4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2013 by Delphix. All rights reserved.
26 /* Portions Copyright 2007 Jeremy Teo */
27 /* Portions Copyright 2010 Robert Milkowski */
29 #include <sys/types.h>
30 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/sysmacros.h>
34 #include <sys/resource.h>
37 #include <sys/vnode.h>
41 #include <sys/taskq.h>
43 #include <sys/atomic.h>
44 #include <sys/namei.h>
46 #include <sys/cmn_err.h>
47 #include <sys/errno.h>
48 #include <sys/unistd.h>
49 #include <sys/zfs_dir.h>
50 #include <sys/zfs_ioctl.h>
51 #include <sys/fs/zfs.h>
53 #include <sys/dmu_objset.h>
59 #include <sys/dirent.h>
60 #include <sys/policy.h>
61 #include <sys/sunddi.h>
62 #include <sys/filio.h>
64 #include <sys/zfs_ctldir.h>
65 #include <sys/zfs_fuid.h>
66 #include <sys/zfs_sa.h>
68 #include <sys/zfs_rlock.h>
69 #include <sys/extdirent.h>
70 #include <sys/kidmap.h>
73 #include <sys/sched.h>
79 * Each vnode op performs some logical unit of work. To do this, the ZPL must
80 * properly lock its in-core state, create a DMU transaction, do the work,
81 * record this work in the intent log (ZIL), commit the DMU transaction,
82 * and wait for the intent log to commit if it is a synchronous operation.
83 * Moreover, the vnode ops must work in both normal and log replay context.
84 * The ordering of events is important to avoid deadlocks and references
85 * to freed memory. The example below illustrates the following Big Rules:
87 * (1) A check must be made in each zfs thread for a mounted file system.
88 * This is done avoiding races using ZFS_ENTER(zfsvfs).
89 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
90 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
91 * can return EIO from the calling function.
93 * (2) VN_RELE() should always be the last thing except for zil_commit()
94 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
95 * First, if it's the last reference, the vnode/znode
96 * can be freed, so the zp may point to freed memory. Second, the last
97 * reference will call zfs_zinactive(), which may induce a lot of work --
98 * pushing cached pages (which acquires range locks) and syncing out
99 * cached atime changes. Third, zfs_zinactive() may require a new tx,
100 * which could deadlock the system if you were already holding one.
101 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
103 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
104 * as they can span dmu_tx_assign() calls.
106 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
107 * dmu_tx_assign(). This is critical because we don't want to block
108 * while holding locks.
110 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
111 * reduces lock contention and CPU usage when we must wait (note that if
112 * throughput is constrained by the storage, nearly every transaction
115 * Note, in particular, that if a lock is sometimes acquired before
116 * the tx assigns, and sometimes after (e.g. z_lock), then failing
117 * to use a non-blocking assign can deadlock the system. The scenario:
119 * Thread A has grabbed a lock before calling dmu_tx_assign().
120 * Thread B is in an already-assigned tx, and blocks for this lock.
121 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
122 * forever, because the previous txg can't quiesce until B's tx commits.
124 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
125 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
126 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
127 * to indicate that this operation has already called dmu_tx_wait().
128 * This will ensure that we don't retry forever, waiting a short bit
131 * (5) If the operation succeeded, generate the intent log entry for it
132 * before dropping locks. This ensures that the ordering of events
133 * in the intent log matches the order in which they actually occurred.
134 * During ZIL replay the zfs_log_* functions will update the sequence
135 * number to indicate the zil transaction has replayed.
137 * (6) At the end of each vnode op, the DMU tx must always commit,
138 * regardless of whether there were any errors.
140 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
141 * to ensure that synchronous semantics are provided when necessary.
143 * In general, this is how things should be ordered in each vnode op:
145 * ZFS_ENTER(zfsvfs); // exit if unmounted
147 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD())
148 * rw_enter(...); // grab any other locks you need
149 * tx = dmu_tx_create(...); // get DMU tx
150 * dmu_tx_hold_*(); // hold each object you might modify
151 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
153 * rw_exit(...); // drop locks
154 * zfs_dirent_unlock(dl); // unlock directory entry
155 * VN_RELE(...); // release held vnodes
156 * if (error == ERESTART) {
162 * dmu_tx_abort(tx); // abort DMU tx
163 * ZFS_EXIT(zfsvfs); // finished in zfs
164 * return (error); // really out of space
166 * error = do_real_work(); // do whatever this VOP does
168 * zfs_log_*(...); // on success, make ZIL entry
169 * dmu_tx_commit(tx); // commit DMU tx -- error or not
170 * rw_exit(...); // drop locks
171 * zfs_dirent_unlock(dl); // unlock directory entry
172 * VN_RELE(...); // release held vnodes
173 * zil_commit(zilog, foid); // synchronous when necessary
174 * ZFS_EXIT(zfsvfs); // finished in zfs
175 * return (error); // done, report error
180 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
182 znode_t *zp = VTOZ(*vpp);
183 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
188 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
189 ((flag & FAPPEND) == 0)) {
191 return (SET_ERROR(EPERM));
194 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
195 ZTOV(zp)->v_type == VREG &&
196 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
197 if (fs_vscan(*vpp, cr, 0) != 0) {
199 return (SET_ERROR(EACCES));
203 /* Keep a count of the synchronous opens in the znode */
204 if (flag & (FSYNC | FDSYNC))
205 atomic_inc_32(&zp->z_sync_cnt);
213 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
214 caller_context_t *ct)
216 znode_t *zp = VTOZ(vp);
217 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
220 * Clean up any locks held by this process on the vp.
222 cleanlocks(vp, ddi_get_pid(), 0);
223 cleanshares(vp, ddi_get_pid());
228 /* Decrement the synchronous opens in the znode */
229 if ((flag & (FSYNC | FDSYNC)) && (count == 1))
230 atomic_dec_32(&zp->z_sync_cnt);
232 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
233 ZTOV(zp)->v_type == VREG &&
234 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
235 VERIFY(fs_vscan(vp, cr, 1) == 0);
242 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
243 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
246 zfs_holey(vnode_t *vp, u_long cmd, offset_t *off)
248 znode_t *zp = VTOZ(vp);
249 uint64_t noff = (uint64_t)*off; /* new offset */
254 file_sz = zp->z_size;
255 if (noff >= file_sz) {
256 return (SET_ERROR(ENXIO));
259 if (cmd == _FIO_SEEK_HOLE)
264 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
267 if ((error == ESRCH) || (noff > file_sz)) {
269 * Handle the virtual hole at the end of file.
275 return (SET_ERROR(ENXIO));
286 zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred,
287 int *rvalp, caller_context_t *ct)
299 * The following two ioctls are used by bfu. Faking out,
300 * necessary to avoid bfu errors.
309 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
310 return (SET_ERROR(EFAULT));
312 off = *(offset_t *)data;
315 zfsvfs = zp->z_zfsvfs;
319 /* offset parameter is in/out */
320 error = zfs_holey(vp, com, &off);
325 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
326 return (SET_ERROR(EFAULT));
328 *(offset_t *)data = off;
332 return (SET_ERROR(ENOTTY));
336 page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes)
343 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE
344 * aligned boundaries, if the range is not aligned. As a result a
345 * DEV_BSIZE subrange with partially dirty data may get marked as clean.
346 * It may happen that all DEV_BSIZE subranges are marked clean and thus
347 * the whole page would be considred clean despite have some dirty data.
348 * For this reason we should shrink the range to DEV_BSIZE aligned
349 * boundaries before calling vm_page_clear_dirty.
351 end = rounddown2(off + nbytes, DEV_BSIZE);
352 off = roundup2(off, DEV_BSIZE);
356 VM_OBJECT_LOCK_ASSERT(obj, MA_OWNED);
359 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
361 if ((pp->oflags & VPO_BUSY) != 0) {
363 * Reference the page before unlocking and
364 * sleeping so that the page daemon is less
365 * likely to reclaim it.
367 vm_page_reference(pp);
368 vm_page_sleep(pp, "zfsmwb");
371 } else if (pp == NULL) {
372 pp = vm_page_alloc(obj, OFF_TO_IDX(start),
373 VM_ALLOC_SYSTEM | VM_ALLOC_IFCACHED |
376 ASSERT(pp != NULL && !pp->valid);
381 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
382 vm_object_pip_add(obj, 1);
383 vm_page_io_start(pp);
384 pmap_remove_write(pp);
386 vm_page_clear_dirty(pp, off, nbytes);
394 page_unbusy(vm_page_t pp)
397 vm_page_io_finish(pp);
398 vm_object_pip_subtract(pp->object, 1);
402 page_hold(vnode_t *vp, int64_t start)
408 VM_OBJECT_LOCK_ASSERT(obj, MA_OWNED);
411 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
413 if ((pp->oflags & VPO_BUSY) != 0) {
415 * Reference the page before unlocking and
416 * sleeping so that the page daemon is less
417 * likely to reclaim it.
419 vm_page_reference(pp);
420 vm_page_sleep(pp, "zfsmwb");
424 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
437 page_unhold(vm_page_t pp)
446 * When a file is memory mapped, we must keep the IO data synchronized
447 * between the DMU cache and the memory mapped pages. What this means:
449 * On Write: If we find a memory mapped page, we write to *both*
450 * the page and the dmu buffer.
453 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid,
454 int segflg, dmu_tx_t *tx)
461 ASSERT(segflg != UIO_NOCOPY);
462 ASSERT(vp->v_mount != NULL);
466 off = start & PAGEOFFSET;
468 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
470 int nbytes = imin(PAGESIZE - off, len);
472 if ((pp = page_busy(vp, start, off, nbytes)) != NULL) {
473 VM_OBJECT_UNLOCK(obj);
475 va = zfs_map_page(pp, &sf);
476 (void) dmu_read(os, oid, start+off, nbytes,
477 va+off, DMU_READ_PREFETCH);;
486 vm_object_pip_wakeupn(obj, 0);
487 VM_OBJECT_UNLOCK(obj);
491 * Read with UIO_NOCOPY flag means that sendfile(2) requests
492 * ZFS to populate a range of page cache pages with data.
494 * NOTE: this function could be optimized to pre-allocate
495 * all pages in advance, drain VPO_BUSY on all of them,
496 * map them into contiguous KVA region and populate them
497 * in one single dmu_read() call.
500 mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio)
502 znode_t *zp = VTOZ(vp);
503 objset_t *os = zp->z_zfsvfs->z_os;
513 ASSERT(uio->uio_segflg == UIO_NOCOPY);
514 ASSERT(vp->v_mount != NULL);
517 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0);
520 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) {
521 int bytes = MIN(PAGESIZE, len);
523 pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_NOBUSY |
524 VM_ALLOC_NORMAL | VM_ALLOC_RETRY | VM_ALLOC_IGN_SBUSY);
525 if (pp->valid == 0) {
526 vm_page_io_start(pp);
527 VM_OBJECT_UNLOCK(obj);
528 va = zfs_map_page(pp, &sf);
529 error = dmu_read(os, zp->z_id, start, bytes, va,
531 if (bytes != PAGESIZE && error == 0)
532 bzero(va + bytes, PAGESIZE - bytes);
535 vm_page_io_finish(pp);
538 if (pp->wire_count == 0 && pp->valid == 0 &&
539 pp->busy == 0 && !(pp->oflags & VPO_BUSY))
542 pp->valid = VM_PAGE_BITS_ALL;
543 vm_page_activate(pp);
547 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
551 uio->uio_resid -= bytes;
552 uio->uio_offset += bytes;
555 VM_OBJECT_UNLOCK(obj);
560 * When a file is memory mapped, we must keep the IO data synchronized
561 * between the DMU cache and the memory mapped pages. What this means:
563 * On Read: We "read" preferentially from memory mapped pages,
564 * else we default from the dmu buffer.
566 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
567 * the file is memory mapped.
570 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
572 znode_t *zp = VTOZ(vp);
573 objset_t *os = zp->z_zfsvfs->z_os;
581 ASSERT(vp->v_mount != NULL);
585 start = uio->uio_loffset;
586 off = start & PAGEOFFSET;
588 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
590 uint64_t bytes = MIN(PAGESIZE - off, len);
592 if (pp = page_hold(vp, start)) {
596 VM_OBJECT_UNLOCK(obj);
597 va = zfs_map_page(pp, &sf);
598 error = uiomove(va + off, bytes, UIO_READ, uio);
603 VM_OBJECT_UNLOCK(obj);
604 error = dmu_read_uio(os, zp->z_id, uio, bytes);
612 VM_OBJECT_UNLOCK(obj);
616 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
619 * Read bytes from specified file into supplied buffer.
621 * IN: vp - vnode of file to be read from.
622 * uio - structure supplying read location, range info,
624 * ioflag - SYNC flags; used to provide FRSYNC semantics.
625 * cr - credentials of caller.
626 * ct - caller context
628 * OUT: uio - updated offset and range, buffer filled.
630 * RETURN: 0 on success, error code on failure.
633 * vp - atime updated if byte count > 0
637 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
639 znode_t *zp = VTOZ(vp);
640 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
651 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
653 return (SET_ERROR(EACCES));
657 * Validate file offset
659 if (uio->uio_loffset < (offset_t)0) {
661 return (SET_ERROR(EINVAL));
665 * Fasttrack empty reads
667 if (uio->uio_resid == 0) {
673 * Check for mandatory locks
675 if (MANDMODE(zp->z_mode)) {
676 if (error = chklock(vp, FREAD,
677 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
684 * If we're in FRSYNC mode, sync out this znode before reading it.
687 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
688 zil_commit(zfsvfs->z_log, zp->z_id);
691 * Lock the range against changes.
693 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
696 * If we are reading past end-of-file we can skip
697 * to the end; but we might still need to set atime.
699 if (uio->uio_loffset >= zp->z_size) {
704 ASSERT(uio->uio_loffset < zp->z_size);
705 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
708 if ((uio->uio_extflg == UIO_XUIO) &&
709 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
711 int blksz = zp->z_blksz;
712 uint64_t offset = uio->uio_loffset;
714 xuio = (xuio_t *)uio;
716 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
719 ASSERT(offset + n <= blksz);
722 (void) dmu_xuio_init(xuio, nblk);
724 if (vn_has_cached_data(vp)) {
726 * For simplicity, we always allocate a full buffer
727 * even if we only expect to read a portion of a block.
729 while (--nblk >= 0) {
730 (void) dmu_xuio_add(xuio,
731 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
739 nbytes = MIN(n, zfs_read_chunk_size -
740 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
743 if (uio->uio_segflg == UIO_NOCOPY)
744 error = mappedread_sf(vp, nbytes, uio);
746 #endif /* __FreeBSD__ */
747 if (vn_has_cached_data(vp))
748 error = mappedread(vp, nbytes, uio);
750 error = dmu_read_uio(os, zp->z_id, uio, nbytes);
752 /* convert checksum errors into IO errors */
754 error = SET_ERROR(EIO);
761 zfs_range_unlock(rl);
763 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
769 * Write the bytes to a file.
771 * IN: vp - vnode of file to be written to.
772 * uio - structure supplying write location, range info,
774 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is
775 * set if in append mode.
776 * cr - credentials of caller.
777 * ct - caller context (NFS/CIFS fem monitor only)
779 * OUT: uio - updated offset and range.
781 * RETURN: 0 on success, error code on failure.
784 * vp - ctime|mtime updated if byte count > 0
789 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
791 znode_t *zp = VTOZ(vp);
792 rlim64_t limit = MAXOFFSET_T;
793 ssize_t start_resid = uio->uio_resid;
797 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
802 int max_blksz = zfsvfs->z_max_blksz;
805 iovec_t *aiov = NULL;
808 int iovcnt = uio->uio_iovcnt;
809 iovec_t *iovp = uio->uio_iov;
812 sa_bulk_attr_t bulk[4];
813 uint64_t mtime[2], ctime[2];
816 * Fasttrack empty write
822 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
828 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
829 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
830 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
832 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
836 * If immutable or not appending then return EPERM
838 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
839 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
840 (uio->uio_loffset < zp->z_size))) {
842 return (SET_ERROR(EPERM));
845 zilog = zfsvfs->z_log;
848 * Validate file offset
850 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
853 return (SET_ERROR(EINVAL));
857 * Check for mandatory locks before calling zfs_range_lock()
858 * in order to prevent a deadlock with locks set via fcntl().
860 if (MANDMODE((mode_t)zp->z_mode) &&
861 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
868 * Pre-fault the pages to ensure slow (eg NFS) pages
870 * Skip this if uio contains loaned arc_buf.
872 if ((uio->uio_extflg == UIO_XUIO) &&
873 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
874 xuio = (xuio_t *)uio;
876 uio_prefaultpages(MIN(n, max_blksz), uio);
880 * If in append mode, set the io offset pointer to eof.
882 if (ioflag & FAPPEND) {
884 * Obtain an appending range lock to guarantee file append
885 * semantics. We reset the write offset once we have the lock.
887 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
889 if (rl->r_len == UINT64_MAX) {
891 * We overlocked the file because this write will cause
892 * the file block size to increase.
893 * Note that zp_size cannot change with this lock held.
