4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
24 * Copyright 2014 Nexenta Systems, Inc. All rights reserved.
27 /* Portions Copyright 2007 Jeremy Teo */
28 /* Portions Copyright 2010 Robert Milkowski */
30 #include <sys/types.h>
31 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/sysmacros.h>
35 #include <sys/resource.h>
38 #include <sys/vnode.h>
42 #include <sys/taskq.h>
44 #include <sys/atomic.h>
45 #include <sys/namei.h>
47 #include <sys/cmn_err.h>
48 #include <sys/errno.h>
49 #include <sys/unistd.h>
50 #include <sys/zfs_dir.h>
51 #include <sys/zfs_ioctl.h>
52 #include <sys/fs/zfs.h>
54 #include <sys/dmu_objset.h>
60 #include <sys/dirent.h>
61 #include <sys/policy.h>
62 #include <sys/sunddi.h>
63 #include <sys/filio.h>
65 #include <sys/zfs_ctldir.h>
66 #include <sys/zfs_fuid.h>
67 #include <sys/zfs_sa.h>
69 #include <sys/zfs_rlock.h>
70 #include <sys/extdirent.h>
71 #include <sys/kidmap.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) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
110 * dmu_tx_assign(). This is critical because we don't want to block
111 * while holding locks.
113 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
114 * reduces lock contention and CPU usage when we must wait (note that if
115 * throughput is constrained by the storage, nearly every transaction
118 * Note, in particular, that if a lock is sometimes acquired before
119 * the tx assigns, and sometimes after (e.g. z_lock), then failing
120 * to use a non-blocking assign can deadlock the system. The scenario:
122 * Thread A has grabbed a lock before calling dmu_tx_assign().
123 * Thread B is in an already-assigned tx, and blocks for this lock.
124 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
125 * forever, because the previous txg can't quiesce until B's tx commits.
127 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
128 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
129 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
130 * to indicate that this operation has already called dmu_tx_wait().
131 * This will ensure that we don't retry forever, waiting a short bit
134 * (5) If the operation succeeded, generate the intent log entry for it
135 * before dropping locks. This ensures that the ordering of events
136 * in the intent log matches the order in which they actually occurred.
137 * During ZIL replay the zfs_log_* functions will update the sequence
138 * number to indicate the zil transaction has replayed.
140 * (6) At the end of each vnode op, the DMU tx must always commit,
141 * regardless of whether there were any errors.
143 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
144 * to ensure that synchronous semantics are provided when necessary.
146 * In general, this is how things should be ordered in each vnode op:
148 * ZFS_ENTER(zfsvfs); // exit if unmounted
150 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD())
151 * rw_enter(...); // grab any other locks you need
152 * tx = dmu_tx_create(...); // get DMU tx
153 * dmu_tx_hold_*(); // hold each object you might modify
154 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
156 * rw_exit(...); // drop locks
157 * zfs_dirent_unlock(dl); // unlock directory entry
158 * VN_RELE(...); // release held vnodes
159 * if (error == ERESTART) {
165 * dmu_tx_abort(tx); // abort DMU tx
166 * ZFS_EXIT(zfsvfs); // finished in zfs
167 * return (error); // really out of space
169 * error = do_real_work(); // do whatever this VOP does
171 * zfs_log_*(...); // on success, make ZIL entry
172 * dmu_tx_commit(tx); // commit DMU tx -- error or not
173 * rw_exit(...); // drop locks
174 * zfs_dirent_unlock(dl); // unlock directory entry
175 * VN_RELE(...); // release held vnodes
176 * zil_commit(zilog, foid); // synchronous when necessary
177 * ZFS_EXIT(zfsvfs); // finished in zfs
178 * return (error); // done, report error
183 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
185 znode_t *zp = VTOZ(*vpp);
186 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
191 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
192 ((flag & FAPPEND) == 0)) {
194 return (SET_ERROR(EPERM));
197 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
198 ZTOV(zp)->v_type == VREG &&
199 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
200 if (fs_vscan(*vpp, cr, 0) != 0) {
202 return (SET_ERROR(EACCES));
206 /* Keep a count of the synchronous opens in the znode */
207 if (flag & (FSYNC | FDSYNC))
208 atomic_inc_32(&zp->z_sync_cnt);
216 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
217 caller_context_t *ct)
219 znode_t *zp = VTOZ(vp);
220 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
223 * Clean up any locks held by this process on the vp.
225 cleanlocks(vp, ddi_get_pid(), 0);
226 cleanshares(vp, ddi_get_pid());
231 /* Decrement the synchronous opens in the znode */
232 if ((flag & (FSYNC | FDSYNC)) && (count == 1))
233 atomic_dec_32(&zp->z_sync_cnt);
235 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
236 ZTOV(zp)->v_type == VREG &&
237 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
238 VERIFY(fs_vscan(vp, cr, 1) == 0);
245 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
246 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
249 zfs_holey(vnode_t *vp, u_long cmd, offset_t *off)
251 znode_t *zp = VTOZ(vp);
252 uint64_t noff = (uint64_t)*off; /* new offset */
257 file_sz = zp->z_size;
258 if (noff >= file_sz) {
259 return (SET_ERROR(ENXIO));
262 if (cmd == _FIO_SEEK_HOLE)
267 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
270 return (SET_ERROR(ENXIO));
273 * We could find a hole that begins after the logical end-of-file,
274 * because dmu_offset_next() only works on whole blocks. If the
275 * EOF falls mid-block, then indicate that the "virtual hole"
276 * at the end of the file begins at the logical EOF, rather than
277 * at the end of the last block.
279 if (noff > file_sz) {
292 zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred,
293 int *rvalp, caller_context_t *ct)
305 * The following two ioctls are used by bfu. Faking out,
306 * necessary to avoid bfu errors.
315 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
316 return (SET_ERROR(EFAULT));
318 off = *(offset_t *)data;
321 zfsvfs = zp->z_zfsvfs;
325 /* offset parameter is in/out */
326 error = zfs_holey(vp, com, &off);
331 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
332 return (SET_ERROR(EFAULT));
334 *(offset_t *)data = off;
338 return (SET_ERROR(ENOTTY));
342 page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes)
349 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE
350 * aligned boundaries, if the range is not aligned. As a result a
351 * DEV_BSIZE subrange with partially dirty data may get marked as clean.
352 * It may happen that all DEV_BSIZE subranges are marked clean and thus
353 * the whole page would be considred clean despite have some dirty data.
354 * For this reason we should shrink the range to DEV_BSIZE aligned
355 * boundaries before calling vm_page_clear_dirty.
357 end = rounddown2(off + nbytes, DEV_BSIZE);
358 off = roundup2(off, DEV_BSIZE);
362 zfs_vmobject_assert_wlocked(obj);
365 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
367 if (vm_page_xbusied(pp)) {
369 * Reference the page before unlocking and
370 * sleeping so that the page daemon is less
371 * likely to reclaim it.
373 vm_page_reference(pp);
375 zfs_vmobject_wunlock(obj);
376 vm_page_busy_sleep(pp, "zfsmwb");
377 zfs_vmobject_wlock(obj);
381 } else if (pp == NULL) {
382 pp = vm_page_alloc(obj, OFF_TO_IDX(start),
383 VM_ALLOC_SYSTEM | VM_ALLOC_IFCACHED |
386 ASSERT(pp != NULL && !pp->valid);
391 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
392 vm_object_pip_add(obj, 1);
393 pmap_remove_write(pp);
395 vm_page_clear_dirty(pp, off, nbytes);
403 page_unbusy(vm_page_t pp)
407 vm_object_pip_subtract(pp->object, 1);
411 page_hold(vnode_t *vp, int64_t start)
417 zfs_vmobject_assert_wlocked(obj);
420 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
422 if (vm_page_xbusied(pp)) {
424 * Reference the page before unlocking and
425 * sleeping so that the page daemon is less
426 * likely to reclaim it.
428 vm_page_reference(pp);
430 zfs_vmobject_wunlock(obj);
431 vm_page_busy_sleep(pp, "zfsmwb");
432 zfs_vmobject_wlock(obj);
436 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
449 page_unhold(vm_page_t pp)
458 * When a file is memory mapped, we must keep the IO data synchronized
459 * between the DMU cache and the memory mapped pages. What this means:
461 * On Write: If we find a memory mapped page, we write to *both*
462 * the page and the dmu buffer.
465 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid,
466 int segflg, dmu_tx_t *tx)
473 ASSERT(segflg != UIO_NOCOPY);
474 ASSERT(vp->v_mount != NULL);
478 off = start & PAGEOFFSET;
479 zfs_vmobject_wlock(obj);
480 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
482 int nbytes = imin(PAGESIZE - off, len);
484 if ((pp = page_busy(vp, start, off, nbytes)) != NULL) {
485 zfs_vmobject_wunlock(obj);
487 va = zfs_map_page(pp, &sf);
488 (void) dmu_read(os, oid, start+off, nbytes,
489 va+off, DMU_READ_PREFETCH);;
492 zfs_vmobject_wlock(obj);
498 vm_object_pip_wakeupn(obj, 0);
499 zfs_vmobject_wunlock(obj);
503 * Read with UIO_NOCOPY flag means that sendfile(2) requests
504 * ZFS to populate a range of page cache pages with data.
506 * NOTE: this function could be optimized to pre-allocate
507 * all pages in advance, drain exclusive busy on all of them,
508 * map them into contiguous KVA region and populate them
509 * in one single dmu_read() call.
512 mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio)
514 znode_t *zp = VTOZ(vp);
515 objset_t *os = zp->z_zfsvfs->z_os;
525 ASSERT(uio->uio_segflg == UIO_NOCOPY);
526 ASSERT(vp->v_mount != NULL);
529 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0);
531 zfs_vmobject_wlock(obj);
532 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) {
533 int bytes = MIN(PAGESIZE, len);
535 pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_SBUSY |
536 VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY);
537 if (pp->valid == 0) {
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);
549 if (pp->wire_count == 0 && pp->valid == 0 &&
553 pp->valid = VM_PAGE_BITS_ALL;
554 vm_page_activate(pp);
558 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
563 uio->uio_resid -= bytes;
564 uio->uio_offset += bytes;
567 zfs_vmobject_wunlock(obj);
572 * When a file is memory mapped, we must keep the IO data synchronized
573 * between the DMU cache and the memory mapped pages. What this means:
575 * On Read: We "read" preferentially from memory mapped pages,
576 * else we default from the dmu buffer.
578 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
579 * the file is memory mapped.
582 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
584 znode_t *zp = VTOZ(vp);
592 ASSERT(vp->v_mount != NULL);
596 start = uio->uio_loffset;
597 off = start & PAGEOFFSET;
598 zfs_vmobject_wlock(obj);
599 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
601 uint64_t bytes = MIN(PAGESIZE - off, len);
603 if (pp = page_hold(vp, start)) {
607 zfs_vmobject_wunlock(obj);
608 va = zfs_map_page(pp, &sf);
609 error = uiomove(va + off, bytes, UIO_READ, uio);
611 zfs_vmobject_wlock(obj);
614 zfs_vmobject_wunlock(obj);
615 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
617 zfs_vmobject_wlock(obj);
624 zfs_vmobject_wunlock(obj);
628 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
631 * Read bytes from specified file into supplied buffer.
633 * IN: vp - vnode of file to be read from.
634 * uio - structure supplying read location, range info,
636 * ioflag - SYNC flags; used to provide FRSYNC semantics.
637 * cr - credentials of caller.
638 * ct - caller context
640 * OUT: uio - updated offset and range, buffer filled.
642 * RETURN: 0 on success, error code on failure.
645 * vp - atime updated if byte count > 0
649 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
651 znode_t *zp = VTOZ(vp);
652 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
661 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
663 return (SET_ERROR(EACCES));
667 * Validate file offset
669 if (uio->uio_loffset < (offset_t)0) {
671 return (SET_ERROR(EINVAL));
675 * Fasttrack empty reads
677 if (uio->uio_resid == 0) {
683 * Check for mandatory locks
685 if (MANDMODE(zp->z_mode)) {
686 if (error = chklock(vp, FREAD,
687 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
694 * If we're in FRSYNC mode, sync out this znode before reading it.
697 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
698 zil_commit(zfsvfs->z_log, zp->z_id);
701 * Lock the range against changes.
703 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
706 * If we are reading past end-of-file we can skip
707 * to the end; but we might still need to set atime.
709 if (uio->uio_loffset >= zp->z_size) {
714 ASSERT(uio->uio_loffset < zp->z_size);
715 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
718 if ((uio->uio_extflg == UIO_XUIO) &&
719 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
721 int blksz = zp->z_blksz;
722 uint64_t offset = uio->uio_loffset;
724 xuio = (xuio_t *)uio;
726 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
729 ASSERT(offset + n <= blksz);
732 (void) dmu_xuio_init(xuio, nblk);
734 if (vn_has_cached_data(vp)) {
736 * For simplicity, we always allocate a full buffer
737 * even if we only expect to read a portion of a block.
739 while (--nblk >= 0) {
740 (void) dmu_xuio_add(xuio,
741 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
749 nbytes = MIN(n, zfs_read_chunk_size -
750 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
753 if (uio->uio_segflg == UIO_NOCOPY)
754 error = mappedread_sf(vp, nbytes, uio);
756 #endif /* __FreeBSD__ */
757 if (vn_has_cached_data(vp)) {
758 error = mappedread(vp, nbytes, uio);
760 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
764 /* convert checksum errors into IO errors */
766 error = SET_ERROR(EIO);
773 zfs_range_unlock(rl);
775 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
781 * Write the bytes to a file.
783 * IN: vp - vnode of file to be written to.
784 * uio - structure supplying write location, range info,
786 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is
787 * set if in append mode.
788 * cr - credentials of caller.
789 * ct - caller context (NFS/CIFS fem monitor only)
791 * OUT: uio - updated offset and range.
793 * RETURN: 0 on success, error code on failure.
796 * vp - ctime|mtime updated if byte count > 0
801 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
803 znode_t *zp = VTOZ(vp);
804 rlim64_t limit = MAXOFFSET_T;
805 ssize_t start_resid = uio->uio_resid;
809 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
814 int max_blksz = zfsvfs->z_max_blksz;
817 iovec_t *aiov = NULL;
820 int iovcnt = uio->uio_iovcnt;
821 iovec_t *iovp = uio->uio_iov;
824 sa_bulk_attr_t bulk[4];
825 uint64_t mtime[2], ctime[2];
828 * Fasttrack empty write
834 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
840 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
841 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
842 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
844 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
848 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
849 * callers might not be able to detect properly that we are read-only,
850 * so check it explicitly here.
852 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
854 return (SET_ERROR(EROFS));
858 * If immutable or not appending then return EPERM
860 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
861 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
862 (uio->uio_loffset < zp->z_size))) {
864 return (SET_ERROR(EPERM));
867 zilog = zfsvfs->z_log;
870 * Validate file offset
872 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
875 return (SET_ERROR(EINVAL));
879 * Check for mandatory locks before calling zfs_range_lock()
880 * in order to prevent a deadlock with locks set via fcntl().
882 if (MANDMODE((mode_t)zp->z_mode) &&
883 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
890 * Pre-fault the pages to ensure slow (eg NFS) pages
892 * Skip this if uio contains loaned arc_buf.
894 if ((uio->uio_extflg == UIO_XUIO) &&
895 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
896 xuio = (xuio_t *)uio;
898 uio_prefaultpages(MIN(n, max_blksz), uio);
902 * If in append mode, set the io offset pointer to eof.
904 if (ioflag & FAPPEND) {
906 * Obtain an appending range lock to guarantee file append
907 * semantics. We reset the write offset once we have the lock.
909 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
911 if (rl->r_len == UINT64_MAX) {
913 * We overlocked the file because this write will cause
914 * the file block size to increase.
915 * Note that zp_size cannot change with this lock held.
919 uio->uio_loffset = woff;
922 * Note that if the file block size will change as a result of
923 * this write, then this range lock will lock the entire file
924 * so that we can re-write the block safely.
926 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
929 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) {
930 zfs_range_unlock(rl);
936 zfs_range_unlock(rl);
938 return (SET_ERROR(EFBIG));
941 if ((woff + n) > limit || woff > (limit - n))
944 /* Will this write extend the file length? */
945 write_eof = (woff + n > zp->z_size);
947 end_size = MAX(zp->z_size, woff + n);
950 * Write the file in reasonable size chunks. Each chunk is written
951 * in a separate transaction; this keeps the intent log records small
952 * and allows us to do more fine-grained space accounting.
956 woff = uio->uio_loffset;
957 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
958 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
960 dmu_return_arcbuf(abuf);
961 error = SET_ERROR(EDQUOT);
965 if (xuio && abuf == NULL) {
966 ASSERT(i_iov < iovcnt);
968 abuf = dmu_xuio_arcbuf(xuio, i_iov);
969 dmu_xuio_clear(xuio, i_iov);
970 DTRACE_PROBE3(zfs_cp_write, int, i_iov,
971 iovec_t *, aiov, arc_buf_t *, abuf);
972 ASSERT((aiov->iov_base == abuf->b_data) ||
973 ((char *)aiov->iov_base - (char *)abuf->b_data +
974 aiov->iov_len == arc_buf_size(abuf)));
976 } else if (abuf == NULL && n >= max_blksz &&
977 woff >= zp->z_size &&
978 P2PHASE(woff, max_blksz) == 0 &&
979 zp->z_blksz == max_blksz) {
981 * This write covers a full block. "Borrow" a buffer
982 * from the dmu so that we can fill it before we enter
983 * a transaction. This avoids the possibility of
984 * holding up the transaction if the data copy hangs
985 * up on a pagefault (e.g., from an NFS server mapping).
