
6118 xvap.xva_xoptattrs.xoa_readonly);
6119 FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM,
6120 xvap.xva_xoptattrs.xoa_system);
6121 FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN,
6122 xvap.xva_xoptattrs.xoa_hidden);
6123 FLAG_CHECK(UF_REPARSE, XAT_REPARSE,
6124 xvap.xva_xoptattrs.xoa_reparse);
6125 FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE,
6126 xvap.xva_xoptattrs.xoa_offline);
6127 FLAG_CHECK(UF_SPARSE, XAT_SPARSE,
6128 xvap.xva_xoptattrs.xoa_sparse);
6131 *vap = xvap.xva_vattr;
6132 vap->va_flags = fflags;
6137 zfs_freebsd_setattr(ap)
6138 struct vop_setattr_args /* {
6140 struct vattr *a_vap;
6141 struct ucred *a_cred;
6144 vnode_t *vp = ap->a_vp;
6145 vattr_t *vap = ap->a_vap;
6146 cred_t *cred = ap->a_cred;
6151 vattr_init_mask(vap);
6152 vap->va_mask &= ~AT_NOSET;
6155 xvap.xva_vattr = *vap;
6157 zflags = VTOZ(vp)->z_pflags;
6159 if (vap->va_flags != VNOVAL) {
6160 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs;
6163 if (zfsvfs->z_use_fuids == B_FALSE)
6164 return (EOPNOTSUPP);
6166 fflags = vap->va_flags;
6169 * We need to figure out whether it makes sense to allow
6170 * UF_REPARSE through, since we don't really have other
6171 * facilities to handle reparse points and zfs_setattr()
6172 * doesn't currently allow setting that attribute anyway.
6174 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE|
6175 UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE|
6176 UF_OFFLINE|UF_SPARSE)) != 0)
6177 return (EOPNOTSUPP);
6179 * Unprivileged processes are not permitted to unset system
6180 * flags, or modify flags if any system flags are set.
6181 * Privileged non-jail processes may not modify system flags
6182 * if securelevel > 0 and any existing system flags are set.
6183 * Privileged jail processes behave like privileged non-jail
6184 * processes if the security.jail.chflags_allowed sysctl is
6185 * is non-zero; otherwise, they behave like unprivileged
6188 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 ||
6189 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) {
6191 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6192 error = securelevel_gt(cred, 0);
6198 * Callers may only modify the file flags on objects they
6199 * have VADMIN rights for.
6201 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0)
6204 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6208 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) {
6213 #define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \
6214 if (((fflags & (fflag)) && !(zflags & (zflag))) || \
6215 ((zflags & (zflag)) && !(fflags & (fflag)))) { \
6216 XVA_SET_REQ(&xvap, (xflag)); \
6217 (xfield) = ((fflags & (fflag)) != 0); \
6220 /* Convert chflags into ZFS-type flags. */
6221 /* XXX: what about SF_SETTABLE?. */
6222 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE,
6223 xvap.xva_xoptattrs.xoa_immutable);
6224 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY,
6225 xvap.xva_xoptattrs.xoa_appendonly);
6226 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK,
6227 xvap.xva_xoptattrs.xoa_nounlink);
6228 FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE,
6229 xvap.xva_xoptattrs.xoa_archive);
6230 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP,
6231 xvap.xva_xoptattrs.xoa_nodump);
6232 FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY,
6233 xvap.xva_xoptattrs.xoa_readonly);
6234 FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM,
6235 xvap.xva_xoptattrs.xoa_system);
6236 FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN,
6237 xvap.xva_xoptattrs.xoa_hidden);
6238 FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE,
6239 xvap.xva_xoptattrs.xoa_hidden);
6240 FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE,
6241 xvap.xva_xoptattrs.xoa_offline);
6242 FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE,
6243 xvap.xva_xoptattrs.xoa_sparse);
6246 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL));
6250 zfs_freebsd_rename(ap)
6251 struct vop_rename_args /* {
6252 struct vnode *a_fdvp;
6253 struct vnode *a_fvp;
6254 struct componentname *a_fcnp;
6255 struct vnode *a_tdvp;
6256 struct vnode *a_tvp;
6257 struct componentname *a_tcnp;
6260 vnode_t *fdvp = ap->a_fdvp;
6261 vnode_t *fvp = ap->a_fvp;
6262 vnode_t *tdvp = ap->a_tdvp;
6263 vnode_t *tvp = ap->a_tvp;
6266 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART));
6267 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART));
6270 * Check for cross-device rename.
