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]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
25 * Copyright (c) 2014 Integros [integros.com]
26 * Copyright 2017 Nexenta Systems, Inc.
29 /* Portions Copyright 2007 Jeremy Teo */
30 /* Portions Copyright 2010 Robert Milkowski */
32 #include <sys/types.h>
33 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/sysmacros.h>
37 #include <sys/resource.h>
40 #include <sys/vnode.h>
44 #include <sys/taskq.h>
46 #include <sys/atomic.h>
47 #include <sys/namei.h>
49 #include <sys/cmn_err.h>
50 #include <sys/errno.h>
51 #include <sys/unistd.h>
52 #include <sys/zfs_dir.h>
53 #include <sys/zfs_ioctl.h>
54 #include <sys/fs/zfs.h>
56 #include <sys/dmu_objset.h>
62 #include <sys/dirent.h>
63 #include <sys/policy.h>
64 #include <sys/sunddi.h>
65 #include <sys/filio.h>
67 #include <sys/zfs_ctldir.h>
68 #include <sys/zfs_fuid.h>
69 #include <sys/zfs_sa.h>
70 #include <sys/zfs_rlock.h>
71 #include <sys/extdirent.h>
72 #include <sys/kidmap.h>
75 #include <sys/sched.h>
77 #include <sys/vmmeter.h>
78 #include <vm/vm_param.h>
83 * Each vnode op performs some logical unit of work. To do this, the ZPL must
84 * properly lock its in-core state, create a DMU transaction, do the work,
85 * record this work in the intent log (ZIL), commit the DMU transaction,
86 * and wait for the intent log to commit if it is a synchronous operation.
87 * Moreover, the vnode ops must work in both normal and log replay context.
88 * The ordering of events is important to avoid deadlocks and references
89 * to freed memory. The example below illustrates the following Big Rules:
91 * (1) A check must be made in each zfs thread for a mounted file system.
92 * This is done avoiding races using ZFS_ENTER(zfsvfs).
93 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
94 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
95 * can return EIO from the calling function.
97 * (2) VN_RELE() should always be the last thing except for zil_commit()
98 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
99 * First, if it's the last reference, the vnode/znode
100 * can be freed, so the zp may point to freed memory. Second, the last
101 * reference will call zfs_zinactive(), which may induce a lot of work --
102 * pushing cached pages (which acquires range locks) and syncing out
103 * cached atime changes. Third, zfs_zinactive() may require a new tx,
104 * which could deadlock the system if you were already holding one.
105 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
107 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
108 * as they can span dmu_tx_assign() calls.
110 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
111 * dmu_tx_assign(). This is critical because we don't want to block
112 * while holding locks.
114 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
115 * reduces lock contention and CPU usage when we must wait (note that if
116 * throughput is constrained by the storage, nearly every transaction
119 * Note, in particular, that if a lock is sometimes acquired before
120 * the tx assigns, and sometimes after (e.g. z_lock), then failing
121 * to use a non-blocking assign can deadlock the system. The scenario:
123 * Thread A has grabbed a lock before calling dmu_tx_assign().
124 * Thread B is in an already-assigned tx, and blocks for this lock.
125 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
126 * forever, because the previous txg can't quiesce until B's tx commits.
128 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
129 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
130 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
131 * to indicate that this operation has already called dmu_tx_wait().
132 * This will ensure that we don't retry forever, waiting a short bit
135 * (5) If the operation succeeded, generate the intent log entry for it
136 * before dropping locks. This ensures that the ordering of events
137 * in the intent log matches the order in which they actually occurred.
138 * During ZIL replay the zfs_log_* functions will update the sequence
139 * number to indicate the zil transaction has replayed.
141 * (6) At the end of each vnode op, the DMU tx must always commit,
142 * regardless of whether there were any errors.
144 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
145 * to ensure that synchronous semantics are provided when necessary.
147 * In general, this is how things should be ordered in each vnode op:
149 * ZFS_ENTER(zfsvfs); // exit if unmounted
151 * zfs_dirent_lookup(&dl, ...) // lock directory entry (may VN_HOLD())
152 * rw_enter(...); // grab any other locks you need
153 * tx = dmu_tx_create(...); // get DMU tx
154 * dmu_tx_hold_*(); // hold each object you might modify
155 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
157 * rw_exit(...); // drop locks
158 * zfs_dirent_unlock(dl); // unlock directory entry
159 * VN_RELE(...); // release held vnodes
160 * if (error == ERESTART) {
166 * dmu_tx_abort(tx); // abort DMU tx
167 * ZFS_EXIT(zfsvfs); // finished in zfs
168 * return (error); // really out of space
170 * error = do_real_work(); // do whatever this VOP does
172 * zfs_log_*(...); // on success, make ZIL entry
173 * dmu_tx_commit(tx); // commit DMU tx -- error or not
174 * rw_exit(...); // drop locks
175 * zfs_dirent_unlock(dl); // unlock directory entry
176 * VN_RELE(...); // release held vnodes
177 * zil_commit(zilog, foid); // synchronous when necessary
178 * ZFS_EXIT(zfsvfs); // finished in zfs
179 * return (error); // done, report error
184 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
186 znode_t *zp = VTOZ(*vpp);
187 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
192 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
193 ((flag & FAPPEND) == 0)) {
195 return (SET_ERROR(EPERM));
198 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
199 ZTOV(zp)->v_type == VREG &&
200 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
201 if (fs_vscan(*vpp, cr, 0) != 0) {
203 return (SET_ERROR(EACCES));
207 /* Keep a count of the synchronous opens in the znode */
208 if (flag & (FSYNC | FDSYNC))
209 atomic_inc_32(&zp->z_sync_cnt);
217 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
218 caller_context_t *ct)
220 znode_t *zp = VTOZ(vp);
221 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
224 * Clean up any locks held by this process on the vp.
226 cleanlocks(vp, ddi_get_pid(), 0);
227 cleanshares(vp, ddi_get_pid());
232 /* Decrement the synchronous opens in the znode */
233 if ((flag & (FSYNC | FDSYNC)) && (count == 1))
234 atomic_dec_32(&zp->z_sync_cnt);
236 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
237 ZTOV(zp)->v_type == VREG &&
238 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
239 VERIFY(fs_vscan(vp, cr, 1) == 0);
246 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
247 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
250 zfs_holey(vnode_t *vp, u_long cmd, offset_t *off)
252 znode_t *zp = VTOZ(vp);
253 uint64_t noff = (uint64_t)*off; /* new offset */
258 file_sz = zp->z_size;
259 if (noff >= file_sz) {
260 return (SET_ERROR(ENXIO));
263 if (cmd == _FIO_SEEK_HOLE)
268 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
271 return (SET_ERROR(ENXIO));
274 * We could find a hole that begins after the logical end-of-file,
275 * because dmu_offset_next() only works on whole blocks. If the
276 * EOF falls mid-block, then indicate that the "virtual hole"
277 * at the end of the file begins at the logical EOF, rather than
278 * at the end of the last block.
280 if (noff > file_sz) {
293 zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred,
294 int *rvalp, caller_context_t *ct)
298 dmu_object_info_t doi;
309 * The following two ioctls are used by bfu. Faking out,
310 * necessary to avoid bfu errors.
323 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
324 return (SET_ERROR(EFAULT));
326 off = *(offset_t *)data;
329 zfsvfs = zp->z_zfsvfs;
333 /* offset parameter is in/out */
334 error = zfs_holey(vp, com, &off);
339 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
340 return (SET_ERROR(EFAULT));
342 *(offset_t *)data = off;
347 case _FIO_COUNT_FILLED:
350 * _FIO_COUNT_FILLED adds a new ioctl command which
351 * exposes the number of filled blocks in a
355 zfsvfs = zp->z_zfsvfs;
360 * Wait for all dirty blocks for this object
361 * to get synced out to disk, and the DMU info
364 error = dmu_object_wait_synced(zfsvfs->z_os, zp->z_id);
371 * Retrieve fill count from DMU object.
373 error = dmu_object_info(zfsvfs->z_os, zp->z_id, &doi);
379 ndata = doi.doi_fill_count;
382 if (ddi_copyout(&ndata, (void *)data, sizeof (ndata), flag))
383 return (SET_ERROR(EFAULT));
388 return (SET_ERROR(ENOTTY));
392 page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes)
399 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE
400 * aligned boundaries, if the range is not aligned. As a result a
401 * DEV_BSIZE subrange with partially dirty data may get marked as clean.
402 * It may happen that all DEV_BSIZE subranges are marked clean and thus
403 * the whole page would be considred clean despite have some dirty data.
404 * For this reason we should shrink the range to DEV_BSIZE aligned
405 * boundaries before calling vm_page_clear_dirty.
407 end = rounddown2(off + nbytes, DEV_BSIZE);
408 off = roundup2(off, DEV_BSIZE);
412 zfs_vmobject_assert_wlocked(obj);
415 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
417 if (vm_page_xbusied(pp)) {
419 * Reference the page before unlocking and
420 * sleeping so that the page daemon is less
421 * likely to reclaim it.
423 vm_page_reference(pp);
425 zfs_vmobject_wunlock(obj);
426 vm_page_busy_sleep(pp, "zfsmwb", true);
427 zfs_vmobject_wlock(obj);
431 } else if (pp != NULL) {
437 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
438 vm_object_pip_add(obj, 1);
439 pmap_remove_write(pp);
441 vm_page_clear_dirty(pp, off, nbytes);
449 page_unbusy(vm_page_t pp)
453 vm_object_pip_subtract(pp->object, 1);
457 page_hold(vnode_t *vp, int64_t start)
463 zfs_vmobject_assert_wlocked(obj);
466 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
468 if (vm_page_xbusied(pp)) {
470 * Reference the page before unlocking and
471 * sleeping so that the page daemon is less
472 * likely to reclaim it.
474 vm_page_reference(pp);
476 zfs_vmobject_wunlock(obj);
477 vm_page_busy_sleep(pp, "zfsmwb", true);
478 zfs_vmobject_wlock(obj);
482 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
495 page_unhold(vm_page_t pp)
504 * When a file is memory mapped, we must keep the IO data synchronized
505 * between the DMU cache and the memory mapped pages. What this means:
507 * On Write: If we find a memory mapped page, we write to *both*
508 * the page and the dmu buffer.
511 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid,
512 int segflg, dmu_tx_t *tx)
519 ASSERT(segflg != UIO_NOCOPY);
520 ASSERT(vp->v_mount != NULL);
524 off = start & PAGEOFFSET;
525 zfs_vmobject_wlock(obj);
526 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
528 int nbytes = imin(PAGESIZE - off, len);
530 if ((pp = page_busy(vp, start, off, nbytes)) != NULL) {
531 zfs_vmobject_wunlock(obj);
533 va = zfs_map_page(pp, &sf);
534 (void) dmu_read(os, oid, start+off, nbytes,
535 va+off, DMU_READ_PREFETCH);;
538 zfs_vmobject_wlock(obj);
544 vm_object_pip_wakeupn(obj, 0);
545 zfs_vmobject_wunlock(obj);
549 * Read with UIO_NOCOPY flag means that sendfile(2) requests
550 * ZFS to populate a range of page cache pages with data.
552 * NOTE: this function could be optimized to pre-allocate
553 * all pages in advance, drain exclusive busy on all of them,
554 * map them into contiguous KVA region and populate them
555 * in one single dmu_read() call.
558 mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio)
560 znode_t *zp = VTOZ(vp);
561 objset_t *os = zp->z_zfsvfs->z_os;
571 ASSERT(uio->uio_segflg == UIO_NOCOPY);
572 ASSERT(vp->v_mount != NULL);
575 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0);
577 zfs_vmobject_wlock(obj);
578 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) {
579 int bytes = MIN(PAGESIZE, len);
581 pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_SBUSY |
582 VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY);
583 if (pp->valid == 0) {
584 zfs_vmobject_wunlock(obj);
585 va = zfs_map_page(pp, &sf);
586 error = dmu_read(os, zp->z_id, start, bytes, va,
588 if (bytes != PAGESIZE && error == 0)
589 bzero(va + bytes, PAGESIZE - bytes);
591 zfs_vmobject_wlock(obj);
595 if (pp->wire_count == 0 && pp->valid == 0 &&
599 pp->valid = VM_PAGE_BITS_ALL;
600 vm_page_activate(pp);
604 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
609 uio->uio_resid -= bytes;
610 uio->uio_offset += bytes;
613 zfs_vmobject_wunlock(obj);
618 * When a file is memory mapped, we must keep the IO data synchronized
619 * between the DMU cache and the memory mapped pages. What this means:
621 * On Read: We "read" preferentially from memory mapped pages,
622 * else we default from the dmu buffer.
624 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
625 * the file is memory mapped.
628 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
630 znode_t *zp = VTOZ(vp);
638 ASSERT(vp->v_mount != NULL);
642 start = uio->uio_loffset;
643 off = start & PAGEOFFSET;
644 zfs_vmobject_wlock(obj);
645 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
647 uint64_t bytes = MIN(PAGESIZE - off, len);
649 if (pp = page_hold(vp, start)) {
653 zfs_vmobject_wunlock(obj);
654 va = zfs_map_page(pp, &sf);
656 error = uiomove(va + off, bytes, UIO_READ, uio);
658 error = vn_io_fault_uiomove(va + off, bytes, uio);
661 zfs_vmobject_wlock(obj);
664 zfs_vmobject_wunlock(obj);
665 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
667 zfs_vmobject_wlock(obj);
674 zfs_vmobject_wunlock(obj);
678 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
681 * Read bytes from specified file into supplied buffer.
683 * IN: vp - vnode of file to be read from.
684 * uio - structure supplying read location, range info,
686 * ioflag - SYNC flags; used to provide FRSYNC semantics.
687 * cr - credentials of caller.
688 * ct - caller context
690 * OUT: uio - updated offset and range, buffer filled.
692 * RETURN: 0 on success, error code on failure.
695 * vp - atime updated if byte count > 0
699 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
701 znode_t *zp = VTOZ(vp);
702 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
711 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
713 return (SET_ERROR(EACCES));
717 * Validate file offset
719 if (uio->uio_loffset < (offset_t)0) {
721 return (SET_ERROR(EINVAL));
725 * Fasttrack empty reads
727 if (uio->uio_resid == 0) {
733 * Check for mandatory locks
735 if (MANDMODE(zp->z_mode)) {
736 if (error = chklock(vp, FREAD,
737 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
744 * If we're in FRSYNC mode, sync out this znode before reading it.
747 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
748 zil_commit(zfsvfs->z_log, zp->z_id);
751 * Lock the range against changes.
753 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
756 * If we are reading past end-of-file we can skip
757 * to the end; but we might still need to set atime.
759 if (uio->uio_loffset >= zp->z_size) {
764 ASSERT(uio->uio_loffset < zp->z_size);
765 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
768 if ((uio->uio_extflg == UIO_XUIO) &&
769 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
771 int blksz = zp->z_blksz;
772 uint64_t offset = uio->uio_loffset;
774 xuio = (xuio_t *)uio;
776 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
779 ASSERT(offset + n <= blksz);
782 (void) dmu_xuio_init(xuio, nblk);
784 if (vn_has_cached_data(vp)) {
786 * For simplicity, we always allocate a full buffer
787 * even if we only expect to read a portion of a block.