989 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
991 ASSERT(abuf != NULL);
992 ASSERT(arc_buf_size(abuf) == max_blksz);
993 if (error = uiocopy(abuf->b_data, max_blksz,
994 UIO_WRITE, uio, &cbytes)) {
995 dmu_return_arcbuf(abuf);
998 ASSERT(cbytes == max_blksz);
1002 * Start a transaction.
1004 tx = dmu_tx_create(zfsvfs->z_os);
1005 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1006 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
1007 zfs_sa_upgrade_txholds(tx, zp);
1008 error = dmu_tx_assign(tx, TXG_WAIT);
1012 dmu_return_arcbuf(abuf);
1017 * If zfs_range_lock() over-locked we grow the blocksize
1018 * and then reduce the lock range. This will only happen
1019 * on the first iteration since zfs_range_reduce() will
1020 * shrink down r_len to the appropriate size.
1022 if (rl->r_len == UINT64_MAX) {
1025 if (zp->z_blksz > max_blksz) {
1027 * File's blocksize is already larger than the
1028 * "recordsize" property. Only let it grow to
1029 * the next power of 2.
1031 ASSERT(!ISP2(zp->z_blksz));
1032 new_blksz = MIN(end_size,
1033 1 << highbit64(zp->z_blksz));
1035 new_blksz = MIN(end_size, max_blksz);
1037 zfs_grow_blocksize(zp, new_blksz, tx);
1038 zfs_range_reduce(rl, woff, n);
1042 * XXX - should we really limit each write to z_max_blksz?
1043 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
1045 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
1047 if (woff + nbytes > zp->z_size)
1048 vnode_pager_setsize(vp, woff + nbytes);
1051 tx_bytes = uio->uio_resid;
1052 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
1054 tx_bytes -= uio->uio_resid;
1057 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
1059 * If this is not a full block write, but we are
1060 * extending the file past EOF and this data starts
1061 * block-aligned, use assign_arcbuf(). Otherwise,
1062 * write via dmu_write().
1064 if (tx_bytes < max_blksz && (!write_eof ||
1065 aiov->iov_base != abuf->b_data)) {
1067 dmu_write(zfsvfs->z_os, zp->z_id, woff,
1068 aiov->iov_len, aiov->iov_base, tx);
1069 dmu_return_arcbuf(abuf);
1070 xuio_stat_wbuf_copied();
1072 ASSERT(xuio || tx_bytes == max_blksz);
1073 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
1076 ASSERT(tx_bytes <= uio->uio_resid);
1077 uioskip(uio, tx_bytes);
1079 if (tx_bytes && vn_has_cached_data(vp)) {
1080 update_pages(vp, woff, tx_bytes, zfsvfs->z_os,
1081 zp->z_id, uio->uio_segflg, tx);
1085 * If we made no progress, we're done. If we made even
1086 * partial progress, update the znode and ZIL accordingly.
1088 if (tx_bytes == 0) {
1089 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
1090 (void *)&zp->z_size, sizeof (uint64_t), tx);
1097 * Clear Set-UID/Set-GID bits on successful write if not
1098 * privileged and at least one of the excute bits is set.
1100 * It would be nice to to this after all writes have
1101 * been done, but that would still expose the ISUID/ISGID
1102 * to another app after the partial write is committed.
1104 * Note: we don't call zfs_fuid_map_id() here because
1105 * user 0 is not an ephemeral uid.
1107 mutex_enter(&zp->z_acl_lock);
1108 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
1109 (S_IXUSR >> 6))) != 0 &&
1110 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
1111 secpolicy_vnode_setid_retain(vp, cr,
1112 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
1114 zp->z_mode &= ~(S_ISUID | S_ISGID);
1115 newmode = zp->z_mode;
1116 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
1117 (void *)&newmode, sizeof (uint64_t), tx);
1119 mutex_exit(&zp->z_acl_lock);
1121 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
1125 * Update the file size (zp_size) if it has changed;
1126 * account for possible concurrent updates.
1128 while ((end_size = zp->z_size) < uio->uio_loffset) {
1129 (void) atomic_cas_64(&zp->z_size, end_size,
1134 * If we are replaying and eof is non zero then force
1135 * the file size to the specified eof. Note, there's no
1136 * concurrency during replay.
1138 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
1139 zp->z_size = zfsvfs->z_replay_eof;
1141 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1143 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
1148 ASSERT(tx_bytes == nbytes);
1153 uio_prefaultpages(MIN(n, max_blksz), uio);
1157 zfs_range_unlock(rl);
1160 * If we're in replay mode, or we made no progress, return error.
1161 * Otherwise, it's at least a partial write, so it's successful.
1163 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1168 if (ioflag & (FSYNC | FDSYNC) ||
1169 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1170 zil_commit(zilog, zp->z_id);
1177 zfs_get_done(zgd_t *zgd, int error)
1179 znode_t *zp = zgd->zgd_private;
1180 objset_t *os = zp->z_zfsvfs->z_os;
1183 dmu_buf_rele(zgd->zgd_db, zgd);
1185 zfs_range_unlock(zgd->zgd_rl);
1188 * Release the vnode asynchronously as we currently have the
1189 * txg stopped from syncing.
1191 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1193 if (error == 0 && zgd->zgd_bp)
1194 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1196 kmem_free(zgd, sizeof (zgd_t));
1200 static int zil_fault_io = 0;
1204 * Get data to generate a TX_WRITE intent log record.
1207 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1209 zfsvfs_t *zfsvfs = arg;
1210 objset_t *os = zfsvfs->z_os;
1212 uint64_t object = lr->lr_foid;
1213 uint64_t offset = lr->lr_offset;
1214 uint64_t size = lr->lr_length;
1215 blkptr_t *bp = &lr->lr_blkptr;
1220 ASSERT(zio != NULL);
1224 * Nothing to do if the file has been removed
1226 if (zfs_zget(zfsvfs, object, &zp) != 0)
1227 return (SET_ERROR(ENOENT));
1228 if (zp->z_unlinked) {
1230 * Release the vnode asynchronously as we currently have the
1231 * txg stopped from syncing.
1233 VN_RELE_ASYNC(ZTOV(zp),
1234 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1235 return (SET_ERROR(ENOENT));
1238 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1239 zgd->zgd_zilog = zfsvfs->z_log;
1240 zgd->zgd_private = zp;
1243 * Write records come in two flavors: immediate and indirect.
1244 * For small writes it's cheaper to store the data with the
1245 * log record (immediate); for large writes it's cheaper to
1246 * sync the data and get a pointer to it (indirect) so that
1247 * we don't have to write the data twice.
1249 if (buf != NULL) { /* immediate write */
1250 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1251 /* test for truncation needs to be done while range locked */
1252 if (offset >= zp->z_size) {
1253 error = SET_ERROR(ENOENT);
1255 error = dmu_read(os, object, offset, size, buf,
1256 DMU_READ_NO_PREFETCH);
1258 ASSERT(error == 0 || error == ENOENT);
1259 } else { /* indirect write */
1261 * Have to lock the whole block to ensure when it's
1262 * written out and it's checksum is being calculated
1263 * that no one can change the data. We need to re-check
1264 * blocksize after we get the lock in case it's changed!
1269 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1271 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1273 if (zp->z_blksz == size)
1276 zfs_range_unlock(zgd->zgd_rl);
1278 /* test for truncation needs to be done while range locked */
1279 if (lr->lr_offset >= zp->z_size)
1280 error = SET_ERROR(ENOENT);
1283 error = SET_ERROR(EIO);
1288 error = dmu_buf_hold(os, object, offset, zgd, &db,
1289 DMU_READ_NO_PREFETCH);
1292 blkptr_t *obp = dmu_buf_get_blkptr(db);
1294 ASSERT(BP_IS_HOLE(bp));
1301 ASSERT(db->db_offset == offset);
1302 ASSERT(db->db_size == size);
1304 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1306 ASSERT(error || lr->lr_length <= zp->z_blksz);
1309 * On success, we need to wait for the write I/O
1310 * initiated by dmu_sync() to complete before we can
1311 * release this dbuf. We will finish everything up
1312 * in the zfs_get_done() callback.
1317 if (error == EALREADY) {
1318 lr->lr_common.lrc_txtype = TX_WRITE2;
1324 zfs_get_done(zgd, error);
1331 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1332 caller_context_t *ct)
1334 znode_t *zp = VTOZ(vp);
1335 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1341 if (flag & V_ACE_MASK)
1342 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1344 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1351 * If vnode is for a device return a specfs vnode instead.
1354 specvp_check(vnode_t **vpp, cred_t *cr)
1358 if (IS_DEVVP(*vpp)) {
1361 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1364 error = SET_ERROR(ENOSYS);
1372 * Lookup an entry in a directory, or an extended attribute directory.
1373 * If it exists, return a held vnode reference for it.
1375 * IN: dvp - vnode of directory to search.
1376 * nm - name of entry to lookup.
1377 * pnp - full pathname to lookup [UNUSED].
1378 * flags - LOOKUP_XATTR set if looking for an attribute.
1379 * rdir - root directory vnode [UNUSED].
1380 * cr - credentials of caller.
1381 * ct - caller context
1382 * direntflags - directory lookup flags
1383 * realpnp - returned pathname.
1385 * OUT: vpp - vnode of located entry, NULL if not found.
1387 * RETURN: 0 on success, error code on failure.
1394 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp,
1395 int nameiop, cred_t *cr, kthread_t *td, int flags)
1397 znode_t *zdp = VTOZ(dvp);
1398 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1400 int *direntflags = NULL;
1401 void *realpnp = NULL;
1404 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1406 if (dvp->v_type != VDIR) {
1407 return (SET_ERROR(ENOTDIR));
1408 } else if (zdp->z_sa_hdl == NULL) {
1409 return (SET_ERROR(EIO));
1412 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1413 error = zfs_fastaccesschk_execute(zdp, cr);
1421 vnode_t *tvp = dnlc_lookup(dvp, nm);
1424 error = zfs_fastaccesschk_execute(zdp, cr);
1429 if (tvp == DNLC_NO_VNODE) {
1431 return (SET_ERROR(ENOENT));
1434 return (specvp_check(vpp, cr));
1440 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1447 if (flags & LOOKUP_XATTR) {
1450 * If the xattr property is off, refuse the lookup request.
1452 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1454 return (SET_ERROR(EINVAL));
1459 * We don't allow recursive attributes..
1460 * Maybe someday we will.
1462 if (zdp->z_pflags & ZFS_XATTR) {
1464 return (SET_ERROR(EINVAL));
1467 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1473 * Do we have permission to get into attribute directory?
1476 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1486 if (dvp->v_type != VDIR) {
1488 return (SET_ERROR(ENOTDIR));
1492 * Check accessibility of directory.
1495 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1500 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1501 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1503 return (SET_ERROR(EILSEQ));
1506 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1508 error = specvp_check(vpp, cr);
1510 /* Translate errors and add SAVENAME when needed. */
1511 if (cnp->cn_flags & ISLASTCN) {
1515 if (error == ENOENT) {
1516 error = EJUSTRETURN;
1517 cnp->cn_flags |= SAVENAME;
1523 cnp->cn_flags |= SAVENAME;
1527 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) {
1530 if (cnp->cn_flags & ISDOTDOT) {
1531 ltype = VOP_ISLOCKED(dvp);
1535 error = vn_lock(*vpp, cnp->cn_lkflags);
1536 if (cnp->cn_flags & ISDOTDOT)
1537 vn_lock(dvp, ltype | LK_RETRY);
1547 #ifdef FREEBSD_NAMECACHE
1549 * Insert name into cache (as non-existent) if appropriate.
1551 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) != 0)
1552 cache_enter(dvp, *vpp, cnp);
1554 * Insert name into cache if appropriate.
1556 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) {
1557 if (!(cnp->cn_flags & ISLASTCN) ||
1558 (nameiop != DELETE && nameiop != RENAME)) {
1559 cache_enter(dvp, *vpp, cnp);
1568 * Attempt to create a new entry in a directory. If the entry
1569 * already exists, truncate the file if permissible, else return
1570 * an error. Return the vp of the created or trunc'd file.
1572 * IN: dvp - vnode of directory to put new file entry in.
1573 * name - name of new file entry.
1574 * vap - attributes of new file.
1575 * excl - flag indicating exclusive or non-exclusive mode.
1576 * mode - mode to open file with.
1577 * cr - credentials of caller.
1578 * flag - large file flag [UNUSED].
1579 * ct - caller context
1580 * vsecp - ACL to be set
1582 * OUT: vpp - vnode of created or trunc'd entry.
1584 * RETURN: 0 on success, error code on failure.
1587 * dvp - ctime|mtime updated if new entry created
1588 * vp - ctime|mtime always, atime if new
1593 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode,
1594 vnode_t **vpp, cred_t *cr, kthread_t *td)
1596 znode_t *zp, *dzp = VTOZ(dvp);
1597 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1605 gid_t gid = crgetgid(cr);
1606 zfs_acl_ids_t acl_ids;
1607 boolean_t fuid_dirtied;
1608 boolean_t have_acl = B_FALSE;
1609 boolean_t waited = B_FALSE;
1614 * If we have an ephemeral id, ACL, or XVATTR then
1615 * make sure file system is at proper version
1618 ksid = crgetsid(cr, KSID_OWNER);
1620 uid = ksid_getid(ksid);
1624 if (zfsvfs->z_use_fuids == B_FALSE &&
1625 (vsecp || (vap->va_mask & AT_XVATTR) ||
1626 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1627 return (SET_ERROR(EINVAL));
1632 zilog = zfsvfs->z_log;
1634 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1635 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1637 return (SET_ERROR(EILSEQ));
1640 if (vap->va_mask & AT_XVATTR) {
1641 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
1642 crgetuid(cr), cr, vap->va_type)) != 0) {
1648 getnewvnode_reserve(1);
1653 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr))
1654 vap->va_mode &= ~S_ISVTX;
1656 if (*name == '\0') {
1658 * Null component name refers to the directory itself.
1665 /* possible VN_HOLD(zp) */
1668 if (flag & FIGNORECASE)
1671 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1675 zfs_acl_ids_free(&acl_ids);
1676 if (strcmp(name, "..") == 0)
1677 error = SET_ERROR(EISDIR);
1678 getnewvnode_drop_reserve();
1688 * Create a new file object and update the directory
1691 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1693 zfs_acl_ids_free(&acl_ids);
1698 * We only support the creation of regular files in
1699 * extended attribute directories.
1702 if ((dzp->z_pflags & ZFS_XATTR) &&
1703 (vap->va_type != VREG)) {
1705 zfs_acl_ids_free(&acl_ids);
1706 error = SET_ERROR(EINVAL);
1710 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1711 cr, vsecp, &acl_ids)) != 0)
1715 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1716 zfs_acl_ids_free(&acl_ids);
1717 error = SET_ERROR(EDQUOT);
1721 tx = dmu_tx_create(os);
1723 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1724 ZFS_SA_BASE_ATTR_SIZE);
1726 fuid_dirtied = zfsvfs->z_fuid_dirty;
1728 zfs_fuid_txhold(zfsvfs, tx);
1729 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1730 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1731 if (!zfsvfs->z_use_sa &&
1732 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1733 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1734 0, acl_ids.z_aclp->z_acl_bytes);
1736 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1738 zfs_dirent_unlock(dl);
1739 if (error == ERESTART) {
1745 zfs_acl_ids_free(&acl_ids);
1747 getnewvnode_drop_reserve();
1751 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1754 zfs_fuid_sync(zfsvfs, tx);
1756 (void) zfs_link_create(dl, zp, tx, ZNEW);
1757 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1758 if (flag & FIGNORECASE)
1760 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1761 vsecp, acl_ids.z_fuidp, vap);
1762 zfs_acl_ids_free(&acl_ids);
1765 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1768 zfs_acl_ids_free(&acl_ids);
1772 * A directory entry already exists for this name.
1775 * Can't truncate an existing file if in exclusive mode.
1778 error = SET_ERROR(EEXIST);
1782 * Can't open a directory for writing.
1784 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1785 error = SET_ERROR(EISDIR);
1789 * Verify requested access to file.
1791 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1795 mutex_enter(&dzp->z_lock);
1797 mutex_exit(&dzp->z_lock);
1800 * Truncate regular files if requested.
1802 if ((ZTOV(zp)->v_type == VREG) &&
1803 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1804 /* we can't hold any locks when calling zfs_freesp() */
1805 zfs_dirent_unlock(dl);
1807 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1809 vnevent_create(ZTOV(zp), ct);
1814 getnewvnode_drop_reserve();
1816 zfs_dirent_unlock(dl);
1823 error = specvp_check(vpp, cr);
1826 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1827 zil_commit(zilog, 0);
1834 * Remove an entry from a directory.
1836 * IN: dvp - vnode of directory to remove entry from.
1837 * name - name of entry to remove.
1838 * cr - credentials of caller.
1839 * ct - caller context
1840 * flags - case flags
1842 * RETURN: 0 on success, error code on failure.