897 uio->uio_loffset = woff;
900 * Note that if the file block size will change as a result of
901 * this write, then this range lock will lock the entire file
902 * so that we can re-write the block safely.
904 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
907 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) {
908 zfs_range_unlock(rl);
914 zfs_range_unlock(rl);
916 return (SET_ERROR(EFBIG));
919 if ((woff + n) > limit || woff > (limit - n))
922 /* Will this write extend the file length? */
923 write_eof = (woff + n > zp->z_size);
925 end_size = MAX(zp->z_size, woff + n);
928 * Write the file in reasonable size chunks. Each chunk is written
929 * in a separate transaction; this keeps the intent log records small
930 * and allows us to do more fine-grained space accounting.
934 woff = uio->uio_loffset;
935 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
936 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
938 dmu_return_arcbuf(abuf);
939 error = SET_ERROR(EDQUOT);
943 if (xuio && abuf == NULL) {
944 ASSERT(i_iov < iovcnt);
946 abuf = dmu_xuio_arcbuf(xuio, i_iov);
947 dmu_xuio_clear(xuio, i_iov);
948 DTRACE_PROBE3(zfs_cp_write, int, i_iov,
949 iovec_t *, aiov, arc_buf_t *, abuf);
950 ASSERT((aiov->iov_base == abuf->b_data) ||
951 ((char *)aiov->iov_base - (char *)abuf->b_data +
952 aiov->iov_len == arc_buf_size(abuf)));
954 } else if (abuf == NULL && n >= max_blksz &&
955 woff >= zp->z_size &&
956 P2PHASE(woff, max_blksz) == 0 &&
957 zp->z_blksz == max_blksz) {
959 * This write covers a full block. "Borrow" a buffer
960 * from the dmu so that we can fill it before we enter
961 * a transaction. This avoids the possibility of
962 * holding up the transaction if the data copy hangs
963 * up on a pagefault (e.g., from an NFS server mapping).
967 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
969 ASSERT(abuf != NULL);
970 ASSERT(arc_buf_size(abuf) == max_blksz);
971 if (error = uiocopy(abuf->b_data, max_blksz,
972 UIO_WRITE, uio, &cbytes)) {
973 dmu_return_arcbuf(abuf);
976 ASSERT(cbytes == max_blksz);
980 * Start a transaction.
982 tx = dmu_tx_create(zfsvfs->z_os);
983 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
984 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
985 zfs_sa_upgrade_txholds(tx, zp);
986 error = dmu_tx_assign(tx, TXG_WAIT);
990 dmu_return_arcbuf(abuf);
995 * If zfs_range_lock() over-locked we grow the blocksize
996 * and then reduce the lock range. This will only happen
997 * on the first iteration since zfs_range_reduce() will
998 * shrink down r_len to the appropriate size.
1000 if (rl->r_len == UINT64_MAX) {
1003 if (zp->z_blksz > max_blksz) {
1004 ASSERT(!ISP2(zp->z_blksz));
1005 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
1007 new_blksz = MIN(end_size, max_blksz);
1009 zfs_grow_blocksize(zp, new_blksz, tx);
1010 zfs_range_reduce(rl, woff, n);
1014 * XXX - should we really limit each write to z_max_blksz?
1015 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
1017 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
1019 if (woff + nbytes > zp->z_size)
1020 vnode_pager_setsize(vp, woff + nbytes);
1023 tx_bytes = uio->uio_resid;
1024 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
1026 tx_bytes -= uio->uio_resid;
1029 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
1031 * If this is not a full block write, but we are
1032 * extending the file past EOF and this data starts
1033 * block-aligned, use assign_arcbuf(). Otherwise,
1034 * write via dmu_write().
1036 if (tx_bytes < max_blksz && (!write_eof ||
1037 aiov->iov_base != abuf->b_data)) {
1039 dmu_write(zfsvfs->z_os, zp->z_id, woff,
1040 aiov->iov_len, aiov->iov_base, tx);
1041 dmu_return_arcbuf(abuf);
1042 xuio_stat_wbuf_copied();
1044 ASSERT(xuio || tx_bytes == max_blksz);
1045 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
1048 ASSERT(tx_bytes <= uio->uio_resid);
1049 uioskip(uio, tx_bytes);
1051 if (tx_bytes && vn_has_cached_data(vp)) {
1052 update_pages(vp, woff, tx_bytes, zfsvfs->z_os,
1053 zp->z_id, uio->uio_segflg, tx);
1057 * If we made no progress, we're done. If we made even
1058 * partial progress, update the znode and ZIL accordingly.
1060 if (tx_bytes == 0) {
1061 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
1062 (void *)&zp->z_size, sizeof (uint64_t), tx);
1069 * Clear Set-UID/Set-GID bits on successful write if not
1070 * privileged and at least one of the excute bits is set.
1072 * It would be nice to to this after all writes have
1073 * been done, but that would still expose the ISUID/ISGID
1074 * to another app after the partial write is committed.
1076 * Note: we don't call zfs_fuid_map_id() here because
1077 * user 0 is not an ephemeral uid.
1079 mutex_enter(&zp->z_acl_lock);
1080 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
1081 (S_IXUSR >> 6))) != 0 &&
1082 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
1083 secpolicy_vnode_setid_retain(vp, cr,
1084 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
1086 zp->z_mode &= ~(S_ISUID | S_ISGID);
1087 newmode = zp->z_mode;
1088 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
1089 (void *)&newmode, sizeof (uint64_t), tx);
1091 mutex_exit(&zp->z_acl_lock);
1093 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
1097 * Update the file size (zp_size) if it has changed;
1098 * account for possible concurrent updates.
1100 while ((end_size = zp->z_size) < uio->uio_loffset) {
1101 (void) atomic_cas_64(&zp->z_size, end_size,
1106 * If we are replaying and eof is non zero then force
1107 * the file size to the specified eof. Note, there's no
1108 * concurrency during replay.
1110 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
1111 zp->z_size = zfsvfs->z_replay_eof;
1113 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1115 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
1120 ASSERT(tx_bytes == nbytes);
1125 uio_prefaultpages(MIN(n, max_blksz), uio);
1129 zfs_range_unlock(rl);
1132 * If we're in replay mode, or we made no progress, return error.
1133 * Otherwise, it's at least a partial write, so it's successful.
1135 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1140 if (ioflag & (FSYNC | FDSYNC) ||
1141 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1142 zil_commit(zilog, zp->z_id);
1149 zfs_get_done(zgd_t *zgd, int error)
1151 znode_t *zp = zgd->zgd_private;
1152 objset_t *os = zp->z_zfsvfs->z_os;
1156 dmu_buf_rele(zgd->zgd_db, zgd);
1158 zfs_range_unlock(zgd->zgd_rl);
1160 vfslocked = VFS_LOCK_GIANT(zp->z_zfsvfs->z_vfs);
1162 * Release the vnode asynchronously as we currently have the
1163 * txg stopped from syncing.
1165 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1167 if (error == 0 && zgd->zgd_bp)
1168 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1170 kmem_free(zgd, sizeof (zgd_t));
1171 VFS_UNLOCK_GIANT(vfslocked);
1175 static int zil_fault_io = 0;
1179 * Get data to generate a TX_WRITE intent log record.
1182 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1184 zfsvfs_t *zfsvfs = arg;
1185 objset_t *os = zfsvfs->z_os;
1187 uint64_t object = lr->lr_foid;
1188 uint64_t offset = lr->lr_offset;
1189 uint64_t size = lr->lr_length;
1190 blkptr_t *bp = &lr->lr_blkptr;
1195 ASSERT(zio != NULL);
1199 * Nothing to do if the file has been removed
1201 if (zfs_zget(zfsvfs, object, &zp) != 0)
1202 return (SET_ERROR(ENOENT));
1203 if (zp->z_unlinked) {
1205 * Release the vnode asynchronously as we currently have the
1206 * txg stopped from syncing.
1208 VN_RELE_ASYNC(ZTOV(zp),
1209 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1210 return (SET_ERROR(ENOENT));
1213 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1214 zgd->zgd_zilog = zfsvfs->z_log;
1215 zgd->zgd_private = zp;
1218 * Write records come in two flavors: immediate and indirect.
1219 * For small writes it's cheaper to store the data with the
1220 * log record (immediate); for large writes it's cheaper to
1221 * sync the data and get a pointer to it (indirect) so that
1222 * we don't have to write the data twice.
1224 if (buf != NULL) { /* immediate write */
1225 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1226 /* test for truncation needs to be done while range locked */
1227 if (offset >= zp->z_size) {
1228 error = SET_ERROR(ENOENT);
1230 error = dmu_read(os, object, offset, size, buf,
1231 DMU_READ_NO_PREFETCH);
1233 ASSERT(error == 0 || error == ENOENT);
1234 } else { /* indirect write */
1236 * Have to lock the whole block to ensure when it's
1237 * written out and it's checksum is being calculated
1238 * that no one can change the data. We need to re-check
1239 * blocksize after we get the lock in case it's changed!
1244 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1246 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1248 if (zp->z_blksz == size)
1251 zfs_range_unlock(zgd->zgd_rl);
1253 /* test for truncation needs to be done while range locked */
1254 if (lr->lr_offset >= zp->z_size)
1255 error = SET_ERROR(ENOENT);
1258 error = SET_ERROR(EIO);
1263 error = dmu_buf_hold(os, object, offset, zgd, &db,
1264 DMU_READ_NO_PREFETCH);
1267 blkptr_t *obp = dmu_buf_get_blkptr(db);
1269 ASSERT(BP_IS_HOLE(bp));
1276 ASSERT(db->db_offset == offset);
1277 ASSERT(db->db_size == size);
1279 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1281 ASSERT(error || lr->lr_length <= zp->z_blksz);
1284 * On success, we need to wait for the write I/O
1285 * initiated by dmu_sync() to complete before we can
1286 * release this dbuf. We will finish everything up
1287 * in the zfs_get_done() callback.
1292 if (error == EALREADY) {
1293 lr->lr_common.lrc_txtype = TX_WRITE2;
1299 zfs_get_done(zgd, error);
1306 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1307 caller_context_t *ct)
1309 znode_t *zp = VTOZ(vp);
1310 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1316 if (flag & V_ACE_MASK)
1317 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1319 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1326 * If vnode is for a device return a specfs vnode instead.
1329 specvp_check(vnode_t **vpp, cred_t *cr)
1333 if (IS_DEVVP(*vpp)) {
1336 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1339 error = SET_ERROR(ENOSYS);
1347 * Lookup an entry in a directory, or an extended attribute directory.
1348 * If it exists, return a held vnode reference for it.
1350 * IN: dvp - vnode of directory to search.
1351 * nm - name of entry to lookup.
1352 * pnp - full pathname to lookup [UNUSED].
1353 * flags - LOOKUP_XATTR set if looking for an attribute.
1354 * rdir - root directory vnode [UNUSED].
1355 * cr - credentials of caller.
1356 * ct - caller context
1357 * direntflags - directory lookup flags
1358 * realpnp - returned pathname.
1360 * OUT: vpp - vnode of located entry, NULL if not found.
1362 * RETURN: 0 on success, error code on failure.
1369 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp,
1370 int nameiop, cred_t *cr, kthread_t *td, int flags)
1372 znode_t *zdp = VTOZ(dvp);
1373 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1375 int *direntflags = NULL;
1376 void *realpnp = NULL;
1379 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1381 if (dvp->v_type != VDIR) {
1382 return (SET_ERROR(ENOTDIR));
1383 } else if (zdp->z_sa_hdl == NULL) {
1384 return (SET_ERROR(EIO));
1387 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1388 error = zfs_fastaccesschk_execute(zdp, cr);
1396 vnode_t *tvp = dnlc_lookup(dvp, nm);
1399 error = zfs_fastaccesschk_execute(zdp, cr);
1404 if (tvp == DNLC_NO_VNODE) {
1406 return (SET_ERROR(ENOENT));
1409 return (specvp_check(vpp, cr));
1415 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1422 if (flags & LOOKUP_XATTR) {
1425 * If the xattr property is off, refuse the lookup request.
1427 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1429 return (SET_ERROR(EINVAL));
1434 * We don't allow recursive attributes..
1435 * Maybe someday we will.
1437 if (zdp->z_pflags & ZFS_XATTR) {
1439 return (SET_ERROR(EINVAL));
1442 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1448 * Do we have permission to get into attribute directory?
1451 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1461 if (dvp->v_type != VDIR) {
1463 return (SET_ERROR(ENOTDIR));
1467 * Check accessibility of directory.
1470 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1475 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1476 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1478 return (SET_ERROR(EILSEQ));
1481 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1483 error = specvp_check(vpp, cr);
1485 /* Translate errors and add SAVENAME when needed. */
1486 if (cnp->cn_flags & ISLASTCN) {
1490 if (error == ENOENT) {
1491 error = EJUSTRETURN;
1492 cnp->cn_flags |= SAVENAME;
1498 cnp->cn_flags |= SAVENAME;
1502 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) {
1505 if (cnp->cn_flags & ISDOTDOT) {
1506 ltype = VOP_ISLOCKED(dvp);
1510 error = vn_lock(*vpp, cnp->cn_lkflags);
1511 if (cnp->cn_flags & ISDOTDOT)
1512 vn_lock(dvp, ltype | LK_RETRY);
1522 #ifdef FREEBSD_NAMECACHE
1524 * Insert name into cache (as non-existent) if appropriate.
1526 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
1527 cache_enter(dvp, *vpp, cnp);
1529 * Insert name into cache if appropriate.
1531 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) {
1532 if (!(cnp->cn_flags & ISLASTCN) ||
1533 (nameiop != DELETE && nameiop != RENAME)) {
1534 cache_enter(dvp, *vpp, cnp);
1543 * Attempt to create a new entry in a directory. If the entry
1544 * already exists, truncate the file if permissible, else return
1545 * an error. Return the vp of the created or trunc'd file.
1547 * IN: dvp - vnode of directory to put new file entry in.
1548 * name - name of new file entry.
1549 * vap - attributes of new file.
1550 * excl - flag indicating exclusive or non-exclusive mode.
1551 * mode - mode to open file with.
1552 * cr - credentials of caller.
1553 * flag - large file flag [UNUSED].
1554 * ct - caller context
1555 * vsecp - ACL to be set
1557 * OUT: vpp - vnode of created or trunc'd entry.
1559 * RETURN: 0 on success, error code on failure.
1562 * dvp - ctime|mtime updated if new entry created
1563 * vp - ctime|mtime always, atime if new
1568 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode,
1569 vnode_t **vpp, cred_t *cr, kthread_t *td)
1571 znode_t *zp, *dzp = VTOZ(dvp);
1572 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1580 gid_t gid = crgetgid(cr);
1581 zfs_acl_ids_t acl_ids;
1582 boolean_t fuid_dirtied;
1583 boolean_t have_acl = B_FALSE;
1584 boolean_t waited = B_FALSE;
1589 * If we have an ephemeral id, ACL, or XVATTR then
1590 * make sure file system is at proper version
1593 ksid = crgetsid(cr, KSID_OWNER);
1595 uid = ksid_getid(ksid);
1599 if (zfsvfs->z_use_fuids == B_FALSE &&
1600 (vsecp || (vap->va_mask & AT_XVATTR) ||
1601 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1602 return (SET_ERROR(EINVAL));
1607 zilog = zfsvfs->z_log;
1609 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1610 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1612 return (SET_ERROR(EILSEQ));
1615 if (vap->va_mask & AT_XVATTR) {
1616 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
1617 crgetuid(cr), cr, vap->va_type)) != 0) {
1623 getnewvnode_reserve(1);
1628 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr))
1629 vap->va_mode &= ~S_ISVTX;
1631 if (*name == '\0') {
1633 * Null component name refers to the directory itself.
1640 /* possible VN_HOLD(zp) */
1643 if (flag & FIGNORECASE)
1646 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1650 zfs_acl_ids_free(&acl_ids);
1651 if (strcmp(name, "..") == 0)
1652 error = SET_ERROR(EISDIR);
1653 getnewvnode_drop_reserve();
1663 * Create a new file object and update the directory
1666 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1668 zfs_acl_ids_free(&acl_ids);
1673 * We only support the creation of regular files in
1674 * extended attribute directories.
1677 if ((dzp->z_pflags & ZFS_XATTR) &&
1678 (vap->va_type != VREG)) {
1680 zfs_acl_ids_free(&acl_ids);
1681 error = SET_ERROR(EINVAL);
1685 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1686 cr, vsecp, &acl_ids)) != 0)
1690 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1691 zfs_acl_ids_free(&acl_ids);
1692 error = SET_ERROR(EDQUOT);
1696 tx = dmu_tx_create(os);
1698 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1699 ZFS_SA_BASE_ATTR_SIZE);
1701 fuid_dirtied = zfsvfs->z_fuid_dirty;
1703 zfs_fuid_txhold(zfsvfs, tx);
1704 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1705 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1706 if (!zfsvfs->z_use_sa &&
1707 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1708 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1709 0, acl_ids.z_aclp->z_acl_bytes);
1711 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1713 zfs_dirent_unlock(dl);
1714 if (error == ERESTART) {
1720 zfs_acl_ids_free(&acl_ids);
1722 getnewvnode_drop_reserve();
1726 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1729 zfs_fuid_sync(zfsvfs, tx);
1731 (void) zfs_link_create(dl, zp, tx, ZNEW);
1732 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1733 if (flag & FIGNORECASE)
1735 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1736 vsecp, acl_ids.z_fuidp, vap);
1737 zfs_acl_ids_free(&acl_ids);
1740 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1743 zfs_acl_ids_free(&acl_ids);
1747 * A directory entry already exists for this name.
