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);
599 error = uiomove(va + off, bytes, UIO_READ, uio);
601 error = vn_io_fault_uiomove(va + off, bytes, uio);
607 VM_OBJECT_UNLOCK(obj);
608 error = dmu_read_uio(os, zp->z_id, uio, bytes);
616 VM_OBJECT_UNLOCK(obj);
620 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
623 * Read bytes from specified file into supplied buffer.
625 * IN: vp - vnode of file to be read from.
626 * uio - structure supplying read location, range info,
628 * ioflag - SYNC flags; used to provide FRSYNC semantics.
629 * cr - credentials of caller.
630 * ct - caller context
632 * OUT: uio - updated offset and range, buffer filled.
634 * RETURN: 0 on success, error code on failure.
637 * vp - atime updated if byte count > 0
641 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
643 znode_t *zp = VTOZ(vp);
644 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
655 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
657 return (SET_ERROR(EACCES));
661 * Validate file offset
663 if (uio->uio_loffset < (offset_t)0) {
665 return (SET_ERROR(EINVAL));
669 * Fasttrack empty reads
671 if (uio->uio_resid == 0) {
677 * Check for mandatory locks
679 if (MANDMODE(zp->z_mode)) {
680 if (error = chklock(vp, FREAD,
681 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
688 * If we're in FRSYNC mode, sync out this znode before reading it.
691 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
692 zil_commit(zfsvfs->z_log, zp->z_id);
695 * Lock the range against changes.
697 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
700 * If we are reading past end-of-file we can skip
701 * to the end; but we might still need to set atime.
703 if (uio->uio_loffset >= zp->z_size) {
708 ASSERT(uio->uio_loffset < zp->z_size);
709 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
712 if ((uio->uio_extflg == UIO_XUIO) &&
713 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
715 int blksz = zp->z_blksz;
716 uint64_t offset = uio->uio_loffset;
718 xuio = (xuio_t *)uio;
720 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
723 ASSERT(offset + n <= blksz);
726 (void) dmu_xuio_init(xuio, nblk);
728 if (vn_has_cached_data(vp)) {
730 * For simplicity, we always allocate a full buffer
731 * even if we only expect to read a portion of a block.
733 while (--nblk >= 0) {
734 (void) dmu_xuio_add(xuio,
735 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
743 nbytes = MIN(n, zfs_read_chunk_size -
744 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
747 if (uio->uio_segflg == UIO_NOCOPY)
748 error = mappedread_sf(vp, nbytes, uio);
750 #endif /* __FreeBSD__ */
751 if (vn_has_cached_data(vp))
752 error = mappedread(vp, nbytes, uio);
754 error = dmu_read_uio(os, zp->z_id, uio, nbytes);
756 /* convert checksum errors into IO errors */
758 error = SET_ERROR(EIO);
765 zfs_range_unlock(rl);
767 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
773 * Write the bytes to a file.
775 * IN: vp - vnode of file to be written to.
776 * uio - structure supplying write location, range info,
778 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is
779 * set if in append mode.
780 * cr - credentials of caller.
781 * ct - caller context (NFS/CIFS fem monitor only)
783 * OUT: uio - updated offset and range.
785 * RETURN: 0 on success, error code on failure.
788 * vp - ctime|mtime updated if byte count > 0
793 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
795 znode_t *zp = VTOZ(vp);
796 rlim64_t limit = MAXOFFSET_T;
797 ssize_t start_resid = uio->uio_resid;
801 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
806 int max_blksz = zfsvfs->z_max_blksz;
809 iovec_t *aiov = NULL;
812 int iovcnt = uio->uio_iovcnt;
813 iovec_t *iovp = uio->uio_iov;
816 sa_bulk_attr_t bulk[4];
817 uint64_t mtime[2], ctime[2];
820 * Fasttrack empty write
826 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
832 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
833 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
834 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
836 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
840 * If immutable or not appending then return EPERM
842 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
843 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
844 (uio->uio_loffset < zp->z_size))) {
846 return (SET_ERROR(EPERM));
849 zilog = zfsvfs->z_log;
852 * Validate file offset
854 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
857 return (SET_ERROR(EINVAL));
861 * Check for mandatory locks before calling zfs_range_lock()
862 * in order to prevent a deadlock with locks set via fcntl().
864 if (MANDMODE((mode_t)zp->z_mode) &&
865 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
872 * Pre-fault the pages to ensure slow (eg NFS) pages
874 * Skip this if uio contains loaned arc_buf.
876 if ((uio->uio_extflg == UIO_XUIO) &&
877 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
878 xuio = (xuio_t *)uio;
880 uio_prefaultpages(MIN(n, max_blksz), uio);
884 * If in append mode, set the io offset pointer to eof.
886 if (ioflag & FAPPEND) {
888 * Obtain an appending range lock to guarantee file append
889 * semantics. We reset the write offset once we have the lock.
891 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
893 if (rl->r_len == UINT64_MAX) {
895 * We overlocked the file because this write will cause
896 * the file block size to increase.
897 * Note that zp_size cannot change with this lock held.
901 uio->uio_loffset = woff;
904 * Note that if the file block size will change as a result of
905 * this write, then this range lock will lock the entire file
906 * so that we can re-write the block safely.
908 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
911 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) {
912 zfs_range_unlock(rl);
918 zfs_range_unlock(rl);
920 return (SET_ERROR(EFBIG));
923 if ((woff + n) > limit || woff > (limit - n))
926 /* Will this write extend the file length? */
927 write_eof = (woff + n > zp->z_size);
929 end_size = MAX(zp->z_size, woff + n);
932 * Write the file in reasonable size chunks. Each chunk is written
933 * in a separate transaction; this keeps the intent log records small
934 * and allows us to do more fine-grained space accounting.
938 woff = uio->uio_loffset;
939 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
940 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
942 dmu_return_arcbuf(abuf);
943 error = SET_ERROR(EDQUOT);
947 if (xuio && abuf == NULL) {
948 ASSERT(i_iov < iovcnt);
950 abuf = dmu_xuio_arcbuf(xuio, i_iov);
951 dmu_xuio_clear(xuio, i_iov);
952 DTRACE_PROBE3(zfs_cp_write, int, i_iov,
953 iovec_t *, aiov, arc_buf_t *, abuf);
954 ASSERT((aiov->iov_base == abuf->b_data) ||
955 ((char *)aiov->iov_base - (char *)abuf->b_data +
956 aiov->iov_len == arc_buf_size(abuf)));
958 } else if (abuf == NULL && n >= max_blksz &&
959 woff >= zp->z_size &&
960 P2PHASE(woff, max_blksz) == 0 &&
961 zp->z_blksz == max_blksz) {
963 * This write covers a full block. "Borrow" a buffer
964 * from the dmu so that we can fill it before we enter
965 * a transaction. This avoids the possibility of
966 * holding up the transaction if the data copy hangs
967 * up on a pagefault (e.g., from an NFS server mapping).
973 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
975 ASSERT(abuf != NULL);
976 ASSERT(arc_buf_size(abuf) == max_blksz);
978 if (error = uiocopy(abuf->b_data, max_blksz,
979 UIO_WRITE, uio, &cbytes)) {
980 dmu_return_arcbuf(abuf);
983 ASSERT(cbytes == max_blksz);
985 ssize_t resid = uio->uio_resid;
986 error = vn_io_fault_uiomove(abuf->b_data, max_blksz, uio);
988 uio->uio_offset -= resid - uio->uio_resid;
989 uio->uio_resid = resid;
990 dmu_return_arcbuf(abuf);
997 * Start a transaction.
999 tx = dmu_tx_create(zfsvfs->z_os);
1000 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1001 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
1002 zfs_sa_upgrade_txholds(tx, zp);
1003 error = dmu_tx_assign(tx, TXG_WAIT);
1007 dmu_return_arcbuf(abuf);
1012 * If zfs_range_lock() over-locked we grow the blocksize
1013 * and then reduce the lock range. This will only happen
1014 * on the first iteration since zfs_range_reduce() will
1015 * shrink down r_len to the appropriate size.
1017 if (rl->r_len == UINT64_MAX) {
1020 if (zp->z_blksz > max_blksz) {
1021 ASSERT(!ISP2(zp->z_blksz));
1022 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
1024 new_blksz = MIN(end_size, max_blksz);
1026 zfs_grow_blocksize(zp, new_blksz, tx);
1027 zfs_range_reduce(rl, woff, n);
1031 * XXX - should we really limit each write to z_max_blksz?
1032 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
1034 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
1036 if (woff + nbytes > zp->z_size)
1037 vnode_pager_setsize(vp, woff + nbytes);
1040 tx_bytes = uio->uio_resid;
1041 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
1043 tx_bytes -= uio->uio_resid;
1046 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
1048 * If this is not a full block write, but we are
1049 * extending the file past EOF and this data starts
1050 * block-aligned, use assign_arcbuf(). Otherwise,
1051 * write via dmu_write().
1053 if (tx_bytes < max_blksz && (!write_eof ||
1054 aiov->iov_base != abuf->b_data)) {
1056 dmu_write(zfsvfs->z_os, zp->z_id, woff,
1057 aiov->iov_len, aiov->iov_base, tx);
1058 dmu_return_arcbuf(abuf);
1059 xuio_stat_wbuf_copied();
1061 ASSERT(xuio || tx_bytes == max_blksz);
1062 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
1066 ASSERT(tx_bytes <= uio->uio_resid);
1067 uioskip(uio, tx_bytes);
1070 if (tx_bytes && vn_has_cached_data(vp)) {
1071 update_pages(vp, woff, tx_bytes, zfsvfs->z_os,
1072 zp->z_id, uio->uio_segflg, tx);
1076 * If we made no progress, we're done. If we made even
1077 * partial progress, update the znode and ZIL accordingly.
1079 if (tx_bytes == 0) {
1080 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
1081 (void *)&zp->z_size, sizeof (uint64_t), tx);
1088 * Clear Set-UID/Set-GID bits on successful write if not
1089 * privileged and at least one of the excute bits is set.
1091 * It would be nice to to this after all writes have
1092 * been done, but that would still expose the ISUID/ISGID
1093 * to another app after the partial write is committed.
1095 * Note: we don't call zfs_fuid_map_id() here because
1096 * user 0 is not an ephemeral uid.
1098 mutex_enter(&zp->z_acl_lock);
1099 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
1100 (S_IXUSR >> 6))) != 0 &&
1101 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
1102 secpolicy_vnode_setid_retain(vp, cr,
1103 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
1105 zp->z_mode &= ~(S_ISUID | S_ISGID);
1106 newmode = zp->z_mode;
1107 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
1108 (void *)&newmode, sizeof (uint64_t), tx);
1110 mutex_exit(&zp->z_acl_lock);
1112 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
1116 * Update the file size (zp_size) if it has changed;
1117 * account for possible concurrent updates.
1119 while ((end_size = zp->z_size) < uio->uio_loffset) {
1120 (void) atomic_cas_64(&zp->z_size, end_size,
1125 ASSERT(error == 0 || error == EFAULT);
1129 * If we are replaying and eof is non zero then force
1130 * the file size to the specified eof. Note, there's no
1131 * concurrency during replay.
1133 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
1134 zp->z_size = zfsvfs->z_replay_eof;
1137 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1139 (void) sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1141 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
1146 ASSERT(tx_bytes == nbytes);
1151 uio_prefaultpages(MIN(n, max_blksz), uio);
1155 zfs_range_unlock(rl);
1158 * If we're in replay mode, or we made no progress, return error.
1159 * Otherwise, it's at least a partial write, so it's successful.
1161 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1168 * EFAULT means that at least one page of the source buffer was not
1169 * available. VFS will re-try remaining I/O upon this error.
1171 if (error == EFAULT) {
1177 if (ioflag & (FSYNC | FDSYNC) ||
1178 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1179 zil_commit(zilog, zp->z_id);
1186 zfs_get_done(zgd_t *zgd, int error)
1188 znode_t *zp = zgd->zgd_private;
1189 objset_t *os = zp->z_zfsvfs->z_os;
1193 dmu_buf_rele(zgd->zgd_db, zgd);
1195 zfs_range_unlock(zgd->zgd_rl);
1197 vfslocked = VFS_LOCK_GIANT(zp->z_zfsvfs->z_vfs);
1199 * Release the vnode asynchronously as we currently have the
1200 * txg stopped from syncing.
1202 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1204 if (error == 0 && zgd->zgd_bp)
1205 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1207 kmem_free(zgd, sizeof (zgd_t));
1208 VFS_UNLOCK_GIANT(vfslocked);
1212 static int zil_fault_io = 0;
1216 * Get data to generate a TX_WRITE intent log record.
1219 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1221 zfsvfs_t *zfsvfs = arg;
1222 objset_t *os = zfsvfs->z_os;
1224 uint64_t object = lr->lr_foid;
1225 uint64_t offset = lr->lr_offset;
1226 uint64_t size = lr->lr_length;
1227 blkptr_t *bp = &lr->lr_blkptr;
1232 ASSERT(zio != NULL);
1236 * Nothing to do if the file has been removed
1238 if (zfs_zget(zfsvfs, object, &zp) != 0)
1239 return (SET_ERROR(ENOENT));
1240 if (zp->z_unlinked) {
1242 * Release the vnode asynchronously as we currently have the
1243 * txg stopped from syncing.
1245 VN_RELE_ASYNC(ZTOV(zp),
1246 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1247 return (SET_ERROR(ENOENT));
1250 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1251 zgd->zgd_zilog = zfsvfs->z_log;
1252 zgd->zgd_private = zp;
1255 * Write records come in two flavors: immediate and indirect.
1256 * For small writes it's cheaper to store the data with the
1257 * log record (immediate); for large writes it's cheaper to
1258 * sync the data and get a pointer to it (indirect) so that
1259 * we don't have to write the data twice.
1261 if (buf != NULL) { /* immediate write */
1262 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1263 /* test for truncation needs to be done while range locked */
1264 if (offset >= zp->z_size) {
1265 error = SET_ERROR(ENOENT);
1267 error = dmu_read(os, object, offset, size, buf,
1268 DMU_READ_NO_PREFETCH);
1270 ASSERT(error == 0 || error == ENOENT);
1271 } else { /* indirect write */
1273 * Have to lock the whole block to ensure when it's
1274 * written out and it's checksum is being calculated
1275 * that no one can change the data. We need to re-check
1276 * blocksize after we get the lock in case it's changed!
1281 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1283 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1285 if (zp->z_blksz == size)
1288 zfs_range_unlock(zgd->zgd_rl);
1290 /* test for truncation needs to be done while range locked */
1291 if (lr->lr_offset >= zp->z_size)
1292 error = SET_ERROR(ENOENT);
1295 error = SET_ERROR(EIO);
1300 error = dmu_buf_hold(os, object, offset, zgd, &db,
1301 DMU_READ_NO_PREFETCH);
1304 blkptr_t *obp = dmu_buf_get_blkptr(db);
1306 ASSERT(BP_IS_HOLE(bp));
1313 ASSERT(db->db_offset == offset);
1314 ASSERT(db->db_size == size);
1316 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1318 ASSERT(error || lr->lr_length <= zp->z_blksz);
1321 * On success, we need to wait for the write I/O
1322 * initiated by dmu_sync() to complete before we can
1323 * release this dbuf. We will finish everything up
1324 * in the zfs_get_done() callback.
1329 if (error == EALREADY) {
1330 lr->lr_common.lrc_txtype = TX_WRITE2;
1336 zfs_get_done(zgd, error);
1343 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1344 caller_context_t *ct)
1346 znode_t *zp = VTOZ(vp);
1347 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1353 if (flag & V_ACE_MASK)
1354 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1356 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1363 * If vnode is for a device return a specfs vnode instead.
1366 specvp_check(vnode_t **vpp, cred_t *cr)
1370 if (IS_DEVVP(*vpp)) {
1373 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1376 error = SET_ERROR(ENOSYS);
1384 * Lookup an entry in a directory, or an extended attribute directory.
