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>
84 * Each vnode op performs some logical unit of work. To do this, the ZPL must
85 * properly lock its in-core state, create a DMU transaction, do the work,
86 * record this work in the intent log (ZIL), commit the DMU transaction,
87 * and wait for the intent log to commit if it is a synchronous operation.
88 * Moreover, the vnode ops must work in both normal and log replay context.
89 * The ordering of events is important to avoid deadlocks and references
90 * to freed memory. The example below illustrates the following Big Rules:
92 * (1) A check must be made in each zfs thread for a mounted file system.
93 * This is done avoiding races using ZFS_ENTER(zfsvfs).
94 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
95 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
96 * can return EIO from the calling function.
98 * (2) VN_RELE() should always be the last thing except for zil_commit()
99 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
100 * First, if it's the last reference, the vnode/znode
101 * can be freed, so the zp may point to freed memory. Second, the last
102 * reference will call zfs_zinactive(), which may induce a lot of work --
103 * pushing cached pages (which acquires range locks) and syncing out
104 * cached atime changes. Third, zfs_zinactive() may require a new tx,
105 * which could deadlock the system if you were already holding one.
106 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
108 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
109 * as they can span dmu_tx_assign() calls.
111 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
112 * dmu_tx_assign(). This is critical because we don't want to block
113 * while holding locks.
115 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
116 * reduces lock contention and CPU usage when we must wait (note that if
117 * throughput is constrained by the storage, nearly every transaction
120 * Note, in particular, that if a lock is sometimes acquired before
121 * the tx assigns, and sometimes after (e.g. z_lock), then failing
122 * to use a non-blocking assign can deadlock the system. The scenario:
124 * Thread A has grabbed a lock before calling dmu_tx_assign().
125 * Thread B is in an already-assigned tx, and blocks for this lock.
126 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
127 * forever, because the previous txg can't quiesce until B's tx commits.
129 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
130 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
131 * calls to dmu_tx_assign(), pass TXG_NOTHROTTLE in addition to TXG_NOWAIT,
132 * to indicate that this operation has already called dmu_tx_wait().
133 * This will ensure that we don't retry forever, waiting a short bit
136 * (5) If the operation succeeded, generate the intent log entry for it
137 * before dropping locks. This ensures that the ordering of events
138 * in the intent log matches the order in which they actually occurred.
139 * During ZIL replay the zfs_log_* functions will update the sequence
140 * number to indicate the zil transaction has replayed.
142 * (6) At the end of each vnode op, the DMU tx must always commit,
143 * regardless of whether there were any errors.
145 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
146 * to ensure that synchronous semantics are provided when necessary.
148 * In general, this is how things should be ordered in each vnode op:
150 * ZFS_ENTER(zfsvfs); // exit if unmounted
152 * zfs_dirent_lookup(&dl, ...) // lock directory entry (may VN_HOLD())
153 * rw_enter(...); // grab any other locks you need
154 * tx = dmu_tx_create(...); // get DMU tx
155 * dmu_tx_hold_*(); // hold each object you might modify
156 * error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
158 * rw_exit(...); // drop locks
159 * zfs_dirent_unlock(dl); // unlock directory entry
160 * VN_RELE(...); // release held vnodes
161 * if (error == ERESTART) {
167 * dmu_tx_abort(tx); // abort DMU tx
168 * ZFS_EXIT(zfsvfs); // finished in zfs
169 * return (error); // really out of space
171 * error = do_real_work(); // do whatever this VOP does
173 * zfs_log_*(...); // on success, make ZIL entry
174 * dmu_tx_commit(tx); // commit DMU tx -- error or not
175 * rw_exit(...); // drop locks
176 * zfs_dirent_unlock(dl); // unlock directory entry
177 * VN_RELE(...); // release held vnodes
178 * zil_commit(zilog, foid); // synchronous when necessary
179 * ZFS_EXIT(zfsvfs); // finished in zfs
180 * return (error); // done, report error
185 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
187 znode_t *zp = VTOZ(*vpp);
188 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
193 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
194 ((flag & FAPPEND) == 0)) {
196 return (SET_ERROR(EPERM));
199 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
200 ZTOV(zp)->v_type == VREG &&
201 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
202 if (fs_vscan(*vpp, cr, 0) != 0) {
204 return (SET_ERROR(EACCES));
208 /* Keep a count of the synchronous opens in the znode */
209 if (flag & (FSYNC | FDSYNC))
210 atomic_inc_32(&zp->z_sync_cnt);
218 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
219 caller_context_t *ct)
221 znode_t *zp = VTOZ(vp);
222 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
225 * Clean up any locks held by this process on the vp.
227 cleanlocks(vp, ddi_get_pid(), 0);
228 cleanshares(vp, ddi_get_pid());
233 /* Decrement the synchronous opens in the znode */
234 if ((flag & (FSYNC | FDSYNC)) && (count == 1))
235 atomic_dec_32(&zp->z_sync_cnt);
237 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
238 ZTOV(zp)->v_type == VREG &&
239 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
240 VERIFY(fs_vscan(vp, cr, 1) == 0);
247 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
248 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
251 zfs_holey(vnode_t *vp, u_long cmd, offset_t *off)
253 znode_t *zp = VTOZ(vp);
254 uint64_t noff = (uint64_t)*off; /* new offset */
259 file_sz = zp->z_size;
260 if (noff >= file_sz) {
261 return (SET_ERROR(ENXIO));
264 if (cmd == _FIO_SEEK_HOLE)
269 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
272 return (SET_ERROR(ENXIO));
275 * We could find a hole that begins after the logical end-of-file,
276 * because dmu_offset_next() only works on whole blocks. If the
277 * EOF falls mid-block, then indicate that the "virtual hole"
278 * at the end of the file begins at the logical EOF, rather than
279 * at the end of the last block.
281 if (noff > file_sz) {
294 zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred,
295 int *rvalp, caller_context_t *ct)
299 dmu_object_info_t doi;
310 * The following two ioctls are used by bfu. Faking out,
311 * necessary to avoid bfu errors.
324 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
325 return (SET_ERROR(EFAULT));
327 off = *(offset_t *)data;
330 zfsvfs = zp->z_zfsvfs;
334 /* offset parameter is in/out */
335 error = zfs_holey(vp, com, &off);
340 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
341 return (SET_ERROR(EFAULT));
343 *(offset_t *)data = off;
348 case _FIO_COUNT_FILLED:
351 * _FIO_COUNT_FILLED adds a new ioctl command which
352 * exposes the number of filled blocks in a
356 zfsvfs = zp->z_zfsvfs;
361 * Wait for all dirty blocks for this object
362 * to get synced out to disk, and the DMU info
365 error = dmu_object_wait_synced(zfsvfs->z_os, zp->z_id);
372 * Retrieve fill count from DMU object.
374 error = dmu_object_info(zfsvfs->z_os, zp->z_id, &doi);
380 ndata = doi.doi_fill_count;
383 if (ddi_copyout(&ndata, (void *)data, sizeof (ndata), flag))
384 return (SET_ERROR(EFAULT));
389 return (SET_ERROR(ENOTTY));
393 page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes)
400 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE
401 * aligned boundaries, if the range is not aligned. As a result a
402 * DEV_BSIZE subrange with partially dirty data may get marked as clean.
403 * It may happen that all DEV_BSIZE subranges are marked clean and thus
404 * the whole page would be considred clean despite have some dirty data.
405 * For this reason we should shrink the range to DEV_BSIZE aligned
406 * boundaries before calling vm_page_clear_dirty.
408 end = rounddown2(off + nbytes, DEV_BSIZE);
409 off = roundup2(off, DEV_BSIZE);
413 zfs_vmobject_assert_wlocked(obj);
416 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
418 if (vm_page_xbusied(pp)) {
420 * Reference the page before unlocking and
421 * sleeping so that the page daemon is less
422 * likely to reclaim it.
424 vm_page_reference(pp);
426 zfs_vmobject_wunlock(obj);
427 vm_page_busy_sleep(pp, "zfsmwb", true);
428 zfs_vmobject_wlock(obj);
432 } else if (pp != NULL) {
438 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
439 vm_object_pip_add(obj, 1);
440 pmap_remove_write(pp);
442 vm_page_clear_dirty(pp, off, nbytes);
450 page_unbusy(vm_page_t pp)
454 vm_object_pip_subtract(pp->object, 1);
458 page_wire(vnode_t *vp, int64_t start)
464 zfs_vmobject_assert_wlocked(obj);
467 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
469 if (vm_page_xbusied(pp)) {
471 * Reference the page before unlocking and
472 * sleeping so that the page daemon is less
473 * likely to reclaim it.
475 vm_page_reference(pp);
477 zfs_vmobject_wunlock(obj);
478 vm_page_busy_sleep(pp, "zfsmwb", true);
479 zfs_vmobject_wlock(obj);
483 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
495 page_unwire(vm_page_t pp)
499 vm_page_unwire(pp, PQ_ACTIVE);
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_wire(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).
1041 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
1043 ASSERT(abuf != NULL);
1044 ASSERT(arc_buf_size(abuf) == max_blksz);
1045 if (error = uiocopy(abuf->b_data, max_blksz,
1046 UIO_WRITE, uio, &cbytes)) {
1047 dmu_return_arcbuf(abuf);
1050 ASSERT(cbytes == max_blksz);
1054 * Start a transaction.
1056 tx = dmu_tx_create(zfsvfs->z_os);
1057 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1058 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
1059 zfs_sa_upgrade_txholds(tx, zp);
1060 error = dmu_tx_assign(tx, TXG_WAIT);
1064 dmu_return_arcbuf(abuf);
1069 * If zfs_range_lock() over-locked we grow the blocksize
1070 * and then reduce the lock range. This will only happen
1071 * on the first iteration since zfs_range_reduce() will
1072 * shrink down r_len to the appropriate size.
1074 if (rl->r_len == UINT64_MAX) {
1077 if (zp->z_blksz > max_blksz) {
1079 * File's blocksize is already larger than the
1080 * "recordsize" property. Only let it grow to
1081 * the next power of 2.
1083 ASSERT(!ISP2(zp->z_blksz));
1084 new_blksz = MIN(end_size,
1085 1 << highbit64(zp->z_blksz));
1087 new_blksz = MIN(end_size, max_blksz);
1089 zfs_grow_blocksize(zp, new_blksz, tx);
1090 zfs_range_reduce(rl, woff, n);
1094 * XXX - should we really limit each write to z_max_blksz?
1095 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
1097 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
1099 if (woff + nbytes > zp->z_size)
1100 vnode_pager_setsize(vp, woff + nbytes);
1103 tx_bytes = uio->uio_resid;
1104 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
1106 tx_bytes -= uio->uio_resid;
1109 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
1111 * If this is not a full block write, but we are
1112 * extending the file past EOF and this data starts
1113 * block-aligned, use assign_arcbuf(). Otherwise,
1114 * write via dmu_write().
1116 if (tx_bytes < max_blksz && (!write_eof ||
1117 aiov->iov_base != abuf->b_data)) {
1119 dmu_write(zfsvfs->z_os, zp->z_id, woff,
1120 aiov->iov_len, aiov->iov_base, tx);
1121 dmu_return_arcbuf(abuf);
1122 xuio_stat_wbuf_copied();
1124 ASSERT(xuio || tx_bytes == max_blksz);
1125 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
1128 ASSERT(tx_bytes <= uio->uio_resid);
1129 uioskip(uio, tx_bytes);
1131 if (tx_bytes && vn_has_cached_data(vp)) {
1132 update_pages(vp, woff, tx_bytes, zfsvfs->z_os,
1133 zp->z_id, uio->uio_segflg, tx);
1137 * If we made no progress, we're done. If we made even
1138 * partial progress, update the znode and ZIL accordingly.
1140 if (tx_bytes == 0) {
1141 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
1142 (void *)&zp->z_size, sizeof (uint64_t), tx);
1149 * Clear Set-UID/Set-GID bits on successful write if not
1150 * privileged and at least one of the excute bits is set.
1152 * It would be nice to to this after all writes have
1153 * been done, but that would still expose the ISUID/ISGID
1154 * to another app after the partial write is committed.
1156 * Note: we don't call zfs_fuid_map_id() here because
1157 * user 0 is not an ephemeral uid.
1159 mutex_enter(&zp->z_acl_lock);
1160 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
1161 (S_IXUSR >> 6))) != 0 &&
1162 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
1163 secpolicy_vnode_setid_retain(vp, cr,
1164 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
1166 zp->z_mode &= ~(S_ISUID | S_ISGID);
1167 newmode = zp->z_mode;
1168 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
1169 (void *)&newmode, sizeof (uint64_t), tx);
1171 mutex_exit(&zp->z_acl_lock);
1173 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
1177 * Update the file size (zp_size) if it has changed;
1178 * account for possible concurrent updates.
1180 while ((end_size = zp->z_size) < uio->uio_loffset) {
1181 (void) atomic_cas_64(&zp->z_size, end_size,
1186 ASSERT(error == 0 || error == EFAULT);
1190 * If we are replaying and eof is non zero then force
1191 * the file size to the specified eof. Note, there's no
1192 * concurrency during replay.
1194 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
1195 zp->z_size = zfsvfs->z_replay_eof;
1198 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1200 (void) sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1202 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
1207 ASSERT(tx_bytes == nbytes);
1212 uio_prefaultpages(MIN(n, max_blksz), uio);
1216 zfs_range_unlock(rl);
1219 * If we're in replay mode, or we made no progress, return error.
1220 * Otherwise, it's at least a partial write, so it's successful.
1222 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1229 * EFAULT means that at least one page of the source buffer was not
1230 * available. VFS will re-try remaining I/O upon this error.
1232 if (error == EFAULT) {
1238 if (ioflag & (FSYNC | FDSYNC) ||
1239 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1240 zil_commit(zilog, zp->z_id);
1248 zfs_get_done(zgd_t *zgd, int error)
1250 znode_t *zp = zgd->zgd_private;
1251 objset_t *os = zp->z_zfsvfs->z_os;
1254 dmu_buf_rele(zgd->zgd_db, zgd);
1256 zfs_range_unlock(zgd->zgd_rl);
1259 * Release the vnode asynchronously as we currently have the
1260 * txg stopped from syncing.
