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>
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_param.h>
77 #include <vm/vm_pageout.h>
82 * Each vnode op performs some logical unit of work. To do this, the ZPL must
83 * properly lock its in-core state, create a DMU transaction, do the work,
84 * record this work in the intent log (ZIL), commit the DMU transaction,
85 * and wait for the intent log to commit if it is a synchronous operation.
86 * Moreover, the vnode ops must work in both normal and log replay context.
87 * The ordering of events is important to avoid deadlocks and references
88 * to freed memory. The example below illustrates the following Big Rules:
90 * (1) A check must be made in each zfs thread for a mounted file system.
91 * This is done avoiding races using ZFS_ENTER(zfsvfs).
92 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
93 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
94 * can return EIO from the calling function.
96 * (2) VN_RELE() should always be the last thing except for zil_commit()
97 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
98 * First, if it's the last reference, the vnode/znode
99 * can be freed, so the zp may point to freed memory. Second, the last
100 * reference will call zfs_zinactive(), which may induce a lot of work --
101 * pushing cached pages (which acquires range locks) and syncing out
102 * cached atime changes. Third, zfs_zinactive() may require a new tx,
103 * which could deadlock the system if you were already holding one.
104 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
106 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
107 * as they can span dmu_tx_assign() calls.
109 * (4) Always pass TXG_NOWAIT as the second argument to dmu_tx_assign().
110 * This is critical because we don't want to block while holding locks.
111 * Note, in particular, that if a lock is sometimes acquired before
112 * the tx assigns, and sometimes after (e.g. z_lock), then failing to
113 * use a non-blocking assign can deadlock the system. The scenario:
115 * Thread A has grabbed a lock before calling dmu_tx_assign().
116 * Thread B is in an already-assigned tx, and blocks for this lock.
117 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
118 * forever, because the previous txg can't quiesce until B's tx commits.
120 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
121 * then drop all locks, call dmu_tx_wait(), and try again.
123 * (5) If the operation succeeded, generate the intent log entry for it
124 * before dropping locks. This ensures that the ordering of events
125 * in the intent log matches the order in which they actually occurred.
126 * During ZIL replay the zfs_log_* functions will update the sequence
127 * number to indicate the zil transaction has replayed.
129 * (6) At the end of each vnode op, the DMU tx must always commit,
130 * regardless of whether there were any errors.
132 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
133 * to ensure that synchronous semantics are provided when necessary.
135 * In general, this is how things should be ordered in each vnode op:
137 * ZFS_ENTER(zfsvfs); // exit if unmounted
139 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD())
140 * rw_enter(...); // grab any other locks you need
141 * tx = dmu_tx_create(...); // get DMU tx
142 * dmu_tx_hold_*(); // hold each object you might modify
143 * error = dmu_tx_assign(tx, TXG_NOWAIT); // try to assign
145 * rw_exit(...); // drop locks
146 * zfs_dirent_unlock(dl); // unlock directory entry
147 * VN_RELE(...); // release held vnodes
148 * if (error == ERESTART) {
153 * dmu_tx_abort(tx); // abort DMU tx
154 * ZFS_EXIT(zfsvfs); // finished in zfs
155 * return (error); // really out of space
157 * error = do_real_work(); // do whatever this VOP does
159 * zfs_log_*(...); // on success, make ZIL entry
160 * dmu_tx_commit(tx); // commit DMU tx -- error or not
161 * rw_exit(...); // drop locks
162 * zfs_dirent_unlock(dl); // unlock directory entry
163 * VN_RELE(...); // release held vnodes
164 * zil_commit(zilog, foid); // synchronous when necessary
165 * ZFS_EXIT(zfsvfs); // finished in zfs
166 * return (error); // done, report error
171 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
173 znode_t *zp = VTOZ(*vpp);
174 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
179 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
180 ((flag & FAPPEND) == 0)) {
182 return (SET_ERROR(EPERM));
185 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
186 ZTOV(zp)->v_type == VREG &&
187 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
188 if (fs_vscan(*vpp, cr, 0) != 0) {
190 return (SET_ERROR(EACCES));
194 /* Keep a count of the synchronous opens in the znode */
195 if (flag & (FSYNC | FDSYNC))
196 atomic_inc_32(&zp->z_sync_cnt);
204 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
205 caller_context_t *ct)
207 znode_t *zp = VTOZ(vp);
208 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
211 * Clean up any locks held by this process on the vp.
213 cleanlocks(vp, ddi_get_pid(), 0);
214 cleanshares(vp, ddi_get_pid());
219 /* Decrement the synchronous opens in the znode */
220 if ((flag & (FSYNC | FDSYNC)) && (count == 1))
221 atomic_dec_32(&zp->z_sync_cnt);
223 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
224 ZTOV(zp)->v_type == VREG &&
225 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
226 VERIFY(fs_vscan(vp, cr, 1) == 0);
233 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
234 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
237 zfs_holey(vnode_t *vp, u_long cmd, offset_t *off)
239 znode_t *zp = VTOZ(vp);
240 uint64_t noff = (uint64_t)*off; /* new offset */
245 file_sz = zp->z_size;
246 if (noff >= file_sz) {
247 return (SET_ERROR(ENXIO));
250 if (cmd == _FIO_SEEK_HOLE)
255 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
258 if ((error == ESRCH) || (noff > file_sz)) {
260 * Handle the virtual hole at the end of file.
266 return (SET_ERROR(ENXIO));
277 zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred,
278 int *rvalp, caller_context_t *ct)
290 * The following two ioctls are used by bfu. Faking out,
291 * necessary to avoid bfu errors.
300 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
301 return (SET_ERROR(EFAULT));
303 off = *(offset_t *)data;
306 zfsvfs = zp->z_zfsvfs;
310 /* offset parameter is in/out */
311 error = zfs_holey(vp, com, &off);
316 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
317 return (SET_ERROR(EFAULT));
319 *(offset_t *)data = off;
323 return (SET_ERROR(ENOTTY));
327 page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes)
333 zfs_vmobject_assert_wlocked(obj);
336 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
338 if ((pp->oflags & VPO_BUSY) != 0) {
340 * Reference the page before unlocking and
341 * sleeping so that the page daemon is less
342 * likely to reclaim it.
344 vm_page_reference(pp);
345 vm_page_sleep(pp, "zfsmwb");
348 } else if (pp == NULL) {
349 pp = vm_page_alloc(obj, OFF_TO_IDX(start),
350 VM_ALLOC_SYSTEM | VM_ALLOC_IFCACHED |
353 ASSERT(pp != NULL && !pp->valid);
358 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
359 vm_object_pip_add(obj, 1);
360 vm_page_io_start(pp);
361 pmap_remove_write(pp);
362 vm_page_clear_dirty(pp, off, nbytes);
370 page_unbusy(vm_page_t pp)
373 vm_page_io_finish(pp);
374 vm_object_pip_subtract(pp->object, 1);
378 page_hold(vnode_t *vp, int64_t start)
384 zfs_vmobject_assert_wlocked(obj);
387 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
389 if ((pp->oflags & VPO_BUSY) != 0) {
391 * Reference the page before unlocking and
392 * sleeping so that the page daemon is less
393 * likely to reclaim it.
395 vm_page_reference(pp);
396 vm_page_sleep(pp, "zfsmwb");
400 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
413 page_unhold(vm_page_t pp)
422 zfs_map_page(vm_page_t pp, struct sf_buf **sfp)
425 *sfp = sf_buf_alloc(pp, 0);
426 return ((caddr_t)sf_buf_kva(*sfp));
430 zfs_unmap_page(struct sf_buf *sf)
437 * When a file is memory mapped, we must keep the IO data synchronized
438 * between the DMU cache and the memory mapped pages. What this means:
440 * On Write: If we find a memory mapped page, we write to *both*
441 * the page and the dmu buffer.
444 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid,
445 int segflg, dmu_tx_t *tx)
452 ASSERT(vp->v_mount != NULL);
456 off = start & PAGEOFFSET;
457 zfs_vmobject_wlock(obj);
458 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
460 int nbytes = imin(PAGESIZE - off, len);
462 if (segflg == UIO_NOCOPY) {
463 pp = vm_page_lookup(obj, OFF_TO_IDX(start));
465 ("zfs update_pages: NULL page in putpages case"));
467 ("zfs update_pages: unaligned data in putpages case"));
468 KASSERT(pp->valid == VM_PAGE_BITS_ALL,
469 ("zfs update_pages: invalid page in putpages case"));
470 KASSERT(pp->busy > 0,
471 ("zfs update_pages: unbusy page in putpages case"));
472 KASSERT(!pmap_page_is_write_mapped(pp),
473 ("zfs update_pages: writable page in putpages case"));
474 zfs_vmobject_wunlock(obj);
476 va = zfs_map_page(pp, &sf);
477 (void) dmu_write(os, oid, start, nbytes, va, tx);
480 zfs_vmobject_wlock(obj);
482 } else if ((pp = page_busy(vp, start, off, nbytes)) != NULL) {
483 zfs_vmobject_wunlock(obj);
485 va = zfs_map_page(pp, &sf);
486 (void) dmu_read(os, oid, start+off, nbytes,
487 va+off, DMU_READ_PREFETCH);;
490 zfs_vmobject_wlock(obj);
496 if (segflg != UIO_NOCOPY)
497 vm_object_pip_wakeupn(obj, 0);
498 zfs_vmobject_wunlock(obj);
502 * Read with UIO_NOCOPY flag means that sendfile(2) requests
503 * ZFS to populate a range of page cache pages with data.
505 * NOTE: this function could be optimized to pre-allocate
506 * all pages in advance, drain VPO_BUSY on all of them,
507 * map them into contiguous KVA region and populate them
508 * in one single dmu_read() call.
511 mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio)
513 znode_t *zp = VTOZ(vp);
514 objset_t *os = zp->z_zfsvfs->z_os;
524 ASSERT(uio->uio_segflg == UIO_NOCOPY);
525 ASSERT(vp->v_mount != NULL);
528 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0);
530 zfs_vmobject_wlock(obj);
531 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) {
532 int bytes = MIN(PAGESIZE, len);
534 pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_NOBUSY |
535 VM_ALLOC_NORMAL | VM_ALLOC_RETRY | VM_ALLOC_IGN_SBUSY);
536 if (pp->valid == 0) {
537 vm_page_io_start(pp);
538 zfs_vmobject_wunlock(obj);
539 va = zfs_map_page(pp, &sf);
540 error = dmu_read(os, zp->z_id, start, bytes, va,
542 if (bytes != PAGESIZE && error == 0)
543 bzero(va + bytes, PAGESIZE - bytes);
545 zfs_vmobject_wlock(obj);
546 vm_page_io_finish(pp);
549 if (pp->wire_count == 0 && pp->valid == 0 &&
550 pp->busy == 0 && !(pp->oflags & VPO_BUSY))
553 pp->valid = VM_PAGE_BITS_ALL;
554 vm_page_activate(pp);
560 uio->uio_resid -= bytes;
561 uio->uio_offset += bytes;
564 zfs_vmobject_wunlock(obj);
569 * When a file is memory mapped, we must keep the IO data synchronized
570 * between the DMU cache and the memory mapped pages. What this means:
572 * On Read: We "read" preferentially from memory mapped pages,
573 * else we default from the dmu buffer.
575 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
576 * the file is memory mapped.
579 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
581 znode_t *zp = VTOZ(vp);
582 objset_t *os = zp->z_zfsvfs->z_os;
590 ASSERT(vp->v_mount != NULL);
594 start = uio->uio_loffset;
595 off = start & PAGEOFFSET;
596 zfs_vmobject_wlock(obj);
597 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
599 uint64_t bytes = MIN(PAGESIZE - off, len);
601 if (pp = page_hold(vp, start)) {
605 zfs_vmobject_wunlock(obj);
606 va = zfs_map_page(pp, &sf);
607 error = uiomove(va + off, bytes, UIO_READ, uio);
609 zfs_vmobject_wlock(obj);
612 zfs_vmobject_wunlock(obj);
613 error = dmu_read_uio(os, zp->z_id, uio, bytes);
614 zfs_vmobject_wlock(obj);
621 zfs_vmobject_wunlock(obj);
625 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
628 * Read bytes from specified file into supplied buffer.
630 * IN: vp - vnode of file to be read from.
631 * uio - structure supplying read location, range info,
633 * ioflag - SYNC flags; used to provide FRSYNC semantics.
634 * cr - credentials of caller.
635 * ct - caller context
637 * OUT: uio - updated offset and range, buffer filled.
639 * RETURN: 0 on success, error code on failure.
642 * vp - atime updated if byte count > 0
646 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
648 znode_t *zp = VTOZ(vp);
649 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
660 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
662 return (SET_ERROR(EACCES));
666 * Validate file offset
668 if (uio->uio_loffset < (offset_t)0) {
670 return (SET_ERROR(EINVAL));
674 * Fasttrack empty reads
676 if (uio->uio_resid == 0) {
682 * Check for mandatory locks
684 if (MANDMODE(zp->z_mode)) {
685 if (error = chklock(vp, FREAD,
686 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
693 * If we're in FRSYNC mode, sync out this znode before reading it.
696 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
697 zil_commit(zfsvfs->z_log, zp->z_id);
700 * Lock the range against changes.
702 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
705 * If we are reading past end-of-file we can skip
706 * to the end; but we might still need to set atime.
708 if (uio->uio_loffset >= zp->z_size) {
713 ASSERT(uio->uio_loffset < zp->z_size);
714 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
717 if ((uio->uio_extflg == UIO_XUIO) &&
718 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
720 int blksz = zp->z_blksz;
721 uint64_t offset = uio->uio_loffset;
723 xuio = (xuio_t *)uio;
725 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
728 ASSERT(offset + n <= blksz);
731 (void) dmu_xuio_init(xuio, nblk);
733 if (vn_has_cached_data(vp)) {
735 * For simplicity, we always allocate a full buffer
736 * even if we only expect to read a portion of a block.
738 while (--nblk >= 0) {
739 (void) dmu_xuio_add(xuio,
740 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
748 nbytes = MIN(n, zfs_read_chunk_size -
749 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
752 if (uio->uio_segflg == UIO_NOCOPY)
753 error = mappedread_sf(vp, nbytes, uio);
755 #endif /* __FreeBSD__ */
756 if (vn_has_cached_data(vp))
757 error = mappedread(vp, nbytes, uio);
759 error = dmu_read_uio(os, zp->z_id, uio, nbytes);
761 /* convert checksum errors into IO errors */
763 error = SET_ERROR(EIO);
770 zfs_range_unlock(rl);
772 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
778 * Write the bytes to a file.
780 * IN: vp - vnode of file to be written to.
781 * uio - structure supplying write location, range info,
783 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is
784 * set if in append mode.
785 * cr - credentials of caller.
786 * ct - caller context (NFS/CIFS fem monitor only)
788 * OUT: uio - updated offset and range.
790 * RETURN: 0 on success, error code on failure.
793 * vp - ctime|mtime updated if byte count > 0
798 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
800 znode_t *zp = VTOZ(vp);
801 rlim64_t limit = MAXOFFSET_T;
802 ssize_t start_resid = uio->uio_resid;
806 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
811 int max_blksz = zfsvfs->z_max_blksz;
814 iovec_t *aiov = NULL;
817 int iovcnt = uio->uio_iovcnt;
818 iovec_t *iovp = uio->uio_iov;
821 sa_bulk_attr_t bulk[4];
822 uint64_t mtime[2], ctime[2];
825 * Fasttrack empty write
831 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
837 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
838 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
839 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
841 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
845 * If immutable or not appending then return EPERM
847 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
848 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
849 (uio->uio_loffset < zp->z_size))) {
851 return (SET_ERROR(EPERM));
854 zilog = zfsvfs->z_log;
857 * Validate file offset
859 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
862 return (SET_ERROR(EINVAL));
866 * Check for mandatory locks before calling zfs_range_lock()
867 * in order to prevent a deadlock with locks set via fcntl().
869 if (MANDMODE((mode_t)zp->z_mode) &&
870 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
877 * Pre-fault the pages to ensure slow (eg NFS) pages
879 * Skip this if uio contains loaned arc_buf.
881 if ((uio->uio_extflg == UIO_XUIO) &&
882 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
883 xuio = (xuio_t *)uio;
885 uio_prefaultpages(MIN(n, max_blksz), uio);
889 * If in append mode, set the io offset pointer to eof.
891 if (ioflag & FAPPEND) {
893 * Obtain an appending range lock to guarantee file append
894 * semantics. We reset the write offset once we have the lock.