1385 * If it exists, return a held vnode reference for it.
1387 * IN: dvp - vnode of directory to search.
1388 * nm - name of entry to lookup.
1389 * pnp - full pathname to lookup [UNUSED].
1390 * flags - LOOKUP_XATTR set if looking for an attribute.
1391 * rdir - root directory vnode [UNUSED].
1392 * cr - credentials of caller.
1393 * ct - caller context
1394 * direntflags - directory lookup flags
1395 * realpnp - returned pathname.
1397 * OUT: vpp - vnode of located entry, NULL if not found.
1399 * RETURN: 0 on success, error code on failure.
1406 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp,
1407 int nameiop, cred_t *cr, kthread_t *td, int flags)
1409 znode_t *zdp = VTOZ(dvp);
1410 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1412 int *direntflags = NULL;
1413 void *realpnp = NULL;
1416 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1418 if (dvp->v_type != VDIR) {
1419 return (SET_ERROR(ENOTDIR));
1420 } else if (zdp->z_sa_hdl == NULL) {
1421 return (SET_ERROR(EIO));
1424 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1425 error = zfs_fastaccesschk_execute(zdp, cr);
1433 vnode_t *tvp = dnlc_lookup(dvp, nm);
1436 error = zfs_fastaccesschk_execute(zdp, cr);
1441 if (tvp == DNLC_NO_VNODE) {
1443 return (SET_ERROR(ENOENT));
1446 return (specvp_check(vpp, cr));
1452 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1459 if (flags & LOOKUP_XATTR) {
1462 * If the xattr property is off, refuse the lookup request.
1464 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1466 return (SET_ERROR(EINVAL));
1471 * We don't allow recursive attributes..
1472 * Maybe someday we will.
1474 if (zdp->z_pflags & ZFS_XATTR) {
1476 return (SET_ERROR(EINVAL));
1479 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1485 * Do we have permission to get into attribute directory?
1488 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1498 if (dvp->v_type != VDIR) {
1500 return (SET_ERROR(ENOTDIR));
1504 * Check accessibility of directory.
1507 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1512 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1513 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1515 return (SET_ERROR(EILSEQ));
1518 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1520 error = specvp_check(vpp, cr);
1522 /* Translate errors and add SAVENAME when needed. */
1523 if (cnp->cn_flags & ISLASTCN) {
1527 if (error == ENOENT) {
1528 error = EJUSTRETURN;
1529 cnp->cn_flags |= SAVENAME;
1535 cnp->cn_flags |= SAVENAME;
1539 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) {
1542 if (cnp->cn_flags & ISDOTDOT) {
1543 ltype = VOP_ISLOCKED(dvp);
1547 error = vn_lock(*vpp, cnp->cn_lkflags);
1548 if (cnp->cn_flags & ISDOTDOT)
1549 vn_lock(dvp, ltype | LK_RETRY);
1559 #ifdef FREEBSD_NAMECACHE
1561 * Insert name into cache (as non-existent) if appropriate.
1563 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
1564 cache_enter(dvp, *vpp, cnp);
1566 * Insert name into cache if appropriate.
1568 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) {
1569 if (!(cnp->cn_flags & ISLASTCN) ||
1570 (nameiop != DELETE && nameiop != RENAME)) {
1571 cache_enter(dvp, *vpp, cnp);
1580 * Attempt to create a new entry in a directory. If the entry
1581 * already exists, truncate the file if permissible, else return
1582 * an error. Return the vp of the created or trunc'd file.
1584 * IN: dvp - vnode of directory to put new file entry in.
1585 * name - name of new file entry.
1586 * vap - attributes of new file.
1587 * excl - flag indicating exclusive or non-exclusive mode.
1588 * mode - mode to open file with.
1589 * cr - credentials of caller.
1590 * flag - large file flag [UNUSED].
1591 * ct - caller context
1592 * vsecp - ACL to be set
1594 * OUT: vpp - vnode of created or trunc'd entry.
1596 * RETURN: 0 on success, error code on failure.
1599 * dvp - ctime|mtime updated if new entry created
1600 * vp - ctime|mtime always, atime if new
1605 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode,
1606 vnode_t **vpp, cred_t *cr, kthread_t *td)
1608 znode_t *zp, *dzp = VTOZ(dvp);
1609 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1617 gid_t gid = crgetgid(cr);
1618 zfs_acl_ids_t acl_ids;
1619 boolean_t fuid_dirtied;
1620 boolean_t have_acl = B_FALSE;
1621 boolean_t waited = B_FALSE;
1626 * If we have an ephemeral id, ACL, or XVATTR then
1627 * make sure file system is at proper version
1630 ksid = crgetsid(cr, KSID_OWNER);
1632 uid = ksid_getid(ksid);
1636 if (zfsvfs->z_use_fuids == B_FALSE &&
1637 (vsecp || (vap->va_mask & AT_XVATTR) ||
1638 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1639 return (SET_ERROR(EINVAL));
1644 zilog = zfsvfs->z_log;
1646 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1647 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1649 return (SET_ERROR(EILSEQ));
1652 if (vap->va_mask & AT_XVATTR) {
1653 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
1654 crgetuid(cr), cr, vap->va_type)) != 0) {
1660 getnewvnode_reserve(1);
1665 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr))
1666 vap->va_mode &= ~S_ISVTX;
1668 if (*name == '\0') {
1670 * Null component name refers to the directory itself.
1677 /* possible VN_HOLD(zp) */
1680 if (flag & FIGNORECASE)
1683 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1687 zfs_acl_ids_free(&acl_ids);
1688 if (strcmp(name, "..") == 0)
1689 error = SET_ERROR(EISDIR);
1690 getnewvnode_drop_reserve();
1700 * Create a new file object and update the directory
1703 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1705 zfs_acl_ids_free(&acl_ids);
1710 * We only support the creation of regular files in
1711 * extended attribute directories.
1714 if ((dzp->z_pflags & ZFS_XATTR) &&
1715 (vap->va_type != VREG)) {
1717 zfs_acl_ids_free(&acl_ids);
1718 error = SET_ERROR(EINVAL);
1722 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1723 cr, vsecp, &acl_ids)) != 0)
1727 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1728 zfs_acl_ids_free(&acl_ids);
1729 error = SET_ERROR(EDQUOT);
1733 tx = dmu_tx_create(os);
1735 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1736 ZFS_SA_BASE_ATTR_SIZE);
1738 fuid_dirtied = zfsvfs->z_fuid_dirty;
1740 zfs_fuid_txhold(zfsvfs, tx);
1741 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1742 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1743 if (!zfsvfs->z_use_sa &&
1744 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1745 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1746 0, acl_ids.z_aclp->z_acl_bytes);
1748 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1750 zfs_dirent_unlock(dl);
1751 if (error == ERESTART) {
1757 zfs_acl_ids_free(&acl_ids);
1759 getnewvnode_drop_reserve();
1763 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1766 zfs_fuid_sync(zfsvfs, tx);
1768 (void) zfs_link_create(dl, zp, tx, ZNEW);
1769 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1770 if (flag & FIGNORECASE)
1772 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1773 vsecp, acl_ids.z_fuidp, vap);
1774 zfs_acl_ids_free(&acl_ids);
1777 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1780 zfs_acl_ids_free(&acl_ids);
1784 * A directory entry already exists for this name.
1787 * Can't truncate an existing file if in exclusive mode.
1790 error = SET_ERROR(EEXIST);
1794 * Can't open a directory for writing.
1796 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1797 error = SET_ERROR(EISDIR);
1801 * Verify requested access to file.
1803 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1807 mutex_enter(&dzp->z_lock);
1809 mutex_exit(&dzp->z_lock);
1812 * Truncate regular files if requested.
1814 if ((ZTOV(zp)->v_type == VREG) &&
1815 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1816 /* we can't hold any locks when calling zfs_freesp() */
1817 zfs_dirent_unlock(dl);
1819 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1821 vnevent_create(ZTOV(zp), ct);
1826 getnewvnode_drop_reserve();
1828 zfs_dirent_unlock(dl);
1835 error = specvp_check(vpp, cr);
1838 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1839 zil_commit(zilog, 0);
1846 * Remove an entry from a directory.
1848 * IN: dvp - vnode of directory to remove entry from.
1849 * name - name of entry to remove.
1850 * cr - credentials of caller.
1851 * ct - caller context
1852 * flags - case flags
1854 * RETURN: 0 on success, error code on failure.
1858 * vp - ctime (if nlink > 0)
1861 uint64_t null_xattr = 0;
1865 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1868 znode_t *zp, *dzp = VTOZ(dvp);
1871 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1873 uint64_t acl_obj, xattr_obj;
1874 uint64_t xattr_obj_unlinked = 0;
1878 boolean_t may_delete_now, delete_now = FALSE;
1879 boolean_t unlinked, toobig = FALSE;
1881 pathname_t *realnmp = NULL;
1885 boolean_t waited = B_FALSE;
1889 zilog = zfsvfs->z_log;
1891 if (flags & FIGNORECASE) {
1901 * Attempt to lock directory; fail if entry doesn't exist.
1903 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1913 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1918 * Need to use rmdir for removing directories.
1920 if (vp->v_type == VDIR) {
1921 error = SET_ERROR(EPERM);
1925 vnevent_remove(vp, dvp, name, ct);
1928 dnlc_remove(dvp, realnmp->pn_buf);
1930 dnlc_remove(dvp, name);
1933 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1937 * We may delete the znode now, or we may put it in the unlinked set;
1938 * it depends on whether we're the last link, and on whether there are
1939 * other holds on the vnode. So we dmu_tx_hold() the right things to
1940 * allow for either case.
1943 tx = dmu_tx_create(zfsvfs->z_os);
1944 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1945 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1946 zfs_sa_upgrade_txholds(tx, zp);
1947 zfs_sa_upgrade_txholds(tx, dzp);
1948 if (may_delete_now) {
1950 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1951 /* if the file is too big, only hold_free a token amount */
1952 dmu_tx_hold_free(tx, zp->z_id, 0,
1953 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1956 /* are there any extended attributes? */
1957 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1958 &xattr_obj, sizeof (xattr_obj));
1959 if (error == 0 && xattr_obj) {
1960 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1962 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1963 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1966 mutex_enter(&zp->z_lock);
1967 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1968 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1969 mutex_exit(&zp->z_lock);
1971 /* charge as an update -- would be nice not to charge at all */
1972 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1974 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1976 zfs_dirent_unlock(dl);
1980 if (error == ERESTART) {
1994 * Remove the directory entry.
1996 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
2006 * Hold z_lock so that we can make sure that the ACL obj
2007 * hasn't changed. Could have been deleted due to
2010 mutex_enter(&zp->z_lock);
2012 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
2013 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
2014 delete_now = may_delete_now && !toobig &&
2015 vp->v_count == 1 && !vn_has_cached_data(vp) &&
2016 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
2023 panic("zfs_remove: delete_now branch taken");
2025 if (xattr_obj_unlinked) {
2026 ASSERT3U(xzp->z_links, ==, 2);
2027 mutex_enter(&xzp->z_lock);
2028 xzp->z_unlinked = 1;
2030 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
2031 &xzp->z_links, sizeof (xzp->z_links), tx);
2032 ASSERT3U(error, ==, 0);
2033 mutex_exit(&xzp->z_lock);
2034 zfs_unlinked_add(xzp, tx);
2037 error = sa_remove(zp->z_sa_hdl,
2038 SA_ZPL_XATTR(zfsvfs), tx);
2040 error = sa_update(zp->z_sa_hdl,
2041 SA_ZPL_XATTR(zfsvfs), &null_xattr,
2042 sizeof (uint64_t), tx);
2047 ASSERT0(vp->v_count);
2049 mutex_exit(&zp->z_lock);
2050 zfs_znode_delete(zp, tx);
2051 } else if (unlinked) {
2052 mutex_exit(&zp->z_lock);
2053 zfs_unlinked_add(zp, tx);
2055 vp->v_vflag |= VV_NOSYNC;
2060 if (flags & FIGNORECASE)
2062 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
2069 zfs_dirent_unlock(dl);
2076 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2077 zil_commit(zilog, 0);
2084 * Create a new directory and insert it into dvp using the name
2085 * provided. Return a pointer to the inserted directory.
2087 * IN: dvp - vnode of directory to add subdir to.
2088 * dirname - name of new directory.
2089 * vap - attributes of new directory.
2090 * cr - credentials of caller.
2091 * ct - caller context
2092 * flags - case flags
2093 * vsecp - ACL to be set
2095 * OUT: vpp - vnode of created directory.
2097 * RETURN: 0 on success, error code on failure.
2100 * dvp - ctime|mtime updated
2101 * vp - ctime|mtime|atime updated
2105 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
2106 caller_context_t *ct, int flags, vsecattr_t *vsecp)
2108 znode_t *zp, *dzp = VTOZ(dvp);
2109 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2118 gid_t gid = crgetgid(cr);
2119 zfs_acl_ids_t acl_ids;
2120 boolean_t fuid_dirtied;
2121 boolean_t waited = B_FALSE;
2123 ASSERT(vap->va_type == VDIR);
2126 * If we have an ephemeral id, ACL, or XVATTR then
2127 * make sure file system is at proper version
2130 ksid = crgetsid(cr, KSID_OWNER);
2132 uid = ksid_getid(ksid);
2135 if (zfsvfs->z_use_fuids == B_FALSE &&
2136 (vsecp || (vap->va_mask & AT_XVATTR) ||
2137 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
2138 return (SET_ERROR(EINVAL));
2142 zilog = zfsvfs->z_log;
2144 if (dzp->z_pflags & ZFS_XATTR) {
2146 return (SET_ERROR(EINVAL));
2149 if (zfsvfs->z_utf8 && u8_validate(dirname,
2150 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2152 return (SET_ERROR(EILSEQ));
2154 if (flags & FIGNORECASE)
2157 if (vap->va_mask & AT_XVATTR) {
2158 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
2159 crgetuid(cr), cr, vap->va_type)) != 0) {
2165 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
2166 vsecp, &acl_ids)) != 0) {
2171 getnewvnode_reserve(1);
2174 * First make sure the new directory doesn't exist.
2176 * Existence is checked first to make sure we don't return
2177 * EACCES instead of EEXIST which can cause some applications
2183 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
2185 zfs_acl_ids_free(&acl_ids);
2186 getnewvnode_drop_reserve();
2191 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
2192 zfs_acl_ids_free(&acl_ids);
2193 zfs_dirent_unlock(dl);
2194 getnewvnode_drop_reserve();
2199 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
2200 zfs_acl_ids_free(&acl_ids);
2201 zfs_dirent_unlock(dl);
2202 getnewvnode_drop_reserve();
2204 return (SET_ERROR(EDQUOT));
2208 * Add a new entry to the directory.
2210 tx = dmu_tx_create(zfsvfs->z_os);
2211 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2212 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2213 fuid_dirtied = zfsvfs->z_fuid_dirty;
2215 zfs_fuid_txhold(zfsvfs, tx);
2216 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2217 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2218 acl_ids.z_aclp->z_acl_bytes);
2221 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2222 ZFS_SA_BASE_ATTR_SIZE);
2224 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2226 zfs_dirent_unlock(dl);
2227 if (error == ERESTART) {
2233 zfs_acl_ids_free(&acl_ids);
2235 getnewvnode_drop_reserve();
2243 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2246 zfs_fuid_sync(zfsvfs, tx);
2249 * Now put new name in parent dir.
2251 (void) zfs_link_create(dl, zp, tx, ZNEW);
2255 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2256 if (flags & FIGNORECASE)
2258 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2259 acl_ids.z_fuidp, vap);
2261 zfs_acl_ids_free(&acl_ids);
2265 getnewvnode_drop_reserve();
2267 zfs_dirent_unlock(dl);
2269 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2270 zil_commit(zilog, 0);
2277 * Remove a directory subdir entry. If the current working
2278 * directory is the same as the subdir to be removed, the
2281 * IN: dvp - vnode of directory to remove from.