1750 * Can't truncate an existing file if in exclusive mode.
1753 error = SET_ERROR(EEXIST);
1757 * Can't open a directory for writing.
1759 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1760 error = SET_ERROR(EISDIR);
1764 * Verify requested access to file.
1766 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1770 mutex_enter(&dzp->z_lock);
1772 mutex_exit(&dzp->z_lock);
1775 * Truncate regular files if requested.
1777 if ((ZTOV(zp)->v_type == VREG) &&
1778 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1779 /* we can't hold any locks when calling zfs_freesp() */
1780 zfs_dirent_unlock(dl);
1782 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1784 vnevent_create(ZTOV(zp), ct);
1789 getnewvnode_drop_reserve();
1791 zfs_dirent_unlock(dl);
1798 error = specvp_check(vpp, cr);
1801 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1802 zil_commit(zilog, 0);
1809 * Remove an entry from a directory.
1811 * IN: dvp - vnode of directory to remove entry from.
1812 * name - name of entry to remove.
1813 * cr - credentials of caller.
1814 * ct - caller context
1815 * flags - case flags
1817 * RETURN: 0 on success, error code on failure.
1821 * vp - ctime (if nlink > 0)
1824 uint64_t null_xattr = 0;
1828 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1831 znode_t *zp, *dzp = VTOZ(dvp);
1834 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1836 uint64_t acl_obj, xattr_obj;
1837 uint64_t xattr_obj_unlinked = 0;
1841 boolean_t may_delete_now, delete_now = FALSE;
1842 boolean_t unlinked, toobig = FALSE;
1844 pathname_t *realnmp = NULL;
1848 boolean_t waited = B_FALSE;
1852 zilog = zfsvfs->z_log;
1854 if (flags & FIGNORECASE) {
1864 * Attempt to lock directory; fail if entry doesn't exist.
1866 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1876 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1881 * Need to use rmdir for removing directories.
1883 if (vp->v_type == VDIR) {
1884 error = SET_ERROR(EPERM);
1888 vnevent_remove(vp, dvp, name, ct);
1891 dnlc_remove(dvp, realnmp->pn_buf);
1893 dnlc_remove(dvp, name);
1896 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1900 * We may delete the znode now, or we may put it in the unlinked set;
1901 * it depends on whether we're the last link, and on whether there are
1902 * other holds on the vnode. So we dmu_tx_hold() the right things to
1903 * allow for either case.
1906 tx = dmu_tx_create(zfsvfs->z_os);
1907 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1908 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1909 zfs_sa_upgrade_txholds(tx, zp);
1910 zfs_sa_upgrade_txholds(tx, dzp);
1911 if (may_delete_now) {
1913 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1914 /* if the file is too big, only hold_free a token amount */
1915 dmu_tx_hold_free(tx, zp->z_id, 0,
1916 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1919 /* are there any extended attributes? */
1920 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1921 &xattr_obj, sizeof (xattr_obj));
1922 if (error == 0 && xattr_obj) {
1923 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1925 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1926 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1929 mutex_enter(&zp->z_lock);
1930 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1931 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1932 mutex_exit(&zp->z_lock);
1934 /* charge as an update -- would be nice not to charge at all */
1935 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1937 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1939 zfs_dirent_unlock(dl);
1943 if (error == ERESTART) {
1957 * Remove the directory entry.
1959 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1969 * Hold z_lock so that we can make sure that the ACL obj
1970 * hasn't changed. Could have been deleted due to
1973 mutex_enter(&zp->z_lock);
1975 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1976 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1977 delete_now = may_delete_now && !toobig &&
1978 vp->v_count == 1 && !vn_has_cached_data(vp) &&
1979 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1986 panic("zfs_remove: delete_now branch taken");
1988 if (xattr_obj_unlinked) {
1989 ASSERT3U(xzp->z_links, ==, 2);
1990 mutex_enter(&xzp->z_lock);
1991 xzp->z_unlinked = 1;
1993 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1994 &xzp->z_links, sizeof (xzp->z_links), tx);
1995 ASSERT3U(error, ==, 0);
1996 mutex_exit(&xzp->z_lock);
1997 zfs_unlinked_add(xzp, tx);
2000 error = sa_remove(zp->z_sa_hdl,
2001 SA_ZPL_XATTR(zfsvfs), tx);
2003 error = sa_update(zp->z_sa_hdl,
2004 SA_ZPL_XATTR(zfsvfs), &null_xattr,
2005 sizeof (uint64_t), tx);
2010 ASSERT0(vp->v_count);
2012 mutex_exit(&zp->z_lock);
2013 zfs_znode_delete(zp, tx);
2014 } else if (unlinked) {
2015 mutex_exit(&zp->z_lock);
2016 zfs_unlinked_add(zp, tx);
2018 vp->v_vflag |= VV_NOSYNC;
2023 if (flags & FIGNORECASE)
2025 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
2032 zfs_dirent_unlock(dl);
2039 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2040 zil_commit(zilog, 0);
2047 * Create a new directory and insert it into dvp using the name
2048 * provided. Return a pointer to the inserted directory.
2050 * IN: dvp - vnode of directory to add subdir to.
2051 * dirname - name of new directory.
2052 * vap - attributes of new directory.
2053 * cr - credentials of caller.
2054 * ct - caller context
2055 * flags - case flags
2056 * vsecp - ACL to be set
2058 * OUT: vpp - vnode of created directory.
2060 * RETURN: 0 on success, error code on failure.
2063 * dvp - ctime|mtime updated
2064 * vp - ctime|mtime|atime updated
2068 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
2069 caller_context_t *ct, int flags, vsecattr_t *vsecp)
2071 znode_t *zp, *dzp = VTOZ(dvp);
2072 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2081 gid_t gid = crgetgid(cr);
2082 zfs_acl_ids_t acl_ids;
2083 boolean_t fuid_dirtied;
2084 boolean_t waited = B_FALSE;
2086 ASSERT(vap->va_type == VDIR);
2089 * If we have an ephemeral id, ACL, or XVATTR then
2090 * make sure file system is at proper version
2093 ksid = crgetsid(cr, KSID_OWNER);
2095 uid = ksid_getid(ksid);
2098 if (zfsvfs->z_use_fuids == B_FALSE &&
2099 (vsecp || (vap->va_mask & AT_XVATTR) ||
2100 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
2101 return (SET_ERROR(EINVAL));
2105 zilog = zfsvfs->z_log;
2107 if (dzp->z_pflags & ZFS_XATTR) {
2109 return (SET_ERROR(EINVAL));
2112 if (zfsvfs->z_utf8 && u8_validate(dirname,
2113 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2115 return (SET_ERROR(EILSEQ));
2117 if (flags & FIGNORECASE)
2120 if (vap->va_mask & AT_XVATTR) {
2121 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
2122 crgetuid(cr), cr, vap->va_type)) != 0) {
2128 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
2129 vsecp, &acl_ids)) != 0) {
2134 getnewvnode_reserve(1);
2137 * First make sure the new directory doesn't exist.
2139 * Existence is checked first to make sure we don't return
2140 * EACCES instead of EEXIST which can cause some applications
2146 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
2148 zfs_acl_ids_free(&acl_ids);
2149 getnewvnode_drop_reserve();
2154 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
2155 zfs_acl_ids_free(&acl_ids);
2156 zfs_dirent_unlock(dl);
2157 getnewvnode_drop_reserve();
2162 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
2163 zfs_acl_ids_free(&acl_ids);
2164 zfs_dirent_unlock(dl);
2165 getnewvnode_drop_reserve();
2167 return (SET_ERROR(EDQUOT));
2171 * Add a new entry to the directory.
2173 tx = dmu_tx_create(zfsvfs->z_os);
2174 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2175 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2176 fuid_dirtied = zfsvfs->z_fuid_dirty;
2178 zfs_fuid_txhold(zfsvfs, tx);
2179 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2180 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2181 acl_ids.z_aclp->z_acl_bytes);
2184 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2185 ZFS_SA_BASE_ATTR_SIZE);
2187 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2189 zfs_dirent_unlock(dl);
2190 if (error == ERESTART) {
2196 zfs_acl_ids_free(&acl_ids);
2198 getnewvnode_drop_reserve();
2206 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2209 zfs_fuid_sync(zfsvfs, tx);
2212 * Now put new name in parent dir.
2214 (void) zfs_link_create(dl, zp, tx, ZNEW);
2218 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2219 if (flags & FIGNORECASE)
2221 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2222 acl_ids.z_fuidp, vap);
2224 zfs_acl_ids_free(&acl_ids);
2228 getnewvnode_drop_reserve();
2230 zfs_dirent_unlock(dl);
2232 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2233 zil_commit(zilog, 0);
2240 * Remove a directory subdir entry. If the current working
2241 * directory is the same as the subdir to be removed, the
2244 * IN: dvp - vnode of directory to remove from.
2245 * name - name of directory to be removed.
2246 * cwd - vnode of current working directory.
2247 * cr - credentials of caller.
2248 * ct - caller context
2249 * flags - case flags
2251 * RETURN: 0 on success, error code on failure.
2254 * dvp - ctime|mtime updated
2258 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2259 caller_context_t *ct, int flags)
2261 znode_t *dzp = VTOZ(dvp);
2264 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2270 boolean_t waited = B_FALSE;
2274 zilog = zfsvfs->z_log;
2276 if (flags & FIGNORECASE)
2282 * Attempt to lock directory; fail if entry doesn't exist.
2284 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2292 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2296 if (vp->v_type != VDIR) {
2297 error = SET_ERROR(ENOTDIR);
2302 error = SET_ERROR(EINVAL);
2306 vnevent_rmdir(vp, dvp, name, ct);
2309 * Grab a lock on the directory to make sure that noone is
2310 * trying to add (or lookup) entries while we are removing it.
2312 rw_enter(&zp->z_name_lock, RW_WRITER);
2315 * Grab a lock on the parent pointer to make sure we play well
2316 * with the treewalk and directory rename code.
2318 rw_enter(&zp->z_parent_lock, RW_WRITER);
2320 tx = dmu_tx_create(zfsvfs->z_os);
2321 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2322 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2323 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2324 zfs_sa_upgrade_txholds(tx, zp);
2325 zfs_sa_upgrade_txholds(tx, dzp);
2326 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2328 rw_exit(&zp->z_parent_lock);
2329 rw_exit(&zp->z_name_lock);
2330 zfs_dirent_unlock(dl);
2332 if (error == ERESTART) {
2343 #ifdef FREEBSD_NAMECACHE
2347 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2350 uint64_t txtype = TX_RMDIR;
2351 if (flags & FIGNORECASE)
2353 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2358 rw_exit(&zp->z_parent_lock);
2359 rw_exit(&zp->z_name_lock);
2360 #ifdef FREEBSD_NAMECACHE
2364 zfs_dirent_unlock(dl);
2368 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2369 zil_commit(zilog, 0);
2376 * Read as many directory entries as will fit into the provided
2377 * buffer from the given directory cursor position (specified in
2378 * the uio structure).
2380 * IN: vp - vnode of directory to read.
2381 * uio - structure supplying read location, range info,
2382 * and return buffer.
2383 * cr - credentials of caller.
2384 * ct - caller context
2385 * flags - case flags
2387 * OUT: uio - updated offset and range, buffer filled.
2388 * eofp - set to true if end-of-file detected.
2390 * RETURN: 0 on success, error code on failure.
2393 * vp - atime updated
2395 * Note that the low 4 bits of the cookie returned by zap is always zero.
2396 * This allows us to use the low range for "special" directory entries:
2397 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2398 * we use the offset 2 for the '.zfs' directory.
2402 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies)
2404 znode_t *zp = VTOZ(vp);
2408 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2413 zap_attribute_t zap;
2414 uint_t bytes_wanted;
2415 uint64_t offset; /* must be unsigned; checks for < 1 */
2421 boolean_t check_sysattrs;
2424 u_long *cooks = NULL;
2430 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2431 &parent, sizeof (parent))) != 0) {
2437 * If we are not given an eof variable,
2444 * Check for valid iov_len.
2446 if (uio->uio_iov->iov_len <= 0) {
2448 return (SET_ERROR(EINVAL));
2452 * Quit if directory has been removed (posix)
2454 if ((*eofp = zp->z_unlinked) != 0) {
2461 offset = uio->uio_loffset;
2462 prefetch = zp->z_zn_prefetch;
2465 * Initialize the iterator cursor.
2469 * Start iteration from the beginning of the directory.
2471 zap_cursor_init(&zc, os, zp->z_id);
2474 * The offset is a serialized cursor.
2476 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2480 * Get space to change directory entries into fs independent format.
2482 iovp = uio->uio_iov;
2483 bytes_wanted = iovp->iov_len;
2484 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2485 bufsize = bytes_wanted;
2486 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2487 odp = (struct dirent64 *)outbuf;
2489 bufsize = bytes_wanted;
2491 odp = (struct dirent64 *)iovp->iov_base;
2493 eodp = (struct edirent *)odp;
2495 if (ncookies != NULL) {
2497 * Minimum entry size is dirent size and 1 byte for a file name.
2499 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1);
2500 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK);
2505 * If this VFS supports the system attribute view interface; and
2506 * we're looking at an extended attribute directory; and we care
2507 * about normalization conflicts on this vfs; then we must check
2508 * for normalization conflicts with the sysattr name space.
2511 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2512 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2513 (flags & V_RDDIR_ENTFLAGS);
2519 * Transform to file-system independent format
2522 while (outcount < bytes_wanted) {
2525 off64_t *next = NULL;
2528 * Special case `.', `..', and `.zfs'.
2531 (void) strcpy(zap.za_name, ".");
2532 zap.za_normalization_conflict = 0;
2535 } else if (offset == 1) {
2536 (void) strcpy(zap.za_name, "..");
2537 zap.za_normalization_conflict = 0;
2540 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2541 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2542 zap.za_normalization_conflict = 0;
2543 objnum = ZFSCTL_INO_ROOT;
2549 if (error = zap_cursor_retrieve(&zc, &zap)) {
2550 if ((*eofp = (error == ENOENT)) != 0)
2556 if (zap.za_integer_length != 8 ||
2557 zap.za_num_integers != 1) {
2558 cmn_err(CE_WARN, "zap_readdir: bad directory "
2559 "entry, obj = %lld, offset = %lld\n",
2560 (u_longlong_t)zp->z_id,
2561 (u_longlong_t)offset);
2562 error = SET_ERROR(ENXIO);
2566 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2568 * MacOS X can extract the object type here such as:
2569 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2571 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2573 if (check_sysattrs && !zap.za_normalization_conflict) {
2575 zap.za_normalization_conflict =
2576 xattr_sysattr_casechk(zap.za_name);
2578 panic("%s:%u: TODO", __func__, __LINE__);
2583 if (flags & V_RDDIR_ACCFILTER) {
2585 * If we have no access at all, don't include
2586 * this entry in the returned information
2589 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2591 if (!zfs_has_access(ezp, cr)) {
2598 if (flags & V_RDDIR_ENTFLAGS)
2599 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2601 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2604 * Will this entry fit in the buffer?
2606 if (outcount + reclen > bufsize) {
2608 * Did we manage to fit anything in the buffer?
2611 error = SET_ERROR(EINVAL);
2616 if (flags & V_RDDIR_ENTFLAGS) {
2618 * Add extended flag entry:
2620 eodp->ed_ino = objnum;
2621 eodp->ed_reclen = reclen;
2622 /* NOTE: ed_off is the offset for the *next* entry */
2623 next = &(eodp->ed_off);
2624 eodp->ed_eflags = zap.za_normalization_conflict ?
2625 ED_CASE_CONFLICT : 0;
2626 (void) strncpy(eodp->ed_name, zap.za_name,
2627 EDIRENT_NAMELEN(reclen));
2628 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2633 odp->d_ino = objnum;
2634 odp->d_reclen = reclen;
2635 odp->d_namlen = strlen(zap.za_name);
2636 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1);
2638 odp = (dirent64_t *)((intptr_t)odp + reclen);
2642 ASSERT(outcount <= bufsize);
2644 /* Prefetch znode */
2646 dmu_prefetch(os, objnum, 0, 0);
2650 * Move to the next entry, fill in the previous offset.
2652 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2653 zap_cursor_advance(&zc);
2654 offset = zap_cursor_serialize(&zc);
2659 if (cooks != NULL) {
2662 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks));
2665 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2667 /* Subtract unused cookies */
2668 if (ncookies != NULL)
2669 *ncookies -= ncooks;
2671 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2672 iovp->iov_base += outcount;
2673 iovp->iov_len -= outcount;
2674 uio->uio_resid -= outcount;
2675 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2677 * Reset the pointer.