896 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
898 if (rl->r_len == UINT64_MAX) {
900 * We overlocked the file because this write will cause
901 * the file block size to increase.
902 * Note that zp_size cannot change with this lock held.
906 uio->uio_loffset = woff;
909 * Note that if the file block size will change as a result of
910 * this write, then this range lock will lock the entire file
911 * so that we can re-write the block safely.
913 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
916 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) {
917 zfs_range_unlock(rl);
923 zfs_range_unlock(rl);
925 return (SET_ERROR(EFBIG));
928 if ((woff + n) > limit || woff > (limit - n))
931 /* Will this write extend the file length? */
932 write_eof = (woff + n > zp->z_size);
934 end_size = MAX(zp->z_size, woff + n);
937 * Write the file in reasonable size chunks. Each chunk is written
938 * in a separate transaction; this keeps the intent log records small
939 * and allows us to do more fine-grained space accounting.
943 woff = uio->uio_loffset;
945 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
946 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
948 dmu_return_arcbuf(abuf);
949 error = SET_ERROR(EDQUOT);
953 if (xuio && abuf == NULL) {
954 ASSERT(i_iov < iovcnt);
956 abuf = dmu_xuio_arcbuf(xuio, i_iov);
957 dmu_xuio_clear(xuio, i_iov);
958 DTRACE_PROBE3(zfs_cp_write, int, i_iov,
959 iovec_t *, aiov, arc_buf_t *, abuf);
960 ASSERT((aiov->iov_base == abuf->b_data) ||
961 ((char *)aiov->iov_base - (char *)abuf->b_data +
962 aiov->iov_len == arc_buf_size(abuf)));
964 } else if (abuf == NULL && n >= max_blksz &&
965 woff >= zp->z_size &&
966 P2PHASE(woff, max_blksz) == 0 &&
967 zp->z_blksz == max_blksz) {
969 * This write covers a full block. "Borrow" a buffer
970 * from the dmu so that we can fill it before we enter
971 * a transaction. This avoids the possibility of
972 * holding up the transaction if the data copy hangs
973 * up on a pagefault (e.g., from an NFS server mapping).
977 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
979 ASSERT(abuf != NULL);
980 ASSERT(arc_buf_size(abuf) == max_blksz);
981 if (error = uiocopy(abuf->b_data, max_blksz,
982 UIO_WRITE, uio, &cbytes)) {
983 dmu_return_arcbuf(abuf);
986 ASSERT(cbytes == max_blksz);
990 * Start a transaction.
992 tx = dmu_tx_create(zfsvfs->z_os);
993 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
994 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
995 zfs_sa_upgrade_txholds(tx, zp);
996 error = dmu_tx_assign(tx, TXG_NOWAIT);
998 if (error == ERESTART) {
1005 dmu_return_arcbuf(abuf);
1010 * If zfs_range_lock() over-locked we grow the blocksize
1011 * and then reduce the lock range. This will only happen
1012 * on the first iteration since zfs_range_reduce() will
1013 * shrink down r_len to the appropriate size.
1015 if (rl->r_len == UINT64_MAX) {
1018 if (zp->z_blksz > max_blksz) {
1019 ASSERT(!ISP2(zp->z_blksz));
1020 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
1022 new_blksz = MIN(end_size, max_blksz);
1024 zfs_grow_blocksize(zp, new_blksz, tx);
1025 zfs_range_reduce(rl, woff, n);
1029 * XXX - should we really limit each write to z_max_blksz?
1030 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
1032 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
1034 if (woff + nbytes > zp->z_size)
1035 vnode_pager_setsize(vp, woff + nbytes);
1038 tx_bytes = uio->uio_resid;
1039 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
1041 tx_bytes -= uio->uio_resid;
1044 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
1046 * If this is not a full block write, but we are
1047 * extending the file past EOF and this data starts
1048 * block-aligned, use assign_arcbuf(). Otherwise,
1049 * write via dmu_write().
1051 if (tx_bytes < max_blksz && (!write_eof ||
1052 aiov->iov_base != abuf->b_data)) {
1054 dmu_write(zfsvfs->z_os, zp->z_id, woff,
1055 aiov->iov_len, aiov->iov_base, tx);
1056 dmu_return_arcbuf(abuf);
1057 xuio_stat_wbuf_copied();
1059 ASSERT(xuio || tx_bytes == max_blksz);
1060 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
1063 ASSERT(tx_bytes <= uio->uio_resid);
1064 uioskip(uio, tx_bytes);
1066 if (tx_bytes && vn_has_cached_data(vp)) {
1067 update_pages(vp, woff, tx_bytes, zfsvfs->z_os,
1068 zp->z_id, uio->uio_segflg, tx);
1072 * If we made no progress, we're done. If we made even
1073 * partial progress, update the znode and ZIL accordingly.
1075 if (tx_bytes == 0) {
1076 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
1077 (void *)&zp->z_size, sizeof (uint64_t), tx);
1084 * Clear Set-UID/Set-GID bits on successful write if not
1085 * privileged and at least one of the excute bits is set.
1087 * It would be nice to to this after all writes have
1088 * been done, but that would still expose the ISUID/ISGID
1089 * to another app after the partial write is committed.
1091 * Note: we don't call zfs_fuid_map_id() here because
1092 * user 0 is not an ephemeral uid.
1094 mutex_enter(&zp->z_acl_lock);
1095 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
1096 (S_IXUSR >> 6))) != 0 &&
1097 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
1098 secpolicy_vnode_setid_retain(vp, cr,
1099 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
1101 zp->z_mode &= ~(S_ISUID | S_ISGID);
1102 newmode = zp->z_mode;
1103 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
1104 (void *)&newmode, sizeof (uint64_t), tx);
1106 mutex_exit(&zp->z_acl_lock);
1108 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
1112 * Update the file size (zp_size) if it has changed;
1113 * account for possible concurrent updates.
1115 while ((end_size = zp->z_size) < uio->uio_loffset) {
1116 (void) atomic_cas_64(&zp->z_size, end_size,
1121 * If we are replaying and eof is non zero then force
1122 * the file size to the specified eof. Note, there's no
1123 * concurrency during replay.
1125 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
1126 zp->z_size = zfsvfs->z_replay_eof;
1128 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1130 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
1135 ASSERT(tx_bytes == nbytes);
1140 uio_prefaultpages(MIN(n, max_blksz), uio);
1144 zfs_range_unlock(rl);
1147 * If we're in replay mode, or we made no progress, return error.
1148 * Otherwise, it's at least a partial write, so it's successful.
1150 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1155 if (ioflag & (FSYNC | FDSYNC) ||
1156 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1157 zil_commit(zilog, zp->z_id);
1164 zfs_get_done(zgd_t *zgd, int error)
1166 znode_t *zp = zgd->zgd_private;
1167 objset_t *os = zp->z_zfsvfs->z_os;
1170 dmu_buf_rele(zgd->zgd_db, zgd);
1172 zfs_range_unlock(zgd->zgd_rl);
1175 * Release the vnode asynchronously as we currently have the
1176 * txg stopped from syncing.
1178 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1180 if (error == 0 && zgd->zgd_bp)
1181 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1183 kmem_free(zgd, sizeof (zgd_t));
1187 static int zil_fault_io = 0;
1191 * Get data to generate a TX_WRITE intent log record.
1194 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1196 zfsvfs_t *zfsvfs = arg;
1197 objset_t *os = zfsvfs->z_os;
1199 uint64_t object = lr->lr_foid;
1200 uint64_t offset = lr->lr_offset;
1201 uint64_t size = lr->lr_length;
1202 blkptr_t *bp = &lr->lr_blkptr;
1207 ASSERT(zio != NULL);
1211 * Nothing to do if the file has been removed
1213 if (zfs_zget(zfsvfs, object, &zp) != 0)
1214 return (SET_ERROR(ENOENT));
1215 if (zp->z_unlinked) {
1217 * Release the vnode asynchronously as we currently have the
1218 * txg stopped from syncing.
1220 VN_RELE_ASYNC(ZTOV(zp),
1221 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1222 return (SET_ERROR(ENOENT));
1225 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1226 zgd->zgd_zilog = zfsvfs->z_log;
1227 zgd->zgd_private = zp;
1230 * Write records come in two flavors: immediate and indirect.
1231 * For small writes it's cheaper to store the data with the
1232 * log record (immediate); for large writes it's cheaper to
1233 * sync the data and get a pointer to it (indirect) so that
1234 * we don't have to write the data twice.
1236 if (buf != NULL) { /* immediate write */
1237 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1238 /* test for truncation needs to be done while range locked */
1239 if (offset >= zp->z_size) {
1240 error = SET_ERROR(ENOENT);
1242 error = dmu_read(os, object, offset, size, buf,
1243 DMU_READ_NO_PREFETCH);
1245 ASSERT(error == 0 || error == ENOENT);
1246 } else { /* indirect write */
1248 * Have to lock the whole block to ensure when it's
1249 * written out and it's checksum is being calculated
1250 * that no one can change the data. We need to re-check
1251 * blocksize after we get the lock in case it's changed!
1256 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1258 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1260 if (zp->z_blksz == size)
1263 zfs_range_unlock(zgd->zgd_rl);
1265 /* test for truncation needs to be done while range locked */
1266 if (lr->lr_offset >= zp->z_size)
1267 error = SET_ERROR(ENOENT);
1270 error = SET_ERROR(EIO);
1275 error = dmu_buf_hold(os, object, offset, zgd, &db,
1276 DMU_READ_NO_PREFETCH);
1279 blkptr_t *obp = dmu_buf_get_blkptr(db);
1281 ASSERT(BP_IS_HOLE(bp));
1288 ASSERT(db->db_offset == offset);
1289 ASSERT(db->db_size == size);
1291 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1293 ASSERT(error || lr->lr_length <= zp->z_blksz);
1296 * On success, we need to wait for the write I/O
1297 * initiated by dmu_sync() to complete before we can
1298 * release this dbuf. We will finish everything up
1299 * in the zfs_get_done() callback.
1304 if (error == EALREADY) {
1305 lr->lr_common.lrc_txtype = TX_WRITE2;
1311 zfs_get_done(zgd, error);
1318 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1319 caller_context_t *ct)
1321 znode_t *zp = VTOZ(vp);
1322 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1328 if (flag & V_ACE_MASK)
1329 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1331 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1338 * If vnode is for a device return a specfs vnode instead.
1341 specvp_check(vnode_t **vpp, cred_t *cr)
1345 if (IS_DEVVP(*vpp)) {
1348 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1351 error = SET_ERROR(ENOSYS);
1359 * Lookup an entry in a directory, or an extended attribute directory.
1360 * If it exists, return a held vnode reference for it.
1362 * IN: dvp - vnode of directory to search.
1363 * nm - name of entry to lookup.
1364 * pnp - full pathname to lookup [UNUSED].
1365 * flags - LOOKUP_XATTR set if looking for an attribute.
1366 * rdir - root directory vnode [UNUSED].
1367 * cr - credentials of caller.
1368 * ct - caller context
1369 * direntflags - directory lookup flags
1370 * realpnp - returned pathname.
1372 * OUT: vpp - vnode of located entry, NULL if not found.
1374 * RETURN: 0 on success, error code on failure.
1381 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp,
1382 int nameiop, cred_t *cr, kthread_t *td, int flags)
1384 znode_t *zdp = VTOZ(dvp);
1385 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1387 int *direntflags = NULL;
1388 void *realpnp = NULL;
1391 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1393 if (dvp->v_type != VDIR) {
1394 return (SET_ERROR(ENOTDIR));
1395 } else if (zdp->z_sa_hdl == NULL) {
1396 return (SET_ERROR(EIO));
1399 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1400 error = zfs_fastaccesschk_execute(zdp, cr);
1408 vnode_t *tvp = dnlc_lookup(dvp, nm);
1411 error = zfs_fastaccesschk_execute(zdp, cr);
1416 if (tvp == DNLC_NO_VNODE) {
1418 return (SET_ERROR(ENOENT));
1421 return (specvp_check(vpp, cr));
1427 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1434 if (flags & LOOKUP_XATTR) {
1437 * If the xattr property is off, refuse the lookup request.
1439 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1441 return (SET_ERROR(EINVAL));
1446 * We don't allow recursive attributes..
1447 * Maybe someday we will.
1449 if (zdp->z_pflags & ZFS_XATTR) {
1451 return (SET_ERROR(EINVAL));
1454 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1460 * Do we have permission to get into attribute directory?
1463 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1473 if (dvp->v_type != VDIR) {
1475 return (SET_ERROR(ENOTDIR));
1479 * Check accessibility of directory.
1482 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1487 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1488 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1490 return (SET_ERROR(EILSEQ));
1493 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1495 error = specvp_check(vpp, cr);
1497 /* Translate errors and add SAVENAME when needed. */
1498 if (cnp->cn_flags & ISLASTCN) {
1502 if (error == ENOENT) {
1503 error = EJUSTRETURN;
1504 cnp->cn_flags |= SAVENAME;
1510 cnp->cn_flags |= SAVENAME;
1514 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) {
1517 if (cnp->cn_flags & ISDOTDOT) {
1518 ltype = VOP_ISLOCKED(dvp);
1522 error = zfs_vnode_lock(*vpp, cnp->cn_lkflags);
1523 if (cnp->cn_flags & ISDOTDOT)
1524 vn_lock(dvp, ltype | LK_RETRY);
1534 #ifdef FREEBSD_NAMECACHE
1536 * Insert name into cache (as non-existent) if appropriate.
1538 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
1539 cache_enter(dvp, *vpp, cnp);
1541 * Insert name into cache if appropriate.
1543 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) {
1544 if (!(cnp->cn_flags & ISLASTCN) ||
1545 (nameiop != DELETE && nameiop != RENAME)) {
1546 cache_enter(dvp, *vpp, cnp);
1555 * Attempt to create a new entry in a directory. If the entry
1556 * already exists, truncate the file if permissible, else return
1557 * an error. Return the vp of the created or trunc'd file.
1559 * IN: dvp - vnode of directory to put new file entry in.
1560 * name - name of new file entry.
1561 * vap - attributes of new file.
1562 * excl - flag indicating exclusive or non-exclusive mode.
1563 * mode - mode to open file with.
1564 * cr - credentials of caller.
1565 * flag - large file flag [UNUSED].
1566 * ct - caller context
1567 * vsecp - ACL to be set
1569 * OUT: vpp - vnode of created or trunc'd entry.
1571 * RETURN: 0 on success, error code on failure.
1574 * dvp - ctime|mtime updated if new entry created
1575 * vp - ctime|mtime always, atime if new
1580 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode,
1581 vnode_t **vpp, cred_t *cr, kthread_t *td)
1583 znode_t *zp, *dzp = VTOZ(dvp);
1584 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1592 gid_t gid = crgetgid(cr);
1593 zfs_acl_ids_t acl_ids;
1594 boolean_t fuid_dirtied;
1595 boolean_t have_acl = B_FALSE;
1600 * If we have an ephemeral id, ACL, or XVATTR then
1601 * make sure file system is at proper version
1604 ksid = crgetsid(cr, KSID_OWNER);
1606 uid = ksid_getid(ksid);
1610 if (zfsvfs->z_use_fuids == B_FALSE &&
1611 (vsecp || (vap->va_mask & AT_XVATTR) ||
1612 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1613 return (SET_ERROR(EINVAL));
1618 zilog = zfsvfs->z_log;
1620 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1621 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1623 return (SET_ERROR(EILSEQ));
1626 if (vap->va_mask & AT_XVATTR) {
1627 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
1628 crgetuid(cr), cr, vap->va_type)) != 0) {
1636 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr))
1637 vap->va_mode &= ~S_ISVTX;
1639 if (*name == '\0') {
1641 * Null component name refers to the directory itself.
1648 /* possible VN_HOLD(zp) */
1651 if (flag & FIGNORECASE)
1654 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1658 zfs_acl_ids_free(&acl_ids);
1659 if (strcmp(name, "..") == 0)
1660 error = SET_ERROR(EISDIR);
1670 * Create a new file object and update the directory
1673 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1675 zfs_acl_ids_free(&acl_ids);
1680 * We only support the creation of regular files in
1681 * extended attribute directories.
