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/sched.h>
76 #include <vm/vm_param.h>
77 #include <vm/vm_pageout.h>
82 * Each vnode op performs some logical unit of work. To do this, the ZPL must
83 * properly lock its in-core state, create a DMU transaction, do the work,
84 * record this work in the intent log (ZIL), commit the DMU transaction,
85 * and wait for the intent log to commit if it is a synchronous operation.
86 * Moreover, the vnode ops must work in both normal and log replay context.
87 * The ordering of events is important to avoid deadlocks and references
88 * to freed memory. The example below illustrates the following Big Rules:
90 * (1) A check must be made in each zfs thread for a mounted file system.
91 * This is done avoiding races using ZFS_ENTER(zfsvfs).
92 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
93 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
94 * can return EIO from the calling function.
96 * (2) VN_RELE() should always be the last thing except for zil_commit()
97 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
98 * First, if it's the last reference, the vnode/znode
99 * can be freed, so the zp may point to freed memory. Second, the last
100 * reference will call zfs_zinactive(), which may induce a lot of work --
101 * pushing cached pages (which acquires range locks) and syncing out
102 * cached atime changes. Third, zfs_zinactive() may require a new tx,
103 * which could deadlock the system if you were already holding one.
104 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
106 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
107 * as they can span dmu_tx_assign() calls.
109 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
110 * dmu_tx_assign(). This is critical because we don't want to block
111 * while holding locks.
113 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
114 * reduces lock contention and CPU usage when we must wait (note that if
115 * throughput is constrained by the storage, nearly every transaction
118 * Note, in particular, that if a lock is sometimes acquired before
119 * the tx assigns, and sometimes after (e.g. z_lock), then failing
120 * to use a non-blocking assign can deadlock the system. The scenario:
122 * Thread A has grabbed a lock before calling dmu_tx_assign().
123 * Thread B is in an already-assigned tx, and blocks for this lock.
124 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
125 * forever, because the previous txg can't quiesce until B's tx commits.
127 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
128 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
129 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
130 * to indicate that this operation has already called dmu_tx_wait().
131 * This will ensure that we don't retry forever, waiting a short bit
134 * (5) If the operation succeeded, generate the intent log entry for it
135 * before dropping locks. This ensures that the ordering of events
136 * in the intent log matches the order in which they actually occurred.
137 * During ZIL replay the zfs_log_* functions will update the sequence
138 * number to indicate the zil transaction has replayed.
140 * (6) At the end of each vnode op, the DMU tx must always commit,
141 * regardless of whether there were any errors.
143 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
144 * to ensure that synchronous semantics are provided when necessary.
146 * In general, this is how things should be ordered in each vnode op:
148 * ZFS_ENTER(zfsvfs); // exit if unmounted
150 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD())
151 * rw_enter(...); // grab any other locks you need
152 * tx = dmu_tx_create(...); // get DMU tx
153 * dmu_tx_hold_*(); // hold each object you might modify
154 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
156 * rw_exit(...); // drop locks
157 * zfs_dirent_unlock(dl); // unlock directory entry
158 * VN_RELE(...); // release held vnodes
159 * if (error == ERESTART) {
165 * dmu_tx_abort(tx); // abort DMU tx
166 * ZFS_EXIT(zfsvfs); // finished in zfs
167 * return (error); // really out of space
169 * error = do_real_work(); // do whatever this VOP does
171 * zfs_log_*(...); // on success, make ZIL entry
172 * dmu_tx_commit(tx); // commit DMU tx -- error or not
173 * rw_exit(...); // drop locks
174 * zfs_dirent_unlock(dl); // unlock directory entry
175 * VN_RELE(...); // release held vnodes
176 * zil_commit(zilog, foid); // synchronous when necessary
177 * ZFS_EXIT(zfsvfs); // finished in zfs
178 * return (error); // done, report error
183 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
185 znode_t *zp = VTOZ(*vpp);
186 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
191 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
192 ((flag & FAPPEND) == 0)) {
194 return (SET_ERROR(EPERM));
197 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
198 ZTOV(zp)->v_type == VREG &&
199 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
200 if (fs_vscan(*vpp, cr, 0) != 0) {
202 return (SET_ERROR(EACCES));
206 /* Keep a count of the synchronous opens in the znode */
207 if (flag & (FSYNC | FDSYNC))
208 atomic_inc_32(&zp->z_sync_cnt);
216 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
217 caller_context_t *ct)
219 znode_t *zp = VTOZ(vp);
220 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
223 * Clean up any locks held by this process on the vp.
225 cleanlocks(vp, ddi_get_pid(), 0);
226 cleanshares(vp, ddi_get_pid());
231 /* Decrement the synchronous opens in the znode */
232 if ((flag & (FSYNC | FDSYNC)) && (count == 1))
233 atomic_dec_32(&zp->z_sync_cnt);
235 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
236 ZTOV(zp)->v_type == VREG &&
237 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
238 VERIFY(fs_vscan(vp, cr, 1) == 0);
245 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
246 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
249 zfs_holey(vnode_t *vp, u_long cmd, offset_t *off)
251 znode_t *zp = VTOZ(vp);
252 uint64_t noff = (uint64_t)*off; /* new offset */
257 file_sz = zp->z_size;
258 if (noff >= file_sz) {
259 return (SET_ERROR(ENXIO));
262 if (cmd == _FIO_SEEK_HOLE)
267 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
270 if ((error == ESRCH) || (noff > file_sz)) {
272 * Handle the virtual hole at the end of file.
278 return (SET_ERROR(ENXIO));
289 zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred,
290 int *rvalp, caller_context_t *ct)
302 * The following two ioctls are used by bfu. Faking out,
303 * necessary to avoid bfu errors.
312 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
313 return (SET_ERROR(EFAULT));
315 off = *(offset_t *)data;
318 zfsvfs = zp->z_zfsvfs;
322 /* offset parameter is in/out */
323 error = zfs_holey(vp, com, &off);
328 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
329 return (SET_ERROR(EFAULT));
331 *(offset_t *)data = off;
335 return (SET_ERROR(ENOTTY));
339 page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes)
346 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE
347 * aligned boundaries, if the range is not aligned. As a result a
348 * DEV_BSIZE subrange with partially dirty data may get marked as clean.
349 * It may happen that all DEV_BSIZE subranges are marked clean and thus
350 * the whole page would be considred clean despite have some dirty data.
351 * For this reason we should shrink the range to DEV_BSIZE aligned
352 * boundaries before calling vm_page_clear_dirty.
354 end = rounddown2(off + nbytes, DEV_BSIZE);
355 off = roundup2(off, DEV_BSIZE);
359 zfs_vmobject_assert_wlocked(obj);
362 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
364 if (vm_page_xbusied(pp)) {
366 * Reference the page before unlocking and
367 * sleeping so that the page daemon is less
368 * likely to reclaim it.
370 vm_page_reference(pp);
372 zfs_vmobject_wunlock(obj);
373 vm_page_busy_sleep(pp, "zfsmwb");
374 zfs_vmobject_wlock(obj);
378 } else if (pp == NULL) {
379 pp = vm_page_alloc(obj, OFF_TO_IDX(start),
380 VM_ALLOC_SYSTEM | VM_ALLOC_IFCACHED |
383 ASSERT(pp != NULL && !pp->valid);
388 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
389 vm_object_pip_add(obj, 1);
390 pmap_remove_write(pp);
392 vm_page_clear_dirty(pp, off, nbytes);
400 page_unbusy(vm_page_t pp)
404 vm_object_pip_subtract(pp->object, 1);
408 page_hold(vnode_t *vp, int64_t start)
414 zfs_vmobject_assert_wlocked(obj);
417 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
419 if (vm_page_xbusied(pp)) {
421 * Reference the page before unlocking and
422 * sleeping so that the page daemon is less
423 * likely to reclaim it.
425 vm_page_reference(pp);
427 zfs_vmobject_wunlock(obj);
428 vm_page_busy_sleep(pp, "zfsmwb");
429 zfs_vmobject_wlock(obj);
433 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
446 page_unhold(vm_page_t pp)
455 * When a file is memory mapped, we must keep the IO data synchronized
456 * between the DMU cache and the memory mapped pages. What this means:
458 * On Write: If we find a memory mapped page, we write to *both*
459 * the page and the dmu buffer.
462 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid,
463 int segflg, dmu_tx_t *tx)
470 ASSERT(segflg != UIO_NOCOPY);
471 ASSERT(vp->v_mount != NULL);
475 off = start & PAGEOFFSET;
476 zfs_vmobject_wlock(obj);
477 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
479 int nbytes = imin(PAGESIZE - off, len);
481 if ((pp = page_busy(vp, start, off, nbytes)) != NULL) {
482 zfs_vmobject_wunlock(obj);
484 va = zfs_map_page(pp, &sf);
485 (void) dmu_read(os, oid, start+off, nbytes,
486 va+off, DMU_READ_PREFETCH);;
489 zfs_vmobject_wlock(obj);
495 vm_object_pip_wakeupn(obj, 0);
496 zfs_vmobject_wunlock(obj);
500 * Read with UIO_NOCOPY flag means that sendfile(2) requests
501 * ZFS to populate a range of page cache pages with data.
503 * NOTE: this function could be optimized to pre-allocate
504 * all pages in advance, drain exclusive busy on all of them,
505 * map them into contiguous KVA region and populate them
506 * in one single dmu_read() call.
509 mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio)
511 znode_t *zp = VTOZ(vp);
512 objset_t *os = zp->z_zfsvfs->z_os;
522 ASSERT(uio->uio_segflg == UIO_NOCOPY);
523 ASSERT(vp->v_mount != NULL);
526 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0);
528 zfs_vmobject_wlock(obj);
529 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) {
530 int bytes = MIN(PAGESIZE, len);
532 pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_SBUSY |
533 VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY);
534 if (pp->valid == 0) {
535 zfs_vmobject_wunlock(obj);
536 va = zfs_map_page(pp, &sf);
537 error = dmu_read(os, zp->z_id, start, bytes, va,
539 if (bytes != PAGESIZE && error == 0)
540 bzero(va + bytes, PAGESIZE - bytes);
542 zfs_vmobject_wlock(obj);
546 if (pp->wire_count == 0 && pp->valid == 0 &&
550 pp->valid = VM_PAGE_BITS_ALL;
551 vm_page_activate(pp);
555 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
560 uio->uio_resid -= bytes;
561 uio->uio_offset += bytes;
564 zfs_vmobject_wunlock(obj);
569 * When a file is memory mapped, we must keep the IO data synchronized
570 * between the DMU cache and the memory mapped pages. What this means:
572 * On Read: We "read" preferentially from memory mapped pages,
573 * else we default from the dmu buffer.
575 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
576 * the file is memory mapped.
579 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
581 znode_t *zp = VTOZ(vp);
582 objset_t *os = zp->z_zfsvfs->z_os;
590 ASSERT(vp->v_mount != NULL);
594 start = uio->uio_loffset;
595 off = start & PAGEOFFSET;
596 zfs_vmobject_wlock(obj);
597 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
599 uint64_t bytes = MIN(PAGESIZE - off, len);
601 if (pp = page_hold(vp, start)) {
605 zfs_vmobject_wunlock(obj);
606 va = zfs_map_page(pp, &sf);
607 error = uiomove(va + off, bytes, UIO_READ, uio);
609 zfs_vmobject_wlock(obj);
612 zfs_vmobject_wunlock(obj);
613 error = dmu_read_uio(os, zp->z_id, uio, bytes);
614 zfs_vmobject_wlock(obj);
621 zfs_vmobject_wunlock(obj);
625 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
628 * Read bytes from specified file into supplied buffer.
630 * IN: vp - vnode of file to be read from.
631 * uio - structure supplying read location, range info,
633 * ioflag - SYNC flags; used to provide FRSYNC semantics.
634 * cr - credentials of caller.
635 * ct - caller context
637 * OUT: uio - updated offset and range, buffer filled.
639 * RETURN: 0 on success, error code on failure.
642 * vp - atime updated if byte count > 0
646 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
648 znode_t *zp = VTOZ(vp);
649 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
660 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
662 return (SET_ERROR(EACCES));
666 * Validate file offset
668 if (uio->uio_loffset < (offset_t)0) {
670 return (SET_ERROR(EINVAL));
674 * Fasttrack empty reads
676 if (uio->uio_resid == 0) {
682 * Check for mandatory locks
684 if (MANDMODE(zp->z_mode)) {
685 if (error = chklock(vp, FREAD,
686 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
693 * If we're in FRSYNC mode, sync out this znode before reading it.
696 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
697 zil_commit(zfsvfs->z_log, zp->z_id);
700 * Lock the range against changes.
702 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
705 * If we are reading past end-of-file we can skip
706 * to the end; but we might still need to set atime.
708 if (uio->uio_loffset >= zp->z_size) {
713 ASSERT(uio->uio_loffset < zp->z_size);
714 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
717 if ((uio->uio_extflg == UIO_XUIO) &&
718 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
720 int blksz = zp->z_blksz;
721 uint64_t offset = uio->uio_loffset;
723 xuio = (xuio_t *)uio;
725 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
728 ASSERT(offset + n <= blksz);
731 (void) dmu_xuio_init(xuio, nblk);
733 if (vn_has_cached_data(vp)) {
735 * For simplicity, we always allocate a full buffer
736 * even if we only expect to read a portion of a block.
738 while (--nblk >= 0) {
739 (void) dmu_xuio_add(xuio,
740 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
748 nbytes = MIN(n, zfs_read_chunk_size -
749 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
752 if (uio->uio_segflg == UIO_NOCOPY)
753 error = mappedread_sf(vp, nbytes, uio);
755 #endif /* __FreeBSD__ */
756 if (vn_has_cached_data(vp))
757 error = mappedread(vp, nbytes, uio);
759 error = dmu_read_uio(os, zp->z_id, uio, nbytes);
761 /* convert checksum errors into IO errors */
763 error = SET_ERROR(EIO);
770 zfs_range_unlock(rl);
772 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
778 * Write the bytes to a file.
780 * IN: vp - vnode of file to be written to.
781 * uio - structure supplying write location, range info,
783 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is
784 * set if in append mode.
785 * cr - credentials of caller.
