4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012 by Delphix. All rights reserved.
26 /* Portions Copyright 2007 Jeremy Teo */
27 /* Portions Copyright 2010 Robert Milkowski */
29 #include <sys/types.h>
30 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/sysmacros.h>
34 #include <sys/resource.h>
35 #include <sys/resourcevar.h>
37 #include <sys/vnode.h>
41 #include <sys/taskq.h>
43 #include <sys/atomic.h>
44 #include <sys/namei.h>
46 #include <sys/cmn_err.h>
47 #include <sys/errno.h>
48 #include <sys/unistd.h>
49 #include <sys/zfs_dir.h>
50 #include <sys/zfs_ioctl.h>
51 #include <sys/fs/zfs.h>
53 #include <sys/dmu_objset.h>
59 #include <sys/dirent.h>
60 #include <sys/policy.h>
61 #include <sys/sunddi.h>
62 #include <sys/filio.h>
64 #include <sys/zfs_ctldir.h>
65 #include <sys/zfs_fuid.h>
66 #include <sys/zfs_sa.h>
68 #include <sys/zfs_rlock.h>
69 #include <sys/extdirent.h>
70 #include <sys/kidmap.h>
73 #include <sys/sf_buf.h>
74 #include <sys/sched.h>
76 #include <vm/vm_pageout.h>
81 * Each vnode op performs some logical unit of work. To do this, the ZPL must
82 * properly lock its in-core state, create a DMU transaction, do the work,
83 * record this work in the intent log (ZIL), commit the DMU transaction,
84 * and wait for the intent log to commit if it is a synchronous operation.
85 * Moreover, the vnode ops must work in both normal and log replay context.
86 * The ordering of events is important to avoid deadlocks and references
87 * to freed memory. The example below illustrates the following Big Rules:
89 * (1) A check must be made in each zfs thread for a mounted file system.
90 * This is done avoiding races using ZFS_ENTER(zfsvfs).
91 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
92 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
93 * can return EIO from the calling function.
95 * (2) VN_RELE() should always be the last thing except for zil_commit()
96 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
97 * First, if it's the last reference, the vnode/znode
98 * can be freed, so the zp may point to freed memory. Second, the last
99 * reference will call zfs_zinactive(), which may induce a lot of work --
100 * pushing cached pages (which acquires range locks) and syncing out
101 * cached atime changes. Third, zfs_zinactive() may require a new tx,
102 * which could deadlock the system if you were already holding one.
103 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
105 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
106 * as they can span dmu_tx_assign() calls.
108 * (4) Always pass TXG_NOWAIT as the second argument to dmu_tx_assign().
109 * This is critical because we don't want to block while holding locks.
110 * Note, in particular, that if a lock is sometimes acquired before
111 * the tx assigns, and sometimes after (e.g. z_lock), then failing to
112 * use a non-blocking assign can deadlock the system. The scenario:
114 * Thread A has grabbed a lock before calling dmu_tx_assign().
115 * Thread B is in an already-assigned tx, and blocks for this lock.
116 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
117 * forever, because the previous txg can't quiesce until B's tx commits.
119 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
120 * then drop all locks, call dmu_tx_wait(), and try again.
122 * (5) If the operation succeeded, generate the intent log entry for it
123 * before dropping locks. This ensures that the ordering of events
124 * in the intent log matches the order in which they actually occurred.
125 * During ZIL replay the zfs_log_* functions will update the sequence
126 * number to indicate the zil transaction has replayed.
128 * (6) At the end of each vnode op, the DMU tx must always commit,
129 * regardless of whether there were any errors.
131 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
132 * to ensure that synchronous semantics are provided when necessary.
134 * In general, this is how things should be ordered in each vnode op:
136 * ZFS_ENTER(zfsvfs); // exit if unmounted
138 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD())
139 * rw_enter(...); // grab any other locks you need
140 * tx = dmu_tx_create(...); // get DMU tx
141 * dmu_tx_hold_*(); // hold each object you might modify
142 * error = dmu_tx_assign(tx, TXG_NOWAIT); // try to assign
144 * rw_exit(...); // drop locks
145 * zfs_dirent_unlock(dl); // unlock directory entry
146 * VN_RELE(...); // release held vnodes
147 * if (error == ERESTART) {
152 * dmu_tx_abort(tx); // abort DMU tx
153 * ZFS_EXIT(zfsvfs); // finished in zfs
154 * return (error); // really out of space
156 * error = do_real_work(); // do whatever this VOP does
158 * zfs_log_*(...); // on success, make ZIL entry
159 * dmu_tx_commit(tx); // commit DMU tx -- error or not
160 * rw_exit(...); // drop locks
161 * zfs_dirent_unlock(dl); // unlock directory entry
162 * VN_RELE(...); // release held vnodes
163 * zil_commit(zilog, foid); // synchronous when necessary
164 * ZFS_EXIT(zfsvfs); // finished in zfs
165 * return (error); // done, report error
170 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
172 znode_t *zp = VTOZ(*vpp);
173 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
178 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
179 ((flag & FAPPEND) == 0)) {
184 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
185 ZTOV(zp)->v_type == VREG &&
186 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
187 if (fs_vscan(*vpp, cr, 0) != 0) {
193 /* Keep a count of the synchronous opens in the znode */
194 if (flag & (FSYNC | FDSYNC))
195 atomic_inc_32(&zp->z_sync_cnt);
203 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
204 caller_context_t *ct)
206 znode_t *zp = VTOZ(vp);
207 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
210 * Clean up any locks held by this process on the vp.
212 cleanlocks(vp, ddi_get_pid(), 0);
213 cleanshares(vp, ddi_get_pid());
218 /* Decrement the synchronous opens in the znode */
219 if ((flag & (FSYNC | FDSYNC)) && (count == 1))
220 atomic_dec_32(&zp->z_sync_cnt);
222 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
223 ZTOV(zp)->v_type == VREG &&
224 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
225 VERIFY(fs_vscan(vp, cr, 1) == 0);
232 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
233 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
236 zfs_holey(vnode_t *vp, u_long cmd, offset_t *off)
238 znode_t *zp = VTOZ(vp);
239 uint64_t noff = (uint64_t)*off; /* new offset */
244 file_sz = zp->z_size;
245 if (noff >= file_sz) {
249 if (cmd == _FIO_SEEK_HOLE)
254 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
257 if ((error == ESRCH) || (noff > file_sz)) {
259 * Handle the virtual hole at the end of file.
276 zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred,
277 int *rvalp, caller_context_t *ct)
289 * The following two ioctls are used by bfu. Faking out,
290 * necessary to avoid bfu errors.
299 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
302 off = *(offset_t *)data;
305 zfsvfs = zp->z_zfsvfs;
309 /* offset parameter is in/out */
310 error = zfs_holey(vp, com, &off);
315 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
318 *(offset_t *)data = off;
326 page_lookup(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes)
332 VM_OBJECT_LOCK_ASSERT(obj, MA_OWNED);
335 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
336 vm_page_is_valid(pp, (vm_offset_t)off, nbytes)) {
337 if (vm_page_sleep_if_busy(pp, FALSE, "zfsmwb"))
340 vm_page_lock_queues();
342 vm_page_unlock_queues();
344 if (__predict_false(obj->cache != NULL)) {
345 vm_page_cache_free(obj, OFF_TO_IDX(start),
346 OFF_TO_IDX(start) + 1);
356 page_unlock(vm_page_t pp)
363 zfs_map_page(vm_page_t pp, struct sf_buf **sfp)
366 *sfp = sf_buf_alloc(pp, 0);
367 return ((caddr_t)sf_buf_kva(*sfp));
371 zfs_unmap_page(struct sf_buf *sf)
378 * When a file is memory mapped, we must keep the IO data synchronized
379 * between the DMU cache and the memory mapped pages. What this means:
381 * On Write: If we find a memory mapped page, we write to *both*
382 * the page and the dmu buffer.
385 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid,
386 int segflg, dmu_tx_t *tx)
392 ASSERT(vp->v_mount != NULL);
396 off = start & PAGEOFFSET;
398 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
400 int nbytes = MIN(PAGESIZE - off, len);
402 if ((pp = page_lookup(vp, start, off, nbytes)) != NULL) {
405 VM_OBJECT_UNLOCK(obj);
406 va = zfs_map_page(pp, &sf);
407 if (segflg == UIO_NOCOPY) {
408 (void) dmu_write(os, oid, start+off, nbytes,
411 (void) dmu_read(os, oid, start+off, nbytes,
412 va+off, DMU_READ_PREFETCH);
421 VM_OBJECT_UNLOCK(obj);
425 * Read with UIO_NOCOPY flag means that sendfile(2) requests
426 * ZFS to populate a range of page cache pages with data.
428 * NOTE: this function could be optimized to pre-allocate
429 * all pages in advance, drain VPO_BUSY on all of them,
430 * map them into contiguous KVA region and populate them
431 * in one single dmu_read() call.
434 mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio)
436 znode_t *zp = VTOZ(vp);
437 objset_t *os = zp->z_zfsvfs->z_os;
447 ASSERT(uio->uio_segflg == UIO_NOCOPY);
448 ASSERT(vp->v_mount != NULL);
451 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0);
454 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) {
455 int bytes = MIN(PAGESIZE, len);
458 pp = vm_page_lookup(obj, OFF_TO_IDX(start));
459 if (pp != NULL && vm_page_sleep_if_busy(pp, FALSE,
463 pp = vm_page_alloc(obj, OFF_TO_IDX(start),
464 VM_ALLOC_NOBUSY | VM_ALLOC_NORMAL);
466 VM_OBJECT_UNLOCK(obj);
472 if (pp->valid == 0) {
473 vm_page_io_start(pp);
474 VM_OBJECT_UNLOCK(obj);
475 va = zfs_map_page(pp, &sf);
476 error = dmu_read(os, zp->z_id, start, bytes, va,
478 if (bytes != PAGESIZE && error == 0)
479 bzero(va + bytes, PAGESIZE - bytes);
482 vm_page_io_finish(pp);
483 vm_page_lock_queues();
487 pp->valid = VM_PAGE_BITS_ALL;
488 vm_page_activate(pp);
490 vm_page_unlock_queues();
494 uio->uio_resid -= bytes;
495 uio->uio_offset += bytes;
498 VM_OBJECT_UNLOCK(obj);
503 * When a file is memory mapped, we must keep the IO data synchronized
504 * between the DMU cache and the memory mapped pages. What this means:
506 * On Read: We "read" preferentially from memory mapped pages,
507 * else we default from the dmu buffer.
509 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
510 * the file is memory mapped.
513 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
515 znode_t *zp = VTOZ(vp);
516 objset_t *os = zp->z_zfsvfs->z_os;
524 ASSERT(vp->v_mount != NULL);
528 start = uio->uio_loffset;
529 off = start & PAGEOFFSET;
531 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
533 uint64_t bytes = MIN(PAGESIZE - off, len);
535 if (pp = page_lookup(vp, start, off, bytes)) {
538 VM_OBJECT_UNLOCK(obj);
539 va = zfs_map_page(pp, &sf);
540 error = uiomove(va + off, bytes, UIO_READ, uio);
545 VM_OBJECT_UNLOCK(obj);
546 error = dmu_read_uio(os, zp->z_id, uio, bytes);
554 VM_OBJECT_UNLOCK(obj);
558 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
561 * Read bytes from specified file into supplied buffer.
563 * IN: vp - vnode of file to be read from.
564 * uio - structure supplying read location, range info,
566 * ioflag - SYNC flags; used to provide FRSYNC semantics.
567 * cr - credentials of caller.
568 * ct - caller context
570 * OUT: uio - updated offset and range, buffer filled.
572 * RETURN: 0 if success
573 * error code if failure
576 * vp - atime updated if byte count > 0
580 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
582 znode_t *zp = VTOZ(vp);
583 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
594 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
600 * Validate file offset
602 if (uio->uio_loffset < (offset_t)0) {
608 * Fasttrack empty reads
610 if (uio->uio_resid == 0) {
616 * Check for mandatory locks
618 if (MANDMODE(zp->z_mode)) {
619 if (error = chklock(vp, FREAD,
620 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
627 * If we're in FRSYNC mode, sync out this znode before reading it.
630 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
631 zil_commit(zfsvfs->z_log, zp->z_id);
634 * Lock the range against changes.
636 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
639 * If we are reading past end-of-file we can skip
640 * to the end; but we might still need to set atime.
642 if (uio->uio_loffset >= zp->z_size) {
647 ASSERT(uio->uio_loffset < zp->z_size);
648 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
651 if ((uio->uio_extflg == UIO_XUIO) &&
652 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
654 int blksz = zp->z_blksz;
655 uint64_t offset = uio->uio_loffset;
657 xuio = (xuio_t *)uio;
659 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
662 ASSERT(offset + n <= blksz);
665 (void) dmu_xuio_init(xuio, nblk);
667 if (vn_has_cached_data(vp)) {
669 * For simplicity, we always allocate a full buffer
670 * even if we only expect to read a portion of a block.
672 while (--nblk >= 0) {
673 (void) dmu_xuio_add(xuio,
674 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
682 nbytes = MIN(n, zfs_read_chunk_size -
683 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
686 if (uio->uio_segflg == UIO_NOCOPY)
687 error = mappedread_sf(vp, nbytes, uio);
689 #endif /* __FreeBSD__ */
690 if (vn_has_cached_data(vp))
691 error = mappedread(vp, nbytes, uio);
693 error = dmu_read_uio(os, zp->z_id, uio, nbytes);
695 /* convert checksum errors into IO errors */
704 zfs_range_unlock(rl);
706 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
712 * Write the bytes to a file.
714 * IN: vp - vnode of file to be written to.
715 * uio - structure supplying write location, range info,
717 * ioflag - FAPPEND flag set if in append mode.
718 * cr - credentials of caller.
719 * ct - caller context (NFS/CIFS fem monitor only)
721 * OUT: uio - updated offset and range.
723 * RETURN: 0 if success
724 * error code if failure
727 * vp - ctime|mtime updated if byte count > 0
732 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
734 znode_t *zp = VTOZ(vp);
735 rlim64_t limit = MAXOFFSET_T;
736 ssize_t start_resid = uio->uio_resid;
740 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
745 int max_blksz = zfsvfs->z_max_blksz;
751 int iovcnt = uio->uio_iovcnt;
752 iovec_t *iovp = uio->uio_iov;
755 sa_bulk_attr_t bulk[4];
756 uint64_t mtime[2], ctime[2];
759 * Fasttrack empty write
765 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
771 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
772 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
773 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
775 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
779 * If immutable or not appending then return EPERM
781 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
782 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
783 (uio->uio_loffset < zp->z_size))) {
788 zilog = zfsvfs->z_log;
791 * Validate file offset
793 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
800 * Check for mandatory locks before calling zfs_range_lock()
801 * in order to prevent a deadlock with locks set via fcntl().
803 if (MANDMODE((mode_t)zp->z_mode) &&
804 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
811 * Pre-fault the pages to ensure slow (eg NFS) pages
813 * Skip this if uio contains loaned arc_buf.
815 if ((uio->uio_extflg == UIO_XUIO) &&
816 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
817 xuio = (xuio_t *)uio;
819 uio_prefaultpages(MIN(n, max_blksz), uio);
823 * If in append mode, set the io offset pointer to eof.
825 if (ioflag & FAPPEND) {
827 * Obtain an appending range lock to guarantee file append
828 * semantics. We reset the write offset once we have the lock.
830 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
832 if (rl->r_len == UINT64_MAX) {
834 * We overlocked the file because this write will cause
835 * the file block size to increase.
836 * Note that zp_size cannot change with this lock held.
840 uio->uio_loffset = woff;
843 * Note that if the file block size will change as a result of
844 * this write, then this range lock will lock the entire file
845 * so that we can re-write the block safely.
