4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
25 /* Portions Copyright 2007 Jeremy Teo */
26 /* Portions Copyright 2010 Robert Milkowski */
28 #include <sys/types.h>
29 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/sysmacros.h>
33 #include <sys/resource.h>
34 #include <sys/resourcevar.h>
36 #include <sys/vnode.h>
40 #include <sys/taskq.h>
42 #include <sys/atomic.h>
43 #include <sys/namei.h>
45 #include <sys/cmn_err.h>
46 #include <sys/errno.h>
47 #include <sys/unistd.h>
48 #include <sys/zfs_dir.h>
49 #include <sys/zfs_ioctl.h>
50 #include <sys/fs/zfs.h>
52 #include <sys/dmu_objset.h>
58 #include <sys/dirent.h>
59 #include <sys/policy.h>
60 #include <sys/sunddi.h>
61 #include <sys/filio.h>
63 #include <sys/zfs_ctldir.h>
64 #include <sys/zfs_fuid.h>
65 #include <sys/zfs_sa.h>
67 #include <sys/zfs_rlock.h>
68 #include <sys/extdirent.h>
69 #include <sys/kidmap.h>
72 #include <sys/sf_buf.h>
73 #include <sys/sched.h>
75 #include <vm/vm_pageout.h>
80 * Each vnode op performs some logical unit of work. To do this, the ZPL must
81 * properly lock its in-core state, create a DMU transaction, do the work,
82 * record this work in the intent log (ZIL), commit the DMU transaction,
83 * and wait for the intent log to commit if it is a synchronous operation.
84 * Moreover, the vnode ops must work in both normal and log replay context.
85 * The ordering of events is important to avoid deadlocks and references
86 * to freed memory. The example below illustrates the following Big Rules:
88 * (1) A check must be made in each zfs thread for a mounted file system.
89 * This is done avoiding races using ZFS_ENTER(zfsvfs).
90 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
91 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
92 * can return EIO from the calling function.
94 * (2) VN_RELE() should always be the last thing except for zil_commit()
95 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
96 * First, if it's the last reference, the vnode/znode
97 * can be freed, so the zp may point to freed memory. Second, the last
98 * reference will call zfs_zinactive(), which may induce a lot of work --
99 * pushing cached pages (which acquires range locks) and syncing out
100 * cached atime changes. Third, zfs_zinactive() may require a new tx,
101 * which could deadlock the system if you were already holding one.
102 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
104 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
105 * as they can span dmu_tx_assign() calls.
107 * (4) Always pass TXG_NOWAIT as the second argument to dmu_tx_assign().
108 * This is critical because we don't want to block while holding locks.
109 * Note, in particular, that if a lock is sometimes acquired before
110 * the tx assigns, and sometimes after (e.g. z_lock), then failing to
111 * use a non-blocking assign can deadlock the system. The scenario:
113 * Thread A has grabbed a lock before calling dmu_tx_assign().
114 * Thread B is in an already-assigned tx, and blocks for this lock.
115 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
116 * forever, because the previous txg can't quiesce until B's tx commits.
118 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
119 * then drop all locks, call dmu_tx_wait(), and try again.
121 * (5) If the operation succeeded, generate the intent log entry for it
122 * before dropping locks. This ensures that the ordering of events
123 * in the intent log matches the order in which they actually occurred.
124 * During ZIL replay the zfs_log_* functions will update the sequence
125 * number to indicate the zil transaction has replayed.
127 * (6) At the end of each vnode op, the DMU tx must always commit,
128 * regardless of whether there were any errors.
130 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
131 * to ensure that synchronous semantics are provided when necessary.
133 * In general, this is how things should be ordered in each vnode op:
135 * ZFS_ENTER(zfsvfs); // exit if unmounted
137 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD())
138 * rw_enter(...); // grab any other locks you need
139 * tx = dmu_tx_create(...); // get DMU tx
140 * dmu_tx_hold_*(); // hold each object you might modify
141 * error = dmu_tx_assign(tx, TXG_NOWAIT); // try to assign
143 * rw_exit(...); // drop locks
144 * zfs_dirent_unlock(dl); // unlock directory entry
145 * VN_RELE(...); // release held vnodes
146 * if (error == ERESTART) {
151 * dmu_tx_abort(tx); // abort DMU tx
152 * ZFS_EXIT(zfsvfs); // finished in zfs
153 * return (error); // really out of space
155 * error = do_real_work(); // do whatever this VOP does
157 * zfs_log_*(...); // on success, make ZIL entry
158 * dmu_tx_commit(tx); // commit DMU tx -- error or not
159 * rw_exit(...); // drop locks
160 * zfs_dirent_unlock(dl); // unlock directory entry
161 * VN_RELE(...); // release held vnodes
162 * zil_commit(zilog, foid); // synchronous when necessary
163 * ZFS_EXIT(zfsvfs); // finished in zfs
164 * return (error); // done, report error
169 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
171 znode_t *zp = VTOZ(*vpp);
172 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
177 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
178 ((flag & FAPPEND) == 0)) {
183 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
184 ZTOV(zp)->v_type == VREG &&
185 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
186 if (fs_vscan(*vpp, cr, 0) != 0) {
192 /* Keep a count of the synchronous opens in the znode */
193 if (flag & (FSYNC | FDSYNC))
194 atomic_inc_32(&zp->z_sync_cnt);
202 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
203 caller_context_t *ct)
205 znode_t *zp = VTOZ(vp);
206 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
209 * Clean up any locks held by this process on the vp.
211 cleanlocks(vp, ddi_get_pid(), 0);
212 cleanshares(vp, ddi_get_pid());
217 /* Decrement the synchronous opens in the znode */
218 if ((flag & (FSYNC | FDSYNC)) && (count == 1))
219 atomic_dec_32(&zp->z_sync_cnt);
221 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
222 ZTOV(zp)->v_type == VREG &&
223 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
224 VERIFY(fs_vscan(vp, cr, 1) == 0);
231 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
232 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
235 zfs_holey(vnode_t *vp, u_long cmd, offset_t *off)
237 znode_t *zp = VTOZ(vp);
238 uint64_t noff = (uint64_t)*off; /* new offset */
243 file_sz = zp->z_size;
244 if (noff >= file_sz) {
248 if (cmd == _FIO_SEEK_HOLE)
253 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
256 if ((error == ESRCH) || (noff > file_sz)) {
258 * Handle the virtual hole at the end of file.
275 zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred,
276 int *rvalp, caller_context_t *ct)
288 * The following two ioctls are used by bfu. Faking out,
289 * necessary to avoid bfu errors.
297 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
301 zfsvfs = zp->z_zfsvfs;
305 /* offset parameter is in/out */
306 error = zfs_holey(vp, com, &off);
310 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
318 page_lookup(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes)
324 VM_OBJECT_LOCK_ASSERT(obj, MA_OWNED);
327 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
328 vm_page_is_valid(pp, (vm_offset_t)off, nbytes)) {
329 if (vm_page_sleep_if_busy(pp, FALSE, "zfsmwb"))
332 vm_page_lock_queues();
334 vm_page_unlock_queues();
336 if (__predict_false(obj->cache != NULL)) {
337 vm_page_cache_free(obj, OFF_TO_IDX(start),
338 OFF_TO_IDX(start) + 1);
348 page_unlock(vm_page_t pp)
355 zfs_map_page(vm_page_t pp, struct sf_buf **sfp)
358 *sfp = sf_buf_alloc(pp, 0);
359 return ((caddr_t)sf_buf_kva(*sfp));
363 zfs_unmap_page(struct sf_buf *sf)
370 * When a file is memory mapped, we must keep the IO data synchronized
371 * between the DMU cache and the memory mapped pages. What this means:
373 * On Write: If we find a memory mapped page, we write to *both*
374 * the page and the dmu buffer.
377 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid,
378 int segflg, dmu_tx_t *tx)
384 ASSERT(vp->v_mount != NULL);
388 off = start & PAGEOFFSET;
390 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
392 int nbytes = MIN(PAGESIZE - off, len);
394 if ((pp = page_lookup(vp, start, off, nbytes)) != NULL) {
397 VM_OBJECT_UNLOCK(obj);
398 va = zfs_map_page(pp, &sf);
399 if (segflg == UIO_NOCOPY) {
400 (void) dmu_write(os, oid, start+off, nbytes,
403 (void) dmu_read(os, oid, start+off, nbytes,
404 va+off, DMU_READ_PREFETCH);
413 VM_OBJECT_UNLOCK(obj);
417 * Read with UIO_NOCOPY flag means that sendfile(2) requests
418 * ZFS to populate a range of page cache pages with data.
420 * NOTE: this function could be optimized to pre-allocate
421 * all pages in advance, drain VPO_BUSY on all of them,
422 * map them into contiguous KVA region and populate them
423 * in one single dmu_read() call.
426 mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio)
428 znode_t *zp = VTOZ(vp);
429 objset_t *os = zp->z_zfsvfs->z_os;
439 ASSERT(uio->uio_segflg == UIO_NOCOPY);
440 ASSERT(vp->v_mount != NULL);
443 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0);
446 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) {
447 int bytes = MIN(PAGESIZE, len);
450 pp = vm_page_lookup(obj, OFF_TO_IDX(start));
451 if (pp != NULL && vm_page_sleep_if_busy(pp, FALSE,
455 pp = vm_page_alloc(obj, OFF_TO_IDX(start),
456 VM_ALLOC_NOBUSY | VM_ALLOC_NORMAL);
458 VM_OBJECT_UNLOCK(obj);
464 if (pp->valid == 0) {
465 vm_page_io_start(pp);
466 VM_OBJECT_UNLOCK(obj);
467 va = zfs_map_page(pp, &sf);
468 error = dmu_read(os, zp->z_id, start, bytes, va,
470 if (bytes != PAGESIZE && error == 0)
471 bzero(va + bytes, PAGESIZE - bytes);
474 vm_page_io_finish(pp);
475 vm_page_lock_queues();
479 pp->valid = VM_PAGE_BITS_ALL;
480 vm_page_activate(pp);
482 vm_page_unlock_queues();
486 uio->uio_resid -= bytes;
487 uio->uio_offset += bytes;
490 VM_OBJECT_UNLOCK(obj);
495 * When a file is memory mapped, we must keep the IO data synchronized
496 * between the DMU cache and the memory mapped pages. What this means:
498 * On Read: We "read" preferentially from memory mapped pages,
499 * else we default from the dmu buffer.
501 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
502 * the file is memory mapped.
505 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
507 znode_t *zp = VTOZ(vp);
508 objset_t *os = zp->z_zfsvfs->z_os;
516 ASSERT(vp->v_mount != NULL);
520 start = uio->uio_loffset;
521 off = start & PAGEOFFSET;
523 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
525 uint64_t bytes = MIN(PAGESIZE - off, len);
527 if (pp = page_lookup(vp, start, off, bytes)) {
530 VM_OBJECT_UNLOCK(obj);
531 va = zfs_map_page(pp, &sf);
532 error = uiomove(va + off, bytes, UIO_READ, uio);
537 VM_OBJECT_UNLOCK(obj);
538 error = dmu_read_uio(os, zp->z_id, uio, bytes);
546 VM_OBJECT_UNLOCK(obj);
550 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
553 * Read bytes from specified file into supplied buffer.
555 * IN: vp - vnode of file to be read from.
556 * uio - structure supplying read location, range info,
558 * ioflag - SYNC flags; used to provide FRSYNC semantics.
559 * cr - credentials of caller.
560 * ct - caller context
562 * OUT: uio - updated offset and range, buffer filled.
564 * RETURN: 0 if success
565 * error code if failure
568 * vp - atime updated if byte count > 0
572 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
574 znode_t *zp = VTOZ(vp);
575 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
586 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
592 * Validate file offset
594 if (uio->uio_loffset < (offset_t)0) {
600 * Fasttrack empty reads
602 if (uio->uio_resid == 0) {
608 * Check for mandatory locks
610 if (MANDMODE(zp->z_mode)) {
611 if (error = chklock(vp, FREAD,
612 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
619 * If we're in FRSYNC mode, sync out this znode before reading it.
622 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
623 zil_commit(zfsvfs->z_log, zp->z_id);
626 * Lock the range against changes.
628 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
631 * If we are reading past end-of-file we can skip
632 * to the end; but we might still need to set atime.
634 if (uio->uio_loffset >= zp->z_size) {
639 ASSERT(uio->uio_loffset < zp->z_size);
640 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
643 if ((uio->uio_extflg == UIO_XUIO) &&
644 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
646 int blksz = zp->z_blksz;
647 uint64_t offset = uio->uio_loffset;
649 xuio = (xuio_t *)uio;
651 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
654 ASSERT(offset + n <= blksz);
657 (void) dmu_xuio_init(xuio, nblk);
659 if (vn_has_cached_data(vp)) {
661 * For simplicity, we always allocate a full buffer
662 * even if we only expect to read a portion of a block.
664 while (--nblk >= 0) {
665 (void) dmu_xuio_add(xuio,
666 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
674 nbytes = MIN(n, zfs_read_chunk_size -
675 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
678 if (uio->uio_segflg == UIO_NOCOPY)
679 error = mappedread_sf(vp, nbytes, uio);
681 #endif /* __FreeBSD__ */
682 if (vn_has_cached_data(vp))
683 error = mappedread(vp, nbytes, uio);
685 error = dmu_read_uio(os, zp->z_id, uio, nbytes);
687 /* convert checksum errors into IO errors */
696 zfs_range_unlock(rl);
698 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
704 * Write the bytes to a file.
706 * IN: vp - vnode of file to be written to.
707 * uio - structure supplying write location, range info,
709 * ioflag - FAPPEND flag set if in append mode.
710 * cr - credentials of caller.
711 * ct - caller context (NFS/CIFS fem monitor only)
713 * OUT: uio - updated offset and range.
715 * RETURN: 0 if success
716 * error code if failure
719 * vp - ctime|mtime updated if byte count > 0
724 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
726 znode_t *zp = VTOZ(vp);
727 rlim64_t limit = MAXOFFSET_T;
728 ssize_t start_resid = uio->uio_resid;
732 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
737 int max_blksz = zfsvfs->z_max_blksz;
743 int iovcnt = uio->uio_iovcnt;
744 iovec_t *iovp = uio->uio_iov;
747 sa_bulk_attr_t bulk[4];
748 uint64_t mtime[2], ctime[2];
751 * Fasttrack empty write
757 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
763 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
764 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
765 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
767 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
771 * If immutable or not appending then return EPERM
773 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
774 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
775 (uio->uio_loffset < zp->z_size))) {
780 zilog = zfsvfs->z_log;
783 * Validate file offset
785 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
792 * Check for mandatory locks before calling zfs_range_lock()
793 * in order to prevent a deadlock with locks set via fcntl().
795 if (MANDMODE((mode_t)zp->z_mode) &&
796 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
803 * Pre-fault the pages to ensure slow (eg NFS) pages
805 * Skip this if uio contains loaned arc_buf.
807 if ((uio->uio_extflg == UIO_XUIO) &&
808 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
809 xuio = (xuio_t *)uio;
811 uio_prefaultpages(MIN(n, max_blksz), uio);
815 * If in append mode, set the io offset pointer to eof.
817 if (ioflag & FAPPEND) {
819 * Obtain an appending range lock to guarantee file append
820 * semantics. We reset the write offset once we have the lock.
822 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
824 if (rl->r_len == UINT64_MAX) {
826 * We overlocked the file because this write will cause
827 * the file block size to increase.
828 * Note that zp_size cannot change with this lock held.
832 uio->uio_loffset = woff;
835 * Note that if the file block size will change as a result of
836 * this write, then this range lock will lock the entire file
837 * so that we can re-write the block safely.
