4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2013 by Delphix. All rights reserved.
24 * Copyright 2013 Nexenta Systems, Inc. All rights reserved.
27 /* Portions Copyright 2007 Jeremy Teo */
28 /* Portions Copyright 2010 Robert Milkowski */
30 #include <sys/types.h>
31 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/sysmacros.h>
35 #include <sys/resource.h>
38 #include <sys/vnode.h>
42 #include <sys/taskq.h>
44 #include <sys/atomic.h>
45 #include <sys/namei.h>
47 #include <sys/cmn_err.h>
48 #include <sys/errno.h>
49 #include <sys/unistd.h>
50 #include <sys/zfs_dir.h>
51 #include <sys/zfs_ioctl.h>
52 #include <sys/fs/zfs.h>
54 #include <sys/dmu_objset.h>
60 #include <sys/dirent.h>
61 #include <sys/policy.h>
62 #include <sys/sunddi.h>
63 #include <sys/filio.h>
65 #include <sys/zfs_ctldir.h>
66 #include <sys/zfs_fuid.h>
67 #include <sys/zfs_sa.h>
69 #include <sys/zfs_rlock.h>
70 #include <sys/extdirent.h>
71 #include <sys/kidmap.h>
74 #include <sys/sf_buf.h>
75 #include <sys/sched.h>
77 #include <vm/vm_param.h>
78 #include <vm/vm_pageout.h>
83 * Each vnode op performs some logical unit of work. To do this, the ZPL must
84 * properly lock its in-core state, create a DMU transaction, do the work,
85 * record this work in the intent log (ZIL), commit the DMU transaction,
86 * and wait for the intent log to commit if it is a synchronous operation.
87 * Moreover, the vnode ops must work in both normal and log replay context.
88 * The ordering of events is important to avoid deadlocks and references
89 * to freed memory. The example below illustrates the following Big Rules:
91 * (1) A check must be made in each zfs thread for a mounted file system.
92 * This is done avoiding races using ZFS_ENTER(zfsvfs).
93 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
94 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
95 * can return EIO from the calling function.
97 * (2) VN_RELE() should always be the last thing except for zil_commit()
98 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
99 * First, if it's the last reference, the vnode/znode
100 * can be freed, so the zp may point to freed memory. Second, the last
101 * reference will call zfs_zinactive(), which may induce a lot of work --
102 * pushing cached pages (which acquires range locks) and syncing out
103 * cached atime changes. Third, zfs_zinactive() may require a new tx,
104 * which could deadlock the system if you were already holding one.
105 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
107 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
108 * as they can span dmu_tx_assign() calls.
110 * (4) Always pass TXG_NOWAIT as the second argument to dmu_tx_assign().
111 * This is critical because we don't want to block while holding locks.
112 * Note, in particular, that if a lock is sometimes acquired before
113 * the tx assigns, and sometimes after (e.g. z_lock), then failing to
114 * use a non-blocking assign can deadlock the system. The scenario:
116 * Thread A has grabbed a lock before calling dmu_tx_assign().
117 * Thread B is in an already-assigned tx, and blocks for this lock.
118 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
119 * forever, because the previous txg can't quiesce until B's tx commits.
121 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
122 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
123 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
124 * to indicate that this operation has already called dmu_tx_wait().
125 * This will ensure that we don't retry forever, waiting a short bit
128 * (5) If the operation succeeded, generate the intent log entry for it
129 * before dropping locks. This ensures that the ordering of events
130 * in the intent log matches the order in which they actually occurred.
131 * During ZIL replay the zfs_log_* functions will update the sequence
132 * number to indicate the zil transaction has replayed.
134 * (6) At the end of each vnode op, the DMU tx must always commit,
135 * regardless of whether there were any errors.
137 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
138 * to ensure that synchronous semantics are provided when necessary.
140 * In general, this is how things should be ordered in each vnode op:
142 * ZFS_ENTER(zfsvfs); // exit if unmounted
144 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD())
145 * rw_enter(...); // grab any other locks you need
146 * tx = dmu_tx_create(...); // get DMU tx
147 * dmu_tx_hold_*(); // hold each object you might modify
148 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
150 * rw_exit(...); // drop locks
151 * zfs_dirent_unlock(dl); // unlock directory entry
152 * VN_RELE(...); // release held vnodes
153 * if (error == ERESTART) {
159 * dmu_tx_abort(tx); // abort DMU tx
160 * ZFS_EXIT(zfsvfs); // finished in zfs
161 * return (error); // really out of space
163 * error = do_real_work(); // do whatever this VOP does
165 * zfs_log_*(...); // on success, make ZIL entry
166 * dmu_tx_commit(tx); // commit DMU tx -- error or not
167 * rw_exit(...); // drop locks
168 * zfs_dirent_unlock(dl); // unlock directory entry
169 * VN_RELE(...); // release held vnodes
170 * zil_commit(zilog, foid); // synchronous when necessary
171 * ZFS_EXIT(zfsvfs); // finished in zfs
172 * return (error); // done, report error
177 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
179 znode_t *zp = VTOZ(*vpp);
180 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
185 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
186 ((flag & FAPPEND) == 0)) {
188 return (SET_ERROR(EPERM));
191 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
192 ZTOV(zp)->v_type == VREG &&
193 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
194 if (fs_vscan(*vpp, cr, 0) != 0) {
196 return (SET_ERROR(EACCES));
200 /* Keep a count of the synchronous opens in the znode */
201 if (flag & (FSYNC | FDSYNC))
202 atomic_inc_32(&zp->z_sync_cnt);
210 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
211 caller_context_t *ct)
213 znode_t *zp = VTOZ(vp);
214 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
217 * Clean up any locks held by this process on the vp.
219 cleanlocks(vp, ddi_get_pid(), 0);
220 cleanshares(vp, ddi_get_pid());
225 /* Decrement the synchronous opens in the znode */
226 if ((flag & (FSYNC | FDSYNC)) && (count == 1))
227 atomic_dec_32(&zp->z_sync_cnt);
229 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
230 ZTOV(zp)->v_type == VREG &&
231 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
232 VERIFY(fs_vscan(vp, cr, 1) == 0);
239 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
240 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
243 zfs_holey(vnode_t *vp, u_long cmd, offset_t *off)
245 znode_t *zp = VTOZ(vp);
246 uint64_t noff = (uint64_t)*off; /* new offset */
251 file_sz = zp->z_size;
252 if (noff >= file_sz) {
253 return (SET_ERROR(ENXIO));
256 if (cmd == _FIO_SEEK_HOLE)
261 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
264 if ((error == ESRCH) || (noff > file_sz)) {
266 * Handle the virtual hole at the end of file.
272 return (SET_ERROR(ENXIO));
283 zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred,
284 int *rvalp, caller_context_t *ct)
296 * The following two ioctls are used by bfu. Faking out,
297 * necessary to avoid bfu errors.
306 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
307 return (SET_ERROR(EFAULT));
309 off = *(offset_t *)data;
312 zfsvfs = zp->z_zfsvfs;
316 /* offset parameter is in/out */
317 error = zfs_holey(vp, com, &off);
322 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
323 return (SET_ERROR(EFAULT));
325 *(offset_t *)data = off;
329 return (SET_ERROR(ENOTTY));
333 page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes)
340 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE
341 * aligned boundaries, if the range is not aligned. As a result a
342 * DEV_BSIZE subrange with partially dirty data may get marked as clean.
343 * It may happen that all DEV_BSIZE subranges are marked clean and thus
344 * the whole page would be considred clean despite have some dirty data.
345 * For this reason we should shrink the range to DEV_BSIZE aligned
346 * boundaries before calling vm_page_clear_dirty.
348 end = rounddown2(off + nbytes, DEV_BSIZE);
349 off = roundup2(off, DEV_BSIZE);
353 zfs_vmobject_assert_wlocked(obj);
356 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
358 if (vm_page_xbusied(pp)) {
360 * Reference the page before unlocking and
361 * sleeping so that the page daemon is less
362 * likely to reclaim it.
364 vm_page_reference(pp);
366 zfs_vmobject_wunlock(obj);
367 vm_page_busy_sleep(pp, "zfsmwb");
368 zfs_vmobject_wlock(obj);
372 } else if (pp == NULL) {
373 pp = vm_page_alloc(obj, OFF_TO_IDX(start),
374 VM_ALLOC_SYSTEM | VM_ALLOC_IFCACHED |
377 ASSERT(pp != NULL && !pp->valid);
382 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
383 vm_object_pip_add(obj, 1);
384 pmap_remove_write(pp);
386 vm_page_clear_dirty(pp, off, nbytes);
394 page_unbusy(vm_page_t pp)
398 vm_object_pip_subtract(pp->object, 1);
402 page_hold(vnode_t *vp, int64_t start)
408 zfs_vmobject_assert_wlocked(obj);
411 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
413 if (vm_page_xbusied(pp)) {
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);
421 zfs_vmobject_wunlock(obj);
422 vm_page_busy_sleep(pp, "zfsmwb");
423 zfs_vmobject_wlock(obj);
427 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
440 page_unhold(vm_page_t pp)
449 zfs_map_page(vm_page_t pp, struct sf_buf **sfp)
452 *sfp = sf_buf_alloc(pp, 0);
453 return ((caddr_t)sf_buf_kva(*sfp));
457 zfs_unmap_page(struct sf_buf *sf)
464 * When a file is memory mapped, we must keep the IO data synchronized
465 * between the DMU cache and the memory mapped pages. What this means:
467 * On Write: If we find a memory mapped page, we write to *both*
468 * the page and the dmu buffer.
471 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid,
472 int segflg, dmu_tx_t *tx)
479 ASSERT(vp->v_mount != NULL);
483 off = start & PAGEOFFSET;
484 zfs_vmobject_wlock(obj);
485 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
487 int nbytes = imin(PAGESIZE - off, len);
489 if (segflg == UIO_NOCOPY) {
490 pp = vm_page_lookup(obj, OFF_TO_IDX(start));
492 ("zfs update_pages: NULL page in putpages case"));
494 ("zfs update_pages: unaligned data in putpages case"));
495 KASSERT(pp->valid == VM_PAGE_BITS_ALL,
496 ("zfs update_pages: invalid page in putpages case"));
497 KASSERT(vm_page_sbusied(pp),
498 ("zfs update_pages: unbusy page in putpages case"));
499 KASSERT(!pmap_page_is_write_mapped(pp),
500 ("zfs update_pages: writable page in putpages case"));
501 zfs_vmobject_wunlock(obj);
503 va = zfs_map_page(pp, &sf);
504 (void) dmu_write(os, oid, start, nbytes, va, tx);
507 zfs_vmobject_wlock(obj);
509 } else if ((pp = page_busy(vp, start, off, nbytes)) != NULL) {
510 zfs_vmobject_wunlock(obj);
512 va = zfs_map_page(pp, &sf);
513 (void) dmu_read(os, oid, start+off, nbytes,
514 va+off, DMU_READ_PREFETCH);;
517 zfs_vmobject_wlock(obj);
523 if (segflg != UIO_NOCOPY)
524 vm_object_pip_wakeupn(obj, 0);
525 zfs_vmobject_wunlock(obj);
529 * Read with UIO_NOCOPY flag means that sendfile(2) requests
530 * ZFS to populate a range of page cache pages with data.
532 * NOTE: this function could be optimized to pre-allocate
533 * all pages in advance, drain exclusive busy on all of them,
534 * map them into contiguous KVA region and populate them
535 * in one single dmu_read() call.
538 mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio)
540 znode_t *zp = VTOZ(vp);
541 objset_t *os = zp->z_zfsvfs->z_os;
551 ASSERT(uio->uio_segflg == UIO_NOCOPY);
552 ASSERT(vp->v_mount != NULL);
555 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0);
557 zfs_vmobject_wlock(obj);
558 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) {
559 int bytes = MIN(PAGESIZE, len);
561 pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_SBUSY |
562 VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY);
563 if (pp->valid == 0) {
564 zfs_vmobject_wunlock(obj);
565 va = zfs_map_page(pp, &sf);
566 error = dmu_read(os, zp->z_id, start, bytes, va,
568 if (bytes != PAGESIZE && error == 0)
569 bzero(va + bytes, PAGESIZE - bytes);
571 zfs_vmobject_wlock(obj);
575 if (pp->wire_count == 0 && pp->valid == 0 &&
579 pp->valid = VM_PAGE_BITS_ALL;
580 vm_page_activate(pp);
587 uio->uio_resid -= bytes;
588 uio->uio_offset += bytes;
591 zfs_vmobject_wunlock(obj);
596 * When a file is memory mapped, we must keep the IO data synchronized
597 * between the DMU cache and the memory mapped pages. What this means:
599 * On Read: We "read" preferentially from memory mapped pages,
600 * else we default from the dmu buffer.
602 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
603 * the file is memory mapped.
606 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
608 znode_t *zp = VTOZ(vp);
609 objset_t *os = zp->z_zfsvfs->z_os;
617 ASSERT(vp->v_mount != NULL);
621 start = uio->uio_loffset;
622 off = start & PAGEOFFSET;
623 zfs_vmobject_wlock(obj);
624 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
626 uint64_t bytes = MIN(PAGESIZE - off, len);
628 if (pp = page_hold(vp, start)) {
632 zfs_vmobject_wunlock(obj);
633 va = zfs_map_page(pp, &sf);
634 error = uiomove(va + off, bytes, UIO_READ, uio);
636 zfs_vmobject_wlock(obj);
639 zfs_vmobject_wunlock(obj);
640 error = dmu_read_uio(os, zp->z_id, uio, bytes);
641 zfs_vmobject_wlock(obj);
648 zfs_vmobject_wunlock(obj);
652 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
655 * Read bytes from specified file into supplied buffer.
657 * IN: vp - vnode of file to be read from.
658 * uio - structure supplying read location, range info,
660 * ioflag - SYNC flags; used to provide FRSYNC semantics.
661 * cr - credentials of caller.
662 * ct - caller context
664 * OUT: uio - updated offset and range, buffer filled.
666 * RETURN: 0 on success, error code on failure.
669 * vp - atime updated if byte count > 0
673 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
675 znode_t *zp = VTOZ(vp);
676 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
687 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
689 return (SET_ERROR(EACCES));
693 * Validate file offset
695 if (uio->uio_loffset < (offset_t)0) {
697 return (SET_ERROR(EINVAL));
701 * Fasttrack empty reads
703 if (uio->uio_resid == 0) {
709 * Check for mandatory locks
711 if (MANDMODE(zp->z_mode)) {
712 if (error = chklock(vp, FREAD,
713 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
720 * If we're in FRSYNC mode, sync out this znode before reading it.
723 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
724 zil_commit(zfsvfs->z_log, zp->z_id);
727 * Lock the range against changes.
729 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
732 * If we are reading past end-of-file we can skip
733 * to the end; but we might still need to set atime.
735 if (uio->uio_loffset >= zp->z_size) {
740 ASSERT(uio->uio_loffset < zp->z_size);
741 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
744 if ((uio->uio_extflg == UIO_XUIO) &&
745 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
747 int blksz = zp->z_blksz;
748 uint64_t offset = uio->uio_loffset;
750 xuio = (xuio_t *)uio;
752 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
755 ASSERT(offset + n <= blksz);
758 (void) dmu_xuio_init(xuio, nblk);
760 if (vn_has_cached_data(vp)) {
762 * For simplicity, we always allocate a full buffer
763 * even if we only expect to read a portion of a block.
765 while (--nblk >= 0) {
766 (void) dmu_xuio_add(xuio,
767 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
775 nbytes = MIN(n, zfs_read_chunk_size -
776 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
779 if (uio->uio_segflg == UIO_NOCOPY)
780 error = mappedread_sf(vp, nbytes, uio);
782 #endif /* __FreeBSD__ */
783 if (vn_has_cached_data(vp))
784 error = mappedread(vp, nbytes, uio);
786 error = dmu_read_uio(os, zp->z_id, uio, nbytes);
788 /* convert checksum errors into IO errors */
790 error = SET_ERROR(EIO);
797 zfs_range_unlock(rl);
799 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
805 * Write the bytes to a file.
807 * IN: vp - vnode of file to be written to.
808 * uio - structure supplying write location, range info,
810 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is
811 * set if in append mode.
812 * cr - credentials of caller.
813 * ct - caller context (NFS/CIFS fem monitor only)
815 * OUT: uio - updated offset and range.
817 * RETURN: 0 on success, error code on failure.
820 * vp - ctime|mtime updated if byte count > 0
825 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
827 znode_t *zp = VTOZ(vp);
828 rlim64_t limit = MAXOFFSET_T;
829 ssize_t start_resid = uio->uio_resid;
833 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
838 int max_blksz = zfsvfs->z_max_blksz;
841 iovec_t *aiov = NULL;
844 int iovcnt = uio->uio_iovcnt;
845 iovec_t *iovp = uio->uio_iov;
848 sa_bulk_attr_t bulk[4];
849 uint64_t mtime[2], ctime[2];
852 * Fasttrack empty write
858 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
864 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
865 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
866 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
868 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
872 * If immutable or not appending then return EPERM
874 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
875 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
876 (uio->uio_loffset < zp->z_size))) {
878 return (SET_ERROR(EPERM));
881 zilog = zfsvfs->z_log;
884 * Validate file offset
886 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
889 return (SET_ERROR(EINVAL));
893 * Check for mandatory locks before calling zfs_range_lock()
894 * in order to prevent a deadlock with locks set via fcntl().