1684 if ((dzp->z_pflags & ZFS_XATTR) &&
1685 (vap->va_type != VREG)) {
1687 zfs_acl_ids_free(&acl_ids);
1688 error = SET_ERROR(EINVAL);
1692 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1693 cr, vsecp, &acl_ids)) != 0)
1697 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1698 zfs_acl_ids_free(&acl_ids);
1699 error = SET_ERROR(EDQUOT);
1703 tx = dmu_tx_create(os);
1705 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1706 ZFS_SA_BASE_ATTR_SIZE);
1708 fuid_dirtied = zfsvfs->z_fuid_dirty;
1710 zfs_fuid_txhold(zfsvfs, tx);
1711 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1712 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1713 if (!zfsvfs->z_use_sa &&
1714 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1715 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1716 0, acl_ids.z_aclp->z_acl_bytes);
1718 error = dmu_tx_assign(tx, TXG_NOWAIT);
1720 zfs_dirent_unlock(dl);
1721 if (error == ERESTART) {
1726 zfs_acl_ids_free(&acl_ids);
1731 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1734 zfs_fuid_sync(zfsvfs, tx);
1736 (void) zfs_link_create(dl, zp, tx, ZNEW);
1737 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1738 if (flag & FIGNORECASE)
1740 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1741 vsecp, acl_ids.z_fuidp, vap);
1742 zfs_acl_ids_free(&acl_ids);
1745 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1748 zfs_acl_ids_free(&acl_ids);
1752 * A directory entry already exists for this name.
1755 * Can't truncate an existing file if in exclusive mode.
1758 error = SET_ERROR(EEXIST);
1762 * Can't open a directory for writing.
1764 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1765 error = SET_ERROR(EISDIR);
1769 * Verify requested access to file.
1771 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1775 mutex_enter(&dzp->z_lock);
1777 mutex_exit(&dzp->z_lock);
1780 * Truncate regular files if requested.
1782 if ((ZTOV(zp)->v_type == VREG) &&
1783 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1784 /* we can't hold any locks when calling zfs_freesp() */
1785 zfs_dirent_unlock(dl);
1787 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1789 vnevent_create(ZTOV(zp), ct);
1795 zfs_dirent_unlock(dl);
1802 error = specvp_check(vpp, cr);
1805 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1806 zil_commit(zilog, 0);
1813 * Remove an entry from a directory.
1815 * IN: dvp - vnode of directory to remove entry from.
1816 * name - name of entry to remove.
1817 * cr - credentials of caller.
1818 * ct - caller context
1819 * flags - case flags
1821 * RETURN: 0 on success, error code on failure.
1825 * vp - ctime (if nlink > 0)
1828 uint64_t null_xattr = 0;
1832 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1835 znode_t *zp, *dzp = VTOZ(dvp);
1838 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1840 uint64_t acl_obj, xattr_obj;
1841 uint64_t xattr_obj_unlinked = 0;
1845 boolean_t may_delete_now, delete_now = FALSE;
1846 boolean_t unlinked, toobig = FALSE;
1848 pathname_t *realnmp = NULL;
1855 zilog = zfsvfs->z_log;
1857 if (flags & FIGNORECASE) {
1867 * Attempt to lock directory; fail if entry doesn't exist.
1869 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1879 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1884 * Need to use rmdir for removing directories.
1886 if (vp->v_type == VDIR) {
1887 error = SET_ERROR(EPERM);
1891 vnevent_remove(vp, dvp, name, ct);
1894 dnlc_remove(dvp, realnmp->pn_buf);
1896 dnlc_remove(dvp, name);
1899 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1903 * We may delete the znode now, or we may put it in the unlinked set;
1904 * it depends on whether we're the last link, and on whether there are
1905 * other holds on the vnode. So we dmu_tx_hold() the right things to
1906 * allow for either case.
1909 tx = dmu_tx_create(zfsvfs->z_os);
1910 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1911 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1912 zfs_sa_upgrade_txholds(tx, zp);
1913 zfs_sa_upgrade_txholds(tx, dzp);
1914 if (may_delete_now) {
1916 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1917 /* if the file is too big, only hold_free a token amount */
1918 dmu_tx_hold_free(tx, zp->z_id, 0,
1919 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1922 /* are there any extended attributes? */
1923 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1924 &xattr_obj, sizeof (xattr_obj));
1925 if (error == 0 && xattr_obj) {
1926 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1928 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1929 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1932 mutex_enter(&zp->z_lock);
1933 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1934 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1935 mutex_exit(&zp->z_lock);
1937 /* charge as an update -- would be nice not to charge at all */
1938 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1940 error = dmu_tx_assign(tx, TXG_NOWAIT);
1942 zfs_dirent_unlock(dl);
1946 if (error == ERESTART) {
1959 * Remove the directory entry.
1961 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1971 * Hold z_lock so that we can make sure that the ACL obj
1972 * hasn't changed. Could have been deleted due to
1975 mutex_enter(&zp->z_lock);
1977 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1978 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1979 delete_now = may_delete_now && !toobig &&
1980 vp->v_count == 1 && !vn_has_cached_data(vp) &&
1981 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1988 panic("zfs_remove: delete_now branch taken");
1990 if (xattr_obj_unlinked) {
1991 ASSERT3U(xzp->z_links, ==, 2);
1992 mutex_enter(&xzp->z_lock);
1993 xzp->z_unlinked = 1;
1995 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1996 &xzp->z_links, sizeof (xzp->z_links), tx);
1997 ASSERT3U(error, ==, 0);
1998 mutex_exit(&xzp->z_lock);
1999 zfs_unlinked_add(xzp, tx);
2002 error = sa_remove(zp->z_sa_hdl,
2003 SA_ZPL_XATTR(zfsvfs), tx);
2005 error = sa_update(zp->z_sa_hdl,
2006 SA_ZPL_XATTR(zfsvfs), &null_xattr,
2007 sizeof (uint64_t), tx);
2012 ASSERT0(vp->v_count);
2014 mutex_exit(&zp->z_lock);
2015 zfs_znode_delete(zp, tx);
2016 } else if (unlinked) {
2017 mutex_exit(&zp->z_lock);
2018 zfs_unlinked_add(zp, tx);
2020 vp->v_vflag |= VV_NOSYNC;
2025 if (flags & FIGNORECASE)
2027 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
2034 zfs_dirent_unlock(dl);
2041 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2042 zil_commit(zilog, 0);
2049 * Create a new directory and insert it into dvp using the name
2050 * provided. Return a pointer to the inserted directory.
2052 * IN: dvp - vnode of directory to add subdir to.
2053 * dirname - name of new directory.
2054 * vap - attributes of new directory.
2055 * cr - credentials of caller.
2056 * ct - caller context
2057 * flags - case flags
2058 * vsecp - ACL to be set
2060 * OUT: vpp - vnode of created directory.
2062 * RETURN: 0 on success, error code on failure.
2065 * dvp - ctime|mtime updated
2066 * vp - ctime|mtime|atime updated
2070 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
2071 caller_context_t *ct, int flags, vsecattr_t *vsecp)
2073 znode_t *zp, *dzp = VTOZ(dvp);
2074 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2083 gid_t gid = crgetgid(cr);
2084 zfs_acl_ids_t acl_ids;
2085 boolean_t fuid_dirtied;
2087 ASSERT(vap->va_type == VDIR);
2090 * If we have an ephemeral id, ACL, or XVATTR then
2091 * make sure file system is at proper version
2094 ksid = crgetsid(cr, KSID_OWNER);
2096 uid = ksid_getid(ksid);
2099 if (zfsvfs->z_use_fuids == B_FALSE &&
2100 (vsecp || (vap->va_mask & AT_XVATTR) ||
2101 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
2102 return (SET_ERROR(EINVAL));
2106 zilog = zfsvfs->z_log;
2108 if (dzp->z_pflags & ZFS_XATTR) {
2110 return (SET_ERROR(EINVAL));
2113 if (zfsvfs->z_utf8 && u8_validate(dirname,
2114 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2116 return (SET_ERROR(EILSEQ));
2118 if (flags & FIGNORECASE)
2121 if (vap->va_mask & AT_XVATTR) {
2122 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
2123 crgetuid(cr), cr, vap->va_type)) != 0) {
2129 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
2130 vsecp, &acl_ids)) != 0) {
2135 * First make sure the new directory doesn't exist.
2137 * Existence is checked first to make sure we don't return
2138 * EACCES instead of EEXIST which can cause some applications
2144 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
2146 zfs_acl_ids_free(&acl_ids);
2151 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
2152 zfs_acl_ids_free(&acl_ids);
2153 zfs_dirent_unlock(dl);
2158 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
2159 zfs_acl_ids_free(&acl_ids);
2160 zfs_dirent_unlock(dl);
2162 return (SET_ERROR(EDQUOT));
2166 * Add a new entry to the directory.
2168 tx = dmu_tx_create(zfsvfs->z_os);
2169 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2170 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2171 fuid_dirtied = zfsvfs->z_fuid_dirty;
2173 zfs_fuid_txhold(zfsvfs, tx);
2174 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2175 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2176 acl_ids.z_aclp->z_acl_bytes);
2179 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2180 ZFS_SA_BASE_ATTR_SIZE);
2182 error = dmu_tx_assign(tx, TXG_NOWAIT);
2184 zfs_dirent_unlock(dl);
2185 if (error == ERESTART) {
2190 zfs_acl_ids_free(&acl_ids);
2199 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2202 zfs_fuid_sync(zfsvfs, tx);
2205 * Now put new name in parent dir.
2207 (void) zfs_link_create(dl, zp, tx, ZNEW);
2211 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2212 if (flags & FIGNORECASE)
2214 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2215 acl_ids.z_fuidp, vap);
2217 zfs_acl_ids_free(&acl_ids);
2221 zfs_dirent_unlock(dl);
2223 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2224 zil_commit(zilog, 0);
2231 * Remove a directory subdir entry. If the current working
2232 * directory is the same as the subdir to be removed, the
2235 * IN: dvp - vnode of directory to remove from.
2236 * name - name of directory to be removed.
2237 * cwd - vnode of current working directory.
2238 * cr - credentials of caller.
2239 * ct - caller context
2240 * flags - case flags
2242 * RETURN: 0 on success, error code on failure.
2245 * dvp - ctime|mtime updated
2249 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2250 caller_context_t *ct, int flags)
2252 znode_t *dzp = VTOZ(dvp);
2255 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2264 zilog = zfsvfs->z_log;
2266 if (flags & FIGNORECASE)
2272 * Attempt to lock directory; fail if entry doesn't exist.
2274 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2282 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2286 if (vp->v_type != VDIR) {
2287 error = SET_ERROR(ENOTDIR);
2292 error = SET_ERROR(EINVAL);
2296 vnevent_rmdir(vp, dvp, name, ct);
2299 * Grab a lock on the directory to make sure that noone is
2300 * trying to add (or lookup) entries while we are removing it.
2302 rw_enter(&zp->z_name_lock, RW_WRITER);
2305 * Grab a lock on the parent pointer to make sure we play well
2306 * with the treewalk and directory rename code.
2308 rw_enter(&zp->z_parent_lock, RW_WRITER);
2310 tx = dmu_tx_create(zfsvfs->z_os);
2311 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2312 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2313 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2314 zfs_sa_upgrade_txholds(tx, zp);
2315 zfs_sa_upgrade_txholds(tx, dzp);
2316 error = dmu_tx_assign(tx, TXG_NOWAIT);
2318 rw_exit(&zp->z_parent_lock);
2319 rw_exit(&zp->z_name_lock);
2320 zfs_dirent_unlock(dl);
2322 if (error == ERESTART) {
2332 #ifdef FREEBSD_NAMECACHE
2336 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2339 uint64_t txtype = TX_RMDIR;
2340 if (flags & FIGNORECASE)
2342 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2347 rw_exit(&zp->z_parent_lock);
2348 rw_exit(&zp->z_name_lock);
2349 #ifdef FREEBSD_NAMECACHE
2353 zfs_dirent_unlock(dl);
2357 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2358 zil_commit(zilog, 0);
2365 * Read as many directory entries as will fit into the provided
2366 * buffer from the given directory cursor position (specified in
2367 * the uio structure).
2369 * IN: vp - vnode of directory to read.
2370 * uio - structure supplying read location, range info,
2371 * and return buffer.
2372 * cr - credentials of caller.
2373 * ct - caller context
2374 * flags - case flags
2376 * OUT: uio - updated offset and range, buffer filled.
2377 * eofp - set to true if end-of-file detected.
2379 * RETURN: 0 on success, error code on failure.
2382 * vp - atime updated
2384 * Note that the low 4 bits of the cookie returned by zap is always zero.
2385 * This allows us to use the low range for "special" directory entries:
2386 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2387 * we use the offset 2 for the '.zfs' directory.
2391 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies)
2393 znode_t *zp = VTOZ(vp);
2397 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2402 zap_attribute_t zap;
2403 uint_t bytes_wanted;
2404 uint64_t offset; /* must be unsigned; checks for < 1 */
2410 boolean_t check_sysattrs;
2413 u_long *cooks = NULL;
2419 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2420 &parent, sizeof (parent))) != 0) {
2426 * If we are not given an eof variable,
2433 * Check for valid iov_len.
2435 if (uio->uio_iov->iov_len <= 0) {
2437 return (SET_ERROR(EINVAL));
2441 * Quit if directory has been removed (posix)
2443 if ((*eofp = zp->z_unlinked) != 0) {
2450 offset = uio->uio_loffset;
2451 prefetch = zp->z_zn_prefetch;
2454 * Initialize the iterator cursor.
2458 * Start iteration from the beginning of the directory.
2460 zap_cursor_init(&zc, os, zp->z_id);
2463 * The offset is a serialized cursor.
2465 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2469 * Get space to change directory entries into fs independent format.
2471 iovp = uio->uio_iov;
2472 bytes_wanted = iovp->iov_len;
2473 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2474 bufsize = bytes_wanted;
2475 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2476 odp = (struct dirent64 *)outbuf;
2478 bufsize = bytes_wanted;
2480 odp = (struct dirent64 *)iovp->iov_base;
2482 eodp = (struct edirent *)odp;
2484 if (ncookies != NULL) {
2486 * Minimum entry size is dirent size and 1 byte for a file name.
2488 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1);
2489 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK);
2494 * If this VFS supports the system attribute view interface; and
2495 * we're looking at an extended attribute directory; and we care
2496 * about normalization conflicts on this vfs; then we must check
2497 * for normalization conflicts with the sysattr name space.
2500 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2501 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2502 (flags & V_RDDIR_ENTFLAGS);
2508 * Transform to file-system independent format
2511 while (outcount < bytes_wanted) {
2514 off64_t *next = NULL;
2517 * Special case `.', `..', and `.zfs'.
2520 (void) strcpy(zap.za_name, ".");
2521 zap.za_normalization_conflict = 0;
2524 } else if (offset == 1) {
2525 (void) strcpy(zap.za_name, "..");
2526 zap.za_normalization_conflict = 0;
2529 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2530 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2531 zap.za_normalization_conflict = 0;
2532 objnum = ZFSCTL_INO_ROOT;
2538 if (error = zap_cursor_retrieve(&zc, &zap)) {
2539 if ((*eofp = (error == ENOENT)) != 0)
2545 if (zap.za_integer_length != 8 ||
2546 zap.za_num_integers != 1) {
2547 cmn_err(CE_WARN, "zap_readdir: bad directory "
2548 "entry, obj = %lld, offset = %lld\n",
2549 (u_longlong_t)zp->z_id,
2550 (u_longlong_t)offset);
2551 error = SET_ERROR(ENXIO);
2555 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2557 * MacOS X can extract the object type here such as:
2558 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2560 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2562 if (check_sysattrs && !zap.za_normalization_conflict) {
2564 zap.za_normalization_conflict =
2565 xattr_sysattr_casechk(zap.za_name);
2567 panic("%s:%u: TODO", __func__, __LINE__);
2572 if (flags & V_RDDIR_ACCFILTER) {
2574 * If we have no access at all, don't include
2575 * this entry in the returned information
2578 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2580 if (!zfs_has_access(ezp, cr)) {
2587 if (flags & V_RDDIR_ENTFLAGS)
2588 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2590 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2593 * Will this entry fit in the buffer?
2595 if (outcount + reclen > bufsize) {
2597 * Did we manage to fit anything in the buffer?
2600 error = SET_ERROR(EINVAL);
2605 if (flags & V_RDDIR_ENTFLAGS) {
2607 * Add extended flag entry:
2609 eodp->ed_ino = objnum;
2610 eodp->ed_reclen = reclen;
2611 /* NOTE: ed_off is the offset for the *next* entry */
2612 next = &(eodp->ed_off);
2613 eodp->ed_eflags = zap.za_normalization_conflict ?
