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, 2015 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)
297 dmu_object_info_t doi;
308 * The following two ioctls are used by bfu. Faking out,
309 * necessary to avoid bfu errors.
322 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
323 return (SET_ERROR(EFAULT));
325 off = *(offset_t *)data;
328 zfsvfs = zp->z_zfsvfs;
332 /* offset parameter is in/out */
333 error = zfs_holey(vp, com, &off);
338 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
339 return (SET_ERROR(EFAULT));
341 *(offset_t *)data = off;
346 case _FIO_COUNT_FILLED:
349 * _FIO_COUNT_FILLED adds a new ioctl command which
350 * exposes the number of filled blocks in a
354 zfsvfs = zp->z_zfsvfs;
359 * Wait for all dirty blocks for this object
360 * to get synced out to disk, and the DMU info
363 error = dmu_object_wait_synced(zfsvfs->z_os, zp->z_id);
370 * Retrieve fill count from DMU object.
372 error = dmu_object_info(zfsvfs->z_os, zp->z_id, &doi);
378 ndata = doi.doi_fill_count;
381 if (ddi_copyout(&ndata, (void *)data, sizeof (ndata), flag))
382 return (SET_ERROR(EFAULT));
387 return (SET_ERROR(ENOTTY));
391 page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes)
398 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE
399 * aligned boundaries, if the range is not aligned. As a result a
400 * DEV_BSIZE subrange with partially dirty data may get marked as clean.
401 * It may happen that all DEV_BSIZE subranges are marked clean and thus
402 * the whole page would be considred clean despite have some dirty data.
403 * For this reason we should shrink the range to DEV_BSIZE aligned
404 * boundaries before calling vm_page_clear_dirty.
406 end = rounddown2(off + nbytes, DEV_BSIZE);
407 off = roundup2(off, DEV_BSIZE);
411 zfs_vmobject_assert_wlocked(obj);
414 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
416 if (vm_page_xbusied(pp)) {
418 * Reference the page before unlocking and
419 * sleeping so that the page daemon is less
420 * likely to reclaim it.
422 vm_page_reference(pp);
424 zfs_vmobject_wunlock(obj);
425 vm_page_busy_sleep(pp, "zfsmwb");
426 zfs_vmobject_wlock(obj);
430 } else if (pp == NULL) {
431 pp = vm_page_alloc(obj, OFF_TO_IDX(start),
432 VM_ALLOC_SYSTEM | VM_ALLOC_IFCACHED |
435 ASSERT(pp != NULL && !pp->valid);
440 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
441 vm_object_pip_add(obj, 1);
442 pmap_remove_write(pp);
444 vm_page_clear_dirty(pp, off, nbytes);
452 page_unbusy(vm_page_t pp)
456 vm_object_pip_subtract(pp->object, 1);
460 page_hold(vnode_t *vp, int64_t start)
466 zfs_vmobject_assert_wlocked(obj);
469 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
471 if (vm_page_xbusied(pp)) {
473 * Reference the page before unlocking and
474 * sleeping so that the page daemon is less
475 * likely to reclaim it.
477 vm_page_reference(pp);
479 zfs_vmobject_wunlock(obj);
480 vm_page_busy_sleep(pp, "zfsmwb");
481 zfs_vmobject_wlock(obj);
485 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
498 page_unhold(vm_page_t pp)
507 * When a file is memory mapped, we must keep the IO data synchronized
508 * between the DMU cache and the memory mapped pages. What this means:
510 * On Write: If we find a memory mapped page, we write to *both*
511 * the page and the dmu buffer.
514 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid,
515 int segflg, dmu_tx_t *tx)
522 ASSERT(segflg != UIO_NOCOPY);
523 ASSERT(vp->v_mount != NULL);
527 off = start & PAGEOFFSET;
528 zfs_vmobject_wlock(obj);
529 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
531 int nbytes = imin(PAGESIZE - off, len);
533 if ((pp = page_busy(vp, start, off, nbytes)) != NULL) {
534 zfs_vmobject_wunlock(obj);
536 va = zfs_map_page(pp, &sf);
537 (void) dmu_read(os, oid, start+off, nbytes,
538 va+off, DMU_READ_PREFETCH);;
541 zfs_vmobject_wlock(obj);
547 vm_object_pip_wakeupn(obj, 0);
548 zfs_vmobject_wunlock(obj);
552 * Read with UIO_NOCOPY flag means that sendfile(2) requests
553 * ZFS to populate a range of page cache pages with data.