2679 offset = uio->uio_loffset;
2683 zap_cursor_fini(&zc);
2684 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2685 kmem_free(outbuf, bufsize);
2687 if (error == ENOENT)
2690 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2692 uio->uio_loffset = offset;
2694 if (error != 0 && cookies != NULL) {
2695 free(*cookies, M_TEMP);
2702 ulong_t zfs_fsync_sync_cnt = 4;
2705 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2707 znode_t *zp = VTOZ(vp);
2708 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2710 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2712 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2715 zil_commit(zfsvfs->z_log, zp->z_id);
2723 * Get the requested file attributes and place them in the provided
2726 * IN: vp - vnode of file.
2727 * vap - va_mask identifies requested attributes.
2728 * If AT_XVATTR set, then optional attrs are requested
2729 * flags - ATTR_NOACLCHECK (CIFS server context)
2730 * cr - credentials of caller.
2731 * ct - caller context
2733 * OUT: vap - attribute values.
2735 * RETURN: 0 (always succeeds).
2739 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2740 caller_context_t *ct)
2742 znode_t *zp = VTOZ(vp);
2743 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2746 u_longlong_t nblocks;
2748 uint64_t mtime[2], ctime[2], crtime[2], rdev;
2749 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2750 xoptattr_t *xoap = NULL;
2751 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2752 sa_bulk_attr_t bulk[4];
2758 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2760 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2761 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2762 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
2763 if (vp->v_type == VBLK || vp->v_type == VCHR)
2764 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
2767 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2773 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2774 * Also, if we are the owner don't bother, since owner should
2775 * always be allowed to read basic attributes of file.
2777 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2778 (vap->va_uid != crgetuid(cr))) {
2779 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2787 * Return all attributes. It's cheaper to provide the answer
2788 * than to determine whether we were asked the question.
2791 mutex_enter(&zp->z_lock);
2792 vap->va_type = IFTOVT(zp->z_mode);
2793 vap->va_mode = zp->z_mode & ~S_IFMT;
2795 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2797 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
2799 vap->va_nodeid = zp->z_id;
2800 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2801 links = zp->z_links + 1;
2803 links = zp->z_links;
2804 vap->va_nlink = MIN(links, LINK_MAX); /* nlink_t limit! */
2805 vap->va_size = zp->z_size;
2807 vap->va_rdev = vp->v_rdev;
2809 if (vp->v_type == VBLK || vp->v_type == VCHR)
2810 vap->va_rdev = zfs_cmpldev(rdev);
2812 vap->va_seq = zp->z_seq;
2813 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */
2814 vap->va_filerev = zp->z_seq;
2817 * Add in any requested optional attributes and the create time.
2818 * Also set the corresponding bits in the returned attribute bitmap.
2820 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2821 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2823 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2824 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2827 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2828 xoap->xoa_readonly =
2829 ((zp->z_pflags & ZFS_READONLY) != 0);
2830 XVA_SET_RTN(xvap, XAT_READONLY);
2833 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2835 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2836 XVA_SET_RTN(xvap, XAT_SYSTEM);
2839 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2841 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2842 XVA_SET_RTN(xvap, XAT_HIDDEN);
2845 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2846 xoap->xoa_nounlink =
2847 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2848 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2851 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2852 xoap->xoa_immutable =
2853 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2854 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2857 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2858 xoap->xoa_appendonly =
2859 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2860 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2863 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2865 ((zp->z_pflags & ZFS_NODUMP) != 0);
2866 XVA_SET_RTN(xvap, XAT_NODUMP);
2869 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2871 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2872 XVA_SET_RTN(xvap, XAT_OPAQUE);
2875 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2876 xoap->xoa_av_quarantined =
2877 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2878 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2881 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2882 xoap->xoa_av_modified =
2883 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2884 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2887 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2888 vp->v_type == VREG) {
2889 zfs_sa_get_scanstamp(zp, xvap);
2892 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2895 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2896 times, sizeof (times));
2897 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2898 XVA_SET_RTN(xvap, XAT_CREATETIME);
2901 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2902 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2903 XVA_SET_RTN(xvap, XAT_REPARSE);
2905 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2906 xoap->xoa_generation = zp->z_gen;
2907 XVA_SET_RTN(xvap, XAT_GEN);
2910 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2912 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2913 XVA_SET_RTN(xvap, XAT_OFFLINE);
2916 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2918 ((zp->z_pflags & ZFS_SPARSE) != 0);
2919 XVA_SET_RTN(xvap, XAT_SPARSE);
2923 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2924 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2925 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2926 ZFS_TIME_DECODE(&vap->va_birthtime, crtime);
2928 mutex_exit(&zp->z_lock);
2930 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2931 vap->va_blksize = blksize;
2932 vap->va_bytes = nblocks << 9; /* nblocks * 512 */
2934 if (zp->z_blksz == 0) {
2936 * Block size hasn't been set; suggest maximal I/O transfers.
2938 vap->va_blksize = zfsvfs->z_max_blksz;
2946 * Set the file attributes to the values contained in the
2949 * IN: vp - vnode of file to be modified.
2950 * vap - new attribute values.
2951 * If AT_XVATTR set, then optional attrs are being set
2952 * flags - ATTR_UTIME set if non-default time values provided.
2953 * - ATTR_NOACLCHECK (CIFS context only).
2954 * cr - credentials of caller.
2955 * ct - caller context
2957 * RETURN: 0 on success, error code on failure.
2960 * vp - ctime updated, mtime updated if size changed.
2964 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2965 caller_context_t *ct)
2967 znode_t *zp = VTOZ(vp);
2968 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2973 uint_t mask = vap->va_mask;
2974 uint_t saved_mask = 0;
2975 uint64_t saved_mode;
2978 uint64_t new_uid, new_gid;
2980 uint64_t mtime[2], ctime[2];
2982 int need_policy = FALSE;
2984 zfs_fuid_info_t *fuidp = NULL;
2985 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2988 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2989 boolean_t fuid_dirtied = B_FALSE;
2990 sa_bulk_attr_t bulk[7], xattr_bulk[7];
2991 int count = 0, xattr_count = 0;
2996 if (mask & AT_NOSET)
2997 return (SET_ERROR(EINVAL));
3002 zilog = zfsvfs->z_log;
3005 * Make sure that if we have ephemeral uid/gid or xvattr specified
3006 * that file system is at proper version level
3009 if (zfsvfs->z_use_fuids == B_FALSE &&
3010 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
3011 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
3012 (mask & AT_XVATTR))) {
3014 return (SET_ERROR(EINVAL));
3017 if (mask & AT_SIZE && vp->v_type == VDIR) {
3019 return (SET_ERROR(EISDIR));
3022 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
3024 return (SET_ERROR(EINVAL));
3028 * If this is an xvattr_t, then get a pointer to the structure of
3029 * optional attributes. If this is NULL, then we have a vattr_t.
3031 xoap = xva_getxoptattr(xvap);
3033 xva_init(&tmpxvattr);
3036 * Immutable files can only alter immutable bit and atime
3038 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
3039 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
3040 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
3042 return (SET_ERROR(EPERM));
3045 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
3047 return (SET_ERROR(EPERM));
3051 * Verify timestamps doesn't overflow 32 bits.
3052 * ZFS can handle large timestamps, but 32bit syscalls can't
3053 * handle times greater than 2039. This check should be removed
3054 * once large timestamps are fully supported.
3056 if (mask & (AT_ATIME | AT_MTIME)) {
3057 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
3058 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
3060 return (SET_ERROR(EOVERFLOW));
3068 /* Can this be moved to before the top label? */
3069 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
3071 return (SET_ERROR(EROFS));
3075 * First validate permissions
3078 if (mask & AT_SIZE) {
3080 * XXX - Note, we are not providing any open
3081 * mode flags here (like FNDELAY), so we may
3082 * block if there are locks present... this
3083 * should be addressed in openat().
3085 /* XXX - would it be OK to generate a log record here? */
3086 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
3093 if (mask & (AT_ATIME|AT_MTIME) ||
3094 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
3095 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
3096 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
3097 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
3098 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
3099 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
3100 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
3101 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
3105 if (mask & (AT_UID|AT_GID)) {
3106 int idmask = (mask & (AT_UID|AT_GID));
3111 * NOTE: even if a new mode is being set,
3112 * we may clear S_ISUID/S_ISGID bits.
3115 if (!(mask & AT_MODE))
3116 vap->va_mode = zp->z_mode;
3119 * Take ownership or chgrp to group we are a member of
3122 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
3123 take_group = (mask & AT_GID) &&
3124 zfs_groupmember(zfsvfs, vap->va_gid, cr);
3127 * If both AT_UID and AT_GID are set then take_owner and
3128 * take_group must both be set in order to allow taking
3131 * Otherwise, send the check through secpolicy_vnode_setattr()
3135 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
3136 ((idmask == AT_UID) && take_owner) ||
3137 ((idmask == AT_GID) && take_group)) {
3138 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
3139 skipaclchk, cr) == 0) {
3141 * Remove setuid/setgid for non-privileged users
3143 secpolicy_setid_clear(vap, vp, cr);
3144 trim_mask = (mask & (AT_UID|AT_GID));
3153 mutex_enter(&zp->z_lock);
3154 oldva.va_mode = zp->z_mode;
3155 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
3156 if (mask & AT_XVATTR) {
3158 * Update xvattr mask to include only those attributes
3159 * that are actually changing.
3161 * the bits will be restored prior to actually setting
3162 * the attributes so the caller thinks they were set.
3164 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
3165 if (xoap->xoa_appendonly !=
3166 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
3169 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
3170 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
3174 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
3175 if (xoap->xoa_nounlink !=
3176 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
3179 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
3180 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
3184 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
3185 if (xoap->xoa_immutable !=
3186 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
3189 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
3190 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
3194 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
3195 if (xoap->xoa_nodump !=
3196 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
3199 XVA_CLR_REQ(xvap, XAT_NODUMP);
3200 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
3204 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3205 if (xoap->xoa_av_modified !=
3206 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3209 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3210 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
3214 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3215 if ((vp->v_type != VREG &&
3216 xoap->xoa_av_quarantined) ||
3217 xoap->xoa_av_quarantined !=
3218 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3221 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3222 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
3226 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3227 mutex_exit(&zp->z_lock);
3229 return (SET_ERROR(EPERM));
3232 if (need_policy == FALSE &&
3233 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3234 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3239 mutex_exit(&zp->z_lock);
3241 if (mask & AT_MODE) {
3242 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3243 err = secpolicy_setid_setsticky_clear(vp, vap,
3249 trim_mask |= AT_MODE;
3257 * If trim_mask is set then take ownership
3258 * has been granted or write_acl is present and user
3259 * has the ability to modify mode. In that case remove
3260 * UID|GID and or MODE from mask so that
3261 * secpolicy_vnode_setattr() doesn't revoke it.
3265 saved_mask = vap->va_mask;
3266 vap->va_mask &= ~trim_mask;
3267 if (trim_mask & AT_MODE) {
3269 * Save the mode, as secpolicy_vnode_setattr()
3270 * will overwrite it with ova.va_mode.
3272 saved_mode = vap->va_mode;
3275 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3276 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3283 vap->va_mask |= saved_mask;
3284 if (trim_mask & AT_MODE) {
3286 * Recover the mode after
3287 * secpolicy_vnode_setattr().
3289 vap->va_mode = saved_mode;
3295 * secpolicy_vnode_setattr, or take ownership may have
3298 mask = vap->va_mask;
3300 if ((mask & (AT_UID | AT_GID))) {
3301 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3302 &xattr_obj, sizeof (xattr_obj));
3304 if (err == 0 && xattr_obj) {
3305 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3309 if (mask & AT_UID) {
3310 new_uid = zfs_fuid_create(zfsvfs,
3311 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3312 if (new_uid != zp->z_uid &&
3313 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3315 VN_RELE(ZTOV(attrzp));
3316 err = SET_ERROR(EDQUOT);
3321 if (mask & AT_GID) {
3322 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3323 cr, ZFS_GROUP, &fuidp);
3324 if (new_gid != zp->z_gid &&
3325 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3327 VN_RELE(ZTOV(attrzp));
3328 err = SET_ERROR(EDQUOT);
3333 tx = dmu_tx_create(zfsvfs->z_os);
3335 if (mask & AT_MODE) {
3336 uint64_t pmode = zp->z_mode;
3338 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3340 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3341 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3342 err = SET_ERROR(EPERM);
3346 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3349 mutex_enter(&zp->z_lock);
3350 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3352 * Are we upgrading ACL from old V0 format
3355 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3356 zfs_znode_acl_version(zp) ==
3357 ZFS_ACL_VERSION_INITIAL) {
3358 dmu_tx_hold_free(tx, acl_obj, 0,
3360 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3361 0, aclp->z_acl_bytes);
3363 dmu_tx_hold_write(tx, acl_obj, 0,
3366 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3367 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3368 0, aclp->z_acl_bytes);
3370 mutex_exit(&zp->z_lock);
3371 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3373 if ((mask & AT_XVATTR) &&
3374 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3375 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3377 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3381 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3384 fuid_dirtied = zfsvfs->z_fuid_dirty;
3386 zfs_fuid_txhold(zfsvfs, tx);
3388 zfs_sa_upgrade_txholds(tx, zp);
3390 err = dmu_tx_assign(tx, TXG_WAIT);
3396 * Set each attribute requested.
3397 * We group settings according to the locks they need to acquire.
3399 * Note: you cannot set ctime directly, although it will be
3400 * updated as a side-effect of calling this function.
3404 if (mask & (AT_UID|AT_GID|AT_MODE))
3405 mutex_enter(&zp->z_acl_lock);
3406 mutex_enter(&zp->z_lock);
3408 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3409 &zp->z_pflags, sizeof (zp->z_pflags));
3412 if (mask & (AT_UID|AT_GID|AT_MODE))
3413 mutex_enter(&attrzp->z_acl_lock);
3414 mutex_enter(&attrzp->z_lock);
3415 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3416 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3417 sizeof (attrzp->z_pflags));
3420 if (mask & (AT_UID|AT_GID)) {
3422 if (mask & AT_UID) {
3423 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3424 &new_uid, sizeof (new_uid));
3425 zp->z_uid = new_uid;
3427 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3428 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3430 attrzp->z_uid = new_uid;
3434 if (mask & AT_GID) {
3435 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3436 NULL, &new_gid, sizeof (new_gid));
3437 zp->z_gid = new_gid;
3439 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3440 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3442 attrzp->z_gid = new_gid;
3445 if (!(mask & AT_MODE)) {
3446 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3447 NULL, &new_mode, sizeof (new_mode));
3448 new_mode = zp->z_mode;
3450 err = zfs_acl_chown_setattr(zp);
3453 err = zfs_acl_chown_setattr(attrzp);
3458 if (mask & AT_MODE) {
3459 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3460 &new_mode, sizeof (new_mode));
3461 zp->z_mode = new_mode;
3462 ASSERT3U((uintptr_t)aclp, !=, 0);
3463 err = zfs_aclset_common(zp, aclp, cr, tx);
3465 if (zp->z_acl_cached)
3466 zfs_acl_free(zp->z_acl_cached);
3467 zp->z_acl_cached = aclp;
3472 if (mask & AT_ATIME) {
3473 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3474 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3475 &zp->z_atime, sizeof (zp->z_atime));
3478 if (mask & AT_MTIME) {
3479 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3480 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3481 mtime, sizeof (mtime));
3484 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3485 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3486 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3487 NULL, mtime, sizeof (mtime));
3488 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3489 &ctime, sizeof (ctime));
3490 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3492 } else if (mask != 0) {
3493 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3494 &ctime, sizeof (ctime));
3495 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3498 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3499 SA_ZPL_CTIME(zfsvfs), NULL,
3500 &ctime, sizeof (ctime));
3501 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3502 mtime, ctime, B_TRUE);
3506 * Do this after setting timestamps to prevent timestamp
3507 * update from toggling bit
3510 if (xoap && (mask & AT_XVATTR)) {
3513 * restore trimmed off masks
3514 * so that return masks can be set for caller.
3517 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3518 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3520 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3521 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3523 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3524 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3526 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3527 XVA_SET_REQ(xvap, XAT_NODUMP);
3529 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3530 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3532 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3533 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3536 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3537 ASSERT(vp->v_type == VREG);
3539 zfs_xvattr_set(zp, xvap, tx);
3543 zfs_fuid_sync(zfsvfs, tx);
3546 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3548 mutex_exit(&zp->z_lock);
3549 if (mask & (AT_UID|AT_GID|AT_MODE))
3550 mutex_exit(&zp->z_acl_lock);
3553 if (mask & (AT_UID|AT_GID|AT_MODE))
3554 mutex_exit(&attrzp->z_acl_lock);
3555 mutex_exit(&attrzp->z_lock);
3558 if (err == 0 && attrzp) {
3559 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3565 VN_RELE(ZTOV(attrzp));
3571 zfs_fuid_info_free(fuidp);
3577 if (err == ERESTART)
3580 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3585 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3586 zil_commit(zilog, 0);
3592 typedef struct zfs_zlock {
3593 krwlock_t *zl_rwlock; /* lock we acquired */
3594 znode_t *zl_znode; /* znode we held */
3595 struct zfs_zlock *zl_next; /* next in list */
3599 * Drop locks and release vnodes that were held by zfs_rename_lock().