2614 ED_CASE_CONFLICT : 0;
2615 (void) strncpy(eodp->ed_name, zap.za_name,
2616 EDIRENT_NAMELEN(reclen));
2617 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2622 odp->d_ino = objnum;
2623 odp->d_reclen = reclen;
2624 odp->d_namlen = strlen(zap.za_name);
2625 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1);
2627 odp = (dirent64_t *)((intptr_t)odp + reclen);
2631 ASSERT(outcount <= bufsize);
2633 /* Prefetch znode */
2635 dmu_prefetch(os, objnum, 0, 0);
2639 * Move to the next entry, fill in the previous offset.
2641 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2642 zap_cursor_advance(&zc);
2643 offset = zap_cursor_serialize(&zc);
2648 if (cooks != NULL) {
2651 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks));
2654 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2656 /* Subtract unused cookies */
2657 if (ncookies != NULL)
2658 *ncookies -= ncooks;
2660 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2661 iovp->iov_base += outcount;
2662 iovp->iov_len -= outcount;
2663 uio->uio_resid -= outcount;
2664 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2666 * Reset the pointer.
2668 offset = uio->uio_loffset;
2672 zap_cursor_fini(&zc);
2673 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2674 kmem_free(outbuf, bufsize);
2676 if (error == ENOENT)
2679 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2681 uio->uio_loffset = offset;
2683 if (error != 0 && cookies != NULL) {
2684 free(*cookies, M_TEMP);
2691 ulong_t zfs_fsync_sync_cnt = 4;
2694 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2696 znode_t *zp = VTOZ(vp);
2697 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2699 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2701 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2704 zil_commit(zfsvfs->z_log, zp->z_id);
2712 * Get the requested file attributes and place them in the provided
2715 * IN: vp - vnode of file.
2716 * vap - va_mask identifies requested attributes.
2717 * If AT_XVATTR set, then optional attrs are requested
2718 * flags - ATTR_NOACLCHECK (CIFS server context)
2719 * cr - credentials of caller.
2720 * ct - caller context
2722 * OUT: vap - attribute values.
2724 * RETURN: 0 (always succeeds).
2728 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2729 caller_context_t *ct)
2731 znode_t *zp = VTOZ(vp);
2732 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2735 u_longlong_t nblocks;
2737 uint64_t mtime[2], ctime[2], crtime[2], rdev;
2738 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2739 xoptattr_t *xoap = NULL;
2740 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2741 sa_bulk_attr_t bulk[4];
2747 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2749 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2750 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2751 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
2752 if (vp->v_type == VBLK || vp->v_type == VCHR)
2753 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
2756 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2762 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2763 * Also, if we are the owner don't bother, since owner should
2764 * always be allowed to read basic attributes of file.
2766 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2767 (vap->va_uid != crgetuid(cr))) {
2768 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2776 * Return all attributes. It's cheaper to provide the answer
2777 * than to determine whether we were asked the question.
2780 mutex_enter(&zp->z_lock);
2781 vap->va_type = IFTOVT(zp->z_mode);
2782 vap->va_mode = zp->z_mode & ~S_IFMT;
2784 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2786 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
2788 vap->va_nodeid = zp->z_id;
2789 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2790 links = zp->z_links + 1;
2792 links = zp->z_links;
2793 vap->va_nlink = MIN(links, LINK_MAX); /* nlink_t limit! */
2794 vap->va_size = zp->z_size;
2796 vap->va_rdev = vp->v_rdev;
2798 if (vp->v_type == VBLK || vp->v_type == VCHR)
2799 vap->va_rdev = zfs_cmpldev(rdev);
2801 vap->va_seq = zp->z_seq;
2802 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */
2805 * Add in any requested optional attributes and the create time.
2806 * Also set the corresponding bits in the returned attribute bitmap.
2808 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2809 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2811 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2812 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2815 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2816 xoap->xoa_readonly =
2817 ((zp->z_pflags & ZFS_READONLY) != 0);
2818 XVA_SET_RTN(xvap, XAT_READONLY);
2821 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2823 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2824 XVA_SET_RTN(xvap, XAT_SYSTEM);
2827 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2829 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2830 XVA_SET_RTN(xvap, XAT_HIDDEN);
2833 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2834 xoap->xoa_nounlink =
2835 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2836 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2839 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2840 xoap->xoa_immutable =
2841 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2842 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2845 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2846 xoap->xoa_appendonly =
2847 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2848 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2851 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2853 ((zp->z_pflags & ZFS_NODUMP) != 0);
2854 XVA_SET_RTN(xvap, XAT_NODUMP);
2857 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2859 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2860 XVA_SET_RTN(xvap, XAT_OPAQUE);
2863 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2864 xoap->xoa_av_quarantined =
2865 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2866 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2869 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2870 xoap->xoa_av_modified =
2871 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2872 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2875 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2876 vp->v_type == VREG) {
2877 zfs_sa_get_scanstamp(zp, xvap);
2880 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2883 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2884 times, sizeof (times));
2885 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2886 XVA_SET_RTN(xvap, XAT_CREATETIME);
2889 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2890 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2891 XVA_SET_RTN(xvap, XAT_REPARSE);
2893 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2894 xoap->xoa_generation = zp->z_gen;
2895 XVA_SET_RTN(xvap, XAT_GEN);
2898 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2900 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2901 XVA_SET_RTN(xvap, XAT_OFFLINE);
2904 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2906 ((zp->z_pflags & ZFS_SPARSE) != 0);
2907 XVA_SET_RTN(xvap, XAT_SPARSE);
2911 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2912 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2913 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2914 ZFS_TIME_DECODE(&vap->va_birthtime, crtime);
2916 mutex_exit(&zp->z_lock);
2918 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2919 vap->va_blksize = blksize;
2920 vap->va_bytes = nblocks << 9; /* nblocks * 512 */
2922 if (zp->z_blksz == 0) {
2924 * Block size hasn't been set; suggest maximal I/O transfers.
2926 vap->va_blksize = zfsvfs->z_max_blksz;
2934 * Set the file attributes to the values contained in the
2937 * IN: vp - vnode of file to be modified.
2938 * vap - new attribute values.
2939 * If AT_XVATTR set, then optional attrs are being set
2940 * flags - ATTR_UTIME set if non-default time values provided.
2941 * - ATTR_NOACLCHECK (CIFS context only).
2942 * cr - credentials of caller.
2943 * ct - caller context
2945 * RETURN: 0 on success, error code on failure.
2948 * vp - ctime updated, mtime updated if size changed.
2952 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2953 caller_context_t *ct)
2955 znode_t *zp = VTOZ(vp);
2956 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2961 uint_t mask = vap->va_mask;
2962 uint_t saved_mask = 0;
2963 uint64_t saved_mode;
2966 uint64_t new_uid, new_gid;
2968 uint64_t mtime[2], ctime[2];
2970 int need_policy = FALSE;
2972 zfs_fuid_info_t *fuidp = NULL;
2973 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2976 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2977 boolean_t fuid_dirtied = B_FALSE;
2978 sa_bulk_attr_t bulk[7], xattr_bulk[7];
2979 int count = 0, xattr_count = 0;
2984 if (mask & AT_NOSET)
2985 return (SET_ERROR(EINVAL));
2990 zilog = zfsvfs->z_log;
2993 * Make sure that if we have ephemeral uid/gid or xvattr specified
2994 * that file system is at proper version level
2997 if (zfsvfs->z_use_fuids == B_FALSE &&
2998 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2999 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
3000 (mask & AT_XVATTR))) {
3002 return (SET_ERROR(EINVAL));
3005 if (mask & AT_SIZE && vp->v_type == VDIR) {
3007 return (SET_ERROR(EISDIR));
3010 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
3012 return (SET_ERROR(EINVAL));
3016 * If this is an xvattr_t, then get a pointer to the structure of
3017 * optional attributes. If this is NULL, then we have a vattr_t.
3019 xoap = xva_getxoptattr(xvap);
3021 xva_init(&tmpxvattr);
3024 * Immutable files can only alter immutable bit and atime
3026 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
3027 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
3028 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
3030 return (SET_ERROR(EPERM));
3033 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
3035 return (SET_ERROR(EPERM));
3039 * Verify timestamps doesn't overflow 32 bits.
3040 * ZFS can handle large timestamps, but 32bit syscalls can't
3041 * handle times greater than 2039. This check should be removed
3042 * once large timestamps are fully supported.
3044 if (mask & (AT_ATIME | AT_MTIME)) {
3045 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
3046 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
3048 return (SET_ERROR(EOVERFLOW));
3056 /* Can this be moved to before the top label? */
3057 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
3059 return (SET_ERROR(EROFS));
3063 * First validate permissions
3066 if (mask & AT_SIZE) {
3068 * XXX - Note, we are not providing any open
3069 * mode flags here (like FNDELAY), so we may
3070 * block if there are locks present... this
3071 * should be addressed in openat().
3073 /* XXX - would it be OK to generate a log record here? */
3074 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
3081 if (mask & (AT_ATIME|AT_MTIME) ||
3082 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
3083 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
3084 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
3085 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
3086 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
3087 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
3088 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
3089 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
3093 if (mask & (AT_UID|AT_GID)) {
3094 int idmask = (mask & (AT_UID|AT_GID));
3099 * NOTE: even if a new mode is being set,
3100 * we may clear S_ISUID/S_ISGID bits.
3103 if (!(mask & AT_MODE))
3104 vap->va_mode = zp->z_mode;
3107 * Take ownership or chgrp to group we are a member of
3110 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
3111 take_group = (mask & AT_GID) &&
3112 zfs_groupmember(zfsvfs, vap->va_gid, cr);
3115 * If both AT_UID and AT_GID are set then take_owner and
3116 * take_group must both be set in order to allow taking
3119 * Otherwise, send the check through secpolicy_vnode_setattr()
3123 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
3124 ((idmask == AT_UID) && take_owner) ||
3125 ((idmask == AT_GID) && take_group)) {
3126 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
3127 skipaclchk, cr) == 0) {
3129 * Remove setuid/setgid for non-privileged users
3131 secpolicy_setid_clear(vap, vp, cr);
3132 trim_mask = (mask & (AT_UID|AT_GID));
3141 mutex_enter(&zp->z_lock);
3142 oldva.va_mode = zp->z_mode;
3143 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
3144 if (mask & AT_XVATTR) {
3146 * Update xvattr mask to include only those attributes
3147 * that are actually changing.
3149 * the bits will be restored prior to actually setting
3150 * the attributes so the caller thinks they were set.
3152 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
3153 if (xoap->xoa_appendonly !=
3154 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
3157 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
3158 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
3162 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
3163 if (xoap->xoa_nounlink !=
3164 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
3167 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
3168 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
3172 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
3173 if (xoap->xoa_immutable !=
3174 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
3177 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
3178 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
3182 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
3183 if (xoap->xoa_nodump !=
3184 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
3187 XVA_CLR_REQ(xvap, XAT_NODUMP);
3188 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
3192 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3193 if (xoap->xoa_av_modified !=
3194 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3197 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3198 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
3202 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3203 if ((vp->v_type != VREG &&
3204 xoap->xoa_av_quarantined) ||
3205 xoap->xoa_av_quarantined !=
3206 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3209 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3210 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
3214 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3215 mutex_exit(&zp->z_lock);
3217 return (SET_ERROR(EPERM));
3220 if (need_policy == FALSE &&
3221 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3222 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3227 mutex_exit(&zp->z_lock);
3229 if (mask & AT_MODE) {
3230 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3231 err = secpolicy_setid_setsticky_clear(vp, vap,
3237 trim_mask |= AT_MODE;
3245 * If trim_mask is set then take ownership
3246 * has been granted or write_acl is present and user
3247 * has the ability to modify mode. In that case remove
3248 * UID|GID and or MODE from mask so that
3249 * secpolicy_vnode_setattr() doesn't revoke it.
3253 saved_mask = vap->va_mask;
3254 vap->va_mask &= ~trim_mask;
3255 if (trim_mask & AT_MODE) {
3257 * Save the mode, as secpolicy_vnode_setattr()
3258 * will overwrite it with ova.va_mode.
3260 saved_mode = vap->va_mode;
3263 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3264 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3271 vap->va_mask |= saved_mask;
3272 if (trim_mask & AT_MODE) {
3274 * Recover the mode after
3275 * secpolicy_vnode_setattr().
3277 vap->va_mode = saved_mode;
3283 * secpolicy_vnode_setattr, or take ownership may have
3286 mask = vap->va_mask;
3288 if ((mask & (AT_UID | AT_GID))) {
3289 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3290 &xattr_obj, sizeof (xattr_obj));
3292 if (err == 0 && xattr_obj) {
3293 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3297 if (mask & AT_UID) {
3298 new_uid = zfs_fuid_create(zfsvfs,
3299 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3300 if (new_uid != zp->z_uid &&
3301 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3303 VN_RELE(ZTOV(attrzp));
3304 err = SET_ERROR(EDQUOT);
3309 if (mask & AT_GID) {
3310 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3311 cr, ZFS_GROUP, &fuidp);
3312 if (new_gid != zp->z_gid &&
3313 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3315 VN_RELE(ZTOV(attrzp));
3316 err = SET_ERROR(EDQUOT);
3321 tx = dmu_tx_create(zfsvfs->z_os);
3323 if (mask & AT_MODE) {
3324 uint64_t pmode = zp->z_mode;
3326 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3328 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3329 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3330 err = SET_ERROR(EPERM);
3334 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3337 mutex_enter(&zp->z_lock);
3338 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3340 * Are we upgrading ACL from old V0 format
3343 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3344 zfs_znode_acl_version(zp) ==
3345 ZFS_ACL_VERSION_INITIAL) {
3346 dmu_tx_hold_free(tx, acl_obj, 0,
3348 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3349 0, aclp->z_acl_bytes);
3351 dmu_tx_hold_write(tx, acl_obj, 0,
3354 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3355 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3356 0, aclp->z_acl_bytes);
3358 mutex_exit(&zp->z_lock);
3359 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3361 if ((mask & AT_XVATTR) &&
3362 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3363 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3365 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3369 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3372 fuid_dirtied = zfsvfs->z_fuid_dirty;
3374 zfs_fuid_txhold(zfsvfs, tx);
3376 zfs_sa_upgrade_txholds(tx, zp);
3378 err = dmu_tx_assign(tx, TXG_NOWAIT);
3380 if (err == ERESTART)
3387 * Set each attribute requested.
3388 * We group settings according to the locks they need to acquire.
3390 * Note: you cannot set ctime directly, although it will be
3391 * updated as a side-effect of calling this function.
3395 if (mask & (AT_UID|AT_GID|AT_MODE))
3396 mutex_enter(&zp->z_acl_lock);
3397 mutex_enter(&zp->z_lock);
3399 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3400 &zp->z_pflags, sizeof (zp->z_pflags));
3403 if (mask & (AT_UID|AT_GID|AT_MODE))
3404 mutex_enter(&attrzp->z_acl_lock);
3405 mutex_enter(&attrzp->z_lock);
3406 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3407 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3408 sizeof (attrzp->z_pflags));
3411 if (mask & (AT_UID|AT_GID)) {
3413 if (mask & AT_UID) {
3414 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3415 &new_uid, sizeof (new_uid));
3416 zp->z_uid = new_uid;
3418 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3419 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3421 attrzp->z_uid = new_uid;
3425 if (mask & AT_GID) {
3426 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3427 NULL, &new_gid, sizeof (new_gid));
3428 zp->z_gid = new_gid;
3430 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3431 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3433 attrzp->z_gid = new_gid;
3436 if (!(mask & AT_MODE)) {
3437 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3438 NULL, &new_mode, sizeof (new_mode));
3439 new_mode = zp->z_mode;
3441 err = zfs_acl_chown_setattr(zp);
3444 err = zfs_acl_chown_setattr(attrzp);
3449 if (mask & AT_MODE) {
3450 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3451 &new_mode, sizeof (new_mode));
3452 zp->z_mode = new_mode;
3453 ASSERT3U((uintptr_t)aclp, !=, 0);
3454 err = zfs_aclset_common(zp, aclp, cr, tx);
3456 if (zp->z_acl_cached)
3457 zfs_acl_free(zp->z_acl_cached);
3458 zp->z_acl_cached = aclp;
3463 if (mask & AT_ATIME) {
3464 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3465 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3466 &zp->z_atime, sizeof (zp->z_atime));
3469 if (mask & AT_MTIME) {
3470 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3471 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3472 mtime, sizeof (mtime));
3475 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3476 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3477 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3478 NULL, mtime, sizeof (mtime));
3479 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3480 &ctime, sizeof (ctime));
3481 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3483 } else if (mask != 0) {
3484 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3485 &ctime, sizeof (ctime));
3486 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3489 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3490 SA_ZPL_CTIME(zfsvfs), NULL,
3491 &ctime, sizeof (ctime));
3492 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3493 mtime, ctime, B_TRUE);
3497 * Do this after setting timestamps to prevent timestamp
3498 * update from toggling bit
3501 if (xoap && (mask & AT_XVATTR)) {
3504 * restore trimmed off masks
3505 * so that return masks can be set for caller.