896 if (MANDMODE((mode_t)zp->z_mode) &&
897 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
904 * Pre-fault the pages to ensure slow (eg NFS) pages
906 * Skip this if uio contains loaned arc_buf.
908 if ((uio->uio_extflg == UIO_XUIO) &&
909 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
910 xuio = (xuio_t *)uio;
912 uio_prefaultpages(MIN(n, max_blksz), uio);
916 * If in append mode, set the io offset pointer to eof.
918 if (ioflag & FAPPEND) {
920 * Obtain an appending range lock to guarantee file append
921 * semantics. We reset the write offset once we have the lock.
923 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
925 if (rl->r_len == UINT64_MAX) {
927 * We overlocked the file because this write will cause
928 * the file block size to increase.
929 * Note that zp_size cannot change with this lock held.
933 uio->uio_loffset = woff;
936 * Note that if the file block size will change as a result of
937 * this write, then this range lock will lock the entire file
938 * so that we can re-write the block safely.
940 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
943 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) {
944 zfs_range_unlock(rl);
950 zfs_range_unlock(rl);
952 return (SET_ERROR(EFBIG));
955 if ((woff + n) > limit || woff > (limit - n))
958 /* Will this write extend the file length? */
959 write_eof = (woff + n > zp->z_size);
961 end_size = MAX(zp->z_size, woff + n);
964 * Write the file in reasonable size chunks. Each chunk is written
965 * in a separate transaction; this keeps the intent log records small
966 * and allows us to do more fine-grained space accounting.
970 woff = uio->uio_loffset;
972 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
973 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
975 dmu_return_arcbuf(abuf);
976 error = SET_ERROR(EDQUOT);
980 if (xuio && abuf == NULL) {
981 ASSERT(i_iov < iovcnt);
983 abuf = dmu_xuio_arcbuf(xuio, i_iov);
984 dmu_xuio_clear(xuio, i_iov);
985 DTRACE_PROBE3(zfs_cp_write, int, i_iov,
986 iovec_t *, aiov, arc_buf_t *, abuf);
987 ASSERT((aiov->iov_base == abuf->b_data) ||
988 ((char *)aiov->iov_base - (char *)abuf->b_data +
989 aiov->iov_len == arc_buf_size(abuf)));
991 } else if (abuf == NULL && n >= max_blksz &&
992 woff >= zp->z_size &&
993 P2PHASE(woff, max_blksz) == 0 &&
994 zp->z_blksz == max_blksz) {
996 * This write covers a full block. "Borrow" a buffer
997 * from the dmu so that we can fill it before we enter
998 * a transaction. This avoids the possibility of
999 * holding up the transaction if the data copy hangs
1000 * up on a pagefault (e.g., from an NFS server mapping).
1004 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
1006 ASSERT(abuf != NULL);
1007 ASSERT(arc_buf_size(abuf) == max_blksz);
1008 if (error = uiocopy(abuf->b_data, max_blksz,
1009 UIO_WRITE, uio, &cbytes)) {
1010 dmu_return_arcbuf(abuf);
1013 ASSERT(cbytes == max_blksz);
1017 * Start a transaction.
1019 tx = dmu_tx_create(zfsvfs->z_os);
1020 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1021 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
1022 zfs_sa_upgrade_txholds(tx, zp);
1023 error = dmu_tx_assign(tx, TXG_NOWAIT);
1025 if (error == ERESTART) {
1032 dmu_return_arcbuf(abuf);
1037 * If zfs_range_lock() over-locked we grow the blocksize
1038 * and then reduce the lock range. This will only happen
1039 * on the first iteration since zfs_range_reduce() will
1040 * shrink down r_len to the appropriate size.
1042 if (rl->r_len == UINT64_MAX) {
1045 if (zp->z_blksz > max_blksz) {
1046 ASSERT(!ISP2(zp->z_blksz));
1047 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
1049 new_blksz = MIN(end_size, max_blksz);
1051 zfs_grow_blocksize(zp, new_blksz, tx);
1052 zfs_range_reduce(rl, woff, n);
1056 * XXX - should we really limit each write to z_max_blksz?
1057 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
1059 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
1061 if (woff + nbytes > zp->z_size)
1062 vnode_pager_setsize(vp, woff + nbytes);
1065 tx_bytes = uio->uio_resid;
1066 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
1068 tx_bytes -= uio->uio_resid;
1071 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
1073 * If this is not a full block write, but we are
1074 * extending the file past EOF and this data starts
1075 * block-aligned, use assign_arcbuf(). Otherwise,
1076 * write via dmu_write().
1078 if (tx_bytes < max_blksz && (!write_eof ||
1079 aiov->iov_base != abuf->b_data)) {
1081 dmu_write(zfsvfs->z_os, zp->z_id, woff,
1082 aiov->iov_len, aiov->iov_base, tx);
1083 dmu_return_arcbuf(abuf);
1084 xuio_stat_wbuf_copied();
1086 ASSERT(xuio || tx_bytes == max_blksz);
1087 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
1090 ASSERT(tx_bytes <= uio->uio_resid);
1091 uioskip(uio, tx_bytes);
1093 if (tx_bytes && vn_has_cached_data(vp)) {
1094 update_pages(vp, woff, tx_bytes, zfsvfs->z_os,
1095 zp->z_id, uio->uio_segflg, tx);
1099 * If we made no progress, we're done. If we made even
1100 * partial progress, update the znode and ZIL accordingly.
1102 if (tx_bytes == 0) {
1103 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
1104 (void *)&zp->z_size, sizeof (uint64_t), tx);
1111 * Clear Set-UID/Set-GID bits on successful write if not
1112 * privileged and at least one of the excute bits is set.
1114 * It would be nice to to this after all writes have
1115 * been done, but that would still expose the ISUID/ISGID
1116 * to another app after the partial write is committed.
1118 * Note: we don't call zfs_fuid_map_id() here because
1119 * user 0 is not an ephemeral uid.
1121 mutex_enter(&zp->z_acl_lock);
1122 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
1123 (S_IXUSR >> 6))) != 0 &&
1124 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
1125 secpolicy_vnode_setid_retain(vp, cr,
1126 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
1128 zp->z_mode &= ~(S_ISUID | S_ISGID);
1129 newmode = zp->z_mode;
1130 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
1131 (void *)&newmode, sizeof (uint64_t), tx);
1133 mutex_exit(&zp->z_acl_lock);
1135 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
1139 * Update the file size (zp_size) if it has changed;
1140 * account for possible concurrent updates.
1142 while ((end_size = zp->z_size) < uio->uio_loffset) {
1143 (void) atomic_cas_64(&zp->z_size, end_size,
1148 * If we are replaying and eof is non zero then force
1149 * the file size to the specified eof. Note, there's no
1150 * concurrency during replay.
1152 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
1153 zp->z_size = zfsvfs->z_replay_eof;
1155 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1157 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
1162 ASSERT(tx_bytes == nbytes);
1167 uio_prefaultpages(MIN(n, max_blksz), uio);
1171 zfs_range_unlock(rl);
1174 * If we're in replay mode, or we made no progress, return error.
1175 * Otherwise, it's at least a partial write, so it's successful.
1177 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1182 if (ioflag & (FSYNC | FDSYNC) ||
1183 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1184 zil_commit(zilog, zp->z_id);
1191 zfs_get_done(zgd_t *zgd, int error)
1193 znode_t *zp = zgd->zgd_private;
1194 objset_t *os = zp->z_zfsvfs->z_os;
1197 dmu_buf_rele(zgd->zgd_db, zgd);
1199 zfs_range_unlock(zgd->zgd_rl);
1202 * Release the vnode asynchronously as we currently have the
1203 * txg stopped from syncing.
1205 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1207 if (error == 0 && zgd->zgd_bp)
1208 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1210 kmem_free(zgd, sizeof (zgd_t));
1214 static int zil_fault_io = 0;
1218 * Get data to generate a TX_WRITE intent log record.
1221 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1223 zfsvfs_t *zfsvfs = arg;
1224 objset_t *os = zfsvfs->z_os;
1226 uint64_t object = lr->lr_foid;
1227 uint64_t offset = lr->lr_offset;
1228 uint64_t size = lr->lr_length;
1229 blkptr_t *bp = &lr->lr_blkptr;
1234 ASSERT(zio != NULL);
1238 * Nothing to do if the file has been removed
1240 if (zfs_zget(zfsvfs, object, &zp) != 0)
1241 return (SET_ERROR(ENOENT));
1242 if (zp->z_unlinked) {
1244 * Release the vnode asynchronously as we currently have the
1245 * txg stopped from syncing.
1247 VN_RELE_ASYNC(ZTOV(zp),
1248 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1249 return (SET_ERROR(ENOENT));
1252 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1253 zgd->zgd_zilog = zfsvfs->z_log;
1254 zgd->zgd_private = zp;
1257 * Write records come in two flavors: immediate and indirect.
1258 * For small writes it's cheaper to store the data with the
1259 * log record (immediate); for large writes it's cheaper to
1260 * sync the data and get a pointer to it (indirect) so that
1261 * we don't have to write the data twice.
1263 if (buf != NULL) { /* immediate write */
1264 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1265 /* test for truncation needs to be done while range locked */
1266 if (offset >= zp->z_size) {
1267 error = SET_ERROR(ENOENT);
1269 error = dmu_read(os, object, offset, size, buf,
1270 DMU_READ_NO_PREFETCH);
1272 ASSERT(error == 0 || error == ENOENT);
1273 } else { /* indirect write */
1275 * Have to lock the whole block to ensure when it's
1276 * written out and it's checksum is being calculated
1277 * that no one can change the data. We need to re-check
1278 * blocksize after we get the lock in case it's changed!
1283 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1285 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1287 if (zp->z_blksz == size)
1290 zfs_range_unlock(zgd->zgd_rl);
1292 /* test for truncation needs to be done while range locked */
1293 if (lr->lr_offset >= zp->z_size)
1294 error = SET_ERROR(ENOENT);
1297 error = SET_ERROR(EIO);
1302 error = dmu_buf_hold(os, object, offset, zgd, &db,
1303 DMU_READ_NO_PREFETCH);
1306 blkptr_t *obp = dmu_buf_get_blkptr(db);
1308 ASSERT(BP_IS_HOLE(bp));
1315 ASSERT(db->db_offset == offset);
1316 ASSERT(db->db_size == size);
1318 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1320 ASSERT(error || lr->lr_length <= zp->z_blksz);
1323 * On success, we need to wait for the write I/O
1324 * initiated by dmu_sync() to complete before we can
1325 * release this dbuf. We will finish everything up
1326 * in the zfs_get_done() callback.
1331 if (error == EALREADY) {
1332 lr->lr_common.lrc_txtype = TX_WRITE2;
1338 zfs_get_done(zgd, error);
1345 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1346 caller_context_t *ct)
1348 znode_t *zp = VTOZ(vp);
1349 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1355 if (flag & V_ACE_MASK)
1356 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1358 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1365 * If vnode is for a device return a specfs vnode instead.
1368 specvp_check(vnode_t **vpp, cred_t *cr)
1372 if (IS_DEVVP(*vpp)) {
1375 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1378 error = SET_ERROR(ENOSYS);
1386 * Lookup an entry in a directory, or an extended attribute directory.
1387 * If it exists, return a held vnode reference for it.
1389 * IN: dvp - vnode of directory to search.
1390 * nm - name of entry to lookup.
1391 * pnp - full pathname to lookup [UNUSED].
1392 * flags - LOOKUP_XATTR set if looking for an attribute.
1393 * rdir - root directory vnode [UNUSED].
1394 * cr - credentials of caller.
1395 * ct - caller context
1396 * direntflags - directory lookup flags
1397 * realpnp - returned pathname.
1399 * OUT: vpp - vnode of located entry, NULL if not found.
1401 * RETURN: 0 on success, error code on failure.
1408 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp,
1409 int nameiop, cred_t *cr, kthread_t *td, int flags)
1411 znode_t *zdp = VTOZ(dvp);
1412 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1414 int *direntflags = NULL;
1415 void *realpnp = NULL;
1418 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1420 if (dvp->v_type != VDIR) {
1421 return (SET_ERROR(ENOTDIR));
1422 } else if (zdp->z_sa_hdl == NULL) {
1423 return (SET_ERROR(EIO));
1426 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1427 error = zfs_fastaccesschk_execute(zdp, cr);
1435 vnode_t *tvp = dnlc_lookup(dvp, nm);
1438 error = zfs_fastaccesschk_execute(zdp, cr);
1443 if (tvp == DNLC_NO_VNODE) {
1445 return (SET_ERROR(ENOENT));
1448 return (specvp_check(vpp, cr));
1454 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1461 if (flags & LOOKUP_XATTR) {
1464 * If the xattr property is off, refuse the lookup request.
1466 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1468 return (SET_ERROR(EINVAL));
1473 * We don't allow recursive attributes..
1474 * Maybe someday we will.
1476 if (zdp->z_pflags & ZFS_XATTR) {
1478 return (SET_ERROR(EINVAL));
1481 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1487 * Do we have permission to get into attribute directory?
1490 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1500 if (dvp->v_type != VDIR) {
1502 return (SET_ERROR(ENOTDIR));
1506 * Check accessibility of directory.
1509 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1514 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1515 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1517 return (SET_ERROR(EILSEQ));
1520 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1522 error = specvp_check(vpp, cr);
1524 /* Translate errors and add SAVENAME when needed. */
1525 if (cnp->cn_flags & ISLASTCN) {
1529 if (error == ENOENT) {
1530 error = EJUSTRETURN;
1531 cnp->cn_flags |= SAVENAME;
1537 cnp->cn_flags |= SAVENAME;
1541 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) {
1544 if (cnp->cn_flags & ISDOTDOT) {
1545 ltype = VOP_ISLOCKED(dvp);
1549 error = vn_lock(*vpp, cnp->cn_lkflags);
1550 if (cnp->cn_flags & ISDOTDOT)
1551 vn_lock(dvp, ltype | LK_RETRY);
1561 #ifdef FREEBSD_NAMECACHE
1563 * Insert name into cache (as non-existent) if appropriate.
1565 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
1566 cache_enter(dvp, *vpp, cnp);
1568 * Insert name into cache if appropriate.
1570 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) {
1571 if (!(cnp->cn_flags & ISLASTCN) ||
1572 (nameiop != DELETE && nameiop != RENAME)) {
1573 cache_enter(dvp, *vpp, cnp);
1582 * Attempt to create a new entry in a directory. If the entry
1583 * already exists, truncate the file if permissible, else return
1584 * an error. Return the vp of the created or trunc'd file.
1586 * IN: dvp - vnode of directory to put new file entry in.
1587 * name - name of new file entry.
1588 * vap - attributes of new file.
1589 * excl - flag indicating exclusive or non-exclusive mode.
1590 * mode - mode to open file with.
1591 * cr - credentials of caller.
1592 * flag - large file flag [UNUSED].
1593 * ct - caller context
1594 * vsecp - ACL to be set
1596 * OUT: vpp - vnode of created or trunc'd entry.
1598 * RETURN: 0 on success, error code on failure.
1601 * dvp - ctime|mtime updated if new entry created
1602 * vp - ctime|mtime always, atime if new
1607 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode,
1608 vnode_t **vpp, cred_t *cr, kthread_t *td)
1610 znode_t *zp, *dzp = VTOZ(dvp);
1611 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1619 gid_t gid = crgetgid(cr);
1620 zfs_acl_ids_t acl_ids;
1621 boolean_t fuid_dirtied;
1622 boolean_t have_acl = B_FALSE;
1623 boolean_t waited = B_FALSE;
1628 * If we have an ephemeral id, ACL, or XVATTR then
1629 * make sure file system is at proper version
1632 ksid = crgetsid(cr, KSID_OWNER);
1634 uid = ksid_getid(ksid);
1638 if (zfsvfs->z_use_fuids == B_FALSE &&
1639 (vsecp || (vap->va_mask & AT_XVATTR) ||
1640 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1641 return (SET_ERROR(EINVAL));
1646 zilog = zfsvfs->z_log;
1648 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1649 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1651 return (SET_ERROR(EILSEQ));
1654 if (vap->va_mask & AT_XVATTR) {
1655 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
1656 crgetuid(cr), cr, vap->va_type)) != 0) {
1664 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr))
1665 vap->va_mode &= ~S_ISVTX;
1667 if (*name == '\0') {
1669 * Null component name refers to the directory itself.
1676 /* possible VN_HOLD(zp) */
1679 if (flag & FIGNORECASE)
1682 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1686 zfs_acl_ids_free(&acl_ids);
1687 if (strcmp(name, "..") == 0)
1688 error = SET_ERROR(EISDIR);
1698 * Create a new file object and update the directory
1701 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1703 zfs_acl_ids_free(&acl_ids);
1708 * We only support the creation of regular files in
1709 * extended attribute directories.
