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) 2013 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) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
109 * dmu_tx_assign(). This is critical because we don't want to block
110 * while holding locks.
112 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
113 * reduces lock contention and CPU usage when we must wait (note that if
114 * throughput is constrained by the storage, nearly every transaction
117 * Note, in particular, that if a lock is sometimes acquired before
118 * the tx assigns, and sometimes after (e.g. z_lock), then failing
119 * to use a non-blocking assign can deadlock the system. The scenario:
121 * Thread A has grabbed a lock before calling dmu_tx_assign().
122 * Thread B is in an already-assigned tx, and blocks for this lock.
123 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
124 * forever, because the previous txg can't quiesce until B's tx commits.
126 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
127 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
128 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
129 * to indicate that this operation has already called dmu_tx_wait().
130 * This will ensure that we don't retry forever, waiting a short bit
133 * (5) If the operation succeeded, generate the intent log entry for it
134 * before dropping locks. This ensures that the ordering of events
135 * in the intent log matches the order in which they actually occurred.
136 * During ZIL replay the zfs_log_* functions will update the sequence
137 * number to indicate the zil transaction has replayed.
139 * (6) At the end of each vnode op, the DMU tx must always commit,
140 * regardless of whether there were any errors.
142 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
143 * to ensure that synchronous semantics are provided when necessary.
145 * In general, this is how things should be ordered in each vnode op:
147 * ZFS_ENTER(zfsvfs); // exit if unmounted
149 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD())
150 * rw_enter(...); // grab any other locks you need
151 * tx = dmu_tx_create(...); // get DMU tx
152 * dmu_tx_hold_*(); // hold each object you might modify
153 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
155 * rw_exit(...); // drop locks
156 * zfs_dirent_unlock(dl); // unlock directory entry
157 * VN_RELE(...); // release held vnodes
158 * if (error == ERESTART) {
164 * dmu_tx_abort(tx); // abort DMU tx
165 * ZFS_EXIT(zfsvfs); // finished in zfs
166 * return (error); // really out of space
168 * error = do_real_work(); // do whatever this VOP does
170 * zfs_log_*(...); // on success, make ZIL entry
171 * dmu_tx_commit(tx); // commit DMU tx -- error or not
172 * rw_exit(...); // drop locks
173 * zfs_dirent_unlock(dl); // unlock directory entry
174 * VN_RELE(...); // release held vnodes
175 * zil_commit(zilog, foid); // synchronous when necessary
176 * ZFS_EXIT(zfsvfs); // finished in zfs
177 * return (error); // done, report error
182 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
184 znode_t *zp = VTOZ(*vpp);
185 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
190 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
191 ((flag & FAPPEND) == 0)) {
193 return (SET_ERROR(EPERM));
196 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
197 ZTOV(zp)->v_type == VREG &&
198 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
199 if (fs_vscan(*vpp, cr, 0) != 0) {
201 return (SET_ERROR(EACCES));
205 /* Keep a count of the synchronous opens in the znode */
206 if (flag & (FSYNC | FDSYNC))
207 atomic_inc_32(&zp->z_sync_cnt);
215 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
216 caller_context_t *ct)
218 znode_t *zp = VTOZ(vp);
219 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
222 * Clean up any locks held by this process on the vp.
224 cleanlocks(vp, ddi_get_pid(), 0);
225 cleanshares(vp, ddi_get_pid());
230 /* Decrement the synchronous opens in the znode */
231 if ((flag & (FSYNC | FDSYNC)) && (count == 1))
232 atomic_dec_32(&zp->z_sync_cnt);
234 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
235 ZTOV(zp)->v_type == VREG &&
236 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
237 VERIFY(fs_vscan(vp, cr, 1) == 0);
244 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
245 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
248 zfs_holey(vnode_t *vp, u_long cmd, offset_t *off)
250 znode_t *zp = VTOZ(vp);
251 uint64_t noff = (uint64_t)*off; /* new offset */
256 file_sz = zp->z_size;
257 if (noff >= file_sz) {
258 return (SET_ERROR(ENXIO));
261 if (cmd == _FIO_SEEK_HOLE)
266 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
269 if ((error == ESRCH) || (noff > file_sz)) {
271 * Handle the virtual hole at the end of file.
277 return (SET_ERROR(ENXIO));
288 zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred,
289 int *rvalp, caller_context_t *ct)
301 * The following two ioctls are used by bfu. Faking out,
302 * necessary to avoid bfu errors.
311 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
312 return (SET_ERROR(EFAULT));
314 off = *(offset_t *)data;
317 zfsvfs = zp->z_zfsvfs;
321 /* offset parameter is in/out */
322 error = zfs_holey(vp, com, &off);
327 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
328 return (SET_ERROR(EFAULT));
330 *(offset_t *)data = off;
334 return (SET_ERROR(ENOTTY));
338 page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes)
345 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE
346 * aligned boundaries, if the range is not aligned. As a result a
347 * DEV_BSIZE subrange with partially dirty data may get marked as clean.
348 * It may happen that all DEV_BSIZE subranges are marked clean and thus
349 * the whole page would be considred clean despite have some dirty data.
350 * For this reason we should shrink the range to DEV_BSIZE aligned
351 * boundaries before calling vm_page_clear_dirty.
353 end = rounddown2(off + nbytes, DEV_BSIZE);
354 off = roundup2(off, DEV_BSIZE);
358 VM_OBJECT_LOCK_ASSERT(obj, MA_OWNED);
361 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
363 if (vm_page_sleep_if_busy(pp, FALSE, "zfsmwb"))
366 pp = vm_page_alloc(obj, OFF_TO_IDX(start),
367 VM_ALLOC_SYSTEM | VM_ALLOC_IFCACHED |
372 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
373 vm_object_pip_add(obj, 1);
374 vm_page_io_start(pp);
375 vm_page_lock_queues();
376 pmap_remove_write(pp);
378 vm_page_clear_dirty(pp, off, nbytes);
379 vm_page_unlock_queues();
387 page_unbusy(vm_page_t pp)
390 vm_page_io_finish(pp);
391 vm_object_pip_subtract(pp->object, 1);
395 page_hold(vnode_t *vp, int64_t start)
401 VM_OBJECT_LOCK_ASSERT(obj, MA_OWNED);
404 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
406 if (vm_page_sleep_if_busy(pp, FALSE, "zfsmwb"))
408 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
409 vm_page_lock_queues();
411 vm_page_unlock_queues();
421 page_unhold(vm_page_t pp)
424 vm_page_lock_queues();
426 vm_page_unlock_queues();
430 zfs_map_page(vm_page_t pp, struct sf_buf **sfp)
433 *sfp = sf_buf_alloc(pp, 0);
434 return ((caddr_t)sf_buf_kva(*sfp));
438 zfs_unmap_page(struct sf_buf *sf)
445 * When a file is memory mapped, we must keep the IO data synchronized
446 * between the DMU cache and the memory mapped pages. What this means:
448 * On Write: If we find a memory mapped page, we write to *both*
449 * the page and the dmu buffer.
452 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid,
453 int segflg, dmu_tx_t *tx)
460 ASSERT(vp->v_mount != NULL);
464 off = start & PAGEOFFSET;
466 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
468 int nbytes = imin(PAGESIZE - off, len);
470 if (segflg == UIO_NOCOPY) {
471 pp = vm_page_lookup(obj, OFF_TO_IDX(start));
473 ("zfs update_pages: NULL page in putpages case"));
475 ("zfs update_pages: unaligned data in putpages case"));
476 KASSERT(pp->valid == VM_PAGE_BITS_ALL,
477 ("zfs update_pages: invalid page in putpages case"));
478 KASSERT(pp->busy > 0,
479 ("zfs update_pages: unbusy page in putpages case"));
480 KASSERT((pp->flags & PG_WRITEABLE) == 0,
481 ("zfs update_pages: writable page in putpages case"));
482 VM_OBJECT_UNLOCK(obj);
484 va = zfs_map_page(pp, &sf);
485 (void) dmu_write(os, oid, start, nbytes, va, tx);
490 } else if ((pp = page_busy(vp, start, off, nbytes)) != NULL) {
491 VM_OBJECT_UNLOCK(obj);
493 va = zfs_map_page(pp, &sf);
494 (void) dmu_read(os, oid, start+off, nbytes,
495 va+off, DMU_READ_PREFETCH);;
504 if (segflg != UIO_NOCOPY)
505 vm_object_pip_wakeupn(obj, 0);
506 VM_OBJECT_UNLOCK(obj);
510 * Read with UIO_NOCOPY flag means that sendfile(2) requests
511 * ZFS to populate a range of page cache pages with data.
513 * NOTE: this function could be optimized to pre-allocate
514 * all pages in advance, drain VPO_BUSY on all of them,
515 * map them into contiguous KVA region and populate them
516 * in one single dmu_read() call.
519 mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio)
521 znode_t *zp = VTOZ(vp);
522 objset_t *os = zp->z_zfsvfs->z_os;
532 ASSERT(uio->uio_segflg == UIO_NOCOPY);
533 ASSERT(vp->v_mount != NULL);
536 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0);
539 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) {
540 int bytes = MIN(PAGESIZE, len);
543 pp = vm_page_lookup(obj, OFF_TO_IDX(start));
544 if (pp != NULL && vm_page_sleep_if_busy(pp, FALSE,
548 pp = vm_page_alloc(obj, OFF_TO_IDX(start),
549 VM_ALLOC_NOBUSY | VM_ALLOC_NORMAL);
551 VM_OBJECT_UNLOCK(obj);
557 if (pp->valid == 0) {
558 vm_page_io_start(pp);
559 VM_OBJECT_UNLOCK(obj);
560 va = zfs_map_page(pp, &sf);
561 error = dmu_read(os, zp->z_id, start, bytes, va,
563 if (bytes != PAGESIZE && error == 0)
564 bzero(va + bytes, PAGESIZE - bytes);
567 vm_page_io_finish(pp);
568 vm_page_lock_queues();
570 if (pp->wire_count == 0 && pp->valid == 0 &&
571 pp->busy == 0 && !(pp->oflags & VPO_BUSY))
574 pp->valid = VM_PAGE_BITS_ALL;
575 vm_page_activate(pp);
577 vm_page_unlock_queues();
579 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
583 uio->uio_resid -= bytes;
584 uio->uio_offset += bytes;
587 VM_OBJECT_UNLOCK(obj);
592 * When a file is memory mapped, we must keep the IO data synchronized
593 * between the DMU cache and the memory mapped pages. What this means:
595 * On Read: We "read" preferentially from memory mapped pages,
596 * else we default from the dmu buffer.
598 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
599 * the file is memory mapped.
602 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
604 znode_t *zp = VTOZ(vp);
605 objset_t *os = zp->z_zfsvfs->z_os;
613 ASSERT(vp->v_mount != NULL);
617 start = uio->uio_loffset;
618 off = start & PAGEOFFSET;
620 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
622 uint64_t bytes = MIN(PAGESIZE - off, len);
624 if (pp = page_hold(vp, start)) {
627 VM_OBJECT_UNLOCK(obj);
628 va = zfs_map_page(pp, &sf);
629 error = uiomove(va + off, bytes, UIO_READ, uio);
634 VM_OBJECT_UNLOCK(obj);
635 error = dmu_read_uio(os, zp->z_id, uio, bytes);
643 VM_OBJECT_UNLOCK(obj);
647 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
650 * Read bytes from specified file into supplied buffer.
652 * IN: vp - vnode of file to be read from.
653 * uio - structure supplying read location, range info,
655 * ioflag - SYNC flags; used to provide FRSYNC semantics.
656 * cr - credentials of caller.
657 * ct - caller context
659 * OUT: uio - updated offset and range, buffer filled.
661 * RETURN: 0 on success, error code on failure.
664 * vp - atime updated if byte count > 0
668 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
670 znode_t *zp = VTOZ(vp);
671 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
682 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
684 return (SET_ERROR(EACCES));
688 * Validate file offset
690 if (uio->uio_loffset < (offset_t)0) {
692 return (SET_ERROR(EINVAL));
696 * Fasttrack empty reads
698 if (uio->uio_resid == 0) {
704 * Check for mandatory locks
706 if (MANDMODE(zp->z_mode)) {
707 if (error = chklock(vp, FREAD,
708 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
715 * If we're in FRSYNC mode, sync out this znode before reading it.
718 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
719 zil_commit(zfsvfs->z_log, zp->z_id);
722 * Lock the range against changes.
724 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
727 * If we are reading past end-of-file we can skip
728 * to the end; but we might still need to set atime.
730 if (uio->uio_loffset >= zp->z_size) {
735 ASSERT(uio->uio_loffset < zp->z_size);
736 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
739 if ((uio->uio_extflg == UIO_XUIO) &&
740 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
742 int blksz = zp->z_blksz;
743 uint64_t offset = uio->uio_loffset;
745 xuio = (xuio_t *)uio;
747 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
750 ASSERT(offset + n <= blksz);
753 (void) dmu_xuio_init(xuio, nblk);
755 if (vn_has_cached_data(vp)) {
757 * For simplicity, we always allocate a full buffer
758 * even if we only expect to read a portion of a block.
760 while (--nblk >= 0) {
761 (void) dmu_xuio_add(xuio,
762 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
770 nbytes = MIN(n, zfs_read_chunk_size -
771 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
774 if (uio->uio_segflg == UIO_NOCOPY)
775 error = mappedread_sf(vp, nbytes, uio);
777 #endif /* __FreeBSD__ */
778 if (vn_has_cached_data(vp))
779 error = mappedread(vp, nbytes, uio);
781 error = dmu_read_uio(os, zp->z_id, uio, nbytes);
783 /* convert checksum errors into IO errors */
785 error = SET_ERROR(EIO);
792 zfs_range_unlock(rl);
794 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
800 * Write the bytes to a file.
802 * IN: vp - vnode of file to be written to.
803 * uio - structure supplying write location, range info,
805 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is
806 * set if in append mode.
807 * cr - credentials of caller.
808 * ct - caller context (NFS/CIFS fem monitor only)
810 * OUT: uio - updated offset and range.
812 * RETURN: 0 on success, error code on failure.
815 * vp - ctime|mtime updated if byte count > 0
820 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
822 znode_t *zp = VTOZ(vp);
823 rlim64_t limit = MAXOFFSET_T;
824 ssize_t start_resid = uio->uio_resid;
828 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
833 int max_blksz = zfsvfs->z_max_blksz;
836 iovec_t *aiov = NULL;
839 int iovcnt = uio->uio_iovcnt;
840 iovec_t *iovp = uio->uio_iov;
843 sa_bulk_attr_t bulk[4];
844 uint64_t mtime[2], ctime[2];
847 * Fasttrack empty write
853 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
859 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
860 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
861 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
863 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
867 * If immutable or not appending then return EPERM
869 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
870 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
871 (uio->uio_loffset < zp->z_size))) {
873 return (SET_ERROR(EPERM));
876 zilog = zfsvfs->z_log;
879 * Validate file offset
881 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
884 return (SET_ERROR(EINVAL));
888 * Check for mandatory locks before calling zfs_range_lock()
889 * in order to prevent a deadlock with locks set via fcntl().
891 if (MANDMODE((mode_t)zp->z_mode) &&
892 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
899 * Pre-fault the pages to ensure slow (eg NFS) pages
901 * Skip this if uio contains loaned arc_buf.
903 if ((uio->uio_extflg == UIO_XUIO) &&
904 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
905 xuio = (xuio_t *)uio;
907 uio_prefaultpages(MIN(n, max_blksz), uio);
911 * If in append mode, set the io offset pointer to eof.
913 if (ioflag & FAPPEND) {
915 * Obtain an appending range lock to guarantee file append
916 * semantics. We reset the write offset once we have the lock.
918 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
920 if (rl->r_len == UINT64_MAX) {
922 * We overlocked the file because this write will cause
923 * the file block size to increase.
924 * Note that zp_size cannot change with this lock held.
928 uio->uio_loffset = woff;
931 * Note that if the file block size will change as a result of
932 * this write, then this range lock will lock the entire file
933 * so that we can re-write the block safely.
935 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
939 zfs_range_unlock(rl);
941 return (SET_ERROR(EFBIG));
944 if ((woff + n) > limit || woff > (limit - n))
947 /* Will this write extend the file length? */
948 write_eof = (woff + n > zp->z_size);
950 end_size = MAX(zp->z_size, woff + n);
953 * Write the file in reasonable size chunks. Each chunk is written
954 * in a separate transaction; this keeps the intent log records small
955 * and allows us to do more fine-grained space accounting.