839 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
843 zfs_range_unlock(rl);
848 if ((woff + n) > limit || woff > (limit - n))
851 /* Will this write extend the file length? */
852 write_eof = (woff + n > zp->z_size);
854 end_size = MAX(zp->z_size, woff + n);
857 * Write the file in reasonable size chunks. Each chunk is written
858 * in a separate transaction; this keeps the intent log records small
859 * and allows us to do more fine-grained space accounting.
863 woff = uio->uio_loffset;
865 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
866 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
868 dmu_return_arcbuf(abuf);
873 if (xuio && abuf == NULL) {
874 ASSERT(i_iov < iovcnt);
876 abuf = dmu_xuio_arcbuf(xuio, i_iov);
877 dmu_xuio_clear(xuio, i_iov);
878 DTRACE_PROBE3(zfs_cp_write, int, i_iov,
879 iovec_t *, aiov, arc_buf_t *, abuf);
880 ASSERT((aiov->iov_base == abuf->b_data) ||
881 ((char *)aiov->iov_base - (char *)abuf->b_data +
882 aiov->iov_len == arc_buf_size(abuf)));
884 } else if (abuf == NULL && n >= max_blksz &&
885 woff >= zp->z_size &&
886 P2PHASE(woff, max_blksz) == 0 &&
887 zp->z_blksz == max_blksz) {
889 * This write covers a full block. "Borrow" a buffer
890 * from the dmu so that we can fill it before we enter
891 * a transaction. This avoids the possibility of
892 * holding up the transaction if the data copy hangs
893 * up on a pagefault (e.g., from an NFS server mapping).
897 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
899 ASSERT(abuf != NULL);
900 ASSERT(arc_buf_size(abuf) == max_blksz);
901 if (error = uiocopy(abuf->b_data, max_blksz,
902 UIO_WRITE, uio, &cbytes)) {
903 dmu_return_arcbuf(abuf);
906 ASSERT(cbytes == max_blksz);
910 * Start a transaction.
912 tx = dmu_tx_create(zfsvfs->z_os);
913 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
914 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
915 zfs_sa_upgrade_txholds(tx, zp);
916 error = dmu_tx_assign(tx, TXG_NOWAIT);
918 if (error == ERESTART) {
925 dmu_return_arcbuf(abuf);
930 * If zfs_range_lock() over-locked we grow the blocksize
931 * and then reduce the lock range. This will only happen
932 * on the first iteration since zfs_range_reduce() will
933 * shrink down r_len to the appropriate size.
935 if (rl->r_len == UINT64_MAX) {
938 if (zp->z_blksz > max_blksz) {
939 ASSERT(!ISP2(zp->z_blksz));
940 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
942 new_blksz = MIN(end_size, max_blksz);
944 zfs_grow_blocksize(zp, new_blksz, tx);
945 zfs_range_reduce(rl, woff, n);
949 * XXX - should we really limit each write to z_max_blksz?
950 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
952 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
954 if (woff + nbytes > zp->z_size)
955 vnode_pager_setsize(vp, woff + nbytes);
958 tx_bytes = uio->uio_resid;
959 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
961 tx_bytes -= uio->uio_resid;
964 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
966 * If this is not a full block write, but we are
967 * extending the file past EOF and this data starts
968 * block-aligned, use assign_arcbuf(). Otherwise,
969 * write via dmu_write().
971 if (tx_bytes < max_blksz && (!write_eof ||
972 aiov->iov_base != abuf->b_data)) {
974 dmu_write(zfsvfs->z_os, zp->z_id, woff,
975 aiov->iov_len, aiov->iov_base, tx);
976 dmu_return_arcbuf(abuf);
977 xuio_stat_wbuf_copied();
979 ASSERT(xuio || tx_bytes == max_blksz);
980 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
983 ASSERT(tx_bytes <= uio->uio_resid);
984 uioskip(uio, tx_bytes);
986 if (tx_bytes && vn_has_cached_data(vp)) {
987 update_pages(vp, woff, tx_bytes, zfsvfs->z_os,
988 zp->z_id, uio->uio_segflg, tx);
992 * If we made no progress, we're done. If we made even
993 * partial progress, update the znode and ZIL accordingly.
996 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
997 (void *)&zp->z_size, sizeof (uint64_t), tx);
1004 * Clear Set-UID/Set-GID bits on successful write if not
1005 * privileged and at least one of the excute bits is set.
1007 * It would be nice to to this after all writes have
1008 * been done, but that would still expose the ISUID/ISGID
1009 * to another app after the partial write is committed.
1011 * Note: we don't call zfs_fuid_map_id() here because
1012 * user 0 is not an ephemeral uid.
1014 mutex_enter(&zp->z_acl_lock);
1015 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
1016 (S_IXUSR >> 6))) != 0 &&
1017 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
1018 secpolicy_vnode_setid_retain(vp, cr,
1019 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
1021 zp->z_mode &= ~(S_ISUID | S_ISGID);
1022 newmode = zp->z_mode;
1023 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
1024 (void *)&newmode, sizeof (uint64_t), tx);
1026 mutex_exit(&zp->z_acl_lock);
1028 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
1032 * Update the file size (zp_size) if it has changed;
1033 * account for possible concurrent updates.
1035 while ((end_size = zp->z_size) < uio->uio_loffset) {
1036 (void) atomic_cas_64(&zp->z_size, end_size,
1041 * If we are replaying and eof is non zero then force
1042 * the file size to the specified eof. Note, there's no
1043 * concurrency during replay.
1045 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
1046 zp->z_size = zfsvfs->z_replay_eof;
1048 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1050 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
1055 ASSERT(tx_bytes == nbytes);
1060 uio_prefaultpages(MIN(n, max_blksz), uio);
1064 zfs_range_unlock(rl);
1067 * If we're in replay mode, or we made no progress, return error.
1068 * Otherwise, it's at least a partial write, so it's successful.
1070 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1075 if (ioflag & (FSYNC | FDSYNC) ||
1076 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1077 zil_commit(zilog, zp->z_id);
1084 zfs_get_done(zgd_t *zgd, int error)
1086 znode_t *zp = zgd->zgd_private;
1087 objset_t *os = zp->z_zfsvfs->z_os;
1091 dmu_buf_rele(zgd->zgd_db, zgd);
1093 zfs_range_unlock(zgd->zgd_rl);
1095 vfslocked = VFS_LOCK_GIANT(zp->z_zfsvfs->z_vfs);
1097 * Release the vnode asynchronously as we currently have the
1098 * txg stopped from syncing.
1100 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1102 if (error == 0 && zgd->zgd_bp)
1103 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1105 kmem_free(zgd, sizeof (zgd_t));
1106 VFS_UNLOCK_GIANT(vfslocked);
1110 static int zil_fault_io = 0;
1114 * Get data to generate a TX_WRITE intent log record.
1117 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1119 zfsvfs_t *zfsvfs = arg;
1120 objset_t *os = zfsvfs->z_os;
1122 uint64_t object = lr->lr_foid;
1123 uint64_t offset = lr->lr_offset;
1124 uint64_t size = lr->lr_length;
1125 blkptr_t *bp = &lr->lr_blkptr;
1130 ASSERT(zio != NULL);
1134 * Nothing to do if the file has been removed
1136 if (zfs_zget(zfsvfs, object, &zp) != 0)
1138 if (zp->z_unlinked) {
1140 * Release the vnode asynchronously as we currently have the
1141 * txg stopped from syncing.
1143 VN_RELE_ASYNC(ZTOV(zp),
1144 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1148 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1149 zgd->zgd_zilog = zfsvfs->z_log;
1150 zgd->zgd_private = zp;
1153 * Write records come in two flavors: immediate and indirect.
1154 * For small writes it's cheaper to store the data with the
1155 * log record (immediate); for large writes it's cheaper to
1156 * sync the data and get a pointer to it (indirect) so that
1157 * we don't have to write the data twice.
1159 if (buf != NULL) { /* immediate write */
1160 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1161 /* test for truncation needs to be done while range locked */
1162 if (offset >= zp->z_size) {
1165 error = dmu_read(os, object, offset, size, buf,
1166 DMU_READ_NO_PREFETCH);
1168 ASSERT(error == 0 || error == ENOENT);
1169 } else { /* indirect write */
1171 * Have to lock the whole block to ensure when it's
1172 * written out and it's checksum is being calculated
1173 * that no one can change the data. We need to re-check
1174 * blocksize after we get the lock in case it's changed!
1179 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1181 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1183 if (zp->z_blksz == size)
1186 zfs_range_unlock(zgd->zgd_rl);
1188 /* test for truncation needs to be done while range locked */
1189 if (lr->lr_offset >= zp->z_size)
1198 error = dmu_buf_hold(os, object, offset, zgd, &db,
1199 DMU_READ_NO_PREFETCH);
1205 ASSERT(db->db_offset == offset);
1206 ASSERT(db->db_size == size);
1208 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1210 ASSERT(error || lr->lr_length <= zp->z_blksz);
1213 * On success, we need to wait for the write I/O
1214 * initiated by dmu_sync() to complete before we can
1215 * release this dbuf. We will finish everything up
1216 * in the zfs_get_done() callback.
1221 if (error == EALREADY) {
1222 lr->lr_common.lrc_txtype = TX_WRITE2;
1228 zfs_get_done(zgd, error);
1235 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1236 caller_context_t *ct)
1238 znode_t *zp = VTOZ(vp);
1239 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1245 if (flag & V_ACE_MASK)
1246 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1248 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1255 * If vnode is for a device return a specfs vnode instead.
1258 specvp_check(vnode_t **vpp, cred_t *cr)
1262 if (IS_DEVVP(*vpp)) {
1265 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1276 * Lookup an entry in a directory, or an extended attribute directory.
1277 * If it exists, return a held vnode reference for it.
1279 * IN: dvp - vnode of directory to search.
1280 * nm - name of entry to lookup.
1281 * pnp - full pathname to lookup [UNUSED].
1282 * flags - LOOKUP_XATTR set if looking for an attribute.
1283 * rdir - root directory vnode [UNUSED].
1284 * cr - credentials of caller.
1285 * ct - caller context
1286 * direntflags - directory lookup flags
1287 * realpnp - returned pathname.
1289 * OUT: vpp - vnode of located entry, NULL if not found.
1291 * RETURN: 0 if success
1292 * error code if failure
1299 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp,
1300 int nameiop, cred_t *cr, kthread_t *td, int flags)
1302 znode_t *zdp = VTOZ(dvp);
1303 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1305 int *direntflags = NULL;
1306 void *realpnp = NULL;
1309 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1311 if (dvp->v_type != VDIR) {
1313 } else if (zdp->z_sa_hdl == NULL) {
1317 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1318 error = zfs_fastaccesschk_execute(zdp, cr);
1326 vnode_t *tvp = dnlc_lookup(dvp, nm);
1329 error = zfs_fastaccesschk_execute(zdp, cr);
1334 if (tvp == DNLC_NO_VNODE) {
1339 return (specvp_check(vpp, cr));
1345 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1352 if (flags & LOOKUP_XATTR) {
1355 * If the xattr property is off, refuse the lookup request.
1357 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1364 * We don't allow recursive attributes..
1365 * Maybe someday we will.
1367 if (zdp->z_pflags & ZFS_XATTR) {
1372 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1378 * Do we have permission to get into attribute directory?
1381 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1391 if (dvp->v_type != VDIR) {
1397 * Check accessibility of directory.
1400 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1405 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1406 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1411 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1413 error = specvp_check(vpp, cr);
1415 /* Translate errors and add SAVENAME when needed. */
1416 if (cnp->cn_flags & ISLASTCN) {
1420 if (error == ENOENT) {
1421 error = EJUSTRETURN;
1422 cnp->cn_flags |= SAVENAME;
1428 cnp->cn_flags |= SAVENAME;
1432 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) {
1435 if (cnp->cn_flags & ISDOTDOT) {
1436 ltype = VOP_ISLOCKED(dvp);
1440 error = zfs_vnode_lock(*vpp, cnp->cn_lkflags);
1441 if (cnp->cn_flags & ISDOTDOT)
1442 vn_lock(dvp, ltype | LK_RETRY);
1452 #ifdef FREEBSD_NAMECACHE
1454 * Insert name into cache (as non-existent) if appropriate.
1456 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
1457 cache_enter(dvp, *vpp, cnp);
1459 * Insert name into cache if appropriate.
1461 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) {
1462 if (!(cnp->cn_flags & ISLASTCN) ||
1463 (nameiop != DELETE && nameiop != RENAME)) {
1464 cache_enter(dvp, *vpp, cnp);
1473 * Attempt to create a new entry in a directory. If the entry
1474 * already exists, truncate the file if permissible, else return
1475 * an error. Return the vp of the created or trunc'd file.
1477 * IN: dvp - vnode of directory to put new file entry in.
1478 * name - name of new file entry.
1479 * vap - attributes of new file.
1480 * excl - flag indicating exclusive or non-exclusive mode.
1481 * mode - mode to open file with.
1482 * cr - credentials of caller.
1483 * flag - large file flag [UNUSED].
1484 * ct - caller context
1485 * vsecp - ACL to be set
1487 * OUT: vpp - vnode of created or trunc'd entry.
1489 * RETURN: 0 if success
1490 * error code if failure
1493 * dvp - ctime|mtime updated if new entry created
1494 * vp - ctime|mtime always, atime if new
1499 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode,
1500 vnode_t **vpp, cred_t *cr, kthread_t *td)
1502 znode_t *zp, *dzp = VTOZ(dvp);
1503 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1511 gid_t gid = crgetgid(cr);
1512 zfs_acl_ids_t acl_ids;
1513 boolean_t fuid_dirtied;
1514 boolean_t have_acl = B_FALSE;
1519 * If we have an ephemeral id, ACL, or XVATTR then
1520 * make sure file system is at proper version
1523 ksid = crgetsid(cr, KSID_OWNER);
1525 uid = ksid_getid(ksid);
1529 if (zfsvfs->z_use_fuids == B_FALSE &&
1530 (vsecp || (vap->va_mask & AT_XVATTR) ||
1531 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1537 zilog = zfsvfs->z_log;
1539 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1540 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1545 if (vap->va_mask & AT_XVATTR) {
1546 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
1547 crgetuid(cr), cr, vap->va_type)) != 0) {
1555 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr))
1556 vap->va_mode &= ~S_ISVTX;
1558 if (*name == '\0') {
1560 * Null component name refers to the directory itself.
1567 /* possible VN_HOLD(zp) */
1570 if (flag & FIGNORECASE)
1573 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1577 zfs_acl_ids_free(&acl_ids);
1578 if (strcmp(name, "..") == 0)
1589 * Create a new file object and update the directory
1592 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1594 zfs_acl_ids_free(&acl_ids);
1599 * We only support the creation of regular files in
1600 * extended attribute directories.
1603 if ((dzp->z_pflags & ZFS_XATTR) &&
1604 (vap->va_type != VREG)) {
1606 zfs_acl_ids_free(&acl_ids);
1611 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1612 cr, vsecp, &acl_ids)) != 0)
1616 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1617 zfs_acl_ids_free(&acl_ids);
1622 tx = dmu_tx_create(os);
1624 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1625 ZFS_SA_BASE_ATTR_SIZE);
1627 fuid_dirtied = zfsvfs->z_fuid_dirty;
1629 zfs_fuid_txhold(zfsvfs, tx);
1630 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1631 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1632 if (!zfsvfs->z_use_sa &&
1633 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1634 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1635 0, acl_ids.z_aclp->z_acl_bytes);
1637 error = dmu_tx_assign(tx, TXG_NOWAIT);
1639 zfs_dirent_unlock(dl);
1640 if (error == ERESTART) {
1645 zfs_acl_ids_free(&acl_ids);
1650 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1653 zfs_fuid_sync(zfsvfs, tx);
1655 (void) zfs_link_create(dl, zp, tx, ZNEW);
1656 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1657 if (flag & FIGNORECASE)
1659 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1660 vsecp, acl_ids.z_fuidp, vap);
1661 zfs_acl_ids_free(&acl_ids);
1664 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1667 zfs_acl_ids_free(&acl_ids);
1671 * A directory entry already exists for this name.