786 * ct - caller context (NFS/CIFS fem monitor only)
788 * OUT: uio - updated offset and range.
790 * RETURN: 0 on success, error code on failure.
793 * vp - ctime|mtime updated if byte count > 0
798 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
800 znode_t *zp = VTOZ(vp);
801 rlim64_t limit = MAXOFFSET_T;
802 ssize_t start_resid = uio->uio_resid;
806 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
811 int max_blksz = zfsvfs->z_max_blksz;
814 iovec_t *aiov = NULL;
817 int iovcnt = uio->uio_iovcnt;
818 iovec_t *iovp = uio->uio_iov;
821 sa_bulk_attr_t bulk[4];
822 uint64_t mtime[2], ctime[2];
825 * Fasttrack empty write
831 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
837 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
838 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
839 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
841 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
845 * If immutable or not appending then return EPERM
847 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
848 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
849 (uio->uio_loffset < zp->z_size))) {
851 return (SET_ERROR(EPERM));
854 zilog = zfsvfs->z_log;
857 * Validate file offset
859 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
862 return (SET_ERROR(EINVAL));
866 * Check for mandatory locks before calling zfs_range_lock()
867 * in order to prevent a deadlock with locks set via fcntl().
869 if (MANDMODE((mode_t)zp->z_mode) &&
870 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
877 * Pre-fault the pages to ensure slow (eg NFS) pages
879 * Skip this if uio contains loaned arc_buf.
881 if ((uio->uio_extflg == UIO_XUIO) &&
882 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
883 xuio = (xuio_t *)uio;
885 uio_prefaultpages(MIN(n, max_blksz), uio);
889 * If in append mode, set the io offset pointer to eof.
891 if (ioflag & FAPPEND) {
893 * Obtain an appending range lock to guarantee file append
894 * semantics. We reset the write offset once we have the lock.
896 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
898 if (rl->r_len == UINT64_MAX) {
900 * We overlocked the file because this write will cause
901 * the file block size to increase.
902 * Note that zp_size cannot change with this lock held.
906 uio->uio_loffset = woff;
909 * Note that if the file block size will change as a result of
910 * this write, then this range lock will lock the entire file
911 * so that we can re-write the block safely.
913 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
916 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) {
917 zfs_range_unlock(rl);
923 zfs_range_unlock(rl);
925 return (SET_ERROR(EFBIG));
928 if ((woff + n) > limit || woff > (limit - n))
931 /* Will this write extend the file length? */
932 write_eof = (woff + n > zp->z_size);
934 end_size = MAX(zp->z_size, woff + n);
937 * Write the file in reasonable size chunks. Each chunk is written
938 * in a separate transaction; this keeps the intent log records small
939 * and allows us to do more fine-grained space accounting.
943 woff = uio->uio_loffset;
944 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
945 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
947 dmu_return_arcbuf(abuf);
948 error = SET_ERROR(EDQUOT);
952 if (xuio && abuf == NULL) {
953 ASSERT(i_iov < iovcnt);
955 abuf = dmu_xuio_arcbuf(xuio, i_iov);
956 dmu_xuio_clear(xuio, i_iov);
957 DTRACE_PROBE3(zfs_cp_write, int, i_iov,
958 iovec_t *, aiov, arc_buf_t *, abuf);
959 ASSERT((aiov->iov_base == abuf->b_data) ||
960 ((char *)aiov->iov_base - (char *)abuf->b_data +
961 aiov->iov_len == arc_buf_size(abuf)));
963 } else if (abuf == NULL && n >= max_blksz &&
964 woff >= zp->z_size &&
965 P2PHASE(woff, max_blksz) == 0 &&
966 zp->z_blksz == max_blksz) {
968 * This write covers a full block. "Borrow" a buffer
969 * from the dmu so that we can fill it before we enter
970 * a transaction. This avoids the possibility of
971 * holding up the transaction if the data copy hangs
972 * up on a pagefault (e.g., from an NFS server mapping).
976 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
978 ASSERT(abuf != NULL);
979 ASSERT(arc_buf_size(abuf) == max_blksz);
980 if (error = uiocopy(abuf->b_data, max_blksz,
981 UIO_WRITE, uio, &cbytes)) {
982 dmu_return_arcbuf(abuf);
985 ASSERT(cbytes == max_blksz);
989 * Start a transaction.
991 tx = dmu_tx_create(zfsvfs->z_os);
992 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
993 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
994 zfs_sa_upgrade_txholds(tx, zp);
995 error = dmu_tx_assign(tx, TXG_WAIT);
999 dmu_return_arcbuf(abuf);
1004 * If zfs_range_lock() over-locked we grow the blocksize
1005 * and then reduce the lock range. This will only happen
1006 * on the first iteration since zfs_range_reduce() will
1007 * shrink down r_len to the appropriate size.
1009 if (rl->r_len == UINT64_MAX) {
1012 if (zp->z_blksz > max_blksz) {
1013 ASSERT(!ISP2(zp->z_blksz));
1014 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
1016 new_blksz = MIN(end_size, max_blksz);
1018 zfs_grow_blocksize(zp, new_blksz, tx);
1019 zfs_range_reduce(rl, woff, n);
1023 * XXX - should we really limit each write to z_max_blksz?
1024 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
1026 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
1028 if (woff + nbytes > zp->z_size)
1029 vnode_pager_setsize(vp, woff + nbytes);
1032 tx_bytes = uio->uio_resid;
1033 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
1035 tx_bytes -= uio->uio_resid;
1038 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
1040 * If this is not a full block write, but we are
1041 * extending the file past EOF and this data starts
1042 * block-aligned, use assign_arcbuf(). Otherwise,
1043 * write via dmu_write().
1045 if (tx_bytes < max_blksz && (!write_eof ||
1046 aiov->iov_base != abuf->b_data)) {
1048 dmu_write(zfsvfs->z_os, zp->z_id, woff,
1049 aiov->iov_len, aiov->iov_base, tx);
1050 dmu_return_arcbuf(abuf);
1051 xuio_stat_wbuf_copied();
1053 ASSERT(xuio || tx_bytes == max_blksz);
1054 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
1057 ASSERT(tx_bytes <= uio->uio_resid);
1058 uioskip(uio, tx_bytes);
1060 if (tx_bytes && vn_has_cached_data(vp)) {
1061 update_pages(vp, woff, tx_bytes, zfsvfs->z_os,
1062 zp->z_id, uio->uio_segflg, tx);
1066 * If we made no progress, we're done. If we made even
1067 * partial progress, update the znode and ZIL accordingly.
1069 if (tx_bytes == 0) {
1070 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
1071 (void *)&zp->z_size, sizeof (uint64_t), tx);
1078 * Clear Set-UID/Set-GID bits on successful write if not
1079 * privileged and at least one of the excute bits is set.
1081 * It would be nice to to this after all writes have
1082 * been done, but that would still expose the ISUID/ISGID
1083 * to another app after the partial write is committed.
1085 * Note: we don't call zfs_fuid_map_id() here because
1086 * user 0 is not an ephemeral uid.
1088 mutex_enter(&zp->z_acl_lock);
1089 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
1090 (S_IXUSR >> 6))) != 0 &&
1091 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
1092 secpolicy_vnode_setid_retain(vp, cr,
1093 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
1095 zp->z_mode &= ~(S_ISUID | S_ISGID);
1096 newmode = zp->z_mode;
1097 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
1098 (void *)&newmode, sizeof (uint64_t), tx);
1100 mutex_exit(&zp->z_acl_lock);
1102 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
1106 * Update the file size (zp_size) if it has changed;
1107 * account for possible concurrent updates.
1109 while ((end_size = zp->z_size) < uio->uio_loffset) {
1110 (void) atomic_cas_64(&zp->z_size, end_size,
1115 * If we are replaying and eof is non zero then force
1116 * the file size to the specified eof. Note, there's no
1117 * concurrency during replay.
1119 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
1120 zp->z_size = zfsvfs->z_replay_eof;
1122 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1124 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
1129 ASSERT(tx_bytes == nbytes);
1134 uio_prefaultpages(MIN(n, max_blksz), uio);
1138 zfs_range_unlock(rl);
1141 * If we're in replay mode, or we made no progress, return error.
1142 * Otherwise, it's at least a partial write, so it's successful.
1144 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1149 if (ioflag & (FSYNC | FDSYNC) ||
1150 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1151 zil_commit(zilog, zp->z_id);
1158 zfs_get_done(zgd_t *zgd, int error)
1160 znode_t *zp = zgd->zgd_private;
1161 objset_t *os = zp->z_zfsvfs->z_os;
1164 dmu_buf_rele(zgd->zgd_db, zgd);
1166 zfs_range_unlock(zgd->zgd_rl);
1169 * Release the vnode asynchronously as we currently have the
1170 * txg stopped from syncing.
1172 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1174 if (error == 0 && zgd->zgd_bp)
1175 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1177 kmem_free(zgd, sizeof (zgd_t));
1181 static int zil_fault_io = 0;
1185 * Get data to generate a TX_WRITE intent log record.
1188 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1190 zfsvfs_t *zfsvfs = arg;
1191 objset_t *os = zfsvfs->z_os;
1193 uint64_t object = lr->lr_foid;
1194 uint64_t offset = lr->lr_offset;
1195 uint64_t size = lr->lr_length;
1196 blkptr_t *bp = &lr->lr_blkptr;
1201 ASSERT(zio != NULL);
1205 * Nothing to do if the file has been removed
1207 if (zfs_zget(zfsvfs, object, &zp) != 0)
1208 return (SET_ERROR(ENOENT));
1209 if (zp->z_unlinked) {
1211 * Release the vnode asynchronously as we currently have the
1212 * txg stopped from syncing.
1214 VN_RELE_ASYNC(ZTOV(zp),
1215 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1216 return (SET_ERROR(ENOENT));
1219 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1220 zgd->zgd_zilog = zfsvfs->z_log;
1221 zgd->zgd_private = zp;
1224 * Write records come in two flavors: immediate and indirect.
1225 * For small writes it's cheaper to store the data with the
1226 * log record (immediate); for large writes it's cheaper to
1227 * sync the data and get a pointer to it (indirect) so that
1228 * we don't have to write the data twice.
1230 if (buf != NULL) { /* immediate write */
1231 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1232 /* test for truncation needs to be done while range locked */
1233 if (offset >= zp->z_size) {
1234 error = SET_ERROR(ENOENT);
1236 error = dmu_read(os, object, offset, size, buf,
1237 DMU_READ_NO_PREFETCH);
1239 ASSERT(error == 0 || error == ENOENT);
1240 } else { /* indirect write */
1242 * Have to lock the whole block to ensure when it's
1243 * written out and it's checksum is being calculated
1244 * that no one can change the data. We need to re-check
1245 * blocksize after we get the lock in case it's changed!
1250 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1252 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1254 if (zp->z_blksz == size)
1257 zfs_range_unlock(zgd->zgd_rl);
1259 /* test for truncation needs to be done while range locked */
1260 if (lr->lr_offset >= zp->z_size)
1261 error = SET_ERROR(ENOENT);
1264 error = SET_ERROR(EIO);
1269 error = dmu_buf_hold(os, object, offset, zgd, &db,
1270 DMU_READ_NO_PREFETCH);
1273 blkptr_t *obp = dmu_buf_get_blkptr(db);
1275 ASSERT(BP_IS_HOLE(bp));
1282 ASSERT(db->db_offset == offset);
1283 ASSERT(db->db_size == size);
1285 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1287 ASSERT(error || lr->lr_length <= zp->z_blksz);
1290 * On success, we need to wait for the write I/O
1291 * initiated by dmu_sync() to complete before we can
1292 * release this dbuf. We will finish everything up
1293 * in the zfs_get_done() callback.
1298 if (error == EALREADY) {
1299 lr->lr_common.lrc_txtype = TX_WRITE2;
1305 zfs_get_done(zgd, error);
1312 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1313 caller_context_t *ct)
1315 znode_t *zp = VTOZ(vp);
1316 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1322 if (flag & V_ACE_MASK)
1323 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1325 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1332 * If vnode is for a device return a specfs vnode instead.
1335 specvp_check(vnode_t **vpp, cred_t *cr)
1339 if (IS_DEVVP(*vpp)) {
1342 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1345 error = SET_ERROR(ENOSYS);
1353 * Lookup an entry in a directory, or an extended attribute directory.
1354 * If it exists, return a held vnode reference for it.
1356 * IN: dvp - vnode of directory to search.
1357 * nm - name of entry to lookup.
1358 * pnp - full pathname to lookup [UNUSED].
1359 * flags - LOOKUP_XATTR set if looking for an attribute.
1360 * rdir - root directory vnode [UNUSED].
1361 * cr - credentials of caller.
1362 * ct - caller context
1363 * direntflags - directory lookup flags
1364 * realpnp - returned pathname.
1366 * OUT: vpp - vnode of located entry, NULL if not found.
1368 * RETURN: 0 on success, error code on failure.
1375 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp,
1376 int nameiop, cred_t *cr, kthread_t *td, int flags)
1378 znode_t *zdp = VTOZ(dvp);
1379 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1381 int *direntflags = NULL;
1382 void *realpnp = NULL;
1385 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1387 if (dvp->v_type != VDIR) {
1388 return (SET_ERROR(ENOTDIR));
1389 } else if (zdp->z_sa_hdl == NULL) {
1390 return (SET_ERROR(EIO));
1393 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1394 error = zfs_fastaccesschk_execute(zdp, cr);
1402 vnode_t *tvp = dnlc_lookup(dvp, nm);
1405 error = zfs_fastaccesschk_execute(zdp, cr);
1410 if (tvp == DNLC_NO_VNODE) {
1412 return (SET_ERROR(ENOENT));
1415 return (specvp_check(vpp, cr));
1421 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1428 if (flags & LOOKUP_XATTR) {
1431 * If the xattr property is off, refuse the lookup request.
1433 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1435 return (SET_ERROR(EINVAL));
1440 * We don't allow recursive attributes..
1441 * Maybe someday we will.
1443 if (zdp->z_pflags & ZFS_XATTR) {
1445 return (SET_ERROR(EINVAL));
1448 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1454 * Do we have permission to get into attribute directory?
1457 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1467 if (dvp->v_type != VDIR) {
1469 return (SET_ERROR(ENOTDIR));
1473 * Check accessibility of directory.