3602 zfs_rename_unlock(zfs_zlock_t **zlpp)
3606 while ((zl = *zlpp) != NULL) {
3607 if (zl->zl_znode != NULL)
3608 VN_RELE(ZTOV(zl->zl_znode));
3609 rw_exit(zl->zl_rwlock);
3610 *zlpp = zl->zl_next;
3611 kmem_free(zl, sizeof (*zl));
3616 * Search back through the directory tree, using the ".." entries.
3617 * Lock each directory in the chain to prevent concurrent renames.
3618 * Fail any attempt to move a directory into one of its own descendants.
3619 * XXX - z_parent_lock can overlap with map or grow locks
3622 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3626 uint64_t rootid = zp->z_zfsvfs->z_root;
3627 uint64_t oidp = zp->z_id;
3628 krwlock_t *rwlp = &szp->z_parent_lock;
3629 krw_t rw = RW_WRITER;
3632 * First pass write-locks szp and compares to zp->z_id.
3633 * Later passes read-lock zp and compare to zp->z_parent.
3636 if (!rw_tryenter(rwlp, rw)) {
3638 * Another thread is renaming in this path.
3639 * Note that if we are a WRITER, we don't have any
3640 * parent_locks held yet.
3642 if (rw == RW_READER && zp->z_id > szp->z_id) {
3644 * Drop our locks and restart
3646 zfs_rename_unlock(&zl);
3650 rwlp = &szp->z_parent_lock;
3655 * Wait for other thread to drop its locks
3661 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3662 zl->zl_rwlock = rwlp;
3663 zl->zl_znode = NULL;
3664 zl->zl_next = *zlpp;
3667 if (oidp == szp->z_id) /* We're a descendant of szp */
3668 return (SET_ERROR(EINVAL));
3670 if (oidp == rootid) /* We've hit the top */
3673 if (rw == RW_READER) { /* i.e. not the first pass */
3674 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3679 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3680 &oidp, sizeof (oidp));
3681 rwlp = &zp->z_parent_lock;
3684 } while (zp->z_id != sdzp->z_id);
3690 * Move an entry from the provided source directory to the target
3691 * directory. Change the entry name as indicated.
3693 * IN: sdvp - Source directory containing the "old entry".
3694 * snm - Old entry name.
3695 * tdvp - Target directory to contain the "new entry".
3696 * tnm - New entry name.
3697 * cr - credentials of caller.
3698 * ct - caller context
3699 * flags - case flags
3701 * RETURN: 0 on success, error code on failure.
3704 * sdvp,tdvp - ctime|mtime updated
3708 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3709 caller_context_t *ct, int flags)
3711 znode_t *tdzp, *szp, *tzp;
3712 znode_t *sdzp = VTOZ(sdvp);
3713 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs;
3716 zfs_dirlock_t *sdl, *tdl;
3719 int cmp, serr, terr;
3722 boolean_t waited = B_FALSE;
3725 ZFS_VERIFY_ZP(sdzp);
3726 zilog = zfsvfs->z_log;
3729 * Make sure we have the real vp for the target directory.
3731 if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3734 if (tdvp->v_vfsp != sdvp->v_vfsp || zfsctl_is_node(tdvp)) {
3736 return (SET_ERROR(EXDEV));
3740 ZFS_VERIFY_ZP(tdzp);
3741 if (zfsvfs->z_utf8 && u8_validate(tnm,
3742 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3744 return (SET_ERROR(EILSEQ));
3747 if (flags & FIGNORECASE)
3756 * This is to prevent the creation of links into attribute space
3757 * by renaming a linked file into/outof an attribute directory.
3758 * See the comment in zfs_link() for why this is considered bad.
3760 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3762 return (SET_ERROR(EINVAL));
3766 * Lock source and target directory entries. To prevent deadlock,
3767 * a lock ordering must be defined. We lock the directory with
3768 * the smallest object id first, or if it's a tie, the one with
3769 * the lexically first name.
3771 if (sdzp->z_id < tdzp->z_id) {
3773 } else if (sdzp->z_id > tdzp->z_id) {
3777 * First compare the two name arguments without
3778 * considering any case folding.
3780 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3782 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3783 ASSERT(error == 0 || !zfsvfs->z_utf8);
3786 * POSIX: "If the old argument and the new argument
3787 * both refer to links to the same existing file,
3788 * the rename() function shall return successfully
3789 * and perform no other action."
3795 * If the file system is case-folding, then we may
3796 * have some more checking to do. A case-folding file
3797 * system is either supporting mixed case sensitivity
3798 * access or is completely case-insensitive. Note
3799 * that the file system is always case preserving.
3801 * In mixed sensitivity mode case sensitive behavior
3802 * is the default. FIGNORECASE must be used to
3803 * explicitly request case insensitive behavior.
3805 * If the source and target names provided differ only
3806 * by case (e.g., a request to rename 'tim' to 'Tim'),
3807 * we will treat this as a special case in the
3808 * case-insensitive mode: as long as the source name
3809 * is an exact match, we will allow this to proceed as
3810 * a name-change request.
3812 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3813 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3814 flags & FIGNORECASE)) &&
3815 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3818 * case preserving rename request, require exact
3827 * If the source and destination directories are the same, we should
3828 * grab the z_name_lock of that directory only once.
3832 rw_enter(&sdzp->z_name_lock, RW_READER);
3836 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3837 ZEXISTS | zflg, NULL, NULL);
3838 terr = zfs_dirent_lock(&tdl,
3839 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3841 terr = zfs_dirent_lock(&tdl,
3842 tdzp, tnm, &tzp, zflg, NULL, NULL);
3843 serr = zfs_dirent_lock(&sdl,
3844 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3850 * Source entry invalid or not there.
3853 zfs_dirent_unlock(tdl);
3859 rw_exit(&sdzp->z_name_lock);
3862 * FreeBSD: In OpenSolaris they only check if rename source is
3863 * ".." here, because "." is handled in their lookup. This is
3864 * not the case for FreeBSD, so we check for "." explicitly.
3866 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0)
3867 serr = SET_ERROR(EINVAL);
3872 zfs_dirent_unlock(sdl);
3876 rw_exit(&sdzp->z_name_lock);
3878 if (strcmp(tnm, "..") == 0)
3879 terr = SET_ERROR(EINVAL);
3885 * Must have write access at the source to remove the old entry
3886 * and write access at the target to create the new entry.
3887 * Note that if target and source are the same, this can be
3888 * done in a single check.
3891 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3894 if (ZTOV(szp)->v_type == VDIR) {
3896 * Check to make sure rename is valid.
3897 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3899 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3904 * Does target exist?
3908 * Source and target must be the same type.
3910 if (ZTOV(szp)->v_type == VDIR) {
3911 if (ZTOV(tzp)->v_type != VDIR) {
3912 error = SET_ERROR(ENOTDIR);
3916 if (ZTOV(tzp)->v_type == VDIR) {
3917 error = SET_ERROR(EISDIR);
3922 * POSIX dictates that when the source and target
3923 * entries refer to the same file object, rename
3924 * must do nothing and exit without error.
3926 if (szp->z_id == tzp->z_id) {
3932 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3934 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3937 * notify the target directory if it is not the same
3938 * as source directory.
3941 vnevent_rename_dest_dir(tdvp, ct);
3944 tx = dmu_tx_create(zfsvfs->z_os);
3945 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3946 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3947 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3948 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3950 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3951 zfs_sa_upgrade_txholds(tx, tdzp);
3954 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3955 zfs_sa_upgrade_txholds(tx, tzp);
3958 zfs_sa_upgrade_txholds(tx, szp);
3959 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3960 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3963 zfs_rename_unlock(&zl);
3964 zfs_dirent_unlock(sdl);
3965 zfs_dirent_unlock(tdl);
3968 rw_exit(&sdzp->z_name_lock);
3973 if (error == ERESTART) {
3984 if (tzp) /* Attempt to remove the existing target */
3985 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3988 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3990 szp->z_pflags |= ZFS_AV_MODIFIED;
3992 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3993 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3996 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3998 zfs_log_rename(zilog, tx, TX_RENAME |
3999 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
4000 sdl->dl_name, tdzp, tdl->dl_name, szp);
4003 * Update path information for the target vnode
4005 vn_renamepath(tdvp, ZTOV(szp), tnm,
4009 * At this point, we have successfully created
4010 * the target name, but have failed to remove
4011 * the source name. Since the create was done
4012 * with the ZRENAMING flag, there are
4013 * complications; for one, the link count is
4014 * wrong. The easiest way to deal with this
4015 * is to remove the newly created target, and
4016 * return the original error. This must
4017 * succeed; fortunately, it is very unlikely to
4018 * fail, since we just created it.
4020 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
4021 ZRENAMING, NULL), ==, 0);
4024 #ifdef FREEBSD_NAMECACHE
4028 cache_purge(ZTOV(szp));
4030 cache_purge(ZTOV(tzp));
4038 zfs_rename_unlock(&zl);
4040 zfs_dirent_unlock(sdl);
4041 zfs_dirent_unlock(tdl);
4044 rw_exit(&sdzp->z_name_lock);
4051 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4052 zil_commit(zilog, 0);
4060 * Insert the indicated symbolic reference entry into the directory.
4062 * IN: dvp - Directory to contain new symbolic link.
4063 * link - Name for new symlink entry.
4064 * vap - Attributes of new entry.
4065 * cr - credentials of caller.
4066 * ct - caller context
4067 * flags - case flags
4069 * RETURN: 0 on success, error code on failure.
4072 * dvp - ctime|mtime updated
4076 zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link,
4077 cred_t *cr, kthread_t *td)
4079 znode_t *zp, *dzp = VTOZ(dvp);
4082 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4084 uint64_t len = strlen(link);
4087 zfs_acl_ids_t acl_ids;
4088 boolean_t fuid_dirtied;
4089 uint64_t txtype = TX_SYMLINK;
4090 boolean_t waited = B_FALSE;
4093 ASSERT(vap->va_type == VLNK);
4097 zilog = zfsvfs->z_log;
4099 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
4100 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4102 return (SET_ERROR(EILSEQ));
4104 if (flags & FIGNORECASE)
4107 if (len > MAXPATHLEN) {
4109 return (SET_ERROR(ENAMETOOLONG));
4112 if ((error = zfs_acl_ids_create(dzp, 0,
4113 vap, cr, NULL, &acl_ids)) != 0) {
4118 getnewvnode_reserve(1);
4122 * Attempt to lock directory; fail if entry already exists.
4124 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
4126 zfs_acl_ids_free(&acl_ids);
4127 getnewvnode_drop_reserve();
4132 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4133 zfs_acl_ids_free(&acl_ids);
4134 zfs_dirent_unlock(dl);
4135 getnewvnode_drop_reserve();
4140 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
4141 zfs_acl_ids_free(&acl_ids);
4142 zfs_dirent_unlock(dl);
4143 getnewvnode_drop_reserve();
4145 return (SET_ERROR(EDQUOT));
4147 tx = dmu_tx_create(zfsvfs->z_os);
4148 fuid_dirtied = zfsvfs->z_fuid_dirty;
4149 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
4150 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4151 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
4152 ZFS_SA_BASE_ATTR_SIZE + len);
4153 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
4154 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
4155 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
4156 acl_ids.z_aclp->z_acl_bytes);
4159 zfs_fuid_txhold(zfsvfs, tx);
4160 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4162 zfs_dirent_unlock(dl);
4163 if (error == ERESTART) {
4169 zfs_acl_ids_free(&acl_ids);
4171 getnewvnode_drop_reserve();
4177 * Create a new object for the symlink.
4178 * for version 4 ZPL datsets the symlink will be an SA attribute
4180 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
4183 zfs_fuid_sync(zfsvfs, tx);
4185 mutex_enter(&zp->z_lock);
4187 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
4190 zfs_sa_symlink(zp, link, len, tx);
4191 mutex_exit(&zp->z_lock);
4194 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
4195 &zp->z_size, sizeof (zp->z_size), tx);
4197 * Insert the new object into the directory.
4199 (void) zfs_link_create(dl, zp, tx, ZNEW);
4201 if (flags & FIGNORECASE)
4203 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
4206 zfs_acl_ids_free(&acl_ids);
4210 getnewvnode_drop_reserve();
4212 zfs_dirent_unlock(dl);
4214 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4215 zil_commit(zilog, 0);
4222 * Return, in the buffer contained in the provided uio structure,
4223 * the symbolic path referred to by vp.
4225 * IN: vp - vnode of symbolic link.
4226 * uio - structure to contain the link path.
4227 * cr - credentials of caller.
4228 * ct - caller context
4230 * OUT: uio - structure containing the link path.
4232 * RETURN: 0 on success, error code on failure.
4235 * vp - atime updated
4239 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
4241 znode_t *zp = VTOZ(vp);
4242 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4248 mutex_enter(&zp->z_lock);
4250 error = sa_lookup_uio(zp->z_sa_hdl,
4251 SA_ZPL_SYMLINK(zfsvfs), uio);
4253 error = zfs_sa_readlink(zp, uio);
4254 mutex_exit(&zp->z_lock);
4256 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4263 * Insert a new entry into directory tdvp referencing svp.
4265 * IN: tdvp - Directory to contain new entry.
4266 * svp - vnode of new entry.
4267 * name - name of new entry.
4268 * cr - credentials of caller.
4269 * ct - caller context
4271 * RETURN: 0 on success, error code on failure.
4274 * tdvp - ctime|mtime updated
4275 * svp - ctime updated
4279 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
4280 caller_context_t *ct, int flags)
4282 znode_t *dzp = VTOZ(tdvp);
4284 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4293 boolean_t waited = B_FALSE;
4295 ASSERT(tdvp->v_type == VDIR);
4299 zilog = zfsvfs->z_log;
4301 if (VOP_REALVP(svp, &realvp, ct) == 0)
4305 * POSIX dictates that we return EPERM here.
4306 * Better choices include ENOTSUP or EISDIR.
4308 if (svp->v_type == VDIR) {
4310 return (SET_ERROR(EPERM));
4313 if (svp->v_vfsp != tdvp->v_vfsp || zfsctl_is_node(svp)) {
4315 return (SET_ERROR(EXDEV));
4321 /* Prevent links to .zfs/shares files */
4323 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4324 &parent, sizeof (uint64_t))) != 0) {
4328 if (parent == zfsvfs->z_shares_dir) {
4330 return (SET_ERROR(EPERM));
4333 if (zfsvfs->z_utf8 && u8_validate(name,
4334 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4336 return (SET_ERROR(EILSEQ));
4338 if (flags & FIGNORECASE)
4342 * We do not support links between attributes and non-attributes
4343 * because of the potential security risk of creating links
4344 * into "normal" file space in order to circumvent restrictions
4345 * imposed in attribute space.
4347 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4349 return (SET_ERROR(EINVAL));
4353 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4354 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) {
4356 return (SET_ERROR(EPERM));
4359 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4366 * Attempt to lock directory; fail if entry already exists.
4368 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4374 tx = dmu_tx_create(zfsvfs->z_os);
4375 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4376 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4377 zfs_sa_upgrade_txholds(tx, szp);
4378 zfs_sa_upgrade_txholds(tx, dzp);
4379 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4381 zfs_dirent_unlock(dl);
4382 if (error == ERESTART) {
4393 error = zfs_link_create(dl, szp, tx, 0);
4396 uint64_t txtype = TX_LINK;
4397 if (flags & FIGNORECASE)
4399 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4404 zfs_dirent_unlock(dl);
4407 vnevent_link(svp, ct);
4410 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4411 zil_commit(zilog, 0);
4419 * zfs_null_putapage() is used when the file system has been force
4420 * unmounted. It just drops the pages.
4424 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4425 size_t *lenp, int flags, cred_t *cr)
4427 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4432 * Push a page out to disk, klustering if possible.
4434 * IN: vp - file to push page to.
4435 * pp - page to push.
4436 * flags - additional flags.
4437 * cr - credentials of caller.
4439 * OUT: offp - start of range pushed.
4440 * lenp - len of range pushed.
4442 * RETURN: 0 on success, error code on failure.
4444 * NOTE: callers must have locked the page to be pushed. On
4445 * exit, the page (and all other pages in the kluster) must be
4450 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4451 size_t *lenp, int flags, cred_t *cr)
4453 znode_t *zp = VTOZ(vp);
4454 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4456 u_offset_t off, koff;
4463 * If our blocksize is bigger than the page size, try to kluster
4464 * multiple pages so that we write a full block (thus avoiding
4465 * a read-modify-write).
4467 if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4468 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4469 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4470 ASSERT(koff <= zp->z_size);
4471 if (koff + klen > zp->z_size)
4472 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4473 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4475 ASSERT3U(btop(len), ==, btopr(len));
4478 * Can't push pages past end-of-file.