1674 * Can't truncate an existing file if in exclusive mode.
1681 * Can't open a directory for writing.
1683 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1688 * Verify requested access to file.
1690 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1694 mutex_enter(&dzp->z_lock);
1696 mutex_exit(&dzp->z_lock);
1699 * Truncate regular files if requested.
1701 if ((ZTOV(zp)->v_type == VREG) &&
1702 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1703 /* we can't hold any locks when calling zfs_freesp() */
1704 zfs_dirent_unlock(dl);
1706 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1708 vnevent_create(ZTOV(zp), ct);
1714 zfs_dirent_unlock(dl);
1721 error = specvp_check(vpp, cr);
1724 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1725 zil_commit(zilog, 0);
1732 * Remove an entry from a directory.
1734 * IN: dvp - vnode of directory to remove entry from.
1735 * name - name of entry to remove.
1736 * cr - credentials of caller.
1737 * ct - caller context
1738 * flags - case flags
1740 * RETURN: 0 if success
1741 * error code if failure
1745 * vp - ctime (if nlink > 0)
1748 uint64_t null_xattr = 0;
1752 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1755 znode_t *zp, *dzp = VTOZ(dvp);
1758 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1760 uint64_t acl_obj, xattr_obj;
1761 uint64_t xattr_obj_unlinked = 0;
1765 boolean_t may_delete_now, delete_now = FALSE;
1766 boolean_t unlinked, toobig = FALSE;
1768 pathname_t *realnmp = NULL;
1775 zilog = zfsvfs->z_log;
1777 if (flags & FIGNORECASE) {
1787 * Attempt to lock directory; fail if entry doesn't exist.
1789 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1799 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1804 * Need to use rmdir for removing directories.
1806 if (vp->v_type == VDIR) {
1811 vnevent_remove(vp, dvp, name, ct);
1814 dnlc_remove(dvp, realnmp->pn_buf);
1816 dnlc_remove(dvp, name);
1819 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1823 * We may delete the znode now, or we may put it in the unlinked set;
1824 * it depends on whether we're the last link, and on whether there are
1825 * other holds on the vnode. So we dmu_tx_hold() the right things to
1826 * allow for either case.
1829 tx = dmu_tx_create(zfsvfs->z_os);
1830 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1831 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1832 zfs_sa_upgrade_txholds(tx, zp);
1833 zfs_sa_upgrade_txholds(tx, dzp);
1834 if (may_delete_now) {
1836 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1837 /* if the file is too big, only hold_free a token amount */
1838 dmu_tx_hold_free(tx, zp->z_id, 0,
1839 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1842 /* are there any extended attributes? */
1843 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1844 &xattr_obj, sizeof (xattr_obj));
1845 if (error == 0 && xattr_obj) {
1846 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1847 ASSERT3U(error, ==, 0);
1848 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1849 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1852 mutex_enter(&zp->z_lock);
1853 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1854 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1855 mutex_exit(&zp->z_lock);
1857 /* charge as an update -- would be nice not to charge at all */
1858 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1860 error = dmu_tx_assign(tx, TXG_NOWAIT);
1862 zfs_dirent_unlock(dl);
1866 if (error == ERESTART) {
1879 * Remove the directory entry.
1881 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1891 * Hold z_lock so that we can make sure that the ACL obj
1892 * hasn't changed. Could have been deleted due to
1895 mutex_enter(&zp->z_lock);
1897 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1898 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1899 delete_now = may_delete_now && !toobig &&
1900 vp->v_count == 1 && !vn_has_cached_data(vp) &&
1901 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1907 if (xattr_obj_unlinked) {
1908 ASSERT3U(xzp->z_links, ==, 2);
1909 mutex_enter(&xzp->z_lock);
1910 xzp->z_unlinked = 1;
1912 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1913 &xzp->z_links, sizeof (xzp->z_links), tx);
1914 ASSERT3U(error, ==, 0);
1915 mutex_exit(&xzp->z_lock);
1916 zfs_unlinked_add(xzp, tx);
1919 error = sa_remove(zp->z_sa_hdl,
1920 SA_ZPL_XATTR(zfsvfs), tx);
1922 error = sa_update(zp->z_sa_hdl,
1923 SA_ZPL_XATTR(zfsvfs), &null_xattr,
1924 sizeof (uint64_t), tx);
1925 ASSERT3U(error, ==, 0);
1929 ASSERT3U(vp->v_count, ==, 0);
1931 mutex_exit(&zp->z_lock);
1932 zfs_znode_delete(zp, tx);
1933 } else if (unlinked) {
1934 mutex_exit(&zp->z_lock);
1935 zfs_unlinked_add(zp, tx);
1939 if (flags & FIGNORECASE)
1941 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1948 zfs_dirent_unlock(dl);
1955 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1956 zil_commit(zilog, 0);
1963 * Create a new directory and insert it into dvp using the name
1964 * provided. Return a pointer to the inserted directory.
1966 * IN: dvp - vnode of directory to add subdir to.
1967 * dirname - name of new directory.
1968 * vap - attributes of new directory.
1969 * cr - credentials of caller.
1970 * ct - caller context
1971 * vsecp - ACL to be set
1973 * OUT: vpp - vnode of created directory.
1975 * RETURN: 0 if success
1976 * error code if failure
1979 * dvp - ctime|mtime updated
1980 * vp - ctime|mtime|atime updated
1984 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
1985 caller_context_t *ct, int flags, vsecattr_t *vsecp)
1987 znode_t *zp, *dzp = VTOZ(dvp);
1988 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1997 gid_t gid = crgetgid(cr);
1998 zfs_acl_ids_t acl_ids;
1999 boolean_t fuid_dirtied;
2001 ASSERT(vap->va_type == VDIR);
2004 * If we have an ephemeral id, ACL, or XVATTR then
2005 * make sure file system is at proper version
2008 ksid = crgetsid(cr, KSID_OWNER);
2010 uid = ksid_getid(ksid);
2013 if (zfsvfs->z_use_fuids == B_FALSE &&
2014 (vsecp || (vap->va_mask & AT_XVATTR) ||
2015 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
2020 zilog = zfsvfs->z_log;
2022 if (dzp->z_pflags & ZFS_XATTR) {
2027 if (zfsvfs->z_utf8 && u8_validate(dirname,
2028 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2032 if (flags & FIGNORECASE)
2035 if (vap->va_mask & AT_XVATTR) {
2036 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
2037 crgetuid(cr), cr, vap->va_type)) != 0) {
2043 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
2044 vsecp, &acl_ids)) != 0) {
2049 * First make sure the new directory doesn't exist.
2051 * Existence is checked first to make sure we don't return
2052 * EACCES instead of EEXIST which can cause some applications
2058 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
2060 zfs_acl_ids_free(&acl_ids);
2065 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
2066 zfs_acl_ids_free(&acl_ids);
2067 zfs_dirent_unlock(dl);
2072 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
2073 zfs_acl_ids_free(&acl_ids);
2074 zfs_dirent_unlock(dl);
2080 * Add a new entry to the directory.
2082 tx = dmu_tx_create(zfsvfs->z_os);
2083 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2084 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2085 fuid_dirtied = zfsvfs->z_fuid_dirty;
2087 zfs_fuid_txhold(zfsvfs, tx);
2088 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2089 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2090 acl_ids.z_aclp->z_acl_bytes);
2093 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2094 ZFS_SA_BASE_ATTR_SIZE);
2096 error = dmu_tx_assign(tx, TXG_NOWAIT);
2098 zfs_dirent_unlock(dl);
2099 if (error == ERESTART) {
2104 zfs_acl_ids_free(&acl_ids);
2113 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2116 zfs_fuid_sync(zfsvfs, tx);
2119 * Now put new name in parent dir.
2121 (void) zfs_link_create(dl, zp, tx, ZNEW);
2125 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2126 if (flags & FIGNORECASE)
2128 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2129 acl_ids.z_fuidp, vap);
2131 zfs_acl_ids_free(&acl_ids);
2135 zfs_dirent_unlock(dl);
2137 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2138 zil_commit(zilog, 0);
2145 * Remove a directory subdir entry. If the current working
2146 * directory is the same as the subdir to be removed, the
2149 * IN: dvp - vnode of directory to remove from.
2150 * name - name of directory to be removed.
2151 * cwd - vnode of current working directory.
2152 * cr - credentials of caller.
2153 * ct - caller context
2154 * flags - case flags
2156 * RETURN: 0 if success
2157 * error code if failure
2160 * dvp - ctime|mtime updated
2164 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2165 caller_context_t *ct, int flags)
2167 znode_t *dzp = VTOZ(dvp);
2170 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2179 zilog = zfsvfs->z_log;
2181 if (flags & FIGNORECASE)
2187 * Attempt to lock directory; fail if entry doesn't exist.
2189 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2197 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2201 if (vp->v_type != VDIR) {
2211 vnevent_rmdir(vp, dvp, name, ct);
2214 * Grab a lock on the directory to make sure that noone is
2215 * trying to add (or lookup) entries while we are removing it.
2217 rw_enter(&zp->z_name_lock, RW_WRITER);
2220 * Grab a lock on the parent pointer to make sure we play well
2221 * with the treewalk and directory rename code.
2223 rw_enter(&zp->z_parent_lock, RW_WRITER);
2225 tx = dmu_tx_create(zfsvfs->z_os);
2226 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2227 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2228 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2229 zfs_sa_upgrade_txholds(tx, zp);
2230 zfs_sa_upgrade_txholds(tx, dzp);
2231 error = dmu_tx_assign(tx, TXG_NOWAIT);
2233 rw_exit(&zp->z_parent_lock);
2234 rw_exit(&zp->z_name_lock);
2235 zfs_dirent_unlock(dl);
2237 if (error == ERESTART) {
2247 #ifdef FREEBSD_NAMECACHE
2251 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2254 uint64_t txtype = TX_RMDIR;
2255 if (flags & FIGNORECASE)
2257 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2262 rw_exit(&zp->z_parent_lock);
2263 rw_exit(&zp->z_name_lock);
2264 #ifdef FREEBSD_NAMECACHE
2268 zfs_dirent_unlock(dl);
2272 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2273 zil_commit(zilog, 0);
2280 * Read as many directory entries as will fit into the provided
2281 * buffer from the given directory cursor position (specified in
2282 * the uio structure.
2284 * IN: vp - vnode of directory to read.
2285 * uio - structure supplying read location, range info,
2286 * and return buffer.
2287 * cr - credentials of caller.
2288 * ct - caller context
2289 * flags - case flags
2291 * OUT: uio - updated offset and range, buffer filled.
2292 * eofp - set to true if end-of-file detected.
2294 * RETURN: 0 if success
2295 * error code if failure
2298 * vp - atime updated
2300 * Note that the low 4 bits of the cookie returned by zap is always zero.
2301 * This allows us to use the low range for "special" directory entries:
2302 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2303 * we use the offset 2 for the '.zfs' directory.
2307 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies)
2309 znode_t *zp = VTOZ(vp);
2313 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2318 zap_attribute_t zap;
2319 uint_t bytes_wanted;
2320 uint64_t offset; /* must be unsigned; checks for < 1 */
2326 boolean_t check_sysattrs;
2329 u_long *cooks = NULL;
2335 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2336 &parent, sizeof (parent))) != 0) {
2342 * If we are not given an eof variable,
2349 * Check for valid iov_len.
2351 if (uio->uio_iov->iov_len <= 0) {
2357 * Quit if directory has been removed (posix)
2359 if ((*eofp = zp->z_unlinked) != 0) {
2366 offset = uio->uio_loffset;
2367 prefetch = zp->z_zn_prefetch;
2370 * Initialize the iterator cursor.
2374 * Start iteration from the beginning of the directory.
2376 zap_cursor_init(&zc, os, zp->z_id);
2379 * The offset is a serialized cursor.
2381 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2385 * Get space to change directory entries into fs independent format.
2387 iovp = uio->uio_iov;
2388 bytes_wanted = iovp->iov_len;
2389 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2390 bufsize = bytes_wanted;
2391 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2392 odp = (struct dirent64 *)outbuf;
2394 bufsize = bytes_wanted;
2395 odp = (struct dirent64 *)iovp->iov_base;
2397 eodp = (struct edirent *)odp;
2399 if (ncookies != NULL) {
2401 * Minimum entry size is dirent size and 1 byte for a file name.
2403 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1);
2404 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK);
2409 * If this VFS supports the system attribute view interface; and
2410 * we're looking at an extended attribute directory; and we care
2411 * about normalization conflicts on this vfs; then we must check
2412 * for normalization conflicts with the sysattr name space.
2415 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2416 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2417 (flags & V_RDDIR_ENTFLAGS);
2423 * Transform to file-system independent format
2426 while (outcount < bytes_wanted) {
2429 off64_t *next = NULL;
2432 * Special case `.', `..', and `.zfs'.
2435 (void) strcpy(zap.za_name, ".");
2436 zap.za_normalization_conflict = 0;
2439 } else if (offset == 1) {
2440 (void) strcpy(zap.za_name, "..");
2441 zap.za_normalization_conflict = 0;
2444 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2445 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2446 zap.za_normalization_conflict = 0;
2447 objnum = ZFSCTL_INO_ROOT;
2453 if (error = zap_cursor_retrieve(&zc, &zap)) {
2454 if ((*eofp = (error == ENOENT)) != 0)
2460 if (zap.za_integer_length != 8 ||
2461 zap.za_num_integers != 1) {
2462 cmn_err(CE_WARN, "zap_readdir: bad directory "
2463 "entry, obj = %lld, offset = %lld\n",
2464 (u_longlong_t)zp->z_id,
2465 (u_longlong_t)offset);
2470 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2472 * MacOS X can extract the object type here such as:
2473 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2475 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2477 if (check_sysattrs && !zap.za_normalization_conflict) {
2479 zap.za_normalization_conflict =
2480 xattr_sysattr_casechk(zap.za_name);
2482 panic("%s:%u: TODO", __func__, __LINE__);
2487 if (flags & V_RDDIR_ACCFILTER) {
2489 * If we have no access at all, don't include
2490 * this entry in the returned information
2493 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2495 if (!zfs_has_access(ezp, cr)) {
2502 if (flags & V_RDDIR_ENTFLAGS)
2503 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2505 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2508 * Will this entry fit in the buffer?
2510 if (outcount + reclen > bufsize) {
2512 * Did we manage to fit anything in the buffer?
2520 if (flags & V_RDDIR_ENTFLAGS) {
2522 * Add extended flag entry:
2524 eodp->ed_ino = objnum;
2525 eodp->ed_reclen = reclen;
2526 /* NOTE: ed_off is the offset for the *next* entry */
2527 next = &(eodp->ed_off);
2528 eodp->ed_eflags = zap.za_normalization_conflict ?
2529 ED_CASE_CONFLICT : 0;
2530 (void) strncpy(eodp->ed_name, zap.za_name,
2531 EDIRENT_NAMELEN(reclen));
2532 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2537 odp->d_ino = objnum;
2538 odp->d_reclen = reclen;
2539 odp->d_namlen = strlen(zap.za_name);
2540 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1);
2542 odp = (dirent64_t *)((intptr_t)odp + reclen);
2546 ASSERT(outcount <= bufsize);
2548 /* Prefetch znode */
2550 dmu_prefetch(os, objnum, 0, 0);
2554 * Move to the next entry, fill in the previous offset.
2556 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2557 zap_cursor_advance(&zc);
2558 offset = zap_cursor_serialize(&zc);
2563 if (cooks != NULL) {
2566 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks));
2569 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2571 /* Subtract unused cookies */
2572 if (ncookies != NULL)
2573 *ncookies -= ncooks;
2575 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2576 iovp->iov_base += outcount;
2577 iovp->iov_len -= outcount;
2578 uio->uio_resid -= outcount;
2579 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2581 * Reset the pointer.