1846 * vp - ctime (if nlink > 0)
1849 uint64_t null_xattr = 0;
1853 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1856 znode_t *zp, *dzp = VTOZ(dvp);
1859 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1861 uint64_t acl_obj, xattr_obj;
1862 uint64_t xattr_obj_unlinked = 0;
1866 boolean_t may_delete_now, delete_now = FALSE;
1867 boolean_t unlinked, toobig = FALSE;
1869 pathname_t *realnmp = NULL;
1873 boolean_t waited = B_FALSE;
1877 zilog = zfsvfs->z_log;
1879 if (flags & FIGNORECASE) {
1889 * Attempt to lock directory; fail if entry doesn't exist.
1891 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1901 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1906 * Need to use rmdir for removing directories.
1908 if (vp->v_type == VDIR) {
1909 error = SET_ERROR(EPERM);
1913 vnevent_remove(vp, dvp, name, ct);
1916 dnlc_remove(dvp, realnmp->pn_buf);
1918 dnlc_remove(dvp, name);
1921 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1925 * We may delete the znode now, or we may put it in the unlinked set;
1926 * it depends on whether we're the last link, and on whether there are
1927 * other holds on the vnode. So we dmu_tx_hold() the right things to
1928 * allow for either case.
1931 tx = dmu_tx_create(zfsvfs->z_os);
1932 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1933 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1934 zfs_sa_upgrade_txholds(tx, zp);
1935 zfs_sa_upgrade_txholds(tx, dzp);
1936 if (may_delete_now) {
1938 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1939 /* if the file is too big, only hold_free a token amount */
1940 dmu_tx_hold_free(tx, zp->z_id, 0,
1941 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1944 /* are there any extended attributes? */
1945 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1946 &xattr_obj, sizeof (xattr_obj));
1947 if (error == 0 && xattr_obj) {
1948 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1950 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1951 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1954 mutex_enter(&zp->z_lock);
1955 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1956 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1957 mutex_exit(&zp->z_lock);
1959 /* charge as an update -- would be nice not to charge at all */
1960 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1963 * Mark this transaction as typically resulting in a net free of
1964 * space, unless object removal will be delayed indefinitely
1965 * (due to active holds on the vnode due to the file being open).
1968 dmu_tx_mark_netfree(tx);
1970 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1972 zfs_dirent_unlock(dl);
1976 if (error == ERESTART) {
1990 * Remove the directory entry.
1992 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
2001 * Hold z_lock so that we can make sure that the ACL obj
2002 * hasn't changed. Could have been deleted due to
2005 mutex_enter(&zp->z_lock);
2007 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
2008 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
2009 delete_now = may_delete_now && !toobig &&
2010 vp->v_count == 1 && !vn_has_cached_data(vp) &&
2011 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
2018 panic("zfs_remove: delete_now branch taken");
2020 if (xattr_obj_unlinked) {
2021 ASSERT3U(xzp->z_links, ==, 2);
2022 mutex_enter(&xzp->z_lock);
2023 xzp->z_unlinked = 1;
2025 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
2026 &xzp->z_links, sizeof (xzp->z_links), tx);
2027 ASSERT3U(error, ==, 0);
2028 mutex_exit(&xzp->z_lock);
2029 zfs_unlinked_add(xzp, tx);
2032 error = sa_remove(zp->z_sa_hdl,
2033 SA_ZPL_XATTR(zfsvfs), tx);
2035 error = sa_update(zp->z_sa_hdl,
2036 SA_ZPL_XATTR(zfsvfs), &null_xattr,
2037 sizeof (uint64_t), tx);
2042 ASSERT0(vp->v_count);
2044 mutex_exit(&zp->z_lock);
2045 zfs_znode_delete(zp, tx);
2046 } else if (unlinked) {
2047 mutex_exit(&zp->z_lock);
2048 zfs_unlinked_add(zp, tx);
2050 vp->v_vflag |= VV_NOSYNC;
2055 if (flags & FIGNORECASE)
2057 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
2064 zfs_dirent_unlock(dl);
2071 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2072 zil_commit(zilog, 0);
2079 * Create a new directory and insert it into dvp using the name
2080 * provided. Return a pointer to the inserted directory.
2082 * IN: dvp - vnode of directory to add subdir to.
2083 * dirname - name of new directory.
2084 * vap - attributes of new directory.
2085 * cr - credentials of caller.
2086 * ct - caller context
2087 * flags - case flags
2088 * vsecp - ACL to be set
2090 * OUT: vpp - vnode of created directory.
2092 * RETURN: 0 on success, error code on failure.
2095 * dvp - ctime|mtime updated
2096 * vp - ctime|mtime|atime updated
2100 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
2101 caller_context_t *ct, int flags, vsecattr_t *vsecp)
2103 znode_t *zp, *dzp = VTOZ(dvp);
2104 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2113 gid_t gid = crgetgid(cr);
2114 zfs_acl_ids_t acl_ids;
2115 boolean_t fuid_dirtied;
2116 boolean_t waited = B_FALSE;
2118 ASSERT(vap->va_type == VDIR);
2121 * If we have an ephemeral id, ACL, or XVATTR then
2122 * make sure file system is at proper version
2125 ksid = crgetsid(cr, KSID_OWNER);
2127 uid = ksid_getid(ksid);
2130 if (zfsvfs->z_use_fuids == B_FALSE &&
2131 (vsecp || (vap->va_mask & AT_XVATTR) ||
2132 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
2133 return (SET_ERROR(EINVAL));
2137 zilog = zfsvfs->z_log;
2139 if (dzp->z_pflags & ZFS_XATTR) {
2141 return (SET_ERROR(EINVAL));
2144 if (zfsvfs->z_utf8 && u8_validate(dirname,
2145 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2147 return (SET_ERROR(EILSEQ));
2149 if (flags & FIGNORECASE)
2152 if (vap->va_mask & AT_XVATTR) {
2153 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
2154 crgetuid(cr), cr, vap->va_type)) != 0) {
2160 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
2161 vsecp, &acl_ids)) != 0) {
2166 getnewvnode_reserve(1);
2169 * First make sure the new directory doesn't exist.
2171 * Existence is checked first to make sure we don't return
2172 * EACCES instead of EEXIST which can cause some applications
2178 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
2180 zfs_acl_ids_free(&acl_ids);
2181 getnewvnode_drop_reserve();
2186 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
2187 zfs_acl_ids_free(&acl_ids);
2188 zfs_dirent_unlock(dl);
2189 getnewvnode_drop_reserve();
2194 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
2195 zfs_acl_ids_free(&acl_ids);
2196 zfs_dirent_unlock(dl);
2197 getnewvnode_drop_reserve();
2199 return (SET_ERROR(EDQUOT));
2203 * Add a new entry to the directory.
2205 tx = dmu_tx_create(zfsvfs->z_os);
2206 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2207 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2208 fuid_dirtied = zfsvfs->z_fuid_dirty;
2210 zfs_fuid_txhold(zfsvfs, tx);
2211 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2212 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2213 acl_ids.z_aclp->z_acl_bytes);
2216 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2217 ZFS_SA_BASE_ATTR_SIZE);
2219 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2221 zfs_dirent_unlock(dl);
2222 if (error == ERESTART) {
2228 zfs_acl_ids_free(&acl_ids);
2230 getnewvnode_drop_reserve();
2238 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2241 zfs_fuid_sync(zfsvfs, tx);
2244 * Now put new name in parent dir.
2246 (void) zfs_link_create(dl, zp, tx, ZNEW);
2250 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2251 if (flags & FIGNORECASE)
2253 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2254 acl_ids.z_fuidp, vap);
2256 zfs_acl_ids_free(&acl_ids);
2260 getnewvnode_drop_reserve();
2262 zfs_dirent_unlock(dl);
2264 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2265 zil_commit(zilog, 0);
2272 * Remove a directory subdir entry. If the current working
2273 * directory is the same as the subdir to be removed, the
2276 * IN: dvp - vnode of directory to remove from.
2277 * name - name of directory to be removed.
2278 * cwd - vnode of current working directory.
2279 * cr - credentials of caller.
2280 * ct - caller context
2281 * flags - case flags
2283 * RETURN: 0 on success, error code on failure.
2286 * dvp - ctime|mtime updated
2290 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2291 caller_context_t *ct, int flags)
2293 znode_t *dzp = VTOZ(dvp);
2296 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2302 boolean_t waited = B_FALSE;
2306 zilog = zfsvfs->z_log;
2308 if (flags & FIGNORECASE)
2314 * Attempt to lock directory; fail if entry doesn't exist.
2316 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2324 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2328 if (vp->v_type != VDIR) {
2329 error = SET_ERROR(ENOTDIR);
2334 error = SET_ERROR(EINVAL);
2338 vnevent_rmdir(vp, dvp, name, ct);
2341 * Grab a lock on the directory to make sure that noone is
2342 * trying to add (or lookup) entries while we are removing it.
2344 rw_enter(&zp->z_name_lock, RW_WRITER);
2347 * Grab a lock on the parent pointer to make sure we play well
2348 * with the treewalk and directory rename code.
2350 rw_enter(&zp->z_parent_lock, RW_WRITER);
2352 tx = dmu_tx_create(zfsvfs->z_os);
2353 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2354 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2355 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2356 zfs_sa_upgrade_txholds(tx, zp);
2357 zfs_sa_upgrade_txholds(tx, dzp);
2358 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2360 rw_exit(&zp->z_parent_lock);
2361 rw_exit(&zp->z_name_lock);
2362 zfs_dirent_unlock(dl);
2364 if (error == ERESTART) {
2375 #ifdef FREEBSD_NAMECACHE
2379 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2382 uint64_t txtype = TX_RMDIR;
2383 if (flags & FIGNORECASE)
2385 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2390 rw_exit(&zp->z_parent_lock);
2391 rw_exit(&zp->z_name_lock);
2392 #ifdef FREEBSD_NAMECACHE
2396 zfs_dirent_unlock(dl);
2400 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2401 zil_commit(zilog, 0);
2408 * Read as many directory entries as will fit into the provided
2409 * buffer from the given directory cursor position (specified in
2410 * the uio structure).
2412 * IN: vp - vnode of directory to read.
2413 * uio - structure supplying read location, range info,
2414 * and return buffer.
2415 * cr - credentials of caller.
2416 * ct - caller context
2417 * flags - case flags
2419 * OUT: uio - updated offset and range, buffer filled.
2420 * eofp - set to true if end-of-file detected.
2422 * RETURN: 0 on success, error code on failure.
2425 * vp - atime updated
2427 * Note that the low 4 bits of the cookie returned by zap is always zero.
2428 * This allows us to use the low range for "special" directory entries:
2429 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2430 * we use the offset 2 for the '.zfs' directory.
2434 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies)
2436 znode_t *zp = VTOZ(vp);
2440 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2445 zap_attribute_t zap;
2446 uint_t bytes_wanted;
2447 uint64_t offset; /* must be unsigned; checks for < 1 */
2453 boolean_t check_sysattrs;
2456 u_long *cooks = NULL;
2462 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2463 &parent, sizeof (parent))) != 0) {
2469 * If we are not given an eof variable,
2476 * Check for valid iov_len.
2478 if (uio->uio_iov->iov_len <= 0) {
2480 return (SET_ERROR(EINVAL));
2484 * Quit if directory has been removed (posix)
2486 if ((*eofp = zp->z_unlinked) != 0) {
2493 offset = uio->uio_loffset;
2494 prefetch = zp->z_zn_prefetch;
2497 * Initialize the iterator cursor.
2501 * Start iteration from the beginning of the directory.
2503 zap_cursor_init(&zc, os, zp->z_id);
2506 * The offset is a serialized cursor.
2508 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2512 * Get space to change directory entries into fs independent format.
2514 iovp = uio->uio_iov;
2515 bytes_wanted = iovp->iov_len;
2516 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2517 bufsize = bytes_wanted;
2518 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2519 odp = (struct dirent64 *)outbuf;
2521 bufsize = bytes_wanted;
2523 odp = (struct dirent64 *)iovp->iov_base;
2525 eodp = (struct edirent *)odp;
2527 if (ncookies != NULL) {
2529 * Minimum entry size is dirent size and 1 byte for a file name.
2531 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1);
2532 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK);
2537 * If this VFS supports the system attribute view interface; and
2538 * we're looking at an extended attribute directory; and we care
2539 * about normalization conflicts on this vfs; then we must check
2540 * for normalization conflicts with the sysattr name space.
2543 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2544 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2545 (flags & V_RDDIR_ENTFLAGS);
2551 * Transform to file-system independent format
2554 while (outcount < bytes_wanted) {
2557 off64_t *next = NULL;
2560 * Special case `.', `..', and `.zfs'.
2563 (void) strcpy(zap.za_name, ".");
2564 zap.za_normalization_conflict = 0;
2567 } else if (offset == 1) {
2568 (void) strcpy(zap.za_name, "..");
2569 zap.za_normalization_conflict = 0;
2572 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2573 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2574 zap.za_normalization_conflict = 0;
2575 objnum = ZFSCTL_INO_ROOT;
2581 if (error = zap_cursor_retrieve(&zc, &zap)) {
2582 if ((*eofp = (error == ENOENT)) != 0)
2588 if (zap.za_integer_length != 8 ||
2589 zap.za_num_integers != 1) {
2590 cmn_err(CE_WARN, "zap_readdir: bad directory "
2591 "entry, obj = %lld, offset = %lld\n",
2592 (u_longlong_t)zp->z_id,
2593 (u_longlong_t)offset);
2594 error = SET_ERROR(ENXIO);
2598 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2600 * MacOS X can extract the object type here such as:
2601 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2603 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2605 if (check_sysattrs && !zap.za_normalization_conflict) {
2607 zap.za_normalization_conflict =
2608 xattr_sysattr_casechk(zap.za_name);
2610 panic("%s:%u: TODO", __func__, __LINE__);
2615 if (flags & V_RDDIR_ACCFILTER) {
2617 * If we have no access at all, don't include
2618 * this entry in the returned information
2621 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2623 if (!zfs_has_access(ezp, cr)) {
2630 if (flags & V_RDDIR_ENTFLAGS)
2631 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2633 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2636 * Will this entry fit in the buffer?
2638 if (outcount + reclen > bufsize) {
2640 * Did we manage to fit anything in the buffer?
2643 error = SET_ERROR(EINVAL);
2648 if (flags & V_RDDIR_ENTFLAGS) {
2650 * Add extended flag entry:
2652 eodp->ed_ino = objnum;
2653 eodp->ed_reclen = reclen;
2654 /* NOTE: ed_off is the offset for the *next* entry */
2655 next = &(eodp->ed_off);
2656 eodp->ed_eflags = zap.za_normalization_conflict ?
2657 ED_CASE_CONFLICT : 0;
2658 (void) strncpy(eodp->ed_name, zap.za_name,
2659 EDIRENT_NAMELEN(reclen));
2660 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2665 odp->d_ino = objnum;
2666 odp->d_reclen = reclen;
2667 odp->d_namlen = strlen(zap.za_name);
2668 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1);
2670 odp = (dirent64_t *)((intptr_t)odp + reclen);
2674 ASSERT(outcount <= bufsize);
2676 /* Prefetch znode */
2678 dmu_prefetch(os, objnum, 0, 0);
2682 * Move to the next entry, fill in the previous offset.
2684 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2685 zap_cursor_advance(&zc);
2686 offset = zap_cursor_serialize(&zc);
2691 if (cooks != NULL) {
2694 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks));
2697 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2699 /* Subtract unused cookies */
2700 if (ncookies != NULL)
2701 *ncookies -= ncooks;
2703 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2704 iovp->iov_base += outcount;
2705 iovp->iov_len -= outcount;
2706 uio->uio_resid -= outcount;
2707 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2709 * Reset the pointer.
2711 offset = uio->uio_loffset;
2715 zap_cursor_fini(&zc);
2716 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2717 kmem_free(outbuf, bufsize);
2719 if (error == ENOENT)
2722 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2724 uio->uio_loffset = offset;
2726 if (error != 0 && cookies != NULL) {
2727 free(*cookies, M_TEMP);
2734 ulong_t zfs_fsync_sync_cnt = 4;
2737 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2739 znode_t *zp = VTOZ(vp);
2740 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2742 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2744 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2747 zil_commit(zfsvfs->z_log, zp->z_id);
2755 * Get the requested file attributes and place them in the provided
2758 * IN: vp - vnode of file.
2759 * vap - va_mask identifies requested attributes.
2760 * If AT_XVATTR set, then optional attrs are requested
2761 * flags - ATTR_NOACLCHECK (CIFS server context)
2762 * cr - credentials of caller.
2763 * ct - caller context
2765 * OUT: vap - attribute values.
2767 * RETURN: 0 (always succeeds).
2771 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2772 caller_context_t *ct)
2774 znode_t *zp = VTOZ(vp);
2775 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2778 u_longlong_t nblocks;
2780 uint64_t mtime[2], ctime[2], crtime[2], rdev;
2781 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2782 xoptattr_t *xoap = NULL;
2783 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2784 sa_bulk_attr_t bulk[4];
2790 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2792 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2793 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2794 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
2795 if (vp->v_type == VBLK || vp->v_type == VCHR)
2796 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
2799 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2805 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2806 * Also, if we are the owner don't bother, since owner should
2807 * always be allowed to read basic attributes of file.
2809 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2810 (vap->va_uid != crgetuid(cr))) {
2811 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2819 * Return all attributes. It's cheaper to provide the answer
2820 * than to determine whether we were asked the question.