4480 if (off >= zp->z_size) {
4481 /* ignore all pages */
4484 } else if (off + len > zp->z_size) {
4485 int npages = btopr(zp->z_size - off);
4488 page_list_break(&pp, &trunc, npages);
4489 /* ignore pages past end of file */
4491 pvn_write_done(trunc, flags);
4492 len = zp->z_size - off;
4495 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4496 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4497 err = SET_ERROR(EDQUOT);
4500 tx = dmu_tx_create(zfsvfs->z_os);
4501 dmu_tx_hold_write(tx, zp->z_id, off, len);
4503 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4504 zfs_sa_upgrade_txholds(tx, zp);
4505 err = dmu_tx_assign(tx, TXG_WAIT);
4511 if (zp->z_blksz <= PAGESIZE) {
4512 caddr_t va = zfs_map_page(pp, S_READ);
4513 ASSERT3U(len, <=, PAGESIZE);
4514 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4515 zfs_unmap_page(pp, va);
4517 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4521 uint64_t mtime[2], ctime[2];
4522 sa_bulk_attr_t bulk[3];
4525 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4527 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4529 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4531 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4533 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4538 pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4548 * Copy the portion of the file indicated from pages into the file.
4549 * The pages are stored in a page list attached to the files vnode.
4551 * IN: vp - vnode of file to push page data to.
4552 * off - position in file to put data.
4553 * len - amount of data to write.
4554 * flags - flags to control the operation.
4555 * cr - credentials of caller.
4556 * ct - caller context.
4558 * RETURN: 0 on success, error code on failure.
4561 * vp - ctime|mtime updated
4565 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4566 caller_context_t *ct)
4568 znode_t *zp = VTOZ(vp);
4569 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4581 * Align this request to the file block size in case we kluster.
4582 * XXX - this can result in pretty aggresive locking, which can
4583 * impact simultanious read/write access. One option might be
4584 * to break up long requests (len == 0) into block-by-block
4585 * operations to get narrower locking.
4587 blksz = zp->z_blksz;
4589 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4592 if (len > 0 && ISP2(blksz))
4593 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4599 * Search the entire vp list for pages >= io_off.
4601 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4602 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4605 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4607 if (off > zp->z_size) {
4608 /* past end of file */
4609 zfs_range_unlock(rl);
4614 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4616 for (off = io_off; io_off < off + len; io_off += io_len) {
4617 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4618 pp = page_lookup(vp, io_off,
4619 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4621 pp = page_lookup_nowait(vp, io_off,
4622 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4625 if (pp != NULL && pvn_getdirty(pp, flags)) {
4629 * Found a dirty page to push
4631 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4639 zfs_range_unlock(rl);
4640 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4641 zil_commit(zfsvfs->z_log, zp->z_id);
4649 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4651 znode_t *zp = VTOZ(vp);
4652 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4655 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4656 if (zp->z_sa_hdl == NULL) {
4658 * The fs has been unmounted, or we did a
4659 * suspend/resume and this file no longer exists.
4661 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4662 vrecycle(vp, curthread);
4666 mutex_enter(&zp->z_lock);
4667 if (zp->z_unlinked) {
4669 * Fast path to recycle a vnode of a removed file.
4671 mutex_exit(&zp->z_lock);
4672 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4673 vrecycle(vp, curthread);
4676 mutex_exit(&zp->z_lock);
4678 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4679 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4681 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4682 zfs_sa_upgrade_txholds(tx, zp);
4683 error = dmu_tx_assign(tx, TXG_WAIT);
4687 mutex_enter(&zp->z_lock);
4688 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4689 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4690 zp->z_atime_dirty = 0;
4691 mutex_exit(&zp->z_lock);
4695 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4700 * Bounds-check the seek operation.
4702 * IN: vp - vnode seeking within
4703 * ooff - old file offset
4704 * noffp - pointer to new file offset
4705 * ct - caller context
4707 * RETURN: 0 on success, EINVAL if new offset invalid.
4711 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4712 caller_context_t *ct)
4714 if (vp->v_type == VDIR)
4716 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4720 * Pre-filter the generic locking function to trap attempts to place
4721 * a mandatory lock on a memory mapped file.
4724 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4725 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4727 znode_t *zp = VTOZ(vp);
4728 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4734 * We are following the UFS semantics with respect to mapcnt
4735 * here: If we see that the file is mapped already, then we will
4736 * return an error, but we don't worry about races between this
4737 * function and zfs_map().
4739 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4741 return (SET_ERROR(EAGAIN));
4744 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4748 * If we can't find a page in the cache, we will create a new page
4749 * and fill it with file data. For efficiency, we may try to fill
4750 * multiple pages at once (klustering) to fill up the supplied page
4751 * list. Note that the pages to be filled are held with an exclusive
4752 * lock to prevent access by other threads while they are being filled.
4755 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4756 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4758 znode_t *zp = VTOZ(vp);
4759 page_t *pp, *cur_pp;
4760 objset_t *os = zp->z_zfsvfs->z_os;
4761 u_offset_t io_off, total;
4765 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4767 * We only have a single page, don't bother klustering
4771 pp = page_create_va(vp, io_off, io_len,
4772 PG_EXCL | PG_WAIT, seg, addr);
4775 * Try to find enough pages to fill the page list
4777 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4778 &io_len, off, plsz, 0);
4782 * The page already exists, nothing to do here.
4789 * Fill the pages in the kluster.
4792 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4795 ASSERT3U(io_off, ==, cur_pp->p_offset);
4796 va = zfs_map_page(cur_pp, S_WRITE);
4797 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4799 zfs_unmap_page(cur_pp, va);
4801 /* On error, toss the entire kluster */
4802 pvn_read_done(pp, B_ERROR);
4803 /* convert checksum errors into IO errors */
4805 err = SET_ERROR(EIO);
4808 cur_pp = cur_pp->p_next;
4812 * Fill in the page list array from the kluster starting
4813 * from the desired offset `off'.
4814 * NOTE: the page list will always be null terminated.
4816 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4817 ASSERT(pl == NULL || (*pl)->p_offset == off);
4823 * Return pointers to the pages for the file region [off, off + len]
4824 * in the pl array. If plsz is greater than len, this function may
4825 * also return page pointers from after the specified region
4826 * (i.e. the region [off, off + plsz]). These additional pages are
4827 * only returned if they are already in the cache, or were created as
4828 * part of a klustered read.
4830 * IN: vp - vnode of file to get data from.
4831 * off - position in file to get data from.
4832 * len - amount of data to retrieve.
4833 * plsz - length of provided page list.
4834 * seg - segment to obtain pages for.
4835 * addr - virtual address of fault.
4836 * rw - mode of created pages.
4837 * cr - credentials of caller.
4838 * ct - caller context.
4840 * OUT: protp - protection mode of created pages.
4841 * pl - list of pages created.
4843 * RETURN: 0 on success, error code on failure.
4846 * vp - atime updated
4850 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4851 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4852 enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4854 znode_t *zp = VTOZ(vp);
4855 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4859 /* we do our own caching, faultahead is unnecessary */
4862 else if (len > plsz)
4865 len = P2ROUNDUP(len, PAGESIZE);
4866 ASSERT(plsz >= len);
4875 * Loop through the requested range [off, off + len) looking
4876 * for pages. If we don't find a page, we will need to create
4877 * a new page and fill it with data from the file.
4880 if (*pl = page_lookup(vp, off, SE_SHARED))
4882 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4885 ASSERT3U((*pl)->p_offset, ==, off);
4889 ASSERT3U(len, >=, PAGESIZE);
4892 ASSERT3U(plsz, >=, PAGESIZE);
4899 * Fill out the page array with any pages already in the cache.
4902 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4909 * Release any pages we have previously locked.
4914 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4924 * Request a memory map for a section of a file. This code interacts
4925 * with common code and the VM system as follows:
4927 * - common code calls mmap(), which ends up in smmap_common()
4928 * - this calls VOP_MAP(), which takes you into (say) zfs
4929 * - zfs_map() calls as_map(), passing segvn_create() as the callback
4930 * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4931 * - zfs_addmap() updates z_mapcnt
4935 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4936 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4937 caller_context_t *ct)
4939 znode_t *zp = VTOZ(vp);
4940 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4941 segvn_crargs_t vn_a;
4947 if ((prot & PROT_WRITE) && (zp->z_pflags &
4948 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4950 return (SET_ERROR(EPERM));
4953 if ((prot & (PROT_READ | PROT_EXEC)) &&
4954 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4956 return (SET_ERROR(EACCES));
4959 if (vp->v_flag & VNOMAP) {
4961 return (SET_ERROR(ENOSYS));
4964 if (off < 0 || len > MAXOFFSET_T - off) {
4966 return (SET_ERROR(ENXIO));
4969 if (vp->v_type != VREG) {
4971 return (SET_ERROR(ENODEV));
4975 * If file is locked, disallow mapping.
4977 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
4979 return (SET_ERROR(EAGAIN));
4983 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4991 vn_a.offset = (u_offset_t)off;
4992 vn_a.type = flags & MAP_TYPE;
4994 vn_a.maxprot = maxprot;
4997 vn_a.flags = flags & ~MAP_TYPE;
4999 vn_a.lgrp_mem_policy_flags = 0;
5001 error = as_map(as, *addrp, len, segvn_create, &vn_a);
5010 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5011 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
5012 caller_context_t *ct)
5014 uint64_t pages = btopr(len);
5016 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
5021 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
5022 * more accurate mtime for the associated file. Since we don't have a way of
5023 * detecting when the data was actually modified, we have to resort to
5024 * heuristics. If an explicit msync() is done, then we mark the mtime when the
5025 * last page is pushed. The problem occurs when the msync() call is omitted,
5026 * which by far the most common case:
5034 * putpage() via fsflush
5036 * If we wait until fsflush to come along, we can have a modification time that
5037 * is some arbitrary point in the future. In order to prevent this in the
5038 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
5043 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5044 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
5045 caller_context_t *ct)
5047 uint64_t pages = btopr(len);
5049 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
5050 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
5052 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
5053 vn_has_cached_data(vp))
5054 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
5060 * Free or allocate space in a file. Currently, this function only
5061 * supports the `F_FREESP' command. However, this command is somewhat
5062 * misnamed, as its functionality includes the ability to allocate as
5063 * well as free space.
5065 * IN: vp - vnode of file to free data in.
5066 * cmd - action to take (only F_FREESP supported).
5067 * bfp - section of file to free/alloc.
5068 * flag - current file open mode flags.
5069 * offset - current file offset.
5070 * cr - credentials of caller [UNUSED].
5071 * ct - caller context.
5073 * RETURN: 0 on success, error code on failure.
5076 * vp - ctime|mtime updated
5080 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
5081 offset_t offset, cred_t *cr, caller_context_t *ct)
5083 znode_t *zp = VTOZ(vp);
5084 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5091 if (cmd != F_FREESP) {
5093 return (SET_ERROR(EINVAL));
5096 if (error = convoff(vp, bfp, 0, offset)) {
5101 if (bfp->l_len < 0) {
5103 return (SET_ERROR(EINVAL));
5107 len = bfp->l_len; /* 0 means from off to end of file */
5109 error = zfs_freesp(zp, off, len, flag, TRUE);
5116 CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid));
5117 CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid));
5121 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
5123 znode_t *zp = VTOZ(vp);
5124 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5127 uint64_t object = zp->z_id;
5134 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
5135 &gen64, sizeof (uint64_t))) != 0) {
5140 gen = (uint32_t)gen64;
5142 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
5145 if (fidp->fid_len < size) {
5146 fidp->fid_len = size;
5148 return (SET_ERROR(ENOSPC));
5151 fidp->fid_len = size;
5154 zfid = (zfid_short_t *)fidp;
5156 zfid->zf_len = size;
5158 for (i = 0; i < sizeof (zfid->zf_object); i++)
5159 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
5161 /* Must have a non-zero generation number to distinguish from .zfs */
5164 for (i = 0; i < sizeof (zfid->zf_gen); i++)
5165 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
5167 if (size == LONG_FID_LEN) {
5168 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
5171 zlfid = (zfid_long_t *)fidp;
5173 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
5174 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
5176 /* XXX - this should be the generation number for the objset */
5177 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
5178 zlfid->zf_setgen[i] = 0;
5186 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
5187 caller_context_t *ct)
5199 case _PC_FILESIZEBITS:
5203 case _PC_XATTR_EXISTS:
5205 zfsvfs = zp->z_zfsvfs;
5209 error = zfs_dirent_lock(&dl, zp, "", &xzp,
5210 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
5212 zfs_dirent_unlock(dl);
5213 if (!zfs_dirempty(xzp))
5216 } else if (error == ENOENT) {
5218 * If there aren't extended attributes, it's the
5219 * same as having zero of them.
5226 case _PC_SATTR_ENABLED:
5227 case _PC_SATTR_EXISTS:
5228 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
5229 (vp->v_type == VREG || vp->v_type == VDIR);
5232 case _PC_ACCESS_FILTERING:
5233 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
5237 case _PC_ACL_ENABLED:
5238 *valp = _ACL_ACE_ENABLED;
5241 case _PC_MIN_HOLE_SIZE:
5242 *valp = (int)SPA_MINBLOCKSIZE;
5245 case _PC_TIMESTAMP_RESOLUTION:
5246 /* nanosecond timestamp resolution */
5250 case _PC_ACL_EXTENDED:
5258 case _PC_ACL_PATH_MAX:
5259 *valp = ACL_MAX_ENTRIES;
5263 return (EOPNOTSUPP);
5269 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5270 caller_context_t *ct)
5272 znode_t *zp = VTOZ(vp);
5273 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5275 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5279 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5287 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5288 caller_context_t *ct)
5290 znode_t *zp = VTOZ(vp);
5291 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5293 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5294 zilog_t *zilog = zfsvfs->z_log;
5299 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5301 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5302 zil_commit(zilog, 0);
5310 * The smallest read we may consider to loan out an arcbuf.
5311 * This must be a power of 2.
5313 int zcr_blksz_min = (1 << 10); /* 1K */
5315 * If set to less than the file block size, allow loaning out of an
5316 * arcbuf for a partial block read. This must be a power of 2.
5318 int zcr_blksz_max = (1 << 17); /* 128K */
5322 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5323 caller_context_t *ct)
5325 znode_t *zp = VTOZ(vp);
5326 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5327 int max_blksz = zfsvfs->z_max_blksz;
5328 uio_t *uio = &xuio->xu_uio;
5329 ssize_t size = uio->uio_resid;
5330 offset_t offset = uio->uio_loffset;
5335 int preamble, postamble;
5337 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5338 return (SET_ERROR(EINVAL));
5345 * Loan out an arc_buf for write if write size is bigger than
5346 * max_blksz, and the file's block size is also max_blksz.
5349 if (size < blksz || zp->z_blksz != blksz) {
5351 return (SET_ERROR(EINVAL));
5354 * Caller requests buffers for write before knowing where the
5355 * write offset might be (e.g. NFS TCP write).
5360 preamble = P2PHASE(offset, blksz);
5362 preamble = blksz - preamble;
5367 postamble = P2PHASE(size, blksz);
5370 fullblk = size / blksz;
5371 (void) dmu_xuio_init(xuio,
5372 (preamble != 0) + fullblk + (postamble != 0));
5373 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5374 int, postamble, int,
5375 (preamble != 0) + fullblk + (postamble != 0));
5378 * Have to fix iov base/len for partial buffers. They
5379 * currently represent full arc_buf's.
5382 /* data begins in the middle of the arc_buf */
5383 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5386 (void) dmu_xuio_add(xuio, abuf,
5387 blksz - preamble, preamble);
5390 for (i = 0; i < fullblk; i++) {
5391 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5394 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5398 /* data ends in the middle of the arc_buf */
5399 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5402 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5407 * Loan out an arc_buf for read if the read size is larger than
5408 * the current file block size. Block alignment is not
5409 * considered. Partial arc_buf will be loaned out for read.
5411 blksz = zp->z_blksz;
5412 if (blksz < zcr_blksz_min)
5413 blksz = zcr_blksz_min;
5414 if (blksz > zcr_blksz_max)
5415 blksz = zcr_blksz_max;
5416 /* avoid potential complexity of dealing with it */
5417 if (blksz > max_blksz) {
5419 return (SET_ERROR(EINVAL));
5422 maxsize = zp->z_size - uio->uio_loffset;
5426 if (size < blksz || vn_has_cached_data(vp)) {
5428 return (SET_ERROR(EINVAL));
5433 return (SET_ERROR(EINVAL));
5436 uio->uio_extflg = UIO_XUIO;
5437 XUIO_XUZC_RW(xuio) = ioflag;
5444 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5448 int ioflag = XUIO_XUZC_RW(xuio);
5450 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5452 i = dmu_xuio_cnt(xuio);
5454 abuf = dmu_xuio_arcbuf(xuio, i);
5456 * if abuf == NULL, it must be a write buffer
5457 * that has been returned in zfs_write().
5460 dmu_return_arcbuf(abuf);
5461 ASSERT(abuf || ioflag == UIO_WRITE);
5464 dmu_xuio_fini(xuio);
5469 * Predeclare these here so that the compiler assumes that
5470 * this is an "old style" function declaration that does
5471 * not include arguments => we won't get type mismatch errors
5472 * in the initializations that follow.
