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);
585 objset_t *os = zp->z_zfsvfs->z_os;
593 ASSERT(vp->v_mount != NULL);
597 start = uio->uio_loffset;
598 off = start & PAGEOFFSET;
599 zfs_vmobject_wlock(obj);
600 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
602 uint64_t bytes = MIN(PAGESIZE - off, len);
604 if (pp = page_hold(vp, start)) {
608 zfs_vmobject_wunlock(obj);
609 va = zfs_map_page(pp, &sf);
610 error = uiomove(va + off, bytes, UIO_READ, uio);
612 zfs_vmobject_wlock(obj);
615 zfs_vmobject_wunlock(obj);
616 error = dmu_read_uio(os, zp->z_id, uio, bytes);
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;
663 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
665 return (SET_ERROR(EACCES));
669 * Validate file offset
671 if (uio->uio_loffset < (offset_t)0) {
673 return (SET_ERROR(EINVAL));
677 * Fasttrack empty reads
679 if (uio->uio_resid == 0) {
685 * Check for mandatory locks
687 if (MANDMODE(zp->z_mode)) {
688 if (error = chklock(vp, FREAD,
689 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
696 * If we're in FRSYNC mode, sync out this znode before reading it.
699 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
700 zil_commit(zfsvfs->z_log, zp->z_id);
703 * Lock the range against changes.
705 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
708 * If we are reading past end-of-file we can skip
709 * to the end; but we might still need to set atime.
711 if (uio->uio_loffset >= zp->z_size) {
716 ASSERT(uio->uio_loffset < zp->z_size);
717 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
720 if ((uio->uio_extflg == UIO_XUIO) &&
721 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
723 int blksz = zp->z_blksz;
724 uint64_t offset = uio->uio_loffset;
726 xuio = (xuio_t *)uio;
728 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
731 ASSERT(offset + n <= blksz);
734 (void) dmu_xuio_init(xuio, nblk);
736 if (vn_has_cached_data(vp)) {
738 * For simplicity, we always allocate a full buffer
739 * even if we only expect to read a portion of a block.
741 while (--nblk >= 0) {
742 (void) dmu_xuio_add(xuio,
743 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
751 nbytes = MIN(n, zfs_read_chunk_size -
752 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
755 if (uio->uio_segflg == UIO_NOCOPY)
756 error = mappedread_sf(vp, nbytes, uio);
758 #endif /* __FreeBSD__ */
759 if (vn_has_cached_data(vp))
760 error = mappedread(vp, nbytes, uio);
762 error = dmu_read_uio(os, zp->z_id, uio, nbytes);
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) {
1016 ASSERT(!ISP2(zp->z_blksz));
1017 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
1019 new_blksz = MIN(end_size, max_blksz);
1021 zfs_grow_blocksize(zp, new_blksz, tx);
1022 zfs_range_reduce(rl, woff, n);
1026 * XXX - should we really limit each write to z_max_blksz?
1027 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
1029 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
1031 if (woff + nbytes > zp->z_size)
1032 vnode_pager_setsize(vp, woff + nbytes);
1035 tx_bytes = uio->uio_resid;
1036 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
1038 tx_bytes -= uio->uio_resid;
1041 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
1043 * If this is not a full block write, but we are
1044 * extending the file past EOF and this data starts
1045 * block-aligned, use assign_arcbuf(). Otherwise,
1046 * write via dmu_write().
1048 if (tx_bytes < max_blksz && (!write_eof ||
1049 aiov->iov_base != abuf->b_data)) {
1051 dmu_write(zfsvfs->z_os, zp->z_id, woff,
1052 aiov->iov_len, aiov->iov_base, tx);
1053 dmu_return_arcbuf(abuf);
1054 xuio_stat_wbuf_copied();
1056 ASSERT(xuio || tx_bytes == max_blksz);
1057 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
1060 ASSERT(tx_bytes <= uio->uio_resid);
1061 uioskip(uio, tx_bytes);
1063 if (tx_bytes && vn_has_cached_data(vp)) {
1064 update_pages(vp, woff, tx_bytes, zfsvfs->z_os,
1065 zp->z_id, uio->uio_segflg, tx);
1069 * If we made no progress, we're done. If we made even
1070 * partial progress, update the znode and ZIL accordingly.
1072 if (tx_bytes == 0) {
1073 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
1074 (void *)&zp->z_size, sizeof (uint64_t), tx);
1081 * Clear Set-UID/Set-GID bits on successful write if not
1082 * privileged and at least one of the excute bits is set.
1084 * It would be nice to to this after all writes have
1085 * been done, but that would still expose the ISUID/ISGID
1086 * to another app after the partial write is committed.
1088 * Note: we don't call zfs_fuid_map_id() here because
1089 * user 0 is not an ephemeral uid.
1091 mutex_enter(&zp->z_acl_lock);
1092 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
1093 (S_IXUSR >> 6))) != 0 &&
1094 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
1095 secpolicy_vnode_setid_retain(vp, cr,
1096 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
1098 zp->z_mode &= ~(S_ISUID | S_ISGID);
1099 newmode = zp->z_mode;
1100 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
1101 (void *)&newmode, sizeof (uint64_t), tx);
1103 mutex_exit(&zp->z_acl_lock);
1105 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
1109 * Update the file size (zp_size) if it has changed;
1110 * account for possible concurrent updates.
1112 while ((end_size = zp->z_size) < uio->uio_loffset) {
1113 (void) atomic_cas_64(&zp->z_size, end_size,
1118 * If we are replaying and eof is non zero then force
1119 * the file size to the specified eof. Note, there's no
1120 * concurrency during replay.
1122 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
1123 zp->z_size = zfsvfs->z_replay_eof;
1125 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1127 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
1132 ASSERT(tx_bytes == nbytes);
1137 uio_prefaultpages(MIN(n, max_blksz), uio);
1141 zfs_range_unlock(rl);
1144 * If we're in replay mode, or we made no progress, return error.
1145 * Otherwise, it's at least a partial write, so it's successful.
1147 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1152 if (ioflag & (FSYNC | FDSYNC) ||
1153 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1154 zil_commit(zilog, zp->z_id);
1161 zfs_get_done(zgd_t *zgd, int error)
1163 znode_t *zp = zgd->zgd_private;
1164 objset_t *os = zp->z_zfsvfs->z_os;
1167 dmu_buf_rele(zgd->zgd_db, zgd);
1169 zfs_range_unlock(zgd->zgd_rl);
1172 * Release the vnode asynchronously as we currently have the
1173 * txg stopped from syncing.
1175 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1177 if (error == 0 && zgd->zgd_bp)
1178 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1180 kmem_free(zgd, sizeof (zgd_t));
1184 static int zil_fault_io = 0;
1188 * Get data to generate a TX_WRITE intent log record.
1191 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1193 zfsvfs_t *zfsvfs = arg;
1194 objset_t *os = zfsvfs->z_os;
1196 uint64_t object = lr->lr_foid;
1197 uint64_t offset = lr->lr_offset;
1198 uint64_t size = lr->lr_length;
1199 blkptr_t *bp = &lr->lr_blkptr;
1204 ASSERT(zio != NULL);
1208 * Nothing to do if the file has been removed
1210 if (zfs_zget(zfsvfs, object, &zp) != 0)
1211 return (SET_ERROR(ENOENT));
1212 if (zp->z_unlinked) {
1214 * Release the vnode asynchronously as we currently have the
1215 * txg stopped from syncing.
1217 VN_RELE_ASYNC(ZTOV(zp),
1218 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1219 return (SET_ERROR(ENOENT));
1222 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1223 zgd->zgd_zilog = zfsvfs->z_log;
1224 zgd->zgd_private = zp;
1227 * Write records come in two flavors: immediate and indirect.
1228 * For small writes it's cheaper to store the data with the
1229 * log record (immediate); for large writes it's cheaper to
1230 * sync the data and get a pointer to it (indirect) so that
1231 * we don't have to write the data twice.
1233 if (buf != NULL) { /* immediate write */
1234 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1235 /* test for truncation needs to be done while range locked */
1236 if (offset >= zp->z_size) {
1237 error = SET_ERROR(ENOENT);
1239 error = dmu_read(os, object, offset, size, buf,
1240 DMU_READ_NO_PREFETCH);
1242 ASSERT(error == 0 || error == ENOENT);
1243 } else { /* indirect write */
1245 * Have to lock the whole block to ensure when it's
1246 * written out and it's checksum is being calculated
1247 * that no one can change the data. We need to re-check
1248 * blocksize after we get the lock in case it's changed!
1253 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1255 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1257 if (zp->z_blksz == size)
1260 zfs_range_unlock(zgd->zgd_rl);
1262 /* test for truncation needs to be done while range locked */
1263 if (lr->lr_offset >= zp->z_size)
1264 error = SET_ERROR(ENOENT);
1267 error = SET_ERROR(EIO);
1272 error = dmu_buf_hold(os, object, offset, zgd, &db,
1273 DMU_READ_NO_PREFETCH);
1276 blkptr_t *obp = dmu_buf_get_blkptr(db);
1278 ASSERT(BP_IS_HOLE(bp));
1285 ASSERT(db->db_offset == offset);
1286 ASSERT(db->db_size == size);
1288 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1290 ASSERT(error || lr->lr_length <= zp->z_blksz);
1293 * On success, we need to wait for the write I/O
1294 * initiated by dmu_sync() to complete before we can
1295 * release this dbuf. We will finish everything up
1296 * in the zfs_get_done() callback.
1301 if (error == EALREADY) {
1302 lr->lr_common.lrc_txtype = TX_WRITE2;
1308 zfs_get_done(zgd, error);
1315 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1316 caller_context_t *ct)
1318 znode_t *zp = VTOZ(vp);
1319 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1325 if (flag & V_ACE_MASK)
1326 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1328 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1335 * If vnode is for a device return a specfs vnode instead.
1338 specvp_check(vnode_t **vpp, cred_t *cr)
1342 if (IS_DEVVP(*vpp)) {
1345 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1348 error = SET_ERROR(ENOSYS);
1356 * Lookup an entry in a directory, or an extended attribute directory.
1357 * If it exists, return a held vnode reference for it.
1359 * IN: dvp - vnode of directory to search.
1360 * nm - name of entry to lookup.
1361 * pnp - full pathname to lookup [UNUSED].
1362 * flags - LOOKUP_XATTR set if looking for an attribute.
1363 * rdir - root directory vnode [UNUSED].
1364 * cr - credentials of caller.
1365 * ct - caller context
1366 * direntflags - directory lookup flags
1367 * realpnp - returned pathname.
1369 * OUT: vpp - vnode of located entry, NULL if not found.
1371 * RETURN: 0 on success, error code on failure.
1378 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp,
1379 int nameiop, cred_t *cr, kthread_t *td, int flags)
1381 znode_t *zdp = VTOZ(dvp);
1382 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1384 int *direntflags = NULL;
1385 void *realpnp = NULL;
1388 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1390 if (dvp->v_type != VDIR) {
1391 return (SET_ERROR(ENOTDIR));
1392 } else if (zdp->z_sa_hdl == NULL) {
1393 return (SET_ERROR(EIO));
1396 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1397 error = zfs_fastaccesschk_execute(zdp, cr);
1405 vnode_t *tvp = dnlc_lookup(dvp, nm);
1408 error = zfs_fastaccesschk_execute(zdp, cr);
1413 if (tvp == DNLC_NO_VNODE) {
1415 return (SET_ERROR(ENOENT));
1418 return (specvp_check(vpp, cr));
1424 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1431 if (flags & LOOKUP_XATTR) {
1434 * If the xattr property is off, refuse the lookup request.
1436 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1438 return (SET_ERROR(EINVAL));
1443 * We don't allow recursive attributes..
1444 * Maybe someday we will.
1446 if (zdp->z_pflags & ZFS_XATTR) {
1448 return (SET_ERROR(EINVAL));
1451 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1457 * Do we have permission to get into attribute directory?
1460 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1470 if (dvp->v_type != VDIR) {
1472 return (SET_ERROR(ENOTDIR));
1476 * Check accessibility of directory.
1479 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1484 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1485 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1487 return (SET_ERROR(EILSEQ));
1490 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1492 error = specvp_check(vpp, cr);
1494 /* Translate errors and add SAVENAME when needed. */
1495 if (cnp->cn_flags & ISLASTCN) {
1499 if (error == ENOENT) {
1500 error = EJUSTRETURN;
1501 cnp->cn_flags |= SAVENAME;
1507 cnp->cn_flags |= SAVENAME;
1511 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) {
1514 if (cnp->cn_flags & ISDOTDOT) {
1515 ltype = VOP_ISLOCKED(dvp);
1519 error = vn_lock(*vpp, cnp->cn_lkflags);
1520 if (cnp->cn_flags & ISDOTDOT)
1521 vn_lock(dvp, ltype | LK_RETRY);
1531 #ifdef FREEBSD_NAMECACHE
1533 * Insert name into cache (as non-existent) if appropriate.
1535 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
1536 cache_enter(dvp, *vpp, cnp);
1538 * Insert name into cache if appropriate.
1540 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) {
1541 if (!(cnp->cn_flags & ISLASTCN) ||
1542 (nameiop != DELETE && nameiop != RENAME)) {
1543 cache_enter(dvp, *vpp, cnp);
1552 * Attempt to create a new entry in a directory. If the entry
1553 * already exists, truncate the file if permissible, else return
1554 * an error. Return the vp of the created or trunc'd file.
1556 * IN: dvp - vnode of directory to put new file entry in.
1557 * name - name of new file entry.
1558 * vap - attributes of new file.
1559 * excl - flag indicating exclusive or non-exclusive mode.
1560 * mode - mode to open file with.
1561 * cr - credentials of caller.
1562 * flag - large file flag [UNUSED].
1563 * ct - caller context
1564 * vsecp - ACL to be set
1566 * OUT: vpp - vnode of created or trunc'd entry.
1568 * RETURN: 0 on success, error code on failure.
1571 * dvp - ctime|mtime updated if new entry created
1572 * vp - ctime|mtime always, atime if new
1577 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode,
1578 vnode_t **vpp, cred_t *cr, kthread_t *td)
1580 znode_t *zp, *dzp = VTOZ(dvp);
1581 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1589 gid_t gid = crgetgid(cr);
1590 zfs_acl_ids_t acl_ids;
1591 boolean_t fuid_dirtied;
1592 boolean_t have_acl = B_FALSE;
1593 boolean_t waited = B_FALSE;
1598 * If we have an ephemeral id, ACL, or XVATTR then
1599 * make sure file system is at proper version
1602 ksid = crgetsid(cr, KSID_OWNER);
1604 uid = ksid_getid(ksid);
1608 if (zfsvfs->z_use_fuids == B_FALSE &&
1609 (vsecp || (vap->va_mask & AT_XVATTR) ||
1610 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1611 return (SET_ERROR(EINVAL));
1616 zilog = zfsvfs->z_log;
1618 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1619 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1621 return (SET_ERROR(EILSEQ));
1624 if (vap->va_mask & AT_XVATTR) {
1625 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
1626 crgetuid(cr), cr, vap->va_type)) != 0) {
1632 getnewvnode_reserve(1);
1637 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr))
1638 vap->va_mode &= ~S_ISVTX;
1640 if (*name == '\0') {
1642 * Null component name refers to the directory itself.