1476 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1481 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1482 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1484 return (SET_ERROR(EILSEQ));
1487 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1489 error = specvp_check(vpp, cr);
1491 /* Translate errors and add SAVENAME when needed. */
1492 if (cnp->cn_flags & ISLASTCN) {
1496 if (error == ENOENT) {
1497 error = EJUSTRETURN;
1498 cnp->cn_flags |= SAVENAME;
1504 cnp->cn_flags |= SAVENAME;
1508 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) {
1511 if (cnp->cn_flags & ISDOTDOT) {
1512 ltype = VOP_ISLOCKED(dvp);
1516 error = vn_lock(*vpp, cnp->cn_lkflags);
1517 if (cnp->cn_flags & ISDOTDOT)
1518 vn_lock(dvp, ltype | LK_RETRY);
1528 #ifdef FREEBSD_NAMECACHE
1530 * Insert name into cache (as non-existent) if appropriate.
1532 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
1533 cache_enter(dvp, *vpp, cnp);
1535 * Insert name into cache if appropriate.
1537 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) {
1538 if (!(cnp->cn_flags & ISLASTCN) ||
1539 (nameiop != DELETE && nameiop != RENAME)) {
1540 cache_enter(dvp, *vpp, cnp);
1549 * Attempt to create a new entry in a directory. If the entry
1550 * already exists, truncate the file if permissible, else return
1551 * an error. Return the vp of the created or trunc'd file.
1553 * IN: dvp - vnode of directory to put new file entry in.
1554 * name - name of new file entry.
1555 * vap - attributes of new file.
1556 * excl - flag indicating exclusive or non-exclusive mode.
1557 * mode - mode to open file with.
1558 * cr - credentials of caller.
1559 * flag - large file flag [UNUSED].
1560 * ct - caller context
1561 * vsecp - ACL to be set
1563 * OUT: vpp - vnode of created or trunc'd entry.
1565 * RETURN: 0 on success, error code on failure.
1568 * dvp - ctime|mtime updated if new entry created
1569 * vp - ctime|mtime always, atime if new
1574 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode,
1575 vnode_t **vpp, cred_t *cr, kthread_t *td)
1577 znode_t *zp, *dzp = VTOZ(dvp);
1578 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1586 gid_t gid = crgetgid(cr);
1587 zfs_acl_ids_t acl_ids;
1588 boolean_t fuid_dirtied;
1589 boolean_t have_acl = B_FALSE;
1590 boolean_t waited = B_FALSE;
1595 * If we have an ephemeral id, ACL, or XVATTR then
1596 * make sure file system is at proper version
1599 ksid = crgetsid(cr, KSID_OWNER);
1601 uid = ksid_getid(ksid);
1605 if (zfsvfs->z_use_fuids == B_FALSE &&
1606 (vsecp || (vap->va_mask & AT_XVATTR) ||
1607 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1608 return (SET_ERROR(EINVAL));
1613 zilog = zfsvfs->z_log;
1615 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1616 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1618 return (SET_ERROR(EILSEQ));
1621 if (vap->va_mask & AT_XVATTR) {
1622 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
1623 crgetuid(cr), cr, vap->va_type)) != 0) {
1629 getnewvnode_reserve(1);
1634 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr))
1635 vap->va_mode &= ~S_ISVTX;
1637 if (*name == '\0') {
1639 * Null component name refers to the directory itself.
1646 /* possible VN_HOLD(zp) */
1649 if (flag & FIGNORECASE)
1652 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1656 zfs_acl_ids_free(&acl_ids);
1657 if (strcmp(name, "..") == 0)
1658 error = SET_ERROR(EISDIR);
1659 getnewvnode_drop_reserve();
1669 * Create a new file object and update the directory
1672 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1674 zfs_acl_ids_free(&acl_ids);
1679 * We only support the creation of regular files in
1680 * extended attribute directories.
1683 if ((dzp->z_pflags & ZFS_XATTR) &&
1684 (vap->va_type != VREG)) {
1686 zfs_acl_ids_free(&acl_ids);
1687 error = SET_ERROR(EINVAL);
1691 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1692 cr, vsecp, &acl_ids)) != 0)
1696 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1697 zfs_acl_ids_free(&acl_ids);
1698 error = SET_ERROR(EDQUOT);
1702 tx = dmu_tx_create(os);
1704 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1705 ZFS_SA_BASE_ATTR_SIZE);
1707 fuid_dirtied = zfsvfs->z_fuid_dirty;
1709 zfs_fuid_txhold(zfsvfs, tx);
1710 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1711 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1712 if (!zfsvfs->z_use_sa &&
1713 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1714 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1715 0, acl_ids.z_aclp->z_acl_bytes);
1717 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1719 zfs_dirent_unlock(dl);
1720 if (error == ERESTART) {
1726 zfs_acl_ids_free(&acl_ids);
1728 getnewvnode_drop_reserve();
1732 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1735 zfs_fuid_sync(zfsvfs, tx);
1737 (void) zfs_link_create(dl, zp, tx, ZNEW);
1738 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1739 if (flag & FIGNORECASE)
1741 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1742 vsecp, acl_ids.z_fuidp, vap);
1743 zfs_acl_ids_free(&acl_ids);
1746 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1749 zfs_acl_ids_free(&acl_ids);
1753 * A directory entry already exists for this name.
1756 * Can't truncate an existing file if in exclusive mode.
1759 error = SET_ERROR(EEXIST);
1763 * Can't open a directory for writing.
1765 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1766 error = SET_ERROR(EISDIR);
1770 * Verify requested access to file.
1772 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1776 mutex_enter(&dzp->z_lock);
1778 mutex_exit(&dzp->z_lock);
1781 * Truncate regular files if requested.
1783 if ((ZTOV(zp)->v_type == VREG) &&
1784 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1785 /* we can't hold any locks when calling zfs_freesp() */
1786 zfs_dirent_unlock(dl);
1788 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1790 vnevent_create(ZTOV(zp), ct);
1795 getnewvnode_drop_reserve();
1797 zfs_dirent_unlock(dl);
1804 error = specvp_check(vpp, cr);
1807 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1808 zil_commit(zilog, 0);
1815 * Remove an entry from a directory.
1817 * IN: dvp - vnode of directory to remove entry from.
1818 * name - name of entry to remove.
1819 * cr - credentials of caller.
1820 * ct - caller context
1821 * flags - case flags
1823 * RETURN: 0 on success, error code on failure.
1827 * vp - ctime (if nlink > 0)
1830 uint64_t null_xattr = 0;
1834 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1837 znode_t *zp, *dzp = VTOZ(dvp);
1840 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1842 uint64_t acl_obj, xattr_obj;
1843 uint64_t xattr_obj_unlinked = 0;
1847 boolean_t may_delete_now, delete_now = FALSE;
1848 boolean_t unlinked, toobig = FALSE;
1850 pathname_t *realnmp = NULL;
1854 boolean_t waited = B_FALSE;
1858 zilog = zfsvfs->z_log;
1860 if (flags & FIGNORECASE) {
1870 * Attempt to lock directory; fail if entry doesn't exist.
1872 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1882 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1887 * Need to use rmdir for removing directories.
1889 if (vp->v_type == VDIR) {
1890 error = SET_ERROR(EPERM);
1894 vnevent_remove(vp, dvp, name, ct);
1897 dnlc_remove(dvp, realnmp->pn_buf);
1899 dnlc_remove(dvp, name);
1902 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1906 * We may delete the znode now, or we may put it in the unlinked set;
1907 * it depends on whether we're the last link, and on whether there are
1908 * other holds on the vnode. So we dmu_tx_hold() the right things to
1909 * allow for either case.
1912 tx = dmu_tx_create(zfsvfs->z_os);
1913 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1914 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1915 zfs_sa_upgrade_txholds(tx, zp);
1916 zfs_sa_upgrade_txholds(tx, dzp);
1917 if (may_delete_now) {
1919 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1920 /* if the file is too big, only hold_free a token amount */
1921 dmu_tx_hold_free(tx, zp->z_id, 0,
1922 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1925 /* are there any extended attributes? */
1926 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1927 &xattr_obj, sizeof (xattr_obj));
1928 if (error == 0 && xattr_obj) {
1929 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1931 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1932 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1935 mutex_enter(&zp->z_lock);
1936 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1937 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1938 mutex_exit(&zp->z_lock);
1940 /* charge as an update -- would be nice not to charge at all */
1941 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1943 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1945 zfs_dirent_unlock(dl);
1949 if (error == ERESTART) {
1963 * Remove the directory entry.
1965 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1975 * Hold z_lock so that we can make sure that the ACL obj
1976 * hasn't changed. Could have been deleted due to
1979 mutex_enter(&zp->z_lock);
1981 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1982 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1983 delete_now = may_delete_now && !toobig &&
1984 vp->v_count == 1 && !vn_has_cached_data(vp) &&
1985 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1992 panic("zfs_remove: delete_now branch taken");
1994 if (xattr_obj_unlinked) {
1995 ASSERT3U(xzp->z_links, ==, 2);
1996 mutex_enter(&xzp->z_lock);
1997 xzp->z_unlinked = 1;
1999 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
2000 &xzp->z_links, sizeof (xzp->z_links), tx);
2001 ASSERT3U(error, ==, 0);
2002 mutex_exit(&xzp->z_lock);
2003 zfs_unlinked_add(xzp, tx);
2006 error = sa_remove(zp->z_sa_hdl,
2007 SA_ZPL_XATTR(zfsvfs), tx);
2009 error = sa_update(zp->z_sa_hdl,
2010 SA_ZPL_XATTR(zfsvfs), &null_xattr,
2011 sizeof (uint64_t), tx);
2016 ASSERT0(vp->v_count);
2018 mutex_exit(&zp->z_lock);
2019 zfs_znode_delete(zp, tx);
2020 } else if (unlinked) {
2021 mutex_exit(&zp->z_lock);
2022 zfs_unlinked_add(zp, tx);
2024 vp->v_vflag |= VV_NOSYNC;
2029 if (flags & FIGNORECASE)
2031 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
2038 zfs_dirent_unlock(dl);
2045 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2046 zil_commit(zilog, 0);
2053 * Create a new directory and insert it into dvp using the name
2054 * provided. Return a pointer to the inserted directory.
2056 * IN: dvp - vnode of directory to add subdir to.
2057 * dirname - name of new directory.
2058 * vap - attributes of new directory.
2059 * cr - credentials of caller.
2060 * ct - caller context
2061 * flags - case flags
2062 * vsecp - ACL to be set
2064 * OUT: vpp - vnode of created directory.
2066 * RETURN: 0 on success, error code on failure.
2069 * dvp - ctime|mtime updated
2070 * vp - ctime|mtime|atime updated
2074 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
2075 caller_context_t *ct, int flags, vsecattr_t *vsecp)
2077 znode_t *zp, *dzp = VTOZ(dvp);
2078 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2087 gid_t gid = crgetgid(cr);
2088 zfs_acl_ids_t acl_ids;
2089 boolean_t fuid_dirtied;
2090 boolean_t waited = B_FALSE;
2092 ASSERT(vap->va_type == VDIR);
2095 * If we have an ephemeral id, ACL, or XVATTR then
2096 * make sure file system is at proper version
2099 ksid = crgetsid(cr, KSID_OWNER);
2101 uid = ksid_getid(ksid);
2104 if (zfsvfs->z_use_fuids == B_FALSE &&
2105 (vsecp || (vap->va_mask & AT_XVATTR) ||
2106 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
2107 return (SET_ERROR(EINVAL));
2111 zilog = zfsvfs->z_log;
2113 if (dzp->z_pflags & ZFS_XATTR) {
2115 return (SET_ERROR(EINVAL));
2118 if (zfsvfs->z_utf8 && u8_validate(dirname,
2119 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2121 return (SET_ERROR(EILSEQ));
2123 if (flags & FIGNORECASE)
2126 if (vap->va_mask & AT_XVATTR) {
2127 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
2128 crgetuid(cr), cr, vap->va_type)) != 0) {
2134 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
2135 vsecp, &acl_ids)) != 0) {
2140 getnewvnode_reserve(1);
2143 * First make sure the new directory doesn't exist.
2145 * Existence is checked first to make sure we don't return
2146 * EACCES instead of EEXIST which can cause some applications
2152 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
2154 zfs_acl_ids_free(&acl_ids);
2155 getnewvnode_drop_reserve();
2160 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
2161 zfs_acl_ids_free(&acl_ids);
2162 zfs_dirent_unlock(dl);
2163 getnewvnode_drop_reserve();
2168 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
2169 zfs_acl_ids_free(&acl_ids);
2170 zfs_dirent_unlock(dl);
2171 getnewvnode_drop_reserve();
2173 return (SET_ERROR(EDQUOT));
2177 * Add a new entry to the directory.
2179 tx = dmu_tx_create(zfsvfs->z_os);
2180 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2181 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2182 fuid_dirtied = zfsvfs->z_fuid_dirty;
2184 zfs_fuid_txhold(zfsvfs, tx);
2185 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2186 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2187 acl_ids.z_aclp->z_acl_bytes);
2190 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2191 ZFS_SA_BASE_ATTR_SIZE);
2193 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2195 zfs_dirent_unlock(dl);
2196 if (error == ERESTART) {
2202 zfs_acl_ids_free(&acl_ids);
2204 getnewvnode_drop_reserve();
2212 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2215 zfs_fuid_sync(zfsvfs, tx);
2218 * Now put new name in parent dir.
2220 (void) zfs_link_create(dl, zp, tx, ZNEW);
2224 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2225 if (flags & FIGNORECASE)
2227 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2228 acl_ids.z_fuidp, vap);
2230 zfs_acl_ids_free(&acl_ids);
2234 getnewvnode_drop_reserve();
2236 zfs_dirent_unlock(dl);
2238 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2239 zil_commit(zilog, 0);
2246 * Remove a directory subdir entry. If the current working
2247 * directory is the same as the subdir to be removed, the
2250 * IN: dvp - vnode of directory to remove from.
2251 * name - name of directory to be removed.
2252 * cwd - vnode of current working directory.
2253 * cr - credentials of caller.
2254 * ct - caller context
2255 * flags - case flags
2257 * RETURN: 0 on success, error code on failure.