789 while (--nblk >= 0) {
790 (void) dmu_xuio_add(xuio,
791 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
799 nbytes = MIN(n, zfs_read_chunk_size -
800 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
803 if (uio->uio_segflg == UIO_NOCOPY)
804 error = mappedread_sf(vp, nbytes, uio);
806 #endif /* __FreeBSD__ */
807 if (vn_has_cached_data(vp)) {
808 error = mappedread(vp, nbytes, uio);
810 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
814 /* convert checksum errors into IO errors */
816 error = SET_ERROR(EIO);
823 zfs_range_unlock(rl);
825 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
831 * Write the bytes to a file.
833 * IN: vp - vnode of file to be written to.
834 * uio - structure supplying write location, range info,
836 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is
837 * set if in append mode.
838 * cr - credentials of caller.
839 * ct - caller context (NFS/CIFS fem monitor only)
841 * OUT: uio - updated offset and range.
843 * RETURN: 0 on success, error code on failure.
846 * vp - ctime|mtime updated if byte count > 0
851 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
853 znode_t *zp = VTOZ(vp);
854 rlim64_t limit = MAXOFFSET_T;
855 ssize_t start_resid = uio->uio_resid;
859 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
864 int max_blksz = zfsvfs->z_max_blksz;
867 iovec_t *aiov = NULL;
870 int iovcnt = uio->uio_iovcnt;
871 iovec_t *iovp = uio->uio_iov;
874 sa_bulk_attr_t bulk[4];
875 uint64_t mtime[2], ctime[2];
878 * Fasttrack empty write
884 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
890 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
891 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
892 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
894 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
898 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
899 * callers might not be able to detect properly that we are read-only,
900 * so check it explicitly here.
902 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
904 return (SET_ERROR(EROFS));
908 * If immutable or not appending then return EPERM.
909 * Intentionally allow ZFS_READONLY through here.
910 * See zfs_zaccess_common()
912 if ((zp->z_pflags & ZFS_IMMUTABLE) ||
913 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
914 (uio->uio_loffset < zp->z_size))) {
916 return (SET_ERROR(EPERM));
919 zilog = zfsvfs->z_log;
922 * Validate file offset
924 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
927 return (SET_ERROR(EINVAL));
931 * Check for mandatory locks before calling zfs_range_lock()
932 * in order to prevent a deadlock with locks set via fcntl().
934 if (MANDMODE((mode_t)zp->z_mode) &&
935 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
942 * Pre-fault the pages to ensure slow (eg NFS) pages
944 * Skip this if uio contains loaned arc_buf.
946 if ((uio->uio_extflg == UIO_XUIO) &&
947 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
948 xuio = (xuio_t *)uio;
950 uio_prefaultpages(MIN(n, max_blksz), uio);
954 * If in append mode, set the io offset pointer to eof.
956 if (ioflag & FAPPEND) {
958 * Obtain an appending range lock to guarantee file append
959 * semantics. We reset the write offset once we have the lock.
961 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
963 if (rl->r_len == UINT64_MAX) {
965 * We overlocked the file because this write will cause
966 * the file block size to increase.
967 * Note that zp_size cannot change with this lock held.
971 uio->uio_loffset = woff;
974 * Note that if the file block size will change as a result of
975 * this write, then this range lock will lock the entire file
976 * so that we can re-write the block safely.
978 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
981 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) {
982 zfs_range_unlock(rl);
988 zfs_range_unlock(rl);
990 return (SET_ERROR(EFBIG));
993 if ((woff + n) > limit || woff > (limit - n))
996 /* Will this write extend the file length? */
997 write_eof = (woff + n > zp->z_size);
999 end_size = MAX(zp->z_size, woff + n);
1002 * Write the file in reasonable size chunks. Each chunk is written
1003 * in a separate transaction; this keeps the intent log records small
1004 * and allows us to do more fine-grained space accounting.
1008 woff = uio->uio_loffset;
1009 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
1010 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
1012 dmu_return_arcbuf(abuf);
1013 error = SET_ERROR(EDQUOT);
1017 if (xuio && abuf == NULL) {
1018 ASSERT(i_iov < iovcnt);
1019 aiov = &iovp[i_iov];
1020 abuf = dmu_xuio_arcbuf(xuio, i_iov);
1021 dmu_xuio_clear(xuio, i_iov);
1022 DTRACE_PROBE3(zfs_cp_write, int, i_iov,
1023 iovec_t *, aiov, arc_buf_t *, abuf);
1024 ASSERT((aiov->iov_base == abuf->b_data) ||
1025 ((char *)aiov->iov_base - (char *)abuf->b_data +
1026 aiov->iov_len == arc_buf_size(abuf)));
1028 } else if (abuf == NULL && n >= max_blksz &&
1029 woff >= zp->z_size &&
1030 P2PHASE(woff, max_blksz) == 0 &&
1031 zp->z_blksz == max_blksz) {
1033 * This write covers a full block. "Borrow" a buffer
1034 * from the dmu so that we can fill it before we enter
1035 * a transaction. This avoids the possibility of
1036 * holding up the transaction if the data copy hangs
1037 * up on a pagefault (e.g., from an NFS server mapping).
1043 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
1045 ASSERT(abuf != NULL);
1046 ASSERT(arc_buf_size(abuf) == max_blksz);
1048 if (error = uiocopy(abuf->b_data, max_blksz,
1049 UIO_WRITE, uio, &cbytes)) {
1050 dmu_return_arcbuf(abuf);
1053 ASSERT(cbytes == max_blksz);
1055 ssize_t resid = uio->uio_resid;
1056 error = vn_io_fault_uiomove(abuf->b_data, max_blksz, uio);
1058 uio->uio_offset -= resid - uio->uio_resid;
1059 uio->uio_resid = resid;
1060 dmu_return_arcbuf(abuf);
1067 * Start a transaction.
1069 tx = dmu_tx_create(zfsvfs->z_os);
1070 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1071 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
1072 zfs_sa_upgrade_txholds(tx, zp);
1073 error = dmu_tx_assign(tx, TXG_WAIT);
1077 dmu_return_arcbuf(abuf);
1082 * If zfs_range_lock() over-locked we grow the blocksize
1083 * and then reduce the lock range. This will only happen
1084 * on the first iteration since zfs_range_reduce() will
1085 * shrink down r_len to the appropriate size.
1087 if (rl->r_len == UINT64_MAX) {
1090 if (zp->z_blksz > max_blksz) {
1092 * File's blocksize is already larger than the
1093 * "recordsize" property. Only let it grow to
1094 * the next power of 2.
1096 ASSERT(!ISP2(zp->z_blksz));
1097 new_blksz = MIN(end_size,
1098 1 << highbit64(zp->z_blksz));
1100 new_blksz = MIN(end_size, max_blksz);
1102 zfs_grow_blocksize(zp, new_blksz, tx);
1103 zfs_range_reduce(rl, woff, n);
1107 * XXX - should we really limit each write to z_max_blksz?
1108 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
1110 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
1112 if (woff + nbytes > zp->z_size)
1113 vnode_pager_setsize(vp, woff + nbytes);
1116 tx_bytes = uio->uio_resid;
1117 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
1119 tx_bytes -= uio->uio_resid;
1122 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
1124 * If this is not a full block write, but we are
1125 * extending the file past EOF and this data starts
1126 * block-aligned, use assign_arcbuf(). Otherwise,
1127 * write via dmu_write().
1129 if (tx_bytes < max_blksz && (!write_eof ||
1130 aiov->iov_base != abuf->b_data)) {
1132 dmu_write(zfsvfs->z_os, zp->z_id, woff,
1133 aiov->iov_len, aiov->iov_base, tx);
1134 dmu_return_arcbuf(abuf);
1135 xuio_stat_wbuf_copied();
1137 ASSERT(xuio || tx_bytes == max_blksz);
1138 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
1142 ASSERT(tx_bytes <= uio->uio_resid);
1143 uioskip(uio, tx_bytes);
1146 if (tx_bytes && vn_has_cached_data(vp)) {
1147 update_pages(vp, woff, tx_bytes, zfsvfs->z_os,
1148 zp->z_id, uio->uio_segflg, tx);
1152 * If we made no progress, we're done. If we made even
1153 * partial progress, update the znode and ZIL accordingly.
1155 if (tx_bytes == 0) {
1156 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
1157 (void *)&zp->z_size, sizeof (uint64_t), tx);
1164 * Clear Set-UID/Set-GID bits on successful write if not
1165 * privileged and at least one of the excute bits is set.
1167 * It would be nice to to this after all writes have
1168 * been done, but that would still expose the ISUID/ISGID
1169 * to another app after the partial write is committed.
1171 * Note: we don't call zfs_fuid_map_id() here because
1172 * user 0 is not an ephemeral uid.
1174 mutex_enter(&zp->z_acl_lock);
1175 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
1176 (S_IXUSR >> 6))) != 0 &&
1177 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
1178 secpolicy_vnode_setid_retain(vp, cr,
1179 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
1181 zp->z_mode &= ~(S_ISUID | S_ISGID);
1182 newmode = zp->z_mode;
1183 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
1184 (void *)&newmode, sizeof (uint64_t), tx);
1186 mutex_exit(&zp->z_acl_lock);
1188 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
1192 * Update the file size (zp_size) if it has changed;
1193 * account for possible concurrent updates.
1195 while ((end_size = zp->z_size) < uio->uio_loffset) {
1196 (void) atomic_cas_64(&zp->z_size, end_size,
1201 ASSERT(error == 0 || error == EFAULT);
1205 * If we are replaying and eof is non zero then force
1206 * the file size to the specified eof. Note, there's no
1207 * concurrency during replay.
1209 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
1210 zp->z_size = zfsvfs->z_replay_eof;
1213 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1215 (void) sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1217 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
1222 ASSERT(tx_bytes == nbytes);
1227 uio_prefaultpages(MIN(n, max_blksz), uio);
1231 zfs_range_unlock(rl);
1234 * If we're in replay mode, or we made no progress, return error.
1235 * Otherwise, it's at least a partial write, so it's successful.
1237 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1244 * EFAULT means that at least one page of the source buffer was not
1245 * available. VFS will re-try remaining I/O upon this error.
1247 if (error == EFAULT) {
1253 if (ioflag & (FSYNC | FDSYNC) ||
1254 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1255 zil_commit(zilog, zp->z_id);
1262 zfs_get_done(zgd_t *zgd, int error)
1264 znode_t *zp = zgd->zgd_private;
1265 objset_t *os = zp->z_zfsvfs->z_os;
1268 dmu_buf_rele(zgd->zgd_db, zgd);
1270 zfs_range_unlock(zgd->zgd_rl);
1273 * Release the vnode asynchronously as we currently have the
1274 * txg stopped from syncing.
1276 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1278 if (error == 0 && zgd->zgd_bp)
1279 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1281 kmem_free(zgd, sizeof (zgd_t));
1285 static int zil_fault_io = 0;
1289 * Get data to generate a TX_WRITE intent log record.
1292 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1294 zfsvfs_t *zfsvfs = arg;
1295 objset_t *os = zfsvfs->z_os;
1297 uint64_t object = lr->lr_foid;
1298 uint64_t offset = lr->lr_offset;
1299 uint64_t size = lr->lr_length;
1304 ASSERT(zio != NULL);
1308 * Nothing to do if the file has been removed
1310 if (zfs_zget(zfsvfs, object, &zp) != 0)
1311 return (SET_ERROR(ENOENT));
1312 if (zp->z_unlinked) {
1314 * Release the vnode asynchronously as we currently have the
1315 * txg stopped from syncing.
1317 VN_RELE_ASYNC(ZTOV(zp),
1318 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1319 return (SET_ERROR(ENOENT));
1322 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1323 zgd->zgd_zilog = zfsvfs->z_log;
1324 zgd->zgd_private = zp;
1327 * Write records come in two flavors: immediate and indirect.
1328 * For small writes it's cheaper to store the data with the
1329 * log record (immediate); for large writes it's cheaper to
1330 * sync the data and get a pointer to it (indirect) so that
1331 * we don't have to write the data twice.
1333 if (buf != NULL) { /* immediate write */
1334 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1335 /* test for truncation needs to be done while range locked */
1336 if (offset >= zp->z_size) {
1337 error = SET_ERROR(ENOENT);
1339 error = dmu_read(os, object, offset, size, buf,
1340 DMU_READ_NO_PREFETCH);
1342 ASSERT(error == 0 || error == ENOENT);
1343 } else { /* indirect write */
1345 * Have to lock the whole block to ensure when it's
1346 * written out and it's checksum is being calculated
1347 * that no one can change the data. We need to re-check
1348 * blocksize after we get the lock in case it's changed!
1353 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1355 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1357 if (zp->z_blksz == size)
1360 zfs_range_unlock(zgd->zgd_rl);
1362 /* test for truncation needs to be done while range locked */
1363 if (lr->lr_offset >= zp->z_size)
1364 error = SET_ERROR(ENOENT);
1367 error = SET_ERROR(EIO);
1372 error = dmu_buf_hold(os, object, offset, zgd, &db,
1373 DMU_READ_NO_PREFETCH);
1376 blkptr_t *bp = &lr->lr_blkptr;
1381 ASSERT(db->db_offset == offset);
1382 ASSERT(db->db_size == size);
1384 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1386 ASSERT(error || lr->lr_length <= size);
1389 * On success, we need to wait for the write I/O
1390 * initiated by dmu_sync() to complete before we can
1391 * release this dbuf. We will finish everything up
1392 * in the zfs_get_done() callback.
1397 if (error == EALREADY) {
1398 lr->lr_common.lrc_txtype = TX_WRITE2;
1404 zfs_get_done(zgd, error);
1411 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1412 caller_context_t *ct)
1414 znode_t *zp = VTOZ(vp);
1415 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1421 if (flag & V_ACE_MASK)
1422 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1424 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1431 zfs_dd_callback(struct mount *mp, void *arg, int lkflags, struct vnode **vpp)
1436 error = vn_lock(*vpp, lkflags);
1443 zfs_lookup_lock(vnode_t *dvp, vnode_t *vp, const char *name, int lkflags)
1445 znode_t *zdp = VTOZ(dvp);
1446 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1450 ASSERT_VOP_LOCKED(dvp, __func__);
1452 if ((zdp->z_pflags & ZFS_XATTR) == 0)
1453 VERIFY(!RRM_LOCK_HELD(&zfsvfs->z_teardown_lock));
1456 if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) {
1457 ASSERT3P(dvp, ==, vp);
1459 ltype = lkflags & LK_TYPE_MASK;
1460 if (ltype != VOP_ISLOCKED(dvp)) {
1461 if (ltype == LK_EXCLUSIVE)
1462 vn_lock(dvp, LK_UPGRADE | LK_RETRY);
1463 else /* if (ltype == LK_SHARED) */
1464 vn_lock(dvp, LK_DOWNGRADE | LK_RETRY);
1467 * Relock for the "." case could leave us with
1470 if (dvp->v_iflag & VI_DOOMED) {
1472 return (SET_ERROR(ENOENT));
1476 } else if (name[0] == '.' && name[1] == '.' && name[2] == 0) {
1478 * Note that in this case, dvp is the child vnode, and we
1479 * are looking up the parent vnode - exactly reverse from
1480 * normal operation. Unlocking dvp requires some rather
1481 * tricky unlock/relock dance to prevent mp from being freed;
1482 * use vn_vget_ino_gen() which takes care of all that.
1484 * XXX Note that there is a time window when both vnodes are
1485 * unlocked. It is possible, although highly unlikely, that
1486 * during that window the parent-child relationship between
1487 * the vnodes may change, for example, get reversed.
1488 * In that case we would have a wrong lock order for the vnodes.