959 woff = uio->uio_loffset;
960 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
961 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
963 dmu_return_arcbuf(abuf);
964 error = SET_ERROR(EDQUOT);
968 if (xuio && abuf == NULL) {
969 ASSERT(i_iov < iovcnt);
971 abuf = dmu_xuio_arcbuf(xuio, i_iov);
972 dmu_xuio_clear(xuio, i_iov);
973 DTRACE_PROBE3(zfs_cp_write, int, i_iov,
974 iovec_t *, aiov, arc_buf_t *, abuf);
975 ASSERT((aiov->iov_base == abuf->b_data) ||
976 ((char *)aiov->iov_base - (char *)abuf->b_data +
977 aiov->iov_len == arc_buf_size(abuf)));
979 } else if (abuf == NULL && n >= max_blksz &&
980 woff >= zp->z_size &&
981 P2PHASE(woff, max_blksz) == 0 &&
982 zp->z_blksz == max_blksz) {
984 * This write covers a full block. "Borrow" a buffer
985 * from the dmu so that we can fill it before we enter
986 * a transaction. This avoids the possibility of
987 * holding up the transaction if the data copy hangs
988 * up on a pagefault (e.g., from an NFS server mapping).
992 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
994 ASSERT(abuf != NULL);
995 ASSERT(arc_buf_size(abuf) == max_blksz);
996 if (error = uiocopy(abuf->b_data, max_blksz,
997 UIO_WRITE, uio, &cbytes)) {
998 dmu_return_arcbuf(abuf);
1001 ASSERT(cbytes == max_blksz);
1005 * Start a transaction.
1007 tx = dmu_tx_create(zfsvfs->z_os);
1008 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1009 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
1010 zfs_sa_upgrade_txholds(tx, zp);
1011 error = dmu_tx_assign(tx, TXG_WAIT);
1015 dmu_return_arcbuf(abuf);
1020 * If zfs_range_lock() over-locked we grow the blocksize
1021 * and then reduce the lock range. This will only happen
1022 * on the first iteration since zfs_range_reduce() will
1023 * shrink down r_len to the appropriate size.
1025 if (rl->r_len == UINT64_MAX) {
1028 if (zp->z_blksz > max_blksz) {
1029 ASSERT(!ISP2(zp->z_blksz));
1030 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
1032 new_blksz = MIN(end_size, max_blksz);
1034 zfs_grow_blocksize(zp, new_blksz, tx);
1035 zfs_range_reduce(rl, woff, n);
1039 * XXX - should we really limit each write to z_max_blksz?
1040 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
1042 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
1044 if (woff + nbytes > zp->z_size)
1045 vnode_pager_setsize(vp, woff + nbytes);
1048 tx_bytes = uio->uio_resid;
1049 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
1051 tx_bytes -= uio->uio_resid;
1054 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
1056 * If this is not a full block write, but we are
1057 * extending the file past EOF and this data starts
1058 * block-aligned, use assign_arcbuf(). Otherwise,
1059 * write via dmu_write().
1061 if (tx_bytes < max_blksz && (!write_eof ||
1062 aiov->iov_base != abuf->b_data)) {
1064 dmu_write(zfsvfs->z_os, zp->z_id, woff,
1065 aiov->iov_len, aiov->iov_base, tx);
1066 dmu_return_arcbuf(abuf);
1067 xuio_stat_wbuf_copied();
1069 ASSERT(xuio || tx_bytes == max_blksz);
1070 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
1073 ASSERT(tx_bytes <= uio->uio_resid);
1074 uioskip(uio, tx_bytes);
1076 if (tx_bytes && vn_has_cached_data(vp)) {
1077 update_pages(vp, woff, tx_bytes, zfsvfs->z_os,
1078 zp->z_id, uio->uio_segflg, tx);
1082 * If we made no progress, we're done. If we made even
1083 * partial progress, update the znode and ZIL accordingly.
1085 if (tx_bytes == 0) {
1086 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
1087 (void *)&zp->z_size, sizeof (uint64_t), tx);
1094 * Clear Set-UID/Set-GID bits on successful write if not
1095 * privileged and at least one of the excute bits is set.
1097 * It would be nice to to this after all writes have
1098 * been done, but that would still expose the ISUID/ISGID
1099 * to another app after the partial write is committed.
1101 * Note: we don't call zfs_fuid_map_id() here because
1102 * user 0 is not an ephemeral uid.
1104 mutex_enter(&zp->z_acl_lock);
1105 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
1106 (S_IXUSR >> 6))) != 0 &&
1107 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
1108 secpolicy_vnode_setid_retain(vp, cr,
1109 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
1111 zp->z_mode &= ~(S_ISUID | S_ISGID);
1112 newmode = zp->z_mode;
1113 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
1114 (void *)&newmode, sizeof (uint64_t), tx);
1116 mutex_exit(&zp->z_acl_lock);
1118 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
1122 * Update the file size (zp_size) if it has changed;
1123 * account for possible concurrent updates.
1125 while ((end_size = zp->z_size) < uio->uio_loffset) {
1126 (void) atomic_cas_64(&zp->z_size, end_size,
1131 * If we are replaying and eof is non zero then force
1132 * the file size to the specified eof. Note, there's no
1133 * concurrency during replay.
1135 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
1136 zp->z_size = zfsvfs->z_replay_eof;
1138 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1140 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
1145 ASSERT(tx_bytes == nbytes);
1150 uio_prefaultpages(MIN(n, max_blksz), uio);
1154 zfs_range_unlock(rl);
1157 * If we're in replay mode, or we made no progress, return error.
1158 * Otherwise, it's at least a partial write, so it's successful.
1160 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1165 if (ioflag & (FSYNC | FDSYNC) ||
1166 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1167 zil_commit(zilog, zp->z_id);
1174 zfs_get_done(zgd_t *zgd, int error)
1176 znode_t *zp = zgd->zgd_private;
1177 objset_t *os = zp->z_zfsvfs->z_os;
1181 dmu_buf_rele(zgd->zgd_db, zgd);
1183 zfs_range_unlock(zgd->zgd_rl);
1185 vfslocked = VFS_LOCK_GIANT(zp->z_zfsvfs->z_vfs);
1187 * Release the vnode asynchronously as we currently have the
1188 * txg stopped from syncing.
1190 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1192 if (error == 0 && zgd->zgd_bp)
1193 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1195 kmem_free(zgd, sizeof (zgd_t));
1196 VFS_UNLOCK_GIANT(vfslocked);
1200 static int zil_fault_io = 0;
1204 * Get data to generate a TX_WRITE intent log record.
1207 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1209 zfsvfs_t *zfsvfs = arg;
1210 objset_t *os = zfsvfs->z_os;
1212 uint64_t object = lr->lr_foid;
1213 uint64_t offset = lr->lr_offset;
1214 uint64_t size = lr->lr_length;
1215 blkptr_t *bp = &lr->lr_blkptr;
1220 ASSERT(zio != NULL);
1224 * Nothing to do if the file has been removed
1226 if (zfs_zget(zfsvfs, object, &zp) != 0)
1227 return (SET_ERROR(ENOENT));
1228 if (zp->z_unlinked) {
1230 * Release the vnode asynchronously as we currently have the
1231 * txg stopped from syncing.
1233 VN_RELE_ASYNC(ZTOV(zp),
1234 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1235 return (SET_ERROR(ENOENT));
1238 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1239 zgd->zgd_zilog = zfsvfs->z_log;
1240 zgd->zgd_private = zp;
1243 * Write records come in two flavors: immediate and indirect.
1244 * For small writes it's cheaper to store the data with the
1245 * log record (immediate); for large writes it's cheaper to
1246 * sync the data and get a pointer to it (indirect) so that
1247 * we don't have to write the data twice.
1249 if (buf != NULL) { /* immediate write */
1250 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1251 /* test for truncation needs to be done while range locked */
1252 if (offset >= zp->z_size) {
1253 error = SET_ERROR(ENOENT);
1255 error = dmu_read(os, object, offset, size, buf,
1256 DMU_READ_NO_PREFETCH);
1258 ASSERT(error == 0 || error == ENOENT);
1259 } else { /* indirect write */
1261 * Have to lock the whole block to ensure when it's
1262 * written out and it's checksum is being calculated
1263 * that no one can change the data. We need to re-check
1264 * blocksize after we get the lock in case it's changed!
1269 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1271 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1273 if (zp->z_blksz == size)
1276 zfs_range_unlock(zgd->zgd_rl);
1278 /* test for truncation needs to be done while range locked */
1279 if (lr->lr_offset >= zp->z_size)
1280 error = SET_ERROR(ENOENT);
1283 error = SET_ERROR(EIO);
1288 error = dmu_buf_hold(os, object, offset, zgd, &db,
1289 DMU_READ_NO_PREFETCH);
1292 blkptr_t *obp = dmu_buf_get_blkptr(db);
1294 ASSERT(BP_IS_HOLE(bp));
1301 ASSERT(db->db_offset == offset);
1302 ASSERT(db->db_size == size);
1304 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1306 ASSERT(error || lr->lr_length <= zp->z_blksz);
1309 * On success, we need to wait for the write I/O
1310 * initiated by dmu_sync() to complete before we can
1311 * release this dbuf. We will finish everything up
1312 * in the zfs_get_done() callback.
1317 if (error == EALREADY) {
1318 lr->lr_common.lrc_txtype = TX_WRITE2;
1324 zfs_get_done(zgd, error);
1331 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1332 caller_context_t *ct)
1334 znode_t *zp = VTOZ(vp);
1335 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1341 if (flag & V_ACE_MASK)
1342 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1344 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1351 * If vnode is for a device return a specfs vnode instead.
1354 specvp_check(vnode_t **vpp, cred_t *cr)
1358 if (IS_DEVVP(*vpp)) {
1361 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1364 error = SET_ERROR(ENOSYS);
1372 * Lookup an entry in a directory, or an extended attribute directory.
1373 * If it exists, return a held vnode reference for it.
1375 * IN: dvp - vnode of directory to search.
1376 * nm - name of entry to lookup.
1377 * pnp - full pathname to lookup [UNUSED].
1378 * flags - LOOKUP_XATTR set if looking for an attribute.
1379 * rdir - root directory vnode [UNUSED].
1380 * cr - credentials of caller.
1381 * ct - caller context
1382 * direntflags - directory lookup flags
1383 * realpnp - returned pathname.
1385 * OUT: vpp - vnode of located entry, NULL if not found.
1387 * RETURN: 0 on success, error code on failure.
1394 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp,
1395 int nameiop, cred_t *cr, kthread_t *td, int flags)
1397 znode_t *zdp = VTOZ(dvp);
1398 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1400 int *direntflags = NULL;
1401 void *realpnp = NULL;
1404 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1406 if (dvp->v_type != VDIR) {
1407 return (SET_ERROR(ENOTDIR));
1408 } else if (zdp->z_sa_hdl == NULL) {
1409 return (SET_ERROR(EIO));
1412 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1413 error = zfs_fastaccesschk_execute(zdp, cr);
1421 vnode_t *tvp = dnlc_lookup(dvp, nm);
1424 error = zfs_fastaccesschk_execute(zdp, cr);
1429 if (tvp == DNLC_NO_VNODE) {
1431 return (SET_ERROR(ENOENT));
1434 return (specvp_check(vpp, cr));
1440 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1447 if (flags & LOOKUP_XATTR) {
1450 * If the xattr property is off, refuse the lookup request.
1452 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1454 return (SET_ERROR(EINVAL));
1459 * We don't allow recursive attributes..
1460 * Maybe someday we will.
1462 if (zdp->z_pflags & ZFS_XATTR) {
1464 return (SET_ERROR(EINVAL));
1467 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1473 * Do we have permission to get into attribute directory?
1476 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1486 if (dvp->v_type != VDIR) {
1488 return (SET_ERROR(ENOTDIR));
1492 * Check accessibility of directory.
1495 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1500 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1501 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1503 return (SET_ERROR(EILSEQ));
1506 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1508 error = specvp_check(vpp, cr);
1510 /* Translate errors and add SAVENAME when needed. */
1511 if (cnp->cn_flags & ISLASTCN) {
1515 if (error == ENOENT) {
1516 error = EJUSTRETURN;
1517 cnp->cn_flags |= SAVENAME;
1523 cnp->cn_flags |= SAVENAME;
1527 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) {
1530 if (cnp->cn_flags & ISDOTDOT) {
1531 ltype = VOP_ISLOCKED(dvp);
1535 error = vn_lock(*vpp, cnp->cn_lkflags);
1536 if (cnp->cn_flags & ISDOTDOT)
1537 vn_lock(dvp, ltype | LK_RETRY);
1547 #ifdef FREEBSD_NAMECACHE
1549 * Insert name into cache (as non-existent) if appropriate.
1551 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
1552 cache_enter(dvp, *vpp, cnp);
1554 * Insert name into cache if appropriate.
1556 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) {
1557 if (!(cnp->cn_flags & ISLASTCN) ||
1558 (nameiop != DELETE && nameiop != RENAME)) {
1559 cache_enter(dvp, *vpp, cnp);
1568 * Attempt to create a new entry in a directory. If the entry
1569 * already exists, truncate the file if permissible, else return
1570 * an error. Return the vp of the created or trunc'd file.
1572 * IN: dvp - vnode of directory to put new file entry in.
1573 * name - name of new file entry.
1574 * vap - attributes of new file.
1575 * excl - flag indicating exclusive or non-exclusive mode.
1576 * mode - mode to open file with.
1577 * cr - credentials of caller.
1578 * flag - large file flag [UNUSED].
1579 * ct - caller context
1580 * vsecp - ACL to be set
1582 * OUT: vpp - vnode of created or trunc'd entry.
1584 * RETURN: 0 on success, error code on failure.
1587 * dvp - ctime|mtime updated if new entry created
1588 * vp - ctime|mtime always, atime if new
1593 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode,
1594 vnode_t **vpp, cred_t *cr, kthread_t *td)
1596 znode_t *zp, *dzp = VTOZ(dvp);
1597 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1605 gid_t gid = crgetgid(cr);
1606 zfs_acl_ids_t acl_ids;
1607 boolean_t fuid_dirtied;
1608 boolean_t have_acl = B_FALSE;
1609 boolean_t waited = B_FALSE;
1614 * If we have an ephemeral id, ACL, or XVATTR then
1615 * make sure file system is at proper version
1618 ksid = crgetsid(cr, KSID_OWNER);
1620 uid = ksid_getid(ksid);
1624 if (zfsvfs->z_use_fuids == B_FALSE &&
1625 (vsecp || (vap->va_mask & AT_XVATTR) ||
1626 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1627 return (SET_ERROR(EINVAL));
1632 zilog = zfsvfs->z_log;
1634 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1635 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1637 return (SET_ERROR(EILSEQ));
1640 if (vap->va_mask & AT_XVATTR) {
1641 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
1642 crgetuid(cr), cr, vap->va_type)) != 0) {
1648 getnewvnode_reserve(1);
1653 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr))
1654 vap->va_mode &= ~S_ISVTX;
1656 if (*name == '\0') {
1658 * Null component name refers to the directory itself.
1665 /* possible VN_HOLD(zp) */
1668 if (flag & FIGNORECASE)
1671 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1675 zfs_acl_ids_free(&acl_ids);
1676 if (strcmp(name, "..") == 0)
1677 error = SET_ERROR(EISDIR);
1678 getnewvnode_drop_reserve();
1688 * Create a new file object and update the directory
1691 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1693 zfs_acl_ids_free(&acl_ids);
1698 * We only support the creation of regular files in
1699 * extended attribute directories.