2823 mutex_enter(&zp->z_lock);
2824 vap->va_type = IFTOVT(zp->z_mode);
2825 vap->va_mode = zp->z_mode & ~S_IFMT;
2827 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2829 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
2831 vap->va_nodeid = zp->z_id;
2832 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2833 links = zp->z_links + 1;
2835 links = zp->z_links;
2836 vap->va_nlink = MIN(links, LINK_MAX); /* nlink_t limit! */
2837 vap->va_size = zp->z_size;
2839 vap->va_rdev = vp->v_rdev;
2841 if (vp->v_type == VBLK || vp->v_type == VCHR)
2842 vap->va_rdev = zfs_cmpldev(rdev);
2844 vap->va_seq = zp->z_seq;
2845 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */
2846 vap->va_filerev = zp->z_seq;
2849 * Add in any requested optional attributes and the create time.
2850 * Also set the corresponding bits in the returned attribute bitmap.
2852 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2853 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2855 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2856 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2859 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2860 xoap->xoa_readonly =
2861 ((zp->z_pflags & ZFS_READONLY) != 0);
2862 XVA_SET_RTN(xvap, XAT_READONLY);
2865 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2867 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2868 XVA_SET_RTN(xvap, XAT_SYSTEM);
2871 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2873 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2874 XVA_SET_RTN(xvap, XAT_HIDDEN);
2877 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2878 xoap->xoa_nounlink =
2879 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2880 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2883 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2884 xoap->xoa_immutable =
2885 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2886 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2889 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2890 xoap->xoa_appendonly =
2891 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2892 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2895 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2897 ((zp->z_pflags & ZFS_NODUMP) != 0);
2898 XVA_SET_RTN(xvap, XAT_NODUMP);
2901 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2903 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2904 XVA_SET_RTN(xvap, XAT_OPAQUE);
2907 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2908 xoap->xoa_av_quarantined =
2909 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2910 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2913 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2914 xoap->xoa_av_modified =
2915 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2916 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2919 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2920 vp->v_type == VREG) {
2921 zfs_sa_get_scanstamp(zp, xvap);
2924 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2927 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2928 times, sizeof (times));
2929 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2930 XVA_SET_RTN(xvap, XAT_CREATETIME);
2933 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2934 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2935 XVA_SET_RTN(xvap, XAT_REPARSE);
2937 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2938 xoap->xoa_generation = zp->z_gen;
2939 XVA_SET_RTN(xvap, XAT_GEN);
2942 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2944 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2945 XVA_SET_RTN(xvap, XAT_OFFLINE);
2948 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2950 ((zp->z_pflags & ZFS_SPARSE) != 0);
2951 XVA_SET_RTN(xvap, XAT_SPARSE);
2955 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2956 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2957 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2958 ZFS_TIME_DECODE(&vap->va_birthtime, crtime);
2960 mutex_exit(&zp->z_lock);
2962 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2963 vap->va_blksize = blksize;
2964 vap->va_bytes = nblocks << 9; /* nblocks * 512 */
2966 if (zp->z_blksz == 0) {
2968 * Block size hasn't been set; suggest maximal I/O transfers.
2970 vap->va_blksize = zfsvfs->z_max_blksz;
2978 * Set the file attributes to the values contained in the
2981 * IN: vp - vnode of file to be modified.
2982 * vap - new attribute values.
2983 * If AT_XVATTR set, then optional attrs are being set
2984 * flags - ATTR_UTIME set if non-default time values provided.
2985 * - ATTR_NOACLCHECK (CIFS context only).
2986 * cr - credentials of caller.
2987 * ct - caller context
2989 * RETURN: 0 on success, error code on failure.
2992 * vp - ctime updated, mtime updated if size changed.
2996 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2997 caller_context_t *ct)
2999 znode_t *zp = VTOZ(vp);
3000 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3005 uint_t mask = vap->va_mask;
3006 uint_t saved_mask = 0;
3007 uint64_t saved_mode;
3010 uint64_t new_uid, new_gid;
3012 uint64_t mtime[2], ctime[2];
3014 int need_policy = FALSE;
3016 zfs_fuid_info_t *fuidp = NULL;
3017 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
3020 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
3021 boolean_t fuid_dirtied = B_FALSE;
3022 sa_bulk_attr_t bulk[7], xattr_bulk[7];
3023 int count = 0, xattr_count = 0;
3028 if (mask & AT_NOSET)
3029 return (SET_ERROR(EINVAL));
3034 zilog = zfsvfs->z_log;
3037 * Make sure that if we have ephemeral uid/gid or xvattr specified
3038 * that file system is at proper version level
3041 if (zfsvfs->z_use_fuids == B_FALSE &&
3042 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
3043 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
3044 (mask & AT_XVATTR))) {
3046 return (SET_ERROR(EINVAL));
3049 if (mask & AT_SIZE && vp->v_type == VDIR) {
3051 return (SET_ERROR(EISDIR));
3054 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
3056 return (SET_ERROR(EINVAL));
3060 * If this is an xvattr_t, then get a pointer to the structure of
3061 * optional attributes. If this is NULL, then we have a vattr_t.
3063 xoap = xva_getxoptattr(xvap);
3065 xva_init(&tmpxvattr);
3068 * Immutable files can only alter immutable bit and atime
3070 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
3071 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
3072 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
3074 return (SET_ERROR(EPERM));
3077 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
3079 return (SET_ERROR(EPERM));
3083 * Verify timestamps doesn't overflow 32 bits.
3084 * ZFS can handle large timestamps, but 32bit syscalls can't
3085 * handle times greater than 2039. This check should be removed
3086 * once large timestamps are fully supported.
3088 if (mask & (AT_ATIME | AT_MTIME)) {
3089 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
3090 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
3092 return (SET_ERROR(EOVERFLOW));
3100 /* Can this be moved to before the top label? */
3101 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
3103 return (SET_ERROR(EROFS));
3107 * First validate permissions
3110 if (mask & AT_SIZE) {
3112 * XXX - Note, we are not providing any open
3113 * mode flags here (like FNDELAY), so we may
3114 * block if there are locks present... this
3115 * should be addressed in openat().
3117 /* XXX - would it be OK to generate a log record here? */
3118 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
3125 if (mask & (AT_ATIME|AT_MTIME) ||
3126 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
3127 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
3128 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
3129 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
3130 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
3131 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
3132 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
3133 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
3137 if (mask & (AT_UID|AT_GID)) {
3138 int idmask = (mask & (AT_UID|AT_GID));
3143 * NOTE: even if a new mode is being set,
3144 * we may clear S_ISUID/S_ISGID bits.
3147 if (!(mask & AT_MODE))
3148 vap->va_mode = zp->z_mode;
3151 * Take ownership or chgrp to group we are a member of
3154 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
3155 take_group = (mask & AT_GID) &&
3156 zfs_groupmember(zfsvfs, vap->va_gid, cr);
3159 * If both AT_UID and AT_GID are set then take_owner and
3160 * take_group must both be set in order to allow taking
3163 * Otherwise, send the check through secpolicy_vnode_setattr()
3167 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
3168 ((idmask == AT_UID) && take_owner) ||
3169 ((idmask == AT_GID) && take_group)) {
3170 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
3171 skipaclchk, cr) == 0) {
3173 * Remove setuid/setgid for non-privileged users
3175 secpolicy_setid_clear(vap, vp, cr);
3176 trim_mask = (mask & (AT_UID|AT_GID));
3185 mutex_enter(&zp->z_lock);
3186 oldva.va_mode = zp->z_mode;
3187 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
3188 if (mask & AT_XVATTR) {
3190 * Update xvattr mask to include only those attributes
3191 * that are actually changing.
3193 * the bits will be restored prior to actually setting
3194 * the attributes so the caller thinks they were set.
3196 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
3197 if (xoap->xoa_appendonly !=
3198 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
3201 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
3202 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
3206 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
3207 if (xoap->xoa_nounlink !=
3208 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
3211 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
3212 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
3216 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
3217 if (xoap->xoa_immutable !=
3218 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
3221 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
3222 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
3226 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
3227 if (xoap->xoa_nodump !=
3228 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
3231 XVA_CLR_REQ(xvap, XAT_NODUMP);
3232 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
3236 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3237 if (xoap->xoa_av_modified !=
3238 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3241 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3242 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
3246 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3247 if ((vp->v_type != VREG &&
3248 xoap->xoa_av_quarantined) ||
3249 xoap->xoa_av_quarantined !=
3250 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3253 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3254 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
3258 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3259 mutex_exit(&zp->z_lock);
3261 return (SET_ERROR(EPERM));
3264 if (need_policy == FALSE &&
3265 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3266 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3271 mutex_exit(&zp->z_lock);
3273 if (mask & AT_MODE) {
3274 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3275 err = secpolicy_setid_setsticky_clear(vp, vap,
3281 trim_mask |= AT_MODE;
3289 * If trim_mask is set then take ownership
3290 * has been granted or write_acl is present and user
3291 * has the ability to modify mode. In that case remove
3292 * UID|GID and or MODE from mask so that
3293 * secpolicy_vnode_setattr() doesn't revoke it.
3297 saved_mask = vap->va_mask;
3298 vap->va_mask &= ~trim_mask;
3299 if (trim_mask & AT_MODE) {
3301 * Save the mode, as secpolicy_vnode_setattr()
3302 * will overwrite it with ova.va_mode.
3304 saved_mode = vap->va_mode;
3307 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3308 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3315 vap->va_mask |= saved_mask;
3316 if (trim_mask & AT_MODE) {
3318 * Recover the mode after
3319 * secpolicy_vnode_setattr().
3321 vap->va_mode = saved_mode;
3327 * secpolicy_vnode_setattr, or take ownership may have
3330 mask = vap->va_mask;
3332 if ((mask & (AT_UID | AT_GID))) {
3333 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3334 &xattr_obj, sizeof (xattr_obj));
3336 if (err == 0 && xattr_obj) {
3337 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3341 if (mask & AT_UID) {
3342 new_uid = zfs_fuid_create(zfsvfs,
3343 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3344 if (new_uid != zp->z_uid &&
3345 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3347 VN_RELE(ZTOV(attrzp));
3348 err = SET_ERROR(EDQUOT);
3353 if (mask & AT_GID) {
3354 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3355 cr, ZFS_GROUP, &fuidp);
3356 if (new_gid != zp->z_gid &&
3357 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3359 VN_RELE(ZTOV(attrzp));
3360 err = SET_ERROR(EDQUOT);
3365 tx = dmu_tx_create(zfsvfs->z_os);
3367 if (mask & AT_MODE) {
3368 uint64_t pmode = zp->z_mode;
3370 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3372 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3373 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3374 err = SET_ERROR(EPERM);
3378 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3381 mutex_enter(&zp->z_lock);
3382 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3384 * Are we upgrading ACL from old V0 format
3387 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3388 zfs_znode_acl_version(zp) ==
3389 ZFS_ACL_VERSION_INITIAL) {
3390 dmu_tx_hold_free(tx, acl_obj, 0,
3392 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3393 0, aclp->z_acl_bytes);
3395 dmu_tx_hold_write(tx, acl_obj, 0,
3398 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3399 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3400 0, aclp->z_acl_bytes);
3402 mutex_exit(&zp->z_lock);
3403 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3405 if ((mask & AT_XVATTR) &&
3406 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3407 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3409 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3413 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3416 fuid_dirtied = zfsvfs->z_fuid_dirty;
3418 zfs_fuid_txhold(zfsvfs, tx);
3420 zfs_sa_upgrade_txholds(tx, zp);
3422 err = dmu_tx_assign(tx, TXG_WAIT);
3428 * Set each attribute requested.
3429 * We group settings according to the locks they need to acquire.
3431 * Note: you cannot set ctime directly, although it will be
3432 * updated as a side-effect of calling this function.
3436 if (mask & (AT_UID|AT_GID|AT_MODE))
3437 mutex_enter(&zp->z_acl_lock);
3438 mutex_enter(&zp->z_lock);
3440 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3441 &zp->z_pflags, sizeof (zp->z_pflags));
3444 if (mask & (AT_UID|AT_GID|AT_MODE))
3445 mutex_enter(&attrzp->z_acl_lock);
3446 mutex_enter(&attrzp->z_lock);
3447 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3448 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3449 sizeof (attrzp->z_pflags));
3452 if (mask & (AT_UID|AT_GID)) {
3454 if (mask & AT_UID) {
3455 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3456 &new_uid, sizeof (new_uid));
3457 zp->z_uid = new_uid;
3459 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3460 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3462 attrzp->z_uid = new_uid;
3466 if (mask & AT_GID) {
3467 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3468 NULL, &new_gid, sizeof (new_gid));
3469 zp->z_gid = new_gid;
3471 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3472 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3474 attrzp->z_gid = new_gid;
3477 if (!(mask & AT_MODE)) {
3478 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3479 NULL, &new_mode, sizeof (new_mode));
3480 new_mode = zp->z_mode;
3482 err = zfs_acl_chown_setattr(zp);
3485 err = zfs_acl_chown_setattr(attrzp);
3490 if (mask & AT_MODE) {
3491 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3492 &new_mode, sizeof (new_mode));
3493 zp->z_mode = new_mode;
3494 ASSERT3U((uintptr_t)aclp, !=, 0);
3495 err = zfs_aclset_common(zp, aclp, cr, tx);
3497 if (zp->z_acl_cached)
3498 zfs_acl_free(zp->z_acl_cached);
3499 zp->z_acl_cached = aclp;
3504 if (mask & AT_ATIME) {
3505 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3506 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3507 &zp->z_atime, sizeof (zp->z_atime));
3510 if (mask & AT_MTIME) {
3511 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3512 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3513 mtime, sizeof (mtime));
3516 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3517 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3518 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3519 NULL, mtime, sizeof (mtime));
3520 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3521 &ctime, sizeof (ctime));
3522 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3524 } else if (mask != 0) {
3525 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3526 &ctime, sizeof (ctime));
3527 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3530 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3531 SA_ZPL_CTIME(zfsvfs), NULL,
3532 &ctime, sizeof (ctime));
3533 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3534 mtime, ctime, B_TRUE);
3538 * Do this after setting timestamps to prevent timestamp
3539 * update from toggling bit
3542 if (xoap && (mask & AT_XVATTR)) {
3545 * restore trimmed off masks
3546 * so that return masks can be set for caller.
3549 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3550 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3552 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3553 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3555 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3556 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3558 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3559 XVA_SET_REQ(xvap, XAT_NODUMP);
3561 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3562 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3564 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3565 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3568 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3569 ASSERT(vp->v_type == VREG);
3571 zfs_xvattr_set(zp, xvap, tx);
3575 zfs_fuid_sync(zfsvfs, tx);
3578 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3580 mutex_exit(&zp->z_lock);
3581 if (mask & (AT_UID|AT_GID|AT_MODE))
3582 mutex_exit(&zp->z_acl_lock);
3585 if (mask & (AT_UID|AT_GID|AT_MODE))
3586 mutex_exit(&attrzp->z_acl_lock);
3587 mutex_exit(&attrzp->z_lock);
3590 if (err == 0 && attrzp) {
3591 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3597 VN_RELE(ZTOV(attrzp));
3603 zfs_fuid_info_free(fuidp);
3609 if (err == ERESTART)
3612 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3617 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3618 zil_commit(zilog, 0);
3624 typedef struct zfs_zlock {
3625 krwlock_t *zl_rwlock; /* lock we acquired */
3626 znode_t *zl_znode; /* znode we held */
3627 struct zfs_zlock *zl_next; /* next in list */
3631 * Drop locks and release vnodes that were held by zfs_rename_lock().
3634 zfs_rename_unlock(zfs_zlock_t **zlpp)
3638 while ((zl = *zlpp) != NULL) {
3639 if (zl->zl_znode != NULL)
3640 VN_RELE(ZTOV(zl->zl_znode));
3641 rw_exit(zl->zl_rwlock);
3642 *zlpp = zl->zl_next;
3643 kmem_free(zl, sizeof (*zl));
3648 * Search back through the directory tree, using the ".." entries.
3649 * Lock each directory in the chain to prevent concurrent renames.
3650 * Fail any attempt to move a directory into one of its own descendants.
3651 * XXX - z_parent_lock can overlap with map or grow locks
3654 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3658 uint64_t rootid = zp->z_zfsvfs->z_root;
3659 uint64_t oidp = zp->z_id;
3660 krwlock_t *rwlp = &szp->z_parent_lock;
3661 krw_t rw = RW_WRITER;
3664 * First pass write-locks szp and compares to zp->z_id.
3665 * Later passes read-lock zp and compare to zp->z_parent.
3668 if (!rw_tryenter(rwlp, rw)) {
3670 * Another thread is renaming in this path.
3671 * Note that if we are a WRITER, we don't have any
3672 * parent_locks held yet.
3674 if (rw == RW_READER && zp->z_id > szp->z_id) {
3676 * Drop our locks and restart
3678 zfs_rename_unlock(&zl);
3682 rwlp = &szp->z_parent_lock;
3687 * Wait for other thread to drop its locks
3693 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3694 zl->zl_rwlock = rwlp;
3695 zl->zl_znode = NULL;
3696 zl->zl_next = *zlpp;
3699 if (oidp == szp->z_id) /* We're a descendant of szp */
3700 return (SET_ERROR(EINVAL));
3702 if (oidp == rootid) /* We've hit the top */
3705 if (rw == RW_READER) { /* i.e. not the first pass */
3706 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3711 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3712 &oidp, sizeof (oidp));
3713 rwlp = &zp->z_parent_lock;
3716 } while (zp->z_id != sdzp->z_id);
3722 * Move an entry from the provided source directory to the target
3723 * directory. Change the entry name as indicated.