2282 * name - name of directory to be removed.
2283 * cwd - vnode of current working directory.
2284 * cr - credentials of caller.
2285 * ct - caller context
2286 * flags - case flags
2288 * RETURN: 0 on success, error code on failure.
2291 * dvp - ctime|mtime updated
2295 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2296 caller_context_t *ct, int flags)
2298 znode_t *dzp = VTOZ(dvp);
2301 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2307 boolean_t waited = B_FALSE;
2311 zilog = zfsvfs->z_log;
2313 if (flags & FIGNORECASE)
2319 * Attempt to lock directory; fail if entry doesn't exist.
2321 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2329 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2333 if (vp->v_type != VDIR) {
2334 error = SET_ERROR(ENOTDIR);
2339 error = SET_ERROR(EINVAL);
2343 vnevent_rmdir(vp, dvp, name, ct);
2346 * Grab a lock on the directory to make sure that noone is
2347 * trying to add (or lookup) entries while we are removing it.
2349 rw_enter(&zp->z_name_lock, RW_WRITER);
2352 * Grab a lock on the parent pointer to make sure we play well
2353 * with the treewalk and directory rename code.
2355 rw_enter(&zp->z_parent_lock, RW_WRITER);
2357 tx = dmu_tx_create(zfsvfs->z_os);
2358 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2359 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2360 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2361 zfs_sa_upgrade_txholds(tx, zp);
2362 zfs_sa_upgrade_txholds(tx, dzp);
2363 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2365 rw_exit(&zp->z_parent_lock);
2366 rw_exit(&zp->z_name_lock);
2367 zfs_dirent_unlock(dl);
2369 if (error == ERESTART) {
2380 #ifdef FREEBSD_NAMECACHE
2384 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2387 uint64_t txtype = TX_RMDIR;
2388 if (flags & FIGNORECASE)
2390 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2395 rw_exit(&zp->z_parent_lock);
2396 rw_exit(&zp->z_name_lock);
2397 #ifdef FREEBSD_NAMECACHE
2401 zfs_dirent_unlock(dl);
2405 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2406 zil_commit(zilog, 0);
2413 * Read as many directory entries as will fit into the provided
2414 * buffer from the given directory cursor position (specified in
2415 * the uio structure).
2417 * IN: vp - vnode of directory to read.
2418 * uio - structure supplying read location, range info,
2419 * and return buffer.
2420 * cr - credentials of caller.
2421 * ct - caller context
2422 * flags - case flags
2424 * OUT: uio - updated offset and range, buffer filled.
2425 * eofp - set to true if end-of-file detected.
2427 * RETURN: 0 on success, error code on failure.
2430 * vp - atime updated
2432 * Note that the low 4 bits of the cookie returned by zap is always zero.
2433 * This allows us to use the low range for "special" directory entries:
2434 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2435 * we use the offset 2 for the '.zfs' directory.
2439 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies)
2441 znode_t *zp = VTOZ(vp);
2445 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2450 zap_attribute_t zap;
2451 uint_t bytes_wanted;
2452 uint64_t offset; /* must be unsigned; checks for < 1 */
2458 boolean_t check_sysattrs;
2461 u_long *cooks = NULL;
2467 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2468 &parent, sizeof (parent))) != 0) {
2474 * If we are not given an eof variable,
2481 * Check for valid iov_len.
2483 if (uio->uio_iov->iov_len <= 0) {
2485 return (SET_ERROR(EINVAL));
2489 * Quit if directory has been removed (posix)
2491 if ((*eofp = zp->z_unlinked) != 0) {
2498 offset = uio->uio_loffset;
2499 prefetch = zp->z_zn_prefetch;
2502 * Initialize the iterator cursor.
2506 * Start iteration from the beginning of the directory.
2508 zap_cursor_init(&zc, os, zp->z_id);
2511 * The offset is a serialized cursor.
2513 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2517 * Get space to change directory entries into fs independent format.
2519 iovp = uio->uio_iov;
2520 bytes_wanted = iovp->iov_len;
2521 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2522 bufsize = bytes_wanted;
2523 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2524 odp = (struct dirent64 *)outbuf;
2526 bufsize = bytes_wanted;
2528 odp = (struct dirent64 *)iovp->iov_base;
2530 eodp = (struct edirent *)odp;
2532 if (ncookies != NULL) {
2534 * Minimum entry size is dirent size and 1 byte for a file name.
2536 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1);
2537 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK);
2542 * If this VFS supports the system attribute view interface; and
2543 * we're looking at an extended attribute directory; and we care
2544 * about normalization conflicts on this vfs; then we must check
2545 * for normalization conflicts with the sysattr name space.
2548 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2549 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2550 (flags & V_RDDIR_ENTFLAGS);
2556 * Transform to file-system independent format
2559 while (outcount < bytes_wanted) {
2562 off64_t *next = NULL;
2565 * Special case `.', `..', and `.zfs'.
2568 (void) strcpy(zap.za_name, ".");
2569 zap.za_normalization_conflict = 0;
2572 } else if (offset == 1) {
2573 (void) strcpy(zap.za_name, "..");
2574 zap.za_normalization_conflict = 0;
2577 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2578 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2579 zap.za_normalization_conflict = 0;
2580 objnum = ZFSCTL_INO_ROOT;
2586 if (error = zap_cursor_retrieve(&zc, &zap)) {
2587 if ((*eofp = (error == ENOENT)) != 0)
2593 if (zap.za_integer_length != 8 ||
2594 zap.za_num_integers != 1) {
2595 cmn_err(CE_WARN, "zap_readdir: bad directory "
2596 "entry, obj = %lld, offset = %lld\n",
2597 (u_longlong_t)zp->z_id,
2598 (u_longlong_t)offset);
2599 error = SET_ERROR(ENXIO);
2603 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2605 * MacOS X can extract the object type here such as:
2606 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2608 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2610 if (check_sysattrs && !zap.za_normalization_conflict) {
2612 zap.za_normalization_conflict =
2613 xattr_sysattr_casechk(zap.za_name);
2615 panic("%s:%u: TODO", __func__, __LINE__);
2620 if (flags & V_RDDIR_ACCFILTER) {
2622 * If we have no access at all, don't include
2623 * this entry in the returned information
2626 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2628 if (!zfs_has_access(ezp, cr)) {
2635 if (flags & V_RDDIR_ENTFLAGS)
2636 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2638 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2641 * Will this entry fit in the buffer?
2643 if (outcount + reclen > bufsize) {
2645 * Did we manage to fit anything in the buffer?
2648 error = SET_ERROR(EINVAL);
2653 if (flags & V_RDDIR_ENTFLAGS) {
2655 * Add extended flag entry:
2657 eodp->ed_ino = objnum;
2658 eodp->ed_reclen = reclen;
2659 /* NOTE: ed_off is the offset for the *next* entry */
2660 next = &(eodp->ed_off);
2661 eodp->ed_eflags = zap.za_normalization_conflict ?
2662 ED_CASE_CONFLICT : 0;
2663 (void) strncpy(eodp->ed_name, zap.za_name,
2664 EDIRENT_NAMELEN(reclen));
2665 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2670 odp->d_ino = objnum;
2671 odp->d_reclen = reclen;
2672 odp->d_namlen = strlen(zap.za_name);
2673 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1);
2675 odp = (dirent64_t *)((intptr_t)odp + reclen);
2679 ASSERT(outcount <= bufsize);
2681 /* Prefetch znode */
2683 dmu_prefetch(os, objnum, 0, 0);
2687 * Move to the next entry, fill in the previous offset.
2689 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2690 zap_cursor_advance(&zc);
2691 offset = zap_cursor_serialize(&zc);
2696 if (cooks != NULL) {
2699 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks));
2702 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2704 /* Subtract unused cookies */
2705 if (ncookies != NULL)
2706 *ncookies -= ncooks;
2708 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2709 iovp->iov_base += outcount;
2710 iovp->iov_len -= outcount;
2711 uio->uio_resid -= outcount;
2712 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2714 * Reset the pointer.
2716 offset = uio->uio_loffset;
2720 zap_cursor_fini(&zc);
2721 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2722 kmem_free(outbuf, bufsize);
2724 if (error == ENOENT)
2727 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2729 uio->uio_loffset = offset;
2731 if (error != 0 && cookies != NULL) {
2732 free(*cookies, M_TEMP);
2739 ulong_t zfs_fsync_sync_cnt = 4;
2742 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2744 znode_t *zp = VTOZ(vp);
2745 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2747 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2749 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2752 zil_commit(zfsvfs->z_log, zp->z_id);
2760 * Get the requested file attributes and place them in the provided
2763 * IN: vp - vnode of file.
2764 * vap - va_mask identifies requested attributes.
2765 * If AT_XVATTR set, then optional attrs are requested
2766 * flags - ATTR_NOACLCHECK (CIFS server context)
2767 * cr - credentials of caller.
2768 * ct - caller context
2770 * OUT: vap - attribute values.
2772 * RETURN: 0 (always succeeds).
2776 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2777 caller_context_t *ct)
2779 znode_t *zp = VTOZ(vp);
2780 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2783 u_longlong_t nblocks;
2785 uint64_t mtime[2], ctime[2], crtime[2], rdev;
2786 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2787 xoptattr_t *xoap = NULL;
2788 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2789 sa_bulk_attr_t bulk[4];
2795 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2797 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2798 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2799 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
2800 if (vp->v_type == VBLK || vp->v_type == VCHR)
2801 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
2804 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2810 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2811 * Also, if we are the owner don't bother, since owner should
2812 * always be allowed to read basic attributes of file.
2814 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2815 (vap->va_uid != crgetuid(cr))) {
2816 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2824 * Return all attributes. It's cheaper to provide the answer
2825 * than to determine whether we were asked the question.
2828 mutex_enter(&zp->z_lock);
2829 vap->va_type = IFTOVT(zp->z_mode);
2830 vap->va_mode = zp->z_mode & ~S_IFMT;
2832 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2834 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
2836 vap->va_nodeid = zp->z_id;
2837 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2838 links = zp->z_links + 1;
2840 links = zp->z_links;
2841 vap->va_nlink = MIN(links, LINK_MAX); /* nlink_t limit! */
2842 vap->va_size = zp->z_size;
2844 vap->va_rdev = vp->v_rdev;
2846 if (vp->v_type == VBLK || vp->v_type == VCHR)
2847 vap->va_rdev = zfs_cmpldev(rdev);
2849 vap->va_seq = zp->z_seq;
2850 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */
2851 vap->va_filerev = zp->z_seq;
2854 * Add in any requested optional attributes and the create time.
2855 * Also set the corresponding bits in the returned attribute bitmap.
2857 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2858 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2860 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2861 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2864 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2865 xoap->xoa_readonly =
2866 ((zp->z_pflags & ZFS_READONLY) != 0);
2867 XVA_SET_RTN(xvap, XAT_READONLY);
2870 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2872 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2873 XVA_SET_RTN(xvap, XAT_SYSTEM);
2876 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2878 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2879 XVA_SET_RTN(xvap, XAT_HIDDEN);
2882 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2883 xoap->xoa_nounlink =
2884 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2885 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2888 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2889 xoap->xoa_immutable =
2890 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2891 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2894 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2895 xoap->xoa_appendonly =
2896 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2897 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2900 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2902 ((zp->z_pflags & ZFS_NODUMP) != 0);
2903 XVA_SET_RTN(xvap, XAT_NODUMP);
2906 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2908 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2909 XVA_SET_RTN(xvap, XAT_OPAQUE);
2912 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2913 xoap->xoa_av_quarantined =
2914 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2915 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2918 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2919 xoap->xoa_av_modified =
2920 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2921 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2924 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2925 vp->v_type == VREG) {
2926 zfs_sa_get_scanstamp(zp, xvap);
2929 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2932 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2933 times, sizeof (times));
2934 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2935 XVA_SET_RTN(xvap, XAT_CREATETIME);
2938 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2939 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2940 XVA_SET_RTN(xvap, XAT_REPARSE);
2942 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2943 xoap->xoa_generation = zp->z_gen;
2944 XVA_SET_RTN(xvap, XAT_GEN);
2947 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2949 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2950 XVA_SET_RTN(xvap, XAT_OFFLINE);
2953 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2955 ((zp->z_pflags & ZFS_SPARSE) != 0);
2956 XVA_SET_RTN(xvap, XAT_SPARSE);
2960 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2961 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2962 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2963 ZFS_TIME_DECODE(&vap->va_birthtime, crtime);
2965 mutex_exit(&zp->z_lock);
2967 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2968 vap->va_blksize = blksize;
2969 vap->va_bytes = nblocks << 9; /* nblocks * 512 */
2971 if (zp->z_blksz == 0) {
2973 * Block size hasn't been set; suggest maximal I/O transfers.
2975 vap->va_blksize = zfsvfs->z_max_blksz;
2983 * Set the file attributes to the values contained in the
2986 * IN: vp - vnode of file to be modified.
2987 * vap - new attribute values.
2988 * If AT_XVATTR set, then optional attrs are being set
2989 * flags - ATTR_UTIME set if non-default time values provided.
2990 * - ATTR_NOACLCHECK (CIFS context only).
2991 * cr - credentials of caller.
2992 * ct - caller context
2994 * RETURN: 0 on success, error code on failure.
2997 * vp - ctime updated, mtime updated if size changed.
3001 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
3002 caller_context_t *ct)
3004 znode_t *zp = VTOZ(vp);
3005 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3010 uint_t mask = vap->va_mask;
3011 uint_t saved_mask = 0;
3012 uint64_t saved_mode;
3015 uint64_t new_uid, new_gid;
3017 uint64_t mtime[2], ctime[2];
3019 int need_policy = FALSE;
3021 zfs_fuid_info_t *fuidp = NULL;
3022 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
3025 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
3026 boolean_t fuid_dirtied = B_FALSE;
3027 sa_bulk_attr_t bulk[7], xattr_bulk[7];
3028 int count = 0, xattr_count = 0;
3033 if (mask & AT_NOSET)
3034 return (SET_ERROR(EINVAL));
3039 zilog = zfsvfs->z_log;
3042 * Make sure that if we have ephemeral uid/gid or xvattr specified
3043 * that file system is at proper version level
3046 if (zfsvfs->z_use_fuids == B_FALSE &&
3047 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
3048 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
3049 (mask & AT_XVATTR))) {
3051 return (SET_ERROR(EINVAL));
3054 if (mask & AT_SIZE && vp->v_type == VDIR) {
3056 return (SET_ERROR(EISDIR));
3059 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
3061 return (SET_ERROR(EINVAL));
3065 * If this is an xvattr_t, then get a pointer to the structure of
3066 * optional attributes. If this is NULL, then we have a vattr_t.
3068 xoap = xva_getxoptattr(xvap);
3070 xva_init(&tmpxvattr);
3073 * Immutable files can only alter immutable bit and atime
3075 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
3076 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
3077 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
3079 return (SET_ERROR(EPERM));
3082 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
3084 return (SET_ERROR(EPERM));
3088 * Verify timestamps doesn't overflow 32 bits.
3089 * ZFS can handle large timestamps, but 32bit syscalls can't
3090 * handle times greater than 2039. This check should be removed
3091 * once large timestamps are fully supported.
3093 if (mask & (AT_ATIME | AT_MTIME)) {
3094 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
3095 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
3097 return (SET_ERROR(EOVERFLOW));
3105 /* Can this be moved to before the top label? */
3106 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
3108 return (SET_ERROR(EROFS));
3112 * First validate permissions
3115 if (mask & AT_SIZE) {
3117 * XXX - Note, we are not providing any open
3118 * mode flags here (like FNDELAY), so we may
3119 * block if there are locks present... this
3120 * should be addressed in openat().