1262 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1264 kmem_free(zgd, sizeof (zgd_t));
1268 static int zil_fault_io = 0;
1272 * Get data to generate a TX_WRITE intent log record.
1275 zfs_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb, zio_t *zio)
1277 zfsvfs_t *zfsvfs = arg;
1278 objset_t *os = zfsvfs->z_os;
1280 uint64_t object = lr->lr_foid;
1281 uint64_t offset = lr->lr_offset;
1282 uint64_t size = lr->lr_length;
1287 ASSERT3P(lwb, !=, NULL);
1288 ASSERT3P(zio, !=, NULL);
1289 ASSERT3U(size, !=, 0);
1292 * Nothing to do if the file has been removed
1294 if (zfs_zget(zfsvfs, object, &zp) != 0)
1295 return (SET_ERROR(ENOENT));
1296 if (zp->z_unlinked) {
1298 * Release the vnode asynchronously as we currently have the
1299 * txg stopped from syncing.
1301 VN_RELE_ASYNC(ZTOV(zp),
1302 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1303 return (SET_ERROR(ENOENT));
1306 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1308 zgd->zgd_private = zp;
1311 * Write records come in two flavors: immediate and indirect.
1312 * For small writes it's cheaper to store the data with the
1313 * log record (immediate); for large writes it's cheaper to
1314 * sync the data and get a pointer to it (indirect) so that
1315 * we don't have to write the data twice.
1317 if (buf != NULL) { /* immediate write */
1318 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1319 /* test for truncation needs to be done while range locked */
1320 if (offset >= zp->z_size) {
1321 error = SET_ERROR(ENOENT);
1323 error = dmu_read(os, object, offset, size, buf,
1324 DMU_READ_NO_PREFETCH);
1326 ASSERT(error == 0 || error == ENOENT);
1327 } else { /* indirect write */
1329 * Have to lock the whole block to ensure when it's
1330 * written out and its checksum is being calculated
1331 * that no one can change the data. We need to re-check
1332 * blocksize after we get the lock in case it's changed!
1337 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1339 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1341 if (zp->z_blksz == size)
1344 zfs_range_unlock(zgd->zgd_rl);
1346 /* test for truncation needs to be done while range locked */
1347 if (lr->lr_offset >= zp->z_size)
1348 error = SET_ERROR(ENOENT);
1351 error = SET_ERROR(EIO);
1356 error = dmu_buf_hold(os, object, offset, zgd, &db,
1357 DMU_READ_NO_PREFETCH);
1360 blkptr_t *bp = &lr->lr_blkptr;
1365 ASSERT(db->db_offset == offset);
1366 ASSERT(db->db_size == size);
1368 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1370 ASSERT(error || lr->lr_length <= size);
1373 * On success, we need to wait for the write I/O
1374 * initiated by dmu_sync() to complete before we can
1375 * release this dbuf. We will finish everything up
1376 * in the zfs_get_done() callback.
1381 if (error == EALREADY) {
1382 lr->lr_common.lrc_txtype = TX_WRITE2;
1384 * TX_WRITE2 relies on the data previously
1385 * written by the TX_WRITE that caused
1386 * EALREADY. We zero out the BP because
1387 * it is the old, currently-on-disk BP.
1396 zfs_get_done(zgd, error);
1403 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1404 caller_context_t *ct)
1406 znode_t *zp = VTOZ(vp);
1407 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1413 if (flag & V_ACE_MASK)
1414 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1416 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1423 zfs_dd_callback(struct mount *mp, void *arg, int lkflags, struct vnode **vpp)
1428 error = vn_lock(*vpp, lkflags);
1435 zfs_lookup_lock(vnode_t *dvp, vnode_t *vp, const char *name, int lkflags)
1437 znode_t *zdp = VTOZ(dvp);
1438 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1442 ASSERT_VOP_LOCKED(dvp, __func__);
1444 if ((zdp->z_pflags & ZFS_XATTR) == 0)
1445 VERIFY(!RRM_LOCK_HELD(&zfsvfs->z_teardown_lock));
1448 if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) {
1449 ASSERT3P(dvp, ==, vp);
1451 ltype = lkflags & LK_TYPE_MASK;
1452 if (ltype != VOP_ISLOCKED(dvp)) {
1453 if (ltype == LK_EXCLUSIVE)
1454 vn_lock(dvp, LK_UPGRADE | LK_RETRY);
1455 else /* if (ltype == LK_SHARED) */
1456 vn_lock(dvp, LK_DOWNGRADE | LK_RETRY);
1459 * Relock for the "." case could leave us with
1462 if (dvp->v_iflag & VI_DOOMED) {
1464 return (SET_ERROR(ENOENT));
1468 } else if (name[0] == '.' && name[1] == '.' && name[2] == 0) {
1470 * Note that in this case, dvp is the child vnode, and we
1471 * are looking up the parent vnode - exactly reverse from
1472 * normal operation. Unlocking dvp requires some rather
1473 * tricky unlock/relock dance to prevent mp from being freed;
1474 * use vn_vget_ino_gen() which takes care of all that.
1476 * XXX Note that there is a time window when both vnodes are
1477 * unlocked. It is possible, although highly unlikely, that
1478 * during that window the parent-child relationship between
1479 * the vnodes may change, for example, get reversed.
1480 * In that case we would have a wrong lock order for the vnodes.
1481 * All other filesystems seem to ignore this problem, so we
1483 * A potential solution could be implemented as follows:
1484 * - using LK_NOWAIT when locking the second vnode and retrying
1486 * - checking that the parent-child relationship still holds
1487 * after locking both vnodes and retrying if it doesn't
1489 error = vn_vget_ino_gen(dvp, zfs_dd_callback, vp, lkflags, &vp);
1492 error = vn_lock(vp, lkflags);
1500 * Lookup an entry in a directory, or an extended attribute directory.
1501 * If it exists, return a held vnode reference for it.
1503 * IN: dvp - vnode of directory to search.
1504 * nm - name of entry to lookup.
1505 * pnp - full pathname to lookup [UNUSED].
1506 * flags - LOOKUP_XATTR set if looking for an attribute.
1507 * rdir - root directory vnode [UNUSED].
1508 * cr - credentials of caller.
1509 * ct - caller context
1511 * OUT: vpp - vnode of located entry, NULL if not found.
1513 * RETURN: 0 on success, error code on failure.
1520 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp,
1521 int nameiop, cred_t *cr, kthread_t *td, int flags)
1523 znode_t *zdp = VTOZ(dvp);
1525 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1529 * Fast path lookup, however we must skip DNLC lookup
1530 * for case folding or normalizing lookups because the
1531 * DNLC code only stores the passed in name. This means
1532 * creating 'a' and removing 'A' on a case insensitive
1533 * file system would work, but DNLC still thinks 'a'
1534 * exists and won't let you create it again on the next
1535 * pass through fast path.
1537 if (!(flags & LOOKUP_XATTR)) {
1538 if (dvp->v_type != VDIR) {
1539 return (SET_ERROR(ENOTDIR));
1540 } else if (zdp->z_sa_hdl == NULL) {
1541 return (SET_ERROR(EIO));
1545 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1552 if (flags & LOOKUP_XATTR) {
1555 * If the xattr property is off, refuse the lookup request.
1557 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1559 return (SET_ERROR(EINVAL));
1564 * We don't allow recursive attributes..
1565 * Maybe someday we will.
1567 if (zdp->z_pflags & ZFS_XATTR) {
1569 return (SET_ERROR(EINVAL));
1572 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1578 * Do we have permission to get into attribute directory?
1580 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1591 * Check accessibility of directory.
1593 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1598 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1599 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1601 return (SET_ERROR(EILSEQ));
1606 * First handle the special cases.
1608 if ((cnp->cn_flags & ISDOTDOT) != 0) {
1610 * If we are a snapshot mounted under .zfs, return
1611 * the vp for the snapshot directory.
1613 if (zdp->z_id == zfsvfs->z_root && zfsvfs->z_parent != zfsvfs) {
1614 struct componentname cn;
1619 ltype = VOP_ISLOCKED(dvp);
1621 error = zfsctl_root(zfsvfs->z_parent, LK_SHARED,
1624 cn.cn_nameptr = "snapshot";
1625 cn.cn_namelen = strlen(cn.cn_nameptr);
1626 cn.cn_nameiop = cnp->cn_nameiop;
1627 cn.cn_flags = cnp->cn_flags & ~ISDOTDOT;
1628 cn.cn_lkflags = cnp->cn_lkflags;
1629 error = VOP_LOOKUP(zfsctl_vp, vpp, &cn);
1632 vn_lock(dvp, ltype | LK_RETRY);
1636 if (zfs_has_ctldir(zdp) && strcmp(nm, ZFS_CTLDIR_NAME) == 0) {
1638 if ((cnp->cn_flags & ISLASTCN) != 0 && nameiop != LOOKUP)
1639 return (SET_ERROR(ENOTSUP));
1640 error = zfsctl_root(zfsvfs, cnp->cn_lkflags, vpp);
1645 * The loop is retry the lookup if the parent-child relationship
1646 * changes during the dot-dot locking complexities.
1651 error = zfs_dirlook(zdp, nm, &zp);
1659 error = zfs_lookup_lock(dvp, *vpp, nm, cnp->cn_lkflags);
1662 * If we've got a locking error, then the vnode
1663 * got reclaimed because of a force unmount.
1664 * We never enter doomed vnodes into the name cache.
1670 if ((cnp->cn_flags & ISDOTDOT) == 0)
1674 if (zdp->z_sa_hdl == NULL) {
1675 error = SET_ERROR(EIO);
1677 error = sa_lookup(zdp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
1678 &parent, sizeof (parent));
1685 if (zp->z_id == parent) {
1696 /* Translate errors and add SAVENAME when needed. */
1697 if (cnp->cn_flags & ISLASTCN) {
1701 if (error == ENOENT) {
1702 error = EJUSTRETURN;
1703 cnp->cn_flags |= SAVENAME;
1709 cnp->cn_flags |= SAVENAME;
1714 /* Insert name into cache (as non-existent) if appropriate. */
1715 if (zfsvfs->z_use_namecache &&
1716 error == ENOENT && (cnp->cn_flags & MAKEENTRY) != 0)
1717 cache_enter(dvp, NULL, cnp);
1719 /* Insert name into cache if appropriate. */
1720 if (zfsvfs->z_use_namecache &&
1721 error == 0 && (cnp->cn_flags & MAKEENTRY)) {
1722 if (!(cnp->cn_flags & ISLASTCN) ||
1723 (nameiop != DELETE && nameiop != RENAME)) {
1724 cache_enter(dvp, *vpp, cnp);
1732 * Attempt to create a new entry in a directory. If the entry
1733 * already exists, truncate the file if permissible, else return
1734 * an error. Return the vp of the created or trunc'd file.
1736 * IN: dvp - vnode of directory to put new file entry in.
1737 * name - name of new file entry.
1738 * vap - attributes of new file.
1739 * excl - flag indicating exclusive or non-exclusive mode.
1740 * mode - mode to open file with.
1741 * cr - credentials of caller.
1742 * flag - large file flag [UNUSED].
1743 * ct - caller context
1744 * vsecp - ACL to be set
1746 * OUT: vpp - vnode of created or trunc'd entry.
1748 * RETURN: 0 on success, error code on failure.
1751 * dvp - ctime|mtime updated if new entry created
1752 * vp - ctime|mtime always, atime if new
1757 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode,
1758 vnode_t **vpp, cred_t *cr, kthread_t *td)
1760 znode_t *zp, *dzp = VTOZ(dvp);
1761 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1768 gid_t gid = crgetgid(cr);
1769 zfs_acl_ids_t acl_ids;
1770 boolean_t fuid_dirtied;
1776 * If we have an ephemeral id, ACL, or XVATTR then
1777 * make sure file system is at proper version
1780 ksid = crgetsid(cr, KSID_OWNER);
1782 uid = ksid_getid(ksid);
1786 if (zfsvfs->z_use_fuids == B_FALSE &&
1787 (vsecp || (vap->va_mask & AT_XVATTR) ||
1788 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1789 return (SET_ERROR(EINVAL));
1794 zilog = zfsvfs->z_log;
1796 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1797 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1799 return (SET_ERROR(EILSEQ));
1802 if (vap->va_mask & AT_XVATTR) {
1803 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
1804 crgetuid(cr), cr, vap->va_type)) != 0) {
1812 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr))
1813 vap->va_mode &= ~S_ISVTX;
1815 error = zfs_dirent_lookup(dzp, name, &zp, ZNEW);
1820 ASSERT3P(zp, ==, NULL);
1823 * Create a new file object and update the directory
1826 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1831 * We only support the creation of regular files in
1832 * extended attribute directories.
1835 if ((dzp->z_pflags & ZFS_XATTR) &&
1836 (vap->va_type != VREG)) {
1837 error = SET_ERROR(EINVAL);
1841 if ((error = zfs_acl_ids_create(dzp, 0, vap,
1842 cr, vsecp, &acl_ids)) != 0)
1845 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1846 zfs_acl_ids_free(&acl_ids);
1847 error = SET_ERROR(EDQUOT);
1851 getnewvnode_reserve(1);
1853 tx = dmu_tx_create(os);
1855 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1856 ZFS_SA_BASE_ATTR_SIZE);
1858 fuid_dirtied = zfsvfs->z_fuid_dirty;
1860 zfs_fuid_txhold(zfsvfs, tx);
1861 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1862 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1863 if (!zfsvfs->z_use_sa &&
1864 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1865 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1866 0, acl_ids.z_aclp->z_acl_bytes);
1868 error = dmu_tx_assign(tx, TXG_WAIT);
1870 zfs_acl_ids_free(&acl_ids);
1872 getnewvnode_drop_reserve();
1876 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1879 zfs_fuid_sync(zfsvfs, tx);
1881 (void) zfs_link_create(dzp, name, zp, tx, ZNEW);
1882 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1883 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1884 vsecp, acl_ids.z_fuidp, vap);
1885 zfs_acl_ids_free(&acl_ids);
1888 getnewvnode_drop_reserve();
1895 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1896 zil_commit(zilog, 0);
1903 * Remove an entry from a directory.
1905 * IN: dvp - vnode of directory to remove entry from.