3725 * IN: sdvp - Source directory containing the "old entry".
3726 * snm - Old entry name.
3727 * tdvp - Target directory to contain the "new entry".
3728 * tnm - New entry name.
3729 * cr - credentials of caller.
3730 * ct - caller context
3731 * flags - case flags
3733 * RETURN: 0 on success, error code on failure.
3736 * sdvp,tdvp - ctime|mtime updated
3740 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3741 caller_context_t *ct, int flags)
3743 znode_t *tdzp, *sdzp, *szp, *tzp;
3747 zfs_dirlock_t *sdl, *tdl;
3750 int cmp, serr, terr;
3753 boolean_t waited = B_FALSE;
3756 ZFS_VERIFY_ZP(tdzp);
3757 zfsvfs = tdzp->z_zfsvfs;
3759 zilog = zfsvfs->z_log;
3763 * In case sdzp is not valid, let's be sure to exit from the right
3766 if (sdzp->z_sa_hdl == NULL) {
3768 return (SET_ERROR(EIO));
3772 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3773 * ctldir appear to have the same v_vfsp.
3775 if (sdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) {
3777 return (SET_ERROR(EXDEV));
3780 if (zfsvfs->z_utf8 && u8_validate(tnm,
3781 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3783 return (SET_ERROR(EILSEQ));
3786 if (flags & FIGNORECASE)
3795 * This is to prevent the creation of links into attribute space
3796 * by renaming a linked file into/outof an attribute directory.
3797 * See the comment in zfs_link() for why this is considered bad.
3799 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3801 return (SET_ERROR(EINVAL));
3805 * Lock source and target directory entries. To prevent deadlock,
3806 * a lock ordering must be defined. We lock the directory with
3807 * the smallest object id first, or if it's a tie, the one with
3808 * the lexically first name.
3810 if (sdzp->z_id < tdzp->z_id) {
3812 } else if (sdzp->z_id > tdzp->z_id) {
3816 * First compare the two name arguments without
3817 * considering any case folding.
3819 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3821 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3822 ASSERT(error == 0 || !zfsvfs->z_utf8);
3825 * POSIX: "If the old argument and the new argument
3826 * both refer to links to the same existing file,
3827 * the rename() function shall return successfully
3828 * and perform no other action."
3834 * If the file system is case-folding, then we may
3835 * have some more checking to do. A case-folding file
3836 * system is either supporting mixed case sensitivity
3837 * access or is completely case-insensitive. Note
3838 * that the file system is always case preserving.
3840 * In mixed sensitivity mode case sensitive behavior
3841 * is the default. FIGNORECASE must be used to
3842 * explicitly request case insensitive behavior.
3844 * If the source and target names provided differ only
3845 * by case (e.g., a request to rename 'tim' to 'Tim'),
3846 * we will treat this as a special case in the
3847 * case-insensitive mode: as long as the source name
3848 * is an exact match, we will allow this to proceed as
3849 * a name-change request.
3851 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3852 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3853 flags & FIGNORECASE)) &&
3854 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3857 * case preserving rename request, require exact
3866 * If the source and destination directories are the same, we should
3867 * grab the z_name_lock of that directory only once.
3871 rw_enter(&sdzp->z_name_lock, RW_READER);
3875 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3876 ZEXISTS | zflg, NULL, NULL);
3877 terr = zfs_dirent_lock(&tdl,
3878 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3880 terr = zfs_dirent_lock(&tdl,
3881 tdzp, tnm, &tzp, zflg, NULL, NULL);
3882 serr = zfs_dirent_lock(&sdl,
3883 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3889 * Source entry invalid or not there.
3892 zfs_dirent_unlock(tdl);
3898 rw_exit(&sdzp->z_name_lock);
3901 * FreeBSD: In OpenSolaris they only check if rename source is
3902 * ".." here, because "." is handled in their lookup. This is
3903 * not the case for FreeBSD, so we check for "." explicitly.
3905 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0)
3906 serr = SET_ERROR(EINVAL);
3911 zfs_dirent_unlock(sdl);
3915 rw_exit(&sdzp->z_name_lock);
3917 if (strcmp(tnm, "..") == 0)
3918 terr = SET_ERROR(EINVAL);
3924 * Must have write access at the source to remove the old entry
3925 * and write access at the target to create the new entry.
3926 * Note that if target and source are the same, this can be
3927 * done in a single check.
3930 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3933 if (ZTOV(szp)->v_type == VDIR) {
3935 * Check to make sure rename is valid.
3936 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3938 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3943 * Does target exist?
3947 * Source and target must be the same type.
3949 if (ZTOV(szp)->v_type == VDIR) {
3950 if (ZTOV(tzp)->v_type != VDIR) {
3951 error = SET_ERROR(ENOTDIR);
3955 if (ZTOV(tzp)->v_type == VDIR) {
3956 error = SET_ERROR(EISDIR);
3961 * POSIX dictates that when the source and target
3962 * entries refer to the same file object, rename
3963 * must do nothing and exit without error.
3965 if (szp->z_id == tzp->z_id) {
3971 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3973 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3976 * notify the target directory if it is not the same
3977 * as source directory.
3980 vnevent_rename_dest_dir(tdvp, ct);
3983 tx = dmu_tx_create(zfsvfs->z_os);
3984 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3985 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3986 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3987 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3989 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3990 zfs_sa_upgrade_txholds(tx, tdzp);
3993 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3994 zfs_sa_upgrade_txholds(tx, tzp);
3997 zfs_sa_upgrade_txholds(tx, szp);
3998 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3999 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4002 zfs_rename_unlock(&zl);
4003 zfs_dirent_unlock(sdl);
4004 zfs_dirent_unlock(tdl);
4007 rw_exit(&sdzp->z_name_lock);
4012 if (error == ERESTART) {
4023 if (tzp) /* Attempt to remove the existing target */
4024 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
4027 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
4029 szp->z_pflags |= ZFS_AV_MODIFIED;
4031 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
4032 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
4035 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
4037 zfs_log_rename(zilog, tx, TX_RENAME |
4038 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
4039 sdl->dl_name, tdzp, tdl->dl_name, szp);
4042 * Update path information for the target vnode
4044 vn_renamepath(tdvp, ZTOV(szp), tnm,
4048 * At this point, we have successfully created
4049 * the target name, but have failed to remove
4050 * the source name. Since the create was done
4051 * with the ZRENAMING flag, there are
4052 * complications; for one, the link count is
4053 * wrong. The easiest way to deal with this
4054 * is to remove the newly created target, and
4055 * return the original error. This must
4056 * succeed; fortunately, it is very unlikely to
4057 * fail, since we just created it.
4059 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
4060 ZRENAMING, NULL), ==, 0);
4063 #ifdef FREEBSD_NAMECACHE
4067 cache_purge(ZTOV(szp));
4069 cache_purge(ZTOV(tzp));
4077 zfs_rename_unlock(&zl);
4079 zfs_dirent_unlock(sdl);
4080 zfs_dirent_unlock(tdl);
4083 rw_exit(&sdzp->z_name_lock);
4090 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4091 zil_commit(zilog, 0);
4099 * Insert the indicated symbolic reference entry into the directory.
4101 * IN: dvp - Directory to contain new symbolic link.
4102 * link - Name for new symlink entry.
4103 * vap - Attributes of new entry.
4104 * cr - credentials of caller.
4105 * ct - caller context
4106 * flags - case flags
4108 * RETURN: 0 on success, error code on failure.
4111 * dvp - ctime|mtime updated
4115 zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link,
4116 cred_t *cr, kthread_t *td)
4118 znode_t *zp, *dzp = VTOZ(dvp);
4121 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4123 uint64_t len = strlen(link);
4126 zfs_acl_ids_t acl_ids;
4127 boolean_t fuid_dirtied;
4128 uint64_t txtype = TX_SYMLINK;
4129 boolean_t waited = B_FALSE;
4132 ASSERT(vap->va_type == VLNK);
4136 zilog = zfsvfs->z_log;
4138 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
4139 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4141 return (SET_ERROR(EILSEQ));
4143 if (flags & FIGNORECASE)
4146 if (len > MAXPATHLEN) {
4148 return (SET_ERROR(ENAMETOOLONG));
4151 if ((error = zfs_acl_ids_create(dzp, 0,
4152 vap, cr, NULL, &acl_ids)) != 0) {
4157 getnewvnode_reserve(1);
4161 * Attempt to lock directory; fail if entry already exists.
4163 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
4165 zfs_acl_ids_free(&acl_ids);
4166 getnewvnode_drop_reserve();
4171 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4172 zfs_acl_ids_free(&acl_ids);
4173 zfs_dirent_unlock(dl);
4174 getnewvnode_drop_reserve();
4179 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
4180 zfs_acl_ids_free(&acl_ids);
4181 zfs_dirent_unlock(dl);
4182 getnewvnode_drop_reserve();
4184 return (SET_ERROR(EDQUOT));
4186 tx = dmu_tx_create(zfsvfs->z_os);
4187 fuid_dirtied = zfsvfs->z_fuid_dirty;
4188 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
4189 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4190 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
4191 ZFS_SA_BASE_ATTR_SIZE + len);
4192 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
4193 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
4194 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
4195 acl_ids.z_aclp->z_acl_bytes);
4198 zfs_fuid_txhold(zfsvfs, tx);
4199 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4201 zfs_dirent_unlock(dl);
4202 if (error == ERESTART) {
4208 zfs_acl_ids_free(&acl_ids);
4210 getnewvnode_drop_reserve();
4216 * Create a new object for the symlink.
4217 * for version 4 ZPL datsets the symlink will be an SA attribute
4219 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
4222 zfs_fuid_sync(zfsvfs, tx);
4224 mutex_enter(&zp->z_lock);
4226 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
4229 zfs_sa_symlink(zp, link, len, tx);
4230 mutex_exit(&zp->z_lock);
4233 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
4234 &zp->z_size, sizeof (zp->z_size), tx);
4236 * Insert the new object into the directory.
4238 (void) zfs_link_create(dl, zp, tx, ZNEW);
4240 if (flags & FIGNORECASE)
4242 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
4245 zfs_acl_ids_free(&acl_ids);
4249 getnewvnode_drop_reserve();
4251 zfs_dirent_unlock(dl);
4253 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4254 zil_commit(zilog, 0);
4261 * Return, in the buffer contained in the provided uio structure,
4262 * the symbolic path referred to by vp.
4264 * IN: vp - vnode of symbolic link.
4265 * uio - structure to contain the link path.
4266 * cr - credentials of caller.
4267 * ct - caller context
4269 * OUT: uio - structure containing the link path.
4271 * RETURN: 0 on success, error code on failure.
4274 * vp - atime updated
4278 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
4280 znode_t *zp = VTOZ(vp);
4281 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4287 mutex_enter(&zp->z_lock);
4289 error = sa_lookup_uio(zp->z_sa_hdl,
4290 SA_ZPL_SYMLINK(zfsvfs), uio);
4292 error = zfs_sa_readlink(zp, uio);
4293 mutex_exit(&zp->z_lock);
4295 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4302 * Insert a new entry into directory tdvp referencing svp.
4304 * IN: tdvp - Directory to contain new entry.
4305 * svp - vnode of new entry.
4306 * name - name of new entry.
4307 * cr - credentials of caller.
4308 * ct - caller context
4310 * RETURN: 0 on success, error code on failure.
4313 * tdvp - ctime|mtime updated
4314 * svp - ctime updated
4318 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
4319 caller_context_t *ct, int flags)
4321 znode_t *dzp = VTOZ(tdvp);
4323 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4332 boolean_t waited = B_FALSE;
4334 ASSERT(tdvp->v_type == VDIR);
4338 zilog = zfsvfs->z_log;
4340 if (VOP_REALVP(svp, &realvp, ct) == 0)
4344 * POSIX dictates that we return EPERM here.
4345 * Better choices include ENOTSUP or EISDIR.
4347 if (svp->v_type == VDIR) {
4349 return (SET_ERROR(EPERM));
4355 if (szp->z_pflags & (ZFS_APPENDONLY | ZFS_IMMUTABLE | ZFS_READONLY)) {
4357 return (SET_ERROR(EPERM));
4361 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
4362 * ctldir appear to have the same v_vfsp.
4364 if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) {
4366 return (SET_ERROR(EXDEV));
4369 /* Prevent links to .zfs/shares files */
4371 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4372 &parent, sizeof (uint64_t))) != 0) {
4376 if (parent == zfsvfs->z_shares_dir) {
4378 return (SET_ERROR(EPERM));
4381 if (zfsvfs->z_utf8 && u8_validate(name,
4382 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4384 return (SET_ERROR(EILSEQ));
4386 if (flags & FIGNORECASE)
4390 * We do not support links between attributes and non-attributes
4391 * because of the potential security risk of creating links
4392 * into "normal" file space in order to circumvent restrictions
4393 * imposed in attribute space.
4395 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4397 return (SET_ERROR(EINVAL));
4401 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4402 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) {
4404 return (SET_ERROR(EPERM));
4407 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4414 * Attempt to lock directory; fail if entry already exists.
4416 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4422 tx = dmu_tx_create(zfsvfs->z_os);
4423 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4424 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4425 zfs_sa_upgrade_txholds(tx, szp);
4426 zfs_sa_upgrade_txholds(tx, dzp);
4427 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4429 zfs_dirent_unlock(dl);
4430 if (error == ERESTART) {
4441 error = zfs_link_create(dl, szp, tx, 0);
4444 uint64_t txtype = TX_LINK;
4445 if (flags & FIGNORECASE)
4447 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4452 zfs_dirent_unlock(dl);
4455 vnevent_link(svp, ct);
4458 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4459 zil_commit(zilog, 0);
4467 * zfs_null_putapage() is used when the file system has been force
4468 * unmounted. It just drops the pages.
4472 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4473 size_t *lenp, int flags, cred_t *cr)
4475 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4480 * Push a page out to disk, klustering if possible.
4482 * IN: vp - file to push page to.
4483 * pp - page to push.
4484 * flags - additional flags.
4485 * cr - credentials of caller.
4487 * OUT: offp - start of range pushed.
4488 * lenp - len of range pushed.
4490 * RETURN: 0 on success, error code on failure.
4492 * NOTE: callers must have locked the page to be pushed. On
4493 * exit, the page (and all other pages in the kluster) must be
4498 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4499 size_t *lenp, int flags, cred_t *cr)
4501 znode_t *zp = VTOZ(vp);
4502 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4504 u_offset_t off, koff;
4511 * If our blocksize is bigger than the page size, try to kluster
4512 * multiple pages so that we write a full block (thus avoiding
4513 * a read-modify-write).
4515 if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4516 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4517 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4518 ASSERT(koff <= zp->z_size);
4519 if (koff + klen > zp->z_size)
4520 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4521 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4523 ASSERT3U(btop(len), ==, btopr(len));
4526 * Can't push pages past end-of-file.
4528 if (off >= zp->z_size) {
4529 /* ignore all pages */
4532 } else if (off + len > zp->z_size) {
4533 int npages = btopr(zp->z_size - off);
4536 page_list_break(&pp, &trunc, npages);
4537 /* ignore pages past end of file */
4539 pvn_write_done(trunc, flags);
4540 len = zp->z_size - off;
4543 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4544 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4545 err = SET_ERROR(EDQUOT);
4548 tx = dmu_tx_create(zfsvfs->z_os);
4549 dmu_tx_hold_write(tx, zp->z_id, off, len);
4551 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4552 zfs_sa_upgrade_txholds(tx, zp);
4553 err = dmu_tx_assign(tx, TXG_WAIT);
4559 if (zp->z_blksz <= PAGESIZE) {
4560 caddr_t va = zfs_map_page(pp, S_READ);
4561 ASSERT3U(len, <=, PAGESIZE);
4562 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4563 zfs_unmap_page(pp, va);
4565 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4569 uint64_t mtime[2], ctime[2];
4570 sa_bulk_attr_t bulk[3];
4573 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4575 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4577 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4579 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4581 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4586 pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4596 * Copy the portion of the file indicated from pages into the file.
4597 * The pages are stored in a page list attached to the files vnode.
4599 * IN: vp - vnode of file to push page data to.
4600 * off - position in file to put data.
4601 * len - amount of data to write.
4602 * flags - flags to control the operation.
4603 * cr - credentials of caller.
4604 * ct - caller context.
4606 * RETURN: 0 on success, error code on failure.
4609 * vp - ctime|mtime updated
4613 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4614 caller_context_t *ct)
4616 znode_t *zp = VTOZ(vp);
4617 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4629 * Align this request to the file block size in case we kluster.
4630 * XXX - this can result in pretty aggresive locking, which can
4631 * impact simultanious read/write access. One option might be
4632 * to break up long requests (len == 0) into block-by-block
4633 * operations to get narrower locking.
4635 blksz = zp->z_blksz;
4637 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4640 if (len > 0 && ISP2(blksz))
4641 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4647 * Search the entire vp list for pages >= io_off.