2583 offset = uio->uio_loffset;
2587 zap_cursor_fini(&zc);
2588 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2589 kmem_free(outbuf, bufsize);
2591 if (error == ENOENT)
2594 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2596 uio->uio_loffset = offset;
2598 if (error != 0 && cookies != NULL) {
2599 free(*cookies, M_TEMP);
2606 ulong_t zfs_fsync_sync_cnt = 4;
2609 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2611 znode_t *zp = VTOZ(vp);
2612 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2614 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2616 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2619 zil_commit(zfsvfs->z_log, zp->z_id);
2627 * Get the requested file attributes and place them in the provided
2630 * IN: vp - vnode of file.
2631 * vap - va_mask identifies requested attributes.
2632 * If AT_XVATTR set, then optional attrs are requested
2633 * flags - ATTR_NOACLCHECK (CIFS server context)
2634 * cr - credentials of caller.
2635 * ct - caller context
2637 * OUT: vap - attribute values.
2639 * RETURN: 0 (always succeeds)
2643 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2644 caller_context_t *ct)
2646 znode_t *zp = VTOZ(vp);
2647 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2650 u_longlong_t nblocks;
2652 uint64_t mtime[2], ctime[2], crtime[2], rdev;
2653 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2654 xoptattr_t *xoap = NULL;
2655 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2656 sa_bulk_attr_t bulk[4];
2662 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2664 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2665 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2666 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &crtime, 16);
2667 if (vp->v_type == VBLK || vp->v_type == VCHR)
2668 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
2671 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2677 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2678 * Also, if we are the owner don't bother, since owner should
2679 * always be allowed to read basic attributes of file.
2681 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2682 (vap->va_uid != crgetuid(cr))) {
2683 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2691 * Return all attributes. It's cheaper to provide the answer
2692 * than to determine whether we were asked the question.
2695 mutex_enter(&zp->z_lock);
2696 vap->va_type = IFTOVT(zp->z_mode);
2697 vap->va_mode = zp->z_mode & ~S_IFMT;
2699 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2701 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
2703 vap->va_nodeid = zp->z_id;
2704 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2705 links = zp->z_links + 1;
2707 links = zp->z_links;
2708 vap->va_nlink = MIN(links, UINT32_MAX); /* nlink_t limit! */
2709 vap->va_size = zp->z_size;
2711 vap->va_rdev = vp->v_rdev;
2713 if (vp->v_type == VBLK || vp->v_type == VCHR)
2714 vap->va_rdev = zfs_cmpldev(rdev);
2716 vap->va_seq = zp->z_seq;
2717 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */
2720 * Add in any requested optional attributes and the create time.
2721 * Also set the corresponding bits in the returned attribute bitmap.
2723 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2724 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2726 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2727 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2730 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2731 xoap->xoa_readonly =
2732 ((zp->z_pflags & ZFS_READONLY) != 0);
2733 XVA_SET_RTN(xvap, XAT_READONLY);
2736 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2738 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2739 XVA_SET_RTN(xvap, XAT_SYSTEM);
2742 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2744 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2745 XVA_SET_RTN(xvap, XAT_HIDDEN);
2748 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2749 xoap->xoa_nounlink =
2750 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2751 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2754 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2755 xoap->xoa_immutable =
2756 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2757 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2760 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2761 xoap->xoa_appendonly =
2762 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2763 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2766 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2768 ((zp->z_pflags & ZFS_NODUMP) != 0);
2769 XVA_SET_RTN(xvap, XAT_NODUMP);
2772 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2774 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2775 XVA_SET_RTN(xvap, XAT_OPAQUE);
2778 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2779 xoap->xoa_av_quarantined =
2780 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2781 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2784 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2785 xoap->xoa_av_modified =
2786 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2787 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2790 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2791 vp->v_type == VREG) {
2792 zfs_sa_get_scanstamp(zp, xvap);
2795 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2798 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2799 times, sizeof (times));
2800 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2801 XVA_SET_RTN(xvap, XAT_CREATETIME);
2804 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2805 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2806 XVA_SET_RTN(xvap, XAT_REPARSE);
2808 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2809 xoap->xoa_generation = zp->z_gen;
2810 XVA_SET_RTN(xvap, XAT_GEN);
2813 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2815 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2816 XVA_SET_RTN(xvap, XAT_OFFLINE);
2819 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2821 ((zp->z_pflags & ZFS_SPARSE) != 0);
2822 XVA_SET_RTN(xvap, XAT_SPARSE);
2826 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2827 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2828 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2829 ZFS_TIME_DECODE(&vap->va_birthtime, crtime);
2831 mutex_exit(&zp->z_lock);
2833 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2834 vap->va_blksize = blksize;
2835 vap->va_bytes = nblocks << 9; /* nblocks * 512 */
2837 if (zp->z_blksz == 0) {
2839 * Block size hasn't been set; suggest maximal I/O transfers.
2841 vap->va_blksize = zfsvfs->z_max_blksz;
2849 * Set the file attributes to the values contained in the
2852 * IN: vp - vnode of file to be modified.
2853 * vap - new attribute values.
2854 * If AT_XVATTR set, then optional attrs are being set
2855 * flags - ATTR_UTIME set if non-default time values provided.
2856 * - ATTR_NOACLCHECK (CIFS context only).
2857 * cr - credentials of caller.
2858 * ct - caller context
2860 * RETURN: 0 if success
2861 * error code if failure
2864 * vp - ctime updated, mtime updated if size changed.
2868 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2869 caller_context_t *ct)
2871 znode_t *zp = VTOZ(vp);
2872 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2877 uint_t mask = vap->va_mask;
2879 uint64_t saved_mode;
2882 uint64_t new_uid, new_gid;
2884 uint64_t mtime[2], ctime[2];
2886 int need_policy = FALSE;
2888 zfs_fuid_info_t *fuidp = NULL;
2889 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2892 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2893 boolean_t fuid_dirtied = B_FALSE;
2894 sa_bulk_attr_t bulk[7], xattr_bulk[7];
2895 int count = 0, xattr_count = 0;
2900 if (mask & AT_NOSET)
2906 zilog = zfsvfs->z_log;
2909 * Make sure that if we have ephemeral uid/gid or xvattr specified
2910 * that file system is at proper version level
2913 if (zfsvfs->z_use_fuids == B_FALSE &&
2914 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2915 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2916 (mask & AT_XVATTR))) {
2921 if (mask & AT_SIZE && vp->v_type == VDIR) {
2926 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2932 * If this is an xvattr_t, then get a pointer to the structure of
2933 * optional attributes. If this is NULL, then we have a vattr_t.
2935 xoap = xva_getxoptattr(xvap);
2937 xva_init(&tmpxvattr);
2940 * Immutable files can only alter immutable bit and atime
2942 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2943 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2944 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2949 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2955 * Verify timestamps doesn't overflow 32 bits.
2956 * ZFS can handle large timestamps, but 32bit syscalls can't
2957 * handle times greater than 2039. This check should be removed
2958 * once large timestamps are fully supported.
2960 if (mask & (AT_ATIME | AT_MTIME)) {
2961 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2962 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2972 /* Can this be moved to before the top label? */
2973 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2979 * First validate permissions
2982 if (mask & AT_SIZE) {
2984 * XXX - Note, we are not providing any open
2985 * mode flags here (like FNDELAY), so we may
2986 * block if there are locks present... this
2987 * should be addressed in openat().
2989 /* XXX - would it be OK to generate a log record here? */
2990 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2997 if (mask & (AT_ATIME|AT_MTIME) ||
2998 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2999 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
3000 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
3001 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
3002 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
3003 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
3004 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
3005 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
3009 if (mask & (AT_UID|AT_GID)) {
3010 int idmask = (mask & (AT_UID|AT_GID));
3015 * NOTE: even if a new mode is being set,
3016 * we may clear S_ISUID/S_ISGID bits.
3019 if (!(mask & AT_MODE))
3020 vap->va_mode = zp->z_mode;
3023 * Take ownership or chgrp to group we are a member of
3026 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
3027 take_group = (mask & AT_GID) &&
3028 zfs_groupmember(zfsvfs, vap->va_gid, cr);
3031 * If both AT_UID and AT_GID are set then take_owner and
3032 * take_group must both be set in order to allow taking
3035 * Otherwise, send the check through secpolicy_vnode_setattr()
3039 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
3040 ((idmask == AT_UID) && take_owner) ||
3041 ((idmask == AT_GID) && take_group)) {
3042 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
3043 skipaclchk, cr) == 0) {
3045 * Remove setuid/setgid for non-privileged users
3047 secpolicy_setid_clear(vap, vp, cr);
3048 trim_mask = (mask & (AT_UID|AT_GID));
3057 mutex_enter(&zp->z_lock);
3058 oldva.va_mode = zp->z_mode;
3059 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
3060 if (mask & AT_XVATTR) {
3062 * Update xvattr mask to include only those attributes
3063 * that are actually changing.
3065 * the bits will be restored prior to actually setting
3066 * the attributes so the caller thinks they were set.
3068 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
3069 if (xoap->xoa_appendonly !=
3070 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
3073 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
3074 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
3078 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
3079 if (xoap->xoa_nounlink !=
3080 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
3083 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
3084 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
3088 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
3089 if (xoap->xoa_immutable !=
3090 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
3093 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
3094 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
3098 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
3099 if (xoap->xoa_nodump !=
3100 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
3103 XVA_CLR_REQ(xvap, XAT_NODUMP);
3104 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
3108 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3109 if (xoap->xoa_av_modified !=
3110 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3113 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3114 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
3118 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3119 if ((vp->v_type != VREG &&
3120 xoap->xoa_av_quarantined) ||
3121 xoap->xoa_av_quarantined !=
3122 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3125 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3126 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
3130 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3131 mutex_exit(&zp->z_lock);
3136 if (need_policy == FALSE &&
3137 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3138 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3143 mutex_exit(&zp->z_lock);
3145 if (mask & AT_MODE) {
3146 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3147 err = secpolicy_setid_setsticky_clear(vp, vap,
3153 trim_mask |= AT_MODE;
3161 * If trim_mask is set then take ownership
3162 * has been granted or write_acl is present and user
3163 * has the ability to modify mode. In that case remove
3164 * UID|GID and or MODE from mask so that
3165 * secpolicy_vnode_setattr() doesn't revoke it.
3169 saved_mask = vap->va_mask;
3170 vap->va_mask &= ~trim_mask;
3171 if (trim_mask & AT_MODE) {
3173 * Save the mode, as secpolicy_vnode_setattr()
3174 * will overwrite it with ova.va_mode.
3176 saved_mode = vap->va_mode;
3179 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3180 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3187 vap->va_mask |= saved_mask;
3188 if (trim_mask & AT_MODE) {
3190 * Recover the mode after
3191 * secpolicy_vnode_setattr().
3193 vap->va_mode = saved_mode;
3199 * secpolicy_vnode_setattr, or take ownership may have
3202 mask = vap->va_mask;
3204 if ((mask & (AT_UID | AT_GID))) {
3205 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3206 &xattr_obj, sizeof (xattr_obj));
3208 if (err == 0 && xattr_obj) {
3209 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3213 if (mask & AT_UID) {
3214 new_uid = zfs_fuid_create(zfsvfs,
3215 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3216 if (new_uid != zp->z_uid &&
3217 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3219 VN_RELE(ZTOV(attrzp));
3225 if (mask & AT_GID) {
3226 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3227 cr, ZFS_GROUP, &fuidp);
3228 if (new_gid != zp->z_gid &&
3229 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3231 VN_RELE(ZTOV(attrzp));
3237 tx = dmu_tx_create(zfsvfs->z_os);
3239 if (mask & AT_MODE) {
3240 uint64_t pmode = zp->z_mode;
3242 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3244 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3247 mutex_enter(&zp->z_lock);
3248 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3250 * Are we upgrading ACL from old V0 format
3253 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3254 zfs_znode_acl_version(zp) ==
3255 ZFS_ACL_VERSION_INITIAL) {
3256 dmu_tx_hold_free(tx, acl_obj, 0,
3258 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3259 0, aclp->z_acl_bytes);
3261 dmu_tx_hold_write(tx, acl_obj, 0,
3264 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3265 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3266 0, aclp->z_acl_bytes);
3268 mutex_exit(&zp->z_lock);
3269 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3271 if ((mask & AT_XVATTR) &&
3272 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3273 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3275 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3279 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3282 fuid_dirtied = zfsvfs->z_fuid_dirty;
3284 zfs_fuid_txhold(zfsvfs, tx);
3286 zfs_sa_upgrade_txholds(tx, zp);
3288 err = dmu_tx_assign(tx, TXG_NOWAIT);
3290 if (err == ERESTART)
3297 * Set each attribute requested.
3298 * We group settings according to the locks they need to acquire.
3300 * Note: you cannot set ctime directly, although it will be
3301 * updated as a side-effect of calling this function.
3305 if (mask & (AT_UID|AT_GID|AT_MODE))
3306 mutex_enter(&zp->z_acl_lock);
3307 mutex_enter(&zp->z_lock);
3309 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3310 &zp->z_pflags, sizeof (zp->z_pflags));
3313 if (mask & (AT_UID|AT_GID|AT_MODE))
3314 mutex_enter(&attrzp->z_acl_lock);
3315 mutex_enter(&attrzp->z_lock);
3316 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3317 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3318 sizeof (attrzp->z_pflags));
3321 if (mask & (AT_UID|AT_GID)) {
3323 if (mask & AT_UID) {
3324 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3325 &new_uid, sizeof (new_uid));
3326 zp->z_uid = new_uid;
3328 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3329 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3331 attrzp->z_uid = new_uid;
3335 if (mask & AT_GID) {
3336 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3337 NULL, &new_gid, sizeof (new_gid));
3338 zp->z_gid = new_gid;
3340 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3341 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3343 attrzp->z_gid = new_gid;
3346 if (!(mask & AT_MODE)) {
3347 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3348 NULL, &new_mode, sizeof (new_mode));
3349 new_mode = zp->z_mode;
3351 err = zfs_acl_chown_setattr(zp);
3354 err = zfs_acl_chown_setattr(attrzp);
3359 if (mask & AT_MODE) {
3360 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3361 &new_mode, sizeof (new_mode));
3362 zp->z_mode = new_mode;
3363 ASSERT3U((uintptr_t)aclp, !=, 0);
3364 err = zfs_aclset_common(zp, aclp, cr, tx);
3365 ASSERT3U(err, ==, 0);
3366 if (zp->z_acl_cached)
3367 zfs_acl_free(zp->z_acl_cached);
3368 zp->z_acl_cached = aclp;
3373 if (mask & AT_ATIME) {
3374 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3375 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3376 &zp->z_atime, sizeof (zp->z_atime));
3379 if (mask & AT_MTIME) {
3380 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3381 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3382 mtime, sizeof (mtime));
3385 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3386 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3387 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3388 NULL, mtime, sizeof (mtime));
3389 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3390 &ctime, sizeof (ctime));
3391 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3393 } else if (mask != 0) {
3394 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3395 &ctime, sizeof (ctime));
3396 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3399 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3400 SA_ZPL_CTIME(zfsvfs), NULL,
3401 &ctime, sizeof (ctime));
3402 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3403 mtime, ctime, B_TRUE);
3407 * Do this after setting timestamps to prevent timestamp
3408 * update from toggling bit
3411 if (xoap && (mask & AT_XVATTR)) {
3414 * restore trimmed off masks
3415 * so that return masks can be set for caller.