1702 if ((dzp->z_pflags & ZFS_XATTR) &&
1703 (vap->va_type != VREG)) {
1705 zfs_acl_ids_free(&acl_ids);
1706 error = SET_ERROR(EINVAL);
1710 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1711 cr, vsecp, &acl_ids)) != 0)
1715 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1716 zfs_acl_ids_free(&acl_ids);
1717 error = SET_ERROR(EDQUOT);
1721 tx = dmu_tx_create(os);
1723 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1724 ZFS_SA_BASE_ATTR_SIZE);
1726 fuid_dirtied = zfsvfs->z_fuid_dirty;
1728 zfs_fuid_txhold(zfsvfs, tx);
1729 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1730 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1731 if (!zfsvfs->z_use_sa &&
1732 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1733 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1734 0, acl_ids.z_aclp->z_acl_bytes);
1736 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1738 zfs_dirent_unlock(dl);
1739 if (error == ERESTART) {
1745 zfs_acl_ids_free(&acl_ids);
1747 getnewvnode_drop_reserve();
1751 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1754 zfs_fuid_sync(zfsvfs, tx);
1756 (void) zfs_link_create(dl, zp, tx, ZNEW);
1757 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1758 if (flag & FIGNORECASE)
1760 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1761 vsecp, acl_ids.z_fuidp, vap);
1762 zfs_acl_ids_free(&acl_ids);
1765 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1768 zfs_acl_ids_free(&acl_ids);
1772 * A directory entry already exists for this name.
1775 * Can't truncate an existing file if in exclusive mode.
1778 error = SET_ERROR(EEXIST);
1782 * Can't open a directory for writing.
1784 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1785 error = SET_ERROR(EISDIR);
1789 * Verify requested access to file.
1791 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1795 mutex_enter(&dzp->z_lock);
1797 mutex_exit(&dzp->z_lock);
1800 * Truncate regular files if requested.
1802 if ((ZTOV(zp)->v_type == VREG) &&
1803 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1804 /* we can't hold any locks when calling zfs_freesp() */
1805 zfs_dirent_unlock(dl);
1807 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1809 vnevent_create(ZTOV(zp), ct);
1814 getnewvnode_drop_reserve();
1816 zfs_dirent_unlock(dl);
1823 error = specvp_check(vpp, cr);
1826 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1827 zil_commit(zilog, 0);
1834 * Remove an entry from a directory.
1836 * IN: dvp - vnode of directory to remove entry from.
1837 * name - name of entry to remove.
1838 * cr - credentials of caller.
1839 * ct - caller context
1840 * flags - case flags
1842 * RETURN: 0 on success, error code on failure.
1846 * vp - ctime (if nlink > 0)
1849 uint64_t null_xattr = 0;
1853 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1856 znode_t *zp, *dzp = VTOZ(dvp);
1859 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1861 uint64_t acl_obj, xattr_obj;
1862 uint64_t xattr_obj_unlinked = 0;
1866 boolean_t may_delete_now, delete_now = FALSE;
1867 boolean_t unlinked, toobig = FALSE;
1869 pathname_t *realnmp = NULL;
1873 boolean_t waited = B_FALSE;
1877 zilog = zfsvfs->z_log;
1879 if (flags & FIGNORECASE) {
1889 * Attempt to lock directory; fail if entry doesn't exist.
1891 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1901 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1906 * Need to use rmdir for removing directories.
1908 if (vp->v_type == VDIR) {
1909 error = SET_ERROR(EPERM);
1913 vnevent_remove(vp, dvp, name, ct);
1916 dnlc_remove(dvp, realnmp->pn_buf);
1918 dnlc_remove(dvp, name);
1921 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1925 * We may delete the znode now, or we may put it in the unlinked set;
1926 * it depends on whether we're the last link, and on whether there are
1927 * other holds on the vnode. So we dmu_tx_hold() the right things to
1928 * allow for either case.
1931 tx = dmu_tx_create(zfsvfs->z_os);
1932 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1933 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1934 zfs_sa_upgrade_txholds(tx, zp);
1935 zfs_sa_upgrade_txholds(tx, dzp);
1936 if (may_delete_now) {
1938 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1939 /* if the file is too big, only hold_free a token amount */
1940 dmu_tx_hold_free(tx, zp->z_id, 0,
1941 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1944 /* are there any extended attributes? */
1945 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1946 &xattr_obj, sizeof (xattr_obj));
1947 if (error == 0 && xattr_obj) {
1948 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1950 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1951 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1954 mutex_enter(&zp->z_lock);
1955 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1956 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1957 mutex_exit(&zp->z_lock);
1959 /* charge as an update -- would be nice not to charge at all */
1960 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1962 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1964 zfs_dirent_unlock(dl);
1968 if (error == ERESTART) {
1982 * Remove the directory entry.
1984 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1994 * Hold z_lock so that we can make sure that the ACL obj
1995 * hasn't changed. Could have been deleted due to
1998 mutex_enter(&zp->z_lock);
2000 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
2001 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
2002 delete_now = may_delete_now && !toobig &&
2003 vp->v_count == 1 && !vn_has_cached_data(vp) &&
2004 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
2011 panic("zfs_remove: delete_now branch taken");
2013 if (xattr_obj_unlinked) {
2014 ASSERT3U(xzp->z_links, ==, 2);
2015 mutex_enter(&xzp->z_lock);
2016 xzp->z_unlinked = 1;
2018 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
2019 &xzp->z_links, sizeof (xzp->z_links), tx);
2020 ASSERT3U(error, ==, 0);
2021 mutex_exit(&xzp->z_lock);
2022 zfs_unlinked_add(xzp, tx);
2025 error = sa_remove(zp->z_sa_hdl,
2026 SA_ZPL_XATTR(zfsvfs), tx);
2028 error = sa_update(zp->z_sa_hdl,
2029 SA_ZPL_XATTR(zfsvfs), &null_xattr,
2030 sizeof (uint64_t), tx);
2035 ASSERT0(vp->v_count);
2037 mutex_exit(&zp->z_lock);
2038 zfs_znode_delete(zp, tx);
2039 } else if (unlinked) {
2040 mutex_exit(&zp->z_lock);
2041 zfs_unlinked_add(zp, tx);
2043 vp->v_vflag |= VV_NOSYNC;
2048 if (flags & FIGNORECASE)
2050 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
2057 zfs_dirent_unlock(dl);
2064 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2065 zil_commit(zilog, 0);
2072 * Create a new directory and insert it into dvp using the name
2073 * provided. Return a pointer to the inserted directory.
2075 * IN: dvp - vnode of directory to add subdir to.
2076 * dirname - name of new directory.
2077 * vap - attributes of new directory.
2078 * cr - credentials of caller.
2079 * ct - caller context
2080 * flags - case flags
2081 * vsecp - ACL to be set
2083 * OUT: vpp - vnode of created directory.
2085 * RETURN: 0 on success, error code on failure.
2088 * dvp - ctime|mtime updated
2089 * vp - ctime|mtime|atime updated
2093 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
2094 caller_context_t *ct, int flags, vsecattr_t *vsecp)
2096 znode_t *zp, *dzp = VTOZ(dvp);
2097 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2106 gid_t gid = crgetgid(cr);
2107 zfs_acl_ids_t acl_ids;
2108 boolean_t fuid_dirtied;
2109 boolean_t waited = B_FALSE;
2111 ASSERT(vap->va_type == VDIR);
2114 * If we have an ephemeral id, ACL, or XVATTR then
2115 * make sure file system is at proper version
2118 ksid = crgetsid(cr, KSID_OWNER);
2120 uid = ksid_getid(ksid);
2123 if (zfsvfs->z_use_fuids == B_FALSE &&
2124 (vsecp || (vap->va_mask & AT_XVATTR) ||
2125 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
2126 return (SET_ERROR(EINVAL));
2130 zilog = zfsvfs->z_log;
2132 if (dzp->z_pflags & ZFS_XATTR) {
2134 return (SET_ERROR(EINVAL));
2137 if (zfsvfs->z_utf8 && u8_validate(dirname,
2138 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2140 return (SET_ERROR(EILSEQ));
2142 if (flags & FIGNORECASE)
2145 if (vap->va_mask & AT_XVATTR) {
2146 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
2147 crgetuid(cr), cr, vap->va_type)) != 0) {
2153 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
2154 vsecp, &acl_ids)) != 0) {
2159 getnewvnode_reserve(1);
2162 * First make sure the new directory doesn't exist.
2164 * Existence is checked first to make sure we don't return
2165 * EACCES instead of EEXIST which can cause some applications
2171 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
2173 zfs_acl_ids_free(&acl_ids);
2174 getnewvnode_drop_reserve();
2179 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
2180 zfs_acl_ids_free(&acl_ids);
2181 zfs_dirent_unlock(dl);
2182 getnewvnode_drop_reserve();
2187 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
2188 zfs_acl_ids_free(&acl_ids);
2189 zfs_dirent_unlock(dl);
2190 getnewvnode_drop_reserve();
2192 return (SET_ERROR(EDQUOT));
2196 * Add a new entry to the directory.
2198 tx = dmu_tx_create(zfsvfs->z_os);
2199 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2200 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2201 fuid_dirtied = zfsvfs->z_fuid_dirty;
2203 zfs_fuid_txhold(zfsvfs, tx);
2204 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2205 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2206 acl_ids.z_aclp->z_acl_bytes);
2209 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2210 ZFS_SA_BASE_ATTR_SIZE);
2212 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2214 zfs_dirent_unlock(dl);
2215 if (error == ERESTART) {
2221 zfs_acl_ids_free(&acl_ids);
2223 getnewvnode_drop_reserve();
2231 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2234 zfs_fuid_sync(zfsvfs, tx);
2237 * Now put new name in parent dir.
2239 (void) zfs_link_create(dl, zp, tx, ZNEW);
2243 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2244 if (flags & FIGNORECASE)
2246 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2247 acl_ids.z_fuidp, vap);
2249 zfs_acl_ids_free(&acl_ids);
2253 getnewvnode_drop_reserve();
2255 zfs_dirent_unlock(dl);
2257 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2258 zil_commit(zilog, 0);
2265 * Remove a directory subdir entry. If the current working
2266 * directory is the same as the subdir to be removed, the
2269 * IN: dvp - vnode of directory to remove from.
2270 * name - name of directory to be removed.
2271 * cwd - vnode of current working directory.
2272 * cr - credentials of caller.
2273 * ct - caller context
2274 * flags - case flags
2276 * RETURN: 0 on success, error code on failure.
2279 * dvp - ctime|mtime updated
2283 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2284 caller_context_t *ct, int flags)
2286 znode_t *dzp = VTOZ(dvp);
2289 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2295 boolean_t waited = B_FALSE;
2299 zilog = zfsvfs->z_log;
2301 if (flags & FIGNORECASE)
2307 * Attempt to lock directory; fail if entry doesn't exist.
2309 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2317 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2321 if (vp->v_type != VDIR) {
2322 error = SET_ERROR(ENOTDIR);
2327 error = SET_ERROR(EINVAL);
2331 vnevent_rmdir(vp, dvp, name, ct);
2334 * Grab a lock on the directory to make sure that noone is
2335 * trying to add (or lookup) entries while we are removing it.
2337 rw_enter(&zp->z_name_lock, RW_WRITER);
2340 * Grab a lock on the parent pointer to make sure we play well
2341 * with the treewalk and directory rename code.
2343 rw_enter(&zp->z_parent_lock, RW_WRITER);
2345 tx = dmu_tx_create(zfsvfs->z_os);
2346 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2347 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2348 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2349 zfs_sa_upgrade_txholds(tx, zp);
2350 zfs_sa_upgrade_txholds(tx, dzp);
2351 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2353 rw_exit(&zp->z_parent_lock);
2354 rw_exit(&zp->z_name_lock);
2355 zfs_dirent_unlock(dl);
2357 if (error == ERESTART) {
2368 #ifdef FREEBSD_NAMECACHE
2372 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2375 uint64_t txtype = TX_RMDIR;
2376 if (flags & FIGNORECASE)
2378 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2383 rw_exit(&zp->z_parent_lock);
2384 rw_exit(&zp->z_name_lock);
2385 #ifdef FREEBSD_NAMECACHE
2389 zfs_dirent_unlock(dl);
2393 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2394 zil_commit(zilog, 0);
2401 * Read as many directory entries as will fit into the provided
2402 * buffer from the given directory cursor position (specified in
2403 * the uio structure).
2405 * IN: vp - vnode of directory to read.
2406 * uio - structure supplying read location, range info,
2407 * and return buffer.
2408 * cr - credentials of caller.
2409 * ct - caller context
2410 * flags - case flags
2412 * OUT: uio - updated offset and range, buffer filled.
2413 * eofp - set to true if end-of-file detected.
2415 * RETURN: 0 on success, error code on failure.
2418 * vp - atime updated
2420 * Note that the low 4 bits of the cookie returned by zap is always zero.
2421 * This allows us to use the low range for "special" directory entries:
2422 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2423 * we use the offset 2 for the '.zfs' directory.
2427 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies)
2429 znode_t *zp = VTOZ(vp);
2433 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2438 zap_attribute_t zap;
2439 uint_t bytes_wanted;
2440 uint64_t offset; /* must be unsigned; checks for < 1 */
2446 boolean_t check_sysattrs;
2449 u_long *cooks = NULL;
2455 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2456 &parent, sizeof (parent))) != 0) {
2462 * If we are not given an eof variable,
2469 * Check for valid iov_len.
2471 if (uio->uio_iov->iov_len <= 0) {
2473 return (SET_ERROR(EINVAL));
2477 * Quit if directory has been removed (posix)
2479 if ((*eofp = zp->z_unlinked) != 0) {
2486 offset = uio->uio_loffset;
2487 prefetch = zp->z_zn_prefetch;
2490 * Initialize the iterator cursor.
2494 * Start iteration from the beginning of the directory.
2496 zap_cursor_init(&zc, os, zp->z_id);
2499 * The offset is a serialized cursor.
2501 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2505 * Get space to change directory entries into fs independent format.
2507 iovp = uio->uio_iov;
2508 bytes_wanted = iovp->iov_len;
2509 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2510 bufsize = bytes_wanted;
2511 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2512 odp = (struct dirent64 *)outbuf;
2514 bufsize = bytes_wanted;
2516 odp = (struct dirent64 *)iovp->iov_base;
2518 eodp = (struct edirent *)odp;
2520 if (ncookies != NULL) {
2522 * Minimum entry size is dirent size and 1 byte for a file name.
2524 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1);
2525 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK);
2530 * If this VFS supports the system attribute view interface; and
2531 * we're looking at an extended attribute directory; and we care
2532 * about normalization conflicts on this vfs; then we must check
2533 * for normalization conflicts with the sysattr name space.
2536 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2537 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2538 (flags & V_RDDIR_ENTFLAGS);
2544 * Transform to file-system independent format
2547 while (outcount < bytes_wanted) {
2550 off64_t *next = NULL;
2553 * Special case `.', `..', and `.zfs'.
2556 (void) strcpy(zap.za_name, ".");
2557 zap.za_normalization_conflict = 0;
2560 } else if (offset == 1) {
2561 (void) strcpy(zap.za_name, "..");
2562 zap.za_normalization_conflict = 0;
2565 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2566 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2567 zap.za_normalization_conflict = 0;
2568 objnum = ZFSCTL_INO_ROOT;
2574 if (error = zap_cursor_retrieve(&zc, &zap)) {
2575 if ((*eofp = (error == ENOENT)) != 0)
2581 if (zap.za_integer_length != 8 ||
2582 zap.za_num_integers != 1) {
2583 cmn_err(CE_WARN, "zap_readdir: bad directory "
2584 "entry, obj = %lld, offset = %lld\n",
2585 (u_longlong_t)zp->z_id,
2586 (u_longlong_t)offset);
2587 error = SET_ERROR(ENXIO);
2591 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2593 * MacOS X can extract the object type here such as:
2594 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2596 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2598 if (check_sysattrs && !zap.za_normalization_conflict) {
2600 zap.za_normalization_conflict =
2601 xattr_sysattr_casechk(zap.za_name);
2603 panic("%s:%u: TODO", __func__, __LINE__);
2608 if (flags & V_RDDIR_ACCFILTER) {
2610 * If we have no access at all, don't include
2611 * this entry in the returned information
2614 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2616 if (!zfs_has_access(ezp, cr)) {
2623 if (flags & V_RDDIR_ENTFLAGS)
2624 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2626 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2629 * Will this entry fit in the buffer?
2631 if (outcount + reclen > bufsize) {
2633 * Did we manage to fit anything in the buffer?
2636 error = SET_ERROR(EINVAL);
2641 if (flags & V_RDDIR_ENTFLAGS) {
2643 * Add extended flag entry:
2645 eodp->ed_ino = objnum;
2646 eodp->ed_reclen = reclen;
2647 /* NOTE: ed_off is the offset for the *next* entry */
2648 next = &(eodp->ed_off);
2649 eodp->ed_eflags = zap.za_normalization_conflict ?