555 * NOTE: this function could be optimized to pre-allocate
556 * all pages in advance, drain exclusive busy on all of them,
557 * map them into contiguous KVA region and populate them
558 * in one single dmu_read() call.
561 mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio)
563 znode_t *zp = VTOZ(vp);
564 objset_t *os = zp->z_zfsvfs->z_os;
574 ASSERT(uio->uio_segflg == UIO_NOCOPY);
575 ASSERT(vp->v_mount != NULL);
578 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0);
580 zfs_vmobject_wlock(obj);
581 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) {
582 int bytes = MIN(PAGESIZE, len);
584 pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_SBUSY |
585 VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY);
586 if (pp->valid == 0) {
587 zfs_vmobject_wunlock(obj);
588 va = zfs_map_page(pp, &sf);
589 error = dmu_read(os, zp->z_id, start, bytes, va,
591 if (bytes != PAGESIZE && error == 0)
592 bzero(va + bytes, PAGESIZE - bytes);
594 zfs_vmobject_wlock(obj);
598 if (pp->wire_count == 0 && pp->valid == 0 &&
602 pp->valid = VM_PAGE_BITS_ALL;
603 vm_page_activate(pp);
607 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
612 uio->uio_resid -= bytes;
613 uio->uio_offset += bytes;
616 zfs_vmobject_wunlock(obj);
621 * When a file is memory mapped, we must keep the IO data synchronized
622 * between the DMU cache and the memory mapped pages. What this means:
624 * On Read: We "read" preferentially from memory mapped pages,
625 * else we default from the dmu buffer.
627 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
628 * the file is memory mapped.
631 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
633 znode_t *zp = VTOZ(vp);
641 ASSERT(vp->v_mount != NULL);
645 start = uio->uio_loffset;
646 off = start & PAGEOFFSET;
647 zfs_vmobject_wlock(obj);
648 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
650 uint64_t bytes = MIN(PAGESIZE - off, len);
652 if (pp = page_hold(vp, start)) {
656 zfs_vmobject_wunlock(obj);
657 va = zfs_map_page(pp, &sf);
658 error = uiomove(va + off, bytes, UIO_READ, uio);
660 zfs_vmobject_wlock(obj);
663 zfs_vmobject_wunlock(obj);
664 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
666 zfs_vmobject_wlock(obj);
673 zfs_vmobject_wunlock(obj);
677 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
680 * Read bytes from specified file into supplied buffer.
682 * IN: vp - vnode of file to be read from.
683 * uio - structure supplying read location, range info,
685 * ioflag - SYNC flags; used to provide FRSYNC semantics.
686 * cr - credentials of caller.
687 * ct - caller context
689 * OUT: uio - updated offset and range, buffer filled.
691 * RETURN: 0 on success, error code on failure.
694 * vp - atime updated if byte count > 0
698 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
700 znode_t *zp = VTOZ(vp);
701 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
710 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
712 return (SET_ERROR(EACCES));
716 * Validate file offset
718 if (uio->uio_loffset < (offset_t)0) {
720 return (SET_ERROR(EINVAL));
724 * Fasttrack empty reads
726 if (uio->uio_resid == 0) {
732 * Check for mandatory locks
734 if (MANDMODE(zp->z_mode)) {
735 if (error = chklock(vp, FREAD,
736 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
743 * If we're in FRSYNC mode, sync out this znode before reading it.
746 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
747 zil_commit(zfsvfs->z_log, zp->z_id);
750 * Lock the range against changes.
752 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
755 * If we are reading past end-of-file we can skip
756 * to the end; but we might still need to set atime.
758 if (uio->uio_loffset >= zp->z_size) {
763 ASSERT(uio->uio_loffset < zp->z_size);
764 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
767 if ((uio->uio_extflg == UIO_XUIO) &&
768 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
770 int blksz = zp->z_blksz;
771 uint64_t offset = uio->uio_loffset;
773 xuio = (xuio_t *)uio;
775 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
778 ASSERT(offset + n <= blksz);
781 (void) dmu_xuio_init(xuio, nblk);
783 if (vn_has_cached_data(vp)) {
785 * For simplicity, we always allocate a full buffer
786 * even if we only expect to read a portion of a block.