1489 * All other filesystems seem to ignore this problem, so we
1491 * A potential solution could be implemented as follows:
1492 * - using LK_NOWAIT when locking the second vnode and retrying
1494 * - checking that the parent-child relationship still holds
1495 * after locking both vnodes and retrying if it doesn't
1497 error = vn_vget_ino_gen(dvp, zfs_dd_callback, vp, lkflags, &vp);
1500 error = vn_lock(vp, lkflags);
1508 * Lookup an entry in a directory, or an extended attribute directory.
1509 * If it exists, return a held vnode reference for it.
1511 * IN: dvp - vnode of directory to search.
1512 * nm - name of entry to lookup.
1513 * pnp - full pathname to lookup [UNUSED].
1514 * flags - LOOKUP_XATTR set if looking for an attribute.
1515 * rdir - root directory vnode [UNUSED].
1516 * cr - credentials of caller.
1517 * ct - caller context
1519 * OUT: vpp - vnode of located entry, NULL if not found.
1521 * RETURN: 0 on success, error code on failure.
1528 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp,
1529 int nameiop, cred_t *cr, kthread_t *td, int flags)
1531 znode_t *zdp = VTOZ(dvp);
1533 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1537 * Fast path lookup, however we must skip DNLC lookup
1538 * for case folding or normalizing lookups because the
1539 * DNLC code only stores the passed in name. This means
1540 * creating 'a' and removing 'A' on a case insensitive
1541 * file system would work, but DNLC still thinks 'a'
1542 * exists and won't let you create it again on the next
1543 * pass through fast path.
1545 if (!(flags & LOOKUP_XATTR)) {
1546 if (dvp->v_type != VDIR) {
1547 return (SET_ERROR(ENOTDIR));
1548 } else if (zdp->z_sa_hdl == NULL) {
1549 return (SET_ERROR(EIO));
1553 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1560 if (flags & LOOKUP_XATTR) {
1563 * If the xattr property is off, refuse the lookup request.
1565 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1567 return (SET_ERROR(EINVAL));
1572 * We don't allow recursive attributes..
1573 * Maybe someday we will.
1575 if (zdp->z_pflags & ZFS_XATTR) {
1577 return (SET_ERROR(EINVAL));
1580 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1586 * Do we have permission to get into attribute directory?
1588 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1599 * Check accessibility of directory.
1601 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1606 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1607 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1609 return (SET_ERROR(EILSEQ));
1614 * First handle the special cases.
1616 if ((cnp->cn_flags & ISDOTDOT) != 0) {
1618 * If we are a snapshot mounted under .zfs, return
1619 * the vp for the snapshot directory.
1621 if (zdp->z_id == zfsvfs->z_root && zfsvfs->z_parent != zfsvfs) {
1622 struct componentname cn;
1627 ltype = VOP_ISLOCKED(dvp);
1629 error = zfsctl_root(zfsvfs->z_parent, LK_SHARED,
1632 cn.cn_nameptr = "snapshot";
1633 cn.cn_namelen = strlen(cn.cn_nameptr);
1634 cn.cn_nameiop = cnp->cn_nameiop;
1635 cn.cn_flags = cnp->cn_flags & ~ISDOTDOT;
1636 cn.cn_lkflags = cnp->cn_lkflags;
1637 error = VOP_LOOKUP(zfsctl_vp, vpp, &cn);
1640 vn_lock(dvp, ltype | LK_RETRY);
1644 if (zfs_has_ctldir(zdp) && strcmp(nm, ZFS_CTLDIR_NAME) == 0) {
1646 if ((cnp->cn_flags & ISLASTCN) != 0 && nameiop != LOOKUP)
1647 return (SET_ERROR(ENOTSUP));
1648 error = zfsctl_root(zfsvfs, cnp->cn_lkflags, vpp);
1653 * The loop is retry the lookup if the parent-child relationship
1654 * changes during the dot-dot locking complexities.
1659 error = zfs_dirlook(zdp, nm, &zp);
1667 error = zfs_lookup_lock(dvp, *vpp, nm, cnp->cn_lkflags);
1670 * If we've got a locking error, then the vnode
1671 * got reclaimed because of a force unmount.
1672 * We never enter doomed vnodes into the name cache.
1678 if ((cnp->cn_flags & ISDOTDOT) == 0)
1682 if (zdp->z_sa_hdl == NULL) {
1683 error = SET_ERROR(EIO);
1685 error = sa_lookup(zdp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
1686 &parent, sizeof (parent));
1693 if (zp->z_id == parent) {
1704 /* Translate errors and add SAVENAME when needed. */
1705 if (cnp->cn_flags & ISLASTCN) {
1709 if (error == ENOENT) {
1710 error = EJUSTRETURN;
1711 cnp->cn_flags |= SAVENAME;
1717 cnp->cn_flags |= SAVENAME;
1722 /* Insert name into cache (as non-existent) if appropriate. */
1723 if (zfsvfs->z_use_namecache &&
1724 error == ENOENT && (cnp->cn_flags & MAKEENTRY) != 0)
1725 cache_enter(dvp, NULL, cnp);
1727 /* Insert name into cache if appropriate. */
1728 if (zfsvfs->z_use_namecache &&
1729 error == 0 && (cnp->cn_flags & MAKEENTRY)) {
1730 if (!(cnp->cn_flags & ISLASTCN) ||
1731 (nameiop != DELETE && nameiop != RENAME)) {
1732 cache_enter(dvp, *vpp, cnp);
1740 * Attempt to create a new entry in a directory. If the entry
1741 * already exists, truncate the file if permissible, else return
1742 * an error. Return the vp of the created or trunc'd file.
1744 * IN: dvp - vnode of directory to put new file entry in.
1745 * name - name of new file entry.
1746 * vap - attributes of new file.
1747 * excl - flag indicating exclusive or non-exclusive mode.
1748 * mode - mode to open file with.
1749 * cr - credentials of caller.
1750 * flag - large file flag [UNUSED].
1751 * ct - caller context
1752 * vsecp - ACL to be set
1754 * OUT: vpp - vnode of created or trunc'd entry.
1756 * RETURN: 0 on success, error code on failure.
1759 * dvp - ctime|mtime updated if new entry created
1760 * vp - ctime|mtime always, atime if new
1765 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode,
1766 vnode_t **vpp, cred_t *cr, kthread_t *td)
1768 znode_t *zp, *dzp = VTOZ(dvp);
1769 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1776 gid_t gid = crgetgid(cr);
1777 zfs_acl_ids_t acl_ids;
1778 boolean_t fuid_dirtied;
1784 * If we have an ephemeral id, ACL, or XVATTR then
1785 * make sure file system is at proper version
1788 ksid = crgetsid(cr, KSID_OWNER);
1790 uid = ksid_getid(ksid);
1794 if (zfsvfs->z_use_fuids == B_FALSE &&
1795 (vsecp || (vap->va_mask & AT_XVATTR) ||
1796 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1797 return (SET_ERROR(EINVAL));
1802 zilog = zfsvfs->z_log;
1804 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1805 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1807 return (SET_ERROR(EILSEQ));
1810 if (vap->va_mask & AT_XVATTR) {
1811 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
1812 crgetuid(cr), cr, vap->va_type)) != 0) {
1820 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr))
1821 vap->va_mode &= ~S_ISVTX;
1823 error = zfs_dirent_lookup(dzp, name, &zp, ZNEW);
1828 ASSERT3P(zp, ==, NULL);
1831 * Create a new file object and update the directory
1834 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1839 * We only support the creation of regular files in
1840 * extended attribute directories.
1843 if ((dzp->z_pflags & ZFS_XATTR) &&
1844 (vap->va_type != VREG)) {
1845 error = SET_ERROR(EINVAL);
1849 if ((error = zfs_acl_ids_create(dzp, 0, vap,
1850 cr, vsecp, &acl_ids)) != 0)
1853 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1854 zfs_acl_ids_free(&acl_ids);
1855 error = SET_ERROR(EDQUOT);
1859 getnewvnode_reserve(1);
1861 tx = dmu_tx_create(os);
1863 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1864 ZFS_SA_BASE_ATTR_SIZE);
1866 fuid_dirtied = zfsvfs->z_fuid_dirty;
1868 zfs_fuid_txhold(zfsvfs, tx);
1869 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1870 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1871 if (!zfsvfs->z_use_sa &&
1872 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1873 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1874 0, acl_ids.z_aclp->z_acl_bytes);
1876 error = dmu_tx_assign(tx, TXG_WAIT);
1878 zfs_acl_ids_free(&acl_ids);
1880 getnewvnode_drop_reserve();
1884 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1887 zfs_fuid_sync(zfsvfs, tx);
1889 (void) zfs_link_create(dzp, name, zp, tx, ZNEW);
1890 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1891 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1892 vsecp, acl_ids.z_fuidp, vap);
1893 zfs_acl_ids_free(&acl_ids);
1896 getnewvnode_drop_reserve();
1903 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1904 zil_commit(zilog, 0);
1911 * Remove an entry from a directory.
1913 * IN: dvp - vnode of directory to remove entry from.
1914 * name - name of entry to remove.
1915 * cr - credentials of caller.
1916 * ct - caller context
1917 * flags - case flags
1919 * RETURN: 0 on success, error code on failure.
1923 * vp - ctime (if nlink > 0)
1928 zfs_remove(vnode_t *dvp, vnode_t *vp, char *name, cred_t *cr)
1930 znode_t *dzp = VTOZ(dvp);
1931 znode_t *zp = VTOZ(vp);
1933 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1935 uint64_t acl_obj, xattr_obj;
1938 boolean_t unlinked, toobig = FALSE;
1945 zilog = zfsvfs->z_log;
1951 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1956 * Need to use rmdir for removing directories.
1958 if (vp->v_type == VDIR) {
1959 error = SET_ERROR(EPERM);
1963 vnevent_remove(vp, dvp, name, ct);
1967 /* are there any extended attributes? */
1968 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1969 &xattr_obj, sizeof (xattr_obj));
1970 if (error == 0 && xattr_obj) {
1971 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1976 * We may delete the znode now, or we may put it in the unlinked set;
1977 * it depends on whether we're the last link, and on whether there are
1978 * other holds on the vnode. So we dmu_tx_hold() the right things to
1979 * allow for either case.
1981 tx = dmu_tx_create(zfsvfs->z_os);
1982 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1983 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1984 zfs_sa_upgrade_txholds(tx, zp);
1985 zfs_sa_upgrade_txholds(tx, dzp);
1988 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1989 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1992 /* charge as an update -- would be nice not to charge at all */
1993 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1996 * Mark this transaction as typically resulting in a net free of space
1998 dmu_tx_mark_netfree(tx);
2000 error = dmu_tx_assign(tx, TXG_WAIT);
2008 * Remove the directory entry.
2010 error = zfs_link_destroy(dzp, name, zp, tx, ZEXISTS, &unlinked);
2018 zfs_unlinked_add(zp, tx);
2019 vp->v_vflag |= VV_NOSYNC;
2023 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
2031 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2032 zil_commit(zilog, 0);
2039 * Create a new directory and insert it into dvp using the name
2040 * provided. Return a pointer to the inserted directory.
2042 * IN: dvp - vnode of directory to add subdir to.
2043 * dirname - name of new directory.
2044 * vap - attributes of new directory.
2045 * cr - credentials of caller.
2046 * ct - caller context
2047 * flags - case flags
2048 * vsecp - ACL to be set
2050 * OUT: vpp - vnode of created directory.
2052 * RETURN: 0 on success, error code on failure.
2055 * dvp - ctime|mtime updated
2056 * vp - ctime|mtime|atime updated
2060 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr)
2062 znode_t *zp, *dzp = VTOZ(dvp);
2063 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2070 gid_t gid = crgetgid(cr);
2071 zfs_acl_ids_t acl_ids;
2072 boolean_t fuid_dirtied;
2074 ASSERT(vap->va_type == VDIR);
2077 * If we have an ephemeral id, ACL, or XVATTR then
2078 * make sure file system is at proper version
2081 ksid = crgetsid(cr, KSID_OWNER);
2083 uid = ksid_getid(ksid);
2086 if (zfsvfs->z_use_fuids == B_FALSE &&
2087 ((vap->va_mask & AT_XVATTR) ||
2088 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
2089 return (SET_ERROR(EINVAL));
2093 zilog = zfsvfs->z_log;
2095 if (dzp->z_pflags & ZFS_XATTR) {
2097 return (SET_ERROR(EINVAL));
2100 if (zfsvfs->z_utf8 && u8_validate(dirname,
2101 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2103 return (SET_ERROR(EILSEQ));
2106 if (vap->va_mask & AT_XVATTR) {
2107 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
2108 crgetuid(cr), cr, vap->va_type)) != 0) {
2114 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
2115 NULL, &acl_ids)) != 0) {
2121 * First make sure the new directory doesn't exist.
2123 * Existence is checked first to make sure we don't return
2124 * EACCES instead of EEXIST which can cause some applications
2129 if (error = zfs_dirent_lookup(dzp, dirname, &zp, ZNEW)) {
2130 zfs_acl_ids_free(&acl_ids);
2134 ASSERT3P(zp, ==, NULL);
2136 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
2137 zfs_acl_ids_free(&acl_ids);
2142 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
2143 zfs_acl_ids_free(&acl_ids);
2145 return (SET_ERROR(EDQUOT));
2149 * Add a new entry to the directory.
2151 getnewvnode_reserve(1);
2152 tx = dmu_tx_create(zfsvfs->z_os);
2153 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2154 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2155 fuid_dirtied = zfsvfs->z_fuid_dirty;
2157 zfs_fuid_txhold(zfsvfs, tx);
2158 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2159 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2160 acl_ids.z_aclp->z_acl_bytes);
2163 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2164 ZFS_SA_BASE_ATTR_SIZE);
2166 error = dmu_tx_assign(tx, TXG_WAIT);
2168 zfs_acl_ids_free(&acl_ids);
2170 getnewvnode_drop_reserve();
2178 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2181 zfs_fuid_sync(zfsvfs, tx);
2184 * Now put new name in parent dir.
2186 (void) zfs_link_create(dzp, dirname, zp, tx, ZNEW);
2190 txtype = zfs_log_create_txtype(Z_DIR, NULL, vap);
2191 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, NULL,
2192 acl_ids.z_fuidp, vap);
2194 zfs_acl_ids_free(&acl_ids);
2198 getnewvnode_drop_reserve();
2200 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2201 zil_commit(zilog, 0);
2208 * Remove a directory subdir entry. If the current working
2209 * directory is the same as the subdir to be removed, the
2212 * IN: dvp - vnode of directory to remove from.
2213 * name - name of directory to be removed.
2214 * cwd - vnode of current working directory.
2215 * cr - credentials of caller.
2216 * ct - caller context
2217 * flags - case flags
2219 * RETURN: 0 on success, error code on failure.
2222 * dvp - ctime|mtime updated
2226 zfs_rmdir(vnode_t *dvp, vnode_t *vp, char *name, cred_t *cr)
2228 znode_t *dzp = VTOZ(dvp);
2229 znode_t *zp = VTOZ(vp);
2230 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2238 zilog = zfsvfs->z_log;
2241 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2245 if (vp->v_type != VDIR) {
2246 error = SET_ERROR(ENOTDIR);
2250 vnevent_rmdir(vp, dvp, name, ct);
2252 tx = dmu_tx_create(zfsvfs->z_os);
2253 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2254 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2255 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2256 zfs_sa_upgrade_txholds(tx, zp);
2257 zfs_sa_upgrade_txholds(tx, dzp);
2258 dmu_tx_mark_netfree(tx);
2259 error = dmu_tx_assign(tx, TXG_WAIT);
2268 error = zfs_link_destroy(dzp, name, zp, tx, ZEXISTS, NULL);
2271 uint64_t txtype = TX_RMDIR;
2272 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2279 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2280 zil_commit(zilog, 0);
2287 * Read as many directory entries as will fit into the provided
2288 * buffer from the given directory cursor position (specified in
2289 * the uio structure).