2650 ED_CASE_CONFLICT : 0;
2651 (void) strncpy(eodp->ed_name, zap.za_name,
2652 EDIRENT_NAMELEN(reclen));
2653 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2658 odp->d_ino = objnum;
2659 odp->d_reclen = reclen;
2660 odp->d_namlen = strlen(zap.za_name);
2661 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1);
2663 odp = (dirent64_t *)((intptr_t)odp + reclen);
2667 ASSERT(outcount <= bufsize);
2669 /* Prefetch znode */
2671 dmu_prefetch(os, objnum, 0, 0);
2675 * Move to the next entry, fill in the previous offset.
2677 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2678 zap_cursor_advance(&zc);
2679 offset = zap_cursor_serialize(&zc);
2684 if (cooks != NULL) {
2687 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks));
2690 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2692 /* Subtract unused cookies */
2693 if (ncookies != NULL)
2694 *ncookies -= ncooks;
2696 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2697 iovp->iov_base += outcount;
2698 iovp->iov_len -= outcount;
2699 uio->uio_resid -= outcount;
2700 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2702 * Reset the pointer.
2704 offset = uio->uio_loffset;
2708 zap_cursor_fini(&zc);
2709 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2710 kmem_free(outbuf, bufsize);
2712 if (error == ENOENT)
2715 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2717 uio->uio_loffset = offset;
2719 if (error != 0 && cookies != NULL) {
2720 free(*cookies, M_TEMP);
2727 ulong_t zfs_fsync_sync_cnt = 4;
2730 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2732 znode_t *zp = VTOZ(vp);
2733 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2735 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2737 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2740 zil_commit(zfsvfs->z_log, zp->z_id);
2748 * Get the requested file attributes and place them in the provided
2751 * IN: vp - vnode of file.
2752 * vap - va_mask identifies requested attributes.
2753 * If AT_XVATTR set, then optional attrs are requested
2754 * flags - ATTR_NOACLCHECK (CIFS server context)
2755 * cr - credentials of caller.
2756 * ct - caller context
2758 * OUT: vap - attribute values.
2760 * RETURN: 0 (always succeeds).
2764 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2765 caller_context_t *ct)
2767 znode_t *zp = VTOZ(vp);
2768 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2771 u_longlong_t nblocks;
2773 uint64_t mtime[2], ctime[2], crtime[2], rdev;
2774 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2775 xoptattr_t *xoap = NULL;
2776 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2777 sa_bulk_attr_t bulk[4];
2783 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2785 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2786 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2787 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
2788 if (vp->v_type == VBLK || vp->v_type == VCHR)
2789 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
2792 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2798 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2799 * Also, if we are the owner don't bother, since owner should
2800 * always be allowed to read basic attributes of file.
2802 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2803 (vap->va_uid != crgetuid(cr))) {
2804 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2812 * Return all attributes. It's cheaper to provide the answer
2813 * than to determine whether we were asked the question.
2816 mutex_enter(&zp->z_lock);
2817 vap->va_type = IFTOVT(zp->z_mode);
2818 vap->va_mode = zp->z_mode & ~S_IFMT;
2820 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2822 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
2824 vap->va_nodeid = zp->z_id;
2825 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2826 links = zp->z_links + 1;
2828 links = zp->z_links;
2829 vap->va_nlink = MIN(links, UINT32_MAX); /* nlink_t limit! */
2830 vap->va_size = zp->z_size;
2832 vap->va_rdev = vp->v_rdev;
2834 if (vp->v_type == VBLK || vp->v_type == VCHR)
2835 vap->va_rdev = zfs_cmpldev(rdev);
2837 vap->va_seq = zp->z_seq;
2838 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */
2839 vap->va_filerev = zp->z_seq;
2842 * Add in any requested optional attributes and the create time.
2843 * Also set the corresponding bits in the returned attribute bitmap.
2845 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2846 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2848 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2849 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2852 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2853 xoap->xoa_readonly =
2854 ((zp->z_pflags & ZFS_READONLY) != 0);
2855 XVA_SET_RTN(xvap, XAT_READONLY);
2858 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2860 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2861 XVA_SET_RTN(xvap, XAT_SYSTEM);
2864 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2866 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2867 XVA_SET_RTN(xvap, XAT_HIDDEN);
2870 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2871 xoap->xoa_nounlink =
2872 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2873 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2876 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2877 xoap->xoa_immutable =
2878 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2879 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2882 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2883 xoap->xoa_appendonly =
2884 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2885 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2888 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2890 ((zp->z_pflags & ZFS_NODUMP) != 0);
2891 XVA_SET_RTN(xvap, XAT_NODUMP);
2894 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2896 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2897 XVA_SET_RTN(xvap, XAT_OPAQUE);
2900 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2901 xoap->xoa_av_quarantined =
2902 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2903 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2906 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2907 xoap->xoa_av_modified =
2908 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2909 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2912 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2913 vp->v_type == VREG) {
2914 zfs_sa_get_scanstamp(zp, xvap);
2917 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2920 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2921 times, sizeof (times));
2922 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2923 XVA_SET_RTN(xvap, XAT_CREATETIME);
2926 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2927 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2928 XVA_SET_RTN(xvap, XAT_REPARSE);
2930 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2931 xoap->xoa_generation = zp->z_gen;
2932 XVA_SET_RTN(xvap, XAT_GEN);
2935 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2937 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2938 XVA_SET_RTN(xvap, XAT_OFFLINE);
2941 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2943 ((zp->z_pflags & ZFS_SPARSE) != 0);
2944 XVA_SET_RTN(xvap, XAT_SPARSE);
2948 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2949 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2950 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2951 ZFS_TIME_DECODE(&vap->va_birthtime, crtime);
2953 mutex_exit(&zp->z_lock);
2955 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2956 vap->va_blksize = blksize;
2957 vap->va_bytes = nblocks << 9; /* nblocks * 512 */
2959 if (zp->z_blksz == 0) {
2961 * Block size hasn't been set; suggest maximal I/O transfers.
2963 vap->va_blksize = zfsvfs->z_max_blksz;
2971 * Set the file attributes to the values contained in the
2974 * IN: vp - vnode of file to be modified.
2975 * vap - new attribute values.
2976 * If AT_XVATTR set, then optional attrs are being set
2977 * flags - ATTR_UTIME set if non-default time values provided.
2978 * - ATTR_NOACLCHECK (CIFS context only).
2979 * cr - credentials of caller.
2980 * ct - caller context
2982 * RETURN: 0 on success, error code on failure.
2985 * vp - ctime updated, mtime updated if size changed.
2989 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2990 caller_context_t *ct)
2992 znode_t *zp = VTOZ(vp);
2993 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2998 uint_t mask = vap->va_mask;
2999 uint_t saved_mask = 0;
3000 uint64_t saved_mode;
3003 uint64_t new_uid, new_gid;
3005 uint64_t mtime[2], ctime[2];
3007 int need_policy = FALSE;
3009 zfs_fuid_info_t *fuidp = NULL;
3010 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
3013 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
3014 boolean_t fuid_dirtied = B_FALSE;
3015 sa_bulk_attr_t bulk[7], xattr_bulk[7];
3016 int count = 0, xattr_count = 0;
3021 if (mask & AT_NOSET)
3022 return (SET_ERROR(EINVAL));
3027 zilog = zfsvfs->z_log;
3030 * Make sure that if we have ephemeral uid/gid or xvattr specified
3031 * that file system is at proper version level
3034 if (zfsvfs->z_use_fuids == B_FALSE &&
3035 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
3036 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
3037 (mask & AT_XVATTR))) {
3039 return (SET_ERROR(EINVAL));
3042 if (mask & AT_SIZE && vp->v_type == VDIR) {
3044 return (SET_ERROR(EISDIR));
3047 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
3049 return (SET_ERROR(EINVAL));
3053 * If this is an xvattr_t, then get a pointer to the structure of
3054 * optional attributes. If this is NULL, then we have a vattr_t.
3056 xoap = xva_getxoptattr(xvap);
3058 xva_init(&tmpxvattr);
3061 * Immutable files can only alter immutable bit and atime
3063 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
3064 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
3065 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
3067 return (SET_ERROR(EPERM));
3070 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
3072 return (SET_ERROR(EPERM));
3076 * Verify timestamps doesn't overflow 32 bits.
3077 * ZFS can handle large timestamps, but 32bit syscalls can't
3078 * handle times greater than 2039. This check should be removed
3079 * once large timestamps are fully supported.
3081 if (mask & (AT_ATIME | AT_MTIME)) {
3082 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
3083 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
3085 return (SET_ERROR(EOVERFLOW));
3093 /* Can this be moved to before the top label? */
3094 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
3096 return (SET_ERROR(EROFS));
3100 * First validate permissions
3103 if (mask & AT_SIZE) {
3105 * XXX - Note, we are not providing any open
3106 * mode flags here (like FNDELAY), so we may
3107 * block if there are locks present... this
3108 * should be addressed in openat().
3110 /* XXX - would it be OK to generate a log record here? */
3111 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
3118 if (mask & (AT_ATIME|AT_MTIME) ||
3119 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
3120 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
3121 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
3122 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
3123 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
3124 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
3125 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
3126 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
3130 if (mask & (AT_UID|AT_GID)) {
3131 int idmask = (mask & (AT_UID|AT_GID));
3136 * NOTE: even if a new mode is being set,
3137 * we may clear S_ISUID/S_ISGID bits.
3140 if (!(mask & AT_MODE))
3141 vap->va_mode = zp->z_mode;
3144 * Take ownership or chgrp to group we are a member of
3147 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
3148 take_group = (mask & AT_GID) &&
3149 zfs_groupmember(zfsvfs, vap->va_gid, cr);
3152 * If both AT_UID and AT_GID are set then take_owner and
3153 * take_group must both be set in order to allow taking
3156 * Otherwise, send the check through secpolicy_vnode_setattr()
3160 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
3161 ((idmask == AT_UID) && take_owner) ||
3162 ((idmask == AT_GID) && take_group)) {
3163 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
3164 skipaclchk, cr) == 0) {
3166 * Remove setuid/setgid for non-privileged users
3168 secpolicy_setid_clear(vap, vp, cr);
3169 trim_mask = (mask & (AT_UID|AT_GID));
3178 mutex_enter(&zp->z_lock);
3179 oldva.va_mode = zp->z_mode;
3180 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
3181 if (mask & AT_XVATTR) {
3183 * Update xvattr mask to include only those attributes
3184 * that are actually changing.
3186 * the bits will be restored prior to actually setting
3187 * the attributes so the caller thinks they were set.
3189 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
3190 if (xoap->xoa_appendonly !=
3191 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
3194 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
3195 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
3199 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
3200 if (xoap->xoa_nounlink !=
3201 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
3204 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
3205 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
3209 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
3210 if (xoap->xoa_immutable !=
3211 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
3214 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
3215 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
3219 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
3220 if (xoap->xoa_nodump !=
3221 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
3224 XVA_CLR_REQ(xvap, XAT_NODUMP);
3225 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
3229 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3230 if (xoap->xoa_av_modified !=
3231 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3234 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3235 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
3239 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3240 if ((vp->v_type != VREG &&
3241 xoap->xoa_av_quarantined) ||
3242 xoap->xoa_av_quarantined !=
3243 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3246 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3247 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
3251 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3252 mutex_exit(&zp->z_lock);
3254 return (SET_ERROR(EPERM));
3257 if (need_policy == FALSE &&
3258 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3259 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3264 mutex_exit(&zp->z_lock);
3266 if (mask & AT_MODE) {
3267 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3268 err = secpolicy_setid_setsticky_clear(vp, vap,
3274 trim_mask |= AT_MODE;
3282 * If trim_mask is set then take ownership
3283 * has been granted or write_acl is present and user
3284 * has the ability to modify mode. In that case remove
3285 * UID|GID and or MODE from mask so that
3286 * secpolicy_vnode_setattr() doesn't revoke it.
3290 saved_mask = vap->va_mask;
3291 vap->va_mask &= ~trim_mask;
3292 if (trim_mask & AT_MODE) {
3294 * Save the mode, as secpolicy_vnode_setattr()
3295 * will overwrite it with ova.va_mode.
3297 saved_mode = vap->va_mode;
3300 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3301 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3308 vap->va_mask |= saved_mask;
3309 if (trim_mask & AT_MODE) {
3311 * Recover the mode after
3312 * secpolicy_vnode_setattr().
3314 vap->va_mode = saved_mode;
3320 * secpolicy_vnode_setattr, or take ownership may have
3323 mask = vap->va_mask;
3325 if ((mask & (AT_UID | AT_GID))) {
3326 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3327 &xattr_obj, sizeof (xattr_obj));
3329 if (err == 0 && xattr_obj) {
3330 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3334 if (mask & AT_UID) {
3335 new_uid = zfs_fuid_create(zfsvfs,
3336 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3337 if (new_uid != zp->z_uid &&
3338 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3340 VN_RELE(ZTOV(attrzp));
3341 err = SET_ERROR(EDQUOT);
3346 if (mask & AT_GID) {
3347 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3348 cr, ZFS_GROUP, &fuidp);
3349 if (new_gid != zp->z_gid &&
3350 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3352 VN_RELE(ZTOV(attrzp));
3353 err = SET_ERROR(EDQUOT);
3358 tx = dmu_tx_create(zfsvfs->z_os);
3360 if (mask & AT_MODE) {
3361 uint64_t pmode = zp->z_mode;
3363 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3365 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3366 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3367 err = SET_ERROR(EPERM);
3371 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3374 mutex_enter(&zp->z_lock);
3375 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3377 * Are we upgrading ACL from old V0 format
3380 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3381 zfs_znode_acl_version(zp) ==
3382 ZFS_ACL_VERSION_INITIAL) {
3383 dmu_tx_hold_free(tx, acl_obj, 0,
3385 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3386 0, aclp->z_acl_bytes);
3388 dmu_tx_hold_write(tx, acl_obj, 0,
3391 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3392 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3393 0, aclp->z_acl_bytes);
3395 mutex_exit(&zp->z_lock);
3396 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3398 if ((mask & AT_XVATTR) &&
3399 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3400 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3402 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3406 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3409 fuid_dirtied = zfsvfs->z_fuid_dirty;
3411 zfs_fuid_txhold(zfsvfs, tx);
3413 zfs_sa_upgrade_txholds(tx, zp);
3415 err = dmu_tx_assign(tx, TXG_WAIT);
3421 * Set each attribute requested.
3422 * We group settings according to the locks they need to acquire.
3424 * Note: you cannot set ctime directly, although it will be
3425 * updated as a side-effect of calling this function.
3429 if (mask & (AT_UID|AT_GID|AT_MODE))
3430 mutex_enter(&zp->z_acl_lock);
3431 mutex_enter(&zp->z_lock);
3433 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3434 &zp->z_pflags, sizeof (zp->z_pflags));
3437 if (mask & (AT_UID|AT_GID|AT_MODE))
3438 mutex_enter(&attrzp->z_acl_lock);
3439 mutex_enter(&attrzp->z_lock);
3440 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3441 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3442 sizeof (attrzp->z_pflags));
3445 if (mask & (AT_UID|AT_GID)) {
3447 if (mask & AT_UID) {
3448 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3449 &new_uid, sizeof (new_uid));
3450 zp->z_uid = new_uid;
3452 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3453 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3455 attrzp->z_uid = new_uid;
3459 if (mask & AT_GID) {
3460 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3461 NULL, &new_gid, sizeof (new_gid));
3462 zp->z_gid = new_gid;
3464 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3465 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3467 attrzp->z_gid = new_gid;
3470 if (!(mask & AT_MODE)) {
3471 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3472 NULL, &new_mode, sizeof (new_mode));
3473 new_mode = zp->z_mode;
3475 err = zfs_acl_chown_setattr(zp);
3478 err = zfs_acl_chown_setattr(attrzp);
3483 if (mask & AT_MODE) {
3484 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3485 &new_mode, sizeof (new_mode));
3486 zp->z_mode = new_mode;
3487 ASSERT3U((uintptr_t)aclp, !=, 0);
3488 err = zfs_aclset_common(zp, aclp, cr, tx);
3490 if (zp->z_acl_cached)
3491 zfs_acl_free(zp->z_acl_cached);
3492 zp->z_acl_cached = aclp;
3497 if (mask & AT_ATIME) {
3498 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3499 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3500 &zp->z_atime, sizeof (zp->z_atime));
3503 if (mask & AT_MTIME) {
3504 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3505 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3506 mtime, sizeof (mtime));
3509 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3510 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3511 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3512 NULL, mtime, sizeof (mtime));
3513 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3514 &ctime, sizeof (ctime));
3515 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3517 } else if (mask != 0) {
3518 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3519 &ctime, sizeof (ctime));
3520 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3523 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3524 SA_ZPL_CTIME(zfsvfs), NULL,
3525 &ctime, sizeof (ctime));
3526 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3527 mtime, ctime, B_TRUE);
3531 * Do this after setting timestamps to prevent timestamp
3532 * update from toggling bit
3535 if (xoap && (mask & AT_XVATTR)) {
3538 * restore trimmed off masks
3539 * so that return masks can be set for caller.