788 while (--nblk >= 0) {
789 (void) dmu_xuio_add(xuio,
790 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
798 nbytes = MIN(n, zfs_read_chunk_size -
799 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
802 if (uio->uio_segflg == UIO_NOCOPY)
803 error = mappedread_sf(vp, nbytes, uio);
805 #endif /* __FreeBSD__ */
806 if (vn_has_cached_data(vp)) {
807 error = mappedread(vp, nbytes, uio);
809 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
813 /* convert checksum errors into IO errors */
815 error = SET_ERROR(EIO);
822 zfs_range_unlock(rl);
824 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
830 * Write the bytes to a file.
832 * IN: vp - vnode of file to be written to.
833 * uio - structure supplying write location, range info,
835 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is
836 * set if in append mode.
837 * cr - credentials of caller.
838 * ct - caller context (NFS/CIFS fem monitor only)
840 * OUT: uio - updated offset and range.
842 * RETURN: 0 on success, error code on failure.
845 * vp - ctime|mtime updated if byte count > 0
850 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
852 znode_t *zp = VTOZ(vp);
853 rlim64_t limit = MAXOFFSET_T;
854 ssize_t start_resid = uio->uio_resid;
858 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
863 int max_blksz = zfsvfs->z_max_blksz;
866 iovec_t *aiov = NULL;
869 int iovcnt = uio->uio_iovcnt;
870 iovec_t *iovp = uio->uio_iov;
873 sa_bulk_attr_t bulk[4];
874 uint64_t mtime[2], ctime[2];
877 * Fasttrack empty write
883 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
889 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
890 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
891 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
893 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
897 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
898 * callers might not be able to detect properly that we are read-only,
899 * so check it explicitly here.
901 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
903 return (SET_ERROR(EROFS));
907 * If immutable or not appending then return EPERM
909 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
910 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
911 (uio->uio_loffset < zp->z_size))) {
913 return (SET_ERROR(EPERM));
916 zilog = zfsvfs->z_log;
919 * Validate file offset
921 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
924 return (SET_ERROR(EINVAL));
928 * Check for mandatory locks before calling zfs_range_lock()
929 * in order to prevent a deadlock with locks set via fcntl().
931 if (MANDMODE((mode_t)zp->z_mode) &&
932 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
939 * Pre-fault the pages to ensure slow (eg NFS) pages
941 * Skip this if uio contains loaned arc_buf.
943 if ((uio->uio_extflg == UIO_XUIO) &&
944 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
945 xuio = (xuio_t *)uio;
947 uio_prefaultpages(MIN(n, max_blksz), uio);
951 * If in append mode, set the io offset pointer to eof.
953 if (ioflag & FAPPEND) {
955 * Obtain an appending range lock to guarantee file append
956 * semantics. We reset the write offset once we have the lock.
958 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
960 if (rl->r_len == UINT64_MAX) {
962 * We overlocked the file because this write will cause
963 * the file block size to increase.
964 * Note that zp_size cannot change with this lock held.
968 uio->uio_loffset = woff;
971 * Note that if the file block size will change as a result of
972 * this write, then this range lock will lock the entire file
973 * so that we can re-write the block safely.
975 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
978 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) {
979 zfs_range_unlock(rl);
985 zfs_range_unlock(rl);
987 return (SET_ERROR(EFBIG));
990 if ((woff + n) > limit || woff > (limit - n))
993 /* Will this write extend the file length? */
994 write_eof = (woff + n > zp->z_size);
996 end_size = MAX(zp->z_size, woff + n);
999 * Write the file in reasonable size chunks. Each chunk is written
1000 * in a separate transaction; this keeps the intent log records small
1001 * and allows us to do more fine-grained space accounting.