1712 if ((dzp->z_pflags & ZFS_XATTR) &&
1713 (vap->va_type != VREG)) {
1715 zfs_acl_ids_free(&acl_ids);
1716 error = SET_ERROR(EINVAL);
1720 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1721 cr, vsecp, &acl_ids)) != 0)
1725 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1726 zfs_acl_ids_free(&acl_ids);
1727 error = SET_ERROR(EDQUOT);
1731 tx = dmu_tx_create(os);
1733 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1734 ZFS_SA_BASE_ATTR_SIZE);
1736 fuid_dirtied = zfsvfs->z_fuid_dirty;
1738 zfs_fuid_txhold(zfsvfs, tx);
1739 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1740 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1741 if (!zfsvfs->z_use_sa &&
1742 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1743 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1744 0, acl_ids.z_aclp->z_acl_bytes);
1746 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1748 zfs_dirent_unlock(dl);
1749 if (error == ERESTART) {
1755 zfs_acl_ids_free(&acl_ids);
1760 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1763 zfs_fuid_sync(zfsvfs, tx);
1765 (void) zfs_link_create(dl, zp, tx, ZNEW);
1766 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1767 if (flag & FIGNORECASE)
1769 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1770 vsecp, acl_ids.z_fuidp, vap);
1771 zfs_acl_ids_free(&acl_ids);
1774 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1777 zfs_acl_ids_free(&acl_ids);
1781 * A directory entry already exists for this name.
1784 * Can't truncate an existing file if in exclusive mode.
1787 error = SET_ERROR(EEXIST);
1791 * Can't open a directory for writing.
1793 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1794 error = SET_ERROR(EISDIR);
1798 * Verify requested access to file.
1800 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1804 mutex_enter(&dzp->z_lock);
1806 mutex_exit(&dzp->z_lock);
1809 * Truncate regular files if requested.
1811 if ((ZTOV(zp)->v_type == VREG) &&
1812 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1813 /* we can't hold any locks when calling zfs_freesp() */
1814 zfs_dirent_unlock(dl);
1816 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1818 vnevent_create(ZTOV(zp), ct);
1824 zfs_dirent_unlock(dl);
1831 error = specvp_check(vpp, cr);
1834 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1835 zil_commit(zilog, 0);
1842 * Remove an entry from a directory.
1844 * IN: dvp - vnode of directory to remove entry from.
1845 * name - name of entry to remove.
1846 * cr - credentials of caller.
1847 * ct - caller context
1848 * flags - case flags
1850 * RETURN: 0 on success, error code on failure.
1854 * vp - ctime (if nlink > 0)
1857 uint64_t null_xattr = 0;
1861 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1864 znode_t *zp, *dzp = VTOZ(dvp);
1867 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1869 uint64_t acl_obj, xattr_obj;
1870 uint64_t xattr_obj_unlinked = 0;
1874 boolean_t may_delete_now, delete_now = FALSE;
1875 boolean_t unlinked, toobig = FALSE;
1877 pathname_t *realnmp = NULL;
1881 boolean_t waited = B_FALSE;
1885 zilog = zfsvfs->z_log;
1887 if (flags & FIGNORECASE) {
1897 * Attempt to lock directory; fail if entry doesn't exist.
1899 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1909 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1914 * Need to use rmdir for removing directories.
1916 if (vp->v_type == VDIR) {
1917 error = SET_ERROR(EPERM);
1921 vnevent_remove(vp, dvp, name, ct);
1924 dnlc_remove(dvp, realnmp->pn_buf);
1926 dnlc_remove(dvp, name);
1929 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1933 * We may delete the znode now, or we may put it in the unlinked set;
1934 * it depends on whether we're the last link, and on whether there are
1935 * other holds on the vnode. So we dmu_tx_hold() the right things to
1936 * allow for either case.
1939 tx = dmu_tx_create(zfsvfs->z_os);
1940 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1941 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1942 zfs_sa_upgrade_txholds(tx, zp);
1943 zfs_sa_upgrade_txholds(tx, dzp);
1944 if (may_delete_now) {
1946 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1947 /* if the file is too big, only hold_free a token amount */
1948 dmu_tx_hold_free(tx, zp->z_id, 0,
1949 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1952 /* are there any extended attributes? */
1953 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1954 &xattr_obj, sizeof (xattr_obj));
1955 if (error == 0 && xattr_obj) {
1956 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1958 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1959 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1962 mutex_enter(&zp->z_lock);
1963 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1964 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1965 mutex_exit(&zp->z_lock);
1967 /* charge as an update -- would be nice not to charge at all */
1968 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1970 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1972 zfs_dirent_unlock(dl);
1976 if (error == ERESTART) {
1990 * Remove the directory entry.
1992 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
2002 * Hold z_lock so that we can make sure that the ACL obj
2003 * hasn't changed. Could have been deleted due to
2006 mutex_enter(&zp->z_lock);
2008 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
2009 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
2010 delete_now = may_delete_now && !toobig &&
2011 vp->v_count == 1 && !vn_has_cached_data(vp) &&
2012 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
2019 panic("zfs_remove: delete_now branch taken");
2021 if (xattr_obj_unlinked) {
2022 ASSERT3U(xzp->z_links, ==, 2);
2023 mutex_enter(&xzp->z_lock);
2024 xzp->z_unlinked = 1;
2026 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
2027 &xzp->z_links, sizeof (xzp->z_links), tx);
2028 ASSERT3U(error, ==, 0);
2029 mutex_exit(&xzp->z_lock);
2030 zfs_unlinked_add(xzp, tx);
2033 error = sa_remove(zp->z_sa_hdl,
2034 SA_ZPL_XATTR(zfsvfs), tx);
2036 error = sa_update(zp->z_sa_hdl,
2037 SA_ZPL_XATTR(zfsvfs), &null_xattr,
2038 sizeof (uint64_t), tx);
2043 ASSERT0(vp->v_count);
2045 mutex_exit(&zp->z_lock);
2046 zfs_znode_delete(zp, tx);
2047 } else if (unlinked) {
2048 mutex_exit(&zp->z_lock);
2049 zfs_unlinked_add(zp, tx);
2051 vp->v_vflag |= VV_NOSYNC;
2056 if (flags & FIGNORECASE)
2058 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
2065 zfs_dirent_unlock(dl);
2072 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2073 zil_commit(zilog, 0);
2080 * Create a new directory and insert it into dvp using the name
2081 * provided. Return a pointer to the inserted directory.
2083 * IN: dvp - vnode of directory to add subdir to.
2084 * dirname - name of new directory.
2085 * vap - attributes of new directory.
2086 * cr - credentials of caller.
2087 * ct - caller context
2088 * flags - case flags
2089 * vsecp - ACL to be set
2091 * OUT: vpp - vnode of created directory.
2093 * RETURN: 0 on success, error code on failure.
2096 * dvp - ctime|mtime updated
2097 * vp - ctime|mtime|atime updated
2101 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
2102 caller_context_t *ct, int flags, vsecattr_t *vsecp)
2104 znode_t *zp, *dzp = VTOZ(dvp);
2105 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2114 gid_t gid = crgetgid(cr);
2115 zfs_acl_ids_t acl_ids;
2116 boolean_t fuid_dirtied;
2117 boolean_t waited = B_FALSE;
2119 ASSERT(vap->va_type == VDIR);
2122 * If we have an ephemeral id, ACL, or XVATTR then
2123 * make sure file system is at proper version
2126 ksid = crgetsid(cr, KSID_OWNER);
2128 uid = ksid_getid(ksid);
2131 if (zfsvfs->z_use_fuids == B_FALSE &&
2132 (vsecp || (vap->va_mask & AT_XVATTR) ||
2133 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
2134 return (SET_ERROR(EINVAL));
2138 zilog = zfsvfs->z_log;
2140 if (dzp->z_pflags & ZFS_XATTR) {
2142 return (SET_ERROR(EINVAL));
2145 if (zfsvfs->z_utf8 && u8_validate(dirname,
2146 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2148 return (SET_ERROR(EILSEQ));
2150 if (flags & FIGNORECASE)
2153 if (vap->va_mask & AT_XVATTR) {
2154 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
2155 crgetuid(cr), cr, vap->va_type)) != 0) {
2161 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
2162 vsecp, &acl_ids)) != 0) {
2167 * First make sure the new directory doesn't exist.
2169 * Existence is checked first to make sure we don't return
2170 * EACCES instead of EEXIST which can cause some applications
2176 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
2178 zfs_acl_ids_free(&acl_ids);
2183 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
2184 zfs_acl_ids_free(&acl_ids);
2185 zfs_dirent_unlock(dl);
2190 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
2191 zfs_acl_ids_free(&acl_ids);
2192 zfs_dirent_unlock(dl);
2194 return (SET_ERROR(EDQUOT));
2198 * Add a new entry to the directory.
2200 tx = dmu_tx_create(zfsvfs->z_os);
2201 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2202 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2203 fuid_dirtied = zfsvfs->z_fuid_dirty;
2205 zfs_fuid_txhold(zfsvfs, tx);
2206 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2207 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2208 acl_ids.z_aclp->z_acl_bytes);
2211 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2212 ZFS_SA_BASE_ATTR_SIZE);
2214 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2216 zfs_dirent_unlock(dl);
2217 if (error == ERESTART) {
2223 zfs_acl_ids_free(&acl_ids);
2232 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2235 zfs_fuid_sync(zfsvfs, tx);
2238 * Now put new name in parent dir.
2240 (void) zfs_link_create(dl, zp, tx, ZNEW);
2244 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2245 if (flags & FIGNORECASE)
2247 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2248 acl_ids.z_fuidp, vap);
2250 zfs_acl_ids_free(&acl_ids);
2254 zfs_dirent_unlock(dl);
2256 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2257 zil_commit(zilog, 0);
2264 * Remove a directory subdir entry. If the current working
2265 * directory is the same as the subdir to be removed, the
2268 * IN: dvp - vnode of directory to remove from.
2269 * name - name of directory to be removed.
2270 * cwd - vnode of current working directory.
2271 * cr - credentials of caller.
2272 * ct - caller context
2273 * flags - case flags
2275 * RETURN: 0 on success, error code on failure.
2278 * dvp - ctime|mtime updated
2282 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2283 caller_context_t *ct, int flags)
2285 znode_t *dzp = VTOZ(dvp);
2288 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2294 boolean_t waited = B_FALSE;
2298 zilog = zfsvfs->z_log;
2300 if (flags & FIGNORECASE)
2306 * Attempt to lock directory; fail if entry doesn't exist.
2308 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2316 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2320 if (vp->v_type != VDIR) {
2321 error = SET_ERROR(ENOTDIR);
2326 error = SET_ERROR(EINVAL);
2330 vnevent_rmdir(vp, dvp, name, ct);
2333 * Grab a lock on the directory to make sure that noone is
2334 * trying to add (or lookup) entries while we are removing it.
2336 rw_enter(&zp->z_name_lock, RW_WRITER);
2339 * Grab a lock on the parent pointer to make sure we play well
2340 * with the treewalk and directory rename code.
2342 rw_enter(&zp->z_parent_lock, RW_WRITER);
2344 tx = dmu_tx_create(zfsvfs->z_os);
2345 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2346 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2347 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2348 zfs_sa_upgrade_txholds(tx, zp);
2349 zfs_sa_upgrade_txholds(tx, dzp);
2350 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2352 rw_exit(&zp->z_parent_lock);
2353 rw_exit(&zp->z_name_lock);
2354 zfs_dirent_unlock(dl);
2356 if (error == ERESTART) {
2367 #ifdef FREEBSD_NAMECACHE
2371 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2374 uint64_t txtype = TX_RMDIR;
2375 if (flags & FIGNORECASE)
2377 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2382 rw_exit(&zp->z_parent_lock);
2383 rw_exit(&zp->z_name_lock);
2384 #ifdef FREEBSD_NAMECACHE
2388 zfs_dirent_unlock(dl);
2392 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2393 zil_commit(zilog, 0);
2400 * Read as many directory entries as will fit into the provided
2401 * buffer from the given directory cursor position (specified in
2402 * the uio structure).
2404 * IN: vp - vnode of directory to read.
2405 * uio - structure supplying read location, range info,
2406 * and return buffer.
2407 * cr - credentials of caller.
2408 * ct - caller context
2409 * flags - case flags
2411 * OUT: uio - updated offset and range, buffer filled.
2412 * eofp - set to true if end-of-file detected.
2414 * RETURN: 0 on success, error code on failure.
2417 * vp - atime updated
2419 * Note that the low 4 bits of the cookie returned by zap is always zero.
2420 * This allows us to use the low range for "special" directory entries:
2421 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2422 * we use the offset 2 for the '.zfs' directory.
2426 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies)
2428 znode_t *zp = VTOZ(vp);
2432 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2437 zap_attribute_t zap;
2438 uint_t bytes_wanted;
2439 uint64_t offset; /* must be unsigned; checks for < 1 */
2445 boolean_t check_sysattrs;
2448 u_long *cooks = NULL;
2454 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2455 &parent, sizeof (parent))) != 0) {
2461 * If we are not given an eof variable,
2468 * Check for valid iov_len.
2470 if (uio->uio_iov->iov_len <= 0) {
2472 return (SET_ERROR(EINVAL));
2476 * Quit if directory has been removed (posix)
2478 if ((*eofp = zp->z_unlinked) != 0) {
2485 offset = uio->uio_loffset;
2486 prefetch = zp->z_zn_prefetch;
2489 * Initialize the iterator cursor.
2493 * Start iteration from the beginning of the directory.
2495 zap_cursor_init(&zc, os, zp->z_id);
2498 * The offset is a serialized cursor.
2500 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2504 * Get space to change directory entries into fs independent format.
2506 iovp = uio->uio_iov;
2507 bytes_wanted = iovp->iov_len;
2508 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2509 bufsize = bytes_wanted;
2510 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2511 odp = (struct dirent64 *)outbuf;
2513 bufsize = bytes_wanted;
2515 odp = (struct dirent64 *)iovp->iov_base;
2517 eodp = (struct edirent *)odp;
2519 if (ncookies != NULL) {
2521 * Minimum entry size is dirent size and 1 byte for a file name.
2523 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1);
2524 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK);
2529 * If this VFS supports the system attribute view interface; and
2530 * we're looking at an extended attribute directory; and we care
2531 * about normalization conflicts on this vfs; then we must check
2532 * for normalization conflicts with the sysattr name space.
2535 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2536 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2537 (flags & V_RDDIR_ENTFLAGS);
2543 * Transform to file-system independent format
2546 while (outcount < bytes_wanted) {
2549 off64_t *next = NULL;
2552 * Special case `.', `..', and `.zfs'.
2555 (void) strcpy(zap.za_name, ".");
2556 zap.za_normalization_conflict = 0;
2559 } else if (offset == 1) {
2560 (void) strcpy(zap.za_name, "..");
2561 zap.za_normalization_conflict = 0;
2564 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2565 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2566 zap.za_normalization_conflict = 0;
2567 objnum = ZFSCTL_INO_ROOT;
2573 if (error = zap_cursor_retrieve(&zc, &zap)) {
2574 if ((*eofp = (error == ENOENT)) != 0)
2580 if (zap.za_integer_length != 8 ||
2581 zap.za_num_integers != 1) {
2582 cmn_err(CE_WARN, "zap_readdir: bad directory "
2583 "entry, obj = %lld, offset = %lld\n",
2584 (u_longlong_t)zp->z_id,
2585 (u_longlong_t)offset);
2586 error = SET_ERROR(ENXIO);
2590 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2592 * MacOS X can extract the object type here such as:
2593 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2595 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2597 if (check_sysattrs && !zap.za_normalization_conflict) {
2599 zap.za_normalization_conflict =
2600 xattr_sysattr_casechk(zap.za_name);
2602 panic("%s:%u: TODO", __func__, __LINE__);
2607 if (flags & V_RDDIR_ACCFILTER) {
2609 * If we have no access at all, don't include
2610 * this entry in the returned information
2613 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2615 if (!zfs_has_access(ezp, cr)) {
2622 if (flags & V_RDDIR_ENTFLAGS)
2623 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2625 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2628 * Will this entry fit in the buffer?
2630 if (outcount + reclen > bufsize) {
2632 * Did we manage to fit anything in the buffer?
2635 error = SET_ERROR(EINVAL);
2640 if (flags & V_RDDIR_ENTFLAGS) {
2642 * Add extended flag entry:
2644 eodp->ed_ino = objnum;
2645 eodp->ed_reclen = reclen;
2646 /* NOTE: ed_off is the offset for the *next* entry */
2647 next = &(eodp->ed_off);
2648 eodp->ed_eflags = zap.za_normalization_conflict ?