847 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
851 zfs_range_unlock(rl);
856 if ((woff + n) > limit || woff > (limit - n))
859 /* Will this write extend the file length? */
860 write_eof = (woff + n > zp->z_size);
862 end_size = MAX(zp->z_size, woff + n);
865 * Write the file in reasonable size chunks. Each chunk is written
866 * in a separate transaction; this keeps the intent log records small
867 * and allows us to do more fine-grained space accounting.
871 woff = uio->uio_loffset;
873 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
874 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
876 dmu_return_arcbuf(abuf);
881 if (xuio && abuf == NULL) {
882 ASSERT(i_iov < iovcnt);
884 abuf = dmu_xuio_arcbuf(xuio, i_iov);
885 dmu_xuio_clear(xuio, i_iov);
886 DTRACE_PROBE3(zfs_cp_write, int, i_iov,
887 iovec_t *, aiov, arc_buf_t *, abuf);
888 ASSERT((aiov->iov_base == abuf->b_data) ||
889 ((char *)aiov->iov_base - (char *)abuf->b_data +
890 aiov->iov_len == arc_buf_size(abuf)));
892 } else if (abuf == NULL && n >= max_blksz &&
893 woff >= zp->z_size &&
894 P2PHASE(woff, max_blksz) == 0 &&
895 zp->z_blksz == max_blksz) {
897 * This write covers a full block. "Borrow" a buffer
898 * from the dmu so that we can fill it before we enter
899 * a transaction. This avoids the possibility of
900 * holding up the transaction if the data copy hangs
901 * up on a pagefault (e.g., from an NFS server mapping).
905 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
907 ASSERT(abuf != NULL);
908 ASSERT(arc_buf_size(abuf) == max_blksz);
909 if (error = uiocopy(abuf->b_data, max_blksz,
910 UIO_WRITE, uio, &cbytes)) {
911 dmu_return_arcbuf(abuf);
914 ASSERT(cbytes == max_blksz);
918 * Start a transaction.
920 tx = dmu_tx_create(zfsvfs->z_os);
921 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
922 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
923 zfs_sa_upgrade_txholds(tx, zp);
924 error = dmu_tx_assign(tx, TXG_NOWAIT);
926 if (error == ERESTART) {
933 dmu_return_arcbuf(abuf);
938 * If zfs_range_lock() over-locked we grow the blocksize
939 * and then reduce the lock range. This will only happen
940 * on the first iteration since zfs_range_reduce() will
941 * shrink down r_len to the appropriate size.
943 if (rl->r_len == UINT64_MAX) {
946 if (zp->z_blksz > max_blksz) {
947 ASSERT(!ISP2(zp->z_blksz));
948 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
950 new_blksz = MIN(end_size, max_blksz);
952 zfs_grow_blocksize(zp, new_blksz, tx);
953 zfs_range_reduce(rl, woff, n);
957 * XXX - should we really limit each write to z_max_blksz?
958 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
960 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
962 if (woff + nbytes > zp->z_size)
963 vnode_pager_setsize(vp, woff + nbytes);
966 tx_bytes = uio->uio_resid;
967 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
969 tx_bytes -= uio->uio_resid;
972 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
974 * If this is not a full block write, but we are
975 * extending the file past EOF and this data starts
976 * block-aligned, use assign_arcbuf(). Otherwise,
977 * write via dmu_write().
979 if (tx_bytes < max_blksz && (!write_eof ||
980 aiov->iov_base != abuf->b_data)) {
982 dmu_write(zfsvfs->z_os, zp->z_id, woff,
983 aiov->iov_len, aiov->iov_base, tx);
984 dmu_return_arcbuf(abuf);
985 xuio_stat_wbuf_copied();
987 ASSERT(xuio || tx_bytes == max_blksz);
988 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
991 ASSERT(tx_bytes <= uio->uio_resid);
992 uioskip(uio, tx_bytes);
994 if (tx_bytes && vn_has_cached_data(vp)) {
995 update_pages(vp, woff, tx_bytes, zfsvfs->z_os,
996 zp->z_id, uio->uio_segflg, tx);
1000 * If we made no progress, we're done. If we made even
1001 * partial progress, update the znode and ZIL accordingly.
1003 if (tx_bytes == 0) {
1004 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
1005 (void *)&zp->z_size, sizeof (uint64_t), tx);
1012 * Clear Set-UID/Set-GID bits on successful write if not
1013 * privileged and at least one of the excute bits is set.
1015 * It would be nice to to this after all writes have
1016 * been done, but that would still expose the ISUID/ISGID
1017 * to another app after the partial write is committed.
1019 * Note: we don't call zfs_fuid_map_id() here because
1020 * user 0 is not an ephemeral uid.
1022 mutex_enter(&zp->z_acl_lock);
1023 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
1024 (S_IXUSR >> 6))) != 0 &&
1025 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
1026 secpolicy_vnode_setid_retain(vp, cr,
1027 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
1029 zp->z_mode &= ~(S_ISUID | S_ISGID);
1030 newmode = zp->z_mode;
1031 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
1032 (void *)&newmode, sizeof (uint64_t), tx);
1034 mutex_exit(&zp->z_acl_lock);
1036 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
1040 * Update the file size (zp_size) if it has changed;
1041 * account for possible concurrent updates.
1043 while ((end_size = zp->z_size) < uio->uio_loffset) {
1044 (void) atomic_cas_64(&zp->z_size, end_size,
1049 * If we are replaying and eof is non zero then force
1050 * the file size to the specified eof. Note, there's no
1051 * concurrency during replay.
1053 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
1054 zp->z_size = zfsvfs->z_replay_eof;
1056 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1058 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
1063 ASSERT(tx_bytes == nbytes);
1068 uio_prefaultpages(MIN(n, max_blksz), uio);
1072 zfs_range_unlock(rl);
1075 * If we're in replay mode, or we made no progress, return error.
1076 * Otherwise, it's at least a partial write, so it's successful.
1078 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1083 if (ioflag & (FSYNC | FDSYNC) ||
1084 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1085 zil_commit(zilog, zp->z_id);
1092 zfs_get_done(zgd_t *zgd, int error)
1094 znode_t *zp = zgd->zgd_private;
1095 objset_t *os = zp->z_zfsvfs->z_os;
1099 dmu_buf_rele(zgd->zgd_db, zgd);
1101 zfs_range_unlock(zgd->zgd_rl);
1103 vfslocked = VFS_LOCK_GIANT(zp->z_zfsvfs->z_vfs);
1105 * Release the vnode asynchronously as we currently have the
1106 * txg stopped from syncing.
1108 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1110 if (error == 0 && zgd->zgd_bp)
1111 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1113 kmem_free(zgd, sizeof (zgd_t));
1114 VFS_UNLOCK_GIANT(vfslocked);
1118 static int zil_fault_io = 0;
1122 * Get data to generate a TX_WRITE intent log record.
1125 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1127 zfsvfs_t *zfsvfs = arg;
1128 objset_t *os = zfsvfs->z_os;
1130 uint64_t object = lr->lr_foid;
1131 uint64_t offset = lr->lr_offset;
1132 uint64_t size = lr->lr_length;
1133 blkptr_t *bp = &lr->lr_blkptr;
1138 ASSERT(zio != NULL);
1142 * Nothing to do if the file has been removed
1144 if (zfs_zget(zfsvfs, object, &zp) != 0)
1146 if (zp->z_unlinked) {
1148 * Release the vnode asynchronously as we currently have the
1149 * txg stopped from syncing.
1151 VN_RELE_ASYNC(ZTOV(zp),
1152 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1156 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1157 zgd->zgd_zilog = zfsvfs->z_log;
1158 zgd->zgd_private = zp;
1161 * Write records come in two flavors: immediate and indirect.
1162 * For small writes it's cheaper to store the data with the
1163 * log record (immediate); for large writes it's cheaper to
1164 * sync the data and get a pointer to it (indirect) so that
1165 * we don't have to write the data twice.
1167 if (buf != NULL) { /* immediate write */
1168 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1169 /* test for truncation needs to be done while range locked */
1170 if (offset >= zp->z_size) {
1173 error = dmu_read(os, object, offset, size, buf,
1174 DMU_READ_NO_PREFETCH);
1176 ASSERT(error == 0 || error == ENOENT);
1177 } else { /* indirect write */
1179 * Have to lock the whole block to ensure when it's
1180 * written out and it's checksum is being calculated
1181 * that no one can change the data. We need to re-check
1182 * blocksize after we get the lock in case it's changed!
1187 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1189 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1191 if (zp->z_blksz == size)
1194 zfs_range_unlock(zgd->zgd_rl);
1196 /* test for truncation needs to be done while range locked */
1197 if (lr->lr_offset >= zp->z_size)
1206 error = dmu_buf_hold(os, object, offset, zgd, &db,
1207 DMU_READ_NO_PREFETCH);
1213 ASSERT(db->db_offset == offset);
1214 ASSERT(db->db_size == size);
1216 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1218 ASSERT(error || lr->lr_length <= zp->z_blksz);
1221 * On success, we need to wait for the write I/O
1222 * initiated by dmu_sync() to complete before we can
1223 * release this dbuf. We will finish everything up
1224 * in the zfs_get_done() callback.
1229 if (error == EALREADY) {
1230 lr->lr_common.lrc_txtype = TX_WRITE2;
1236 zfs_get_done(zgd, error);
1243 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1244 caller_context_t *ct)
1246 znode_t *zp = VTOZ(vp);
1247 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1253 if (flag & V_ACE_MASK)
1254 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1256 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1263 * If vnode is for a device return a specfs vnode instead.
1266 specvp_check(vnode_t **vpp, cred_t *cr)
1270 if (IS_DEVVP(*vpp)) {
1273 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1284 * Lookup an entry in a directory, or an extended attribute directory.
1285 * If it exists, return a held vnode reference for it.
1287 * IN: dvp - vnode of directory to search.
1288 * nm - name of entry to lookup.
1289 * pnp - full pathname to lookup [UNUSED].
1290 * flags - LOOKUP_XATTR set if looking for an attribute.
1291 * rdir - root directory vnode [UNUSED].
1292 * cr - credentials of caller.
1293 * ct - caller context
1294 * direntflags - directory lookup flags
1295 * realpnp - returned pathname.
1297 * OUT: vpp - vnode of located entry, NULL if not found.
1299 * RETURN: 0 if success
1300 * error code if failure
1307 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp,
1308 int nameiop, cred_t *cr, kthread_t *td, int flags)
1310 znode_t *zdp = VTOZ(dvp);
1311 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1313 int *direntflags = NULL;
1314 void *realpnp = NULL;
1317 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1319 if (dvp->v_type != VDIR) {
1321 } else if (zdp->z_sa_hdl == NULL) {
1325 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1326 error = zfs_fastaccesschk_execute(zdp, cr);
1334 vnode_t *tvp = dnlc_lookup(dvp, nm);
1337 error = zfs_fastaccesschk_execute(zdp, cr);
1342 if (tvp == DNLC_NO_VNODE) {
1347 return (specvp_check(vpp, cr));
1353 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1360 if (flags & LOOKUP_XATTR) {
1363 * If the xattr property is off, refuse the lookup request.
1365 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1372 * We don't allow recursive attributes..
1373 * Maybe someday we will.
1375 if (zdp->z_pflags & ZFS_XATTR) {
1380 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1386 * Do we have permission to get into attribute directory?
1389 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1399 if (dvp->v_type != VDIR) {
1405 * Check accessibility of directory.
1408 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1413 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1414 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1419 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1421 error = specvp_check(vpp, cr);
1423 /* Translate errors and add SAVENAME when needed. */
1424 if (cnp->cn_flags & ISLASTCN) {
1428 if (error == ENOENT) {
1429 error = EJUSTRETURN;
1430 cnp->cn_flags |= SAVENAME;
1436 cnp->cn_flags |= SAVENAME;
1440 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) {
1443 if (cnp->cn_flags & ISDOTDOT) {
1444 ltype = VOP_ISLOCKED(dvp);
1448 error = zfs_vnode_lock(*vpp, cnp->cn_lkflags);
1449 if (cnp->cn_flags & ISDOTDOT)
1450 vn_lock(dvp, ltype | LK_RETRY);
1460 #ifdef FREEBSD_NAMECACHE
1462 * Insert name into cache (as non-existent) if appropriate.
1464 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
1465 cache_enter(dvp, *vpp, cnp);
1467 * Insert name into cache if appropriate.
1469 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) {
1470 if (!(cnp->cn_flags & ISLASTCN) ||
1471 (nameiop != DELETE && nameiop != RENAME)) {
1472 cache_enter(dvp, *vpp, cnp);
1481 * Attempt to create a new entry in a directory. If the entry
1482 * already exists, truncate the file if permissible, else return
1483 * an error. Return the vp of the created or trunc'd file.
1485 * IN: dvp - vnode of directory to put new file entry in.
1486 * name - name of new file entry.
1487 * vap - attributes of new file.
1488 * excl - flag indicating exclusive or non-exclusive mode.
1489 * mode - mode to open file with.
1490 * cr - credentials of caller.
1491 * flag - large file flag [UNUSED].
1492 * ct - caller context
1493 * vsecp - ACL to be set
1495 * OUT: vpp - vnode of created or trunc'd entry.
1497 * RETURN: 0 if success
1498 * error code if failure
1501 * dvp - ctime|mtime updated if new entry created
1502 * vp - ctime|mtime always, atime if new
1507 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode,
1508 vnode_t **vpp, cred_t *cr, kthread_t *td)
1510 znode_t *zp, *dzp = VTOZ(dvp);
1511 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1519 gid_t gid = crgetgid(cr);
1520 zfs_acl_ids_t acl_ids;
1521 boolean_t fuid_dirtied;
1522 boolean_t have_acl = B_FALSE;
1527 * If we have an ephemeral id, ACL, or XVATTR then
1528 * make sure file system is at proper version
1531 ksid = crgetsid(cr, KSID_OWNER);
1533 uid = ksid_getid(ksid);
1537 if (zfsvfs->z_use_fuids == B_FALSE &&
1538 (vsecp || (vap->va_mask & AT_XVATTR) ||
1539 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1545 zilog = zfsvfs->z_log;
1547 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1548 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1553 if (vap->va_mask & AT_XVATTR) {
1554 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
1555 crgetuid(cr), cr, vap->va_type)) != 0) {
1563 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr))
1564 vap->va_mode &= ~S_ISVTX;
1566 if (*name == '\0') {
1568 * Null component name refers to the directory itself.
1575 /* possible VN_HOLD(zp) */
1578 if (flag & FIGNORECASE)
1581 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1585 zfs_acl_ids_free(&acl_ids);
1586 if (strcmp(name, "..") == 0)
1597 * Create a new file object and update the directory
1600 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1602 zfs_acl_ids_free(&acl_ids);
1607 * We only support the creation of regular files in
1608 * extended attribute directories.
1611 if ((dzp->z_pflags & ZFS_XATTR) &&
1612 (vap->va_type != VREG)) {
1614 zfs_acl_ids_free(&acl_ids);
1619 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1620 cr, vsecp, &acl_ids)) != 0)
1624 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1625 zfs_acl_ids_free(&acl_ids);
1630 tx = dmu_tx_create(os);
1632 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1633 ZFS_SA_BASE_ATTR_SIZE);
1635 fuid_dirtied = zfsvfs->z_fuid_dirty;
1637 zfs_fuid_txhold(zfsvfs, tx);
1638 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1639 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1640 if (!zfsvfs->z_use_sa &&
1641 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1642 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1643 0, acl_ids.z_aclp->z_acl_bytes);
1645 error = dmu_tx_assign(tx, TXG_NOWAIT);
1647 zfs_dirent_unlock(dl);
1648 if (error == ERESTART) {
1653 zfs_acl_ids_free(&acl_ids);
1658 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1661 zfs_fuid_sync(zfsvfs, tx);
1663 (void) zfs_link_create(dl, zp, tx, ZNEW);
1664 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1665 if (flag & FIGNORECASE)
1667 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1668 vsecp, acl_ids.z_fuidp, vap);
1669 zfs_acl_ids_free(&acl_ids);
1672 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1675 zfs_acl_ids_free(&acl_ids);
1679 * A directory entry already exists for this name.