6272 if ((fdvp->v_mount != tdvp->v_mount) ||
6273 (tvp && (fdvp->v_mount != tvp->v_mount)))
6276 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp,
6277 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0);
6291 zfs_freebsd_symlink(ap)
6292 struct vop_symlink_args /* {
6293 struct vnode *a_dvp;
6294 struct vnode **a_vpp;
6295 struct componentname *a_cnp;
6296 struct vattr *a_vap;
6300 struct componentname *cnp = ap->a_cnp;
6301 vattr_t *vap = ap->a_vap;
6303 ASSERT(cnp->cn_flags & SAVENAME);
6305 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */
6306 vattr_init_mask(vap);
6308 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap,
6309 ap->a_target, cnp->cn_cred, cnp->cn_thread));
6313 zfs_freebsd_readlink(ap)
6314 struct vop_readlink_args /* {
6317 struct ucred *a_cred;
6321 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL));
6325 zfs_freebsd_link(ap)
6326 struct vop_link_args /* {
6327 struct vnode *a_tdvp;
6329 struct componentname *a_cnp;
6332 struct componentname *cnp = ap->a_cnp;
6333 vnode_t *vp = ap->a_vp;
6334 vnode_t *tdvp = ap->a_tdvp;
6336 if (tdvp->v_mount != vp->v_mount)
6339 ASSERT(cnp->cn_flags & SAVENAME);
6341 return (zfs_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0));
6345 zfs_freebsd_inactive(ap)
6346 struct vop_inactive_args /* {
6348 struct thread *a_td;
6351 vnode_t *vp = ap->a_vp;
6353 zfs_inactive(vp, ap->a_td->td_ucred, NULL);
6358 zfs_freebsd_reclaim(ap)
6359 struct vop_reclaim_args /* {
6361 struct thread *a_td;
6364 vnode_t *vp = ap->a_vp;
6365 znode_t *zp = VTOZ(vp);
6366 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
6370 /* Destroy the vm object and flush associated pages. */
6371 vnode_destroy_vobject(vp);
6374 * z_teardown_inactive_lock protects from a race with
6375 * zfs_znode_dmu_fini in zfsvfs_teardown during
6378 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
6379 if (zp->z_sa_hdl == NULL)
6383 rw_exit(&zfsvfs->z_teardown_inactive_lock);
6391 struct vop_fid_args /* {
6397 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL));
6401 zfs_freebsd_pathconf(ap)
6402 struct vop_pathconf_args /* {
6405 register_t *a_retval;
6411 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL);
6413 *ap->a_retval = val;
6414 else if (error == EOPNOTSUPP)
6415 error = vop_stdpathconf(ap);
6420 zfs_freebsd_fifo_pathconf(ap)
6421 struct vop_pathconf_args /* {
6424 register_t *a_retval;
6428 switch (ap->a_name) {
6429 case _PC_ACL_EXTENDED:
6431 case _PC_ACL_PATH_MAX:
6432 case _PC_MAC_PRESENT:
6433 return (zfs_freebsd_pathconf(ap));
6435 return (fifo_specops.vop_pathconf(ap));
6440 * FreeBSD's extended attributes namespace defines file name prefix for ZFS'
6441 * extended attribute name:
6444 * system freebsd:system:
6445 * user (none, can be used to access ZFS fsattr(5) attributes
6446 * created on Solaris)
6449 zfs_create_attrname(int attrnamespace, const char *name, char *attrname,
6452 const char *namespace, *prefix, *suffix;
6454 /* We don't allow '/' character in attribute name. */
6455 if (strchr(name, '/') != NULL)
6457 /* We don't allow attribute names that start with "freebsd:" string. */
6458 if (strncmp(name, "freebsd:", 8) == 0)
6461 bzero(attrname, size);
6463 switch (attrnamespace) {
6464 case EXTATTR_NAMESPACE_USER:
6466 prefix = "freebsd:";
6467 namespace = EXTATTR_NAMESPACE_USER_STRING;
6471 * This is the default namespace by which we can access all
6472 * attributes created on Solaris.