1649 /* possible VN_HOLD(zp) */
1652 if (flag & FIGNORECASE)
1655 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1659 zfs_acl_ids_free(&acl_ids);
1660 if (strcmp(name, "..") == 0)
1661 error = SET_ERROR(EISDIR);
1662 getnewvnode_drop_reserve();
1672 * Create a new file object and update the directory
1675 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1677 zfs_acl_ids_free(&acl_ids);
1682 * We only support the creation of regular files in
1683 * extended attribute directories.
1686 if ((dzp->z_pflags & ZFS_XATTR) &&
1687 (vap->va_type != VREG)) {
1689 zfs_acl_ids_free(&acl_ids);
1690 error = SET_ERROR(EINVAL);
1694 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1695 cr, vsecp, &acl_ids)) != 0)
1699 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1700 zfs_acl_ids_free(&acl_ids);
1701 error = SET_ERROR(EDQUOT);
1705 tx = dmu_tx_create(os);
1707 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1708 ZFS_SA_BASE_ATTR_SIZE);
1710 fuid_dirtied = zfsvfs->z_fuid_dirty;
1712 zfs_fuid_txhold(zfsvfs, tx);
1713 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1714 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1715 if (!zfsvfs->z_use_sa &&
1716 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1717 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1718 0, acl_ids.z_aclp->z_acl_bytes);
1720 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1722 zfs_dirent_unlock(dl);
1723 if (error == ERESTART) {
1729 zfs_acl_ids_free(&acl_ids);
1731 getnewvnode_drop_reserve();
1735 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1738 zfs_fuid_sync(zfsvfs, tx);
1740 (void) zfs_link_create(dl, zp, tx, ZNEW);
1741 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1742 if (flag & FIGNORECASE)
1744 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1745 vsecp, acl_ids.z_fuidp, vap);
1746 zfs_acl_ids_free(&acl_ids);
1749 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1752 zfs_acl_ids_free(&acl_ids);
1756 * A directory entry already exists for this name.
1759 * Can't truncate an existing file if in exclusive mode.
1762 error = SET_ERROR(EEXIST);
1766 * Can't open a directory for writing.
1768 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1769 error = SET_ERROR(EISDIR);
1773 * Verify requested access to file.
1775 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1779 mutex_enter(&dzp->z_lock);
1781 mutex_exit(&dzp->z_lock);
1784 * Truncate regular files if requested.
1786 if ((ZTOV(zp)->v_type == VREG) &&
1787 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1788 /* we can't hold any locks when calling zfs_freesp() */
1789 zfs_dirent_unlock(dl);
1791 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1793 vnevent_create(ZTOV(zp), ct);
1798 getnewvnode_drop_reserve();
1800 zfs_dirent_unlock(dl);
1807 error = specvp_check(vpp, cr);
1810 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1811 zil_commit(zilog, 0);
1818 * Remove an entry from a directory.
1820 * IN: dvp - vnode of directory to remove entry from.
1821 * name - name of entry to remove.
1822 * cr - credentials of caller.
1823 * ct - caller context
1824 * flags - case flags
1826 * RETURN: 0 on success, error code on failure.
1830 * vp - ctime (if nlink > 0)
1833 uint64_t null_xattr = 0;
1837 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1840 znode_t *zp, *dzp = VTOZ(dvp);
1843 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1845 uint64_t acl_obj, xattr_obj;
1846 uint64_t xattr_obj_unlinked = 0;
1850 boolean_t may_delete_now, delete_now = FALSE;
1851 boolean_t unlinked, toobig = FALSE;
1853 pathname_t *realnmp = NULL;
1857 boolean_t waited = B_FALSE;
1861 zilog = zfsvfs->z_log;
1863 if (flags & FIGNORECASE) {
1873 * Attempt to lock directory; fail if entry doesn't exist.
1875 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1885 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1890 * Need to use rmdir for removing directories.
1892 if (vp->v_type == VDIR) {
1893 error = SET_ERROR(EPERM);
1897 vnevent_remove(vp, dvp, name, ct);
1900 dnlc_remove(dvp, realnmp->pn_buf);
1902 dnlc_remove(dvp, name);
1905 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1909 * We may delete the znode now, or we may put it in the unlinked set;
1910 * it depends on whether we're the last link, and on whether there are
1911 * other holds on the vnode. So we dmu_tx_hold() the right things to
1912 * allow for either case.
1915 tx = dmu_tx_create(zfsvfs->z_os);
1916 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1917 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1918 zfs_sa_upgrade_txholds(tx, zp);
1919 zfs_sa_upgrade_txholds(tx, dzp);
1920 if (may_delete_now) {
1922 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1923 /* if the file is too big, only hold_free a token amount */
1924 dmu_tx_hold_free(tx, zp->z_id, 0,
1925 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1928 /* are there any extended attributes? */
1929 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1930 &xattr_obj, sizeof (xattr_obj));
1931 if (error == 0 && xattr_obj) {
1932 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1934 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1935 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1938 mutex_enter(&zp->z_lock);
1939 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1940 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1941 mutex_exit(&zp->z_lock);
1943 /* charge as an update -- would be nice not to charge at all */
1944 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1947 * Mark this transaction as typically resulting in a net free of
1948 * space, unless object removal will be delayed indefinitely
1949 * (due to active holds on the vnode due to the file being open).
1952 dmu_tx_mark_netfree(tx);
1954 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1956 zfs_dirent_unlock(dl);
1960 if (error == ERESTART) {
1974 * Remove the directory entry.
1976 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1985 * Hold z_lock so that we can make sure that the ACL obj
1986 * hasn't changed. Could have been deleted due to
1989 mutex_enter(&zp->z_lock);
1991 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1992 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1993 delete_now = may_delete_now && !toobig &&
1994 vp->v_count == 1 && !vn_has_cached_data(vp) &&
1995 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
2002 panic("zfs_remove: delete_now branch taken");
2004 if (xattr_obj_unlinked) {
2005 ASSERT3U(xzp->z_links, ==, 2);
2006 mutex_enter(&xzp->z_lock);
2007 xzp->z_unlinked = 1;
2009 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
2010 &xzp->z_links, sizeof (xzp->z_links), tx);
2011 ASSERT3U(error, ==, 0);
2012 mutex_exit(&xzp->z_lock);
2013 zfs_unlinked_add(xzp, tx);
2016 error = sa_remove(zp->z_sa_hdl,
2017 SA_ZPL_XATTR(zfsvfs), tx);
2019 error = sa_update(zp->z_sa_hdl,
2020 SA_ZPL_XATTR(zfsvfs), &null_xattr,
2021 sizeof (uint64_t), tx);
2026 ASSERT0(vp->v_count);
2028 mutex_exit(&zp->z_lock);
2029 zfs_znode_delete(zp, tx);
2030 } else if (unlinked) {
2031 mutex_exit(&zp->z_lock);
2032 zfs_unlinked_add(zp, tx);
2034 vp->v_vflag |= VV_NOSYNC;
2039 if (flags & FIGNORECASE)
2041 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
2048 zfs_dirent_unlock(dl);
2055 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2056 zil_commit(zilog, 0);
2063 * Create a new directory and insert it into dvp using the name
2064 * provided. Return a pointer to the inserted directory.
2066 * IN: dvp - vnode of directory to add subdir to.
2067 * dirname - name of new directory.
2068 * vap - attributes of new directory.
2069 * cr - credentials of caller.
2070 * ct - caller context
2071 * flags - case flags
2072 * vsecp - ACL to be set
2074 * OUT: vpp - vnode of created directory.
2076 * RETURN: 0 on success, error code on failure.
2079 * dvp - ctime|mtime updated
2080 * vp - ctime|mtime|atime updated
2084 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
2085 caller_context_t *ct, int flags, vsecattr_t *vsecp)
2087 znode_t *zp, *dzp = VTOZ(dvp);
2088 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2097 gid_t gid = crgetgid(cr);
2098 zfs_acl_ids_t acl_ids;
2099 boolean_t fuid_dirtied;
2100 boolean_t waited = B_FALSE;
2102 ASSERT(vap->va_type == VDIR);
2105 * If we have an ephemeral id, ACL, or XVATTR then
2106 * make sure file system is at proper version
2109 ksid = crgetsid(cr, KSID_OWNER);
2111 uid = ksid_getid(ksid);
2114 if (zfsvfs->z_use_fuids == B_FALSE &&
2115 (vsecp || (vap->va_mask & AT_XVATTR) ||
2116 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
2117 return (SET_ERROR(EINVAL));
2121 zilog = zfsvfs->z_log;
2123 if (dzp->z_pflags & ZFS_XATTR) {
2125 return (SET_ERROR(EINVAL));
2128 if (zfsvfs->z_utf8 && u8_validate(dirname,
2129 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2131 return (SET_ERROR(EILSEQ));
2133 if (flags & FIGNORECASE)
2136 if (vap->va_mask & AT_XVATTR) {
2137 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
2138 crgetuid(cr), cr, vap->va_type)) != 0) {
2144 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
2145 vsecp, &acl_ids)) != 0) {
2150 getnewvnode_reserve(1);
2153 * First make sure the new directory doesn't exist.
2155 * Existence is checked first to make sure we don't return
2156 * EACCES instead of EEXIST which can cause some applications
2162 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
2164 zfs_acl_ids_free(&acl_ids);
2165 getnewvnode_drop_reserve();
2170 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
2171 zfs_acl_ids_free(&acl_ids);
2172 zfs_dirent_unlock(dl);
2173 getnewvnode_drop_reserve();
2178 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
2179 zfs_acl_ids_free(&acl_ids);
2180 zfs_dirent_unlock(dl);
2181 getnewvnode_drop_reserve();
2183 return (SET_ERROR(EDQUOT));
2187 * Add a new entry to the directory.
2189 tx = dmu_tx_create(zfsvfs->z_os);
2190 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2191 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2192 fuid_dirtied = zfsvfs->z_fuid_dirty;
2194 zfs_fuid_txhold(zfsvfs, tx);
2195 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2196 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2197 acl_ids.z_aclp->z_acl_bytes);
2200 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2201 ZFS_SA_BASE_ATTR_SIZE);
2203 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2205 zfs_dirent_unlock(dl);
2206 if (error == ERESTART) {
2212 zfs_acl_ids_free(&acl_ids);
2214 getnewvnode_drop_reserve();
2222 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2225 zfs_fuid_sync(zfsvfs, tx);
2228 * Now put new name in parent dir.
2230 (void) zfs_link_create(dl, zp, tx, ZNEW);
2234 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2235 if (flags & FIGNORECASE)
2237 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2238 acl_ids.z_fuidp, vap);
2240 zfs_acl_ids_free(&acl_ids);
2244 getnewvnode_drop_reserve();
2246 zfs_dirent_unlock(dl);
2248 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2249 zil_commit(zilog, 0);
2256 * Remove a directory subdir entry. If the current working
2257 * directory is the same as the subdir to be removed, the
2260 * IN: dvp - vnode of directory to remove from.
2261 * name - name of directory to be removed.
2262 * cwd - vnode of current working directory.
2263 * cr - credentials of caller.
2264 * ct - caller context
2265 * flags - case flags
2267 * RETURN: 0 on success, error code on failure.
2270 * dvp - ctime|mtime updated
2274 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2275 caller_context_t *ct, int flags)
2277 znode_t *dzp = VTOZ(dvp);
2280 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2286 boolean_t waited = B_FALSE;
2290 zilog = zfsvfs->z_log;
2292 if (flags & FIGNORECASE)
2298 * Attempt to lock directory; fail if entry doesn't exist.
2300 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2308 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2312 if (vp->v_type != VDIR) {
2313 error = SET_ERROR(ENOTDIR);
2318 error = SET_ERROR(EINVAL);
2322 vnevent_rmdir(vp, dvp, name, ct);
2325 * Grab a lock on the directory to make sure that noone is
2326 * trying to add (or lookup) entries while we are removing it.
2328 rw_enter(&zp->z_name_lock, RW_WRITER);
2331 * Grab a lock on the parent pointer to make sure we play well
2332 * with the treewalk and directory rename code.
2334 rw_enter(&zp->z_parent_lock, RW_WRITER);
2336 tx = dmu_tx_create(zfsvfs->z_os);
2337 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2338 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2339 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2340 zfs_sa_upgrade_txholds(tx, zp);
2341 zfs_sa_upgrade_txholds(tx, dzp);
2342 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2344 rw_exit(&zp->z_parent_lock);
2345 rw_exit(&zp->z_name_lock);
2346 zfs_dirent_unlock(dl);
2348 if (error == ERESTART) {
2359 #ifdef FREEBSD_NAMECACHE
2363 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2366 uint64_t txtype = TX_RMDIR;
2367 if (flags & FIGNORECASE)
2369 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2374 rw_exit(&zp->z_parent_lock);
2375 rw_exit(&zp->z_name_lock);
2376 #ifdef FREEBSD_NAMECACHE
2380 zfs_dirent_unlock(dl);
2384 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2385 zil_commit(zilog, 0);
2392 * Read as many directory entries as will fit into the provided
2393 * buffer from the given directory cursor position (specified in
2394 * the uio structure).
2396 * IN: vp - vnode of directory to read.
2397 * uio - structure supplying read location, range info,
2398 * and return buffer.
2399 * cr - credentials of caller.
2400 * ct - caller context
2401 * flags - case flags
2403 * OUT: uio - updated offset and range, buffer filled.
2404 * eofp - set to true if end-of-file detected.
2406 * RETURN: 0 on success, error code on failure.
2409 * vp - atime updated
2411 * Note that the low 4 bits of the cookie returned by zap is always zero.
2412 * This allows us to use the low range for "special" directory entries:
2413 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2414 * we use the offset 2 for the '.zfs' directory.
2418 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies)
2420 znode_t *zp = VTOZ(vp);
2424 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2429 zap_attribute_t zap;
2430 uint_t bytes_wanted;
2431 uint64_t offset; /* must be unsigned; checks for < 1 */
2437 boolean_t check_sysattrs;
2440 u_long *cooks = NULL;
2446 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2447 &parent, sizeof (parent))) != 0) {
2453 * If we are not given an eof variable,
2460 * Check for valid iov_len.
2462 if (uio->uio_iov->iov_len <= 0) {
2464 return (SET_ERROR(EINVAL));
2468 * Quit if directory has been removed (posix)
2470 if ((*eofp = zp->z_unlinked) != 0) {
2477 offset = uio->uio_loffset;
2478 prefetch = zp->z_zn_prefetch;
2481 * Initialize the iterator cursor.
2485 * Start iteration from the beginning of the directory.
2487 zap_cursor_init(&zc, os, zp->z_id);
2490 * The offset is a serialized cursor.
2492 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2496 * Get space to change directory entries into fs independent format.
2498 iovp = uio->uio_iov;
2499 bytes_wanted = iovp->iov_len;
2500 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2501 bufsize = bytes_wanted;
2502 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2503 odp = (struct dirent64 *)outbuf;
2505 bufsize = bytes_wanted;
2507 odp = (struct dirent64 *)iovp->iov_base;
2509 eodp = (struct edirent *)odp;
2511 if (ncookies != NULL) {
2513 * Minimum entry size is dirent size and 1 byte for a file name.
2515 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1);
2516 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK);
2521 * If this VFS supports the system attribute view interface; and
2522 * we're looking at an extended attribute directory; and we care
2523 * about normalization conflicts on this vfs; then we must check
2524 * for normalization conflicts with the sysattr name space.