1682 * Can't truncate an existing file if in exclusive mode.
1689 * Can't open a directory for writing.
1691 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1696 * Verify requested access to file.
1698 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1702 mutex_enter(&dzp->z_lock);
1704 mutex_exit(&dzp->z_lock);
1707 * Truncate regular files if requested.
1709 if ((ZTOV(zp)->v_type == VREG) &&
1710 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1711 /* we can't hold any locks when calling zfs_freesp() */
1712 zfs_dirent_unlock(dl);
1714 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1716 vnevent_create(ZTOV(zp), ct);
1722 zfs_dirent_unlock(dl);
1729 error = specvp_check(vpp, cr);
1732 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1733 zil_commit(zilog, 0);
1740 * Remove an entry from a directory.
1742 * IN: dvp - vnode of directory to remove entry from.
1743 * name - name of entry to remove.
1744 * cr - credentials of caller.
1745 * ct - caller context
1746 * flags - case flags
1748 * RETURN: 0 if success
1749 * error code if failure
1753 * vp - ctime (if nlink > 0)
1756 uint64_t null_xattr = 0;
1760 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1763 znode_t *zp, *dzp = VTOZ(dvp);
1766 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1768 uint64_t acl_obj, xattr_obj;
1769 uint64_t xattr_obj_unlinked = 0;
1773 boolean_t may_delete_now, delete_now = FALSE;
1774 boolean_t unlinked, toobig = FALSE;
1776 pathname_t *realnmp = NULL;
1783 zilog = zfsvfs->z_log;
1785 if (flags & FIGNORECASE) {
1795 * Attempt to lock directory; fail if entry doesn't exist.
1797 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1807 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1812 * Need to use rmdir for removing directories.
1814 if (vp->v_type == VDIR) {
1819 vnevent_remove(vp, dvp, name, ct);
1822 dnlc_remove(dvp, realnmp->pn_buf);
1824 dnlc_remove(dvp, name);
1827 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1831 * We may delete the znode now, or we may put it in the unlinked set;
1832 * it depends on whether we're the last link, and on whether there are
1833 * other holds on the vnode. So we dmu_tx_hold() the right things to
1834 * allow for either case.
1837 tx = dmu_tx_create(zfsvfs->z_os);
1838 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1839 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1840 zfs_sa_upgrade_txholds(tx, zp);
1841 zfs_sa_upgrade_txholds(tx, dzp);
1842 if (may_delete_now) {
1844 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1845 /* if the file is too big, only hold_free a token amount */
1846 dmu_tx_hold_free(tx, zp->z_id, 0,
1847 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1850 /* are there any extended attributes? */
1851 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1852 &xattr_obj, sizeof (xattr_obj));
1853 if (error == 0 && xattr_obj) {
1854 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1856 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1857 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1860 mutex_enter(&zp->z_lock);
1861 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1862 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1863 mutex_exit(&zp->z_lock);
1865 /* charge as an update -- would be nice not to charge at all */
1866 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1868 error = dmu_tx_assign(tx, TXG_NOWAIT);
1870 zfs_dirent_unlock(dl);
1874 if (error == ERESTART) {
1887 * Remove the directory entry.
1889 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1899 * Hold z_lock so that we can make sure that the ACL obj
1900 * hasn't changed. Could have been deleted due to
1903 mutex_enter(&zp->z_lock);
1905 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1906 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1907 delete_now = may_delete_now && !toobig &&
1908 vp->v_count == 1 && !vn_has_cached_data(vp) &&
1909 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1916 panic("zfs_remove: delete_now branch taken");
1918 if (xattr_obj_unlinked) {
1919 ASSERT3U(xzp->z_links, ==, 2);
1920 mutex_enter(&xzp->z_lock);
1921 xzp->z_unlinked = 1;
1923 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1924 &xzp->z_links, sizeof (xzp->z_links), tx);
1925 ASSERT3U(error, ==, 0);
1926 mutex_exit(&xzp->z_lock);
1927 zfs_unlinked_add(xzp, tx);
1930 error = sa_remove(zp->z_sa_hdl,
1931 SA_ZPL_XATTR(zfsvfs), tx);
1933 error = sa_update(zp->z_sa_hdl,
1934 SA_ZPL_XATTR(zfsvfs), &null_xattr,
1935 sizeof (uint64_t), tx);
1940 ASSERT0(vp->v_count);
1942 mutex_exit(&zp->z_lock);
1943 zfs_znode_delete(zp, tx);
1944 } else if (unlinked) {
1945 mutex_exit(&zp->z_lock);
1946 zfs_unlinked_add(zp, tx);
1948 vp->v_vflag |= VV_NOSYNC;
1953 if (flags & FIGNORECASE)
1955 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1962 zfs_dirent_unlock(dl);
1969 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1970 zil_commit(zilog, 0);
1977 * Create a new directory and insert it into dvp using the name
1978 * provided. Return a pointer to the inserted directory.
1980 * IN: dvp - vnode of directory to add subdir to.
1981 * dirname - name of new directory.
1982 * vap - attributes of new directory.
1983 * cr - credentials of caller.
1984 * ct - caller context
1985 * vsecp - ACL to be set
1987 * OUT: vpp - vnode of created directory.
1989 * RETURN: 0 if success
1990 * error code if failure
1993 * dvp - ctime|mtime updated
1994 * vp - ctime|mtime|atime updated
1998 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
1999 caller_context_t *ct, int flags, vsecattr_t *vsecp)
2001 znode_t *zp, *dzp = VTOZ(dvp);
2002 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2011 gid_t gid = crgetgid(cr);
2012 zfs_acl_ids_t acl_ids;
2013 boolean_t fuid_dirtied;
2015 ASSERT(vap->va_type == VDIR);
2018 * If we have an ephemeral id, ACL, or XVATTR then
2019 * make sure file system is at proper version
2022 ksid = crgetsid(cr, KSID_OWNER);
2024 uid = ksid_getid(ksid);
2027 if (zfsvfs->z_use_fuids == B_FALSE &&
2028 (vsecp || (vap->va_mask & AT_XVATTR) ||
2029 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
2034 zilog = zfsvfs->z_log;
2036 if (dzp->z_pflags & ZFS_XATTR) {
2041 if (zfsvfs->z_utf8 && u8_validate(dirname,
2042 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2046 if (flags & FIGNORECASE)
2049 if (vap->va_mask & AT_XVATTR) {
2050 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
2051 crgetuid(cr), cr, vap->va_type)) != 0) {
2057 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
2058 vsecp, &acl_ids)) != 0) {
2063 * First make sure the new directory doesn't exist.
2065 * Existence is checked first to make sure we don't return
2066 * EACCES instead of EEXIST which can cause some applications
2072 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
2074 zfs_acl_ids_free(&acl_ids);
2079 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
2080 zfs_acl_ids_free(&acl_ids);
2081 zfs_dirent_unlock(dl);
2086 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
2087 zfs_acl_ids_free(&acl_ids);
2088 zfs_dirent_unlock(dl);
2094 * Add a new entry to the directory.
2096 tx = dmu_tx_create(zfsvfs->z_os);
2097 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2098 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2099 fuid_dirtied = zfsvfs->z_fuid_dirty;
2101 zfs_fuid_txhold(zfsvfs, tx);
2102 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2103 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2104 acl_ids.z_aclp->z_acl_bytes);
2107 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2108 ZFS_SA_BASE_ATTR_SIZE);
2110 error = dmu_tx_assign(tx, TXG_NOWAIT);
2112 zfs_dirent_unlock(dl);
2113 if (error == ERESTART) {
2118 zfs_acl_ids_free(&acl_ids);
2127 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2130 zfs_fuid_sync(zfsvfs, tx);
2133 * Now put new name in parent dir.
2135 (void) zfs_link_create(dl, zp, tx, ZNEW);
2139 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2140 if (flags & FIGNORECASE)
2142 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2143 acl_ids.z_fuidp, vap);
2145 zfs_acl_ids_free(&acl_ids);
2149 zfs_dirent_unlock(dl);
2151 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2152 zil_commit(zilog, 0);
2159 * Remove a directory subdir entry. If the current working
2160 * directory is the same as the subdir to be removed, the
2163 * IN: dvp - vnode of directory to remove from.
2164 * name - name of directory to be removed.
2165 * cwd - vnode of current working directory.
2166 * cr - credentials of caller.
2167 * ct - caller context
2168 * flags - case flags
2170 * RETURN: 0 if success
2171 * error code if failure
2174 * dvp - ctime|mtime updated
2178 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2179 caller_context_t *ct, int flags)
2181 znode_t *dzp = VTOZ(dvp);
2184 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2193 zilog = zfsvfs->z_log;
2195 if (flags & FIGNORECASE)
2201 * Attempt to lock directory; fail if entry doesn't exist.
2203 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2211 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2215 if (vp->v_type != VDIR) {
2225 vnevent_rmdir(vp, dvp, name, ct);
2228 * Grab a lock on the directory to make sure that noone is
2229 * trying to add (or lookup) entries while we are removing it.
2231 rw_enter(&zp->z_name_lock, RW_WRITER);
2234 * Grab a lock on the parent pointer to make sure we play well
2235 * with the treewalk and directory rename code.
2237 rw_enter(&zp->z_parent_lock, RW_WRITER);
2239 tx = dmu_tx_create(zfsvfs->z_os);
2240 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2241 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2242 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2243 zfs_sa_upgrade_txholds(tx, zp);
2244 zfs_sa_upgrade_txholds(tx, dzp);
2245 error = dmu_tx_assign(tx, TXG_NOWAIT);
2247 rw_exit(&zp->z_parent_lock);
2248 rw_exit(&zp->z_name_lock);
2249 zfs_dirent_unlock(dl);
2251 if (error == ERESTART) {
2261 #ifdef FREEBSD_NAMECACHE
2265 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2268 uint64_t txtype = TX_RMDIR;
2269 if (flags & FIGNORECASE)
2271 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2276 rw_exit(&zp->z_parent_lock);
2277 rw_exit(&zp->z_name_lock);
2278 #ifdef FREEBSD_NAMECACHE
2282 zfs_dirent_unlock(dl);
2286 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2287 zil_commit(zilog, 0);
2294 * Read as many directory entries as will fit into the provided
2295 * buffer from the given directory cursor position (specified in
2296 * the uio structure.
2298 * IN: vp - vnode of directory to read.
2299 * uio - structure supplying read location, range info,
2300 * and return buffer.
2301 * cr - credentials of caller.
2302 * ct - caller context
2303 * flags - case flags
2305 * OUT: uio - updated offset and range, buffer filled.
2306 * eofp - set to true if end-of-file detected.
2308 * RETURN: 0 if success
2309 * error code if failure
2312 * vp - atime updated
2314 * Note that the low 4 bits of the cookie returned by zap is always zero.
2315 * This allows us to use the low range for "special" directory entries:
2316 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2317 * we use the offset 2 for the '.zfs' directory.
2321 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies)
2323 znode_t *zp = VTOZ(vp);
2327 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2332 zap_attribute_t zap;
2333 uint_t bytes_wanted;
2334 uint64_t offset; /* must be unsigned; checks for < 1 */
2340 boolean_t check_sysattrs;
2343 u_long *cooks = NULL;
2349 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2350 &parent, sizeof (parent))) != 0) {
2356 * If we are not given an eof variable,
2363 * Check for valid iov_len.
2365 if (uio->uio_iov->iov_len <= 0) {
2371 * Quit if directory has been removed (posix)
2373 if ((*eofp = zp->z_unlinked) != 0) {
2380 offset = uio->uio_loffset;
2381 prefetch = zp->z_zn_prefetch;
2384 * Initialize the iterator cursor.
2388 * Start iteration from the beginning of the directory.
2390 zap_cursor_init(&zc, os, zp->z_id);
2393 * The offset is a serialized cursor.
2395 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2399 * Get space to change directory entries into fs independent format.
2401 iovp = uio->uio_iov;
2402 bytes_wanted = iovp->iov_len;
2403 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2404 bufsize = bytes_wanted;
2405 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2406 odp = (struct dirent64 *)outbuf;
2408 bufsize = bytes_wanted;
2409 odp = (struct dirent64 *)iovp->iov_base;
2411 eodp = (struct edirent *)odp;
2413 if (ncookies != NULL) {
2415 * Minimum entry size is dirent size and 1 byte for a file name.
2417 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1);
2418 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK);
2423 * If this VFS supports the system attribute view interface; and
2424 * we're looking at an extended attribute directory; and we care
2425 * about normalization conflicts on this vfs; then we must check
2426 * for normalization conflicts with the sysattr name space.
2429 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2430 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2431 (flags & V_RDDIR_ENTFLAGS);
2437 * Transform to file-system independent format
2440 while (outcount < bytes_wanted) {
2443 off64_t *next = NULL;
2446 * Special case `.', `..', and `.zfs'.
2449 (void) strcpy(zap.za_name, ".");
2450 zap.za_normalization_conflict = 0;
2453 } else if (offset == 1) {
2454 (void) strcpy(zap.za_name, "..");
2455 zap.za_normalization_conflict = 0;
2458 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2459 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2460 zap.za_normalization_conflict = 0;
2461 objnum = ZFSCTL_INO_ROOT;
2467 if (error = zap_cursor_retrieve(&zc, &zap)) {
2468 if ((*eofp = (error == ENOENT)) != 0)
2474 if (zap.za_integer_length != 8 ||
2475 zap.za_num_integers != 1) {
2476 cmn_err(CE_WARN, "zap_readdir: bad directory "
2477 "entry, obj = %lld, offset = %lld\n",
2478 (u_longlong_t)zp->z_id,
2479 (u_longlong_t)offset);
2484 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2486 * MacOS X can extract the object type here such as:
2487 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2489 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2491 if (check_sysattrs && !zap.za_normalization_conflict) {
2493 zap.za_normalization_conflict =
2494 xattr_sysattr_casechk(zap.za_name);
2496 panic("%s:%u: TODO", __func__, __LINE__);
2501 if (flags & V_RDDIR_ACCFILTER) {
2503 * If we have no access at all, don't include
2504 * this entry in the returned information
2507 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2509 if (!zfs_has_access(ezp, cr)) {
2516 if (flags & V_RDDIR_ENTFLAGS)
2517 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2519 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2522 * Will this entry fit in the buffer?
2524 if (outcount + reclen > bufsize) {
2526 * Did we manage to fit anything in the buffer?
2534 if (flags & V_RDDIR_ENTFLAGS) {
2536 * Add extended flag entry:
2538 eodp->ed_ino = objnum;
2539 eodp->ed_reclen = reclen;
2540 /* NOTE: ed_off is the offset for the *next* entry */
2541 next = &(eodp->ed_off);
2542 eodp->ed_eflags = zap.za_normalization_conflict ?
2543 ED_CASE_CONFLICT : 0;
2544 (void) strncpy(eodp->ed_name, zap.za_name,
2545 EDIRENT_NAMELEN(reclen));
2546 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2551 odp->d_ino = objnum;
2552 odp->d_reclen = reclen;
2553 odp->d_namlen = strlen(zap.za_name);
2554 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1);
2556 odp = (dirent64_t *)((intptr_t)odp + reclen);
2560 ASSERT(outcount <= bufsize);
2562 /* Prefetch znode */
2564 dmu_prefetch(os, objnum, 0, 0);
2568 * Move to the next entry, fill in the previous offset.
2570 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2571 zap_cursor_advance(&zc);
2572 offset = zap_cursor_serialize(&zc);
2577 if (cooks != NULL) {
2580 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks));
2583 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2585 /* Subtract unused cookies */
2586 if (ncookies != NULL)
2587 *ncookies -= ncooks;
2589 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2590 iovp->iov_base += outcount;
2591 iovp->iov_len -= outcount;
2592 uio->uio_resid -= outcount;
2593 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2595 * Reset the pointer.