2291 * IN: vp - vnode of directory to read.
2292 * uio - structure supplying read location, range info,
2293 * and return buffer.
2294 * cr - credentials of caller.
2295 * ct - caller context
2296 * flags - case flags
2298 * OUT: uio - updated offset and range, buffer filled.
2299 * eofp - set to true if end-of-file detected.
2301 * RETURN: 0 on success, error code on failure.
2304 * vp - atime updated
2306 * Note that the low 4 bits of the cookie returned by zap is always zero.
2307 * This allows us to use the low range for "special" directory entries:
2308 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2309 * we use the offset 2 for the '.zfs' directory.
2313 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies)
2315 znode_t *zp = VTOZ(vp);
2319 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2324 zap_attribute_t zap;
2325 uint_t bytes_wanted;
2326 uint64_t offset; /* must be unsigned; checks for < 1 */
2332 boolean_t check_sysattrs;
2335 u_long *cooks = NULL;
2341 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2342 &parent, sizeof (parent))) != 0) {
2348 * If we are not given an eof variable,
2355 * Check for valid iov_len.
2357 if (uio->uio_iov->iov_len <= 0) {
2359 return (SET_ERROR(EINVAL));
2363 * Quit if directory has been removed (posix)
2365 if ((*eofp = zp->z_unlinked) != 0) {
2372 offset = uio->uio_loffset;
2373 prefetch = zp->z_zn_prefetch;
2376 * Initialize the iterator cursor.
2380 * Start iteration from the beginning of the directory.
2382 zap_cursor_init(&zc, os, zp->z_id);
2385 * The offset is a serialized cursor.
2387 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2391 * Get space to change directory entries into fs independent format.
2393 iovp = uio->uio_iov;
2394 bytes_wanted = iovp->iov_len;
2395 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2396 bufsize = bytes_wanted;
2397 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2398 odp = (struct dirent64 *)outbuf;
2400 bufsize = bytes_wanted;
2402 odp = (struct dirent64 *)iovp->iov_base;
2404 eodp = (struct edirent *)odp;
2406 if (ncookies != NULL) {
2408 * Minimum entry size is dirent size and 1 byte for a file name.
2410 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1);
2411 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK);
2416 * If this VFS supports the system attribute view interface; and
2417 * we're looking at an extended attribute directory; and we care
2418 * about normalization conflicts on this vfs; then we must check
2419 * for normalization conflicts with the sysattr name space.
2422 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2423 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2424 (flags & V_RDDIR_ENTFLAGS);
2430 * Transform to file-system independent format
2433 while (outcount < bytes_wanted) {
2436 off64_t *next = NULL;
2439 * Special case `.', `..', and `.zfs'.
2442 (void) strcpy(zap.za_name, ".");
2443 zap.za_normalization_conflict = 0;
2446 } else if (offset == 1) {
2447 (void) strcpy(zap.za_name, "..");
2448 zap.za_normalization_conflict = 0;
2451 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2452 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2453 zap.za_normalization_conflict = 0;
2454 objnum = ZFSCTL_INO_ROOT;
2460 if (error = zap_cursor_retrieve(&zc, &zap)) {
2461 if ((*eofp = (error == ENOENT)) != 0)
2467 if (zap.za_integer_length != 8 ||
2468 zap.za_num_integers != 1) {
2469 cmn_err(CE_WARN, "zap_readdir: bad directory "
2470 "entry, obj = %lld, offset = %lld\n",
2471 (u_longlong_t)zp->z_id,
2472 (u_longlong_t)offset);
2473 error = SET_ERROR(ENXIO);
2477 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2479 * MacOS X can extract the object type here such as:
2480 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2482 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2484 if (check_sysattrs && !zap.za_normalization_conflict) {
2486 zap.za_normalization_conflict =
2487 xattr_sysattr_casechk(zap.za_name);
2489 panic("%s:%u: TODO", __func__, __LINE__);
2494 if (flags & V_RDDIR_ACCFILTER) {
2496 * If we have no access at all, don't include
2497 * this entry in the returned information
2500 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2502 if (!zfs_has_access(ezp, cr)) {
2509 if (flags & V_RDDIR_ENTFLAGS)
2510 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2512 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2515 * Will this entry fit in the buffer?
2517 if (outcount + reclen > bufsize) {
2519 * Did we manage to fit anything in the buffer?
2522 error = SET_ERROR(EINVAL);
2527 if (flags & V_RDDIR_ENTFLAGS) {
2529 * Add extended flag entry:
2531 eodp->ed_ino = objnum;
2532 eodp->ed_reclen = reclen;
2533 /* NOTE: ed_off is the offset for the *next* entry */
2534 next = &(eodp->ed_off);
2535 eodp->ed_eflags = zap.za_normalization_conflict ?
2536 ED_CASE_CONFLICT : 0;
2537 (void) strncpy(eodp->ed_name, zap.za_name,
2538 EDIRENT_NAMELEN(reclen));
2539 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2544 odp->d_ino = objnum;
2545 odp->d_reclen = reclen;
2546 odp->d_namlen = strlen(zap.za_name);
2547 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1);
2549 odp = (dirent64_t *)((intptr_t)odp + reclen);
2553 ASSERT(outcount <= bufsize);
2555 /* Prefetch znode */
2557 dmu_prefetch(os, objnum, 0, 0, 0,
2558 ZIO_PRIORITY_SYNC_READ);
2562 * Move to the next entry, fill in the previous offset.
2564 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2565 zap_cursor_advance(&zc);
2566 offset = zap_cursor_serialize(&zc);
2571 if (cooks != NULL) {
2574 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks));
2577 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2579 /* Subtract unused cookies */
2580 if (ncookies != NULL)
2581 *ncookies -= ncooks;
2583 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2584 iovp->iov_base += outcount;
2585 iovp->iov_len -= outcount;
2586 uio->uio_resid -= outcount;
2587 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2589 * Reset the pointer.
2591 offset = uio->uio_loffset;
2595 zap_cursor_fini(&zc);
2596 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2597 kmem_free(outbuf, bufsize);
2599 if (error == ENOENT)
2602 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2604 uio->uio_loffset = offset;
2606 if (error != 0 && cookies != NULL) {
2607 free(*cookies, M_TEMP);
2614 ulong_t zfs_fsync_sync_cnt = 4;
2617 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2619 znode_t *zp = VTOZ(vp);
2620 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2622 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2624 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2627 zil_commit(zfsvfs->z_log, zp->z_id);
2635 * Get the requested file attributes and place them in the provided
2638 * IN: vp - vnode of file.
2639 * vap - va_mask identifies requested attributes.
2640 * If AT_XVATTR set, then optional attrs are requested
2641 * flags - ATTR_NOACLCHECK (CIFS server context)
2642 * cr - credentials of caller.
2643 * ct - caller context
2645 * OUT: vap - attribute values.
2647 * RETURN: 0 (always succeeds).
2651 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2652 caller_context_t *ct)
2654 znode_t *zp = VTOZ(vp);
2655 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2658 u_longlong_t nblocks;
2660 uint64_t mtime[2], ctime[2], crtime[2], rdev;
2661 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2662 xoptattr_t *xoap = NULL;
2663 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2664 sa_bulk_attr_t bulk[4];
2670 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2672 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2673 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2674 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
2675 if (vp->v_type == VBLK || vp->v_type == VCHR)
2676 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
2679 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2685 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2686 * Also, if we are the owner don't bother, since owner should
2687 * always be allowed to read basic attributes of file.
2689 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2690 (vap->va_uid != crgetuid(cr))) {
2691 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2699 * Return all attributes. It's cheaper to provide the answer
2700 * than to determine whether we were asked the question.
2703 vap->va_type = IFTOVT(zp->z_mode);
2704 vap->va_mode = zp->z_mode & ~S_IFMT;
2706 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2710 vap->va_nodeid = zp->z_id;
2711 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2712 links = zp->z_links + 1;
2714 links = zp->z_links;
2715 vap->va_nlink = MIN(links, LINK_MAX); /* nlink_t limit! */
2716 vap->va_size = zp->z_size;
2718 vap->va_rdev = vp->v_rdev;
2720 if (vp->v_type == VBLK || vp->v_type == VCHR)
2721 vap->va_rdev = zfs_cmpldev(rdev);
2723 vap->va_seq = zp->z_seq;
2724 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */
2725 vap->va_filerev = zp->z_seq;
2728 * Add in any requested optional attributes and the create time.
2729 * Also set the corresponding bits in the returned attribute bitmap.
2731 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2732 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2734 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2735 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2738 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2739 xoap->xoa_readonly =
2740 ((zp->z_pflags & ZFS_READONLY) != 0);
2741 XVA_SET_RTN(xvap, XAT_READONLY);
2744 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2746 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2747 XVA_SET_RTN(xvap, XAT_SYSTEM);
2750 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2752 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2753 XVA_SET_RTN(xvap, XAT_HIDDEN);
2756 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2757 xoap->xoa_nounlink =
2758 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2759 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2762 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2763 xoap->xoa_immutable =
2764 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2765 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2768 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2769 xoap->xoa_appendonly =
2770 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2771 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2774 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2776 ((zp->z_pflags & ZFS_NODUMP) != 0);
2777 XVA_SET_RTN(xvap, XAT_NODUMP);
2780 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2782 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2783 XVA_SET_RTN(xvap, XAT_OPAQUE);
2786 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2787 xoap->xoa_av_quarantined =
2788 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2789 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2792 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2793 xoap->xoa_av_modified =
2794 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2795 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2798 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2799 vp->v_type == VREG) {
2800 zfs_sa_get_scanstamp(zp, xvap);
2803 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2804 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2805 XVA_SET_RTN(xvap, XAT_REPARSE);
2807 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2808 xoap->xoa_generation = zp->z_gen;
2809 XVA_SET_RTN(xvap, XAT_GEN);
2812 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2814 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2815 XVA_SET_RTN(xvap, XAT_OFFLINE);
2818 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2820 ((zp->z_pflags & ZFS_SPARSE) != 0);
2821 XVA_SET_RTN(xvap, XAT_SPARSE);
2825 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2826 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2827 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2828 ZFS_TIME_DECODE(&vap->va_birthtime, crtime);
2831 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2832 vap->va_blksize = blksize;
2833 vap->va_bytes = nblocks << 9; /* nblocks * 512 */
2835 if (zp->z_blksz == 0) {
2837 * Block size hasn't been set; suggest maximal I/O transfers.
2839 vap->va_blksize = zfsvfs->z_max_blksz;
2847 * Set the file attributes to the values contained in the
2850 * IN: vp - vnode of file to be modified.
2851 * vap - new attribute values.
2852 * If AT_XVATTR set, then optional attrs are being set
2853 * flags - ATTR_UTIME set if non-default time values provided.
2854 * - ATTR_NOACLCHECK (CIFS context only).
2855 * cr - credentials of caller.
2856 * ct - caller context
2858 * RETURN: 0 on success, error code on failure.
2861 * vp - ctime updated, mtime updated if size changed.
2865 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2866 caller_context_t *ct)
2868 znode_t *zp = VTOZ(vp);
2869 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2874 uint_t mask = vap->va_mask;
2875 uint_t saved_mask = 0;
2876 uint64_t saved_mode;
2879 uint64_t new_uid, new_gid;
2881 uint64_t mtime[2], ctime[2];
2883 int need_policy = FALSE;
2885 zfs_fuid_info_t *fuidp = NULL;
2886 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2889 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2890 boolean_t fuid_dirtied = B_FALSE;
2891 sa_bulk_attr_t bulk[7], xattr_bulk[7];
2892 int count = 0, xattr_count = 0;
2897 if (mask & AT_NOSET)
2898 return (SET_ERROR(EINVAL));
2903 zilog = zfsvfs->z_log;
2906 * Make sure that if we have ephemeral uid/gid or xvattr specified
2907 * that file system is at proper version level
2910 if (zfsvfs->z_use_fuids == B_FALSE &&
2911 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2912 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2913 (mask & AT_XVATTR))) {
2915 return (SET_ERROR(EINVAL));
2918 if (mask & AT_SIZE && vp->v_type == VDIR) {
2920 return (SET_ERROR(EISDIR));
2923 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2925 return (SET_ERROR(EINVAL));
2929 * If this is an xvattr_t, then get a pointer to the structure of
2930 * optional attributes. If this is NULL, then we have a vattr_t.
2932 xoap = xva_getxoptattr(xvap);
2934 xva_init(&tmpxvattr);
2937 * Immutable files can only alter immutable bit and atime
2939 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2940 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2941 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2943 return (SET_ERROR(EPERM));
2947 * Note: ZFS_READONLY is handled in zfs_zaccess_common.
2951 * Verify timestamps doesn't overflow 32 bits.
2952 * ZFS can handle large timestamps, but 32bit syscalls can't
2953 * handle times greater than 2039. This check should be removed
2954 * once large timestamps are fully supported.
2956 if (mask & (AT_ATIME | AT_MTIME)) {
2957 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2958 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2960 return (SET_ERROR(EOVERFLOW));
2963 if (xoap && (mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME) &&
2964 TIMESPEC_OVERFLOW(&vap->va_birthtime)) {
2966 return (SET_ERROR(EOVERFLOW));
2972 /* Can this be moved to before the top label? */
2973 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2975 return (SET_ERROR(EROFS));
2979 * First validate permissions
2982 if (mask & AT_SIZE) {
2984 * XXX - Note, we are not providing any open
2985 * mode flags here (like FNDELAY), so we may
2986 * block if there are locks present... this
2987 * should be addressed in openat().
2989 /* XXX - would it be OK to generate a log record here? */
2990 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2997 if (mask & (AT_ATIME|AT_MTIME) ||
2998 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2999 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
3000 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
3001 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
3002 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
3003 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
3004 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
3005 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
3009 if (mask & (AT_UID|AT_GID)) {
3010 int idmask = (mask & (AT_UID|AT_GID));
3015 * NOTE: even if a new mode is being set,
3016 * we may clear S_ISUID/S_ISGID bits.
3019 if (!(mask & AT_MODE))
3020 vap->va_mode = zp->z_mode;
3023 * Take ownership or chgrp to group we are a member of
3026 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
3027 take_group = (mask & AT_GID) &&
3028 zfs_groupmember(zfsvfs, vap->va_gid, cr);
3031 * If both AT_UID and AT_GID are set then take_owner and
3032 * take_group must both be set in order to allow taking
3035 * Otherwise, send the check through secpolicy_vnode_setattr()
3039 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
3040 ((idmask == AT_UID) && take_owner) ||
3041 ((idmask == AT_GID) && take_group)) {
3042 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
3043 skipaclchk, cr) == 0) {
3045 * Remove setuid/setgid for non-privileged users
3047 secpolicy_setid_clear(vap, vp, cr);
3048 trim_mask = (mask & (AT_UID|AT_GID));
3057 oldva.va_mode = zp->z_mode;
3058 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
3059 if (mask & AT_XVATTR) {
3061 * Update xvattr mask to include only those attributes
3062 * that are actually changing.
3064 * the bits will be restored prior to actually setting
3065 * the attributes so the caller thinks they were set.