2260 * dvp - ctime|mtime updated
2264 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2265 caller_context_t *ct, int flags)
2267 znode_t *dzp = VTOZ(dvp);
2270 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2276 boolean_t waited = B_FALSE;
2280 zilog = zfsvfs->z_log;
2282 if (flags & FIGNORECASE)
2288 * Attempt to lock directory; fail if entry doesn't exist.
2290 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2298 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2302 if (vp->v_type != VDIR) {
2303 error = SET_ERROR(ENOTDIR);
2308 error = SET_ERROR(EINVAL);
2312 vnevent_rmdir(vp, dvp, name, ct);
2315 * Grab a lock on the directory to make sure that noone is
2316 * trying to add (or lookup) entries while we are removing it.
2318 rw_enter(&zp->z_name_lock, RW_WRITER);
2321 * Grab a lock on the parent pointer to make sure we play well
2322 * with the treewalk and directory rename code.
2324 rw_enter(&zp->z_parent_lock, RW_WRITER);
2326 tx = dmu_tx_create(zfsvfs->z_os);
2327 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2328 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2329 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2330 zfs_sa_upgrade_txholds(tx, zp);
2331 zfs_sa_upgrade_txholds(tx, dzp);
2332 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2334 rw_exit(&zp->z_parent_lock);
2335 rw_exit(&zp->z_name_lock);
2336 zfs_dirent_unlock(dl);
2338 if (error == ERESTART) {
2349 #ifdef FREEBSD_NAMECACHE
2353 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2356 uint64_t txtype = TX_RMDIR;
2357 if (flags & FIGNORECASE)
2359 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2364 rw_exit(&zp->z_parent_lock);
2365 rw_exit(&zp->z_name_lock);
2366 #ifdef FREEBSD_NAMECACHE
2370 zfs_dirent_unlock(dl);
2374 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2375 zil_commit(zilog, 0);
2382 * Read as many directory entries as will fit into the provided
2383 * buffer from the given directory cursor position (specified in
2384 * the uio structure).
2386 * IN: vp - vnode of directory to read.
2387 * uio - structure supplying read location, range info,
2388 * and return buffer.
2389 * cr - credentials of caller.
2390 * ct - caller context
2391 * flags - case flags
2393 * OUT: uio - updated offset and range, buffer filled.
2394 * eofp - set to true if end-of-file detected.
2396 * RETURN: 0 on success, error code on failure.
2399 * vp - atime updated
2401 * Note that the low 4 bits of the cookie returned by zap is always zero.
2402 * This allows us to use the low range for "special" directory entries:
2403 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2404 * we use the offset 2 for the '.zfs' directory.
2408 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies)
2410 znode_t *zp = VTOZ(vp);
2414 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2419 zap_attribute_t zap;
2420 uint_t bytes_wanted;
2421 uint64_t offset; /* must be unsigned; checks for < 1 */
2427 boolean_t check_sysattrs;
2430 u_long *cooks = NULL;
2436 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2437 &parent, sizeof (parent))) != 0) {
2443 * If we are not given an eof variable,
2450 * Check for valid iov_len.
2452 if (uio->uio_iov->iov_len <= 0) {
2454 return (SET_ERROR(EINVAL));
2458 * Quit if directory has been removed (posix)
2460 if ((*eofp = zp->z_unlinked) != 0) {
2467 offset = uio->uio_loffset;
2468 prefetch = zp->z_zn_prefetch;
2471 * Initialize the iterator cursor.
2475 * Start iteration from the beginning of the directory.
2477 zap_cursor_init(&zc, os, zp->z_id);
2480 * The offset is a serialized cursor.
2482 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2486 * Get space to change directory entries into fs independent format.
2488 iovp = uio->uio_iov;
2489 bytes_wanted = iovp->iov_len;
2490 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2491 bufsize = bytes_wanted;
2492 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2493 odp = (struct dirent64 *)outbuf;
2495 bufsize = bytes_wanted;
2497 odp = (struct dirent64 *)iovp->iov_base;
2499 eodp = (struct edirent *)odp;
2501 if (ncookies != NULL) {
2503 * Minimum entry size is dirent size and 1 byte for a file name.
2505 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1);
2506 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK);
2511 * If this VFS supports the system attribute view interface; and
2512 * we're looking at an extended attribute directory; and we care
2513 * about normalization conflicts on this vfs; then we must check
2514 * for normalization conflicts with the sysattr name space.
2517 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2518 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2519 (flags & V_RDDIR_ENTFLAGS);
2525 * Transform to file-system independent format
2528 while (outcount < bytes_wanted) {
2531 off64_t *next = NULL;
2534 * Special case `.', `..', and `.zfs'.
2537 (void) strcpy(zap.za_name, ".");
2538 zap.za_normalization_conflict = 0;
2541 } else if (offset == 1) {
2542 (void) strcpy(zap.za_name, "..");
2543 zap.za_normalization_conflict = 0;
2546 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2547 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2548 zap.za_normalization_conflict = 0;
2549 objnum = ZFSCTL_INO_ROOT;
2555 if (error = zap_cursor_retrieve(&zc, &zap)) {
2556 if ((*eofp = (error == ENOENT)) != 0)
2562 if (zap.za_integer_length != 8 ||
2563 zap.za_num_integers != 1) {
2564 cmn_err(CE_WARN, "zap_readdir: bad directory "
2565 "entry, obj = %lld, offset = %lld\n",
2566 (u_longlong_t)zp->z_id,
2567 (u_longlong_t)offset);
2568 error = SET_ERROR(ENXIO);
2572 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2574 * MacOS X can extract the object type here such as:
2575 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2577 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2579 if (check_sysattrs && !zap.za_normalization_conflict) {
2581 zap.za_normalization_conflict =
2582 xattr_sysattr_casechk(zap.za_name);
2584 panic("%s:%u: TODO", __func__, __LINE__);
2589 if (flags & V_RDDIR_ACCFILTER) {
2591 * If we have no access at all, don't include
2592 * this entry in the returned information
2595 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2597 if (!zfs_has_access(ezp, cr)) {
2604 if (flags & V_RDDIR_ENTFLAGS)
2605 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2607 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2610 * Will this entry fit in the buffer?
2612 if (outcount + reclen > bufsize) {
2614 * Did we manage to fit anything in the buffer?
2617 error = SET_ERROR(EINVAL);
2622 if (flags & V_RDDIR_ENTFLAGS) {
2624 * Add extended flag entry:
2626 eodp->ed_ino = objnum;
2627 eodp->ed_reclen = reclen;
2628 /* NOTE: ed_off is the offset for the *next* entry */
2629 next = &(eodp->ed_off);
2630 eodp->ed_eflags = zap.za_normalization_conflict ?
2631 ED_CASE_CONFLICT : 0;
2632 (void) strncpy(eodp->ed_name, zap.za_name,
2633 EDIRENT_NAMELEN(reclen));
2634 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2639 odp->d_ino = objnum;
2640 odp->d_reclen = reclen;
2641 odp->d_namlen = strlen(zap.za_name);
2642 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1);
2644 odp = (dirent64_t *)((intptr_t)odp + reclen);
2648 ASSERT(outcount <= bufsize);
2650 /* Prefetch znode */
2652 dmu_prefetch(os, objnum, 0, 0);
2656 * Move to the next entry, fill in the previous offset.
2658 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2659 zap_cursor_advance(&zc);
2660 offset = zap_cursor_serialize(&zc);
2665 if (cooks != NULL) {
2668 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks));
2671 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2673 /* Subtract unused cookies */
2674 if (ncookies != NULL)
2675 *ncookies -= ncooks;
2677 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2678 iovp->iov_base += outcount;
2679 iovp->iov_len -= outcount;
2680 uio->uio_resid -= outcount;
2681 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2683 * Reset the pointer.
2685 offset = uio->uio_loffset;
2689 zap_cursor_fini(&zc);
2690 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2691 kmem_free(outbuf, bufsize);
2693 if (error == ENOENT)
2696 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2698 uio->uio_loffset = offset;
2700 if (error != 0 && cookies != NULL) {
2701 free(*cookies, M_TEMP);
2708 ulong_t zfs_fsync_sync_cnt = 4;
2711 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2713 znode_t *zp = VTOZ(vp);
2714 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2716 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2718 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2721 zil_commit(zfsvfs->z_log, zp->z_id);
2729 * Get the requested file attributes and place them in the provided
2732 * IN: vp - vnode of file.
2733 * vap - va_mask identifies requested attributes.
2734 * If AT_XVATTR set, then optional attrs are requested
2735 * flags - ATTR_NOACLCHECK (CIFS server context)
2736 * cr - credentials of caller.
2737 * ct - caller context
2739 * OUT: vap - attribute values.
2741 * RETURN: 0 (always succeeds).
2745 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2746 caller_context_t *ct)
2748 znode_t *zp = VTOZ(vp);
2749 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2752 u_longlong_t nblocks;
2754 uint64_t mtime[2], ctime[2], crtime[2], rdev;
2755 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2756 xoptattr_t *xoap = NULL;
2757 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2758 sa_bulk_attr_t bulk[4];
2764 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2766 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2767 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2768 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
2769 if (vp->v_type == VBLK || vp->v_type == VCHR)
2770 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
2773 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2779 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2780 * Also, if we are the owner don't bother, since owner should
2781 * always be allowed to read basic attributes of file.
2783 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2784 (vap->va_uid != crgetuid(cr))) {
2785 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2793 * Return all attributes. It's cheaper to provide the answer
2794 * than to determine whether we were asked the question.
2797 mutex_enter(&zp->z_lock);
2798 vap->va_type = IFTOVT(zp->z_mode);
2799 vap->va_mode = zp->z_mode & ~S_IFMT;
2801 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2803 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
2805 vap->va_nodeid = zp->z_id;
2806 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2807 links = zp->z_links + 1;
2809 links = zp->z_links;
2810 vap->va_nlink = MIN(links, LINK_MAX); /* nlink_t limit! */
2811 vap->va_size = zp->z_size;
2813 vap->va_rdev = vp->v_rdev;
2815 if (vp->v_type == VBLK || vp->v_type == VCHR)
2816 vap->va_rdev = zfs_cmpldev(rdev);
2818 vap->va_seq = zp->z_seq;
2819 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */
2822 * Add in any requested optional attributes and the create time.
2823 * Also set the corresponding bits in the returned attribute bitmap.
2825 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2826 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2828 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2829 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2832 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2833 xoap->xoa_readonly =
2834 ((zp->z_pflags & ZFS_READONLY) != 0);
2835 XVA_SET_RTN(xvap, XAT_READONLY);
2838 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2840 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2841 XVA_SET_RTN(xvap, XAT_SYSTEM);
2844 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2846 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2847 XVA_SET_RTN(xvap, XAT_HIDDEN);
2850 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2851 xoap->xoa_nounlink =
2852 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2853 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2856 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2857 xoap->xoa_immutable =
2858 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2859 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2862 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2863 xoap->xoa_appendonly =
2864 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2865 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2868 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2870 ((zp->z_pflags & ZFS_NODUMP) != 0);
2871 XVA_SET_RTN(xvap, XAT_NODUMP);
2874 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2876 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2877 XVA_SET_RTN(xvap, XAT_OPAQUE);
2880 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2881 xoap->xoa_av_quarantined =
2882 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2883 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2886 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2887 xoap->xoa_av_modified =
2888 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2889 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2892 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2893 vp->v_type == VREG) {
2894 zfs_sa_get_scanstamp(zp, xvap);
2897 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2900 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2901 times, sizeof (times));
2902 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2903 XVA_SET_RTN(xvap, XAT_CREATETIME);
2906 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2907 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2908 XVA_SET_RTN(xvap, XAT_REPARSE);
2910 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2911 xoap->xoa_generation = zp->z_gen;
2912 XVA_SET_RTN(xvap, XAT_GEN);
2915 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2917 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2918 XVA_SET_RTN(xvap, XAT_OFFLINE);
2921 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2923 ((zp->z_pflags & ZFS_SPARSE) != 0);
2924 XVA_SET_RTN(xvap, XAT_SPARSE);
2928 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2929 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2930 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2931 ZFS_TIME_DECODE(&vap->va_birthtime, crtime);
2933 mutex_exit(&zp->z_lock);
2935 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2936 vap->va_blksize = blksize;
2937 vap->va_bytes = nblocks << 9; /* nblocks * 512 */
2939 if (zp->z_blksz == 0) {
2941 * Block size hasn't been set; suggest maximal I/O transfers.
2943 vap->va_blksize = zfsvfs->z_max_blksz;
2951 * Set the file attributes to the values contained in the
2954 * IN: vp - vnode of file to be modified.
2955 * vap - new attribute values.
2956 * If AT_XVATTR set, then optional attrs are being set
2957 * flags - ATTR_UTIME set if non-default time values provided.
2958 * - ATTR_NOACLCHECK (CIFS context only).
2959 * cr - credentials of caller.
2960 * ct - caller context
2962 * RETURN: 0 on success, error code on failure.
2965 * vp - ctime updated, mtime updated if size changed.
2969 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2970 caller_context_t *ct)
2972 znode_t *zp = VTOZ(vp);
2973 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2978 uint_t mask = vap->va_mask;
2979 uint_t saved_mask = 0;
2980 uint64_t saved_mode;
2983 uint64_t new_uid, new_gid;
2985 uint64_t mtime[2], ctime[2];
2987 int need_policy = FALSE;
2989 zfs_fuid_info_t *fuidp = NULL;
2990 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2993 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2994 boolean_t fuid_dirtied = B_FALSE;
2995 sa_bulk_attr_t bulk[7], xattr_bulk[7];
2996 int count = 0, xattr_count = 0;
3001 if (mask & AT_NOSET)
3002 return (SET_ERROR(EINVAL));
3007 zilog = zfsvfs->z_log;
3010 * Make sure that if we have ephemeral uid/gid or xvattr specified
3011 * that file system is at proper version level
3014 if (zfsvfs->z_use_fuids == B_FALSE &&
3015 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
3016 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
3017 (mask & AT_XVATTR))) {
3019 return (SET_ERROR(EINVAL));
3022 if (mask & AT_SIZE && vp->v_type == VDIR) {
3024 return (SET_ERROR(EISDIR));
3027 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
3029 return (SET_ERROR(EINVAL));
3033 * If this is an xvattr_t, then get a pointer to the structure of
3034 * optional attributes. If this is NULL, then we have a vattr_t.