3122 /* XXX - would it be OK to generate a log record here? */
3123 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
3130 if (mask & (AT_ATIME|AT_MTIME) ||
3131 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
3132 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
3133 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
3134 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
3135 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
3136 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
3137 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
3138 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
3142 if (mask & (AT_UID|AT_GID)) {
3143 int idmask = (mask & (AT_UID|AT_GID));
3148 * NOTE: even if a new mode is being set,
3149 * we may clear S_ISUID/S_ISGID bits.
3152 if (!(mask & AT_MODE))
3153 vap->va_mode = zp->z_mode;
3156 * Take ownership or chgrp to group we are a member of
3159 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
3160 take_group = (mask & AT_GID) &&
3161 zfs_groupmember(zfsvfs, vap->va_gid, cr);
3164 * If both AT_UID and AT_GID are set then take_owner and
3165 * take_group must both be set in order to allow taking
3168 * Otherwise, send the check through secpolicy_vnode_setattr()
3172 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
3173 ((idmask == AT_UID) && take_owner) ||
3174 ((idmask == AT_GID) && take_group)) {
3175 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
3176 skipaclchk, cr) == 0) {
3178 * Remove setuid/setgid for non-privileged users
3180 secpolicy_setid_clear(vap, vp, cr);
3181 trim_mask = (mask & (AT_UID|AT_GID));
3190 mutex_enter(&zp->z_lock);
3191 oldva.va_mode = zp->z_mode;
3192 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
3193 if (mask & AT_XVATTR) {
3195 * Update xvattr mask to include only those attributes
3196 * that are actually changing.
3198 * the bits will be restored prior to actually setting
3199 * the attributes so the caller thinks they were set.
3201 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
3202 if (xoap->xoa_appendonly !=
3203 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
3206 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
3207 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
3211 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
3212 if (xoap->xoa_nounlink !=
3213 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
3216 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
3217 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
3221 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
3222 if (xoap->xoa_immutable !=
3223 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
3226 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
3227 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
3231 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
3232 if (xoap->xoa_nodump !=
3233 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
3236 XVA_CLR_REQ(xvap, XAT_NODUMP);
3237 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
3241 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3242 if (xoap->xoa_av_modified !=
3243 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3246 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3247 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
3251 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3252 if ((vp->v_type != VREG &&
3253 xoap->xoa_av_quarantined) ||
3254 xoap->xoa_av_quarantined !=
3255 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3258 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3259 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
3263 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3264 mutex_exit(&zp->z_lock);
3266 return (SET_ERROR(EPERM));
3269 if (need_policy == FALSE &&
3270 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3271 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3276 mutex_exit(&zp->z_lock);
3278 if (mask & AT_MODE) {
3279 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3280 err = secpolicy_setid_setsticky_clear(vp, vap,
3286 trim_mask |= AT_MODE;
3294 * If trim_mask is set then take ownership
3295 * has been granted or write_acl is present and user
3296 * has the ability to modify mode. In that case remove
3297 * UID|GID and or MODE from mask so that
3298 * secpolicy_vnode_setattr() doesn't revoke it.
3302 saved_mask = vap->va_mask;
3303 vap->va_mask &= ~trim_mask;
3304 if (trim_mask & AT_MODE) {
3306 * Save the mode, as secpolicy_vnode_setattr()
3307 * will overwrite it with ova.va_mode.
3309 saved_mode = vap->va_mode;
3312 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3313 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3320 vap->va_mask |= saved_mask;
3321 if (trim_mask & AT_MODE) {
3323 * Recover the mode after
3324 * secpolicy_vnode_setattr().
3326 vap->va_mode = saved_mode;
3332 * secpolicy_vnode_setattr, or take ownership may have
3335 mask = vap->va_mask;
3337 if ((mask & (AT_UID | AT_GID))) {
3338 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3339 &xattr_obj, sizeof (xattr_obj));
3341 if (err == 0 && xattr_obj) {
3342 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3346 if (mask & AT_UID) {
3347 new_uid = zfs_fuid_create(zfsvfs,
3348 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3349 if (new_uid != zp->z_uid &&
3350 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3352 VN_RELE(ZTOV(attrzp));
3353 err = SET_ERROR(EDQUOT);
3358 if (mask & AT_GID) {
3359 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3360 cr, ZFS_GROUP, &fuidp);
3361 if (new_gid != zp->z_gid &&
3362 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3364 VN_RELE(ZTOV(attrzp));
3365 err = SET_ERROR(EDQUOT);
3370 tx = dmu_tx_create(zfsvfs->z_os);
3372 if (mask & AT_MODE) {
3373 uint64_t pmode = zp->z_mode;
3375 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3377 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3378 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3379 err = SET_ERROR(EPERM);
3383 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3386 mutex_enter(&zp->z_lock);
3387 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3389 * Are we upgrading ACL from old V0 format
3392 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3393 zfs_znode_acl_version(zp) ==
3394 ZFS_ACL_VERSION_INITIAL) {
3395 dmu_tx_hold_free(tx, acl_obj, 0,
3397 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3398 0, aclp->z_acl_bytes);
3400 dmu_tx_hold_write(tx, acl_obj, 0,
3403 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3404 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3405 0, aclp->z_acl_bytes);
3407 mutex_exit(&zp->z_lock);
3408 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3410 if ((mask & AT_XVATTR) &&
3411 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3412 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3414 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3418 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3421 fuid_dirtied = zfsvfs->z_fuid_dirty;
3423 zfs_fuid_txhold(zfsvfs, tx);
3425 zfs_sa_upgrade_txholds(tx, zp);
3427 err = dmu_tx_assign(tx, TXG_WAIT);
3433 * Set each attribute requested.
3434 * We group settings according to the locks they need to acquire.
3436 * Note: you cannot set ctime directly, although it will be
3437 * updated as a side-effect of calling this function.
3441 if (mask & (AT_UID|AT_GID|AT_MODE))
3442 mutex_enter(&zp->z_acl_lock);
3443 mutex_enter(&zp->z_lock);
3445 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3446 &zp->z_pflags, sizeof (zp->z_pflags));
3449 if (mask & (AT_UID|AT_GID|AT_MODE))
3450 mutex_enter(&attrzp->z_acl_lock);
3451 mutex_enter(&attrzp->z_lock);
3452 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3453 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3454 sizeof (attrzp->z_pflags));
3457 if (mask & (AT_UID|AT_GID)) {
3459 if (mask & AT_UID) {
3460 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3461 &new_uid, sizeof (new_uid));
3462 zp->z_uid = new_uid;
3464 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3465 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3467 attrzp->z_uid = new_uid;
3471 if (mask & AT_GID) {
3472 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3473 NULL, &new_gid, sizeof (new_gid));
3474 zp->z_gid = new_gid;
3476 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3477 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3479 attrzp->z_gid = new_gid;
3482 if (!(mask & AT_MODE)) {
3483 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3484 NULL, &new_mode, sizeof (new_mode));
3485 new_mode = zp->z_mode;
3487 err = zfs_acl_chown_setattr(zp);
3490 err = zfs_acl_chown_setattr(attrzp);
3495 if (mask & AT_MODE) {
3496 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3497 &new_mode, sizeof (new_mode));
3498 zp->z_mode = new_mode;
3499 ASSERT3U((uintptr_t)aclp, !=, 0);
3500 err = zfs_aclset_common(zp, aclp, cr, tx);
3502 if (zp->z_acl_cached)
3503 zfs_acl_free(zp->z_acl_cached);
3504 zp->z_acl_cached = aclp;
3509 if (mask & AT_ATIME) {
3510 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3511 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3512 &zp->z_atime, sizeof (zp->z_atime));
3515 if (mask & AT_MTIME) {
3516 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3517 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3518 mtime, sizeof (mtime));
3521 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3522 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3523 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3524 NULL, mtime, sizeof (mtime));
3525 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3526 &ctime, sizeof (ctime));
3527 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3529 } else if (mask != 0) {
3530 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3531 &ctime, sizeof (ctime));
3532 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3535 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3536 SA_ZPL_CTIME(zfsvfs), NULL,
3537 &ctime, sizeof (ctime));
3538 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3539 mtime, ctime, B_TRUE);
3543 * Do this after setting timestamps to prevent timestamp
3544 * update from toggling bit
3547 if (xoap && (mask & AT_XVATTR)) {
3550 * restore trimmed off masks
3551 * so that return masks can be set for caller.
3554 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3555 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3557 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3558 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3560 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3561 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3563 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3564 XVA_SET_REQ(xvap, XAT_NODUMP);
3566 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3567 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3569 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3570 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3573 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3574 ASSERT(vp->v_type == VREG);
3576 zfs_xvattr_set(zp, xvap, tx);
3580 zfs_fuid_sync(zfsvfs, tx);
3583 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3585 mutex_exit(&zp->z_lock);
3586 if (mask & (AT_UID|AT_GID|AT_MODE))
3587 mutex_exit(&zp->z_acl_lock);
3590 if (mask & (AT_UID|AT_GID|AT_MODE))
3591 mutex_exit(&attrzp->z_acl_lock);
3592 mutex_exit(&attrzp->z_lock);
3595 if (err == 0 && attrzp) {
3596 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3602 VN_RELE(ZTOV(attrzp));
3608 zfs_fuid_info_free(fuidp);
3614 if (err == ERESTART)
3617 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3622 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3623 zil_commit(zilog, 0);
3629 typedef struct zfs_zlock {
3630 krwlock_t *zl_rwlock; /* lock we acquired */
3631 znode_t *zl_znode; /* znode we held */
3632 struct zfs_zlock *zl_next; /* next in list */
3636 * Drop locks and release vnodes that were held by zfs_rename_lock().
3639 zfs_rename_unlock(zfs_zlock_t **zlpp)
3643 while ((zl = *zlpp) != NULL) {
3644 if (zl->zl_znode != NULL)
3645 VN_RELE(ZTOV(zl->zl_znode));
3646 rw_exit(zl->zl_rwlock);
3647 *zlpp = zl->zl_next;
3648 kmem_free(zl, sizeof (*zl));
3653 * Search back through the directory tree, using the ".." entries.
3654 * Lock each directory in the chain to prevent concurrent renames.
3655 * Fail any attempt to move a directory into one of its own descendants.
3656 * XXX - z_parent_lock can overlap with map or grow locks
3659 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3663 uint64_t rootid = zp->z_zfsvfs->z_root;
3664 uint64_t oidp = zp->z_id;
3665 krwlock_t *rwlp = &szp->z_parent_lock;
3666 krw_t rw = RW_WRITER;
3669 * First pass write-locks szp and compares to zp->z_id.
3670 * Later passes read-lock zp and compare to zp->z_parent.
3673 if (!rw_tryenter(rwlp, rw)) {
3675 * Another thread is renaming in this path.
3676 * Note that if we are a WRITER, we don't have any
3677 * parent_locks held yet.
3679 if (rw == RW_READER && zp->z_id > szp->z_id) {
3681 * Drop our locks and restart
3683 zfs_rename_unlock(&zl);
3687 rwlp = &szp->z_parent_lock;
3692 * Wait for other thread to drop its locks
3698 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3699 zl->zl_rwlock = rwlp;
3700 zl->zl_znode = NULL;
3701 zl->zl_next = *zlpp;
3704 if (oidp == szp->z_id) /* We're a descendant of szp */
3705 return (SET_ERROR(EINVAL));
3707 if (oidp == rootid) /* We've hit the top */
3710 if (rw == RW_READER) { /* i.e. not the first pass */
3711 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3716 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3717 &oidp, sizeof (oidp));
3718 rwlp = &zp->z_parent_lock;
3721 } while (zp->z_id != sdzp->z_id);
3727 * Move an entry from the provided source directory to the target
3728 * directory. Change the entry name as indicated.
3730 * IN: sdvp - Source directory containing the "old entry".
3731 * snm - Old entry name.
3732 * tdvp - Target directory to contain the "new entry".
3733 * tnm - New entry name.
3734 * cr - credentials of caller.
3735 * ct - caller context
3736 * flags - case flags
3738 * RETURN: 0 on success, error code on failure.
3741 * sdvp,tdvp - ctime|mtime updated
3745 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3746 caller_context_t *ct, int flags)
3748 znode_t *tdzp, *szp, *tzp;
3749 znode_t *sdzp = VTOZ(sdvp);
3750 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs;
3753 zfs_dirlock_t *sdl, *tdl;
3756 int cmp, serr, terr;
3759 boolean_t waited = B_FALSE;
3762 ZFS_VERIFY_ZP(sdzp);
3763 zilog = zfsvfs->z_log;
3766 * Make sure we have the real vp for the target directory.
3768 if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3771 if (tdvp->v_vfsp != sdvp->v_vfsp || zfsctl_is_node(tdvp)) {
3773 return (SET_ERROR(EXDEV));
3777 ZFS_VERIFY_ZP(tdzp);
3778 if (zfsvfs->z_utf8 && u8_validate(tnm,
3779 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3781 return (SET_ERROR(EILSEQ));
3784 if (flags & FIGNORECASE)
3793 * This is to prevent the creation of links into attribute space
3794 * by renaming a linked file into/outof an attribute directory.
3795 * See the comment in zfs_link() for why this is considered bad.
3797 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3799 return (SET_ERROR(EINVAL));
3803 * Lock source and target directory entries. To prevent deadlock,
3804 * a lock ordering must be defined. We lock the directory with
3805 * the smallest object id first, or if it's a tie, the one with
3806 * the lexically first name.
3808 if (sdzp->z_id < tdzp->z_id) {
3810 } else if (sdzp->z_id > tdzp->z_id) {
3814 * First compare the two name arguments without
3815 * considering any case folding.
3817 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3819 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3820 ASSERT(error == 0 || !zfsvfs->z_utf8);
3823 * POSIX: "If the old argument and the new argument
3824 * both refer to links to the same existing file,
3825 * the rename() function shall return successfully
3826 * and perform no other action."
3832 * If the file system is case-folding, then we may
3833 * have some more checking to do. A case-folding file
3834 * system is either supporting mixed case sensitivity
3835 * access or is completely case-insensitive. Note
3836 * that the file system is always case preserving.
3838 * In mixed sensitivity mode case sensitive behavior
3839 * is the default. FIGNORECASE must be used to
3840 * explicitly request case insensitive behavior.
3842 * If the source and target names provided differ only
3843 * by case (e.g., a request to rename 'tim' to 'Tim'),
3844 * we will treat this as a special case in the
3845 * case-insensitive mode: as long as the source name
3846 * is an exact match, we will allow this to proceed as
3847 * a name-change request.
3849 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3850 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3851 flags & FIGNORECASE)) &&
3852 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3855 * case preserving rename request, require exact
3864 * If the source and destination directories are the same, we should
3865 * grab the z_name_lock of that directory only once.
3869 rw_enter(&sdzp->z_name_lock, RW_READER);
3873 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3874 ZEXISTS | zflg, NULL, NULL);
3875 terr = zfs_dirent_lock(&tdl,
3876 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3878 terr = zfs_dirent_lock(&tdl,
3879 tdzp, tnm, &tzp, zflg, NULL, NULL);
3880 serr = zfs_dirent_lock(&sdl,
3881 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3887 * Source entry invalid or not there.
3890 zfs_dirent_unlock(tdl);
3896 rw_exit(&sdzp->z_name_lock);
3899 * FreeBSD: In OpenSolaris they only check if rename source is
3900 * ".." here, because "." is handled in their lookup. This is
3901 * not the case for FreeBSD, so we check for "." explicitly.
3903 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0)
3904 serr = SET_ERROR(EINVAL);
3909 zfs_dirent_unlock(sdl);
3913 rw_exit(&sdzp->z_name_lock);
3915 if (strcmp(tnm, "..") == 0)
3916 terr = SET_ERROR(EINVAL);
3922 * Must have write access at the source to remove the old entry
3923 * and write access at the target to create the new entry.
3924 * Note that if target and source are the same, this can be
3925 * done in a single check.
3928 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3931 if (ZTOV(szp)->v_type == VDIR) {
3933 * Check to make sure rename is valid.