1906 * name - name of entry to remove.
1907 * cr - credentials of caller.
1908 * ct - caller context
1909 * flags - case flags
1911 * RETURN: 0 on success, error code on failure.
1915 * vp - ctime (if nlink > 0)
1920 zfs_remove(vnode_t *dvp, vnode_t *vp, char *name, cred_t *cr)
1922 znode_t *dzp = VTOZ(dvp);
1923 znode_t *zp = VTOZ(vp);
1925 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1927 uint64_t acl_obj, xattr_obj;
1930 boolean_t unlinked, toobig = FALSE;
1937 zilog = zfsvfs->z_log;
1943 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1948 * Need to use rmdir for removing directories.
1950 if (vp->v_type == VDIR) {
1951 error = SET_ERROR(EPERM);
1955 vnevent_remove(vp, dvp, name, ct);
1959 /* are there any extended attributes? */
1960 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1961 &xattr_obj, sizeof (xattr_obj));
1962 if (error == 0 && xattr_obj) {
1963 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1968 * We may delete the znode now, or we may put it in the unlinked set;
1969 * it depends on whether we're the last link, and on whether there are
1970 * other holds on the vnode. So we dmu_tx_hold() the right things to
1971 * allow for either case.
1973 tx = dmu_tx_create(zfsvfs->z_os);
1974 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1975 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1976 zfs_sa_upgrade_txholds(tx, zp);
1977 zfs_sa_upgrade_txholds(tx, dzp);
1980 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1981 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1984 /* charge as an update -- would be nice not to charge at all */
1985 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1988 * Mark this transaction as typically resulting in a net free of space
1990 dmu_tx_mark_netfree(tx);
1992 error = dmu_tx_assign(tx, TXG_WAIT);
2000 * Remove the directory entry.
2002 error = zfs_link_destroy(dzp, name, zp, tx, ZEXISTS, &unlinked);
2010 zfs_unlinked_add(zp, tx);
2011 vp->v_vflag |= VV_NOSYNC;
2015 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
2023 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2024 zil_commit(zilog, 0);
2031 * Create a new directory and insert it into dvp using the name
2032 * provided. Return a pointer to the inserted directory.
2034 * IN: dvp - vnode of directory to add subdir to.
2035 * dirname - name of new directory.
2036 * vap - attributes of new directory.
2037 * cr - credentials of caller.
2038 * ct - caller context
2039 * flags - case flags
2040 * vsecp - ACL to be set
2042 * OUT: vpp - vnode of created directory.
2044 * RETURN: 0 on success, error code on failure.
2047 * dvp - ctime|mtime updated
2048 * vp - ctime|mtime|atime updated
2052 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr)
2054 znode_t *zp, *dzp = VTOZ(dvp);
2055 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2062 gid_t gid = crgetgid(cr);
2063 zfs_acl_ids_t acl_ids;
2064 boolean_t fuid_dirtied;
2066 ASSERT(vap->va_type == VDIR);
2069 * If we have an ephemeral id, ACL, or XVATTR then
2070 * make sure file system is at proper version
2073 ksid = crgetsid(cr, KSID_OWNER);
2075 uid = ksid_getid(ksid);
2078 if (zfsvfs->z_use_fuids == B_FALSE &&
2079 ((vap->va_mask & AT_XVATTR) ||
2080 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
2081 return (SET_ERROR(EINVAL));
2085 zilog = zfsvfs->z_log;
2087 if (dzp->z_pflags & ZFS_XATTR) {
2089 return (SET_ERROR(EINVAL));
2092 if (zfsvfs->z_utf8 && u8_validate(dirname,
2093 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2095 return (SET_ERROR(EILSEQ));
2098 if (vap->va_mask & AT_XVATTR) {
2099 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
2100 crgetuid(cr), cr, vap->va_type)) != 0) {
2106 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
2107 NULL, &acl_ids)) != 0) {
2113 * First make sure the new directory doesn't exist.
2115 * Existence is checked first to make sure we don't return
2116 * EACCES instead of EEXIST which can cause some applications
2121 if (error = zfs_dirent_lookup(dzp, dirname, &zp, ZNEW)) {
2122 zfs_acl_ids_free(&acl_ids);
2126 ASSERT3P(zp, ==, NULL);
2128 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
2129 zfs_acl_ids_free(&acl_ids);
2134 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
2135 zfs_acl_ids_free(&acl_ids);
2137 return (SET_ERROR(EDQUOT));
2141 * Add a new entry to the directory.
2143 getnewvnode_reserve(1);
2144 tx = dmu_tx_create(zfsvfs->z_os);
2145 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2146 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2147 fuid_dirtied = zfsvfs->z_fuid_dirty;
2149 zfs_fuid_txhold(zfsvfs, tx);
2150 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2151 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2152 acl_ids.z_aclp->z_acl_bytes);
2155 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2156 ZFS_SA_BASE_ATTR_SIZE);
2158 error = dmu_tx_assign(tx, TXG_WAIT);
2160 zfs_acl_ids_free(&acl_ids);
2162 getnewvnode_drop_reserve();
2170 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2173 zfs_fuid_sync(zfsvfs, tx);
2176 * Now put new name in parent dir.
2178 (void) zfs_link_create(dzp, dirname, zp, tx, ZNEW);
2182 txtype = zfs_log_create_txtype(Z_DIR, NULL, vap);
2183 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, NULL,
2184 acl_ids.z_fuidp, vap);
2186 zfs_acl_ids_free(&acl_ids);
2190 getnewvnode_drop_reserve();
2192 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2193 zil_commit(zilog, 0);
2200 * Remove a directory subdir entry. If the current working
2201 * directory is the same as the subdir to be removed, the
2204 * IN: dvp - vnode of directory to remove from.
2205 * name - name of directory to be removed.
2206 * cwd - vnode of current working directory.
2207 * cr - credentials of caller.
2208 * ct - caller context
2209 * flags - case flags
2211 * RETURN: 0 on success, error code on failure.
2214 * dvp - ctime|mtime updated
2218 zfs_rmdir(vnode_t *dvp, vnode_t *vp, char *name, cred_t *cr)
2220 znode_t *dzp = VTOZ(dvp);
2221 znode_t *zp = VTOZ(vp);
2222 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2230 zilog = zfsvfs->z_log;
2233 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2237 if (vp->v_type != VDIR) {
2238 error = SET_ERROR(ENOTDIR);
2242 vnevent_rmdir(vp, dvp, name, ct);
2244 tx = dmu_tx_create(zfsvfs->z_os);
2245 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2246 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2247 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2248 zfs_sa_upgrade_txholds(tx, zp);
2249 zfs_sa_upgrade_txholds(tx, dzp);
2250 dmu_tx_mark_netfree(tx);
2251 error = dmu_tx_assign(tx, TXG_WAIT);
2260 error = zfs_link_destroy(dzp, name, zp, tx, ZEXISTS, NULL);
2263 uint64_t txtype = TX_RMDIR;
2264 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2271 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2272 zil_commit(zilog, 0);
2279 * Read as many directory entries as will fit into the provided
2280 * buffer from the given directory cursor position (specified in
2281 * the uio structure).
2283 * IN: vp - vnode of directory to read.
2284 * uio - structure supplying read location, range info,
2285 * and return buffer.
2286 * cr - credentials of caller.
2287 * ct - caller context
2288 * flags - case flags
2290 * OUT: uio - updated offset and range, buffer filled.
2291 * eofp - set to true if end-of-file detected.
2293 * RETURN: 0 on success, error code on failure.
2296 * vp - atime updated
2298 * Note that the low 4 bits of the cookie returned by zap is always zero.
2299 * This allows us to use the low range for "special" directory entries:
2300 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2301 * we use the offset 2 for the '.zfs' directory.
2305 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies)
2307 znode_t *zp = VTOZ(vp);
2311 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2316 zap_attribute_t zap;
2317 uint_t bytes_wanted;
2318 uint64_t offset; /* must be unsigned; checks for < 1 */
2324 boolean_t check_sysattrs;
2327 u_long *cooks = NULL;
2333 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2334 &parent, sizeof (parent))) != 0) {
2340 * If we are not given an eof variable,
2347 * Check for valid iov_len.
2349 if (uio->uio_iov->iov_len <= 0) {
2351 return (SET_ERROR(EINVAL));
2355 * Quit if directory has been removed (posix)
2357 if ((*eofp = zp->z_unlinked) != 0) {
2364 offset = uio->uio_loffset;
2365 prefetch = zp->z_zn_prefetch;
2368 * Initialize the iterator cursor.
2372 * Start iteration from the beginning of the directory.
2374 zap_cursor_init(&zc, os, zp->z_id);
2377 * The offset is a serialized cursor.
2379 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2383 * Get space to change directory entries into fs independent format.
2385 iovp = uio->uio_iov;
2386 bytes_wanted = iovp->iov_len;
2387 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2388 bufsize = bytes_wanted;
2389 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2390 odp = (struct dirent64 *)outbuf;
2392 bufsize = bytes_wanted;
2394 odp = (struct dirent64 *)iovp->iov_base;
2396 eodp = (struct edirent *)odp;
2398 if (ncookies != NULL) {
2400 * Minimum entry size is dirent size and 1 byte for a file name.
2402 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1);
2403 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK);
2408 * If this VFS supports the system attribute view interface; and
2409 * we're looking at an extended attribute directory; and we care
2410 * about normalization conflicts on this vfs; then we must check
2411 * for normalization conflicts with the sysattr name space.
2414 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2415 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2416 (flags & V_RDDIR_ENTFLAGS);
2422 * Transform to file-system independent format
2425 while (outcount < bytes_wanted) {
2428 off64_t *next = NULL;
2431 * Special case `.', `..', and `.zfs'.
2434 (void) strcpy(zap.za_name, ".");
2435 zap.za_normalization_conflict = 0;
2438 } else if (offset == 1) {
2439 (void) strcpy(zap.za_name, "..");
2440 zap.za_normalization_conflict = 0;
2443 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2444 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2445 zap.za_normalization_conflict = 0;
2446 objnum = ZFSCTL_INO_ROOT;
2452 if (error = zap_cursor_retrieve(&zc, &zap)) {
2453 if ((*eofp = (error == ENOENT)) != 0)
2459 if (zap.za_integer_length != 8 ||
2460 zap.za_num_integers != 1) {
2461 cmn_err(CE_WARN, "zap_readdir: bad directory "
2462 "entry, obj = %lld, offset = %lld\n",
2463 (u_longlong_t)zp->z_id,
2464 (u_longlong_t)offset);
2465 error = SET_ERROR(ENXIO);
2469 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2471 * MacOS X can extract the object type here such as:
2472 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2474 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2476 if (check_sysattrs && !zap.za_normalization_conflict) {
2478 zap.za_normalization_conflict =
2479 xattr_sysattr_casechk(zap.za_name);
2481 panic("%s:%u: TODO", __func__, __LINE__);
2486 if (flags & V_RDDIR_ACCFILTER) {
2488 * If we have no access at all, don't include
2489 * this entry in the returned information
2492 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2494 if (!zfs_has_access(ezp, cr)) {
2501 if (flags & V_RDDIR_ENTFLAGS)
2502 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2504 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2507 * Will this entry fit in the buffer?
2509 if (outcount + reclen > bufsize) {
2511 * Did we manage to fit anything in the buffer?
2514 error = SET_ERROR(EINVAL);
2519 if (flags & V_RDDIR_ENTFLAGS) {
2521 * Add extended flag entry:
2523 eodp->ed_ino = objnum;
2524 eodp->ed_reclen = reclen;
2525 /* NOTE: ed_off is the offset for the *next* entry. */
2526 next = &eodp->ed_off;
2527 eodp->ed_eflags = zap.za_normalization_conflict ?
2528 ED_CASE_CONFLICT : 0;
2529 (void) strncpy(eodp->ed_name, zap.za_name,
2530 EDIRENT_NAMELEN(reclen));
2531 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2536 odp->d_ino = objnum;
2537 odp->d_reclen = reclen;
2538 odp->d_namlen = strlen(zap.za_name);
2539 /* NOTE: d_off is the offset for the *next* entry. */
2541 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1);
2543 dirent_terminate(odp);
2544 odp = (dirent64_t *)((intptr_t)odp + reclen);
2548 ASSERT(outcount <= bufsize);
2550 /* Prefetch znode */
2552 dmu_prefetch(os, objnum, 0, 0, 0,
2553 ZIO_PRIORITY_SYNC_READ);
2557 * Move to the next entry, fill in the previous offset.
2559 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2560 zap_cursor_advance(&zc);
2561 offset = zap_cursor_serialize(&zc);
2566 /* Fill the offset right after advancing the cursor. */
2569 if (cooks != NULL) {
2572 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks));
2575 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2577 /* Subtract unused cookies */
2578 if (ncookies != NULL)
2579 *ncookies -= ncooks;
2581 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2582 iovp->iov_base += outcount;
2583 iovp->iov_len -= outcount;
2584 uio->uio_resid -= outcount;
2585 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2587 * Reset the pointer.
2589 offset = uio->uio_loffset;
2593 zap_cursor_fini(&zc);
2594 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2595 kmem_free(outbuf, bufsize);
2597 if (error == ENOENT)
2600 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2602 uio->uio_loffset = offset;
2604 if (error != 0 && cookies != NULL) {
2605 free(*cookies, M_TEMP);
2612 ulong_t zfs_fsync_sync_cnt = 4;
2615 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2617 znode_t *zp = VTOZ(vp);
2618 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2620 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2622 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2625 zil_commit(zfsvfs->z_log, zp->z_id);
2633 * Get the requested file attributes and place them in the provided
2636 * IN: vp - vnode of file.
2637 * vap - va_mask identifies requested attributes.
2638 * If AT_XVATTR set, then optional attrs are requested
2639 * flags - ATTR_NOACLCHECK (CIFS server context)
2640 * cr - credentials of caller.
2641 * ct - caller context
2643 * OUT: vap - attribute values.
2645 * RETURN: 0 (always succeeds).