3036 xoap = xva_getxoptattr(xvap);
3038 xva_init(&tmpxvattr);
3041 * Immutable files can only alter immutable bit and atime
3043 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
3044 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
3045 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
3047 return (SET_ERROR(EPERM));
3050 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
3052 return (SET_ERROR(EPERM));
3056 * Verify timestamps doesn't overflow 32 bits.
3057 * ZFS can handle large timestamps, but 32bit syscalls can't
3058 * handle times greater than 2039. This check should be removed
3059 * once large timestamps are fully supported.
3061 if (mask & (AT_ATIME | AT_MTIME)) {
3062 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
3063 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
3065 return (SET_ERROR(EOVERFLOW));
3073 /* Can this be moved to before the top label? */
3074 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
3076 return (SET_ERROR(EROFS));
3080 * First validate permissions
3083 if (mask & AT_SIZE) {
3085 * XXX - Note, we are not providing any open
3086 * mode flags here (like FNDELAY), so we may
3087 * block if there are locks present... this
3088 * should be addressed in openat().
3090 /* XXX - would it be OK to generate a log record here? */
3091 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
3098 if (mask & (AT_ATIME|AT_MTIME) ||
3099 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
3100 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
3101 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
3102 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
3103 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
3104 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
3105 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
3106 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
3110 if (mask & (AT_UID|AT_GID)) {
3111 int idmask = (mask & (AT_UID|AT_GID));
3116 * NOTE: even if a new mode is being set,
3117 * we may clear S_ISUID/S_ISGID bits.
3120 if (!(mask & AT_MODE))
3121 vap->va_mode = zp->z_mode;
3124 * Take ownership or chgrp to group we are a member of
3127 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
3128 take_group = (mask & AT_GID) &&
3129 zfs_groupmember(zfsvfs, vap->va_gid, cr);
3132 * If both AT_UID and AT_GID are set then take_owner and
3133 * take_group must both be set in order to allow taking
3136 * Otherwise, send the check through secpolicy_vnode_setattr()
3140 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
3141 ((idmask == AT_UID) && take_owner) ||
3142 ((idmask == AT_GID) && take_group)) {
3143 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
3144 skipaclchk, cr) == 0) {
3146 * Remove setuid/setgid for non-privileged users
3148 secpolicy_setid_clear(vap, vp, cr);
3149 trim_mask = (mask & (AT_UID|AT_GID));
3158 mutex_enter(&zp->z_lock);
3159 oldva.va_mode = zp->z_mode;
3160 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
3161 if (mask & AT_XVATTR) {
3163 * Update xvattr mask to include only those attributes
3164 * that are actually changing.
3166 * the bits will be restored prior to actually setting
3167 * the attributes so the caller thinks they were set.
3169 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
3170 if (xoap->xoa_appendonly !=
3171 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
3174 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
3175 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
3179 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
3180 if (xoap->xoa_nounlink !=
3181 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
3184 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
3185 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
3189 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
3190 if (xoap->xoa_immutable !=
3191 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
3194 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
3195 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
3199 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
3200 if (xoap->xoa_nodump !=
3201 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
3204 XVA_CLR_REQ(xvap, XAT_NODUMP);
3205 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
3209 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3210 if (xoap->xoa_av_modified !=
3211 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3214 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3215 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
3219 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3220 if ((vp->v_type != VREG &&
3221 xoap->xoa_av_quarantined) ||
3222 xoap->xoa_av_quarantined !=
3223 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3226 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3227 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
3231 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3232 mutex_exit(&zp->z_lock);
3234 return (SET_ERROR(EPERM));
3237 if (need_policy == FALSE &&
3238 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3239 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3244 mutex_exit(&zp->z_lock);
3246 if (mask & AT_MODE) {
3247 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3248 err = secpolicy_setid_setsticky_clear(vp, vap,
3254 trim_mask |= AT_MODE;
3262 * If trim_mask is set then take ownership
3263 * has been granted or write_acl is present and user
3264 * has the ability to modify mode. In that case remove
3265 * UID|GID and or MODE from mask so that
3266 * secpolicy_vnode_setattr() doesn't revoke it.
3270 saved_mask = vap->va_mask;
3271 vap->va_mask &= ~trim_mask;
3272 if (trim_mask & AT_MODE) {
3274 * Save the mode, as secpolicy_vnode_setattr()
3275 * will overwrite it with ova.va_mode.
3277 saved_mode = vap->va_mode;
3280 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3281 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3288 vap->va_mask |= saved_mask;
3289 if (trim_mask & AT_MODE) {
3291 * Recover the mode after
3292 * secpolicy_vnode_setattr().
3294 vap->va_mode = saved_mode;
3300 * secpolicy_vnode_setattr, or take ownership may have
3303 mask = vap->va_mask;
3305 if ((mask & (AT_UID | AT_GID))) {
3306 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3307 &xattr_obj, sizeof (xattr_obj));
3309 if (err == 0 && xattr_obj) {
3310 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3314 if (mask & AT_UID) {
3315 new_uid = zfs_fuid_create(zfsvfs,
3316 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3317 if (new_uid != zp->z_uid &&
3318 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3320 VN_RELE(ZTOV(attrzp));
3321 err = SET_ERROR(EDQUOT);
3326 if (mask & AT_GID) {
3327 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3328 cr, ZFS_GROUP, &fuidp);
3329 if (new_gid != zp->z_gid &&
3330 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3332 VN_RELE(ZTOV(attrzp));
3333 err = SET_ERROR(EDQUOT);
3338 tx = dmu_tx_create(zfsvfs->z_os);
3340 if (mask & AT_MODE) {
3341 uint64_t pmode = zp->z_mode;
3343 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3345 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3346 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3347 err = SET_ERROR(EPERM);
3351 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3354 mutex_enter(&zp->z_lock);
3355 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3357 * Are we upgrading ACL from old V0 format
3360 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3361 zfs_znode_acl_version(zp) ==
3362 ZFS_ACL_VERSION_INITIAL) {
3363 dmu_tx_hold_free(tx, acl_obj, 0,
3365 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3366 0, aclp->z_acl_bytes);
3368 dmu_tx_hold_write(tx, acl_obj, 0,
3371 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3372 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3373 0, aclp->z_acl_bytes);
3375 mutex_exit(&zp->z_lock);
3376 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3378 if ((mask & AT_XVATTR) &&
3379 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3380 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3382 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3386 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3389 fuid_dirtied = zfsvfs->z_fuid_dirty;
3391 zfs_fuid_txhold(zfsvfs, tx);
3393 zfs_sa_upgrade_txholds(tx, zp);
3395 err = dmu_tx_assign(tx, TXG_WAIT);
3401 * Set each attribute requested.
3402 * We group settings according to the locks they need to acquire.
3404 * Note: you cannot set ctime directly, although it will be
3405 * updated as a side-effect of calling this function.
3409 if (mask & (AT_UID|AT_GID|AT_MODE))
3410 mutex_enter(&zp->z_acl_lock);
3411 mutex_enter(&zp->z_lock);
3413 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3414 &zp->z_pflags, sizeof (zp->z_pflags));
3417 if (mask & (AT_UID|AT_GID|AT_MODE))
3418 mutex_enter(&attrzp->z_acl_lock);
3419 mutex_enter(&attrzp->z_lock);
3420 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3421 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3422 sizeof (attrzp->z_pflags));
3425 if (mask & (AT_UID|AT_GID)) {
3427 if (mask & AT_UID) {
3428 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3429 &new_uid, sizeof (new_uid));
3430 zp->z_uid = new_uid;
3432 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3433 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3435 attrzp->z_uid = new_uid;
3439 if (mask & AT_GID) {
3440 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3441 NULL, &new_gid, sizeof (new_gid));
3442 zp->z_gid = new_gid;
3444 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3445 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3447 attrzp->z_gid = new_gid;
3450 if (!(mask & AT_MODE)) {
3451 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3452 NULL, &new_mode, sizeof (new_mode));
3453 new_mode = zp->z_mode;
3455 err = zfs_acl_chown_setattr(zp);
3458 err = zfs_acl_chown_setattr(attrzp);
3463 if (mask & AT_MODE) {
3464 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3465 &new_mode, sizeof (new_mode));
3466 zp->z_mode = new_mode;
3467 ASSERT3U((uintptr_t)aclp, !=, 0);
3468 err = zfs_aclset_common(zp, aclp, cr, tx);
3470 if (zp->z_acl_cached)
3471 zfs_acl_free(zp->z_acl_cached);
3472 zp->z_acl_cached = aclp;
3477 if (mask & AT_ATIME) {
3478 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3479 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3480 &zp->z_atime, sizeof (zp->z_atime));
3483 if (mask & AT_MTIME) {
3484 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3485 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3486 mtime, sizeof (mtime));
3489 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3490 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3491 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3492 NULL, mtime, sizeof (mtime));
3493 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3494 &ctime, sizeof (ctime));
3495 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3497 } else if (mask != 0) {
3498 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3499 &ctime, sizeof (ctime));
3500 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3503 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3504 SA_ZPL_CTIME(zfsvfs), NULL,
3505 &ctime, sizeof (ctime));
3506 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3507 mtime, ctime, B_TRUE);
3511 * Do this after setting timestamps to prevent timestamp
3512 * update from toggling bit
3515 if (xoap && (mask & AT_XVATTR)) {
3518 * restore trimmed off masks
3519 * so that return masks can be set for caller.
3522 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3523 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3525 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3526 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3528 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3529 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3531 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3532 XVA_SET_REQ(xvap, XAT_NODUMP);
3534 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3535 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3537 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3538 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3541 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3542 ASSERT(vp->v_type == VREG);
3544 zfs_xvattr_set(zp, xvap, tx);
3548 zfs_fuid_sync(zfsvfs, tx);
3551 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3553 mutex_exit(&zp->z_lock);
3554 if (mask & (AT_UID|AT_GID|AT_MODE))
3555 mutex_exit(&zp->z_acl_lock);
3558 if (mask & (AT_UID|AT_GID|AT_MODE))
3559 mutex_exit(&attrzp->z_acl_lock);
3560 mutex_exit(&attrzp->z_lock);
3563 if (err == 0 && attrzp) {
3564 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3570 VN_RELE(ZTOV(attrzp));
3576 zfs_fuid_info_free(fuidp);
3582 if (err == ERESTART)
3585 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3590 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3591 zil_commit(zilog, 0);
3597 typedef struct zfs_zlock {
3598 krwlock_t *zl_rwlock; /* lock we acquired */
3599 znode_t *zl_znode; /* znode we held */
3600 struct zfs_zlock *zl_next; /* next in list */
3604 * Drop locks and release vnodes that were held by zfs_rename_lock().
3607 zfs_rename_unlock(zfs_zlock_t **zlpp)
3611 while ((zl = *zlpp) != NULL) {
3612 if (zl->zl_znode != NULL)
3613 VN_RELE(ZTOV(zl->zl_znode));
3614 rw_exit(zl->zl_rwlock);
3615 *zlpp = zl->zl_next;
3616 kmem_free(zl, sizeof (*zl));
3621 * Search back through the directory tree, using the ".." entries.
3622 * Lock each directory in the chain to prevent concurrent renames.
3623 * Fail any attempt to move a directory into one of its own descendants.
3624 * XXX - z_parent_lock can overlap with map or grow locks
3627 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3631 uint64_t rootid = zp->z_zfsvfs->z_root;
3632 uint64_t oidp = zp->z_id;
3633 krwlock_t *rwlp = &szp->z_parent_lock;
3634 krw_t rw = RW_WRITER;
3637 * First pass write-locks szp and compares to zp->z_id.
3638 * Later passes read-lock zp and compare to zp->z_parent.
3641 if (!rw_tryenter(rwlp, rw)) {
3643 * Another thread is renaming in this path.
3644 * Note that if we are a WRITER, we don't have any
3645 * parent_locks held yet.
3647 if (rw == RW_READER && zp->z_id > szp->z_id) {
3649 * Drop our locks and restart
3651 zfs_rename_unlock(&zl);
3655 rwlp = &szp->z_parent_lock;
3660 * Wait for other thread to drop its locks
3666 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3667 zl->zl_rwlock = rwlp;
3668 zl->zl_znode = NULL;
3669 zl->zl_next = *zlpp;
3672 if (oidp == szp->z_id) /* We're a descendant of szp */
3673 return (SET_ERROR(EINVAL));
3675 if (oidp == rootid) /* We've hit the top */
3678 if (rw == RW_READER) { /* i.e. not the first pass */
3679 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3684 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3685 &oidp, sizeof (oidp));
3686 rwlp = &zp->z_parent_lock;
3689 } while (zp->z_id != sdzp->z_id);
3695 * Move an entry from the provided source directory to the target
3696 * directory. Change the entry name as indicated.
3698 * IN: sdvp - Source directory containing the "old entry".
3699 * snm - Old entry name.
3700 * tdvp - Target directory to contain the "new entry".
3701 * tnm - New entry name.
3702 * cr - credentials of caller.
3703 * ct - caller context
3704 * flags - case flags
3706 * RETURN: 0 on success, error code on failure.
3709 * sdvp,tdvp - ctime|mtime updated
3713 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3714 caller_context_t *ct, int flags)
3716 znode_t *tdzp, *szp, *tzp;
3717 znode_t *sdzp = VTOZ(sdvp);
3718 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs;
3721 zfs_dirlock_t *sdl, *tdl;
3724 int cmp, serr, terr;
3727 boolean_t waited = B_FALSE;
3730 ZFS_VERIFY_ZP(sdzp);
3731 zilog = zfsvfs->z_log;
3734 * Make sure we have the real vp for the target directory.
3736 if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3740 ZFS_VERIFY_ZP(tdzp);
3743 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3744 * ctldir appear to have the same v_vfsp.
3746 if (tdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) {
3748 return (SET_ERROR(EXDEV));
3751 if (zfsvfs->z_utf8 && u8_validate(tnm,
3752 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3754 return (SET_ERROR(EILSEQ));
3757 if (flags & FIGNORECASE)
3766 * This is to prevent the creation of links into attribute space
3767 * by renaming a linked file into/outof an attribute directory.