3542 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3543 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3545 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3546 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3548 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3549 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3551 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3552 XVA_SET_REQ(xvap, XAT_NODUMP);
3554 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3555 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3557 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3558 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3561 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3562 ASSERT(vp->v_type == VREG);
3564 zfs_xvattr_set(zp, xvap, tx);
3568 zfs_fuid_sync(zfsvfs, tx);
3571 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3573 mutex_exit(&zp->z_lock);
3574 if (mask & (AT_UID|AT_GID|AT_MODE))
3575 mutex_exit(&zp->z_acl_lock);
3578 if (mask & (AT_UID|AT_GID|AT_MODE))
3579 mutex_exit(&attrzp->z_acl_lock);
3580 mutex_exit(&attrzp->z_lock);
3583 if (err == 0 && attrzp) {
3584 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3590 VN_RELE(ZTOV(attrzp));
3596 zfs_fuid_info_free(fuidp);
3602 if (err == ERESTART)
3605 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3610 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3611 zil_commit(zilog, 0);
3617 typedef struct zfs_zlock {
3618 krwlock_t *zl_rwlock; /* lock we acquired */
3619 znode_t *zl_znode; /* znode we held */
3620 struct zfs_zlock *zl_next; /* next in list */
3624 * Drop locks and release vnodes that were held by zfs_rename_lock().
3627 zfs_rename_unlock(zfs_zlock_t **zlpp)
3631 while ((zl = *zlpp) != NULL) {
3632 if (zl->zl_znode != NULL)
3633 VN_RELE(ZTOV(zl->zl_znode));
3634 rw_exit(zl->zl_rwlock);
3635 *zlpp = zl->zl_next;
3636 kmem_free(zl, sizeof (*zl));
3641 * Search back through the directory tree, using the ".." entries.
3642 * Lock each directory in the chain to prevent concurrent renames.
3643 * Fail any attempt to move a directory into one of its own descendants.
3644 * XXX - z_parent_lock can overlap with map or grow locks
3647 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3651 uint64_t rootid = zp->z_zfsvfs->z_root;
3652 uint64_t oidp = zp->z_id;
3653 krwlock_t *rwlp = &szp->z_parent_lock;
3654 krw_t rw = RW_WRITER;
3657 * First pass write-locks szp and compares to zp->z_id.
3658 * Later passes read-lock zp and compare to zp->z_parent.
3661 if (!rw_tryenter(rwlp, rw)) {
3663 * Another thread is renaming in this path.
3664 * Note that if we are a WRITER, we don't have any
3665 * parent_locks held yet.
3667 if (rw == RW_READER && zp->z_id > szp->z_id) {
3669 * Drop our locks and restart
3671 zfs_rename_unlock(&zl);
3675 rwlp = &szp->z_parent_lock;
3680 * Wait for other thread to drop its locks
3686 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3687 zl->zl_rwlock = rwlp;
3688 zl->zl_znode = NULL;
3689 zl->zl_next = *zlpp;
3692 if (oidp == szp->z_id) /* We're a descendant of szp */
3693 return (SET_ERROR(EINVAL));
3695 if (oidp == rootid) /* We've hit the top */
3698 if (rw == RW_READER) { /* i.e. not the first pass */
3699 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3704 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3705 &oidp, sizeof (oidp));
3706 rwlp = &zp->z_parent_lock;
3709 } while (zp->z_id != sdzp->z_id);
3715 * Move an entry from the provided source directory to the target
3716 * directory. Change the entry name as indicated.
3718 * IN: sdvp - Source directory containing the "old entry".
3719 * snm - Old entry name.
3720 * tdvp - Target directory to contain the "new entry".
3721 * tnm - New entry name.
3722 * cr - credentials of caller.
3723 * ct - caller context
3724 * flags - case flags
3726 * RETURN: 0 on success, error code on failure.
3729 * sdvp,tdvp - ctime|mtime updated
3733 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3734 caller_context_t *ct, int flags)
3736 znode_t *tdzp, *szp, *tzp;
3737 znode_t *sdzp = VTOZ(sdvp);
3738 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs;
3741 zfs_dirlock_t *sdl, *tdl;
3744 int cmp, serr, terr;
3747 boolean_t waited = B_FALSE;
3750 ZFS_VERIFY_ZP(sdzp);
3751 zilog = zfsvfs->z_log;
3754 * Make sure we have the real vp for the target directory.
3756 if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3759 if (tdvp->v_vfsp != sdvp->v_vfsp || zfsctl_is_node(tdvp)) {
3761 return (SET_ERROR(EXDEV));
3765 ZFS_VERIFY_ZP(tdzp);
3766 if (zfsvfs->z_utf8 && u8_validate(tnm,
3767 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3769 return (SET_ERROR(EILSEQ));
3772 if (flags & FIGNORECASE)
3781 * This is to prevent the creation of links into attribute space
3782 * by renaming a linked file into/outof an attribute directory.
3783 * See the comment in zfs_link() for why this is considered bad.
3785 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3787 return (SET_ERROR(EINVAL));
3791 * Lock source and target directory entries. To prevent deadlock,
3792 * a lock ordering must be defined. We lock the directory with
3793 * the smallest object id first, or if it's a tie, the one with
3794 * the lexically first name.
3796 if (sdzp->z_id < tdzp->z_id) {
3798 } else if (sdzp->z_id > tdzp->z_id) {
3802 * First compare the two name arguments without
3803 * considering any case folding.
3805 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3807 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3808 ASSERT(error == 0 || !zfsvfs->z_utf8);
3811 * POSIX: "If the old argument and the new argument
3812 * both refer to links to the same existing file,
3813 * the rename() function shall return successfully
3814 * and perform no other action."
3820 * If the file system is case-folding, then we may
3821 * have some more checking to do. A case-folding file
3822 * system is either supporting mixed case sensitivity
3823 * access or is completely case-insensitive. Note
3824 * that the file system is always case preserving.
3826 * In mixed sensitivity mode case sensitive behavior
3827 * is the default. FIGNORECASE must be used to
3828 * explicitly request case insensitive behavior.
3830 * If the source and target names provided differ only
3831 * by case (e.g., a request to rename 'tim' to 'Tim'),
3832 * we will treat this as a special case in the
3833 * case-insensitive mode: as long as the source name
3834 * is an exact match, we will allow this to proceed as
3835 * a name-change request.
3837 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3838 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3839 flags & FIGNORECASE)) &&
3840 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3843 * case preserving rename request, require exact
3852 * If the source and destination directories are the same, we should
3853 * grab the z_name_lock of that directory only once.
3857 rw_enter(&sdzp->z_name_lock, RW_READER);
3861 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3862 ZEXISTS | zflg, NULL, NULL);
3863 terr = zfs_dirent_lock(&tdl,
3864 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3866 terr = zfs_dirent_lock(&tdl,
3867 tdzp, tnm, &tzp, zflg, NULL, NULL);
3868 serr = zfs_dirent_lock(&sdl,
3869 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3875 * Source entry invalid or not there.
3878 zfs_dirent_unlock(tdl);
3884 rw_exit(&sdzp->z_name_lock);
3887 * FreeBSD: In OpenSolaris they only check if rename source is
3888 * ".." here, because "." is handled in their lookup. This is
3889 * not the case for FreeBSD, so we check for "." explicitly.
3891 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0)
3892 serr = SET_ERROR(EINVAL);
3897 zfs_dirent_unlock(sdl);
3901 rw_exit(&sdzp->z_name_lock);
3903 if (strcmp(tnm, "..") == 0)
3904 terr = SET_ERROR(EINVAL);
3910 * Must have write access at the source to remove the old entry
3911 * and write access at the target to create the new entry.
3912 * Note that if target and source are the same, this can be
3913 * done in a single check.
3916 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3919 if (ZTOV(szp)->v_type == VDIR) {
3921 * Check to make sure rename is valid.
3922 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3924 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3929 * Does target exist?
3933 * Source and target must be the same type.
3935 if (ZTOV(szp)->v_type == VDIR) {
3936 if (ZTOV(tzp)->v_type != VDIR) {
3937 error = SET_ERROR(ENOTDIR);
3941 if (ZTOV(tzp)->v_type == VDIR) {
3942 error = SET_ERROR(EISDIR);
3947 * POSIX dictates that when the source and target
3948 * entries refer to the same file object, rename
3949 * must do nothing and exit without error.
3951 if (szp->z_id == tzp->z_id) {
3957 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3959 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3962 * notify the target directory if it is not the same
3963 * as source directory.
3966 vnevent_rename_dest_dir(tdvp, ct);
3969 tx = dmu_tx_create(zfsvfs->z_os);
3970 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3971 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3972 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3973 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3975 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3976 zfs_sa_upgrade_txholds(tx, tdzp);
3979 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3980 zfs_sa_upgrade_txholds(tx, tzp);
3983 zfs_sa_upgrade_txholds(tx, szp);
3984 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3985 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3988 zfs_rename_unlock(&zl);
3989 zfs_dirent_unlock(sdl);
3990 zfs_dirent_unlock(tdl);
3993 rw_exit(&sdzp->z_name_lock);
3998 if (error == ERESTART) {
4009 if (tzp) /* Attempt to remove the existing target */
4010 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
4013 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
4015 szp->z_pflags |= ZFS_AV_MODIFIED;
4017 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
4018 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
4021 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
4023 zfs_log_rename(zilog, tx, TX_RENAME |
4024 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
4025 sdl->dl_name, tdzp, tdl->dl_name, szp);
4028 * Update path information for the target vnode
4030 vn_renamepath(tdvp, ZTOV(szp), tnm,
4034 * At this point, we have successfully created
4035 * the target name, but have failed to remove
4036 * the source name. Since the create was done
4037 * with the ZRENAMING flag, there are
4038 * complications; for one, the link count is
4039 * wrong. The easiest way to deal with this
4040 * is to remove the newly created target, and
4041 * return the original error. This must
4042 * succeed; fortunately, it is very unlikely to
4043 * fail, since we just created it.
4045 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
4046 ZRENAMING, NULL), ==, 0);
4049 #ifdef FREEBSD_NAMECACHE
4053 cache_purge(ZTOV(szp));
4055 cache_purge(ZTOV(tzp));
4063 zfs_rename_unlock(&zl);
4065 zfs_dirent_unlock(sdl);
4066 zfs_dirent_unlock(tdl);
4069 rw_exit(&sdzp->z_name_lock);
4076 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4077 zil_commit(zilog, 0);
4085 * Insert the indicated symbolic reference entry into the directory.
4087 * IN: dvp - Directory to contain new symbolic link.
4088 * link - Name for new symlink entry.
4089 * vap - Attributes of new entry.
4090 * cr - credentials of caller.
4091 * ct - caller context
4092 * flags - case flags
4094 * RETURN: 0 on success, error code on failure.
4097 * dvp - ctime|mtime updated
4101 zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link,
4102 cred_t *cr, kthread_t *td)
4104 znode_t *zp, *dzp = VTOZ(dvp);
4107 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4109 uint64_t len = strlen(link);
4112 zfs_acl_ids_t acl_ids;
4113 boolean_t fuid_dirtied;
4114 uint64_t txtype = TX_SYMLINK;
4115 boolean_t waited = B_FALSE;
4118 ASSERT(vap->va_type == VLNK);
4122 zilog = zfsvfs->z_log;
4124 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
4125 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4127 return (SET_ERROR(EILSEQ));
4129 if (flags & FIGNORECASE)
4132 if (len > MAXPATHLEN) {
4134 return (SET_ERROR(ENAMETOOLONG));
4137 if ((error = zfs_acl_ids_create(dzp, 0,
4138 vap, cr, NULL, &acl_ids)) != 0) {
4143 getnewvnode_reserve(1);
4147 * Attempt to lock directory; fail if entry already exists.
4149 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
4151 zfs_acl_ids_free(&acl_ids);
4152 getnewvnode_drop_reserve();
4157 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4158 zfs_acl_ids_free(&acl_ids);
4159 zfs_dirent_unlock(dl);
4160 getnewvnode_drop_reserve();
4165 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
4166 zfs_acl_ids_free(&acl_ids);
4167 zfs_dirent_unlock(dl);
4168 getnewvnode_drop_reserve();
4170 return (SET_ERROR(EDQUOT));
4172 tx = dmu_tx_create(zfsvfs->z_os);
4173 fuid_dirtied = zfsvfs->z_fuid_dirty;
4174 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
4175 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4176 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
4177 ZFS_SA_BASE_ATTR_SIZE + len);
4178 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
4179 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
4180 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
4181 acl_ids.z_aclp->z_acl_bytes);
4184 zfs_fuid_txhold(zfsvfs, tx);
4185 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4187 zfs_dirent_unlock(dl);
4188 if (error == ERESTART) {
4194 zfs_acl_ids_free(&acl_ids);
4196 getnewvnode_drop_reserve();
4202 * Create a new object for the symlink.
4203 * for version 4 ZPL datsets the symlink will be an SA attribute
4205 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
4208 zfs_fuid_sync(zfsvfs, tx);
4210 mutex_enter(&zp->z_lock);
4212 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
4215 zfs_sa_symlink(zp, link, len, tx);
4216 mutex_exit(&zp->z_lock);
4219 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
4220 &zp->z_size, sizeof (zp->z_size), tx);
4222 * Insert the new object into the directory.
4224 (void) zfs_link_create(dl, zp, tx, ZNEW);
4226 if (flags & FIGNORECASE)
4228 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
4231 zfs_acl_ids_free(&acl_ids);
4235 getnewvnode_drop_reserve();
4237 zfs_dirent_unlock(dl);
4239 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4240 zil_commit(zilog, 0);
4247 * Return, in the buffer contained in the provided uio structure,
4248 * the symbolic path referred to by vp.
4250 * IN: vp - vnode of symbolic link.
4251 * uio - structure to contain the link path.
4252 * cr - credentials of caller.
4253 * ct - caller context
4255 * OUT: uio - structure containing the link path.
4257 * RETURN: 0 on success, error code on failure.
4260 * vp - atime updated
4264 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
4266 znode_t *zp = VTOZ(vp);
4267 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4273 mutex_enter(&zp->z_lock);
4275 error = sa_lookup_uio(zp->z_sa_hdl,
4276 SA_ZPL_SYMLINK(zfsvfs), uio);
4278 error = zfs_sa_readlink(zp, uio);
4279 mutex_exit(&zp->z_lock);
4281 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4288 * Insert a new entry into directory tdvp referencing svp.
4290 * IN: tdvp - Directory to contain new entry.
4291 * svp - vnode of new entry.
4292 * name - name of new entry.
4293 * cr - credentials of caller.
4294 * ct - caller context
4296 * RETURN: 0 on success, error code on failure.
4299 * tdvp - ctime|mtime updated
4300 * svp - ctime updated
4304 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
4305 caller_context_t *ct, int flags)
4307 znode_t *dzp = VTOZ(tdvp);
4309 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4318 boolean_t waited = B_FALSE;
4320 ASSERT(tdvp->v_type == VDIR);
4324 zilog = zfsvfs->z_log;
4326 if (VOP_REALVP(svp, &realvp, ct) == 0)
4330 * POSIX dictates that we return EPERM here.
4331 * Better choices include ENOTSUP or EISDIR.