5474 static int zfs_inval();
5475 static int zfs_isdir();
5480 return (SET_ERROR(EINVAL));
5486 return (SET_ERROR(EISDIR));
5489 * Directory vnode operations template
5491 vnodeops_t *zfs_dvnodeops;
5492 const fs_operation_def_t zfs_dvnodeops_template[] = {
5493 VOPNAME_OPEN, { .vop_open = zfs_open },
5494 VOPNAME_CLOSE, { .vop_close = zfs_close },
5495 VOPNAME_READ, { .error = zfs_isdir },
5496 VOPNAME_WRITE, { .error = zfs_isdir },
5497 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5498 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5499 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5500 VOPNAME_ACCESS, { .vop_access = zfs_access },
5501 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5502 VOPNAME_CREATE, { .vop_create = zfs_create },
5503 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5504 VOPNAME_LINK, { .vop_link = zfs_link },
5505 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5506 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir },
5507 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5508 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5509 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink },
5510 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5511 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5512 VOPNAME_FID, { .vop_fid = zfs_fid },
5513 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5514 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5515 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5516 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5517 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5522 * Regular file vnode operations template
5524 vnodeops_t *zfs_fvnodeops;
5525 const fs_operation_def_t zfs_fvnodeops_template[] = {
5526 VOPNAME_OPEN, { .vop_open = zfs_open },
5527 VOPNAME_CLOSE, { .vop_close = zfs_close },
5528 VOPNAME_READ, { .vop_read = zfs_read },
5529 VOPNAME_WRITE, { .vop_write = zfs_write },
5530 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5531 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5532 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5533 VOPNAME_ACCESS, { .vop_access = zfs_access },
5534 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5535 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5536 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5537 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5538 VOPNAME_FID, { .vop_fid = zfs_fid },
5539 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5540 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock },
5541 VOPNAME_SPACE, { .vop_space = zfs_space },
5542 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage },
5543 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage },
5544 VOPNAME_MAP, { .vop_map = zfs_map },
5545 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap },
5546 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap },
5547 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5548 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5549 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5550 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5551 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf },
5552 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf },
5557 * Symbolic link vnode operations template
5559 vnodeops_t *zfs_symvnodeops;
5560 const fs_operation_def_t zfs_symvnodeops_template[] = {
5561 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5562 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5563 VOPNAME_ACCESS, { .vop_access = zfs_access },
5564 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5565 VOPNAME_READLINK, { .vop_readlink = zfs_readlink },
5566 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5567 VOPNAME_FID, { .vop_fid = zfs_fid },
5568 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5569 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5574 * special share hidden files vnode operations template
5576 vnodeops_t *zfs_sharevnodeops;
5577 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5578 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5579 VOPNAME_ACCESS, { .vop_access = zfs_access },
5580 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5581 VOPNAME_FID, { .vop_fid = zfs_fid },
5582 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5583 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5584 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5585 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5590 * Extended attribute directory vnode operations template
5592 * This template is identical to the directory vnodes
5593 * operation template except for restricted operations:
5597 * Note that there are other restrictions embedded in:
5598 * zfs_create() - restrict type to VREG
5599 * zfs_link() - no links into/out of attribute space
5600 * zfs_rename() - no moves into/out of attribute space
5602 vnodeops_t *zfs_xdvnodeops;
5603 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5604 VOPNAME_OPEN, { .vop_open = zfs_open },
5605 VOPNAME_CLOSE, { .vop_close = zfs_close },
5606 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5607 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5608 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5609 VOPNAME_ACCESS, { .vop_access = zfs_access },
5610 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5611 VOPNAME_CREATE, { .vop_create = zfs_create },
5612 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5613 VOPNAME_LINK, { .vop_link = zfs_link },
5614 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5615 VOPNAME_MKDIR, { .error = zfs_inval },
5616 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5617 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5618 VOPNAME_SYMLINK, { .error = zfs_inval },
5619 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5620 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5621 VOPNAME_FID, { .vop_fid = zfs_fid },
5622 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5623 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5624 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5625 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5626 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5631 * Error vnode operations template
5633 vnodeops_t *zfs_evnodeops;
5634 const fs_operation_def_t zfs_evnodeops_template[] = {
5635 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5636 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5642 ioflags(int ioflags)
5646 if (ioflags & IO_APPEND)
5648 if (ioflags & IO_NDELAY)
5650 if (ioflags & IO_SYNC)
5651 flags |= (FSYNC | FDSYNC | FRSYNC);
5657 zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage)
5659 znode_t *zp = VTOZ(vp);
5660 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5661 objset_t *os = zp->z_zfsvfs->z_os;
5662 vm_page_t mfirst, mlast, mreq;
5666 off_t startoff, endoff;
5668 vm_pindex_t reqstart, reqend;
5669 int pcount, lsize, reqsize, size;
5674 pcount = OFF_TO_IDX(round_page(count));
5676 object = mreq->object;
5679 KASSERT(vp->v_object == object, ("mismatching object"));
5681 if (pcount > 1 && zp->z_blksz > PAGESIZE) {
5682 startoff = rounddown(IDX_TO_OFF(mreq->pindex), zp->z_blksz);
5683 reqstart = OFF_TO_IDX(round_page(startoff));
5684 if (reqstart < m[0]->pindex)
5687 reqstart = reqstart - m[0]->pindex;
5688 endoff = roundup(IDX_TO_OFF(mreq->pindex) + PAGE_SIZE,
5690 reqend = OFF_TO_IDX(trunc_page(endoff)) - 1;
5691 if (reqend > m[pcount - 1]->pindex)
5692 reqend = m[pcount - 1]->pindex;
5693 reqsize = reqend - m[reqstart]->pindex + 1;
5694 KASSERT(reqstart <= reqpage && reqpage < reqstart + reqsize,
5695 ("reqpage beyond [reqstart, reqstart + reqsize[ bounds"));
5700 mfirst = m[reqstart];
5701 mlast = m[reqstart + reqsize - 1];
5703 VM_OBJECT_LOCK(object);
5705 for (i = 0; i < reqstart; i++) {
5708 vm_page_unlock(m[i]);
5710 for (i = reqstart + reqsize; i < pcount; i++) {
5713 vm_page_unlock(m[i]);
5716 if (mreq->valid && reqsize == 1) {
5717 if (mreq->valid != VM_PAGE_BITS_ALL)
5718 vm_page_zero_invalid(mreq, TRUE);
5719 VM_OBJECT_UNLOCK(object);
5721 return (VM_PAGER_OK);
5724 PCPU_INC(cnt.v_vnodein);
5725 PCPU_ADD(cnt.v_vnodepgsin, reqsize);
5727 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) {
5728 for (i = reqstart; i < reqstart + reqsize; i++) {
5732 vm_page_unlock(m[i]);
5735 VM_OBJECT_UNLOCK(object);
5737 return (VM_PAGER_BAD);
5741 if (IDX_TO_OFF(mlast->pindex) + lsize > object->un_pager.vnp.vnp_size)
5742 lsize = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mlast->pindex);
5744 VM_OBJECT_UNLOCK(object);
5746 for (i = reqstart; i < reqstart + reqsize; i++) {
5748 if (i == (reqstart + reqsize - 1))
5750 va = zfs_map_page(m[i], &sf);
5751 error = dmu_read(os, zp->z_id, IDX_TO_OFF(m[i]->pindex),
5752 size, va, DMU_READ_PREFETCH);
5753 if (size != PAGE_SIZE)
5754 bzero(va + size, PAGE_SIZE - size);
5760 VM_OBJECT_LOCK(object);
5762 for (i = reqstart; i < reqstart + reqsize; i++) {
5764 m[i]->valid = VM_PAGE_BITS_ALL;
5765 KASSERT(m[i]->dirty == 0, ("zfs_getpages: page %p is dirty", m[i]));
5767 vm_page_readahead_finish(m[i]);
5770 VM_OBJECT_UNLOCK(object);
5772 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
5774 return (error ? VM_PAGER_ERROR : VM_PAGER_OK);
5778 zfs_freebsd_getpages(ap)
5779 struct vop_getpages_args /* {
5784 vm_ooffset_t a_offset;
5788 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage));
5792 zfs_putpages(struct vnode *vp, vm_page_t *ma, size_t len, int flags,
5795 znode_t *zp = VTOZ(vp);
5796 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5805 vm_ooffset_t lo_off;
5816 object = vp->v_object;
5820 KASSERT(ma[0]->object == object, ("mismatching object"));
5821 KASSERT(len > 0 && (len & PAGE_MASK) == 0, ("unexpected length"));
5823 for (i = 0; i < pcount; i++)
5824 rtvals[i] = VM_PAGER_ERROR;
5826 off = IDX_TO_OFF(ma[0]->pindex);
5827 blksz = zp->z_blksz;
5828 lo_off = rounddown(off, blksz);
5829 lo_len = roundup(len + (off - lo_off), blksz);
5830 rl = zfs_range_lock(zp, lo_off, lo_len, RL_WRITER);
5832 VM_OBJECT_LOCK(object);
5833 if (len + off > object->un_pager.vnp.vnp_size) {
5834 if (object->un_pager.vnp.vnp_size > off) {
5837 len = object->un_pager.vnp.vnp_size - off;
5839 if ((pgoff = (int)len & PAGE_MASK) != 0) {
5841 * If the object is locked and the following
5842 * conditions hold, then the page's dirty
5843 * field cannot be concurrently changed by a
5847 KASSERT(m->busy > 0,
5848 ("zfs_putpages: page %p is not busy", m));
5849 KASSERT(!pmap_page_is_write_mapped(m),
5850 ("zfs_putpages: page %p is not read-only", m));
5851 vm_page_clear_dirty(m, pgoff, PAGE_SIZE -
5858 if (ncount < pcount) {
5859 for (i = ncount; i < pcount; i++) {
5860 rtvals[i] = VM_PAGER_BAD;
5864 VM_OBJECT_UNLOCK(object);
5869 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
5870 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
5875 tx = dmu_tx_create(zfsvfs->z_os);
5876 dmu_tx_hold_write(tx, zp->z_id, off, len);
5878 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
5879 zfs_sa_upgrade_txholds(tx, zp);
5880 err = dmu_tx_assign(tx, TXG_NOWAIT);
5882 if (err == ERESTART) {
5891 if (zp->z_blksz < PAGE_SIZE) {
5893 for (i = 0; len > 0; off += tocopy, len -= tocopy, i++) {
5894 tocopy = len > PAGE_SIZE ? PAGE_SIZE : len;
5895 va = zfs_map_page(ma[i], &sf);
5896 dmu_write(zfsvfs->z_os, zp->z_id, off, tocopy, va, tx);
5900 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, ma, tx);
5904 uint64_t mtime[2], ctime[2];
5905 sa_bulk_attr_t bulk[3];
5908 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
5910 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
5912 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
5914 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
5916 (void)sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
5917 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
5919 VM_OBJECT_LOCK(object);
5920 for (i = 0; i < ncount; i++) {
5921 rtvals[i] = VM_PAGER_OK;
5922 vm_page_undirty(ma[i]);
5924 VM_OBJECT_UNLOCK(object);
5925 PCPU_INC(cnt.v_vnodeout);
5926 PCPU_ADD(cnt.v_vnodepgsout, ncount);
5931 zfs_range_unlock(rl);
5932 if ((flags & (VM_PAGER_PUT_SYNC | VM_PAGER_PUT_INVAL)) != 0 ||
5933 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5934 zil_commit(zfsvfs->z_log, zp->z_id);
5940 zfs_freebsd_putpages(ap)
5941 struct vop_putpages_args /* {
5947 vm_ooffset_t a_offset;
5951 return (zfs_putpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_sync,
5956 zfs_freebsd_bmap(ap)
5957 struct vop_bmap_args /* {
5960 struct bufobj **a_bop;
5967 if (ap->a_bop != NULL)
5968 *ap->a_bop = &ap->a_vp->v_bufobj;
5969 if (ap->a_bnp != NULL)
5970 *ap->a_bnp = ap->a_bn;
5971 if (ap->a_runp != NULL)
5973 if (ap->a_runb != NULL)
5980 zfs_freebsd_open(ap)
5981 struct vop_open_args /* {
5984 struct ucred *a_cred;
5985 struct thread *a_td;
5988 vnode_t *vp = ap->a_vp;
5989 znode_t *zp = VTOZ(vp);
5992 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL);
5994 vnode_create_vobject(vp, zp->z_size, ap->a_td);
5999 zfs_freebsd_close(ap)
6000 struct vop_close_args /* {
6003 struct ucred *a_cred;
6004 struct thread *a_td;
6008 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL));
6012 zfs_freebsd_ioctl(ap)
6013 struct vop_ioctl_args /* {
6023 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data,
6024 ap->a_fflag, ap->a_cred, NULL, NULL));
6028 zfs_freebsd_read(ap)
6029 struct vop_read_args /* {
6033 struct ucred *a_cred;
6037 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
6042 zfs_freebsd_write(ap)
6043 struct vop_write_args /* {
6047 struct ucred *a_cred;
6051 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
6056 zfs_freebsd_access(ap)
6057 struct vop_access_args /* {
6059 accmode_t a_accmode;
6060 struct ucred *a_cred;
6061 struct thread *a_td;
6064 vnode_t *vp = ap->a_vp;
6065 znode_t *zp = VTOZ(vp);
6070 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND,
6072 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND);
6074 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL);
6077 * VADMIN has to be handled by vaccess().
6080 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND);
6082 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
6083 zp->z_gid, accmode, ap->a_cred, NULL);
6088 * For VEXEC, ensure that at least one execute bit is set for
6091 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR &&
6092 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
6100 zfs_freebsd_lookup(ap)
6101 struct vop_lookup_args /* {
6102 struct vnode *a_dvp;
6103 struct vnode **a_vpp;
6104 struct componentname *a_cnp;
6107 struct componentname *cnp = ap->a_cnp;
6108 char nm[NAME_MAX + 1];
6110 ASSERT(cnp->cn_namelen < sizeof(nm));
6111 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm)));
6113 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop,
6114 cnp->cn_cred, cnp->cn_thread, 0));
6118 zfs_freebsd_create(ap)
6119 struct vop_create_args /* {
6120 struct vnode *a_dvp;
6121 struct vnode **a_vpp;
6122 struct componentname *a_cnp;
6123 struct vattr *a_vap;
6126 struct componentname *cnp = ap->a_cnp;
6127 vattr_t *vap = ap->a_vap;
6130 ASSERT(cnp->cn_flags & SAVENAME);
6132 vattr_init_mask(vap);
6133 mode = vap->va_mode & ALLPERMS;
6135 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode,
6136 ap->a_vpp, cnp->cn_cred, cnp->cn_thread));
6140 zfs_freebsd_remove(ap)
6141 struct vop_remove_args /* {
6142 struct vnode *a_dvp;
6144 struct componentname *a_cnp;
6148 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6150 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr,
6151 ap->a_cnp->cn_cred, NULL, 0));
6155 zfs_freebsd_mkdir(ap)
6156 struct vop_mkdir_args /* {
6157 struct vnode *a_dvp;
6158 struct vnode **a_vpp;
6159 struct componentname *a_cnp;
6160 struct vattr *a_vap;
6163 vattr_t *vap = ap->a_vap;
6165 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6167 vattr_init_mask(vap);
6169 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp,
6170 ap->a_cnp->cn_cred, NULL, 0, NULL));
6174 zfs_freebsd_rmdir(ap)
6175 struct vop_rmdir_args /* {
6176 struct vnode *a_dvp;
6178 struct componentname *a_cnp;
6181 struct componentname *cnp = ap->a_cnp;
6183 ASSERT(cnp->cn_flags & SAVENAME);
6185 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0));
6189 zfs_freebsd_readdir(ap)
6190 struct vop_readdir_args /* {
6193 struct ucred *a_cred;
6200 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag,
6201 ap->a_ncookies, ap->a_cookies));
6205 zfs_freebsd_fsync(ap)
6206 struct vop_fsync_args /* {
6209 struct thread *a_td;
6214 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL));
6218 zfs_freebsd_getattr(ap)
6219 struct vop_getattr_args /* {
6221 struct vattr *a_vap;
6222 struct ucred *a_cred;
6225 vattr_t *vap = ap->a_vap;
6231 xvap.xva_vattr = *vap;
6232 xvap.xva_vattr.va_mask |= AT_XVATTR;
6234 /* Convert chflags into ZFS-type flags. */
6235 /* XXX: what about SF_SETTABLE?. */
6236 XVA_SET_REQ(&xvap, XAT_IMMUTABLE);
6237 XVA_SET_REQ(&xvap, XAT_APPENDONLY);
6238 XVA_SET_REQ(&xvap, XAT_NOUNLINK);
6239 XVA_SET_REQ(&xvap, XAT_NODUMP);
6240 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL);
6244 /* Convert ZFS xattr into chflags. */
6245 #define FLAG_CHECK(fflag, xflag, xfield) do { \
6246 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \
6247 fflags |= (fflag); \
6249 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE,
6250 xvap.xva_xoptattrs.xoa_immutable);
6251 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY,
6252 xvap.xva_xoptattrs.xoa_appendonly);
6253 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK,
6254 xvap.xva_xoptattrs.xoa_nounlink);
6255 FLAG_CHECK(UF_NODUMP, XAT_NODUMP,
6256 xvap.xva_xoptattrs.xoa_nodump);
6258 *vap = xvap.xva_vattr;
6259 vap->va_flags = fflags;
6264 zfs_freebsd_setattr(ap)
6265 struct vop_setattr_args /* {
6267 struct vattr *a_vap;
6268 struct ucred *a_cred;
6271 vnode_t *vp = ap->a_vp;
6272 vattr_t *vap = ap->a_vap;
6273 cred_t *cred = ap->a_cred;
6278 vattr_init_mask(vap);
6279 vap->va_mask &= ~AT_NOSET;
6282 xvap.xva_vattr = *vap;
6284 zflags = VTOZ(vp)->z_pflags;
6286 if (vap->va_flags != VNOVAL) {
6287 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs;
6290 if (zfsvfs->z_use_fuids == B_FALSE)
6291 return (EOPNOTSUPP);
6293 fflags = vap->va_flags;
6294 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_NODUMP)) != 0)
6295 return (EOPNOTSUPP);
6297 * Unprivileged processes are not permitted to unset system
6298 * flags, or modify flags if any system flags are set.