2649 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2650 caller_context_t *ct)
2652 znode_t *zp = VTOZ(vp);
2653 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2656 u_longlong_t nblocks;
2657 uint64_t mtime[2], ctime[2], crtime[2], rdev;
2658 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2659 xoptattr_t *xoap = NULL;
2660 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2661 sa_bulk_attr_t bulk[4];
2667 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2669 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2670 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2671 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
2672 if (vp->v_type == VBLK || vp->v_type == VCHR)
2673 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
2676 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2682 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2683 * Also, if we are the owner don't bother, since owner should
2684 * always be allowed to read basic attributes of file.
2686 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2687 (vap->va_uid != crgetuid(cr))) {
2688 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2696 * Return all attributes. It's cheaper to provide the answer
2697 * than to determine whether we were asked the question.
2700 vap->va_type = IFTOVT(zp->z_mode);
2701 vap->va_mode = zp->z_mode & ~S_IFMT;
2703 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2707 vap->va_nodeid = zp->z_id;
2708 vap->va_nlink = zp->z_links;
2709 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp) &&
2710 zp->z_links < ZFS_LINK_MAX)
2712 vap->va_size = zp->z_size;
2714 vap->va_rdev = vp->v_rdev;
2716 if (vp->v_type == VBLK || vp->v_type == VCHR)
2717 vap->va_rdev = zfs_cmpldev(rdev);
2719 vap->va_seq = zp->z_seq;
2720 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */
2721 vap->va_filerev = zp->z_seq;
2724 * Add in any requested optional attributes and the create time.
2725 * Also set the corresponding bits in the returned attribute bitmap.
2727 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2728 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2730 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2731 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2734 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2735 xoap->xoa_readonly =
2736 ((zp->z_pflags & ZFS_READONLY) != 0);
2737 XVA_SET_RTN(xvap, XAT_READONLY);
2740 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2742 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2743 XVA_SET_RTN(xvap, XAT_SYSTEM);
2746 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2748 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2749 XVA_SET_RTN(xvap, XAT_HIDDEN);
2752 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2753 xoap->xoa_nounlink =
2754 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2755 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2758 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2759 xoap->xoa_immutable =
2760 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2761 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2764 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2765 xoap->xoa_appendonly =
2766 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2767 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2770 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2772 ((zp->z_pflags & ZFS_NODUMP) != 0);
2773 XVA_SET_RTN(xvap, XAT_NODUMP);
2776 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2778 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2779 XVA_SET_RTN(xvap, XAT_OPAQUE);
2782 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2783 xoap->xoa_av_quarantined =
2784 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2785 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2788 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2789 xoap->xoa_av_modified =
2790 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2791 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2794 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2795 vp->v_type == VREG) {
2796 zfs_sa_get_scanstamp(zp, xvap);
2799 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2800 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2801 XVA_SET_RTN(xvap, XAT_REPARSE);
2803 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2804 xoap->xoa_generation = zp->z_gen;
2805 XVA_SET_RTN(xvap, XAT_GEN);
2808 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2810 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2811 XVA_SET_RTN(xvap, XAT_OFFLINE);
2814 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2816 ((zp->z_pflags & ZFS_SPARSE) != 0);
2817 XVA_SET_RTN(xvap, XAT_SPARSE);
2821 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2822 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2823 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2824 ZFS_TIME_DECODE(&vap->va_birthtime, crtime);
2827 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2828 vap->va_blksize = blksize;
2829 vap->va_bytes = nblocks << 9; /* nblocks * 512 */
2831 if (zp->z_blksz == 0) {
2833 * Block size hasn't been set; suggest maximal I/O transfers.
2835 vap->va_blksize = zfsvfs->z_max_blksz;
2843 * Set the file attributes to the values contained in the
2846 * IN: vp - vnode of file to be modified.
2847 * vap - new attribute values.
2848 * If AT_XVATTR set, then optional attrs are being set
2849 * flags - ATTR_UTIME set if non-default time values provided.
2850 * - ATTR_NOACLCHECK (CIFS context only).
2851 * cr - credentials of caller.
2852 * ct - caller context
2854 * RETURN: 0 on success, error code on failure.
2857 * vp - ctime updated, mtime updated if size changed.
2861 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2862 caller_context_t *ct)
2864 znode_t *zp = VTOZ(vp);
2865 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2870 uint_t mask = vap->va_mask;
2871 uint_t saved_mask = 0;
2872 uint64_t saved_mode;
2875 uint64_t new_uid, new_gid;
2877 uint64_t mtime[2], ctime[2];
2879 int need_policy = FALSE;
2881 zfs_fuid_info_t *fuidp = NULL;
2882 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2885 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2886 boolean_t fuid_dirtied = B_FALSE;
2887 sa_bulk_attr_t bulk[7], xattr_bulk[7];
2888 int count = 0, xattr_count = 0;
2893 if (mask & AT_NOSET)
2894 return (SET_ERROR(EINVAL));
2899 zilog = zfsvfs->z_log;
2902 * Make sure that if we have ephemeral uid/gid or xvattr specified
2903 * that file system is at proper version level
2906 if (zfsvfs->z_use_fuids == B_FALSE &&
2907 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2908 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2909 (mask & AT_XVATTR))) {
2911 return (SET_ERROR(EINVAL));
2914 if (mask & AT_SIZE && vp->v_type == VDIR) {
2916 return (SET_ERROR(EISDIR));
2919 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2921 return (SET_ERROR(EINVAL));
2925 * If this is an xvattr_t, then get a pointer to the structure of
2926 * optional attributes. If this is NULL, then we have a vattr_t.
2928 xoap = xva_getxoptattr(xvap);
2930 xva_init(&tmpxvattr);
2933 * Immutable files can only alter immutable bit and atime
2935 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2936 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2937 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2939 return (SET_ERROR(EPERM));
2943 * Note: ZFS_READONLY is handled in zfs_zaccess_common.
2947 * Verify timestamps doesn't overflow 32 bits.
2948 * ZFS can handle large timestamps, but 32bit syscalls can't
2949 * handle times greater than 2039. This check should be removed
2950 * once large timestamps are fully supported.
2952 if (mask & (AT_ATIME | AT_MTIME)) {
2953 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2954 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2956 return (SET_ERROR(EOVERFLOW));
2959 if (xoap && (mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME) &&
2960 TIMESPEC_OVERFLOW(&vap->va_birthtime)) {
2962 return (SET_ERROR(EOVERFLOW));
2968 /* Can this be moved to before the top label? */
2969 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2971 return (SET_ERROR(EROFS));
2975 * First validate permissions
2978 if (mask & AT_SIZE) {
2980 * XXX - Note, we are not providing any open
2981 * mode flags here (like FNDELAY), so we may
2982 * block if there are locks present... this
2983 * should be addressed in openat().
2985 /* XXX - would it be OK to generate a log record here? */
2986 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2993 if (mask & (AT_ATIME|AT_MTIME) ||
2994 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2995 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2996 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2997 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2998 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2999 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
3000 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
3001 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
3005 if (mask & (AT_UID|AT_GID)) {
3006 int idmask = (mask & (AT_UID|AT_GID));
3011 * NOTE: even if a new mode is being set,
3012 * we may clear S_ISUID/S_ISGID bits.
3015 if (!(mask & AT_MODE))
3016 vap->va_mode = zp->z_mode;
3019 * Take ownership or chgrp to group we are a member of
3022 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
3023 take_group = (mask & AT_GID) &&
3024 zfs_groupmember(zfsvfs, vap->va_gid, cr);
3027 * If both AT_UID and AT_GID are set then take_owner and
3028 * take_group must both be set in order to allow taking
3031 * Otherwise, send the check through secpolicy_vnode_setattr()
3035 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
3036 ((idmask == AT_UID) && take_owner) ||
3037 ((idmask == AT_GID) && take_group)) {
3038 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
3039 skipaclchk, cr) == 0) {
3041 * Remove setuid/setgid for non-privileged users
3043 secpolicy_setid_clear(vap, vp, cr);
3044 trim_mask = (mask & (AT_UID|AT_GID));
3053 oldva.va_mode = zp->z_mode;
3054 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
3055 if (mask & AT_XVATTR) {
3057 * Update xvattr mask to include only those attributes
3058 * that are actually changing.
3060 * the bits will be restored prior to actually setting
3061 * the attributes so the caller thinks they were set.
3063 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
3064 if (xoap->xoa_appendonly !=
3065 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
3068 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
3069 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
3073 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
3074 if (xoap->xoa_nounlink !=
3075 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
3078 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
3079 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
3083 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
3084 if (xoap->xoa_immutable !=
3085 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
3088 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
3089 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
3093 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
3094 if (xoap->xoa_nodump !=
3095 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
3098 XVA_CLR_REQ(xvap, XAT_NODUMP);
3099 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
3103 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3104 if (xoap->xoa_av_modified !=
3105 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3108 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3109 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
3113 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3114 if ((vp->v_type != VREG &&
3115 xoap->xoa_av_quarantined) ||
3116 xoap->xoa_av_quarantined !=
3117 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3120 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3121 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
3125 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3127 return (SET_ERROR(EPERM));
3130 if (need_policy == FALSE &&
3131 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3132 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3137 if (mask & AT_MODE) {
3138 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3139 err = secpolicy_setid_setsticky_clear(vp, vap,
3145 trim_mask |= AT_MODE;
3153 * If trim_mask is set then take ownership
3154 * has been granted or write_acl is present and user
3155 * has the ability to modify mode. In that case remove
3156 * UID|GID and or MODE from mask so that
3157 * secpolicy_vnode_setattr() doesn't revoke it.
3161 saved_mask = vap->va_mask;
3162 vap->va_mask &= ~trim_mask;
3163 if (trim_mask & AT_MODE) {
3165 * Save the mode, as secpolicy_vnode_setattr()
3166 * will overwrite it with ova.va_mode.
3168 saved_mode = vap->va_mode;
3171 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3172 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3179 vap->va_mask |= saved_mask;
3180 if (trim_mask & AT_MODE) {
3182 * Recover the mode after
3183 * secpolicy_vnode_setattr().
3185 vap->va_mode = saved_mode;
3191 * secpolicy_vnode_setattr, or take ownership may have
3194 mask = vap->va_mask;
3196 if ((mask & (AT_UID | AT_GID))) {
3197 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3198 &xattr_obj, sizeof (xattr_obj));
3200 if (err == 0 && xattr_obj) {
3201 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3203 err = vn_lock(ZTOV(attrzp), LK_EXCLUSIVE);
3205 vrele(ZTOV(attrzp));
3210 if (mask & AT_UID) {
3211 new_uid = zfs_fuid_create(zfsvfs,
3212 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3213 if (new_uid != zp->z_uid &&
3214 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3217 err = SET_ERROR(EDQUOT);
3222 if (mask & AT_GID) {
3223 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3224 cr, ZFS_GROUP, &fuidp);
3225 if (new_gid != zp->z_gid &&
3226 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3229 err = SET_ERROR(EDQUOT);
3234 tx = dmu_tx_create(zfsvfs->z_os);
3236 if (mask & AT_MODE) {
3237 uint64_t pmode = zp->z_mode;
3239 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3241 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3242 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3243 err = SET_ERROR(EPERM);
3247 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3250 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3252 * Are we upgrading ACL from old V0 format
3255 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3256 zfs_znode_acl_version(zp) ==
3257 ZFS_ACL_VERSION_INITIAL) {
3258 dmu_tx_hold_free(tx, acl_obj, 0,
3260 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3261 0, aclp->z_acl_bytes);
3263 dmu_tx_hold_write(tx, acl_obj, 0,
3266 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3267 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3268 0, aclp->z_acl_bytes);
3270 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3272 if ((mask & AT_XVATTR) &&
3273 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3274 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3276 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3280 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3283 fuid_dirtied = zfsvfs->z_fuid_dirty;
3285 zfs_fuid_txhold(zfsvfs, tx);
3287 zfs_sa_upgrade_txholds(tx, zp);
3289 err = dmu_tx_assign(tx, TXG_WAIT);
3295 * Set each attribute requested.
3296 * We group settings according to the locks they need to acquire.
3298 * Note: you cannot set ctime directly, although it will be
3299 * updated as a side-effect of calling this function.
3302 if (mask & (AT_UID|AT_GID|AT_MODE))
3303 mutex_enter(&zp->z_acl_lock);
3305 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3306 &zp->z_pflags, sizeof (zp->z_pflags));
3309 if (mask & (AT_UID|AT_GID|AT_MODE))
3310 mutex_enter(&attrzp->z_acl_lock);
3311 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3312 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3313 sizeof (attrzp->z_pflags));
3316 if (mask & (AT_UID|AT_GID)) {
3318 if (mask & AT_UID) {
3319 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3320 &new_uid, sizeof (new_uid));
3321 zp->z_uid = new_uid;
3323 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3324 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3326 attrzp->z_uid = new_uid;
3330 if (mask & AT_GID) {
3331 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3332 NULL, &new_gid, sizeof (new_gid));
3333 zp->z_gid = new_gid;
3335 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3336 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3338 attrzp->z_gid = new_gid;
3341 if (!(mask & AT_MODE)) {
3342 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3343 NULL, &new_mode, sizeof (new_mode));
3344 new_mode = zp->z_mode;
3346 err = zfs_acl_chown_setattr(zp);
3349 err = zfs_acl_chown_setattr(attrzp);
3354 if (mask & AT_MODE) {
3355 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3356 &new_mode, sizeof (new_mode));
3357 zp->z_mode = new_mode;
3358 ASSERT3U((uintptr_t)aclp, !=, 0);
3359 err = zfs_aclset_common(zp, aclp, cr, tx);
3361 if (zp->z_acl_cached)
3362 zfs_acl_free(zp->z_acl_cached);
3363 zp->z_acl_cached = aclp;
3368 if (mask & AT_ATIME) {
3369 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3370 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3371 &zp->z_atime, sizeof (zp->z_atime));
3374 if (mask & AT_MTIME) {
3375 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3376 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3377 mtime, sizeof (mtime));
3380 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3381 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3382 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3383 NULL, mtime, sizeof (mtime));
3384 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3385 &ctime, sizeof (ctime));
3386 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3388 } else if (mask != 0) {
3389 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3390 &ctime, sizeof (ctime));
3391 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3394 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3395 SA_ZPL_CTIME(zfsvfs), NULL,
3396 &ctime, sizeof (ctime));
3397 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3398 mtime, ctime, B_TRUE);
3402 * Do this after setting timestamps to prevent timestamp
3403 * update from toggling bit
3406 if (xoap && (mask & AT_XVATTR)) {
3408 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME))
3409 xoap->xoa_createtime = vap->va_birthtime;
3411 * restore trimmed off masks
3412 * so that return masks can be set for caller.
