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) 2012, 2014 by Delphix. All rights reserved.
24 * Copyright 2013 Nexenta Systems, Inc. All rights reserved.
27 /* Portions Copyright 2007 Jeremy Teo */
28 /* Portions Copyright 2010 Robert Milkowski */
30 #include <sys/types.h>
31 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/sysmacros.h>
35 #include <sys/resource.h>
38 #include <sys/vnode.h>
42 #include <sys/taskq.h>
44 #include <sys/atomic.h>
45 #include <sys/namei.h>
47 #include <sys/cmn_err.h>
48 #include <sys/errno.h>
49 #include <sys/unistd.h>
50 #include <sys/zfs_dir.h>
51 #include <sys/zfs_ioctl.h>
52 #include <sys/fs/zfs.h>
54 #include <sys/dmu_objset.h>
60 #include <sys/dirent.h>
61 #include <sys/policy.h>
62 #include <sys/sunddi.h>
63 #include <sys/filio.h>
65 #include <sys/zfs_ctldir.h>
66 #include <sys/zfs_fuid.h>
67 #include <sys/zfs_sa.h>
69 #include <sys/zfs_rlock.h>
70 #include <sys/extdirent.h>
71 #include <sys/kidmap.h>
74 #include <sys/sched.h>
76 #include <vm/vm_param.h>
77 #include <vm/vm_pageout.h>
82 * Each vnode op performs some logical unit of work. To do this, the ZPL must
83 * properly lock its in-core state, create a DMU transaction, do the work,
84 * record this work in the intent log (ZIL), commit the DMU transaction,
85 * and wait for the intent log to commit if it is a synchronous operation.
86 * Moreover, the vnode ops must work in both normal and log replay context.
87 * The ordering of events is important to avoid deadlocks and references
88 * to freed memory. The example below illustrates the following Big Rules:
90 * (1) A check must be made in each zfs thread for a mounted file system.
91 * This is done avoiding races using ZFS_ENTER(zfsvfs).
92 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
93 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
94 * can return EIO from the calling function.
96 * (2) VN_RELE() should always be the last thing except for zil_commit()
97 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
98 * First, if it's the last reference, the vnode/znode
99 * can be freed, so the zp may point to freed memory. Second, the last
100 * reference will call zfs_zinactive(), which may induce a lot of work --
101 * pushing cached pages (which acquires range locks) and syncing out
102 * cached atime changes. Third, zfs_zinactive() may require a new tx,
103 * which could deadlock the system if you were already holding one.
104 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
106 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
107 * as they can span dmu_tx_assign() calls.
109 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
110 * dmu_tx_assign(). This is critical because we don't want to block
111 * while holding locks.
113 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
114 * reduces lock contention and CPU usage when we must wait (note that if
115 * throughput is constrained by the storage, nearly every transaction
118 * Note, in particular, that if a lock is sometimes acquired before
119 * the tx assigns, and sometimes after (e.g. z_lock), then failing
120 * to use a non-blocking assign can deadlock the system. The scenario:
122 * Thread A has grabbed a lock before calling dmu_tx_assign().
123 * Thread B is in an already-assigned tx, and blocks for this lock.
124 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
125 * forever, because the previous txg can't quiesce until B's tx commits.
127 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
128 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
129 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
130 * to indicate that this operation has already called dmu_tx_wait().
131 * This will ensure that we don't retry forever, waiting a short bit
134 * (5) If the operation succeeded, generate the intent log entry for it
135 * before dropping locks. This ensures that the ordering of events
136 * in the intent log matches the order in which they actually occurred.
137 * During ZIL replay the zfs_log_* functions will update the sequence
138 * number to indicate the zil transaction has replayed.
140 * (6) At the end of each vnode op, the DMU tx must always commit,
141 * regardless of whether there were any errors.
143 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
144 * to ensure that synchronous semantics are provided when necessary.
146 * In general, this is how things should be ordered in each vnode op:
148 * ZFS_ENTER(zfsvfs); // exit if unmounted
150 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD())
151 * rw_enter(...); // grab any other locks you need
152 * tx = dmu_tx_create(...); // get DMU tx
153 * dmu_tx_hold_*(); // hold each object you might modify
154 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
156 * rw_exit(...); // drop locks
157 * zfs_dirent_unlock(dl); // unlock directory entry
158 * VN_RELE(...); // release held vnodes
159 * if (error == ERESTART) {
165 * dmu_tx_abort(tx); // abort DMU tx
166 * ZFS_EXIT(zfsvfs); // finished in zfs
167 * return (error); // really out of space
169 * error = do_real_work(); // do whatever this VOP does
171 * zfs_log_*(...); // on success, make ZIL entry
172 * dmu_tx_commit(tx); // commit DMU tx -- error or not
173 * rw_exit(...); // drop locks
174 * zfs_dirent_unlock(dl); // unlock directory entry
175 * VN_RELE(...); // release held vnodes
176 * zil_commit(zilog, foid); // synchronous when necessary
177 * ZFS_EXIT(zfsvfs); // finished in zfs
178 * return (error); // done, report error
183 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
185 znode_t *zp = VTOZ(*vpp);
186 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
191 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
192 ((flag & FAPPEND) == 0)) {
194 return (SET_ERROR(EPERM));
197 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
198 ZTOV(zp)->v_type == VREG &&
199 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
200 if (fs_vscan(*vpp, cr, 0) != 0) {
202 return (SET_ERROR(EACCES));
206 /* Keep a count of the synchronous opens in the znode */
207 if (flag & (FSYNC | FDSYNC))
208 atomic_inc_32(&zp->z_sync_cnt);
216 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
217 caller_context_t *ct)
219 znode_t *zp = VTOZ(vp);
220 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
223 * Clean up any locks held by this process on the vp.
225 cleanlocks(vp, ddi_get_pid(), 0);
226 cleanshares(vp, ddi_get_pid());
231 /* Decrement the synchronous opens in the znode */
232 if ((flag & (FSYNC | FDSYNC)) && (count == 1))
233 atomic_dec_32(&zp->z_sync_cnt);
235 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
236 ZTOV(zp)->v_type == VREG &&
237 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
238 VERIFY(fs_vscan(vp, cr, 1) == 0);
245 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
246 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
249 zfs_holey(vnode_t *vp, u_long cmd, offset_t *off)
251 znode_t *zp = VTOZ(vp);
252 uint64_t noff = (uint64_t)*off; /* new offset */
257 file_sz = zp->z_size;
258 if (noff >= file_sz) {
259 return (SET_ERROR(ENXIO));
262 if (cmd == _FIO_SEEK_HOLE)
267 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
270 return (SET_ERROR(ENXIO));
273 * We could find a hole that begins after the logical end-of-file,
274 * because dmu_offset_next() only works on whole blocks. If the
275 * EOF falls mid-block, then indicate that the "virtual hole"
276 * at the end of the file begins at the logical EOF, rather than
277 * at the end of the last block.
279 if (noff > file_sz) {
292 zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred,
293 int *rvalp, caller_context_t *ct)
305 * The following two ioctls are used by bfu. Faking out,
306 * necessary to avoid bfu errors.
315 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
316 return (SET_ERROR(EFAULT));
318 off = *(offset_t *)data;
321 zfsvfs = zp->z_zfsvfs;
325 /* offset parameter is in/out */
326 error = zfs_holey(vp, com, &off);
331 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
332 return (SET_ERROR(EFAULT));
334 *(offset_t *)data = off;
338 return (SET_ERROR(ENOTTY));
342 page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes)
349 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE
350 * aligned boundaries, if the range is not aligned. As a result a
351 * DEV_BSIZE subrange with partially dirty data may get marked as clean.
352 * It may happen that all DEV_BSIZE subranges are marked clean and thus
353 * the whole page would be considred clean despite have some dirty data.
354 * For this reason we should shrink the range to DEV_BSIZE aligned
355 * boundaries before calling vm_page_clear_dirty.
357 end = rounddown2(off + nbytes, DEV_BSIZE);
358 off = roundup2(off, DEV_BSIZE);
362 zfs_vmobject_assert_wlocked(obj);
365 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
367 if (vm_page_xbusied(pp)) {
369 * Reference the page before unlocking and
370 * sleeping so that the page daemon is less
371 * likely to reclaim it.
373 vm_page_reference(pp);
375 zfs_vmobject_wunlock(obj);
376 vm_page_busy_sleep(pp, "zfsmwb");
377 zfs_vmobject_wlock(obj);
381 } else if (pp == NULL) {
382 pp = vm_page_alloc(obj, OFF_TO_IDX(start),
383 VM_ALLOC_SYSTEM | VM_ALLOC_IFCACHED |
386 ASSERT(pp != NULL && !pp->valid);
391 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
392 vm_object_pip_add(obj, 1);
393 pmap_remove_write(pp);
395 vm_page_clear_dirty(pp, off, nbytes);
403 page_unbusy(vm_page_t pp)
407 vm_object_pip_subtract(pp->object, 1);
411 page_hold(vnode_t *vp, int64_t start)
417 zfs_vmobject_assert_wlocked(obj);
420 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
422 if (vm_page_xbusied(pp)) {
424 * Reference the page before unlocking and
425 * sleeping so that the page daemon is less
426 * likely to reclaim it.
428 vm_page_reference(pp);
430 zfs_vmobject_wunlock(obj);
431 vm_page_busy_sleep(pp, "zfsmwb");
432 zfs_vmobject_wlock(obj);
436 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
449 page_unhold(vm_page_t pp)
458 * When a file is memory mapped, we must keep the IO data synchronized
459 * between the DMU cache and the memory mapped pages. What this means:
461 * On Write: If we find a memory mapped page, we write to *both*
462 * the page and the dmu buffer.
465 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid,
466 int segflg, dmu_tx_t *tx)
473 ASSERT(segflg != UIO_NOCOPY);
474 ASSERT(vp->v_mount != NULL);
478 off = start & PAGEOFFSET;
479 zfs_vmobject_wlock(obj);
480 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
482 int nbytes = imin(PAGESIZE - off, len);
484 if ((pp = page_busy(vp, start, off, nbytes)) != NULL) {
485 zfs_vmobject_wunlock(obj);
487 va = zfs_map_page(pp, &sf);
488 (void) dmu_read(os, oid, start+off, nbytes,
489 va+off, DMU_READ_PREFETCH);;
492 zfs_vmobject_wlock(obj);
498 vm_object_pip_wakeupn(obj, 0);
499 zfs_vmobject_wunlock(obj);
503 * Read with UIO_NOCOPY flag means that sendfile(2) requests
504 * ZFS to populate a range of page cache pages with data.
506 * NOTE: this function could be optimized to pre-allocate
507 * all pages in advance, drain exclusive busy on all of them,
508 * map them into contiguous KVA region and populate them
509 * in one single dmu_read() call.
512 mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio)
514 znode_t *zp = VTOZ(vp);
515 objset_t *os = zp->z_zfsvfs->z_os;
525 ASSERT(uio->uio_segflg == UIO_NOCOPY);
526 ASSERT(vp->v_mount != NULL);
529 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0);
531 zfs_vmobject_wlock(obj);
532 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) {
533 int bytes = MIN(PAGESIZE, len);
535 pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_SBUSY |
536 VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY);
537 if (pp->valid == 0) {
538 zfs_vmobject_wunlock(obj);
539 va = zfs_map_page(pp, &sf);
540 error = dmu_read(os, zp->z_id, start, bytes, va,
542 if (bytes != PAGESIZE && error == 0)
543 bzero(va + bytes, PAGESIZE - bytes);
545 zfs_vmobject_wlock(obj);
549 if (pp->wire_count == 0 && pp->valid == 0 &&
553 pp->valid = VM_PAGE_BITS_ALL;
554 vm_page_activate(pp);
558 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
563 uio->uio_resid -= bytes;
564 uio->uio_offset += bytes;
567 zfs_vmobject_wunlock(obj);
572 * When a file is memory mapped, we must keep the IO data synchronized
573 * between the DMU cache and the memory mapped pages. What this means:
575 * On Read: We "read" preferentially from memory mapped pages,
576 * else we default from the dmu buffer.
578 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
579 * the file is memory mapped.
582 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
584 znode_t *zp = VTOZ(vp);
592 ASSERT(vp->v_mount != NULL);
596 start = uio->uio_loffset;
597 off = start & PAGEOFFSET;
598 zfs_vmobject_wlock(obj);
599 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
601 uint64_t bytes = MIN(PAGESIZE - off, len);
603 if (pp = page_hold(vp, start)) {
607 zfs_vmobject_wunlock(obj);
608 va = zfs_map_page(pp, &sf);
609 error = uiomove(va + off, bytes, UIO_READ, uio);
611 zfs_vmobject_wlock(obj);
614 zfs_vmobject_wunlock(obj);
615 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
617 zfs_vmobject_wlock(obj);
624 zfs_vmobject_wunlock(obj);
628 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
631 * Read bytes from specified file into supplied buffer.
633 * IN: vp - vnode of file to be read from.
634 * uio - structure supplying read location, range info,
636 * ioflag - SYNC flags; used to provide FRSYNC semantics.
637 * cr - credentials of caller.
638 * ct - caller context
640 * OUT: uio - updated offset and range, buffer filled.
642 * RETURN: 0 on success, error code on failure.
645 * vp - atime updated if byte count > 0
649 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
651 znode_t *zp = VTOZ(vp);
652 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
661 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
663 return (SET_ERROR(EACCES));
667 * Validate file offset
669 if (uio->uio_loffset < (offset_t)0) {
671 return (SET_ERROR(EINVAL));
675 * Fasttrack empty reads
677 if (uio->uio_resid == 0) {
683 * Check for mandatory locks
685 if (MANDMODE(zp->z_mode)) {
686 if (error = chklock(vp, FREAD,
687 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
694 * If we're in FRSYNC mode, sync out this znode before reading it.
697 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
698 zil_commit(zfsvfs->z_log, zp->z_id);
701 * Lock the range against changes.
703 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
706 * If we are reading past end-of-file we can skip
707 * to the end; but we might still need to set atime.
709 if (uio->uio_loffset >= zp->z_size) {
714 ASSERT(uio->uio_loffset < zp->z_size);
715 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
718 if ((uio->uio_extflg == UIO_XUIO) &&
719 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
721 int blksz = zp->z_blksz;
722 uint64_t offset = uio->uio_loffset;
724 xuio = (xuio_t *)uio;
726 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
729 ASSERT(offset + n <= blksz);
732 (void) dmu_xuio_init(xuio, nblk);
734 if (vn_has_cached_data(vp)) {
736 * For simplicity, we always allocate a full buffer
737 * even if we only expect to read a portion of a block.
739 while (--nblk >= 0) {
740 (void) dmu_xuio_add(xuio,
741 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
749 nbytes = MIN(n, zfs_read_chunk_size -
750 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
753 if (uio->uio_segflg == UIO_NOCOPY)
754 error = mappedread_sf(vp, nbytes, uio);
756 #endif /* __FreeBSD__ */
757 if (vn_has_cached_data(vp)) {
758 error = mappedread(vp, nbytes, uio);
760 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
764 /* convert checksum errors into IO errors */
766 error = SET_ERROR(EIO);
773 zfs_range_unlock(rl);
775 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
781 * Write the bytes to a file.
783 * IN: vp - vnode of file to be written to.
784 * uio - structure supplying write location, range info,
786 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is
787 * set if in append mode.
788 * cr - credentials of caller.
789 * ct - caller context (NFS/CIFS fem monitor only)
791 * OUT: uio - updated offset and range.
793 * RETURN: 0 on success, error code on failure.
796 * vp - ctime|mtime updated if byte count > 0
801 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
803 znode_t *zp = VTOZ(vp);
804 rlim64_t limit = MAXOFFSET_T;
805 ssize_t start_resid = uio->uio_resid;
809 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
814 int max_blksz = zfsvfs->z_max_blksz;
817 iovec_t *aiov = NULL;
820 int iovcnt = uio->uio_iovcnt;
821 iovec_t *iovp = uio->uio_iov;
824 sa_bulk_attr_t bulk[4];
825 uint64_t mtime[2], ctime[2];
828 * Fasttrack empty write
834 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
840 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
841 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
842 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
844 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
848 * If immutable or not appending then return EPERM
850 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
851 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
852 (uio->uio_loffset < zp->z_size))) {
854 return (SET_ERROR(EPERM));
857 zilog = zfsvfs->z_log;
860 * Validate file offset
862 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
865 return (SET_ERROR(EINVAL));
869 * Check for mandatory locks before calling zfs_range_lock()
870 * in order to prevent a deadlock with locks set via fcntl().
872 if (MANDMODE((mode_t)zp->z_mode) &&
873 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
880 * Pre-fault the pages to ensure slow (eg NFS) pages
882 * Skip this if uio contains loaned arc_buf.
884 if ((uio->uio_extflg == UIO_XUIO) &&
885 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
886 xuio = (xuio_t *)uio;
888 uio_prefaultpages(MIN(n, max_blksz), uio);
892 * If in append mode, set the io offset pointer to eof.
894 if (ioflag & FAPPEND) {
896 * Obtain an appending range lock to guarantee file append
897 * semantics. We reset the write offset once we have the lock.
899 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
901 if (rl->r_len == UINT64_MAX) {
903 * We overlocked the file because this write will cause
904 * the file block size to increase.
905 * Note that zp_size cannot change with this lock held.
909 uio->uio_loffset = woff;
912 * Note that if the file block size will change as a result of
913 * this write, then this range lock will lock the entire file
914 * so that we can re-write the block safely.