1005 woff = uio->uio_loffset;
1006 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
1007 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
1009 dmu_return_arcbuf(abuf);
1010 error = SET_ERROR(EDQUOT);
1014 if (xuio && abuf == NULL) {
1015 ASSERT(i_iov < iovcnt);
1016 aiov = &iovp[i_iov];
1017 abuf = dmu_xuio_arcbuf(xuio, i_iov);
1018 dmu_xuio_clear(xuio, i_iov);
1019 DTRACE_PROBE3(zfs_cp_write, int, i_iov,
1020 iovec_t *, aiov, arc_buf_t *, abuf);
1021 ASSERT((aiov->iov_base == abuf->b_data) ||
1022 ((char *)aiov->iov_base - (char *)abuf->b_data +
1023 aiov->iov_len == arc_buf_size(abuf)));
1025 } else if (abuf == NULL && n >= max_blksz &&
1026 woff >= zp->z_size &&
1027 P2PHASE(woff, max_blksz) == 0 &&
1028 zp->z_blksz == max_blksz) {
1030 * This write covers a full block. "Borrow" a buffer
1031 * from the dmu so that we can fill it before we enter
1032 * a transaction. This avoids the possibility of
1033 * holding up the transaction if the data copy hangs
1034 * up on a pagefault (e.g., from an NFS server mapping).
1038 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
1040 ASSERT(abuf != NULL);
1041 ASSERT(arc_buf_size(abuf) == max_blksz);
1042 if (error = uiocopy(abuf->b_data, max_blksz,
1043 UIO_WRITE, uio, &cbytes)) {
1044 dmu_return_arcbuf(abuf);
1047 ASSERT(cbytes == max_blksz);
1051 * Start a transaction.
1053 tx = dmu_tx_create(zfsvfs->z_os);
1054 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1055 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
1056 zfs_sa_upgrade_txholds(tx, zp);
1057 error = dmu_tx_assign(tx, TXG_WAIT);
1061 dmu_return_arcbuf(abuf);
1066 * If zfs_range_lock() over-locked we grow the blocksize
1067 * and then reduce the lock range. This will only happen
1068 * on the first iteration since zfs_range_reduce() will
1069 * shrink down r_len to the appropriate size.
1071 if (rl->r_len == UINT64_MAX) {
1074 if (zp->z_blksz > max_blksz) {
1076 * File's blocksize is already larger than the
1077 * "recordsize" property. Only let it grow to
1078 * the next power of 2.
1080 ASSERT(!ISP2(zp->z_blksz));
1081 new_blksz = MIN(end_size,
1082 1 << highbit64(zp->z_blksz));
1084 new_blksz = MIN(end_size, max_blksz);
1086 zfs_grow_blocksize(zp, new_blksz, tx);
1087 zfs_range_reduce(rl, woff, n);
1091 * XXX - should we really limit each write to z_max_blksz?
1092 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
1094 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
1096 if (woff + nbytes > zp->z_size)
1097 vnode_pager_setsize(vp, woff + nbytes);
1100 tx_bytes = uio->uio_resid;
1101 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
1103 tx_bytes -= uio->uio_resid;
1106 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
1108 * If this is not a full block write, but we are
1109 * extending the file past EOF and this data starts
1110 * block-aligned, use assign_arcbuf(). Otherwise,
1111 * write via dmu_write().
1113 if (tx_bytes < max_blksz && (!write_eof ||
1114 aiov->iov_base != abuf->b_data)) {
1116 dmu_write(zfsvfs->z_os, zp->z_id, woff,
1117 aiov->iov_len, aiov->iov_base, tx);
1118 dmu_return_arcbuf(abuf);
1119 xuio_stat_wbuf_copied();
1121 ASSERT(xuio || tx_bytes == max_blksz);
1122 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
1125 ASSERT(tx_bytes <= uio->uio_resid);
1126 uioskip(uio, tx_bytes);
1128 if (tx_bytes && vn_has_cached_data(vp)) {
1129 update_pages(vp, woff, tx_bytes, zfsvfs->z_os,
1130 zp->z_id, uio->uio_segflg, tx);
1134 * If we made no progress, we're done. If we made even
1135 * partial progress, update the znode and ZIL accordingly.
1137 if (tx_bytes == 0) {
1138 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
1139 (void *)&zp->z_size, sizeof (uint64_t), tx);
1146 * Clear Set-UID/Set-GID bits on successful write if not
1147 * privileged and at least one of the excute bits is set.
1149 * It would be nice to to this after all writes have
1150 * been done, but that would still expose the ISUID/ISGID
1151 * to another app after the partial write is committed.