2527 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2528 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2529 (flags & V_RDDIR_ENTFLAGS);
2535 * Transform to file-system independent format
2538 while (outcount < bytes_wanted) {
2541 off64_t *next = NULL;
2544 * Special case `.', `..', and `.zfs'.
2547 (void) strcpy(zap.za_name, ".");
2548 zap.za_normalization_conflict = 0;
2551 } else if (offset == 1) {
2552 (void) strcpy(zap.za_name, "..");
2553 zap.za_normalization_conflict = 0;
2556 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2557 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2558 zap.za_normalization_conflict = 0;
2559 objnum = ZFSCTL_INO_ROOT;
2565 if (error = zap_cursor_retrieve(&zc, &zap)) {
2566 if ((*eofp = (error == ENOENT)) != 0)
2572 if (zap.za_integer_length != 8 ||
2573 zap.za_num_integers != 1) {
2574 cmn_err(CE_WARN, "zap_readdir: bad directory "
2575 "entry, obj = %lld, offset = %lld\n",
2576 (u_longlong_t)zp->z_id,
2577 (u_longlong_t)offset);
2578 error = SET_ERROR(ENXIO);
2582 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2584 * MacOS X can extract the object type here such as:
2585 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2587 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2589 if (check_sysattrs && !zap.za_normalization_conflict) {
2591 zap.za_normalization_conflict =
2592 xattr_sysattr_casechk(zap.za_name);
2594 panic("%s:%u: TODO", __func__, __LINE__);
2599 if (flags & V_RDDIR_ACCFILTER) {
2601 * If we have no access at all, don't include
2602 * this entry in the returned information
2605 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2607 if (!zfs_has_access(ezp, cr)) {
2614 if (flags & V_RDDIR_ENTFLAGS)
2615 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2617 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2620 * Will this entry fit in the buffer?
2622 if (outcount + reclen > bufsize) {
2624 * Did we manage to fit anything in the buffer?
2627 error = SET_ERROR(EINVAL);
2632 if (flags & V_RDDIR_ENTFLAGS) {
2634 * Add extended flag entry:
2636 eodp->ed_ino = objnum;
2637 eodp->ed_reclen = reclen;
2638 /* NOTE: ed_off is the offset for the *next* entry */
2639 next = &(eodp->ed_off);
2640 eodp->ed_eflags = zap.za_normalization_conflict ?
2641 ED_CASE_CONFLICT : 0;
2642 (void) strncpy(eodp->ed_name, zap.za_name,
2643 EDIRENT_NAMELEN(reclen));
2644 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2649 odp->d_ino = objnum;
2650 odp->d_reclen = reclen;
2651 odp->d_namlen = strlen(zap.za_name);
2652 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1);
2654 odp = (dirent64_t *)((intptr_t)odp + reclen);
2658 ASSERT(outcount <= bufsize);
2660 /* Prefetch znode */
2662 dmu_prefetch(os, objnum, 0, 0);
2666 * Move to the next entry, fill in the previous offset.
2668 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2669 zap_cursor_advance(&zc);
2670 offset = zap_cursor_serialize(&zc);
2675 if (cooks != NULL) {
2678 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks));
2681 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2683 /* Subtract unused cookies */
2684 if (ncookies != NULL)
2685 *ncookies -= ncooks;
2687 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2688 iovp->iov_base += outcount;
2689 iovp->iov_len -= outcount;
2690 uio->uio_resid -= outcount;
2691 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2693 * Reset the pointer.
2695 offset = uio->uio_loffset;
2699 zap_cursor_fini(&zc);
2700 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2701 kmem_free(outbuf, bufsize);
2703 if (error == ENOENT)
2706 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2708 uio->uio_loffset = offset;
2710 if (error != 0 && cookies != NULL) {
2711 free(*cookies, M_TEMP);
2718 ulong_t zfs_fsync_sync_cnt = 4;
2721 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2723 znode_t *zp = VTOZ(vp);
2724 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2726 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2728 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2731 zil_commit(zfsvfs->z_log, zp->z_id);
2739 * Get the requested file attributes and place them in the provided
2742 * IN: vp - vnode of file.
2743 * vap - va_mask identifies requested attributes.
2744 * If AT_XVATTR set, then optional attrs are requested
2745 * flags - ATTR_NOACLCHECK (CIFS server context)
2746 * cr - credentials of caller.
2747 * ct - caller context
2749 * OUT: vap - attribute values.
2751 * RETURN: 0 (always succeeds).
2755 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2756 caller_context_t *ct)
2758 znode_t *zp = VTOZ(vp);
2759 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2762 u_longlong_t nblocks;
2764 uint64_t mtime[2], ctime[2], crtime[2], rdev;
2765 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2766 xoptattr_t *xoap = NULL;
2767 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2768 sa_bulk_attr_t bulk[4];
2774 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2776 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2777 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2778 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
2779 if (vp->v_type == VBLK || vp->v_type == VCHR)
2780 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
2783 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2789 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2790 * Also, if we are the owner don't bother, since owner should
2791 * always be allowed to read basic attributes of file.
2793 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2794 (vap->va_uid != crgetuid(cr))) {
2795 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2803 * Return all attributes. It's cheaper to provide the answer
2804 * than to determine whether we were asked the question.
2807 mutex_enter(&zp->z_lock);
2808 vap->va_type = IFTOVT(zp->z_mode);
2809 vap->va_mode = zp->z_mode & ~S_IFMT;
2811 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2813 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
2815 vap->va_nodeid = zp->z_id;
2816 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2817 links = zp->z_links + 1;
2819 links = zp->z_links;
2820 vap->va_nlink = MIN(links, LINK_MAX); /* nlink_t limit! */
2821 vap->va_size = zp->z_size;
2823 vap->va_rdev = vp->v_rdev;
2825 if (vp->v_type == VBLK || vp->v_type == VCHR)
2826 vap->va_rdev = zfs_cmpldev(rdev);
2828 vap->va_seq = zp->z_seq;
2829 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */
2830 vap->va_filerev = zp->z_seq;
2833 * Add in any requested optional attributes and the create time.
2834 * Also set the corresponding bits in the returned attribute bitmap.
2836 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2837 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2839 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2840 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2843 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2844 xoap->xoa_readonly =
2845 ((zp->z_pflags & ZFS_READONLY) != 0);
2846 XVA_SET_RTN(xvap, XAT_READONLY);
2849 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2851 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2852 XVA_SET_RTN(xvap, XAT_SYSTEM);
2855 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2857 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2858 XVA_SET_RTN(xvap, XAT_HIDDEN);
2861 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2862 xoap->xoa_nounlink =
2863 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2864 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2867 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2868 xoap->xoa_immutable =
2869 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2870 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2873 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2874 xoap->xoa_appendonly =
2875 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2876 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2879 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2881 ((zp->z_pflags & ZFS_NODUMP) != 0);
2882 XVA_SET_RTN(xvap, XAT_NODUMP);
2885 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2887 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2888 XVA_SET_RTN(xvap, XAT_OPAQUE);
2891 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2892 xoap->xoa_av_quarantined =
2893 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2894 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2897 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2898 xoap->xoa_av_modified =
2899 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2900 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2903 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2904 vp->v_type == VREG) {
2905 zfs_sa_get_scanstamp(zp, xvap);
2908 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2911 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2912 times, sizeof (times));
2913 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2914 XVA_SET_RTN(xvap, XAT_CREATETIME);
2917 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2918 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2919 XVA_SET_RTN(xvap, XAT_REPARSE);
2921 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2922 xoap->xoa_generation = zp->z_gen;
2923 XVA_SET_RTN(xvap, XAT_GEN);
2926 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2928 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2929 XVA_SET_RTN(xvap, XAT_OFFLINE);
2932 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2934 ((zp->z_pflags & ZFS_SPARSE) != 0);
2935 XVA_SET_RTN(xvap, XAT_SPARSE);
2939 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2940 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2941 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2942 ZFS_TIME_DECODE(&vap->va_birthtime, crtime);
2944 mutex_exit(&zp->z_lock);
2946 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2947 vap->va_blksize = blksize;
2948 vap->va_bytes = nblocks << 9; /* nblocks * 512 */
2950 if (zp->z_blksz == 0) {
2952 * Block size hasn't been set; suggest maximal I/O transfers.
2954 vap->va_blksize = zfsvfs->z_max_blksz;
2962 * Set the file attributes to the values contained in the
2965 * IN: vp - vnode of file to be modified.
2966 * vap - new attribute values.
2967 * If AT_XVATTR set, then optional attrs are being set
2968 * flags - ATTR_UTIME set if non-default time values provided.
2969 * - ATTR_NOACLCHECK (CIFS context only).
2970 * cr - credentials of caller.
2971 * ct - caller context
2973 * RETURN: 0 on success, error code on failure.
2976 * vp - ctime updated, mtime updated if size changed.
2980 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2981 caller_context_t *ct)
2983 znode_t *zp = VTOZ(vp);
2984 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2989 uint_t mask = vap->va_mask;
2990 uint_t saved_mask = 0;
2991 uint64_t saved_mode;
2994 uint64_t new_uid, new_gid;
2996 uint64_t mtime[2], ctime[2];
2998 int need_policy = FALSE;
3000 zfs_fuid_info_t *fuidp = NULL;
3001 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
3004 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
3005 boolean_t fuid_dirtied = B_FALSE;
3006 sa_bulk_attr_t bulk[7], xattr_bulk[7];
3007 int count = 0, xattr_count = 0;
3012 if (mask & AT_NOSET)
3013 return (SET_ERROR(EINVAL));
3018 zilog = zfsvfs->z_log;
3021 * Make sure that if we have ephemeral uid/gid or xvattr specified
3022 * that file system is at proper version level
3025 if (zfsvfs->z_use_fuids == B_FALSE &&
3026 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
3027 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
3028 (mask & AT_XVATTR))) {
3030 return (SET_ERROR(EINVAL));
3033 if (mask & AT_SIZE && vp->v_type == VDIR) {
3035 return (SET_ERROR(EISDIR));
3038 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
3040 return (SET_ERROR(EINVAL));
3044 * If this is an xvattr_t, then get a pointer to the structure of
3045 * optional attributes. If this is NULL, then we have a vattr_t.
3047 xoap = xva_getxoptattr(xvap);
3049 xva_init(&tmpxvattr);
3052 * Immutable files can only alter immutable bit and atime
3054 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
3055 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
3056 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
3058 return (SET_ERROR(EPERM));
3061 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
3063 return (SET_ERROR(EPERM));
3067 * Verify timestamps doesn't overflow 32 bits.
3068 * ZFS can handle large timestamps, but 32bit syscalls can't
3069 * handle times greater than 2039. This check should be removed
3070 * once large timestamps are fully supported.
3072 if (mask & (AT_ATIME | AT_MTIME)) {
3073 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
3074 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
3076 return (SET_ERROR(EOVERFLOW));
3084 /* Can this be moved to before the top label? */
3085 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
3087 return (SET_ERROR(EROFS));
3091 * First validate permissions
3094 if (mask & AT_SIZE) {
3096 * XXX - Note, we are not providing any open
3097 * mode flags here (like FNDELAY), so we may
3098 * block if there are locks present... this
3099 * should be addressed in openat().
3101 /* XXX - would it be OK to generate a log record here? */
3102 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
3109 if (mask & (AT_ATIME|AT_MTIME) ||
3110 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
3111 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
3112 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
3113 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
3114 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
3115 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
3116 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
3117 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
3121 if (mask & (AT_UID|AT_GID)) {
3122 int idmask = (mask & (AT_UID|AT_GID));
3127 * NOTE: even if a new mode is being set,
3128 * we may clear S_ISUID/S_ISGID bits.
3131 if (!(mask & AT_MODE))
3132 vap->va_mode = zp->z_mode;
3135 * Take ownership or chgrp to group we are a member of
3138 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
3139 take_group = (mask & AT_GID) &&
3140 zfs_groupmember(zfsvfs, vap->va_gid, cr);
3143 * If both AT_UID and AT_GID are set then take_owner and
3144 * take_group must both be set in order to allow taking
3147 * Otherwise, send the check through secpolicy_vnode_setattr()
3151 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
3152 ((idmask == AT_UID) && take_owner) ||
3153 ((idmask == AT_GID) && take_group)) {
3154 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
3155 skipaclchk, cr) == 0) {
3157 * Remove setuid/setgid for non-privileged users
3159 secpolicy_setid_clear(vap, vp, cr);
3160 trim_mask = (mask & (AT_UID|AT_GID));
3169 mutex_enter(&zp->z_lock);
3170 oldva.va_mode = zp->z_mode;
3171 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
3172 if (mask & AT_XVATTR) {
3174 * Update xvattr mask to include only those attributes
3175 * that are actually changing.
3177 * the bits will be restored prior to actually setting
3178 * the attributes so the caller thinks they were set.
3180 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
3181 if (xoap->xoa_appendonly !=
3182 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
3185 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
3186 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
3190 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
3191 if (xoap->xoa_nounlink !=
3192 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
3195 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
3196 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
3200 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
3201 if (xoap->xoa_immutable !=
3202 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
3205 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
3206 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
3210 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
3211 if (xoap->xoa_nodump !=
3212 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
3215 XVA_CLR_REQ(xvap, XAT_NODUMP);
3216 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
3220 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3221 if (xoap->xoa_av_modified !=
3222 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3225 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3226 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
3230 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3231 if ((vp->v_type != VREG &&
3232 xoap->xoa_av_quarantined) ||
3233 xoap->xoa_av_quarantined !=
3234 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3237 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3238 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
3242 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3243 mutex_exit(&zp->z_lock);
3245 return (SET_ERROR(EPERM));
3248 if (need_policy == FALSE &&
3249 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3250 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3255 mutex_exit(&zp->z_lock);
3257 if (mask & AT_MODE) {
3258 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3259 err = secpolicy_setid_setsticky_clear(vp, vap,
3265 trim_mask |= AT_MODE;
3273 * If trim_mask is set then take ownership
3274 * has been granted or write_acl is present and user
3275 * has the ability to modify mode. In that case remove
3276 * UID|GID and or MODE from mask so that
3277 * secpolicy_vnode_setattr() doesn't revoke it.
3281 saved_mask = vap->va_mask;
3282 vap->va_mask &= ~trim_mask;
3283 if (trim_mask & AT_MODE) {
3285 * Save the mode, as secpolicy_vnode_setattr()
3286 * will overwrite it with ova.va_mode.
3288 saved_mode = vap->va_mode;
3291 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3292 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3299 vap->va_mask |= saved_mask;
3300 if (trim_mask & AT_MODE) {
3302 * Recover the mode after
3303 * secpolicy_vnode_setattr().