3067 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
3068 if (xoap->xoa_appendonly !=
3069 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
3072 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
3073 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
3077 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
3078 if (xoap->xoa_nounlink !=
3079 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
3082 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
3083 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
3087 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
3088 if (xoap->xoa_immutable !=
3089 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
3092 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
3093 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
3097 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
3098 if (xoap->xoa_nodump !=
3099 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
3102 XVA_CLR_REQ(xvap, XAT_NODUMP);
3103 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
3107 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3108 if (xoap->xoa_av_modified !=
3109 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3112 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3113 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
3117 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3118 if ((vp->v_type != VREG &&
3119 xoap->xoa_av_quarantined) ||
3120 xoap->xoa_av_quarantined !=
3121 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3124 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3125 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
3129 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3131 return (SET_ERROR(EPERM));
3134 if (need_policy == FALSE &&
3135 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3136 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3141 if (mask & AT_MODE) {
3142 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3143 err = secpolicy_setid_setsticky_clear(vp, vap,
3149 trim_mask |= AT_MODE;
3157 * If trim_mask is set then take ownership
3158 * has been granted or write_acl is present and user
3159 * has the ability to modify mode. In that case remove
3160 * UID|GID and or MODE from mask so that
3161 * secpolicy_vnode_setattr() doesn't revoke it.
3165 saved_mask = vap->va_mask;
3166 vap->va_mask &= ~trim_mask;
3167 if (trim_mask & AT_MODE) {
3169 * Save the mode, as secpolicy_vnode_setattr()
3170 * will overwrite it with ova.va_mode.
3172 saved_mode = vap->va_mode;
3175 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3176 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3183 vap->va_mask |= saved_mask;
3184 if (trim_mask & AT_MODE) {
3186 * Recover the mode after
3187 * secpolicy_vnode_setattr().
3189 vap->va_mode = saved_mode;
3195 * secpolicy_vnode_setattr, or take ownership may have
3198 mask = vap->va_mask;
3200 if ((mask & (AT_UID | AT_GID))) {
3201 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3202 &xattr_obj, sizeof (xattr_obj));
3204 if (err == 0 && xattr_obj) {
3205 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3207 err = vn_lock(ZTOV(attrzp), LK_EXCLUSIVE);
3209 vrele(ZTOV(attrzp));
3214 if (mask & AT_UID) {
3215 new_uid = zfs_fuid_create(zfsvfs,
3216 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3217 if (new_uid != zp->z_uid &&
3218 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3221 err = SET_ERROR(EDQUOT);
3226 if (mask & AT_GID) {
3227 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3228 cr, ZFS_GROUP, &fuidp);
3229 if (new_gid != zp->z_gid &&
3230 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3233 err = SET_ERROR(EDQUOT);
3238 tx = dmu_tx_create(zfsvfs->z_os);
3240 if (mask & AT_MODE) {
3241 uint64_t pmode = zp->z_mode;
3243 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3245 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3246 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3247 err = SET_ERROR(EPERM);
3251 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3254 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3256 * Are we upgrading ACL from old V0 format
3259 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3260 zfs_znode_acl_version(zp) ==
3261 ZFS_ACL_VERSION_INITIAL) {
3262 dmu_tx_hold_free(tx, acl_obj, 0,
3264 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3265 0, aclp->z_acl_bytes);
3267 dmu_tx_hold_write(tx, acl_obj, 0,
3270 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3271 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3272 0, aclp->z_acl_bytes);
3274 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3276 if ((mask & AT_XVATTR) &&
3277 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3278 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3280 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3284 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3287 fuid_dirtied = zfsvfs->z_fuid_dirty;
3289 zfs_fuid_txhold(zfsvfs, tx);
3291 zfs_sa_upgrade_txholds(tx, zp);
3293 err = dmu_tx_assign(tx, TXG_WAIT);
3299 * Set each attribute requested.
3300 * We group settings according to the locks they need to acquire.
3302 * Note: you cannot set ctime directly, although it will be
3303 * updated as a side-effect of calling this function.
3306 if (mask & (AT_UID|AT_GID|AT_MODE))
3307 mutex_enter(&zp->z_acl_lock);
3309 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3310 &zp->z_pflags, sizeof (zp->z_pflags));
3313 if (mask & (AT_UID|AT_GID|AT_MODE))
3314 mutex_enter(&attrzp->z_acl_lock);
3315 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3316 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3317 sizeof (attrzp->z_pflags));
3320 if (mask & (AT_UID|AT_GID)) {
3322 if (mask & AT_UID) {
3323 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3324 &new_uid, sizeof (new_uid));
3325 zp->z_uid = new_uid;
3327 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3328 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3330 attrzp->z_uid = new_uid;
3334 if (mask & AT_GID) {
3335 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3336 NULL, &new_gid, sizeof (new_gid));
3337 zp->z_gid = new_gid;
3339 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3340 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3342 attrzp->z_gid = new_gid;
3345 if (!(mask & AT_MODE)) {
3346 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3347 NULL, &new_mode, sizeof (new_mode));
3348 new_mode = zp->z_mode;
3350 err = zfs_acl_chown_setattr(zp);
3353 err = zfs_acl_chown_setattr(attrzp);
3358 if (mask & AT_MODE) {
3359 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3360 &new_mode, sizeof (new_mode));
3361 zp->z_mode = new_mode;
3362 ASSERT3U((uintptr_t)aclp, !=, 0);
3363 err = zfs_aclset_common(zp, aclp, cr, tx);
3365 if (zp->z_acl_cached)
3366 zfs_acl_free(zp->z_acl_cached);
3367 zp->z_acl_cached = aclp;
3372 if (mask & AT_ATIME) {
3373 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3374 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3375 &zp->z_atime, sizeof (zp->z_atime));
3378 if (mask & AT_MTIME) {
3379 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3380 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3381 mtime, sizeof (mtime));
3384 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3385 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3386 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3387 NULL, mtime, sizeof (mtime));
3388 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3389 &ctime, sizeof (ctime));
3390 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3392 } else if (mask != 0) {
3393 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3394 &ctime, sizeof (ctime));
3395 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3398 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3399 SA_ZPL_CTIME(zfsvfs), NULL,
3400 &ctime, sizeof (ctime));
3401 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3402 mtime, ctime, B_TRUE);
3406 * Do this after setting timestamps to prevent timestamp
3407 * update from toggling bit
3410 if (xoap && (mask & AT_XVATTR)) {
3412 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME))
3413 xoap->xoa_createtime = vap->va_birthtime;
3415 * restore trimmed off masks
3416 * so that return masks can be set for caller.
3419 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3420 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3422 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3423 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3425 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3426 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3428 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3429 XVA_SET_REQ(xvap, XAT_NODUMP);
3431 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3432 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3434 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3435 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3438 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3439 ASSERT(vp->v_type == VREG);
3441 zfs_xvattr_set(zp, xvap, tx);
3445 zfs_fuid_sync(zfsvfs, tx);
3448 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3450 if (mask & (AT_UID|AT_GID|AT_MODE))
3451 mutex_exit(&zp->z_acl_lock);
3454 if (mask & (AT_UID|AT_GID|AT_MODE))
3455 mutex_exit(&attrzp->z_acl_lock);
3458 if (err == 0 && attrzp) {
3459 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3471 zfs_fuid_info_free(fuidp);
3478 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3483 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3484 zil_commit(zilog, 0);
3491 * We acquire all but fdvp locks using non-blocking acquisitions. If we
3492 * fail to acquire any lock in the path we will drop all held locks,
3493 * acquire the new lock in a blocking fashion, and then release it and
3494 * restart the rename. This acquire/release step ensures that we do not
3495 * spin on a lock waiting for release. On error release all vnode locks
3496 * and decrement references the way tmpfs_rename() would do.
3499 zfs_rename_relock(struct vnode *sdvp, struct vnode **svpp,
3500 struct vnode *tdvp, struct vnode **tvpp,
3501 const struct componentname *scnp, const struct componentname *tcnp)
3504 struct vnode *nvp, *svp, *tvp;
3505 znode_t *sdzp, *tdzp, *szp, *tzp;
3506 const char *snm = scnp->cn_nameptr;
3507 const char *tnm = tcnp->cn_nameptr;
3510 VOP_UNLOCK(tdvp, 0);
3511 if (*tvpp != NULL && *tvpp != tdvp)
3512 VOP_UNLOCK(*tvpp, 0);
3515 error = vn_lock(sdvp, LK_EXCLUSIVE);
3520 error = vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT);
3522 VOP_UNLOCK(sdvp, 0);
3525 error = vn_lock(tdvp, LK_EXCLUSIVE);
3528 VOP_UNLOCK(tdvp, 0);
3534 * Before using sdzp and tdzp we must ensure that they are live.
3535 * As a porting legacy from illumos we have two things to worry
3536 * about. One is typical for FreeBSD and it is that the vnode is
3537 * not reclaimed (doomed). The other is that the znode is live.
3538 * The current code can invalidate the znode without acquiring the
3539 * corresponding vnode lock if the object represented by the znode
3540 * and vnode is no longer valid after a rollback or receive operation.
3541 * z_teardown_lock hidden behind ZFS_ENTER and ZFS_EXIT is the lock
3542 * that protects the znodes from the invalidation.
3544 zfsvfs = sdzp->z_zfsvfs;
3545 ASSERT3P(zfsvfs, ==, tdzp->z_zfsvfs);
3549 * We can not use ZFS_VERIFY_ZP() here because it could directly return
3550 * bypassing the cleanup code in the case of an error.
3552 if (tdzp->z_sa_hdl == NULL || sdzp->z_sa_hdl == NULL) {
3554 VOP_UNLOCK(sdvp, 0);
3555 VOP_UNLOCK(tdvp, 0);
3556 error = SET_ERROR(EIO);
3561 * Re-resolve svp to be certain it still exists and fetch the
3564 error = zfs_dirent_lookup(sdzp, snm, &szp, ZEXISTS);
3566 /* Source entry invalid or not there. */
3568 VOP_UNLOCK(sdvp, 0);
3569 VOP_UNLOCK(tdvp, 0);
3570 if ((scnp->cn_flags & ISDOTDOT) != 0 ||
3571 (scnp->cn_namelen == 1 && scnp->cn_nameptr[0] == '.'))
3572 error = SET_ERROR(EINVAL);
3578 * Re-resolve tvp, if it disappeared we just carry on.
3580 error = zfs_dirent_lookup(tdzp, tnm, &tzp, 0);
3583 VOP_UNLOCK(sdvp, 0);
3584 VOP_UNLOCK(tdvp, 0);
3586 if ((tcnp->cn_flags & ISDOTDOT) != 0)
3587 error = SET_ERROR(EINVAL);
3596 * At present the vnode locks must be acquired before z_teardown_lock,
3597 * although it would be more logical to use the opposite order.
3602 * Now try acquire locks on svp and tvp.
3605 error = vn_lock(nvp, LK_EXCLUSIVE | LK_NOWAIT);
3607 VOP_UNLOCK(sdvp, 0);
3608 VOP_UNLOCK(tdvp, 0);
3611 if (error != EBUSY) {
3615 error = vn_lock(nvp, LK_EXCLUSIVE);
3622 * Concurrent rename race.
3627 error = SET_ERROR(EINVAL);
3642 error = vn_lock(nvp, LK_EXCLUSIVE | LK_NOWAIT);
3644 VOP_UNLOCK(sdvp, 0);
3645 VOP_UNLOCK(tdvp, 0);
3646 VOP_UNLOCK(*svpp, 0);
3647 if (error != EBUSY) {
3651 error = vn_lock(nvp, LK_EXCLUSIVE);
3669 * Note that we must use VRELE_ASYNC in this function as it walks
3670 * up the directory tree and vrele may need to acquire an exclusive
3671 * lock if a last reference to a vnode is dropped.
3674 zfs_rename_check(znode_t *szp, znode_t *sdzp, znode_t *tdzp)
3681 zfsvfs = tdzp->z_zfsvfs;
3683 return (SET_ERROR(EINVAL));
3686 if (tdzp->z_id == zfsvfs->z_root)
3690 ASSERT(!zp->z_unlinked);
3691 if ((error = sa_lookup(zp->z_sa_hdl,
3692 SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0)
3695 if (parent == szp->z_id) {
3696 error = SET_ERROR(EINVAL);
3699 if (parent == zfsvfs->z_root)
3701 if (parent == sdzp->z_id)
3704 error = zfs_zget(zfsvfs, parent, &zp1);
3709 VN_RELE_ASYNC(ZTOV(zp),
3710 dsl_pool_vnrele_taskq(dmu_objset_pool(zfsvfs->z_os)));
3714 if (error == ENOTDIR)
3715 panic("checkpath: .. not a directory\n");
3717 VN_RELE_ASYNC(ZTOV(zp),
3718 dsl_pool_vnrele_taskq(dmu_objset_pool(zfsvfs->z_os)));
3723 * Move an entry from the provided source directory to the target
3724 * directory. Change the entry name as indicated.
3726 * IN: sdvp - Source directory containing the "old entry".
3727 * snm - Old entry name.
3728 * tdvp - Target directory to contain the "new entry".
3729 * tnm - New entry name.
3730 * cr - credentials of caller.
3731 * ct - caller context
3732 * flags - case flags
3734 * RETURN: 0 on success, error code on failure.
3737 * sdvp,tdvp - ctime|mtime updated
3741 zfs_rename(vnode_t *sdvp, vnode_t **svpp, struct componentname *scnp,
3742 vnode_t *tdvp, vnode_t **tvpp, struct componentname *tcnp,
3746 znode_t *sdzp, *tdzp, *szp, *tzp;
3747 zilog_t *zilog = NULL;
3749 char *snm = scnp->cn_nameptr;
3750 char *tnm = tcnp->cn_nameptr;
3753 /* Reject renames across filesystems. */
3754 if ((*svpp)->v_mount != tdvp->v_mount ||
3755 ((*tvpp) != NULL && (*svpp)->v_mount != (*tvpp)->v_mount)) {
3756 error = SET_ERROR(EXDEV);
3760 if (zfsctl_is_node(tdvp)) {
3761 error = SET_ERROR(EXDEV);
3766 * Lock all four vnodes to ensure safety and semantics of renaming.
3768 error = zfs_rename_relock(sdvp, svpp, tdvp, tvpp, scnp, tcnp);
3770 /* no vnodes are locked in the case of error here */
3776 zfsvfs = tdzp->z_zfsvfs;
3777 zilog = zfsvfs->z_log;
3780 * After we re-enter ZFS_ENTER() we will have to revalidate all
3785 if (zfsvfs->z_utf8 && u8_validate(tnm,
3786 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3787 error = SET_ERROR(EILSEQ);
3791 /* If source and target are the same file, there is nothing to do. */
3792 if ((*svpp) == (*tvpp)) {
3797 if (((*svpp)->v_type == VDIR && (*svpp)->v_mountedhere != NULL) ||
3798 ((*tvpp) != NULL && (*tvpp)->v_type == VDIR &&
3799 (*tvpp)->v_mountedhere != NULL)) {
3800 error = SET_ERROR(EXDEV);
3805 * We can not use ZFS_VERIFY_ZP() here because it could directly return
3806 * bypassing the cleanup code in the case of an error.
3808 if (tdzp->z_sa_hdl == NULL || sdzp->z_sa_hdl == NULL) {
3809 error = SET_ERROR(EIO);
3814 tzp = *tvpp == NULL ? NULL : VTOZ(*tvpp);
3815 if (szp->z_sa_hdl == NULL || (tzp != NULL && tzp->z_sa_hdl == NULL)) {
3816 error = SET_ERROR(EIO);
3821 * This is to prevent the creation of links into attribute space
3822 * by renaming a linked file into/outof an attribute directory.