3934 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3936 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3941 * Does target exist?
3945 * Source and target must be the same type.
3947 if (ZTOV(szp)->v_type == VDIR) {
3948 if (ZTOV(tzp)->v_type != VDIR) {
3949 error = SET_ERROR(ENOTDIR);
3953 if (ZTOV(tzp)->v_type == VDIR) {
3954 error = SET_ERROR(EISDIR);
3959 * POSIX dictates that when the source and target
3960 * entries refer to the same file object, rename
3961 * must do nothing and exit without error.
3963 if (szp->z_id == tzp->z_id) {
3969 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3971 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3974 * notify the target directory if it is not the same
3975 * as source directory.
3978 vnevent_rename_dest_dir(tdvp, ct);
3981 tx = dmu_tx_create(zfsvfs->z_os);
3982 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3983 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3984 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3985 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3987 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3988 zfs_sa_upgrade_txholds(tx, tdzp);
3991 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3992 zfs_sa_upgrade_txholds(tx, tzp);
3995 zfs_sa_upgrade_txholds(tx, szp);
3996 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3997 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4000 zfs_rename_unlock(&zl);
4001 zfs_dirent_unlock(sdl);
4002 zfs_dirent_unlock(tdl);
4005 rw_exit(&sdzp->z_name_lock);
4010 if (error == ERESTART) {
4021 if (tzp) /* Attempt to remove the existing target */
4022 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
4025 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
4027 szp->z_pflags |= ZFS_AV_MODIFIED;
4029 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
4030 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
4033 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
4035 zfs_log_rename(zilog, tx, TX_RENAME |
4036 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
4037 sdl->dl_name, tdzp, tdl->dl_name, szp);
4040 * Update path information for the target vnode
4042 vn_renamepath(tdvp, ZTOV(szp), tnm,
4046 * At this point, we have successfully created
4047 * the target name, but have failed to remove
4048 * the source name. Since the create was done
4049 * with the ZRENAMING flag, there are
4050 * complications; for one, the link count is
4051 * wrong. The easiest way to deal with this
4052 * is to remove the newly created target, and
4053 * return the original error. This must
4054 * succeed; fortunately, it is very unlikely to
4055 * fail, since we just created it.
4057 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
4058 ZRENAMING, NULL), ==, 0);
4061 #ifdef FREEBSD_NAMECACHE
4065 cache_purge(ZTOV(szp));
4067 cache_purge(ZTOV(tzp));
4075 zfs_rename_unlock(&zl);
4077 zfs_dirent_unlock(sdl);
4078 zfs_dirent_unlock(tdl);
4081 rw_exit(&sdzp->z_name_lock);
4088 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4089 zil_commit(zilog, 0);
4097 * Insert the indicated symbolic reference entry into the directory.
4099 * IN: dvp - Directory to contain new symbolic link.
4100 * link - Name for new symlink entry.
4101 * vap - Attributes of new entry.
4102 * cr - credentials of caller.
4103 * ct - caller context
4104 * flags - case flags
4106 * RETURN: 0 on success, error code on failure.
4109 * dvp - ctime|mtime updated
4113 zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link,
4114 cred_t *cr, kthread_t *td)
4116 znode_t *zp, *dzp = VTOZ(dvp);
4119 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4121 uint64_t len = strlen(link);
4124 zfs_acl_ids_t acl_ids;
4125 boolean_t fuid_dirtied;
4126 uint64_t txtype = TX_SYMLINK;
4127 boolean_t waited = B_FALSE;
4130 ASSERT(vap->va_type == VLNK);
4134 zilog = zfsvfs->z_log;
4136 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
4137 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4139 return (SET_ERROR(EILSEQ));
4141 if (flags & FIGNORECASE)
4144 if (len > MAXPATHLEN) {
4146 return (SET_ERROR(ENAMETOOLONG));
4149 if ((error = zfs_acl_ids_create(dzp, 0,
4150 vap, cr, NULL, &acl_ids)) != 0) {
4155 getnewvnode_reserve(1);
4159 * Attempt to lock directory; fail if entry already exists.
4161 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
4163 zfs_acl_ids_free(&acl_ids);
4164 getnewvnode_drop_reserve();
4169 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4170 zfs_acl_ids_free(&acl_ids);
4171 zfs_dirent_unlock(dl);
4172 getnewvnode_drop_reserve();
4177 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
4178 zfs_acl_ids_free(&acl_ids);
4179 zfs_dirent_unlock(dl);
4180 getnewvnode_drop_reserve();
4182 return (SET_ERROR(EDQUOT));
4184 tx = dmu_tx_create(zfsvfs->z_os);
4185 fuid_dirtied = zfsvfs->z_fuid_dirty;
4186 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
4187 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4188 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
4189 ZFS_SA_BASE_ATTR_SIZE + len);
4190 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
4191 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
4192 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
4193 acl_ids.z_aclp->z_acl_bytes);
4196 zfs_fuid_txhold(zfsvfs, tx);
4197 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4199 zfs_dirent_unlock(dl);
4200 if (error == ERESTART) {
4206 zfs_acl_ids_free(&acl_ids);
4208 getnewvnode_drop_reserve();
4214 * Create a new object for the symlink.
4215 * for version 4 ZPL datsets the symlink will be an SA attribute
4217 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
4220 zfs_fuid_sync(zfsvfs, tx);
4222 mutex_enter(&zp->z_lock);
4224 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
4227 zfs_sa_symlink(zp, link, len, tx);
4228 mutex_exit(&zp->z_lock);
4231 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
4232 &zp->z_size, sizeof (zp->z_size), tx);
4234 * Insert the new object into the directory.
4236 (void) zfs_link_create(dl, zp, tx, ZNEW);
4238 if (flags & FIGNORECASE)
4240 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
4243 zfs_acl_ids_free(&acl_ids);
4247 getnewvnode_drop_reserve();
4249 zfs_dirent_unlock(dl);
4251 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4252 zil_commit(zilog, 0);
4259 * Return, in the buffer contained in the provided uio structure,
4260 * the symbolic path referred to by vp.
4262 * IN: vp - vnode of symbolic link.
4263 * uio - structure to contain the link path.
4264 * cr - credentials of caller.
4265 * ct - caller context
4267 * OUT: uio - structure containing the link path.
4269 * RETURN: 0 on success, error code on failure.
4272 * vp - atime updated
4276 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
4278 znode_t *zp = VTOZ(vp);
4279 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4285 mutex_enter(&zp->z_lock);
4287 error = sa_lookup_uio(zp->z_sa_hdl,
4288 SA_ZPL_SYMLINK(zfsvfs), uio);
4290 error = zfs_sa_readlink(zp, uio);
4291 mutex_exit(&zp->z_lock);
4293 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4300 * Insert a new entry into directory tdvp referencing svp.
4302 * IN: tdvp - Directory to contain new entry.
4303 * svp - vnode of new entry.
4304 * name - name of new entry.
4305 * cr - credentials of caller.
4306 * ct - caller context
4308 * RETURN: 0 on success, error code on failure.
4311 * tdvp - ctime|mtime updated
4312 * svp - ctime updated
4316 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
4317 caller_context_t *ct, int flags)
4319 znode_t *dzp = VTOZ(tdvp);
4321 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4330 boolean_t waited = B_FALSE;
4332 ASSERT(tdvp->v_type == VDIR);
4336 zilog = zfsvfs->z_log;
4338 if (VOP_REALVP(svp, &realvp, ct) == 0)
4342 * POSIX dictates that we return EPERM here.
4343 * Better choices include ENOTSUP or EISDIR.
4345 if (svp->v_type == VDIR) {
4347 return (SET_ERROR(EPERM));
4350 if (svp->v_vfsp != tdvp->v_vfsp || zfsctl_is_node(svp)) {
4352 return (SET_ERROR(EXDEV));
4358 /* Prevent links to .zfs/shares files */
4360 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4361 &parent, sizeof (uint64_t))) != 0) {
4365 if (parent == zfsvfs->z_shares_dir) {
4367 return (SET_ERROR(EPERM));
4370 if (zfsvfs->z_utf8 && u8_validate(name,
4371 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4373 return (SET_ERROR(EILSEQ));
4375 if (flags & FIGNORECASE)
4379 * We do not support links between attributes and non-attributes
4380 * because of the potential security risk of creating links
4381 * into "normal" file space in order to circumvent restrictions
4382 * imposed in attribute space.
4384 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4386 return (SET_ERROR(EINVAL));
4390 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4391 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) {
4393 return (SET_ERROR(EPERM));
4396 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4403 * Attempt to lock directory; fail if entry already exists.
4405 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4411 tx = dmu_tx_create(zfsvfs->z_os);
4412 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4413 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4414 zfs_sa_upgrade_txholds(tx, szp);
4415 zfs_sa_upgrade_txholds(tx, dzp);
4416 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4418 zfs_dirent_unlock(dl);
4419 if (error == ERESTART) {
4430 error = zfs_link_create(dl, szp, tx, 0);
4433 uint64_t txtype = TX_LINK;
4434 if (flags & FIGNORECASE)
4436 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4441 zfs_dirent_unlock(dl);
4444 vnevent_link(svp, ct);
4447 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4448 zil_commit(zilog, 0);
4456 * zfs_null_putapage() is used when the file system has been force
4457 * unmounted. It just drops the pages.
4461 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4462 size_t *lenp, int flags, cred_t *cr)
4464 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4469 * Push a page out to disk, klustering if possible.
4471 * IN: vp - file to push page to.
4472 * pp - page to push.
4473 * flags - additional flags.
4474 * cr - credentials of caller.
4476 * OUT: offp - start of range pushed.
4477 * lenp - len of range pushed.
4479 * RETURN: 0 on success, error code on failure.
4481 * NOTE: callers must have locked the page to be pushed. On
4482 * exit, the page (and all other pages in the kluster) must be
4487 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4488 size_t *lenp, int flags, cred_t *cr)
4490 znode_t *zp = VTOZ(vp);
4491 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4493 u_offset_t off, koff;
4500 * If our blocksize is bigger than the page size, try to kluster
4501 * multiple pages so that we write a full block (thus avoiding
4502 * a read-modify-write).
4504 if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4505 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4506 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4507 ASSERT(koff <= zp->z_size);
4508 if (koff + klen > zp->z_size)
4509 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4510 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4512 ASSERT3U(btop(len), ==, btopr(len));
4515 * Can't push pages past end-of-file.
4517 if (off >= zp->z_size) {
4518 /* ignore all pages */
4521 } else if (off + len > zp->z_size) {
4522 int npages = btopr(zp->z_size - off);
4525 page_list_break(&pp, &trunc, npages);
4526 /* ignore pages past end of file */
4528 pvn_write_done(trunc, flags);
4529 len = zp->z_size - off;
4532 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4533 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4534 err = SET_ERROR(EDQUOT);
4537 tx = dmu_tx_create(zfsvfs->z_os);
4538 dmu_tx_hold_write(tx, zp->z_id, off, len);
4540 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4541 zfs_sa_upgrade_txholds(tx, zp);
4542 err = dmu_tx_assign(tx, TXG_WAIT);
4548 if (zp->z_blksz <= PAGESIZE) {
4549 caddr_t va = zfs_map_page(pp, S_READ);
4550 ASSERT3U(len, <=, PAGESIZE);
4551 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4552 zfs_unmap_page(pp, va);
4554 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4558 uint64_t mtime[2], ctime[2];
4559 sa_bulk_attr_t bulk[3];
4562 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4564 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4566 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4568 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4570 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4575 pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4585 * Copy the portion of the file indicated from pages into the file.
4586 * The pages are stored in a page list attached to the files vnode.
4588 * IN: vp - vnode of file to push page data to.
4589 * off - position in file to put data.
4590 * len - amount of data to write.
4591 * flags - flags to control the operation.
4592 * cr - credentials of caller.
4593 * ct - caller context.
4595 * RETURN: 0 on success, error code on failure.
4598 * vp - ctime|mtime updated
4602 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4603 caller_context_t *ct)
4605 znode_t *zp = VTOZ(vp);
4606 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4618 * Align this request to the file block size in case we kluster.
4619 * XXX - this can result in pretty aggresive locking, which can
4620 * impact simultanious read/write access. One option might be
4621 * to break up long requests (len == 0) into block-by-block
4622 * operations to get narrower locking.
4624 blksz = zp->z_blksz;
4626 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4629 if (len > 0 && ISP2(blksz))
4630 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4636 * Search the entire vp list for pages >= io_off.
4638 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4639 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4642 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4644 if (off > zp->z_size) {
4645 /* past end of file */
4646 zfs_range_unlock(rl);
4651 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4653 for (off = io_off; io_off < off + len; io_off += io_len) {
4654 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4655 pp = page_lookup(vp, io_off,
4656 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4658 pp = page_lookup_nowait(vp, io_off,
4659 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4662 if (pp != NULL && pvn_getdirty(pp, flags)) {
4666 * Found a dirty page to push
4668 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4676 zfs_range_unlock(rl);
4677 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4678 zil_commit(zfsvfs->z_log, zp->z_id);
4686 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4688 znode_t *zp = VTOZ(vp);
4689 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4692 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4693 if (zp->z_sa_hdl == NULL) {
4695 * The fs has been unmounted, or we did a
4696 * suspend/resume and this file no longer exists.
4698 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4699 vrecycle(vp, curthread);
4703 mutex_enter(&zp->z_lock);
4704 if (zp->z_unlinked) {
4706 * Fast path to recycle a vnode of a removed file.
4708 mutex_exit(&zp->z_lock);
4709 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4710 vrecycle(vp, curthread);
4713 mutex_exit(&zp->z_lock);
4715 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4716 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4718 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4719 zfs_sa_upgrade_txholds(tx, zp);
4720 error = dmu_tx_assign(tx, TXG_WAIT);
4724 mutex_enter(&zp->z_lock);
4725 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4726 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4727 zp->z_atime_dirty = 0;
4728 mutex_exit(&zp->z_lock);
4732 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4737 * Bounds-check the seek operation.
4739 * IN: vp - vnode seeking within
4740 * ooff - old file offset
4741 * noffp - pointer to new file offset
4742 * ct - caller context
4744 * RETURN: 0 on success, EINVAL if new offset invalid.
4748 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4749 caller_context_t *ct)
4751 if (vp->v_type == VDIR)
4753 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4757 * Pre-filter the generic locking function to trap attempts to place
4758 * a mandatory lock on a memory mapped file.
4761 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4762 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4764 znode_t *zp = VTOZ(vp);
4765 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4771 * We are following the UFS semantics with respect to mapcnt
4772 * here: If we see that the file is mapped already, then we will
4773 * return an error, but we don't worry about races between this
4774 * function and zfs_map().
4776 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4778 return (SET_ERROR(EAGAIN));
4781 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4785 * If we can't find a page in the cache, we will create a new page
4786 * and fill it with file data. For efficiency, we may try to fill
4787 * multiple pages at once (klustering) to fill up the supplied page
4788 * list. Note that the pages to be filled are held with an exclusive
4789 * lock to prevent access by other threads while they are being filled.
4792 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4793 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4795 znode_t *zp = VTOZ(vp);
4796 page_t *pp, *cur_pp;
4797 objset_t *os = zp->z_zfsvfs->z_os;
4798 u_offset_t io_off, total;
4802 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4804 * We only have a single page, don't bother klustering
4808 pp = page_create_va(vp, io_off, io_len,
4809 PG_EXCL | PG_WAIT, seg, addr);
4812 * Try to find enough pages to fill the page list
4814 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4815 &io_len, off, plsz, 0);
4819 * The page already exists, nothing to do here.
4826 * Fill the pages in the kluster.