3305 vap->va_mode = saved_mode;
3311 * secpolicy_vnode_setattr, or take ownership may have
3314 mask = vap->va_mask;
3316 if ((mask & (AT_UID | AT_GID))) {
3317 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3318 &xattr_obj, sizeof (xattr_obj));
3320 if (err == 0 && xattr_obj) {
3321 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3325 if (mask & AT_UID) {
3326 new_uid = zfs_fuid_create(zfsvfs,
3327 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3328 if (new_uid != zp->z_uid &&
3329 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3331 VN_RELE(ZTOV(attrzp));
3332 err = SET_ERROR(EDQUOT);
3337 if (mask & AT_GID) {
3338 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3339 cr, ZFS_GROUP, &fuidp);
3340 if (new_gid != zp->z_gid &&
3341 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3343 VN_RELE(ZTOV(attrzp));
3344 err = SET_ERROR(EDQUOT);
3349 tx = dmu_tx_create(zfsvfs->z_os);
3351 if (mask & AT_MODE) {
3352 uint64_t pmode = zp->z_mode;
3354 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3356 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3357 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3358 err = SET_ERROR(EPERM);
3362 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3365 mutex_enter(&zp->z_lock);
3366 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3368 * Are we upgrading ACL from old V0 format
3371 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3372 zfs_znode_acl_version(zp) ==
3373 ZFS_ACL_VERSION_INITIAL) {
3374 dmu_tx_hold_free(tx, acl_obj, 0,
3376 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3377 0, aclp->z_acl_bytes);
3379 dmu_tx_hold_write(tx, acl_obj, 0,
3382 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3383 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3384 0, aclp->z_acl_bytes);
3386 mutex_exit(&zp->z_lock);
3387 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3389 if ((mask & AT_XVATTR) &&
3390 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3391 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3393 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3397 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3400 fuid_dirtied = zfsvfs->z_fuid_dirty;
3402 zfs_fuid_txhold(zfsvfs, tx);
3404 zfs_sa_upgrade_txholds(tx, zp);
3406 err = dmu_tx_assign(tx, TXG_WAIT);
3412 * Set each attribute requested.
3413 * We group settings according to the locks they need to acquire.
3415 * Note: you cannot set ctime directly, although it will be
3416 * updated as a side-effect of calling this function.
3420 if (mask & (AT_UID|AT_GID|AT_MODE))
3421 mutex_enter(&zp->z_acl_lock);
3422 mutex_enter(&zp->z_lock);
3424 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3425 &zp->z_pflags, sizeof (zp->z_pflags));
3428 if (mask & (AT_UID|AT_GID|AT_MODE))
3429 mutex_enter(&attrzp->z_acl_lock);
3430 mutex_enter(&attrzp->z_lock);
3431 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3432 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3433 sizeof (attrzp->z_pflags));
3436 if (mask & (AT_UID|AT_GID)) {
3438 if (mask & AT_UID) {
3439 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3440 &new_uid, sizeof (new_uid));
3441 zp->z_uid = new_uid;
3443 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3444 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3446 attrzp->z_uid = new_uid;
3450 if (mask & AT_GID) {
3451 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3452 NULL, &new_gid, sizeof (new_gid));
3453 zp->z_gid = new_gid;
3455 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3456 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3458 attrzp->z_gid = new_gid;
3461 if (!(mask & AT_MODE)) {
3462 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3463 NULL, &new_mode, sizeof (new_mode));
3464 new_mode = zp->z_mode;
3466 err = zfs_acl_chown_setattr(zp);
3469 err = zfs_acl_chown_setattr(attrzp);
3474 if (mask & AT_MODE) {
3475 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3476 &new_mode, sizeof (new_mode));
3477 zp->z_mode = new_mode;
3478 ASSERT3U((uintptr_t)aclp, !=, 0);
3479 err = zfs_aclset_common(zp, aclp, cr, tx);
3481 if (zp->z_acl_cached)
3482 zfs_acl_free(zp->z_acl_cached);
3483 zp->z_acl_cached = aclp;
3488 if (mask & AT_ATIME) {
3489 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3490 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3491 &zp->z_atime, sizeof (zp->z_atime));
3494 if (mask & AT_MTIME) {
3495 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3496 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3497 mtime, sizeof (mtime));
3500 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3501 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3502 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3503 NULL, mtime, sizeof (mtime));
3504 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3505 &ctime, sizeof (ctime));
3506 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3508 } else if (mask != 0) {
3509 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3510 &ctime, sizeof (ctime));
3511 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3514 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3515 SA_ZPL_CTIME(zfsvfs), NULL,
3516 &ctime, sizeof (ctime));
3517 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3518 mtime, ctime, B_TRUE);
3522 * Do this after setting timestamps to prevent timestamp
3523 * update from toggling bit
3526 if (xoap && (mask & AT_XVATTR)) {
3529 * restore trimmed off masks
3530 * so that return masks can be set for caller.
3533 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3534 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3536 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3537 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3539 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3540 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3542 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3543 XVA_SET_REQ(xvap, XAT_NODUMP);
3545 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3546 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3548 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3549 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3552 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3553 ASSERT(vp->v_type == VREG);
3555 zfs_xvattr_set(zp, xvap, tx);
3559 zfs_fuid_sync(zfsvfs, tx);
3562 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3564 mutex_exit(&zp->z_lock);
3565 if (mask & (AT_UID|AT_GID|AT_MODE))
3566 mutex_exit(&zp->z_acl_lock);
3569 if (mask & (AT_UID|AT_GID|AT_MODE))
3570 mutex_exit(&attrzp->z_acl_lock);
3571 mutex_exit(&attrzp->z_lock);
3574 if (err == 0 && attrzp) {
3575 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3581 VN_RELE(ZTOV(attrzp));
3587 zfs_fuid_info_free(fuidp);
3593 if (err == ERESTART)
3596 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3601 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3602 zil_commit(zilog, 0);
3608 typedef struct zfs_zlock {
3609 krwlock_t *zl_rwlock; /* lock we acquired */
3610 znode_t *zl_znode; /* znode we held */
3611 struct zfs_zlock *zl_next; /* next in list */
3615 * Drop locks and release vnodes that were held by zfs_rename_lock().
3618 zfs_rename_unlock(zfs_zlock_t **zlpp)
3622 while ((zl = *zlpp) != NULL) {
3623 if (zl->zl_znode != NULL)
3624 VN_RELE(ZTOV(zl->zl_znode));
3625 rw_exit(zl->zl_rwlock);
3626 *zlpp = zl->zl_next;
3627 kmem_free(zl, sizeof (*zl));
3632 * Search back through the directory tree, using the ".." entries.
3633 * Lock each directory in the chain to prevent concurrent renames.
3634 * Fail any attempt to move a directory into one of its own descendants.
3635 * XXX - z_parent_lock can overlap with map or grow locks
3638 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3642 uint64_t rootid = zp->z_zfsvfs->z_root;
3643 uint64_t oidp = zp->z_id;
3644 krwlock_t *rwlp = &szp->z_parent_lock;
3645 krw_t rw = RW_WRITER;
3648 * First pass write-locks szp and compares to zp->z_id.
3649 * Later passes read-lock zp and compare to zp->z_parent.
3652 if (!rw_tryenter(rwlp, rw)) {
3654 * Another thread is renaming in this path.
3655 * Note that if we are a WRITER, we don't have any
3656 * parent_locks held yet.
3658 if (rw == RW_READER && zp->z_id > szp->z_id) {
3660 * Drop our locks and restart
3662 zfs_rename_unlock(&zl);
3666 rwlp = &szp->z_parent_lock;
3671 * Wait for other thread to drop its locks
3677 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3678 zl->zl_rwlock = rwlp;
3679 zl->zl_znode = NULL;
3680 zl->zl_next = *zlpp;
3683 if (oidp == szp->z_id) /* We're a descendant of szp */
3684 return (SET_ERROR(EINVAL));
3686 if (oidp == rootid) /* We've hit the top */
3689 if (rw == RW_READER) { /* i.e. not the first pass */
3690 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3695 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3696 &oidp, sizeof (oidp));
3697 rwlp = &zp->z_parent_lock;
3700 } while (zp->z_id != sdzp->z_id);
3706 * Move an entry from the provided source directory to the target
3707 * directory. Change the entry name as indicated.
3709 * IN: sdvp - Source directory containing the "old entry".
3710 * snm - Old entry name.
3711 * tdvp - Target directory to contain the "new entry".
3712 * tnm - New entry name.
3713 * cr - credentials of caller.
3714 * ct - caller context
3715 * flags - case flags
3717 * RETURN: 0 on success, error code on failure.
3720 * sdvp,tdvp - ctime|mtime updated
3724 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3725 caller_context_t *ct, int flags)
3727 znode_t *tdzp, *szp, *tzp;
3728 znode_t *sdzp = VTOZ(sdvp);
3729 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs;
3732 zfs_dirlock_t *sdl, *tdl;
3735 int cmp, serr, terr;
3738 boolean_t waited = B_FALSE;
3741 ZFS_VERIFY_ZP(sdzp);
3742 zilog = zfsvfs->z_log;
3745 * Make sure we have the real vp for the target directory.
3747 if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3751 ZFS_VERIFY_ZP(tdzp);
3754 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3755 * ctldir appear to have the same v_vfsp.
3757 if (tdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) {
3759 return (SET_ERROR(EXDEV));
3762 if (zfsvfs->z_utf8 && u8_validate(tnm,
3763 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3765 return (SET_ERROR(EILSEQ));
3768 if (flags & FIGNORECASE)
3777 * This is to prevent the creation of links into attribute space
3778 * by renaming a linked file into/outof an attribute directory.
3779 * See the comment in zfs_link() for why this is considered bad.
3781 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3783 return (SET_ERROR(EINVAL));
3787 * Lock source and target directory entries. To prevent deadlock,
3788 * a lock ordering must be defined. We lock the directory with
3789 * the smallest object id first, or if it's a tie, the one with
3790 * the lexically first name.
3792 if (sdzp->z_id < tdzp->z_id) {
3794 } else if (sdzp->z_id > tdzp->z_id) {
3798 * First compare the two name arguments without
3799 * considering any case folding.
3801 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3803 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3804 ASSERT(error == 0 || !zfsvfs->z_utf8);
3807 * POSIX: "If the old argument and the new argument
3808 * both refer to links to the same existing file,
3809 * the rename() function shall return successfully
3810 * and perform no other action."
3816 * If the file system is case-folding, then we may
3817 * have some more checking to do. A case-folding file
3818 * system is either supporting mixed case sensitivity
3819 * access or is completely case-insensitive. Note
3820 * that the file system is always case preserving.
3822 * In mixed sensitivity mode case sensitive behavior
3823 * is the default. FIGNORECASE must be used to
3824 * explicitly request case insensitive behavior.
3826 * If the source and target names provided differ only
3827 * by case (e.g., a request to rename 'tim' to 'Tim'),
3828 * we will treat this as a special case in the
3829 * case-insensitive mode: as long as the source name
3830 * is an exact match, we will allow this to proceed as
3831 * a name-change request.
3833 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3834 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3835 flags & FIGNORECASE)) &&
3836 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3839 * case preserving rename request, require exact
3848 * If the source and destination directories are the same, we should
3849 * grab the z_name_lock of that directory only once.
3853 rw_enter(&sdzp->z_name_lock, RW_READER);
3857 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3858 ZEXISTS | zflg, NULL, NULL);
3859 terr = zfs_dirent_lock(&tdl,
3860 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3862 terr = zfs_dirent_lock(&tdl,
3863 tdzp, tnm, &tzp, zflg, NULL, NULL);
3864 serr = zfs_dirent_lock(&sdl,
3865 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3871 * Source entry invalid or not there.
3874 zfs_dirent_unlock(tdl);
3880 rw_exit(&sdzp->z_name_lock);
3883 * FreeBSD: In OpenSolaris they only check if rename source is
3884 * ".." here, because "." is handled in their lookup. This is
3885 * not the case for FreeBSD, so we check for "." explicitly.
3887 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0)
3888 serr = SET_ERROR(EINVAL);
3893 zfs_dirent_unlock(sdl);
3897 rw_exit(&sdzp->z_name_lock);
3899 if (strcmp(tnm, "..") == 0)
3900 terr = SET_ERROR(EINVAL);
3906 * Must have write access at the source to remove the old entry
3907 * and write access at the target to create the new entry.
3908 * Note that if target and source are the same, this can be
3909 * done in a single check.
3912 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3915 if (ZTOV(szp)->v_type == VDIR) {
3917 * Check to make sure rename is valid.
3918 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3920 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3925 * Does target exist?
3929 * Source and target must be the same type.
3931 if (ZTOV(szp)->v_type == VDIR) {
3932 if (ZTOV(tzp)->v_type != VDIR) {
3933 error = SET_ERROR(ENOTDIR);
3937 if (ZTOV(tzp)->v_type == VDIR) {
3938 error = SET_ERROR(EISDIR);
3943 * POSIX dictates that when the source and target
3944 * entries refer to the same file object, rename
3945 * must do nothing and exit without error.
3947 if (szp->z_id == tzp->z_id) {
3953 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3955 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3958 * notify the target directory if it is not the same
3959 * as source directory.
3962 vnevent_rename_dest_dir(tdvp, ct);
3965 tx = dmu_tx_create(zfsvfs->z_os);
3966 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3967 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3968 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3969 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3971 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3972 zfs_sa_upgrade_txholds(tx, tdzp);
3975 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3976 zfs_sa_upgrade_txholds(tx, tzp);
3979 zfs_sa_upgrade_txholds(tx, szp);
3980 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3981 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3984 zfs_rename_unlock(&zl);
3985 zfs_dirent_unlock(sdl);
3986 zfs_dirent_unlock(tdl);
3989 rw_exit(&sdzp->z_name_lock);
3994 if (error == ERESTART) {
4005 if (tzp) /* Attempt to remove the existing target */
4006 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
4009 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
4011 szp->z_pflags |= ZFS_AV_MODIFIED;
4013 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
4014 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
4017 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
4019 zfs_log_rename(zilog, tx, TX_RENAME |
4020 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
4021 sdl->dl_name, tdzp, tdl->dl_name, szp);
4024 * Update path information for the target vnode
4026 vn_renamepath(tdvp, ZTOV(szp), tnm,
4030 * At this point, we have successfully created
4031 * the target name, but have failed to remove
4032 * the source name. Since the create was done
4033 * with the ZRENAMING flag, there are
4034 * complications; for one, the link count is
4035 * wrong. The easiest way to deal with this
4036 * is to remove the newly created target, and
4037 * return the original error. This must
4038 * succeed; fortunately, it is very unlikely to
4039 * fail, since we just created it.
4041 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
4042 ZRENAMING, NULL), ==, 0);
4045 #ifdef FREEBSD_NAMECACHE
4049 cache_purge(ZTOV(szp));
4051 cache_purge(ZTOV(tzp));
4059 zfs_rename_unlock(&zl);
4061 zfs_dirent_unlock(sdl);
4062 zfs_dirent_unlock(tdl);
4065 rw_exit(&sdzp->z_name_lock);
4072 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4073 zil_commit(zilog, 0);
4081 * Insert the indicated symbolic reference entry into the directory.
4083 * IN: dvp - Directory to contain new symbolic link.
4084 * link - Name for new symlink entry.
4085 * vap - Attributes of new entry.
4086 * cr - credentials of caller.
4087 * ct - caller context
4088 * flags - case flags
4090 * RETURN: 0 on success, error code on failure.