2597 offset = uio->uio_loffset;
2601 zap_cursor_fini(&zc);
2602 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2603 kmem_free(outbuf, bufsize);
2605 if (error == ENOENT)
2608 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2610 uio->uio_loffset = offset;
2612 if (error != 0 && cookies != NULL) {
2613 free(*cookies, M_TEMP);
2620 ulong_t zfs_fsync_sync_cnt = 4;
2623 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2625 znode_t *zp = VTOZ(vp);
2626 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2628 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2630 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2633 zil_commit(zfsvfs->z_log, zp->z_id);
2641 * Get the requested file attributes and place them in the provided
2644 * IN: vp - vnode of file.
2645 * vap - va_mask identifies requested attributes.
2646 * If AT_XVATTR set, then optional attrs are requested
2647 * flags - ATTR_NOACLCHECK (CIFS server context)
2648 * cr - credentials of caller.
2649 * ct - caller context
2651 * OUT: vap - attribute values.
2653 * RETURN: 0 (always succeeds)
2657 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2658 caller_context_t *ct)
2660 znode_t *zp = VTOZ(vp);
2661 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2664 u_longlong_t nblocks;
2666 uint64_t mtime[2], ctime[2], crtime[2], rdev;
2667 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2668 xoptattr_t *xoap = NULL;
2669 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2670 sa_bulk_attr_t bulk[4];
2676 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2678 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2679 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2680 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &crtime, 16);
2681 if (vp->v_type == VBLK || vp->v_type == VCHR)
2682 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
2685 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2691 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2692 * Also, if we are the owner don't bother, since owner should
2693 * always be allowed to read basic attributes of file.
2695 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2696 (vap->va_uid != crgetuid(cr))) {
2697 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2705 * Return all attributes. It's cheaper to provide the answer
2706 * than to determine whether we were asked the question.
2709 mutex_enter(&zp->z_lock);
2710 vap->va_type = IFTOVT(zp->z_mode);
2711 vap->va_mode = zp->z_mode & ~S_IFMT;
2713 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2715 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
2717 vap->va_nodeid = zp->z_id;
2718 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2719 links = zp->z_links + 1;
2721 links = zp->z_links;
2722 vap->va_nlink = MIN(links, UINT32_MAX); /* nlink_t limit! */
2723 vap->va_size = zp->z_size;
2725 vap->va_rdev = vp->v_rdev;
2727 if (vp->v_type == VBLK || vp->v_type == VCHR)
2728 vap->va_rdev = zfs_cmpldev(rdev);
2730 vap->va_seq = zp->z_seq;
2731 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */
2734 * Add in any requested optional attributes and the create time.
2735 * Also set the corresponding bits in the returned attribute bitmap.
2737 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2738 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2740 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2741 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2744 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2745 xoap->xoa_readonly =
2746 ((zp->z_pflags & ZFS_READONLY) != 0);
2747 XVA_SET_RTN(xvap, XAT_READONLY);
2750 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2752 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2753 XVA_SET_RTN(xvap, XAT_SYSTEM);
2756 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2758 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2759 XVA_SET_RTN(xvap, XAT_HIDDEN);
2762 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2763 xoap->xoa_nounlink =
2764 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2765 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2768 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2769 xoap->xoa_immutable =
2770 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2771 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2774 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2775 xoap->xoa_appendonly =
2776 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2777 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2780 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2782 ((zp->z_pflags & ZFS_NODUMP) != 0);
2783 XVA_SET_RTN(xvap, XAT_NODUMP);
2786 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2788 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2789 XVA_SET_RTN(xvap, XAT_OPAQUE);
2792 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2793 xoap->xoa_av_quarantined =
2794 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2795 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2798 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2799 xoap->xoa_av_modified =
2800 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2801 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2804 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2805 vp->v_type == VREG) {
2806 zfs_sa_get_scanstamp(zp, xvap);
2809 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2812 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2813 times, sizeof (times));
2814 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2815 XVA_SET_RTN(xvap, XAT_CREATETIME);
2818 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2819 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2820 XVA_SET_RTN(xvap, XAT_REPARSE);
2822 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2823 xoap->xoa_generation = zp->z_gen;
2824 XVA_SET_RTN(xvap, XAT_GEN);
2827 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2829 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2830 XVA_SET_RTN(xvap, XAT_OFFLINE);
2833 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2835 ((zp->z_pflags & ZFS_SPARSE) != 0);
2836 XVA_SET_RTN(xvap, XAT_SPARSE);
2840 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2841 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2842 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2843 ZFS_TIME_DECODE(&vap->va_birthtime, crtime);
2845 mutex_exit(&zp->z_lock);
2847 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2848 vap->va_blksize = blksize;
2849 vap->va_bytes = nblocks << 9; /* nblocks * 512 */
2851 if (zp->z_blksz == 0) {
2853 * Block size hasn't been set; suggest maximal I/O transfers.
2855 vap->va_blksize = zfsvfs->z_max_blksz;
2863 * Set the file attributes to the values contained in the
2866 * IN: vp - vnode of file to be modified.
2867 * vap - new attribute values.
2868 * If AT_XVATTR set, then optional attrs are being set
2869 * flags - ATTR_UTIME set if non-default time values provided.
2870 * - ATTR_NOACLCHECK (CIFS context only).
2871 * cr - credentials of caller.
2872 * ct - caller context
2874 * RETURN: 0 if success
2875 * error code if failure
2878 * vp - ctime updated, mtime updated if size changed.
2882 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2883 caller_context_t *ct)
2885 znode_t *zp = VTOZ(vp);
2886 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2891 uint_t mask = vap->va_mask;
2893 uint64_t saved_mode;
2896 uint64_t new_uid, new_gid;
2898 uint64_t mtime[2], ctime[2];
2900 int need_policy = FALSE;
2902 zfs_fuid_info_t *fuidp = NULL;
2903 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2906 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2907 boolean_t fuid_dirtied = B_FALSE;
2908 sa_bulk_attr_t bulk[7], xattr_bulk[7];
2909 int count = 0, xattr_count = 0;
2914 if (mask & AT_NOSET)
2920 zilog = zfsvfs->z_log;
2923 * Make sure that if we have ephemeral uid/gid or xvattr specified
2924 * that file system is at proper version level
2927 if (zfsvfs->z_use_fuids == B_FALSE &&
2928 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2929 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2930 (mask & AT_XVATTR))) {
2935 if (mask & AT_SIZE && vp->v_type == VDIR) {
2940 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2946 * If this is an xvattr_t, then get a pointer to the structure of
2947 * optional attributes. If this is NULL, then we have a vattr_t.
2949 xoap = xva_getxoptattr(xvap);
2951 xva_init(&tmpxvattr);
2954 * Immutable files can only alter immutable bit and atime
2956 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2957 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2958 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2963 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2969 * Verify timestamps doesn't overflow 32 bits.
2970 * ZFS can handle large timestamps, but 32bit syscalls can't
2971 * handle times greater than 2039. This check should be removed
2972 * once large timestamps are fully supported.
2974 if (mask & (AT_ATIME | AT_MTIME)) {
2975 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2976 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2986 /* Can this be moved to before the top label? */
2987 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2993 * First validate permissions
2996 if (mask & AT_SIZE) {
2998 * XXX - Note, we are not providing any open
2999 * mode flags here (like FNDELAY), so we may
3000 * block if there are locks present... this
3001 * should be addressed in openat().
3003 /* XXX - would it be OK to generate a log record here? */
3004 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
3011 if (mask & (AT_ATIME|AT_MTIME) ||
3012 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
3013 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
3014 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
3015 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
3016 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
3017 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
3018 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
3019 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
3023 if (mask & (AT_UID|AT_GID)) {
3024 int idmask = (mask & (AT_UID|AT_GID));
3029 * NOTE: even if a new mode is being set,
3030 * we may clear S_ISUID/S_ISGID bits.
3033 if (!(mask & AT_MODE))
3034 vap->va_mode = zp->z_mode;
3037 * Take ownership or chgrp to group we are a member of
3040 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
3041 take_group = (mask & AT_GID) &&
3042 zfs_groupmember(zfsvfs, vap->va_gid, cr);
3045 * If both AT_UID and AT_GID are set then take_owner and
3046 * take_group must both be set in order to allow taking
3049 * Otherwise, send the check through secpolicy_vnode_setattr()
3053 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
3054 ((idmask == AT_UID) && take_owner) ||
3055 ((idmask == AT_GID) && take_group)) {
3056 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
3057 skipaclchk, cr) == 0) {
3059 * Remove setuid/setgid for non-privileged users
3061 secpolicy_setid_clear(vap, vp, cr);
3062 trim_mask = (mask & (AT_UID|AT_GID));
3071 mutex_enter(&zp->z_lock);
3072 oldva.va_mode = zp->z_mode;
3073 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
3074 if (mask & AT_XVATTR) {
3076 * Update xvattr mask to include only those attributes
3077 * that are actually changing.
3079 * the bits will be restored prior to actually setting
3080 * the attributes so the caller thinks they were set.
3082 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
3083 if (xoap->xoa_appendonly !=
3084 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
3087 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
3088 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
3092 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
3093 if (xoap->xoa_nounlink !=
3094 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
3097 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
3098 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
3102 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
3103 if (xoap->xoa_immutable !=
3104 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
3107 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
3108 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
3112 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
3113 if (xoap->xoa_nodump !=
3114 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
3117 XVA_CLR_REQ(xvap, XAT_NODUMP);
3118 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
3122 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3123 if (xoap->xoa_av_modified !=
3124 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3127 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3128 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
3132 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3133 if ((vp->v_type != VREG &&
3134 xoap->xoa_av_quarantined) ||
3135 xoap->xoa_av_quarantined !=
3136 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3139 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3140 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
3144 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3145 mutex_exit(&zp->z_lock);
3150 if (need_policy == FALSE &&
3151 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3152 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3157 mutex_exit(&zp->z_lock);
3159 if (mask & AT_MODE) {
3160 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3161 err = secpolicy_setid_setsticky_clear(vp, vap,
3167 trim_mask |= AT_MODE;
3175 * If trim_mask is set then take ownership
3176 * has been granted or write_acl is present and user
3177 * has the ability to modify mode. In that case remove
3178 * UID|GID and or MODE from mask so that
3179 * secpolicy_vnode_setattr() doesn't revoke it.
3183 saved_mask = vap->va_mask;
3184 vap->va_mask &= ~trim_mask;
3185 if (trim_mask & AT_MODE) {
3187 * Save the mode, as secpolicy_vnode_setattr()
3188 * will overwrite it with ova.va_mode.
3190 saved_mode = vap->va_mode;
3193 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3194 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3201 vap->va_mask |= saved_mask;
3202 if (trim_mask & AT_MODE) {
3204 * Recover the mode after
3205 * secpolicy_vnode_setattr().
3207 vap->va_mode = saved_mode;
3213 * secpolicy_vnode_setattr, or take ownership may have
3216 mask = vap->va_mask;
3218 if ((mask & (AT_UID | AT_GID))) {
3219 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3220 &xattr_obj, sizeof (xattr_obj));
3222 if (err == 0 && xattr_obj) {
3223 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3227 if (mask & AT_UID) {
3228 new_uid = zfs_fuid_create(zfsvfs,
3229 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3230 if (new_uid != zp->z_uid &&
3231 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3233 VN_RELE(ZTOV(attrzp));
3239 if (mask & AT_GID) {
3240 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3241 cr, ZFS_GROUP, &fuidp);
3242 if (new_gid != zp->z_gid &&
3243 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3245 VN_RELE(ZTOV(attrzp));
3251 tx = dmu_tx_create(zfsvfs->z_os);
3253 if (mask & AT_MODE) {
3254 uint64_t pmode = zp->z_mode;
3256 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3258 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3261 mutex_enter(&zp->z_lock);
3262 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3264 * Are we upgrading ACL from old V0 format
3267 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3268 zfs_znode_acl_version(zp) ==
3269 ZFS_ACL_VERSION_INITIAL) {
3270 dmu_tx_hold_free(tx, acl_obj, 0,
3272 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3273 0, aclp->z_acl_bytes);
3275 dmu_tx_hold_write(tx, acl_obj, 0,
3278 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3279 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3280 0, aclp->z_acl_bytes);
3282 mutex_exit(&zp->z_lock);
3283 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3285 if ((mask & AT_XVATTR) &&
3286 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3287 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3289 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3293 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3296 fuid_dirtied = zfsvfs->z_fuid_dirty;
3298 zfs_fuid_txhold(zfsvfs, tx);
3300 zfs_sa_upgrade_txholds(tx, zp);
3302 err = dmu_tx_assign(tx, TXG_NOWAIT);
3304 if (err == ERESTART)
3311 * Set each attribute requested.
3312 * We group settings according to the locks they need to acquire.
3314 * Note: you cannot set ctime directly, although it will be
3315 * updated as a side-effect of calling this function.
3319 if (mask & (AT_UID|AT_GID|AT_MODE))
3320 mutex_enter(&zp->z_acl_lock);
3321 mutex_enter(&zp->z_lock);
3323 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3324 &zp->z_pflags, sizeof (zp->z_pflags));
3327 if (mask & (AT_UID|AT_GID|AT_MODE))
3328 mutex_enter(&attrzp->z_acl_lock);
3329 mutex_enter(&attrzp->z_lock);
3330 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3331 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3332 sizeof (attrzp->z_pflags));
3335 if (mask & (AT_UID|AT_GID)) {
3337 if (mask & AT_UID) {
3338 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3339 &new_uid, sizeof (new_uid));
3340 zp->z_uid = new_uid;
3342 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3343 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3345 attrzp->z_uid = new_uid;
3349 if (mask & AT_GID) {
3350 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3351 NULL, &new_gid, sizeof (new_gid));
3352 zp->z_gid = new_gid;
3354 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3355 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3357 attrzp->z_gid = new_gid;
3360 if (!(mask & AT_MODE)) {
3361 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3362 NULL, &new_mode, sizeof (new_mode));
3363 new_mode = zp->z_mode;
3365 err = zfs_acl_chown_setattr(zp);
3368 err = zfs_acl_chown_setattr(attrzp);
3373 if (mask & AT_MODE) {
3374 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3375 &new_mode, sizeof (new_mode));
3376 zp->z_mode = new_mode;
3377 ASSERT3U((uintptr_t)aclp, !=, 0);
3378 err = zfs_aclset_common(zp, aclp, cr, tx);
3380 if (zp->z_acl_cached)
3381 zfs_acl_free(zp->z_acl_cached);
3382 zp->z_acl_cached = aclp;
3387 if (mask & AT_ATIME) {
3388 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3389 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3390 &zp->z_atime, sizeof (zp->z_atime));
3393 if (mask & AT_MTIME) {
3394 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3395 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3396 mtime, sizeof (mtime));
3399 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3400 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3401 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3402 NULL, mtime, sizeof (mtime));
3403 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3404 &ctime, sizeof (ctime));
3405 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3407 } else if (mask != 0) {
3408 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3409 &ctime, sizeof (ctime));
3410 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3413 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3414 SA_ZPL_CTIME(zfsvfs), NULL,
3415 &ctime, sizeof (ctime));
3416 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3417 mtime, ctime, B_TRUE);
3421 * Do this after setting timestamps to prevent timestamp
3422 * update from toggling bit
3425 if (xoap && (mask & AT_XVATTR)) {
3428 * restore trimmed off masks
3429 * so that return masks can be set for caller.