1153 * Note: we don't call zfs_fuid_map_id() here because
1154 * user 0 is not an ephemeral uid.
1156 mutex_enter(&zp->z_acl_lock);
1157 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
1158 (S_IXUSR >> 6))) != 0 &&
1159 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
1160 secpolicy_vnode_setid_retain(vp, cr,
1161 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
1163 zp->z_mode &= ~(S_ISUID | S_ISGID);
1164 newmode = zp->z_mode;
1165 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
1166 (void *)&newmode, sizeof (uint64_t), tx);
1168 mutex_exit(&zp->z_acl_lock);
1170 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
1174 * Update the file size (zp_size) if it has changed;
1175 * account for possible concurrent updates.
1177 while ((end_size = zp->z_size) < uio->uio_loffset) {
1178 (void) atomic_cas_64(&zp->z_size, end_size,
1183 * If we are replaying and eof is non zero then force
1184 * the file size to the specified eof. Note, there's no
1185 * concurrency during replay.
1187 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
1188 zp->z_size = zfsvfs->z_replay_eof;
1190 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1192 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
1197 ASSERT(tx_bytes == nbytes);
1202 uio_prefaultpages(MIN(n, max_blksz), uio);
1206 zfs_range_unlock(rl);
1209 * If we're in replay mode, or we made no progress, return error.
1210 * Otherwise, it's at least a partial write, so it's successful.
1212 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1217 if (ioflag & (FSYNC | FDSYNC) ||
1218 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1219 zil_commit(zilog, zp->z_id);
1226 zfs_get_done(zgd_t *zgd, int error)
1228 znode_t *zp = zgd->zgd_private;
1229 objset_t *os = zp->z_zfsvfs->z_os;
1232 dmu_buf_rele(zgd->zgd_db, zgd);
1234 zfs_range_unlock(zgd->zgd_rl);
1237 * Release the vnode asynchronously as we currently have the
1238 * txg stopped from syncing.
1240 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1242 if (error == 0 && zgd->zgd_bp)
1243 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1245 kmem_free(zgd, sizeof (zgd_t));
1249 static int zil_fault_io = 0;
1253 * Get data to generate a TX_WRITE intent log record.
1256 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1258 zfsvfs_t *zfsvfs = arg;
1259 objset_t *os = zfsvfs->z_os;
1261 uint64_t object = lr->lr_foid;
1262 uint64_t offset = lr->lr_offset;
1263 uint64_t size = lr->lr_length;
1264 blkptr_t *bp = &lr->lr_blkptr;
1269 ASSERT(zio != NULL);
1273 * Nothing to do if the file has been removed
1275 if (zfs_zget(zfsvfs, object, &zp) != 0)
1276 return (SET_ERROR(ENOENT));
1277 if (zp->z_unlinked) {
1279 * Release the vnode asynchronously as we currently have the
1280 * txg stopped from syncing.
1282 VN_RELE_ASYNC(ZTOV(zp),
1283 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1284 return (SET_ERROR(ENOENT));
1287 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1288 zgd->zgd_zilog = zfsvfs->z_log;
1289 zgd->zgd_private = zp;
1292 * Write records come in two flavors: immediate and indirect.
1293 * For small writes it's cheaper to store the data with the
1294 * log record (immediate); for large writes it's cheaper to
1295 * sync the data and get a pointer to it (indirect) so that
1296 * we don't have to write the data twice.
1298 if (buf != NULL) { /* immediate write */
1299 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1300 /* test for truncation needs to be done while range locked */
1301 if (offset >= zp->z_size) {
1302 error = SET_ERROR(ENOENT);
1304 error = dmu_read(os, object, offset, size, buf,
1305 DMU_READ_NO_PREFETCH);
1307 ASSERT(error == 0 || error == ENOENT);
1308 } else { /* indirect write */
1310 * Have to lock the whole block to ensure when it's
1311 * written out and it's checksum is being calculated
1312 * that no one can change the data. We need to re-check
1313 * blocksize after we get the lock in case it's changed!