3508 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3509 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3511 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3512 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3514 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3515 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3517 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3518 XVA_SET_REQ(xvap, XAT_NODUMP);
3520 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3521 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3523 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3524 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3527 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3528 ASSERT(vp->v_type == VREG);
3530 zfs_xvattr_set(zp, xvap, tx);
3534 zfs_fuid_sync(zfsvfs, tx);
3537 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3539 mutex_exit(&zp->z_lock);
3540 if (mask & (AT_UID|AT_GID|AT_MODE))
3541 mutex_exit(&zp->z_acl_lock);
3544 if (mask & (AT_UID|AT_GID|AT_MODE))
3545 mutex_exit(&attrzp->z_acl_lock);
3546 mutex_exit(&attrzp->z_lock);
3549 if (err == 0 && attrzp) {
3550 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3556 VN_RELE(ZTOV(attrzp));
3562 zfs_fuid_info_free(fuidp);
3568 if (err == ERESTART)
3571 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3576 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3577 zil_commit(zilog, 0);
3583 typedef struct zfs_zlock {
3584 krwlock_t *zl_rwlock; /* lock we acquired */
3585 znode_t *zl_znode; /* znode we held */
3586 struct zfs_zlock *zl_next; /* next in list */
3590 * Drop locks and release vnodes that were held by zfs_rename_lock().
3593 zfs_rename_unlock(zfs_zlock_t **zlpp)
3597 while ((zl = *zlpp) != NULL) {
3598 if (zl->zl_znode != NULL)
3599 VN_RELE(ZTOV(zl->zl_znode));
3600 rw_exit(zl->zl_rwlock);
3601 *zlpp = zl->zl_next;
3602 kmem_free(zl, sizeof (*zl));
3607 * Search back through the directory tree, using the ".." entries.
3608 * Lock each directory in the chain to prevent concurrent renames.
3609 * Fail any attempt to move a directory into one of its own descendants.
3610 * XXX - z_parent_lock can overlap with map or grow locks
3613 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3617 uint64_t rootid = zp->z_zfsvfs->z_root;
3618 uint64_t oidp = zp->z_id;
3619 krwlock_t *rwlp = &szp->z_parent_lock;
3620 krw_t rw = RW_WRITER;
3623 * First pass write-locks szp and compares to zp->z_id.
3624 * Later passes read-lock zp and compare to zp->z_parent.
3627 if (!rw_tryenter(rwlp, rw)) {
3629 * Another thread is renaming in this path.
3630 * Note that if we are a WRITER, we don't have any
3631 * parent_locks held yet.
3633 if (rw == RW_READER && zp->z_id > szp->z_id) {
3635 * Drop our locks and restart
3637 zfs_rename_unlock(&zl);
3641 rwlp = &szp->z_parent_lock;
3646 * Wait for other thread to drop its locks
3652 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3653 zl->zl_rwlock = rwlp;
3654 zl->zl_znode = NULL;
3655 zl->zl_next = *zlpp;
3658 if (oidp == szp->z_id) /* We're a descendant of szp */
3659 return (SET_ERROR(EINVAL));
3661 if (oidp == rootid) /* We've hit the top */
3664 if (rw == RW_READER) { /* i.e. not the first pass */
3665 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3670 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3671 &oidp, sizeof (oidp));
3672 rwlp = &zp->z_parent_lock;
3675 } while (zp->z_id != sdzp->z_id);
3681 * Move an entry from the provided source directory to the target
3682 * directory. Change the entry name as indicated.
3684 * IN: sdvp - Source directory containing the "old entry".
3685 * snm - Old entry name.
3686 * tdvp - Target directory to contain the "new entry".
3687 * tnm - New entry name.
3688 * cr - credentials of caller.
3689 * ct - caller context
3690 * flags - case flags
3692 * RETURN: 0 on success, error code on failure.
3695 * sdvp,tdvp - ctime|mtime updated
3699 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3700 caller_context_t *ct, int flags)
3702 znode_t *tdzp, *szp, *tzp;
3703 znode_t *sdzp = VTOZ(sdvp);
3704 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs;
3707 zfs_dirlock_t *sdl, *tdl;
3710 int cmp, serr, terr;
3715 ZFS_VERIFY_ZP(sdzp);
3716 zilog = zfsvfs->z_log;
3719 * Make sure we have the real vp for the target directory.
3721 if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3724 if (tdvp->v_vfsp != sdvp->v_vfsp || zfsctl_is_node(tdvp)) {
3726 return (SET_ERROR(EXDEV));
3730 ZFS_VERIFY_ZP(tdzp);
3731 if (zfsvfs->z_utf8 && u8_validate(tnm,
3732 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3734 return (SET_ERROR(EILSEQ));
3737 if (flags & FIGNORECASE)
3746 * This is to prevent the creation of links into attribute space
3747 * by renaming a linked file into/outof an attribute directory.
3748 * See the comment in zfs_link() for why this is considered bad.
3750 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3752 return (SET_ERROR(EINVAL));
3756 * Lock source and target directory entries. To prevent deadlock,
3757 * a lock ordering must be defined. We lock the directory with
3758 * the smallest object id first, or if it's a tie, the one with
3759 * the lexically first name.
3761 if (sdzp->z_id < tdzp->z_id) {
3763 } else if (sdzp->z_id > tdzp->z_id) {
3767 * First compare the two name arguments without
3768 * considering any case folding.
3770 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3772 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3773 ASSERT(error == 0 || !zfsvfs->z_utf8);
3776 * POSIX: "If the old argument and the new argument
3777 * both refer to links to the same existing file,
3778 * the rename() function shall return successfully
3779 * and perform no other action."
3785 * If the file system is case-folding, then we may
3786 * have some more checking to do. A case-folding file
3787 * system is either supporting mixed case sensitivity
3788 * access or is completely case-insensitive. Note
3789 * that the file system is always case preserving.
3791 * In mixed sensitivity mode case sensitive behavior
3792 * is the default. FIGNORECASE must be used to
3793 * explicitly request case insensitive behavior.
3795 * If the source and target names provided differ only
3796 * by case (e.g., a request to rename 'tim' to 'Tim'),
3797 * we will treat this as a special case in the
3798 * case-insensitive mode: as long as the source name
3799 * is an exact match, we will allow this to proceed as
3800 * a name-change request.
3802 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3803 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3804 flags & FIGNORECASE)) &&
3805 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3808 * case preserving rename request, require exact
3817 * If the source and destination directories are the same, we should
3818 * grab the z_name_lock of that directory only once.
3822 rw_enter(&sdzp->z_name_lock, RW_READER);
3826 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3827 ZEXISTS | zflg, NULL, NULL);
3828 terr = zfs_dirent_lock(&tdl,
3829 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3831 terr = zfs_dirent_lock(&tdl,
3832 tdzp, tnm, &tzp, zflg, NULL, NULL);
3833 serr = zfs_dirent_lock(&sdl,
3834 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3840 * Source entry invalid or not there.
3843 zfs_dirent_unlock(tdl);
3849 rw_exit(&sdzp->z_name_lock);
3852 * FreeBSD: In OpenSolaris they only check if rename source is
3853 * ".." here, because "." is handled in their lookup. This is
3854 * not the case for FreeBSD, so we check for "." explicitly.
3856 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0)
3857 serr = SET_ERROR(EINVAL);
3862 zfs_dirent_unlock(sdl);
3866 rw_exit(&sdzp->z_name_lock);
3868 if (strcmp(tnm, "..") == 0)
3869 terr = SET_ERROR(EINVAL);
3875 * Must have write access at the source to remove the old entry
3876 * and write access at the target to create the new entry.
3877 * Note that if target and source are the same, this can be
3878 * done in a single check.
3881 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3884 if (ZTOV(szp)->v_type == VDIR) {
3886 * Check to make sure rename is valid.
3887 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3889 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3894 * Does target exist?
3898 * Source and target must be the same type.
3900 if (ZTOV(szp)->v_type == VDIR) {
3901 if (ZTOV(tzp)->v_type != VDIR) {
3902 error = SET_ERROR(ENOTDIR);
3906 if (ZTOV(tzp)->v_type == VDIR) {
3907 error = SET_ERROR(EISDIR);
3912 * POSIX dictates that when the source and target
3913 * entries refer to the same file object, rename
3914 * must do nothing and exit without error.
3916 if (szp->z_id == tzp->z_id) {
3922 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3924 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3927 * notify the target directory if it is not the same
3928 * as source directory.
3931 vnevent_rename_dest_dir(tdvp, ct);
3934 tx = dmu_tx_create(zfsvfs->z_os);
3935 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3936 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3937 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3938 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3940 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3941 zfs_sa_upgrade_txholds(tx, tdzp);
3944 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3945 zfs_sa_upgrade_txholds(tx, tzp);
3948 zfs_sa_upgrade_txholds(tx, szp);
3949 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3950 error = dmu_tx_assign(tx, TXG_NOWAIT);
3953 zfs_rename_unlock(&zl);
3954 zfs_dirent_unlock(sdl);
3955 zfs_dirent_unlock(tdl);
3958 rw_exit(&sdzp->z_name_lock);
3963 if (error == ERESTART) {
3973 if (tzp) /* Attempt to remove the existing target */
3974 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3977 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3979 szp->z_pflags |= ZFS_AV_MODIFIED;
3981 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3982 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3985 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3987 zfs_log_rename(zilog, tx, TX_RENAME |
3988 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3989 sdl->dl_name, tdzp, tdl->dl_name, szp);
3992 * Update path information for the target vnode
3994 vn_renamepath(tdvp, ZTOV(szp), tnm,
3998 * At this point, we have successfully created
3999 * the target name, but have failed to remove
4000 * the source name. Since the create was done
4001 * with the ZRENAMING flag, there are
4002 * complications; for one, the link count is
4003 * wrong. The easiest way to deal with this
4004 * is to remove the newly created target, and
4005 * return the original error. This must
4006 * succeed; fortunately, it is very unlikely to
4007 * fail, since we just created it.
4009 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
4010 ZRENAMING, NULL), ==, 0);
4013 #ifdef FREEBSD_NAMECACHE
4017 cache_purge(ZTOV(szp));
4019 cache_purge(ZTOV(tzp));
4027 zfs_rename_unlock(&zl);
4029 zfs_dirent_unlock(sdl);
4030 zfs_dirent_unlock(tdl);
4033 rw_exit(&sdzp->z_name_lock);
4040 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4041 zil_commit(zilog, 0);
4049 * Insert the indicated symbolic reference entry into the directory.
4051 * IN: dvp - Directory to contain new symbolic link.
4052 * link - Name for new symlink entry.
4053 * vap - Attributes of new entry.
4054 * cr - credentials of caller.
4055 * ct - caller context
4056 * flags - case flags
4058 * RETURN: 0 on success, error code on failure.
4061 * dvp - ctime|mtime updated
4065 zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link,
4066 cred_t *cr, kthread_t *td)
4068 znode_t *zp, *dzp = VTOZ(dvp);
4071 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4073 uint64_t len = strlen(link);
4076 zfs_acl_ids_t acl_ids;
4077 boolean_t fuid_dirtied;
4078 uint64_t txtype = TX_SYMLINK;
4081 ASSERT(vap->va_type == VLNK);
4085 zilog = zfsvfs->z_log;
4087 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
4088 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4090 return (SET_ERROR(EILSEQ));
4092 if (flags & FIGNORECASE)
4095 if (len > MAXPATHLEN) {
4097 return (SET_ERROR(ENAMETOOLONG));
4100 if ((error = zfs_acl_ids_create(dzp, 0,
4101 vap, cr, NULL, &acl_ids)) != 0) {
4107 * Attempt to lock directory; fail if entry already exists.
4109 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
4111 zfs_acl_ids_free(&acl_ids);
4116 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4117 zfs_acl_ids_free(&acl_ids);
4118 zfs_dirent_unlock(dl);
4123 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
4124 zfs_acl_ids_free(&acl_ids);
4125 zfs_dirent_unlock(dl);
4127 return (SET_ERROR(EDQUOT));
4129 tx = dmu_tx_create(zfsvfs->z_os);
4130 fuid_dirtied = zfsvfs->z_fuid_dirty;
4131 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
4132 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4133 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
4134 ZFS_SA_BASE_ATTR_SIZE + len);
4135 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
4136 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
4137 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
4138 acl_ids.z_aclp->z_acl_bytes);
4141 zfs_fuid_txhold(zfsvfs, tx);
4142 error = dmu_tx_assign(tx, TXG_NOWAIT);
4144 zfs_dirent_unlock(dl);
4145 if (error == ERESTART) {
4150 zfs_acl_ids_free(&acl_ids);
4157 * Create a new object for the symlink.
4158 * for version 4 ZPL datsets the symlink will be an SA attribute
4160 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
4163 zfs_fuid_sync(zfsvfs, tx);
4165 mutex_enter(&zp->z_lock);
4167 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
4170 zfs_sa_symlink(zp, link, len, tx);
4171 mutex_exit(&zp->z_lock);
4174 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
4175 &zp->z_size, sizeof (zp->z_size), tx);
4177 * Insert the new object into the directory.
4179 (void) zfs_link_create(dl, zp, tx, ZNEW);
4181 if (flags & FIGNORECASE)
4183 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
4186 zfs_acl_ids_free(&acl_ids);
4190 zfs_dirent_unlock(dl);
4192 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4193 zil_commit(zilog, 0);
4200 * Return, in the buffer contained in the provided uio structure,
4201 * the symbolic path referred to by vp.
4203 * IN: vp - vnode of symbolic link.
4204 * uio - structure to contain the link path.
4205 * cr - credentials of caller.
4206 * ct - caller context
4208 * OUT: uio - structure containing the link path.
4210 * RETURN: 0 on success, error code on failure.
4213 * vp - atime updated
4217 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
4219 znode_t *zp = VTOZ(vp);
4220 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4226 mutex_enter(&zp->z_lock);
4228 error = sa_lookup_uio(zp->z_sa_hdl,
4229 SA_ZPL_SYMLINK(zfsvfs), uio);
4231 error = zfs_sa_readlink(zp, uio);
4232 mutex_exit(&zp->z_lock);
4234 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4241 * Insert a new entry into directory tdvp referencing svp.
4243 * IN: tdvp - Directory to contain new entry.
4244 * svp - vnode of new entry.
4245 * name - name of new entry.
4246 * cr - credentials of caller.
4247 * ct - caller context
4249 * RETURN: 0 on success, error code on failure.
4252 * tdvp - ctime|mtime updated
4253 * svp - ctime updated
4257 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
4258 caller_context_t *ct, int flags)
4260 znode_t *dzp = VTOZ(tdvp);
4262 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4272 ASSERT(tdvp->v_type == VDIR);
4276 zilog = zfsvfs->z_log;
4278 if (VOP_REALVP(svp, &realvp, ct) == 0)
4282 * POSIX dictates that we return EPERM here.
4283 * Better choices include ENOTSUP or EISDIR.
4285 if (svp->v_type == VDIR) {
4287 return (SET_ERROR(EPERM));
4290 if (svp->v_vfsp != tdvp->v_vfsp || zfsctl_is_node(svp)) {
4292 return (SET_ERROR(EXDEV));
4298 /* Prevent links to .zfs/shares files */
4300 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4301 &parent, sizeof (uint64_t))) != 0) {
4305 if (parent == zfsvfs->z_shares_dir) {
4307 return (SET_ERROR(EPERM));
4310 if (zfsvfs->z_utf8 && u8_validate(name,
4311 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4313 return (SET_ERROR(EILSEQ));
4315 if (flags & FIGNORECASE)
4319 * We do not support links between attributes and non-attributes
4320 * because of the potential security risk of creating links
4321 * into "normal" file space in order to circumvent restrictions
4322 * imposed in attribute space.
4324 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4326 return (SET_ERROR(EINVAL));
4330 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4331 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) {
4333 return (SET_ERROR(EPERM));
4336 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4343 * Attempt to lock directory; fail if entry already exists.
4345 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4351 tx = dmu_tx_create(zfsvfs->z_os);
4352 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4353 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4354 zfs_sa_upgrade_txholds(tx, szp);
4355 zfs_sa_upgrade_txholds(tx, dzp);
4356 error = dmu_tx_assign(tx, TXG_NOWAIT);
4358 zfs_dirent_unlock(dl);
4359 if (error == ERESTART) {
4369 error = zfs_link_create(dl, szp, tx, 0);
4372 uint64_t txtype = TX_LINK;
4373 if (flags & FIGNORECASE)
4375 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4380 zfs_dirent_unlock(dl);
4383 vnevent_link(svp, ct);
4386 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4387 zil_commit(zilog, 0);
4395 * zfs_null_putapage() is used when the file system has been force
4396 * unmounted. It just drops the pages.