3418 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3419 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3421 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3422 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3424 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3425 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3427 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3428 XVA_SET_REQ(xvap, XAT_NODUMP);
3430 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3431 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3433 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3434 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3437 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3438 ASSERT(vp->v_type == VREG);
3440 zfs_xvattr_set(zp, xvap, tx);
3444 zfs_fuid_sync(zfsvfs, tx);
3447 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3449 mutex_exit(&zp->z_lock);
3450 if (mask & (AT_UID|AT_GID|AT_MODE))
3451 mutex_exit(&zp->z_acl_lock);
3454 if (mask & (AT_UID|AT_GID|AT_MODE))
3455 mutex_exit(&attrzp->z_acl_lock);
3456 mutex_exit(&attrzp->z_lock);
3459 if (err == 0 && attrzp) {
3460 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3466 VN_RELE(ZTOV(attrzp));
3471 zfs_fuid_info_free(fuidp);
3477 if (err == ERESTART)
3480 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3485 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3486 zil_commit(zilog, 0);
3492 typedef struct zfs_zlock {
3493 krwlock_t *zl_rwlock; /* lock we acquired */
3494 znode_t *zl_znode; /* znode we held */
3495 struct zfs_zlock *zl_next; /* next in list */
3499 * Drop locks and release vnodes that were held by zfs_rename_lock().
3502 zfs_rename_unlock(zfs_zlock_t **zlpp)
3506 while ((zl = *zlpp) != NULL) {
3507 if (zl->zl_znode != NULL)
3508 VN_RELE(ZTOV(zl->zl_znode));
3509 rw_exit(zl->zl_rwlock);
3510 *zlpp = zl->zl_next;
3511 kmem_free(zl, sizeof (*zl));
3516 * Search back through the directory tree, using the ".." entries.
3517 * Lock each directory in the chain to prevent concurrent renames.
3518 * Fail any attempt to move a directory into one of its own descendants.
3519 * XXX - z_parent_lock can overlap with map or grow locks
3522 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3526 uint64_t rootid = zp->z_zfsvfs->z_root;
3527 uint64_t oidp = zp->z_id;
3528 krwlock_t *rwlp = &szp->z_parent_lock;
3529 krw_t rw = RW_WRITER;
3532 * First pass write-locks szp and compares to zp->z_id.
3533 * Later passes read-lock zp and compare to zp->z_parent.
3536 if (!rw_tryenter(rwlp, rw)) {
3538 * Another thread is renaming in this path.
3539 * Note that if we are a WRITER, we don't have any
3540 * parent_locks held yet.
3542 if (rw == RW_READER && zp->z_id > szp->z_id) {
3544 * Drop our locks and restart
3546 zfs_rename_unlock(&zl);
3550 rwlp = &szp->z_parent_lock;
3555 * Wait for other thread to drop its locks
3561 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3562 zl->zl_rwlock = rwlp;
3563 zl->zl_znode = NULL;
3564 zl->zl_next = *zlpp;
3567 if (oidp == szp->z_id) /* We're a descendant of szp */
3570 if (oidp == rootid) /* We've hit the top */
3573 if (rw == RW_READER) { /* i.e. not the first pass */
3574 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3579 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3580 &oidp, sizeof (oidp));
3581 rwlp = &zp->z_parent_lock;
3584 } while (zp->z_id != sdzp->z_id);
3590 * Move an entry from the provided source directory to the target
3591 * directory. Change the entry name as indicated.
3593 * IN: sdvp - Source directory containing the "old entry".
3594 * snm - Old entry name.
3595 * tdvp - Target directory to contain the "new entry".
3596 * tnm - New entry name.
3597 * cr - credentials of caller.
3598 * ct - caller context
3599 * flags - case flags
3601 * RETURN: 0 if success
3602 * error code if failure
3605 * sdvp,tdvp - ctime|mtime updated
3609 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3610 caller_context_t *ct, int flags)
3612 znode_t *tdzp, *szp, *tzp;
3613 znode_t *sdzp = VTOZ(sdvp);
3614 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs;
3617 zfs_dirlock_t *sdl, *tdl;
3620 int cmp, serr, terr;
3625 ZFS_VERIFY_ZP(sdzp);
3626 zilog = zfsvfs->z_log;
3629 * Make sure we have the real vp for the target directory.
3631 if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3634 if (tdvp->v_vfsp != sdvp->v_vfsp || zfsctl_is_node(tdvp)) {
3640 ZFS_VERIFY_ZP(tdzp);
3641 if (zfsvfs->z_utf8 && u8_validate(tnm,
3642 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3647 if (flags & FIGNORECASE)
3656 * This is to prevent the creation of links into attribute space
3657 * by renaming a linked file into/outof an attribute directory.
3658 * See the comment in zfs_link() for why this is considered bad.
3660 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3666 * Lock source and target directory entries. To prevent deadlock,
3667 * a lock ordering must be defined. We lock the directory with
3668 * the smallest object id first, or if it's a tie, the one with
3669 * the lexically first name.
3671 if (sdzp->z_id < tdzp->z_id) {
3673 } else if (sdzp->z_id > tdzp->z_id) {
3677 * First compare the two name arguments without
3678 * considering any case folding.
3680 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3682 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3683 ASSERT(error == 0 || !zfsvfs->z_utf8);
3686 * POSIX: "If the old argument and the new argument
3687 * both refer to links to the same existing file,
3688 * the rename() function shall return successfully
3689 * and perform no other action."
3695 * If the file system is case-folding, then we may
3696 * have some more checking to do. A case-folding file
3697 * system is either supporting mixed case sensitivity
3698 * access or is completely case-insensitive. Note
3699 * that the file system is always case preserving.
3701 * In mixed sensitivity mode case sensitive behavior
3702 * is the default. FIGNORECASE must be used to
3703 * explicitly request case insensitive behavior.
3705 * If the source and target names provided differ only
3706 * by case (e.g., a request to rename 'tim' to 'Tim'),
3707 * we will treat this as a special case in the
3708 * case-insensitive mode: as long as the source name
3709 * is an exact match, we will allow this to proceed as
3710 * a name-change request.
3712 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3713 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3714 flags & FIGNORECASE)) &&
3715 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3718 * case preserving rename request, require exact
3727 * If the source and destination directories are the same, we should
3728 * grab the z_name_lock of that directory only once.
3732 rw_enter(&sdzp->z_name_lock, RW_READER);
3736 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3737 ZEXISTS | zflg, NULL, NULL);
3738 terr = zfs_dirent_lock(&tdl,
3739 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3741 terr = zfs_dirent_lock(&tdl,
3742 tdzp, tnm, &tzp, zflg, NULL, NULL);
3743 serr = zfs_dirent_lock(&sdl,
3744 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3750 * Source entry invalid or not there.
3753 zfs_dirent_unlock(tdl);
3759 rw_exit(&sdzp->z_name_lock);
3762 * FreeBSD: In OpenSolaris they only check if rename source is
3763 * ".." here, because "." is handled in their lookup. This is
3764 * not the case for FreeBSD, so we check for "." explicitly.
3766 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0)
3772 zfs_dirent_unlock(sdl);
3776 rw_exit(&sdzp->z_name_lock);
3778 if (strcmp(tnm, "..") == 0)
3785 * Must have write access at the source to remove the old entry
3786 * and write access at the target to create the new entry.
3787 * Note that if target and source are the same, this can be
3788 * done in a single check.
3791 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3794 if (ZTOV(szp)->v_type == VDIR) {
3796 * Check to make sure rename is valid.
3797 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3799 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3804 * Does target exist?
3808 * Source and target must be the same type.
3810 if (ZTOV(szp)->v_type == VDIR) {
3811 if (ZTOV(tzp)->v_type != VDIR) {
3816 if (ZTOV(tzp)->v_type == VDIR) {
3822 * POSIX dictates that when the source and target
3823 * entries refer to the same file object, rename
3824 * must do nothing and exit without error.
3826 if (szp->z_id == tzp->z_id) {
3832 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3834 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3837 * notify the target directory if it is not the same
3838 * as source directory.
3841 vnevent_rename_dest_dir(tdvp, ct);
3844 tx = dmu_tx_create(zfsvfs->z_os);
3845 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3846 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3847 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3848 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3850 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3851 zfs_sa_upgrade_txholds(tx, tdzp);
3854 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3855 zfs_sa_upgrade_txholds(tx, tzp);
3858 zfs_sa_upgrade_txholds(tx, szp);
3859 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3860 error = dmu_tx_assign(tx, TXG_NOWAIT);
3863 zfs_rename_unlock(&zl);
3864 zfs_dirent_unlock(sdl);
3865 zfs_dirent_unlock(tdl);
3868 rw_exit(&sdzp->z_name_lock);
3873 if (error == ERESTART) {
3883 if (tzp) /* Attempt to remove the existing target */
3884 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3887 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3889 szp->z_pflags |= ZFS_AV_MODIFIED;
3891 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3892 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3893 ASSERT3U(error, ==, 0);
3895 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3897 zfs_log_rename(zilog, tx, TX_RENAME |
3898 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3899 sdl->dl_name, tdzp, tdl->dl_name, szp);
3902 * Update path information for the target vnode
3904 vn_renamepath(tdvp, ZTOV(szp), tnm,
3908 * At this point, we have successfully created
3909 * the target name, but have failed to remove
3910 * the source name. Since the create was done
3911 * with the ZRENAMING flag, there are
3912 * complications; for one, the link count is
3913 * wrong. The easiest way to deal with this
3914 * is to remove the newly created target, and
3915 * return the original error. This must
3916 * succeed; fortunately, it is very unlikely to
3917 * fail, since we just created it.
3919 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3920 ZRENAMING, NULL), ==, 0);
3923 #ifdef FREEBSD_NAMECACHE
3934 zfs_rename_unlock(&zl);
3936 zfs_dirent_unlock(sdl);
3937 zfs_dirent_unlock(tdl);
3940 rw_exit(&sdzp->z_name_lock);
3947 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3948 zil_commit(zilog, 0);
3956 * Insert the indicated symbolic reference entry into the directory.
3958 * IN: dvp - Directory to contain new symbolic link.
3959 * link - Name for new symlink entry.
3960 * vap - Attributes of new entry.
3961 * target - Target path of new symlink.
3962 * cr - credentials of caller.
3963 * ct - caller context
3964 * flags - case flags
3966 * RETURN: 0 if success
3967 * error code if failure
3970 * dvp - ctime|mtime updated
3974 zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link,
3975 cred_t *cr, kthread_t *td)
3977 znode_t *zp, *dzp = VTOZ(dvp);
3980 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
3982 uint64_t len = strlen(link);
3985 zfs_acl_ids_t acl_ids;
3986 boolean_t fuid_dirtied;
3987 uint64_t txtype = TX_SYMLINK;
3990 ASSERT(vap->va_type == VLNK);
3994 zilog = zfsvfs->z_log;
3996 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3997 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4001 if (flags & FIGNORECASE)
4004 if (len > MAXPATHLEN) {
4006 return (ENAMETOOLONG);
4009 if ((error = zfs_acl_ids_create(dzp, 0,
4010 vap, cr, NULL, &acl_ids)) != 0) {
4016 * Attempt to lock directory; fail if entry already exists.
4018 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
4020 zfs_acl_ids_free(&acl_ids);
4025 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4026 zfs_acl_ids_free(&acl_ids);
4027 zfs_dirent_unlock(dl);
4032 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
4033 zfs_acl_ids_free(&acl_ids);
4034 zfs_dirent_unlock(dl);
4038 tx = dmu_tx_create(zfsvfs->z_os);
4039 fuid_dirtied = zfsvfs->z_fuid_dirty;
4040 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
4041 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4042 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
4043 ZFS_SA_BASE_ATTR_SIZE + len);
4044 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
4045 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
4046 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
4047 acl_ids.z_aclp->z_acl_bytes);
4050 zfs_fuid_txhold(zfsvfs, tx);
4051 error = dmu_tx_assign(tx, TXG_NOWAIT);
4053 zfs_dirent_unlock(dl);
4054 if (error == ERESTART) {
4059 zfs_acl_ids_free(&acl_ids);
4066 * Create a new object for the symlink.
4067 * for version 4 ZPL datsets the symlink will be an SA attribute
4069 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
4072 zfs_fuid_sync(zfsvfs, tx);
4074 mutex_enter(&zp->z_lock);
4076 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
4079 zfs_sa_symlink(zp, link, len, tx);
4080 mutex_exit(&zp->z_lock);
4083 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
4084 &zp->z_size, sizeof (zp->z_size), tx);
4086 * Insert the new object into the directory.
4088 (void) zfs_link_create(dl, zp, tx, ZNEW);
4090 if (flags & FIGNORECASE)
4092 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
4095 zfs_acl_ids_free(&acl_ids);
4099 zfs_dirent_unlock(dl);
4101 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4102 zil_commit(zilog, 0);
4109 * Return, in the buffer contained in the provided uio structure,
4110 * the symbolic path referred to by vp.
4112 * IN: vp - vnode of symbolic link.
4113 * uoip - structure to contain the link path.
4114 * cr - credentials of caller.
4115 * ct - caller context
4117 * OUT: uio - structure to contain the link path.
4119 * RETURN: 0 if success
4120 * error code if failure
4123 * vp - atime updated
4127 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
4129 znode_t *zp = VTOZ(vp);
4130 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4136 mutex_enter(&zp->z_lock);
4138 error = sa_lookup_uio(zp->z_sa_hdl,
4139 SA_ZPL_SYMLINK(zfsvfs), uio);
4141 error = zfs_sa_readlink(zp, uio);
4142 mutex_exit(&zp->z_lock);
4144 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4151 * Insert a new entry into directory tdvp referencing svp.
4153 * IN: tdvp - Directory to contain new entry.
4154 * svp - vnode of new entry.
4155 * name - name of new entry.
4156 * cr - credentials of caller.
4157 * ct - caller context
4159 * RETURN: 0 if success
4160 * error code if failure
4163 * tdvp - ctime|mtime updated
4164 * svp - ctime updated
4168 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
4169 caller_context_t *ct, int flags)
4171 znode_t *dzp = VTOZ(tdvp);
4173 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4183 ASSERT(tdvp->v_type == VDIR);
4187 zilog = zfsvfs->z_log;
4189 if (VOP_REALVP(svp, &realvp, ct) == 0)
4193 * POSIX dictates that we return EPERM here.
4194 * Better choices include ENOTSUP or EISDIR.
4196 if (svp->v_type == VDIR) {
4201 if (svp->v_vfsp != tdvp->v_vfsp || zfsctl_is_node(svp)) {
4209 /* Prevent links to .zfs/shares files */
4211 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4212 &parent, sizeof (uint64_t))) != 0) {
4216 if (parent == zfsvfs->z_shares_dir) {
4221 if (zfsvfs->z_utf8 && u8_validate(name,
4222 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4226 if (flags & FIGNORECASE)
4230 * We do not support links between attributes and non-attributes
4231 * because of the potential security risk of creating links
4232 * into "normal" file space in order to circumvent restrictions
4233 * imposed in attribute space.
4235 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4241 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4242 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) {
4247 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4254 * Attempt to lock directory; fail if entry already exists.
4256 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4262 tx = dmu_tx_create(zfsvfs->z_os);
4263 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4264 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4265 zfs_sa_upgrade_txholds(tx, szp);
4266 zfs_sa_upgrade_txholds(tx, dzp);
4267 error = dmu_tx_assign(tx, TXG_NOWAIT);
4269 zfs_dirent_unlock(dl);
4270 if (error == ERESTART) {
4280 error = zfs_link_create(dl, szp, tx, 0);
4283 uint64_t txtype = TX_LINK;
4284 if (flags & FIGNORECASE)
4286 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4291 zfs_dirent_unlock(dl);
4294 vnevent_link(svp, ct);
4297 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4298 zil_commit(zilog, 0);
4306 * zfs_null_putapage() is used when the file system has been force
4307 * unmounted. It just drops the pages.
4311 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4312 size_t *lenp, int flags, cred_t *cr)
4314 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4319 * Push a page out to disk, klustering if possible.
4321 * IN: vp - file to push page to.
4322 * pp - page to push.
4323 * flags - additional flags.
4324 * cr - credentials of caller.
4326 * OUT: offp - start of range pushed.