3432 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3433 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3435 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3436 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3438 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3439 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3441 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3442 XVA_SET_REQ(xvap, XAT_NODUMP);
3444 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3445 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3447 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3448 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3451 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3452 ASSERT(vp->v_type == VREG);
3454 zfs_xvattr_set(zp, xvap, tx);
3458 zfs_fuid_sync(zfsvfs, tx);
3461 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3463 mutex_exit(&zp->z_lock);
3464 if (mask & (AT_UID|AT_GID|AT_MODE))
3465 mutex_exit(&zp->z_acl_lock);
3468 if (mask & (AT_UID|AT_GID|AT_MODE))
3469 mutex_exit(&attrzp->z_acl_lock);
3470 mutex_exit(&attrzp->z_lock);
3473 if (err == 0 && attrzp) {
3474 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3480 VN_RELE(ZTOV(attrzp));
3485 zfs_fuid_info_free(fuidp);
3491 if (err == ERESTART)
3494 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3499 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3500 zil_commit(zilog, 0);
3506 typedef struct zfs_zlock {
3507 krwlock_t *zl_rwlock; /* lock we acquired */
3508 znode_t *zl_znode; /* znode we held */
3509 struct zfs_zlock *zl_next; /* next in list */
3513 * Drop locks and release vnodes that were held by zfs_rename_lock().
3516 zfs_rename_unlock(zfs_zlock_t **zlpp)
3520 while ((zl = *zlpp) != NULL) {
3521 if (zl->zl_znode != NULL)
3522 VN_RELE(ZTOV(zl->zl_znode));
3523 rw_exit(zl->zl_rwlock);
3524 *zlpp = zl->zl_next;
3525 kmem_free(zl, sizeof (*zl));
3530 * Search back through the directory tree, using the ".." entries.
3531 * Lock each directory in the chain to prevent concurrent renames.
3532 * Fail any attempt to move a directory into one of its own descendants.
3533 * XXX - z_parent_lock can overlap with map or grow locks
3536 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3540 uint64_t rootid = zp->z_zfsvfs->z_root;
3541 uint64_t oidp = zp->z_id;
3542 krwlock_t *rwlp = &szp->z_parent_lock;
3543 krw_t rw = RW_WRITER;
3546 * First pass write-locks szp and compares to zp->z_id.
3547 * Later passes read-lock zp and compare to zp->z_parent.
3550 if (!rw_tryenter(rwlp, rw)) {
3552 * Another thread is renaming in this path.
3553 * Note that if we are a WRITER, we don't have any
3554 * parent_locks held yet.
3556 if (rw == RW_READER && zp->z_id > szp->z_id) {
3558 * Drop our locks and restart
3560 zfs_rename_unlock(&zl);
3564 rwlp = &szp->z_parent_lock;
3569 * Wait for other thread to drop its locks
3575 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3576 zl->zl_rwlock = rwlp;
3577 zl->zl_znode = NULL;
3578 zl->zl_next = *zlpp;
3581 if (oidp == szp->z_id) /* We're a descendant of szp */
3584 if (oidp == rootid) /* We've hit the top */
3587 if (rw == RW_READER) { /* i.e. not the first pass */
3588 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3593 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3594 &oidp, sizeof (oidp));
3595 rwlp = &zp->z_parent_lock;
3598 } while (zp->z_id != sdzp->z_id);
3604 * Move an entry from the provided source directory to the target
3605 * directory. Change the entry name as indicated.
3607 * IN: sdvp - Source directory containing the "old entry".
3608 * snm - Old entry name.
3609 * tdvp - Target directory to contain the "new entry".
3610 * tnm - New entry name.
3611 * cr - credentials of caller.
3612 * ct - caller context
3613 * flags - case flags
3615 * RETURN: 0 if success
3616 * error code if failure
3619 * sdvp,tdvp - ctime|mtime updated
3623 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3624 caller_context_t *ct, int flags)
3626 znode_t *tdzp, *szp, *tzp;
3627 znode_t *sdzp = VTOZ(sdvp);
3628 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs;
3631 zfs_dirlock_t *sdl, *tdl;
3634 int cmp, serr, terr;
3639 ZFS_VERIFY_ZP(sdzp);
3640 zilog = zfsvfs->z_log;
3643 * Make sure we have the real vp for the target directory.
3645 if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3648 if (tdvp->v_vfsp != sdvp->v_vfsp || zfsctl_is_node(tdvp)) {
3654 ZFS_VERIFY_ZP(tdzp);
3655 if (zfsvfs->z_utf8 && u8_validate(tnm,
3656 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3661 if (flags & FIGNORECASE)
3670 * This is to prevent the creation of links into attribute space
3671 * by renaming a linked file into/outof an attribute directory.
3672 * See the comment in zfs_link() for why this is considered bad.
3674 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3680 * Lock source and target directory entries. To prevent deadlock,
3681 * a lock ordering must be defined. We lock the directory with
3682 * the smallest object id first, or if it's a tie, the one with
3683 * the lexically first name.
3685 if (sdzp->z_id < tdzp->z_id) {
3687 } else if (sdzp->z_id > tdzp->z_id) {
3691 * First compare the two name arguments without
3692 * considering any case folding.
3694 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3696 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3697 ASSERT(error == 0 || !zfsvfs->z_utf8);
3700 * POSIX: "If the old argument and the new argument
3701 * both refer to links to the same existing file,
3702 * the rename() function shall return successfully
3703 * and perform no other action."
3709 * If the file system is case-folding, then we may
3710 * have some more checking to do. A case-folding file
3711 * system is either supporting mixed case sensitivity
3712 * access or is completely case-insensitive. Note
3713 * that the file system is always case preserving.
3715 * In mixed sensitivity mode case sensitive behavior
3716 * is the default. FIGNORECASE must be used to
3717 * explicitly request case insensitive behavior.
3719 * If the source and target names provided differ only
3720 * by case (e.g., a request to rename 'tim' to 'Tim'),
3721 * we will treat this as a special case in the
3722 * case-insensitive mode: as long as the source name
3723 * is an exact match, we will allow this to proceed as
3724 * a name-change request.
3726 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3727 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3728 flags & FIGNORECASE)) &&
3729 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3732 * case preserving rename request, require exact
3741 * If the source and destination directories are the same, we should
3742 * grab the z_name_lock of that directory only once.
3746 rw_enter(&sdzp->z_name_lock, RW_READER);
3750 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3751 ZEXISTS | zflg, NULL, NULL);
3752 terr = zfs_dirent_lock(&tdl,
3753 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3755 terr = zfs_dirent_lock(&tdl,
3756 tdzp, tnm, &tzp, zflg, NULL, NULL);
3757 serr = zfs_dirent_lock(&sdl,
3758 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3764 * Source entry invalid or not there.
3767 zfs_dirent_unlock(tdl);
3773 rw_exit(&sdzp->z_name_lock);
3776 * FreeBSD: In OpenSolaris they only check if rename source is
3777 * ".." here, because "." is handled in their lookup. This is
3778 * not the case for FreeBSD, so we check for "." explicitly.
3780 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0)
3786 zfs_dirent_unlock(sdl);
3790 rw_exit(&sdzp->z_name_lock);
3792 if (strcmp(tnm, "..") == 0)
3799 * Must have write access at the source to remove the old entry
3800 * and write access at the target to create the new entry.
3801 * Note that if target and source are the same, this can be
3802 * done in a single check.
3805 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3808 if (ZTOV(szp)->v_type == VDIR) {
3810 * Check to make sure rename is valid.
3811 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3813 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3818 * Does target exist?
3822 * Source and target must be the same type.
3824 if (ZTOV(szp)->v_type == VDIR) {
3825 if (ZTOV(tzp)->v_type != VDIR) {
3830 if (ZTOV(tzp)->v_type == VDIR) {
3836 * POSIX dictates that when the source and target
3837 * entries refer to the same file object, rename
3838 * must do nothing and exit without error.
3840 if (szp->z_id == tzp->z_id) {
3846 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3848 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3851 * notify the target directory if it is not the same
3852 * as source directory.
3855 vnevent_rename_dest_dir(tdvp, ct);
3858 tx = dmu_tx_create(zfsvfs->z_os);
3859 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3860 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3861 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3862 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3864 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3865 zfs_sa_upgrade_txholds(tx, tdzp);
3868 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3869 zfs_sa_upgrade_txholds(tx, tzp);
3872 zfs_sa_upgrade_txholds(tx, szp);
3873 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3874 error = dmu_tx_assign(tx, TXG_NOWAIT);
3877 zfs_rename_unlock(&zl);
3878 zfs_dirent_unlock(sdl);
3879 zfs_dirent_unlock(tdl);
3882 rw_exit(&sdzp->z_name_lock);
3887 if (error == ERESTART) {
3897 if (tzp) /* Attempt to remove the existing target */
3898 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3901 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3903 szp->z_pflags |= ZFS_AV_MODIFIED;
3905 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3906 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3909 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3911 zfs_log_rename(zilog, tx, TX_RENAME |
3912 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3913 sdl->dl_name, tdzp, tdl->dl_name, szp);
3916 * Update path information for the target vnode
3918 vn_renamepath(tdvp, ZTOV(szp), tnm,
3922 * At this point, we have successfully created
3923 * the target name, but have failed to remove
3924 * the source name. Since the create was done
3925 * with the ZRENAMING flag, there are
3926 * complications; for one, the link count is
3927 * wrong. The easiest way to deal with this
3928 * is to remove the newly created target, and
3929 * return the original error. This must
3930 * succeed; fortunately, it is very unlikely to
3931 * fail, since we just created it.
3933 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3934 ZRENAMING, NULL), ==, 0);
3937 #ifdef FREEBSD_NAMECACHE
3948 zfs_rename_unlock(&zl);
3950 zfs_dirent_unlock(sdl);
3951 zfs_dirent_unlock(tdl);
3954 rw_exit(&sdzp->z_name_lock);
3961 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3962 zil_commit(zilog, 0);
3970 * Insert the indicated symbolic reference entry into the directory.
3972 * IN: dvp - Directory to contain new symbolic link.
3973 * link - Name for new symlink entry.
3974 * vap - Attributes of new entry.
3975 * target - Target path of new symlink.
3976 * cr - credentials of caller.
3977 * ct - caller context
3978 * flags - case flags
3980 * RETURN: 0 if success
3981 * error code if failure
3984 * dvp - ctime|mtime updated
3988 zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link,
3989 cred_t *cr, kthread_t *td)
3991 znode_t *zp, *dzp = VTOZ(dvp);
3994 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
3996 uint64_t len = strlen(link);
3999 zfs_acl_ids_t acl_ids;
4000 boolean_t fuid_dirtied;
4001 uint64_t txtype = TX_SYMLINK;
4004 ASSERT(vap->va_type == VLNK);
4008 zilog = zfsvfs->z_log;
4010 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
4011 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4015 if (flags & FIGNORECASE)
4018 if (len > MAXPATHLEN) {
4020 return (ENAMETOOLONG);
4023 if ((error = zfs_acl_ids_create(dzp, 0,
4024 vap, cr, NULL, &acl_ids)) != 0) {
4030 * Attempt to lock directory; fail if entry already exists.
4032 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
4034 zfs_acl_ids_free(&acl_ids);
4039 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4040 zfs_acl_ids_free(&acl_ids);
4041 zfs_dirent_unlock(dl);
4046 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
4047 zfs_acl_ids_free(&acl_ids);
4048 zfs_dirent_unlock(dl);
4052 tx = dmu_tx_create(zfsvfs->z_os);
4053 fuid_dirtied = zfsvfs->z_fuid_dirty;
4054 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
4055 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4056 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
4057 ZFS_SA_BASE_ATTR_SIZE + len);
4058 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
4059 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
4060 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
4061 acl_ids.z_aclp->z_acl_bytes);
4064 zfs_fuid_txhold(zfsvfs, tx);
4065 error = dmu_tx_assign(tx, TXG_NOWAIT);
4067 zfs_dirent_unlock(dl);
4068 if (error == ERESTART) {
4073 zfs_acl_ids_free(&acl_ids);
4080 * Create a new object for the symlink.
4081 * for version 4 ZPL datsets the symlink will be an SA attribute
4083 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
4086 zfs_fuid_sync(zfsvfs, tx);
4088 mutex_enter(&zp->z_lock);
4090 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
4093 zfs_sa_symlink(zp, link, len, tx);
4094 mutex_exit(&zp->z_lock);
4097 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
4098 &zp->z_size, sizeof (zp->z_size), tx);
4100 * Insert the new object into the directory.
4102 (void) zfs_link_create(dl, zp, tx, ZNEW);
4104 if (flags & FIGNORECASE)
4106 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
4109 zfs_acl_ids_free(&acl_ids);
4113 zfs_dirent_unlock(dl);
4115 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4116 zil_commit(zilog, 0);
4123 * Return, in the buffer contained in the provided uio structure,
4124 * the symbolic path referred to by vp.
4126 * IN: vp - vnode of symbolic link.
4127 * uoip - structure to contain the link path.
4128 * cr - credentials of caller.
4129 * ct - caller context
4131 * OUT: uio - structure to contain the link path.
4133 * RETURN: 0 if success
4134 * error code if failure
4137 * vp - atime updated
4141 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
4143 znode_t *zp = VTOZ(vp);
4144 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4150 mutex_enter(&zp->z_lock);
4152 error = sa_lookup_uio(zp->z_sa_hdl,
4153 SA_ZPL_SYMLINK(zfsvfs), uio);
4155 error = zfs_sa_readlink(zp, uio);
4156 mutex_exit(&zp->z_lock);
4158 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4165 * Insert a new entry into directory tdvp referencing svp.
4167 * IN: tdvp - Directory to contain new entry.
4168 * svp - vnode of new entry.
4169 * name - name of new entry.
4170 * cr - credentials of caller.
4171 * ct - caller context
4173 * RETURN: 0 if success
4174 * error code if failure
4177 * tdvp - ctime|mtime updated
4178 * svp - ctime updated
4182 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
4183 caller_context_t *ct, int flags)
4185 znode_t *dzp = VTOZ(tdvp);
4187 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4197 ASSERT(tdvp->v_type == VDIR);
4201 zilog = zfsvfs->z_log;
4203 if (VOP_REALVP(svp, &realvp, ct) == 0)
4207 * POSIX dictates that we return EPERM here.
4208 * Better choices include ENOTSUP or EISDIR.
4210 if (svp->v_type == VDIR) {
4215 if (svp->v_vfsp != tdvp->v_vfsp || zfsctl_is_node(svp)) {
4223 /* Prevent links to .zfs/shares files */
4225 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4226 &parent, sizeof (uint64_t))) != 0) {
4230 if (parent == zfsvfs->z_shares_dir) {
4235 if (zfsvfs->z_utf8 && u8_validate(name,
4236 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4240 if (flags & FIGNORECASE)
4244 * We do not support links between attributes and non-attributes
4245 * because of the potential security risk of creating links
4246 * into "normal" file space in order to circumvent restrictions
4247 * imposed in attribute space.
4249 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4255 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4256 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) {
4261 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4268 * Attempt to lock directory; fail if entry already exists.
4270 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4276 tx = dmu_tx_create(zfsvfs->z_os);
4277 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4278 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4279 zfs_sa_upgrade_txholds(tx, szp);
4280 zfs_sa_upgrade_txholds(tx, dzp);
4281 error = dmu_tx_assign(tx, TXG_NOWAIT);
4283 zfs_dirent_unlock(dl);
4284 if (error == ERESTART) {
4294 error = zfs_link_create(dl, szp, tx, 0);
4297 uint64_t txtype = TX_LINK;
4298 if (flags & FIGNORECASE)
4300 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4305 zfs_dirent_unlock(dl);
4308 vnevent_link(svp, ct);
4311 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4312 zil_commit(zilog, 0);
4320 * zfs_null_putapage() is used when the file system has been force
4321 * unmounted. It just drops the pages.
4325 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4326 size_t *lenp, int flags, cred_t *cr)
4328 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4333 * Push a page out to disk, klustering if possible.
4335 * IN: vp - file to push page to.
4336 * pp - page to push.