4829 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4832 ASSERT3U(io_off, ==, cur_pp->p_offset);
4833 va = zfs_map_page(cur_pp, S_WRITE);
4834 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4836 zfs_unmap_page(cur_pp, va);
4838 /* On error, toss the entire kluster */
4839 pvn_read_done(pp, B_ERROR);
4840 /* convert checksum errors into IO errors */
4842 err = SET_ERROR(EIO);
4845 cur_pp = cur_pp->p_next;
4849 * Fill in the page list array from the kluster starting
4850 * from the desired offset `off'.
4851 * NOTE: the page list will always be null terminated.
4853 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4854 ASSERT(pl == NULL || (*pl)->p_offset == off);
4860 * Return pointers to the pages for the file region [off, off + len]
4861 * in the pl array. If plsz is greater than len, this function may
4862 * also return page pointers from after the specified region
4863 * (i.e. the region [off, off + plsz]). These additional pages are
4864 * only returned if they are already in the cache, or were created as
4865 * part of a klustered read.
4867 * IN: vp - vnode of file to get data from.
4868 * off - position in file to get data from.
4869 * len - amount of data to retrieve.
4870 * plsz - length of provided page list.
4871 * seg - segment to obtain pages for.
4872 * addr - virtual address of fault.
4873 * rw - mode of created pages.
4874 * cr - credentials of caller.
4875 * ct - caller context.
4877 * OUT: protp - protection mode of created pages.
4878 * pl - list of pages created.
4880 * RETURN: 0 on success, error code on failure.
4883 * vp - atime updated
4887 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4888 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4889 enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4891 znode_t *zp = VTOZ(vp);
4892 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4896 /* we do our own caching, faultahead is unnecessary */
4899 else if (len > plsz)
4902 len = P2ROUNDUP(len, PAGESIZE);
4903 ASSERT(plsz >= len);
4912 * Loop through the requested range [off, off + len) looking
4913 * for pages. If we don't find a page, we will need to create
4914 * a new page and fill it with data from the file.
4917 if (*pl = page_lookup(vp, off, SE_SHARED))
4919 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4922 ASSERT3U((*pl)->p_offset, ==, off);
4926 ASSERT3U(len, >=, PAGESIZE);
4929 ASSERT3U(plsz, >=, PAGESIZE);
4936 * Fill out the page array with any pages already in the cache.
4939 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4946 * Release any pages we have previously locked.
4951 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4961 * Request a memory map for a section of a file. This code interacts
4962 * with common code and the VM system as follows:
4964 * - common code calls mmap(), which ends up in smmap_common()
4965 * - this calls VOP_MAP(), which takes you into (say) zfs
4966 * - zfs_map() calls as_map(), passing segvn_create() as the callback
4967 * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4968 * - zfs_addmap() updates z_mapcnt
4972 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4973 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4974 caller_context_t *ct)
4976 znode_t *zp = VTOZ(vp);
4977 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4978 segvn_crargs_t vn_a;
4984 if ((prot & PROT_WRITE) && (zp->z_pflags &
4985 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4987 return (SET_ERROR(EPERM));
4990 if ((prot & (PROT_READ | PROT_EXEC)) &&
4991 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4993 return (SET_ERROR(EACCES));
4996 if (vp->v_flag & VNOMAP) {
4998 return (SET_ERROR(ENOSYS));
5001 if (off < 0 || len > MAXOFFSET_T - off) {
5003 return (SET_ERROR(ENXIO));
5006 if (vp->v_type != VREG) {
5008 return (SET_ERROR(ENODEV));
5012 * If file is locked, disallow mapping.
5014 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
5016 return (SET_ERROR(EAGAIN));
5020 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
5028 vn_a.offset = (u_offset_t)off;
5029 vn_a.type = flags & MAP_TYPE;
5031 vn_a.maxprot = maxprot;
5034 vn_a.flags = flags & ~MAP_TYPE;
5036 vn_a.lgrp_mem_policy_flags = 0;
5038 error = as_map(as, *addrp, len, segvn_create, &vn_a);
5047 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5048 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
5049 caller_context_t *ct)
5051 uint64_t pages = btopr(len);
5053 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
5058 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
5059 * more accurate mtime for the associated file. Since we don't have a way of
5060 * detecting when the data was actually modified, we have to resort to
5061 * heuristics. If an explicit msync() is done, then we mark the mtime when the
5062 * last page is pushed. The problem occurs when the msync() call is omitted,
5063 * which by far the most common case:
5071 * putpage() via fsflush
5073 * If we wait until fsflush to come along, we can have a modification time that
5074 * is some arbitrary point in the future. In order to prevent this in the
5075 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
5080 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5081 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
5082 caller_context_t *ct)
5084 uint64_t pages = btopr(len);
5086 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
5087 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
5089 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
5090 vn_has_cached_data(vp))
5091 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
5097 * Free or allocate space in a file. Currently, this function only
5098 * supports the `F_FREESP' command. However, this command is somewhat
5099 * misnamed, as its functionality includes the ability to allocate as
5100 * well as free space.
5102 * IN: vp - vnode of file to free data in.
5103 * cmd - action to take (only F_FREESP supported).
5104 * bfp - section of file to free/alloc.
5105 * flag - current file open mode flags.
5106 * offset - current file offset.
5107 * cr - credentials of caller [UNUSED].
5108 * ct - caller context.
5110 * RETURN: 0 on success, error code on failure.
5113 * vp - ctime|mtime updated
5117 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
5118 offset_t offset, cred_t *cr, caller_context_t *ct)
5120 znode_t *zp = VTOZ(vp);
5121 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5128 if (cmd != F_FREESP) {
5130 return (SET_ERROR(EINVAL));
5133 if (error = convoff(vp, bfp, 0, offset)) {
5138 if (bfp->l_len < 0) {
5140 return (SET_ERROR(EINVAL));
5144 len = bfp->l_len; /* 0 means from off to end of file */
5146 error = zfs_freesp(zp, off, len, flag, TRUE);
5153 CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid));
5154 CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid));
5158 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
5160 znode_t *zp = VTOZ(vp);
5161 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5164 uint64_t object = zp->z_id;
5171 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
5172 &gen64, sizeof (uint64_t))) != 0) {
5177 gen = (uint32_t)gen64;
5179 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
5182 if (fidp->fid_len < size) {
5183 fidp->fid_len = size;
5185 return (SET_ERROR(ENOSPC));
5188 fidp->fid_len = size;
5191 zfid = (zfid_short_t *)fidp;
5193 zfid->zf_len = size;
5195 for (i = 0; i < sizeof (zfid->zf_object); i++)
5196 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
5198 /* Must have a non-zero generation number to distinguish from .zfs */
5201 for (i = 0; i < sizeof (zfid->zf_gen); i++)
5202 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
5204 if (size == LONG_FID_LEN) {
5205 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
5208 zlfid = (zfid_long_t *)fidp;
5210 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
5211 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
5213 /* XXX - this should be the generation number for the objset */
5214 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
5215 zlfid->zf_setgen[i] = 0;
5223 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
5224 caller_context_t *ct)
5236 case _PC_FILESIZEBITS:
5240 case _PC_XATTR_EXISTS:
5242 zfsvfs = zp->z_zfsvfs;
5246 error = zfs_dirent_lock(&dl, zp, "", &xzp,
5247 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
5249 zfs_dirent_unlock(dl);
5250 if (!zfs_dirempty(xzp))
5253 } else if (error == ENOENT) {
5255 * If there aren't extended attributes, it's the
5256 * same as having zero of them.
5263 case _PC_SATTR_ENABLED:
5264 case _PC_SATTR_EXISTS:
5265 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
5266 (vp->v_type == VREG || vp->v_type == VDIR);
5269 case _PC_ACCESS_FILTERING:
5270 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
5274 case _PC_ACL_ENABLED:
5275 *valp = _ACL_ACE_ENABLED;
5278 case _PC_MIN_HOLE_SIZE:
5279 *valp = (int)SPA_MINBLOCKSIZE;
5282 case _PC_TIMESTAMP_RESOLUTION:
5283 /* nanosecond timestamp resolution */
5287 case _PC_ACL_EXTENDED:
5295 case _PC_ACL_PATH_MAX:
5296 *valp = ACL_MAX_ENTRIES;
5300 return (EOPNOTSUPP);
5306 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5307 caller_context_t *ct)
5309 znode_t *zp = VTOZ(vp);
5310 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5312 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5316 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5324 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5325 caller_context_t *ct)
5327 znode_t *zp = VTOZ(vp);
5328 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5330 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5331 zilog_t *zilog = zfsvfs->z_log;
5336 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5338 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5339 zil_commit(zilog, 0);
5347 * The smallest read we may consider to loan out an arcbuf.
5348 * This must be a power of 2.
5350 int zcr_blksz_min = (1 << 10); /* 1K */
5352 * If set to less than the file block size, allow loaning out of an
5353 * arcbuf for a partial block read. This must be a power of 2.
5355 int zcr_blksz_max = (1 << 17); /* 128K */
5359 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5360 caller_context_t *ct)
5362 znode_t *zp = VTOZ(vp);
5363 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5364 int max_blksz = zfsvfs->z_max_blksz;
5365 uio_t *uio = &xuio->xu_uio;
5366 ssize_t size = uio->uio_resid;
5367 offset_t offset = uio->uio_loffset;
5372 int preamble, postamble;
5374 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5375 return (SET_ERROR(EINVAL));
5382 * Loan out an arc_buf for write if write size is bigger than
5383 * max_blksz, and the file's block size is also max_blksz.
5386 if (size < blksz || zp->z_blksz != blksz) {
5388 return (SET_ERROR(EINVAL));
5391 * Caller requests buffers for write before knowing where the
5392 * write offset might be (e.g. NFS TCP write).
5397 preamble = P2PHASE(offset, blksz);
5399 preamble = blksz - preamble;
5404 postamble = P2PHASE(size, blksz);
5407 fullblk = size / blksz;
5408 (void) dmu_xuio_init(xuio,
5409 (preamble != 0) + fullblk + (postamble != 0));
5410 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5411 int, postamble, int,
5412 (preamble != 0) + fullblk + (postamble != 0));
5415 * Have to fix iov base/len for partial buffers. They
5416 * currently represent full arc_buf's.
5419 /* data begins in the middle of the arc_buf */
5420 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5423 (void) dmu_xuio_add(xuio, abuf,
5424 blksz - preamble, preamble);
5427 for (i = 0; i < fullblk; i++) {
5428 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5431 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5435 /* data ends in the middle of the arc_buf */
5436 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5439 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5444 * Loan out an arc_buf for read if the read size is larger than
5445 * the current file block size. Block alignment is not
5446 * considered. Partial arc_buf will be loaned out for read.
5448 blksz = zp->z_blksz;
5449 if (blksz < zcr_blksz_min)
5450 blksz = zcr_blksz_min;
5451 if (blksz > zcr_blksz_max)
5452 blksz = zcr_blksz_max;
5453 /* avoid potential complexity of dealing with it */
5454 if (blksz > max_blksz) {
5456 return (SET_ERROR(EINVAL));
5459 maxsize = zp->z_size - uio->uio_loffset;
5463 if (size < blksz || vn_has_cached_data(vp)) {
5465 return (SET_ERROR(EINVAL));
5470 return (SET_ERROR(EINVAL));
5473 uio->uio_extflg = UIO_XUIO;
5474 XUIO_XUZC_RW(xuio) = ioflag;
5481 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5485 int ioflag = XUIO_XUZC_RW(xuio);
5487 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5489 i = dmu_xuio_cnt(xuio);
5491 abuf = dmu_xuio_arcbuf(xuio, i);
5493 * if abuf == NULL, it must be a write buffer
5494 * that has been returned in zfs_write().
5497 dmu_return_arcbuf(abuf);
5498 ASSERT(abuf || ioflag == UIO_WRITE);
5501 dmu_xuio_fini(xuio);
5506 * Predeclare these here so that the compiler assumes that
5507 * this is an "old style" function declaration that does
5508 * not include arguments => we won't get type mismatch errors
5509 * in the initializations that follow.