3415 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3416 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3418 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3419 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3421 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3422 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3424 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3425 XVA_SET_REQ(xvap, XAT_NODUMP);
3427 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3428 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3430 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3431 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3434 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3435 ASSERT(vp->v_type == VREG);
3437 zfs_xvattr_set(zp, xvap, tx);
3441 zfs_fuid_sync(zfsvfs, tx);
3444 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3446 if (mask & (AT_UID|AT_GID|AT_MODE))
3447 mutex_exit(&zp->z_acl_lock);
3450 if (mask & (AT_UID|AT_GID|AT_MODE))
3451 mutex_exit(&attrzp->z_acl_lock);
3454 if (err == 0 && attrzp) {
3455 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3467 zfs_fuid_info_free(fuidp);
3474 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3479 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3480 zil_commit(zilog, 0);
3487 * We acquire all but fdvp locks using non-blocking acquisitions. If we
3488 * fail to acquire any lock in the path we will drop all held locks,
3489 * acquire the new lock in a blocking fashion, and then release it and
3490 * restart the rename. This acquire/release step ensures that we do not
3491 * spin on a lock waiting for release. On error release all vnode locks
3492 * and decrement references the way tmpfs_rename() would do.
3495 zfs_rename_relock(struct vnode *sdvp, struct vnode **svpp,
3496 struct vnode *tdvp, struct vnode **tvpp,
3497 const struct componentname *scnp, const struct componentname *tcnp)
3500 struct vnode *nvp, *svp, *tvp;
3501 znode_t *sdzp, *tdzp, *szp, *tzp;
3502 const char *snm = scnp->cn_nameptr;
3503 const char *tnm = tcnp->cn_nameptr;
3506 VOP_UNLOCK(tdvp, 0);
3507 if (*tvpp != NULL && *tvpp != tdvp)
3508 VOP_UNLOCK(*tvpp, 0);
3511 error = vn_lock(sdvp, LK_EXCLUSIVE);
3516 error = vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT);
3518 VOP_UNLOCK(sdvp, 0);
3521 error = vn_lock(tdvp, LK_EXCLUSIVE);
3524 VOP_UNLOCK(tdvp, 0);
3530 * Before using sdzp and tdzp we must ensure that they are live.
3531 * As a porting legacy from illumos we have two things to worry
3532 * about. One is typical for FreeBSD and it is that the vnode is
3533 * not reclaimed (doomed). The other is that the znode is live.
3534 * The current code can invalidate the znode without acquiring the
3535 * corresponding vnode lock if the object represented by the znode
3536 * and vnode is no longer valid after a rollback or receive operation.
3537 * z_teardown_lock hidden behind ZFS_ENTER and ZFS_EXIT is the lock
3538 * that protects the znodes from the invalidation.
3540 zfsvfs = sdzp->z_zfsvfs;
3541 ASSERT3P(zfsvfs, ==, tdzp->z_zfsvfs);
3545 * We can not use ZFS_VERIFY_ZP() here because it could directly return
3546 * bypassing the cleanup code in the case of an error.
3548 if (tdzp->z_sa_hdl == NULL || sdzp->z_sa_hdl == NULL) {
3550 VOP_UNLOCK(sdvp, 0);
3551 VOP_UNLOCK(tdvp, 0);
3552 error = SET_ERROR(EIO);
3557 * Re-resolve svp to be certain it still exists and fetch the
3560 error = zfs_dirent_lookup(sdzp, snm, &szp, ZEXISTS);
3562 /* Source entry invalid or not there. */
3564 VOP_UNLOCK(sdvp, 0);
3565 VOP_UNLOCK(tdvp, 0);
3566 if ((scnp->cn_flags & ISDOTDOT) != 0 ||
3567 (scnp->cn_namelen == 1 && scnp->cn_nameptr[0] == '.'))
3568 error = SET_ERROR(EINVAL);
3574 * Re-resolve tvp, if it disappeared we just carry on.
3576 error = zfs_dirent_lookup(tdzp, tnm, &tzp, 0);
3579 VOP_UNLOCK(sdvp, 0);
3580 VOP_UNLOCK(tdvp, 0);
3582 if ((tcnp->cn_flags & ISDOTDOT) != 0)
3583 error = SET_ERROR(EINVAL);
3592 * At present the vnode locks must be acquired before z_teardown_lock,
3593 * although it would be more logical to use the opposite order.
3598 * Now try acquire locks on svp and tvp.
3601 error = vn_lock(nvp, LK_EXCLUSIVE | LK_NOWAIT);
3603 VOP_UNLOCK(sdvp, 0);
3604 VOP_UNLOCK(tdvp, 0);
3607 if (error != EBUSY) {
3611 error = vn_lock(nvp, LK_EXCLUSIVE);
3618 * Concurrent rename race.
3623 error = SET_ERROR(EINVAL);
3638 error = vn_lock(nvp, LK_EXCLUSIVE | LK_NOWAIT);
3640 VOP_UNLOCK(sdvp, 0);
3641 VOP_UNLOCK(tdvp, 0);
3642 VOP_UNLOCK(*svpp, 0);
3643 if (error != EBUSY) {
3647 error = vn_lock(nvp, LK_EXCLUSIVE);
3665 * Note that we must use VRELE_ASYNC in this function as it walks
3666 * up the directory tree and vrele may need to acquire an exclusive
3667 * lock if a last reference to a vnode is dropped.
3670 zfs_rename_check(znode_t *szp, znode_t *sdzp, znode_t *tdzp)
3677 zfsvfs = tdzp->z_zfsvfs;
3679 return (SET_ERROR(EINVAL));
3682 if (tdzp->z_id == zfsvfs->z_root)
3686 ASSERT(!zp->z_unlinked);
3687 if ((error = sa_lookup(zp->z_sa_hdl,
3688 SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0)
3691 if (parent == szp->z_id) {
3692 error = SET_ERROR(EINVAL);
3695 if (parent == zfsvfs->z_root)
3697 if (parent == sdzp->z_id)
3700 error = zfs_zget(zfsvfs, parent, &zp1);
3705 VN_RELE_ASYNC(ZTOV(zp),
3706 dsl_pool_vnrele_taskq(dmu_objset_pool(zfsvfs->z_os)));
3710 if (error == ENOTDIR)
3711 panic("checkpath: .. not a directory\n");
3713 VN_RELE_ASYNC(ZTOV(zp),
3714 dsl_pool_vnrele_taskq(dmu_objset_pool(zfsvfs->z_os)));
3719 * Move an entry from the provided source directory to the target
3720 * directory. Change the entry name as indicated.
3722 * IN: sdvp - Source directory containing the "old entry".
3723 * snm - Old entry name.
3724 * tdvp - Target directory to contain the "new entry".
3725 * tnm - New entry name.
3726 * cr - credentials of caller.
3727 * ct - caller context
3728 * flags - case flags
3730 * RETURN: 0 on success, error code on failure.
3733 * sdvp,tdvp - ctime|mtime updated
3737 zfs_rename(vnode_t *sdvp, vnode_t **svpp, struct componentname *scnp,
3738 vnode_t *tdvp, vnode_t **tvpp, struct componentname *tcnp,
3742 znode_t *sdzp, *tdzp, *szp, *tzp;
3743 zilog_t *zilog = NULL;
3745 char *snm = scnp->cn_nameptr;
3746 char *tnm = tcnp->cn_nameptr;
3749 /* Reject renames across filesystems. */
3750 if ((*svpp)->v_mount != tdvp->v_mount ||
3751 ((*tvpp) != NULL && (*svpp)->v_mount != (*tvpp)->v_mount)) {
3752 error = SET_ERROR(EXDEV);
3756 if (zfsctl_is_node(tdvp)) {
3757 error = SET_ERROR(EXDEV);
3762 * Lock all four vnodes to ensure safety and semantics of renaming.
3764 error = zfs_rename_relock(sdvp, svpp, tdvp, tvpp, scnp, tcnp);
3766 /* no vnodes are locked in the case of error here */
3772 zfsvfs = tdzp->z_zfsvfs;
3773 zilog = zfsvfs->z_log;
3776 * After we re-enter ZFS_ENTER() we will have to revalidate all
3781 if (zfsvfs->z_utf8 && u8_validate(tnm,
3782 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3783 error = SET_ERROR(EILSEQ);
3787 /* If source and target are the same file, there is nothing to do. */
3788 if ((*svpp) == (*tvpp)) {
3793 if (((*svpp)->v_type == VDIR && (*svpp)->v_mountedhere != NULL) ||
3794 ((*tvpp) != NULL && (*tvpp)->v_type == VDIR &&
3795 (*tvpp)->v_mountedhere != NULL)) {
3796 error = SET_ERROR(EXDEV);
3801 * We can not use ZFS_VERIFY_ZP() here because it could directly return
3802 * bypassing the cleanup code in the case of an error.
3804 if (tdzp->z_sa_hdl == NULL || sdzp->z_sa_hdl == NULL) {
3805 error = SET_ERROR(EIO);
3810 tzp = *tvpp == NULL ? NULL : VTOZ(*tvpp);
3811 if (szp->z_sa_hdl == NULL || (tzp != NULL && tzp->z_sa_hdl == NULL)) {
3812 error = SET_ERROR(EIO);
3817 * This is to prevent the creation of links into attribute space
3818 * by renaming a linked file into/outof an attribute directory.
3819 * See the comment in zfs_link() for why this is considered bad.
3821 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3822 error = SET_ERROR(EINVAL);
3827 * Must have write access at the source to remove the old entry
3828 * and write access at the target to create the new entry.
3829 * Note that if target and source are the same, this can be
3830 * done in a single check.
3832 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3835 if ((*svpp)->v_type == VDIR) {
3837 * Avoid ".", "..", and aliases of "." for obvious reasons.
3839 if ((scnp->cn_namelen == 1 && scnp->cn_nameptr[0] == '.') ||
3841 (scnp->cn_flags | tcnp->cn_flags) & ISDOTDOT) {
3847 * Check to make sure rename is valid.
3848 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3850 if (error = zfs_rename_check(szp, sdzp, tdzp))
3855 * Does target exist?
3859 * Source and target must be the same type.
3861 if ((*svpp)->v_type == VDIR) {
3862 if ((*tvpp)->v_type != VDIR) {
3863 error = SET_ERROR(ENOTDIR);
3871 if ((*tvpp)->v_type == VDIR) {
3872 error = SET_ERROR(EISDIR);
3878 vnevent_rename_src(*svpp, sdvp, scnp->cn_nameptr, ct);
3880 vnevent_rename_dest(*tvpp, tdvp, tnm, ct);
3883 * notify the target directory if it is not the same
3884 * as source directory.
3887 vnevent_rename_dest_dir(tdvp, ct);
3890 tx = dmu_tx_create(zfsvfs->z_os);
3891 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3892 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3893 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3894 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3896 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3897 zfs_sa_upgrade_txholds(tx, tdzp);
3900 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3901 zfs_sa_upgrade_txholds(tx, tzp);
3904 zfs_sa_upgrade_txholds(tx, szp);
3905 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3906 error = dmu_tx_assign(tx, TXG_WAIT);
3913 if (tzp) /* Attempt to remove the existing target */
3914 error = zfs_link_destroy(tdzp, tnm, tzp, tx, 0, NULL);
3917 error = zfs_link_create(tdzp, tnm, szp, tx, ZRENAMING);
3919 szp->z_pflags |= ZFS_AV_MODIFIED;
3921 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3922 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3925 error = zfs_link_destroy(sdzp, snm, szp, tx, ZRENAMING,
3928 zfs_log_rename(zilog, tx, TX_RENAME, sdzp,
3929 snm, tdzp, tnm, szp);
3932 * Update path information for the target vnode
3934 vn_renamepath(tdvp, *svpp, tnm, strlen(tnm));
3937 * At this point, we have successfully created
3938 * the target name, but have failed to remove
3939 * the source name. Since the create was done
3940 * with the ZRENAMING flag, there are
3941 * complications; for one, the link count is
3942 * wrong. The easiest way to deal with this
3943 * is to remove the newly created target, and
3944 * return the original error. This must
3945 * succeed; fortunately, it is very unlikely to
3946 * fail, since we just created it.
3948 VERIFY3U(zfs_link_destroy(tdzp, tnm, szp, tx,
3949 ZRENAMING, NULL), ==, 0);
3956 cache_purge_negative(tdvp);
3962 unlockout: /* all 4 vnodes are locked, ZFS_ENTER called */
3964 VOP_UNLOCK(*svpp, 0);
3965 VOP_UNLOCK(sdvp, 0);
3967 out: /* original two vnodes are locked */
3968 if (error == 0 && zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3969 zil_commit(zilog, 0);
3972 VOP_UNLOCK(*tvpp, 0);
3974 VOP_UNLOCK(tdvp, 0);
3979 * Insert the indicated symbolic reference entry into the directory.
3981 * IN: dvp - Directory to contain new symbolic link.
3982 * link - Name for new symlink entry.
3983 * vap - Attributes of new entry.
3984 * cr - credentials of caller.
3985 * ct - caller context
3986 * flags - case flags
3988 * RETURN: 0 on success, error code on failure.
3991 * dvp - ctime|mtime updated
3995 zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link,
3996 cred_t *cr, kthread_t *td)
3998 znode_t *zp, *dzp = VTOZ(dvp);
4000 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4002 uint64_t len = strlen(link);
4004 zfs_acl_ids_t acl_ids;
4005 boolean_t fuid_dirtied;
4006 uint64_t txtype = TX_SYMLINK;
4009 ASSERT(vap->va_type == VLNK);
4013 zilog = zfsvfs->z_log;
4015 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
4016 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4018 return (SET_ERROR(EILSEQ));
4021 if (len > MAXPATHLEN) {
4023 return (SET_ERROR(ENAMETOOLONG));
4026 if ((error = zfs_acl_ids_create(dzp, 0,
4027 vap, cr, NULL, &acl_ids)) != 0) {
4033 * Attempt to lock directory; fail if entry already exists.