3823 * See the comment in zfs_link() for why this is considered bad.
3825 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3826 error = SET_ERROR(EINVAL);
3831 * Must have write access at the source to remove the old entry
3832 * and write access at the target to create the new entry.
3833 * Note that if target and source are the same, this can be
3834 * done in a single check.
3836 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3839 if ((*svpp)->v_type == VDIR) {
3841 * Avoid ".", "..", and aliases of "." for obvious reasons.
3843 if ((scnp->cn_namelen == 1 && scnp->cn_nameptr[0] == '.') ||
3845 (scnp->cn_flags | tcnp->cn_flags) & ISDOTDOT) {
3851 * Check to make sure rename is valid.
3852 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3854 if (error = zfs_rename_check(szp, sdzp, tdzp))
3859 * Does target exist?
3863 * Source and target must be the same type.
3865 if ((*svpp)->v_type == VDIR) {
3866 if ((*tvpp)->v_type != VDIR) {
3867 error = SET_ERROR(ENOTDIR);
3875 if ((*tvpp)->v_type == VDIR) {
3876 error = SET_ERROR(EISDIR);
3882 vnevent_rename_src(*svpp, sdvp, scnp->cn_nameptr, ct);
3884 vnevent_rename_dest(*tvpp, tdvp, tnm, ct);
3887 * notify the target directory if it is not the same
3888 * as source directory.
3891 vnevent_rename_dest_dir(tdvp, ct);
3894 tx = dmu_tx_create(zfsvfs->z_os);
3895 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3896 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3897 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3898 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3900 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3901 zfs_sa_upgrade_txholds(tx, tdzp);
3904 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3905 zfs_sa_upgrade_txholds(tx, tzp);
3908 zfs_sa_upgrade_txholds(tx, szp);
3909 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3910 error = dmu_tx_assign(tx, TXG_WAIT);
3917 if (tzp) /* Attempt to remove the existing target */
3918 error = zfs_link_destroy(tdzp, tnm, tzp, tx, 0, NULL);
3921 error = zfs_link_create(tdzp, tnm, szp, tx, ZRENAMING);
3923 szp->z_pflags |= ZFS_AV_MODIFIED;
3925 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3926 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3929 error = zfs_link_destroy(sdzp, snm, szp, tx, ZRENAMING,
3932 zfs_log_rename(zilog, tx, TX_RENAME, sdzp,
3933 snm, tdzp, tnm, szp);
3936 * Update path information for the target vnode
3938 vn_renamepath(tdvp, *svpp, tnm, strlen(tnm));
3941 * At this point, we have successfully created
3942 * the target name, but have failed to remove
3943 * the source name. Since the create was done
3944 * with the ZRENAMING flag, there are
3945 * complications; for one, the link count is
3946 * wrong. The easiest way to deal with this
3947 * is to remove the newly created target, and
3948 * return the original error. This must
3949 * succeed; fortunately, it is very unlikely to
3950 * fail, since we just created it.
3952 VERIFY3U(zfs_link_destroy(tdzp, tnm, szp, tx,
3953 ZRENAMING, NULL), ==, 0);
3960 cache_purge_negative(tdvp);
3966 unlockout: /* all 4 vnodes are locked, ZFS_ENTER called */
3968 VOP_UNLOCK(*svpp, 0);
3969 VOP_UNLOCK(sdvp, 0);
3971 out: /* original two vnodes are locked */
3972 if (error == 0 && zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3973 zil_commit(zilog, 0);
3976 VOP_UNLOCK(*tvpp, 0);
3978 VOP_UNLOCK(tdvp, 0);
3983 * Insert the indicated symbolic reference entry into the directory.
3985 * IN: dvp - Directory to contain new symbolic link.
3986 * link - Name for new symlink entry.
3987 * vap - Attributes of new entry.
3988 * cr - credentials of caller.
3989 * ct - caller context
3990 * flags - case flags
3992 * RETURN: 0 on success, error code on failure.
3995 * dvp - ctime|mtime updated
3999 zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link,
4000 cred_t *cr, kthread_t *td)
4002 znode_t *zp, *dzp = VTOZ(dvp);
4004 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4006 uint64_t len = strlen(link);
4008 zfs_acl_ids_t acl_ids;
4009 boolean_t fuid_dirtied;
4010 uint64_t txtype = TX_SYMLINK;
4013 ASSERT(vap->va_type == VLNK);
4017 zilog = zfsvfs->z_log;
4019 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
4020 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4022 return (SET_ERROR(EILSEQ));
4025 if (len > MAXPATHLEN) {
4027 return (SET_ERROR(ENAMETOOLONG));
4030 if ((error = zfs_acl_ids_create(dzp, 0,
4031 vap, cr, NULL, &acl_ids)) != 0) {
4037 * Attempt to lock directory; fail if entry already exists.
4039 error = zfs_dirent_lookup(dzp, name, &zp, ZNEW);
4041 zfs_acl_ids_free(&acl_ids);
4046 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4047 zfs_acl_ids_free(&acl_ids);
4052 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
4053 zfs_acl_ids_free(&acl_ids);
4055 return (SET_ERROR(EDQUOT));
4058 getnewvnode_reserve(1);
4059 tx = dmu_tx_create(zfsvfs->z_os);
4060 fuid_dirtied = zfsvfs->z_fuid_dirty;
4061 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
4062 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4063 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
4064 ZFS_SA_BASE_ATTR_SIZE + len);
4065 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
4066 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
4067 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
4068 acl_ids.z_aclp->z_acl_bytes);
4071 zfs_fuid_txhold(zfsvfs, tx);
4072 error = dmu_tx_assign(tx, TXG_WAIT);
4074 zfs_acl_ids_free(&acl_ids);
4076 getnewvnode_drop_reserve();
4082 * Create a new object for the symlink.
4083 * for version 4 ZPL datsets the symlink will be an SA attribute
4085 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
4088 zfs_fuid_sync(zfsvfs, tx);
4091 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
4094 zfs_sa_symlink(zp, link, len, tx);
4097 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
4098 &zp->z_size, sizeof (zp->z_size), tx);
4100 * Insert the new object into the directory.
4102 (void) zfs_link_create(dzp, name, zp, tx, ZNEW);
4104 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
4107 zfs_acl_ids_free(&acl_ids);
4111 getnewvnode_drop_reserve();
4113 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4114 zil_commit(zilog, 0);
4121 * Return, in the buffer contained in the provided uio structure,
4122 * the symbolic path referred to by vp.
4124 * IN: vp - vnode of symbolic link.
4125 * uio - structure to contain the link path.
4126 * cr - credentials of caller.
4127 * ct - caller context
4129 * OUT: uio - structure containing the link path.
4131 * RETURN: 0 on success, error code on failure.
4134 * vp - atime updated
4138 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
4140 znode_t *zp = VTOZ(vp);
4141 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4148 error = sa_lookup_uio(zp->z_sa_hdl,
4149 SA_ZPL_SYMLINK(zfsvfs), uio);
4151 error = zfs_sa_readlink(zp, uio);
4153 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4160 * Insert a new entry into directory tdvp referencing svp.
4162 * IN: tdvp - Directory to contain new entry.
4163 * svp - vnode of new entry.
4164 * name - name of new entry.
4165 * cr - credentials of caller.
4166 * ct - caller context
4168 * RETURN: 0 on success, error code on failure.
4171 * tdvp - ctime|mtime updated
4172 * svp - ctime updated
4176 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
4177 caller_context_t *ct, int flags)
4179 znode_t *dzp = VTOZ(tdvp);
4181 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4188 ASSERT(tdvp->v_type == VDIR);
4192 zilog = zfsvfs->z_log;
4195 * POSIX dictates that we return EPERM here.
4196 * Better choices include ENOTSUP or EISDIR.
4198 if (svp->v_type == VDIR) {
4200 return (SET_ERROR(EPERM));
4206 if (szp->z_pflags & (ZFS_APPENDONLY | ZFS_IMMUTABLE | ZFS_READONLY)) {
4208 return (SET_ERROR(EPERM));
4211 /* Prevent links to .zfs/shares files */
4213 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4214 &parent, sizeof (uint64_t))) != 0) {
4218 if (parent == zfsvfs->z_shares_dir) {
4220 return (SET_ERROR(EPERM));
4223 if (zfsvfs->z_utf8 && u8_validate(name,
4224 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4226 return (SET_ERROR(EILSEQ));
4230 * We do not support links between attributes and non-attributes
4231 * because of the potential security risk of creating links
4232 * into "normal" file space in order to circumvent restrictions
4233 * imposed in attribute space.
4235 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4237 return (SET_ERROR(EINVAL));
4241 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4242 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) {
4244 return (SET_ERROR(EPERM));
4247 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4253 * Attempt to lock directory; fail if entry already exists.
4255 error = zfs_dirent_lookup(dzp, name, &tzp, ZNEW);
4261 tx = dmu_tx_create(zfsvfs->z_os);
4262 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4263 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4264 zfs_sa_upgrade_txholds(tx, szp);
4265 zfs_sa_upgrade_txholds(tx, dzp);
4266 error = dmu_tx_assign(tx, TXG_WAIT);
4273 error = zfs_link_create(dzp, name, szp, tx, 0);
4276 uint64_t txtype = TX_LINK;
4277 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4283 vnevent_link(svp, ct);
4286 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4287 zil_commit(zilog, 0);
4296 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4298 znode_t *zp = VTOZ(vp);
4299 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4302 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4303 if (zp->z_sa_hdl == NULL) {
4305 * The fs has been unmounted, or we did a
4306 * suspend/resume and this file no longer exists.
4308 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4313 if (zp->z_unlinked) {
4315 * Fast path to recycle a vnode of a removed file.
4317 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4322 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4323 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4325 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4326 zfs_sa_upgrade_txholds(tx, zp);
4327 error = dmu_tx_assign(tx, TXG_WAIT);
4331 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4332 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4333 zp->z_atime_dirty = 0;
4337 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4341 CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid));
4342 CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid));
4346 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
4348 znode_t *zp = VTOZ(vp);
4349 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4352 uint64_t object = zp->z_id;
4359 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
4360 &gen64, sizeof (uint64_t))) != 0) {
4365 gen = (uint32_t)gen64;
4367 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
4370 if (fidp->fid_len < size) {
4371 fidp->fid_len = size;
4373 return (SET_ERROR(ENOSPC));
4376 fidp->fid_len = size;
4379 zfid = (zfid_short_t *)fidp;
4381 zfid->zf_len = size;
4383 for (i = 0; i < sizeof (zfid->zf_object); i++)
4384 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4386 /* Must have a non-zero generation number to distinguish from .zfs */
4389 for (i = 0; i < sizeof (zfid->zf_gen); i++)
4390 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4392 if (size == LONG_FID_LEN) {
4393 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
4396 zlfid = (zfid_long_t *)fidp;
4398 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4399 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4401 /* XXX - this should be the generation number for the objset */
4402 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4403 zlfid->zf_setgen[i] = 0;
4411 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4412 caller_context_t *ct)
4423 case _PC_FILESIZEBITS:
4427 case _PC_XATTR_EXISTS:
4429 zfsvfs = zp->z_zfsvfs;
4433 error = zfs_dirent_lookup(zp, "", &xzp,
4434 ZXATTR | ZEXISTS | ZSHARED);
4436 if (!zfs_dirempty(xzp))
4439 } else if (error == ENOENT) {
4441 * If there aren't extended attributes, it's the
4442 * same as having zero of them.