4649 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4650 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4653 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4655 if (off > zp->z_size) {
4656 /* past end of file */
4657 zfs_range_unlock(rl);
4662 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4664 for (off = io_off; io_off < off + len; io_off += io_len) {
4665 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4666 pp = page_lookup(vp, io_off,
4667 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4669 pp = page_lookup_nowait(vp, io_off,
4670 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4673 if (pp != NULL && pvn_getdirty(pp, flags)) {
4677 * Found a dirty page to push
4679 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4687 zfs_range_unlock(rl);
4688 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4689 zil_commit(zfsvfs->z_log, zp->z_id);
4693 #endif /* illumos */
4697 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4699 znode_t *zp = VTOZ(vp);
4700 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4703 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4704 if (zp->z_sa_hdl == NULL) {
4706 * The fs has been unmounted, or we did a
4707 * suspend/resume and this file no longer exists.
4709 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4714 mutex_enter(&zp->z_lock);
4715 if (zp->z_unlinked) {
4717 * Fast path to recycle a vnode of a removed file.
4719 mutex_exit(&zp->z_lock);
4720 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4724 mutex_exit(&zp->z_lock);
4726 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4727 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4729 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4730 zfs_sa_upgrade_txholds(tx, zp);
4731 error = dmu_tx_assign(tx, TXG_WAIT);
4735 mutex_enter(&zp->z_lock);
4736 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4737 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4738 zp->z_atime_dirty = 0;
4739 mutex_exit(&zp->z_lock);
4743 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4748 * Bounds-check the seek operation.
4750 * IN: vp - vnode seeking within
4751 * ooff - old file offset
4752 * noffp - pointer to new file offset
4753 * ct - caller context
4755 * RETURN: 0 on success, EINVAL if new offset invalid.
4759 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4760 caller_context_t *ct)
4762 if (vp->v_type == VDIR)
4764 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4768 * Pre-filter the generic locking function to trap attempts to place
4769 * a mandatory lock on a memory mapped file.
4772 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4773 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4775 znode_t *zp = VTOZ(vp);
4776 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4782 * We are following the UFS semantics with respect to mapcnt
4783 * here: If we see that the file is mapped already, then we will
4784 * return an error, but we don't worry about races between this
4785 * function and zfs_map().
4787 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4789 return (SET_ERROR(EAGAIN));
4792 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4796 * If we can't find a page in the cache, we will create a new page
4797 * and fill it with file data. For efficiency, we may try to fill
4798 * multiple pages at once (klustering) to fill up the supplied page
4799 * list. Note that the pages to be filled are held with an exclusive
4800 * lock to prevent access by other threads while they are being filled.
4803 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4804 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4806 znode_t *zp = VTOZ(vp);
4807 page_t *pp, *cur_pp;
4808 objset_t *os = zp->z_zfsvfs->z_os;
4809 u_offset_t io_off, total;
4813 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4815 * We only have a single page, don't bother klustering
4819 pp = page_create_va(vp, io_off, io_len,
4820 PG_EXCL | PG_WAIT, seg, addr);
4823 * Try to find enough pages to fill the page list
4825 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4826 &io_len, off, plsz, 0);
4830 * The page already exists, nothing to do here.
4837 * Fill the pages in the kluster.
4840 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4843 ASSERT3U(io_off, ==, cur_pp->p_offset);
4844 va = zfs_map_page(cur_pp, S_WRITE);
4845 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4847 zfs_unmap_page(cur_pp, va);
4849 /* On error, toss the entire kluster */
4850 pvn_read_done(pp, B_ERROR);
4851 /* convert checksum errors into IO errors */
4853 err = SET_ERROR(EIO);
4856 cur_pp = cur_pp->p_next;
4860 * Fill in the page list array from the kluster starting
4861 * from the desired offset `off'.
4862 * NOTE: the page list will always be null terminated.
4864 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4865 ASSERT(pl == NULL || (*pl)->p_offset == off);
4871 * Return pointers to the pages for the file region [off, off + len]
4872 * in the pl array. If plsz is greater than len, this function may
4873 * also return page pointers from after the specified region
4874 * (i.e. the region [off, off + plsz]). These additional pages are
4875 * only returned if they are already in the cache, or were created as
4876 * part of a klustered read.
4878 * IN: vp - vnode of file to get data from.
4879 * off - position in file to get data from.
4880 * len - amount of data to retrieve.
4881 * plsz - length of provided page list.
4882 * seg - segment to obtain pages for.
4883 * addr - virtual address of fault.
4884 * rw - mode of created pages.
4885 * cr - credentials of caller.
4886 * ct - caller context.
4888 * OUT: protp - protection mode of created pages.
4889 * pl - list of pages created.
4891 * RETURN: 0 on success, error code on failure.
4894 * vp - atime updated
4898 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4899 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4900 enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4902 znode_t *zp = VTOZ(vp);
4903 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4907 /* we do our own caching, faultahead is unnecessary */
4910 else if (len > plsz)
4913 len = P2ROUNDUP(len, PAGESIZE);
4914 ASSERT(plsz >= len);
4923 * Loop through the requested range [off, off + len) looking
4924 * for pages. If we don't find a page, we will need to create
4925 * a new page and fill it with data from the file.
4928 if (*pl = page_lookup(vp, off, SE_SHARED))
4930 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4933 ASSERT3U((*pl)->p_offset, ==, off);
4937 ASSERT3U(len, >=, PAGESIZE);
4940 ASSERT3U(plsz, >=, PAGESIZE);
4947 * Fill out the page array with any pages already in the cache.
4950 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4957 * Release any pages we have previously locked.
4962 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4972 * Request a memory map for a section of a file. This code interacts
4973 * with common code and the VM system as follows:
4975 * - common code calls mmap(), which ends up in smmap_common()
4976 * - this calls VOP_MAP(), which takes you into (say) zfs
4977 * - zfs_map() calls as_map(), passing segvn_create() as the callback
4978 * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4979 * - zfs_addmap() updates z_mapcnt
4983 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4984 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4985 caller_context_t *ct)
4987 znode_t *zp = VTOZ(vp);
4988 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4989 segvn_crargs_t vn_a;
4995 if ((prot & PROT_WRITE) && (zp->z_pflags &
4996 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4998 return (SET_ERROR(EPERM));
5001 if ((prot & (PROT_READ | PROT_EXEC)) &&
5002 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
5004 return (SET_ERROR(EACCES));
5007 if (vp->v_flag & VNOMAP) {
5009 return (SET_ERROR(ENOSYS));
5012 if (off < 0 || len > MAXOFFSET_T - off) {
5014 return (SET_ERROR(ENXIO));
5017 if (vp->v_type != VREG) {
5019 return (SET_ERROR(ENODEV));
5023 * If file is locked, disallow mapping.
5025 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
5027 return (SET_ERROR(EAGAIN));
5031 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
5039 vn_a.offset = (u_offset_t)off;
5040 vn_a.type = flags & MAP_TYPE;
5042 vn_a.maxprot = maxprot;
5045 vn_a.flags = flags & ~MAP_TYPE;
5047 vn_a.lgrp_mem_policy_flags = 0;
5049 error = as_map(as, *addrp, len, segvn_create, &vn_a);
5058 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5059 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
5060 caller_context_t *ct)
5062 uint64_t pages = btopr(len);
5064 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
5069 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
5070 * more accurate mtime for the associated file. Since we don't have a way of
5071 * detecting when the data was actually modified, we have to resort to
5072 * heuristics. If an explicit msync() is done, then we mark the mtime when the
5073 * last page is pushed. The problem occurs when the msync() call is omitted,
5074 * which by far the most common case:
5082 * putpage() via fsflush
5084 * If we wait until fsflush to come along, we can have a modification time that
5085 * is some arbitrary point in the future. In order to prevent this in the
5086 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
5091 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5092 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
5093 caller_context_t *ct)
5095 uint64_t pages = btopr(len);
5097 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
5098 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
5100 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
5101 vn_has_cached_data(vp))
5102 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
5108 * Free or allocate space in a file. Currently, this function only
5109 * supports the `F_FREESP' command. However, this command is somewhat
5110 * misnamed, as its functionality includes the ability to allocate as
5111 * well as free space.
5113 * IN: vp - vnode of file to free data in.
5114 * cmd - action to take (only F_FREESP supported).
5115 * bfp - section of file to free/alloc.
5116 * flag - current file open mode flags.
5117 * offset - current file offset.
5118 * cr - credentials of caller [UNUSED].
5119 * ct - caller context.
5121 * RETURN: 0 on success, error code on failure.
5124 * vp - ctime|mtime updated
5128 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
5129 offset_t offset, cred_t *cr, caller_context_t *ct)
5131 znode_t *zp = VTOZ(vp);
5132 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5139 if (cmd != F_FREESP) {
5141 return (SET_ERROR(EINVAL));
5145 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
5146 * callers might not be able to detect properly that we are read-only,
5147 * so check it explicitly here.
5149 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
5151 return (SET_ERROR(EROFS));
5154 if (error = convoff(vp, bfp, 0, offset)) {
5159 if (bfp->l_len < 0) {
5161 return (SET_ERROR(EINVAL));
5165 len = bfp->l_len; /* 0 means from off to end of file */
5167 error = zfs_freesp(zp, off, len, flag, TRUE);
5172 #endif /* illumos */
5174 CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid));
5175 CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid));
5179 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
5181 znode_t *zp = VTOZ(vp);
5182 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5185 uint64_t object = zp->z_id;
5192 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
5193 &gen64, sizeof (uint64_t))) != 0) {
5198 gen = (uint32_t)gen64;
5200 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
5203 if (fidp->fid_len < size) {
5204 fidp->fid_len = size;
5206 return (SET_ERROR(ENOSPC));
5209 fidp->fid_len = size;
5212 zfid = (zfid_short_t *)fidp;
5214 zfid->zf_len = size;
5216 for (i = 0; i < sizeof (zfid->zf_object); i++)
5217 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
5219 /* Must have a non-zero generation number to distinguish from .zfs */
5222 for (i = 0; i < sizeof (zfid->zf_gen); i++)
5223 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
5225 if (size == LONG_FID_LEN) {
5226 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
5229 zlfid = (zfid_long_t *)fidp;
5231 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
5232 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
5234 /* XXX - this should be the generation number for the objset */
5235 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
5236 zlfid->zf_setgen[i] = 0;
5244 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
5245 caller_context_t *ct)
5257 case _PC_FILESIZEBITS:
5261 case _PC_XATTR_EXISTS:
5263 zfsvfs = zp->z_zfsvfs;
5267 error = zfs_dirent_lock(&dl, zp, "", &xzp,
5268 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
5270 zfs_dirent_unlock(dl);
5271 if (!zfs_dirempty(xzp))
5274 } else if (error == ENOENT) {
5276 * If there aren't extended attributes, it's the
5277 * same as having zero of them.
5284 case _PC_SATTR_ENABLED:
5285 case _PC_SATTR_EXISTS:
5286 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
5287 (vp->v_type == VREG || vp->v_type == VDIR);
5290 case _PC_ACCESS_FILTERING:
5291 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
5295 case _PC_ACL_ENABLED:
5296 *valp = _ACL_ACE_ENABLED;
5298 #endif /* illumos */
5299 case _PC_MIN_HOLE_SIZE:
5300 *valp = (int)SPA_MINBLOCKSIZE;
5303 case _PC_TIMESTAMP_RESOLUTION:
5304 /* nanosecond timestamp resolution */
5308 case _PC_ACL_EXTENDED:
5316 case _PC_ACL_PATH_MAX:
5317 *valp = ACL_MAX_ENTRIES;
5321 return (EOPNOTSUPP);
5327 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5328 caller_context_t *ct)
5330 znode_t *zp = VTOZ(vp);
5331 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5333 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5337 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5345 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5346 caller_context_t *ct)
5348 znode_t *zp = VTOZ(vp);
5349 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5351 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5352 zilog_t *zilog = zfsvfs->z_log;
5357 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5359 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5360 zil_commit(zilog, 0);
5368 * The smallest read we may consider to loan out an arcbuf.
5369 * This must be a power of 2.
5371 int zcr_blksz_min = (1 << 10); /* 1K */
5373 * If set to less than the file block size, allow loaning out of an
5374 * arcbuf for a partial block read. This must be a power of 2.
5376 int zcr_blksz_max = (1 << 17); /* 128K */
5380 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5381 caller_context_t *ct)
5383 znode_t *zp = VTOZ(vp);
5384 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5385 int max_blksz = zfsvfs->z_max_blksz;
5386 uio_t *uio = &xuio->xu_uio;
5387 ssize_t size = uio->uio_resid;
5388 offset_t offset = uio->uio_loffset;
5393 int preamble, postamble;
5395 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5396 return (SET_ERROR(EINVAL));
5403 * Loan out an arc_buf for write if write size is bigger than
5404 * max_blksz, and the file's block size is also max_blksz.
5407 if (size < blksz || zp->z_blksz != blksz) {
5409 return (SET_ERROR(EINVAL));
5412 * Caller requests buffers for write before knowing where the
5413 * write offset might be (e.g. NFS TCP write).
5418 preamble = P2PHASE(offset, blksz);
5420 preamble = blksz - preamble;
5425 postamble = P2PHASE(size, blksz);
5428 fullblk = size / blksz;
5429 (void) dmu_xuio_init(xuio,
5430 (preamble != 0) + fullblk + (postamble != 0));
5431 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5432 int, postamble, int,
5433 (preamble != 0) + fullblk + (postamble != 0));
5436 * Have to fix iov base/len for partial buffers. They
5437 * currently represent full arc_buf's.
5440 /* data begins in the middle of the arc_buf */
5441 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5444 (void) dmu_xuio_add(xuio, abuf,
5445 blksz - preamble, preamble);
5448 for (i = 0; i < fullblk; i++) {
5449 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5452 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5456 /* data ends in the middle of the arc_buf */
5457 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5460 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5465 * Loan out an arc_buf for read if the read size is larger than
5466 * the current file block size. Block alignment is not
5467 * considered. Partial arc_buf will be loaned out for read.
5469 blksz = zp->z_blksz;
5470 if (blksz < zcr_blksz_min)
5471 blksz = zcr_blksz_min;
5472 if (blksz > zcr_blksz_max)
5473 blksz = zcr_blksz_max;
5474 /* avoid potential complexity of dealing with it */
5475 if (blksz > max_blksz) {
5477 return (SET_ERROR(EINVAL));
5480 maxsize = zp->z_size - uio->uio_loffset;
5484 if (size < blksz || vn_has_cached_data(vp)) {
5486 return (SET_ERROR(EINVAL));
5491 return (SET_ERROR(EINVAL));
5494 uio->uio_extflg = UIO_XUIO;
5495 XUIO_XUZC_RW(xuio) = ioflag;
5502 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5506 int ioflag = XUIO_XUZC_RW(xuio);
5508 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5510 i = dmu_xuio_cnt(xuio);
5512 abuf = dmu_xuio_arcbuf(xuio, i);
5514 * if abuf == NULL, it must be a write buffer
5515 * that has been returned in zfs_write().
5518 dmu_return_arcbuf(abuf);
5519 ASSERT(abuf || ioflag == UIO_WRITE);
5522 dmu_xuio_fini(xuio);
5527 * Predeclare these here so that the compiler assumes that
5528 * this is an "old style" function declaration that does
5529 * not include arguments => we won't get type mismatch errors
5530 * in the initializations that follow.