4035 error = zfs_dirent_lookup(dzp, name, &zp, ZNEW);
4037 zfs_acl_ids_free(&acl_ids);
4042 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4043 zfs_acl_ids_free(&acl_ids);
4048 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
4049 zfs_acl_ids_free(&acl_ids);
4051 return (SET_ERROR(EDQUOT));
4054 getnewvnode_reserve(1);
4055 tx = dmu_tx_create(zfsvfs->z_os);
4056 fuid_dirtied = zfsvfs->z_fuid_dirty;
4057 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
4058 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4059 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
4060 ZFS_SA_BASE_ATTR_SIZE + len);
4061 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
4062 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
4063 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
4064 acl_ids.z_aclp->z_acl_bytes);
4067 zfs_fuid_txhold(zfsvfs, tx);
4068 error = dmu_tx_assign(tx, TXG_WAIT);
4070 zfs_acl_ids_free(&acl_ids);
4072 getnewvnode_drop_reserve();
4078 * Create a new object for the symlink.
4079 * for version 4 ZPL datsets the symlink will be an SA attribute
4081 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
4084 zfs_fuid_sync(zfsvfs, tx);
4087 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
4090 zfs_sa_symlink(zp, link, len, tx);
4093 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
4094 &zp->z_size, sizeof (zp->z_size), tx);
4096 * Insert the new object into the directory.
4098 (void) zfs_link_create(dzp, name, zp, tx, ZNEW);
4100 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
4103 zfs_acl_ids_free(&acl_ids);
4107 getnewvnode_drop_reserve();
4109 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4110 zil_commit(zilog, 0);
4117 * Return, in the buffer contained in the provided uio structure,
4118 * the symbolic path referred to by vp.
4120 * IN: vp - vnode of symbolic link.
4121 * uio - structure to contain the link path.
4122 * cr - credentials of caller.
4123 * ct - caller context
4125 * OUT: uio - structure containing the link path.
4127 * RETURN: 0 on success, error code on failure.
4130 * vp - atime updated
4134 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
4136 znode_t *zp = VTOZ(vp);
4137 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4144 error = sa_lookup_uio(zp->z_sa_hdl,
4145 SA_ZPL_SYMLINK(zfsvfs), uio);
4147 error = zfs_sa_readlink(zp, uio);
4149 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4156 * Insert a new entry into directory tdvp referencing svp.
4158 * IN: tdvp - Directory to contain new entry.
4159 * svp - vnode of new entry.
4160 * name - name of new entry.
4161 * cr - credentials of caller.
4162 * ct - caller context
4164 * RETURN: 0 on success, error code on failure.
4167 * tdvp - ctime|mtime updated
4168 * svp - ctime updated
4172 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
4173 caller_context_t *ct, int flags)
4175 znode_t *dzp = VTOZ(tdvp);
4177 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4184 ASSERT(tdvp->v_type == VDIR);
4188 zilog = zfsvfs->z_log;
4191 * POSIX dictates that we return EPERM here.
4192 * Better choices include ENOTSUP or EISDIR.
4194 if (svp->v_type == VDIR) {
4196 return (SET_ERROR(EPERM));
4202 if (szp->z_pflags & (ZFS_APPENDONLY | ZFS_IMMUTABLE | ZFS_READONLY)) {
4204 return (SET_ERROR(EPERM));
4207 /* Prevent links to .zfs/shares files */
4209 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4210 &parent, sizeof (uint64_t))) != 0) {
4214 if (parent == zfsvfs->z_shares_dir) {
4216 return (SET_ERROR(EPERM));
4219 if (zfsvfs->z_utf8 && u8_validate(name,
4220 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4222 return (SET_ERROR(EILSEQ));
4226 * We do not support links between attributes and non-attributes
4227 * because of the potential security risk of creating links
4228 * into "normal" file space in order to circumvent restrictions
4229 * imposed in attribute space.
4231 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4233 return (SET_ERROR(EINVAL));
4237 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4238 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) {
4240 return (SET_ERROR(EPERM));
4243 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4249 * Attempt to lock directory; fail if entry already exists.
4251 error = zfs_dirent_lookup(dzp, name, &tzp, ZNEW);
4257 tx = dmu_tx_create(zfsvfs->z_os);
4258 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4259 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4260 zfs_sa_upgrade_txholds(tx, szp);
4261 zfs_sa_upgrade_txholds(tx, dzp);
4262 error = dmu_tx_assign(tx, TXG_WAIT);
4269 error = zfs_link_create(dzp, name, szp, tx, 0);
4272 uint64_t txtype = TX_LINK;
4273 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4279 vnevent_link(svp, ct);
4282 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4283 zil_commit(zilog, 0);
4292 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4294 znode_t *zp = VTOZ(vp);
4295 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4298 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4299 if (zp->z_sa_hdl == NULL) {
4301 * The fs has been unmounted, or we did a
4302 * suspend/resume and this file no longer exists.
4304 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4309 if (zp->z_unlinked) {
4311 * Fast path to recycle a vnode of a removed file.
4313 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4318 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4319 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4321 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4322 zfs_sa_upgrade_txholds(tx, zp);
4323 error = dmu_tx_assign(tx, TXG_WAIT);
4327 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4328 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4329 zp->z_atime_dirty = 0;
4333 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4337 CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid));
4338 CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid));
4342 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
4344 znode_t *zp = VTOZ(vp);
4345 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4348 uint64_t object = zp->z_id;
4355 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
4356 &gen64, sizeof (uint64_t))) != 0) {
4361 gen = (uint32_t)gen64;
4363 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
4366 if (fidp->fid_len < size) {
4367 fidp->fid_len = size;
4369 return (SET_ERROR(ENOSPC));
4372 fidp->fid_len = size;
4375 zfid = (zfid_short_t *)fidp;
4377 zfid->zf_len = size;
4379 for (i = 0; i < sizeof (zfid->zf_object); i++)
4380 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4382 /* Must have a non-zero generation number to distinguish from .zfs */
4385 for (i = 0; i < sizeof (zfid->zf_gen); i++)
4386 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4388 if (size == LONG_FID_LEN) {
4389 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
4392 zlfid = (zfid_long_t *)fidp;
4394 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4395 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4397 /* XXX - this should be the generation number for the objset */
4398 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4399 zlfid->zf_setgen[i] = 0;
4407 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4408 caller_context_t *ct)
4416 *valp = MIN(LONG_MAX, ZFS_LINK_MAX);
4419 case _PC_FILESIZEBITS:
4423 case _PC_XATTR_EXISTS:
4425 zfsvfs = zp->z_zfsvfs;
4429 error = zfs_dirent_lookup(zp, "", &xzp,
4430 ZXATTR | ZEXISTS | ZSHARED);
4432 if (!zfs_dirempty(xzp))
4435 } else if (error == ENOENT) {
4437 * If there aren't extended attributes, it's the
4438 * same as having zero of them.
4445 case _PC_SATTR_ENABLED:
4446 case _PC_SATTR_EXISTS:
4447 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
4448 (vp->v_type == VREG || vp->v_type == VDIR);
4451 case _PC_ACCESS_FILTERING:
4452 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
4456 case _PC_ACL_ENABLED:
4457 *valp = _ACL_ACE_ENABLED;
4459 #endif /* illumos */
4460 case _PC_MIN_HOLE_SIZE:
4461 *valp = (int)SPA_MINBLOCKSIZE;
4464 case _PC_TIMESTAMP_RESOLUTION:
4465 /* nanosecond timestamp resolution */
4469 case _PC_ACL_EXTENDED:
4477 case _PC_ACL_PATH_MAX:
4478 *valp = ACL_MAX_ENTRIES;
4482 return (EOPNOTSUPP);
4488 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4489 caller_context_t *ct)
4491 znode_t *zp = VTOZ(vp);
4492 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4494 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4498 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4506 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4507 caller_context_t *ct)
4509 znode_t *zp = VTOZ(vp);
4510 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4512 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4513 zilog_t *zilog = zfsvfs->z_log;
4518 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
4520 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4521 zil_commit(zilog, 0);
4528 zfs_getpages(struct vnode *vp, vm_page_t *ma, int count, int *rbehind,
4531 znode_t *zp = VTOZ(vp);
4532 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4533 objset_t *os = zp->z_zfsvfs->z_os;
4536 off_t start, end, obj_size;
4538 int pgsin_b, pgsin_a;
4544 start = IDX_TO_OFF(ma[0]->pindex);
4545 end = IDX_TO_OFF(ma[count - 1]->pindex + 1);
4548 * Lock a range covering all required and optional pages.
4549 * Note that we need to handle the case of the block size growing.
4552 blksz = zp->z_blksz;
4553 rl = zfs_range_lock(zp, rounddown(start, blksz),
4554 roundup(end, blksz) - rounddown(start, blksz), RL_READER);
4555 if (blksz == zp->z_blksz)
4557 zfs_range_unlock(rl);
4560 object = ma[0]->object;
4561 zfs_vmobject_wlock(object);
4562 obj_size = object->un_pager.vnp.vnp_size;
4563 zfs_vmobject_wunlock(object);
4564 if (IDX_TO_OFF(ma[count - 1]->pindex) >= obj_size) {
4565 zfs_range_unlock(rl);
4567 return (zfs_vm_pagerret_bad);
4571 if (rbehind != NULL) {
4572 pgsin_b = OFF_TO_IDX(start - rounddown(start, blksz));
4573 pgsin_b = MIN(*rbehind, pgsin_b);
4577 if (rahead != NULL) {
4578 pgsin_a = OFF_TO_IDX(roundup(end, blksz) - end);
4579 if (end + IDX_TO_OFF(pgsin_a) >= obj_size)
4580 pgsin_a = OFF_TO_IDX(round_page(obj_size) - end);
4581 pgsin_a = MIN(*rahead, pgsin_a);
4585 * NB: we need to pass the exact byte size of the data that we expect
4586 * to read after accounting for the file size. This is required because
4587 * ZFS will panic if we request DMU to read beyond the end of the last
4590 error = dmu_read_pages(os, zp->z_id, ma, count, &pgsin_b, &pgsin_a,
4591 MIN(end, obj_size) - (end - PAGE_SIZE));
4593 zfs_range_unlock(rl);
4594 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4598 return (zfs_vm_pagerret_error);
4600 VM_CNT_INC(v_vnodein);
4601 VM_CNT_ADD(v_vnodepgsin, count + pgsin_b + pgsin_a);
4602 if (rbehind != NULL)
4606 return (zfs_vm_pagerret_ok);
4610 zfs_freebsd_getpages(ap)
4611 struct vop_getpages_args /* {
4620 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_rbehind,
4625 zfs_putpages(struct vnode *vp, vm_page_t *ma, size_t len, int flags,
4628 znode_t *zp = VTOZ(vp);
4629 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4638 vm_ooffset_t lo_off;
4649 object = vp->v_object;
4653 KASSERT(ma[0]->object == object, ("mismatching object"));
4654 KASSERT(len > 0 && (len & PAGE_MASK) == 0, ("unexpected length"));
4656 for (i = 0; i < pcount; i++)
4657 rtvals[i] = zfs_vm_pagerret_error;
4659 off = IDX_TO_OFF(ma[0]->pindex);
4660 blksz = zp->z_blksz;
4661 lo_off = rounddown(off, blksz);
4662 lo_len = roundup(len + (off - lo_off), blksz);
4663 rl = zfs_range_lock(zp, lo_off, lo_len, RL_WRITER);
4665 zfs_vmobject_wlock(object);
4666 if (len + off > object->un_pager.vnp.vnp_size) {
4667 if (object->un_pager.vnp.vnp_size > off) {
4670 len = object->un_pager.vnp.vnp_size - off;
4672 if ((pgoff = (int)len & PAGE_MASK) != 0) {
4674 * If the object is locked and the following
4675 * conditions hold, then the page's dirty
4676 * field cannot be concurrently changed by a
4680 vm_page_assert_sbusied(m);
4681 KASSERT(!pmap_page_is_write_mapped(m),
4682 ("zfs_putpages: page %p is not read-only", m));
4683 vm_page_clear_dirty(m, pgoff, PAGE_SIZE -
4690 if (ncount < pcount) {
4691 for (i = ncount; i < pcount; i++) {
4692 rtvals[i] = zfs_vm_pagerret_bad;
4696 zfs_vmobject_wunlock(object);
4701 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4702 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4706 tx = dmu_tx_create(zfsvfs->z_os);
4707 dmu_tx_hold_write(tx, zp->z_id, off, len);
4709 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4710 zfs_sa_upgrade_txholds(tx, zp);
4711 err = dmu_tx_assign(tx, TXG_WAIT);
4717 if (zp->z_blksz < PAGE_SIZE) {
4718 for (i = 0; len > 0; off += tocopy, len -= tocopy, i++) {
4719 tocopy = len > PAGE_SIZE ? PAGE_SIZE : len;
4720 va = zfs_map_page(ma[i], &sf);
4721 dmu_write(zfsvfs->z_os, zp->z_id, off, tocopy, va, tx);
4725 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, ma, tx);
4729 uint64_t mtime[2], ctime[2];
4730 sa_bulk_attr_t bulk[3];
4733 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4735 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4737 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4739 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4741 err = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
4743 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4745 zfs_vmobject_wlock(object);
4746 for (i = 0; i < ncount; i++) {
4747 rtvals[i] = zfs_vm_pagerret_ok;
4748 vm_page_undirty(ma[i]);
4750 zfs_vmobject_wunlock(object);
4751 VM_CNT_INC(v_vnodeout);
4752 VM_CNT_ADD(v_vnodepgsout, ncount);
4757 zfs_range_unlock(rl);
4758 if ((flags & (zfs_vm_pagerput_sync | zfs_vm_pagerput_inval)) != 0 ||
4759 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4760 zil_commit(zfsvfs->z_log, zp->z_id);
4766 zfs_freebsd_putpages(ap)
4767 struct vop_putpages_args /* {
4776 return (zfs_putpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_sync,
4781 zfs_freebsd_bmap(ap)
4782 struct vop_bmap_args /* {
4785 struct bufobj **a_bop;
4792 if (ap->a_bop != NULL)
4793 *ap->a_bop = &ap->a_vp->v_bufobj;
4794 if (ap->a_bnp != NULL)
4795 *ap->a_bnp = ap->a_bn;
4796 if (ap->a_runp != NULL)
4798 if (ap->a_runb != NULL)
4805 zfs_freebsd_open(ap)
4806 struct vop_open_args /* {
4809 struct ucred *a_cred;
4810 struct thread *a_td;
4813 vnode_t *vp = ap->a_vp;
4814 znode_t *zp = VTOZ(vp);
4817 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL);
4819 vnode_create_vobject(vp, zp->z_size, ap->a_td);
4824 zfs_freebsd_close(ap)
4825 struct vop_close_args /* {
4828 struct ucred *a_cred;
4829 struct thread *a_td;
4833 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL));
4837 zfs_freebsd_ioctl(ap)
4838 struct vop_ioctl_args /* {
4848 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data,
4849 ap->a_fflag, ap->a_cred, NULL, NULL));
4853 ioflags(int ioflags)
4857 if (ioflags & IO_APPEND)
4859 if (ioflags & IO_NDELAY)
4861 if (ioflags & IO_SYNC)
4862 flags |= (FSYNC | FDSYNC | FRSYNC);
4868 zfs_freebsd_read(ap)
4869 struct vop_read_args /* {
4873 struct ucred *a_cred;
4877 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
4882 zfs_freebsd_write(ap)
4883 struct vop_write_args /* {
4887 struct ucred *a_cred;
4891 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
4896 zfs_freebsd_access(ap)
4897 struct vop_access_args /* {
4899 accmode_t a_accmode;
4900 struct ucred *a_cred;
4901 struct thread *a_td;
4904 vnode_t *vp = ap->a_vp;
4905 znode_t *zp = VTOZ(vp);
4910 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND,
4912 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND);
4914 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL);
4917 * VADMIN has to be handled by vaccess().