4400 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4401 size_t *lenp, int flags, cred_t *cr)
4403 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4408 * Push a page out to disk, klustering if possible.
4410 * IN: vp - file to push page to.
4411 * pp - page to push.
4412 * flags - additional flags.
4413 * cr - credentials of caller.
4415 * OUT: offp - start of range pushed.
4416 * lenp - len of range pushed.
4418 * RETURN: 0 on success, error code on failure.
4420 * NOTE: callers must have locked the page to be pushed. On
4421 * exit, the page (and all other pages in the kluster) must be
4426 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4427 size_t *lenp, int flags, cred_t *cr)
4429 znode_t *zp = VTOZ(vp);
4430 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4432 u_offset_t off, koff;
4439 * If our blocksize is bigger than the page size, try to kluster
4440 * multiple pages so that we write a full block (thus avoiding
4441 * a read-modify-write).
4443 if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4444 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4445 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4446 ASSERT(koff <= zp->z_size);
4447 if (koff + klen > zp->z_size)
4448 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4449 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4451 ASSERT3U(btop(len), ==, btopr(len));
4454 * Can't push pages past end-of-file.
4456 if (off >= zp->z_size) {
4457 /* ignore all pages */
4460 } else if (off + len > zp->z_size) {
4461 int npages = btopr(zp->z_size - off);
4464 page_list_break(&pp, &trunc, npages);
4465 /* ignore pages past end of file */
4467 pvn_write_done(trunc, flags);
4468 len = zp->z_size - off;
4471 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4472 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4473 err = SET_ERROR(EDQUOT);
4477 tx = dmu_tx_create(zfsvfs->z_os);
4478 dmu_tx_hold_write(tx, zp->z_id, off, len);
4480 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4481 zfs_sa_upgrade_txholds(tx, zp);
4482 err = dmu_tx_assign(tx, TXG_NOWAIT);
4484 if (err == ERESTART) {
4493 if (zp->z_blksz <= PAGESIZE) {
4494 caddr_t va = zfs_map_page(pp, S_READ);
4495 ASSERT3U(len, <=, PAGESIZE);
4496 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4497 zfs_unmap_page(pp, va);
4499 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4503 uint64_t mtime[2], ctime[2];
4504 sa_bulk_attr_t bulk[3];
4507 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4509 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4511 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4513 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4515 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4520 pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4530 * Copy the portion of the file indicated from pages into the file.
4531 * The pages are stored in a page list attached to the files vnode.
4533 * IN: vp - vnode of file to push page data to.
4534 * off - position in file to put data.
4535 * len - amount of data to write.
4536 * flags - flags to control the operation.
4537 * cr - credentials of caller.
4538 * ct - caller context.
4540 * RETURN: 0 on success, error code on failure.
4543 * vp - ctime|mtime updated
4547 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4548 caller_context_t *ct)
4550 znode_t *zp = VTOZ(vp);
4551 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4563 * Align this request to the file block size in case we kluster.
4564 * XXX - this can result in pretty aggresive locking, which can
4565 * impact simultanious read/write access. One option might be
4566 * to break up long requests (len == 0) into block-by-block
4567 * operations to get narrower locking.
4569 blksz = zp->z_blksz;
4571 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4574 if (len > 0 && ISP2(blksz))
4575 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4581 * Search the entire vp list for pages >= io_off.
4583 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4584 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4587 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4589 if (off > zp->z_size) {
4590 /* past end of file */
4591 zfs_range_unlock(rl);
4596 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4598 for (off = io_off; io_off < off + len; io_off += io_len) {
4599 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4600 pp = page_lookup(vp, io_off,
4601 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4603 pp = page_lookup_nowait(vp, io_off,
4604 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4607 if (pp != NULL && pvn_getdirty(pp, flags)) {
4611 * Found a dirty page to push
4613 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4621 zfs_range_unlock(rl);
4622 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4623 zil_commit(zfsvfs->z_log, zp->z_id);
4631 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4633 znode_t *zp = VTOZ(vp);
4634 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4637 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4638 if (zp->z_sa_hdl == NULL) {
4640 * The fs has been unmounted, or we did a
4641 * suspend/resume and this file no longer exists.
4643 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4648 mutex_enter(&zp->z_lock);
4649 if (zp->z_unlinked) {
4651 * Fast path to recycle a vnode of a removed file.
4653 mutex_exit(&zp->z_lock);
4654 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4658 mutex_exit(&zp->z_lock);
4660 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4661 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4663 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4664 zfs_sa_upgrade_txholds(tx, zp);
4665 error = dmu_tx_assign(tx, TXG_WAIT);
4669 mutex_enter(&zp->z_lock);
4670 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4671 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4672 zp->z_atime_dirty = 0;
4673 mutex_exit(&zp->z_lock);
4677 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4682 * Bounds-check the seek operation.
4684 * IN: vp - vnode seeking within
4685 * ooff - old file offset
4686 * noffp - pointer to new file offset
4687 * ct - caller context
4689 * RETURN: 0 on success, EINVAL if new offset invalid.
4693 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4694 caller_context_t *ct)
4696 if (vp->v_type == VDIR)
4698 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4702 * Pre-filter the generic locking function to trap attempts to place
4703 * a mandatory lock on a memory mapped file.
4706 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4707 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4709 znode_t *zp = VTOZ(vp);
4710 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4716 * We are following the UFS semantics with respect to mapcnt
4717 * here: If we see that the file is mapped already, then we will
4718 * return an error, but we don't worry about races between this
4719 * function and zfs_map().
4721 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4723 return (SET_ERROR(EAGAIN));
4726 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4730 * If we can't find a page in the cache, we will create a new page
4731 * and fill it with file data. For efficiency, we may try to fill
4732 * multiple pages at once (klustering) to fill up the supplied page
4733 * list. Note that the pages to be filled are held with an exclusive
4734 * lock to prevent access by other threads while they are being filled.
4737 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4738 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4740 znode_t *zp = VTOZ(vp);
4741 page_t *pp, *cur_pp;
4742 objset_t *os = zp->z_zfsvfs->z_os;
4743 u_offset_t io_off, total;
4747 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4749 * We only have a single page, don't bother klustering
4753 pp = page_create_va(vp, io_off, io_len,
4754 PG_EXCL | PG_WAIT, seg, addr);
4757 * Try to find enough pages to fill the page list
4759 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4760 &io_len, off, plsz, 0);
4764 * The page already exists, nothing to do here.
4771 * Fill the pages in the kluster.
4774 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4777 ASSERT3U(io_off, ==, cur_pp->p_offset);
4778 va = zfs_map_page(cur_pp, S_WRITE);
4779 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4781 zfs_unmap_page(cur_pp, va);
4783 /* On error, toss the entire kluster */
4784 pvn_read_done(pp, B_ERROR);
4785 /* convert checksum errors into IO errors */
4787 err = SET_ERROR(EIO);
4790 cur_pp = cur_pp->p_next;
4794 * Fill in the page list array from the kluster starting
4795 * from the desired offset `off'.
4796 * NOTE: the page list will always be null terminated.
4798 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4799 ASSERT(pl == NULL || (*pl)->p_offset == off);
4805 * Return pointers to the pages for the file region [off, off + len]
4806 * in the pl array. If plsz is greater than len, this function may
4807 * also return page pointers from after the specified region
4808 * (i.e. the region [off, off + plsz]). These additional pages are
4809 * only returned if they are already in the cache, or were created as
4810 * part of a klustered read.
4812 * IN: vp - vnode of file to get data from.
4813 * off - position in file to get data from.
4814 * len - amount of data to retrieve.
4815 * plsz - length of provided page list.
4816 * seg - segment to obtain pages for.
4817 * addr - virtual address of fault.
4818 * rw - mode of created pages.
4819 * cr - credentials of caller.
4820 * ct - caller context.
4822 * OUT: protp - protection mode of created pages.
4823 * pl - list of pages created.
4825 * RETURN: 0 on success, error code on failure.
4828 * vp - atime updated
4832 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4833 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4834 enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4836 znode_t *zp = VTOZ(vp);
4837 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4841 /* we do our own caching, faultahead is unnecessary */
4844 else if (len > plsz)
4847 len = P2ROUNDUP(len, PAGESIZE);
4848 ASSERT(plsz >= len);
4857 * Loop through the requested range [off, off + len) looking
4858 * for pages. If we don't find a page, we will need to create
4859 * a new page and fill it with data from the file.
4862 if (*pl = page_lookup(vp, off, SE_SHARED))
4864 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4867 ASSERT3U((*pl)->p_offset, ==, off);
4871 ASSERT3U(len, >=, PAGESIZE);
4874 ASSERT3U(plsz, >=, PAGESIZE);
4881 * Fill out the page array with any pages already in the cache.
4884 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4891 * Release any pages we have previously locked.
4896 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4906 * Request a memory map for a section of a file. This code interacts
4907 * with common code and the VM system as follows:
4909 * - common code calls mmap(), which ends up in smmap_common()
4910 * - this calls VOP_MAP(), which takes you into (say) zfs
4911 * - zfs_map() calls as_map(), passing segvn_create() as the callback
4912 * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4913 * - zfs_addmap() updates z_mapcnt
4917 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4918 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4919 caller_context_t *ct)
4921 znode_t *zp = VTOZ(vp);
4922 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4923 segvn_crargs_t vn_a;
4929 if ((prot & PROT_WRITE) && (zp->z_pflags &
4930 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4932 return (SET_ERROR(EPERM));
4935 if ((prot & (PROT_READ | PROT_EXEC)) &&
4936 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4938 return (SET_ERROR(EACCES));
4941 if (vp->v_flag & VNOMAP) {
4943 return (SET_ERROR(ENOSYS));
4946 if (off < 0 || len > MAXOFFSET_T - off) {
4948 return (SET_ERROR(ENXIO));
4951 if (vp->v_type != VREG) {
4953 return (SET_ERROR(ENODEV));
4957 * If file is locked, disallow mapping.
4959 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
4961 return (SET_ERROR(EAGAIN));
4965 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4973 vn_a.offset = (u_offset_t)off;
4974 vn_a.type = flags & MAP_TYPE;
4976 vn_a.maxprot = maxprot;
4979 vn_a.flags = flags & ~MAP_TYPE;
4981 vn_a.lgrp_mem_policy_flags = 0;
4983 error = as_map(as, *addrp, len, segvn_create, &vn_a);
4992 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4993 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4994 caller_context_t *ct)
4996 uint64_t pages = btopr(len);
4998 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
5003 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
5004 * more accurate mtime for the associated file. Since we don't have a way of
5005 * detecting when the data was actually modified, we have to resort to
5006 * heuristics. If an explicit msync() is done, then we mark the mtime when the
5007 * last page is pushed. The problem occurs when the msync() call is omitted,
5008 * which by far the most common case:
5016 * putpage() via fsflush
5018 * If we wait until fsflush to come along, we can have a modification time that
5019 * is some arbitrary point in the future. In order to prevent this in the
5020 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
5025 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5026 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
5027 caller_context_t *ct)
5029 uint64_t pages = btopr(len);
5031 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
5032 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
5034 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
5035 vn_has_cached_data(vp))
5036 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
5042 * Free or allocate space in a file. Currently, this function only
5043 * supports the `F_FREESP' command. However, this command is somewhat
5044 * misnamed, as its functionality includes the ability to allocate as
5045 * well as free space.
5047 * IN: vp - vnode of file to free data in.
5048 * cmd - action to take (only F_FREESP supported).
5049 * bfp - section of file to free/alloc.
5050 * flag - current file open mode flags.
5051 * offset - current file offset.
5052 * cr - credentials of caller [UNUSED].
5053 * ct - caller context.
5055 * RETURN: 0 on success, error code on failure.
5058 * vp - ctime|mtime updated
5062 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
5063 offset_t offset, cred_t *cr, caller_context_t *ct)
5065 znode_t *zp = VTOZ(vp);
5066 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5073 if (cmd != F_FREESP) {
5075 return (SET_ERROR(EINVAL));
5078 if (error = convoff(vp, bfp, 0, offset)) {
5083 if (bfp->l_len < 0) {
5085 return (SET_ERROR(EINVAL));
5089 len = bfp->l_len; /* 0 means from off to end of file */
5091 error = zfs_freesp(zp, off, len, flag, TRUE);
5098 CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid));
5099 CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid));
5103 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
5105 znode_t *zp = VTOZ(vp);
5106 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5109 uint64_t object = zp->z_id;
5116 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
5117 &gen64, sizeof (uint64_t))) != 0) {
5122 gen = (uint32_t)gen64;
5124 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
5127 if (fidp->fid_len < size) {
5128 fidp->fid_len = size;
5130 return (SET_ERROR(ENOSPC));
5133 fidp->fid_len = size;
5136 zfid = (zfid_short_t *)fidp;
5138 zfid->zf_len = size;
5140 for (i = 0; i < sizeof (zfid->zf_object); i++)
5141 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
5143 /* Must have a non-zero generation number to distinguish from .zfs */
5146 for (i = 0; i < sizeof (zfid->zf_gen); i++)
5147 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
5149 if (size == LONG_FID_LEN) {
5150 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
5153 zlfid = (zfid_long_t *)fidp;
5155 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
5156 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
5158 /* XXX - this should be the generation number for the objset */
5159 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
5160 zlfid->zf_setgen[i] = 0;
5168 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
5169 caller_context_t *ct)
5181 case _PC_FILESIZEBITS:
5185 case _PC_XATTR_EXISTS:
5187 zfsvfs = zp->z_zfsvfs;
5191 error = zfs_dirent_lock(&dl, zp, "", &xzp,
5192 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
5194 zfs_dirent_unlock(dl);
5195 if (!zfs_dirempty(xzp))
5198 } else if (error == ENOENT) {
5200 * If there aren't extended attributes, it's the
5201 * same as having zero of them.
5208 case _PC_SATTR_ENABLED:
5209 case _PC_SATTR_EXISTS:
5210 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
5211 (vp->v_type == VREG || vp->v_type == VDIR);
5214 case _PC_ACCESS_FILTERING:
5215 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
5219 case _PC_ACL_ENABLED:
5220 *valp = _ACL_ACE_ENABLED;
5223 case _PC_MIN_HOLE_SIZE:
5224 *valp = (int)SPA_MINBLOCKSIZE;
5227 case _PC_TIMESTAMP_RESOLUTION:
5228 /* nanosecond timestamp resolution */
5232 case _PC_ACL_EXTENDED:
5240 case _PC_ACL_PATH_MAX:
5241 *valp = ACL_MAX_ENTRIES;
5245 return (EOPNOTSUPP);
5251 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5252 caller_context_t *ct)
5254 znode_t *zp = VTOZ(vp);
5255 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5257 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5261 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5269 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5270 caller_context_t *ct)
5272 znode_t *zp = VTOZ(vp);
5273 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5275 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5276 zilog_t *zilog = zfsvfs->z_log;
5281 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5283 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5284 zil_commit(zilog, 0);
5292 * The smallest read we may consider to loan out an arcbuf.
5293 * This must be a power of 2.
5295 int zcr_blksz_min = (1 << 10); /* 1K */
5297 * If set to less than the file block size, allow loaning out of an
5298 * arcbuf for a partial block read. This must be a power of 2.
5300 int zcr_blksz_max = (1 << 17); /* 128K */
5304 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5305 caller_context_t *ct)
5307 znode_t *zp = VTOZ(vp);
5308 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5309 int max_blksz = zfsvfs->z_max_blksz;
5310 uio_t *uio = &xuio->xu_uio;
5311 ssize_t size = uio->uio_resid;
5312 offset_t offset = uio->uio_loffset;
5317 int preamble, postamble;
5319 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5320 return (SET_ERROR(EINVAL));
5327 * Loan out an arc_buf for write if write size is bigger than
5328 * max_blksz, and the file's block size is also max_blksz.
5331 if (size < blksz || zp->z_blksz != blksz) {
5333 return (SET_ERROR(EINVAL));
5336 * Caller requests buffers for write before knowing where the
5337 * write offset might be (e.g. NFS TCP write).