4337 * flags - additional flags.
4338 * cr - credentials of caller.
4340 * OUT: offp - start of range pushed.
4341 * lenp - len of range pushed.
4343 * RETURN: 0 if success
4344 * error code if failure
4346 * NOTE: callers must have locked the page to be pushed. On
4347 * exit, the page (and all other pages in the kluster) must be
4352 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4353 size_t *lenp, int flags, cred_t *cr)
4355 znode_t *zp = VTOZ(vp);
4356 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4358 u_offset_t off, koff;
4365 * If our blocksize is bigger than the page size, try to kluster
4366 * multiple pages so that we write a full block (thus avoiding
4367 * a read-modify-write).
4369 if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4370 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4371 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4372 ASSERT(koff <= zp->z_size);
4373 if (koff + klen > zp->z_size)
4374 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4375 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4377 ASSERT3U(btop(len), ==, btopr(len));
4380 * Can't push pages past end-of-file.
4382 if (off >= zp->z_size) {
4383 /* ignore all pages */
4386 } else if (off + len > zp->z_size) {
4387 int npages = btopr(zp->z_size - off);
4390 page_list_break(&pp, &trunc, npages);
4391 /* ignore pages past end of file */
4393 pvn_write_done(trunc, flags);
4394 len = zp->z_size - off;
4397 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4398 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4403 tx = dmu_tx_create(zfsvfs->z_os);
4404 dmu_tx_hold_write(tx, zp->z_id, off, len);
4406 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4407 zfs_sa_upgrade_txholds(tx, zp);
4408 err = dmu_tx_assign(tx, TXG_NOWAIT);
4410 if (err == ERESTART) {
4419 if (zp->z_blksz <= PAGESIZE) {
4420 caddr_t va = zfs_map_page(pp, S_READ);
4421 ASSERT3U(len, <=, PAGESIZE);
4422 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4423 zfs_unmap_page(pp, va);
4425 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4429 uint64_t mtime[2], ctime[2];
4430 sa_bulk_attr_t bulk[3];
4433 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4435 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4437 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4439 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4441 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4446 pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4456 * Copy the portion of the file indicated from pages into the file.
4457 * The pages are stored in a page list attached to the files vnode.
4459 * IN: vp - vnode of file to push page data to.
4460 * off - position in file to put data.
4461 * len - amount of data to write.
4462 * flags - flags to control the operation.
4463 * cr - credentials of caller.
4464 * ct - caller context.
4466 * RETURN: 0 if success
4467 * error code if failure
4470 * vp - ctime|mtime updated
4474 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4475 caller_context_t *ct)
4477 znode_t *zp = VTOZ(vp);
4478 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4490 * Align this request to the file block size in case we kluster.
4491 * XXX - this can result in pretty aggresive locking, which can
4492 * impact simultanious read/write access. One option might be
4493 * to break up long requests (len == 0) into block-by-block
4494 * operations to get narrower locking.
4496 blksz = zp->z_blksz;
4498 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4501 if (len > 0 && ISP2(blksz))
4502 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4508 * Search the entire vp list for pages >= io_off.
4510 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4511 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4514 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4516 if (off > zp->z_size) {
4517 /* past end of file */
4518 zfs_range_unlock(rl);
4523 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4525 for (off = io_off; io_off < off + len; io_off += io_len) {
4526 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4527 pp = page_lookup(vp, io_off,
4528 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4530 pp = page_lookup_nowait(vp, io_off,
4531 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4534 if (pp != NULL && pvn_getdirty(pp, flags)) {
4538 * Found a dirty page to push
4540 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4548 zfs_range_unlock(rl);
4549 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4550 zil_commit(zfsvfs->z_log, zp->z_id);
4558 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4560 znode_t *zp = VTOZ(vp);
4561 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4564 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4565 if (zp->z_sa_hdl == NULL) {
4567 * The fs has been unmounted, or we did a
4568 * suspend/resume and this file no longer exists.
4571 ASSERT(vp->v_count <= 1);
4574 vrecycle(vp, curthread);
4575 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4579 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4580 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4582 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4583 zfs_sa_upgrade_txholds(tx, zp);
4584 error = dmu_tx_assign(tx, TXG_WAIT);
4588 mutex_enter(&zp->z_lock);
4589 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4590 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4591 zp->z_atime_dirty = 0;
4592 mutex_exit(&zp->z_lock);
4598 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4603 * Bounds-check the seek operation.
4605 * IN: vp - vnode seeking within
4606 * ooff - old file offset
4607 * noffp - pointer to new file offset
4608 * ct - caller context
4610 * RETURN: 0 if success
4611 * EINVAL if new offset invalid
4615 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4616 caller_context_t *ct)
4618 if (vp->v_type == VDIR)
4620 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4624 * Pre-filter the generic locking function to trap attempts to place
4625 * a mandatory lock on a memory mapped file.
4628 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4629 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4631 znode_t *zp = VTOZ(vp);
4632 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4638 * We are following the UFS semantics with respect to mapcnt
4639 * here: If we see that the file is mapped already, then we will
4640 * return an error, but we don't worry about races between this
4641 * function and zfs_map().
4643 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4648 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4652 * If we can't find a page in the cache, we will create a new page
4653 * and fill it with file data. For efficiency, we may try to fill
4654 * multiple pages at once (klustering) to fill up the supplied page
4655 * list. Note that the pages to be filled are held with an exclusive
4656 * lock to prevent access by other threads while they are being filled.
4659 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4660 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4662 znode_t *zp = VTOZ(vp);
4663 page_t *pp, *cur_pp;
4664 objset_t *os = zp->z_zfsvfs->z_os;
4665 u_offset_t io_off, total;
4669 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4671 * We only have a single page, don't bother klustering
4675 pp = page_create_va(vp, io_off, io_len,
4676 PG_EXCL | PG_WAIT, seg, addr);
4679 * Try to find enough pages to fill the page list
4681 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4682 &io_len, off, plsz, 0);
4686 * The page already exists, nothing to do here.
4693 * Fill the pages in the kluster.
4696 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4699 ASSERT3U(io_off, ==, cur_pp->p_offset);
4700 va = zfs_map_page(cur_pp, S_WRITE);
4701 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4703 zfs_unmap_page(cur_pp, va);
4705 /* On error, toss the entire kluster */
4706 pvn_read_done(pp, B_ERROR);
4707 /* convert checksum errors into IO errors */
4712 cur_pp = cur_pp->p_next;
4716 * Fill in the page list array from the kluster starting
4717 * from the desired offset `off'.
4718 * NOTE: the page list will always be null terminated.
4720 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4721 ASSERT(pl == NULL || (*pl)->p_offset == off);
4727 * Return pointers to the pages for the file region [off, off + len]
4728 * in the pl array. If plsz is greater than len, this function may
4729 * also return page pointers from after the specified region
4730 * (i.e. the region [off, off + plsz]). These additional pages are
4731 * only returned if they are already in the cache, or were created as
4732 * part of a klustered read.
4734 * IN: vp - vnode of file to get data from.
4735 * off - position in file to get data from.
4736 * len - amount of data to retrieve.
4737 * plsz - length of provided page list.
4738 * seg - segment to obtain pages for.
4739 * addr - virtual address of fault.
4740 * rw - mode of created pages.
4741 * cr - credentials of caller.
4742 * ct - caller context.
4744 * OUT: protp - protection mode of created pages.
4745 * pl - list of pages created.
4747 * RETURN: 0 if success
4748 * error code if failure
4751 * vp - atime updated
4755 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4756 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4757 enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4759 znode_t *zp = VTOZ(vp);
4760 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4764 /* we do our own caching, faultahead is unnecessary */
4767 else if (len > plsz)
4770 len = P2ROUNDUP(len, PAGESIZE);
4771 ASSERT(plsz >= len);
4780 * Loop through the requested range [off, off + len) looking
4781 * for pages. If we don't find a page, we will need to create
4782 * a new page and fill it with data from the file.
4785 if (*pl = page_lookup(vp, off, SE_SHARED))
4787 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4790 ASSERT3U((*pl)->p_offset, ==, off);
4794 ASSERT3U(len, >=, PAGESIZE);
4797 ASSERT3U(plsz, >=, PAGESIZE);
4804 * Fill out the page array with any pages already in the cache.
4807 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4814 * Release any pages we have previously locked.
4819 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4829 * Request a memory map for a section of a file. This code interacts
4830 * with common code and the VM system as follows:
4832 * common code calls mmap(), which ends up in smmap_common()
4834 * this calls VOP_MAP(), which takes you into (say) zfs
4836 * zfs_map() calls as_map(), passing segvn_create() as the callback
4838 * segvn_create() creates the new segment and calls VOP_ADDMAP()
4840 * zfs_addmap() updates z_mapcnt
4844 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4845 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4846 caller_context_t *ct)
4848 znode_t *zp = VTOZ(vp);
4849 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4850 segvn_crargs_t vn_a;
4856 if ((prot & PROT_WRITE) && (zp->z_pflags &
4857 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4862 if ((prot & (PROT_READ | PROT_EXEC)) &&
4863 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4868 if (vp->v_flag & VNOMAP) {
4873 if (off < 0 || len > MAXOFFSET_T - off) {
4878 if (vp->v_type != VREG) {
4884 * If file is locked, disallow mapping.
4886 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
4892 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4900 vn_a.offset = (u_offset_t)off;
4901 vn_a.type = flags & MAP_TYPE;
4903 vn_a.maxprot = maxprot;
4906 vn_a.flags = flags & ~MAP_TYPE;
4908 vn_a.lgrp_mem_policy_flags = 0;
4910 error = as_map(as, *addrp, len, segvn_create, &vn_a);
4919 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4920 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4921 caller_context_t *ct)
4923 uint64_t pages = btopr(len);
4925 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
4930 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4931 * more accurate mtime for the associated file. Since we don't have a way of
4932 * detecting when the data was actually modified, we have to resort to
4933 * heuristics. If an explicit msync() is done, then we mark the mtime when the
4934 * last page is pushed. The problem occurs when the msync() call is omitted,
4935 * which by far the most common case:
4943 * putpage() via fsflush
4945 * If we wait until fsflush to come along, we can have a modification time that
4946 * is some arbitrary point in the future. In order to prevent this in the
4947 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
4952 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4953 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
4954 caller_context_t *ct)
4956 uint64_t pages = btopr(len);
4958 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
4959 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
4961 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
4962 vn_has_cached_data(vp))
4963 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
4969 * Free or allocate space in a file. Currently, this function only
4970 * supports the `F_FREESP' command. However, this command is somewhat
4971 * misnamed, as its functionality includes the ability to allocate as
4972 * well as free space.
4974 * IN: vp - vnode of file to free data in.
4975 * cmd - action to take (only F_FREESP supported).
4976 * bfp - section of file to free/alloc.
4977 * flag - current file open mode flags.
4978 * offset - current file offset.
4979 * cr - credentials of caller [UNUSED].
4980 * ct - caller context.
4982 * RETURN: 0 if success
4983 * error code if failure
4986 * vp - ctime|mtime updated
4990 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
4991 offset_t offset, cred_t *cr, caller_context_t *ct)
4993 znode_t *zp = VTOZ(vp);
4994 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5001 if (cmd != F_FREESP) {
5006 if (error = convoff(vp, bfp, 0, offset)) {
5011 if (bfp->l_len < 0) {
5017 len = bfp->l_len; /* 0 means from off to end of file */
5019 error = zfs_freesp(zp, off, len, flag, TRUE);
5026 CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid));
5027 CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid));
5031 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
5033 znode_t *zp = VTOZ(vp);
5034 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5037 uint64_t object = zp->z_id;
5044 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
5045 &gen64, sizeof (uint64_t))) != 0) {
5050 gen = (uint32_t)gen64;
5052 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
5053 fidp->fid_len = size;
5055 zfid = (zfid_short_t *)fidp;
5057 zfid->zf_len = size;
5059 for (i = 0; i < sizeof (zfid->zf_object); i++)
5060 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
5062 /* Must have a non-zero generation number to distinguish from .zfs */
5065 for (i = 0; i < sizeof (zfid->zf_gen); i++)
5066 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
5068 if (size == LONG_FID_LEN) {
5069 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
5072 zlfid = (zfid_long_t *)fidp;
5074 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
5075 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
5077 /* XXX - this should be the generation number for the objset */
5078 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
5079 zlfid->zf_setgen[i] = 0;
5087 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
5088 caller_context_t *ct)
5100 case _PC_FILESIZEBITS:
5104 case _PC_XATTR_EXISTS:
5106 zfsvfs = zp->z_zfsvfs;
5110 error = zfs_dirent_lock(&dl, zp, "", &xzp,
5111 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
5113 zfs_dirent_unlock(dl);
5114 if (!zfs_dirempty(xzp))
5117 } else if (error == ENOENT) {
5119 * If there aren't extended attributes, it's the
5120 * same as having zero of them.
5127 case _PC_SATTR_ENABLED:
5128 case _PC_SATTR_EXISTS:
5129 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
5130 (vp->v_type == VREG || vp->v_type == VDIR);
5133 case _PC_ACCESS_FILTERING:
5134 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
5138 case _PC_ACL_ENABLED:
5139 *valp = _ACL_ACE_ENABLED;
5142 case _PC_MIN_HOLE_SIZE:
5143 *valp = (int)SPA_MINBLOCKSIZE;
5146 case _PC_TIMESTAMP_RESOLUTION:
5147 /* nanosecond timestamp resolution */
5151 case _PC_ACL_EXTENDED:
5159 case _PC_ACL_PATH_MAX:
5160 *valp = ACL_MAX_ENTRIES;
5164 return (EOPNOTSUPP);
5170 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5171 caller_context_t *ct)
5173 znode_t *zp = VTOZ(vp);
5174 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5176 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5180 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5188 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5189 caller_context_t *ct)
5191 znode_t *zp = VTOZ(vp);
5192 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5194 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5195 zilog_t *zilog = zfsvfs->z_log;
5200 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5202 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5203 zil_commit(zilog, 0);
5211 * Tunable, both must be a power of 2.
5213 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
5214 * zcr_blksz_max: if set to less than the file block size, allow loaning out of
5215 * an arcbuf for a partial block read
5217 int zcr_blksz_min = (1 << 10); /* 1K */
5218 int zcr_blksz_max = (1 << 17); /* 128K */
5222 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5223 caller_context_t *ct)
5225 znode_t *zp = VTOZ(vp);
5226 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5227 int max_blksz = zfsvfs->z_max_blksz;
5228 uio_t *uio = &xuio->xu_uio;
5229 ssize_t size = uio->uio_resid;
5230 offset_t offset = uio->uio_loffset;
5235 int preamble, postamble;
5237 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5245 * Loan out an arc_buf for write if write size is bigger than
5246 * max_blksz, and the file's block size is also max_blksz.
5249 if (size < blksz || zp->z_blksz != blksz) {
5254 * Caller requests buffers for write before knowing where the
5255 * write offset might be (e.g. NFS TCP write).
5260 preamble = P2PHASE(offset, blksz);
5262 preamble = blksz - preamble;
5267 postamble = P2PHASE(size, blksz);
5270 fullblk = size / blksz;
5271 (void) dmu_xuio_init(xuio,
5272 (preamble != 0) + fullblk + (postamble != 0));
5273 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5274 int, postamble, int,
5275 (preamble != 0) + fullblk + (postamble != 0));
5278 * Have to fix iov base/len for partial buffers. They
5279 * currently represent full arc_buf's.
5282 /* data begins in the middle of the arc_buf */
5283 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5286 (void) dmu_xuio_add(xuio, abuf,
5287 blksz - preamble, preamble);
5290 for (i = 0; i < fullblk; i++) {
5291 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5294 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5298 /* data ends in the middle of the arc_buf */
5299 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5302 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5307 * Loan out an arc_buf for read if the read size is larger than
5308 * the current file block size. Block alignment is not
5309 * considered. Partial arc_buf will be loaned out for read.