4920 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND);
4922 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
4923 zp->z_gid, accmode, ap->a_cred, NULL);
4928 * For VEXEC, ensure that at least one execute bit is set for
4931 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR &&
4932 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
4940 zfs_freebsd_lookup(ap)
4941 struct vop_lookup_args /* {
4942 struct vnode *a_dvp;
4943 struct vnode **a_vpp;
4944 struct componentname *a_cnp;
4947 struct componentname *cnp = ap->a_cnp;
4948 char nm[NAME_MAX + 1];
4950 ASSERT(cnp->cn_namelen < sizeof(nm));
4951 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm)));
4953 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop,
4954 cnp->cn_cred, cnp->cn_thread, 0));
4958 zfs_cache_lookup(ap)
4959 struct vop_lookup_args /* {
4960 struct vnode *a_dvp;
4961 struct vnode **a_vpp;
4962 struct componentname *a_cnp;
4967 zfsvfs = ap->a_dvp->v_mount->mnt_data;
4968 if (zfsvfs->z_use_namecache)
4969 return (vfs_cache_lookup(ap));
4971 return (zfs_freebsd_lookup(ap));
4975 zfs_freebsd_create(ap)
4976 struct vop_create_args /* {
4977 struct vnode *a_dvp;
4978 struct vnode **a_vpp;
4979 struct componentname *a_cnp;
4980 struct vattr *a_vap;
4984 struct componentname *cnp = ap->a_cnp;
4985 vattr_t *vap = ap->a_vap;
4988 ASSERT(cnp->cn_flags & SAVENAME);
4990 vattr_init_mask(vap);
4991 mode = vap->va_mode & ALLPERMS;
4992 zfsvfs = ap->a_dvp->v_mount->mnt_data;
4994 error = zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode,
4995 ap->a_vpp, cnp->cn_cred, cnp->cn_thread);
4996 if (zfsvfs->z_use_namecache &&
4997 error == 0 && (cnp->cn_flags & MAKEENTRY) != 0)
4998 cache_enter(ap->a_dvp, *ap->a_vpp, cnp);
5003 zfs_freebsd_remove(ap)
5004 struct vop_remove_args /* {
5005 struct vnode *a_dvp;
5007 struct componentname *a_cnp;
5011 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
5013 return (zfs_remove(ap->a_dvp, ap->a_vp, ap->a_cnp->cn_nameptr,
5014 ap->a_cnp->cn_cred));
5018 zfs_freebsd_mkdir(ap)
5019 struct vop_mkdir_args /* {
5020 struct vnode *a_dvp;
5021 struct vnode **a_vpp;
5022 struct componentname *a_cnp;
5023 struct vattr *a_vap;
5026 vattr_t *vap = ap->a_vap;
5028 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
5030 vattr_init_mask(vap);
5032 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp,
5033 ap->a_cnp->cn_cred));
5037 zfs_freebsd_rmdir(ap)
5038 struct vop_rmdir_args /* {
5039 struct vnode *a_dvp;
5041 struct componentname *a_cnp;
5044 struct componentname *cnp = ap->a_cnp;
5046 ASSERT(cnp->cn_flags & SAVENAME);
5048 return (zfs_rmdir(ap->a_dvp, ap->a_vp, cnp->cn_nameptr, cnp->cn_cred));
5052 zfs_freebsd_readdir(ap)
5053 struct vop_readdir_args /* {
5056 struct ucred *a_cred;
5063 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag,
5064 ap->a_ncookies, ap->a_cookies));
5068 zfs_freebsd_fsync(ap)
5069 struct vop_fsync_args /* {
5072 struct thread *a_td;
5077 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL));
5081 zfs_freebsd_getattr(ap)
5082 struct vop_getattr_args /* {
5084 struct vattr *a_vap;
5085 struct ucred *a_cred;
5088 vattr_t *vap = ap->a_vap;
5094 xvap.xva_vattr = *vap;
5095 xvap.xva_vattr.va_mask |= AT_XVATTR;
5097 /* Convert chflags into ZFS-type flags. */
5098 /* XXX: what about SF_SETTABLE?. */
5099 XVA_SET_REQ(&xvap, XAT_IMMUTABLE);
5100 XVA_SET_REQ(&xvap, XAT_APPENDONLY);
5101 XVA_SET_REQ(&xvap, XAT_NOUNLINK);
5102 XVA_SET_REQ(&xvap, XAT_NODUMP);
5103 XVA_SET_REQ(&xvap, XAT_READONLY);
5104 XVA_SET_REQ(&xvap, XAT_ARCHIVE);
5105 XVA_SET_REQ(&xvap, XAT_SYSTEM);
5106 XVA_SET_REQ(&xvap, XAT_HIDDEN);
5107 XVA_SET_REQ(&xvap, XAT_REPARSE);
5108 XVA_SET_REQ(&xvap, XAT_OFFLINE);
5109 XVA_SET_REQ(&xvap, XAT_SPARSE);
5111 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL);
5115 /* Convert ZFS xattr into chflags. */
5116 #define FLAG_CHECK(fflag, xflag, xfield) do { \
5117 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \
5118 fflags |= (fflag); \
5120 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE,
5121 xvap.xva_xoptattrs.xoa_immutable);
5122 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY,
5123 xvap.xva_xoptattrs.xoa_appendonly);
5124 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK,
5125 xvap.xva_xoptattrs.xoa_nounlink);
5126 FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE,
5127 xvap.xva_xoptattrs.xoa_archive);
5128 FLAG_CHECK(UF_NODUMP, XAT_NODUMP,
5129 xvap.xva_xoptattrs.xoa_nodump);
5130 FLAG_CHECK(UF_READONLY, XAT_READONLY,
5131 xvap.xva_xoptattrs.xoa_readonly);
5132 FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM,
5133 xvap.xva_xoptattrs.xoa_system);
5134 FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN,
5135 xvap.xva_xoptattrs.xoa_hidden);
5136 FLAG_CHECK(UF_REPARSE, XAT_REPARSE,
5137 xvap.xva_xoptattrs.xoa_reparse);
5138 FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE,
5139 xvap.xva_xoptattrs.xoa_offline);
5140 FLAG_CHECK(UF_SPARSE, XAT_SPARSE,
5141 xvap.xva_xoptattrs.xoa_sparse);
5144 *vap = xvap.xva_vattr;
5145 vap->va_flags = fflags;
5150 zfs_freebsd_setattr(ap)
5151 struct vop_setattr_args /* {
5153 struct vattr *a_vap;
5154 struct ucred *a_cred;
5157 vnode_t *vp = ap->a_vp;
5158 vattr_t *vap = ap->a_vap;
5159 cred_t *cred = ap->a_cred;
5164 vattr_init_mask(vap);
5165 vap->va_mask &= ~AT_NOSET;
5168 xvap.xva_vattr = *vap;
5170 zflags = VTOZ(vp)->z_pflags;
5172 if (vap->va_flags != VNOVAL) {
5173 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs;
5176 if (zfsvfs->z_use_fuids == B_FALSE)
5177 return (EOPNOTSUPP);
5179 fflags = vap->va_flags;
5182 * We need to figure out whether it makes sense to allow
5183 * UF_REPARSE through, since we don't really have other
5184 * facilities to handle reparse points and zfs_setattr()
5185 * doesn't currently allow setting that attribute anyway.
5187 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE|
5188 UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE|
5189 UF_OFFLINE|UF_SPARSE)) != 0)
5190 return (EOPNOTSUPP);
5192 * Unprivileged processes are not permitted to unset system
5193 * flags, or modify flags if any system flags are set.
5194 * Privileged non-jail processes may not modify system flags
5195 * if securelevel > 0 and any existing system flags are set.
5196 * Privileged jail processes behave like privileged non-jail
5197 * processes if the PR_ALLOW_CHFLAGS permission bit is set;
5198 * otherwise, they behave like unprivileged processes.
5200 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 ||
5201 priv_check_cred(cred, PRIV_VFS_SYSFLAGS) == 0) {
5203 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
5204 error = securelevel_gt(cred, 0);
5210 * Callers may only modify the file flags on objects they
5211 * have VADMIN rights for.
5213 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0)
5216 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
5220 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) {
5225 #define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \
5226 if (((fflags & (fflag)) && !(zflags & (zflag))) || \
5227 ((zflags & (zflag)) && !(fflags & (fflag)))) { \
5228 XVA_SET_REQ(&xvap, (xflag)); \
5229 (xfield) = ((fflags & (fflag)) != 0); \
5232 /* Convert chflags into ZFS-type flags. */
5233 /* XXX: what about SF_SETTABLE?. */
5234 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE,
5235 xvap.xva_xoptattrs.xoa_immutable);
5236 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY,
5237 xvap.xva_xoptattrs.xoa_appendonly);
5238 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK,
5239 xvap.xva_xoptattrs.xoa_nounlink);
5240 FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE,
5241 xvap.xva_xoptattrs.xoa_archive);
5242 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP,
5243 xvap.xva_xoptattrs.xoa_nodump);
5244 FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY,
5245 xvap.xva_xoptattrs.xoa_readonly);
5246 FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM,
5247 xvap.xva_xoptattrs.xoa_system);
5248 FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN,
5249 xvap.xva_xoptattrs.xoa_hidden);
5250 FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE,
5251 xvap.xva_xoptattrs.xoa_hidden);
5252 FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE,
5253 xvap.xva_xoptattrs.xoa_offline);
5254 FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE,
5255 xvap.xva_xoptattrs.xoa_sparse);
5258 if (vap->va_birthtime.tv_sec != VNOVAL) {
5259 xvap.xva_vattr.va_mask |= AT_XVATTR;
5260 XVA_SET_REQ(&xvap, XAT_CREATETIME);
5262 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL));
5266 zfs_freebsd_rename(ap)
5267 struct vop_rename_args /* {
5268 struct vnode *a_fdvp;
5269 struct vnode *a_fvp;
5270 struct componentname *a_fcnp;
5271 struct vnode *a_tdvp;
5272 struct vnode *a_tvp;
5273 struct componentname *a_tcnp;
5276 vnode_t *fdvp = ap->a_fdvp;
5277 vnode_t *fvp = ap->a_fvp;
5278 vnode_t *tdvp = ap->a_tdvp;
5279 vnode_t *tvp = ap->a_tvp;
5282 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART));
5283 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART));
5285 error = zfs_rename(fdvp, &fvp, ap->a_fcnp, tdvp, &tvp,
5286 ap->a_tcnp, ap->a_fcnp->cn_cred);
5298 zfs_freebsd_symlink(ap)
5299 struct vop_symlink_args /* {
5300 struct vnode *a_dvp;
5301 struct vnode **a_vpp;
5302 struct componentname *a_cnp;
5303 struct vattr *a_vap;
5307 struct componentname *cnp = ap->a_cnp;
5308 vattr_t *vap = ap->a_vap;
5310 ASSERT(cnp->cn_flags & SAVENAME);
5312 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */
5313 vattr_init_mask(vap);
5315 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap,
5316 __DECONST(char *, ap->a_target), cnp->cn_cred, cnp->cn_thread));
5320 zfs_freebsd_readlink(ap)
5321 struct vop_readlink_args /* {
5324 struct ucred *a_cred;
5328 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL));
5332 zfs_freebsd_link(ap)
5333 struct vop_link_args /* {
5334 struct vnode *a_tdvp;
5336 struct componentname *a_cnp;
5339 struct componentname *cnp = ap->a_cnp;
5340 vnode_t *vp = ap->a_vp;
5341 vnode_t *tdvp = ap->a_tdvp;
5343 if (tdvp->v_mount != vp->v_mount)
5346 ASSERT(cnp->cn_flags & SAVENAME);
5348 return (zfs_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0));
5352 zfs_freebsd_inactive(ap)
5353 struct vop_inactive_args /* {
5355 struct thread *a_td;
5358 vnode_t *vp = ap->a_vp;
5360 zfs_inactive(vp, ap->a_td->td_ucred, NULL);
5365 zfs_freebsd_reclaim(ap)
5366 struct vop_reclaim_args /* {
5368 struct thread *a_td;
5371 vnode_t *vp = ap->a_vp;
5372 znode_t *zp = VTOZ(vp);
5373 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5377 /* Destroy the vm object and flush associated pages. */
5378 vnode_destroy_vobject(vp);
5381 * z_teardown_inactive_lock protects from a race with
5382 * zfs_znode_dmu_fini in zfsvfs_teardown during
5385 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
5386 if (zp->z_sa_hdl == NULL)
5390 rw_exit(&zfsvfs->z_teardown_inactive_lock);
5398 struct vop_fid_args /* {
5404 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL));
5408 zfs_freebsd_pathconf(ap)
5409 struct vop_pathconf_args /* {
5412 register_t *a_retval;
5418 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL);
5420 *ap->a_retval = val;
5423 if (error != EOPNOTSUPP)
5426 switch (ap->a_name) {
5428 *ap->a_retval = NAME_MAX;
5431 if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO) {
5432 *ap->a_retval = PIPE_BUF;
5437 return (vop_stdpathconf(ap));
5442 * FreeBSD's extended attributes namespace defines file name prefix for ZFS'
5443 * extended attribute name:
5446 * system freebsd:system:
5447 * user (none, can be used to access ZFS fsattr(5) attributes
5448 * created on Solaris)
5451 zfs_create_attrname(int attrnamespace, const char *name, char *attrname,
5454 const char *namespace, *prefix, *suffix;
5456 /* We don't allow '/' character in attribute name. */
5457 if (strchr(name, '/') != NULL)
5459 /* We don't allow attribute names that start with "freebsd:" string. */
5460 if (strncmp(name, "freebsd:", 8) == 0)
5463 bzero(attrname, size);
5465 switch (attrnamespace) {
5466 case EXTATTR_NAMESPACE_USER:
5468 prefix = "freebsd:";
5469 namespace = EXTATTR_NAMESPACE_USER_STRING;
5473 * This is the default namespace by which we can access all
5474 * attributes created on Solaris.