916 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
919 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) {
920 zfs_range_unlock(rl);
926 zfs_range_unlock(rl);
928 return (SET_ERROR(EFBIG));
931 if ((woff + n) > limit || woff > (limit - n))
934 /* Will this write extend the file length? */
935 write_eof = (woff + n > zp->z_size);
937 end_size = MAX(zp->z_size, woff + n);
940 * Write the file in reasonable size chunks. Each chunk is written
941 * in a separate transaction; this keeps the intent log records small
942 * and allows us to do more fine-grained space accounting.
946 woff = uio->uio_loffset;
947 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
948 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
950 dmu_return_arcbuf(abuf);
951 error = SET_ERROR(EDQUOT);
955 if (xuio && abuf == NULL) {
956 ASSERT(i_iov < iovcnt);
958 abuf = dmu_xuio_arcbuf(xuio, i_iov);
959 dmu_xuio_clear(xuio, i_iov);
960 DTRACE_PROBE3(zfs_cp_write, int, i_iov,
961 iovec_t *, aiov, arc_buf_t *, abuf);
962 ASSERT((aiov->iov_base == abuf->b_data) ||
963 ((char *)aiov->iov_base - (char *)abuf->b_data +
964 aiov->iov_len == arc_buf_size(abuf)));
966 } else if (abuf == NULL && n >= max_blksz &&
967 woff >= zp->z_size &&
968 P2PHASE(woff, max_blksz) == 0 &&
969 zp->z_blksz == max_blksz) {
971 * This write covers a full block. "Borrow" a buffer
972 * from the dmu so that we can fill it before we enter
973 * a transaction. This avoids the possibility of
974 * holding up the transaction if the data copy hangs
975 * up on a pagefault (e.g., from an NFS server mapping).
979 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
981 ASSERT(abuf != NULL);
982 ASSERT(arc_buf_size(abuf) == max_blksz);
983 if (error = uiocopy(abuf->b_data, max_blksz,
984 UIO_WRITE, uio, &cbytes)) {
985 dmu_return_arcbuf(abuf);
988 ASSERT(cbytes == max_blksz);
992 * Start a transaction.
994 tx = dmu_tx_create(zfsvfs->z_os);
995 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
996 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
997 zfs_sa_upgrade_txholds(tx, zp);
998 error = dmu_tx_assign(tx, TXG_WAIT);
1002 dmu_return_arcbuf(abuf);
1007 * If zfs_range_lock() over-locked we grow the blocksize
1008 * and then reduce the lock range. This will only happen
1009 * on the first iteration since zfs_range_reduce() will
1010 * shrink down r_len to the appropriate size.
1012 if (rl->r_len == UINT64_MAX) {
1015 if (zp->z_blksz > max_blksz) {
1017 * File's blocksize is already larger than the
1018 * "recordsize" property. Only let it grow to
1019 * the next power of 2.
1021 ASSERT(!ISP2(zp->z_blksz));
1022 new_blksz = MIN(end_size,
1023 1 << highbit64(zp->z_blksz));
1025 new_blksz = MIN(end_size, max_blksz);
1027 zfs_grow_blocksize(zp, new_blksz, tx);
1028 zfs_range_reduce(rl, woff, n);
1032 * XXX - should we really limit each write to z_max_blksz?
1033 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
1035 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
1037 if (woff + nbytes > zp->z_size)
1038 vnode_pager_setsize(vp, woff + nbytes);
1041 tx_bytes = uio->uio_resid;
1042 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
1044 tx_bytes -= uio->uio_resid;
1047 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
1049 * If this is not a full block write, but we are
1050 * extending the file past EOF and this data starts
1051 * block-aligned, use assign_arcbuf(). Otherwise,
1052 * write via dmu_write().
1054 if (tx_bytes < max_blksz && (!write_eof ||
1055 aiov->iov_base != abuf->b_data)) {
1057 dmu_write(zfsvfs->z_os, zp->z_id, woff,
1058 aiov->iov_len, aiov->iov_base, tx);
1059 dmu_return_arcbuf(abuf);
1060 xuio_stat_wbuf_copied();
1062 ASSERT(xuio || tx_bytes == max_blksz);
1063 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
1066 ASSERT(tx_bytes <= uio->uio_resid);
1067 uioskip(uio, tx_bytes);
1069 if (tx_bytes && vn_has_cached_data(vp)) {
1070 update_pages(vp, woff, tx_bytes, zfsvfs->z_os,
1071 zp->z_id, uio->uio_segflg, tx);
1075 * If we made no progress, we're done. If we made even
1076 * partial progress, update the znode and ZIL accordingly.
1078 if (tx_bytes == 0) {
1079 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
1080 (void *)&zp->z_size, sizeof (uint64_t), tx);
1087 * Clear Set-UID/Set-GID bits on successful write if not
1088 * privileged and at least one of the excute bits is set.
1090 * It would be nice to to this after all writes have
1091 * been done, but that would still expose the ISUID/ISGID
1092 * to another app after the partial write is committed.
1094 * Note: we don't call zfs_fuid_map_id() here because
1095 * user 0 is not an ephemeral uid.
1097 mutex_enter(&zp->z_acl_lock);
1098 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
1099 (S_IXUSR >> 6))) != 0 &&
1100 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
1101 secpolicy_vnode_setid_retain(vp, cr,
1102 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
1104 zp->z_mode &= ~(S_ISUID | S_ISGID);
1105 newmode = zp->z_mode;
1106 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
1107 (void *)&newmode, sizeof (uint64_t), tx);
1109 mutex_exit(&zp->z_acl_lock);
1111 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
1115 * Update the file size (zp_size) if it has changed;
1116 * account for possible concurrent updates.
1118 while ((end_size = zp->z_size) < uio->uio_loffset) {
1119 (void) atomic_cas_64(&zp->z_size, end_size,
1124 * If we are replaying and eof is non zero then force
1125 * the file size to the specified eof. Note, there's no
1126 * concurrency during replay.
1128 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
1129 zp->z_size = zfsvfs->z_replay_eof;
1131 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1133 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
1138 ASSERT(tx_bytes == nbytes);
1143 uio_prefaultpages(MIN(n, max_blksz), uio);
1147 zfs_range_unlock(rl);
1150 * If we're in replay mode, or we made no progress, return error.
1151 * Otherwise, it's at least a partial write, so it's successful.
1153 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1158 if (ioflag & (FSYNC | FDSYNC) ||
1159 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1160 zil_commit(zilog, zp->z_id);
1167 zfs_get_done(zgd_t *zgd, int error)
1169 znode_t *zp = zgd->zgd_private;
1170 objset_t *os = zp->z_zfsvfs->z_os;
1173 dmu_buf_rele(zgd->zgd_db, zgd);
1175 zfs_range_unlock(zgd->zgd_rl);
1178 * Release the vnode asynchronously as we currently have the
1179 * txg stopped from syncing.
1181 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1183 if (error == 0 && zgd->zgd_bp)
1184 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1186 kmem_free(zgd, sizeof (zgd_t));
1190 static int zil_fault_io = 0;
1194 * Get data to generate a TX_WRITE intent log record.
1197 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1199 zfsvfs_t *zfsvfs = arg;
1200 objset_t *os = zfsvfs->z_os;
1202 uint64_t object = lr->lr_foid;
1203 uint64_t offset = lr->lr_offset;
1204 uint64_t size = lr->lr_length;
1205 blkptr_t *bp = &lr->lr_blkptr;
1210 ASSERT(zio != NULL);
1214 * Nothing to do if the file has been removed
1216 if (zfs_zget(zfsvfs, object, &zp) != 0)
1217 return (SET_ERROR(ENOENT));
1218 if (zp->z_unlinked) {
1220 * Release the vnode asynchronously as we currently have the
1221 * txg stopped from syncing.
1223 VN_RELE_ASYNC(ZTOV(zp),
1224 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1225 return (SET_ERROR(ENOENT));
1228 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1229 zgd->zgd_zilog = zfsvfs->z_log;
1230 zgd->zgd_private = zp;
1233 * Write records come in two flavors: immediate and indirect.
1234 * For small writes it's cheaper to store the data with the
1235 * log record (immediate); for large writes it's cheaper to
1236 * sync the data and get a pointer to it (indirect) so that
1237 * we don't have to write the data twice.
1239 if (buf != NULL) { /* immediate write */
1240 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1241 /* test for truncation needs to be done while range locked */
1242 if (offset >= zp->z_size) {
1243 error = SET_ERROR(ENOENT);
1245 error = dmu_read(os, object, offset, size, buf,
1246 DMU_READ_NO_PREFETCH);
1248 ASSERT(error == 0 || error == ENOENT);
1249 } else { /* indirect write */
1251 * Have to lock the whole block to ensure when it's
1252 * written out and it's checksum is being calculated
1253 * that no one can change the data. We need to re-check
1254 * blocksize after we get the lock in case it's changed!
1259 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1261 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1263 if (zp->z_blksz == size)
1266 zfs_range_unlock(zgd->zgd_rl);
1268 /* test for truncation needs to be done while range locked */
1269 if (lr->lr_offset >= zp->z_size)
1270 error = SET_ERROR(ENOENT);
1273 error = SET_ERROR(EIO);
1278 error = dmu_buf_hold(os, object, offset, zgd, &db,
1279 DMU_READ_NO_PREFETCH);
1282 blkptr_t *obp = dmu_buf_get_blkptr(db);
1284 ASSERT(BP_IS_HOLE(bp));
1291 ASSERT(db->db_offset == offset);
1292 ASSERT(db->db_size == size);
1294 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1296 ASSERT(error || lr->lr_length <= zp->z_blksz);
1299 * On success, we need to wait for the write I/O
1300 * initiated by dmu_sync() to complete before we can
1301 * release this dbuf. We will finish everything up
1302 * in the zfs_get_done() callback.
1307 if (error == EALREADY) {
1308 lr->lr_common.lrc_txtype = TX_WRITE2;
1314 zfs_get_done(zgd, error);
1321 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1322 caller_context_t *ct)
1324 znode_t *zp = VTOZ(vp);
1325 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1331 if (flag & V_ACE_MASK)
1332 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1334 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1341 * If vnode is for a device return a specfs vnode instead.
1344 specvp_check(vnode_t **vpp, cred_t *cr)
1348 if (IS_DEVVP(*vpp)) {
1351 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1354 error = SET_ERROR(ENOSYS);
1362 * Lookup an entry in a directory, or an extended attribute directory.
1363 * If it exists, return a held vnode reference for it.
1365 * IN: dvp - vnode of directory to search.
1366 * nm - name of entry to lookup.
1367 * pnp - full pathname to lookup [UNUSED].
1368 * flags - LOOKUP_XATTR set if looking for an attribute.
1369 * rdir - root directory vnode [UNUSED].
1370 * cr - credentials of caller.
1371 * ct - caller context
1372 * direntflags - directory lookup flags
1373 * realpnp - returned pathname.
1375 * OUT: vpp - vnode of located entry, NULL if not found.
1377 * RETURN: 0 on success, error code on failure.
1384 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp,
1385 int nameiop, cred_t *cr, kthread_t *td, int flags)
1387 znode_t *zdp = VTOZ(dvp);
1388 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1390 int *direntflags = NULL;
1391 void *realpnp = NULL;
1394 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1396 if (dvp->v_type != VDIR) {
1397 return (SET_ERROR(ENOTDIR));
1398 } else if (zdp->z_sa_hdl == NULL) {
1399 return (SET_ERROR(EIO));
1402 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1403 error = zfs_fastaccesschk_execute(zdp, cr);
1411 vnode_t *tvp = dnlc_lookup(dvp, nm);
1414 error = zfs_fastaccesschk_execute(zdp, cr);
1419 if (tvp == DNLC_NO_VNODE) {
1421 return (SET_ERROR(ENOENT));
1424 return (specvp_check(vpp, cr));
1430 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1437 if (flags & LOOKUP_XATTR) {
1440 * If the xattr property is off, refuse the lookup request.
1442 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1444 return (SET_ERROR(EINVAL));
1449 * We don't allow recursive attributes..
1450 * Maybe someday we will.
1452 if (zdp->z_pflags & ZFS_XATTR) {
1454 return (SET_ERROR(EINVAL));
1457 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1463 * Do we have permission to get into attribute directory?
1466 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1476 if (dvp->v_type != VDIR) {
1478 return (SET_ERROR(ENOTDIR));
1482 * Check accessibility of directory.
1485 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1490 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1491 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1493 return (SET_ERROR(EILSEQ));
1496 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1498 error = specvp_check(vpp, cr);
1500 /* Translate errors and add SAVENAME when needed. */
1501 if (cnp->cn_flags & ISLASTCN) {
1505 if (error == ENOENT) {
1506 error = EJUSTRETURN;
1507 cnp->cn_flags |= SAVENAME;
1513 cnp->cn_flags |= SAVENAME;
1517 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) {
1520 if (cnp->cn_flags & ISDOTDOT) {
1521 ltype = VOP_ISLOCKED(dvp);
1525 error = vn_lock(*vpp, cnp->cn_lkflags);
1526 if (cnp->cn_flags & ISDOTDOT)
1527 vn_lock(dvp, ltype | LK_RETRY);
1537 #ifdef FREEBSD_NAMECACHE
1539 * Insert name into cache (as non-existent) if appropriate.
1541 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) != 0)
1542 cache_enter(dvp, *vpp, cnp);
1544 * Insert name into cache if appropriate.
1546 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) {
1547 if (!(cnp->cn_flags & ISLASTCN) ||
1548 (nameiop != DELETE && nameiop != RENAME)) {
1549 cache_enter(dvp, *vpp, cnp);
1558 * Attempt to create a new entry in a directory. If the entry
1559 * already exists, truncate the file if permissible, else return
1560 * an error. Return the vp of the created or trunc'd file.
1562 * IN: dvp - vnode of directory to put new file entry in.
1563 * name - name of new file entry.
1564 * vap - attributes of new file.
1565 * excl - flag indicating exclusive or non-exclusive mode.
1566 * mode - mode to open file with.
1567 * cr - credentials of caller.
1568 * flag - large file flag [UNUSED].
1569 * ct - caller context
1570 * vsecp - ACL to be set
1572 * OUT: vpp - vnode of created or trunc'd entry.
1574 * RETURN: 0 on success, error code on failure.
1577 * dvp - ctime|mtime updated if new entry created
1578 * vp - ctime|mtime always, atime if new
1583 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode,
1584 vnode_t **vpp, cred_t *cr, kthread_t *td)
1586 znode_t *zp, *dzp = VTOZ(dvp);
1587 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1595 gid_t gid = crgetgid(cr);
1596 zfs_acl_ids_t acl_ids;
1597 boolean_t fuid_dirtied;
1598 boolean_t have_acl = B_FALSE;
1599 boolean_t waited = B_FALSE;
1604 * If we have an ephemeral id, ACL, or XVATTR then
1605 * make sure file system is at proper version
1608 ksid = crgetsid(cr, KSID_OWNER);
1610 uid = ksid_getid(ksid);
1614 if (zfsvfs->z_use_fuids == B_FALSE &&
1615 (vsecp || (vap->va_mask & AT_XVATTR) ||
1616 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1617 return (SET_ERROR(EINVAL));
1622 zilog = zfsvfs->z_log;
1624 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1625 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1627 return (SET_ERROR(EILSEQ));
1630 if (vap->va_mask & AT_XVATTR) {
1631 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
1632 crgetuid(cr), cr, vap->va_type)) != 0) {
1638 getnewvnode_reserve(1);
1643 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr))
1644 vap->va_mode &= ~S_ISVTX;
1646 if (*name == '\0') {
1648 * Null component name refers to the directory itself.
1655 /* possible VN_HOLD(zp) */
1658 if (flag & FIGNORECASE)
1661 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1665 zfs_acl_ids_free(&acl_ids);
1666 if (strcmp(name, "..") == 0)
1667 error = SET_ERROR(EISDIR);
1668 getnewvnode_drop_reserve();
1678 * Create a new file object and update the directory
1681 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1683 zfs_acl_ids_free(&acl_ids);
1688 * We only support the creation of regular files in
1689 * extended attribute directories.
1692 if ((dzp->z_pflags & ZFS_XATTR) &&
1693 (vap->va_type != VREG)) {
1695 zfs_acl_ids_free(&acl_ids);
1696 error = SET_ERROR(EINVAL);
1700 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1701 cr, vsecp, &acl_ids)) != 0)
1705 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1706 zfs_acl_ids_free(&acl_ids);
1707 error = SET_ERROR(EDQUOT);
1711 tx = dmu_tx_create(os);
1713 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1714 ZFS_SA_BASE_ATTR_SIZE);
1716 fuid_dirtied = zfsvfs->z_fuid_dirty;
1718 zfs_fuid_txhold(zfsvfs, tx);
1719 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1720 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1721 if (!zfsvfs->z_use_sa &&
1722 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1723 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1724 0, acl_ids.z_aclp->z_acl_bytes);
1726 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1728 zfs_dirent_unlock(dl);
1729 if (error == ERESTART) {
1735 zfs_acl_ids_free(&acl_ids);
1737 getnewvnode_drop_reserve();
1741 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1744 zfs_fuid_sync(zfsvfs, tx);
1746 (void) zfs_link_create(dl, zp, tx, ZNEW);
1747 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1748 if (flag & FIGNORECASE)
1750 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1751 vsecp, acl_ids.z_fuidp, vap);
1752 zfs_acl_ids_free(&acl_ids);
1755 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1758 zfs_acl_ids_free(&acl_ids);
1762 * A directory entry already exists for this name.