4449 case _PC_SATTR_ENABLED:
4450 case _PC_SATTR_EXISTS:
4451 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
4452 (vp->v_type == VREG || vp->v_type == VDIR);
4455 case _PC_ACCESS_FILTERING:
4456 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
4460 case _PC_ACL_ENABLED:
4461 *valp = _ACL_ACE_ENABLED;
4463 #endif /* illumos */
4464 case _PC_MIN_HOLE_SIZE:
4465 *valp = (int)SPA_MINBLOCKSIZE;
4468 case _PC_TIMESTAMP_RESOLUTION:
4469 /* nanosecond timestamp resolution */
4473 case _PC_ACL_EXTENDED:
4481 case _PC_ACL_PATH_MAX:
4482 *valp = ACL_MAX_ENTRIES;
4486 return (EOPNOTSUPP);
4492 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4493 caller_context_t *ct)
4495 znode_t *zp = VTOZ(vp);
4496 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4498 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4502 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4510 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4511 caller_context_t *ct)
4513 znode_t *zp = VTOZ(vp);
4514 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4516 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4517 zilog_t *zilog = zfsvfs->z_log;
4522 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
4524 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4525 zil_commit(zilog, 0);
4532 ioflags(int ioflags)
4536 if (ioflags & IO_APPEND)
4538 if (ioflags & IO_NDELAY)
4540 if (ioflags & IO_SYNC)
4541 flags |= (FSYNC | FDSYNC | FRSYNC);
4547 zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int *rbehind,
4550 znode_t *zp = VTOZ(vp);
4551 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4552 objset_t *os = zp->z_zfsvfs->z_os;
4557 off_t startoff, endoff;
4559 vm_pindex_t reqstart, reqend;
4562 object = m[0]->object;
4568 zfs_vmobject_wlock(object);
4569 if (m[count - 1]->valid != 0 && --count == 0) {
4570 zfs_vmobject_wunlock(object);
4574 mlast = m[count - 1];
4576 if (IDX_TO_OFF(mlast->pindex) >=
4577 object->un_pager.vnp.vnp_size) {
4578 zfs_vmobject_wunlock(object);
4580 return (zfs_vm_pagerret_bad);
4583 VM_CNT_INC(v_vnodein);
4584 VM_CNT_ADD(v_vnodepgsin, count);
4587 if (IDX_TO_OFF(mlast->pindex) + lsize > object->un_pager.vnp.vnp_size)
4588 lsize = object->un_pager.vnp.vnp_size -
4589 IDX_TO_OFF(mlast->pindex);
4590 zfs_vmobject_wunlock(object);
4592 for (i = 0; i < count; i++) {
4596 va = zfs_map_page(m[i], &sf);
4597 error = dmu_read(os, zp->z_id, IDX_TO_OFF(m[i]->pindex),
4598 size, va, DMU_READ_PREFETCH);
4599 if (size != PAGE_SIZE)
4600 bzero(va + size, PAGE_SIZE - size);
4606 zfs_vmobject_wlock(object);
4607 for (i = 0; i < count; i++)
4608 m[i]->valid = VM_PAGE_BITS_ALL;
4609 zfs_vmobject_wunlock(object);
4612 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4619 return (zfs_vm_pagerret_ok);
4621 return (zfs_vm_pagerret_error);
4625 zfs_freebsd_getpages(ap)
4626 struct vop_getpages_args /* {
4635 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_rbehind,
4640 zfs_putpages(struct vnode *vp, vm_page_t *ma, size_t len, int flags,
4643 znode_t *zp = VTOZ(vp);
4644 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4653 vm_ooffset_t lo_off;
4664 object = vp->v_object;
4668 KASSERT(ma[0]->object == object, ("mismatching object"));
4669 KASSERT(len > 0 && (len & PAGE_MASK) == 0, ("unexpected length"));
4671 for (i = 0; i < pcount; i++)
4672 rtvals[i] = zfs_vm_pagerret_error;
4674 off = IDX_TO_OFF(ma[0]->pindex);
4675 blksz = zp->z_blksz;
4676 lo_off = rounddown(off, blksz);
4677 lo_len = roundup(len + (off - lo_off), blksz);
4678 rl = zfs_range_lock(zp, lo_off, lo_len, RL_WRITER);
4680 zfs_vmobject_wlock(object);
4681 if (len + off > object->un_pager.vnp.vnp_size) {
4682 if (object->un_pager.vnp.vnp_size > off) {
4685 len = object->un_pager.vnp.vnp_size - off;
4687 if ((pgoff = (int)len & PAGE_MASK) != 0) {
4689 * If the object is locked and the following
4690 * conditions hold, then the page's dirty
4691 * field cannot be concurrently changed by a
4695 vm_page_assert_sbusied(m);
4696 KASSERT(!pmap_page_is_write_mapped(m),
4697 ("zfs_putpages: page %p is not read-only", m));
4698 vm_page_clear_dirty(m, pgoff, PAGE_SIZE -
4705 if (ncount < pcount) {
4706 for (i = ncount; i < pcount; i++) {
4707 rtvals[i] = zfs_vm_pagerret_bad;
4711 zfs_vmobject_wunlock(object);
4716 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4717 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4721 tx = dmu_tx_create(zfsvfs->z_os);
4722 dmu_tx_hold_write(tx, zp->z_id, off, len);
4724 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4725 zfs_sa_upgrade_txholds(tx, zp);
4726 err = dmu_tx_assign(tx, TXG_WAIT);
4732 if (zp->z_blksz < PAGE_SIZE) {
4734 for (i = 0; len > 0; off += tocopy, len -= tocopy, i++) {
4735 tocopy = len > PAGE_SIZE ? PAGE_SIZE : len;
4736 va = zfs_map_page(ma[i], &sf);
4737 dmu_write(zfsvfs->z_os, zp->z_id, off, tocopy, va, tx);
4741 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, ma, tx);
4745 uint64_t mtime[2], ctime[2];
4746 sa_bulk_attr_t bulk[3];
4749 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4751 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4753 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4755 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4757 err = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
4759 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4761 zfs_vmobject_wlock(object);
4762 for (i = 0; i < ncount; i++) {
4763 rtvals[i] = zfs_vm_pagerret_ok;
4764 vm_page_undirty(ma[i]);
4766 zfs_vmobject_wunlock(object);
4767 VM_CNT_INC(v_vnodeout);
4768 VM_CNT_ADD(v_vnodepgsout, ncount);
4773 zfs_range_unlock(rl);
4774 if ((flags & (zfs_vm_pagerput_sync | zfs_vm_pagerput_inval)) != 0 ||
4775 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4776 zil_commit(zfsvfs->z_log, zp->z_id);
4782 zfs_freebsd_putpages(ap)
4783 struct vop_putpages_args /* {
4792 return (zfs_putpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_sync,
4797 zfs_freebsd_bmap(ap)
4798 struct vop_bmap_args /* {
4801 struct bufobj **a_bop;
4808 if (ap->a_bop != NULL)
4809 *ap->a_bop = &ap->a_vp->v_bufobj;
4810 if (ap->a_bnp != NULL)
4811 *ap->a_bnp = ap->a_bn;
4812 if (ap->a_runp != NULL)
4814 if (ap->a_runb != NULL)
4821 zfs_freebsd_open(ap)
4822 struct vop_open_args /* {
4825 struct ucred *a_cred;
4826 struct thread *a_td;
4829 vnode_t *vp = ap->a_vp;
4830 znode_t *zp = VTOZ(vp);
4833 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL);
4835 vnode_create_vobject(vp, zp->z_size, ap->a_td);
4840 zfs_freebsd_close(ap)
4841 struct vop_close_args /* {
4844 struct ucred *a_cred;
4845 struct thread *a_td;
4849 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL));
4853 zfs_freebsd_ioctl(ap)
4854 struct vop_ioctl_args /* {
4864 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data,
4865 ap->a_fflag, ap->a_cred, NULL, NULL));
4869 zfs_freebsd_read(ap)
4870 struct vop_read_args /* {
4874 struct ucred *a_cred;
4878 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
4883 zfs_freebsd_write(ap)
4884 struct vop_write_args /* {
4888 struct ucred *a_cred;
4892 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
4897 zfs_freebsd_access(ap)
4898 struct vop_access_args /* {
4900 accmode_t a_accmode;
4901 struct ucred *a_cred;
4902 struct thread *a_td;
4905 vnode_t *vp = ap->a_vp;
4906 znode_t *zp = VTOZ(vp);
4911 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND,
4913 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND);
4915 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL);
4918 * VADMIN has to be handled by vaccess().
4921 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND);
4923 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
4924 zp->z_gid, accmode, ap->a_cred, NULL);
4929 * For VEXEC, ensure that at least one execute bit is set for
4932 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR &&
4933 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
4941 zfs_freebsd_lookup(ap)
4942 struct vop_lookup_args /* {
4943 struct vnode *a_dvp;
4944 struct vnode **a_vpp;
4945 struct componentname *a_cnp;
4948 struct componentname *cnp = ap->a_cnp;
4949 char nm[NAME_MAX + 1];
4951 ASSERT(cnp->cn_namelen < sizeof(nm));
4952 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm)));
4954 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop,
4955 cnp->cn_cred, cnp->cn_thread, 0));
4959 zfs_cache_lookup(ap)
4960 struct vop_lookup_args /* {
4961 struct vnode *a_dvp;
4962 struct vnode **a_vpp;
4963 struct componentname *a_cnp;
4968 zfsvfs = ap->a_dvp->v_mount->mnt_data;
4969 if (zfsvfs->z_use_namecache)
4970 return (vfs_cache_lookup(ap));
4972 return (zfs_freebsd_lookup(ap));
4976 zfs_freebsd_create(ap)
4977 struct vop_create_args /* {
4978 struct vnode *a_dvp;
4979 struct vnode **a_vpp;
4980 struct componentname *a_cnp;
4981 struct vattr *a_vap;
4985 struct componentname *cnp = ap->a_cnp;
4986 vattr_t *vap = ap->a_vap;
4989 ASSERT(cnp->cn_flags & SAVENAME);
4991 vattr_init_mask(vap);
4992 mode = vap->va_mode & ALLPERMS;
4993 zfsvfs = ap->a_dvp->v_mount->mnt_data;
4995 error = zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode,
4996 ap->a_vpp, cnp->cn_cred, cnp->cn_thread);
4997 if (zfsvfs->z_use_namecache &&
4998 error == 0 && (cnp->cn_flags & MAKEENTRY) != 0)
4999 cache_enter(ap->a_dvp, *ap->a_vpp, cnp);
5004 zfs_freebsd_remove(ap)
5005 struct vop_remove_args /* {
5006 struct vnode *a_dvp;
5008 struct componentname *a_cnp;
5012 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
5014 return (zfs_remove(ap->a_dvp, ap->a_vp, ap->a_cnp->cn_nameptr,
5015 ap->a_cnp->cn_cred));
5019 zfs_freebsd_mkdir(ap)
5020 struct vop_mkdir_args /* {
5021 struct vnode *a_dvp;
5022 struct vnode **a_vpp;
5023 struct componentname *a_cnp;
5024 struct vattr *a_vap;
5027 vattr_t *vap = ap->a_vap;
5029 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
5031 vattr_init_mask(vap);
5033 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp,
5034 ap->a_cnp->cn_cred));
5038 zfs_freebsd_rmdir(ap)
5039 struct vop_rmdir_args /* {
5040 struct vnode *a_dvp;
5042 struct componentname *a_cnp;
5045 struct componentname *cnp = ap->a_cnp;
5047 ASSERT(cnp->cn_flags & SAVENAME);
5049 return (zfs_rmdir(ap->a_dvp, ap->a_vp, cnp->cn_nameptr, cnp->cn_cred));
5053 zfs_freebsd_readdir(ap)
5054 struct vop_readdir_args /* {
5057 struct ucred *a_cred;
5064 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag,
5065 ap->a_ncookies, ap->a_cookies));
5069 zfs_freebsd_fsync(ap)
5070 struct vop_fsync_args /* {
5073 struct thread *a_td;
5078 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL));
5082 zfs_freebsd_getattr(ap)
5083 struct vop_getattr_args /* {
5085 struct vattr *a_vap;
5086 struct ucred *a_cred;
5089 vattr_t *vap = ap->a_vap;
5095 xvap.xva_vattr = *vap;
5096 xvap.xva_vattr.va_mask |= AT_XVATTR;
5098 /* Convert chflags into ZFS-type flags. */
5099 /* XXX: what about SF_SETTABLE?. */
5100 XVA_SET_REQ(&xvap, XAT_IMMUTABLE);
5101 XVA_SET_REQ(&xvap, XAT_APPENDONLY);
5102 XVA_SET_REQ(&xvap, XAT_NOUNLINK);
5103 XVA_SET_REQ(&xvap, XAT_NODUMP);
5104 XVA_SET_REQ(&xvap, XAT_READONLY);
5105 XVA_SET_REQ(&xvap, XAT_ARCHIVE);
5106 XVA_SET_REQ(&xvap, XAT_SYSTEM);
5107 XVA_SET_REQ(&xvap, XAT_HIDDEN);
5108 XVA_SET_REQ(&xvap, XAT_REPARSE);
5109 XVA_SET_REQ(&xvap, XAT_OFFLINE);
5110 XVA_SET_REQ(&xvap, XAT_SPARSE);
5112 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL);
5116 /* Convert ZFS xattr into chflags. */
5117 #define FLAG_CHECK(fflag, xflag, xfield) do { \
5118 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \
5119 fflags |= (fflag); \
5121 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE,
5122 xvap.xva_xoptattrs.xoa_immutable);
5123 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY,
5124 xvap.xva_xoptattrs.xoa_appendonly);
5125 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK,
5126 xvap.xva_xoptattrs.xoa_nounlink);
5127 FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE,
5128 xvap.xva_xoptattrs.xoa_archive);
5129 FLAG_CHECK(UF_NODUMP, XAT_NODUMP,
5130 xvap.xva_xoptattrs.xoa_nodump);
5131 FLAG_CHECK(UF_READONLY, XAT_READONLY,
5132 xvap.xva_xoptattrs.xoa_readonly);
5133 FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM,
5134 xvap.xva_xoptattrs.xoa_system);
5135 FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN,
5136 xvap.xva_xoptattrs.xoa_hidden);
5137 FLAG_CHECK(UF_REPARSE, XAT_REPARSE,
5138 xvap.xva_xoptattrs.xoa_reparse);
5139 FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE,
5140 xvap.xva_xoptattrs.xoa_offline);
5141 FLAG_CHECK(UF_SPARSE, XAT_SPARSE,
5142 xvap.xva_xoptattrs.xoa_sparse);
5145 *vap = xvap.xva_vattr;
5146 vap->va_flags = fflags;
5151 zfs_freebsd_setattr(ap)
5152 struct vop_setattr_args /* {
5154 struct vattr *a_vap;
5155 struct ucred *a_cred;
5158 vnode_t *vp = ap->a_vp;
5159 vattr_t *vap = ap->a_vap;
5160 cred_t *cred = ap->a_cred;
5165 vattr_init_mask(vap);
5166 vap->va_mask &= ~AT_NOSET;
5169 xvap.xva_vattr = *vap;
5171 zflags = VTOZ(vp)->z_pflags;
5173 if (vap->va_flags != VNOVAL) {
5174 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs;
5177 if (zfsvfs->z_use_fuids == B_FALSE)
5178 return (EOPNOTSUPP);
5180 fflags = vap->va_flags;
5183 * We need to figure out whether it makes sense to allow
5184 * UF_REPARSE through, since we don't really have other
5185 * facilities to handle reparse points and zfs_setattr()
5186 * doesn't currently allow setting that attribute anyway.
5188 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE|
5189 UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE|
5190 UF_OFFLINE|UF_SPARSE)) != 0)
5191 return (EOPNOTSUPP);
5193 * Unprivileged processes are not permitted to unset system
5194 * flags, or modify flags if any system flags are set.
5195 * Privileged non-jail processes may not modify system flags
5196 * if securelevel > 0 and any existing system flags are set.
5197 * Privileged jail processes behave like privileged non-jail
5198 * processes if the security.jail.chflags_allowed sysctl is
5199 * is non-zero; otherwise, they behave like unprivileged
5202 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 ||
5203 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) {
5205 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
5206 error = securelevel_gt(cred, 0);
5212 * Callers may only modify the file flags on objects they
5213 * have VADMIN rights for.
5215 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0)
5218 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
5222 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) {
5227 #define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \
5228 if (((fflags & (fflag)) && !(zflags & (zflag))) || \
5229 ((zflags & (zflag)) && !(fflags & (fflag)))) { \
5230 XVA_SET_REQ(&xvap, (xflag)); \
5231 (xfield) = ((fflags & (fflag)) != 0); \
5234 /* Convert chflags into ZFS-type flags. */
5235 /* XXX: what about SF_SETTABLE?. */
5236 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE,
5237 xvap.xva_xoptattrs.xoa_immutable);
5238 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY,
5239 xvap.xva_xoptattrs.xoa_appendonly);
5240 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK,
5241 xvap.xva_xoptattrs.xoa_nounlink);
5242 FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE,
5243 xvap.xva_xoptattrs.xoa_archive);
5244 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP,
5245 xvap.xva_xoptattrs.xoa_nodump);
5246 FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY,
5247 xvap.xva_xoptattrs.xoa_readonly);
5248 FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM,
5249 xvap.xva_xoptattrs.xoa_system);
5250 FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN,
5251 xvap.xva_xoptattrs.xoa_hidden);
5252 FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE,
5253 xvap.xva_xoptattrs.xoa_hidden);
5254 FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE,
5255 xvap.xva_xoptattrs.xoa_offline);
5256 FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE,
5257 xvap.xva_xoptattrs.xoa_sparse);
5260 if (vap->va_birthtime.tv_sec != VNOVAL) {
5261 xvap.xva_vattr.va_mask |= AT_XVATTR;
5262 XVA_SET_REQ(&xvap, XAT_CREATETIME);
5264 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL));
5268 zfs_freebsd_rename(ap)
5269 struct vop_rename_args /* {
5270 struct vnode *a_fdvp;
5271 struct vnode *a_fvp;
5272 struct componentname *a_fcnp;
5273 struct vnode *a_tdvp;
5274 struct vnode *a_tvp;
5275 struct componentname *a_tcnp;
5278 vnode_t *fdvp = ap->a_fdvp;
5279 vnode_t *fvp = ap->a_fvp;
5280 vnode_t *tdvp = ap->a_tdvp;
5281 vnode_t *tvp = ap->a_tvp;
5284 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART));
5285 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART));
5287 error = zfs_rename(fdvp, &fvp, ap->a_fcnp, tdvp, &tvp,
5288 ap->a_tcnp, ap->a_fcnp->cn_cred);
5300 zfs_freebsd_symlink(ap)
5301 struct vop_symlink_args /* {
5302 struct vnode *a_dvp;
5303 struct vnode **a_vpp;
5304 struct componentname *a_cnp;
5305 struct vattr *a_vap;
5309 struct componentname *cnp = ap->a_cnp;
5310 vattr_t *vap = ap->a_vap;
5312 ASSERT(cnp->cn_flags & SAVENAME);
5314 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */
5315 vattr_init_mask(vap);
5317 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap,
5318 ap->a_target, cnp->cn_cred, cnp->cn_thread));
5322 zfs_freebsd_readlink(ap)
5323 struct vop_readlink_args /* {
5326 struct ucred *a_cred;
5330 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL));
5334 zfs_freebsd_link(ap)
5335 struct vop_link_args /* {
5336 struct vnode *a_tdvp;
5338 struct componentname *a_cnp;
5341 struct componentname *cnp = ap->a_cnp;
5342 vnode_t *vp = ap->a_vp;
5343 vnode_t *tdvp = ap->a_tdvp;
5345 if (tdvp->v_mount != vp->v_mount)
5348 ASSERT(cnp->cn_flags & SAVENAME);
5350 return (zfs_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0));
5354 zfs_freebsd_inactive(ap)
5355 struct vop_inactive_args /* {
5357 struct thread *a_td;
5360 vnode_t *vp = ap->a_vp;
5362 zfs_inactive(vp, ap->a_td->td_ucred, NULL);
5367 zfs_freebsd_reclaim(ap)
5368 struct vop_reclaim_args /* {
5370 struct thread *a_td;
5373 vnode_t *vp = ap->a_vp;
5374 znode_t *zp = VTOZ(vp);
5375 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5379 /* Destroy the vm object and flush associated pages. */
5380 vnode_destroy_vobject(vp);
5383 * z_teardown_inactive_lock protects from a race with
5384 * zfs_znode_dmu_fini in zfsvfs_teardown during
5387 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
5388 if (zp->z_sa_hdl == NULL)
5392 rw_exit(&zfsvfs->z_teardown_inactive_lock);
5400 struct vop_fid_args /* {
5406 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL));
5410 zfs_freebsd_pathconf(ap)
5411 struct vop_pathconf_args /* {
5414 register_t *a_retval;
5420 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL);
5422 *ap->a_retval = val;
5423 else if (error == EOPNOTSUPP)
5424 error = vop_stdpathconf(ap);
5429 zfs_freebsd_fifo_pathconf(ap)
5430 struct vop_pathconf_args /* {
5433 register_t *a_retval;
5437 switch (ap->a_name) {
5438 case _PC_ACL_EXTENDED:
5440 case _PC_ACL_PATH_MAX:
5441 case _PC_MAC_PRESENT:
5442 return (zfs_freebsd_pathconf(ap));
5444 return (fifo_specops.vop_pathconf(ap));
5449 * FreeBSD's extended attributes namespace defines file name prefix for ZFS'
5450 * extended attribute name:
5453 * system freebsd:system:
5454 * user (none, can be used to access ZFS fsattr(5) attributes
5455 * created on Solaris)
5458 zfs_create_attrname(int attrnamespace, const char *name, char *attrname,
5461 const char *namespace, *prefix, *suffix;
5463 /* We don't allow '/' character in attribute name. */
5464 if (strchr(name, '/') != NULL)
5466 /* We don't allow attribute names that start with "freebsd:" string. */
5467 if (strncmp(name, "freebsd:", 8) == 0)
5470 bzero(attrname, size);
5472 switch (attrnamespace) {
5473 case EXTATTR_NAMESPACE_USER:
5475 prefix = "freebsd:";
5476 namespace = EXTATTR_NAMESPACE_USER_STRING;
5480 * This is the default namespace by which we can access all
5481 * attributes created on Solaris.