5342 preamble = P2PHASE(offset, blksz);
5344 preamble = blksz - preamble;
5349 postamble = P2PHASE(size, blksz);
5352 fullblk = size / blksz;
5353 (void) dmu_xuio_init(xuio,
5354 (preamble != 0) + fullblk + (postamble != 0));
5355 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5356 int, postamble, int,
5357 (preamble != 0) + fullblk + (postamble != 0));
5360 * Have to fix iov base/len for partial buffers. They
5361 * currently represent full arc_buf's.
5364 /* data begins in the middle of the arc_buf */
5365 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5368 (void) dmu_xuio_add(xuio, abuf,
5369 blksz - preamble, preamble);
5372 for (i = 0; i < fullblk; i++) {
5373 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5376 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5380 /* data ends in the middle of the arc_buf */
5381 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5384 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5389 * Loan out an arc_buf for read if the read size is larger than
5390 * the current file block size. Block alignment is not
5391 * considered. Partial arc_buf will be loaned out for read.
5393 blksz = zp->z_blksz;
5394 if (blksz < zcr_blksz_min)
5395 blksz = zcr_blksz_min;
5396 if (blksz > zcr_blksz_max)
5397 blksz = zcr_blksz_max;
5398 /* avoid potential complexity of dealing with it */
5399 if (blksz > max_blksz) {
5401 return (SET_ERROR(EINVAL));
5404 maxsize = zp->z_size - uio->uio_loffset;
5408 if (size < blksz || vn_has_cached_data(vp)) {
5410 return (SET_ERROR(EINVAL));
5415 return (SET_ERROR(EINVAL));
5418 uio->uio_extflg = UIO_XUIO;
5419 XUIO_XUZC_RW(xuio) = ioflag;
5426 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5430 int ioflag = XUIO_XUZC_RW(xuio);
5432 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5434 i = dmu_xuio_cnt(xuio);
5436 abuf = dmu_xuio_arcbuf(xuio, i);
5438 * if abuf == NULL, it must be a write buffer
5439 * that has been returned in zfs_write().
5442 dmu_return_arcbuf(abuf);
5443 ASSERT(abuf || ioflag == UIO_WRITE);
5446 dmu_xuio_fini(xuio);
5451 * Predeclare these here so that the compiler assumes that
5452 * this is an "old style" function declaration that does
5453 * not include arguments => we won't get type mismatch errors
5454 * in the initializations that follow.
5456 static int zfs_inval();
5457 static int zfs_isdir();
5462 return (SET_ERROR(EINVAL));
5468 return (SET_ERROR(EISDIR));
5471 * Directory vnode operations template
5473 vnodeops_t *zfs_dvnodeops;
5474 const fs_operation_def_t zfs_dvnodeops_template[] = {
5475 VOPNAME_OPEN, { .vop_open = zfs_open },
5476 VOPNAME_CLOSE, { .vop_close = zfs_close },
5477 VOPNAME_READ, { .error = zfs_isdir },
5478 VOPNAME_WRITE, { .error = zfs_isdir },
5479 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5480 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5481 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5482 VOPNAME_ACCESS, { .vop_access = zfs_access },
5483 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5484 VOPNAME_CREATE, { .vop_create = zfs_create },
5485 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5486 VOPNAME_LINK, { .vop_link = zfs_link },
5487 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5488 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir },
5489 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5490 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5491 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink },
5492 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5493 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5494 VOPNAME_FID, { .vop_fid = zfs_fid },
5495 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5496 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5497 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5498 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5499 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5504 * Regular file vnode operations template
5506 vnodeops_t *zfs_fvnodeops;
5507 const fs_operation_def_t zfs_fvnodeops_template[] = {
5508 VOPNAME_OPEN, { .vop_open = zfs_open },
5509 VOPNAME_CLOSE, { .vop_close = zfs_close },
5510 VOPNAME_READ, { .vop_read = zfs_read },
5511 VOPNAME_WRITE, { .vop_write = zfs_write },
5512 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5513 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5514 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5515 VOPNAME_ACCESS, { .vop_access = zfs_access },
5516 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5517 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5518 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5519 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5520 VOPNAME_FID, { .vop_fid = zfs_fid },
5521 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5522 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock },
5523 VOPNAME_SPACE, { .vop_space = zfs_space },
5524 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage },
5525 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage },
5526 VOPNAME_MAP, { .vop_map = zfs_map },
5527 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap },
5528 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap },
5529 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5530 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5531 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5532 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5533 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf },
5534 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf },
5539 * Symbolic link vnode operations template
5541 vnodeops_t *zfs_symvnodeops;
5542 const fs_operation_def_t zfs_symvnodeops_template[] = {
5543 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5544 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5545 VOPNAME_ACCESS, { .vop_access = zfs_access },
5546 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5547 VOPNAME_READLINK, { .vop_readlink = zfs_readlink },
5548 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5549 VOPNAME_FID, { .vop_fid = zfs_fid },
5550 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5551 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5556 * special share hidden files vnode operations template
5558 vnodeops_t *zfs_sharevnodeops;
5559 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5560 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5561 VOPNAME_ACCESS, { .vop_access = zfs_access },
5562 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5563 VOPNAME_FID, { .vop_fid = zfs_fid },
5564 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5565 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5566 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5567 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5572 * Extended attribute directory vnode operations template
5574 * This template is identical to the directory vnodes
5575 * operation template except for restricted operations:
5579 * Note that there are other restrictions embedded in:
5580 * zfs_create() - restrict type to VREG
5581 * zfs_link() - no links into/out of attribute space
5582 * zfs_rename() - no moves into/out of attribute space
5584 vnodeops_t *zfs_xdvnodeops;
5585 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5586 VOPNAME_OPEN, { .vop_open = zfs_open },
5587 VOPNAME_CLOSE, { .vop_close = zfs_close },
5588 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5589 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5590 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5591 VOPNAME_ACCESS, { .vop_access = zfs_access },
5592 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5593 VOPNAME_CREATE, { .vop_create = zfs_create },
5594 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5595 VOPNAME_LINK, { .vop_link = zfs_link },
5596 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5597 VOPNAME_MKDIR, { .error = zfs_inval },
5598 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5599 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5600 VOPNAME_SYMLINK, { .error = zfs_inval },
5601 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5602 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5603 VOPNAME_FID, { .vop_fid = zfs_fid },
5604 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5605 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5606 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5607 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5608 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5613 * Error vnode operations template
5615 vnodeops_t *zfs_evnodeops;
5616 const fs_operation_def_t zfs_evnodeops_template[] = {
5617 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5618 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5624 ioflags(int ioflags)
5628 if (ioflags & IO_APPEND)
5630 if (ioflags & IO_NDELAY)
5632 if (ioflags & IO_SYNC)
5633 flags |= (FSYNC | FDSYNC | FRSYNC);
5639 zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage)
5641 znode_t *zp = VTOZ(vp);
5642 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5643 objset_t *os = zp->z_zfsvfs->z_os;
5644 vm_page_t mfirst, mlast, mreq;
5648 off_t startoff, endoff;
5650 vm_pindex_t reqstart, reqend;
5651 int pcount, lsize, reqsize, size;
5656 pcount = OFF_TO_IDX(round_page(count));
5658 object = mreq->object;
5661 KASSERT(vp->v_object == object, ("mismatching object"));
5663 if (pcount > 1 && zp->z_blksz > PAGESIZE) {
5664 startoff = rounddown(IDX_TO_OFF(mreq->pindex), zp->z_blksz);
5665 reqstart = OFF_TO_IDX(round_page(startoff));
5666 if (reqstart < m[0]->pindex)
5669 reqstart = reqstart - m[0]->pindex;
5670 endoff = roundup(IDX_TO_OFF(mreq->pindex) + PAGE_SIZE,
5672 reqend = OFF_TO_IDX(trunc_page(endoff)) - 1;
5673 if (reqend > m[pcount - 1]->pindex)
5674 reqend = m[pcount - 1]->pindex;
5675 reqsize = reqend - m[reqstart]->pindex + 1;
5676 KASSERT(reqstart <= reqpage && reqpage < reqstart + reqsize,
5677 ("reqpage beyond [reqstart, reqstart + reqsize[ bounds"));
5682 mfirst = m[reqstart];
5683 mlast = m[reqstart + reqsize - 1];
5685 zfs_vmobject_wlock(object);
5687 for (i = 0; i < reqstart; i++) {
5690 vm_page_unlock(m[i]);
5692 for (i = reqstart + reqsize; i < pcount; i++) {
5695 vm_page_unlock(m[i]);
5698 if (mreq->valid && reqsize == 1) {
5699 if (mreq->valid != VM_PAGE_BITS_ALL)
5700 vm_page_zero_invalid(mreq, TRUE);
5701 zfs_vmobject_wunlock(object);
5703 return (zfs_vm_pagerret_ok);
5706 PCPU_INC(cnt.v_vnodein);
5707 PCPU_ADD(cnt.v_vnodepgsin, reqsize);
5709 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) {
5710 for (i = reqstart; i < reqstart + reqsize; i++) {
5714 vm_page_unlock(m[i]);
5717 zfs_vmobject_wunlock(object);
5719 return (zfs_vm_pagerret_bad);
5723 if (IDX_TO_OFF(mlast->pindex) + lsize > object->un_pager.vnp.vnp_size)
5724 lsize = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mlast->pindex);
5726 zfs_vmobject_wunlock(object);
5728 for (i = reqstart; i < reqstart + reqsize; i++) {
5730 if (i == (reqstart + reqsize - 1))
5732 va = zfs_map_page(m[i], &sf);
5733 error = dmu_read(os, zp->z_id, IDX_TO_OFF(m[i]->pindex),
5734 size, va, DMU_READ_PREFETCH);
5735 if (size != PAGE_SIZE)
5736 bzero(va + size, PAGE_SIZE - size);
5742 zfs_vmobject_wlock(object);
5744 for (i = reqstart; i < reqstart + reqsize; i++) {
5746 m[i]->valid = VM_PAGE_BITS_ALL;
5747 KASSERT(m[i]->dirty == 0, ("zfs_getpages: page %p is dirty", m[i]));
5749 vm_page_readahead_finish(m[i]);
5752 zfs_vmobject_wunlock(object);
5754 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
5756 return (error ? zfs_vm_pagerret_error : zfs_vm_pagerret_ok);
5760 zfs_freebsd_getpages(ap)
5761 struct vop_getpages_args /* {
5766 vm_ooffset_t a_offset;
5770 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage));
5774 zfs_freebsd_bmap(ap)
5775 struct vop_bmap_args /* {
5778 struct bufobj **a_bop;
5785 if (ap->a_bop != NULL)
5786 *ap->a_bop = &ap->a_vp->v_bufobj;
5787 if (ap->a_bnp != NULL)
5788 *ap->a_bnp = ap->a_bn;
5789 if (ap->a_runp != NULL)
5791 if (ap->a_runb != NULL)
5798 zfs_freebsd_open(ap)
5799 struct vop_open_args /* {
5802 struct ucred *a_cred;
5803 struct thread *a_td;
5806 vnode_t *vp = ap->a_vp;
5807 znode_t *zp = VTOZ(vp);
5810 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL);
5812 vnode_create_vobject(vp, zp->z_size, ap->a_td);
5817 zfs_freebsd_close(ap)
5818 struct vop_close_args /* {
5821 struct ucred *a_cred;
5822 struct thread *a_td;
5826 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL));
5830 zfs_freebsd_ioctl(ap)
5831 struct vop_ioctl_args /* {
5841 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data,
5842 ap->a_fflag, ap->a_cred, NULL, NULL));
5846 zfs_freebsd_read(ap)
5847 struct vop_read_args /* {
5851 struct ucred *a_cred;
5855 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5860 zfs_freebsd_write(ap)
5861 struct vop_write_args /* {
5865 struct ucred *a_cred;
5869 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5874 zfs_freebsd_access(ap)
5875 struct vop_access_args /* {
5877 accmode_t a_accmode;
5878 struct ucred *a_cred;
5879 struct thread *a_td;
5882 vnode_t *vp = ap->a_vp;
5883 znode_t *zp = VTOZ(vp);
5888 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND,
5890 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND);
5892 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL);
5895 * VADMIN has to be handled by vaccess().
5898 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND);
5900 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
5901 zp->z_gid, accmode, ap->a_cred, NULL);
5906 * For VEXEC, ensure that at least one execute bit is set for
5909 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR &&
5910 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
5918 zfs_freebsd_lookup(ap)
5919 struct vop_lookup_args /* {
5920 struct vnode *a_dvp;
5921 struct vnode **a_vpp;
5922 struct componentname *a_cnp;
5925 struct componentname *cnp = ap->a_cnp;
5926 char nm[NAME_MAX + 1];
5928 ASSERT(cnp->cn_namelen < sizeof(nm));
5929 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm)));
5931 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop,
5932 cnp->cn_cred, cnp->cn_thread, 0));
5936 zfs_freebsd_create(ap)
5937 struct vop_create_args /* {
5938 struct vnode *a_dvp;
5939 struct vnode **a_vpp;
5940 struct componentname *a_cnp;
5941 struct vattr *a_vap;
5944 struct componentname *cnp = ap->a_cnp;
5945 vattr_t *vap = ap->a_vap;
5948 ASSERT(cnp->cn_flags & SAVENAME);
5950 vattr_init_mask(vap);
5951 mode = vap->va_mode & ALLPERMS;
5953 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode,
5954 ap->a_vpp, cnp->cn_cred, cnp->cn_thread));
5958 zfs_freebsd_remove(ap)
5959 struct vop_remove_args /* {
5960 struct vnode *a_dvp;
5962 struct componentname *a_cnp;
5966 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
5968 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr,
5969 ap->a_cnp->cn_cred, NULL, 0));
5973 zfs_freebsd_mkdir(ap)
5974 struct vop_mkdir_args /* {
5975 struct vnode *a_dvp;
5976 struct vnode **a_vpp;
5977 struct componentname *a_cnp;
5978 struct vattr *a_vap;
5981 vattr_t *vap = ap->a_vap;
5983 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
5985 vattr_init_mask(vap);
5987 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp,
5988 ap->a_cnp->cn_cred, NULL, 0, NULL));
5992 zfs_freebsd_rmdir(ap)
5993 struct vop_rmdir_args /* {
5994 struct vnode *a_dvp;
5996 struct componentname *a_cnp;
5999 struct componentname *cnp = ap->a_cnp;
6001 ASSERT(cnp->cn_flags & SAVENAME);
6003 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0));
6007 zfs_freebsd_readdir(ap)
6008 struct vop_readdir_args /* {
6011 struct ucred *a_cred;
6018 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag,
6019 ap->a_ncookies, ap->a_cookies));
6023 zfs_freebsd_fsync(ap)
6024 struct vop_fsync_args /* {
6027 struct thread *a_td;
6032 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL));
6036 zfs_freebsd_getattr(ap)
6037 struct vop_getattr_args /* {
6039 struct vattr *a_vap;
6040 struct ucred *a_cred;
6043 vattr_t *vap = ap->a_vap;
6049 xvap.xva_vattr = *vap;
6050 xvap.xva_vattr.va_mask |= AT_XVATTR;
6052 /* Convert chflags into ZFS-type flags. */
6053 /* XXX: what about SF_SETTABLE?. */
6054 XVA_SET_REQ(&xvap, XAT_IMMUTABLE);
6055 XVA_SET_REQ(&xvap, XAT_APPENDONLY);
6056 XVA_SET_REQ(&xvap, XAT_NOUNLINK);
6057 XVA_SET_REQ(&xvap, XAT_NODUMP);
6058 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL);
6062 /* Convert ZFS xattr into chflags. */
6063 #define FLAG_CHECK(fflag, xflag, xfield) do { \
6064 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \
6065 fflags |= (fflag); \
6067 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE,
6068 xvap.xva_xoptattrs.xoa_immutable);
6069 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY,
6070 xvap.xva_xoptattrs.xoa_appendonly);
6071 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK,
6072 xvap.xva_xoptattrs.xoa_nounlink);
6073 FLAG_CHECK(UF_NODUMP, XAT_NODUMP,
6074 xvap.xva_xoptattrs.xoa_nodump);
6076 *vap = xvap.xva_vattr;
6077 vap->va_flags = fflags;
6082 zfs_freebsd_setattr(ap)
6083 struct vop_setattr_args /* {
6085 struct vattr *a_vap;
6086 struct ucred *a_cred;
6089 vnode_t *vp = ap->a_vp;
6090 vattr_t *vap = ap->a_vap;
6091 cred_t *cred = ap->a_cred;
6096 vattr_init_mask(vap);
6097 vap->va_mask &= ~AT_NOSET;
6100 xvap.xva_vattr = *vap;
6102 zflags = VTOZ(vp)->z_pflags;
6104 if (vap->va_flags != VNOVAL) {
6105 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs;
6108 if (zfsvfs->z_use_fuids == B_FALSE)
6109 return (EOPNOTSUPP);
6111 fflags = vap->va_flags;
6112 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_NODUMP)) != 0)
6113 return (EOPNOTSUPP);
6115 * Unprivileged processes are not permitted to unset system
6116 * flags, or modify flags if any system flags are set.