3768 * See the comment in zfs_link() for why this is considered bad.
3770 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3772 return (SET_ERROR(EINVAL));
3776 * Lock source and target directory entries. To prevent deadlock,
3777 * a lock ordering must be defined. We lock the directory with
3778 * the smallest object id first, or if it's a tie, the one with
3779 * the lexically first name.
3781 if (sdzp->z_id < tdzp->z_id) {
3783 } else if (sdzp->z_id > tdzp->z_id) {
3787 * First compare the two name arguments without
3788 * considering any case folding.
3790 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3792 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3793 ASSERT(error == 0 || !zfsvfs->z_utf8);
3796 * POSIX: "If the old argument and the new argument
3797 * both refer to links to the same existing file,
3798 * the rename() function shall return successfully
3799 * and perform no other action."
3805 * If the file system is case-folding, then we may
3806 * have some more checking to do. A case-folding file
3807 * system is either supporting mixed case sensitivity
3808 * access or is completely case-insensitive. Note
3809 * that the file system is always case preserving.
3811 * In mixed sensitivity mode case sensitive behavior
3812 * is the default. FIGNORECASE must be used to
3813 * explicitly request case insensitive behavior.
3815 * If the source and target names provided differ only
3816 * by case (e.g., a request to rename 'tim' to 'Tim'),
3817 * we will treat this as a special case in the
3818 * case-insensitive mode: as long as the source name
3819 * is an exact match, we will allow this to proceed as
3820 * a name-change request.
3822 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3823 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3824 flags & FIGNORECASE)) &&
3825 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3828 * case preserving rename request, require exact
3837 * If the source and destination directories are the same, we should
3838 * grab the z_name_lock of that directory only once.
3842 rw_enter(&sdzp->z_name_lock, RW_READER);
3846 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3847 ZEXISTS | zflg, NULL, NULL);
3848 terr = zfs_dirent_lock(&tdl,
3849 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3851 terr = zfs_dirent_lock(&tdl,
3852 tdzp, tnm, &tzp, zflg, NULL, NULL);
3853 serr = zfs_dirent_lock(&sdl,
3854 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3860 * Source entry invalid or not there.
3863 zfs_dirent_unlock(tdl);
3869 rw_exit(&sdzp->z_name_lock);
3872 * FreeBSD: In OpenSolaris they only check if rename source is
3873 * ".." here, because "." is handled in their lookup. This is
3874 * not the case for FreeBSD, so we check for "." explicitly.
3876 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0)
3877 serr = SET_ERROR(EINVAL);
3882 zfs_dirent_unlock(sdl);
3886 rw_exit(&sdzp->z_name_lock);
3888 if (strcmp(tnm, "..") == 0)
3889 terr = SET_ERROR(EINVAL);
3895 * Must have write access at the source to remove the old entry
3896 * and write access at the target to create the new entry.
3897 * Note that if target and source are the same, this can be
3898 * done in a single check.
3901 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3904 if (ZTOV(szp)->v_type == VDIR) {
3906 * Check to make sure rename is valid.
3907 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3909 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3914 * Does target exist?
3918 * Source and target must be the same type.
3920 if (ZTOV(szp)->v_type == VDIR) {
3921 if (ZTOV(tzp)->v_type != VDIR) {
3922 error = SET_ERROR(ENOTDIR);
3926 if (ZTOV(tzp)->v_type == VDIR) {
3927 error = SET_ERROR(EISDIR);
3932 * POSIX dictates that when the source and target
3933 * entries refer to the same file object, rename
3934 * must do nothing and exit without error.
3936 if (szp->z_id == tzp->z_id) {
3942 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3944 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3947 * notify the target directory if it is not the same
3948 * as source directory.
3951 vnevent_rename_dest_dir(tdvp, ct);
3954 tx = dmu_tx_create(zfsvfs->z_os);
3955 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3956 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3957 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3958 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3960 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3961 zfs_sa_upgrade_txholds(tx, tdzp);
3964 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3965 zfs_sa_upgrade_txholds(tx, tzp);
3968 zfs_sa_upgrade_txholds(tx, szp);
3969 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3970 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3973 zfs_rename_unlock(&zl);
3974 zfs_dirent_unlock(sdl);
3975 zfs_dirent_unlock(tdl);
3978 rw_exit(&sdzp->z_name_lock);
3983 if (error == ERESTART) {
3994 if (tzp) /* Attempt to remove the existing target */
3995 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3998 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
4000 szp->z_pflags |= ZFS_AV_MODIFIED;
4002 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
4003 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
4006 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
4008 zfs_log_rename(zilog, tx, TX_RENAME |
4009 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
4010 sdl->dl_name, tdzp, tdl->dl_name, szp);
4013 * Update path information for the target vnode
4015 vn_renamepath(tdvp, ZTOV(szp), tnm,
4019 * At this point, we have successfully created
4020 * the target name, but have failed to remove
4021 * the source name. Since the create was done
4022 * with the ZRENAMING flag, there are
4023 * complications; for one, the link count is
4024 * wrong. The easiest way to deal with this
4025 * is to remove the newly created target, and
4026 * return the original error. This must
4027 * succeed; fortunately, it is very unlikely to
4028 * fail, since we just created it.
4030 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
4031 ZRENAMING, NULL), ==, 0);
4034 #ifdef FREEBSD_NAMECACHE
4038 cache_purge(ZTOV(szp));
4040 cache_purge(ZTOV(tzp));
4048 zfs_rename_unlock(&zl);
4050 zfs_dirent_unlock(sdl);
4051 zfs_dirent_unlock(tdl);
4054 rw_exit(&sdzp->z_name_lock);
4061 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4062 zil_commit(zilog, 0);
4070 * Insert the indicated symbolic reference entry into the directory.
4072 * IN: dvp - Directory to contain new symbolic link.
4073 * link - Name for new symlink entry.
4074 * vap - Attributes of new entry.
4075 * cr - credentials of caller.
4076 * ct - caller context
4077 * flags - case flags
4079 * RETURN: 0 on success, error code on failure.
4082 * dvp - ctime|mtime updated
4086 zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link,
4087 cred_t *cr, kthread_t *td)
4089 znode_t *zp, *dzp = VTOZ(dvp);
4092 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4094 uint64_t len = strlen(link);
4097 zfs_acl_ids_t acl_ids;
4098 boolean_t fuid_dirtied;
4099 uint64_t txtype = TX_SYMLINK;
4100 boolean_t waited = B_FALSE;
4103 ASSERT(vap->va_type == VLNK);
4107 zilog = zfsvfs->z_log;
4109 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
4110 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4112 return (SET_ERROR(EILSEQ));
4114 if (flags & FIGNORECASE)
4117 if (len > MAXPATHLEN) {
4119 return (SET_ERROR(ENAMETOOLONG));
4122 if ((error = zfs_acl_ids_create(dzp, 0,
4123 vap, cr, NULL, &acl_ids)) != 0) {
4128 getnewvnode_reserve(1);
4132 * Attempt to lock directory; fail if entry already exists.
4134 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
4136 zfs_acl_ids_free(&acl_ids);
4137 getnewvnode_drop_reserve();
4142 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4143 zfs_acl_ids_free(&acl_ids);
4144 zfs_dirent_unlock(dl);
4145 getnewvnode_drop_reserve();
4150 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
4151 zfs_acl_ids_free(&acl_ids);
4152 zfs_dirent_unlock(dl);
4153 getnewvnode_drop_reserve();
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, waited ? TXG_WAITED : TXG_NOWAIT);
4172 zfs_dirent_unlock(dl);
4173 if (error == ERESTART) {
4179 zfs_acl_ids_free(&acl_ids);
4181 getnewvnode_drop_reserve();
4187 * Create a new object for the symlink.
4188 * for version 4 ZPL datsets the symlink will be an SA attribute
4190 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
4193 zfs_fuid_sync(zfsvfs, tx);
4195 mutex_enter(&zp->z_lock);
4197 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
4200 zfs_sa_symlink(zp, link, len, tx);
4201 mutex_exit(&zp->z_lock);
4204 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
4205 &zp->z_size, sizeof (zp->z_size), tx);
4207 * Insert the new object into the directory.
4209 (void) zfs_link_create(dl, zp, tx, ZNEW);
4211 if (flags & FIGNORECASE)
4213 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
4216 zfs_acl_ids_free(&acl_ids);
4220 getnewvnode_drop_reserve();
4222 zfs_dirent_unlock(dl);
4224 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4225 zil_commit(zilog, 0);
4232 * Return, in the buffer contained in the provided uio structure,
4233 * the symbolic path referred to by vp.
4235 * IN: vp - vnode of symbolic link.
4236 * uio - structure to contain the link path.
4237 * cr - credentials of caller.
4238 * ct - caller context
4240 * OUT: uio - structure containing the link path.
4242 * RETURN: 0 on success, error code on failure.
4245 * vp - atime updated
4249 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
4251 znode_t *zp = VTOZ(vp);
4252 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4258 mutex_enter(&zp->z_lock);
4260 error = sa_lookup_uio(zp->z_sa_hdl,
4261 SA_ZPL_SYMLINK(zfsvfs), uio);
4263 error = zfs_sa_readlink(zp, uio);
4264 mutex_exit(&zp->z_lock);
4266 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4273 * Insert a new entry into directory tdvp referencing svp.
4275 * IN: tdvp - Directory to contain new entry.
4276 * svp - vnode of new entry.
4277 * name - name of new entry.
4278 * cr - credentials of caller.
4279 * ct - caller context
4281 * RETURN: 0 on success, error code on failure.
4284 * tdvp - ctime|mtime updated
4285 * svp - ctime updated
4289 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
4290 caller_context_t *ct, int flags)
4292 znode_t *dzp = VTOZ(tdvp);
4294 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4303 boolean_t waited = B_FALSE;
4305 ASSERT(tdvp->v_type == VDIR);
4309 zilog = zfsvfs->z_log;
4311 if (VOP_REALVP(svp, &realvp, ct) == 0)
4315 * POSIX dictates that we return EPERM here.
4316 * Better choices include ENOTSUP or EISDIR.
4318 if (svp->v_type == VDIR) {
4320 return (SET_ERROR(EPERM));
4327 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
4328 * ctldir appear to have the same v_vfsp.
4330 if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) {
4332 return (SET_ERROR(EXDEV));
4335 /* Prevent links to .zfs/shares files */
4337 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4338 &parent, sizeof (uint64_t))) != 0) {
4342 if (parent == zfsvfs->z_shares_dir) {
4344 return (SET_ERROR(EPERM));
4347 if (zfsvfs->z_utf8 && u8_validate(name,
4348 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4350 return (SET_ERROR(EILSEQ));
4352 if (flags & FIGNORECASE)
4356 * We do not support links between attributes and non-attributes
4357 * because of the potential security risk of creating links
4358 * into "normal" file space in order to circumvent restrictions
4359 * imposed in attribute space.
4361 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4363 return (SET_ERROR(EINVAL));
4367 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4368 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) {
4370 return (SET_ERROR(EPERM));
4373 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4380 * Attempt to lock directory; fail if entry already exists.
4382 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4388 tx = dmu_tx_create(zfsvfs->z_os);
4389 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4390 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4391 zfs_sa_upgrade_txholds(tx, szp);
4392 zfs_sa_upgrade_txholds(tx, dzp);
4393 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4395 zfs_dirent_unlock(dl);
4396 if (error == ERESTART) {
4407 error = zfs_link_create(dl, szp, tx, 0);
4410 uint64_t txtype = TX_LINK;
4411 if (flags & FIGNORECASE)
4413 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4418 zfs_dirent_unlock(dl);
4421 vnevent_link(svp, ct);
4424 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4425 zil_commit(zilog, 0);
4433 * zfs_null_putapage() is used when the file system has been force
4434 * unmounted. It just drops the pages.
4438 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4439 size_t *lenp, int flags, cred_t *cr)
4441 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4446 * Push a page out to disk, klustering if possible.
4448 * IN: vp - file to push page to.
4449 * pp - page to push.
4450 * flags - additional flags.
4451 * cr - credentials of caller.
4453 * OUT: offp - start of range pushed.
4454 * lenp - len of range pushed.
4456 * RETURN: 0 on success, error code on failure.
4458 * NOTE: callers must have locked the page to be pushed. On
4459 * exit, the page (and all other pages in the kluster) must be
4464 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4465 size_t *lenp, int flags, cred_t *cr)
4467 znode_t *zp = VTOZ(vp);
4468 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4470 u_offset_t off, koff;
4477 * If our blocksize is bigger than the page size, try to kluster
4478 * multiple pages so that we write a full block (thus avoiding
4479 * a read-modify-write).
4481 if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4482 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4483 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4484 ASSERT(koff <= zp->z_size);
4485 if (koff + klen > zp->z_size)
4486 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4487 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4489 ASSERT3U(btop(len), ==, btopr(len));
4492 * Can't push pages past end-of-file.