4093 * dvp - ctime|mtime updated
4097 zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link,
4098 cred_t *cr, kthread_t *td)
4100 znode_t *zp, *dzp = VTOZ(dvp);
4103 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4105 uint64_t len = strlen(link);
4108 zfs_acl_ids_t acl_ids;
4109 boolean_t fuid_dirtied;
4110 uint64_t txtype = TX_SYMLINK;
4111 boolean_t waited = B_FALSE;
4114 ASSERT(vap->va_type == VLNK);
4118 zilog = zfsvfs->z_log;
4120 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
4121 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4123 return (SET_ERROR(EILSEQ));
4125 if (flags & FIGNORECASE)
4128 if (len > MAXPATHLEN) {
4130 return (SET_ERROR(ENAMETOOLONG));
4133 if ((error = zfs_acl_ids_create(dzp, 0,
4134 vap, cr, NULL, &acl_ids)) != 0) {
4139 getnewvnode_reserve(1);
4143 * Attempt to lock directory; fail if entry already exists.
4145 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
4147 zfs_acl_ids_free(&acl_ids);
4148 getnewvnode_drop_reserve();
4153 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4154 zfs_acl_ids_free(&acl_ids);
4155 zfs_dirent_unlock(dl);
4156 getnewvnode_drop_reserve();
4161 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
4162 zfs_acl_ids_free(&acl_ids);
4163 zfs_dirent_unlock(dl);
4164 getnewvnode_drop_reserve();
4166 return (SET_ERROR(EDQUOT));
4168 tx = dmu_tx_create(zfsvfs->z_os);
4169 fuid_dirtied = zfsvfs->z_fuid_dirty;
4170 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
4171 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4172 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
4173 ZFS_SA_BASE_ATTR_SIZE + len);
4174 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
4175 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
4176 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
4177 acl_ids.z_aclp->z_acl_bytes);
4180 zfs_fuid_txhold(zfsvfs, tx);
4181 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4183 zfs_dirent_unlock(dl);
4184 if (error == ERESTART) {
4190 zfs_acl_ids_free(&acl_ids);
4192 getnewvnode_drop_reserve();
4198 * Create a new object for the symlink.
4199 * for version 4 ZPL datsets the symlink will be an SA attribute
4201 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
4204 zfs_fuid_sync(zfsvfs, tx);
4206 mutex_enter(&zp->z_lock);
4208 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
4211 zfs_sa_symlink(zp, link, len, tx);
4212 mutex_exit(&zp->z_lock);
4215 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
4216 &zp->z_size, sizeof (zp->z_size), tx);
4218 * Insert the new object into the directory.
4220 (void) zfs_link_create(dl, zp, tx, ZNEW);
4222 if (flags & FIGNORECASE)
4224 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
4227 zfs_acl_ids_free(&acl_ids);
4231 getnewvnode_drop_reserve();
4233 zfs_dirent_unlock(dl);
4235 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4236 zil_commit(zilog, 0);
4243 * Return, in the buffer contained in the provided uio structure,
4244 * the symbolic path referred to by vp.
4246 * IN: vp - vnode of symbolic link.
4247 * uio - structure to contain the link path.
4248 * cr - credentials of caller.
4249 * ct - caller context
4251 * OUT: uio - structure containing the link path.
4253 * RETURN: 0 on success, error code on failure.
4256 * vp - atime updated
4260 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
4262 znode_t *zp = VTOZ(vp);
4263 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4269 mutex_enter(&zp->z_lock);
4271 error = sa_lookup_uio(zp->z_sa_hdl,
4272 SA_ZPL_SYMLINK(zfsvfs), uio);
4274 error = zfs_sa_readlink(zp, uio);
4275 mutex_exit(&zp->z_lock);
4277 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4284 * Insert a new entry into directory tdvp referencing svp.
4286 * IN: tdvp - Directory to contain new entry.
4287 * svp - vnode of new entry.
4288 * name - name of new entry.
4289 * cr - credentials of caller.
4290 * ct - caller context
4292 * RETURN: 0 on success, error code on failure.
4295 * tdvp - ctime|mtime updated
4296 * svp - ctime updated
4300 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
4301 caller_context_t *ct, int flags)
4303 znode_t *dzp = VTOZ(tdvp);
4305 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4314 boolean_t waited = B_FALSE;
4316 ASSERT(tdvp->v_type == VDIR);
4320 zilog = zfsvfs->z_log;
4322 if (VOP_REALVP(svp, &realvp, ct) == 0)
4326 * POSIX dictates that we return EPERM here.
4327 * Better choices include ENOTSUP or EISDIR.
4329 if (svp->v_type == VDIR) {
4331 return (SET_ERROR(EPERM));
4338 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
4339 * ctldir appear to have the same v_vfsp.
4341 if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) {
4343 return (SET_ERROR(EXDEV));
4346 /* Prevent links to .zfs/shares files */
4348 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4349 &parent, sizeof (uint64_t))) != 0) {
4353 if (parent == zfsvfs->z_shares_dir) {
4355 return (SET_ERROR(EPERM));
4358 if (zfsvfs->z_utf8 && u8_validate(name,
4359 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4361 return (SET_ERROR(EILSEQ));
4363 if (flags & FIGNORECASE)
4367 * We do not support links between attributes and non-attributes
4368 * because of the potential security risk of creating links
4369 * into "normal" file space in order to circumvent restrictions
4370 * imposed in attribute space.
4372 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4374 return (SET_ERROR(EINVAL));
4378 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4379 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) {
4381 return (SET_ERROR(EPERM));
4384 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4391 * Attempt to lock directory; fail if entry already exists.
4393 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4399 tx = dmu_tx_create(zfsvfs->z_os);
4400 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4401 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4402 zfs_sa_upgrade_txholds(tx, szp);
4403 zfs_sa_upgrade_txholds(tx, dzp);
4404 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4406 zfs_dirent_unlock(dl);
4407 if (error == ERESTART) {
4418 error = zfs_link_create(dl, szp, tx, 0);
4421 uint64_t txtype = TX_LINK;
4422 if (flags & FIGNORECASE)
4424 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4429 zfs_dirent_unlock(dl);
4432 vnevent_link(svp, ct);
4435 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4436 zil_commit(zilog, 0);
4444 * zfs_null_putapage() is used when the file system has been force
4445 * unmounted. It just drops the pages.
4449 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4450 size_t *lenp, int flags, cred_t *cr)
4452 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4457 * Push a page out to disk, klustering if possible.
4459 * IN: vp - file to push page to.
4460 * pp - page to push.
4461 * flags - additional flags.
4462 * cr - credentials of caller.
4464 * OUT: offp - start of range pushed.
4465 * lenp - len of range pushed.
4467 * RETURN: 0 on success, error code on failure.
4469 * NOTE: callers must have locked the page to be pushed. On
4470 * exit, the page (and all other pages in the kluster) must be
4475 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4476 size_t *lenp, int flags, cred_t *cr)
4478 znode_t *zp = VTOZ(vp);
4479 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4481 u_offset_t off, koff;
4488 * If our blocksize is bigger than the page size, try to kluster
4489 * multiple pages so that we write a full block (thus avoiding
4490 * a read-modify-write).
4492 if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4493 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4494 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4495 ASSERT(koff <= zp->z_size);
4496 if (koff + klen > zp->z_size)
4497 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4498 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4500 ASSERT3U(btop(len), ==, btopr(len));
4503 * Can't push pages past end-of-file.
4505 if (off >= zp->z_size) {
4506 /* ignore all pages */
4509 } else if (off + len > zp->z_size) {
4510 int npages = btopr(zp->z_size - off);
4513 page_list_break(&pp, &trunc, npages);
4514 /* ignore pages past end of file */
4516 pvn_write_done(trunc, flags);
4517 len = zp->z_size - off;
4520 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4521 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4522 err = SET_ERROR(EDQUOT);
4525 tx = dmu_tx_create(zfsvfs->z_os);
4526 dmu_tx_hold_write(tx, zp->z_id, off, len);
4528 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4529 zfs_sa_upgrade_txholds(tx, zp);
4530 err = dmu_tx_assign(tx, TXG_WAIT);
4536 if (zp->z_blksz <= PAGESIZE) {
4537 caddr_t va = zfs_map_page(pp, S_READ);
4538 ASSERT3U(len, <=, PAGESIZE);
4539 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4540 zfs_unmap_page(pp, va);
4542 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4546 uint64_t mtime[2], ctime[2];
4547 sa_bulk_attr_t bulk[3];
4550 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4552 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4554 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4556 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4558 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4563 pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4573 * Copy the portion of the file indicated from pages into the file.
4574 * The pages are stored in a page list attached to the files vnode.
4576 * IN: vp - vnode of file to push page data to.
4577 * off - position in file to put data.
4578 * len - amount of data to write.
4579 * flags - flags to control the operation.
4580 * cr - credentials of caller.
4581 * ct - caller context.
4583 * RETURN: 0 on success, error code on failure.
4586 * vp - ctime|mtime updated
4590 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4591 caller_context_t *ct)
4593 znode_t *zp = VTOZ(vp);
4594 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4606 * Align this request to the file block size in case we kluster.
4607 * XXX - this can result in pretty aggresive locking, which can
4608 * impact simultanious read/write access. One option might be
4609 * to break up long requests (len == 0) into block-by-block
4610 * operations to get narrower locking.
4612 blksz = zp->z_blksz;
4614 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4617 if (len > 0 && ISP2(blksz))
4618 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4624 * Search the entire vp list for pages >= io_off.
4626 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4627 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4630 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4632 if (off > zp->z_size) {
4633 /* past end of file */
4634 zfs_range_unlock(rl);
4639 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4641 for (off = io_off; io_off < off + len; io_off += io_len) {
4642 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4643 pp = page_lookup(vp, io_off,
4644 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4646 pp = page_lookup_nowait(vp, io_off,
4647 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4650 if (pp != NULL && pvn_getdirty(pp, flags)) {
4654 * Found a dirty page to push
4656 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4664 zfs_range_unlock(rl);
4665 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4666 zil_commit(zfsvfs->z_log, zp->z_id);
4674 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4676 znode_t *zp = VTOZ(vp);
4677 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4680 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4681 if (zp->z_sa_hdl == NULL) {
4683 * The fs has been unmounted, or we did a
4684 * suspend/resume and this file no longer exists.
4686 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4691 mutex_enter(&zp->z_lock);
4692 if (zp->z_unlinked) {
4694 * Fast path to recycle a vnode of a removed file.
4696 mutex_exit(&zp->z_lock);
4697 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4701 mutex_exit(&zp->z_lock);
4703 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4704 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4706 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4707 zfs_sa_upgrade_txholds(tx, zp);
4708 error = dmu_tx_assign(tx, TXG_WAIT);
4712 mutex_enter(&zp->z_lock);
4713 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4714 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4715 zp->z_atime_dirty = 0;
4716 mutex_exit(&zp->z_lock);
4720 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4725 * Bounds-check the seek operation.
4727 * IN: vp - vnode seeking within
4728 * ooff - old file offset
4729 * noffp - pointer to new file offset
4730 * ct - caller context
4732 * RETURN: 0 on success, EINVAL if new offset invalid.
4736 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4737 caller_context_t *ct)
4739 if (vp->v_type == VDIR)
4741 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4745 * Pre-filter the generic locking function to trap attempts to place
4746 * a mandatory lock on a memory mapped file.
4749 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4750 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4752 znode_t *zp = VTOZ(vp);
4753 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4759 * We are following the UFS semantics with respect to mapcnt
4760 * here: If we see that the file is mapped already, then we will
4761 * return an error, but we don't worry about races between this
4762 * function and zfs_map().
4764 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4766 return (SET_ERROR(EAGAIN));
4769 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4773 * If we can't find a page in the cache, we will create a new page
4774 * and fill it with file data. For efficiency, we may try to fill
4775 * multiple pages at once (klustering) to fill up the supplied page
4776 * list. Note that the pages to be filled are held with an exclusive
4777 * lock to prevent access by other threads while they are being filled.
4780 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4781 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4783 znode_t *zp = VTOZ(vp);
4784 page_t *pp, *cur_pp;
4785 objset_t *os = zp->z_zfsvfs->z_os;
4786 u_offset_t io_off, total;
4790 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4792 * We only have a single page, don't bother klustering
4796 pp = page_create_va(vp, io_off, io_len,
4797 PG_EXCL | PG_WAIT, seg, addr);
4800 * Try to find enough pages to fill the page list
4802 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4803 &io_len, off, plsz, 0);
4807 * The page already exists, nothing to do here.
4814 * Fill the pages in the kluster.
4817 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4820 ASSERT3U(io_off, ==, cur_pp->p_offset);
4821 va = zfs_map_page(cur_pp, S_WRITE);
4822 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4824 zfs_unmap_page(cur_pp, va);
4826 /* On error, toss the entire kluster */
4827 pvn_read_done(pp, B_ERROR);
4828 /* convert checksum errors into IO errors */
4830 err = SET_ERROR(EIO);
4833 cur_pp = cur_pp->p_next;
4837 * Fill in the page list array from the kluster starting
4838 * from the desired offset `off'.
4839 * NOTE: the page list will always be null terminated.
4841 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4842 ASSERT(pl == NULL || (*pl)->p_offset == off);
4848 * Return pointers to the pages for the file region [off, off + len]
4849 * in the pl array. If plsz is greater than len, this function may
4850 * also return page pointers from after the specified region
4851 * (i.e. the region [off, off + plsz]). These additional pages are
4852 * only returned if they are already in the cache, or were created as
4853 * part of a klustered read.
4855 * IN: vp - vnode of file to get data from.
4856 * off - position in file to get data from.
4857 * len - amount of data to retrieve.
4858 * plsz - length of provided page list.
4859 * seg - segment to obtain pages for.
4860 * addr - virtual address of fault.
4861 * rw - mode of created pages.
4862 * cr - credentials of caller.
4863 * ct - caller context.
4865 * OUT: protp - protection mode of created pages.
4866 * pl - list of pages created.
4868 * RETURN: 0 on success, error code on failure.
4871 * vp - atime updated
4875 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4876 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4877 enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4879 znode_t *zp = VTOZ(vp);
4880 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4884 /* we do our own caching, faultahead is unnecessary */
4887 else if (len > plsz)
4890 len = P2ROUNDUP(len, PAGESIZE);
4891 ASSERT(plsz >= len);
4900 * Loop through the requested range [off, off + len) looking
4901 * for pages. If we don't find a page, we will need to create
4902 * a new page and fill it with data from the file.
4905 if (*pl = page_lookup(vp, off, SE_SHARED))
4907 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4910 ASSERT3U((*pl)->p_offset, ==, off);
4914 ASSERT3U(len, >=, PAGESIZE);
4917 ASSERT3U(plsz, >=, PAGESIZE);
4924 * Fill out the page array with any pages already in the cache.
4927 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4934 * Release any pages we have previously locked.
4939 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4949 * Request a memory map for a section of a file. This code interacts
4950 * with common code and the VM system as follows:
4952 * - common code calls mmap(), which ends up in smmap_common()
4953 * - this calls VOP_MAP(), which takes you into (say) zfs
4954 * - zfs_map() calls as_map(), passing segvn_create() as the callback
4955 * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4956 * - zfs_addmap() updates z_mapcnt
4960 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4961 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4962 caller_context_t *ct)
4964 znode_t *zp = VTOZ(vp);
4965 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4966 segvn_crargs_t vn_a;
4972 if ((prot & PROT_WRITE) && (zp->z_pflags &
4973 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4975 return (SET_ERROR(EPERM));
4978 if ((prot & (PROT_READ | PROT_EXEC)) &&
4979 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4981 return (SET_ERROR(EACCES));
4984 if (vp->v_flag & VNOMAP) {
4986 return (SET_ERROR(ENOSYS));
4989 if (off < 0 || len > MAXOFFSET_T - off) {
4991 return (SET_ERROR(ENXIO));
4994 if (vp->v_type != VREG) {
4996 return (SET_ERROR(ENODEV));
5000 * If file is locked, disallow mapping.