6299 * Privileged non-jail processes may not modify system flags
6300 * if securelevel > 0 and any existing system flags are set.
6301 * Privileged jail processes behave like privileged non-jail
6302 * processes if the security.jail.chflags_allowed sysctl is
6303 * is non-zero; otherwise, they behave like unprivileged
6306 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 ||
6307 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) {
6309 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6310 error = securelevel_gt(cred, 0);
6316 * Callers may only modify the file flags on objects they
6317 * have VADMIN rights for.
6319 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0)
6322 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6326 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) {
6331 #define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \
6332 if (((fflags & (fflag)) && !(zflags & (zflag))) || \
6333 ((zflags & (zflag)) && !(fflags & (fflag)))) { \
6334 XVA_SET_REQ(&xvap, (xflag)); \
6335 (xfield) = ((fflags & (fflag)) != 0); \
6338 /* Convert chflags into ZFS-type flags. */
6339 /* XXX: what about SF_SETTABLE?. */
6340 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE,
6341 xvap.xva_xoptattrs.xoa_immutable);
6342 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY,
6343 xvap.xva_xoptattrs.xoa_appendonly);
6344 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK,
6345 xvap.xva_xoptattrs.xoa_nounlink);
6346 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP,
6347 xvap.xva_xoptattrs.xoa_nodump);
6350 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL));
6354 zfs_freebsd_rename(ap)
6355 struct vop_rename_args /* {
6356 struct vnode *a_fdvp;
6357 struct vnode *a_fvp;
6358 struct componentname *a_fcnp;
6359 struct vnode *a_tdvp;
6360 struct vnode *a_tvp;
6361 struct componentname *a_tcnp;
6364 vnode_t *fdvp = ap->a_fdvp;
6365 vnode_t *fvp = ap->a_fvp;
6366 vnode_t *tdvp = ap->a_tdvp;
6367 vnode_t *tvp = ap->a_tvp;
6370 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART));
6371 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART));
6373 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp,
6374 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0);
6389 zfs_freebsd_symlink(ap)
6390 struct vop_symlink_args /* {
6391 struct vnode *a_dvp;
6392 struct vnode **a_vpp;
6393 struct componentname *a_cnp;
6394 struct vattr *a_vap;
6398 struct componentname *cnp = ap->a_cnp;
6399 vattr_t *vap = ap->a_vap;
6401 ASSERT(cnp->cn_flags & SAVENAME);
6403 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */
6404 vattr_init_mask(vap);
6406 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap,
6407 ap->a_target, cnp->cn_cred, cnp->cn_thread));
6411 zfs_freebsd_readlink(ap)
6412 struct vop_readlink_args /* {
6415 struct ucred *a_cred;
6419 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL));
6423 zfs_freebsd_link(ap)
6424 struct vop_link_args /* {
6425 struct vnode *a_tdvp;
6427 struct componentname *a_cnp;
6430 struct componentname *cnp = ap->a_cnp;
6432 ASSERT(cnp->cn_flags & SAVENAME);
6434 return (zfs_link(ap->a_tdvp, ap->a_vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0));
6438 zfs_freebsd_inactive(ap)
6439 struct vop_inactive_args /* {
6441 struct thread *a_td;
6444 vnode_t *vp = ap->a_vp;
6446 zfs_inactive(vp, ap->a_td->td_ucred, NULL);
6451 zfs_freebsd_reclaim(ap)
6452 struct vop_reclaim_args /* {
6454 struct thread *a_td;
6457 vnode_t *vp = ap->a_vp;
6458 znode_t *zp = VTOZ(vp);
6459 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
6463 /* Destroy the vm object and flush associated pages. */
6464 vnode_destroy_vobject(vp);
6467 * z_teardown_inactive_lock protects from a race with
6468 * zfs_znode_dmu_fini in zfsvfs_teardown during
6471 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
6472 if (zp->z_sa_hdl == NULL)
6476 rw_exit(&zfsvfs->z_teardown_inactive_lock);
6484 struct vop_fid_args /* {
6490 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL));
6494 zfs_freebsd_pathconf(ap)
6495 struct vop_pathconf_args /* {
6498 register_t *a_retval;
6504 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL);
6506 *ap->a_retval = val;
6507 else if (error == EOPNOTSUPP)
6508 error = vop_stdpathconf(ap);
6513 zfs_freebsd_fifo_pathconf(ap)
6514 struct vop_pathconf_args /* {
6517 register_t *a_retval;
6521 switch (ap->a_name) {
6522 case _PC_ACL_EXTENDED:
6524 case _PC_ACL_PATH_MAX:
6525 case _PC_MAC_PRESENT:
6526 return (zfs_freebsd_pathconf(ap));
6528 return (fifo_specops.vop_pathconf(ap));
6533 * FreeBSD's extended attributes namespace defines file name prefix for ZFS'
6534 * extended attribute name:
6537 * system freebsd:system:
6538 * user (none, can be used to access ZFS fsattr(5) attributes
6539 * created on Solaris)
6542 zfs_create_attrname(int attrnamespace, const char *name, char *attrname,
6545 const char *namespace, *prefix, *suffix;
6547 /* We don't allow '/' character in attribute name. */
6548 if (strchr(name, '/') != NULL)
6550 /* We don't allow attribute names that start with "freebsd:" string. */
6551 if (strncmp(name, "freebsd:", 8) == 0)
6554 bzero(attrname, size);
6556 switch (attrnamespace) {
6557 case EXTATTR_NAMESPACE_USER:
6559 prefix = "freebsd:";
6560 namespace = EXTATTR_NAMESPACE_USER_STRING;
6564 * This is the default namespace by which we can access all
6565 * attributes created on Solaris.
6567 prefix = namespace = suffix = "";
6570 case EXTATTR_NAMESPACE_SYSTEM:
6571 prefix = "freebsd:";
6572 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING;
6575 case EXTATTR_NAMESPACE_EMPTY:
6579 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix,
6581 return (ENAMETOOLONG);
6587 * Vnode operating to retrieve a named extended attribute.
6590 zfs_getextattr(struct vop_getextattr_args *ap)
6593 IN struct vnode *a_vp;
6594 IN int a_attrnamespace;
6595 IN const char *a_name;
6596 INOUT struct uio *a_uio;
6598 IN struct ucred *a_cred;
6599 IN struct thread *a_td;
6603 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6604 struct thread *td = ap->a_td;
6605 struct nameidata nd;
6608 vnode_t *xvp = NULL, *vp;
6611 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6612 ap->a_cred, ap->a_td, VREAD);
6616 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6623 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6631 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_SYSSPACE, attrname,
6633 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL);
6635 NDFREE(&nd, NDF_ONLY_PNBUF);
6638 if (error == ENOENT)
6643 if (ap->a_size != NULL) {
6644 error = VOP_GETATTR(vp, &va, ap->a_cred);
6646 *ap->a_size = (size_t)va.va_size;
6647 } else if (ap->a_uio != NULL)
6648 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6651 vn_close(vp, flags, ap->a_cred, td);
6658 * Vnode operation to remove a named attribute.
6661 zfs_deleteextattr(struct vop_deleteextattr_args *ap)
6664 IN struct vnode *a_vp;
6665 IN int a_attrnamespace;
6666 IN const char *a_name;
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, VWRITE);
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,
6699 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF | MPSAFE,
6700 UIO_SYSSPACE, attrname, xvp, td);
6705 NDFREE(&nd, NDF_ONLY_PNBUF);
6706 if (error == ENOENT)
6711 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
6712 NDFREE(&nd, NDF_ONLY_PNBUF);
6715 if (vp == nd.ni_dvp)
6725 * Vnode operation to set a named attribute.
6728 zfs_setextattr(struct vop_setextattr_args *ap)
6731 IN struct vnode *a_vp;
6732 IN int a_attrnamespace;
6733 IN const char *a_name;
6734 INOUT struct uio *a_uio;
6735 IN struct ucred *a_cred;
6736 IN struct thread *a_td;
6740 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6741 struct thread *td = ap->a_td;
6742 struct nameidata nd;
6745 vnode_t *xvp = NULL, *vp;
6748 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6749 ap->a_cred, ap->a_td, VWRITE);
6753 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6760 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6761 LOOKUP_XATTR | CREATE_XATTR_DIR);
6767 flags = FFLAGS(O_WRONLY | O_CREAT);
6768 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_SYSSPACE, attrname,
6770 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL);
6772 NDFREE(&nd, NDF_ONLY_PNBUF);
6780 error = VOP_SETATTR(vp, &va, ap->a_cred);
6782 VOP_WRITE(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6785 vn_close(vp, flags, ap->a_cred, td);
6792 * Vnode operation to retrieve extended attributes on a vnode.
6795 zfs_listextattr(struct vop_listextattr_args *ap)
6798 IN struct vnode *a_vp;
6799 IN int a_attrnamespace;
6800 INOUT struct uio *a_uio;
6802 IN struct ucred *a_cred;
6803 IN struct thread *a_td;
6807 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6808 struct thread *td = ap->a_td;
6809 struct nameidata nd;
6810 char attrprefix[16];
6811 u_char dirbuf[sizeof(struct dirent)];
6814 struct uio auio, *uio = ap->a_uio;
6815 size_t *sizep = ap->a_size;
6817 vnode_t *xvp = NULL, *vp;
6818 int done, error, eof, pos;
6820 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6821 ap->a_cred, ap->a_td, VREAD);
6825 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix,
6826 sizeof(attrprefix));
6829 plen = strlen(attrprefix);
6836 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6841 * ENOATTR means that the EA directory does not yet exist,
6842 * i.e. there are no extended attributes there.
6844 if (error == ENOATTR)
6849 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED | MPSAFE,
6850 UIO_SYSSPACE, ".", xvp, td);
6853 NDFREE(&nd, NDF_ONLY_PNBUF);
6859 auio.uio_iov = &aiov;
6860 auio.uio_iovcnt = 1;
6861 auio.uio_segflg = UIO_SYSSPACE;
6863 auio.uio_rw = UIO_READ;
6864 auio.uio_offset = 0;
6869 aiov.iov_base = (void *)dirbuf;
6870 aiov.iov_len = sizeof(dirbuf);
6871 auio.uio_resid = sizeof(dirbuf);
6872 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL);
6873 done = sizeof(dirbuf) - auio.uio_resid;
6876 for (pos = 0; pos < done;) {
6877 dp = (struct dirent *)(dirbuf + pos);
6878 pos += dp->d_reclen;
6880 * XXX: Temporarily we also accept DT_UNKNOWN, as this
6881 * is what we get when attribute was created on Solaris.
6883 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN)
6885 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0)
6887 else if (strncmp(dp->d_name, attrprefix, plen) != 0)
6889 nlen = dp->d_namlen - plen;
6892 else if (uio != NULL) {
6894 * Format of extattr name entry is one byte for
6895 * length and the rest for name.
6897 error = uiomove(&nlen, 1, uio->uio_rw, uio);
6899 error = uiomove(dp->d_name + plen, nlen,
6906 } while (!eof && error == 0);
6915 zfs_freebsd_getacl(ap)
6916 struct vop_getacl_args /* {
6925 vsecattr_t vsecattr;
6927 if (ap->a_type != ACL_TYPE_NFS4)
6930 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
6931 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL))
6934 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt);
6935 if (vsecattr.vsa_aclentp != NULL)
6936 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz);
6942 zfs_freebsd_setacl(ap)
6943 struct vop_setacl_args /* {
6952 vsecattr_t vsecattr;
6953 int aclbsize; /* size of acl list in bytes */
6956 if (ap->a_type != ACL_TYPE_NFS4)
6959 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES)
6963 * With NFSv4 ACLs, chmod(2) may need to add additional entries,
6964 * splitting every entry into two and appending "canonical six"
6965 * entries at the end. Don't allow for setting an ACL that would
6966 * cause chmod(2) to run out of ACL entries.
6968 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES)
6971 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR);
6975 vsecattr.vsa_mask = VSA_ACE;
6976 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t);
6977 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP);
6978 aaclp = vsecattr.vsa_aclentp;
6979 vsecattr.vsa_aclentsz = aclbsize;
6981 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp);
6982 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL);
6983 kmem_free(aaclp, aclbsize);
6989 zfs_freebsd_aclcheck(ap)
6990 struct vop_aclcheck_args /* {
6999 return (EOPNOTSUPP);
7002 struct vop_vector zfs_vnodeops;
7003 struct vop_vector zfs_fifoops;
7004 struct vop_vector zfs_shareops;
7006 struct vop_vector zfs_vnodeops = {
7007 .vop_default = &default_vnodeops,
7008 .vop_inactive = zfs_freebsd_inactive,
7009 .vop_reclaim = zfs_freebsd_reclaim,
7010 .vop_access = zfs_freebsd_access,
7011 #ifdef FREEBSD_NAMECACHE
7012 .vop_lookup = vfs_cache_lookup,
7013 .vop_cachedlookup = zfs_freebsd_lookup,
7015 .vop_lookup = zfs_freebsd_lookup,
7017 .vop_getattr = zfs_freebsd_getattr,
7018 .vop_setattr = zfs_freebsd_setattr,
7019 .vop_create = zfs_freebsd_create,
7020 .vop_mknod = zfs_freebsd_create,
7021 .vop_mkdir = zfs_freebsd_mkdir,
7022 .vop_readdir = zfs_freebsd_readdir,
7023 .vop_fsync = zfs_freebsd_fsync,
7024 .vop_open = zfs_freebsd_open,
7025 .vop_close = zfs_freebsd_close,
7026 .vop_rmdir = zfs_freebsd_rmdir,
7027 .vop_ioctl = zfs_freebsd_ioctl,
7028 .vop_link = zfs_freebsd_link,
7029 .vop_symlink = zfs_freebsd_symlink,
7030 .vop_readlink = zfs_freebsd_readlink,
7031 .vop_read = zfs_freebsd_read,
7032 .vop_write = zfs_freebsd_write,
7033 .vop_remove = zfs_freebsd_remove,
7034 .vop_rename = zfs_freebsd_rename,
7035 .vop_pathconf = zfs_freebsd_pathconf,
7036 .vop_bmap = zfs_freebsd_bmap,
7037 .vop_fid = zfs_freebsd_fid,
7038 .vop_getextattr = zfs_getextattr,
7039 .vop_deleteextattr = zfs_deleteextattr,
7040 .vop_setextattr = zfs_setextattr,
7041 .vop_listextattr = zfs_listextattr,
7042 .vop_getacl = zfs_freebsd_getacl,
7043 .vop_setacl = zfs_freebsd_setacl,
7044 .vop_aclcheck = zfs_freebsd_aclcheck,
7045 .vop_getpages = zfs_freebsd_getpages,
7046 .vop_putpages = zfs_freebsd_putpages,
7049 struct vop_vector zfs_fifoops = {
7050 .vop_default = &fifo_specops,
7051 .vop_fsync = zfs_freebsd_fsync,
7052 .vop_access = zfs_freebsd_access,
7053 .vop_getattr = zfs_freebsd_getattr,
7054 .vop_inactive = zfs_freebsd_inactive,
7055 .vop_read = VOP_PANIC,
7056 .vop_reclaim = zfs_freebsd_reclaim,
7057 .vop_setattr = zfs_freebsd_setattr,
7058 .vop_write = VOP_PANIC,
7059 .vop_pathconf = zfs_freebsd_fifo_pathconf,
7060 .vop_fid = zfs_freebsd_fid,
7061 .vop_getacl = zfs_freebsd_getacl,
7062 .vop_setacl = zfs_freebsd_setacl,
7063 .vop_aclcheck = zfs_freebsd_aclcheck,
7067 * special share hidden files vnode operations template
7069 struct vop_vector zfs_shareops = {
7070 .vop_default = &default_vnodeops,
7071 .vop_access = zfs_freebsd_access,
7072 .vop_inactive = zfs_freebsd_inactive,
7073 .vop_reclaim = zfs_freebsd_reclaim,
7074 .vop_fid = zfs_freebsd_fid,
7075 .vop_pathconf = zfs_freebsd_pathconf,