4494 if (off >= zp->z_size) {
4495 /* ignore all pages */
4498 } else if (off + len > zp->z_size) {
4499 int npages = btopr(zp->z_size - off);
4502 page_list_break(&pp, &trunc, npages);
4503 /* ignore pages past end of file */
4505 pvn_write_done(trunc, flags);
4506 len = zp->z_size - off;
4509 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4510 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4511 err = SET_ERROR(EDQUOT);
4514 tx = dmu_tx_create(zfsvfs->z_os);
4515 dmu_tx_hold_write(tx, zp->z_id, off, len);
4517 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4518 zfs_sa_upgrade_txholds(tx, zp);
4519 err = dmu_tx_assign(tx, TXG_WAIT);
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_putpages(struct vnode *vp, vm_page_t *ma, size_t len, int flags,
5809 znode_t *zp = VTOZ(vp);
5810 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5819 vm_ooffset_t lo_off;
5830 object = vp->v_object;
5834 KASSERT(ma[0]->object == object, ("mismatching object"));
5835 KASSERT(len > 0 && (len & PAGE_MASK) == 0, ("unexpected length"));
5837 for (i = 0; i < pcount; i++)
5838 rtvals[i] = zfs_vm_pagerret_error;
5840 off = IDX_TO_OFF(ma[0]->pindex);
5841 blksz = zp->z_blksz;
5842 lo_off = rounddown(off, blksz);
5843 lo_len = roundup(len + (off - lo_off), blksz);
5844 rl = zfs_range_lock(zp, lo_off, lo_len, RL_WRITER);
5846 zfs_vmobject_wlock(object);
5847 if (len + off > object->un_pager.vnp.vnp_size) {
5848 if (object->un_pager.vnp.vnp_size > off) {
5851 len = object->un_pager.vnp.vnp_size - off;
5853 if ((pgoff = (int)len & PAGE_MASK) != 0) {
5855 * If the object is locked and the following
5856 * conditions hold, then the page's dirty
5857 * field cannot be concurrently changed by a
5861 vm_page_assert_sbusied(m);
5862 KASSERT(!pmap_page_is_write_mapped(m),
5863 ("zfs_putpages: page %p is not read-only", m));
5864 vm_page_clear_dirty(m, pgoff, PAGE_SIZE -
5871 if (ncount < pcount) {
5872 for (i = ncount; i < pcount; i++) {
5873 rtvals[i] = zfs_vm_pagerret_bad;
5877 zfs_vmobject_wunlock(object);
5882 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
5883 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
5888 tx = dmu_tx_create(zfsvfs->z_os);
5889 dmu_tx_hold_write(tx, zp->z_id, off, len);
5891 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
5892 zfs_sa_upgrade_txholds(tx, zp);
5893 err = dmu_tx_assign(tx, TXG_NOWAIT);
5895 if (err == ERESTART) {
5904 if (zp->z_blksz < PAGE_SIZE) {
5906 for (i = 0; len > 0; off += tocopy, len -= tocopy, i++) {
5907 tocopy = len > PAGE_SIZE ? PAGE_SIZE : len;
5908 va = zfs_map_page(ma[i], &sf);
5909 dmu_write(zfsvfs->z_os, zp->z_id, off, tocopy, va, tx);
5913 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, ma, tx);
5917 uint64_t mtime[2], ctime[2];
5918 sa_bulk_attr_t bulk[3];
5921 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
5923 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
5925 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
5927 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
5929 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
5931 zfs_vmobject_wlock(object);
5932 for (i = 0; i < ncount; i++) {
5933 rtvals[i] = zfs_vm_pagerret_ok;
5934 vm_page_undirty(ma[i]);
5936 zfs_vmobject_wunlock(object);
5937 PCPU_INC(cnt.v_vnodeout);
5938 PCPU_ADD(cnt.v_vnodepgsout, ncount);
5943 zfs_range_unlock(rl);
5944 if ((flags & (zfs_vm_pagerput_sync | zfs_vm_pagerput_inval)) != 0 ||
5945 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5946 zil_commit(zfsvfs->z_log, zp->z_id);
5952 zfs_freebsd_putpages(ap)
5953 struct vop_putpages_args /* {
5959 vm_ooffset_t a_offset;
5963 return (zfs_putpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_sync,
5968 zfs_freebsd_bmap(ap)
5969 struct vop_bmap_args /* {
5972 struct bufobj **a_bop;
5979 if (ap->a_bop != NULL)
5980 *ap->a_bop = &ap->a_vp->v_bufobj;
5981 if (ap->a_bnp != NULL)
5982 *ap->a_bnp = ap->a_bn;
5983 if (ap->a_runp != NULL)
5985 if (ap->a_runb != NULL)
5992 zfs_freebsd_open(ap)
5993 struct vop_open_args /* {
5996 struct ucred *a_cred;
5997 struct thread *a_td;
6000 vnode_t *vp = ap->a_vp;
6001 znode_t *zp = VTOZ(vp);
6004 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL);
6006 vnode_create_vobject(vp, zp->z_size, ap->a_td);
6011 zfs_freebsd_close(ap)
6012 struct vop_close_args /* {
6015 struct ucred *a_cred;
6016 struct thread *a_td;
6020 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL));
6024 zfs_freebsd_ioctl(ap)
6025 struct vop_ioctl_args /* {
6035 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data,
6036 ap->a_fflag, ap->a_cred, NULL, NULL));
6040 zfs_freebsd_read(ap)
6041 struct vop_read_args /* {
6045 struct ucred *a_cred;
6049 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
6054 zfs_freebsd_write(ap)
6055 struct vop_write_args /* {
6059 struct ucred *a_cred;
6063 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
6068 zfs_freebsd_access(ap)
6069 struct vop_access_args /* {
6071 accmode_t a_accmode;
6072 struct ucred *a_cred;
6073 struct thread *a_td;
6076 vnode_t *vp = ap->a_vp;
6077 znode_t *zp = VTOZ(vp);
6082 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND,
6084 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND);
6086 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL);
6089 * VADMIN has to be handled by vaccess().
6092 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND);
6094 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
6095 zp->z_gid, accmode, ap->a_cred, NULL);
6100 * For VEXEC, ensure that at least one execute bit is set for
6103 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR &&
6104 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
6112 zfs_freebsd_lookup(ap)
6113 struct vop_lookup_args /* {
6114 struct vnode *a_dvp;
6115 struct vnode **a_vpp;
6116 struct componentname *a_cnp;
6119 struct componentname *cnp = ap->a_cnp;
6120 char nm[NAME_MAX + 1];
6122 ASSERT(cnp->cn_namelen < sizeof(nm));
6123 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm)));
6125 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop,
6126 cnp->cn_cred, cnp->cn_thread, 0));
6130 zfs_freebsd_create(ap)
6131 struct vop_create_args /* {
6132 struct vnode *a_dvp;
6133 struct vnode **a_vpp;
6134 struct componentname *a_cnp;
6135 struct vattr *a_vap;
6138 struct componentname *cnp = ap->a_cnp;
6139 vattr_t *vap = ap->a_vap;
6142 ASSERT(cnp->cn_flags & SAVENAME);
6144 vattr_init_mask(vap);
6145 mode = vap->va_mode & ALLPERMS;
6147 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode,
6148 ap->a_vpp, cnp->cn_cred, cnp->cn_thread));
6152 zfs_freebsd_remove(ap)
6153 struct vop_remove_args /* {
6154 struct vnode *a_dvp;
6156 struct componentname *a_cnp;
6160 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6162 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr,
6163 ap->a_cnp->cn_cred, NULL, 0));
6167 zfs_freebsd_mkdir(ap)
6168 struct vop_mkdir_args /* {
6169 struct vnode *a_dvp;
6170 struct vnode **a_vpp;
6171 struct componentname *a_cnp;
6172 struct vattr *a_vap;
6175 vattr_t *vap = ap->a_vap;
6177 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6179 vattr_init_mask(vap);
6181 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp,
6182 ap->a_cnp->cn_cred, NULL, 0, NULL));
6186 zfs_freebsd_rmdir(ap)
6187 struct vop_rmdir_args /* {
6188 struct vnode *a_dvp;
6190 struct componentname *a_cnp;
6193 struct componentname *cnp = ap->a_cnp;
6195 ASSERT(cnp->cn_flags & SAVENAME);
6197 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0));
6201 zfs_freebsd_readdir(ap)
6202 struct vop_readdir_args /* {
6205 struct ucred *a_cred;
6212 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag,
6213 ap->a_ncookies, ap->a_cookies));
6217 zfs_freebsd_fsync(ap)
6218 struct vop_fsync_args /* {
6221 struct thread *a_td;
6226 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL));
6230 zfs_freebsd_getattr(ap)
6231 struct vop_getattr_args /* {
6233 struct vattr *a_vap;
6234 struct ucred *a_cred;
6237 vattr_t *vap = ap->a_vap;
6243 xvap.xva_vattr = *vap;
6244 xvap.xva_vattr.va_mask |= AT_XVATTR;
6246 /* Convert chflags into ZFS-type flags. */
6247 /* XXX: what about SF_SETTABLE?. */
6248 XVA_SET_REQ(&xvap, XAT_IMMUTABLE);
6249 XVA_SET_REQ(&xvap, XAT_APPENDONLY);
6250 XVA_SET_REQ(&xvap, XAT_NOUNLINK);
6251 XVA_SET_REQ(&xvap, XAT_NODUMP);
6252 XVA_SET_REQ(&xvap, XAT_READONLY);
6253 XVA_SET_REQ(&xvap, XAT_ARCHIVE);
6254 XVA_SET_REQ(&xvap, XAT_SYSTEM);
6255 XVA_SET_REQ(&xvap, XAT_HIDDEN);
6256 XVA_SET_REQ(&xvap, XAT_REPARSE);
6257 XVA_SET_REQ(&xvap, XAT_OFFLINE);
6258 XVA_SET_REQ(&xvap, XAT_SPARSE);
6260 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL);
6264 /* Convert ZFS xattr into chflags. */
6265 #define FLAG_CHECK(fflag, xflag, xfield) do { \
6266 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \
6267 fflags |= (fflag); \
6269 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE,
6270 xvap.xva_xoptattrs.xoa_immutable);
6271 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY,
6272 xvap.xva_xoptattrs.xoa_appendonly);
6273 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK,
6274 xvap.xva_xoptattrs.xoa_nounlink);
6275 FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE,
6276 xvap.xva_xoptattrs.xoa_archive);
6277 FLAG_CHECK(UF_NODUMP, XAT_NODUMP,
6278 xvap.xva_xoptattrs.xoa_nodump);
6279 FLAG_CHECK(UF_READONLY, XAT_READONLY,
6280 xvap.xva_xoptattrs.xoa_readonly);
6281 FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM,
6282 xvap.xva_xoptattrs.xoa_system);
6283 FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN,
6284 xvap.xva_xoptattrs.xoa_hidden);
6285 FLAG_CHECK(UF_REPARSE, XAT_REPARSE,
6286 xvap.xva_xoptattrs.xoa_reparse);
6287 FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE,
6288 xvap.xva_xoptattrs.xoa_offline);
6289 FLAG_CHECK(UF_SPARSE, XAT_SPARSE,
6290 xvap.xva_xoptattrs.xoa_sparse);
6293 *vap = xvap.xva_vattr;
6294 vap->va_flags = fflags;
6299 zfs_freebsd_setattr(ap)
6300 struct vop_setattr_args /* {
6302 struct vattr *a_vap;
6303 struct ucred *a_cred;
6306 vnode_t *vp = ap->a_vp;
6307 vattr_t *vap = ap->a_vap;
6308 cred_t *cred = ap->a_cred;
6313 vattr_init_mask(vap);
6314 vap->va_mask &= ~AT_NOSET;
6317 xvap.xva_vattr = *vap;
6319 zflags = VTOZ(vp)->z_pflags;
6321 if (vap->va_flags != VNOVAL) {
6322 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs;
6325 if (zfsvfs->z_use_fuids == B_FALSE)
6326 return (EOPNOTSUPP);
6328 fflags = vap->va_flags;
6331 * We need to figure out whether it makes sense to allow
6332 * UF_REPARSE through, since we don't really have other
6333 * facilities to handle reparse points and zfs_setattr()
6334 * doesn't currently allow setting that attribute anyway.
6336 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE|
6337 UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE|
6338 UF_OFFLINE|UF_SPARSE)) != 0)
6339 return (EOPNOTSUPP);
6341 * Unprivileged processes are not permitted to unset system
6342 * flags, or modify flags if any system flags are set.
6343 * Privileged non-jail processes may not modify system flags
6344 * if securelevel > 0 and any existing system flags are set.
6345 * Privileged jail processes behave like privileged non-jail
6346 * processes if the security.jail.chflags_allowed sysctl is
6347 * is non-zero; otherwise, they behave like unprivileged
6350 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 ||
6351 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) {
6353 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6354 error = securelevel_gt(cred, 0);
6360 * Callers may only modify the file flags on objects they
6361 * have VADMIN rights for.
6363 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0)
6366 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6370 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) {
6375 #define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \
6376 if (((fflags & (fflag)) && !(zflags & (zflag))) || \
6377 ((zflags & (zflag)) && !(fflags & (fflag)))) { \
6378 XVA_SET_REQ(&xvap, (xflag)); \
6379 (xfield) = ((fflags & (fflag)) != 0); \
6382 /* Convert chflags into ZFS-type flags. */
6383 /* XXX: what about SF_SETTABLE?. */
6384 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE,
6385 xvap.xva_xoptattrs.xoa_immutable);
6386 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY,
6387 xvap.xva_xoptattrs.xoa_appendonly);
6388 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK,
6389 xvap.xva_xoptattrs.xoa_nounlink);
6390 FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE,
6391 xvap.xva_xoptattrs.xoa_archive);
6392 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP,
6393 xvap.xva_xoptattrs.xoa_nodump);
6394 FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY,
6395 xvap.xva_xoptattrs.xoa_readonly);
6396 FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM,
6397 xvap.xva_xoptattrs.xoa_system);
6398 FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN,
6399 xvap.xva_xoptattrs.xoa_hidden);
6400 FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE,
6401 xvap.xva_xoptattrs.xoa_hidden);
6402 FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE,
6403 xvap.xva_xoptattrs.xoa_offline);
6404 FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE,
6405 xvap.xva_xoptattrs.xoa_sparse);
6408 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL));
6412 zfs_freebsd_rename(ap)
6413 struct vop_rename_args /* {
6414 struct vnode *a_fdvp;
6415 struct vnode *a_fvp;
6416 struct componentname *a_fcnp;
6417 struct vnode *a_tdvp;
6418 struct vnode *a_tvp;
6419 struct componentname *a_tcnp;
6422 vnode_t *fdvp = ap->a_fdvp;
6423 vnode_t *fvp = ap->a_fvp;
6424 vnode_t *tdvp = ap->a_tdvp;
6425 vnode_t *tvp = ap->a_tvp;
6428 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART));
6429 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART));
6432 * Check for cross-device rename.