5002 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
5004 return (SET_ERROR(EAGAIN));
5008 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
5016 vn_a.offset = (u_offset_t)off;
5017 vn_a.type = flags & MAP_TYPE;
5019 vn_a.maxprot = maxprot;
5022 vn_a.flags = flags & ~MAP_TYPE;
5024 vn_a.lgrp_mem_policy_flags = 0;
5026 error = as_map(as, *addrp, len, segvn_create, &vn_a);
5035 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5036 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
5037 caller_context_t *ct)
5039 uint64_t pages = btopr(len);
5041 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
5046 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
5047 * more accurate mtime for the associated file. Since we don't have a way of
5048 * detecting when the data was actually modified, we have to resort to
5049 * heuristics. If an explicit msync() is done, then we mark the mtime when the
5050 * last page is pushed. The problem occurs when the msync() call is omitted,
5051 * which by far the most common case:
5059 * putpage() via fsflush
5061 * If we wait until fsflush to come along, we can have a modification time that
5062 * is some arbitrary point in the future. In order to prevent this in the
5063 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
5068 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5069 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
5070 caller_context_t *ct)
5072 uint64_t pages = btopr(len);
5074 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
5075 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
5077 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
5078 vn_has_cached_data(vp))
5079 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
5085 * Free or allocate space in a file. Currently, this function only
5086 * supports the `F_FREESP' command. However, this command is somewhat
5087 * misnamed, as its functionality includes the ability to allocate as
5088 * well as free space.
5090 * IN: vp - vnode of file to free data in.
5091 * cmd - action to take (only F_FREESP supported).
5092 * bfp - section of file to free/alloc.
5093 * flag - current file open mode flags.
5094 * offset - current file offset.
5095 * cr - credentials of caller [UNUSED].
5096 * ct - caller context.
5098 * RETURN: 0 on success, error code on failure.
5101 * vp - ctime|mtime updated
5105 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
5106 offset_t offset, cred_t *cr, caller_context_t *ct)
5108 znode_t *zp = VTOZ(vp);
5109 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5116 if (cmd != F_FREESP) {
5118 return (SET_ERROR(EINVAL));
5121 if (error = convoff(vp, bfp, 0, offset)) {
5126 if (bfp->l_len < 0) {
5128 return (SET_ERROR(EINVAL));
5132 len = bfp->l_len; /* 0 means from off to end of file */
5134 error = zfs_freesp(zp, off, len, flag, TRUE);
5141 CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid));
5142 CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid));
5146 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
5148 znode_t *zp = VTOZ(vp);
5149 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5152 uint64_t object = zp->z_id;
5159 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
5160 &gen64, sizeof (uint64_t))) != 0) {
5165 gen = (uint32_t)gen64;
5167 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
5170 if (fidp->fid_len < size) {
5171 fidp->fid_len = size;
5173 return (SET_ERROR(ENOSPC));
5176 fidp->fid_len = size;
5179 zfid = (zfid_short_t *)fidp;
5181 zfid->zf_len = size;
5183 for (i = 0; i < sizeof (zfid->zf_object); i++)
5184 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
5186 /* Must have a non-zero generation number to distinguish from .zfs */
5189 for (i = 0; i < sizeof (zfid->zf_gen); i++)
5190 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
5192 if (size == LONG_FID_LEN) {
5193 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
5196 zlfid = (zfid_long_t *)fidp;
5198 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
5199 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
5201 /* XXX - this should be the generation number for the objset */
5202 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
5203 zlfid->zf_setgen[i] = 0;
5211 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
5212 caller_context_t *ct)
5224 case _PC_FILESIZEBITS:
5228 case _PC_XATTR_EXISTS:
5230 zfsvfs = zp->z_zfsvfs;
5234 error = zfs_dirent_lock(&dl, zp, "", &xzp,
5235 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
5237 zfs_dirent_unlock(dl);
5238 if (!zfs_dirempty(xzp))
5241 } else if (error == ENOENT) {
5243 * If there aren't extended attributes, it's the
5244 * same as having zero of them.
5251 case _PC_SATTR_ENABLED:
5252 case _PC_SATTR_EXISTS:
5253 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
5254 (vp->v_type == VREG || vp->v_type == VDIR);
5257 case _PC_ACCESS_FILTERING:
5258 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
5262 case _PC_ACL_ENABLED:
5263 *valp = _ACL_ACE_ENABLED;
5266 case _PC_MIN_HOLE_SIZE:
5267 *valp = (int)SPA_MINBLOCKSIZE;
5270 case _PC_TIMESTAMP_RESOLUTION:
5271 /* nanosecond timestamp resolution */
5275 case _PC_ACL_EXTENDED:
5283 case _PC_ACL_PATH_MAX:
5284 *valp = ACL_MAX_ENTRIES;
5288 return (EOPNOTSUPP);
5294 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5295 caller_context_t *ct)
5297 znode_t *zp = VTOZ(vp);
5298 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5300 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5304 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5312 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5313 caller_context_t *ct)
5315 znode_t *zp = VTOZ(vp);
5316 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5318 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5319 zilog_t *zilog = zfsvfs->z_log;
5324 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5326 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5327 zil_commit(zilog, 0);
5335 * The smallest read we may consider to loan out an arcbuf.
5336 * This must be a power of 2.
5338 int zcr_blksz_min = (1 << 10); /* 1K */
5340 * If set to less than the file block size, allow loaning out of an
5341 * arcbuf for a partial block read. This must be a power of 2.
5343 int zcr_blksz_max = (1 << 17); /* 128K */
5347 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5348 caller_context_t *ct)
5350 znode_t *zp = VTOZ(vp);
5351 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5352 int max_blksz = zfsvfs->z_max_blksz;
5353 uio_t *uio = &xuio->xu_uio;
5354 ssize_t size = uio->uio_resid;
5355 offset_t offset = uio->uio_loffset;
5360 int preamble, postamble;
5362 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5363 return (SET_ERROR(EINVAL));
5370 * Loan out an arc_buf for write if write size is bigger than
5371 * max_blksz, and the file's block size is also max_blksz.
5374 if (size < blksz || zp->z_blksz != blksz) {
5376 return (SET_ERROR(EINVAL));
5379 * Caller requests buffers for write before knowing where the
5380 * write offset might be (e.g. NFS TCP write).
5385 preamble = P2PHASE(offset, blksz);
5387 preamble = blksz - preamble;
5392 postamble = P2PHASE(size, blksz);
5395 fullblk = size / blksz;
5396 (void) dmu_xuio_init(xuio,
5397 (preamble != 0) + fullblk + (postamble != 0));
5398 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5399 int, postamble, int,
5400 (preamble != 0) + fullblk + (postamble != 0));
5403 * Have to fix iov base/len for partial buffers. They
5404 * currently represent full arc_buf's.
5407 /* data begins in the middle of the arc_buf */
5408 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5411 (void) dmu_xuio_add(xuio, abuf,
5412 blksz - preamble, preamble);
5415 for (i = 0; i < fullblk; i++) {
5416 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5419 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5423 /* data ends in the middle of the arc_buf */
5424 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5427 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5432 * Loan out an arc_buf for read if the read size is larger than
5433 * the current file block size. Block alignment is not
5434 * considered. Partial arc_buf will be loaned out for read.
5436 blksz = zp->z_blksz;
5437 if (blksz < zcr_blksz_min)
5438 blksz = zcr_blksz_min;
5439 if (blksz > zcr_blksz_max)
5440 blksz = zcr_blksz_max;
5441 /* avoid potential complexity of dealing with it */
5442 if (blksz > max_blksz) {
5444 return (SET_ERROR(EINVAL));
5447 maxsize = zp->z_size - uio->uio_loffset;
5451 if (size < blksz || vn_has_cached_data(vp)) {
5453 return (SET_ERROR(EINVAL));
5458 return (SET_ERROR(EINVAL));
5461 uio->uio_extflg = UIO_XUIO;
5462 XUIO_XUZC_RW(xuio) = ioflag;
5469 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5473 int ioflag = XUIO_XUZC_RW(xuio);
5475 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5477 i = dmu_xuio_cnt(xuio);
5479 abuf = dmu_xuio_arcbuf(xuio, i);
5481 * if abuf == NULL, it must be a write buffer
5482 * that has been returned in zfs_write().
5485 dmu_return_arcbuf(abuf);
5486 ASSERT(abuf || ioflag == UIO_WRITE);
5489 dmu_xuio_fini(xuio);
5494 * Predeclare these here so that the compiler assumes that
5495 * this is an "old style" function declaration that does
5496 * not include arguments => we won't get type mismatch errors
5497 * in the initializations that follow.
5499 static int zfs_inval();
5500 static int zfs_isdir();
5505 return (SET_ERROR(EINVAL));
5511 return (SET_ERROR(EISDIR));
5514 * Directory vnode operations template
5516 vnodeops_t *zfs_dvnodeops;
5517 const fs_operation_def_t zfs_dvnodeops_template[] = {
5518 VOPNAME_OPEN, { .vop_open = zfs_open },
5519 VOPNAME_CLOSE, { .vop_close = zfs_close },
5520 VOPNAME_READ, { .error = zfs_isdir },
5521 VOPNAME_WRITE, { .error = zfs_isdir },
5522 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5523 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5524 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5525 VOPNAME_ACCESS, { .vop_access = zfs_access },
5526 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5527 VOPNAME_CREATE, { .vop_create = zfs_create },
5528 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5529 VOPNAME_LINK, { .vop_link = zfs_link },
5530 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5531 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir },
5532 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5533 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5534 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink },
5535 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5536 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5537 VOPNAME_FID, { .vop_fid = zfs_fid },
5538 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5539 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5540 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5541 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5542 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5547 * Regular file vnode operations template
5549 vnodeops_t *zfs_fvnodeops;
5550 const fs_operation_def_t zfs_fvnodeops_template[] = {
5551 VOPNAME_OPEN, { .vop_open = zfs_open },
5552 VOPNAME_CLOSE, { .vop_close = zfs_close },
5553 VOPNAME_READ, { .vop_read = zfs_read },
5554 VOPNAME_WRITE, { .vop_write = zfs_write },
5555 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5556 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5557 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5558 VOPNAME_ACCESS, { .vop_access = zfs_access },
5559 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5560 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5561 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5562 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5563 VOPNAME_FID, { .vop_fid = zfs_fid },
5564 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5565 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock },
5566 VOPNAME_SPACE, { .vop_space = zfs_space },
5567 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage },
5568 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage },
5569 VOPNAME_MAP, { .vop_map = zfs_map },
5570 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap },
5571 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap },
5572 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5573 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5574 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5575 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5576 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf },
5577 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf },
5582 * Symbolic link vnode operations template
5584 vnodeops_t *zfs_symvnodeops;
5585 const fs_operation_def_t zfs_symvnodeops_template[] = {
5586 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5587 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5588 VOPNAME_ACCESS, { .vop_access = zfs_access },
5589 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5590 VOPNAME_READLINK, { .vop_readlink = zfs_readlink },
5591 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5592 VOPNAME_FID, { .vop_fid = zfs_fid },
5593 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5594 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5599 * special share hidden files vnode operations template
5601 vnodeops_t *zfs_sharevnodeops;
5602 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5603 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5604 VOPNAME_ACCESS, { .vop_access = zfs_access },
5605 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5606 VOPNAME_FID, { .vop_fid = zfs_fid },
5607 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5608 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5609 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5610 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5615 * Extended attribute directory vnode operations template
5617 * This template is identical to the directory vnodes
5618 * operation template except for restricted operations:
5622 * Note that there are other restrictions embedded in:
5623 * zfs_create() - restrict type to VREG
5624 * zfs_link() - no links into/out of attribute space
5625 * zfs_rename() - no moves into/out of attribute space
5627 vnodeops_t *zfs_xdvnodeops;
5628 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5629 VOPNAME_OPEN, { .vop_open = zfs_open },
5630 VOPNAME_CLOSE, { .vop_close = zfs_close },
5631 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5632 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5633 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5634 VOPNAME_ACCESS, { .vop_access = zfs_access },
5635 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5636 VOPNAME_CREATE, { .vop_create = zfs_create },
5637 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5638 VOPNAME_LINK, { .vop_link = zfs_link },
5639 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5640 VOPNAME_MKDIR, { .error = zfs_inval },
5641 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5642 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5643 VOPNAME_SYMLINK, { .error = zfs_inval },
5644 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5645 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5646 VOPNAME_FID, { .vop_fid = zfs_fid },
5647 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5648 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5649 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5650 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5651 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5656 * Error vnode operations template
5658 vnodeops_t *zfs_evnodeops;
5659 const fs_operation_def_t zfs_evnodeops_template[] = {
5660 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5661 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5667 ioflags(int ioflags)
5671 if (ioflags & IO_APPEND)
5673 if (ioflags & IO_NDELAY)
5675 if (ioflags & IO_SYNC)