5311 blksz = zp->z_blksz;
5312 if (blksz < zcr_blksz_min)
5313 blksz = zcr_blksz_min;
5314 if (blksz > zcr_blksz_max)
5315 blksz = zcr_blksz_max;
5316 /* avoid potential complexity of dealing with it */
5317 if (blksz > max_blksz) {
5322 maxsize = zp->z_size - uio->uio_loffset;
5326 if (size < blksz || vn_has_cached_data(vp)) {
5336 uio->uio_extflg = UIO_XUIO;
5337 XUIO_XUZC_RW(xuio) = ioflag;
5344 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5348 int ioflag = XUIO_XUZC_RW(xuio);
5350 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5352 i = dmu_xuio_cnt(xuio);
5354 abuf = dmu_xuio_arcbuf(xuio, i);
5356 * if abuf == NULL, it must be a write buffer
5357 * that has been returned in zfs_write().
5360 dmu_return_arcbuf(abuf);
5361 ASSERT(abuf || ioflag == UIO_WRITE);
5364 dmu_xuio_fini(xuio);
5369 * Predeclare these here so that the compiler assumes that
5370 * this is an "old style" function declaration that does
5371 * not include arguments => we won't get type mismatch errors
5372 * in the initializations that follow.
5374 static int zfs_inval();
5375 static int zfs_isdir();
5389 * Directory vnode operations template
5391 vnodeops_t *zfs_dvnodeops;
5392 const fs_operation_def_t zfs_dvnodeops_template[] = {
5393 VOPNAME_OPEN, { .vop_open = zfs_open },
5394 VOPNAME_CLOSE, { .vop_close = zfs_close },
5395 VOPNAME_READ, { .error = zfs_isdir },
5396 VOPNAME_WRITE, { .error = zfs_isdir },
5397 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5398 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5399 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5400 VOPNAME_ACCESS, { .vop_access = zfs_access },
5401 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5402 VOPNAME_CREATE, { .vop_create = zfs_create },
5403 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5404 VOPNAME_LINK, { .vop_link = zfs_link },
5405 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5406 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir },
5407 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5408 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5409 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink },
5410 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5411 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5412 VOPNAME_FID, { .vop_fid = zfs_fid },
5413 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5414 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5415 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5416 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5417 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5422 * Regular file vnode operations template
5424 vnodeops_t *zfs_fvnodeops;
5425 const fs_operation_def_t zfs_fvnodeops_template[] = {
5426 VOPNAME_OPEN, { .vop_open = zfs_open },
5427 VOPNAME_CLOSE, { .vop_close = zfs_close },
5428 VOPNAME_READ, { .vop_read = zfs_read },
5429 VOPNAME_WRITE, { .vop_write = zfs_write },
5430 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5431 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5432 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5433 VOPNAME_ACCESS, { .vop_access = zfs_access },
5434 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5435 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5436 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5437 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5438 VOPNAME_FID, { .vop_fid = zfs_fid },
5439 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5440 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock },
5441 VOPNAME_SPACE, { .vop_space = zfs_space },
5442 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage },
5443 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage },
5444 VOPNAME_MAP, { .vop_map = zfs_map },
5445 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap },
5446 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap },
5447 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5448 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5449 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5450 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5451 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf },
5452 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf },
5457 * Symbolic link vnode operations template
5459 vnodeops_t *zfs_symvnodeops;
5460 const fs_operation_def_t zfs_symvnodeops_template[] = {
5461 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5462 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5463 VOPNAME_ACCESS, { .vop_access = zfs_access },
5464 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5465 VOPNAME_READLINK, { .vop_readlink = zfs_readlink },
5466 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5467 VOPNAME_FID, { .vop_fid = zfs_fid },
5468 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5469 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5474 * special share hidden files vnode operations template
5476 vnodeops_t *zfs_sharevnodeops;
5477 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5478 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5479 VOPNAME_ACCESS, { .vop_access = zfs_access },
5480 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5481 VOPNAME_FID, { .vop_fid = zfs_fid },
5482 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5483 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5484 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5485 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5490 * Extended attribute directory vnode operations template
5491 * This template is identical to the directory vnodes
5492 * operation template except for restricted operations:
5495 * Note that there are other restrictions embedded in:
5496 * zfs_create() - restrict type to VREG
5497 * zfs_link() - no links into/out of attribute space
5498 * zfs_rename() - no moves into/out of attribute space
5500 vnodeops_t *zfs_xdvnodeops;
5501 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5502 VOPNAME_OPEN, { .vop_open = zfs_open },
5503 VOPNAME_CLOSE, { .vop_close = zfs_close },
5504 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5505 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5506 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5507 VOPNAME_ACCESS, { .vop_access = zfs_access },
5508 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5509 VOPNAME_CREATE, { .vop_create = zfs_create },
5510 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5511 VOPNAME_LINK, { .vop_link = zfs_link },
5512 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5513 VOPNAME_MKDIR, { .error = zfs_inval },
5514 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5515 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5516 VOPNAME_SYMLINK, { .error = zfs_inval },
5517 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5518 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5519 VOPNAME_FID, { .vop_fid = zfs_fid },
5520 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5521 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5522 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5523 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5524 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5529 * Error vnode operations template
5531 vnodeops_t *zfs_evnodeops;
5532 const fs_operation_def_t zfs_evnodeops_template[] = {
5533 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5534 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5540 ioflags(int ioflags)
5544 if (ioflags & IO_APPEND)
5546 if (ioflags & IO_NDELAY)
5548 if (ioflags & IO_SYNC)
5549 flags |= (FSYNC | FDSYNC | FRSYNC);
5555 zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage)
5557 znode_t *zp = VTOZ(vp);
5558 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5559 objset_t *os = zp->z_zfsvfs->z_os;
5570 pcount = round_page(count) / PAGE_SIZE;
5572 object = mreq->object;
5575 KASSERT(vp->v_object == object, ("mismatching object"));
5577 VM_OBJECT_LOCK(object);
5578 vm_page_lock_queues();
5579 for (i = 0; i < pcount; i++) {
5584 vm_page_unlock_queues();
5587 if (mreq->valid != VM_PAGE_BITS_ALL)
5588 vm_page_zero_invalid(mreq, TRUE);
5589 VM_OBJECT_UNLOCK(object);
5591 return (VM_PAGER_OK);
5594 PCPU_INC(cnt.v_vnodein);
5595 PCPU_INC(cnt.v_vnodepgsin);
5597 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) {
5598 VM_OBJECT_UNLOCK(object);
5600 return (VM_PAGER_BAD);
5604 if (IDX_TO_OFF(mreq->pindex) + size > object->un_pager.vnp.vnp_size)
5605 size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mreq->pindex);
5607 VM_OBJECT_UNLOCK(object);
5608 va = zfs_map_page(mreq, &sf);
5609 error = dmu_read(os, zp->z_id, IDX_TO_OFF(mreq->pindex),
5610 size, va, DMU_READ_PREFETCH);
5611 if (size != PAGE_SIZE)
5612 bzero(va + size, PAGE_SIZE - size);
5614 VM_OBJECT_LOCK(object);
5617 mreq->valid = VM_PAGE_BITS_ALL;
5618 KASSERT(mreq->dirty == 0, ("zfs_getpages: page %p is dirty", mreq));
5620 VM_OBJECT_UNLOCK(object);
5622 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
5624 return (error ? VM_PAGER_ERROR : VM_PAGER_OK);
5628 zfs_freebsd_getpages(ap)
5629 struct vop_getpages_args /* {
5634 vm_ooffset_t a_offset;
5638 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage));
5642 zfs_freebsd_open(ap)
5643 struct vop_open_args /* {
5646 struct ucred *a_cred;
5647 struct thread *a_td;
5650 vnode_t *vp = ap->a_vp;
5651 znode_t *zp = VTOZ(vp);
5654 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL);
5656 vnode_create_vobject(vp, zp->z_size, ap->a_td);
5661 zfs_freebsd_close(ap)
5662 struct vop_close_args /* {
5665 struct ucred *a_cred;
5666 struct thread *a_td;
5670 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL));
5674 zfs_freebsd_ioctl(ap)
5675 struct vop_ioctl_args /* {
5685 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data,
5686 ap->a_fflag, ap->a_cred, NULL, NULL));
5690 zfs_freebsd_read(ap)
5691 struct vop_read_args /* {
5695 struct ucred *a_cred;
5699 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5704 zfs_freebsd_write(ap)
5705 struct vop_write_args /* {
5709 struct ucred *a_cred;
5713 if (vn_rlimit_fsize(ap->a_vp, ap->a_uio, ap->a_uio->uio_td))
5716 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5721 zfs_freebsd_access(ap)
5722 struct vop_access_args /* {
5724 accmode_t a_accmode;
5725 struct ucred *a_cred;
5726 struct thread *a_td;
5733 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND,
5735 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND);
5737 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL);
5740 * VADMIN has to be handled by vaccess().
5743 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND);
5745 vnode_t *vp = ap->a_vp;
5746 znode_t *zp = VTOZ(vp);
5748 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
5749 zp->z_gid, accmode, ap->a_cred, NULL);
5757 zfs_freebsd_lookup(ap)
5758 struct vop_lookup_args /* {
5759 struct vnode *a_dvp;
5760 struct vnode **a_vpp;
5761 struct componentname *a_cnp;
5764 struct componentname *cnp = ap->a_cnp;
5765 char nm[NAME_MAX + 1];
5767 ASSERT(cnp->cn_namelen < sizeof(nm));
5768 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm)));
5770 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop,
5771 cnp->cn_cred, cnp->cn_thread, 0));
5775 zfs_freebsd_create(ap)
5776 struct vop_create_args /* {
5777 struct vnode *a_dvp;
5778 struct vnode **a_vpp;
5779 struct componentname *a_cnp;
5780 struct vattr *a_vap;
5783 struct componentname *cnp = ap->a_cnp;
5784 vattr_t *vap = ap->a_vap;
5787 ASSERT(cnp->cn_flags & SAVENAME);
5789 vattr_init_mask(vap);
5790 mode = vap->va_mode & ALLPERMS;
5792 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode,
5793 ap->a_vpp, cnp->cn_cred, cnp->cn_thread));
5797 zfs_freebsd_remove(ap)
5798 struct vop_remove_args /* {
5799 struct vnode *a_dvp;
5801 struct componentname *a_cnp;
5805 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
5807 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr,
5808 ap->a_cnp->cn_cred, NULL, 0));
5812 zfs_freebsd_mkdir(ap)
5813 struct vop_mkdir_args /* {
5814 struct vnode *a_dvp;
5815 struct vnode **a_vpp;
5816 struct componentname *a_cnp;
5817 struct vattr *a_vap;
5820 vattr_t *vap = ap->a_vap;
5822 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
5824 vattr_init_mask(vap);
5826 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp,
5827 ap->a_cnp->cn_cred, NULL, 0, NULL));
5831 zfs_freebsd_rmdir(ap)
5832 struct vop_rmdir_args /* {
5833 struct vnode *a_dvp;
5835 struct componentname *a_cnp;
5838 struct componentname *cnp = ap->a_cnp;
5840 ASSERT(cnp->cn_flags & SAVENAME);
5842 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0));
5846 zfs_freebsd_readdir(ap)
5847 struct vop_readdir_args /* {
5850 struct ucred *a_cred;
5857 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag,
5858 ap->a_ncookies, ap->a_cookies));
5862 zfs_freebsd_fsync(ap)
5863 struct vop_fsync_args /* {
5866 struct thread *a_td;
5871 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL));
5875 zfs_freebsd_getattr(ap)
5876 struct vop_getattr_args /* {
5878 struct vattr *a_vap;
5879 struct ucred *a_cred;
5882 vattr_t *vap = ap->a_vap;
5888 xvap.xva_vattr = *vap;
5889 xvap.xva_vattr.va_mask |= AT_XVATTR;
5891 /* Convert chflags into ZFS-type flags. */
5892 /* XXX: what about SF_SETTABLE?. */
5893 XVA_SET_REQ(&xvap, XAT_IMMUTABLE);
5894 XVA_SET_REQ(&xvap, XAT_APPENDONLY);
5895 XVA_SET_REQ(&xvap, XAT_NOUNLINK);
5896 XVA_SET_REQ(&xvap, XAT_NODUMP);
5897 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL);
5901 /* Convert ZFS xattr into chflags. */
5902 #define FLAG_CHECK(fflag, xflag, xfield) do { \
5903 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \
5904 fflags |= (fflag); \
5906 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE,
5907 xvap.xva_xoptattrs.xoa_immutable);
5908 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY,
5909 xvap.xva_xoptattrs.xoa_appendonly);
5910 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK,
5911 xvap.xva_xoptattrs.xoa_nounlink);
5912 FLAG_CHECK(UF_NODUMP, XAT_NODUMP,
5913 xvap.xva_xoptattrs.xoa_nodump);
5915 *vap = xvap.xva_vattr;
5916 vap->va_flags = fflags;
5921 zfs_freebsd_setattr(ap)
5922 struct vop_setattr_args /* {
5924 struct vattr *a_vap;
5925 struct ucred *a_cred;
5928 vnode_t *vp = ap->a_vp;
5929 vattr_t *vap = ap->a_vap;
5930 cred_t *cred = ap->a_cred;
5935 vattr_init_mask(vap);
5936 vap->va_mask &= ~AT_NOSET;
5939 xvap.xva_vattr = *vap;
5941 zflags = VTOZ(vp)->z_pflags;
5943 if (vap->va_flags != VNOVAL) {
5944 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs;
5947 if (zfsvfs->z_use_fuids == B_FALSE)
5948 return (EOPNOTSUPP);
5950 fflags = vap->va_flags;
5951 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_NODUMP)) != 0)
5952 return (EOPNOTSUPP);
5954 * Unprivileged processes are not permitted to unset system
5955 * flags, or modify flags if any system flags are set.
5956 * Privileged non-jail processes may not modify system flags
5957 * if securelevel > 0 and any existing system flags are set.
5958 * Privileged jail processes behave like privileged non-jail
5959 * processes if the security.jail.chflags_allowed sysctl is
5960 * is non-zero; otherwise, they behave like unprivileged
5963 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 ||
5964 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) {
5966 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
5967 error = securelevel_gt(cred, 0);
5973 * Callers may only modify the file flags on objects they
5974 * have VADMIN rights for.