2649 ED_CASE_CONFLICT : 0;
2650 (void) strncpy(eodp->ed_name, zap.za_name,
2651 EDIRENT_NAMELEN(reclen));
2652 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2657 odp->d_ino = objnum;
2658 odp->d_reclen = reclen;
2659 odp->d_namlen = strlen(zap.za_name);
2660 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1);
2662 odp = (dirent64_t *)((intptr_t)odp + reclen);
2666 ASSERT(outcount <= bufsize);
2668 /* Prefetch znode */
2670 dmu_prefetch(os, objnum, 0, 0);
2674 * Move to the next entry, fill in the previous offset.
2676 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2677 zap_cursor_advance(&zc);
2678 offset = zap_cursor_serialize(&zc);
2683 if (cooks != NULL) {
2686 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks));
2689 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2691 /* Subtract unused cookies */
2692 if (ncookies != NULL)
2693 *ncookies -= ncooks;
2695 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2696 iovp->iov_base += outcount;
2697 iovp->iov_len -= outcount;
2698 uio->uio_resid -= outcount;
2699 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2701 * Reset the pointer.
2703 offset = uio->uio_loffset;
2707 zap_cursor_fini(&zc);
2708 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2709 kmem_free(outbuf, bufsize);
2711 if (error == ENOENT)
2714 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2716 uio->uio_loffset = offset;
2718 if (error != 0 && cookies != NULL) {
2719 free(*cookies, M_TEMP);
2726 ulong_t zfs_fsync_sync_cnt = 4;
2729 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2731 znode_t *zp = VTOZ(vp);
2732 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2734 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2736 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2739 zil_commit(zfsvfs->z_log, zp->z_id);
2747 * Get the requested file attributes and place them in the provided
2750 * IN: vp - vnode of file.
2751 * vap - va_mask identifies requested attributes.
2752 * If AT_XVATTR set, then optional attrs are requested
2753 * flags - ATTR_NOACLCHECK (CIFS server context)
2754 * cr - credentials of caller.
2755 * ct - caller context
2757 * OUT: vap - attribute values.
2759 * RETURN: 0 (always succeeds).
2763 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2764 caller_context_t *ct)
2766 znode_t *zp = VTOZ(vp);
2767 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2770 u_longlong_t nblocks;
2772 uint64_t mtime[2], ctime[2], crtime[2], rdev;
2773 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2774 xoptattr_t *xoap = NULL;
2775 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2776 sa_bulk_attr_t bulk[4];
2782 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2784 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2785 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2786 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
2787 if (vp->v_type == VBLK || vp->v_type == VCHR)
2788 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
2791 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2797 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2798 * Also, if we are the owner don't bother, since owner should
2799 * always be allowed to read basic attributes of file.
2801 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2802 (vap->va_uid != crgetuid(cr))) {
2803 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2811 * Return all attributes. It's cheaper to provide the answer
2812 * than to determine whether we were asked the question.
2815 mutex_enter(&zp->z_lock);
2816 vap->va_type = IFTOVT(zp->z_mode);
2817 vap->va_mode = zp->z_mode & ~S_IFMT;
2819 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2821 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
2823 vap->va_nodeid = zp->z_id;
2824 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2825 links = zp->z_links + 1;
2827 links = zp->z_links;
2828 vap->va_nlink = MIN(links, LINK_MAX); /* nlink_t limit! */
2829 vap->va_size = zp->z_size;
2831 vap->va_rdev = vp->v_rdev;
2833 if (vp->v_type == VBLK || vp->v_type == VCHR)
2834 vap->va_rdev = zfs_cmpldev(rdev);
2836 vap->va_seq = zp->z_seq;
2837 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */
2840 * Add in any requested optional attributes and the create time.
2841 * Also set the corresponding bits in the returned attribute bitmap.
2843 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2844 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2846 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2847 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2850 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2851 xoap->xoa_readonly =
2852 ((zp->z_pflags & ZFS_READONLY) != 0);
2853 XVA_SET_RTN(xvap, XAT_READONLY);
2856 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2858 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2859 XVA_SET_RTN(xvap, XAT_SYSTEM);
2862 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2864 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2865 XVA_SET_RTN(xvap, XAT_HIDDEN);
2868 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2869 xoap->xoa_nounlink =
2870 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2871 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2874 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2875 xoap->xoa_immutable =
2876 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2877 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2880 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2881 xoap->xoa_appendonly =
2882 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2883 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2886 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2888 ((zp->z_pflags & ZFS_NODUMP) != 0);
2889 XVA_SET_RTN(xvap, XAT_NODUMP);
2892 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2894 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2895 XVA_SET_RTN(xvap, XAT_OPAQUE);
2898 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2899 xoap->xoa_av_quarantined =
2900 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2901 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2904 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2905 xoap->xoa_av_modified =
2906 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2907 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2910 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2911 vp->v_type == VREG) {
2912 zfs_sa_get_scanstamp(zp, xvap);
2915 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2918 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2919 times, sizeof (times));
2920 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2921 XVA_SET_RTN(xvap, XAT_CREATETIME);
2924 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2925 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2926 XVA_SET_RTN(xvap, XAT_REPARSE);
2928 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2929 xoap->xoa_generation = zp->z_gen;
2930 XVA_SET_RTN(xvap, XAT_GEN);
2933 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2935 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2936 XVA_SET_RTN(xvap, XAT_OFFLINE);
2939 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2941 ((zp->z_pflags & ZFS_SPARSE) != 0);
2942 XVA_SET_RTN(xvap, XAT_SPARSE);
2946 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2947 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2948 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2949 ZFS_TIME_DECODE(&vap->va_birthtime, crtime);
2951 mutex_exit(&zp->z_lock);
2953 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2954 vap->va_blksize = blksize;
2955 vap->va_bytes = nblocks << 9; /* nblocks * 512 */
2957 if (zp->z_blksz == 0) {
2959 * Block size hasn't been set; suggest maximal I/O transfers.
2961 vap->va_blksize = zfsvfs->z_max_blksz;
2969 * Set the file attributes to the values contained in the
2972 * IN: vp - vnode of file to be modified.
2973 * vap - new attribute values.
2974 * If AT_XVATTR set, then optional attrs are being set
2975 * flags - ATTR_UTIME set if non-default time values provided.
2976 * - ATTR_NOACLCHECK (CIFS context only).
2977 * cr - credentials of caller.
2978 * ct - caller context
2980 * RETURN: 0 on success, error code on failure.
2983 * vp - ctime updated, mtime updated if size changed.
2987 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2988 caller_context_t *ct)
2990 znode_t *zp = VTOZ(vp);
2991 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2996 uint_t mask = vap->va_mask;
2997 uint_t saved_mask = 0;
2998 uint64_t saved_mode;
3001 uint64_t new_uid, new_gid;
3003 uint64_t mtime[2], ctime[2];
3005 int need_policy = FALSE;
3007 zfs_fuid_info_t *fuidp = NULL;
3008 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
3011 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
3012 boolean_t fuid_dirtied = B_FALSE;
3013 sa_bulk_attr_t bulk[7], xattr_bulk[7];
3014 int count = 0, xattr_count = 0;
3019 if (mask & AT_NOSET)
3020 return (SET_ERROR(EINVAL));
3025 zilog = zfsvfs->z_log;
3028 * Make sure that if we have ephemeral uid/gid or xvattr specified
3029 * that file system is at proper version level
3032 if (zfsvfs->z_use_fuids == B_FALSE &&
3033 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
3034 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
3035 (mask & AT_XVATTR))) {
3037 return (SET_ERROR(EINVAL));
3040 if (mask & AT_SIZE && vp->v_type == VDIR) {
3042 return (SET_ERROR(EISDIR));
3045 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
3047 return (SET_ERROR(EINVAL));
3051 * If this is an xvattr_t, then get a pointer to the structure of
3052 * optional attributes. If this is NULL, then we have a vattr_t.
3054 xoap = xva_getxoptattr(xvap);
3056 xva_init(&tmpxvattr);
3059 * Immutable files can only alter immutable bit and atime
3061 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
3062 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
3063 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
3065 return (SET_ERROR(EPERM));
3068 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
3070 return (SET_ERROR(EPERM));
3074 * Verify timestamps doesn't overflow 32 bits.
3075 * ZFS can handle large timestamps, but 32bit syscalls can't
3076 * handle times greater than 2039. This check should be removed
3077 * once large timestamps are fully supported.
3079 if (mask & (AT_ATIME | AT_MTIME)) {
3080 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
3081 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
3083 return (SET_ERROR(EOVERFLOW));
3091 /* Can this be moved to before the top label? */
3092 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
3094 return (SET_ERROR(EROFS));
3098 * First validate permissions
3101 if (mask & AT_SIZE) {
3103 * XXX - Note, we are not providing any open
3104 * mode flags here (like FNDELAY), so we may
3105 * block if there are locks present... this
3106 * should be addressed in openat().
3108 /* XXX - would it be OK to generate a log record here? */
3109 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
3116 if (mask & (AT_ATIME|AT_MTIME) ||
3117 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
3118 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
3119 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
3120 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
3121 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
3122 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
3123 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
3124 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
3128 if (mask & (AT_UID|AT_GID)) {
3129 int idmask = (mask & (AT_UID|AT_GID));
3134 * NOTE: even if a new mode is being set,
3135 * we may clear S_ISUID/S_ISGID bits.
3138 if (!(mask & AT_MODE))
3139 vap->va_mode = zp->z_mode;
3142 * Take ownership or chgrp to group we are a member of
3145 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
3146 take_group = (mask & AT_GID) &&
3147 zfs_groupmember(zfsvfs, vap->va_gid, cr);
3150 * If both AT_UID and AT_GID are set then take_owner and
3151 * take_group must both be set in order to allow taking
3154 * Otherwise, send the check through secpolicy_vnode_setattr()
3158 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
3159 ((idmask == AT_UID) && take_owner) ||
3160 ((idmask == AT_GID) && take_group)) {
3161 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
3162 skipaclchk, cr) == 0) {
3164 * Remove setuid/setgid for non-privileged users
3166 secpolicy_setid_clear(vap, vp, cr);
3167 trim_mask = (mask & (AT_UID|AT_GID));
3176 mutex_enter(&zp->z_lock);
3177 oldva.va_mode = zp->z_mode;
3178 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
3179 if (mask & AT_XVATTR) {
3181 * Update xvattr mask to include only those attributes
3182 * that are actually changing.
3184 * the bits will be restored prior to actually setting
3185 * the attributes so the caller thinks they were set.
3187 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
3188 if (xoap->xoa_appendonly !=
3189 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
3192 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
3193 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
3197 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
3198 if (xoap->xoa_nounlink !=
3199 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
3202 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
3203 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
3207 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
3208 if (xoap->xoa_immutable !=
3209 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
3212 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
3213 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
3217 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
3218 if (xoap->xoa_nodump !=
3219 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
3222 XVA_CLR_REQ(xvap, XAT_NODUMP);
3223 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
3227 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3228 if (xoap->xoa_av_modified !=
3229 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3232 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3233 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
3237 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3238 if ((vp->v_type != VREG &&
3239 xoap->xoa_av_quarantined) ||
3240 xoap->xoa_av_quarantined !=
3241 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3244 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3245 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
3249 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3250 mutex_exit(&zp->z_lock);
3252 return (SET_ERROR(EPERM));
3255 if (need_policy == FALSE &&
3256 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3257 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3262 mutex_exit(&zp->z_lock);
3264 if (mask & AT_MODE) {
3265 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3266 err = secpolicy_setid_setsticky_clear(vp, vap,
3272 trim_mask |= AT_MODE;
3280 * If trim_mask is set then take ownership
3281 * has been granted or write_acl is present and user
3282 * has the ability to modify mode. In that case remove
3283 * UID|GID and or MODE from mask so that
3284 * secpolicy_vnode_setattr() doesn't revoke it.
3288 saved_mask = vap->va_mask;
3289 vap->va_mask &= ~trim_mask;
3290 if (trim_mask & AT_MODE) {
3292 * Save the mode, as secpolicy_vnode_setattr()
3293 * will overwrite it with ova.va_mode.
3295 saved_mode = vap->va_mode;
3298 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3299 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3306 vap->va_mask |= saved_mask;
3307 if (trim_mask & AT_MODE) {
3309 * Recover the mode after
3310 * secpolicy_vnode_setattr().
3312 vap->va_mode = saved_mode;
3318 * secpolicy_vnode_setattr, or take ownership may have
3321 mask = vap->va_mask;
3323 if ((mask & (AT_UID | AT_GID))) {
3324 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3325 &xattr_obj, sizeof (xattr_obj));
3327 if (err == 0 && xattr_obj) {
3328 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3332 if (mask & AT_UID) {
3333 new_uid = zfs_fuid_create(zfsvfs,
3334 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3335 if (new_uid != zp->z_uid &&
3336 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3338 VN_RELE(ZTOV(attrzp));
3339 err = SET_ERROR(EDQUOT);
3344 if (mask & AT_GID) {
3345 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3346 cr, ZFS_GROUP, &fuidp);
3347 if (new_gid != zp->z_gid &&
3348 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3350 VN_RELE(ZTOV(attrzp));
3351 err = SET_ERROR(EDQUOT);
3356 tx = dmu_tx_create(zfsvfs->z_os);
3358 if (mask & AT_MODE) {
3359 uint64_t pmode = zp->z_mode;
3361 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3363 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3364 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3365 err = SET_ERROR(EPERM);
3369 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3372 mutex_enter(&zp->z_lock);
3373 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3375 * Are we upgrading ACL from old V0 format
3378 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3379 zfs_znode_acl_version(zp) ==
3380 ZFS_ACL_VERSION_INITIAL) {
3381 dmu_tx_hold_free(tx, acl_obj, 0,
3383 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3384 0, aclp->z_acl_bytes);
3386 dmu_tx_hold_write(tx, acl_obj, 0,
3389 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3390 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3391 0, aclp->z_acl_bytes);
3393 mutex_exit(&zp->z_lock);
3394 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3396 if ((mask & AT_XVATTR) &&
3397 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3398 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3400 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3404 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3407 fuid_dirtied = zfsvfs->z_fuid_dirty;
3409 zfs_fuid_txhold(zfsvfs, tx);
3411 zfs_sa_upgrade_txholds(tx, zp);
3413 err = dmu_tx_assign(tx, TXG_NOWAIT);
3415 if (err == ERESTART)
3422 * Set each attribute requested.
3423 * We group settings according to the locks they need to acquire.
3425 * Note: you cannot set ctime directly, although it will be
3426 * updated as a side-effect of calling this function.
3430 if (mask & (AT_UID|AT_GID|AT_MODE))
3431 mutex_enter(&zp->z_acl_lock);
3432 mutex_enter(&zp->z_lock);
3434 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3435 &zp->z_pflags, sizeof (zp->z_pflags));
3438 if (mask & (AT_UID|AT_GID|AT_MODE))
3439 mutex_enter(&attrzp->z_acl_lock);
3440 mutex_enter(&attrzp->z_lock);
3441 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3442 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3443 sizeof (attrzp->z_pflags));
3446 if (mask & (AT_UID|AT_GID)) {
3448 if (mask & AT_UID) {
3449 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3450 &new_uid, sizeof (new_uid));
3451 zp->z_uid = new_uid;
3453 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3454 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3456 attrzp->z_uid = new_uid;
3460 if (mask & AT_GID) {
3461 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3462 NULL, &new_gid, sizeof (new_gid));
3463 zp->z_gid = new_gid;
3465 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3466 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3468 attrzp->z_gid = new_gid;
3471 if (!(mask & AT_MODE)) {
3472 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3473 NULL, &new_mode, sizeof (new_mode));
3474 new_mode = zp->z_mode;
3476 err = zfs_acl_chown_setattr(zp);
3479 err = zfs_acl_chown_setattr(attrzp);
3484 if (mask & AT_MODE) {
3485 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3486 &new_mode, sizeof (new_mode));
3487 zp->z_mode = new_mode;
3488 ASSERT3U((uintptr_t)aclp, !=, 0);
3489 err = zfs_aclset_common(zp, aclp, cr, tx);
3491 if (zp->z_acl_cached)
3492 zfs_acl_free(zp->z_acl_cached);
3493 zp->z_acl_cached = aclp;
3498 if (mask & AT_ATIME) {
3499 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3500 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3501 &zp->z_atime, sizeof (zp->z_atime));
3504 if (mask & AT_MTIME) {
3505 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3506 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3507 mtime, sizeof (mtime));
3510 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3511 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3512 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3513 NULL, mtime, sizeof (mtime));
3514 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3515 &ctime, sizeof (ctime));
3516 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3518 } else if (mask != 0) {
3519 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3520 &ctime, sizeof (ctime));
3521 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3524 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3525 SA_ZPL_CTIME(zfsvfs), NULL,
3526 &ctime, sizeof (ctime));
3527 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3528 mtime, ctime, B_TRUE);
3532 * Do this after setting timestamps to prevent timestamp
3533 * update from toggling bit
3536 if (xoap && (mask & AT_XVATTR)) {
3539 * restore trimmed off masks
3540 * so that return masks can be set for caller.