5511 static int zfs_inval();
5512 static int zfs_isdir();
5517 return (SET_ERROR(EINVAL));
5523 return (SET_ERROR(EISDIR));
5526 * Directory vnode operations template
5528 vnodeops_t *zfs_dvnodeops;
5529 const fs_operation_def_t zfs_dvnodeops_template[] = {
5530 VOPNAME_OPEN, { .vop_open = zfs_open },
5531 VOPNAME_CLOSE, { .vop_close = zfs_close },
5532 VOPNAME_READ, { .error = zfs_isdir },
5533 VOPNAME_WRITE, { .error = zfs_isdir },
5534 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5535 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5536 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5537 VOPNAME_ACCESS, { .vop_access = zfs_access },
5538 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5539 VOPNAME_CREATE, { .vop_create = zfs_create },
5540 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5541 VOPNAME_LINK, { .vop_link = zfs_link },
5542 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5543 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir },
5544 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5545 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5546 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink },
5547 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5548 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5549 VOPNAME_FID, { .vop_fid = zfs_fid },
5550 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5551 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5552 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5553 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5554 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5559 * Regular file vnode operations template
5561 vnodeops_t *zfs_fvnodeops;
5562 const fs_operation_def_t zfs_fvnodeops_template[] = {
5563 VOPNAME_OPEN, { .vop_open = zfs_open },
5564 VOPNAME_CLOSE, { .vop_close = zfs_close },
5565 VOPNAME_READ, { .vop_read = zfs_read },
5566 VOPNAME_WRITE, { .vop_write = zfs_write },
5567 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5568 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5569 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5570 VOPNAME_ACCESS, { .vop_access = zfs_access },
5571 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5572 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5573 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5574 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5575 VOPNAME_FID, { .vop_fid = zfs_fid },
5576 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5577 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock },
5578 VOPNAME_SPACE, { .vop_space = zfs_space },
5579 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage },
5580 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage },
5581 VOPNAME_MAP, { .vop_map = zfs_map },
5582 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap },
5583 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap },
5584 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5585 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5586 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5587 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5588 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf },
5589 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf },
5594 * Symbolic link vnode operations template
5596 vnodeops_t *zfs_symvnodeops;
5597 const fs_operation_def_t zfs_symvnodeops_template[] = {
5598 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5599 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5600 VOPNAME_ACCESS, { .vop_access = zfs_access },
5601 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5602 VOPNAME_READLINK, { .vop_readlink = zfs_readlink },
5603 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5604 VOPNAME_FID, { .vop_fid = zfs_fid },
5605 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5606 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5611 * special share hidden files vnode operations template
5613 vnodeops_t *zfs_sharevnodeops;
5614 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5615 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5616 VOPNAME_ACCESS, { .vop_access = zfs_access },
5617 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5618 VOPNAME_FID, { .vop_fid = zfs_fid },
5619 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5620 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5621 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5622 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5627 * Extended attribute directory vnode operations template
5629 * This template is identical to the directory vnodes
5630 * operation template except for restricted operations:
5634 * Note that there are other restrictions embedded in:
5635 * zfs_create() - restrict type to VREG
5636 * zfs_link() - no links into/out of attribute space
5637 * zfs_rename() - no moves into/out of attribute space
5639 vnodeops_t *zfs_xdvnodeops;
5640 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5641 VOPNAME_OPEN, { .vop_open = zfs_open },
5642 VOPNAME_CLOSE, { .vop_close = zfs_close },
5643 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5644 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5645 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5646 VOPNAME_ACCESS, { .vop_access = zfs_access },
5647 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5648 VOPNAME_CREATE, { .vop_create = zfs_create },
5649 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5650 VOPNAME_LINK, { .vop_link = zfs_link },
5651 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5652 VOPNAME_MKDIR, { .error = zfs_inval },
5653 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5654 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5655 VOPNAME_SYMLINK, { .error = zfs_inval },
5656 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5657 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5658 VOPNAME_FID, { .vop_fid = zfs_fid },
5659 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5660 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5661 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5662 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5663 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5668 * Error vnode operations template
5670 vnodeops_t *zfs_evnodeops;
5671 const fs_operation_def_t zfs_evnodeops_template[] = {
5672 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5673 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5679 ioflags(int ioflags)
5683 if (ioflags & IO_APPEND)
5685 if (ioflags & IO_NDELAY)
5687 if (ioflags & IO_SYNC)
5688 flags |= (FSYNC | FDSYNC | FRSYNC);
5694 zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage)
5696 znode_t *zp = VTOZ(vp);
5697 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5698 objset_t *os = zp->z_zfsvfs->z_os;
5699 vm_page_t mfirst, mlast, mreq;
5703 off_t startoff, endoff;
5705 vm_pindex_t reqstart, reqend;
5706 int pcount, lsize, reqsize, size;
5711 pcount = OFF_TO_IDX(round_page(count));
5713 object = mreq->object;
5716 KASSERT(vp->v_object == object, ("mismatching object"));
5718 if (pcount > 1 && zp->z_blksz > PAGESIZE) {
5719 startoff = rounddown(IDX_TO_OFF(mreq->pindex), zp->z_blksz);
5720 reqstart = OFF_TO_IDX(round_page(startoff));
5721 if (reqstart < m[0]->pindex)
5724 reqstart = reqstart - m[0]->pindex;
5725 endoff = roundup(IDX_TO_OFF(mreq->pindex) + PAGE_SIZE,
5727 reqend = OFF_TO_IDX(trunc_page(endoff)) - 1;
5728 if (reqend > m[pcount - 1]->pindex)
5729 reqend = m[pcount - 1]->pindex;
5730 reqsize = reqend - m[reqstart]->pindex + 1;
5731 KASSERT(reqstart <= reqpage && reqpage < reqstart + reqsize,
5732 ("reqpage beyond [reqstart, reqstart + reqsize[ bounds"));
5737 mfirst = m[reqstart];
5738 mlast = m[reqstart + reqsize - 1];
5740 VM_OBJECT_LOCK(object);
5742 for (i = 0; i < reqstart; i++) {
5745 vm_page_unlock(m[i]);
5747 for (i = reqstart + reqsize; i < pcount; i++) {
5750 vm_page_unlock(m[i]);
5753 if (mreq->valid && reqsize == 1) {
5754 if (mreq->valid != VM_PAGE_BITS_ALL)
5755 vm_page_zero_invalid(mreq, TRUE);
5756 VM_OBJECT_UNLOCK(object);
5758 return (VM_PAGER_OK);
5761 PCPU_INC(cnt.v_vnodein);
5762 PCPU_ADD(cnt.v_vnodepgsin, reqsize);
5764 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) {
5765 for (i = reqstart; i < reqstart + reqsize; i++) {
5769 vm_page_unlock(m[i]);
5772 VM_OBJECT_UNLOCK(object);
5774 return (VM_PAGER_BAD);
5778 if (IDX_TO_OFF(mlast->pindex) + lsize > object->un_pager.vnp.vnp_size)
5779 lsize = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mlast->pindex);
5781 VM_OBJECT_UNLOCK(object);
5783 for (i = reqstart; i < reqstart + reqsize; i++) {
5785 if (i == (reqstart + reqsize - 1))
5787 va = zfs_map_page(m[i], &sf);
5788 error = dmu_read(os, zp->z_id, IDX_TO_OFF(m[i]->pindex),
5789 size, va, DMU_READ_PREFETCH);
5790 if (size != PAGE_SIZE)
5791 bzero(va + size, PAGE_SIZE - size);
5797 VM_OBJECT_LOCK(object);
5799 for (i = reqstart; i < reqstart + reqsize; i++) {
5801 m[i]->valid = VM_PAGE_BITS_ALL;
5802 KASSERT(m[i]->dirty == 0, ("zfs_getpages: page %p is dirty", m[i]));
5804 vm_page_readahead_finish(m[i]);
5807 VM_OBJECT_UNLOCK(object);
5809 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
5811 return (error ? VM_PAGER_ERROR : VM_PAGER_OK);
5815 zfs_freebsd_getpages(ap)
5816 struct vop_getpages_args /* {
5821 vm_ooffset_t a_offset;
5825 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage));
5829 zfs_putpages(struct vnode *vp, vm_page_t *ma, size_t len, int flags,
5832 znode_t *zp = VTOZ(vp);
5833 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5842 vm_ooffset_t lo_off;
5853 object = vp->v_object;
5857 KASSERT(ma[0]->object == object, ("mismatching object"));
5858 KASSERT(len > 0 && (len & PAGE_MASK) == 0, ("unexpected length"));
5860 for (i = 0; i < pcount; i++)
5861 rtvals[i] = VM_PAGER_ERROR;
5863 off = IDX_TO_OFF(ma[0]->pindex);
5864 blksz = zp->z_blksz;
5865 lo_off = rounddown(off, blksz);
5866 lo_len = roundup(len + (off - lo_off), blksz);
5867 rl = zfs_range_lock(zp, lo_off, lo_len, RL_WRITER);
5869 VM_OBJECT_LOCK(object);
5870 if (len + off > object->un_pager.vnp.vnp_size) {
5871 if (object->un_pager.vnp.vnp_size > off) {
5874 len = object->un_pager.vnp.vnp_size - off;
5876 if ((pgoff = (int)len & PAGE_MASK) != 0) {
5878 * If the object is locked and the following
5879 * conditions hold, then the page's dirty
5880 * field cannot be concurrently changed by a
5884 KASSERT(m->busy > 0,
5885 ("zfs_putpages: page %p is not busy", m));
5886 KASSERT(!pmap_page_is_write_mapped(m),
5887 ("zfs_putpages: page %p is not read-only", m));
5888 vm_page_clear_dirty(m, pgoff, PAGE_SIZE -
5895 if (ncount < pcount) {
5896 for (i = ncount; i < pcount; i++) {
5897 rtvals[i] = VM_PAGER_BAD;
5901 VM_OBJECT_UNLOCK(object);
5906 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
5907 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
5912 tx = dmu_tx_create(zfsvfs->z_os);
5913 dmu_tx_hold_write(tx, zp->z_id, off, len);
5915 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
5916 zfs_sa_upgrade_txholds(tx, zp);
5917 err = dmu_tx_assign(tx, TXG_NOWAIT);
5919 if (err == ERESTART) {
5928 if (zp->z_blksz < PAGE_SIZE) {
5930 for (i = 0; len > 0; off += tocopy, len -= tocopy, i++) {
5931 tocopy = len > PAGE_SIZE ? PAGE_SIZE : len;
5932 va = zfs_map_page(ma[i], &sf);
5933 dmu_write(zfsvfs->z_os, zp->z_id, off, tocopy, va, tx);
5937 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, ma, tx);
5941 uint64_t mtime[2], ctime[2];
5942 sa_bulk_attr_t bulk[3];
5945 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
5947 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
5949 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
5951 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
5953 (void)sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
5954 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
5956 VM_OBJECT_LOCK(object);
5957 for (i = 0; i < ncount; i++) {
5958 rtvals[i] = VM_PAGER_OK;
5959 vm_page_undirty(ma[i]);
5961 VM_OBJECT_UNLOCK(object);
5962 PCPU_INC(cnt.v_vnodeout);
5963 PCPU_ADD(cnt.v_vnodepgsout, ncount);
5968 zfs_range_unlock(rl);
5969 if ((flags & (VM_PAGER_PUT_SYNC | VM_PAGER_PUT_INVAL)) != 0 ||
5970 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5971 zil_commit(zfsvfs->z_log, zp->z_id);
5977 zfs_freebsd_putpages(ap)
5978 struct vop_putpages_args /* {
5984 vm_ooffset_t a_offset;
5988 return (zfs_putpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_sync,
5993 zfs_freebsd_bmap(ap)
5994 struct vop_bmap_args /* {
5997 struct bufobj **a_bop;
6004 if (ap->a_bop != NULL)
6005 *ap->a_bop = &ap->a_vp->v_bufobj;
6006 if (ap->a_bnp != NULL)
6007 *ap->a_bnp = ap->a_bn;
6008 if (ap->a_runp != NULL)
6010 if (ap->a_runb != NULL)
6017 zfs_freebsd_open(ap)
6018 struct vop_open_args /* {
6021 struct ucred *a_cred;
6022 struct thread *a_td;
6025 vnode_t *vp = ap->a_vp;
6026 znode_t *zp = VTOZ(vp);
6029 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL);
6031 vnode_create_vobject(vp, zp->z_size, ap->a_td);
6036 zfs_freebsd_close(ap)
6037 struct vop_close_args /* {
6040 struct ucred *a_cred;
6041 struct thread *a_td;
6045 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL));
6049 zfs_freebsd_ioctl(ap)
6050 struct vop_ioctl_args /* {
6060 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data,
6061 ap->a_fflag, ap->a_cred, NULL, NULL));
6065 zfs_freebsd_read(ap)
6066 struct vop_read_args /* {
6070 struct ucred *a_cred;
6074 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
6079 zfs_freebsd_write(ap)
6080 struct vop_write_args /* {
6084 struct ucred *a_cred;
6088 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
6093 zfs_freebsd_access(ap)
6094 struct vop_access_args /* {
6096 accmode_t a_accmode;
6097 struct ucred *a_cred;
6098 struct thread *a_td;
6101 vnode_t *vp = ap->a_vp;
6102 znode_t *zp = VTOZ(vp);
6107 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND,
6109 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND);
6111 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL);
6114 * VADMIN has to be handled by vaccess().
6117 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND);
6119 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
6120 zp->z_gid, accmode, ap->a_cred, NULL);
6125 * For VEXEC, ensure that at least one execute bit is set for
6128 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR &&
6129 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
6137 zfs_freebsd_lookup(ap)
6138 struct vop_lookup_args /* {
6139 struct vnode *a_dvp;
6140 struct vnode **a_vpp;
6141 struct componentname *a_cnp;
6144 struct componentname *cnp = ap->a_cnp;
6145 char nm[NAME_MAX + 1];
6147 ASSERT(cnp->cn_namelen < sizeof(nm));
6148 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm)));
6150 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop,
6151 cnp->cn_cred, cnp->cn_thread, 0));
6155 zfs_freebsd_create(ap)
6156 struct vop_create_args /* {
6157 struct vnode *a_dvp;
6158 struct vnode **a_vpp;
6159 struct componentname *a_cnp;
6160 struct vattr *a_vap;
6163 struct componentname *cnp = ap->a_cnp;
6164 vattr_t *vap = ap->a_vap;
6167 ASSERT(cnp->cn_flags & SAVENAME);
6169 vattr_init_mask(vap);
6170 mode = vap->va_mode & ALLPERMS;
6172 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode,
6173 ap->a_vpp, cnp->cn_cred, cnp->cn_thread));
6177 zfs_freebsd_remove(ap)
6178 struct vop_remove_args /* {
6179 struct vnode *a_dvp;
6181 struct componentname *a_cnp;
6185 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6187 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr,
6188 ap->a_cnp->cn_cred, NULL, 0));
6192 zfs_freebsd_mkdir(ap)
6193 struct vop_mkdir_args /* {
6194 struct vnode *a_dvp;
6195 struct vnode **a_vpp;
6196 struct componentname *a_cnp;
6197 struct vattr *a_vap;
6200 vattr_t *vap = ap->a_vap;
6202 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6204 vattr_init_mask(vap);
6206 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp,
6207 ap->a_cnp->cn_cred, NULL, 0, NULL));
6211 zfs_freebsd_rmdir(ap)
6212 struct vop_rmdir_args /* {
6213 struct vnode *a_dvp;
6215 struct componentname *a_cnp;
6218 struct componentname *cnp = ap->a_cnp;
6220 ASSERT(cnp->cn_flags & SAVENAME);
6222 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0));
6226 zfs_freebsd_readdir(ap)
6227 struct vop_readdir_args /* {
6230 struct ucred *a_cred;
6237 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag,
6238 ap->a_ncookies, ap->a_cookies));
6242 zfs_freebsd_fsync(ap)
6243 struct vop_fsync_args /* {
6246 struct thread *a_td;
6251 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL));
6255 zfs_freebsd_getattr(ap)
6256 struct vop_getattr_args /* {
6258 struct vattr *a_vap;
6259 struct ucred *a_cred;
6262 vattr_t *vap = ap->a_vap;
6268 xvap.xva_vattr = *vap;
6269 xvap.xva_vattr.va_mask |= AT_XVATTR;
6271 /* Convert chflags into ZFS-type flags. */
6272 /* XXX: what about SF_SETTABLE?. */
6273 XVA_SET_REQ(&xvap, XAT_IMMUTABLE);
6274 XVA_SET_REQ(&xvap, XAT_APPENDONLY);
6275 XVA_SET_REQ(&xvap, XAT_NOUNLINK);
6276 XVA_SET_REQ(&xvap, XAT_NODUMP);
6277 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL);
6281 /* Convert ZFS xattr into chflags. */
6282 #define FLAG_CHECK(fflag, xflag, xfield) do { \
6283 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \
6284 fflags |= (fflag); \
6286 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE,
6287 xvap.xva_xoptattrs.xoa_immutable);
6288 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY,
6289 xvap.xva_xoptattrs.xoa_appendonly);
6290 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK,
6291 xvap.xva_xoptattrs.xoa_nounlink);
6292 FLAG_CHECK(UF_NODUMP, XAT_NODUMP,
6293 xvap.xva_xoptattrs.xoa_nodump);
6295 *vap = xvap.xva_vattr;
6296 vap->va_flags = fflags;
6301 zfs_freebsd_setattr(ap)
6302 struct vop_setattr_args /* {
6304 struct vattr *a_vap;
6305 struct ucred *a_cred;
6308 vnode_t *vp = ap->a_vp;
6309 vattr_t *vap = ap->a_vap;
6310 cred_t *cred = ap->a_cred;
6315 vattr_init_mask(vap);
6316 vap->va_mask &= ~AT_NOSET;
6319 xvap.xva_vattr = *vap;
6321 zflags = VTOZ(vp)->z_pflags;
6323 if (vap->va_flags != VNOVAL) {
6324 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs;
6327 if (zfsvfs->z_use_fuids == B_FALSE)
6328 return (EOPNOTSUPP);
6330 fflags = vap->va_flags;
6331 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_NODUMP)) != 0)
6332 return (EOPNOTSUPP);
6334 * Unprivileged processes are not permitted to unset system
6335 * flags, or modify flags if any system flags are set.
6336 * Privileged non-jail processes may not modify system flags
6337 * if securelevel > 0 and any existing system flags are set.
6338 * Privileged jail processes behave like privileged non-jail
6339 * processes if the security.jail.chflags_allowed sysctl is
6340 * is non-zero; otherwise, they behave like unprivileged
6343 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 ||
6344 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) {
6346 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6347 error = securelevel_gt(cred, 0);
6353 * Callers may only modify the file flags on objects they
6354 * have VADMIN rights for.