4333 if (svp->v_type == VDIR) {
4335 return (SET_ERROR(EPERM));
4338 if (svp->v_vfsp != tdvp->v_vfsp || zfsctl_is_node(svp)) {
4340 return (SET_ERROR(EXDEV));
4346 /* Prevent links to .zfs/shares files */
4348 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4349 &parent, sizeof (uint64_t))) != 0) {
4353 if (parent == zfsvfs->z_shares_dir) {
4355 return (SET_ERROR(EPERM));
4358 if (zfsvfs->z_utf8 && u8_validate(name,
4359 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4361 return (SET_ERROR(EILSEQ));
4363 if (flags & FIGNORECASE)
4367 * We do not support links between attributes and non-attributes
4368 * because of the potential security risk of creating links
4369 * into "normal" file space in order to circumvent restrictions
4370 * imposed in attribute space.
4372 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4374 return (SET_ERROR(EINVAL));
4378 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4379 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) {
4381 return (SET_ERROR(EPERM));
4384 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4391 * Attempt to lock directory; fail if entry already exists.
4393 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4399 tx = dmu_tx_create(zfsvfs->z_os);
4400 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4401 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4402 zfs_sa_upgrade_txholds(tx, szp);
4403 zfs_sa_upgrade_txholds(tx, dzp);
4404 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4406 zfs_dirent_unlock(dl);
4407 if (error == ERESTART) {
4418 error = zfs_link_create(dl, szp, tx, 0);
4421 uint64_t txtype = TX_LINK;
4422 if (flags & FIGNORECASE)
4424 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4429 zfs_dirent_unlock(dl);
4432 vnevent_link(svp, ct);
4435 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4436 zil_commit(zilog, 0);
4444 * zfs_null_putapage() is used when the file system has been force
4445 * unmounted. It just drops the pages.
4449 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4450 size_t *lenp, int flags, cred_t *cr)
4452 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4457 * Push a page out to disk, klustering if possible.
4459 * IN: vp - file to push page to.
4460 * pp - page to push.
4461 * flags - additional flags.
4462 * cr - credentials of caller.
4464 * OUT: offp - start of range pushed.
4465 * lenp - len of range pushed.
4467 * RETURN: 0 on success, error code on failure.
4469 * NOTE: callers must have locked the page to be pushed. On
4470 * exit, the page (and all other pages in the kluster) must be
4475 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4476 size_t *lenp, int flags, cred_t *cr)
4478 znode_t *zp = VTOZ(vp);
4479 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4481 u_offset_t off, koff;
4488 * If our blocksize is bigger than the page size, try to kluster
4489 * multiple pages so that we write a full block (thus avoiding
4490 * a read-modify-write).
4492 if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4493 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4494 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4495 ASSERT(koff <= zp->z_size);
4496 if (koff + klen > zp->z_size)
4497 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4498 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4500 ASSERT3U(btop(len), ==, btopr(len));
4503 * Can't push pages past end-of-file.
4505 if (off >= zp->z_size) {
4506 /* ignore all pages */
4509 } else if (off + len > zp->z_size) {
4510 int npages = btopr(zp->z_size - off);
4513 page_list_break(&pp, &trunc, npages);
4514 /* ignore pages past end of file */
4516 pvn_write_done(trunc, flags);
4517 len = zp->z_size - off;
4520 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4521 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4522 err = SET_ERROR(EDQUOT);
4525 tx = dmu_tx_create(zfsvfs->z_os);
4526 dmu_tx_hold_write(tx, zp->z_id, off, len);
4528 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4529 zfs_sa_upgrade_txholds(tx, zp);
4530 err = dmu_tx_assign(tx, TXG_WAIT);
4536 if (zp->z_blksz <= PAGESIZE) {
4537 caddr_t va = zfs_map_page(pp, S_READ);
4538 ASSERT3U(len, <=, PAGESIZE);
4539 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4540 zfs_unmap_page(pp, va);
4542 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4546 uint64_t mtime[2], ctime[2];
4547 sa_bulk_attr_t bulk[3];
4550 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4552 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4554 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4556 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4558 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4563 pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4573 * Copy the portion of the file indicated from pages into the file.
4574 * The pages are stored in a page list attached to the files vnode.
4576 * IN: vp - vnode of file to push page data to.
4577 * off - position in file to put data.
4578 * len - amount of data to write.
4579 * flags - flags to control the operation.
4580 * cr - credentials of caller.
4581 * ct - caller context.
4583 * RETURN: 0 on success, error code on failure.
4586 * vp - ctime|mtime updated
4590 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4591 caller_context_t *ct)
4593 znode_t *zp = VTOZ(vp);
4594 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4606 * Align this request to the file block size in case we kluster.
4607 * XXX - this can result in pretty aggresive locking, which can
4608 * impact simultanious read/write access. One option might be
4609 * to break up long requests (len == 0) into block-by-block
4610 * operations to get narrower locking.
4612 blksz = zp->z_blksz;
4614 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4617 if (len > 0 && ISP2(blksz))
4618 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4624 * Search the entire vp list for pages >= io_off.
4626 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4627 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4630 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4632 if (off > zp->z_size) {
4633 /* past end of file */
4634 zfs_range_unlock(rl);
4639 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4641 for (off = io_off; io_off < off + len; io_off += io_len) {
4642 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4643 pp = page_lookup(vp, io_off,
4644 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4646 pp = page_lookup_nowait(vp, io_off,
4647 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4650 if (pp != NULL && pvn_getdirty(pp, flags)) {
4654 * Found a dirty page to push
4656 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4664 zfs_range_unlock(rl);
4665 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4666 zil_commit(zfsvfs->z_log, zp->z_id);
4674 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4676 znode_t *zp = VTOZ(vp);
4677 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4680 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4681 if (zp->z_sa_hdl == NULL) {
4683 * The fs has been unmounted, or we did a
4684 * suspend/resume and this file no longer exists.
4686 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4687 vrecycle(vp, curthread);
4691 mutex_enter(&zp->z_lock);
4692 if (zp->z_unlinked) {
4694 * Fast path to recycle a vnode of a removed file.
4696 mutex_exit(&zp->z_lock);
4697 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4698 vrecycle(vp, curthread);
4701 mutex_exit(&zp->z_lock);
4703 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4704 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4706 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4707 zfs_sa_upgrade_txholds(tx, zp);
4708 error = dmu_tx_assign(tx, TXG_WAIT);
4712 mutex_enter(&zp->z_lock);
4713 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4714 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4715 zp->z_atime_dirty = 0;
4716 mutex_exit(&zp->z_lock);
4720 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4725 * Bounds-check the seek operation.
4727 * IN: vp - vnode seeking within
4728 * ooff - old file offset
4729 * noffp - pointer to new file offset
4730 * ct - caller context
4732 * RETURN: 0 on success, EINVAL if new offset invalid.
4736 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4737 caller_context_t *ct)
4739 if (vp->v_type == VDIR)
4741 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4745 * Pre-filter the generic locking function to trap attempts to place
4746 * a mandatory lock on a memory mapped file.
4749 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4750 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4752 znode_t *zp = VTOZ(vp);
4753 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4759 * We are following the UFS semantics with respect to mapcnt
4760 * here: If we see that the file is mapped already, then we will
4761 * return an error, but we don't worry about races between this
4762 * function and zfs_map().
4764 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4766 return (SET_ERROR(EAGAIN));
4769 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4773 * If we can't find a page in the cache, we will create a new page
4774 * and fill it with file data. For efficiency, we may try to fill
4775 * multiple pages at once (klustering) to fill up the supplied page
4776 * list. Note that the pages to be filled are held with an exclusive
4777 * lock to prevent access by other threads while they are being filled.
4780 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4781 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4783 znode_t *zp = VTOZ(vp);
4784 page_t *pp, *cur_pp;
4785 objset_t *os = zp->z_zfsvfs->z_os;
4786 u_offset_t io_off, total;
4790 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4792 * We only have a single page, don't bother klustering
4796 pp = page_create_va(vp, io_off, io_len,
4797 PG_EXCL | PG_WAIT, seg, addr);
4800 * Try to find enough pages to fill the page list
4802 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4803 &io_len, off, plsz, 0);
4807 * The page already exists, nothing to do here.
4814 * Fill the pages in the kluster.
4817 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4820 ASSERT3U(io_off, ==, cur_pp->p_offset);
4821 va = zfs_map_page(cur_pp, S_WRITE);
4822 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4824 zfs_unmap_page(cur_pp, va);
4826 /* On error, toss the entire kluster */
4827 pvn_read_done(pp, B_ERROR);
4828 /* convert checksum errors into IO errors */
4830 err = SET_ERROR(EIO);
4833 cur_pp = cur_pp->p_next;
4837 * Fill in the page list array from the kluster starting
4838 * from the desired offset `off'.
4839 * NOTE: the page list will always be null terminated.
4841 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4842 ASSERT(pl == NULL || (*pl)->p_offset == off);
4848 * Return pointers to the pages for the file region [off, off + len]
4849 * in the pl array. If plsz is greater than len, this function may
4850 * also return page pointers from after the specified region
4851 * (i.e. the region [off, off + plsz]). These additional pages are
4852 * only returned if they are already in the cache, or were created as
4853 * part of a klustered read.
4855 * IN: vp - vnode of file to get data from.
4856 * off - position in file to get data from.
4857 * len - amount of data to retrieve.
4858 * plsz - length of provided page list.
4859 * seg - segment to obtain pages for.
4860 * addr - virtual address of fault.
4861 * rw - mode of created pages.
4862 * cr - credentials of caller.
4863 * ct - caller context.
4865 * OUT: protp - protection mode of created pages.
4866 * pl - list of pages created.
4868 * RETURN: 0 on success, error code on failure.
4871 * vp - atime updated
4875 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4876 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4877 enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4879 znode_t *zp = VTOZ(vp);
4880 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4884 /* we do our own caching, faultahead is unnecessary */
4887 else if (len > plsz)
4890 len = P2ROUNDUP(len, PAGESIZE);
4891 ASSERT(plsz >= len);
4900 * Loop through the requested range [off, off + len) looking
4901 * for pages. If we don't find a page, we will need to create
4902 * a new page and fill it with data from the file.
4905 if (*pl = page_lookup(vp, off, SE_SHARED))
4907 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4910 ASSERT3U((*pl)->p_offset, ==, off);
4914 ASSERT3U(len, >=, PAGESIZE);
4917 ASSERT3U(plsz, >=, PAGESIZE);
4924 * Fill out the page array with any pages already in the cache.
4927 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4934 * Release any pages we have previously locked.
4939 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4949 * Request a memory map for a section of a file. This code interacts
4950 * with common code and the VM system as follows:
4952 * - common code calls mmap(), which ends up in smmap_common()
4953 * - this calls VOP_MAP(), which takes you into (say) zfs
4954 * - zfs_map() calls as_map(), passing segvn_create() as the callback
4955 * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4956 * - zfs_addmap() updates z_mapcnt
4960 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4961 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4962 caller_context_t *ct)
4964 znode_t *zp = VTOZ(vp);
4965 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4966 segvn_crargs_t vn_a;
4972 if ((prot & PROT_WRITE) && (zp->z_pflags &
4973 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4975 return (SET_ERROR(EPERM));
4978 if ((prot & (PROT_READ | PROT_EXEC)) &&
4979 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4981 return (SET_ERROR(EACCES));
4984 if (vp->v_flag & VNOMAP) {
4986 return (SET_ERROR(ENOSYS));
4989 if (off < 0 || len > MAXOFFSET_T - off) {
4991 return (SET_ERROR(ENXIO));
4994 if (vp->v_type != VREG) {
4996 return (SET_ERROR(ENODEV));
5000 * If file is locked, disallow mapping.
5002 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
5004 return (SET_ERROR(EAGAIN));
5008 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
5016 vn_a.offset = (u_offset_t)off;
5017 vn_a.type = flags & MAP_TYPE;
5019 vn_a.maxprot = maxprot;
5022 vn_a.flags = flags & ~MAP_TYPE;
5024 vn_a.lgrp_mem_policy_flags = 0;
5026 error = as_map(as, *addrp, len, segvn_create, &vn_a);
5035 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5036 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
5037 caller_context_t *ct)
5039 uint64_t pages = btopr(len);
5041 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
5046 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
5047 * more accurate mtime for the associated file. Since we don't have a way of
5048 * detecting when the data was actually modified, we have to resort to
5049 * heuristics. If an explicit msync() is done, then we mark the mtime when the
5050 * last page is pushed. The problem occurs when the msync() call is omitted,
5051 * which by far the most common case:
5059 * putpage() via fsflush
5061 * If we wait until fsflush to come along, we can have a modification time that
5062 * is some arbitrary point in the future. In order to prevent this in the
5063 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
5068 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5069 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
5070 caller_context_t *ct)
5072 uint64_t pages = btopr(len);
5074 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
5075 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
5077 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
5078 vn_has_cached_data(vp))
5079 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
5085 * Free or allocate space in a file. Currently, this function only
5086 * supports the `F_FREESP' command. However, this command is somewhat
5087 * misnamed, as its functionality includes the ability to allocate as
5088 * well as free space.
5090 * IN: vp - vnode of file to free data in.
5091 * cmd - action to take (only F_FREESP supported).
5092 * bfp - section of file to free/alloc.
5093 * flag - current file open mode flags.
5094 * offset - current file offset.
5095 * cr - credentials of caller [UNUSED].
5096 * ct - caller context.
5098 * RETURN: 0 on success, error code on failure.
5101 * vp - ctime|mtime updated
5105 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
5106 offset_t offset, cred_t *cr, caller_context_t *ct)
5108 znode_t *zp = VTOZ(vp);
5109 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5116 if (cmd != F_FREESP) {
5118 return (SET_ERROR(EINVAL));
5121 if (error = convoff(vp, bfp, 0, offset)) {
5126 if (bfp->l_len < 0) {
5128 return (SET_ERROR(EINVAL));
5132 len = bfp->l_len; /* 0 means from off to end of file */
5134 error = zfs_freesp(zp, off, len, flag, TRUE);
5141 CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid));
5142 CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid));
5146 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
5148 znode_t *zp = VTOZ(vp);
5149 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5152 uint64_t object = zp->z_id;
5159 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
5160 &gen64, sizeof (uint64_t))) != 0) {
5165 gen = (uint32_t)gen64;
5167 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
5170 if (fidp->fid_len < size) {
5171 fidp->fid_len = size;
5173 return (SET_ERROR(ENOSPC));
5176 fidp->fid_len = size;
5179 zfid = (zfid_short_t *)fidp;
5181 zfid->zf_len = size;
5183 for (i = 0; i < sizeof (zfid->zf_object); i++)
5184 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
5186 /* Must have a non-zero generation number to distinguish from .zfs */
5189 for (i = 0; i < sizeof (zfid->zf_gen); i++)
5190 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
5192 if (size == LONG_FID_LEN) {
5193 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
5196 zlfid = (zfid_long_t *)fidp;
5198 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
5199 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
5201 /* XXX - this should be the generation number for the objset */
5202 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
5203 zlfid->zf_setgen[i] = 0;
5211 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
5212 caller_context_t *ct)
5224 case _PC_FILESIZEBITS:
5228 case _PC_XATTR_EXISTS:
5230 zfsvfs = zp->z_zfsvfs;
5234 error = zfs_dirent_lock(&dl, zp, "", &xzp,
5235 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
5237 zfs_dirent_unlock(dl);
5238 if (!zfs_dirempty(xzp))
5241 } else if (error == ENOENT) {
5243 * If there aren't extended attributes, it's the
5244 * same as having zero of them.
5251 case _PC_SATTR_ENABLED:
5252 case _PC_SATTR_EXISTS:
5253 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
5254 (vp->v_type == VREG || vp->v_type == VDIR);
5257 case _PC_ACCESS_FILTERING:
5258 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
5262 case _PC_ACL_ENABLED:
5263 *valp = _ACL_ACE_ENABLED;
5266 case _PC_MIN_HOLE_SIZE:
5267 *valp = (int)SPA_MINBLOCKSIZE;
5270 case _PC_TIMESTAMP_RESOLUTION:
5271 /* nanosecond timestamp resolution */
5275 case _PC_ACL_EXTENDED:
5283 case _PC_ACL_PATH_MAX:
5284 *valp = ACL_MAX_ENTRIES;
5288 return (EOPNOTSUPP);
5294 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5295 caller_context_t *ct)
5297 znode_t *zp = VTOZ(vp);
5298 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5300 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5304 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5312 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5313 caller_context_t *ct)
5315 znode_t *zp = VTOZ(vp);
5316 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5318 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5319 zilog_t *zilog = zfsvfs->z_log;
5324 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5326 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5327 zil_commit(zilog, 0);
5335 * The smallest read we may consider to loan out an arcbuf.