6474 prefix = namespace = suffix = "";
6477 case EXTATTR_NAMESPACE_SYSTEM:
6478 prefix = "freebsd:";
6479 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING;
6482 case EXTATTR_NAMESPACE_EMPTY:
6486 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix,
6488 return (ENAMETOOLONG);
6494 * Vnode operating to retrieve a named extended attribute.
6497 zfs_getextattr(struct vop_getextattr_args *ap)
6500 IN struct vnode *a_vp;
6501 IN int a_attrnamespace;
6502 IN const char *a_name;
6503 INOUT struct uio *a_uio;
6505 IN struct ucred *a_cred;
6506 IN struct thread *a_td;
6510 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6511 struct thread *td = ap->a_td;
6512 struct nameidata nd;
6515 vnode_t *xvp = NULL, *vp;
6518 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6519 ap->a_cred, ap->a_td, VREAD);
6523 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6530 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6538 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6540 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL);
6542 NDFREE(&nd, NDF_ONLY_PNBUF);
6545 if (error == ENOENT)
6550 if (ap->a_size != NULL) {
6551 error = VOP_GETATTR(vp, &va, ap->a_cred);
6553 *ap->a_size = (size_t)va.va_size;
6554 } else if (ap->a_uio != NULL)
6555 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6558 vn_close(vp, flags, ap->a_cred, td);
6565 * Vnode operation to remove a named attribute.
6568 zfs_deleteextattr(struct vop_deleteextattr_args *ap)
6571 IN struct vnode *a_vp;
6572 IN int a_attrnamespace;
6573 IN const char *a_name;
6574 IN struct ucred *a_cred;
6575 IN struct thread *a_td;
6579 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6580 struct thread *td = ap->a_td;
6581 struct nameidata nd;
6584 vnode_t *xvp = NULL, *vp;
6587 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6588 ap->a_cred, ap->a_td, VWRITE);
6592 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6599 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6606 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF,
6607 UIO_SYSSPACE, attrname, xvp, td);
6610 NDFREE(&nd, NDF_ONLY_PNBUF);
6613 if (error == ENOENT)
6617 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
6620 if (vp == nd.ni_dvp)
6630 * Vnode operation to set a named attribute.
6633 zfs_setextattr(struct vop_setextattr_args *ap)
6636 IN struct vnode *a_vp;
6637 IN int a_attrnamespace;
6638 IN const char *a_name;
6639 INOUT struct uio *a_uio;
6640 IN struct ucred *a_cred;
6641 IN struct thread *a_td;
6645 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6646 struct thread *td = ap->a_td;
6647 struct nameidata nd;
6650 vnode_t *xvp = NULL, *vp;
6653 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6654 ap->a_cred, ap->a_td, VWRITE);
6658 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6665 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6666 LOOKUP_XATTR | CREATE_XATTR_DIR);
6672 flags = FFLAGS(O_WRONLY | O_CREAT);
6673 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6675 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL);
6677 NDFREE(&nd, NDF_ONLY_PNBUF);
6685 error = VOP_SETATTR(vp, &va, ap->a_cred);
6687 VOP_WRITE(vp, ap->a_uio, IO_UNIT | IO_SYNC, ap->a_cred);
6690 vn_close(vp, flags, ap->a_cred, td);
6697 * Vnode operation to retrieve extended attributes on a vnode.