1765 * Can't truncate an existing file if in exclusive mode.
1768 error = SET_ERROR(EEXIST);
1772 * Can't open a directory for writing.
1774 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1775 error = SET_ERROR(EISDIR);
1779 * Verify requested access to file.
1781 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1785 mutex_enter(&dzp->z_lock);
1787 mutex_exit(&dzp->z_lock);
1790 * Truncate regular files if requested.
1792 if ((ZTOV(zp)->v_type == VREG) &&
1793 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1794 /* we can't hold any locks when calling zfs_freesp() */
1795 zfs_dirent_unlock(dl);
1797 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1799 vnevent_create(ZTOV(zp), ct);
1804 getnewvnode_drop_reserve();
1806 zfs_dirent_unlock(dl);
1813 error = specvp_check(vpp, cr);
1816 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1817 zil_commit(zilog, 0);
1824 * Remove an entry from a directory.
1826 * IN: dvp - vnode of directory to remove entry from.
1827 * name - name of entry to remove.
1828 * cr - credentials of caller.
1829 * ct - caller context
1830 * flags - case flags
1832 * RETURN: 0 on success, error code on failure.
1836 * vp - ctime (if nlink > 0)
1839 uint64_t null_xattr = 0;
1843 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1846 znode_t *zp, *dzp = VTOZ(dvp);
1849 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1851 uint64_t acl_obj, xattr_obj;
1852 uint64_t xattr_obj_unlinked = 0;
1856 boolean_t may_delete_now, delete_now = FALSE;
1857 boolean_t unlinked, toobig = FALSE;
1859 pathname_t *realnmp = NULL;
1863 boolean_t waited = B_FALSE;
1867 zilog = zfsvfs->z_log;
1869 if (flags & FIGNORECASE) {
1879 * Attempt to lock directory; fail if entry doesn't exist.
1881 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1891 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1896 * Need to use rmdir for removing directories.
1898 if (vp->v_type == VDIR) {
1899 error = SET_ERROR(EPERM);
1903 vnevent_remove(vp, dvp, name, ct);
1906 dnlc_remove(dvp, realnmp->pn_buf);
1908 dnlc_remove(dvp, name);
1911 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1915 * We may delete the znode now, or we may put it in the unlinked set;
1916 * it depends on whether we're the last link, and on whether there are
1917 * other holds on the vnode. So we dmu_tx_hold() the right things to
1918 * allow for either case.
1921 tx = dmu_tx_create(zfsvfs->z_os);
1922 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1923 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1924 zfs_sa_upgrade_txholds(tx, zp);
1925 zfs_sa_upgrade_txholds(tx, dzp);
1926 if (may_delete_now) {
1928 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1929 /* if the file is too big, only hold_free a token amount */
1930 dmu_tx_hold_free(tx, zp->z_id, 0,
1931 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1934 /* are there any extended attributes? */
1935 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1936 &xattr_obj, sizeof (xattr_obj));
1937 if (error == 0 && xattr_obj) {
1938 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1940 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1941 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1944 mutex_enter(&zp->z_lock);
1945 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1946 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1947 mutex_exit(&zp->z_lock);
1949 /* charge as an update -- would be nice not to charge at all */
1950 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1953 * Mark this transaction as typically resulting in a net free of
1954 * space, unless object removal will be delayed indefinitely
1955 * (due to active holds on the vnode due to the file being open).
1958 dmu_tx_mark_netfree(tx);
1960 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1962 zfs_dirent_unlock(dl);
1966 if (error == ERESTART) {
1980 * Remove the directory entry.
1982 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1991 * Hold z_lock so that we can make sure that the ACL obj
1992 * hasn't changed. Could have been deleted due to
1995 mutex_enter(&zp->z_lock);
1997 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1998 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1999 delete_now = may_delete_now && !toobig &&
2000 vp->v_count == 1 && !vn_has_cached_data(vp) &&
2001 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
2008 panic("zfs_remove: delete_now branch taken");
2010 if (xattr_obj_unlinked) {
2011 ASSERT3U(xzp->z_links, ==, 2);
2012 mutex_enter(&xzp->z_lock);
2013 xzp->z_unlinked = 1;
2015 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
2016 &xzp->z_links, sizeof (xzp->z_links), tx);
2017 ASSERT3U(error, ==, 0);
2018 mutex_exit(&xzp->z_lock);
2019 zfs_unlinked_add(xzp, tx);
2022 error = sa_remove(zp->z_sa_hdl,
2023 SA_ZPL_XATTR(zfsvfs), tx);
2025 error = sa_update(zp->z_sa_hdl,
2026 SA_ZPL_XATTR(zfsvfs), &null_xattr,
2027 sizeof (uint64_t), tx);
2032 ASSERT0(vp->v_count);
2034 mutex_exit(&zp->z_lock);
2035 zfs_znode_delete(zp, tx);
2036 } else if (unlinked) {
2037 mutex_exit(&zp->z_lock);
2038 zfs_unlinked_add(zp, tx);
2040 vp->v_vflag |= VV_NOSYNC;
2045 if (flags & FIGNORECASE)
2047 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
2054 zfs_dirent_unlock(dl);
2061 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2062 zil_commit(zilog, 0);
2069 * Create a new directory and insert it into dvp using the name
2070 * provided. Return a pointer to the inserted directory.
2072 * IN: dvp - vnode of directory to add subdir to.
2073 * dirname - name of new directory.
2074 * vap - attributes of new directory.
2075 * cr - credentials of caller.
2076 * ct - caller context
2077 * flags - case flags
2078 * vsecp - ACL to be set
2080 * OUT: vpp - vnode of created directory.
2082 * RETURN: 0 on success, error code on failure.
2085 * dvp - ctime|mtime updated
2086 * vp - ctime|mtime|atime updated
2090 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
2091 caller_context_t *ct, int flags, vsecattr_t *vsecp)
2093 znode_t *zp, *dzp = VTOZ(dvp);
2094 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2103 gid_t gid = crgetgid(cr);
2104 zfs_acl_ids_t acl_ids;
2105 boolean_t fuid_dirtied;
2106 boolean_t waited = B_FALSE;
2108 ASSERT(vap->va_type == VDIR);
2111 * If we have an ephemeral id, ACL, or XVATTR then
2112 * make sure file system is at proper version
2115 ksid = crgetsid(cr, KSID_OWNER);
2117 uid = ksid_getid(ksid);
2120 if (zfsvfs->z_use_fuids == B_FALSE &&
2121 (vsecp || (vap->va_mask & AT_XVATTR) ||
2122 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
2123 return (SET_ERROR(EINVAL));
2127 zilog = zfsvfs->z_log;
2129 if (dzp->z_pflags & ZFS_XATTR) {
2131 return (SET_ERROR(EINVAL));
2134 if (zfsvfs->z_utf8 && u8_validate(dirname,
2135 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2137 return (SET_ERROR(EILSEQ));
2139 if (flags & FIGNORECASE)
2142 if (vap->va_mask & AT_XVATTR) {
2143 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
2144 crgetuid(cr), cr, vap->va_type)) != 0) {
2150 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
2151 vsecp, &acl_ids)) != 0) {
2156 getnewvnode_reserve(1);
2159 * First make sure the new directory doesn't exist.
2161 * Existence is checked first to make sure we don't return
2162 * EACCES instead of EEXIST which can cause some applications
2168 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
2170 zfs_acl_ids_free(&acl_ids);
2171 getnewvnode_drop_reserve();
2176 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
2177 zfs_acl_ids_free(&acl_ids);
2178 zfs_dirent_unlock(dl);
2179 getnewvnode_drop_reserve();
2184 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
2185 zfs_acl_ids_free(&acl_ids);
2186 zfs_dirent_unlock(dl);
2187 getnewvnode_drop_reserve();
2189 return (SET_ERROR(EDQUOT));
2193 * Add a new entry to the directory.
2195 tx = dmu_tx_create(zfsvfs->z_os);
2196 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2197 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2198 fuid_dirtied = zfsvfs->z_fuid_dirty;
2200 zfs_fuid_txhold(zfsvfs, tx);
2201 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2202 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2203 acl_ids.z_aclp->z_acl_bytes);
2206 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2207 ZFS_SA_BASE_ATTR_SIZE);
2209 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2211 zfs_dirent_unlock(dl);
2212 if (error == ERESTART) {
2218 zfs_acl_ids_free(&acl_ids);
2220 getnewvnode_drop_reserve();
2228 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2231 zfs_fuid_sync(zfsvfs, tx);
2234 * Now put new name in parent dir.
2236 (void) zfs_link_create(dl, zp, tx, ZNEW);
2240 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2241 if (flags & FIGNORECASE)
2243 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2244 acl_ids.z_fuidp, vap);
2246 zfs_acl_ids_free(&acl_ids);
2250 getnewvnode_drop_reserve();
2252 zfs_dirent_unlock(dl);
2254 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2255 zil_commit(zilog, 0);
2262 * Remove a directory subdir entry. If the current working
2263 * directory is the same as the subdir to be removed, the
2266 * IN: dvp - vnode of directory to remove from.
2267 * name - name of directory to be removed.
2268 * cwd - vnode of current working directory.
2269 * cr - credentials of caller.
2270 * ct - caller context
2271 * flags - case flags
2273 * RETURN: 0 on success, error code on failure.
2276 * dvp - ctime|mtime updated
2280 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2281 caller_context_t *ct, int flags)
2283 znode_t *dzp = VTOZ(dvp);
2286 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2292 boolean_t waited = B_FALSE;
2296 zilog = zfsvfs->z_log;
2298 if (flags & FIGNORECASE)
2304 * Attempt to lock directory; fail if entry doesn't exist.
2306 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2314 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2318 if (vp->v_type != VDIR) {
2319 error = SET_ERROR(ENOTDIR);
2324 error = SET_ERROR(EINVAL);
2328 vnevent_rmdir(vp, dvp, name, ct);
2331 * Grab a lock on the directory to make sure that noone is
2332 * trying to add (or lookup) entries while we are removing it.
2334 rw_enter(&zp->z_name_lock, RW_WRITER);
2337 * Grab a lock on the parent pointer to make sure we play well
2338 * with the treewalk and directory rename code.
2340 rw_enter(&zp->z_parent_lock, RW_WRITER);
2342 tx = dmu_tx_create(zfsvfs->z_os);
2343 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2344 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2345 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2346 zfs_sa_upgrade_txholds(tx, zp);
2347 zfs_sa_upgrade_txholds(tx, dzp);
2348 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2350 rw_exit(&zp->z_parent_lock);
2351 rw_exit(&zp->z_name_lock);
2352 zfs_dirent_unlock(dl);
2354 if (error == ERESTART) {
2365 #ifdef FREEBSD_NAMECACHE
2369 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2372 uint64_t txtype = TX_RMDIR;
2373 if (flags & FIGNORECASE)
2375 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2380 rw_exit(&zp->z_parent_lock);
2381 rw_exit(&zp->z_name_lock);
2382 #ifdef FREEBSD_NAMECACHE
2386 zfs_dirent_unlock(dl);
2390 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2391 zil_commit(zilog, 0);
2398 * Read as many directory entries as will fit into the provided
2399 * buffer from the given directory cursor position (specified in
2400 * the uio structure).
2402 * IN: vp - vnode of directory to read.
2403 * uio - structure supplying read location, range info,
2404 * and return buffer.
2405 * cr - credentials of caller.
2406 * ct - caller context
2407 * flags - case flags
2409 * OUT: uio - updated offset and range, buffer filled.
2410 * eofp - set to true if end-of-file detected.
2412 * RETURN: 0 on success, error code on failure.
2415 * vp - atime updated
2417 * Note that the low 4 bits of the cookie returned by zap is always zero.
2418 * This allows us to use the low range for "special" directory entries:
2419 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2420 * we use the offset 2 for the '.zfs' directory.
2424 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies)
2426 znode_t *zp = VTOZ(vp);
2430 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2435 zap_attribute_t zap;
2436 uint_t bytes_wanted;
2437 uint64_t offset; /* must be unsigned; checks for < 1 */
2443 boolean_t check_sysattrs;
2446 u_long *cooks = NULL;
2452 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2453 &parent, sizeof (parent))) != 0) {
2459 * If we are not given an eof variable,
2466 * Check for valid iov_len.
2468 if (uio->uio_iov->iov_len <= 0) {
2470 return (SET_ERROR(EINVAL));
2474 * Quit if directory has been removed (posix)
2476 if ((*eofp = zp->z_unlinked) != 0) {
2483 offset = uio->uio_loffset;
2484 prefetch = zp->z_zn_prefetch;
2487 * Initialize the iterator cursor.
2491 * Start iteration from the beginning of the directory.
2493 zap_cursor_init(&zc, os, zp->z_id);
2496 * The offset is a serialized cursor.
2498 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2502 * Get space to change directory entries into fs independent format.
2504 iovp = uio->uio_iov;
2505 bytes_wanted = iovp->iov_len;
2506 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2507 bufsize = bytes_wanted;
2508 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2509 odp = (struct dirent64 *)outbuf;
2511 bufsize = bytes_wanted;
2513 odp = (struct dirent64 *)iovp->iov_base;
2515 eodp = (struct edirent *)odp;
2517 if (ncookies != NULL) {
2519 * Minimum entry size is dirent size and 1 byte for a file name.
2521 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1);
2522 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK);
2527 * If this VFS supports the system attribute view interface; and
2528 * we're looking at an extended attribute directory; and we care
2529 * about normalization conflicts on this vfs; then we must check
2530 * for normalization conflicts with the sysattr name space.
2533 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2534 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2535 (flags & V_RDDIR_ENTFLAGS);
2541 * Transform to file-system independent format
2544 while (outcount < bytes_wanted) {
2547 off64_t *next = NULL;
2550 * Special case `.', `..', and `.zfs'.
2553 (void) strcpy(zap.za_name, ".");
2554 zap.za_normalization_conflict = 0;
2557 } else if (offset == 1) {
2558 (void) strcpy(zap.za_name, "..");
2559 zap.za_normalization_conflict = 0;
2562 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2563 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2564 zap.za_normalization_conflict = 0;
2565 objnum = ZFSCTL_INO_ROOT;
2571 if (error = zap_cursor_retrieve(&zc, &zap)) {
2572 if ((*eofp = (error == ENOENT)) != 0)
2578 if (zap.za_integer_length != 8 ||
2579 zap.za_num_integers != 1) {
2580 cmn_err(CE_WARN, "zap_readdir: bad directory "
2581 "entry, obj = %lld, offset = %lld\n",
2582 (u_longlong_t)zp->z_id,
2583 (u_longlong_t)offset);
2584 error = SET_ERROR(ENXIO);
2588 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2590 * MacOS X can extract the object type here such as:
2591 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2593 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2595 if (check_sysattrs && !zap.za_normalization_conflict) {
2597 zap.za_normalization_conflict =
2598 xattr_sysattr_casechk(zap.za_name);
2600 panic("%s:%u: TODO", __func__, __LINE__);
2605 if (flags & V_RDDIR_ACCFILTER) {
2607 * If we have no access at all, don't include
2608 * this entry in the returned information
2611 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2613 if (!zfs_has_access(ezp, cr)) {
2620 if (flags & V_RDDIR_ENTFLAGS)
2621 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2623 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2626 * Will this entry fit in the buffer?
2628 if (outcount + reclen > bufsize) {
2630 * Did we manage to fit anything in the buffer?
2633 error = SET_ERROR(EINVAL);
2638 if (flags & V_RDDIR_ENTFLAGS) {
2640 * Add extended flag entry:
2642 eodp->ed_ino = objnum;
2643 eodp->ed_reclen = reclen;
2644 /* NOTE: ed_off is the offset for the *next* entry */
2645 next = &(eodp->ed_off);
2646 eodp->ed_eflags = zap.za_normalization_conflict ?
2647 ED_CASE_CONFLICT : 0;
2648 (void) strncpy(eodp->ed_name, zap.za_name,
2649 EDIRENT_NAMELEN(reclen));
2650 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2655 odp->d_ino = objnum;
2656 odp->d_reclen = reclen;
2657 odp->d_namlen = strlen(zap.za_name);
2658 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1);
2660 odp = (dirent64_t *)((intptr_t)odp + reclen);
2664 ASSERT(outcount <= bufsize);
2666 /* Prefetch znode */
2668 dmu_prefetch(os, objnum, 0, 0);
2672 * Move to the next entry, fill in the previous offset.
2674 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2675 zap_cursor_advance(&zc);
2676 offset = zap_cursor_serialize(&zc);
2681 if (cooks != NULL) {
2684 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks));
2687 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2689 /* Subtract unused cookies */
2690 if (ncookies != NULL)
2691 *ncookies -= ncooks;
2693 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2694 iovp->iov_base += outcount;
2695 iovp->iov_len -= outcount;
2696 uio->uio_resid -= outcount;
2697 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2699 * Reset the pointer.
2701 offset = uio->uio_loffset;
2705 zap_cursor_fini(&zc);
2706 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2707 kmem_free(outbuf, bufsize);
2709 if (error == ENOENT)
2712 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2714 uio->uio_loffset = offset;
2716 if (error != 0 && cookies != NULL) {
2717 free(*cookies, M_TEMP);
2724 ulong_t zfs_fsync_sync_cnt = 4;
2727 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2729 znode_t *zp = VTOZ(vp);
2730 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2732 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2734 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2737 zil_commit(zfsvfs->z_log, zp->z_id);
2745 * Get the requested file attributes and place them in the provided
2748 * IN: vp - vnode of file.
2749 * vap - va_mask identifies requested attributes.