5336 * This must be a power of 2.
5338 int zcr_blksz_min = (1 << 10); /* 1K */
5340 * If set to less than the file block size, allow loaning out of an
5341 * arcbuf for a partial block read. This must be a power of 2.
5343 int zcr_blksz_max = (1 << 17); /* 128K */
5347 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5348 caller_context_t *ct)
5350 znode_t *zp = VTOZ(vp);
5351 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5352 int max_blksz = zfsvfs->z_max_blksz;
5353 uio_t *uio = &xuio->xu_uio;
5354 ssize_t size = uio->uio_resid;
5355 offset_t offset = uio->uio_loffset;
5360 int preamble, postamble;
5362 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5363 return (SET_ERROR(EINVAL));
5370 * Loan out an arc_buf for write if write size is bigger than
5371 * max_blksz, and the file's block size is also max_blksz.
5374 if (size < blksz || zp->z_blksz != blksz) {
5376 return (SET_ERROR(EINVAL));
5379 * Caller requests buffers for write before knowing where the
5380 * write offset might be (e.g. NFS TCP write).
5385 preamble = P2PHASE(offset, blksz);
5387 preamble = blksz - preamble;
5392 postamble = P2PHASE(size, blksz);
5395 fullblk = size / blksz;
5396 (void) dmu_xuio_init(xuio,
5397 (preamble != 0) + fullblk + (postamble != 0));
5398 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5399 int, postamble, int,
5400 (preamble != 0) + fullblk + (postamble != 0));
5403 * Have to fix iov base/len for partial buffers. They
5404 * currently represent full arc_buf's.
5407 /* data begins in the middle of the arc_buf */
5408 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5411 (void) dmu_xuio_add(xuio, abuf,
5412 blksz - preamble, preamble);
5415 for (i = 0; i < fullblk; i++) {
5416 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5419 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5423 /* data ends in the middle of the arc_buf */
5424 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5427 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5432 * Loan out an arc_buf for read if the read size is larger than
5433 * the current file block size. Block alignment is not
5434 * considered. Partial arc_buf will be loaned out for read.
5436 blksz = zp->z_blksz;
5437 if (blksz < zcr_blksz_min)
5438 blksz = zcr_blksz_min;
5439 if (blksz > zcr_blksz_max)
5440 blksz = zcr_blksz_max;
5441 /* avoid potential complexity of dealing with it */
5442 if (blksz > max_blksz) {
5444 return (SET_ERROR(EINVAL));
5447 maxsize = zp->z_size - uio->uio_loffset;
5451 if (size < blksz || vn_has_cached_data(vp)) {
5453 return (SET_ERROR(EINVAL));
5458 return (SET_ERROR(EINVAL));
5461 uio->uio_extflg = UIO_XUIO;
5462 XUIO_XUZC_RW(xuio) = ioflag;
5469 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5473 int ioflag = XUIO_XUZC_RW(xuio);
5475 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5477 i = dmu_xuio_cnt(xuio);
5479 abuf = dmu_xuio_arcbuf(xuio, i);
5481 * if abuf == NULL, it must be a write buffer
5482 * that has been returned in zfs_write().
5485 dmu_return_arcbuf(abuf);
5486 ASSERT(abuf || ioflag == UIO_WRITE);
5489 dmu_xuio_fini(xuio);
5494 * Predeclare these here so that the compiler assumes that
5495 * this is an "old style" function declaration that does
5496 * not include arguments => we won't get type mismatch errors
5497 * in the initializations that follow.
5499 static int zfs_inval();
5500 static int zfs_isdir();
5505 return (SET_ERROR(EINVAL));
5511 return (SET_ERROR(EISDIR));
5514 * Directory vnode operations template
5516 vnodeops_t *zfs_dvnodeops;
5517 const fs_operation_def_t zfs_dvnodeops_template[] = {
5518 VOPNAME_OPEN, { .vop_open = zfs_open },
5519 VOPNAME_CLOSE, { .vop_close = zfs_close },
5520 VOPNAME_READ, { .error = zfs_isdir },
5521 VOPNAME_WRITE, { .error = zfs_isdir },
5522 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5523 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5524 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5525 VOPNAME_ACCESS, { .vop_access = zfs_access },
5526 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5527 VOPNAME_CREATE, { .vop_create = zfs_create },
5528 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5529 VOPNAME_LINK, { .vop_link = zfs_link },
5530 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5531 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir },
5532 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5533 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5534 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink },
5535 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5536 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5537 VOPNAME_FID, { .vop_fid = zfs_fid },
5538 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5539 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5540 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5541 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5542 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5547 * Regular file vnode operations template
5549 vnodeops_t *zfs_fvnodeops;
5550 const fs_operation_def_t zfs_fvnodeops_template[] = {
5551 VOPNAME_OPEN, { .vop_open = zfs_open },
5552 VOPNAME_CLOSE, { .vop_close = zfs_close },
5553 VOPNAME_READ, { .vop_read = zfs_read },
5554 VOPNAME_WRITE, { .vop_write = zfs_write },
5555 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5556 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5557 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5558 VOPNAME_ACCESS, { .vop_access = zfs_access },
5559 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5560 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5561 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5562 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5563 VOPNAME_FID, { .vop_fid = zfs_fid },
5564 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5565 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock },
5566 VOPNAME_SPACE, { .vop_space = zfs_space },
5567 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage },
5568 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage },
5569 VOPNAME_MAP, { .vop_map = zfs_map },
5570 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap },
5571 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap },
5572 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5573 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5574 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5575 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5576 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf },
5577 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf },
5582 * Symbolic link vnode operations template
5584 vnodeops_t *zfs_symvnodeops;
5585 const fs_operation_def_t zfs_symvnodeops_template[] = {
5586 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5587 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5588 VOPNAME_ACCESS, { .vop_access = zfs_access },
5589 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5590 VOPNAME_READLINK, { .vop_readlink = zfs_readlink },
5591 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5592 VOPNAME_FID, { .vop_fid = zfs_fid },
5593 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5594 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5599 * special share hidden files vnode operations template
5601 vnodeops_t *zfs_sharevnodeops;
5602 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5603 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5604 VOPNAME_ACCESS, { .vop_access = zfs_access },
5605 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5606 VOPNAME_FID, { .vop_fid = zfs_fid },
5607 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5608 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5609 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5610 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5615 * Extended attribute directory vnode operations template
5617 * This template is identical to the directory vnodes
5618 * operation template except for restricted operations:
5622 * Note that there are other restrictions embedded in:
5623 * zfs_create() - restrict type to VREG
5624 * zfs_link() - no links into/out of attribute space
5625 * zfs_rename() - no moves into/out of attribute space
5627 vnodeops_t *zfs_xdvnodeops;
5628 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5629 VOPNAME_OPEN, { .vop_open = zfs_open },
5630 VOPNAME_CLOSE, { .vop_close = zfs_close },
5631 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5632 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5633 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5634 VOPNAME_ACCESS, { .vop_access = zfs_access },
5635 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5636 VOPNAME_CREATE, { .vop_create = zfs_create },
5637 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5638 VOPNAME_LINK, { .vop_link = zfs_link },
5639 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5640 VOPNAME_MKDIR, { .error = zfs_inval },
5641 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5642 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5643 VOPNAME_SYMLINK, { .error = zfs_inval },
5644 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5645 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5646 VOPNAME_FID, { .vop_fid = zfs_fid },
5647 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5648 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5649 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5650 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5651 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5656 * Error vnode operations template
5658 vnodeops_t *zfs_evnodeops;
5659 const fs_operation_def_t zfs_evnodeops_template[] = {
5660 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5661 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5667 ioflags(int ioflags)
5671 if (ioflags & IO_APPEND)
5673 if (ioflags & IO_NDELAY)
5675 if (ioflags & IO_SYNC)
5676 flags |= (FSYNC | FDSYNC | FRSYNC);
5682 zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage)
5684 znode_t *zp = VTOZ(vp);
5685 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5686 objset_t *os = zp->z_zfsvfs->z_os;
5697 pcount = round_page(count) / PAGE_SIZE;
5699 object = mreq->object;
5702 KASSERT(vp->v_object == object, ("mismatching object"));
5704 VM_OBJECT_LOCK(object);
5705 vm_page_lock_queues();
5706 for (i = 0; i < pcount; i++) {
5711 vm_page_unlock_queues();
5714 if (mreq->valid != VM_PAGE_BITS_ALL)
5715 vm_page_zero_invalid(mreq, TRUE);
5716 VM_OBJECT_UNLOCK(object);
5718 return (VM_PAGER_OK);
5721 PCPU_INC(cnt.v_vnodein);
5722 PCPU_INC(cnt.v_vnodepgsin);
5724 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) {
5725 VM_OBJECT_UNLOCK(object);
5727 return (VM_PAGER_BAD);
5731 if (IDX_TO_OFF(mreq->pindex) + size > object->un_pager.vnp.vnp_size)
5732 size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mreq->pindex);
5734 VM_OBJECT_UNLOCK(object);
5735 va = zfs_map_page(mreq, &sf);
5736 error = dmu_read(os, zp->z_id, IDX_TO_OFF(mreq->pindex),
5737 size, va, DMU_READ_PREFETCH);
5738 if (size != PAGE_SIZE)
5739 bzero(va + size, PAGE_SIZE - size);
5741 VM_OBJECT_LOCK(object);
5744 mreq->valid = VM_PAGE_BITS_ALL;
5745 KASSERT(mreq->dirty == 0, ("zfs_getpages: page %p is dirty", mreq));
5747 VM_OBJECT_UNLOCK(object);
5749 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
5751 return (error ? VM_PAGER_ERROR : VM_PAGER_OK);
5755 zfs_freebsd_getpages(ap)
5756 struct vop_getpages_args /* {
5761 vm_ooffset_t a_offset;
5765 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage));
5769 zfs_freebsd_open(ap)
5770 struct vop_open_args /* {
5773 struct ucred *a_cred;
5774 struct thread *a_td;
5777 vnode_t *vp = ap->a_vp;
5778 znode_t *zp = VTOZ(vp);
5781 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL);
5783 vnode_create_vobject(vp, zp->z_size, ap->a_td);
5788 zfs_freebsd_close(ap)
5789 struct vop_close_args /* {
5792 struct ucred *a_cred;
5793 struct thread *a_td;
5797 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL));
5801 zfs_freebsd_ioctl(ap)
5802 struct vop_ioctl_args /* {
5812 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data,
5813 ap->a_fflag, ap->a_cred, NULL, NULL));
5817 zfs_freebsd_read(ap)
5818 struct vop_read_args /* {
5822 struct ucred *a_cred;
5826 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5831 zfs_freebsd_write(ap)
5832 struct vop_write_args /* {
5836 struct ucred *a_cred;
5840 if (vn_rlimit_fsize(ap->a_vp, ap->a_uio, ap->a_uio->uio_td))
5843 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5848 zfs_freebsd_access(ap)
5849 struct vop_access_args /* {
5851 accmode_t a_accmode;
5852 struct ucred *a_cred;
5853 struct thread *a_td;
5860 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND,
5862 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND);
5864 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL);
5867 * VADMIN has to be handled by vaccess().
5870 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND);
5872 vnode_t *vp = ap->a_vp;
5873 znode_t *zp = VTOZ(vp);
5875 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
5876 zp->z_gid, accmode, ap->a_cred, NULL);
5884 zfs_freebsd_lookup(ap)
5885 struct vop_lookup_args /* {
5886 struct vnode *a_dvp;
5887 struct vnode **a_vpp;
5888 struct componentname *a_cnp;
5891 struct componentname *cnp = ap->a_cnp;
5892 char nm[NAME_MAX + 1];
5894 ASSERT(cnp->cn_namelen < sizeof(nm));
5895 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm)));
5897 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop,
5898 cnp->cn_cred, cnp->cn_thread, 0));
5902 zfs_freebsd_create(ap)
5903 struct vop_create_args /* {
5904 struct vnode *a_dvp;
5905 struct vnode **a_vpp;
5906 struct componentname *a_cnp;
5907 struct vattr *a_vap;
5910 struct componentname *cnp = ap->a_cnp;
5911 vattr_t *vap = ap->a_vap;
5914 ASSERT(cnp->cn_flags & SAVENAME);
5916 vattr_init_mask(vap);
5917 mode = vap->va_mode & ALLPERMS;
5919 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode,
5920 ap->a_vpp, cnp->cn_cred, cnp->cn_thread));
5924 zfs_freebsd_remove(ap)
5925 struct vop_remove_args /* {
5926 struct vnode *a_dvp;
5928 struct componentname *a_cnp;
5932 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
5934 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr,
5935 ap->a_cnp->cn_cred, NULL, 0));
5939 zfs_freebsd_mkdir(ap)
5940 struct vop_mkdir_args /* {
5941 struct vnode *a_dvp;
5942 struct vnode **a_vpp;
5943 struct componentname *a_cnp;
5944 struct vattr *a_vap;
5947 vattr_t *vap = ap->a_vap;
5949 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
5951 vattr_init_mask(vap);
5953 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp,
5954 ap->a_cnp->cn_cred, NULL, 0, NULL));
5958 zfs_freebsd_rmdir(ap)
5959 struct vop_rmdir_args /* {
5960 struct vnode *a_dvp;
5962 struct componentname *a_cnp;
5965 struct componentname *cnp = ap->a_cnp;
5967 ASSERT(cnp->cn_flags & SAVENAME);
5969 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0));
5973 zfs_freebsd_readdir(ap)
5974 struct vop_readdir_args /* {
5977 struct ucred *a_cred;
5984 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag,
5985 ap->a_ncookies, ap->a_cookies));
5989 zfs_freebsd_fsync(ap)
5990 struct vop_fsync_args /* {
5993 struct thread *a_td;
5998 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL));
6002 zfs_freebsd_getattr(ap)
6003 struct vop_getattr_args /* {
6005 struct vattr *a_vap;
6006 struct ucred *a_cred;
6009 vattr_t *vap = ap->a_vap;
6015 xvap.xva_vattr = *vap;
6016 xvap.xva_vattr.va_mask |= AT_XVATTR;
6018 /* Convert chflags into ZFS-type flags. */
6019 /* XXX: what about SF_SETTABLE?. */
6020 XVA_SET_REQ(&xvap, XAT_IMMUTABLE);
6021 XVA_SET_REQ(&xvap, XAT_APPENDONLY);
6022 XVA_SET_REQ(&xvap, XAT_NOUNLINK);
6023 XVA_SET_REQ(&xvap, XAT_NODUMP);
6024 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL);
6028 /* Convert ZFS xattr into chflags. */
6029 #define FLAG_CHECK(fflag, xflag, xfield) do { \
6030 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \
6031 fflags |= (fflag); \
6033 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE,
6034 xvap.xva_xoptattrs.xoa_immutable);
6035 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY,
6036 xvap.xva_xoptattrs.xoa_appendonly);
6037 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK,
6038 xvap.xva_xoptattrs.xoa_nounlink);
6039 FLAG_CHECK(UF_NODUMP, XAT_NODUMP,
6040 xvap.xva_xoptattrs.xoa_nodump);
6042 *vap = xvap.xva_vattr;
6043 vap->va_flags = fflags;
6048 zfs_freebsd_setattr(ap)
6049 struct vop_setattr_args /* {
6051 struct vattr *a_vap;
6052 struct ucred *a_cred;
6055 vnode_t *vp = ap->a_vp;
6056 vattr_t *vap = ap->a_vap;
6057 cred_t *cred = ap->a_cred;
6062 vattr_init_mask(vap);
6063 vap->va_mask &= ~AT_NOSET;
6066 xvap.xva_vattr = *vap;
6068 zflags = VTOZ(vp)->z_pflags;
6070 if (vap->va_flags != VNOVAL) {
6071 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs;
6074 if (zfsvfs->z_use_fuids == B_FALSE)
6075 return (EOPNOTSUPP);
6077 fflags = vap->va_flags;
6078 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_NODUMP)) != 0)
6079 return (EOPNOTSUPP);
6081 * Unprivileged processes are not permitted to unset system
6082 * flags, or modify flags if any system flags are set.