3543 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3544 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3546 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3547 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3549 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3550 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3552 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3553 XVA_SET_REQ(xvap, XAT_NODUMP);
3555 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3556 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3558 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3559 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3562 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3563 ASSERT(vp->v_type == VREG);
3565 zfs_xvattr_set(zp, xvap, tx);
3569 zfs_fuid_sync(zfsvfs, tx);
3572 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3574 mutex_exit(&zp->z_lock);
3575 if (mask & (AT_UID|AT_GID|AT_MODE))
3576 mutex_exit(&zp->z_acl_lock);
3579 if (mask & (AT_UID|AT_GID|AT_MODE))
3580 mutex_exit(&attrzp->z_acl_lock);
3581 mutex_exit(&attrzp->z_lock);
3584 if (err == 0 && attrzp) {
3585 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3591 VN_RELE(ZTOV(attrzp));
3597 zfs_fuid_info_free(fuidp);
3603 if (err == ERESTART)
3606 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3611 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3612 zil_commit(zilog, 0);
3618 typedef struct zfs_zlock {
3619 krwlock_t *zl_rwlock; /* lock we acquired */
3620 znode_t *zl_znode; /* znode we held */
3621 struct zfs_zlock *zl_next; /* next in list */
3625 * Drop locks and release vnodes that were held by zfs_rename_lock().
3628 zfs_rename_unlock(zfs_zlock_t **zlpp)
3632 while ((zl = *zlpp) != NULL) {
3633 if (zl->zl_znode != NULL)
3634 VN_RELE(ZTOV(zl->zl_znode));
3635 rw_exit(zl->zl_rwlock);
3636 *zlpp = zl->zl_next;
3637 kmem_free(zl, sizeof (*zl));
3642 * Search back through the directory tree, using the ".." entries.
3643 * Lock each directory in the chain to prevent concurrent renames.
3644 * Fail any attempt to move a directory into one of its own descendants.
3645 * XXX - z_parent_lock can overlap with map or grow locks
3648 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3652 uint64_t rootid = zp->z_zfsvfs->z_root;
3653 uint64_t oidp = zp->z_id;
3654 krwlock_t *rwlp = &szp->z_parent_lock;
3655 krw_t rw = RW_WRITER;
3658 * First pass write-locks szp and compares to zp->z_id.
3659 * Later passes read-lock zp and compare to zp->z_parent.
3662 if (!rw_tryenter(rwlp, rw)) {
3664 * Another thread is renaming in this path.
3665 * Note that if we are a WRITER, we don't have any
3666 * parent_locks held yet.
3668 if (rw == RW_READER && zp->z_id > szp->z_id) {
3670 * Drop our locks and restart
3672 zfs_rename_unlock(&zl);
3676 rwlp = &szp->z_parent_lock;
3681 * Wait for other thread to drop its locks
3687 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3688 zl->zl_rwlock = rwlp;
3689 zl->zl_znode = NULL;
3690 zl->zl_next = *zlpp;
3693 if (oidp == szp->z_id) /* We're a descendant of szp */
3694 return (SET_ERROR(EINVAL));
3696 if (oidp == rootid) /* We've hit the top */
3699 if (rw == RW_READER) { /* i.e. not the first pass */
3700 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3705 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3706 &oidp, sizeof (oidp));
3707 rwlp = &zp->z_parent_lock;
3710 } while (zp->z_id != sdzp->z_id);
3716 * Move an entry from the provided source directory to the target
3717 * directory. Change the entry name as indicated.
3719 * IN: sdvp - Source directory containing the "old entry".
3720 * snm - Old entry name.
3721 * tdvp - Target directory to contain the "new entry".
3722 * tnm - New entry name.
3723 * cr - credentials of caller.
3724 * ct - caller context
3725 * flags - case flags
3727 * RETURN: 0 on success, error code on failure.
3730 * sdvp,tdvp - ctime|mtime updated
3734 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3735 caller_context_t *ct, int flags)
3737 znode_t *tdzp, *szp, *tzp;
3738 znode_t *sdzp = VTOZ(sdvp);
3739 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs;
3742 zfs_dirlock_t *sdl, *tdl;
3745 int cmp, serr, terr;
3748 boolean_t waited = B_FALSE;
3751 ZFS_VERIFY_ZP(sdzp);
3752 zilog = zfsvfs->z_log;
3755 * Make sure we have the real vp for the target directory.
3757 if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3761 ZFS_VERIFY_ZP(tdzp);
3764 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3765 * ctldir appear to have the same v_vfsp.
3767 if (tdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) {
3769 return (SET_ERROR(EXDEV));
3772 if (zfsvfs->z_utf8 && u8_validate(tnm,
3773 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3775 return (SET_ERROR(EILSEQ));
3778 if (flags & FIGNORECASE)
3787 * This is to prevent the creation of links into attribute space
3788 * by renaming a linked file into/outof an attribute directory.
3789 * See the comment in zfs_link() for why this is considered bad.
3791 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3793 return (SET_ERROR(EINVAL));
3797 * Lock source and target directory entries. To prevent deadlock,
3798 * a lock ordering must be defined. We lock the directory with
3799 * the smallest object id first, or if it's a tie, the one with
3800 * the lexically first name.
3802 if (sdzp->z_id < tdzp->z_id) {
3804 } else if (sdzp->z_id > tdzp->z_id) {
3808 * First compare the two name arguments without
3809 * considering any case folding.
3811 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3813 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3814 ASSERT(error == 0 || !zfsvfs->z_utf8);
3817 * POSIX: "If the old argument and the new argument
3818 * both refer to links to the same existing file,
3819 * the rename() function shall return successfully
3820 * and perform no other action."
3826 * If the file system is case-folding, then we may
3827 * have some more checking to do. A case-folding file
3828 * system is either supporting mixed case sensitivity
3829 * access or is completely case-insensitive. Note
3830 * that the file system is always case preserving.
3832 * In mixed sensitivity mode case sensitive behavior
3833 * is the default. FIGNORECASE must be used to
3834 * explicitly request case insensitive behavior.
3836 * If the source and target names provided differ only
3837 * by case (e.g., a request to rename 'tim' to 'Tim'),
3838 * we will treat this as a special case in the
3839 * case-insensitive mode: as long as the source name
3840 * is an exact match, we will allow this to proceed as
3841 * a name-change request.
3843 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3844 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3845 flags & FIGNORECASE)) &&
3846 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3849 * case preserving rename request, require exact
3858 * If the source and destination directories are the same, we should
3859 * grab the z_name_lock of that directory only once.
3863 rw_enter(&sdzp->z_name_lock, RW_READER);
3867 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3868 ZEXISTS | zflg, NULL, NULL);
3869 terr = zfs_dirent_lock(&tdl,
3870 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3872 terr = zfs_dirent_lock(&tdl,
3873 tdzp, tnm, &tzp, zflg, NULL, NULL);
3874 serr = zfs_dirent_lock(&sdl,
3875 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3881 * Source entry invalid or not there.
3884 zfs_dirent_unlock(tdl);
3890 rw_exit(&sdzp->z_name_lock);
3893 * FreeBSD: In OpenSolaris they only check if rename source is
3894 * ".." here, because "." is handled in their lookup. This is
3895 * not the case for FreeBSD, so we check for "." explicitly.
3897 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0)
3898 serr = SET_ERROR(EINVAL);
3903 zfs_dirent_unlock(sdl);
3907 rw_exit(&sdzp->z_name_lock);
3909 if (strcmp(tnm, "..") == 0)
3910 terr = SET_ERROR(EINVAL);
3916 * Must have write access at the source to remove the old entry
3917 * and write access at the target to create the new entry.
3918 * Note that if target and source are the same, this can be
3919 * done in a single check.
3922 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3925 if (ZTOV(szp)->v_type == VDIR) {
3927 * Check to make sure rename is valid.
3928 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3930 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3935 * Does target exist?
3939 * Source and target must be the same type.
3941 if (ZTOV(szp)->v_type == VDIR) {
3942 if (ZTOV(tzp)->v_type != VDIR) {
3943 error = SET_ERROR(ENOTDIR);
3947 if (ZTOV(tzp)->v_type == VDIR) {
3948 error = SET_ERROR(EISDIR);
3953 * POSIX dictates that when the source and target
3954 * entries refer to the same file object, rename
3955 * must do nothing and exit without error.
3957 if (szp->z_id == tzp->z_id) {
3963 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3965 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3968 * notify the target directory if it is not the same
3969 * as source directory.
3972 vnevent_rename_dest_dir(tdvp, ct);
3975 tx = dmu_tx_create(zfsvfs->z_os);
3976 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3977 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3978 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3979 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3981 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3982 zfs_sa_upgrade_txholds(tx, tdzp);
3985 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3986 zfs_sa_upgrade_txholds(tx, tzp);
3989 zfs_sa_upgrade_txholds(tx, szp);
3990 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3991 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3994 zfs_rename_unlock(&zl);
3995 zfs_dirent_unlock(sdl);
3996 zfs_dirent_unlock(tdl);
3999 rw_exit(&sdzp->z_name_lock);
4004 if (error == ERESTART) {
4015 if (tzp) /* Attempt to remove the existing target */
4016 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
4019 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
4021 szp->z_pflags |= ZFS_AV_MODIFIED;
4023 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
4024 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
4027 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
4029 zfs_log_rename(zilog, tx, TX_RENAME |
4030 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
4031 sdl->dl_name, tdzp, tdl->dl_name, szp);
4034 * Update path information for the target vnode
4036 vn_renamepath(tdvp, ZTOV(szp), tnm,
4040 * At this point, we have successfully created
4041 * the target name, but have failed to remove
4042 * the source name. Since the create was done
4043 * with the ZRENAMING flag, there are
4044 * complications; for one, the link count is
4045 * wrong. The easiest way to deal with this
4046 * is to remove the newly created target, and
4047 * return the original error. This must
4048 * succeed; fortunately, it is very unlikely to
4049 * fail, since we just created it.
4051 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
4052 ZRENAMING, NULL), ==, 0);
4055 #ifdef FREEBSD_NAMECACHE
4059 cache_purge(ZTOV(szp));
4061 cache_purge(ZTOV(tzp));
4069 zfs_rename_unlock(&zl);
4071 zfs_dirent_unlock(sdl);
4072 zfs_dirent_unlock(tdl);
4075 rw_exit(&sdzp->z_name_lock);
4082 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4083 zil_commit(zilog, 0);
4091 * Insert the indicated symbolic reference entry into the directory.
4093 * IN: dvp - Directory to contain new symbolic link.
4094 * link - Name for new symlink entry.
4095 * vap - Attributes of new entry.
4096 * cr - credentials of caller.
4097 * ct - caller context
4098 * flags - case flags
4100 * RETURN: 0 on success, error code on failure.
4103 * dvp - ctime|mtime updated
4107 zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link,
4108 cred_t *cr, kthread_t *td)
4110 znode_t *zp, *dzp = VTOZ(dvp);
4113 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4115 uint64_t len = strlen(link);
4118 zfs_acl_ids_t acl_ids;
4119 boolean_t fuid_dirtied;
4120 uint64_t txtype = TX_SYMLINK;
4121 boolean_t waited = B_FALSE;
4124 ASSERT(vap->va_type == VLNK);
4128 zilog = zfsvfs->z_log;
4130 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
4131 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4133 return (SET_ERROR(EILSEQ));
4135 if (flags & FIGNORECASE)
4138 if (len > MAXPATHLEN) {
4140 return (SET_ERROR(ENAMETOOLONG));
4143 if ((error = zfs_acl_ids_create(dzp, 0,
4144 vap, cr, NULL, &acl_ids)) != 0) {
4150 * Attempt to lock directory; fail if entry already exists.
4152 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
4154 zfs_acl_ids_free(&acl_ids);
4159 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4160 zfs_acl_ids_free(&acl_ids);
4161 zfs_dirent_unlock(dl);
4166 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
4167 zfs_acl_ids_free(&acl_ids);
4168 zfs_dirent_unlock(dl);
4170 return (SET_ERROR(EDQUOT));
4172 tx = dmu_tx_create(zfsvfs->z_os);
4173 fuid_dirtied = zfsvfs->z_fuid_dirty;
4174 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
4175 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4176 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
4177 ZFS_SA_BASE_ATTR_SIZE + len);
4178 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
4179 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
4180 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
4181 acl_ids.z_aclp->z_acl_bytes);
4184 zfs_fuid_txhold(zfsvfs, tx);
4185 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4187 zfs_dirent_unlock(dl);
4188 if (error == ERESTART) {
4194 zfs_acl_ids_free(&acl_ids);
4201 * Create a new object for the symlink.
4202 * for version 4 ZPL datsets the symlink will be an SA attribute
4204 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
4207 zfs_fuid_sync(zfsvfs, tx);
4209 mutex_enter(&zp->z_lock);
4211 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
4214 zfs_sa_symlink(zp, link, len, tx);
4215 mutex_exit(&zp->z_lock);
4218 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
4219 &zp->z_size, sizeof (zp->z_size), tx);
4221 * Insert the new object into the directory.
4223 (void) zfs_link_create(dl, zp, tx, ZNEW);
4225 if (flags & FIGNORECASE)
4227 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
4230 zfs_acl_ids_free(&acl_ids);
4234 zfs_dirent_unlock(dl);
4236 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4237 zil_commit(zilog, 0);
4244 * Return, in the buffer contained in the provided uio structure,
4245 * the symbolic path referred to by vp.
4247 * IN: vp - vnode of symbolic link.
4248 * uio - structure to contain the link path.
4249 * cr - credentials of caller.
4250 * ct - caller context
4252 * OUT: uio - structure containing the link path.
4254 * RETURN: 0 on success, error code on failure.
4257 * vp - atime updated
4261 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
4263 znode_t *zp = VTOZ(vp);
4264 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4270 mutex_enter(&zp->z_lock);
4272 error = sa_lookup_uio(zp->z_sa_hdl,
4273 SA_ZPL_SYMLINK(zfsvfs), uio);
4275 error = zfs_sa_readlink(zp, uio);
4276 mutex_exit(&zp->z_lock);
4278 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4285 * Insert a new entry into directory tdvp referencing svp.
4287 * IN: tdvp - Directory to contain new entry.
4288 * svp - vnode of new entry.
4289 * name - name of new entry.
4290 * cr - credentials of caller.
4291 * ct - caller context
4293 * RETURN: 0 on success, error code on failure.
4296 * tdvp - ctime|mtime updated
4297 * svp - ctime updated
4301 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
4302 caller_context_t *ct, int flags)
4304 znode_t *dzp = VTOZ(tdvp);
4306 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4315 boolean_t waited = B_FALSE;
4317 ASSERT(tdvp->v_type == VDIR);
4321 zilog = zfsvfs->z_log;
4323 if (VOP_REALVP(svp, &realvp, ct) == 0)
4327 * POSIX dictates that we return EPERM here.
4328 * Better choices include ENOTSUP or EISDIR.
4330 if (svp->v_type == VDIR) {
4332 return (SET_ERROR(EPERM));
4339 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
4340 * ctldir appear to have the same v_vfsp.
4342 if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) {
4344 return (SET_ERROR(EXDEV));
4347 /* Prevent links to .zfs/shares files */
4349 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4350 &parent, sizeof (uint64_t))) != 0) {
4354 if (parent == zfsvfs->z_shares_dir) {
4356 return (SET_ERROR(EPERM));
4359 if (zfsvfs->z_utf8 && u8_validate(name,
4360 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4362 return (SET_ERROR(EILSEQ));
4364 if (flags & FIGNORECASE)
4368 * We do not support links between attributes and non-attributes
4369 * because of the potential security risk of creating links
4370 * into "normal" file space in order to circumvent restrictions
4371 * imposed in attribute space.
4373 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4375 return (SET_ERROR(EINVAL));
4379 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4380 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) {
4382 return (SET_ERROR(EPERM));
4385 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4392 * Attempt to lock directory; fail if entry already exists.
4394 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4400 tx = dmu_tx_create(zfsvfs->z_os);
4401 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4402 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4403 zfs_sa_upgrade_txholds(tx, szp);
4404 zfs_sa_upgrade_txholds(tx, dzp);
4405 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4407 zfs_dirent_unlock(dl);
4408 if (error == ERESTART) {
4419 error = zfs_link_create(dl, szp, tx, 0);
4422 uint64_t txtype = TX_LINK;
4423 if (flags & FIGNORECASE)
4425 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4430 zfs_dirent_unlock(dl);
4433 vnevent_link(svp, ct);
4436 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4437 zil_commit(zilog, 0);
4445 * zfs_null_putapage() is used when the file system has been force
4446 * unmounted. It just drops the pages.
4450 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4451 size_t *lenp, int flags, cred_t *cr)
4453 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4458 * Push a page out to disk, klustering if possible.
4460 * IN: vp - file to push page to.
4461 * pp - page to push.
4462 * flags - additional flags.
4463 * cr - credentials of caller.
4465 * OUT: offp - start of range pushed.
4466 * lenp - len of range pushed.
4468 * RETURN: 0 on success, error code on failure.