4327 * lenp - len of range pushed.
4329 * RETURN: 0 if success
4330 * error code if failure
4332 * NOTE: callers must have locked the page to be pushed. On
4333 * exit, the page (and all other pages in the kluster) must be
4338 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4339 size_t *lenp, int flags, cred_t *cr)
4341 znode_t *zp = VTOZ(vp);
4342 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4344 u_offset_t off, koff;
4351 * If our blocksize is bigger than the page size, try to kluster
4352 * multiple pages so that we write a full block (thus avoiding
4353 * a read-modify-write).
4355 if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4356 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4357 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4358 ASSERT(koff <= zp->z_size);
4359 if (koff + klen > zp->z_size)
4360 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4361 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4363 ASSERT3U(btop(len), ==, btopr(len));
4366 * Can't push pages past end-of-file.
4368 if (off >= zp->z_size) {
4369 /* ignore all pages */
4372 } else if (off + len > zp->z_size) {
4373 int npages = btopr(zp->z_size - off);
4376 page_list_break(&pp, &trunc, npages);
4377 /* ignore pages past end of file */
4379 pvn_write_done(trunc, flags);
4380 len = zp->z_size - off;
4383 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4384 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4389 tx = dmu_tx_create(zfsvfs->z_os);
4390 dmu_tx_hold_write(tx, zp->z_id, off, len);
4392 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4393 zfs_sa_upgrade_txholds(tx, zp);
4394 err = dmu_tx_assign(tx, TXG_NOWAIT);
4396 if (err == ERESTART) {
4405 if (zp->z_blksz <= PAGESIZE) {
4406 caddr_t va = zfs_map_page(pp, S_READ);
4407 ASSERT3U(len, <=, PAGESIZE);
4408 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4409 zfs_unmap_page(pp, va);
4411 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4415 uint64_t mtime[2], ctime[2];
4416 sa_bulk_attr_t bulk[3];
4419 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4421 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4423 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4425 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4427 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4432 pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4442 * Copy the portion of the file indicated from pages into the file.
4443 * The pages are stored in a page list attached to the files vnode.
4445 * IN: vp - vnode of file to push page data to.
4446 * off - position in file to put data.
4447 * len - amount of data to write.
4448 * flags - flags to control the operation.
4449 * cr - credentials of caller.
4450 * ct - caller context.
4452 * RETURN: 0 if success
4453 * error code if failure
4456 * vp - ctime|mtime updated
4460 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4461 caller_context_t *ct)
4463 znode_t *zp = VTOZ(vp);
4464 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4476 * Align this request to the file block size in case we kluster.
4477 * XXX - this can result in pretty aggresive locking, which can
4478 * impact simultanious read/write access. One option might be
4479 * to break up long requests (len == 0) into block-by-block
4480 * operations to get narrower locking.
4482 blksz = zp->z_blksz;
4484 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4487 if (len > 0 && ISP2(blksz))
4488 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4494 * Search the entire vp list for pages >= io_off.
4496 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4497 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4500 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4502 if (off > zp->z_size) {
4503 /* past end of file */
4504 zfs_range_unlock(rl);
4509 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4511 for (off = io_off; io_off < off + len; io_off += io_len) {
4512 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4513 pp = page_lookup(vp, io_off,
4514 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4516 pp = page_lookup_nowait(vp, io_off,
4517 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4520 if (pp != NULL && pvn_getdirty(pp, flags)) {
4524 * Found a dirty page to push
4526 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4534 zfs_range_unlock(rl);
4535 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4536 zil_commit(zfsvfs->z_log, zp->z_id);
4544 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4546 znode_t *zp = VTOZ(vp);
4547 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4550 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4551 if (zp->z_sa_hdl == NULL) {
4553 * The fs has been unmounted, or we did a
4554 * suspend/resume and this file no longer exists.
4557 ASSERT(vp->v_count <= 1);
4560 vrecycle(vp, curthread);
4561 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4565 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4566 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4568 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4569 zfs_sa_upgrade_txholds(tx, zp);
4570 error = dmu_tx_assign(tx, TXG_WAIT);
4574 mutex_enter(&zp->z_lock);
4575 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4576 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4577 zp->z_atime_dirty = 0;
4578 mutex_exit(&zp->z_lock);
4584 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4589 * Bounds-check the seek operation.
4591 * IN: vp - vnode seeking within
4592 * ooff - old file offset
4593 * noffp - pointer to new file offset
4594 * ct - caller context
4596 * RETURN: 0 if success
4597 * EINVAL if new offset invalid
4601 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4602 caller_context_t *ct)
4604 if (vp->v_type == VDIR)
4606 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4610 * Pre-filter the generic locking function to trap attempts to place
4611 * a mandatory lock on a memory mapped file.
4614 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4615 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4617 znode_t *zp = VTOZ(vp);
4618 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4624 * We are following the UFS semantics with respect to mapcnt
4625 * here: If we see that the file is mapped already, then we will
4626 * return an error, but we don't worry about races between this
4627 * function and zfs_map().
4629 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4634 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4638 * If we can't find a page in the cache, we will create a new page
4639 * and fill it with file data. For efficiency, we may try to fill
4640 * multiple pages at once (klustering) to fill up the supplied page
4641 * list. Note that the pages to be filled are held with an exclusive
4642 * lock to prevent access by other threads while they are being filled.
4645 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4646 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4648 znode_t *zp = VTOZ(vp);
4649 page_t *pp, *cur_pp;
4650 objset_t *os = zp->z_zfsvfs->z_os;
4651 u_offset_t io_off, total;
4655 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4657 * We only have a single page, don't bother klustering
4661 pp = page_create_va(vp, io_off, io_len,
4662 PG_EXCL | PG_WAIT, seg, addr);
4665 * Try to find enough pages to fill the page list
4667 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4668 &io_len, off, plsz, 0);
4672 * The page already exists, nothing to do here.
4679 * Fill the pages in the kluster.
4682 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4685 ASSERT3U(io_off, ==, cur_pp->p_offset);
4686 va = zfs_map_page(cur_pp, S_WRITE);
4687 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4689 zfs_unmap_page(cur_pp, va);
4691 /* On error, toss the entire kluster */
4692 pvn_read_done(pp, B_ERROR);
4693 /* convert checksum errors into IO errors */
4698 cur_pp = cur_pp->p_next;
4702 * Fill in the page list array from the kluster starting
4703 * from the desired offset `off'.
4704 * NOTE: the page list will always be null terminated.
4706 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4707 ASSERT(pl == NULL || (*pl)->p_offset == off);
4713 * Return pointers to the pages for the file region [off, off + len]
4714 * in the pl array. If plsz is greater than len, this function may
4715 * also return page pointers from after the specified region
4716 * (i.e. the region [off, off + plsz]). These additional pages are
4717 * only returned if they are already in the cache, or were created as
4718 * part of a klustered read.
4720 * IN: vp - vnode of file to get data from.
4721 * off - position in file to get data from.
4722 * len - amount of data to retrieve.
4723 * plsz - length of provided page list.
4724 * seg - segment to obtain pages for.
4725 * addr - virtual address of fault.
4726 * rw - mode of created pages.
4727 * cr - credentials of caller.
4728 * ct - caller context.
4730 * OUT: protp - protection mode of created pages.
4731 * pl - list of pages created.
4733 * RETURN: 0 if success
4734 * error code if failure
4737 * vp - atime updated
4741 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4742 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4743 enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4745 znode_t *zp = VTOZ(vp);
4746 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4750 /* we do our own caching, faultahead is unnecessary */
4753 else if (len > plsz)
4756 len = P2ROUNDUP(len, PAGESIZE);
4757 ASSERT(plsz >= len);
4766 * Loop through the requested range [off, off + len) looking
4767 * for pages. If we don't find a page, we will need to create
4768 * a new page and fill it with data from the file.
4771 if (*pl = page_lookup(vp, off, SE_SHARED))
4773 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4776 ASSERT3U((*pl)->p_offset, ==, off);
4780 ASSERT3U(len, >=, PAGESIZE);
4783 ASSERT3U(plsz, >=, PAGESIZE);
4790 * Fill out the page array with any pages already in the cache.
4793 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4800 * Release any pages we have previously locked.
4805 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4815 * Request a memory map for a section of a file. This code interacts
4816 * with common code and the VM system as follows:
4818 * common code calls mmap(), which ends up in smmap_common()
4820 * this calls VOP_MAP(), which takes you into (say) zfs
4822 * zfs_map() calls as_map(), passing segvn_create() as the callback
4824 * segvn_create() creates the new segment and calls VOP_ADDMAP()
4826 * zfs_addmap() updates z_mapcnt
4830 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4831 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4832 caller_context_t *ct)
4834 znode_t *zp = VTOZ(vp);
4835 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4836 segvn_crargs_t vn_a;
4842 if ((prot & PROT_WRITE) && (zp->z_pflags &
4843 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4848 if ((prot & (PROT_READ | PROT_EXEC)) &&
4849 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4854 if (vp->v_flag & VNOMAP) {
4859 if (off < 0 || len > MAXOFFSET_T - off) {
4864 if (vp->v_type != VREG) {
4870 * If file is locked, disallow mapping.
4872 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
4878 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4886 vn_a.offset = (u_offset_t)off;
4887 vn_a.type = flags & MAP_TYPE;
4889 vn_a.maxprot = maxprot;
4892 vn_a.flags = flags & ~MAP_TYPE;
4894 vn_a.lgrp_mem_policy_flags = 0;
4896 error = as_map(as, *addrp, len, segvn_create, &vn_a);
4905 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4906 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4907 caller_context_t *ct)
4909 uint64_t pages = btopr(len);
4911 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
4916 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4917 * more accurate mtime for the associated file. Since we don't have a way of
4918 * detecting when the data was actually modified, we have to resort to
4919 * heuristics. If an explicit msync() is done, then we mark the mtime when the
4920 * last page is pushed. The problem occurs when the msync() call is omitted,
4921 * which by far the most common case:
4929 * putpage() via fsflush
4931 * If we wait until fsflush to come along, we can have a modification time that
4932 * is some arbitrary point in the future. In order to prevent this in the
4933 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
4938 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4939 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
4940 caller_context_t *ct)
4942 uint64_t pages = btopr(len);
4944 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
4945 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
4947 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
4948 vn_has_cached_data(vp))
4949 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
4955 * Free or allocate space in a file. Currently, this function only
4956 * supports the `F_FREESP' command. However, this command is somewhat
4957 * misnamed, as its functionality includes the ability to allocate as
4958 * well as free space.
4960 * IN: vp - vnode of file to free data in.
4961 * cmd - action to take (only F_FREESP supported).
4962 * bfp - section of file to free/alloc.
4963 * flag - current file open mode flags.
4964 * offset - current file offset.
4965 * cr - credentials of caller [UNUSED].
4966 * ct - caller context.
4968 * RETURN: 0 if success
4969 * error code if failure
4972 * vp - ctime|mtime updated
4976 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
4977 offset_t offset, cred_t *cr, caller_context_t *ct)
4979 znode_t *zp = VTOZ(vp);
4980 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4987 if (cmd != F_FREESP) {
4992 if (error = convoff(vp, bfp, 0, offset)) {
4997 if (bfp->l_len < 0) {
5003 len = bfp->l_len; /* 0 means from off to end of file */
5005 error = zfs_freesp(zp, off, len, flag, TRUE);
5012 CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid));
5013 CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid));
5017 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
5019 znode_t *zp = VTOZ(vp);
5020 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5023 uint64_t object = zp->z_id;
5030 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
5031 &gen64, sizeof (uint64_t))) != 0) {
5036 gen = (uint32_t)gen64;
5038 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
5039 fidp->fid_len = size;
5041 zfid = (zfid_short_t *)fidp;
5043 zfid->zf_len = size;
5045 for (i = 0; i < sizeof (zfid->zf_object); i++)
5046 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
5048 /* Must have a non-zero generation number to distinguish from .zfs */
5051 for (i = 0; i < sizeof (zfid->zf_gen); i++)
5052 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
5054 if (size == LONG_FID_LEN) {
5055 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
5058 zlfid = (zfid_long_t *)fidp;
5060 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
5061 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
5063 /* XXX - this should be the generation number for the objset */
5064 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
5065 zlfid->zf_setgen[i] = 0;
5073 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
5074 caller_context_t *ct)
5086 case _PC_FILESIZEBITS:
5090 case _PC_XATTR_EXISTS:
5092 zfsvfs = zp->z_zfsvfs;
5096 error = zfs_dirent_lock(&dl, zp, "", &xzp,
5097 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
5099 zfs_dirent_unlock(dl);
5100 if (!zfs_dirempty(xzp))
5103 } else if (error == ENOENT) {
5105 * If there aren't extended attributes, it's the
5106 * same as having zero of them.
5113 case _PC_SATTR_ENABLED:
5114 case _PC_SATTR_EXISTS:
5115 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
5116 (vp->v_type == VREG || vp->v_type == VDIR);
5119 case _PC_ACCESS_FILTERING:
5120 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
5124 case _PC_ACL_ENABLED:
5125 *valp = _ACL_ACE_ENABLED;
5128 case _PC_MIN_HOLE_SIZE:
5129 *valp = (int)SPA_MINBLOCKSIZE;
5132 case _PC_TIMESTAMP_RESOLUTION:
5133 /* nanosecond timestamp resolution */
5137 case _PC_ACL_EXTENDED:
5145 case _PC_ACL_PATH_MAX:
5146 *valp = ACL_MAX_ENTRIES;
5150 return (EOPNOTSUPP);
5156 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5157 caller_context_t *ct)
5159 znode_t *zp = VTOZ(vp);
5160 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5162 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5166 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5174 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5175 caller_context_t *ct)
5177 znode_t *zp = VTOZ(vp);
5178 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5180 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5181 zilog_t *zilog = zfsvfs->z_log;
5186 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5188 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5189 zil_commit(zilog, 0);
5197 * Tunable, both must be a power of 2.
5199 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
5200 * zcr_blksz_max: if set to less than the file block size, allow loaning out of
5201 * an arcbuf for a partial block read
5203 int zcr_blksz_min = (1 << 10); /* 1K */
5204 int zcr_blksz_max = (1 << 17); /* 128K */
5208 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5209 caller_context_t *ct)
5211 znode_t *zp = VTOZ(vp);
5212 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5213 int max_blksz = zfsvfs->z_max_blksz;
5214 uio_t *uio = &xuio->xu_uio;
5215 ssize_t size = uio->uio_resid;
5216 offset_t offset = uio->uio_loffset;
5221 int preamble, postamble;
5223 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5231 * Loan out an arc_buf for write if write size is bigger than
5232 * max_blksz, and the file's block size is also max_blksz.
5235 if (size < blksz || zp->z_blksz != blksz) {
5240 * Caller requests buffers for write before knowing where the
5241 * write offset might be (e.g. NFS TCP write).
5246 preamble = P2PHASE(offset, blksz);
5248 preamble = blksz - preamble;
5253 postamble = P2PHASE(size, blksz);
5256 fullblk = size / blksz;
5257 (void) dmu_xuio_init(xuio,
5258 (preamble != 0) + fullblk + (postamble != 0));
5259 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5260 int, postamble, int,
5261 (preamble != 0) + fullblk + (postamble != 0));
5264 * Have to fix iov base/len for partial buffers. They
5265 * currently represent full arc_buf's.
5268 /* data begins in the middle of the arc_buf */
5269 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5272 (void) dmu_xuio_add(xuio, abuf,
5273 blksz - preamble, preamble);
5276 for (i = 0; i < fullblk; i++) {
5277 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5280 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5284 /* data ends in the middle of the arc_buf */
5285 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5288 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5293 * Loan out an arc_buf for read if the read size is larger than
5294 * the current file block size. Block alignment is not
5295 * considered. Partial arc_buf will be loaned out for read.