6117 * Privileged non-jail processes may not modify system flags
6118 * if securelevel > 0 and any existing system flags are set.
6119 * Privileged jail processes behave like privileged non-jail
6120 * processes if the security.jail.chflags_allowed sysctl is
6121 * is non-zero; otherwise, they behave like unprivileged
6124 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 ||
6125 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) {
6127 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6128 error = securelevel_gt(cred, 0);
6134 * Callers may only modify the file flags on objects they
6135 * have VADMIN rights for.
6137 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0)
6140 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6144 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) {
6149 #define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \
6150 if (((fflags & (fflag)) && !(zflags & (zflag))) || \
6151 ((zflags & (zflag)) && !(fflags & (fflag)))) { \
6152 XVA_SET_REQ(&xvap, (xflag)); \
6153 (xfield) = ((fflags & (fflag)) != 0); \
6156 /* Convert chflags into ZFS-type flags. */
6157 /* XXX: what about SF_SETTABLE?. */
6158 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE,
6159 xvap.xva_xoptattrs.xoa_immutable);
6160 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY,
6161 xvap.xva_xoptattrs.xoa_appendonly);
6162 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK,
6163 xvap.xva_xoptattrs.xoa_nounlink);
6164 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP,
6165 xvap.xva_xoptattrs.xoa_nodump);
6168 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL));
6172 zfs_freebsd_rename(ap)
6173 struct vop_rename_args /* {
6174 struct vnode *a_fdvp;
6175 struct vnode *a_fvp;
6176 struct componentname *a_fcnp;
6177 struct vnode *a_tdvp;
6178 struct vnode *a_tvp;
6179 struct componentname *a_tcnp;
6182 vnode_t *fdvp = ap->a_fdvp;
6183 vnode_t *fvp = ap->a_fvp;
6184 vnode_t *tdvp = ap->a_tdvp;
6185 vnode_t *tvp = ap->a_tvp;
6188 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART));
6189 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART));
6191 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp,
6192 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0);
6207 zfs_freebsd_symlink(ap)
6208 struct vop_symlink_args /* {
6209 struct vnode *a_dvp;
6210 struct vnode **a_vpp;
6211 struct componentname *a_cnp;
6212 struct vattr *a_vap;
6216 struct componentname *cnp = ap->a_cnp;
6217 vattr_t *vap = ap->a_vap;
6219 ASSERT(cnp->cn_flags & SAVENAME);
6221 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */
6222 vattr_init_mask(vap);
6224 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap,
6225 ap->a_target, cnp->cn_cred, cnp->cn_thread));
6229 zfs_freebsd_readlink(ap)
6230 struct vop_readlink_args /* {
6233 struct ucred *a_cred;
6237 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL));
6241 zfs_freebsd_link(ap)
6242 struct vop_link_args /* {
6243 struct vnode *a_tdvp;
6245 struct componentname *a_cnp;
6248 struct componentname *cnp = ap->a_cnp;
6250 ASSERT(cnp->cn_flags & SAVENAME);
6252 return (zfs_link(ap->a_tdvp, ap->a_vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0));
6256 zfs_freebsd_inactive(ap)
6257 struct vop_inactive_args /* {
6259 struct thread *a_td;
6262 vnode_t *vp = ap->a_vp;
6264 zfs_inactive(vp, ap->a_td->td_ucred, NULL);
6269 zfs_freebsd_reclaim(ap)
6270 struct vop_reclaim_args /* {
6272 struct thread *a_td;
6275 vnode_t *vp = ap->a_vp;
6276 znode_t *zp = VTOZ(vp);
6277 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
6281 /* Destroy the vm object and flush associated pages. */
6282 vnode_destroy_vobject(vp);
6285 * z_teardown_inactive_lock protects from a race with
6286 * zfs_znode_dmu_fini in zfsvfs_teardown during
6289 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
6290 if (zp->z_sa_hdl == NULL)
6294 rw_exit(&zfsvfs->z_teardown_inactive_lock);
6302 struct vop_fid_args /* {
6308 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL));
6312 zfs_freebsd_pathconf(ap)
6313 struct vop_pathconf_args /* {
6316 register_t *a_retval;
6322 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL);
6324 *ap->a_retval = val;
6325 else if (error == EOPNOTSUPP)
6326 error = vop_stdpathconf(ap);
6331 zfs_freebsd_fifo_pathconf(ap)
6332 struct vop_pathconf_args /* {
6335 register_t *a_retval;
6339 switch (ap->a_name) {
6340 case _PC_ACL_EXTENDED:
6342 case _PC_ACL_PATH_MAX:
6343 case _PC_MAC_PRESENT:
6344 return (zfs_freebsd_pathconf(ap));
6346 return (fifo_specops.vop_pathconf(ap));
6351 * FreeBSD's extended attributes namespace defines file name prefix for ZFS'
6352 * extended attribute name:
6355 * system freebsd:system:
6356 * user (none, can be used to access ZFS fsattr(5) attributes
6357 * created on Solaris)
6360 zfs_create_attrname(int attrnamespace, const char *name, char *attrname,
6363 const char *namespace, *prefix, *suffix;
6365 /* We don't allow '/' character in attribute name. */
6366 if (strchr(name, '/') != NULL)
6368 /* We don't allow attribute names that start with "freebsd:" string. */
6369 if (strncmp(name, "freebsd:", 8) == 0)
6372 bzero(attrname, size);
6374 switch (attrnamespace) {
6375 case EXTATTR_NAMESPACE_USER:
6377 prefix = "freebsd:";
6378 namespace = EXTATTR_NAMESPACE_USER_STRING;
6382 * This is the default namespace by which we can access all
6383 * attributes created on Solaris.
6385 prefix = namespace = suffix = "";
6388 case EXTATTR_NAMESPACE_SYSTEM:
6389 prefix = "freebsd:";
6390 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING;
6393 case EXTATTR_NAMESPACE_EMPTY:
6397 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix,
6399 return (ENAMETOOLONG);
6405 * Vnode operating to retrieve a named extended attribute.
6408 zfs_getextattr(struct vop_getextattr_args *ap)
6411 IN struct vnode *a_vp;
6412 IN int a_attrnamespace;
6413 IN const char *a_name;
6414 INOUT struct uio *a_uio;
6416 IN struct ucred *a_cred;
6417 IN struct thread *a_td;
6421 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6422 struct thread *td = ap->a_td;
6423 struct nameidata nd;
6426 vnode_t *xvp = NULL, *vp;
6429 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6430 ap->a_cred, ap->a_td, VREAD);
6434 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6441 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6449 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6451 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL);
6453 NDFREE(&nd, NDF_ONLY_PNBUF);
6456 if (error == ENOENT)
6461 if (ap->a_size != NULL) {
6462 error = VOP_GETATTR(vp, &va, ap->a_cred);
6464 *ap->a_size = (size_t)va.va_size;
6465 } else if (ap->a_uio != NULL)
6466 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6469 vn_close(vp, flags, ap->a_cred, td);
6476 * Vnode operation to remove a named attribute.
6479 zfs_deleteextattr(struct vop_deleteextattr_args *ap)
6482 IN struct vnode *a_vp;
6483 IN int a_attrnamespace;
6484 IN const char *a_name;
6485 IN struct ucred *a_cred;
6486 IN struct thread *a_td;
6490 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6491 struct thread *td = ap->a_td;
6492 struct nameidata nd;
6495 vnode_t *xvp = NULL, *vp;
6498 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6499 ap->a_cred, ap->a_td, VWRITE);
6503 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6510 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6517 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF,
6518 UIO_SYSSPACE, attrname, xvp, td);
6521 NDFREE(&nd, NDF_ONLY_PNBUF);
6524 if (error == ENOENT)
6528 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
6531 if (vp == nd.ni_dvp)
6541 * Vnode operation to set a named attribute.
6544 zfs_setextattr(struct vop_setextattr_args *ap)
6547 IN struct vnode *a_vp;
6548 IN int a_attrnamespace;
6549 IN const char *a_name;
6550 INOUT struct uio *a_uio;
6551 IN struct ucred *a_cred;
6552 IN struct thread *a_td;
6556 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6557 struct thread *td = ap->a_td;
6558 struct nameidata nd;
6561 vnode_t *xvp = NULL, *vp;
6564 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6565 ap->a_cred, ap->a_td, VWRITE);
6569 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6576 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6577 LOOKUP_XATTR | CREATE_XATTR_DIR);
6583 flags = FFLAGS(O_WRONLY | O_CREAT);
6584 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6586 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL);
6588 NDFREE(&nd, NDF_ONLY_PNBUF);
6596 error = VOP_SETATTR(vp, &va, ap->a_cred);
6598 VOP_WRITE(vp, ap->a_uio, IO_UNIT | IO_SYNC, ap->a_cred);
6601 vn_close(vp, flags, ap->a_cred, td);
6608 * Vnode operation to retrieve extended attributes on a vnode.
6611 zfs_listextattr(struct vop_listextattr_args *ap)
6614 IN struct vnode *a_vp;
6615 IN int a_attrnamespace;
6616 INOUT struct uio *a_uio;
6618 IN struct ucred *a_cred;
6619 IN struct thread *a_td;
6623 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6624 struct thread *td = ap->a_td;
6625 struct nameidata nd;
6626 char attrprefix[16];
6627 u_char dirbuf[sizeof(struct dirent)];
6630 struct uio auio, *uio = ap->a_uio;
6631 size_t *sizep = ap->a_size;
6633 vnode_t *xvp = NULL, *vp;
6634 int done, error, eof, pos;
6636 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6637 ap->a_cred, ap->a_td, VREAD);
6641 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix,
6642 sizeof(attrprefix));
6645 plen = strlen(attrprefix);
6652 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6657 * ENOATTR means that the EA directory does not yet exist,
6658 * i.e. there are no extended attributes there.
6660 if (error == ENOATTR)
6665 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED,
6666 UIO_SYSSPACE, ".", xvp, td);
6669 NDFREE(&nd, NDF_ONLY_PNBUF);
6675 auio.uio_iov = &aiov;
6676 auio.uio_iovcnt = 1;
6677 auio.uio_segflg = UIO_SYSSPACE;
6679 auio.uio_rw = UIO_READ;
6680 auio.uio_offset = 0;
6685 aiov.iov_base = (void *)dirbuf;
6686 aiov.iov_len = sizeof(dirbuf);
6687 auio.uio_resid = sizeof(dirbuf);
6688 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL);
6689 done = sizeof(dirbuf) - auio.uio_resid;
6692 for (pos = 0; pos < done;) {
6693 dp = (struct dirent *)(dirbuf + pos);
6694 pos += dp->d_reclen;
6696 * XXX: Temporarily we also accept DT_UNKNOWN, as this
6697 * is what we get when attribute was created on Solaris.
6699 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN)
6701 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0)
6703 else if (strncmp(dp->d_name, attrprefix, plen) != 0)
6705 nlen = dp->d_namlen - plen;
6708 else if (uio != NULL) {
6710 * Format of extattr name entry is one byte for
6711 * length and the rest for name.
6713 error = uiomove(&nlen, 1, uio->uio_rw, uio);
6715 error = uiomove(dp->d_name + plen, nlen,
6722 } while (!eof && error == 0);
6731 zfs_freebsd_getacl(ap)
6732 struct vop_getacl_args /* {
6741 vsecattr_t vsecattr;
6743 if (ap->a_type != ACL_TYPE_NFS4)
6746 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
6747 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL))
6750 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt);
6751 if (vsecattr.vsa_aclentp != NULL)
6752 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz);
6758 zfs_freebsd_setacl(ap)
6759 struct vop_setacl_args /* {
6768 vsecattr_t vsecattr;
6769 int aclbsize; /* size of acl list in bytes */
6772 if (ap->a_type != ACL_TYPE_NFS4)
6775 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES)
6779 * With NFSv4 ACLs, chmod(2) may need to add additional entries,
6780 * splitting every entry into two and appending "canonical six"
6781 * entries at the end. Don't allow for setting an ACL that would
6782 * cause chmod(2) to run out of ACL entries.
6784 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES)
6787 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR);
6791 vsecattr.vsa_mask = VSA_ACE;
6792 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t);
6793 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP);
6794 aaclp = vsecattr.vsa_aclentp;
6795 vsecattr.vsa_aclentsz = aclbsize;
6797 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp);
6798 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL);
6799 kmem_free(aaclp, aclbsize);
6805 zfs_freebsd_aclcheck(ap)
6806 struct vop_aclcheck_args /* {
6815 return (EOPNOTSUPP);
6818 struct vop_vector zfs_vnodeops;
6819 struct vop_vector zfs_fifoops;
6820 struct vop_vector zfs_shareops;
6822 struct vop_vector zfs_vnodeops = {
6823 .vop_default = &default_vnodeops,
6824 .vop_inactive = zfs_freebsd_inactive,
6825 .vop_reclaim = zfs_freebsd_reclaim,
6826 .vop_access = zfs_freebsd_access,
6827 #ifdef FREEBSD_NAMECACHE
6828 .vop_lookup = vfs_cache_lookup,
6829 .vop_cachedlookup = zfs_freebsd_lookup,
6831 .vop_lookup = zfs_freebsd_lookup,
6833 .vop_getattr = zfs_freebsd_getattr,
6834 .vop_setattr = zfs_freebsd_setattr,
6835 .vop_create = zfs_freebsd_create,
6836 .vop_mknod = zfs_freebsd_create,
6837 .vop_mkdir = zfs_freebsd_mkdir,
6838 .vop_readdir = zfs_freebsd_readdir,
6839 .vop_fsync = zfs_freebsd_fsync,
6840 .vop_open = zfs_freebsd_open,
6841 .vop_close = zfs_freebsd_close,
6842 .vop_rmdir = zfs_freebsd_rmdir,
6843 .vop_ioctl = zfs_freebsd_ioctl,
6844 .vop_link = zfs_freebsd_link,
6845 .vop_symlink = zfs_freebsd_symlink,
6846 .vop_readlink = zfs_freebsd_readlink,
6847 .vop_read = zfs_freebsd_read,
6848 .vop_write = zfs_freebsd_write,
6849 .vop_remove = zfs_freebsd_remove,
6850 .vop_rename = zfs_freebsd_rename,
6851 .vop_pathconf = zfs_freebsd_pathconf,
6852 .vop_bmap = zfs_freebsd_bmap,
6853 .vop_fid = zfs_freebsd_fid,
6854 .vop_getextattr = zfs_getextattr,
6855 .vop_deleteextattr = zfs_deleteextattr,
6856 .vop_setextattr = zfs_setextattr,
6857 .vop_listextattr = zfs_listextattr,
6858 .vop_getacl = zfs_freebsd_getacl,
6859 .vop_setacl = zfs_freebsd_setacl,
6860 .vop_aclcheck = zfs_freebsd_aclcheck,
6861 .vop_getpages = zfs_freebsd_getpages,
6864 struct vop_vector zfs_fifoops = {
6865 .vop_default = &fifo_specops,
6866 .vop_fsync = zfs_freebsd_fsync,
6867 .vop_access = zfs_freebsd_access,
6868 .vop_getattr = zfs_freebsd_getattr,
6869 .vop_inactive = zfs_freebsd_inactive,
6870 .vop_read = VOP_PANIC,
6871 .vop_reclaim = zfs_freebsd_reclaim,
6872 .vop_setattr = zfs_freebsd_setattr,
6873 .vop_write = VOP_PANIC,
6874 .vop_pathconf = zfs_freebsd_fifo_pathconf,
6875 .vop_fid = zfs_freebsd_fid,
6876 .vop_getacl = zfs_freebsd_getacl,
6877 .vop_setacl = zfs_freebsd_setacl,
6878 .vop_aclcheck = zfs_freebsd_aclcheck,
6882 * special share hidden files vnode operations template
6884 struct vop_vector zfs_shareops = {
6885 .vop_default = &default_vnodeops,
6886 .vop_access = zfs_freebsd_access,
6887 .vop_inactive = zfs_freebsd_inactive,
6888 .vop_reclaim = zfs_freebsd_reclaim,
6889 .vop_fid = zfs_freebsd_fid,
6890 .vop_pathconf = zfs_freebsd_pathconf,