1318 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1320 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1322 if (zp->z_blksz == size)
1325 zfs_range_unlock(zgd->zgd_rl);
1327 /* test for truncation needs to be done while range locked */
1328 if (lr->lr_offset >= zp->z_size)
1329 error = SET_ERROR(ENOENT);
1332 error = SET_ERROR(EIO);
1337 error = dmu_buf_hold(os, object, offset, zgd, &db,
1338 DMU_READ_NO_PREFETCH);
1341 blkptr_t *obp = dmu_buf_get_blkptr(db);
1343 ASSERT(BP_IS_HOLE(bp));
1350 ASSERT(db->db_offset == offset);
1351 ASSERT(db->db_size == size);
1353 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1355 ASSERT(error || lr->lr_length <= zp->z_blksz);
1358 * On success, we need to wait for the write I/O
1359 * initiated by dmu_sync() to complete before we can
1360 * release this dbuf. We will finish everything up
1361 * in the zfs_get_done() callback.
1366 if (error == EALREADY) {
1367 lr->lr_common.lrc_txtype = TX_WRITE2;
1373 zfs_get_done(zgd, error);
1380 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1381 caller_context_t *ct)
1383 znode_t *zp = VTOZ(vp);
1384 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1390 if (flag & V_ACE_MASK)
1391 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1393 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1400 * If vnode is for a device return a specfs vnode instead.
1403 specvp_check(vnode_t **vpp, cred_t *cr)
1407 if (IS_DEVVP(*vpp)) {
1410 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1413 error = SET_ERROR(ENOSYS);
1421 * Lookup an entry in a directory, or an extended attribute directory.
1422 * If it exists, return a held vnode reference for it.
1424 * IN: dvp - vnode of directory to search.
1425 * nm - name of entry to lookup.
1426 * pnp - full pathname to lookup [UNUSED].
1427 * flags - LOOKUP_XATTR set if looking for an attribute.
1428 * rdir - root directory vnode [UNUSED].
1429 * cr - credentials of caller.
1430 * ct - caller context
1431 * direntflags - directory lookup flags
1432 * realpnp - returned pathname.
1434 * OUT: vpp - vnode of located entry, NULL if not found.
1436 * RETURN: 0 on success, error code on failure.
1443 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp,
1444 int nameiop, cred_t *cr, kthread_t *td, int flags)
1446 znode_t *zdp = VTOZ(dvp);
1447 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1449 int *direntflags = NULL;
1450 void *realpnp = NULL;
1453 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1455 if (dvp->v_type != VDIR) {
1456 return (SET_ERROR(ENOTDIR));
1457 } else if (zdp->z_sa_hdl == NULL) {
1458 return (SET_ERROR(EIO));
1461 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1462 error = zfs_fastaccesschk_execute(zdp, cr);
1470 vnode_t *tvp = dnlc_lookup(dvp, nm);
1473 error = zfs_fastaccesschk_execute(zdp, cr);
1478 if (tvp == DNLC_NO_VNODE) {
1480 return (SET_ERROR(ENOENT));
1483 return (specvp_check(vpp, cr));
1489 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1496 if (flags & LOOKUP_XATTR) {
1499 * If the xattr property is off, refuse the lookup request.
1501 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1503 return (SET_ERROR(EINVAL));
1508 * We don't allow recursive attributes..
1509 * Maybe someday we will.
1511 if (zdp->z_pflags & ZFS_XATTR) {
1513 return (SET_ERROR(EINVAL));
1516 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1522 * Do we have permission to get into attribute directory?
1525 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1535 if (dvp->v_type != VDIR) {
1537 return (SET_ERROR(ENOTDIR));
1541 * Check accessibility of directory.
1544 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1549 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1550 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1552 return (SET_ERROR(EILSEQ));
1555 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1557 error = specvp_check(vpp, cr);
1559 /* Translate errors and add SAVENAME when needed. */
1560 if (cnp->cn_flags & ISLASTCN) {
1564 if (error == ENOENT) {
1565 error = EJUSTRETURN;
1566 cnp->cn_flags |= SAVENAME;
1572 cnp->cn_flags |= SAVENAME;
1576 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) {
1579 if (cnp->cn_flags & ISDOTDOT) {
1580 ltype = VOP_ISLOCKED(dvp);
1584 error = vn_lock(*vpp, cnp->cn_lkflags);
1585 if (cnp->cn_flags & ISDOTDOT)
1586 vn_lock(dvp, ltype | LK_RETRY);
1596 #ifdef FREEBSD_NAMECACHE
1598 * Insert name into cache (as non-existent) if appropriate.