6434 if ((fdvp->v_mount != tdvp->v_mount) ||
6435 (tvp && (fdvp->v_mount != tvp->v_mount)))
6438 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp,
6439 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0);
6453 zfs_freebsd_symlink(ap)
6454 struct vop_symlink_args /* {
6455 struct vnode *a_dvp;
6456 struct vnode **a_vpp;
6457 struct componentname *a_cnp;
6458 struct vattr *a_vap;
6462 struct componentname *cnp = ap->a_cnp;
6463 vattr_t *vap = ap->a_vap;
6465 ASSERT(cnp->cn_flags & SAVENAME);
6467 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */
6468 vattr_init_mask(vap);
6470 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap,
6471 ap->a_target, cnp->cn_cred, cnp->cn_thread));
6475 zfs_freebsd_readlink(ap)
6476 struct vop_readlink_args /* {
6479 struct ucred *a_cred;
6483 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL));
6487 zfs_freebsd_link(ap)
6488 struct vop_link_args /* {
6489 struct vnode *a_tdvp;
6491 struct componentname *a_cnp;
6494 struct componentname *cnp = ap->a_cnp;
6495 vnode_t *vp = ap->a_vp;
6496 vnode_t *tdvp = ap->a_tdvp;
6498 if (tdvp->v_mount != vp->v_mount)
6501 ASSERT(cnp->cn_flags & SAVENAME);
6503 return (zfs_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0));
6507 zfs_freebsd_inactive(ap)
6508 struct vop_inactive_args /* {
6510 struct thread *a_td;
6513 vnode_t *vp = ap->a_vp;
6515 zfs_inactive(vp, ap->a_td->td_ucred, NULL);
6520 zfs_freebsd_reclaim(ap)
6521 struct vop_reclaim_args /* {
6523 struct thread *a_td;
6526 vnode_t *vp = ap->a_vp;
6527 znode_t *zp = VTOZ(vp);
6528 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
6532 /* Destroy the vm object and flush associated pages. */
6533 vnode_destroy_vobject(vp);
6536 * z_teardown_inactive_lock protects from a race with
6537 * zfs_znode_dmu_fini in zfsvfs_teardown during
6540 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
6541 if (zp->z_sa_hdl == NULL)
6545 rw_exit(&zfsvfs->z_teardown_inactive_lock);
6553 struct vop_fid_args /* {
6559 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL));
6563 zfs_freebsd_pathconf(ap)
6564 struct vop_pathconf_args /* {
6567 register_t *a_retval;
6573 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL);
6575 *ap->a_retval = val;
6576 else if (error == EOPNOTSUPP)
6577 error = vop_stdpathconf(ap);
6582 zfs_freebsd_fifo_pathconf(ap)
6583 struct vop_pathconf_args /* {
6586 register_t *a_retval;
6590 switch (ap->a_name) {
6591 case _PC_ACL_EXTENDED:
6593 case _PC_ACL_PATH_MAX:
6594 case _PC_MAC_PRESENT:
6595 return (zfs_freebsd_pathconf(ap));
6597 return (fifo_specops.vop_pathconf(ap));
6602 * FreeBSD's extended attributes namespace defines file name prefix for ZFS'
6603 * extended attribute name:
6606 * system freebsd:system:
6607 * user (none, can be used to access ZFS fsattr(5) attributes
6608 * created on Solaris)
6611 zfs_create_attrname(int attrnamespace, const char *name, char *attrname,
6614 const char *namespace, *prefix, *suffix;
6616 /* We don't allow '/' character in attribute name. */
6617 if (strchr(name, '/') != NULL)
6619 /* We don't allow attribute names that start with "freebsd:" string. */
6620 if (strncmp(name, "freebsd:", 8) == 0)
6623 bzero(attrname, size);
6625 switch (attrnamespace) {
6626 case EXTATTR_NAMESPACE_USER:
6628 prefix = "freebsd:";
6629 namespace = EXTATTR_NAMESPACE_USER_STRING;
6633 * This is the default namespace by which we can access all
6634 * attributes created on Solaris.
6636 prefix = namespace = suffix = "";
6639 case EXTATTR_NAMESPACE_SYSTEM:
6640 prefix = "freebsd:";
6641 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING;
6644 case EXTATTR_NAMESPACE_EMPTY:
6648 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix,
6650 return (ENAMETOOLONG);
6656 * Vnode operating to retrieve a named extended attribute.
6659 zfs_getextattr(struct vop_getextattr_args *ap)
6662 IN struct vnode *a_vp;
6663 IN int a_attrnamespace;
6664 IN const char *a_name;
6665 INOUT struct uio *a_uio;
6667 IN struct ucred *a_cred;
6668 IN struct thread *a_td;
6672 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6673 struct thread *td = ap->a_td;
6674 struct nameidata nd;
6677 vnode_t *xvp = NULL, *vp;
6680 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6681 ap->a_cred, ap->a_td, VREAD);
6685 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6692 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6700 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6702 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL);
6704 NDFREE(&nd, NDF_ONLY_PNBUF);
6707 if (error == ENOENT)
6712 if (ap->a_size != NULL) {
6713 error = VOP_GETATTR(vp, &va, ap->a_cred);
6715 *ap->a_size = (size_t)va.va_size;
6716 } else if (ap->a_uio != NULL)
6717 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6720 vn_close(vp, flags, ap->a_cred, td);
6727 * Vnode operation to remove a named attribute.
6730 zfs_deleteextattr(struct vop_deleteextattr_args *ap)
6733 IN struct vnode *a_vp;
6734 IN int a_attrnamespace;
6735 IN const char *a_name;
6736 IN struct ucred *a_cred;
6737 IN struct thread *a_td;
6741 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6742 struct thread *td = ap->a_td;
6743 struct nameidata nd;
6746 vnode_t *xvp = NULL, *vp;
6749 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6750 ap->a_cred, ap->a_td, VWRITE);
6754 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6761 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6768 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF,
6769 UIO_SYSSPACE, attrname, xvp, td);
6774 NDFREE(&nd, NDF_ONLY_PNBUF);
6775 if (error == ENOENT)
6780 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
6781 NDFREE(&nd, NDF_ONLY_PNBUF);
6784 if (vp == nd.ni_dvp)
6794 * Vnode operation to set a named attribute.
6797 zfs_setextattr(struct vop_setextattr_args *ap)
6800 IN struct vnode *a_vp;
6801 IN int a_attrnamespace;
6802 IN const char *a_name;
6803 INOUT struct uio *a_uio;
6804 IN struct ucred *a_cred;
6805 IN struct thread *a_td;
6809 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6810 struct thread *td = ap->a_td;
6811 struct nameidata nd;
6814 vnode_t *xvp = NULL, *vp;
6817 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6818 ap->a_cred, ap->a_td, VWRITE);
6822 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6829 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6830 LOOKUP_XATTR | CREATE_XATTR_DIR);
6836 flags = FFLAGS(O_WRONLY | O_CREAT);
6837 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6839 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL);
6841 NDFREE(&nd, NDF_ONLY_PNBUF);
6849 error = VOP_SETATTR(vp, &va, ap->a_cred);
6851 VOP_WRITE(vp, ap->a_uio, IO_UNIT | IO_SYNC, ap->a_cred);
6854 vn_close(vp, flags, ap->a_cred, td);
6861 * Vnode operation to retrieve extended attributes on a vnode.
6864 zfs_listextattr(struct vop_listextattr_args *ap)
6867 IN struct vnode *a_vp;
6868 IN int a_attrnamespace;
6869 INOUT struct uio *a_uio;
6871 IN struct ucred *a_cred;
6872 IN struct thread *a_td;
6876 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6877 struct thread *td = ap->a_td;
6878 struct nameidata nd;
6879 char attrprefix[16];
6880 u_char dirbuf[sizeof(struct dirent)];
6883 struct uio auio, *uio = ap->a_uio;
6884 size_t *sizep = ap->a_size;
6886 vnode_t *xvp = NULL, *vp;
6887 int done, error, eof, pos;
6889 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6890 ap->a_cred, ap->a_td, VREAD);
6894 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix,
6895 sizeof(attrprefix));
6898 plen = strlen(attrprefix);
6905 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6910 * ENOATTR means that the EA directory does not yet exist,
6911 * i.e. there are no extended attributes there.
6913 if (error == ENOATTR)
6918 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED,
6919 UIO_SYSSPACE, ".", xvp, td);
6922 NDFREE(&nd, NDF_ONLY_PNBUF);
6928 auio.uio_iov = &aiov;
6929 auio.uio_iovcnt = 1;
6930 auio.uio_segflg = UIO_SYSSPACE;
6932 auio.uio_rw = UIO_READ;
6933 auio.uio_offset = 0;
6938 aiov.iov_base = (void *)dirbuf;
6939 aiov.iov_len = sizeof(dirbuf);
6940 auio.uio_resid = sizeof(dirbuf);
6941 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL);
6942 done = sizeof(dirbuf) - auio.uio_resid;
6945 for (pos = 0; pos < done;) {
6946 dp = (struct dirent *)(dirbuf + pos);
6947 pos += dp->d_reclen;
6949 * XXX: Temporarily we also accept DT_UNKNOWN, as this
6950 * is what we get when attribute was created on Solaris.
6952 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN)
6954 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0)
6956 else if (strncmp(dp->d_name, attrprefix, plen) != 0)
6958 nlen = dp->d_namlen - plen;
6961 else if (uio != NULL) {
6963 * Format of extattr name entry is one byte for
6964 * length and the rest for name.
6966 error = uiomove(&nlen, 1, uio->uio_rw, uio);
6968 error = uiomove(dp->d_name + plen, nlen,
6975 } while (!eof && error == 0);
6984 zfs_freebsd_getacl(ap)
6985 struct vop_getacl_args /* {
6994 vsecattr_t vsecattr;
6996 if (ap->a_type != ACL_TYPE_NFS4)
6999 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
7000 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL))
7003 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt);
7004 if (vsecattr.vsa_aclentp != NULL)
7005 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz);
7011 zfs_freebsd_setacl(ap)
7012 struct vop_setacl_args /* {
7021 vsecattr_t vsecattr;
7022 int aclbsize; /* size of acl list in bytes */
7025 if (ap->a_type != ACL_TYPE_NFS4)
7028 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES)
7032 * With NFSv4 ACLs, chmod(2) may need to add additional entries,
7033 * splitting every entry into two and appending "canonical six"
7034 * entries at the end. Don't allow for setting an ACL that would
7035 * cause chmod(2) to run out of ACL entries.
7037 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES)
7040 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR);
7044 vsecattr.vsa_mask = VSA_ACE;
7045 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t);
7046 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP);
7047 aaclp = vsecattr.vsa_aclentp;
7048 vsecattr.vsa_aclentsz = aclbsize;
7050 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp);
7051 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL);
7052 kmem_free(aaclp, aclbsize);
7058 zfs_freebsd_aclcheck(ap)
7059 struct vop_aclcheck_args /* {
7068 return (EOPNOTSUPP);
7071 struct vop_vector zfs_vnodeops;
7072 struct vop_vector zfs_fifoops;
7073 struct vop_vector zfs_shareops;
7075 struct vop_vector zfs_vnodeops = {
7076 .vop_default = &default_vnodeops,
7077 .vop_inactive = zfs_freebsd_inactive,
7078 .vop_reclaim = zfs_freebsd_reclaim,
7079 .vop_access = zfs_freebsd_access,
7080 #ifdef FREEBSD_NAMECACHE
7081 .vop_lookup = vfs_cache_lookup,
7082 .vop_cachedlookup = zfs_freebsd_lookup,
7084 .vop_lookup = zfs_freebsd_lookup,
7086 .vop_getattr = zfs_freebsd_getattr,
7087 .vop_setattr = zfs_freebsd_setattr,
7088 .vop_create = zfs_freebsd_create,
7089 .vop_mknod = zfs_freebsd_create,
7090 .vop_mkdir = zfs_freebsd_mkdir,
7091 .vop_readdir = zfs_freebsd_readdir,
7092 .vop_fsync = zfs_freebsd_fsync,
7093 .vop_open = zfs_freebsd_open,
7094 .vop_close = zfs_freebsd_close,
7095 .vop_rmdir = zfs_freebsd_rmdir,
7096 .vop_ioctl = zfs_freebsd_ioctl,
7097 .vop_link = zfs_freebsd_link,
7098 .vop_symlink = zfs_freebsd_symlink,
7099 .vop_readlink = zfs_freebsd_readlink,
7100 .vop_read = zfs_freebsd_read,
7101 .vop_write = zfs_freebsd_write,
7102 .vop_remove = zfs_freebsd_remove,
7103 .vop_rename = zfs_freebsd_rename,
7104 .vop_pathconf = zfs_freebsd_pathconf,
7105 .vop_bmap = zfs_freebsd_bmap,
7106 .vop_fid = zfs_freebsd_fid,
7107 .vop_getextattr = zfs_getextattr,
7108 .vop_deleteextattr = zfs_deleteextattr,
7109 .vop_setextattr = zfs_setextattr,
7110 .vop_listextattr = zfs_listextattr,
7111 .vop_getacl = zfs_freebsd_getacl,
7112 .vop_setacl = zfs_freebsd_setacl,
7113 .vop_aclcheck = zfs_freebsd_aclcheck,
7114 .vop_getpages = zfs_freebsd_getpages,
7115 .vop_putpages = zfs_freebsd_putpages,
7118 struct vop_vector zfs_fifoops = {
7119 .vop_default = &fifo_specops,
7120 .vop_fsync = zfs_freebsd_fsync,
7121 .vop_access = zfs_freebsd_access,
7122 .vop_getattr = zfs_freebsd_getattr,
7123 .vop_inactive = zfs_freebsd_inactive,
7124 .vop_read = VOP_PANIC,
7125 .vop_reclaim = zfs_freebsd_reclaim,
7126 .vop_setattr = zfs_freebsd_setattr,
7127 .vop_write = VOP_PANIC,
7128 .vop_pathconf = zfs_freebsd_fifo_pathconf,
7129 .vop_fid = zfs_freebsd_fid,
7130 .vop_getacl = zfs_freebsd_getacl,
7131 .vop_setacl = zfs_freebsd_setacl,
7132 .vop_aclcheck = zfs_freebsd_aclcheck,
7136 * special share hidden files vnode operations template
7138 struct vop_vector zfs_shareops = {
7139 .vop_default = &default_vnodeops,
7140 .vop_access = zfs_freebsd_access,
7141 .vop_inactive = zfs_freebsd_inactive,
7142 .vop_reclaim = zfs_freebsd_reclaim,
7143 .vop_fid = zfs_freebsd_fid,
7144 .vop_pathconf = zfs_freebsd_pathconf,