6700 zfs_listextattr(struct vop_listextattr_args *ap)
6703 IN struct vnode *a_vp;
6704 IN int a_attrnamespace;
6705 INOUT struct uio *a_uio;
6707 IN struct ucred *a_cred;
6708 IN struct thread *a_td;
6712 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6713 struct thread *td = ap->a_td;
6714 struct nameidata nd;
6715 char attrprefix[16];
6716 u_char dirbuf[sizeof(struct dirent)];
6719 struct uio auio, *uio = ap->a_uio;
6720 size_t *sizep = ap->a_size;
6722 vnode_t *xvp = NULL, *vp;
6723 int done, error, eof, pos;
6725 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6726 ap->a_cred, ap->a_td, VREAD);
6730 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix,
6731 sizeof(attrprefix));
6734 plen = strlen(attrprefix);
6741 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6746 * ENOATTR means that the EA directory does not yet exist,
6747 * i.e. there are no extended attributes there.
6749 if (error == ENOATTR)
6754 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED,
6755 UIO_SYSSPACE, ".", xvp, td);
6758 NDFREE(&nd, NDF_ONLY_PNBUF);
6764 auio.uio_iov = &aiov;
6765 auio.uio_iovcnt = 1;
6766 auio.uio_segflg = UIO_SYSSPACE;
6768 auio.uio_rw = UIO_READ;
6769 auio.uio_offset = 0;
6774 aiov.iov_base = (void *)dirbuf;
6775 aiov.iov_len = sizeof(dirbuf);
6776 auio.uio_resid = sizeof(dirbuf);
6777 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL);
6778 done = sizeof(dirbuf) - auio.uio_resid;
6781 for (pos = 0; pos < done;) {
6782 dp = (struct dirent *)(dirbuf + pos);
6783 pos += dp->d_reclen;
6785 * XXX: Temporarily we also accept DT_UNKNOWN, as this
6786 * is what we get when attribute was created on Solaris.
6788 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN)
6790 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0)
6792 else if (strncmp(dp->d_name, attrprefix, plen) != 0)
6794 nlen = dp->d_namlen - plen;
6797 else if (uio != NULL) {
6799 * Format of extattr name entry is one byte for
6800 * length and the rest for name.
6802 error = uiomove(&nlen, 1, uio->uio_rw, uio);
6804 error = uiomove(dp->d_name + plen, nlen,
6811 } while (!eof && error == 0);
6820 zfs_freebsd_getacl(ap)
6821 struct vop_getacl_args /* {
6830 vsecattr_t vsecattr;
6832 if (ap->a_type != ACL_TYPE_NFS4)
6835 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
6836 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL))
6839 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt);
6840 if (vsecattr.vsa_aclentp != NULL)
6841 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz);
6847 zfs_freebsd_setacl(ap)
6848 struct vop_setacl_args /* {
6857 vsecattr_t vsecattr;
6858 int aclbsize; /* size of acl list in bytes */
6861 if (ap->a_type != ACL_TYPE_NFS4)
6864 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES)
6868 * With NFSv4 ACLs, chmod(2) may need to add additional entries,
6869 * splitting every entry into two and appending "canonical six"
6870 * entries at the end. Don't allow for setting an ACL that would
6871 * cause chmod(2) to run out of ACL entries.