1600 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) != 0)
1601 cache_enter(dvp, *vpp, cnp);
1603 * Insert name into cache if appropriate.
1605 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) {
1606 if (!(cnp->cn_flags & ISLASTCN) ||
1607 (nameiop != DELETE && nameiop != RENAME)) {
1608 cache_enter(dvp, *vpp, cnp);
1617 * Attempt to create a new entry in a directory. If the entry
1618 * already exists, truncate the file if permissible, else return
1619 * an error. Return the vp of the created or trunc'd file.
1621 * IN: dvp - vnode of directory to put new file entry in.
1622 * name - name of new file entry.
1623 * vap - attributes of new file.
1624 * excl - flag indicating exclusive or non-exclusive mode.
1625 * mode - mode to open file with.
1626 * cr - credentials of caller.
1627 * flag - large file flag [UNUSED].
1628 * ct - caller context
1629 * vsecp - ACL to be set
1631 * OUT: vpp - vnode of created or trunc'd entry.
1633 * RETURN: 0 on success, error code on failure.
1636 * dvp - ctime|mtime updated if new entry created
1637 * vp - ctime|mtime always, atime if new
1642 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode,
1643 vnode_t **vpp, cred_t *cr, kthread_t *td)
1645 znode_t *zp, *dzp = VTOZ(dvp);
1646 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1654 gid_t gid = crgetgid(cr);
1655 zfs_acl_ids_t acl_ids;
1656 boolean_t fuid_dirtied;
1657 boolean_t have_acl = B_FALSE;
1658 boolean_t waited = B_FALSE;
1663 * If we have an ephemeral id, ACL, or XVATTR then
1664 * make sure file system is at proper version
1667 ksid = crgetsid(cr, KSID_OWNER);
1669 uid = ksid_getid(ksid);
1673 if (zfsvfs->z_use_fuids == B_FALSE &&
1674 (vsecp || (vap->va_mask & AT_XVATTR) ||
1675 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1676 return (SET_ERROR(EINVAL));
1681 zilog = zfsvfs->z_log;
1683 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1684 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1686 return (SET_ERROR(EILSEQ));
1689 if (vap->va_mask & AT_XVATTR) {
1690 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
1691 crgetuid(cr), cr, vap->va_type)) != 0) {
1697 getnewvnode_reserve(1);
1702 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr))
1703 vap->va_mode &= ~S_ISVTX;
1705 if (*name == '\0') {
1707 * Null component name refers to the directory itself.
1714 /* possible VN_HOLD(zp) */
1717 if (flag & FIGNORECASE)
1720 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1724 zfs_acl_ids_free(&acl_ids);
1725 if (strcmp(name, "..") == 0)
1726 error = SET_ERROR(EISDIR);
1727 getnewvnode_drop_reserve();
1737 * Create a new file object and update the directory
1740 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1742 zfs_acl_ids_free(&acl_ids);
1747 * We only support the creation of regular files in
1748 * extended attribute directories.
1751 if ((dzp->z_pflags & ZFS_XATTR) &&
1752 (vap->va_type != VREG)) {
1754 zfs_acl_ids_free(&acl_ids);
1755 error = SET_ERROR(EINVAL);
1759 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1760 cr, vsecp, &acl_ids)) != 0)
1764 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1765 zfs_acl_ids_free(&acl_ids);
1766 error = SET_ERROR(EDQUOT);
1770 tx = dmu_tx_create(os);
1772 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1773 ZFS_SA_BASE_ATTR_SIZE);
1775 fuid_dirtied = zfsvfs->z_fuid_dirty;
1777 zfs_fuid_txhold(zfsvfs, tx);
1778 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1779 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1780 if (!zfsvfs->z_use_sa &&
1781 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1782 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1783 0, acl_ids.z_aclp->z_acl_bytes);
1785 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1787 zfs_dirent_unlock(dl);
1788 if (error == ERESTART) {
1794 zfs_acl_ids_free(&acl_ids);
1796 getnewvnode_drop_reserve();
1800 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1803 zfs_fuid_sync(zfsvfs, tx);
1805 (void) zfs_link_create(dl, zp, tx, ZNEW);
1806 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1807 if (flag & FIGNORECASE)
1809 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1810 vsecp, acl_ids.z_fuidp, vap);
1811 zfs_acl_ids_free(&acl_ids);
1814 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1817 zfs_acl_ids_free(&acl_ids);
1821 * A directory entry already exists for this name.