5532 static int zfs_inval();
5533 static int zfs_isdir();
5538 return (SET_ERROR(EINVAL));
5544 return (SET_ERROR(EISDIR));
5547 * Directory vnode operations template
5549 vnodeops_t *zfs_dvnodeops;
5550 const fs_operation_def_t zfs_dvnodeops_template[] = {
5551 VOPNAME_OPEN, { .vop_open = zfs_open },
5552 VOPNAME_CLOSE, { .vop_close = zfs_close },
5553 VOPNAME_READ, { .error = zfs_isdir },
5554 VOPNAME_WRITE, { .error = zfs_isdir },
5555 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5556 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5557 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5558 VOPNAME_ACCESS, { .vop_access = zfs_access },
5559 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5560 VOPNAME_CREATE, { .vop_create = zfs_create },
5561 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5562 VOPNAME_LINK, { .vop_link = zfs_link },
5563 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5564 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir },
5565 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5566 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5567 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink },
5568 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5569 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5570 VOPNAME_FID, { .vop_fid = zfs_fid },
5571 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5572 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5573 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5574 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5575 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5580 * Regular file vnode operations template
5582 vnodeops_t *zfs_fvnodeops;
5583 const fs_operation_def_t zfs_fvnodeops_template[] = {
5584 VOPNAME_OPEN, { .vop_open = zfs_open },
5585 VOPNAME_CLOSE, { .vop_close = zfs_close },
5586 VOPNAME_READ, { .vop_read = zfs_read },
5587 VOPNAME_WRITE, { .vop_write = zfs_write },
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_RENAME, { .vop_rename = zfs_rename },
5594 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5595 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5596 VOPNAME_FID, { .vop_fid = zfs_fid },
5597 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5598 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock },
5599 VOPNAME_SPACE, { .vop_space = zfs_space },
5600 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage },
5601 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage },
5602 VOPNAME_MAP, { .vop_map = zfs_map },
5603 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap },
5604 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap },
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 },
5609 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf },
5610 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf },
5615 * Symbolic link vnode operations template
5617 vnodeops_t *zfs_symvnodeops;
5618 const fs_operation_def_t zfs_symvnodeops_template[] = {
5619 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5620 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5621 VOPNAME_ACCESS, { .vop_access = zfs_access },
5622 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5623 VOPNAME_READLINK, { .vop_readlink = zfs_readlink },
5624 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5625 VOPNAME_FID, { .vop_fid = zfs_fid },
5626 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5627 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5632 * special share hidden files vnode operations template
5634 vnodeops_t *zfs_sharevnodeops;
5635 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5636 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5637 VOPNAME_ACCESS, { .vop_access = zfs_access },
5638 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5639 VOPNAME_FID, { .vop_fid = zfs_fid },
5640 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5641 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5642 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5643 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5648 * Extended attribute directory vnode operations template
5650 * This template is identical to the directory vnodes
5651 * operation template except for restricted operations:
5655 * Note that there are other restrictions embedded in:
5656 * zfs_create() - restrict type to VREG
5657 * zfs_link() - no links into/out of attribute space
5658 * zfs_rename() - no moves into/out of attribute space
5660 vnodeops_t *zfs_xdvnodeops;
5661 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5662 VOPNAME_OPEN, { .vop_open = zfs_open },
5663 VOPNAME_CLOSE, { .vop_close = zfs_close },
5664 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5665 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5666 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5667 VOPNAME_ACCESS, { .vop_access = zfs_access },
5668 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5669 VOPNAME_CREATE, { .vop_create = zfs_create },
5670 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5671 VOPNAME_LINK, { .vop_link = zfs_link },
5672 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5673 VOPNAME_MKDIR, { .error = zfs_inval },
5674 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5675 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5676 VOPNAME_SYMLINK, { .error = zfs_inval },
5677 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5678 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5679 VOPNAME_FID, { .vop_fid = zfs_fid },
5680 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5681 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5682 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5683 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5684 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5689 * Error vnode operations template
5691 vnodeops_t *zfs_evnodeops;
5692 const fs_operation_def_t zfs_evnodeops_template[] = {
5693 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5694 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5697 #endif /* illumos */
5700 ioflags(int ioflags)
5704 if (ioflags & IO_APPEND)
5706 if (ioflags & IO_NDELAY)
5708 if (ioflags & IO_SYNC)
5709 flags |= (FSYNC | FDSYNC | FRSYNC);
5715 zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage)
5717 znode_t *zp = VTOZ(vp);
5718 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5719 objset_t *os = zp->z_zfsvfs->z_os;
5720 vm_page_t mfirst, mlast, mreq;
5724 off_t startoff, endoff;
5726 vm_pindex_t reqstart, reqend;
5727 int pcount, lsize, reqsize, size;
5732 pcount = OFF_TO_IDX(round_page(count));
5734 object = mreq->object;
5737 KASSERT(vp->v_object == object, ("mismatching object"));
5739 if (pcount > 1 && zp->z_blksz > PAGESIZE) {
5740 startoff = rounddown(IDX_TO_OFF(mreq->pindex), zp->z_blksz);
5741 reqstart = OFF_TO_IDX(round_page(startoff));
5742 if (reqstart < m[0]->pindex)
5745 reqstart = reqstart - m[0]->pindex;
5746 endoff = roundup(IDX_TO_OFF(mreq->pindex) + PAGE_SIZE,
5748 reqend = OFF_TO_IDX(trunc_page(endoff)) - 1;
5749 if (reqend > m[pcount - 1]->pindex)
5750 reqend = m[pcount - 1]->pindex;
5751 reqsize = reqend - m[reqstart]->pindex + 1;
5752 KASSERT(reqstart <= reqpage && reqpage < reqstart + reqsize,
5753 ("reqpage beyond [reqstart, reqstart + reqsize[ bounds"));
5758 mfirst = m[reqstart];
5759 mlast = m[reqstart + reqsize - 1];
5761 zfs_vmobject_wlock(object);
5763 for (i = 0; i < reqstart; i++) {
5766 vm_page_unlock(m[i]);
5768 for (i = reqstart + reqsize; i < pcount; i++) {
5771 vm_page_unlock(m[i]);
5774 if (mreq->valid && reqsize == 1) {
5775 if (mreq->valid != VM_PAGE_BITS_ALL)
5776 vm_page_zero_invalid(mreq, TRUE);
5777 zfs_vmobject_wunlock(object);
5779 return (zfs_vm_pagerret_ok);
5782 PCPU_INC(cnt.v_vnodein);
5783 PCPU_ADD(cnt.v_vnodepgsin, reqsize);
5785 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) {
5786 for (i = reqstart; i < reqstart + reqsize; i++) {
5790 vm_page_unlock(m[i]);
5793 zfs_vmobject_wunlock(object);
5795 return (zfs_vm_pagerret_bad);
5799 if (IDX_TO_OFF(mlast->pindex) + lsize > object->un_pager.vnp.vnp_size)
5800 lsize = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mlast->pindex);
5802 zfs_vmobject_wunlock(object);
5804 for (i = reqstart; i < reqstart + reqsize; i++) {
5806 if (i == (reqstart + reqsize - 1))
5808 va = zfs_map_page(m[i], &sf);
5809 error = dmu_read(os, zp->z_id, IDX_TO_OFF(m[i]->pindex),
5810 size, va, DMU_READ_PREFETCH);
5811 if (size != PAGE_SIZE)
5812 bzero(va + size, PAGE_SIZE - size);
5818 zfs_vmobject_wlock(object);
5820 for (i = reqstart; i < reqstart + reqsize; i++) {
5822 m[i]->valid = VM_PAGE_BITS_ALL;
5823 KASSERT(m[i]->dirty == 0, ("zfs_getpages: page %p is dirty", m[i]));
5825 vm_page_readahead_finish(m[i]);
5828 zfs_vmobject_wunlock(object);
5830 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
5832 return (error ? zfs_vm_pagerret_error : zfs_vm_pagerret_ok);
5836 zfs_freebsd_getpages(ap)
5837 struct vop_getpages_args /* {
5845 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage));
5849 zfs_putpages(struct vnode *vp, vm_page_t *ma, size_t len, int flags,
5852 znode_t *zp = VTOZ(vp);
5853 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5862 vm_ooffset_t lo_off;
5873 object = vp->v_object;
5877 KASSERT(ma[0]->object == object, ("mismatching object"));
5878 KASSERT(len > 0 && (len & PAGE_MASK) == 0, ("unexpected length"));
5880 for (i = 0; i < pcount; i++)
5881 rtvals[i] = zfs_vm_pagerret_error;
5883 off = IDX_TO_OFF(ma[0]->pindex);
5884 blksz = zp->z_blksz;
5885 lo_off = rounddown(off, blksz);
5886 lo_len = roundup(len + (off - lo_off), blksz);
5887 rl = zfs_range_lock(zp, lo_off, lo_len, RL_WRITER);
5889 zfs_vmobject_wlock(object);
5890 if (len + off > object->un_pager.vnp.vnp_size) {
5891 if (object->un_pager.vnp.vnp_size > off) {
5894 len = object->un_pager.vnp.vnp_size - off;
5896 if ((pgoff = (int)len & PAGE_MASK) != 0) {
5898 * If the object is locked and the following
5899 * conditions hold, then the page's dirty
5900 * field cannot be concurrently changed by a
5904 vm_page_assert_sbusied(m);
5905 KASSERT(!pmap_page_is_write_mapped(m),
5906 ("zfs_putpages: page %p is not read-only", m));
5907 vm_page_clear_dirty(m, pgoff, PAGE_SIZE -
5914 if (ncount < pcount) {
5915 for (i = ncount; i < pcount; i++) {
5916 rtvals[i] = zfs_vm_pagerret_bad;
5920 zfs_vmobject_wunlock(object);
5925 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
5926 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
5931 tx = dmu_tx_create(zfsvfs->z_os);
5932 dmu_tx_hold_write(tx, zp->z_id, off, len);
5934 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
5935 zfs_sa_upgrade_txholds(tx, zp);
5936 err = dmu_tx_assign(tx, TXG_NOWAIT);
5938 if (err == ERESTART) {
5947 if (zp->z_blksz < PAGE_SIZE) {
5949 for (i = 0; len > 0; off += tocopy, len -= tocopy, i++) {
5950 tocopy = len > PAGE_SIZE ? PAGE_SIZE : len;
5951 va = zfs_map_page(ma[i], &sf);
5952 dmu_write(zfsvfs->z_os, zp->z_id, off, tocopy, va, tx);
5956 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, ma, tx);
5960 uint64_t mtime[2], ctime[2];
5961 sa_bulk_attr_t bulk[3];
5964 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
5966 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
5968 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
5970 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
5972 (void)sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
5973 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
5975 zfs_vmobject_wlock(object);
5976 for (i = 0; i < ncount; i++) {
5977 rtvals[i] = zfs_vm_pagerret_ok;
5978 vm_page_undirty(ma[i]);
5980 zfs_vmobject_wunlock(object);
5981 PCPU_INC(cnt.v_vnodeout);
5982 PCPU_ADD(cnt.v_vnodepgsout, ncount);
5987 zfs_range_unlock(rl);
5988 if ((flags & (zfs_vm_pagerput_sync | zfs_vm_pagerput_inval)) != 0 ||
5989 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5990 zil_commit(zfsvfs->z_log, zp->z_id);
5996 zfs_freebsd_putpages(ap)
5997 struct vop_putpages_args /* {
6006 return (zfs_putpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_sync,
6011 zfs_freebsd_bmap(ap)
6012 struct vop_bmap_args /* {
6015 struct bufobj **a_bop;
6022 if (ap->a_bop != NULL)
6023 *ap->a_bop = &ap->a_vp->v_bufobj;
6024 if (ap->a_bnp != NULL)
6025 *ap->a_bnp = ap->a_bn;
6026 if (ap->a_runp != NULL)
6028 if (ap->a_runb != NULL)
6035 zfs_freebsd_open(ap)
6036 struct vop_open_args /* {
6039 struct ucred *a_cred;
6040 struct thread *a_td;
6043 vnode_t *vp = ap->a_vp;
6044 znode_t *zp = VTOZ(vp);
6047 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL);
6049 vnode_create_vobject(vp, zp->z_size, ap->a_td);
6054 zfs_freebsd_close(ap)
6055 struct vop_close_args /* {
6058 struct ucred *a_cred;
6059 struct thread *a_td;
6063 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL));
6067 zfs_freebsd_ioctl(ap)
6068 struct vop_ioctl_args /* {
6078 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data,
6079 ap->a_fflag, ap->a_cred, NULL, NULL));
6083 zfs_freebsd_read(ap)
6084 struct vop_read_args /* {
6088 struct ucred *a_cred;
6092 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
6097 zfs_freebsd_write(ap)
6098 struct vop_write_args /* {
6102 struct ucred *a_cred;
6106 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
6111 zfs_freebsd_access(ap)
6112 struct vop_access_args /* {
6114 accmode_t a_accmode;
6115 struct ucred *a_cred;
6116 struct thread *a_td;
6119 vnode_t *vp = ap->a_vp;
6120 znode_t *zp = VTOZ(vp);
6125 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND,
6127 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND);
6129 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL);
6132 * VADMIN has to be handled by vaccess().
6135 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND);
6137 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
6138 zp->z_gid, accmode, ap->a_cred, NULL);
6143 * For VEXEC, ensure that at least one execute bit is set for
6146 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR &&
6147 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
6155 zfs_freebsd_lookup(ap)
6156 struct vop_lookup_args /* {
6157 struct vnode *a_dvp;
6158 struct vnode **a_vpp;
6159 struct componentname *a_cnp;
6162 struct componentname *cnp = ap->a_cnp;
6163 char nm[NAME_MAX + 1];
6165 ASSERT(cnp->cn_namelen < sizeof(nm));
6166 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm)));
6168 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop,
6169 cnp->cn_cred, cnp->cn_thread, 0));
6173 zfs_freebsd_create(ap)
6174 struct vop_create_args /* {
6175 struct vnode *a_dvp;
6176 struct vnode **a_vpp;
6177 struct componentname *a_cnp;
6178 struct vattr *a_vap;
6181 struct componentname *cnp = ap->a_cnp;
6182 vattr_t *vap = ap->a_vap;
6185 ASSERT(cnp->cn_flags & SAVENAME);
6187 vattr_init_mask(vap);
6188 mode = vap->va_mode & ALLPERMS;
6190 error = zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode,
6191 ap->a_vpp, cnp->cn_cred, cnp->cn_thread);
6192 #ifdef FREEBSD_NAMECACHE
6193 if (error == 0 && (cnp->cn_flags & MAKEENTRY) != 0)
6194 cache_enter(ap->a_dvp, *ap->a_vpp, cnp);
6200 zfs_freebsd_remove(ap)
6201 struct vop_remove_args /* {
6202 struct vnode *a_dvp;
6204 struct componentname *a_cnp;
6208 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6210 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr,
6211 ap->a_cnp->cn_cred, NULL, 0));
6215 zfs_freebsd_mkdir(ap)
6216 struct vop_mkdir_args /* {
6217 struct vnode *a_dvp;
6218 struct vnode **a_vpp;
6219 struct componentname *a_cnp;
6220 struct vattr *a_vap;
6223 vattr_t *vap = ap->a_vap;
6225 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6227 vattr_init_mask(vap);
6229 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp,
6230 ap->a_cnp->cn_cred, NULL, 0, NULL));
6234 zfs_freebsd_rmdir(ap)
6235 struct vop_rmdir_args /* {
6236 struct vnode *a_dvp;
6238 struct componentname *a_cnp;
6241 struct componentname *cnp = ap->a_cnp;
6243 ASSERT(cnp->cn_flags & SAVENAME);
6245 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0));
6249 zfs_freebsd_readdir(ap)
6250 struct vop_readdir_args /* {
6253 struct ucred *a_cred;
6260 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag,
6261 ap->a_ncookies, ap->a_cookies));
6265 zfs_freebsd_fsync(ap)
6266 struct vop_fsync_args /* {
6269 struct thread *a_td;
6274 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL));
6278 zfs_freebsd_getattr(ap)
6279 struct vop_getattr_args /* {
6281 struct vattr *a_vap;
6282 struct ucred *a_cred;
6285 vattr_t *vap = ap->a_vap;
6291 xvap.xva_vattr = *vap;
6292 xvap.xva_vattr.va_mask |= AT_XVATTR;
6294 /* Convert chflags into ZFS-type flags. */
6295 /* XXX: what about SF_SETTABLE?. */
6296 XVA_SET_REQ(&xvap, XAT_IMMUTABLE);
6297 XVA_SET_REQ(&xvap, XAT_APPENDONLY);
6298 XVA_SET_REQ(&xvap, XAT_NOUNLINK);
6299 XVA_SET_REQ(&xvap, XAT_NODUMP);
6300 XVA_SET_REQ(&xvap, XAT_READONLY);
6301 XVA_SET_REQ(&xvap, XAT_ARCHIVE);
6302 XVA_SET_REQ(&xvap, XAT_SYSTEM);
6303 XVA_SET_REQ(&xvap, XAT_HIDDEN);
6304 XVA_SET_REQ(&xvap, XAT_REPARSE);
6305 XVA_SET_REQ(&xvap, XAT_OFFLINE);
6306 XVA_SET_REQ(&xvap, XAT_SPARSE);
6308 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL);
6312 /* Convert ZFS xattr into chflags. */
6313 #define FLAG_CHECK(fflag, xflag, xfield) do { \
6314 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \
6315 fflags |= (fflag); \
6317 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE,
6318 xvap.xva_xoptattrs.xoa_immutable);
6319 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY,
6320 xvap.xva_xoptattrs.xoa_appendonly);
6321 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK,
6322 xvap.xva_xoptattrs.xoa_nounlink);
6323 FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE,
6324 xvap.xva_xoptattrs.xoa_archive);
6325 FLAG_CHECK(UF_NODUMP, XAT_NODUMP,
6326 xvap.xva_xoptattrs.xoa_nodump);
6327 FLAG_CHECK(UF_READONLY, XAT_READONLY,
6328 xvap.xva_xoptattrs.xoa_readonly);
6329 FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM,
6330 xvap.xva_xoptattrs.xoa_system);
6331 FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN,
6332 xvap.xva_xoptattrs.xoa_hidden);
6333 FLAG_CHECK(UF_REPARSE, XAT_REPARSE,
6334 xvap.xva_xoptattrs.xoa_reparse);
6335 FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE,
6336 xvap.xva_xoptattrs.xoa_offline);
6337 FLAG_CHECK(UF_SPARSE, XAT_SPARSE,
6338 xvap.xva_xoptattrs.xoa_sparse);
6341 *vap = xvap.xva_vattr;
6342 vap->va_flags = fflags;
6347 zfs_freebsd_setattr(ap)
6348 struct vop_setattr_args /* {
6350 struct vattr *a_vap;
6351 struct ucred *a_cred;
6354 vnode_t *vp = ap->a_vp;
6355 vattr_t *vap = ap->a_vap;
6356 cred_t *cred = ap->a_cred;
6361 vattr_init_mask(vap);
6362 vap->va_mask &= ~AT_NOSET;
6365 xvap.xva_vattr = *vap;
6367 zflags = VTOZ(vp)->z_pflags;
6369 if (vap->va_flags != VNOVAL) {
6370 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs;
6373 if (zfsvfs->z_use_fuids == B_FALSE)
6374 return (EOPNOTSUPP);
6376 fflags = vap->va_flags;
6379 * We need to figure out whether it makes sense to allow
6380 * UF_REPARSE through, since we don't really have other
6381 * facilities to handle reparse points and zfs_setattr()
6382 * doesn't currently allow setting that attribute anyway.