5483 prefix = namespace = suffix = "";
5486 case EXTATTR_NAMESPACE_SYSTEM:
5487 prefix = "freebsd:";
5488 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING;
5491 case EXTATTR_NAMESPACE_EMPTY:
5495 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix,
5497 return (ENAMETOOLONG);
5503 * Vnode operating to retrieve a named extended attribute.
5506 zfs_getextattr(struct vop_getextattr_args *ap)
5509 IN struct vnode *a_vp;
5510 IN int a_attrnamespace;
5511 IN const char *a_name;
5512 INOUT struct uio *a_uio;
5514 IN struct ucred *a_cred;
5515 IN struct thread *a_td;
5519 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
5520 struct thread *td = ap->a_td;
5521 struct nameidata nd;
5524 vnode_t *xvp = NULL, *vp;
5527 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
5528 ap->a_cred, ap->a_td, VREAD);
5532 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
5539 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
5547 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
5549 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL);
5551 NDFREE(&nd, NDF_ONLY_PNBUF);
5554 if (error == ENOENT)
5559 if (ap->a_size != NULL) {
5560 error = VOP_GETATTR(vp, &va, ap->a_cred);
5562 *ap->a_size = (size_t)va.va_size;
5563 } else if (ap->a_uio != NULL)
5564 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred);
5567 vn_close(vp, flags, ap->a_cred, td);
5574 * Vnode operation to remove a named attribute.
5577 zfs_deleteextattr(struct vop_deleteextattr_args *ap)
5580 IN struct vnode *a_vp;
5581 IN int a_attrnamespace;
5582 IN const char *a_name;
5583 IN struct ucred *a_cred;
5584 IN struct thread *a_td;
5588 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
5589 struct thread *td = ap->a_td;
5590 struct nameidata nd;
5593 vnode_t *xvp = NULL, *vp;
5596 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
5597 ap->a_cred, ap->a_td, VWRITE);
5601 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
5608 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
5615 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF,
5616 UIO_SYSSPACE, attrname, xvp, td);
5621 NDFREE(&nd, NDF_ONLY_PNBUF);
5622 if (error == ENOENT)
5627 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
5628 NDFREE(&nd, NDF_ONLY_PNBUF);
5631 if (vp == nd.ni_dvp)
5641 * Vnode operation to set a named attribute.
5644 zfs_setextattr(struct vop_setextattr_args *ap)
5647 IN struct vnode *a_vp;
5648 IN int a_attrnamespace;
5649 IN const char *a_name;
5650 INOUT struct uio *a_uio;
5651 IN struct ucred *a_cred;
5652 IN struct thread *a_td;
5656 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
5657 struct thread *td = ap->a_td;
5658 struct nameidata nd;
5661 vnode_t *xvp = NULL, *vp;
5664 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
5665 ap->a_cred, ap->a_td, VWRITE);
5669 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
5676 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
5677 LOOKUP_XATTR | CREATE_XATTR_DIR);
5683 flags = FFLAGS(O_WRONLY | O_CREAT);
5684 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
5686 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL);
5688 NDFREE(&nd, NDF_ONLY_PNBUF);
5696 error = VOP_SETATTR(vp, &va, ap->a_cred);
5698 VOP_WRITE(vp, ap->a_uio, IO_UNIT, ap->a_cred);
5701 vn_close(vp, flags, ap->a_cred, td);
5708 * Vnode operation to retrieve extended attributes on a vnode.
5711 zfs_listextattr(struct vop_listextattr_args *ap)
5714 IN struct vnode *a_vp;
5715 IN int a_attrnamespace;
5716 INOUT struct uio *a_uio;
5718 IN struct ucred *a_cred;
5719 IN struct thread *a_td;
5723 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
5724 struct thread *td = ap->a_td;
5725 struct nameidata nd;
5726 char attrprefix[16];
5727 u_char dirbuf[sizeof(struct dirent)];
5730 struct uio auio, *uio = ap->a_uio;
5731 size_t *sizep = ap->a_size;
5733 vnode_t *xvp = NULL, *vp;
5734 int done, error, eof, pos;
5736 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
5737 ap->a_cred, ap->a_td, VREAD);
5741 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix,
5742 sizeof(attrprefix));
5745 plen = strlen(attrprefix);
5752 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
5757 * ENOATTR means that the EA directory does not yet exist,
5758 * i.e. there are no extended attributes there.
5760 if (error == ENOATTR)
5765 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED,
5766 UIO_SYSSPACE, ".", xvp, td);
5769 NDFREE(&nd, NDF_ONLY_PNBUF);
5775 auio.uio_iov = &aiov;
5776 auio.uio_iovcnt = 1;
5777 auio.uio_segflg = UIO_SYSSPACE;
5779 auio.uio_rw = UIO_READ;
5780 auio.uio_offset = 0;
5785 aiov.iov_base = (void *)dirbuf;
5786 aiov.iov_len = sizeof(dirbuf);
5787 auio.uio_resid = sizeof(dirbuf);
5788 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL);
5789 done = sizeof(dirbuf) - auio.uio_resid;
5792 for (pos = 0; pos < done;) {
5793 dp = (struct dirent *)(dirbuf + pos);
5794 pos += dp->d_reclen;
5796 * XXX: Temporarily we also accept DT_UNKNOWN, as this
5797 * is what we get when attribute was created on Solaris.
5799 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN)
5801 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0)
5803 else if (strncmp(dp->d_name, attrprefix, plen) != 0)
5805 nlen = dp->d_namlen - plen;
5808 else if (uio != NULL) {
5810 * Format of extattr name entry is one byte for
5811 * length and the rest for name.
5813 error = uiomove(&nlen, 1, uio->uio_rw, uio);
5815 error = uiomove(dp->d_name + plen, nlen,
5822 } while (!eof && error == 0);
5831 zfs_freebsd_getacl(ap)
5832 struct vop_getacl_args /* {
5841 vsecattr_t vsecattr;
5843 if (ap->a_type != ACL_TYPE_NFS4)
5846 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
5847 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL))
5850 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt);
5851 if (vsecattr.vsa_aclentp != NULL)
5852 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz);
5858 zfs_freebsd_setacl(ap)
5859 struct vop_setacl_args /* {
5868 vsecattr_t vsecattr;
5869 int aclbsize; /* size of acl list in bytes */
5872 if (ap->a_type != ACL_TYPE_NFS4)
5875 if (ap->a_aclp == NULL)
5878 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES)
5882 * With NFSv4 ACLs, chmod(2) may need to add additional entries,
5883 * splitting every entry into two and appending "canonical six"
5884 * entries at the end. Don't allow for setting an ACL that would
5885 * cause chmod(2) to run out of ACL entries.
5887 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES)
5890 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR);
5894 vsecattr.vsa_mask = VSA_ACE;
5895 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t);
5896 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP);
5897 aaclp = vsecattr.vsa_aclentp;
5898 vsecattr.vsa_aclentsz = aclbsize;
5900 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp);
5901 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL);
5902 kmem_free(aaclp, aclbsize);
5908 zfs_freebsd_aclcheck(ap)
5909 struct vop_aclcheck_args /* {
5918 return (EOPNOTSUPP);
5922 zfs_vptocnp(struct vop_vptocnp_args *ap)
5924 vnode_t *covered_vp;
5925 vnode_t *vp = ap->a_vp;;
5926 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
5927 znode_t *zp = VTOZ(vp);
5935 * If we are a snapshot mounted under .zfs, run the operation
5936 * on the covered vnode.
5938 if (zp->z_id != zfsvfs->z_root || zfsvfs->z_parent == zfsvfs) {
5939 char name[MAXNAMLEN + 1];
5943 error = zfs_znode_parent_and_name(zp, &dzp, name);
5946 if (*ap->a_buflen < len)
5947 error = SET_ERROR(ENOMEM);
5950 *ap->a_buflen -= len;
5951 bcopy(name, ap->a_buf + *ap->a_buflen, len);
5952 *ap->a_vpp = ZTOV(dzp);
5959 covered_vp = vp->v_mount->mnt_vnodecovered;
5961 ltype = VOP_ISLOCKED(vp);
5963 error = vget(covered_vp, LK_SHARED | LK_VNHELD, curthread);
5965 error = VOP_VPTOCNP(covered_vp, ap->a_vpp, ap->a_cred,
5966 ap->a_buf, ap->a_buflen);
5969 vn_lock(vp, ltype | LK_RETRY);
5970 if ((vp->v_iflag & VI_DOOMED) != 0)
5971 error = SET_ERROR(ENOENT);
5978 struct vop_lock1_args /* {
5989 err = vop_stdlock(ap);
5990 if (err == 0 && (ap->a_flags & LK_NOWAIT) == 0) {
5993 if (vp->v_mount != NULL && (vp->v_iflag & VI_DOOMED) == 0 &&
5994 zp != NULL && (zp->z_pflags & ZFS_XATTR) == 0)
5995 VERIFY(!RRM_LOCK_HELD(&zp->z_zfsvfs->z_teardown_lock));
6001 struct vop_vector zfs_vnodeops;
6002 struct vop_vector zfs_fifoops;
6003 struct vop_vector zfs_shareops;
6005 struct vop_vector zfs_vnodeops = {
6006 .vop_default = &default_vnodeops,
6007 .vop_inactive = zfs_freebsd_inactive,
6008 .vop_reclaim = zfs_freebsd_reclaim,
6009 .vop_access = zfs_freebsd_access,
6010 .vop_lookup = zfs_cache_lookup,
6011 .vop_cachedlookup = zfs_freebsd_lookup,
6012 .vop_getattr = zfs_freebsd_getattr,
6013 .vop_setattr = zfs_freebsd_setattr,
6014 .vop_create = zfs_freebsd_create,
6015 .vop_mknod = zfs_freebsd_create,
6016 .vop_mkdir = zfs_freebsd_mkdir,
6017 .vop_readdir = zfs_freebsd_readdir,
6018 .vop_fsync = zfs_freebsd_fsync,
6019 .vop_open = zfs_freebsd_open,
6020 .vop_close = zfs_freebsd_close,
6021 .vop_rmdir = zfs_freebsd_rmdir,
6022 .vop_ioctl = zfs_freebsd_ioctl,
6023 .vop_link = zfs_freebsd_link,
6024 .vop_symlink = zfs_freebsd_symlink,
6025 .vop_readlink = zfs_freebsd_readlink,
6026 .vop_read = zfs_freebsd_read,
6027 .vop_write = zfs_freebsd_write,
6028 .vop_remove = zfs_freebsd_remove,
6029 .vop_rename = zfs_freebsd_rename,
6030 .vop_pathconf = zfs_freebsd_pathconf,
6031 .vop_bmap = zfs_freebsd_bmap,
6032 .vop_fid = zfs_freebsd_fid,
6033 .vop_getextattr = zfs_getextattr,
6034 .vop_deleteextattr = zfs_deleteextattr,
6035 .vop_setextattr = zfs_setextattr,
6036 .vop_listextattr = zfs_listextattr,
6037 .vop_getacl = zfs_freebsd_getacl,
6038 .vop_setacl = zfs_freebsd_setacl,
6039 .vop_aclcheck = zfs_freebsd_aclcheck,
6040 .vop_getpages = zfs_freebsd_getpages,
6041 .vop_putpages = zfs_freebsd_putpages,
6042 .vop_vptocnp = zfs_vptocnp,
6044 .vop_lock1 = zfs_lock,
6048 struct vop_vector zfs_fifoops = {
6049 .vop_default = &fifo_specops,
6050 .vop_fsync = zfs_freebsd_fsync,
6051 .vop_access = zfs_freebsd_access,
6052 .vop_getattr = zfs_freebsd_getattr,
6053 .vop_inactive = zfs_freebsd_inactive,
6054 .vop_read = VOP_PANIC,
6055 .vop_reclaim = zfs_freebsd_reclaim,
6056 .vop_setattr = zfs_freebsd_setattr,
6057 .vop_write = VOP_PANIC,
6058 .vop_pathconf = zfs_freebsd_fifo_pathconf,
6059 .vop_fid = zfs_freebsd_fid,
6060 .vop_getacl = zfs_freebsd_getacl,
6061 .vop_setacl = zfs_freebsd_setacl,
6062 .vop_aclcheck = zfs_freebsd_aclcheck,
6066 * special share hidden files vnode operations template
6068 struct vop_vector zfs_shareops = {
6069 .vop_default = &default_vnodeops,
6070 .vop_access = zfs_freebsd_access,
6071 .vop_inactive = zfs_freebsd_inactive,
6072 .vop_reclaim = zfs_freebsd_reclaim,
6073 .vop_fid = zfs_freebsd_fid,
6074 .vop_pathconf = zfs_freebsd_pathconf,