6083 * Privileged non-jail processes may not modify system flags
6084 * if securelevel > 0 and any existing system flags are set.
6085 * Privileged jail processes behave like privileged non-jail
6086 * processes if the security.jail.chflags_allowed sysctl is
6087 * is non-zero; otherwise, they behave like unprivileged
6090 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 ||
6091 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) {
6093 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6094 error = securelevel_gt(cred, 0);
6100 * Callers may only modify the file flags on objects they
6101 * have VADMIN rights for.
6103 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0)
6106 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6110 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) {
6115 #define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \
6116 if (((fflags & (fflag)) && !(zflags & (zflag))) || \
6117 ((zflags & (zflag)) && !(fflags & (fflag)))) { \
6118 XVA_SET_REQ(&xvap, (xflag)); \
6119 (xfield) = ((fflags & (fflag)) != 0); \
6122 /* Convert chflags into ZFS-type flags. */
6123 /* XXX: what about SF_SETTABLE?. */
6124 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE,
6125 xvap.xva_xoptattrs.xoa_immutable);
6126 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY,
6127 xvap.xva_xoptattrs.xoa_appendonly);
6128 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK,
6129 xvap.xva_xoptattrs.xoa_nounlink);
6130 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP,
6131 xvap.xva_xoptattrs.xoa_nodump);
6134 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL));
6138 zfs_freebsd_rename(ap)
6139 struct vop_rename_args /* {
6140 struct vnode *a_fdvp;
6141 struct vnode *a_fvp;
6142 struct componentname *a_fcnp;
6143 struct vnode *a_tdvp;
6144 struct vnode *a_tvp;
6145 struct componentname *a_tcnp;
6148 vnode_t *fdvp = ap->a_fdvp;
6149 vnode_t *fvp = ap->a_fvp;
6150 vnode_t *tdvp = ap->a_tdvp;
6151 vnode_t *tvp = ap->a_tvp;
6154 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART));
6155 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART));
6157 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp,
6158 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0);
6173 zfs_freebsd_symlink(ap)
6174 struct vop_symlink_args /* {
6175 struct vnode *a_dvp;
6176 struct vnode **a_vpp;
6177 struct componentname *a_cnp;
6178 struct vattr *a_vap;
6182 struct componentname *cnp = ap->a_cnp;
6183 vattr_t *vap = ap->a_vap;
6185 ASSERT(cnp->cn_flags & SAVENAME);
6187 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */
6188 vattr_init_mask(vap);
6190 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap,
6191 ap->a_target, cnp->cn_cred, cnp->cn_thread));
6195 zfs_freebsd_readlink(ap)
6196 struct vop_readlink_args /* {
6199 struct ucred *a_cred;
6203 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL));
6207 zfs_freebsd_link(ap)
6208 struct vop_link_args /* {
6209 struct vnode *a_tdvp;
6211 struct componentname *a_cnp;
6214 struct componentname *cnp = ap->a_cnp;
6216 ASSERT(cnp->cn_flags & SAVENAME);
6218 return (zfs_link(ap->a_tdvp, ap->a_vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0));
6222 zfs_freebsd_inactive(ap)
6223 struct vop_inactive_args /* {
6225 struct thread *a_td;
6228 vnode_t *vp = ap->a_vp;
6230 zfs_inactive(vp, ap->a_td->td_ucred, NULL);
6235 zfs_freebsd_reclaim(ap)
6236 struct vop_reclaim_args /* {
6238 struct thread *a_td;
6241 vnode_t *vp = ap->a_vp;
6242 znode_t *zp = VTOZ(vp);
6243 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
6247 /* Destroy the vm object and flush associated pages. */
6248 vnode_destroy_vobject(vp);
6251 * z_teardown_inactive_lock protects from a race with
6252 * zfs_znode_dmu_fini in zfsvfs_teardown during
6255 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
6256 if (zp->z_sa_hdl == NULL)
6260 rw_exit(&zfsvfs->z_teardown_inactive_lock);
6268 struct vop_fid_args /* {
6274 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL));
6278 zfs_freebsd_pathconf(ap)
6279 struct vop_pathconf_args /* {
6282 register_t *a_retval;
6288 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL);
6290 *ap->a_retval = val;
6291 else if (error == EOPNOTSUPP)
6292 error = vop_stdpathconf(ap);
6297 zfs_freebsd_fifo_pathconf(ap)
6298 struct vop_pathconf_args /* {
6301 register_t *a_retval;
6305 switch (ap->a_name) {
6306 case _PC_ACL_EXTENDED:
6308 case _PC_ACL_PATH_MAX:
6309 case _PC_MAC_PRESENT:
6310 return (zfs_freebsd_pathconf(ap));
6312 return (fifo_specops.vop_pathconf(ap));
6317 * FreeBSD's extended attributes namespace defines file name prefix for ZFS'
6318 * extended attribute name:
6321 * system freebsd:system:
6322 * user (none, can be used to access ZFS fsattr(5) attributes
6323 * created on Solaris)
6326 zfs_create_attrname(int attrnamespace, const char *name, char *attrname,
6329 const char *namespace, *prefix, *suffix;
6331 /* We don't allow '/' character in attribute name. */
6332 if (strchr(name, '/') != NULL)
6334 /* We don't allow attribute names that start with "freebsd:" string. */
6335 if (strncmp(name, "freebsd:", 8) == 0)
6338 bzero(attrname, size);
6340 switch (attrnamespace) {
6341 case EXTATTR_NAMESPACE_USER:
6343 prefix = "freebsd:";
6344 namespace = EXTATTR_NAMESPACE_USER_STRING;
6348 * This is the default namespace by which we can access all
6349 * attributes created on Solaris.
6351 prefix = namespace = suffix = "";
6354 case EXTATTR_NAMESPACE_SYSTEM:
6355 prefix = "freebsd:";
6356 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING;
6359 case EXTATTR_NAMESPACE_EMPTY:
6363 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix,
6365 return (ENAMETOOLONG);
6371 * Vnode operating to retrieve a named extended attribute.
6374 zfs_getextattr(struct vop_getextattr_args *ap)
6377 IN struct vnode *a_vp;
6378 IN int a_attrnamespace;
6379 IN const char *a_name;
6380 INOUT struct uio *a_uio;
6382 IN struct ucred *a_cred;
6383 IN struct thread *a_td;
6387 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6388 struct thread *td = ap->a_td;
6389 struct nameidata nd;
6392 vnode_t *xvp = NULL, *vp;
6395 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6396 ap->a_cred, ap->a_td, VREAD);
6400 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6407 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6415 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_SYSSPACE, attrname,
6417 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL);
6419 NDFREE(&nd, NDF_ONLY_PNBUF);
6422 if (error == ENOENT)
6427 if (ap->a_size != NULL) {
6428 error = VOP_GETATTR(vp, &va, ap->a_cred);
6430 *ap->a_size = (size_t)va.va_size;
6431 } else if (ap->a_uio != NULL)
6432 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6435 vn_close(vp, flags, ap->a_cred, td);
6442 * Vnode operation to remove a named attribute.
6445 zfs_deleteextattr(struct vop_deleteextattr_args *ap)
6448 IN struct vnode *a_vp;
6449 IN int a_attrnamespace;
6450 IN const char *a_name;
6451 IN struct ucred *a_cred;
6452 IN struct thread *a_td;
6456 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6457 struct thread *td = ap->a_td;
6458 struct nameidata nd;
6461 vnode_t *xvp = NULL, *vp;
6464 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6465 ap->a_cred, ap->a_td, VWRITE);
6469 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6476 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6483 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF | MPSAFE,
6484 UIO_SYSSPACE, attrname, xvp, td);
6489 NDFREE(&nd, NDF_ONLY_PNBUF);
6490 if (error == ENOENT)
6495 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
6496 NDFREE(&nd, NDF_ONLY_PNBUF);
6499 if (vp == nd.ni_dvp)
6509 * Vnode operation to set a named attribute.
6512 zfs_setextattr(struct vop_setextattr_args *ap)
6515 IN struct vnode *a_vp;
6516 IN int a_attrnamespace;
6517 IN const char *a_name;
6518 INOUT struct uio *a_uio;
6519 IN struct ucred *a_cred;
6520 IN struct thread *a_td;
6524 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6525 struct thread *td = ap->a_td;
6526 struct nameidata nd;
6529 vnode_t *xvp = NULL, *vp;
6532 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6533 ap->a_cred, ap->a_td, VWRITE);
6537 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6544 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6545 LOOKUP_XATTR | CREATE_XATTR_DIR);
6551 flags = FFLAGS(O_WRONLY | O_CREAT);
6552 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_SYSSPACE, attrname,
6554 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL);
6556 NDFREE(&nd, NDF_ONLY_PNBUF);
6564 error = VOP_SETATTR(vp, &va, ap->a_cred);
6566 VOP_WRITE(vp, ap->a_uio, IO_UNIT | IO_SYNC, ap->a_cred);
6569 vn_close(vp, flags, ap->a_cred, td);
6576 * Vnode operation to retrieve extended attributes on a vnode.
6579 zfs_listextattr(struct vop_listextattr_args *ap)
6582 IN struct vnode *a_vp;
6583 IN int a_attrnamespace;
6584 INOUT struct uio *a_uio;
6586 IN struct ucred *a_cred;
6587 IN struct thread *a_td;
6591 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6592 struct thread *td = ap->a_td;
6593 struct nameidata nd;
6594 char attrprefix[16];
6595 u_char dirbuf[sizeof(struct dirent)];
6598 struct uio auio, *uio = ap->a_uio;
6599 size_t *sizep = ap->a_size;
6601 vnode_t *xvp = NULL, *vp;
6602 int done, error, eof, pos;
6604 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6605 ap->a_cred, ap->a_td, VREAD);
6609 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix,
6610 sizeof(attrprefix));
6613 plen = strlen(attrprefix);
6620 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6625 * ENOATTR means that the EA directory does not yet exist,
6626 * i.e. there are no extended attributes there.
6628 if (error == ENOATTR)
6633 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED | MPSAFE,
6634 UIO_SYSSPACE, ".", xvp, td);
6637 NDFREE(&nd, NDF_ONLY_PNBUF);
6643 auio.uio_iov = &aiov;
6644 auio.uio_iovcnt = 1;
6645 auio.uio_segflg = UIO_SYSSPACE;
6647 auio.uio_rw = UIO_READ;
6648 auio.uio_offset = 0;
6653 aiov.iov_base = (void *)dirbuf;
6654 aiov.iov_len = sizeof(dirbuf);
6655 auio.uio_resid = sizeof(dirbuf);
6656 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL);
6657 done = sizeof(dirbuf) - auio.uio_resid;
6660 for (pos = 0; pos < done;) {
6661 dp = (struct dirent *)(dirbuf + pos);
6662 pos += dp->d_reclen;
6664 * XXX: Temporarily we also accept DT_UNKNOWN, as this
6665 * is what we get when attribute was created on Solaris.
6667 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN)
6669 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0)
6671 else if (strncmp(dp->d_name, attrprefix, plen) != 0)
6673 nlen = dp->d_namlen - plen;
6676 else if (uio != NULL) {
6678 * Format of extattr name entry is one byte for
6679 * length and the rest for name.
6681 error = uiomove(&nlen, 1, uio->uio_rw, uio);
6683 error = uiomove(dp->d_name + plen, nlen,
6690 } while (!eof && error == 0);
6699 zfs_freebsd_getacl(ap)
6700 struct vop_getacl_args /* {
6709 vsecattr_t vsecattr;
6711 if (ap->a_type != ACL_TYPE_NFS4)
6714 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
6715 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL))
6718 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt);
6719 if (vsecattr.vsa_aclentp != NULL)
6720 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz);
6726 zfs_freebsd_setacl(ap)
6727 struct vop_setacl_args /* {
6736 vsecattr_t vsecattr;
6737 int aclbsize; /* size of acl list in bytes */
6740 if (ap->a_type != ACL_TYPE_NFS4)
6743 if (ap->a_aclp == NULL)
6746 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES)
6750 * With NFSv4 ACLs, chmod(2) may need to add additional entries,
6751 * splitting every entry into two and appending "canonical six"
6752 * entries at the end. Don't allow for setting an ACL that would
6753 * cause chmod(2) to run out of ACL entries.
6755 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES)
6758 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR);
6762 vsecattr.vsa_mask = VSA_ACE;
6763 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t);
6764 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP);
6765 aaclp = vsecattr.vsa_aclentp;
6766 vsecattr.vsa_aclentsz = aclbsize;
6768 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp);
6769 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL);
6770 kmem_free(aaclp, aclbsize);
6776 zfs_freebsd_aclcheck(ap)
6777 struct vop_aclcheck_args /* {
6786 return (EOPNOTSUPP);
6789 struct vop_vector zfs_vnodeops;
6790 struct vop_vector zfs_fifoops;
6791 struct vop_vector zfs_shareops;
6793 struct vop_vector zfs_vnodeops = {
6794 .vop_default = &default_vnodeops,
6795 .vop_inactive = zfs_freebsd_inactive,
6796 .vop_reclaim = zfs_freebsd_reclaim,
6797 .vop_access = zfs_freebsd_access,
6798 #ifdef FREEBSD_NAMECACHE
6799 .vop_lookup = vfs_cache_lookup,
6800 .vop_cachedlookup = zfs_freebsd_lookup,
6802 .vop_lookup = zfs_freebsd_lookup,
6804 .vop_getattr = zfs_freebsd_getattr,
6805 .vop_setattr = zfs_freebsd_setattr,
6806 .vop_create = zfs_freebsd_create,
6807 .vop_mknod = zfs_freebsd_create,
6808 .vop_mkdir = zfs_freebsd_mkdir,
6809 .vop_readdir = zfs_freebsd_readdir,
6810 .vop_fsync = zfs_freebsd_fsync,
6811 .vop_open = zfs_freebsd_open,
6812 .vop_close = zfs_freebsd_close,
6813 .vop_rmdir = zfs_freebsd_rmdir,
6814 .vop_ioctl = zfs_freebsd_ioctl,
6815 .vop_link = zfs_freebsd_link,
6816 .vop_symlink = zfs_freebsd_symlink,
6817 .vop_readlink = zfs_freebsd_readlink,
6818 .vop_read = zfs_freebsd_read,
6819 .vop_write = zfs_freebsd_write,
6820 .vop_remove = zfs_freebsd_remove,
6821 .vop_rename = zfs_freebsd_rename,
6822 .vop_pathconf = zfs_freebsd_pathconf,
6823 .vop_bmap = VOP_EOPNOTSUPP,
6824 .vop_fid = zfs_freebsd_fid,
6825 .vop_getextattr = zfs_getextattr,
6826 .vop_deleteextattr = zfs_deleteextattr,
6827 .vop_setextattr = zfs_setextattr,
6828 .vop_listextattr = zfs_listextattr,
6829 .vop_getacl = zfs_freebsd_getacl,
6830 .vop_setacl = zfs_freebsd_setacl,
6831 .vop_aclcheck = zfs_freebsd_aclcheck,
6832 .vop_getpages = zfs_freebsd_getpages,
6835 struct vop_vector zfs_fifoops = {
6836 .vop_default = &fifo_specops,
6837 .vop_fsync = zfs_freebsd_fsync,
6838 .vop_access = zfs_freebsd_access,
6839 .vop_getattr = zfs_freebsd_getattr,
6840 .vop_inactive = zfs_freebsd_inactive,
6841 .vop_read = VOP_PANIC,
6842 .vop_reclaim = zfs_freebsd_reclaim,
6843 .vop_setattr = zfs_freebsd_setattr,
6844 .vop_write = VOP_PANIC,
6845 .vop_pathconf = zfs_freebsd_fifo_pathconf,
6846 .vop_fid = zfs_freebsd_fid,
6847 .vop_getacl = zfs_freebsd_getacl,
6848 .vop_setacl = zfs_freebsd_setacl,
6849 .vop_aclcheck = zfs_freebsd_aclcheck,
6853 * special share hidden files vnode operations template
6855 struct vop_vector zfs_shareops = {
6856 .vop_default = &default_vnodeops,
6857 .vop_access = zfs_freebsd_access,
6858 .vop_inactive = zfs_freebsd_inactive,
6859 .vop_reclaim = zfs_freebsd_reclaim,
6860 .vop_fid = zfs_freebsd_fid,
6861 .vop_pathconf = zfs_freebsd_pathconf,