6384 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE|
6385 UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE|
6386 UF_OFFLINE|UF_SPARSE)) != 0)
6387 return (EOPNOTSUPP);
6389 * Unprivileged processes are not permitted to unset system
6390 * flags, or modify flags if any system flags are set.
6391 * Privileged non-jail processes may not modify system flags
6392 * if securelevel > 0 and any existing system flags are set.
6393 * Privileged jail processes behave like privileged non-jail
6394 * processes if the security.jail.chflags_allowed sysctl is
6395 * is non-zero; otherwise, they behave like unprivileged
6398 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 ||
6399 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) {
6401 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6402 error = securelevel_gt(cred, 0);
6408 * Callers may only modify the file flags on objects they
6409 * have VADMIN rights for.
6411 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0)
6414 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6418 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) {
6423 #define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \
6424 if (((fflags & (fflag)) && !(zflags & (zflag))) || \
6425 ((zflags & (zflag)) && !(fflags & (fflag)))) { \
6426 XVA_SET_REQ(&xvap, (xflag)); \
6427 (xfield) = ((fflags & (fflag)) != 0); \
6430 /* Convert chflags into ZFS-type flags. */
6431 /* XXX: what about SF_SETTABLE?. */
6432 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE,
6433 xvap.xva_xoptattrs.xoa_immutable);
6434 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY,
6435 xvap.xva_xoptattrs.xoa_appendonly);
6436 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK,
6437 xvap.xva_xoptattrs.xoa_nounlink);
6438 FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE,
6439 xvap.xva_xoptattrs.xoa_archive);
6440 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP,
6441 xvap.xva_xoptattrs.xoa_nodump);
6442 FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY,
6443 xvap.xva_xoptattrs.xoa_readonly);
6444 FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM,
6445 xvap.xva_xoptattrs.xoa_system);
6446 FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN,
6447 xvap.xva_xoptattrs.xoa_hidden);
6448 FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE,
6449 xvap.xva_xoptattrs.xoa_hidden);
6450 FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE,
6451 xvap.xva_xoptattrs.xoa_offline);
6452 FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE,
6453 xvap.xva_xoptattrs.xoa_sparse);
6456 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL));
6460 zfs_freebsd_rename(ap)
6461 struct vop_rename_args /* {
6462 struct vnode *a_fdvp;
6463 struct vnode *a_fvp;
6464 struct componentname *a_fcnp;
6465 struct vnode *a_tdvp;
6466 struct vnode *a_tvp;
6467 struct componentname *a_tcnp;
6470 vnode_t *fdvp = ap->a_fdvp;
6471 vnode_t *fvp = ap->a_fvp;
6472 vnode_t *tdvp = ap->a_tdvp;
6473 vnode_t *tvp = ap->a_tvp;
6476 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART));
6477 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART));
6480 * Check for cross-device rename.
6482 if ((fdvp->v_mount != tdvp->v_mount) ||
6483 (tvp && (fdvp->v_mount != tvp->v_mount)))
6486 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp,
6487 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0);
6501 zfs_freebsd_symlink(ap)
6502 struct vop_symlink_args /* {
6503 struct vnode *a_dvp;
6504 struct vnode **a_vpp;
6505 struct componentname *a_cnp;
6506 struct vattr *a_vap;
6510 struct componentname *cnp = ap->a_cnp;
6511 vattr_t *vap = ap->a_vap;
6513 ASSERT(cnp->cn_flags & SAVENAME);
6515 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */
6516 vattr_init_mask(vap);
6518 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap,
6519 ap->a_target, cnp->cn_cred, cnp->cn_thread));
6523 zfs_freebsd_readlink(ap)
6524 struct vop_readlink_args /* {
6527 struct ucred *a_cred;
6531 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL));
6535 zfs_freebsd_link(ap)
6536 struct vop_link_args /* {
6537 struct vnode *a_tdvp;
6539 struct componentname *a_cnp;
6542 struct componentname *cnp = ap->a_cnp;
6543 vnode_t *vp = ap->a_vp;
6544 vnode_t *tdvp = ap->a_tdvp;
6546 if (tdvp->v_mount != vp->v_mount)
6549 ASSERT(cnp->cn_flags & SAVENAME);
6551 return (zfs_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0));
6555 zfs_freebsd_inactive(ap)
6556 struct vop_inactive_args /* {
6558 struct thread *a_td;
6561 vnode_t *vp = ap->a_vp;
6563 zfs_inactive(vp, ap->a_td->td_ucred, NULL);
6568 zfs_freebsd_reclaim(ap)
6569 struct vop_reclaim_args /* {
6571 struct thread *a_td;
6574 vnode_t *vp = ap->a_vp;
6575 znode_t *zp = VTOZ(vp);
6576 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
6580 /* Destroy the vm object and flush associated pages. */
6581 vnode_destroy_vobject(vp);
6584 * z_teardown_inactive_lock protects from a race with
6585 * zfs_znode_dmu_fini in zfsvfs_teardown during
6588 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
6589 if (zp->z_sa_hdl == NULL)
6593 rw_exit(&zfsvfs->z_teardown_inactive_lock);
6601 struct vop_fid_args /* {
6607 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL));
6611 zfs_freebsd_pathconf(ap)
6612 struct vop_pathconf_args /* {
6615 register_t *a_retval;
6621 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL);
6623 *ap->a_retval = val;
6624 else if (error == EOPNOTSUPP)
6625 error = vop_stdpathconf(ap);
6630 zfs_freebsd_fifo_pathconf(ap)
6631 struct vop_pathconf_args /* {
6634 register_t *a_retval;
6638 switch (ap->a_name) {
6639 case _PC_ACL_EXTENDED:
6641 case _PC_ACL_PATH_MAX:
6642 case _PC_MAC_PRESENT:
6643 return (zfs_freebsd_pathconf(ap));
6645 return (fifo_specops.vop_pathconf(ap));
6650 * FreeBSD's extended attributes namespace defines file name prefix for ZFS'
6651 * extended attribute name:
6654 * system freebsd:system:
6655 * user (none, can be used to access ZFS fsattr(5) attributes
6656 * created on Solaris)
6659 zfs_create_attrname(int attrnamespace, const char *name, char *attrname,
6662 const char *namespace, *prefix, *suffix;
6664 /* We don't allow '/' character in attribute name. */
6665 if (strchr(name, '/') != NULL)
6667 /* We don't allow attribute names that start with "freebsd:" string. */
6668 if (strncmp(name, "freebsd:", 8) == 0)
6671 bzero(attrname, size);
6673 switch (attrnamespace) {
6674 case EXTATTR_NAMESPACE_USER:
6676 prefix = "freebsd:";
6677 namespace = EXTATTR_NAMESPACE_USER_STRING;
6681 * This is the default namespace by which we can access all
6682 * attributes created on Solaris.
6684 prefix = namespace = suffix = "";
6687 case EXTATTR_NAMESPACE_SYSTEM:
6688 prefix = "freebsd:";
6689 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING;
6692 case EXTATTR_NAMESPACE_EMPTY:
6696 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix,
6698 return (ENAMETOOLONG);
6704 * Vnode operating to retrieve a named extended attribute.
6707 zfs_getextattr(struct vop_getextattr_args *ap)
6710 IN struct vnode *a_vp;
6711 IN int a_attrnamespace;
6712 IN const char *a_name;
6713 INOUT struct uio *a_uio;
6715 IN struct ucred *a_cred;
6716 IN struct thread *a_td;
6720 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6721 struct thread *td = ap->a_td;
6722 struct nameidata nd;
6725 vnode_t *xvp = NULL, *vp;
6728 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6729 ap->a_cred, ap->a_td, VREAD);
6733 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6740 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6748 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6750 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL);
6752 NDFREE(&nd, NDF_ONLY_PNBUF);
6755 if (error == ENOENT)
6760 if (ap->a_size != NULL) {
6761 error = VOP_GETATTR(vp, &va, ap->a_cred);
6763 *ap->a_size = (size_t)va.va_size;
6764 } else if (ap->a_uio != NULL)
6765 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6768 vn_close(vp, flags, ap->a_cred, td);
6775 * Vnode operation to remove a named attribute.
6778 zfs_deleteextattr(struct vop_deleteextattr_args *ap)
6781 IN struct vnode *a_vp;
6782 IN int a_attrnamespace;
6783 IN const char *a_name;
6784 IN struct ucred *a_cred;
6785 IN struct thread *a_td;
6789 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6790 struct thread *td = ap->a_td;
6791 struct nameidata nd;
6794 vnode_t *xvp = NULL, *vp;
6797 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6798 ap->a_cred, ap->a_td, VWRITE);
6802 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6809 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6816 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF,
6817 UIO_SYSSPACE, attrname, xvp, td);
6822 NDFREE(&nd, NDF_ONLY_PNBUF);
6823 if (error == ENOENT)
6828 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
6829 NDFREE(&nd, NDF_ONLY_PNBUF);
6832 if (vp == nd.ni_dvp)
6842 * Vnode operation to set a named attribute.
6845 zfs_setextattr(struct vop_setextattr_args *ap)
6848 IN struct vnode *a_vp;
6849 IN int a_attrnamespace;
6850 IN const char *a_name;
6851 INOUT struct uio *a_uio;
6852 IN struct ucred *a_cred;
6853 IN struct thread *a_td;
6857 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6858 struct thread *td = ap->a_td;
6859 struct nameidata nd;
6862 vnode_t *xvp = NULL, *vp;
6865 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6866 ap->a_cred, ap->a_td, VWRITE);
6870 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6877 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6878 LOOKUP_XATTR | CREATE_XATTR_DIR);
6884 flags = FFLAGS(O_WRONLY | O_CREAT);
6885 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6887 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL);
6889 NDFREE(&nd, NDF_ONLY_PNBUF);
6897 error = VOP_SETATTR(vp, &va, ap->a_cred);
6899 VOP_WRITE(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6902 vn_close(vp, flags, ap->a_cred, td);
6909 * Vnode operation to retrieve extended attributes on a vnode.
6912 zfs_listextattr(struct vop_listextattr_args *ap)
6915 IN struct vnode *a_vp;
6916 IN int a_attrnamespace;
6917 INOUT struct uio *a_uio;
6919 IN struct ucred *a_cred;
6920 IN struct thread *a_td;
6924 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6925 struct thread *td = ap->a_td;
6926 struct nameidata nd;
6927 char attrprefix[16];
6928 u_char dirbuf[sizeof(struct dirent)];
6931 struct uio auio, *uio = ap->a_uio;
6932 size_t *sizep = ap->a_size;
6934 vnode_t *xvp = NULL, *vp;
6935 int done, error, eof, pos;
6937 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6938 ap->a_cred, ap->a_td, VREAD);
6942 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix,
6943 sizeof(attrprefix));
6946 plen = strlen(attrprefix);
6953 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6958 * ENOATTR means that the EA directory does not yet exist,
6959 * i.e. there are no extended attributes there.
6961 if (error == ENOATTR)
6966 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED,
6967 UIO_SYSSPACE, ".", xvp, td);
6970 NDFREE(&nd, NDF_ONLY_PNBUF);
6976 auio.uio_iov = &aiov;
6977 auio.uio_iovcnt = 1;
6978 auio.uio_segflg = UIO_SYSSPACE;
6980 auio.uio_rw = UIO_READ;
6981 auio.uio_offset = 0;
6986 aiov.iov_base = (void *)dirbuf;
6987 aiov.iov_len = sizeof(dirbuf);
6988 auio.uio_resid = sizeof(dirbuf);
6989 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL);
6990 done = sizeof(dirbuf) - auio.uio_resid;
6993 for (pos = 0; pos < done;) {
6994 dp = (struct dirent *)(dirbuf + pos);
6995 pos += dp->d_reclen;
6997 * XXX: Temporarily we also accept DT_UNKNOWN, as this
6998 * is what we get when attribute was created on Solaris.
7000 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN)
7002 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0)
7004 else if (strncmp(dp->d_name, attrprefix, plen) != 0)
7006 nlen = dp->d_namlen - plen;
7009 else if (uio != NULL) {
7011 * Format of extattr name entry is one byte for
7012 * length and the rest for name.
7014 error = uiomove(&nlen, 1, uio->uio_rw, uio);
7016 error = uiomove(dp->d_name + plen, nlen,
7023 } while (!eof && error == 0);
7032 zfs_freebsd_getacl(ap)
7033 struct vop_getacl_args /* {
7042 vsecattr_t vsecattr;
7044 if (ap->a_type != ACL_TYPE_NFS4)
7047 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
7048 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL))
7051 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt);
7052 if (vsecattr.vsa_aclentp != NULL)
7053 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz);
7059 zfs_freebsd_setacl(ap)
7060 struct vop_setacl_args /* {
7069 vsecattr_t vsecattr;
7070 int aclbsize; /* size of acl list in bytes */
7073 if (ap->a_type != ACL_TYPE_NFS4)
7076 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES)
7080 * With NFSv4 ACLs, chmod(2) may need to add additional entries,
7081 * splitting every entry into two and appending "canonical six"
7082 * entries at the end. Don't allow for setting an ACL that would
7083 * cause chmod(2) to run out of ACL entries.
7085 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES)
7088 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR);
7092 vsecattr.vsa_mask = VSA_ACE;
7093 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t);
7094 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP);
7095 aaclp = vsecattr.vsa_aclentp;
7096 vsecattr.vsa_aclentsz = aclbsize;
7098 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp);
7099 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL);
7100 kmem_free(aaclp, aclbsize);
7106 zfs_freebsd_aclcheck(ap)
7107 struct vop_aclcheck_args /* {
7116 return (EOPNOTSUPP);
7119 struct vop_vector zfs_vnodeops;
7120 struct vop_vector zfs_fifoops;
7121 struct vop_vector zfs_shareops;
7123 struct vop_vector zfs_vnodeops = {
7124 .vop_default = &default_vnodeops,
7125 .vop_inactive = zfs_freebsd_inactive,
7126 .vop_reclaim = zfs_freebsd_reclaim,
7127 .vop_access = zfs_freebsd_access,
7128 #ifdef FREEBSD_NAMECACHE
7129 .vop_lookup = vfs_cache_lookup,
7130 .vop_cachedlookup = zfs_freebsd_lookup,
7132 .vop_lookup = zfs_freebsd_lookup,
7134 .vop_getattr = zfs_freebsd_getattr,
7135 .vop_setattr = zfs_freebsd_setattr,
7136 .vop_create = zfs_freebsd_create,
7137 .vop_mknod = zfs_freebsd_create,
7138 .vop_mkdir = zfs_freebsd_mkdir,
7139 .vop_readdir = zfs_freebsd_readdir,
7140 .vop_fsync = zfs_freebsd_fsync,
7141 .vop_open = zfs_freebsd_open,
7142 .vop_close = zfs_freebsd_close,
7143 .vop_rmdir = zfs_freebsd_rmdir,
7144 .vop_ioctl = zfs_freebsd_ioctl,
7145 .vop_link = zfs_freebsd_link,
7146 .vop_symlink = zfs_freebsd_symlink,
7147 .vop_readlink = zfs_freebsd_readlink,
7148 .vop_read = zfs_freebsd_read,
7149 .vop_write = zfs_freebsd_write,
7150 .vop_remove = zfs_freebsd_remove,
7151 .vop_rename = zfs_freebsd_rename,
7152 .vop_pathconf = zfs_freebsd_pathconf,
7153 .vop_bmap = zfs_freebsd_bmap,
7154 .vop_fid = zfs_freebsd_fid,
7155 .vop_getextattr = zfs_getextattr,
7156 .vop_deleteextattr = zfs_deleteextattr,
7157 .vop_setextattr = zfs_setextattr,
7158 .vop_listextattr = zfs_listextattr,
7159 .vop_getacl = zfs_freebsd_getacl,
7160 .vop_setacl = zfs_freebsd_setacl,
7161 .vop_aclcheck = zfs_freebsd_aclcheck,
7162 .vop_getpages = zfs_freebsd_getpages,
7163 .vop_putpages = zfs_freebsd_putpages,
7166 struct vop_vector zfs_fifoops = {
7167 .vop_default = &fifo_specops,
7168 .vop_fsync = zfs_freebsd_fsync,
7169 .vop_access = zfs_freebsd_access,
7170 .vop_getattr = zfs_freebsd_getattr,
7171 .vop_inactive = zfs_freebsd_inactive,
7172 .vop_read = VOP_PANIC,
7173 .vop_reclaim = zfs_freebsd_reclaim,
7174 .vop_setattr = zfs_freebsd_setattr,
7175 .vop_write = VOP_PANIC,
7176 .vop_pathconf = zfs_freebsd_fifo_pathconf,
7177 .vop_fid = zfs_freebsd_fid,
7178 .vop_getacl = zfs_freebsd_getacl,
7179 .vop_setacl = zfs_freebsd_setacl,
7180 .vop_aclcheck = zfs_freebsd_aclcheck,
7184 * special share hidden files vnode operations template
7186 struct vop_vector zfs_shareops = {
7187 .vop_default = &default_vnodeops,
7188 .vop_access = zfs_freebsd_access,
7189 .vop_inactive = zfs_freebsd_inactive,
7190 .vop_reclaim = zfs_freebsd_reclaim,
7191 .vop_fid = zfs_freebsd_fid,
7192 .vop_pathconf = zfs_freebsd_pathconf,