5976 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0)
5979 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
5983 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) {
5988 #define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \
5989 if (((fflags & (fflag)) && !(zflags & (zflag))) || \
5990 ((zflags & (zflag)) && !(fflags & (fflag)))) { \
5991 XVA_SET_REQ(&xvap, (xflag)); \
5992 (xfield) = ((fflags & (fflag)) != 0); \
5995 /* Convert chflags into ZFS-type flags. */
5996 /* XXX: what about SF_SETTABLE?. */
5997 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE,
5998 xvap.xva_xoptattrs.xoa_immutable);
5999 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY,
6000 xvap.xva_xoptattrs.xoa_appendonly);
6001 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK,
6002 xvap.xva_xoptattrs.xoa_nounlink);
6003 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP,
6004 xvap.xva_xoptattrs.xoa_nodump);
6007 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL));
6011 zfs_freebsd_rename(ap)
6012 struct vop_rename_args /* {
6013 struct vnode *a_fdvp;
6014 struct vnode *a_fvp;
6015 struct componentname *a_fcnp;
6016 struct vnode *a_tdvp;
6017 struct vnode *a_tvp;
6018 struct componentname *a_tcnp;
6021 vnode_t *fdvp = ap->a_fdvp;
6022 vnode_t *fvp = ap->a_fvp;
6023 vnode_t *tdvp = ap->a_tdvp;
6024 vnode_t *tvp = ap->a_tvp;
6027 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART));
6028 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART));
6030 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp,
6031 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0);
6046 zfs_freebsd_symlink(ap)
6047 struct vop_symlink_args /* {
6048 struct vnode *a_dvp;
6049 struct vnode **a_vpp;
6050 struct componentname *a_cnp;
6051 struct vattr *a_vap;
6055 struct componentname *cnp = ap->a_cnp;
6056 vattr_t *vap = ap->a_vap;
6058 ASSERT(cnp->cn_flags & SAVENAME);
6060 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */
6061 vattr_init_mask(vap);
6063 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap,
6064 ap->a_target, cnp->cn_cred, cnp->cn_thread));
6068 zfs_freebsd_readlink(ap)
6069 struct vop_readlink_args /* {
6072 struct ucred *a_cred;
6076 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL));
6080 zfs_freebsd_link(ap)
6081 struct vop_link_args /* {
6082 struct vnode *a_tdvp;
6084 struct componentname *a_cnp;
6087 struct componentname *cnp = ap->a_cnp;
6089 ASSERT(cnp->cn_flags & SAVENAME);
6091 return (zfs_link(ap->a_tdvp, ap->a_vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0));
6095 zfs_freebsd_inactive(ap)
6096 struct vop_inactive_args /* {
6098 struct thread *a_td;
6101 vnode_t *vp = ap->a_vp;
6103 zfs_inactive(vp, ap->a_td->td_ucred, NULL);
6108 zfs_reclaim_complete(void *arg, int pending)
6111 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
6113 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
6114 if (zp->z_sa_hdl != NULL) {
6115 ZFS_OBJ_HOLD_ENTER(zfsvfs, zp->z_id);
6116 zfs_znode_dmu_fini(zp);
6117 ZFS_OBJ_HOLD_EXIT(zfsvfs, zp->z_id);
6120 rw_exit(&zfsvfs->z_teardown_inactive_lock);
6122 * If the file system is being unmounted, there is a process waiting
6123 * for us, wake it up.
6125 if (zfsvfs->z_unmounted)
6130 zfs_freebsd_reclaim(ap)
6131 struct vop_reclaim_args /* {
6133 struct thread *a_td;
6136 vnode_t *vp = ap->a_vp;
6137 znode_t *zp = VTOZ(vp);
6138 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
6141 rlocked = rw_tryenter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
6146 * Destroy the vm object and flush associated pages.
6148 vnode_destroy_vobject(vp);
6150 mutex_enter(&zp->z_lock);
6152 mutex_exit(&zp->z_lock);
6154 if (zp->z_unlinked) {
6156 } else if (!rlocked) {
6157 TASK_INIT(&zp->z_task, 0, zfs_reclaim_complete, zp);
6158 taskqueue_enqueue(taskqueue_thread, &zp->z_task);
6159 } else if (zp->z_sa_hdl == NULL) {
6161 } else /* if (!zp->z_unlinked && zp->z_dbuf != NULL) */ {
6164 locked = MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id)) ? 2 :
6165 ZFS_OBJ_HOLD_TRYENTER(zfsvfs, zp->z_id);
6168 * Lock can't be obtained due to deadlock possibility,
6169 * so defer znode destruction.
6171 TASK_INIT(&zp->z_task, 0, zfs_reclaim_complete, zp);
6172 taskqueue_enqueue(taskqueue_thread, &zp->z_task);
6174 zfs_znode_dmu_fini(zp);
6176 ZFS_OBJ_HOLD_EXIT(zfsvfs, zp->z_id);
6182 ASSERT(vp->v_holdcnt >= 1);
6185 rw_exit(&zfsvfs->z_teardown_inactive_lock);
6191 struct vop_fid_args /* {
6197 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL));
6201 zfs_freebsd_pathconf(ap)
6202 struct vop_pathconf_args /* {
6205 register_t *a_retval;
6211 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL);
6213 *ap->a_retval = val;
6214 else if (error == EOPNOTSUPP)
6215 error = vop_stdpathconf(ap);
6220 zfs_freebsd_fifo_pathconf(ap)
6221 struct vop_pathconf_args /* {
6224 register_t *a_retval;
6228 switch (ap->a_name) {
6229 case _PC_ACL_EXTENDED:
6231 case _PC_ACL_PATH_MAX:
6232 case _PC_MAC_PRESENT:
6233 return (zfs_freebsd_pathconf(ap));
6235 return (fifo_specops.vop_pathconf(ap));
6240 * FreeBSD's extended attributes namespace defines file name prefix for ZFS'
6241 * extended attribute name:
6244 * system freebsd:system:
6245 * user (none, can be used to access ZFS fsattr(5) attributes
6246 * created on Solaris)
6249 zfs_create_attrname(int attrnamespace, const char *name, char *attrname,
6252 const char *namespace, *prefix, *suffix;
6254 /* We don't allow '/' character in attribute name. */
6255 if (strchr(name, '/') != NULL)
6257 /* We don't allow attribute names that start with "freebsd:" string. */
6258 if (strncmp(name, "freebsd:", 8) == 0)
6261 bzero(attrname, size);
6263 switch (attrnamespace) {
6264 case EXTATTR_NAMESPACE_USER:
6266 prefix = "freebsd:";
6267 namespace = EXTATTR_NAMESPACE_USER_STRING;
6271 * This is the default namespace by which we can access all
6272 * attributes created on Solaris.
6274 prefix = namespace = suffix = "";
6277 case EXTATTR_NAMESPACE_SYSTEM:
6278 prefix = "freebsd:";
6279 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING;
6282 case EXTATTR_NAMESPACE_EMPTY:
6286 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix,
6288 return (ENAMETOOLONG);
6294 * Vnode operating to retrieve a named extended attribute.
6297 zfs_getextattr(struct vop_getextattr_args *ap)
6300 IN struct vnode *a_vp;
6301 IN int a_attrnamespace;
6302 IN const char *a_name;
6303 INOUT struct uio *a_uio;
6305 IN struct ucred *a_cred;
6306 IN struct thread *a_td;
6310 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6311 struct thread *td = ap->a_td;
6312 struct nameidata nd;
6315 vnode_t *xvp = NULL, *vp;
6318 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6319 ap->a_cred, ap->a_td, VREAD);
6323 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6330 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6338 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_SYSSPACE, attrname,
6340 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL);
6342 NDFREE(&nd, NDF_ONLY_PNBUF);
6345 if (error == ENOENT)
6350 if (ap->a_size != NULL) {
6351 error = VOP_GETATTR(vp, &va, ap->a_cred);
6353 *ap->a_size = (size_t)va.va_size;
6354 } else if (ap->a_uio != NULL)
6355 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6358 vn_close(vp, flags, ap->a_cred, td);
6365 * Vnode operation to remove a named attribute.
6368 zfs_deleteextattr(struct vop_deleteextattr_args *ap)
6371 IN struct vnode *a_vp;
6372 IN int a_attrnamespace;
6373 IN const char *a_name;
6374 IN struct ucred *a_cred;
6375 IN struct thread *a_td;
6379 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6380 struct thread *td = ap->a_td;
6381 struct nameidata nd;
6384 vnode_t *xvp = NULL, *vp;
6387 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6388 ap->a_cred, ap->a_td, VWRITE);
6392 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6399 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6406 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF | MPSAFE,
6407 UIO_SYSSPACE, attrname, xvp, td);
6410 NDFREE(&nd, NDF_ONLY_PNBUF);
6413 if (error == ENOENT)
6417 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
6420 if (vp == nd.ni_dvp)
6430 * Vnode operation to set a named attribute.
6433 zfs_setextattr(struct vop_setextattr_args *ap)
6436 IN struct vnode *a_vp;
6437 IN int a_attrnamespace;
6438 IN const char *a_name;
6439 INOUT struct uio *a_uio;
6440 IN struct ucred *a_cred;
6441 IN struct thread *a_td;
6445 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6446 struct thread *td = ap->a_td;
6447 struct nameidata nd;
6450 vnode_t *xvp = NULL, *vp;
6453 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6454 ap->a_cred, ap->a_td, VWRITE);
6458 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6465 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6466 LOOKUP_XATTR | CREATE_XATTR_DIR);
6472 flags = FFLAGS(O_WRONLY | O_CREAT);
6473 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_SYSSPACE, attrname,
6475 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL);
6477 NDFREE(&nd, NDF_ONLY_PNBUF);
6485 error = VOP_SETATTR(vp, &va, ap->a_cred);
6487 VOP_WRITE(vp, ap->a_uio, IO_UNIT | IO_SYNC, ap->a_cred);
6490 vn_close(vp, flags, ap->a_cred, td);
6497 * Vnode operation to retrieve extended attributes on a vnode.
6500 zfs_listextattr(struct vop_listextattr_args *ap)
6503 IN struct vnode *a_vp;
6504 IN int a_attrnamespace;
6505 INOUT struct uio *a_uio;
6507 IN struct ucred *a_cred;
6508 IN struct thread *a_td;
6512 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6513 struct thread *td = ap->a_td;
6514 struct nameidata nd;
6515 char attrprefix[16];
6516 u_char dirbuf[sizeof(struct dirent)];
6519 struct uio auio, *uio = ap->a_uio;
6520 size_t *sizep = ap->a_size;
6522 vnode_t *xvp = NULL, *vp;
6523 int done, error, eof, pos;
6525 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6526 ap->a_cred, ap->a_td, VREAD);
6530 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix,
6531 sizeof(attrprefix));
6534 plen = strlen(attrprefix);
6541 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6546 * ENOATTR means that the EA directory does not yet exist,
6547 * i.e. there are no extended attributes there.
6549 if (error == ENOATTR)
6554 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED | MPSAFE,
6555 UIO_SYSSPACE, ".", xvp, td);
6558 NDFREE(&nd, NDF_ONLY_PNBUF);
6564 auio.uio_iov = &aiov;
6565 auio.uio_iovcnt = 1;
6566 auio.uio_segflg = UIO_SYSSPACE;
6568 auio.uio_rw = UIO_READ;
6569 auio.uio_offset = 0;
6574 aiov.iov_base = (void *)dirbuf;
6575 aiov.iov_len = sizeof(dirbuf);
6576 auio.uio_resid = sizeof(dirbuf);
6577 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL);
6578 done = sizeof(dirbuf) - auio.uio_resid;
6581 for (pos = 0; pos < done;) {
6582 dp = (struct dirent *)(dirbuf + pos);
6583 pos += dp->d_reclen;
6585 * XXX: Temporarily we also accept DT_UNKNOWN, as this
6586 * is what we get when attribute was created on Solaris.
6588 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN)
6590 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0)
6592 else if (strncmp(dp->d_name, attrprefix, plen) != 0)
6594 nlen = dp->d_namlen - plen;
6597 else if (uio != NULL) {
6599 * Format of extattr name entry is one byte for
6600 * length and the rest for name.
6602 error = uiomove(&nlen, 1, uio->uio_rw, uio);
6604 error = uiomove(dp->d_name + plen, nlen,
6611 } while (!eof && error == 0);
6620 zfs_freebsd_getacl(ap)
6621 struct vop_getacl_args /* {
6630 vsecattr_t vsecattr;
6632 if (ap->a_type != ACL_TYPE_NFS4)
6635 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
6636 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL))
6639 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt);
6640 if (vsecattr.vsa_aclentp != NULL)
6641 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz);
6647 zfs_freebsd_setacl(ap)
6648 struct vop_setacl_args /* {
6657 vsecattr_t vsecattr;
6658 int aclbsize; /* size of acl list in bytes */
6661 if (ap->a_type != ACL_TYPE_NFS4)
6664 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES)
6668 * With NFSv4 ACLs, chmod(2) may need to add additional entries,
6669 * splitting every entry into two and appending "canonical six"
6670 * entries at the end. Don't allow for setting an ACL that would
6671 * cause chmod(2) to run out of ACL entries.
6673 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES)
6676 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR);
6680 vsecattr.vsa_mask = VSA_ACE;
6681 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t);
6682 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP);
6683 aaclp = vsecattr.vsa_aclentp;
6684 vsecattr.vsa_aclentsz = aclbsize;
6686 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp);
6687 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL);
6688 kmem_free(aaclp, aclbsize);
6694 zfs_freebsd_aclcheck(ap)
6695 struct vop_aclcheck_args /* {
6704 return (EOPNOTSUPP);
6707 struct vop_vector zfs_vnodeops;
6708 struct vop_vector zfs_fifoops;
6709 struct vop_vector zfs_shareops;
6711 struct vop_vector zfs_vnodeops = {
6712 .vop_default = &default_vnodeops,
6713 .vop_inactive = zfs_freebsd_inactive,
6714 .vop_reclaim = zfs_freebsd_reclaim,
6715 .vop_access = zfs_freebsd_access,
6716 #ifdef FREEBSD_NAMECACHE
6717 .vop_lookup = vfs_cache_lookup,
6718 .vop_cachedlookup = zfs_freebsd_lookup,
6720 .vop_lookup = zfs_freebsd_lookup,
6722 .vop_getattr = zfs_freebsd_getattr,
6723 .vop_setattr = zfs_freebsd_setattr,
6724 .vop_create = zfs_freebsd_create,
6725 .vop_mknod = zfs_freebsd_create,
6726 .vop_mkdir = zfs_freebsd_mkdir,
6727 .vop_readdir = zfs_freebsd_readdir,
6728 .vop_fsync = zfs_freebsd_fsync,
6729 .vop_open = zfs_freebsd_open,
6730 .vop_close = zfs_freebsd_close,
6731 .vop_rmdir = zfs_freebsd_rmdir,
6732 .vop_ioctl = zfs_freebsd_ioctl,
6733 .vop_link = zfs_freebsd_link,
6734 .vop_symlink = zfs_freebsd_symlink,
6735 .vop_readlink = zfs_freebsd_readlink,
6736 .vop_read = zfs_freebsd_read,
6737 .vop_write = zfs_freebsd_write,
6738 .vop_remove = zfs_freebsd_remove,
6739 .vop_rename = zfs_freebsd_rename,
6740 .vop_pathconf = zfs_freebsd_pathconf,
6741 .vop_bmap = VOP_EOPNOTSUPP,
6742 .vop_fid = zfs_freebsd_fid,
6743 .vop_getextattr = zfs_getextattr,
6744 .vop_deleteextattr = zfs_deleteextattr,
6745 .vop_setextattr = zfs_setextattr,
6746 .vop_listextattr = zfs_listextattr,
6747 .vop_getacl = zfs_freebsd_getacl,
6748 .vop_setacl = zfs_freebsd_setacl,
6749 .vop_aclcheck = zfs_freebsd_aclcheck,
6750 .vop_getpages = zfs_freebsd_getpages,
6753 struct vop_vector zfs_fifoops = {
6754 .vop_default = &fifo_specops,
6755 .vop_fsync = zfs_freebsd_fsync,
6756 .vop_access = zfs_freebsd_access,
6757 .vop_getattr = zfs_freebsd_getattr,
6758 .vop_inactive = zfs_freebsd_inactive,
6759 .vop_read = VOP_PANIC,
6760 .vop_reclaim = zfs_freebsd_reclaim,
6761 .vop_setattr = zfs_freebsd_setattr,
6762 .vop_write = VOP_PANIC,
6763 .vop_pathconf = zfs_freebsd_fifo_pathconf,
6764 .vop_fid = zfs_freebsd_fid,
6765 .vop_getacl = zfs_freebsd_getacl,
6766 .vop_setacl = zfs_freebsd_setacl,
6767 .vop_aclcheck = zfs_freebsd_aclcheck,
6771 * special share hidden files vnode operations template
6773 struct vop_vector zfs_shareops = {
6774 .vop_default = &default_vnodeops,
6775 .vop_access = zfs_freebsd_access,
6776 .vop_inactive = zfs_freebsd_inactive,
6777 .vop_reclaim = zfs_freebsd_reclaim,
6778 .vop_fid = zfs_freebsd_fid,
6779 .vop_pathconf = zfs_freebsd_pathconf,