2750 * If AT_XVATTR set, then optional attrs are requested
2751 * flags - ATTR_NOACLCHECK (CIFS server context)
2752 * cr - credentials of caller.
2753 * ct - caller context
2755 * OUT: vap - attribute values.
2757 * RETURN: 0 (always succeeds).
2761 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2762 caller_context_t *ct)
2764 znode_t *zp = VTOZ(vp);
2765 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2768 u_longlong_t nblocks;
2770 uint64_t mtime[2], ctime[2], crtime[2], rdev;
2771 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2772 xoptattr_t *xoap = NULL;
2773 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2774 sa_bulk_attr_t bulk[4];
2780 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2782 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2783 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2784 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
2785 if (vp->v_type == VBLK || vp->v_type == VCHR)
2786 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
2789 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2795 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2796 * Also, if we are the owner don't bother, since owner should
2797 * always be allowed to read basic attributes of file.
2799 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2800 (vap->va_uid != crgetuid(cr))) {
2801 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2809 * Return all attributes. It's cheaper to provide the answer
2810 * than to determine whether we were asked the question.
2813 mutex_enter(&zp->z_lock);
2814 vap->va_type = IFTOVT(zp->z_mode);
2815 vap->va_mode = zp->z_mode & ~S_IFMT;
2817 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2819 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
2821 vap->va_nodeid = zp->z_id;
2822 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2823 links = zp->z_links + 1;
2825 links = zp->z_links;
2826 vap->va_nlink = MIN(links, LINK_MAX); /* nlink_t limit! */
2827 vap->va_size = zp->z_size;
2829 vap->va_rdev = vp->v_rdev;
2831 if (vp->v_type == VBLK || vp->v_type == VCHR)
2832 vap->va_rdev = zfs_cmpldev(rdev);
2834 vap->va_seq = zp->z_seq;
2835 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */
2836 vap->va_filerev = zp->z_seq;
2839 * Add in any requested optional attributes and the create time.
2840 * Also set the corresponding bits in the returned attribute bitmap.
2842 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2843 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2845 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2846 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2849 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2850 xoap->xoa_readonly =
2851 ((zp->z_pflags & ZFS_READONLY) != 0);
2852 XVA_SET_RTN(xvap, XAT_READONLY);
2855 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2857 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2858 XVA_SET_RTN(xvap, XAT_SYSTEM);
2861 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2863 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2864 XVA_SET_RTN(xvap, XAT_HIDDEN);
2867 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2868 xoap->xoa_nounlink =
2869 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2870 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2873 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2874 xoap->xoa_immutable =
2875 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2876 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2879 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2880 xoap->xoa_appendonly =
2881 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2882 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2885 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2887 ((zp->z_pflags & ZFS_NODUMP) != 0);
2888 XVA_SET_RTN(xvap, XAT_NODUMP);
2891 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2893 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2894 XVA_SET_RTN(xvap, XAT_OPAQUE);
2897 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2898 xoap->xoa_av_quarantined =
2899 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2900 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2903 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2904 xoap->xoa_av_modified =
2905 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2906 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2909 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2910 vp->v_type == VREG) {
2911 zfs_sa_get_scanstamp(zp, xvap);
2914 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2917 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2918 times, sizeof (times));
2919 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2920 XVA_SET_RTN(xvap, XAT_CREATETIME);
2923 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2924 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2925 XVA_SET_RTN(xvap, XAT_REPARSE);
2927 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2928 xoap->xoa_generation = zp->z_gen;
2929 XVA_SET_RTN(xvap, XAT_GEN);
2932 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2934 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2935 XVA_SET_RTN(xvap, XAT_OFFLINE);
2938 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2940 ((zp->z_pflags & ZFS_SPARSE) != 0);
2941 XVA_SET_RTN(xvap, XAT_SPARSE);
2945 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2946 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2947 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2948 ZFS_TIME_DECODE(&vap->va_birthtime, crtime);
2950 mutex_exit(&zp->z_lock);
2952 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2953 vap->va_blksize = blksize;
2954 vap->va_bytes = nblocks << 9; /* nblocks * 512 */
2956 if (zp->z_blksz == 0) {
2958 * Block size hasn't been set; suggest maximal I/O transfers.
2960 vap->va_blksize = zfsvfs->z_max_blksz;
2968 * Set the file attributes to the values contained in the
2971 * IN: vp - vnode of file to be modified.
2972 * vap - new attribute values.
2973 * If AT_XVATTR set, then optional attrs are being set
2974 * flags - ATTR_UTIME set if non-default time values provided.
2975 * - ATTR_NOACLCHECK (CIFS context only).
2976 * cr - credentials of caller.
2977 * ct - caller context
2979 * RETURN: 0 on success, error code on failure.
2982 * vp - ctime updated, mtime updated if size changed.
2986 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2987 caller_context_t *ct)
2989 znode_t *zp = VTOZ(vp);
2990 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2995 uint_t mask = vap->va_mask;
2996 uint_t saved_mask = 0;
2997 uint64_t saved_mode;
3000 uint64_t new_uid, new_gid;
3002 uint64_t mtime[2], ctime[2];
3004 int need_policy = FALSE;
3006 zfs_fuid_info_t *fuidp = NULL;
3007 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
3010 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
3011 boolean_t fuid_dirtied = B_FALSE;
3012 sa_bulk_attr_t bulk[7], xattr_bulk[7];
3013 int count = 0, xattr_count = 0;
3018 if (mask & AT_NOSET)
3019 return (SET_ERROR(EINVAL));
3024 zilog = zfsvfs->z_log;
3027 * Make sure that if we have ephemeral uid/gid or xvattr specified
3028 * that file system is at proper version level
3031 if (zfsvfs->z_use_fuids == B_FALSE &&
3032 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
3033 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
3034 (mask & AT_XVATTR))) {
3036 return (SET_ERROR(EINVAL));
3039 if (mask & AT_SIZE && vp->v_type == VDIR) {
3041 return (SET_ERROR(EISDIR));
3044 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
3046 return (SET_ERROR(EINVAL));
3050 * If this is an xvattr_t, then get a pointer to the structure of
3051 * optional attributes. If this is NULL, then we have a vattr_t.
3053 xoap = xva_getxoptattr(xvap);
3055 xva_init(&tmpxvattr);
3058 * Immutable files can only alter immutable bit and atime
3060 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
3061 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
3062 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
3064 return (SET_ERROR(EPERM));
3067 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
3069 return (SET_ERROR(EPERM));
3073 * Verify timestamps doesn't overflow 32 bits.
3074 * ZFS can handle large timestamps, but 32bit syscalls can't
3075 * handle times greater than 2039. This check should be removed
3076 * once large timestamps are fully supported.
3078 if (mask & (AT_ATIME | AT_MTIME)) {
3079 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
3080 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
3082 return (SET_ERROR(EOVERFLOW));
3090 /* Can this be moved to before the top label? */
3091 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
3093 return (SET_ERROR(EROFS));
3097 * First validate permissions
3100 if (mask & AT_SIZE) {
3102 * XXX - Note, we are not providing any open
3103 * mode flags here (like FNDELAY), so we may
3104 * block if there are locks present... this
3105 * should be addressed in openat().
3107 /* XXX - would it be OK to generate a log record here? */
3108 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
3115 if (mask & (AT_ATIME|AT_MTIME) ||
3116 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
3117 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
3118 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
3119 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
3120 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
3121 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
3122 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
3123 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
3127 if (mask & (AT_UID|AT_GID)) {
3128 int idmask = (mask & (AT_UID|AT_GID));
3133 * NOTE: even if a new mode is being set,
3134 * we may clear S_ISUID/S_ISGID bits.
3137 if (!(mask & AT_MODE))
3138 vap->va_mode = zp->z_mode;
3141 * Take ownership or chgrp to group we are a member of
3144 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
3145 take_group = (mask & AT_GID) &&
3146 zfs_groupmember(zfsvfs, vap->va_gid, cr);
3149 * If both AT_UID and AT_GID are set then take_owner and
3150 * take_group must both be set in order to allow taking
3153 * Otherwise, send the check through secpolicy_vnode_setattr()
3157 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
3158 ((idmask == AT_UID) && take_owner) ||
3159 ((idmask == AT_GID) && take_group)) {
3160 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
3161 skipaclchk, cr) == 0) {
3163 * Remove setuid/setgid for non-privileged users
3165 secpolicy_setid_clear(vap, vp, cr);
3166 trim_mask = (mask & (AT_UID|AT_GID));
3175 mutex_enter(&zp->z_lock);
3176 oldva.va_mode = zp->z_mode;
3177 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
3178 if (mask & AT_XVATTR) {
3180 * Update xvattr mask to include only those attributes
3181 * that are actually changing.
3183 * the bits will be restored prior to actually setting
3184 * the attributes so the caller thinks they were set.
3186 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
3187 if (xoap->xoa_appendonly !=
3188 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
3191 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
3192 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
3196 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
3197 if (xoap->xoa_nounlink !=
3198 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
3201 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
3202 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
3206 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
3207 if (xoap->xoa_immutable !=
3208 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
3211 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
3212 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
3216 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
3217 if (xoap->xoa_nodump !=
3218 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
3221 XVA_CLR_REQ(xvap, XAT_NODUMP);
3222 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
3226 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3227 if (xoap->xoa_av_modified !=
3228 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3231 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3232 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
3236 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3237 if ((vp->v_type != VREG &&
3238 xoap->xoa_av_quarantined) ||
3239 xoap->xoa_av_quarantined !=
3240 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3243 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3244 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
3248 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3249 mutex_exit(&zp->z_lock);
3251 return (SET_ERROR(EPERM));
3254 if (need_policy == FALSE &&
3255 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3256 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3261 mutex_exit(&zp->z_lock);
3263 if (mask & AT_MODE) {
3264 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3265 err = secpolicy_setid_setsticky_clear(vp, vap,
3271 trim_mask |= AT_MODE;
3279 * If trim_mask is set then take ownership
3280 * has been granted or write_acl is present and user
3281 * has the ability to modify mode. In that case remove
3282 * UID|GID and or MODE from mask so that
3283 * secpolicy_vnode_setattr() doesn't revoke it.
3287 saved_mask = vap->va_mask;
3288 vap->va_mask &= ~trim_mask;
3289 if (trim_mask & AT_MODE) {
3291 * Save the mode, as secpolicy_vnode_setattr()
3292 * will overwrite it with ova.va_mode.
3294 saved_mode = vap->va_mode;
3297 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3298 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3305 vap->va_mask |= saved_mask;
3306 if (trim_mask & AT_MODE) {
3308 * Recover the mode after
3309 * secpolicy_vnode_setattr().
3311 vap->va_mode = saved_mode;
3317 * secpolicy_vnode_setattr, or take ownership may have
3320 mask = vap->va_mask;
3322 if ((mask & (AT_UID | AT_GID))) {
3323 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3324 &xattr_obj, sizeof (xattr_obj));
3326 if (err == 0 && xattr_obj) {
3327 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3331 if (mask & AT_UID) {
3332 new_uid = zfs_fuid_create(zfsvfs,
3333 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3334 if (new_uid != zp->z_uid &&
3335 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3337 VN_RELE(ZTOV(attrzp));
3338 err = SET_ERROR(EDQUOT);
3343 if (mask & AT_GID) {
3344 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3345 cr, ZFS_GROUP, &fuidp);
3346 if (new_gid != zp->z_gid &&
3347 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3349 VN_RELE(ZTOV(attrzp));
3350 err = SET_ERROR(EDQUOT);
3355 tx = dmu_tx_create(zfsvfs->z_os);
3357 if (mask & AT_MODE) {
3358 uint64_t pmode = zp->z_mode;
3360 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3362 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3363 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3364 err = SET_ERROR(EPERM);
3368 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3371 mutex_enter(&zp->z_lock);
3372 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3374 * Are we upgrading ACL from old V0 format
3377 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3378 zfs_znode_acl_version(zp) ==
3379 ZFS_ACL_VERSION_INITIAL) {
3380 dmu_tx_hold_free(tx, acl_obj, 0,
3382 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3383 0, aclp->z_acl_bytes);
3385 dmu_tx_hold_write(tx, acl_obj, 0,
3388 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3389 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3390 0, aclp->z_acl_bytes);
3392 mutex_exit(&zp->z_lock);
3393 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3395 if ((mask & AT_XVATTR) &&
3396 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3397 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3399 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3403 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3406 fuid_dirtied = zfsvfs->z_fuid_dirty;
3408 zfs_fuid_txhold(zfsvfs, tx);
3410 zfs_sa_upgrade_txholds(tx, zp);
3412 err = dmu_tx_assign(tx, TXG_WAIT);
3418 * Set each attribute requested.
3419 * We group settings according to the locks they need to acquire.
3421 * Note: you cannot set ctime directly, although it will be
3422 * updated as a side-effect of calling this function.
3426 if (mask & (AT_UID|AT_GID|AT_MODE))
3427 mutex_enter(&zp->z_acl_lock);
3428 mutex_enter(&zp->z_lock);
3430 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3431 &zp->z_pflags, sizeof (zp->z_pflags));
3434 if (mask & (AT_UID|AT_GID|AT_MODE))
3435 mutex_enter(&attrzp->z_acl_lock);
3436 mutex_enter(&attrzp->z_lock);
3437 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3438 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3439 sizeof (attrzp->z_pflags));
3442 if (mask & (AT_UID|AT_GID)) {
3444 if (mask & AT_UID) {
3445 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3446 &new_uid, sizeof (new_uid));
3447 zp->z_uid = new_uid;
3449 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3450 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3452 attrzp->z_uid = new_uid;
3456 if (mask & AT_GID) {
3457 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3458 NULL, &new_gid, sizeof (new_gid));
3459 zp->z_gid = new_gid;
3461 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3462 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3464 attrzp->z_gid = new_gid;
3467 if (!(mask & AT_MODE)) {
3468 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3469 NULL, &new_mode, sizeof (new_mode));
3470 new_mode = zp->z_mode;
3472 err = zfs_acl_chown_setattr(zp);
3475 err = zfs_acl_chown_setattr(attrzp);
3480 if (mask & AT_MODE) {
3481 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3482 &new_mode, sizeof (new_mode));
3483 zp->z_mode = new_mode;
3484 ASSERT3U((uintptr_t)aclp, !=, 0);
3485 err = zfs_aclset_common(zp, aclp, cr, tx);
3487 if (zp->z_acl_cached)
3488 zfs_acl_free(zp->z_acl_cached);
3489 zp->z_acl_cached = aclp;
3494 if (mask & AT_ATIME) {
3495 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3496 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3497 &zp->z_atime, sizeof (zp->z_atime));
3500 if (mask & AT_MTIME) {
3501 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3502 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3503 mtime, sizeof (mtime));
3506 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3507 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3508 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3509 NULL, mtime, sizeof (mtime));
3510 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3511 &ctime, sizeof (ctime));
3512 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3514 } else if (mask != 0) {
3515 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3516 &ctime, sizeof (ctime));
3517 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3520 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3521 SA_ZPL_CTIME(zfsvfs), NULL,
3522 &ctime, sizeof (ctime));
3523 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3524 mtime, ctime, B_TRUE);
3528 * Do this after setting timestamps to prevent timestamp
3529 * update from toggling bit
3532 if (xoap && (mask & AT_XVATTR)) {
3535 * restore trimmed off masks
3536 * so that return masks can be set for caller.
3539 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3540 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3542 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3543 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3545 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3546 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3548 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3549 XVA_SET_REQ(xvap, XAT_NODUMP);
3551 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3552 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3554 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3555 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3558 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3559 ASSERT(vp->v_type == VREG);
3561 zfs_xvattr_set(zp, xvap, tx);
3565 zfs_fuid_sync(zfsvfs, tx);
3568 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3570 mutex_exit(&zp->z_lock);
3571 if (mask & (AT_UID|AT_GID|AT_MODE))
3572 mutex_exit(&zp->z_acl_lock);
3575 if (mask & (AT_UID|AT_GID|AT_MODE))
3576 mutex_exit(&attrzp->z_acl_lock);
3577 mutex_exit(&attrzp->z_lock);
3580 if (err == 0 && attrzp) {
3581 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3587 VN_RELE(ZTOV(attrzp));
3593 zfs_fuid_info_free(fuidp);
3599 if (err == ERESTART)
3602 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3607 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3608 zil_commit(zilog, 0);
3614 typedef struct zfs_zlock {
3615 krwlock_t *zl_rwlock; /* lock we acquired */
3616 znode_t *zl_znode; /* znode we held */
3617 struct zfs_zlock *zl_next; /* next in list */
3621 * Drop locks and release vnodes that were held by zfs_rename_lock().
3624 zfs_rename_unlock(zfs_zlock_t **zlpp)
3628 while ((zl = *zlpp) != NULL) {
3629 if (zl->zl_znode != NULL)
3630 VN_RELE(ZTOV(zl->zl_znode));
3631 rw_exit(zl->zl_rwlock);
3632 *zlpp = zl->zl_next;
3633 kmem_free(zl, sizeof (*zl));
3638 * Search back through the directory tree, using the ".." entries.
3639 * Lock each directory in the chain to prevent concurrent renames.
3640 * Fail any attempt to move a directory into one of its own descendants.
3641 * XXX - z_parent_lock can overlap with map or grow locks
3644 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3648 uint64_t rootid = zp->z_zfsvfs->z_root;
3649 uint64_t oidp = zp->z_id;
3650 krwlock_t *rwlp = &szp->z_parent_lock;
3651 krw_t rw = RW_WRITER;
3654 * First pass write-locks szp and compares to zp->z_id.