6873 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES)
6876 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR);
6880 vsecattr.vsa_mask = VSA_ACE;
6881 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t);
6882 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP);
6883 aaclp = vsecattr.vsa_aclentp;
6884 vsecattr.vsa_aclentsz = aclbsize;
6886 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp);
6887 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL);
6888 kmem_free(aaclp, aclbsize);
6894 zfs_freebsd_aclcheck(ap)
6895 struct vop_aclcheck_args /* {
6904 return (EOPNOTSUPP);
6907 struct vop_vector zfs_vnodeops;
6908 struct vop_vector zfs_fifoops;
6909 struct vop_vector zfs_shareops;
6911 struct vop_vector zfs_vnodeops = {
6912 .vop_default = &default_vnodeops,
6913 .vop_inactive = zfs_freebsd_inactive,
6914 .vop_reclaim = zfs_freebsd_reclaim,
6915 .vop_access = zfs_freebsd_access,
6916 #ifdef FREEBSD_NAMECACHE
6917 .vop_lookup = vfs_cache_lookup,
6918 .vop_cachedlookup = zfs_freebsd_lookup,
6920 .vop_lookup = zfs_freebsd_lookup,
6922 .vop_getattr = zfs_freebsd_getattr,
6923 .vop_setattr = zfs_freebsd_setattr,
6924 .vop_create = zfs_freebsd_create,
6925 .vop_mknod = zfs_freebsd_create,
6926 .vop_mkdir = zfs_freebsd_mkdir,
6927 .vop_readdir = zfs_freebsd_readdir,
6928 .vop_fsync = zfs_freebsd_fsync,
6929 .vop_open = zfs_freebsd_open,
6930 .vop_close = zfs_freebsd_close,
6931 .vop_rmdir = zfs_freebsd_rmdir,
6932 .vop_ioctl = zfs_freebsd_ioctl,
6933 .vop_link = zfs_freebsd_link,
6934 .vop_symlink = zfs_freebsd_symlink,
6935 .vop_readlink = zfs_freebsd_readlink,
6936 .vop_read = zfs_freebsd_read,
6937 .vop_write = zfs_freebsd_write,
6938 .vop_remove = zfs_freebsd_remove,
6939 .vop_rename = zfs_freebsd_rename,
6940 .vop_pathconf = zfs_freebsd_pathconf,
6941 .vop_bmap = zfs_freebsd_bmap,
6942 .vop_fid = zfs_freebsd_fid,
6943 .vop_getextattr = zfs_getextattr,
6944 .vop_deleteextattr = zfs_deleteextattr,
6945 .vop_setextattr = zfs_setextattr,
6946 .vop_listextattr = zfs_listextattr,
6947 .vop_getacl = zfs_freebsd_getacl,
6948 .vop_setacl = zfs_freebsd_setacl,
6949 .vop_aclcheck = zfs_freebsd_aclcheck,
6950 .vop_getpages = zfs_freebsd_getpages,
6953 struct vop_vector zfs_fifoops = {
6954 .vop_default = &fifo_specops,
6955 .vop_fsync = zfs_freebsd_fsync,
6956 .vop_access = zfs_freebsd_access,
6957 .vop_getattr = zfs_freebsd_getattr,
6958 .vop_inactive = zfs_freebsd_inactive,
6959 .vop_read = VOP_PANIC,
6960 .vop_reclaim = zfs_freebsd_reclaim,
6961 .vop_setattr = zfs_freebsd_setattr,
6962 .vop_write = VOP_PANIC,
6963 .vop_pathconf = zfs_freebsd_fifo_pathconf,
6964 .vop_fid = zfs_freebsd_fid,
6965 .vop_getacl = zfs_freebsd_getacl,
6966 .vop_setacl = zfs_freebsd_setacl,
6967 .vop_aclcheck = zfs_freebsd_aclcheck,
6971 * special share hidden files vnode operations template
6973 struct vop_vector zfs_shareops = {
6974 .vop_default = &default_vnodeops,
6975 .vop_access = zfs_freebsd_access,
6976 .vop_inactive = zfs_freebsd_inactive,
6977 .vop_reclaim = zfs_freebsd_reclaim,
6978 .vop_fid = zfs_freebsd_fid,
6979 .vop_pathconf = zfs_freebsd_pathconf,