4470 * NOTE: callers must have locked the page to be pushed. On
4471 * exit, the page (and all other pages in the kluster) must be
4476 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4477 size_t *lenp, int flags, cred_t *cr)
4479 znode_t *zp = VTOZ(vp);
4480 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4482 u_offset_t off, koff;
4489 * If our blocksize is bigger than the page size, try to kluster
4490 * multiple pages so that we write a full block (thus avoiding
4491 * a read-modify-write).
4493 if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4494 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4495 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4496 ASSERT(koff <= zp->z_size);
4497 if (koff + klen > zp->z_size)
4498 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4499 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4501 ASSERT3U(btop(len), ==, btopr(len));
4504 * Can't push pages past end-of-file.
4506 if (off >= zp->z_size) {
4507 /* ignore all pages */
4510 } else if (off + len > zp->z_size) {
4511 int npages = btopr(zp->z_size - off);
4514 page_list_break(&pp, &trunc, npages);
4515 /* ignore pages past end of file */
4517 pvn_write_done(trunc, flags);
4518 len = zp->z_size - off;
4521 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4522 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4523 err = SET_ERROR(EDQUOT);
4527 tx = dmu_tx_create(zfsvfs->z_os);
4528 dmu_tx_hold_write(tx, zp->z_id, off, len);
4530 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4531 zfs_sa_upgrade_txholds(tx, zp);
4532 err = dmu_tx_assign(tx, TXG_NOWAIT);
4534 if (err == ERESTART) {
4543 if (zp->z_blksz <= PAGESIZE) {
4544 caddr_t va = zfs_map_page(pp, S_READ);
4545 ASSERT3U(len, <=, PAGESIZE);
4546 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4547 zfs_unmap_page(pp, va);
4549 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4553 uint64_t mtime[2], ctime[2];
4554 sa_bulk_attr_t bulk[3];
4557 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4559 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4561 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4563 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4565 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4570 pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4580 * Copy the portion of the file indicated from pages into the file.
4581 * The pages are stored in a page list attached to the files vnode.
4583 * IN: vp - vnode of file to push page data to.
4584 * off - position in file to put data.
4585 * len - amount of data to write.
4586 * flags - flags to control the operation.
4587 * cr - credentials of caller.
4588 * ct - caller context.
4590 * RETURN: 0 on success, error code on failure.
4593 * vp - ctime|mtime updated
4597 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4598 caller_context_t *ct)
4600 znode_t *zp = VTOZ(vp);
4601 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4613 * Align this request to the file block size in case we kluster.
4614 * XXX - this can result in pretty aggresive locking, which can
4615 * impact simultanious read/write access. One option might be
4616 * to break up long requests (len == 0) into block-by-block
4617 * operations to get narrower locking.
4619 blksz = zp->z_blksz;
4621 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4624 if (len > 0 && ISP2(blksz))
4625 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4631 * Search the entire vp list for pages >= io_off.
4633 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4634 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4637 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4639 if (off > zp->z_size) {
4640 /* past end of file */
4641 zfs_range_unlock(rl);
4646 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4648 for (off = io_off; io_off < off + len; io_off += io_len) {
4649 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4650 pp = page_lookup(vp, io_off,
4651 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4653 pp = page_lookup_nowait(vp, io_off,
4654 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4657 if (pp != NULL && pvn_getdirty(pp, flags)) {
4661 * Found a dirty page to push
4663 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4671 zfs_range_unlock(rl);
4672 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4673 zil_commit(zfsvfs->z_log, zp->z_id);
4681 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4683 znode_t *zp = VTOZ(vp);
4684 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4687 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4688 if (zp->z_sa_hdl == NULL) {
4690 * The fs has been unmounted, or we did a
4691 * suspend/resume and this file no longer exists.
4693 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4698 mutex_enter(&zp->z_lock);
4699 if (zp->z_unlinked) {
4701 * Fast path to recycle a vnode of a removed file.
4703 mutex_exit(&zp->z_lock);
4704 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4708 mutex_exit(&zp->z_lock);
4710 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4711 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4713 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4714 zfs_sa_upgrade_txholds(tx, zp);
4715 error = dmu_tx_assign(tx, TXG_WAIT);
4719 mutex_enter(&zp->z_lock);
4720 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4721 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4722 zp->z_atime_dirty = 0;
4723 mutex_exit(&zp->z_lock);
4727 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4732 * Bounds-check the seek operation.
4734 * IN: vp - vnode seeking within
4735 * ooff - old file offset
4736 * noffp - pointer to new file offset
4737 * ct - caller context
4739 * RETURN: 0 on success, EINVAL if new offset invalid.
4743 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4744 caller_context_t *ct)
4746 if (vp->v_type == VDIR)
4748 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4752 * Pre-filter the generic locking function to trap attempts to place
4753 * a mandatory lock on a memory mapped file.
4756 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4757 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4759 znode_t *zp = VTOZ(vp);
4760 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4766 * We are following the UFS semantics with respect to mapcnt
4767 * here: If we see that the file is mapped already, then we will
4768 * return an error, but we don't worry about races between this
4769 * function and zfs_map().
4771 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4773 return (SET_ERROR(EAGAIN));
4776 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4780 * If we can't find a page in the cache, we will create a new page
4781 * and fill it with file data. For efficiency, we may try to fill
4782 * multiple pages at once (klustering) to fill up the supplied page
4783 * list. Note that the pages to be filled are held with an exclusive
4784 * lock to prevent access by other threads while they are being filled.
4787 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4788 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4790 znode_t *zp = VTOZ(vp);
4791 page_t *pp, *cur_pp;
4792 objset_t *os = zp->z_zfsvfs->z_os;
4793 u_offset_t io_off, total;
4797 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4799 * We only have a single page, don't bother klustering
4803 pp = page_create_va(vp, io_off, io_len,
4804 PG_EXCL | PG_WAIT, seg, addr);
4807 * Try to find enough pages to fill the page list
4809 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4810 &io_len, off, plsz, 0);
4814 * The page already exists, nothing to do here.
4821 * Fill the pages in the kluster.
4824 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4827 ASSERT3U(io_off, ==, cur_pp->p_offset);
4828 va = zfs_map_page(cur_pp, S_WRITE);
4829 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4831 zfs_unmap_page(cur_pp, va);
4833 /* On error, toss the entire kluster */
4834 pvn_read_done(pp, B_ERROR);
4835 /* convert checksum errors into IO errors */
4837 err = SET_ERROR(EIO);
4840 cur_pp = cur_pp->p_next;
4844 * Fill in the page list array from the kluster starting
4845 * from the desired offset `off'.
4846 * NOTE: the page list will always be null terminated.
4848 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4849 ASSERT(pl == NULL || (*pl)->p_offset == off);
4855 * Return pointers to the pages for the file region [off, off + len]
4856 * in the pl array. If plsz is greater than len, this function may
4857 * also return page pointers from after the specified region
4858 * (i.e. the region [off, off + plsz]). These additional pages are
4859 * only returned if they are already in the cache, or were created as
4860 * part of a klustered read.
4862 * IN: vp - vnode of file to get data from.
4863 * off - position in file to get data from.
4864 * len - amount of data to retrieve.
4865 * plsz - length of provided page list.
4866 * seg - segment to obtain pages for.
4867 * addr - virtual address of fault.
4868 * rw - mode of created pages.
4869 * cr - credentials of caller.
4870 * ct - caller context.
4872 * OUT: protp - protection mode of created pages.
4873 * pl - list of pages created.
4875 * RETURN: 0 on success, error code on failure.
4878 * vp - atime updated
4882 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4883 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4884 enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4886 znode_t *zp = VTOZ(vp);
4887 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4891 /* we do our own caching, faultahead is unnecessary */
4894 else if (len > plsz)
4897 len = P2ROUNDUP(len, PAGESIZE);
4898 ASSERT(plsz >= len);
4907 * Loop through the requested range [off, off + len) looking
4908 * for pages. If we don't find a page, we will need to create
4909 * a new page and fill it with data from the file.
4912 if (*pl = page_lookup(vp, off, SE_SHARED))
4914 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4917 ASSERT3U((*pl)->p_offset, ==, off);
4921 ASSERT3U(len, >=, PAGESIZE);
4924 ASSERT3U(plsz, >=, PAGESIZE);
4931 * Fill out the page array with any pages already in the cache.
4934 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4941 * Release any pages we have previously locked.
4946 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4956 * Request a memory map for a section of a file. This code interacts
4957 * with common code and the VM system as follows:
4959 * - common code calls mmap(), which ends up in smmap_common()
4960 * - this calls VOP_MAP(), which takes you into (say) zfs
4961 * - zfs_map() calls as_map(), passing segvn_create() as the callback
4962 * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4963 * - zfs_addmap() updates z_mapcnt
4967 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4968 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4969 caller_context_t *ct)
4971 znode_t *zp = VTOZ(vp);
4972 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4973 segvn_crargs_t vn_a;
4979 if ((prot & PROT_WRITE) && (zp->z_pflags &
4980 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4982 return (SET_ERROR(EPERM));
4985 if ((prot & (PROT_READ | PROT_EXEC)) &&
4986 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4988 return (SET_ERROR(EACCES));
4991 if (vp->v_flag & VNOMAP) {
4993 return (SET_ERROR(ENOSYS));
4996 if (off < 0 || len > MAXOFFSET_T - off) {
4998 return (SET_ERROR(ENXIO));
5001 if (vp->v_type != VREG) {
5003 return (SET_ERROR(ENODEV));
5007 * If file is locked, disallow mapping.
5009 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
5011 return (SET_ERROR(EAGAIN));
5015 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
5023 vn_a.offset = (u_offset_t)off;
5024 vn_a.type = flags & MAP_TYPE;
5026 vn_a.maxprot = maxprot;
5029 vn_a.flags = flags & ~MAP_TYPE;
5031 vn_a.lgrp_mem_policy_flags = 0;
5033 error = as_map(as, *addrp, len, segvn_create, &vn_a);
5042 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5043 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
5044 caller_context_t *ct)
5046 uint64_t pages = btopr(len);
5048 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
5053 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
5054 * more accurate mtime for the associated file. Since we don't have a way of
5055 * detecting when the data was actually modified, we have to resort to
5056 * heuristics. If an explicit msync() is done, then we mark the mtime when the
5057 * last page is pushed. The problem occurs when the msync() call is omitted,
5058 * which by far the most common case:
5066 * putpage() via fsflush
5068 * If we wait until fsflush to come along, we can have a modification time that
5069 * is some arbitrary point in the future. In order to prevent this in the
5070 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
5075 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5076 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
5077 caller_context_t *ct)
5079 uint64_t pages = btopr(len);
5081 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
5082 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
5084 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
5085 vn_has_cached_data(vp))
5086 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
5092 * Free or allocate space in a file. Currently, this function only
5093 * supports the `F_FREESP' command. However, this command is somewhat
5094 * misnamed, as its functionality includes the ability to allocate as
5095 * well as free space.
5097 * IN: vp - vnode of file to free data in.
5098 * cmd - action to take (only F_FREESP supported).
5099 * bfp - section of file to free/alloc.
5100 * flag - current file open mode flags.
5101 * offset - current file offset.
5102 * cr - credentials of caller [UNUSED].
5103 * ct - caller context.
5105 * RETURN: 0 on success, error code on failure.
5108 * vp - ctime|mtime updated
5112 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
5113 offset_t offset, cred_t *cr, caller_context_t *ct)
5115 znode_t *zp = VTOZ(vp);
5116 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5123 if (cmd != F_FREESP) {
5125 return (SET_ERROR(EINVAL));
5128 if (error = convoff(vp, bfp, 0, offset)) {
5133 if (bfp->l_len < 0) {
5135 return (SET_ERROR(EINVAL));
5139 len = bfp->l_len; /* 0 means from off to end of file */
5141 error = zfs_freesp(zp, off, len, flag, TRUE);
5148 CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid));
5149 CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid));
5153 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
5155 znode_t *zp = VTOZ(vp);
5156 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5159 uint64_t object = zp->z_id;
5166 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
5167 &gen64, sizeof (uint64_t))) != 0) {
5172 gen = (uint32_t)gen64;
5174 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
5177 if (fidp->fid_len < size) {
5178 fidp->fid_len = size;
5180 return (SET_ERROR(ENOSPC));
5183 fidp->fid_len = size;
5186 zfid = (zfid_short_t *)fidp;
5188 zfid->zf_len = size;
5190 for (i = 0; i < sizeof (zfid->zf_object); i++)
5191 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
5193 /* Must have a non-zero generation number to distinguish from .zfs */
5196 for (i = 0; i < sizeof (zfid->zf_gen); i++)
5197 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
5199 if (size == LONG_FID_LEN) {
5200 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
5203 zlfid = (zfid_long_t *)fidp;
5205 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
5206 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
5208 /* XXX - this should be the generation number for the objset */
5209 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
5210 zlfid->zf_setgen[i] = 0;
5218 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
5219 caller_context_t *ct)
5231 case _PC_FILESIZEBITS:
5235 case _PC_XATTR_EXISTS:
5237 zfsvfs = zp->z_zfsvfs;
5241 error = zfs_dirent_lock(&dl, zp, "", &xzp,
5242 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
5244 zfs_dirent_unlock(dl);
5245 if (!zfs_dirempty(xzp))
5248 } else if (error == ENOENT) {
5250 * If there aren't extended attributes, it's the
5251 * same as having zero of them.
5258 case _PC_SATTR_ENABLED:
5259 case _PC_SATTR_EXISTS:
5260 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
5261 (vp->v_type == VREG || vp->v_type == VDIR);
5264 case _PC_ACCESS_FILTERING:
5265 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
5269 case _PC_ACL_ENABLED:
5270 *valp = _ACL_ACE_ENABLED;
5273 case _PC_MIN_HOLE_SIZE:
5274 *valp = (int)SPA_MINBLOCKSIZE;
5277 case _PC_TIMESTAMP_RESOLUTION:
5278 /* nanosecond timestamp resolution */
5282 case _PC_ACL_EXTENDED:
5290 case _PC_ACL_PATH_MAX:
5291 *valp = ACL_MAX_ENTRIES;
5295 return (EOPNOTSUPP);
5301 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5302 caller_context_t *ct)
5304 znode_t *zp = VTOZ(vp);
5305 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5307 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5311 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5319 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5320 caller_context_t *ct)
5322 znode_t *zp = VTOZ(vp);
5323 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5325 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5326 zilog_t *zilog = zfsvfs->z_log;
5331 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5333 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5334 zil_commit(zilog, 0);
5342 * The smallest read we may consider to loan out an arcbuf.
5343 * This must be a power of 2.
5345 int zcr_blksz_min = (1 << 10); /* 1K */
5347 * If set to less than the file block size, allow loaning out of an
5348 * arcbuf for a partial block read. This must be a power of 2.
5350 int zcr_blksz_max = (1 << 17); /* 128K */
5354 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5355 caller_context_t *ct)
5357 znode_t *zp = VTOZ(vp);
5358 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5359 int max_blksz = zfsvfs->z_max_blksz;
5360 uio_t *uio = &xuio->xu_uio;
5361 ssize_t size = uio->uio_resid;
5362 offset_t offset = uio->uio_loffset;
5367 int preamble, postamble;
5369 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5370 return (SET_ERROR(EINVAL));
5377 * Loan out an arc_buf for write if write size is bigger than
5378 * max_blksz, and the file's block size is also max_blksz.
5381 if (size < blksz || zp->z_blksz != blksz) {
5383 return (SET_ERROR(EINVAL));
5386 * Caller requests buffers for write before knowing where the
5387 * write offset might be (e.g. NFS TCP write).
5392 preamble = P2PHASE(offset, blksz);
5394 preamble = blksz - preamble;
5399 postamble = P2PHASE(size, blksz);
5402 fullblk = size / blksz;
5403 (void) dmu_xuio_init(xuio,
5404 (preamble != 0) + fullblk + (postamble != 0));
5405 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5406 int, postamble, int,
5407 (preamble != 0) + fullblk + (postamble != 0));
5410 * Have to fix iov base/len for partial buffers. They
5411 * currently represent full arc_buf's.
5414 /* data begins in the middle of the arc_buf */
5415 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5418 (void) dmu_xuio_add(xuio, abuf,
5419 blksz - preamble, preamble);
5422 for (i = 0; i < fullblk; i++) {
5423 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5426 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5430 /* data ends in the middle of the arc_buf */
5431 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5434 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5439 * Loan out an arc_buf for read if the read size is larger than
5440 * the current file block size. Block alignment is not
5441 * considered. Partial arc_buf will be loaned out for read.
5443 blksz = zp->z_blksz;
5444 if (blksz < zcr_blksz_min)
5445 blksz = zcr_blksz_min;
5446 if (blksz > zcr_blksz_max)
5447 blksz = zcr_blksz_max;
5448 /* avoid potential complexity of dealing with it */
5449 if (blksz > max_blksz) {
5451 return (SET_ERROR(EINVAL));
5454 maxsize = zp->z_size - uio->uio_loffset;
5458 if (size < blksz || vn_has_cached_data(vp)) {
5460 return (SET_ERROR(EINVAL));
5465 return (SET_ERROR(EINVAL));
5468 uio->uio_extflg = UIO_XUIO;
5469 XUIO_XUZC_RW(xuio) = ioflag;
5476 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5480 int ioflag = XUIO_XUZC_RW(xuio);
5482 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5484 i = dmu_xuio_cnt(xuio);
5486 abuf = dmu_xuio_arcbuf(xuio, i);
5488 * if abuf == NULL, it must be a write buffer
5489 * that has been returned in zfs_write().