5676 flags |= (FSYNC | FDSYNC | FRSYNC);
5682 zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage)
5684 znode_t *zp = VTOZ(vp);
5685 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5686 objset_t *os = zp->z_zfsvfs->z_os;
5687 vm_page_t mfirst, mlast, mreq;
5691 off_t startoff, endoff;
5693 vm_pindex_t reqstart, reqend;
5694 int pcount, lsize, reqsize, size;
5699 pcount = OFF_TO_IDX(round_page(count));
5701 object = mreq->object;
5704 KASSERT(vp->v_object == object, ("mismatching object"));
5706 if (pcount > 1 && zp->z_blksz > PAGESIZE) {
5707 startoff = rounddown(IDX_TO_OFF(mreq->pindex), zp->z_blksz);
5708 reqstart = OFF_TO_IDX(round_page(startoff));
5709 if (reqstart < m[0]->pindex)
5712 reqstart = reqstart - m[0]->pindex;
5713 endoff = roundup(IDX_TO_OFF(mreq->pindex) + PAGE_SIZE,
5715 reqend = OFF_TO_IDX(trunc_page(endoff)) - 1;
5716 if (reqend > m[pcount - 1]->pindex)
5717 reqend = m[pcount - 1]->pindex;
5718 reqsize = reqend - m[reqstart]->pindex + 1;
5719 KASSERT(reqstart <= reqpage && reqpage < reqstart + reqsize,
5720 ("reqpage beyond [reqstart, reqstart + reqsize[ bounds"));
5725 mfirst = m[reqstart];
5726 mlast = m[reqstart + reqsize - 1];
5728 zfs_vmobject_wlock(object);
5730 for (i = 0; i < reqstart; i++) {
5733 vm_page_unlock(m[i]);
5735 for (i = reqstart + reqsize; i < pcount; i++) {
5738 vm_page_unlock(m[i]);
5741 if (mreq->valid && reqsize == 1) {
5742 if (mreq->valid != VM_PAGE_BITS_ALL)
5743 vm_page_zero_invalid(mreq, TRUE);
5744 zfs_vmobject_wunlock(object);
5746 return (zfs_vm_pagerret_ok);
5749 PCPU_INC(cnt.v_vnodein);
5750 PCPU_ADD(cnt.v_vnodepgsin, reqsize);
5752 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) {
5753 for (i = reqstart; i < reqstart + reqsize; i++) {
5757 vm_page_unlock(m[i]);
5760 zfs_vmobject_wunlock(object);
5762 return (zfs_vm_pagerret_bad);
5766 if (IDX_TO_OFF(mlast->pindex) + lsize > object->un_pager.vnp.vnp_size)
5767 lsize = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mlast->pindex);
5769 zfs_vmobject_wunlock(object);
5771 for (i = reqstart; i < reqstart + reqsize; i++) {
5773 if (i == (reqstart + reqsize - 1))
5775 va = zfs_map_page(m[i], &sf);
5776 error = dmu_read(os, zp->z_id, IDX_TO_OFF(m[i]->pindex),
5777 size, va, DMU_READ_PREFETCH);
5778 if (size != PAGE_SIZE)
5779 bzero(va + size, PAGE_SIZE - size);
5785 zfs_vmobject_wlock(object);
5787 for (i = reqstart; i < reqstart + reqsize; i++) {
5789 m[i]->valid = VM_PAGE_BITS_ALL;
5790 KASSERT(m[i]->dirty == 0, ("zfs_getpages: page %p is dirty", m[i]));
5792 vm_page_readahead_finish(m[i]);
5795 zfs_vmobject_wunlock(object);
5797 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
5799 return (error ? zfs_vm_pagerret_error : zfs_vm_pagerret_ok);
5803 zfs_freebsd_getpages(ap)
5804 struct vop_getpages_args /* {
5809 vm_ooffset_t a_offset;
5813 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage));
5817 zfs_putpages(struct vnode *vp, vm_page_t *ma, size_t len, int flags,
5820 znode_t *zp = VTOZ(vp);
5821 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5830 vm_ooffset_t lo_off;
5841 object = vp->v_object;
5845 KASSERT(ma[0]->object == object, ("mismatching object"));
5846 KASSERT(len > 0 && (len & PAGE_MASK) == 0, ("unexpected length"));
5848 for (i = 0; i < pcount; i++)
5849 rtvals[i] = zfs_vm_pagerret_error;
5851 off = IDX_TO_OFF(ma[0]->pindex);
5852 blksz = zp->z_blksz;
5853 lo_off = rounddown(off, blksz);
5854 lo_len = roundup(len + (off - lo_off), blksz);
5855 rl = zfs_range_lock(zp, lo_off, lo_len, RL_WRITER);
5857 zfs_vmobject_wlock(object);
5858 if (len + off > object->un_pager.vnp.vnp_size) {
5859 if (object->un_pager.vnp.vnp_size > off) {
5862 len = object->un_pager.vnp.vnp_size - off;
5864 if ((pgoff = (int)len & PAGE_MASK) != 0) {
5866 * If the object is locked and the following
5867 * conditions hold, then the page's dirty
5868 * field cannot be concurrently changed by a
5872 vm_page_assert_sbusied(m);
5873 KASSERT(!pmap_page_is_write_mapped(m),
5874 ("zfs_putpages: page %p is not read-only", m));
5875 vm_page_clear_dirty(m, pgoff, PAGE_SIZE -
5882 if (ncount < pcount) {
5883 for (i = ncount; i < pcount; i++) {
5884 rtvals[i] = zfs_vm_pagerret_bad;
5888 zfs_vmobject_wunlock(object);
5893 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
5894 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
5899 tx = dmu_tx_create(zfsvfs->z_os);
5900 dmu_tx_hold_write(tx, zp->z_id, off, len);
5902 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
5903 zfs_sa_upgrade_txholds(tx, zp);
5904 err = dmu_tx_assign(tx, TXG_NOWAIT);
5906 if (err == ERESTART) {
5915 if (zp->z_blksz < PAGE_SIZE) {
5917 for (i = 0; len > 0; off += tocopy, len -= tocopy, i++) {
5918 tocopy = len > PAGE_SIZE ? PAGE_SIZE : len;
5919 va = zfs_map_page(ma[i], &sf);
5920 dmu_write(zfsvfs->z_os, zp->z_id, off, tocopy, va, tx);
5924 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, ma, tx);
5928 uint64_t mtime[2], ctime[2];
5929 sa_bulk_attr_t bulk[3];
5932 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
5934 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
5936 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
5938 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
5940 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
5942 zfs_vmobject_wlock(object);
5943 for (i = 0; i < ncount; i++) {
5944 rtvals[i] = zfs_vm_pagerret_ok;
5945 vm_page_undirty(ma[i]);
5947 zfs_vmobject_wunlock(object);
5948 PCPU_INC(cnt.v_vnodeout);
5949 PCPU_ADD(cnt.v_vnodepgsout, ncount);
5954 zfs_range_unlock(rl);
5955 if ((flags & (zfs_vm_pagerput_sync | zfs_vm_pagerput_inval)) != 0 ||
5956 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5957 zil_commit(zfsvfs->z_log, zp->z_id);
5963 zfs_freebsd_putpages(ap)
5964 struct vop_putpages_args /* {
5970 vm_ooffset_t a_offset;
5974 return (zfs_putpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_sync,
5979 zfs_freebsd_bmap(ap)
5980 struct vop_bmap_args /* {
5983 struct bufobj **a_bop;
5990 if (ap->a_bop != NULL)
5991 *ap->a_bop = &ap->a_vp->v_bufobj;
5992 if (ap->a_bnp != NULL)
5993 *ap->a_bnp = ap->a_bn;
5994 if (ap->a_runp != NULL)
5996 if (ap->a_runb != NULL)
6003 zfs_freebsd_open(ap)
6004 struct vop_open_args /* {
6007 struct ucred *a_cred;
6008 struct thread *a_td;
6011 vnode_t *vp = ap->a_vp;
6012 znode_t *zp = VTOZ(vp);
6015 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL);
6017 vnode_create_vobject(vp, zp->z_size, ap->a_td);
6022 zfs_freebsd_close(ap)
6023 struct vop_close_args /* {
6026 struct ucred *a_cred;
6027 struct thread *a_td;
6031 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL));
6035 zfs_freebsd_ioctl(ap)
6036 struct vop_ioctl_args /* {
6046 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data,
6047 ap->a_fflag, ap->a_cred, NULL, NULL));
6051 zfs_freebsd_read(ap)
6052 struct vop_read_args /* {
6056 struct ucred *a_cred;
6060 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
6065 zfs_freebsd_write(ap)
6066 struct vop_write_args /* {
6070 struct ucred *a_cred;
6074 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
6079 zfs_freebsd_access(ap)
6080 struct vop_access_args /* {
6082 accmode_t a_accmode;
6083 struct ucred *a_cred;
6084 struct thread *a_td;
6087 vnode_t *vp = ap->a_vp;
6088 znode_t *zp = VTOZ(vp);
6093 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND,
6095 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND);
6097 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL);
6100 * VADMIN has to be handled by vaccess().
6103 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND);
6105 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
6106 zp->z_gid, accmode, ap->a_cred, NULL);
6111 * For VEXEC, ensure that at least one execute bit is set for
6114 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR &&
6115 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
6123 zfs_freebsd_lookup(ap)
6124 struct vop_lookup_args /* {
6125 struct vnode *a_dvp;
6126 struct vnode **a_vpp;
6127 struct componentname *a_cnp;
6130 struct componentname *cnp = ap->a_cnp;
6131 char nm[NAME_MAX + 1];
6133 ASSERT(cnp->cn_namelen < sizeof(nm));
6134 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm)));
6136 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop,
6137 cnp->cn_cred, cnp->cn_thread, 0));
6141 zfs_freebsd_create(ap)
6142 struct vop_create_args /* {
6143 struct vnode *a_dvp;
6144 struct vnode **a_vpp;
6145 struct componentname *a_cnp;
6146 struct vattr *a_vap;
6149 struct componentname *cnp = ap->a_cnp;
6150 vattr_t *vap = ap->a_vap;
6153 ASSERT(cnp->cn_flags & SAVENAME);
6155 vattr_init_mask(vap);
6156 mode = vap->va_mode & ALLPERMS;
6158 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode,
6159 ap->a_vpp, cnp->cn_cred, cnp->cn_thread));
6163 zfs_freebsd_remove(ap)
6164 struct vop_remove_args /* {
6165 struct vnode *a_dvp;
6167 struct componentname *a_cnp;
6171 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6173 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr,
6174 ap->a_cnp->cn_cred, NULL, 0));
6178 zfs_freebsd_mkdir(ap)
6179 struct vop_mkdir_args /* {
6180 struct vnode *a_dvp;
6181 struct vnode **a_vpp;
6182 struct componentname *a_cnp;
6183 struct vattr *a_vap;
6186 vattr_t *vap = ap->a_vap;
6188 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6190 vattr_init_mask(vap);
6192 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp,
6193 ap->a_cnp->cn_cred, NULL, 0, NULL));
6197 zfs_freebsd_rmdir(ap)
6198 struct vop_rmdir_args /* {
6199 struct vnode *a_dvp;
6201 struct componentname *a_cnp;
6204 struct componentname *cnp = ap->a_cnp;
6206 ASSERT(cnp->cn_flags & SAVENAME);
6208 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0));
6212 zfs_freebsd_readdir(ap)
6213 struct vop_readdir_args /* {
6216 struct ucred *a_cred;
6223 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag,
6224 ap->a_ncookies, ap->a_cookies));
6228 zfs_freebsd_fsync(ap)
6229 struct vop_fsync_args /* {
6232 struct thread *a_td;
6237 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL));
6241 zfs_freebsd_getattr(ap)
6242 struct vop_getattr_args /* {
6244 struct vattr *a_vap;
6245 struct ucred *a_cred;
6248 vattr_t *vap = ap->a_vap;
6254 xvap.xva_vattr = *vap;
6255 xvap.xva_vattr.va_mask |= AT_XVATTR;
6257 /* Convert chflags into ZFS-type flags. */
6258 /* XXX: what about SF_SETTABLE?. */
6259 XVA_SET_REQ(&xvap, XAT_IMMUTABLE);
6260 XVA_SET_REQ(&xvap, XAT_APPENDONLY);
6261 XVA_SET_REQ(&xvap, XAT_NOUNLINK);
6262 XVA_SET_REQ(&xvap, XAT_NODUMP);
6263 XVA_SET_REQ(&xvap, XAT_READONLY);
6264 XVA_SET_REQ(&xvap, XAT_ARCHIVE);
6265 XVA_SET_REQ(&xvap, XAT_SYSTEM);
6266 XVA_SET_REQ(&xvap, XAT_HIDDEN);
6267 XVA_SET_REQ(&xvap, XAT_REPARSE);
6268 XVA_SET_REQ(&xvap, XAT_OFFLINE);
6269 XVA_SET_REQ(&xvap, XAT_SPARSE);
6271 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL);
6275 /* Convert ZFS xattr into chflags. */
6276 #define FLAG_CHECK(fflag, xflag, xfield) do { \
6277 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \
6278 fflags |= (fflag); \
6280 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE,
6281 xvap.xva_xoptattrs.xoa_immutable);
6282 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY,
6283 xvap.xva_xoptattrs.xoa_appendonly);
6284 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK,
6285 xvap.xva_xoptattrs.xoa_nounlink);
6286 FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE,
6287 xvap.xva_xoptattrs.xoa_archive);
6288 FLAG_CHECK(UF_NODUMP, XAT_NODUMP,
6289 xvap.xva_xoptattrs.xoa_nodump);
6290 FLAG_CHECK(UF_READONLY, XAT_READONLY,
6291 xvap.xva_xoptattrs.xoa_readonly);
6292 FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM,
6293 xvap.xva_xoptattrs.xoa_system);
6294 FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN,
6295 xvap.xva_xoptattrs.xoa_hidden);
6296 FLAG_CHECK(UF_REPARSE, XAT_REPARSE,
6297 xvap.xva_xoptattrs.xoa_reparse);
6298 FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE,
6299 xvap.xva_xoptattrs.xoa_offline);
6300 FLAG_CHECK(UF_SPARSE, XAT_SPARSE,
6301 xvap.xva_xoptattrs.xoa_sparse);
6304 *vap = xvap.xva_vattr;
6305 vap->va_flags = fflags;
6310 zfs_freebsd_setattr(ap)
6311 struct vop_setattr_args /* {
6313 struct vattr *a_vap;
6314 struct ucred *a_cred;
6317 vnode_t *vp = ap->a_vp;
6318 vattr_t *vap = ap->a_vap;
6319 cred_t *cred = ap->a_cred;
6324 vattr_init_mask(vap);
6325 vap->va_mask &= ~AT_NOSET;
6328 xvap.xva_vattr = *vap;
6330 zflags = VTOZ(vp)->z_pflags;
6332 if (vap->va_flags != VNOVAL) {
6333 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs;
6336 if (zfsvfs->z_use_fuids == B_FALSE)
6337 return (EOPNOTSUPP);
6339 fflags = vap->va_flags;
6342 * We need to figure out whether it makes sense to allow
6343 * UF_REPARSE through, since we don't really have other
6344 * facilities to handle reparse points and zfs_setattr()
6345 * doesn't currently allow setting that attribute anyway.
6347 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE|
6348 UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE|
6349 UF_OFFLINE|UF_SPARSE)) != 0)
6350 return (EOPNOTSUPP);
6352 * Unprivileged processes are not permitted to unset system
6353 * flags, or modify flags if any system flags are set.
6354 * Privileged non-jail processes may not modify system flags
6355 * if securelevel > 0 and any existing system flags are set.
6356 * Privileged jail processes behave like privileged non-jail
6357 * processes if the security.jail.chflags_allowed sysctl is
6358 * is non-zero; otherwise, they behave like unprivileged
6361 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 ||
6362 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) {
6364 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6365 error = securelevel_gt(cred, 0);
6371 * Callers may only modify the file flags on objects they
6372 * have VADMIN rights for.
6374 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0)
6377 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6381 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) {
6386 #define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \
6387 if (((fflags & (fflag)) && !(zflags & (zflag))) || \
6388 ((zflags & (zflag)) && !(fflags & (fflag)))) { \
6389 XVA_SET_REQ(&xvap, (xflag)); \
6390 (xfield) = ((fflags & (fflag)) != 0); \
6393 /* Convert chflags into ZFS-type flags. */
6394 /* XXX: what about SF_SETTABLE?. */
6395 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE,
6396 xvap.xva_xoptattrs.xoa_immutable);
6397 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY,
6398 xvap.xva_xoptattrs.xoa_appendonly);
6399 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK,
6400 xvap.xva_xoptattrs.xoa_nounlink);
6401 FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE,
6402 xvap.xva_xoptattrs.xoa_archive);
6403 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP,
6404 xvap.xva_xoptattrs.xoa_nodump);
6405 FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY,
6406 xvap.xva_xoptattrs.xoa_readonly);
6407 FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM,
6408 xvap.xva_xoptattrs.xoa_system);
6409 FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN,
6410 xvap.xva_xoptattrs.xoa_hidden);
6411 FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE,
6412 xvap.xva_xoptattrs.xoa_hidden);
6413 FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE,
6414 xvap.xva_xoptattrs.xoa_offline);
6415 FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE,
6416 xvap.xva_xoptattrs.xoa_sparse);
6419 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL));
6423 zfs_freebsd_rename(ap)
6424 struct vop_rename_args /* {
6425 struct vnode *a_fdvp;
6426 struct vnode *a_fvp;
6427 struct componentname *a_fcnp;
6428 struct vnode *a_tdvp;
6429 struct vnode *a_tvp;
6430 struct componentname *a_tcnp;
6433 vnode_t *fdvp = ap->a_fdvp;
6434 vnode_t *fvp = ap->a_fvp;
6435 vnode_t *tdvp = ap->a_tdvp;
6436 vnode_t *tvp = ap->a_tvp;
6439 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART));
6440 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART));
6443 * Check for cross-device rename.