5297 blksz = zp->z_blksz;
5298 if (blksz < zcr_blksz_min)
5299 blksz = zcr_blksz_min;
5300 if (blksz > zcr_blksz_max)
5301 blksz = zcr_blksz_max;
5302 /* avoid potential complexity of dealing with it */
5303 if (blksz > max_blksz) {
5308 maxsize = zp->z_size - uio->uio_loffset;
5312 if (size < blksz || vn_has_cached_data(vp)) {
5322 uio->uio_extflg = UIO_XUIO;
5323 XUIO_XUZC_RW(xuio) = ioflag;
5330 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5334 int ioflag = XUIO_XUZC_RW(xuio);
5336 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5338 i = dmu_xuio_cnt(xuio);
5340 abuf = dmu_xuio_arcbuf(xuio, i);
5342 * if abuf == NULL, it must be a write buffer
5343 * that has been returned in zfs_write().
5346 dmu_return_arcbuf(abuf);
5347 ASSERT(abuf || ioflag == UIO_WRITE);
5350 dmu_xuio_fini(xuio);
5355 * Predeclare these here so that the compiler assumes that
5356 * this is an "old style" function declaration that does
5357 * not include arguments => we won't get type mismatch errors
5358 * in the initializations that follow.
5360 static int zfs_inval();
5361 static int zfs_isdir();
5375 * Directory vnode operations template
5377 vnodeops_t *zfs_dvnodeops;
5378 const fs_operation_def_t zfs_dvnodeops_template[] = {
5379 VOPNAME_OPEN, { .vop_open = zfs_open },
5380 VOPNAME_CLOSE, { .vop_close = zfs_close },
5381 VOPNAME_READ, { .error = zfs_isdir },
5382 VOPNAME_WRITE, { .error = zfs_isdir },
5383 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5384 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5385 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5386 VOPNAME_ACCESS, { .vop_access = zfs_access },
5387 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5388 VOPNAME_CREATE, { .vop_create = zfs_create },
5389 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5390 VOPNAME_LINK, { .vop_link = zfs_link },
5391 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5392 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir },
5393 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5394 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5395 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink },
5396 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5397 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5398 VOPNAME_FID, { .vop_fid = zfs_fid },
5399 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5400 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5401 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5402 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5403 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5408 * Regular file vnode operations template
5410 vnodeops_t *zfs_fvnodeops;
5411 const fs_operation_def_t zfs_fvnodeops_template[] = {
5412 VOPNAME_OPEN, { .vop_open = zfs_open },
5413 VOPNAME_CLOSE, { .vop_close = zfs_close },
5414 VOPNAME_READ, { .vop_read = zfs_read },
5415 VOPNAME_WRITE, { .vop_write = zfs_write },
5416 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5417 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5418 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5419 VOPNAME_ACCESS, { .vop_access = zfs_access },
5420 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5421 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5422 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5423 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5424 VOPNAME_FID, { .vop_fid = zfs_fid },
5425 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5426 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock },
5427 VOPNAME_SPACE, { .vop_space = zfs_space },
5428 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage },
5429 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage },
5430 VOPNAME_MAP, { .vop_map = zfs_map },
5431 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap },
5432 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap },
5433 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5434 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5435 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5436 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5437 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf },
5438 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf },
5443 * Symbolic link vnode operations template
5445 vnodeops_t *zfs_symvnodeops;
5446 const fs_operation_def_t zfs_symvnodeops_template[] = {
5447 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5448 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5449 VOPNAME_ACCESS, { .vop_access = zfs_access },
5450 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5451 VOPNAME_READLINK, { .vop_readlink = zfs_readlink },
5452 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5453 VOPNAME_FID, { .vop_fid = zfs_fid },
5454 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5455 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5460 * special share hidden files vnode operations template
5462 vnodeops_t *zfs_sharevnodeops;
5463 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5464 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5465 VOPNAME_ACCESS, { .vop_access = zfs_access },
5466 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5467 VOPNAME_FID, { .vop_fid = zfs_fid },
5468 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5469 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5470 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5471 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5476 * Extended attribute directory vnode operations template
5477 * This template is identical to the directory vnodes
5478 * operation template except for restricted operations:
5481 * Note that there are other restrictions embedded in:
5482 * zfs_create() - restrict type to VREG
5483 * zfs_link() - no links into/out of attribute space
5484 * zfs_rename() - no moves into/out of attribute space
5486 vnodeops_t *zfs_xdvnodeops;
5487 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5488 VOPNAME_OPEN, { .vop_open = zfs_open },
5489 VOPNAME_CLOSE, { .vop_close = zfs_close },
5490 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5491 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5492 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5493 VOPNAME_ACCESS, { .vop_access = zfs_access },
5494 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5495 VOPNAME_CREATE, { .vop_create = zfs_create },
5496 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5497 VOPNAME_LINK, { .vop_link = zfs_link },
5498 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5499 VOPNAME_MKDIR, { .error = zfs_inval },
5500 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5501 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5502 VOPNAME_SYMLINK, { .error = zfs_inval },
5503 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5504 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5505 VOPNAME_FID, { .vop_fid = zfs_fid },
5506 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5507 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5508 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5509 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5510 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5515 * Error vnode operations template
5517 vnodeops_t *zfs_evnodeops;
5518 const fs_operation_def_t zfs_evnodeops_template[] = {
5519 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5520 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5526 ioflags(int ioflags)
5530 if (ioflags & IO_APPEND)
5532 if (ioflags & IO_NDELAY)
5534 if (ioflags & IO_SYNC)
5535 flags |= (FSYNC | FDSYNC | FRSYNC);
5541 zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage)
5543 znode_t *zp = VTOZ(vp);
5544 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5545 objset_t *os = zp->z_zfsvfs->z_os;
5556 pcount = round_page(count) / PAGE_SIZE;
5558 object = mreq->object;
5561 KASSERT(vp->v_object == object, ("mismatching object"));
5563 VM_OBJECT_LOCK(object);
5564 vm_page_lock_queues();
5565 for (i = 0; i < pcount; i++) {
5570 vm_page_unlock_queues();
5573 if (mreq->valid != VM_PAGE_BITS_ALL)
5574 vm_page_zero_invalid(mreq, TRUE);
5575 VM_OBJECT_UNLOCK(object);
5577 return (VM_PAGER_OK);
5580 PCPU_INC(cnt.v_vnodein);
5581 PCPU_INC(cnt.v_vnodepgsin);
5583 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) {
5584 VM_OBJECT_UNLOCK(object);
5586 return (VM_PAGER_BAD);
5590 if (IDX_TO_OFF(mreq->pindex) + size > object->un_pager.vnp.vnp_size)
5591 size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mreq->pindex);
5593 VM_OBJECT_UNLOCK(object);
5594 va = zfs_map_page(mreq, &sf);
5595 error = dmu_read(os, zp->z_id, IDX_TO_OFF(mreq->pindex),
5596 size, va, DMU_READ_PREFETCH);
5597 if (size != PAGE_SIZE)
5598 bzero(va + size, PAGE_SIZE - size);
5600 VM_OBJECT_LOCK(object);
5603 mreq->valid = VM_PAGE_BITS_ALL;
5604 KASSERT(mreq->dirty == 0, ("zfs_getpages: page %p is dirty", mreq));
5606 VM_OBJECT_UNLOCK(object);
5608 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
5610 return (error ? VM_PAGER_ERROR : VM_PAGER_OK);
5614 zfs_freebsd_getpages(ap)
5615 struct vop_getpages_args /* {
5620 vm_ooffset_t a_offset;
5624 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage));
5628 zfs_freebsd_open(ap)
5629 struct vop_open_args /* {
5632 struct ucred *a_cred;
5633 struct thread *a_td;
5636 vnode_t *vp = ap->a_vp;
5637 znode_t *zp = VTOZ(vp);
5640 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL);
5642 vnode_create_vobject(vp, zp->z_size, ap->a_td);
5647 zfs_freebsd_close(ap)
5648 struct vop_close_args /* {
5651 struct ucred *a_cred;
5652 struct thread *a_td;
5656 return (zfs_close(ap->a_vp, ap->a_fflag, 0, 0, ap->a_cred, NULL));
5660 zfs_freebsd_ioctl(ap)
5661 struct vop_ioctl_args /* {
5671 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data,
5672 ap->a_fflag, ap->a_cred, NULL, NULL));
5676 zfs_freebsd_read(ap)
5677 struct vop_read_args /* {
5681 struct ucred *a_cred;
5685 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5690 zfs_freebsd_write(ap)
5691 struct vop_write_args /* {
5695 struct ucred *a_cred;
5699 if (vn_rlimit_fsize(ap->a_vp, ap->a_uio, ap->a_uio->uio_td))
5702 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5707 zfs_freebsd_access(ap)
5708 struct vop_access_args /* {
5710 accmode_t a_accmode;
5711 struct ucred *a_cred;
5712 struct thread *a_td;
5719 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND,
5721 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND);
5723 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL);
5726 * VADMIN has to be handled by vaccess().
5729 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND);
5731 vnode_t *vp = ap->a_vp;
5732 znode_t *zp = VTOZ(vp);
5734 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
5735 zp->z_gid, accmode, ap->a_cred, NULL);
5743 zfs_freebsd_lookup(ap)
5744 struct vop_lookup_args /* {
5745 struct vnode *a_dvp;
5746 struct vnode **a_vpp;
5747 struct componentname *a_cnp;
5750 struct componentname *cnp = ap->a_cnp;
5751 char nm[NAME_MAX + 1];
5753 ASSERT(cnp->cn_namelen < sizeof(nm));
5754 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm)));
5756 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop,
5757 cnp->cn_cred, cnp->cn_thread, 0));
5761 zfs_freebsd_create(ap)
5762 struct vop_create_args /* {
5763 struct vnode *a_dvp;
5764 struct vnode **a_vpp;
5765 struct componentname *a_cnp;
5766 struct vattr *a_vap;
5769 struct componentname *cnp = ap->a_cnp;
5770 vattr_t *vap = ap->a_vap;
5773 ASSERT(cnp->cn_flags & SAVENAME);
5775 vattr_init_mask(vap);
5776 mode = vap->va_mode & ALLPERMS;
5778 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode,
5779 ap->a_vpp, cnp->cn_cred, cnp->cn_thread));
5783 zfs_freebsd_remove(ap)
5784 struct vop_remove_args /* {
5785 struct vnode *a_dvp;
5787 struct componentname *a_cnp;
5791 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
5793 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr,
5794 ap->a_cnp->cn_cred, NULL, 0));
5798 zfs_freebsd_mkdir(ap)
5799 struct vop_mkdir_args /* {
5800 struct vnode *a_dvp;
5801 struct vnode **a_vpp;
5802 struct componentname *a_cnp;
5803 struct vattr *a_vap;
5806 vattr_t *vap = ap->a_vap;
5808 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
5810 vattr_init_mask(vap);
5812 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp,
5813 ap->a_cnp->cn_cred, NULL, 0, NULL));
5817 zfs_freebsd_rmdir(ap)
5818 struct vop_rmdir_args /* {
5819 struct vnode *a_dvp;
5821 struct componentname *a_cnp;
5824 struct componentname *cnp = ap->a_cnp;
5826 ASSERT(cnp->cn_flags & SAVENAME);
5828 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0));
5832 zfs_freebsd_readdir(ap)
5833 struct vop_readdir_args /* {
5836 struct ucred *a_cred;
5843 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag,
5844 ap->a_ncookies, ap->a_cookies));
5848 zfs_freebsd_fsync(ap)
5849 struct vop_fsync_args /* {
5852 struct thread *a_td;
5857 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL));
5861 zfs_freebsd_getattr(ap)
5862 struct vop_getattr_args /* {
5864 struct vattr *a_vap;
5865 struct ucred *a_cred;
5868 vattr_t *vap = ap->a_vap;
5874 xvap.xva_vattr = *vap;
5875 xvap.xva_vattr.va_mask |= AT_XVATTR;
5877 /* Convert chflags into ZFS-type flags. */
5878 /* XXX: what about SF_SETTABLE?. */
5879 XVA_SET_REQ(&xvap, XAT_IMMUTABLE);
5880 XVA_SET_REQ(&xvap, XAT_APPENDONLY);
5881 XVA_SET_REQ(&xvap, XAT_NOUNLINK);
5882 XVA_SET_REQ(&xvap, XAT_NODUMP);
5883 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL);
5887 /* Convert ZFS xattr into chflags. */
5888 #define FLAG_CHECK(fflag, xflag, xfield) do { \
5889 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \
5890 fflags |= (fflag); \
5892 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE,
5893 xvap.xva_xoptattrs.xoa_immutable);
5894 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY,
5895 xvap.xva_xoptattrs.xoa_appendonly);
5896 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK,
5897 xvap.xva_xoptattrs.xoa_nounlink);
5898 FLAG_CHECK(UF_NODUMP, XAT_NODUMP,
5899 xvap.xva_xoptattrs.xoa_nodump);
5901 *vap = xvap.xva_vattr;
5902 vap->va_flags = fflags;
5907 zfs_freebsd_setattr(ap)
5908 struct vop_setattr_args /* {
5910 struct vattr *a_vap;
5911 struct ucred *a_cred;
5914 vnode_t *vp = ap->a_vp;
5915 vattr_t *vap = ap->a_vap;
5916 cred_t *cred = ap->a_cred;
5921 vattr_init_mask(vap);
5922 vap->va_mask &= ~AT_NOSET;
5925 xvap.xva_vattr = *vap;
5927 zflags = VTOZ(vp)->z_pflags;
5929 if (vap->va_flags != VNOVAL) {
5930 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs;
5933 if (zfsvfs->z_use_fuids == B_FALSE)
5934 return (EOPNOTSUPP);
5936 fflags = vap->va_flags;
5937 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_NODUMP)) != 0)
5938 return (EOPNOTSUPP);
5940 * Unprivileged processes are not permitted to unset system
5941 * flags, or modify flags if any system flags are set.
5942 * Privileged non-jail processes may not modify system flags
5943 * if securelevel > 0 and any existing system flags are set.
5944 * Privileged jail processes behave like privileged non-jail
5945 * processes if the security.jail.chflags_allowed sysctl is
5946 * is non-zero; otherwise, they behave like unprivileged
5949 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 ||
5950 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) {
5952 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
5953 error = securelevel_gt(cred, 0);
5959 * Callers may only modify the file flags on objects they
5960 * have VADMIN rights for.