6356 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0)
6359 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6363 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) {
6368 #define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \
6369 if (((fflags & (fflag)) && !(zflags & (zflag))) || \
6370 ((zflags & (zflag)) && !(fflags & (fflag)))) { \
6371 XVA_SET_REQ(&xvap, (xflag)); \
6372 (xfield) = ((fflags & (fflag)) != 0); \
6375 /* Convert chflags into ZFS-type flags. */
6376 /* XXX: what about SF_SETTABLE?. */
6377 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE,
6378 xvap.xva_xoptattrs.xoa_immutable);
6379 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY,
6380 xvap.xva_xoptattrs.xoa_appendonly);
6381 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK,
6382 xvap.xva_xoptattrs.xoa_nounlink);
6383 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP,
6384 xvap.xva_xoptattrs.xoa_nodump);
6387 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL));
6391 zfs_freebsd_rename(ap)
6392 struct vop_rename_args /* {
6393 struct vnode *a_fdvp;
6394 struct vnode *a_fvp;
6395 struct componentname *a_fcnp;
6396 struct vnode *a_tdvp;
6397 struct vnode *a_tvp;
6398 struct componentname *a_tcnp;
6401 vnode_t *fdvp = ap->a_fdvp;
6402 vnode_t *fvp = ap->a_fvp;
6403 vnode_t *tdvp = ap->a_tdvp;
6404 vnode_t *tvp = ap->a_tvp;
6407 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART));
6408 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART));
6410 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp,
6411 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0);
6426 zfs_freebsd_symlink(ap)
6427 struct vop_symlink_args /* {
6428 struct vnode *a_dvp;
6429 struct vnode **a_vpp;
6430 struct componentname *a_cnp;
6431 struct vattr *a_vap;
6435 struct componentname *cnp = ap->a_cnp;
6436 vattr_t *vap = ap->a_vap;
6438 ASSERT(cnp->cn_flags & SAVENAME);
6440 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */
6441 vattr_init_mask(vap);
6443 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap,
6444 ap->a_target, cnp->cn_cred, cnp->cn_thread));
6448 zfs_freebsd_readlink(ap)
6449 struct vop_readlink_args /* {
6452 struct ucred *a_cred;
6456 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL));
6460 zfs_freebsd_link(ap)
6461 struct vop_link_args /* {
6462 struct vnode *a_tdvp;
6464 struct componentname *a_cnp;
6467 struct componentname *cnp = ap->a_cnp;
6469 ASSERT(cnp->cn_flags & SAVENAME);
6471 return (zfs_link(ap->a_tdvp, ap->a_vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0));
6475 zfs_freebsd_inactive(ap)
6476 struct vop_inactive_args /* {
6478 struct thread *a_td;
6481 vnode_t *vp = ap->a_vp;
6483 zfs_inactive(vp, ap->a_td->td_ucred, NULL);
6488 zfs_freebsd_reclaim(ap)
6489 struct vop_reclaim_args /* {
6491 struct thread *a_td;
6494 vnode_t *vp = ap->a_vp;
6495 znode_t *zp = VTOZ(vp);
6496 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
6500 /* Destroy the vm object and flush associated pages. */
6501 vnode_destroy_vobject(vp);
6504 * z_teardown_inactive_lock protects from a race with
6505 * zfs_znode_dmu_fini in zfsvfs_teardown during
6508 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
6509 if (zp->z_sa_hdl == NULL)
6513 rw_exit(&zfsvfs->z_teardown_inactive_lock);
6521 struct vop_fid_args /* {
6527 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL));
6531 zfs_freebsd_pathconf(ap)
6532 struct vop_pathconf_args /* {
6535 register_t *a_retval;
6541 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL);
6543 *ap->a_retval = val;
6544 else if (error == EOPNOTSUPP)
6545 error = vop_stdpathconf(ap);
6550 zfs_freebsd_fifo_pathconf(ap)
6551 struct vop_pathconf_args /* {
6554 register_t *a_retval;
6558 switch (ap->a_name) {
6559 case _PC_ACL_EXTENDED:
6561 case _PC_ACL_PATH_MAX:
6562 case _PC_MAC_PRESENT:
6563 return (zfs_freebsd_pathconf(ap));
6565 return (fifo_specops.vop_pathconf(ap));
6570 * FreeBSD's extended attributes namespace defines file name prefix for ZFS'
6571 * extended attribute name:
6574 * system freebsd:system:
6575 * user (none, can be used to access ZFS fsattr(5) attributes
6576 * created on Solaris)
6579 zfs_create_attrname(int attrnamespace, const char *name, char *attrname,
6582 const char *namespace, *prefix, *suffix;
6584 /* We don't allow '/' character in attribute name. */
6585 if (strchr(name, '/') != NULL)
6587 /* We don't allow attribute names that start with "freebsd:" string. */
6588 if (strncmp(name, "freebsd:", 8) == 0)
6591 bzero(attrname, size);
6593 switch (attrnamespace) {
6594 case EXTATTR_NAMESPACE_USER:
6596 prefix = "freebsd:";
6597 namespace = EXTATTR_NAMESPACE_USER_STRING;
6601 * This is the default namespace by which we can access all
6602 * attributes created on Solaris.
6604 prefix = namespace = suffix = "";
6607 case EXTATTR_NAMESPACE_SYSTEM:
6608 prefix = "freebsd:";
6609 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING;
6612 case EXTATTR_NAMESPACE_EMPTY:
6616 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix,
6618 return (ENAMETOOLONG);
6624 * Vnode operating to retrieve a named extended attribute.
6627 zfs_getextattr(struct vop_getextattr_args *ap)
6630 IN struct vnode *a_vp;
6631 IN int a_attrnamespace;
6632 IN const char *a_name;
6633 INOUT struct uio *a_uio;
6635 IN struct ucred *a_cred;
6636 IN struct thread *a_td;
6640 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6641 struct thread *td = ap->a_td;
6642 struct nameidata nd;
6645 vnode_t *xvp = NULL, *vp;
6648 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6649 ap->a_cred, ap->a_td, VREAD);
6653 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6660 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6668 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_SYSSPACE, attrname,
6670 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL);
6672 NDFREE(&nd, NDF_ONLY_PNBUF);
6675 if (error == ENOENT)
6680 if (ap->a_size != NULL) {
6681 error = VOP_GETATTR(vp, &va, ap->a_cred);
6683 *ap->a_size = (size_t)va.va_size;
6684 } else if (ap->a_uio != NULL)
6685 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6688 vn_close(vp, flags, ap->a_cred, td);
6695 * Vnode operation to remove a named attribute.
6698 zfs_deleteextattr(struct vop_deleteextattr_args *ap)
6701 IN struct vnode *a_vp;
6702 IN int a_attrnamespace;
6703 IN const char *a_name;
6704 IN struct ucred *a_cred;
6705 IN struct thread *a_td;
6709 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6710 struct thread *td = ap->a_td;
6711 struct nameidata nd;
6714 vnode_t *xvp = NULL, *vp;
6717 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6718 ap->a_cred, ap->a_td, VWRITE);
6722 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6729 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6736 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF | MPSAFE,
6737 UIO_SYSSPACE, attrname, xvp, td);
6742 NDFREE(&nd, NDF_ONLY_PNBUF);
6743 if (error == ENOENT)
6748 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
6749 NDFREE(&nd, NDF_ONLY_PNBUF);
6752 if (vp == nd.ni_dvp)
6762 * Vnode operation to set a named attribute.
6765 zfs_setextattr(struct vop_setextattr_args *ap)
6768 IN struct vnode *a_vp;
6769 IN int a_attrnamespace;
6770 IN const char *a_name;
6771 INOUT struct uio *a_uio;
6772 IN struct ucred *a_cred;
6773 IN struct thread *a_td;
6777 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6778 struct thread *td = ap->a_td;
6779 struct nameidata nd;
6782 vnode_t *xvp = NULL, *vp;
6785 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6786 ap->a_cred, ap->a_td, VWRITE);
6790 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6797 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6798 LOOKUP_XATTR | CREATE_XATTR_DIR);
6804 flags = FFLAGS(O_WRONLY | O_CREAT);
6805 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_SYSSPACE, attrname,
6807 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL);
6809 NDFREE(&nd, NDF_ONLY_PNBUF);
6817 error = VOP_SETATTR(vp, &va, ap->a_cred);
6819 VOP_WRITE(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6822 vn_close(vp, flags, ap->a_cred, td);
6829 * Vnode operation to retrieve extended attributes on a vnode.
6832 zfs_listextattr(struct vop_listextattr_args *ap)
6835 IN struct vnode *a_vp;
6836 IN int a_attrnamespace;
6837 INOUT struct uio *a_uio;
6839 IN struct ucred *a_cred;
6840 IN struct thread *a_td;
6844 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6845 struct thread *td = ap->a_td;
6846 struct nameidata nd;
6847 char attrprefix[16];
6848 u_char dirbuf[sizeof(struct dirent)];
6851 struct uio auio, *uio = ap->a_uio;
6852 size_t *sizep = ap->a_size;
6854 vnode_t *xvp = NULL, *vp;
6855 int done, error, eof, pos;
6857 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6858 ap->a_cred, ap->a_td, VREAD);
6862 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix,
6863 sizeof(attrprefix));
6866 plen = strlen(attrprefix);
6873 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6878 * ENOATTR means that the EA directory does not yet exist,
6879 * i.e. there are no extended attributes there.
6881 if (error == ENOATTR)
6886 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED | MPSAFE,
6887 UIO_SYSSPACE, ".", xvp, td);
6890 NDFREE(&nd, NDF_ONLY_PNBUF);
6896 auio.uio_iov = &aiov;
6897 auio.uio_iovcnt = 1;
6898 auio.uio_segflg = UIO_SYSSPACE;
6900 auio.uio_rw = UIO_READ;
6901 auio.uio_offset = 0;
6906 aiov.iov_base = (void *)dirbuf;
6907 aiov.iov_len = sizeof(dirbuf);
6908 auio.uio_resid = sizeof(dirbuf);
6909 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL);
6910 done = sizeof(dirbuf) - auio.uio_resid;
6913 for (pos = 0; pos < done;) {
6914 dp = (struct dirent *)(dirbuf + pos);
6915 pos += dp->d_reclen;
6917 * XXX: Temporarily we also accept DT_UNKNOWN, as this
6918 * is what we get when attribute was created on Solaris.
6920 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN)
6922 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0)
6924 else if (strncmp(dp->d_name, attrprefix, plen) != 0)
6926 nlen = dp->d_namlen - plen;
6929 else if (uio != NULL) {
6931 * Format of extattr name entry is one byte for
6932 * length and the rest for name.
6934 error = uiomove(&nlen, 1, uio->uio_rw, uio);
6936 error = uiomove(dp->d_name + plen, nlen,
6943 } while (!eof && error == 0);
6952 zfs_freebsd_getacl(ap)
6953 struct vop_getacl_args /* {
6962 vsecattr_t vsecattr;
6964 if (ap->a_type != ACL_TYPE_NFS4)
6967 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
6968 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL))
6971 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt);
6972 if (vsecattr.vsa_aclentp != NULL)
6973 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz);
6979 zfs_freebsd_setacl(ap)
6980 struct vop_setacl_args /* {
6989 vsecattr_t vsecattr;
6990 int aclbsize; /* size of acl list in bytes */
6993 if (ap->a_type != ACL_TYPE_NFS4)
6996 if (ap->a_aclp == NULL)
6999 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES)
7003 * With NFSv4 ACLs, chmod(2) may need to add additional entries,
7004 * splitting every entry into two and appending "canonical six"
7005 * entries at the end. Don't allow for setting an ACL that would
7006 * cause chmod(2) to run out of ACL entries.
7008 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES)
7011 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR);
7015 vsecattr.vsa_mask = VSA_ACE;
7016 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t);
7017 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP);
7018 aaclp = vsecattr.vsa_aclentp;
7019 vsecattr.vsa_aclentsz = aclbsize;
7021 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp);
7022 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL);
7023 kmem_free(aaclp, aclbsize);
7029 zfs_freebsd_aclcheck(ap)
7030 struct vop_aclcheck_args /* {
7039 return (EOPNOTSUPP);
7043 zfs_vptocnp(struct vop_vptocnp_args *ap)
7045 vnode_t *covered_vp;
7046 vnode_t *vp = ap->a_vp;;
7047 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
7048 znode_t *zp = VTOZ(vp);
7057 * If we are a snapshot mounted under .zfs, run the operation
7058 * on the covered vnode.
7060 if ((error = sa_lookup(zp->z_sa_hdl,
7061 SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0) {
7066 if (zp->z_id != parent || zfsvfs->z_parent == zfsvfs) {
7068 return (vop_stdvptocnp(ap));
7072 covered_vp = vp->v_mount->mnt_vnodecovered;
7074 ltype = VOP_ISLOCKED(vp);
7076 error = vget(covered_vp, LK_EXCLUSIVE, curthread);
7079 error = VOP_VPTOCNP(covered_vp, ap->a_vpp, ap->a_cred,
7080 ap->a_buf, ap->a_buflen);
7083 vn_lock(vp, ltype | LK_RETRY);
7084 if ((vp->v_iflag & VI_DOOMED) != 0)
7085 error = SET_ERROR(ENOENT);
7089 struct vop_vector zfs_vnodeops;
7090 struct vop_vector zfs_fifoops;
7091 struct vop_vector zfs_shareops;
7093 struct vop_vector zfs_vnodeops = {
7094 .vop_default = &default_vnodeops,
7095 .vop_inactive = zfs_freebsd_inactive,
7096 .vop_reclaim = zfs_freebsd_reclaim,
7097 .vop_access = zfs_freebsd_access,
7098 #ifdef FREEBSD_NAMECACHE
7099 .vop_lookup = vfs_cache_lookup,
7100 .vop_cachedlookup = zfs_freebsd_lookup,
7102 .vop_lookup = zfs_freebsd_lookup,
7104 .vop_getattr = zfs_freebsd_getattr,
7105 .vop_setattr = zfs_freebsd_setattr,
7106 .vop_create = zfs_freebsd_create,
7107 .vop_mknod = zfs_freebsd_create,
7108 .vop_mkdir = zfs_freebsd_mkdir,
7109 .vop_readdir = zfs_freebsd_readdir,
7110 .vop_fsync = zfs_freebsd_fsync,
7111 .vop_open = zfs_freebsd_open,
7112 .vop_close = zfs_freebsd_close,
7113 .vop_rmdir = zfs_freebsd_rmdir,
7114 .vop_ioctl = zfs_freebsd_ioctl,
7115 .vop_link = zfs_freebsd_link,
7116 .vop_symlink = zfs_freebsd_symlink,
7117 .vop_readlink = zfs_freebsd_readlink,
7118 .vop_read = zfs_freebsd_read,
7119 .vop_write = zfs_freebsd_write,
7120 .vop_remove = zfs_freebsd_remove,
7121 .vop_rename = zfs_freebsd_rename,
7122 .vop_pathconf = zfs_freebsd_pathconf,
7123 .vop_bmap = zfs_freebsd_bmap,
7124 .vop_fid = zfs_freebsd_fid,
7125 .vop_getextattr = zfs_getextattr,
7126 .vop_deleteextattr = zfs_deleteextattr,
7127 .vop_setextattr = zfs_setextattr,
7128 .vop_listextattr = zfs_listextattr,
7129 .vop_getacl = zfs_freebsd_getacl,
7130 .vop_setacl = zfs_freebsd_setacl,
7131 .vop_aclcheck = zfs_freebsd_aclcheck,
7132 .vop_getpages = zfs_freebsd_getpages,
7133 .vop_putpages = zfs_freebsd_putpages,
7134 .vop_vptocnp = zfs_vptocnp,
7137 struct vop_vector zfs_fifoops = {
7138 .vop_default = &fifo_specops,
7139 .vop_fsync = zfs_freebsd_fsync,
7140 .vop_access = zfs_freebsd_access,
7141 .vop_getattr = zfs_freebsd_getattr,
7142 .vop_inactive = zfs_freebsd_inactive,
7143 .vop_read = VOP_PANIC,
7144 .vop_reclaim = zfs_freebsd_reclaim,
7145 .vop_setattr = zfs_freebsd_setattr,
7146 .vop_write = VOP_PANIC,
7147 .vop_pathconf = zfs_freebsd_fifo_pathconf,
7148 .vop_fid = zfs_freebsd_fid,
7149 .vop_getacl = zfs_freebsd_getacl,
7150 .vop_setacl = zfs_freebsd_setacl,
7151 .vop_aclcheck = zfs_freebsd_aclcheck,
7155 * special share hidden files vnode operations template
7157 struct vop_vector zfs_shareops = {
7158 .vop_default = &default_vnodeops,
7159 .vop_access = zfs_freebsd_access,
7160 .vop_inactive = zfs_freebsd_inactive,
7161 .vop_reclaim = zfs_freebsd_reclaim,
7162 .vop_fid = zfs_freebsd_fid,
7163 .vop_pathconf = zfs_freebsd_pathconf,