3655 * Later passes read-lock zp and compare to zp->z_parent.
3658 if (!rw_tryenter(rwlp, rw)) {
3660 * Another thread is renaming in this path.
3661 * Note that if we are a WRITER, we don't have any
3662 * parent_locks held yet.
3664 if (rw == RW_READER && zp->z_id > szp->z_id) {
3666 * Drop our locks and restart
3668 zfs_rename_unlock(&zl);
3672 rwlp = &szp->z_parent_lock;
3677 * Wait for other thread to drop its locks
3683 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3684 zl->zl_rwlock = rwlp;
3685 zl->zl_znode = NULL;
3686 zl->zl_next = *zlpp;
3689 if (oidp == szp->z_id) /* We're a descendant of szp */
3690 return (SET_ERROR(EINVAL));
3692 if (oidp == rootid) /* We've hit the top */
3695 if (rw == RW_READER) { /* i.e. not the first pass */
3696 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3701 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3702 &oidp, sizeof (oidp));
3703 rwlp = &zp->z_parent_lock;
3706 } while (zp->z_id != sdzp->z_id);
3712 * Move an entry from the provided source directory to the target
3713 * directory. Change the entry name as indicated.
3715 * IN: sdvp - Source directory containing the "old entry".
3716 * snm - Old entry name.
3717 * tdvp - Target directory to contain the "new entry".
3718 * tnm - New entry name.
3719 * cr - credentials of caller.
3720 * ct - caller context
3721 * flags - case flags
3723 * RETURN: 0 on success, error code on failure.
3726 * sdvp,tdvp - ctime|mtime updated
3730 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3731 caller_context_t *ct, int flags)
3733 znode_t *tdzp, *szp, *tzp;
3734 znode_t *sdzp = VTOZ(sdvp);
3735 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs;
3738 zfs_dirlock_t *sdl, *tdl;
3741 int cmp, serr, terr;
3744 boolean_t waited = B_FALSE;
3747 ZFS_VERIFY_ZP(sdzp);
3748 zilog = zfsvfs->z_log;
3751 * Make sure we have the real vp for the target directory.
3753 if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3757 ZFS_VERIFY_ZP(tdzp);
3760 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3761 * ctldir appear to have the same v_vfsp.
3763 if (tdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) {
3765 return (SET_ERROR(EXDEV));
3768 if (zfsvfs->z_utf8 && u8_validate(tnm,
3769 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3771 return (SET_ERROR(EILSEQ));
3774 if (flags & FIGNORECASE)
3783 * This is to prevent the creation of links into attribute space
3784 * by renaming a linked file into/outof an attribute directory.
3785 * See the comment in zfs_link() for why this is considered bad.
3787 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3789 return (SET_ERROR(EINVAL));
3793 * Lock source and target directory entries. To prevent deadlock,
3794 * a lock ordering must be defined. We lock the directory with
3795 * the smallest object id first, or if it's a tie, the one with
3796 * the lexically first name.
3798 if (sdzp->z_id < tdzp->z_id) {
3800 } else if (sdzp->z_id > tdzp->z_id) {
3804 * First compare the two name arguments without
3805 * considering any case folding.
3807 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3809 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3810 ASSERT(error == 0 || !zfsvfs->z_utf8);
3813 * POSIX: "If the old argument and the new argument
3814 * both refer to links to the same existing file,
3815 * the rename() function shall return successfully
3816 * and perform no other action."
3822 * If the file system is case-folding, then we may
3823 * have some more checking to do. A case-folding file
3824 * system is either supporting mixed case sensitivity
3825 * access or is completely case-insensitive. Note
3826 * that the file system is always case preserving.
3828 * In mixed sensitivity mode case sensitive behavior
3829 * is the default. FIGNORECASE must be used to
3830 * explicitly request case insensitive behavior.
3832 * If the source and target names provided differ only
3833 * by case (e.g., a request to rename 'tim' to 'Tim'),
3834 * we will treat this as a special case in the
3835 * case-insensitive mode: as long as the source name
3836 * is an exact match, we will allow this to proceed as
3837 * a name-change request.
3839 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3840 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3841 flags & FIGNORECASE)) &&
3842 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3845 * case preserving rename request, require exact
3854 * If the source and destination directories are the same, we should
3855 * grab the z_name_lock of that directory only once.
3859 rw_enter(&sdzp->z_name_lock, RW_READER);
3863 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3864 ZEXISTS | zflg, NULL, NULL);
3865 terr = zfs_dirent_lock(&tdl,
3866 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3868 terr = zfs_dirent_lock(&tdl,
3869 tdzp, tnm, &tzp, zflg, NULL, NULL);
3870 serr = zfs_dirent_lock(&sdl,
3871 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3877 * Source entry invalid or not there.
3880 zfs_dirent_unlock(tdl);
3886 rw_exit(&sdzp->z_name_lock);
3889 * FreeBSD: In OpenSolaris they only check if rename source is
3890 * ".." here, because "." is handled in their lookup. This is
3891 * not the case for FreeBSD, so we check for "." explicitly.
3893 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0)
3894 serr = SET_ERROR(EINVAL);
3899 zfs_dirent_unlock(sdl);
3903 rw_exit(&sdzp->z_name_lock);
3905 if (strcmp(tnm, "..") == 0)
3906 terr = SET_ERROR(EINVAL);
3912 * Must have write access at the source to remove the old entry
3913 * and write access at the target to create the new entry.
3914 * Note that if target and source are the same, this can be
3915 * done in a single check.
3918 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3921 if (ZTOV(szp)->v_type == VDIR) {
3923 * Check to make sure rename is valid.
3924 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3926 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3931 * Does target exist?
3935 * Source and target must be the same type.
3937 if (ZTOV(szp)->v_type == VDIR) {
3938 if (ZTOV(tzp)->v_type != VDIR) {
3939 error = SET_ERROR(ENOTDIR);
3943 if (ZTOV(tzp)->v_type == VDIR) {
3944 error = SET_ERROR(EISDIR);
3949 * POSIX dictates that when the source and target
3950 * entries refer to the same file object, rename
3951 * must do nothing and exit without error.
3953 if (szp->z_id == tzp->z_id) {
3959 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3961 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3964 * notify the target directory if it is not the same
3965 * as source directory.
3968 vnevent_rename_dest_dir(tdvp, ct);
3971 tx = dmu_tx_create(zfsvfs->z_os);
3972 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3973 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3974 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3975 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3977 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3978 zfs_sa_upgrade_txholds(tx, tdzp);
3981 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3982 zfs_sa_upgrade_txholds(tx, tzp);
3985 zfs_sa_upgrade_txholds(tx, szp);
3986 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3987 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3990 zfs_rename_unlock(&zl);
3991 zfs_dirent_unlock(sdl);
3992 zfs_dirent_unlock(tdl);
3995 rw_exit(&sdzp->z_name_lock);
4000 if (error == ERESTART) {
4011 if (tzp) /* Attempt to remove the existing target */
4012 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
4015 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
4017 szp->z_pflags |= ZFS_AV_MODIFIED;
4019 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
4020 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
4023 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
4025 zfs_log_rename(zilog, tx, TX_RENAME |
4026 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
4027 sdl->dl_name, tdzp, tdl->dl_name, szp);
4030 * Update path information for the target vnode
4032 vn_renamepath(tdvp, ZTOV(szp), tnm,
4036 * At this point, we have successfully created
4037 * the target name, but have failed to remove
4038 * the source name. Since the create was done
4039 * with the ZRENAMING flag, there are
4040 * complications; for one, the link count is
4041 * wrong. The easiest way to deal with this
4042 * is to remove the newly created target, and
4043 * return the original error. This must
4044 * succeed; fortunately, it is very unlikely to
4045 * fail, since we just created it.
4047 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
4048 ZRENAMING, NULL), ==, 0);
4051 #ifdef FREEBSD_NAMECACHE
4055 cache_purge(ZTOV(szp));
4057 cache_purge(ZTOV(tzp));
4065 zfs_rename_unlock(&zl);
4067 zfs_dirent_unlock(sdl);
4068 zfs_dirent_unlock(tdl);
4071 rw_exit(&sdzp->z_name_lock);
4078 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4079 zil_commit(zilog, 0);
4087 * Insert the indicated symbolic reference entry into the directory.
4089 * IN: dvp - Directory to contain new symbolic link.
4090 * link - Name for new symlink entry.
4091 * vap - Attributes of new entry.
4092 * cr - credentials of caller.
4093 * ct - caller context
4094 * flags - case flags
4096 * RETURN: 0 on success, error code on failure.
4099 * dvp - ctime|mtime updated
4103 zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link,
4104 cred_t *cr, kthread_t *td)
4106 znode_t *zp, *dzp = VTOZ(dvp);
4109 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4111 uint64_t len = strlen(link);
4114 zfs_acl_ids_t acl_ids;
4115 boolean_t fuid_dirtied;
4116 uint64_t txtype = TX_SYMLINK;
4117 boolean_t waited = B_FALSE;
4120 ASSERT(vap->va_type == VLNK);
4124 zilog = zfsvfs->z_log;
4126 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
4127 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4129 return (SET_ERROR(EILSEQ));
4131 if (flags & FIGNORECASE)
4134 if (len > MAXPATHLEN) {
4136 return (SET_ERROR(ENAMETOOLONG));
4139 if ((error = zfs_acl_ids_create(dzp, 0,
4140 vap, cr, NULL, &acl_ids)) != 0) {
4145 getnewvnode_reserve(1);
4149 * Attempt to lock directory; fail if entry already exists.
4151 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
4153 zfs_acl_ids_free(&acl_ids);
4154 getnewvnode_drop_reserve();
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);
4162 getnewvnode_drop_reserve();
4167 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
4168 zfs_acl_ids_free(&acl_ids);
4169 zfs_dirent_unlock(dl);
4170 getnewvnode_drop_reserve();
4172 return (SET_ERROR(EDQUOT));
4174 tx = dmu_tx_create(zfsvfs->z_os);
4175 fuid_dirtied = zfsvfs->z_fuid_dirty;
4176 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
4177 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4178 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
4179 ZFS_SA_BASE_ATTR_SIZE + len);
4180 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
4181 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
4182 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
4183 acl_ids.z_aclp->z_acl_bytes);
4186 zfs_fuid_txhold(zfsvfs, tx);
4187 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4189 zfs_dirent_unlock(dl);
4190 if (error == ERESTART) {
4196 zfs_acl_ids_free(&acl_ids);
4198 getnewvnode_drop_reserve();
4204 * Create a new object for the symlink.
4205 * for version 4 ZPL datsets the symlink will be an SA attribute
4207 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
4210 zfs_fuid_sync(zfsvfs, tx);
4212 mutex_enter(&zp->z_lock);
4214 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
4217 zfs_sa_symlink(zp, link, len, tx);
4218 mutex_exit(&zp->z_lock);
4221 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
4222 &zp->z_size, sizeof (zp->z_size), tx);
4224 * Insert the new object into the directory.
4226 (void) zfs_link_create(dl, zp, tx, ZNEW);
4228 if (flags & FIGNORECASE)
4230 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
4233 zfs_acl_ids_free(&acl_ids);
4237 getnewvnode_drop_reserve();
4239 zfs_dirent_unlock(dl);
4241 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4242 zil_commit(zilog, 0);
4249 * Return, in the buffer contained in the provided uio structure,
4250 * the symbolic path referred to by vp.
4252 * IN: vp - vnode of symbolic link.
4253 * uio - structure to contain the link path.
4254 * cr - credentials of caller.
4255 * ct - caller context
4257 * OUT: uio - structure containing the link path.
4259 * RETURN: 0 on success, error code on failure.
4262 * vp - atime updated
4266 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
4268 znode_t *zp = VTOZ(vp);
4269 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4275 mutex_enter(&zp->z_lock);
4277 error = sa_lookup_uio(zp->z_sa_hdl,
4278 SA_ZPL_SYMLINK(zfsvfs), uio);
4280 error = zfs_sa_readlink(zp, uio);
4281 mutex_exit(&zp->z_lock);
4283 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4290 * Insert a new entry into directory tdvp referencing svp.
4292 * IN: tdvp - Directory to contain new entry.
4293 * svp - vnode of new entry.
4294 * name - name of new entry.
4295 * cr - credentials of caller.
4296 * ct - caller context
4298 * RETURN: 0 on success, error code on failure.
4301 * tdvp - ctime|mtime updated
4302 * svp - ctime updated
4306 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
4307 caller_context_t *ct, int flags)
4309 znode_t *dzp = VTOZ(tdvp);
4311 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4320 boolean_t waited = B_FALSE;
4322 ASSERT(tdvp->v_type == VDIR);
4326 zilog = zfsvfs->z_log;
4328 if (VOP_REALVP(svp, &realvp, ct) == 0)
4332 * POSIX dictates that we return EPERM here.
4333 * Better choices include ENOTSUP or EISDIR.
4335 if (svp->v_type == VDIR) {
4337 return (SET_ERROR(EPERM));
4344 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
4345 * ctldir appear to have the same v_vfsp.
4347 if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) {
4349 return (SET_ERROR(EXDEV));
4352 /* Prevent links to .zfs/shares files */
4354 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4355 &parent, sizeof (uint64_t))) != 0) {
4359 if (parent == zfsvfs->z_shares_dir) {
4361 return (SET_ERROR(EPERM));
4364 if (zfsvfs->z_utf8 && u8_validate(name,
4365 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4367 return (SET_ERROR(EILSEQ));
4369 if (flags & FIGNORECASE)
4373 * We do not support links between attributes and non-attributes
4374 * because of the potential security risk of creating links
4375 * into "normal" file space in order to circumvent restrictions
4376 * imposed in attribute space.
4378 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4380 return (SET_ERROR(EINVAL));
4384 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4385 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) {
4387 return (SET_ERROR(EPERM));
4390 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4397 * Attempt to lock directory; fail if entry already exists.
4399 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4405 tx = dmu_tx_create(zfsvfs->z_os);
4406 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4407 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4408 zfs_sa_upgrade_txholds(tx, szp);
4409 zfs_sa_upgrade_txholds(tx, dzp);
4410 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4412 zfs_dirent_unlock(dl);
4413 if (error == ERESTART) {
4424 error = zfs_link_create(dl, szp, tx, 0);
4427 uint64_t txtype = TX_LINK;
4428 if (flags & FIGNORECASE)
4430 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4435 zfs_dirent_unlock(dl);
4438 vnevent_link(svp, ct);
4441 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4442 zil_commit(zilog, 0);
4450 * zfs_null_putapage() is used when the file system has been force
4451 * unmounted. It just drops the pages.
4455 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4456 size_t *lenp, int flags, cred_t *cr)
4458 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4463 * Push a page out to disk, klustering if possible.
4465 * IN: vp - file to push page to.
4466 * pp - page to push.
4467 * flags - additional flags.
4468 * cr - credentials of caller.
4470 * OUT: offp - start of range pushed.
4471 * lenp - len of range pushed.
4473 * RETURN: 0 on success, error code on failure.
4475 * NOTE: callers must have locked the page to be pushed. On
4476 * exit, the page (and all other pages in the kluster) must be
4481 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4482 size_t *lenp, int flags, cred_t *cr)
4484 znode_t *zp = VTOZ(vp);
4485 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4487 u_offset_t off, koff;
4494 * If our blocksize is bigger than the page size, try to kluster
4495 * multiple pages so that we write a full block (thus avoiding
4496 * a read-modify-write).
4498 if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4499 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4500 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4501 ASSERT(koff <= zp->z_size);
4502 if (koff + klen > zp->z_size)
4503 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4504 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4506 ASSERT3U(btop(len), ==, btopr(len));
4509 * Can't push pages past end-of-file.
4511 if (off >= zp->z_size) {
4512 /* ignore all pages */
4515 } else if (off + len > zp->z_size) {
4516 int npages = btopr(zp->z_size - off);
4519 page_list_break(&pp, &trunc, npages);
4520 /* ignore pages past end of file */
4522 pvn_write_done(trunc, flags);
4523 len = zp->z_size - off;
4526 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4527 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4528 err = SET_ERROR(EDQUOT);
4531 tx = dmu_tx_create(zfsvfs->z_os);
4532 dmu_tx_hold_write(tx, zp->z_id, off, len);
4534 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4535 zfs_sa_upgrade_txholds(tx, zp);
4536 err = dmu_tx_assign(tx, TXG_WAIT);
4542 if (zp->z_blksz <= PAGESIZE) {
4543 caddr_t va = zfs_map_page(pp, S_READ);
4544 ASSERT3U(len, <=, PAGESIZE);
4545 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4546 zfs_unmap_page(pp, va);
4548 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4552 uint64_t mtime[2], ctime[2];
4553 sa_bulk_attr_t bulk[3];
4556 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4558 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4560 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4562 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4564 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4569 pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4579 * Copy the portion of the file indicated from pages into the file.
4580 * The pages are stored in a page list attached to the files vnode.
4582 * IN: vp - vnode of file to push page data to.
4583 * off - position in file to put data.
4584 * len - amount of data to write.
4585 * flags - flags to control the operation.
4586 * cr - credentials of caller.
4587 * ct - caller context.
4589 * RETURN: 0 on success, error code on failure.
4592 * vp - ctime|mtime updated
4596 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4597 caller_context_t *ct)
4599 znode_t *zp = VTOZ(vp);
4600 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4612 * Align this request to the file block size in case we kluster.