6445 if ((fdvp->v_mount != tdvp->v_mount) ||
6446 (tvp && (fdvp->v_mount != tvp->v_mount)))
6449 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp,
6450 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0);
6464 zfs_freebsd_symlink(ap)
6465 struct vop_symlink_args /* {
6466 struct vnode *a_dvp;
6467 struct vnode **a_vpp;
6468 struct componentname *a_cnp;
6469 struct vattr *a_vap;
6473 struct componentname *cnp = ap->a_cnp;
6474 vattr_t *vap = ap->a_vap;
6476 ASSERT(cnp->cn_flags & SAVENAME);
6478 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */
6479 vattr_init_mask(vap);
6481 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap,
6482 ap->a_target, cnp->cn_cred, cnp->cn_thread));
6486 zfs_freebsd_readlink(ap)
6487 struct vop_readlink_args /* {
6490 struct ucred *a_cred;
6494 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL));
6498 zfs_freebsd_link(ap)
6499 struct vop_link_args /* {
6500 struct vnode *a_tdvp;
6502 struct componentname *a_cnp;
6505 struct componentname *cnp = ap->a_cnp;
6506 vnode_t *vp = ap->a_vp;
6507 vnode_t *tdvp = ap->a_tdvp;
6509 if (tdvp->v_mount != vp->v_mount)
6512 ASSERT(cnp->cn_flags & SAVENAME);
6514 return (zfs_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0));
6518 zfs_freebsd_inactive(ap)
6519 struct vop_inactive_args /* {
6521 struct thread *a_td;
6524 vnode_t *vp = ap->a_vp;
6526 zfs_inactive(vp, ap->a_td->td_ucred, NULL);
6531 zfs_freebsd_reclaim(ap)
6532 struct vop_reclaim_args /* {
6534 struct thread *a_td;
6537 vnode_t *vp = ap->a_vp;
6538 znode_t *zp = VTOZ(vp);
6539 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
6543 /* Destroy the vm object and flush associated pages. */
6544 vnode_destroy_vobject(vp);
6547 * z_teardown_inactive_lock protects from a race with
6548 * zfs_znode_dmu_fini in zfsvfs_teardown during
6551 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
6552 if (zp->z_sa_hdl == NULL)
6556 rw_exit(&zfsvfs->z_teardown_inactive_lock);
6564 struct vop_fid_args /* {
6570 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL));
6574 zfs_freebsd_pathconf(ap)
6575 struct vop_pathconf_args /* {
6578 register_t *a_retval;
6584 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL);
6586 *ap->a_retval = val;
6587 else if (error == EOPNOTSUPP)
6588 error = vop_stdpathconf(ap);
6593 zfs_freebsd_fifo_pathconf(ap)
6594 struct vop_pathconf_args /* {
6597 register_t *a_retval;
6601 switch (ap->a_name) {
6602 case _PC_ACL_EXTENDED:
6604 case _PC_ACL_PATH_MAX:
6605 case _PC_MAC_PRESENT:
6606 return (zfs_freebsd_pathconf(ap));
6608 return (fifo_specops.vop_pathconf(ap));
6613 * FreeBSD's extended attributes namespace defines file name prefix for ZFS'
6614 * extended attribute name:
6617 * system freebsd:system:
6618 * user (none, can be used to access ZFS fsattr(5) attributes
6619 * created on Solaris)
6622 zfs_create_attrname(int attrnamespace, const char *name, char *attrname,
6625 const char *namespace, *prefix, *suffix;
6627 /* We don't allow '/' character in attribute name. */
6628 if (strchr(name, '/') != NULL)
6630 /* We don't allow attribute names that start with "freebsd:" string. */
6631 if (strncmp(name, "freebsd:", 8) == 0)
6634 bzero(attrname, size);
6636 switch (attrnamespace) {
6637 case EXTATTR_NAMESPACE_USER:
6639 prefix = "freebsd:";
6640 namespace = EXTATTR_NAMESPACE_USER_STRING;
6644 * This is the default namespace by which we can access all
6645 * attributes created on Solaris.
6647 prefix = namespace = suffix = "";
6650 case EXTATTR_NAMESPACE_SYSTEM:
6651 prefix = "freebsd:";
6652 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING;
6655 case EXTATTR_NAMESPACE_EMPTY:
6659 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix,
6661 return (ENAMETOOLONG);
6667 * Vnode operating to retrieve a named extended attribute.
6670 zfs_getextattr(struct vop_getextattr_args *ap)
6673 IN struct vnode *a_vp;
6674 IN int a_attrnamespace;
6675 IN const char *a_name;
6676 INOUT struct uio *a_uio;
6678 IN struct ucred *a_cred;
6679 IN struct thread *a_td;
6683 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6684 struct thread *td = ap->a_td;
6685 struct nameidata nd;
6688 vnode_t *xvp = NULL, *vp;
6691 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6692 ap->a_cred, ap->a_td, VREAD);
6696 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6703 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6711 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6713 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL);
6715 NDFREE(&nd, NDF_ONLY_PNBUF);
6718 if (error == ENOENT)
6723 if (ap->a_size != NULL) {
6724 error = VOP_GETATTR(vp, &va, ap->a_cred);
6726 *ap->a_size = (size_t)va.va_size;
6727 } else if (ap->a_uio != NULL)
6728 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6731 vn_close(vp, flags, ap->a_cred, td);
6738 * Vnode operation to remove a named attribute.
6741 zfs_deleteextattr(struct vop_deleteextattr_args *ap)
6744 IN struct vnode *a_vp;
6745 IN int a_attrnamespace;
6746 IN const char *a_name;
6747 IN struct ucred *a_cred;
6748 IN struct thread *a_td;
6752 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6753 struct thread *td = ap->a_td;
6754 struct nameidata nd;
6757 vnode_t *xvp = NULL, *vp;
6760 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6761 ap->a_cred, ap->a_td, VWRITE);
6765 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6772 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6779 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF,
6780 UIO_SYSSPACE, attrname, xvp, td);
6785 NDFREE(&nd, NDF_ONLY_PNBUF);
6786 if (error == ENOENT)
6791 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
6792 NDFREE(&nd, NDF_ONLY_PNBUF);
6795 if (vp == nd.ni_dvp)
6805 * Vnode operation to set a named attribute.
6808 zfs_setextattr(struct vop_setextattr_args *ap)
6811 IN struct vnode *a_vp;
6812 IN int a_attrnamespace;
6813 IN const char *a_name;
6814 INOUT struct uio *a_uio;
6815 IN struct ucred *a_cred;
6816 IN struct thread *a_td;
6820 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6821 struct thread *td = ap->a_td;
6822 struct nameidata nd;
6825 vnode_t *xvp = NULL, *vp;
6828 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6829 ap->a_cred, ap->a_td, VWRITE);
6833 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6840 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6841 LOOKUP_XATTR | CREATE_XATTR_DIR);
6847 flags = FFLAGS(O_WRONLY | O_CREAT);
6848 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6850 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL);
6852 NDFREE(&nd, NDF_ONLY_PNBUF);
6860 error = VOP_SETATTR(vp, &va, ap->a_cred);
6862 VOP_WRITE(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6865 vn_close(vp, flags, ap->a_cred, td);
6872 * Vnode operation to retrieve extended attributes on a vnode.
6875 zfs_listextattr(struct vop_listextattr_args *ap)
6878 IN struct vnode *a_vp;
6879 IN int a_attrnamespace;
6880 INOUT struct uio *a_uio;
6882 IN struct ucred *a_cred;
6883 IN struct thread *a_td;
6887 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6888 struct thread *td = ap->a_td;
6889 struct nameidata nd;
6890 char attrprefix[16];
6891 u_char dirbuf[sizeof(struct dirent)];
6894 struct uio auio, *uio = ap->a_uio;
6895 size_t *sizep = ap->a_size;
6897 vnode_t *xvp = NULL, *vp;
6898 int done, error, eof, pos;
6900 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6901 ap->a_cred, ap->a_td, VREAD);
6905 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix,
6906 sizeof(attrprefix));
6909 plen = strlen(attrprefix);
6916 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6921 * ENOATTR means that the EA directory does not yet exist,
6922 * i.e. there are no extended attributes there.
6924 if (error == ENOATTR)
6929 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED,
6930 UIO_SYSSPACE, ".", xvp, td);
6933 NDFREE(&nd, NDF_ONLY_PNBUF);
6939 auio.uio_iov = &aiov;
6940 auio.uio_iovcnt = 1;
6941 auio.uio_segflg = UIO_SYSSPACE;
6943 auio.uio_rw = UIO_READ;
6944 auio.uio_offset = 0;
6949 aiov.iov_base = (void *)dirbuf;
6950 aiov.iov_len = sizeof(dirbuf);
6951 auio.uio_resid = sizeof(dirbuf);
6952 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL);
6953 done = sizeof(dirbuf) - auio.uio_resid;
6956 for (pos = 0; pos < done;) {
6957 dp = (struct dirent *)(dirbuf + pos);
6958 pos += dp->d_reclen;
6960 * XXX: Temporarily we also accept DT_UNKNOWN, as this
6961 * is what we get when attribute was created on Solaris.
6963 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN)
6965 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0)
6967 else if (strncmp(dp->d_name, attrprefix, plen) != 0)
6969 nlen = dp->d_namlen - plen;
6972 else if (uio != NULL) {
6974 * Format of extattr name entry is one byte for
6975 * length and the rest for name.
6977 error = uiomove(&nlen, 1, uio->uio_rw, uio);
6979 error = uiomove(dp->d_name + plen, nlen,
6986 } while (!eof && error == 0);
6995 zfs_freebsd_getacl(ap)
6996 struct vop_getacl_args /* {
7005 vsecattr_t vsecattr;
7007 if (ap->a_type != ACL_TYPE_NFS4)
7010 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
7011 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL))
7014 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt);
7015 if (vsecattr.vsa_aclentp != NULL)
7016 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz);
7022 zfs_freebsd_setacl(ap)
7023 struct vop_setacl_args /* {
7032 vsecattr_t vsecattr;
7033 int aclbsize; /* size of acl list in bytes */
7036 if (ap->a_type != ACL_TYPE_NFS4)
7039 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES)
7043 * With NFSv4 ACLs, chmod(2) may need to add additional entries,
7044 * splitting every entry into two and appending "canonical six"
7045 * entries at the end. Don't allow for setting an ACL that would
7046 * cause chmod(2) to run out of ACL entries.
7048 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES)
7051 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR);
7055 vsecattr.vsa_mask = VSA_ACE;
7056 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t);
7057 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP);
7058 aaclp = vsecattr.vsa_aclentp;
7059 vsecattr.vsa_aclentsz = aclbsize;
7061 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp);
7062 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL);
7063 kmem_free(aaclp, aclbsize);
7069 zfs_freebsd_aclcheck(ap)
7070 struct vop_aclcheck_args /* {
7079 return (EOPNOTSUPP);
7082 struct vop_vector zfs_vnodeops;
7083 struct vop_vector zfs_fifoops;
7084 struct vop_vector zfs_shareops;
7086 struct vop_vector zfs_vnodeops = {
7087 .vop_default = &default_vnodeops,
7088 .vop_inactive = zfs_freebsd_inactive,
7089 .vop_reclaim = zfs_freebsd_reclaim,
7090 .vop_access = zfs_freebsd_access,
7091 #ifdef FREEBSD_NAMECACHE
7092 .vop_lookup = vfs_cache_lookup,
7093 .vop_cachedlookup = zfs_freebsd_lookup,
7095 .vop_lookup = zfs_freebsd_lookup,
7097 .vop_getattr = zfs_freebsd_getattr,
7098 .vop_setattr = zfs_freebsd_setattr,
7099 .vop_create = zfs_freebsd_create,
7100 .vop_mknod = zfs_freebsd_create,
7101 .vop_mkdir = zfs_freebsd_mkdir,
7102 .vop_readdir = zfs_freebsd_readdir,
7103 .vop_fsync = zfs_freebsd_fsync,
7104 .vop_open = zfs_freebsd_open,
7105 .vop_close = zfs_freebsd_close,
7106 .vop_rmdir = zfs_freebsd_rmdir,
7107 .vop_ioctl = zfs_freebsd_ioctl,
7108 .vop_link = zfs_freebsd_link,
7109 .vop_symlink = zfs_freebsd_symlink,
7110 .vop_readlink = zfs_freebsd_readlink,
7111 .vop_read = zfs_freebsd_read,
7112 .vop_write = zfs_freebsd_write,
7113 .vop_remove = zfs_freebsd_remove,
7114 .vop_rename = zfs_freebsd_rename,
7115 .vop_pathconf = zfs_freebsd_pathconf,
7116 .vop_bmap = zfs_freebsd_bmap,
7117 .vop_fid = zfs_freebsd_fid,
7118 .vop_getextattr = zfs_getextattr,
7119 .vop_deleteextattr = zfs_deleteextattr,
7120 .vop_setextattr = zfs_setextattr,
7121 .vop_listextattr = zfs_listextattr,
7122 .vop_getacl = zfs_freebsd_getacl,
7123 .vop_setacl = zfs_freebsd_setacl,
7124 .vop_aclcheck = zfs_freebsd_aclcheck,
7125 .vop_getpages = zfs_freebsd_getpages,
7126 .vop_putpages = zfs_freebsd_putpages,
7129 struct vop_vector zfs_fifoops = {
7130 .vop_default = &fifo_specops,
7131 .vop_fsync = zfs_freebsd_fsync,
7132 .vop_access = zfs_freebsd_access,
7133 .vop_getattr = zfs_freebsd_getattr,
7134 .vop_inactive = zfs_freebsd_inactive,
7135 .vop_read = VOP_PANIC,
7136 .vop_reclaim = zfs_freebsd_reclaim,
7137 .vop_setattr = zfs_freebsd_setattr,
7138 .vop_write = VOP_PANIC,
7139 .vop_pathconf = zfs_freebsd_fifo_pathconf,
7140 .vop_fid = zfs_freebsd_fid,
7141 .vop_getacl = zfs_freebsd_getacl,
7142 .vop_setacl = zfs_freebsd_setacl,
7143 .vop_aclcheck = zfs_freebsd_aclcheck,
7147 * special share hidden files vnode operations template
7149 struct vop_vector zfs_shareops = {
7150 .vop_default = &default_vnodeops,
7151 .vop_access = zfs_freebsd_access,
7152 .vop_inactive = zfs_freebsd_inactive,
7153 .vop_reclaim = zfs_freebsd_reclaim,
7154 .vop_fid = zfs_freebsd_fid,
7155 .vop_pathconf = zfs_freebsd_pathconf,