5492 dmu_return_arcbuf(abuf);
5493 ASSERT(abuf || ioflag == UIO_WRITE);
5496 dmu_xuio_fini(xuio);
5501 * Predeclare these here so that the compiler assumes that
5502 * this is an "old style" function declaration that does
5503 * not include arguments => we won't get type mismatch errors
5504 * in the initializations that follow.
5506 static int zfs_inval();
5507 static int zfs_isdir();
5512 return (SET_ERROR(EINVAL));
5518 return (SET_ERROR(EISDIR));
5521 * Directory vnode operations template
5523 vnodeops_t *zfs_dvnodeops;
5524 const fs_operation_def_t zfs_dvnodeops_template[] = {
5525 VOPNAME_OPEN, { .vop_open = zfs_open },
5526 VOPNAME_CLOSE, { .vop_close = zfs_close },
5527 VOPNAME_READ, { .error = zfs_isdir },
5528 VOPNAME_WRITE, { .error = zfs_isdir },
5529 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5530 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5531 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5532 VOPNAME_ACCESS, { .vop_access = zfs_access },
5533 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5534 VOPNAME_CREATE, { .vop_create = zfs_create },
5535 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5536 VOPNAME_LINK, { .vop_link = zfs_link },
5537 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5538 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir },
5539 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5540 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5541 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink },
5542 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5543 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5544 VOPNAME_FID, { .vop_fid = zfs_fid },
5545 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5546 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5547 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5548 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5549 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5554 * Regular file vnode operations template
5556 vnodeops_t *zfs_fvnodeops;
5557 const fs_operation_def_t zfs_fvnodeops_template[] = {
5558 VOPNAME_OPEN, { .vop_open = zfs_open },
5559 VOPNAME_CLOSE, { .vop_close = zfs_close },
5560 VOPNAME_READ, { .vop_read = zfs_read },
5561 VOPNAME_WRITE, { .vop_write = zfs_write },
5562 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5563 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5564 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5565 VOPNAME_ACCESS, { .vop_access = zfs_access },
5566 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5567 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5568 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5569 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5570 VOPNAME_FID, { .vop_fid = zfs_fid },
5571 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5572 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock },
5573 VOPNAME_SPACE, { .vop_space = zfs_space },
5574 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage },
5575 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage },
5576 VOPNAME_MAP, { .vop_map = zfs_map },
5577 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap },
5578 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap },
5579 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5580 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5581 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5582 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5583 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf },
5584 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf },
5589 * Symbolic link vnode operations template
5591 vnodeops_t *zfs_symvnodeops;
5592 const fs_operation_def_t zfs_symvnodeops_template[] = {
5593 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5594 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5595 VOPNAME_ACCESS, { .vop_access = zfs_access },
5596 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5597 VOPNAME_READLINK, { .vop_readlink = zfs_readlink },
5598 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5599 VOPNAME_FID, { .vop_fid = zfs_fid },
5600 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5601 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5606 * special share hidden files vnode operations template
5608 vnodeops_t *zfs_sharevnodeops;
5609 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5610 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5611 VOPNAME_ACCESS, { .vop_access = zfs_access },
5612 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5613 VOPNAME_FID, { .vop_fid = zfs_fid },
5614 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5615 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5616 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5617 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5622 * Extended attribute directory vnode operations template
5624 * This template is identical to the directory vnodes
5625 * operation template except for restricted operations:
5629 * Note that there are other restrictions embedded in:
5630 * zfs_create() - restrict type to VREG
5631 * zfs_link() - no links into/out of attribute space
5632 * zfs_rename() - no moves into/out of attribute space
5634 vnodeops_t *zfs_xdvnodeops;
5635 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5636 VOPNAME_OPEN, { .vop_open = zfs_open },
5637 VOPNAME_CLOSE, { .vop_close = zfs_close },
5638 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5639 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5640 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5641 VOPNAME_ACCESS, { .vop_access = zfs_access },
5642 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5643 VOPNAME_CREATE, { .vop_create = zfs_create },
5644 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5645 VOPNAME_LINK, { .vop_link = zfs_link },
5646 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5647 VOPNAME_MKDIR, { .error = zfs_inval },
5648 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5649 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5650 VOPNAME_SYMLINK, { .error = zfs_inval },
5651 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5652 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5653 VOPNAME_FID, { .vop_fid = zfs_fid },
5654 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5655 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5656 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5657 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5658 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5663 * Error vnode operations template
5665 vnodeops_t *zfs_evnodeops;
5666 const fs_operation_def_t zfs_evnodeops_template[] = {
5667 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5668 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5674 ioflags(int ioflags)
5678 if (ioflags & IO_APPEND)
5680 if (ioflags & IO_NDELAY)
5682 if (ioflags & IO_SYNC)
5683 flags |= (FSYNC | FDSYNC | FRSYNC);
5689 zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage)
5691 znode_t *zp = VTOZ(vp);
5692 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5693 objset_t *os = zp->z_zfsvfs->z_os;
5694 vm_page_t mfirst, mlast, mreq;
5698 off_t startoff, endoff;
5700 vm_pindex_t reqstart, reqend;
5701 int pcount, lsize, reqsize, size;
5706 pcount = OFF_TO_IDX(round_page(count));
5708 object = mreq->object;
5711 KASSERT(vp->v_object == object, ("mismatching object"));
5713 if (pcount > 1 && zp->z_blksz > PAGESIZE) {
5714 startoff = rounddown(IDX_TO_OFF(mreq->pindex), zp->z_blksz);
5715 reqstart = OFF_TO_IDX(round_page(startoff));
5716 if (reqstart < m[0]->pindex)
5719 reqstart = reqstart - m[0]->pindex;
5720 endoff = roundup(IDX_TO_OFF(mreq->pindex) + PAGE_SIZE,
5722 reqend = OFF_TO_IDX(trunc_page(endoff)) - 1;
5723 if (reqend > m[pcount - 1]->pindex)
5724 reqend = m[pcount - 1]->pindex;
5725 reqsize = reqend - m[reqstart]->pindex + 1;
5726 KASSERT(reqstart <= reqpage && reqpage < reqstart + reqsize,
5727 ("reqpage beyond [reqstart, reqstart + reqsize[ bounds"));
5732 mfirst = m[reqstart];
5733 mlast = m[reqstart + reqsize - 1];
5735 zfs_vmobject_wlock(object);
5737 for (i = 0; i < reqstart; i++) {
5740 vm_page_unlock(m[i]);
5742 for (i = reqstart + reqsize; i < pcount; i++) {
5745 vm_page_unlock(m[i]);
5748 if (mreq->valid && reqsize == 1) {
5749 if (mreq->valid != VM_PAGE_BITS_ALL)
5750 vm_page_zero_invalid(mreq, TRUE);
5751 zfs_vmobject_wunlock(object);
5753 return (zfs_vm_pagerret_ok);
5756 PCPU_INC(cnt.v_vnodein);
5757 PCPU_ADD(cnt.v_vnodepgsin, reqsize);
5759 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) {
5760 for (i = reqstart; i < reqstart + reqsize; i++) {
5764 vm_page_unlock(m[i]);
5767 zfs_vmobject_wunlock(object);
5769 return (zfs_vm_pagerret_bad);
5773 if (IDX_TO_OFF(mlast->pindex) + lsize > object->un_pager.vnp.vnp_size)
5774 lsize = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mlast->pindex);
5776 zfs_vmobject_wunlock(object);
5778 for (i = reqstart; i < reqstart + reqsize; i++) {
5780 if (i == (reqstart + reqsize - 1))
5782 va = zfs_map_page(m[i], &sf);
5783 error = dmu_read(os, zp->z_id, IDX_TO_OFF(m[i]->pindex),
5784 size, va, DMU_READ_PREFETCH);
5785 if (size != PAGE_SIZE)
5786 bzero(va + size, PAGE_SIZE - size);
5792 zfs_vmobject_wlock(object);
5794 for (i = reqstart; i < reqstart + reqsize; i++) {
5796 m[i]->valid = VM_PAGE_BITS_ALL;
5797 KASSERT(m[i]->dirty == 0, ("zfs_getpages: page %p is dirty", m[i]));
5799 vm_page_readahead_finish(m[i]);
5802 zfs_vmobject_wunlock(object);
5804 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
5806 return (error ? zfs_vm_pagerret_error : zfs_vm_pagerret_ok);
5810 zfs_freebsd_getpages(ap)
5811 struct vop_getpages_args /* {
5816 vm_ooffset_t a_offset;
5820 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage));
5824 zfs_freebsd_bmap(ap)
5825 struct vop_bmap_args /* {
5828 struct bufobj **a_bop;
5835 if (ap->a_bop != NULL)
5836 *ap->a_bop = &ap->a_vp->v_bufobj;
5837 if (ap->a_bnp != NULL)
5838 *ap->a_bnp = ap->a_bn;
5839 if (ap->a_runp != NULL)
5841 if (ap->a_runb != NULL)
5848 zfs_freebsd_open(ap)
5849 struct vop_open_args /* {
5852 struct ucred *a_cred;
5853 struct thread *a_td;
5856 vnode_t *vp = ap->a_vp;
5857 znode_t *zp = VTOZ(vp);
5860 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL);
5862 vnode_create_vobject(vp, zp->z_size, ap->a_td);
5867 zfs_freebsd_close(ap)
5868 struct vop_close_args /* {
5871 struct ucred *a_cred;
5872 struct thread *a_td;
5876 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL));
5880 zfs_freebsd_ioctl(ap)
5881 struct vop_ioctl_args /* {
5891 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data,
5892 ap->a_fflag, ap->a_cred, NULL, NULL));
5896 zfs_freebsd_read(ap)
5897 struct vop_read_args /* {
5901 struct ucred *a_cred;
5905 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5910 zfs_freebsd_write(ap)
5911 struct vop_write_args /* {
5915 struct ucred *a_cred;
5919 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5924 zfs_freebsd_access(ap)
5925 struct vop_access_args /* {
5927 accmode_t a_accmode;
5928 struct ucred *a_cred;
5929 struct thread *a_td;
5932 vnode_t *vp = ap->a_vp;
5933 znode_t *zp = VTOZ(vp);
5938 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND,
5940 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND);
5942 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL);
5945 * VADMIN has to be handled by vaccess().
5948 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND);
5950 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
5951 zp->z_gid, accmode, ap->a_cred, NULL);
5956 * For VEXEC, ensure that at least one execute bit is set for
5959 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR &&
5960 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
5968 zfs_freebsd_lookup(ap)
5969 struct vop_lookup_args /* {
5970 struct vnode *a_dvp;
5971 struct vnode **a_vpp;
5972 struct componentname *a_cnp;
5975 struct componentname *cnp = ap->a_cnp;
5976 char nm[NAME_MAX + 1];
5978 ASSERT(cnp->cn_namelen < sizeof(nm));
5979 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm)));
5981 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop,
5982 cnp->cn_cred, cnp->cn_thread, 0));
5986 zfs_freebsd_create(ap)
5987 struct vop_create_args /* {
5988 struct vnode *a_dvp;
5989 struct vnode **a_vpp;
5990 struct componentname *a_cnp;
5991 struct vattr *a_vap;
5994 struct componentname *cnp = ap->a_cnp;
5995 vattr_t *vap = ap->a_vap;
5998 ASSERT(cnp->cn_flags & SAVENAME);
6000 vattr_init_mask(vap);
6001 mode = vap->va_mode & ALLPERMS;
6003 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode,
6004 ap->a_vpp, cnp->cn_cred, cnp->cn_thread));
6008 zfs_freebsd_remove(ap)
6009 struct vop_remove_args /* {
6010 struct vnode *a_dvp;
6012 struct componentname *a_cnp;
6016 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6018 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr,
6019 ap->a_cnp->cn_cred, NULL, 0));
6023 zfs_freebsd_mkdir(ap)
6024 struct vop_mkdir_args /* {
6025 struct vnode *a_dvp;
6026 struct vnode **a_vpp;
6027 struct componentname *a_cnp;
6028 struct vattr *a_vap;
6031 vattr_t *vap = ap->a_vap;
6033 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6035 vattr_init_mask(vap);
6037 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp,
6038 ap->a_cnp->cn_cred, NULL, 0, NULL));
6042 zfs_freebsd_rmdir(ap)
6043 struct vop_rmdir_args /* {
6044 struct vnode *a_dvp;
6046 struct componentname *a_cnp;
6049 struct componentname *cnp = ap->a_cnp;
6051 ASSERT(cnp->cn_flags & SAVENAME);
6053 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0));
6057 zfs_freebsd_readdir(ap)
6058 struct vop_readdir_args /* {
6061 struct ucred *a_cred;
6068 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag,
6069 ap->a_ncookies, ap->a_cookies));
6073 zfs_freebsd_fsync(ap)
6074 struct vop_fsync_args /* {
6077 struct thread *a_td;
6082 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL));
6086 zfs_freebsd_getattr(ap)
6087 struct vop_getattr_args /* {
6089 struct vattr *a_vap;
6090 struct ucred *a_cred;
6093 vattr_t *vap = ap->a_vap;
6099 xvap.xva_vattr = *vap;
6100 xvap.xva_vattr.va_mask |= AT_XVATTR;
6102 /* Convert chflags into ZFS-type flags. */
6103 /* XXX: what about SF_SETTABLE?. */
6104 XVA_SET_REQ(&xvap, XAT_IMMUTABLE);
6105 XVA_SET_REQ(&xvap, XAT_APPENDONLY);
6106 XVA_SET_REQ(&xvap, XAT_NOUNLINK);
6107 XVA_SET_REQ(&xvap, XAT_NODUMP);
6108 XVA_SET_REQ(&xvap, XAT_READONLY);
6109 XVA_SET_REQ(&xvap, XAT_ARCHIVE);
6110 XVA_SET_REQ(&xvap, XAT_SYSTEM);
6111 XVA_SET_REQ(&xvap, XAT_HIDDEN);
6112 XVA_SET_REQ(&xvap, XAT_REPARSE);
6113 XVA_SET_REQ(&xvap, XAT_OFFLINE);
6114 XVA_SET_REQ(&xvap, XAT_SPARSE);
6116 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL);
6120 /* Convert ZFS xattr into chflags. */
6121 #define FLAG_CHECK(fflag, xflag, xfield) do { \
6122 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \
6123 fflags |= (fflag); \
6125 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE,
6126 xvap.xva_xoptattrs.xoa_immutable);
6127 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY,
6128 xvap.xva_xoptattrs.xoa_appendonly);
6129 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK,
6130 xvap.xva_xoptattrs.xoa_nounlink);
6131 FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE,
6132 xvap.xva_xoptattrs.xoa_archive);
6133 FLAG_CHECK(UF_NODUMP, XAT_NODUMP,
6134 xvap.xva_xoptattrs.xoa_nodump);
6135 FLAG_CHECK(UF_READONLY, XAT_READONLY,
6136 xvap.xva_xoptattrs.xoa_readonly);
6137 FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM,
6138 xvap.xva_xoptattrs.xoa_system);
6139 FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN,
6140 xvap.xva_xoptattrs.xoa_hidden);
6141 FLAG_CHECK(UF_REPARSE, XAT_REPARSE,
6142 xvap.xva_xoptattrs.xoa_reparse);
6143 FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE,
6144 xvap.xva_xoptattrs.xoa_offline);
6145 FLAG_CHECK(UF_SPARSE, XAT_SPARSE,
6146 xvap.xva_xoptattrs.xoa_sparse);
6149 *vap = xvap.xva_vattr;
6150 vap->va_flags = fflags;
6155 zfs_freebsd_setattr(ap)
6156 struct vop_setattr_args /* {
6158 struct vattr *a_vap;
6159 struct ucred *a_cred;
6162 vnode_t *vp = ap->a_vp;
6163 vattr_t *vap = ap->a_vap;
6164 cred_t *cred = ap->a_cred;
6169 vattr_init_mask(vap);
6170 vap->va_mask &= ~AT_NOSET;
6173 xvap.xva_vattr = *vap;
6175 zflags = VTOZ(vp)->z_pflags;
6177 if (vap->va_flags != VNOVAL) {
6178 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs;
6181 if (zfsvfs->z_use_fuids == B_FALSE)
6182 return (EOPNOTSUPP);
6184 fflags = vap->va_flags;
6187 * We need to figure out whether it makes sense to allow
6188 * UF_REPARSE through, since we don't really have other
6189 * facilities to handle reparse points and zfs_setattr()
6190 * doesn't currently allow setting that attribute anyway.
6192 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE|
6193 UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE|
6194 UF_OFFLINE|UF_SPARSE)) != 0)
6195 return (EOPNOTSUPP);
6197 * Unprivileged processes are not permitted to unset system
6198 * flags, or modify flags if any system flags are set.