4613 * XXX - this can result in pretty aggresive locking, which can
4614 * impact simultanious read/write access. One option might be
4615 * to break up long requests (len == 0) into block-by-block
4616 * operations to get narrower locking.
4618 blksz = zp->z_blksz;
4620 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4623 if (len > 0 && ISP2(blksz))
4624 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4630 * Search the entire vp list for pages >= io_off.
4632 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4633 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4636 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4638 if (off > zp->z_size) {
4639 /* past end of file */
4640 zfs_range_unlock(rl);
4645 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4647 for (off = io_off; io_off < off + len; io_off += io_len) {
4648 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4649 pp = page_lookup(vp, io_off,
4650 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4652 pp = page_lookup_nowait(vp, io_off,
4653 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4656 if (pp != NULL && pvn_getdirty(pp, flags)) {
4660 * Found a dirty page to push
4662 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4670 zfs_range_unlock(rl);
4671 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4672 zil_commit(zfsvfs->z_log, zp->z_id);
4680 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4682 znode_t *zp = VTOZ(vp);
4683 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4686 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4687 if (zp->z_sa_hdl == NULL) {
4689 * The fs has been unmounted, or we did a
4690 * suspend/resume and this file no longer exists.
4692 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4697 mutex_enter(&zp->z_lock);
4698 if (zp->z_unlinked) {
4700 * Fast path to recycle a vnode of a removed file.
4702 mutex_exit(&zp->z_lock);
4703 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4707 mutex_exit(&zp->z_lock);
4709 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4710 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4712 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4713 zfs_sa_upgrade_txholds(tx, zp);
4714 error = dmu_tx_assign(tx, TXG_WAIT);
4718 mutex_enter(&zp->z_lock);
4719 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4720 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4721 zp->z_atime_dirty = 0;
4722 mutex_exit(&zp->z_lock);
4726 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4731 * Bounds-check the seek operation.
4733 * IN: vp - vnode seeking within
4734 * ooff - old file offset
4735 * noffp - pointer to new file offset
4736 * ct - caller context
4738 * RETURN: 0 on success, EINVAL if new offset invalid.
4742 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4743 caller_context_t *ct)
4745 if (vp->v_type == VDIR)
4747 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4751 * Pre-filter the generic locking function to trap attempts to place
4752 * a mandatory lock on a memory mapped file.
4755 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4756 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4758 znode_t *zp = VTOZ(vp);
4759 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4765 * We are following the UFS semantics with respect to mapcnt
4766 * here: If we see that the file is mapped already, then we will
4767 * return an error, but we don't worry about races between this
4768 * function and zfs_map().
4770 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4772 return (SET_ERROR(EAGAIN));
4775 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4779 * If we can't find a page in the cache, we will create a new page
4780 * and fill it with file data. For efficiency, we may try to fill
4781 * multiple pages at once (klustering) to fill up the supplied page
4782 * list. Note that the pages to be filled are held with an exclusive
4783 * lock to prevent access by other threads while they are being filled.
4786 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4787 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4789 znode_t *zp = VTOZ(vp);
4790 page_t *pp, *cur_pp;
4791 objset_t *os = zp->z_zfsvfs->z_os;
4792 u_offset_t io_off, total;
4796 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4798 * We only have a single page, don't bother klustering
4802 pp = page_create_va(vp, io_off, io_len,
4803 PG_EXCL | PG_WAIT, seg, addr);
4806 * Try to find enough pages to fill the page list
4808 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4809 &io_len, off, plsz, 0);
4813 * The page already exists, nothing to do here.
4820 * Fill the pages in the kluster.
4823 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4826 ASSERT3U(io_off, ==, cur_pp->p_offset);
4827 va = zfs_map_page(cur_pp, S_WRITE);
4828 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4830 zfs_unmap_page(cur_pp, va);
4832 /* On error, toss the entire kluster */
4833 pvn_read_done(pp, B_ERROR);
4834 /* convert checksum errors into IO errors */
4836 err = SET_ERROR(EIO);
4839 cur_pp = cur_pp->p_next;
4843 * Fill in the page list array from the kluster starting
4844 * from the desired offset `off'.
4845 * NOTE: the page list will always be null terminated.
4847 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4848 ASSERT(pl == NULL || (*pl)->p_offset == off);
4854 * Return pointers to the pages for the file region [off, off + len]
4855 * in the pl array. If plsz is greater than len, this function may
4856 * also return page pointers from after the specified region
4857 * (i.e. the region [off, off + plsz]). These additional pages are
4858 * only returned if they are already in the cache, or were created as
4859 * part of a klustered read.
4861 * IN: vp - vnode of file to get data from.
4862 * off - position in file to get data from.
4863 * len - amount of data to retrieve.
4864 * plsz - length of provided page list.
4865 * seg - segment to obtain pages for.
4866 * addr - virtual address of fault.
4867 * rw - mode of created pages.
4868 * cr - credentials of caller.
4869 * ct - caller context.
4871 * OUT: protp - protection mode of created pages.
4872 * pl - list of pages created.
4874 * RETURN: 0 on success, error code on failure.
4877 * vp - atime updated
4881 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4882 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4883 enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4885 znode_t *zp = VTOZ(vp);
4886 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4890 /* we do our own caching, faultahead is unnecessary */
4893 else if (len > plsz)
4896 len = P2ROUNDUP(len, PAGESIZE);
4897 ASSERT(plsz >= len);
4906 * Loop through the requested range [off, off + len) looking
4907 * for pages. If we don't find a page, we will need to create
4908 * a new page and fill it with data from the file.
4911 if (*pl = page_lookup(vp, off, SE_SHARED))
4913 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4916 ASSERT3U((*pl)->p_offset, ==, off);
4920 ASSERT3U(len, >=, PAGESIZE);
4923 ASSERT3U(plsz, >=, PAGESIZE);
4930 * Fill out the page array with any pages already in the cache.
4933 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4940 * Release any pages we have previously locked.
4945 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4955 * Request a memory map for a section of a file. This code interacts
4956 * with common code and the VM system as follows:
4958 * - common code calls mmap(), which ends up in smmap_common()
4959 * - this calls VOP_MAP(), which takes you into (say) zfs
4960 * - zfs_map() calls as_map(), passing segvn_create() as the callback
4961 * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4962 * - zfs_addmap() updates z_mapcnt
4966 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4967 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4968 caller_context_t *ct)
4970 znode_t *zp = VTOZ(vp);
4971 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4972 segvn_crargs_t vn_a;
4978 if ((prot & PROT_WRITE) && (zp->z_pflags &
4979 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4981 return (SET_ERROR(EPERM));
4984 if ((prot & (PROT_READ | PROT_EXEC)) &&
4985 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4987 return (SET_ERROR(EACCES));
4990 if (vp->v_flag & VNOMAP) {
4992 return (SET_ERROR(ENOSYS));
4995 if (off < 0 || len > MAXOFFSET_T - off) {
4997 return (SET_ERROR(ENXIO));
5000 if (vp->v_type != VREG) {
5002 return (SET_ERROR(ENODEV));
5006 * If file is locked, disallow mapping.
5008 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
5010 return (SET_ERROR(EAGAIN));
5014 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
5022 vn_a.offset = (u_offset_t)off;
5023 vn_a.type = flags & MAP_TYPE;
5025 vn_a.maxprot = maxprot;
5028 vn_a.flags = flags & ~MAP_TYPE;
5030 vn_a.lgrp_mem_policy_flags = 0;
5032 error = as_map(as, *addrp, len, segvn_create, &vn_a);
5041 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5042 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
5043 caller_context_t *ct)
5045 uint64_t pages = btopr(len);
5047 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
5052 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
5053 * more accurate mtime for the associated file. Since we don't have a way of
5054 * detecting when the data was actually modified, we have to resort to
5055 * heuristics. If an explicit msync() is done, then we mark the mtime when the
5056 * last page is pushed. The problem occurs when the msync() call is omitted,
5057 * which by far the most common case:
5065 * putpage() via fsflush
5067 * If we wait until fsflush to come along, we can have a modification time that
5068 * is some arbitrary point in the future. In order to prevent this in the
5069 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
5074 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5075 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
5076 caller_context_t *ct)
5078 uint64_t pages = btopr(len);
5080 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
5081 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
5083 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
5084 vn_has_cached_data(vp))
5085 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
5091 * Free or allocate space in a file. Currently, this function only
5092 * supports the `F_FREESP' command. However, this command is somewhat
5093 * misnamed, as its functionality includes the ability to allocate as
5094 * well as free space.
5096 * IN: vp - vnode of file to free data in.
5097 * cmd - action to take (only F_FREESP supported).
5098 * bfp - section of file to free/alloc.
5099 * flag - current file open mode flags.
5100 * offset - current file offset.
5101 * cr - credentials of caller [UNUSED].
5102 * ct - caller context.
5104 * RETURN: 0 on success, error code on failure.
5107 * vp - ctime|mtime updated
5111 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
5112 offset_t offset, cred_t *cr, caller_context_t *ct)
5114 znode_t *zp = VTOZ(vp);
5115 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5122 if (cmd != F_FREESP) {
5124 return (SET_ERROR(EINVAL));
5127 if (error = convoff(vp, bfp, 0, offset)) {
5132 if (bfp->l_len < 0) {
5134 return (SET_ERROR(EINVAL));
5138 len = bfp->l_len; /* 0 means from off to end of file */
5140 error = zfs_freesp(zp, off, len, flag, TRUE);
5147 CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid));
5148 CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid));
5152 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
5154 znode_t *zp = VTOZ(vp);
5155 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5158 uint64_t object = zp->z_id;
5165 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
5166 &gen64, sizeof (uint64_t))) != 0) {
5171 gen = (uint32_t)gen64;
5173 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
5176 if (fidp->fid_len < size) {
5177 fidp->fid_len = size;
5179 return (SET_ERROR(ENOSPC));
5182 fidp->fid_len = size;
5185 zfid = (zfid_short_t *)fidp;
5187 zfid->zf_len = size;
5189 for (i = 0; i < sizeof (zfid->zf_object); i++)
5190 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
5192 /* Must have a non-zero generation number to distinguish from .zfs */
5195 for (i = 0; i < sizeof (zfid->zf_gen); i++)
5196 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
5198 if (size == LONG_FID_LEN) {
5199 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
5202 zlfid = (zfid_long_t *)fidp;
5204 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
5205 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
5207 /* XXX - this should be the generation number for the objset */
5208 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
5209 zlfid->zf_setgen[i] = 0;
5217 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
5218 caller_context_t *ct)
5230 case _PC_FILESIZEBITS:
5234 case _PC_XATTR_EXISTS:
5236 zfsvfs = zp->z_zfsvfs;
5240 error = zfs_dirent_lock(&dl, zp, "", &xzp,
5241 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
5243 zfs_dirent_unlock(dl);
5244 if (!zfs_dirempty(xzp))
5247 } else if (error == ENOENT) {
5249 * If there aren't extended attributes, it's the
5250 * same as having zero of them.
5257 case _PC_SATTR_ENABLED:
5258 case _PC_SATTR_EXISTS:
5259 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
5260 (vp->v_type == VREG || vp->v_type == VDIR);
5263 case _PC_ACCESS_FILTERING:
5264 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
5268 case _PC_ACL_ENABLED:
5269 *valp = _ACL_ACE_ENABLED;
5272 case _PC_MIN_HOLE_SIZE:
5273 *valp = (int)SPA_MINBLOCKSIZE;
5276 case _PC_TIMESTAMP_RESOLUTION:
5277 /* nanosecond timestamp resolution */
5281 case _PC_ACL_EXTENDED:
5289 case _PC_ACL_PATH_MAX:
5290 *valp = ACL_MAX_ENTRIES;
5294 return (EOPNOTSUPP);
5300 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5301 caller_context_t *ct)
5303 znode_t *zp = VTOZ(vp);
5304 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5306 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5310 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5318 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5319 caller_context_t *ct)
5321 znode_t *zp = VTOZ(vp);
5322 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5324 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5325 zilog_t *zilog = zfsvfs->z_log;
5330 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5332 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5333 zil_commit(zilog, 0);
5341 * The smallest read we may consider to loan out an arcbuf.
5342 * This must be a power of 2.
5344 int zcr_blksz_min = (1 << 10); /* 1K */
5346 * If set to less than the file block size, allow loaning out of an
5347 * arcbuf for a partial block read. This must be a power of 2.
5349 int zcr_blksz_max = (1 << 17); /* 128K */
5353 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5354 caller_context_t *ct)
5356 znode_t *zp = VTOZ(vp);
5357 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5358 int max_blksz = zfsvfs->z_max_blksz;
5359 uio_t *uio = &xuio->xu_uio;
5360 ssize_t size = uio->uio_resid;
5361 offset_t offset = uio->uio_loffset;
5366 int preamble, postamble;
5368 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5369 return (SET_ERROR(EINVAL));
5376 * Loan out an arc_buf for write if write size is bigger than
5377 * max_blksz, and the file's block size is also max_blksz.
5380 if (size < blksz || zp->z_blksz != blksz) {
5382 return (SET_ERROR(EINVAL));
5385 * Caller requests buffers for write before knowing where the
5386 * write offset might be (e.g. NFS TCP write).
5391 preamble = P2PHASE(offset, blksz);
5393 preamble = blksz - preamble;
5398 postamble = P2PHASE(size, blksz);
5401 fullblk = size / blksz;
5402 (void) dmu_xuio_init(xuio,
5403 (preamble != 0) + fullblk + (postamble != 0));
5404 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5405 int, postamble, int,
5406 (preamble != 0) + fullblk + (postamble != 0));
5409 * Have to fix iov base/len for partial buffers. They
5410 * currently represent full arc_buf's.
5413 /* data begins in the middle of the arc_buf */
5414 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5417 (void) dmu_xuio_add(xuio, abuf,
5418 blksz - preamble, preamble);
5421 for (i = 0; i < fullblk; i++) {
5422 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5425 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5429 /* data ends in the middle of the arc_buf */
5430 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5433 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5438 * Loan out an arc_buf for read if the read size is larger than
5439 * the current file block size. Block alignment is not
5440 * considered. Partial arc_buf will be loaned out for read.
5442 blksz = zp->z_blksz;
5443 if (blksz < zcr_blksz_min)
5444 blksz = zcr_blksz_min;
5445 if (blksz > zcr_blksz_max)
5446 blksz = zcr_blksz_max;
5447 /* avoid potential complexity of dealing with it */
5448 if (blksz > max_blksz) {
5450 return (SET_ERROR(EINVAL));
5453 maxsize = zp->z_size - uio->uio_loffset;
5457 if (size < blksz || vn_has_cached_data(vp)) {
5459 return (SET_ERROR(EINVAL));
5464 return (SET_ERROR(EINVAL));
5467 uio->uio_extflg = UIO_XUIO;
5468 XUIO_XUZC_RW(xuio) = ioflag;
5475 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5479 int ioflag = XUIO_XUZC_RW(xuio);
5481 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5483 i = dmu_xuio_cnt(xuio);
5485 abuf = dmu_xuio_arcbuf(xuio, i);
5487 * if abuf == NULL, it must be a write buffer
5488 * that has been returned in zfs_write().
5491 dmu_return_arcbuf(abuf);
5492 ASSERT(abuf || ioflag == UIO_WRITE);
5495 dmu_xuio_fini(xuio);
5500 * Predeclare these here so that the compiler assumes that
5501 * this is an "old style" function declaration that does
5502 * not include arguments => we won't get type mismatch errors
5503 * in the initializations that follow.