5962 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0)
5965 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
5969 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) {
5974 #define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \
5975 if (((fflags & (fflag)) && !(zflags & (zflag))) || \
5976 ((zflags & (zflag)) && !(fflags & (fflag)))) { \
5977 XVA_SET_REQ(&xvap, (xflag)); \
5978 (xfield) = ((fflags & (fflag)) != 0); \
5981 /* Convert chflags into ZFS-type flags. */
5982 /* XXX: what about SF_SETTABLE?. */
5983 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE,
5984 xvap.xva_xoptattrs.xoa_immutable);
5985 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY,
5986 xvap.xva_xoptattrs.xoa_appendonly);
5987 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK,
5988 xvap.xva_xoptattrs.xoa_nounlink);
5989 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP,
5990 xvap.xva_xoptattrs.xoa_nodump);
5993 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL));
5997 zfs_freebsd_rename(ap)
5998 struct vop_rename_args /* {
5999 struct vnode *a_fdvp;
6000 struct vnode *a_fvp;
6001 struct componentname *a_fcnp;
6002 struct vnode *a_tdvp;
6003 struct vnode *a_tvp;
6004 struct componentname *a_tcnp;
6007 vnode_t *fdvp = ap->a_fdvp;
6008 vnode_t *fvp = ap->a_fvp;
6009 vnode_t *tdvp = ap->a_tdvp;
6010 vnode_t *tvp = ap->a_tvp;
6013 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART));
6014 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART));
6016 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp,
6017 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0);
6032 zfs_freebsd_symlink(ap)
6033 struct vop_symlink_args /* {
6034 struct vnode *a_dvp;
6035 struct vnode **a_vpp;
6036 struct componentname *a_cnp;
6037 struct vattr *a_vap;
6041 struct componentname *cnp = ap->a_cnp;
6042 vattr_t *vap = ap->a_vap;
6044 ASSERT(cnp->cn_flags & SAVENAME);
6046 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */
6047 vattr_init_mask(vap);
6049 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap,
6050 ap->a_target, cnp->cn_cred, cnp->cn_thread));
6054 zfs_freebsd_readlink(ap)
6055 struct vop_readlink_args /* {
6058 struct ucred *a_cred;
6062 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL));
6066 zfs_freebsd_link(ap)
6067 struct vop_link_args /* {
6068 struct vnode *a_tdvp;
6070 struct componentname *a_cnp;
6073 struct componentname *cnp = ap->a_cnp;
6075 ASSERT(cnp->cn_flags & SAVENAME);
6077 return (zfs_link(ap->a_tdvp, ap->a_vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0));
6081 zfs_freebsd_inactive(ap)
6082 struct vop_inactive_args /* {
6084 struct thread *a_td;
6087 vnode_t *vp = ap->a_vp;
6089 zfs_inactive(vp, ap->a_td->td_ucred, NULL);
6094 zfs_reclaim_complete(void *arg, int pending)
6097 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
6099 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
6100 if (zp->z_sa_hdl != NULL) {
6101 ZFS_OBJ_HOLD_ENTER(zfsvfs, zp->z_id);
6102 zfs_znode_dmu_fini(zp);
6103 ZFS_OBJ_HOLD_EXIT(zfsvfs, zp->z_id);
6106 rw_exit(&zfsvfs->z_teardown_inactive_lock);
6108 * If the file system is being unmounted, there is a process waiting
6109 * for us, wake it up.
6111 if (zfsvfs->z_unmounted)
6116 zfs_freebsd_reclaim(ap)
6117 struct vop_reclaim_args /* {
6119 struct thread *a_td;
6122 vnode_t *vp = ap->a_vp;
6123 znode_t *zp = VTOZ(vp);
6124 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
6127 rlocked = rw_tryenter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
6132 * Destroy the vm object and flush associated pages.
6134 vnode_destroy_vobject(vp);
6136 mutex_enter(&zp->z_lock);
6138 mutex_exit(&zp->z_lock);
6140 if (zp->z_unlinked) {
6142 } else if (!rlocked) {
6143 TASK_INIT(&zp->z_task, 0, zfs_reclaim_complete, zp);
6144 taskqueue_enqueue(taskqueue_thread, &zp->z_task);
6145 } else if (zp->z_sa_hdl == NULL) {
6147 } else /* if (!zp->z_unlinked && zp->z_dbuf != NULL) */ {
6150 locked = MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id)) ? 2 :
6151 ZFS_OBJ_HOLD_TRYENTER(zfsvfs, zp->z_id);
6154 * Lock can't be obtained due to deadlock possibility,
6155 * so defer znode destruction.
6157 TASK_INIT(&zp->z_task, 0, zfs_reclaim_complete, zp);
6158 taskqueue_enqueue(taskqueue_thread, &zp->z_task);
6160 zfs_znode_dmu_fini(zp);
6162 ZFS_OBJ_HOLD_EXIT(zfsvfs, zp->z_id);
6168 ASSERT(vp->v_holdcnt >= 1);
6171 rw_exit(&zfsvfs->z_teardown_inactive_lock);
6177 struct vop_fid_args /* {
6183 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL));
6187 zfs_freebsd_pathconf(ap)
6188 struct vop_pathconf_args /* {
6191 register_t *a_retval;
6197 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL);
6199 *ap->a_retval = val;
6200 else if (error == EOPNOTSUPP)
6201 error = vop_stdpathconf(ap);
6206 zfs_freebsd_fifo_pathconf(ap)
6207 struct vop_pathconf_args /* {
6210 register_t *a_retval;
6214 switch (ap->a_name) {
6215 case _PC_ACL_EXTENDED:
6217 case _PC_ACL_PATH_MAX:
6218 case _PC_MAC_PRESENT:
6219 return (zfs_freebsd_pathconf(ap));
6221 return (fifo_specops.vop_pathconf(ap));
6226 * FreeBSD's extended attributes namespace defines file name prefix for ZFS'
6227 * extended attribute name:
6230 * system freebsd:system:
6231 * user (none, can be used to access ZFS fsattr(5) attributes
6232 * created on Solaris)
6235 zfs_create_attrname(int attrnamespace, const char *name, char *attrname,
6238 const char *namespace, *prefix, *suffix;
6240 /* We don't allow '/' character in attribute name. */
6241 if (strchr(name, '/') != NULL)
6243 /* We don't allow attribute names that start with "freebsd:" string. */
6244 if (strncmp(name, "freebsd:", 8) == 0)
6247 bzero(attrname, size);
6249 switch (attrnamespace) {
6250 case EXTATTR_NAMESPACE_USER:
6252 prefix = "freebsd:";
6253 namespace = EXTATTR_NAMESPACE_USER_STRING;
6257 * This is the default namespace by which we can access all
6258 * attributes created on Solaris.
6260 prefix = namespace = suffix = "";
6263 case EXTATTR_NAMESPACE_SYSTEM:
6264 prefix = "freebsd:";
6265 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING;
6268 case EXTATTR_NAMESPACE_EMPTY:
6272 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix,
6274 return (ENAMETOOLONG);
6280 * Vnode operating to retrieve a named extended attribute.
6283 zfs_getextattr(struct vop_getextattr_args *ap)
6286 IN struct vnode *a_vp;
6287 IN int a_attrnamespace;
6288 IN const char *a_name;
6289 INOUT struct uio *a_uio;
6291 IN struct ucred *a_cred;
6292 IN struct thread *a_td;
6296 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6297 struct thread *td = ap->a_td;
6298 struct nameidata nd;
6301 vnode_t *xvp = NULL, *vp;
6304 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6305 ap->a_cred, ap->a_td, VREAD);
6309 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6316 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6324 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_SYSSPACE, attrname,
6326 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL);
6328 NDFREE(&nd, NDF_ONLY_PNBUF);
6331 if (error == ENOENT)
6336 if (ap->a_size != NULL) {
6337 error = VOP_GETATTR(vp, &va, ap->a_cred);
6339 *ap->a_size = (size_t)va.va_size;
6340 } else if (ap->a_uio != NULL)
6341 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6344 vn_close(vp, flags, ap->a_cred, td);
6351 * Vnode operation to remove a named attribute.
6354 zfs_deleteextattr(struct vop_deleteextattr_args *ap)
6357 IN struct vnode *a_vp;
6358 IN int a_attrnamespace;
6359 IN const char *a_name;
6360 IN struct ucred *a_cred;
6361 IN struct thread *a_td;
6365 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6366 struct thread *td = ap->a_td;
6367 struct nameidata nd;
6370 vnode_t *xvp = NULL, *vp;
6373 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6374 ap->a_cred, ap->a_td, VWRITE);
6378 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6385 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6392 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF | MPSAFE,
6393 UIO_SYSSPACE, attrname, xvp, td);
6396 NDFREE(&nd, NDF_ONLY_PNBUF);
6399 if (error == ENOENT)
6403 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
6406 if (vp == nd.ni_dvp)
6416 * Vnode operation to set a named attribute.
6419 zfs_setextattr(struct vop_setextattr_args *ap)
6422 IN struct vnode *a_vp;
6423 IN int a_attrnamespace;
6424 IN const char *a_name;
6425 INOUT struct uio *a_uio;
6426 IN struct ucred *a_cred;
6427 IN struct thread *a_td;
6431 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6432 struct thread *td = ap->a_td;
6433 struct nameidata nd;
6436 vnode_t *xvp = NULL, *vp;
6439 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6440 ap->a_cred, ap->a_td, VWRITE);
6444 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6451 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6452 LOOKUP_XATTR | CREATE_XATTR_DIR);
6458 flags = FFLAGS(O_WRONLY | O_CREAT);
6459 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_SYSSPACE, attrname,
6461 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL);
6463 NDFREE(&nd, NDF_ONLY_PNBUF);
6471 error = VOP_SETATTR(vp, &va, ap->a_cred);
6473 VOP_WRITE(vp, ap->a_uio, IO_UNIT | IO_SYNC, ap->a_cred);
6476 vn_close(vp, flags, ap->a_cred, td);
6483 * Vnode operation to retrieve extended attributes on a vnode.
6486 zfs_listextattr(struct vop_listextattr_args *ap)
6489 IN struct vnode *a_vp;
6490 IN int a_attrnamespace;
6491 INOUT struct uio *a_uio;
6493 IN struct ucred *a_cred;
6494 IN struct thread *a_td;
6498 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6499 struct thread *td = ap->a_td;
6500 struct nameidata nd;
6501 char attrprefix[16];
6502 u_char dirbuf[sizeof(struct dirent)];
6505 struct uio auio, *uio = ap->a_uio;
6506 size_t *sizep = ap->a_size;
6508 vnode_t *xvp = NULL, *vp;
6509 int done, error, eof, pos;
6511 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6512 ap->a_cred, ap->a_td, VREAD);
6516 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix,
6517 sizeof(attrprefix));
6520 plen = strlen(attrprefix);
6527 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6532 * ENOATTR means that the EA directory does not yet exist,
6533 * i.e. there are no extended attributes there.
6535 if (error == ENOATTR)
6540 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED | MPSAFE,
6541 UIO_SYSSPACE, ".", xvp, td);
6544 NDFREE(&nd, NDF_ONLY_PNBUF);
6550 auio.uio_iov = &aiov;
6551 auio.uio_iovcnt = 1;
6552 auio.uio_segflg = UIO_SYSSPACE;
6554 auio.uio_rw = UIO_READ;
6555 auio.uio_offset = 0;
6560 aiov.iov_base = (void *)dirbuf;
6561 aiov.iov_len = sizeof(dirbuf);
6562 auio.uio_resid = sizeof(dirbuf);
6563 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL);
6564 done = sizeof(dirbuf) - auio.uio_resid;
6567 for (pos = 0; pos < done;) {
6568 dp = (struct dirent *)(dirbuf + pos);
6569 pos += dp->d_reclen;
6571 * XXX: Temporarily we also accept DT_UNKNOWN, as this
6572 * is what we get when attribute was created on Solaris.
6574 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN)
6576 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0)
6578 else if (strncmp(dp->d_name, attrprefix, plen) != 0)
6580 nlen = dp->d_namlen - plen;
6583 else if (uio != NULL) {
6585 * Format of extattr name entry is one byte for
6586 * length and the rest for name.
6588 error = uiomove(&nlen, 1, uio->uio_rw, uio);
6590 error = uiomove(dp->d_name + plen, nlen,
6597 } while (!eof && error == 0);
6606 zfs_freebsd_getacl(ap)
6607 struct vop_getacl_args /* {
6616 vsecattr_t vsecattr;
6618 if (ap->a_type != ACL_TYPE_NFS4)
6621 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
6622 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL))
6625 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt);
6626 if (vsecattr.vsa_aclentp != NULL)
6627 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz);
6633 zfs_freebsd_setacl(ap)
6634 struct vop_setacl_args /* {
6643 vsecattr_t vsecattr;
6644 int aclbsize; /* size of acl list in bytes */
6647 if (ap->a_type != ACL_TYPE_NFS4)
6650 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES)
6654 * With NFSv4 ACLs, chmod(2) may need to add additional entries,
6655 * splitting every entry into two and appending "canonical six"
6656 * entries at the end. Don't allow for setting an ACL that would
6657 * cause chmod(2) to run out of ACL entries.
6659 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES)
6662 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR);
6666 vsecattr.vsa_mask = VSA_ACE;
6667 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t);
6668 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP);
6669 aaclp = vsecattr.vsa_aclentp;
6670 vsecattr.vsa_aclentsz = aclbsize;
6672 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp);
6673 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL);
6674 kmem_free(aaclp, aclbsize);
6680 zfs_freebsd_aclcheck(ap)
6681 struct vop_aclcheck_args /* {
6690 return (EOPNOTSUPP);
6693 struct vop_vector zfs_vnodeops;
6694 struct vop_vector zfs_fifoops;
6695 struct vop_vector zfs_shareops;
6697 struct vop_vector zfs_vnodeops = {
6698 .vop_default = &default_vnodeops,
6699 .vop_inactive = zfs_freebsd_inactive,
6700 .vop_reclaim = zfs_freebsd_reclaim,
6701 .vop_access = zfs_freebsd_access,
6702 #ifdef FREEBSD_NAMECACHE
6703 .vop_lookup = vfs_cache_lookup,
6704 .vop_cachedlookup = zfs_freebsd_lookup,
6706 .vop_lookup = zfs_freebsd_lookup,
6708 .vop_getattr = zfs_freebsd_getattr,
6709 .vop_setattr = zfs_freebsd_setattr,
6710 .vop_create = zfs_freebsd_create,
6711 .vop_mknod = zfs_freebsd_create,
6712 .vop_mkdir = zfs_freebsd_mkdir,
6713 .vop_readdir = zfs_freebsd_readdir,
6714 .vop_fsync = zfs_freebsd_fsync,
6715 .vop_open = zfs_freebsd_open,
6716 .vop_close = zfs_freebsd_close,
6717 .vop_rmdir = zfs_freebsd_rmdir,
6718 .vop_ioctl = zfs_freebsd_ioctl,
6719 .vop_link = zfs_freebsd_link,
6720 .vop_symlink = zfs_freebsd_symlink,
6721 .vop_readlink = zfs_freebsd_readlink,
6722 .vop_read = zfs_freebsd_read,
6723 .vop_write = zfs_freebsd_write,
6724 .vop_remove = zfs_freebsd_remove,
6725 .vop_rename = zfs_freebsd_rename,
6726 .vop_pathconf = zfs_freebsd_pathconf,
6727 .vop_bmap = VOP_EOPNOTSUPP,
6728 .vop_fid = zfs_freebsd_fid,
6729 .vop_getextattr = zfs_getextattr,
6730 .vop_deleteextattr = zfs_deleteextattr,
6731 .vop_setextattr = zfs_setextattr,
6732 .vop_listextattr = zfs_listextattr,
6733 .vop_getacl = zfs_freebsd_getacl,
6734 .vop_setacl = zfs_freebsd_setacl,
6735 .vop_aclcheck = zfs_freebsd_aclcheck,
6736 .vop_getpages = zfs_freebsd_getpages,
6739 struct vop_vector zfs_fifoops = {
6740 .vop_default = &fifo_specops,
6741 .vop_fsync = zfs_freebsd_fsync,
6742 .vop_access = zfs_freebsd_access,
6743 .vop_getattr = zfs_freebsd_getattr,
6744 .vop_inactive = zfs_freebsd_inactive,
6745 .vop_read = VOP_PANIC,
6746 .vop_reclaim = zfs_freebsd_reclaim,
6747 .vop_setattr = zfs_freebsd_setattr,
6748 .vop_write = VOP_PANIC,
6749 .vop_pathconf = zfs_freebsd_fifo_pathconf,
6750 .vop_fid = zfs_freebsd_fid,
6751 .vop_getacl = zfs_freebsd_getacl,
6752 .vop_setacl = zfs_freebsd_setacl,
6753 .vop_aclcheck = zfs_freebsd_aclcheck,
6757 * special share hidden files vnode operations template
6759 struct vop_vector zfs_shareops = {
6760 .vop_default = &default_vnodeops,
6761 .vop_access = zfs_freebsd_access,
6762 .vop_inactive = zfs_freebsd_inactive,
6763 .vop_reclaim = zfs_freebsd_reclaim,
6764 .vop_fid = zfs_freebsd_fid,
6765 .vop_pathconf = zfs_freebsd_pathconf,