6199 * Privileged non-jail processes may not modify system flags
6200 * if securelevel > 0 and any existing system flags are set.
6201 * Privileged jail processes behave like privileged non-jail
6202 * processes if the security.jail.chflags_allowed sysctl is
6203 * is non-zero; otherwise, they behave like unprivileged
6206 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 ||
6207 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) {
6209 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6210 error = securelevel_gt(cred, 0);
6216 * Callers may only modify the file flags on objects they
6217 * have VADMIN rights for.
6219 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0)
6222 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6226 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) {
6231 #define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \
6232 if (((fflags & (fflag)) && !(zflags & (zflag))) || \
6233 ((zflags & (zflag)) && !(fflags & (fflag)))) { \
6234 XVA_SET_REQ(&xvap, (xflag)); \
6235 (xfield) = ((fflags & (fflag)) != 0); \
6238 /* Convert chflags into ZFS-type flags. */
6239 /* XXX: what about SF_SETTABLE?. */
6240 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE,
6241 xvap.xva_xoptattrs.xoa_immutable);
6242 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY,
6243 xvap.xva_xoptattrs.xoa_appendonly);
6244 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK,
6245 xvap.xva_xoptattrs.xoa_nounlink);
6246 FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE,
6247 xvap.xva_xoptattrs.xoa_archive);
6248 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP,
6249 xvap.xva_xoptattrs.xoa_nodump);
6250 FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY,
6251 xvap.xva_xoptattrs.xoa_readonly);
6252 FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM,
6253 xvap.xva_xoptattrs.xoa_system);
6254 FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN,
6255 xvap.xva_xoptattrs.xoa_hidden);
6256 FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE,
6257 xvap.xva_xoptattrs.xoa_hidden);
6258 FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE,
6259 xvap.xva_xoptattrs.xoa_offline);
6260 FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE,
6261 xvap.xva_xoptattrs.xoa_sparse);
6264 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL));
6268 zfs_freebsd_rename(ap)
6269 struct vop_rename_args /* {
6270 struct vnode *a_fdvp;
6271 struct vnode *a_fvp;
6272 struct componentname *a_fcnp;
6273 struct vnode *a_tdvp;
6274 struct vnode *a_tvp;
6275 struct componentname *a_tcnp;
6278 vnode_t *fdvp = ap->a_fdvp;
6279 vnode_t *fvp = ap->a_fvp;
6280 vnode_t *tdvp = ap->a_tdvp;
6281 vnode_t *tvp = ap->a_tvp;
6284 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART));
6285 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART));
6288 * Check for cross-device rename.
6290 if ((fdvp->v_mount != tdvp->v_mount) ||
6291 (tvp && (fdvp->v_mount != tvp->v_mount)))
6294 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp,
6295 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0);
6309 zfs_freebsd_symlink(ap)
6310 struct vop_symlink_args /* {
6311 struct vnode *a_dvp;
6312 struct vnode **a_vpp;
6313 struct componentname *a_cnp;
6314 struct vattr *a_vap;
6318 struct componentname *cnp = ap->a_cnp;
6319 vattr_t *vap = ap->a_vap;
6321 ASSERT(cnp->cn_flags & SAVENAME);
6323 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */
6324 vattr_init_mask(vap);
6326 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap,
6327 ap->a_target, cnp->cn_cred, cnp->cn_thread));
6331 zfs_freebsd_readlink(ap)
6332 struct vop_readlink_args /* {
6335 struct ucred *a_cred;
6339 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL));
6343 zfs_freebsd_link(ap)
6344 struct vop_link_args /* {
6345 struct vnode *a_tdvp;
6347 struct componentname *a_cnp;
6350 struct componentname *cnp = ap->a_cnp;
6351 vnode_t *vp = ap->a_vp;
6352 vnode_t *tdvp = ap->a_tdvp;
6354 if (tdvp->v_mount != vp->v_mount)
6357 ASSERT(cnp->cn_flags & SAVENAME);
6359 return (zfs_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0));
6363 zfs_freebsd_inactive(ap)
6364 struct vop_inactive_args /* {
6366 struct thread *a_td;
6369 vnode_t *vp = ap->a_vp;
6371 zfs_inactive(vp, ap->a_td->td_ucred, NULL);
6376 zfs_freebsd_reclaim(ap)
6377 struct vop_reclaim_args /* {
6379 struct thread *a_td;
6382 vnode_t *vp = ap->a_vp;
6383 znode_t *zp = VTOZ(vp);
6384 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
6388 /* Destroy the vm object and flush associated pages. */
6389 vnode_destroy_vobject(vp);
6392 * z_teardown_inactive_lock protects from a race with
6393 * zfs_znode_dmu_fini in zfsvfs_teardown during
6396 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
6397 if (zp->z_sa_hdl == NULL)
6401 rw_exit(&zfsvfs->z_teardown_inactive_lock);
6409 struct vop_fid_args /* {
6415 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL));
6419 zfs_freebsd_pathconf(ap)
6420 struct vop_pathconf_args /* {
6423 register_t *a_retval;
6429 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL);
6431 *ap->a_retval = val;
6432 else if (error == EOPNOTSUPP)
6433 error = vop_stdpathconf(ap);
6438 zfs_freebsd_fifo_pathconf(ap)
6439 struct vop_pathconf_args /* {
6442 register_t *a_retval;
6446 switch (ap->a_name) {
6447 case _PC_ACL_EXTENDED:
6449 case _PC_ACL_PATH_MAX:
6450 case _PC_MAC_PRESENT:
6451 return (zfs_freebsd_pathconf(ap));
6453 return (fifo_specops.vop_pathconf(ap));
6458 * FreeBSD's extended attributes namespace defines file name prefix for ZFS'
6459 * extended attribute name:
6462 * system freebsd:system:
6463 * user (none, can be used to access ZFS fsattr(5) attributes
6464 * created on Solaris)
6467 zfs_create_attrname(int attrnamespace, const char *name, char *attrname,
6470 const char *namespace, *prefix, *suffix;
6472 /* We don't allow '/' character in attribute name. */
6473 if (strchr(name, '/') != NULL)
6475 /* We don't allow attribute names that start with "freebsd:" string. */
6476 if (strncmp(name, "freebsd:", 8) == 0)
6479 bzero(attrname, size);
6481 switch (attrnamespace) {
6482 case EXTATTR_NAMESPACE_USER:
6484 prefix = "freebsd:";
6485 namespace = EXTATTR_NAMESPACE_USER_STRING;
6489 * This is the default namespace by which we can access all
6490 * attributes created on Solaris.
6492 prefix = namespace = suffix = "";
6495 case EXTATTR_NAMESPACE_SYSTEM:
6496 prefix = "freebsd:";
6497 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING;
6500 case EXTATTR_NAMESPACE_EMPTY:
6504 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix,
6506 return (ENAMETOOLONG);
6512 * Vnode operating to retrieve a named extended attribute.
6515 zfs_getextattr(struct vop_getextattr_args *ap)
6518 IN struct vnode *a_vp;
6519 IN int a_attrnamespace;
6520 IN const char *a_name;
6521 INOUT struct uio *a_uio;
6523 IN struct ucred *a_cred;
6524 IN struct thread *a_td;
6528 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6529 struct thread *td = ap->a_td;
6530 struct nameidata nd;
6533 vnode_t *xvp = NULL, *vp;
6536 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6537 ap->a_cred, ap->a_td, VREAD);
6541 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6548 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6556 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6558 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL);
6560 NDFREE(&nd, NDF_ONLY_PNBUF);
6563 if (error == ENOENT)
6568 if (ap->a_size != NULL) {
6569 error = VOP_GETATTR(vp, &va, ap->a_cred);
6571 *ap->a_size = (size_t)va.va_size;
6572 } else if (ap->a_uio != NULL)
6573 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6576 vn_close(vp, flags, ap->a_cred, td);
6583 * Vnode operation to remove a named attribute.
6586 zfs_deleteextattr(struct vop_deleteextattr_args *ap)
6589 IN struct vnode *a_vp;
6590 IN int a_attrnamespace;
6591 IN const char *a_name;
6592 IN struct ucred *a_cred;
6593 IN struct thread *a_td;
6597 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6598 struct thread *td = ap->a_td;
6599 struct nameidata nd;
6602 vnode_t *xvp = NULL, *vp;
6605 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6606 ap->a_cred, ap->a_td, VWRITE);
6610 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6617 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6624 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF,
6625 UIO_SYSSPACE, attrname, xvp, td);
6628 NDFREE(&nd, NDF_ONLY_PNBUF);
6631 if (error == ENOENT)
6635 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
6638 if (vp == nd.ni_dvp)
6648 * Vnode operation to set a named attribute.
6651 zfs_setextattr(struct vop_setextattr_args *ap)
6654 IN struct vnode *a_vp;
6655 IN int a_attrnamespace;
6656 IN const char *a_name;
6657 INOUT struct uio *a_uio;
6658 IN struct ucred *a_cred;
6659 IN struct thread *a_td;
6663 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6664 struct thread *td = ap->a_td;
6665 struct nameidata nd;
6668 vnode_t *xvp = NULL, *vp;
6671 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6672 ap->a_cred, ap->a_td, VWRITE);
6676 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6683 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6684 LOOKUP_XATTR | CREATE_XATTR_DIR);
6690 flags = FFLAGS(O_WRONLY | O_CREAT);
6691 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6693 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL);
6695 NDFREE(&nd, NDF_ONLY_PNBUF);
6703 error = VOP_SETATTR(vp, &va, ap->a_cred);
6705 VOP_WRITE(vp, ap->a_uio, IO_UNIT | IO_SYNC, ap->a_cred);
6708 vn_close(vp, flags, ap->a_cred, td);
6715 * Vnode operation to retrieve extended attributes on a vnode.
6718 zfs_listextattr(struct vop_listextattr_args *ap)
6721 IN struct vnode *a_vp;
6722 IN int a_attrnamespace;
6723 INOUT struct uio *a_uio;
6725 IN struct ucred *a_cred;
6726 IN struct thread *a_td;
6730 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6731 struct thread *td = ap->a_td;
6732 struct nameidata nd;
6733 char attrprefix[16];
6734 u_char dirbuf[sizeof(struct dirent)];
6737 struct uio auio, *uio = ap->a_uio;
6738 size_t *sizep = ap->a_size;
6740 vnode_t *xvp = NULL, *vp;
6741 int done, error, eof, pos;
6743 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6744 ap->a_cred, ap->a_td, VREAD);
6748 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix,
6749 sizeof(attrprefix));
6752 plen = strlen(attrprefix);
6759 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6764 * ENOATTR means that the EA directory does not yet exist,
6765 * i.e. there are no extended attributes there.
6767 if (error == ENOATTR)
6772 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED,
6773 UIO_SYSSPACE, ".", xvp, td);
6776 NDFREE(&nd, NDF_ONLY_PNBUF);
6782 auio.uio_iov = &aiov;
6783 auio.uio_iovcnt = 1;
6784 auio.uio_segflg = UIO_SYSSPACE;
6786 auio.uio_rw = UIO_READ;
6787 auio.uio_offset = 0;
6792 aiov.iov_base = (void *)dirbuf;
6793 aiov.iov_len = sizeof(dirbuf);
6794 auio.uio_resid = sizeof(dirbuf);
6795 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL);
6796 done = sizeof(dirbuf) - auio.uio_resid;
6799 for (pos = 0; pos < done;) {
6800 dp = (struct dirent *)(dirbuf + pos);
6801 pos += dp->d_reclen;
6803 * XXX: Temporarily we also accept DT_UNKNOWN, as this
6804 * is what we get when attribute was created on Solaris.
6806 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN)
6808 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0)
6810 else if (strncmp(dp->d_name, attrprefix, plen) != 0)
6812 nlen = dp->d_namlen - plen;
6815 else if (uio != NULL) {
6817 * Format of extattr name entry is one byte for
6818 * length and the rest for name.
6820 error = uiomove(&nlen, 1, uio->uio_rw, uio);
6822 error = uiomove(dp->d_name + plen, nlen,
6829 } while (!eof && error == 0);
6838 zfs_freebsd_getacl(ap)
6839 struct vop_getacl_args /* {
6848 vsecattr_t vsecattr;
6850 if (ap->a_type != ACL_TYPE_NFS4)
6853 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
6854 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL))
6857 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt);
6858 if (vsecattr.vsa_aclentp != NULL)
6859 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz);
6865 zfs_freebsd_setacl(ap)
6866 struct vop_setacl_args /* {
6875 vsecattr_t vsecattr;
6876 int aclbsize; /* size of acl list in bytes */
6879 if (ap->a_type != ACL_TYPE_NFS4)
6882 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES)
6886 * With NFSv4 ACLs, chmod(2) may need to add additional entries,
6887 * splitting every entry into two and appending "canonical six"
6888 * entries at the end. Don't allow for setting an ACL that would
6889 * cause chmod(2) to run out of ACL entries.
6891 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES)
6894 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR);
6898 vsecattr.vsa_mask = VSA_ACE;
6899 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t);
6900 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP);
6901 aaclp = vsecattr.vsa_aclentp;
6902 vsecattr.vsa_aclentsz = aclbsize;
6904 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp);
6905 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL);
6906 kmem_free(aaclp, aclbsize);
6912 zfs_freebsd_aclcheck(ap)
6913 struct vop_aclcheck_args /* {
6922 return (EOPNOTSUPP);
6925 struct vop_vector zfs_vnodeops;
6926 struct vop_vector zfs_fifoops;
6927 struct vop_vector zfs_shareops;
6929 struct vop_vector zfs_vnodeops = {
6930 .vop_default = &default_vnodeops,
6931 .vop_inactive = zfs_freebsd_inactive,
6932 .vop_reclaim = zfs_freebsd_reclaim,
6933 .vop_access = zfs_freebsd_access,
6934 #ifdef FREEBSD_NAMECACHE
6935 .vop_lookup = vfs_cache_lookup,
6936 .vop_cachedlookup = zfs_freebsd_lookup,
6938 .vop_lookup = zfs_freebsd_lookup,
6940 .vop_getattr = zfs_freebsd_getattr,
6941 .vop_setattr = zfs_freebsd_setattr,
6942 .vop_create = zfs_freebsd_create,
6943 .vop_mknod = zfs_freebsd_create,
6944 .vop_mkdir = zfs_freebsd_mkdir,
6945 .vop_readdir = zfs_freebsd_readdir,
6946 .vop_fsync = zfs_freebsd_fsync,
6947 .vop_open = zfs_freebsd_open,
6948 .vop_close = zfs_freebsd_close,
6949 .vop_rmdir = zfs_freebsd_rmdir,
6950 .vop_ioctl = zfs_freebsd_ioctl,
6951 .vop_link = zfs_freebsd_link,
6952 .vop_symlink = zfs_freebsd_symlink,
6953 .vop_readlink = zfs_freebsd_readlink,
6954 .vop_read = zfs_freebsd_read,
6955 .vop_write = zfs_freebsd_write,
6956 .vop_remove = zfs_freebsd_remove,
6957 .vop_rename = zfs_freebsd_rename,
6958 .vop_pathconf = zfs_freebsd_pathconf,
6959 .vop_bmap = zfs_freebsd_bmap,
6960 .vop_fid = zfs_freebsd_fid,
6961 .vop_getextattr = zfs_getextattr,
6962 .vop_deleteextattr = zfs_deleteextattr,
6963 .vop_setextattr = zfs_setextattr,
6964 .vop_listextattr = zfs_listextattr,
6965 .vop_getacl = zfs_freebsd_getacl,
6966 .vop_setacl = zfs_freebsd_setacl,
6967 .vop_aclcheck = zfs_freebsd_aclcheck,
6968 .vop_getpages = zfs_freebsd_getpages,
6971 struct vop_vector zfs_fifoops = {
6972 .vop_default = &fifo_specops,
6973 .vop_fsync = zfs_freebsd_fsync,
6974 .vop_access = zfs_freebsd_access,
6975 .vop_getattr = zfs_freebsd_getattr,
6976 .vop_inactive = zfs_freebsd_inactive,
6977 .vop_read = VOP_PANIC,
6978 .vop_reclaim = zfs_freebsd_reclaim,
6979 .vop_setattr = zfs_freebsd_setattr,
6980 .vop_write = VOP_PANIC,
6981 .vop_pathconf = zfs_freebsd_fifo_pathconf,
6982 .vop_fid = zfs_freebsd_fid,
6983 .vop_getacl = zfs_freebsd_getacl,
6984 .vop_setacl = zfs_freebsd_setacl,
6985 .vop_aclcheck = zfs_freebsd_aclcheck,
6989 * special share hidden files vnode operations template
6991 struct vop_vector zfs_shareops = {
6992 .vop_default = &default_vnodeops,
6993 .vop_access = zfs_freebsd_access,
6994 .vop_inactive = zfs_freebsd_inactive,
6995 .vop_reclaim = zfs_freebsd_reclaim,
6996 .vop_fid = zfs_freebsd_fid,
6997 .vop_pathconf = zfs_freebsd_pathconf,