5505 static int zfs_inval();
5506 static int zfs_isdir();
5511 return (SET_ERROR(EINVAL));
5517 return (SET_ERROR(EISDIR));
5520 * Directory vnode operations template
5522 vnodeops_t *zfs_dvnodeops;
5523 const fs_operation_def_t zfs_dvnodeops_template[] = {
5524 VOPNAME_OPEN, { .vop_open = zfs_open },
5525 VOPNAME_CLOSE, { .vop_close = zfs_close },
5526 VOPNAME_READ, { .error = zfs_isdir },
5527 VOPNAME_WRITE, { .error = zfs_isdir },
5528 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5529 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5530 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5531 VOPNAME_ACCESS, { .vop_access = zfs_access },
5532 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5533 VOPNAME_CREATE, { .vop_create = zfs_create },
5534 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5535 VOPNAME_LINK, { .vop_link = zfs_link },
5536 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5537 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir },
5538 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5539 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5540 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink },
5541 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5542 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5543 VOPNAME_FID, { .vop_fid = zfs_fid },
5544 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5545 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5546 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5547 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5548 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5553 * Regular file vnode operations template
5555 vnodeops_t *zfs_fvnodeops;
5556 const fs_operation_def_t zfs_fvnodeops_template[] = {
5557 VOPNAME_OPEN, { .vop_open = zfs_open },
5558 VOPNAME_CLOSE, { .vop_close = zfs_close },
5559 VOPNAME_READ, { .vop_read = zfs_read },
5560 VOPNAME_WRITE, { .vop_write = zfs_write },
5561 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5562 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5563 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5564 VOPNAME_ACCESS, { .vop_access = zfs_access },
5565 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5566 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5567 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5568 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5569 VOPNAME_FID, { .vop_fid = zfs_fid },
5570 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5571 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock },
5572 VOPNAME_SPACE, { .vop_space = zfs_space },
5573 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage },
5574 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage },
5575 VOPNAME_MAP, { .vop_map = zfs_map },
5576 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap },
5577 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap },
5578 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5579 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5580 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5581 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5582 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf },
5583 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf },
5588 * Symbolic link vnode operations template
5590 vnodeops_t *zfs_symvnodeops;
5591 const fs_operation_def_t zfs_symvnodeops_template[] = {
5592 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5593 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5594 VOPNAME_ACCESS, { .vop_access = zfs_access },
5595 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5596 VOPNAME_READLINK, { .vop_readlink = zfs_readlink },
5597 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5598 VOPNAME_FID, { .vop_fid = zfs_fid },
5599 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5600 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5605 * special share hidden files vnode operations template
5607 vnodeops_t *zfs_sharevnodeops;
5608 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5609 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5610 VOPNAME_ACCESS, { .vop_access = zfs_access },
5611 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5612 VOPNAME_FID, { .vop_fid = zfs_fid },
5613 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5614 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5615 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5616 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5621 * Extended attribute directory vnode operations template
5623 * This template is identical to the directory vnodes
5624 * operation template except for restricted operations:
5628 * Note that there are other restrictions embedded in:
5629 * zfs_create() - restrict type to VREG
5630 * zfs_link() - no links into/out of attribute space
5631 * zfs_rename() - no moves into/out of attribute space
5633 vnodeops_t *zfs_xdvnodeops;
5634 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5635 VOPNAME_OPEN, { .vop_open = zfs_open },
5636 VOPNAME_CLOSE, { .vop_close = zfs_close },
5637 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5638 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5639 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5640 VOPNAME_ACCESS, { .vop_access = zfs_access },
5641 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5642 VOPNAME_CREATE, { .vop_create = zfs_create },
5643 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5644 VOPNAME_LINK, { .vop_link = zfs_link },
5645 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5646 VOPNAME_MKDIR, { .error = zfs_inval },
5647 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5648 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5649 VOPNAME_SYMLINK, { .error = zfs_inval },
5650 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5651 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5652 VOPNAME_FID, { .vop_fid = zfs_fid },
5653 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5654 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5655 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5656 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5657 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5662 * Error vnode operations template
5664 vnodeops_t *zfs_evnodeops;
5665 const fs_operation_def_t zfs_evnodeops_template[] = {
5666 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5667 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5673 ioflags(int ioflags)
5677 if (ioflags & IO_APPEND)
5679 if (ioflags & IO_NDELAY)
5681 if (ioflags & IO_SYNC)
5682 flags |= (FSYNC | FDSYNC | FRSYNC);
5688 zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage)
5690 znode_t *zp = VTOZ(vp);
5691 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5692 objset_t *os = zp->z_zfsvfs->z_os;
5693 vm_page_t mfirst, mlast, mreq;
5697 off_t startoff, endoff;
5699 vm_pindex_t reqstart, reqend;
5700 int pcount, lsize, reqsize, size;
5705 pcount = OFF_TO_IDX(round_page(count));
5707 object = mreq->object;
5710 KASSERT(vp->v_object == object, ("mismatching object"));
5712 if (pcount > 1 && zp->z_blksz > PAGESIZE) {
5713 startoff = rounddown(IDX_TO_OFF(mreq->pindex), zp->z_blksz);
5714 reqstart = OFF_TO_IDX(round_page(startoff));
5715 if (reqstart < m[0]->pindex)
5718 reqstart = reqstart - m[0]->pindex;
5719 endoff = roundup(IDX_TO_OFF(mreq->pindex) + PAGE_SIZE,
5721 reqend = OFF_TO_IDX(trunc_page(endoff)) - 1;
5722 if (reqend > m[pcount - 1]->pindex)
5723 reqend = m[pcount - 1]->pindex;
5724 reqsize = reqend - m[reqstart]->pindex + 1;
5725 KASSERT(reqstart <= reqpage && reqpage < reqstart + reqsize,
5726 ("reqpage beyond [reqstart, reqstart + reqsize[ bounds"));
5731 mfirst = m[reqstart];
5732 mlast = m[reqstart + reqsize - 1];
5734 zfs_vmobject_wlock(object);
5736 for (i = 0; i < reqstart; i++) {
5739 vm_page_unlock(m[i]);
5741 for (i = reqstart + reqsize; i < pcount; i++) {
5744 vm_page_unlock(m[i]);
5747 if (mreq->valid && reqsize == 1) {
5748 if (mreq->valid != VM_PAGE_BITS_ALL)
5749 vm_page_zero_invalid(mreq, TRUE);
5750 zfs_vmobject_wunlock(object);
5752 return (zfs_vm_pagerret_ok);
5755 PCPU_INC(cnt.v_vnodein);
5756 PCPU_ADD(cnt.v_vnodepgsin, reqsize);
5758 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) {
5759 for (i = reqstart; i < reqstart + reqsize; i++) {
5763 vm_page_unlock(m[i]);
5766 zfs_vmobject_wunlock(object);
5768 return (zfs_vm_pagerret_bad);
5772 if (IDX_TO_OFF(mlast->pindex) + lsize > object->un_pager.vnp.vnp_size)
5773 lsize = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mlast->pindex);
5775 zfs_vmobject_wunlock(object);
5777 for (i = reqstart; i < reqstart + reqsize; i++) {
5779 if (i == (reqstart + reqsize - 1))
5781 va = zfs_map_page(m[i], &sf);
5782 error = dmu_read(os, zp->z_id, IDX_TO_OFF(m[i]->pindex),
5783 size, va, DMU_READ_PREFETCH);
5784 if (size != PAGE_SIZE)
5785 bzero(va + size, PAGE_SIZE - size);
5791 zfs_vmobject_wlock(object);
5793 for (i = reqstart; i < reqstart + reqsize; i++) {
5795 m[i]->valid = VM_PAGE_BITS_ALL;
5796 KASSERT(m[i]->dirty == 0, ("zfs_getpages: page %p is dirty", m[i]));
5798 vm_page_readahead_finish(m[i]);
5801 zfs_vmobject_wunlock(object);
5803 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
5805 return (error ? zfs_vm_pagerret_error : zfs_vm_pagerret_ok);
5809 zfs_freebsd_getpages(ap)
5810 struct vop_getpages_args /* {
5815 vm_ooffset_t a_offset;
5819 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage));
5823 zfs_putpages(struct vnode *vp, vm_page_t *ma, size_t len, int flags,
5826 znode_t *zp = VTOZ(vp);
5827 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5836 vm_ooffset_t lo_off;
5847 object = vp->v_object;
5851 KASSERT(ma[0]->object == object, ("mismatching object"));
5852 KASSERT(len > 0 && (len & PAGE_MASK) == 0, ("unexpected length"));
5854 for (i = 0; i < pcount; i++)
5855 rtvals[i] = zfs_vm_pagerret_error;
5857 off = IDX_TO_OFF(ma[0]->pindex);
5858 blksz = zp->z_blksz;
5859 lo_off = rounddown(off, blksz);
5860 lo_len = roundup(len + (off - lo_off), blksz);
5861 rl = zfs_range_lock(zp, lo_off, lo_len, RL_WRITER);
5863 zfs_vmobject_wlock(object);
5864 if (len + off > object->un_pager.vnp.vnp_size) {
5865 if (object->un_pager.vnp.vnp_size > off) {
5868 len = object->un_pager.vnp.vnp_size - off;
5870 if ((pgoff = (int)len & PAGE_MASK) != 0) {
5872 * If the object is locked and the following
5873 * conditions hold, then the page's dirty
5874 * field cannot be concurrently changed by a
5878 vm_page_assert_sbusied(m);
5879 KASSERT(!pmap_page_is_write_mapped(m),
5880 ("zfs_putpages: page %p is not read-only", m));
5881 vm_page_clear_dirty(m, pgoff, PAGE_SIZE -
5888 if (ncount < pcount) {
5889 for (i = ncount; i < pcount; i++) {
5890 rtvals[i] = zfs_vm_pagerret_bad;
5894 zfs_vmobject_wunlock(object);
5899 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
5900 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
5905 tx = dmu_tx_create(zfsvfs->z_os);
5906 dmu_tx_hold_write(tx, zp->z_id, off, len);
5908 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
5909 zfs_sa_upgrade_txholds(tx, zp);
5910 err = dmu_tx_assign(tx, TXG_NOWAIT);
5912 if (err == ERESTART) {
5921 if (zp->z_blksz < PAGE_SIZE) {
5923 for (i = 0; len > 0; off += tocopy, len -= tocopy, i++) {
5924 tocopy = len > PAGE_SIZE ? PAGE_SIZE : len;
5925 va = zfs_map_page(ma[i], &sf);
5926 dmu_write(zfsvfs->z_os, zp->z_id, off, tocopy, va, tx);
5930 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, ma, tx);
5934 uint64_t mtime[2], ctime[2];
5935 sa_bulk_attr_t bulk[3];
5938 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
5940 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
5942 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
5944 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
5946 (void)sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
5947 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
5949 zfs_vmobject_wlock(object);
5950 for (i = 0; i < ncount; i++) {
5951 rtvals[i] = zfs_vm_pagerret_ok;
5952 vm_page_undirty(ma[i]);
5954 zfs_vmobject_wunlock(object);
5955 PCPU_INC(cnt.v_vnodeout);
5956 PCPU_ADD(cnt.v_vnodepgsout, ncount);
5961 zfs_range_unlock(rl);
5962 if ((flags & (zfs_vm_pagerput_sync | zfs_vm_pagerput_inval)) != 0 ||
5963 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5964 zil_commit(zfsvfs->z_log, zp->z_id);
5970 zfs_freebsd_putpages(ap)
5971 struct vop_putpages_args /* {
5977 vm_ooffset_t a_offset;
5981 return (zfs_putpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_sync,
5986 zfs_freebsd_bmap(ap)
5987 struct vop_bmap_args /* {
5990 struct bufobj **a_bop;
5997 if (ap->a_bop != NULL)
5998 *ap->a_bop = &ap->a_vp->v_bufobj;
5999 if (ap->a_bnp != NULL)
6000 *ap->a_bnp = ap->a_bn;
6001 if (ap->a_runp != NULL)
6003 if (ap->a_runb != NULL)
6010 zfs_freebsd_open(ap)
6011 struct vop_open_args /* {
6014 struct ucred *a_cred;
6015 struct thread *a_td;
6018 vnode_t *vp = ap->a_vp;
6019 znode_t *zp = VTOZ(vp);
6022 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL);
6024 vnode_create_vobject(vp, zp->z_size, ap->a_td);
6029 zfs_freebsd_close(ap)
6030 struct vop_close_args /* {
6033 struct ucred *a_cred;
6034 struct thread *a_td;
6038 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL));
6042 zfs_freebsd_ioctl(ap)
6043 struct vop_ioctl_args /* {
6053 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data,
6054 ap->a_fflag, ap->a_cred, NULL, NULL));
6058 zfs_freebsd_read(ap)
6059 struct vop_read_args /* {
6063 struct ucred *a_cred;
6067 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
6072 zfs_freebsd_write(ap)
6073 struct vop_write_args /* {
6077 struct ucred *a_cred;
6081 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
6086 zfs_freebsd_access(ap)
6087 struct vop_access_args /* {
6089 accmode_t a_accmode;
6090 struct ucred *a_cred;
6091 struct thread *a_td;
6094 vnode_t *vp = ap->a_vp;
6095 znode_t *zp = VTOZ(vp);
6100 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND,
6102 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND);
6104 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL);
6107 * VADMIN has to be handled by vaccess().
6110 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND);
6112 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
6113 zp->z_gid, accmode, ap->a_cred, NULL);
6118 * For VEXEC, ensure that at least one execute bit is set for
6121 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR &&
6122 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
6130 zfs_freebsd_lookup(ap)
6131 struct vop_lookup_args /* {
6132 struct vnode *a_dvp;
6133 struct vnode **a_vpp;
6134 struct componentname *a_cnp;
6137 struct componentname *cnp = ap->a_cnp;
6138 char nm[NAME_MAX + 1];
6140 ASSERT(cnp->cn_namelen < sizeof(nm));
6141 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm)));
6143 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop,
6144 cnp->cn_cred, cnp->cn_thread, 0));
6148 zfs_freebsd_create(ap)
6149 struct vop_create_args /* {
6150 struct vnode *a_dvp;
6151 struct vnode **a_vpp;
6152 struct componentname *a_cnp;
6153 struct vattr *a_vap;
6156 struct componentname *cnp = ap->a_cnp;
6157 vattr_t *vap = ap->a_vap;
6160 ASSERT(cnp->cn_flags & SAVENAME);
6162 vattr_init_mask(vap);
6163 mode = vap->va_mode & ALLPERMS;
6165 error = zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode,
6166 ap->a_vpp, cnp->cn_cred, cnp->cn_thread);
6167 #ifdef FREEBSD_NAMECACHE
6168 if (error == 0 && (cnp->cn_flags & MAKEENTRY) != 0)
6169 cache_enter(ap->a_dvp, *ap->a_vpp, cnp);
6175 zfs_freebsd_remove(ap)
6176 struct vop_remove_args /* {
6177 struct vnode *a_dvp;
6179 struct componentname *a_cnp;
6183 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6185 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr,
6186 ap->a_cnp->cn_cred, NULL, 0));
6190 zfs_freebsd_mkdir(ap)
6191 struct vop_mkdir_args /* {
6192 struct vnode *a_dvp;
6193 struct vnode **a_vpp;
6194 struct componentname *a_cnp;
6195 struct vattr *a_vap;
6198 vattr_t *vap = ap->a_vap;
6200 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6202 vattr_init_mask(vap);
6204 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp,
6205 ap->a_cnp->cn_cred, NULL, 0, NULL));
6209 zfs_freebsd_rmdir(ap)
6210 struct vop_rmdir_args /* {
6211 struct vnode *a_dvp;
6213 struct componentname *a_cnp;
6216 struct componentname *cnp = ap->a_cnp;
6218 ASSERT(cnp->cn_flags & SAVENAME);
6220 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0));
6224 zfs_freebsd_readdir(ap)
6225 struct vop_readdir_args /* {
6228 struct ucred *a_cred;
6235 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag,
6236 ap->a_ncookies, ap->a_cookies));
6240 zfs_freebsd_fsync(ap)
6241 struct vop_fsync_args /* {
6244 struct thread *a_td;
6249 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL));
6253 zfs_freebsd_getattr(ap)
6254 struct vop_getattr_args /* {
6256 struct vattr *a_vap;
6257 struct ucred *a_cred;
6260 vattr_t *vap = ap->a_vap;
6266 xvap.xva_vattr = *vap;
6267 xvap.xva_vattr.va_mask |= AT_XVATTR;
6269 /* Convert chflags into ZFS-type flags. */
6270 /* XXX: what about SF_SETTABLE?. */
6271 XVA_SET_REQ(&xvap, XAT_IMMUTABLE);
6272 XVA_SET_REQ(&xvap, XAT_APPENDONLY);
6273 XVA_SET_REQ(&xvap, XAT_NOUNLINK);
6274 XVA_SET_REQ(&xvap, XAT_NODUMP);
6275 XVA_SET_REQ(&xvap, XAT_READONLY);
6276 XVA_SET_REQ(&xvap, XAT_ARCHIVE);
6277 XVA_SET_REQ(&xvap, XAT_SYSTEM);
6278 XVA_SET_REQ(&xvap, XAT_HIDDEN);
6279 XVA_SET_REQ(&xvap, XAT_REPARSE);
6280 XVA_SET_REQ(&xvap, XAT_OFFLINE);
6281 XVA_SET_REQ(&xvap, XAT_SPARSE);
6283 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL);
6287 /* Convert ZFS xattr into chflags. */
6288 #define FLAG_CHECK(fflag, xflag, xfield) do { \
6289 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \
6290 fflags |= (fflag); \
6292 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE,
6293 xvap.xva_xoptattrs.xoa_immutable);
6294 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY,
6295 xvap.xva_xoptattrs.xoa_appendonly);
6296 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK,
6297 xvap.xva_xoptattrs.xoa_nounlink);
6298 FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE,
6299 xvap.xva_xoptattrs.xoa_archive);
6300 FLAG_CHECK(UF_NODUMP, XAT_NODUMP,
6301 xvap.xva_xoptattrs.xoa_nodump);
6302 FLAG_CHECK(UF_READONLY, XAT_READONLY,
6303 xvap.xva_xoptattrs.xoa_readonly);
6304 FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM,
6305 xvap.xva_xoptattrs.xoa_system);
6306 FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN,
6307 xvap.xva_xoptattrs.xoa_hidden);
6308 FLAG_CHECK(UF_REPARSE, XAT_REPARSE,
6309 xvap.xva_xoptattrs.xoa_reparse);
6310 FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE,
6311 xvap.xva_xoptattrs.xoa_offline);
6312 FLAG_CHECK(UF_SPARSE, XAT_SPARSE,
6313 xvap.xva_xoptattrs.xoa_sparse);
6316 *vap = xvap.xva_vattr;
6317 vap->va_flags = fflags;
6322 zfs_freebsd_setattr(ap)
6323 struct vop_setattr_args /* {
6325 struct vattr *a_vap;
6326 struct ucred *a_cred;
6329 vnode_t *vp = ap->a_vp;
6330 vattr_t *vap = ap->a_vap;
6331 cred_t *cred = ap->a_cred;
6336 vattr_init_mask(vap);
6337 vap->va_mask &= ~AT_NOSET;
6340 xvap.xva_vattr = *vap;
6342 zflags = VTOZ(vp)->z_pflags;
6344 if (vap->va_flags != VNOVAL) {
6345 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs;
6348 if (zfsvfs->z_use_fuids == B_FALSE)
6349 return (EOPNOTSUPP);
6351 fflags = vap->va_flags;
6354 * We need to figure out whether it makes sense to allow
6355 * UF_REPARSE through, since we don't really have other
6356 * facilities to handle reparse points and zfs_setattr()
6357 * doesn't currently allow setting that attribute anyway.