5476 prefix = namespace = suffix = "";
5479 case EXTATTR_NAMESPACE_SYSTEM:
5480 prefix = "freebsd:";
5481 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING;
5484 case EXTATTR_NAMESPACE_EMPTY:
5488 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix,
5490 return (ENAMETOOLONG);
5496 * Vnode operating to retrieve a named extended attribute.
5499 zfs_getextattr(struct vop_getextattr_args *ap)
5502 IN struct vnode *a_vp;
5503 IN int a_attrnamespace;
5504 IN const char *a_name;
5505 INOUT struct uio *a_uio;
5507 IN struct ucred *a_cred;
5508 IN struct thread *a_td;
5512 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
5513 struct thread *td = ap->a_td;
5514 struct nameidata nd;
5517 vnode_t *xvp = NULL, *vp;
5520 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
5521 ap->a_cred, ap->a_td, VREAD);
5525 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
5532 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
5540 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
5542 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL);
5544 NDFREE(&nd, NDF_ONLY_PNBUF);
5547 if (error == ENOENT)
5552 if (ap->a_size != NULL) {
5553 error = VOP_GETATTR(vp, &va, ap->a_cred);
5555 *ap->a_size = (size_t)va.va_size;
5556 } else if (ap->a_uio != NULL)
5557 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred);
5560 vn_close(vp, flags, ap->a_cred, td);
5567 * Vnode operation to remove a named attribute.
5570 zfs_deleteextattr(struct vop_deleteextattr_args *ap)
5573 IN struct vnode *a_vp;
5574 IN int a_attrnamespace;
5575 IN const char *a_name;
5576 IN struct ucred *a_cred;
5577 IN struct thread *a_td;
5581 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
5582 struct thread *td = ap->a_td;
5583 struct nameidata nd;
5586 vnode_t *xvp = NULL, *vp;
5589 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
5590 ap->a_cred, ap->a_td, VWRITE);
5594 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
5601 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
5608 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF,
5609 UIO_SYSSPACE, attrname, xvp, td);
5614 NDFREE(&nd, NDF_ONLY_PNBUF);
5615 if (error == ENOENT)
5620 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
5621 NDFREE(&nd, NDF_ONLY_PNBUF);
5624 if (vp == nd.ni_dvp)
5634 * Vnode operation to set a named attribute.
5637 zfs_setextattr(struct vop_setextattr_args *ap)
5640 IN struct vnode *a_vp;
5641 IN int a_attrnamespace;
5642 IN const char *a_name;
5643 INOUT struct uio *a_uio;
5644 IN struct ucred *a_cred;
5645 IN struct thread *a_td;
5649 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
5650 struct thread *td = ap->a_td;
5651 struct nameidata nd;
5654 vnode_t *xvp = NULL, *vp;
5657 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
5658 ap->a_cred, ap->a_td, VWRITE);
5662 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
5669 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
5670 LOOKUP_XATTR | CREATE_XATTR_DIR);
5676 flags = FFLAGS(O_WRONLY | O_CREAT);
5677 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
5679 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL);
5681 NDFREE(&nd, NDF_ONLY_PNBUF);
5689 error = VOP_SETATTR(vp, &va, ap->a_cred);
5691 VOP_WRITE(vp, ap->a_uio, IO_UNIT, ap->a_cred);
5694 vn_close(vp, flags, ap->a_cred, td);
5701 * Vnode operation to retrieve extended attributes on a vnode.
5704 zfs_listextattr(struct vop_listextattr_args *ap)
5707 IN struct vnode *a_vp;
5708 IN int a_attrnamespace;
5709 INOUT struct uio *a_uio;
5711 IN struct ucred *a_cred;
5712 IN struct thread *a_td;
5716 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
5717 struct thread *td = ap->a_td;
5718 struct nameidata nd;
5719 char attrprefix[16];
5720 u_char dirbuf[sizeof(struct dirent)];
5723 struct uio auio, *uio = ap->a_uio;
5724 size_t *sizep = ap->a_size;
5726 vnode_t *xvp = NULL, *vp;
5727 int done, error, eof, pos;
5729 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
5730 ap->a_cred, ap->a_td, VREAD);
5734 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix,
5735 sizeof(attrprefix));
5738 plen = strlen(attrprefix);
5745 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
5750 * ENOATTR means that the EA directory does not yet exist,
5751 * i.e. there are no extended attributes there.
5753 if (error == ENOATTR)
5758 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED,
5759 UIO_SYSSPACE, ".", xvp, td);
5762 NDFREE(&nd, NDF_ONLY_PNBUF);
5768 auio.uio_iov = &aiov;
5769 auio.uio_iovcnt = 1;
5770 auio.uio_segflg = UIO_SYSSPACE;
5772 auio.uio_rw = UIO_READ;
5773 auio.uio_offset = 0;
5778 aiov.iov_base = (void *)dirbuf;
5779 aiov.iov_len = sizeof(dirbuf);
5780 auio.uio_resid = sizeof(dirbuf);
5781 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL);
5782 done = sizeof(dirbuf) - auio.uio_resid;
5785 for (pos = 0; pos < done;) {
5786 dp = (struct dirent *)(dirbuf + pos);
5787 pos += dp->d_reclen;
5789 * XXX: Temporarily we also accept DT_UNKNOWN, as this
5790 * is what we get when attribute was created on Solaris.
5792 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN)
5794 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0)
5796 else if (strncmp(dp->d_name, attrprefix, plen) != 0)
5798 nlen = dp->d_namlen - plen;
5801 else if (uio != NULL) {
5803 * Format of extattr name entry is one byte for
5804 * length and the rest for name.
5806 error = uiomove(&nlen, 1, uio->uio_rw, uio);
5808 error = uiomove(dp->d_name + plen, nlen,
5815 } while (!eof && error == 0);
5824 zfs_freebsd_getacl(ap)
5825 struct vop_getacl_args /* {
5834 vsecattr_t vsecattr;
5836 if (ap->a_type != ACL_TYPE_NFS4)
5839 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
5840 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL))
5843 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt);
5844 if (vsecattr.vsa_aclentp != NULL)
5845 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz);
5851 zfs_freebsd_setacl(ap)
5852 struct vop_setacl_args /* {
5861 vsecattr_t vsecattr;
5862 int aclbsize; /* size of acl list in bytes */
5865 if (ap->a_type != ACL_TYPE_NFS4)
5868 if (ap->a_aclp == NULL)
5871 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES)
5875 * With NFSv4 ACLs, chmod(2) may need to add additional entries,
5876 * splitting every entry into two and appending "canonical six"
5877 * entries at the end. Don't allow for setting an ACL that would
5878 * cause chmod(2) to run out of ACL entries.
5880 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES)
5883 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR);
5887 vsecattr.vsa_mask = VSA_ACE;
5888 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t);
5889 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP);
5890 aaclp = vsecattr.vsa_aclentp;
5891 vsecattr.vsa_aclentsz = aclbsize;
5893 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp);
5894 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL);
5895 kmem_free(aaclp, aclbsize);
5901 zfs_freebsd_aclcheck(ap)
5902 struct vop_aclcheck_args /* {
5911 return (EOPNOTSUPP);
5915 zfs_vptocnp(struct vop_vptocnp_args *ap)
5917 vnode_t *covered_vp;
5918 vnode_t *vp = ap->a_vp;;
5919 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
5920 znode_t *zp = VTOZ(vp);
5928 * If we are a snapshot mounted under .zfs, run the operation
5929 * on the covered vnode.
5931 if (zp->z_id != zfsvfs->z_root || zfsvfs->z_parent == zfsvfs) {
5932 char name[MAXNAMLEN + 1];
5936 error = zfs_znode_parent_and_name(zp, &dzp, name);
5939 if (*ap->a_buflen < len)
5940 error = SET_ERROR(ENOMEM);
5943 *ap->a_buflen -= len;
5944 bcopy(name, ap->a_buf + *ap->a_buflen, len);
5945 *ap->a_vpp = ZTOV(dzp);
5952 covered_vp = vp->v_mount->mnt_vnodecovered;
5954 ltype = VOP_ISLOCKED(vp);
5956 error = vget(covered_vp, LK_SHARED | LK_VNHELD, curthread);
5958 error = VOP_VPTOCNP(covered_vp, ap->a_vpp, ap->a_cred,
5959 ap->a_buf, ap->a_buflen);
5962 vn_lock(vp, ltype | LK_RETRY);
5963 if ((vp->v_iflag & VI_DOOMED) != 0)
5964 error = SET_ERROR(ENOENT);
5971 struct vop_lock1_args /* {
5982 err = vop_stdlock(ap);
5983 if (err == 0 && (ap->a_flags & LK_NOWAIT) == 0) {
5986 if (vp->v_mount != NULL && (vp->v_iflag & VI_DOOMED) == 0 &&
5987 zp != NULL && (zp->z_pflags & ZFS_XATTR) == 0)
5988 VERIFY(!RRM_LOCK_HELD(&zp->z_zfsvfs->z_teardown_lock));
5994 struct vop_vector zfs_vnodeops;
5995 struct vop_vector zfs_fifoops;
5996 struct vop_vector zfs_shareops;
5998 struct vop_vector zfs_vnodeops = {
5999 .vop_default = &default_vnodeops,
6000 .vop_inactive = zfs_freebsd_inactive,
6001 .vop_reclaim = zfs_freebsd_reclaim,
6002 .vop_access = zfs_freebsd_access,
6003 .vop_allocate = VOP_EINVAL,
6004 .vop_lookup = zfs_cache_lookup,
6005 .vop_cachedlookup = zfs_freebsd_lookup,
6006 .vop_getattr = zfs_freebsd_getattr,
6007 .vop_setattr = zfs_freebsd_setattr,
6008 .vop_create = zfs_freebsd_create,
6009 .vop_mknod = zfs_freebsd_create,
6010 .vop_mkdir = zfs_freebsd_mkdir,
6011 .vop_readdir = zfs_freebsd_readdir,
6012 .vop_fsync = zfs_freebsd_fsync,
6013 .vop_open = zfs_freebsd_open,
6014 .vop_close = zfs_freebsd_close,
6015 .vop_rmdir = zfs_freebsd_rmdir,
6016 .vop_ioctl = zfs_freebsd_ioctl,
6017 .vop_link = zfs_freebsd_link,
6018 .vop_symlink = zfs_freebsd_symlink,
6019 .vop_readlink = zfs_freebsd_readlink,
6020 .vop_read = zfs_freebsd_read,
6021 .vop_write = zfs_freebsd_write,
6022 .vop_remove = zfs_freebsd_remove,
6023 .vop_rename = zfs_freebsd_rename,
6024 .vop_pathconf = zfs_freebsd_pathconf,
6025 .vop_bmap = zfs_freebsd_bmap,
6026 .vop_fid = zfs_freebsd_fid,
6027 .vop_getextattr = zfs_getextattr,
6028 .vop_deleteextattr = zfs_deleteextattr,
6029 .vop_setextattr = zfs_setextattr,
6030 .vop_listextattr = zfs_listextattr,
6031 .vop_getacl = zfs_freebsd_getacl,
6032 .vop_setacl = zfs_freebsd_setacl,
6033 .vop_aclcheck = zfs_freebsd_aclcheck,
6034 .vop_getpages = zfs_freebsd_getpages,
6035 .vop_putpages = zfs_freebsd_putpages,
6036 .vop_vptocnp = zfs_vptocnp,
6038 .vop_lock1 = zfs_lock,
6042 struct vop_vector zfs_fifoops = {
6043 .vop_default = &fifo_specops,
6044 .vop_fsync = zfs_freebsd_fsync,
6045 .vop_access = zfs_freebsd_access,
6046 .vop_getattr = zfs_freebsd_getattr,
6047 .vop_inactive = zfs_freebsd_inactive,
6048 .vop_read = VOP_PANIC,
6049 .vop_reclaim = zfs_freebsd_reclaim,
6050 .vop_setattr = zfs_freebsd_setattr,
6051 .vop_write = VOP_PANIC,
6052 .vop_pathconf = zfs_freebsd_pathconf,
6053 .vop_fid = zfs_freebsd_fid,
6054 .vop_getacl = zfs_freebsd_getacl,
6055 .vop_setacl = zfs_freebsd_setacl,
6056 .vop_aclcheck = zfs_freebsd_aclcheck,
6060 * special share hidden files vnode operations template
6062 struct vop_vector zfs_shareops = {
6063 .vop_default = &default_vnodeops,
6064 .vop_access = zfs_freebsd_access,
6065 .vop_inactive = zfs_freebsd_inactive,
6066 .vop_reclaim = zfs_freebsd_reclaim,
6067 .vop_fid = zfs_freebsd_fid,
6068 .vop_pathconf = zfs_freebsd_pathconf,