6359 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE|
6360 UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE|
6361 UF_OFFLINE|UF_SPARSE)) != 0)
6362 return (EOPNOTSUPP);
6364 * Unprivileged processes are not permitted to unset system
6365 * flags, or modify flags if any system flags are set.
6366 * Privileged non-jail processes may not modify system flags
6367 * if securelevel > 0 and any existing system flags are set.
6368 * Privileged jail processes behave like privileged non-jail
6369 * processes if the security.jail.chflags_allowed sysctl is
6370 * is non-zero; otherwise, they behave like unprivileged
6373 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 ||
6374 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) {
6376 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6377 error = securelevel_gt(cred, 0);
6383 * Callers may only modify the file flags on objects they
6384 * have VADMIN rights for.
6386 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0)
6389 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6393 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) {
6398 #define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \
6399 if (((fflags & (fflag)) && !(zflags & (zflag))) || \
6400 ((zflags & (zflag)) && !(fflags & (fflag)))) { \
6401 XVA_SET_REQ(&xvap, (xflag)); \
6402 (xfield) = ((fflags & (fflag)) != 0); \
6405 /* Convert chflags into ZFS-type flags. */
6406 /* XXX: what about SF_SETTABLE?. */
6407 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE,
6408 xvap.xva_xoptattrs.xoa_immutable);
6409 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY,
6410 xvap.xva_xoptattrs.xoa_appendonly);
6411 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK,
6412 xvap.xva_xoptattrs.xoa_nounlink);
6413 FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE,
6414 xvap.xva_xoptattrs.xoa_archive);
6415 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP,
6416 xvap.xva_xoptattrs.xoa_nodump);
6417 FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY,
6418 xvap.xva_xoptattrs.xoa_readonly);
6419 FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM,
6420 xvap.xva_xoptattrs.xoa_system);
6421 FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN,
6422 xvap.xva_xoptattrs.xoa_hidden);
6423 FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE,
6424 xvap.xva_xoptattrs.xoa_hidden);
6425 FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE,
6426 xvap.xva_xoptattrs.xoa_offline);
6427 FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE,
6428 xvap.xva_xoptattrs.xoa_sparse);
6431 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL));
6435 zfs_freebsd_rename(ap)
6436 struct vop_rename_args /* {
6437 struct vnode *a_fdvp;
6438 struct vnode *a_fvp;
6439 struct componentname *a_fcnp;
6440 struct vnode *a_tdvp;
6441 struct vnode *a_tvp;
6442 struct componentname *a_tcnp;
6445 vnode_t *fdvp = ap->a_fdvp;
6446 vnode_t *fvp = ap->a_fvp;
6447 vnode_t *tdvp = ap->a_tdvp;
6448 vnode_t *tvp = ap->a_tvp;
6451 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART));
6452 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART));
6455 * Check for cross-device rename.
6457 if ((fdvp->v_mount != tdvp->v_mount) ||
6458 (tvp && (fdvp->v_mount != tvp->v_mount)))
6461 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp,
6462 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0);
6476 zfs_freebsd_symlink(ap)
6477 struct vop_symlink_args /* {
6478 struct vnode *a_dvp;
6479 struct vnode **a_vpp;
6480 struct componentname *a_cnp;
6481 struct vattr *a_vap;
6485 struct componentname *cnp = ap->a_cnp;
6486 vattr_t *vap = ap->a_vap;
6488 ASSERT(cnp->cn_flags & SAVENAME);
6490 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */
6491 vattr_init_mask(vap);
6493 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap,
6494 ap->a_target, cnp->cn_cred, cnp->cn_thread));
6498 zfs_freebsd_readlink(ap)
6499 struct vop_readlink_args /* {
6502 struct ucred *a_cred;
6506 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL));
6510 zfs_freebsd_link(ap)
6511 struct vop_link_args /* {
6512 struct vnode *a_tdvp;
6514 struct componentname *a_cnp;
6517 struct componentname *cnp = ap->a_cnp;
6518 vnode_t *vp = ap->a_vp;
6519 vnode_t *tdvp = ap->a_tdvp;
6521 if (tdvp->v_mount != vp->v_mount)
6524 ASSERT(cnp->cn_flags & SAVENAME);
6526 return (zfs_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0));
6530 zfs_freebsd_inactive(ap)
6531 struct vop_inactive_args /* {
6533 struct thread *a_td;
6536 vnode_t *vp = ap->a_vp;
6538 zfs_inactive(vp, ap->a_td->td_ucred, NULL);
6543 zfs_freebsd_reclaim(ap)
6544 struct vop_reclaim_args /* {
6546 struct thread *a_td;
6549 vnode_t *vp = ap->a_vp;
6550 znode_t *zp = VTOZ(vp);
6551 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
6555 /* Destroy the vm object and flush associated pages. */
6556 vnode_destroy_vobject(vp);
6559 * z_teardown_inactive_lock protects from a race with
6560 * zfs_znode_dmu_fini in zfsvfs_teardown during
6563 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
6564 if (zp->z_sa_hdl == NULL)
6568 rw_exit(&zfsvfs->z_teardown_inactive_lock);
6576 struct vop_fid_args /* {
6582 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL));
6586 zfs_freebsd_pathconf(ap)
6587 struct vop_pathconf_args /* {
6590 register_t *a_retval;
6596 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL);
6598 *ap->a_retval = val;
6599 else if (error == EOPNOTSUPP)
6600 error = vop_stdpathconf(ap);
6605 zfs_freebsd_fifo_pathconf(ap)
6606 struct vop_pathconf_args /* {
6609 register_t *a_retval;
6613 switch (ap->a_name) {
6614 case _PC_ACL_EXTENDED:
6616 case _PC_ACL_PATH_MAX:
6617 case _PC_MAC_PRESENT:
6618 return (zfs_freebsd_pathconf(ap));
6620 return (fifo_specops.vop_pathconf(ap));
6625 * FreeBSD's extended attributes namespace defines file name prefix for ZFS'
6626 * extended attribute name:
6629 * system freebsd:system:
6630 * user (none, can be used to access ZFS fsattr(5) attributes
6631 * created on Solaris)
6634 zfs_create_attrname(int attrnamespace, const char *name, char *attrname,
6637 const char *namespace, *prefix, *suffix;
6639 /* We don't allow '/' character in attribute name. */
6640 if (strchr(name, '/') != NULL)
6642 /* We don't allow attribute names that start with "freebsd:" string. */
6643 if (strncmp(name, "freebsd:", 8) == 0)
6646 bzero(attrname, size);
6648 switch (attrnamespace) {
6649 case EXTATTR_NAMESPACE_USER:
6651 prefix = "freebsd:";
6652 namespace = EXTATTR_NAMESPACE_USER_STRING;
6656 * This is the default namespace by which we can access all
6657 * attributes created on Solaris.
6659 prefix = namespace = suffix = "";
6662 case EXTATTR_NAMESPACE_SYSTEM:
6663 prefix = "freebsd:";
6664 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING;
6667 case EXTATTR_NAMESPACE_EMPTY:
6671 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix,
6673 return (ENAMETOOLONG);
6679 * Vnode operating to retrieve a named extended attribute.
6682 zfs_getextattr(struct vop_getextattr_args *ap)
6685 IN struct vnode *a_vp;
6686 IN int a_attrnamespace;
6687 IN const char *a_name;
6688 INOUT struct uio *a_uio;
6690 IN struct ucred *a_cred;
6691 IN struct thread *a_td;
6695 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6696 struct thread *td = ap->a_td;
6697 struct nameidata nd;
6700 vnode_t *xvp = NULL, *vp;
6703 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6704 ap->a_cred, ap->a_td, VREAD);
6708 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6715 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6723 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6725 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL);
6727 NDFREE(&nd, NDF_ONLY_PNBUF);
6730 if (error == ENOENT)
6735 if (ap->a_size != NULL) {
6736 error = VOP_GETATTR(vp, &va, ap->a_cred);
6738 *ap->a_size = (size_t)va.va_size;
6739 } else if (ap->a_uio != NULL)
6740 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6743 vn_close(vp, flags, ap->a_cred, td);
6750 * Vnode operation to remove a named attribute.
6753 zfs_deleteextattr(struct vop_deleteextattr_args *ap)
6756 IN struct vnode *a_vp;
6757 IN int a_attrnamespace;
6758 IN const char *a_name;
6759 IN struct ucred *a_cred;
6760 IN struct thread *a_td;
6764 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6765 struct thread *td = ap->a_td;
6766 struct nameidata nd;
6769 vnode_t *xvp = NULL, *vp;
6772 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6773 ap->a_cred, ap->a_td, VWRITE);
6777 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6784 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6791 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF,
6792 UIO_SYSSPACE, attrname, xvp, td);
6797 NDFREE(&nd, NDF_ONLY_PNBUF);
6798 if (error == ENOENT)
6803 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
6804 NDFREE(&nd, NDF_ONLY_PNBUF);
6807 if (vp == nd.ni_dvp)
6817 * Vnode operation to set a named attribute.
6820 zfs_setextattr(struct vop_setextattr_args *ap)
6823 IN struct vnode *a_vp;
6824 IN int a_attrnamespace;
6825 IN const char *a_name;
6826 INOUT struct uio *a_uio;
6827 IN struct ucred *a_cred;
6828 IN struct thread *a_td;
6832 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6833 struct thread *td = ap->a_td;
6834 struct nameidata nd;
6837 vnode_t *xvp = NULL, *vp;
6840 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6841 ap->a_cred, ap->a_td, VWRITE);
6845 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6852 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6853 LOOKUP_XATTR | CREATE_XATTR_DIR);
6859 flags = FFLAGS(O_WRONLY | O_CREAT);
6860 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6862 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL);
6864 NDFREE(&nd, NDF_ONLY_PNBUF);
6872 error = VOP_SETATTR(vp, &va, ap->a_cred);
6874 VOP_WRITE(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6877 vn_close(vp, flags, ap->a_cred, td);
6884 * Vnode operation to retrieve extended attributes on a vnode.
6887 zfs_listextattr(struct vop_listextattr_args *ap)
6890 IN struct vnode *a_vp;
6891 IN int a_attrnamespace;
6892 INOUT struct uio *a_uio;
6894 IN struct ucred *a_cred;
6895 IN struct thread *a_td;
6899 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6900 struct thread *td = ap->a_td;
6901 struct nameidata nd;
6902 char attrprefix[16];
6903 u_char dirbuf[sizeof(struct dirent)];
6906 struct uio auio, *uio = ap->a_uio;
6907 size_t *sizep = ap->a_size;
6909 vnode_t *xvp = NULL, *vp;
6910 int done, error, eof, pos;
6912 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6913 ap->a_cred, ap->a_td, VREAD);
6917 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix,
6918 sizeof(attrprefix));
6921 plen = strlen(attrprefix);
6928 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6933 * ENOATTR means that the EA directory does not yet exist,
6934 * i.e. there are no extended attributes there.
6936 if (error == ENOATTR)
6941 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED,
6942 UIO_SYSSPACE, ".", xvp, td);
6945 NDFREE(&nd, NDF_ONLY_PNBUF);
6951 auio.uio_iov = &aiov;
6952 auio.uio_iovcnt = 1;
6953 auio.uio_segflg = UIO_SYSSPACE;
6955 auio.uio_rw = UIO_READ;
6956 auio.uio_offset = 0;
6961 aiov.iov_base = (void *)dirbuf;
6962 aiov.iov_len = sizeof(dirbuf);
6963 auio.uio_resid = sizeof(dirbuf);
6964 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL);
6965 done = sizeof(dirbuf) - auio.uio_resid;
6968 for (pos = 0; pos < done;) {
6969 dp = (struct dirent *)(dirbuf + pos);
6970 pos += dp->d_reclen;
6972 * XXX: Temporarily we also accept DT_UNKNOWN, as this
6973 * is what we get when attribute was created on Solaris.
6975 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN)
6977 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0)
6979 else if (strncmp(dp->d_name, attrprefix, plen) != 0)
6981 nlen = dp->d_namlen - plen;
6984 else if (uio != NULL) {
6986 * Format of extattr name entry is one byte for
6987 * length and the rest for name.
6989 error = uiomove(&nlen, 1, uio->uio_rw, uio);
6991 error = uiomove(dp->d_name + plen, nlen,
6998 } while (!eof && error == 0);
7007 zfs_freebsd_getacl(ap)
7008 struct vop_getacl_args /* {
7017 vsecattr_t vsecattr;
7019 if (ap->a_type != ACL_TYPE_NFS4)
7022 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
7023 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL))
7026 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt);
7027 if (vsecattr.vsa_aclentp != NULL)
7028 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz);
7034 zfs_freebsd_setacl(ap)
7035 struct vop_setacl_args /* {
7044 vsecattr_t vsecattr;
7045 int aclbsize; /* size of acl list in bytes */
7048 if (ap->a_type != ACL_TYPE_NFS4)
7051 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES)
7055 * With NFSv4 ACLs, chmod(2) may need to add additional entries,
7056 * splitting every entry into two and appending "canonical six"
7057 * entries at the end. Don't allow for setting an ACL that would
7058 * cause chmod(2) to run out of ACL entries.
7060 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES)
7063 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR);
7067 vsecattr.vsa_mask = VSA_ACE;
7068 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t);
7069 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP);
7070 aaclp = vsecattr.vsa_aclentp;
7071 vsecattr.vsa_aclentsz = aclbsize;
7073 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp);
7074 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL);
7075 kmem_free(aaclp, aclbsize);
7081 zfs_freebsd_aclcheck(ap)
7082 struct vop_aclcheck_args /* {
7091 return (EOPNOTSUPP);
7094 struct vop_vector zfs_vnodeops;
7095 struct vop_vector zfs_fifoops;
7096 struct vop_vector zfs_shareops;
7098 struct vop_vector zfs_vnodeops = {
7099 .vop_default = &default_vnodeops,
7100 .vop_inactive = zfs_freebsd_inactive,
7101 .vop_reclaim = zfs_freebsd_reclaim,
7102 .vop_access = zfs_freebsd_access,
7103 #ifdef FREEBSD_NAMECACHE
7104 .vop_lookup = vfs_cache_lookup,
7105 .vop_cachedlookup = zfs_freebsd_lookup,
7107 .vop_lookup = zfs_freebsd_lookup,
7109 .vop_getattr = zfs_freebsd_getattr,
7110 .vop_setattr = zfs_freebsd_setattr,
7111 .vop_create = zfs_freebsd_create,
7112 .vop_mknod = zfs_freebsd_create,
7113 .vop_mkdir = zfs_freebsd_mkdir,
7114 .vop_readdir = zfs_freebsd_readdir,
7115 .vop_fsync = zfs_freebsd_fsync,
7116 .vop_open = zfs_freebsd_open,
7117 .vop_close = zfs_freebsd_close,
7118 .vop_rmdir = zfs_freebsd_rmdir,
7119 .vop_ioctl = zfs_freebsd_ioctl,
7120 .vop_link = zfs_freebsd_link,
7121 .vop_symlink = zfs_freebsd_symlink,
7122 .vop_readlink = zfs_freebsd_readlink,
7123 .vop_read = zfs_freebsd_read,
7124 .vop_write = zfs_freebsd_write,
7125 .vop_remove = zfs_freebsd_remove,
7126 .vop_rename = zfs_freebsd_rename,
7127 .vop_pathconf = zfs_freebsd_pathconf,
7128 .vop_bmap = zfs_freebsd_bmap,
7129 .vop_fid = zfs_freebsd_fid,
7130 .vop_getextattr = zfs_getextattr,
7131 .vop_deleteextattr = zfs_deleteextattr,
7132 .vop_setextattr = zfs_setextattr,
7133 .vop_listextattr = zfs_listextattr,
7134 .vop_getacl = zfs_freebsd_getacl,
7135 .vop_setacl = zfs_freebsd_setacl,
7136 .vop_aclcheck = zfs_freebsd_aclcheck,
7137 .vop_getpages = zfs_freebsd_getpages,
7138 .vop_putpages = zfs_freebsd_putpages,
7141 struct vop_vector zfs_fifoops = {
7142 .vop_default = &fifo_specops,
7143 .vop_fsync = zfs_freebsd_fsync,
7144 .vop_access = zfs_freebsd_access,
7145 .vop_getattr = zfs_freebsd_getattr,
7146 .vop_inactive = zfs_freebsd_inactive,
7147 .vop_read = VOP_PANIC,
7148 .vop_reclaim = zfs_freebsd_reclaim,
7149 .vop_setattr = zfs_freebsd_setattr,
7150 .vop_write = VOP_PANIC,
7151 .vop_pathconf = zfs_freebsd_fifo_pathconf,
7152 .vop_fid = zfs_freebsd_fid,
7153 .vop_getacl = zfs_freebsd_getacl,
7154 .vop_setacl = zfs_freebsd_setacl,
7155 .vop_aclcheck = zfs_freebsd_aclcheck,
7159 * special share hidden files vnode operations template
7161 struct vop_vector zfs_shareops = {
7162 .vop_default = &default_vnodeops,
7163 .vop_access = zfs_freebsd_access,
7164 .vop_inactive = zfs_freebsd_inactive,
7165 .vop_reclaim = zfs_freebsd_reclaim,
7166 .vop_fid = zfs_freebsd_fid,
7167 .vop_pathconf = zfs_freebsd_pathconf,