1824 * Can't truncate an existing file if in exclusive mode.
1827 error = SET_ERROR(EEXIST);
1831 * Can't open a directory for writing.
1833 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1834 error = SET_ERROR(EISDIR);
1838 * Verify requested access to file.
1840 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1844 mutex_enter(&dzp->z_lock);
1846 mutex_exit(&dzp->z_lock);
1849 * Truncate regular files if requested.
1851 if ((ZTOV(zp)->v_type == VREG) &&
1852 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1853 /* we can't hold any locks when calling zfs_freesp() */
1854 zfs_dirent_unlock(dl);
1856 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1858 vnevent_create(ZTOV(zp), ct);
1863 getnewvnode_drop_reserve();
1865 zfs_dirent_unlock(dl);
1872 error = specvp_check(vpp, cr);
1875 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1876 zil_commit(zilog, 0);
1883 * Remove an entry from a directory.
1885 * IN: dvp - vnode of directory to remove entry from.
1886 * name - name of entry to remove.
1887 * cr - credentials of caller.
1888 * ct - caller context
1889 * flags - case flags
1891 * RETURN: 0 on success, error code on failure.
1895 * vp - ctime (if nlink > 0)
1898 uint64_t null_xattr = 0;
1902 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1905 znode_t *zp, *dzp = VTOZ(dvp);
1908 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1910 uint64_t acl_obj, xattr_obj;
1911 uint64_t xattr_obj_unlinked = 0;
1915 boolean_t may_delete_now, delete_now = FALSE;
1916 boolean_t unlinked, toobig = FALSE;
1918 pathname_t *realnmp = NULL;
1922 boolean_t waited = B_FALSE;
1926 zilog = zfsvfs->z_log;
1928 if (flags & FIGNORECASE) {
1938 * Attempt to lock directory; fail if entry doesn't exist.
1940 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1950 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1955 * Need to use rmdir for removing directories.
1957 if (vp->v_type == VDIR) {
1958 error = SET_ERROR(EPERM);
1962 vnevent_remove(vp, dvp, name, ct);
1965 dnlc_remove(dvp, realnmp->pn_buf);
1967 dnlc_remove(dvp, name);
1970 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1974 * We may delete the znode now, or we may put it in the unlinked set;
1975 * it depends on whether we're the last link, and on whether there are
1976 * other holds on the vnode. So we dmu_tx_hold() the right things to
1977 * allow for either case.
1980 tx = dmu_tx_create(zfsvfs->z_os);
1981 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1982 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1983 zfs_sa_upgrade_txholds(tx, zp);
1984 zfs_sa_upgrade_txholds(tx, dzp);
1985 if (may_delete_now) {
1987 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1988 /* if the file is too big, only hold_free a token amount */
1989 dmu_tx_hold_free(tx, zp->z_id, 0,
1990 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1993 /* are there any extended attributes? */
1994 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1995 &xattr_obj, sizeof (xattr_obj));
1996 if (error == 0 && xattr_obj) {
1997 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1999 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2000 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
2003 mutex_enter(&zp->z_lock);
2004 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
2005 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
2006 mutex_exit(&zp->z_lock);
2008 /* charge as an update -- would be nice not to charge at all */
2009 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2012 * Mark this transaction as typically resulting in a net free of
2013 * space, unless object removal will be delayed indefinitely
2014 * (due to active holds on the vnode due to the file being open).
2017 dmu_tx_mark_netfree(tx);
2019 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2021 zfs_dirent_unlock(dl);
2025 if (error == ERESTART) {
2039 * Remove the directory entry.
2041 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
2050 * Hold z_lock so that we can make sure that the ACL obj
2051 * hasn't changed. Could have been deleted due to
2054 mutex_enter(&zp->z_lock);
2056 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
2057 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
2058 delete_now = may_delete_now && !toobig &&
2059 vp->v_count == 1 && !vn_has_cached_data(vp) &&
2060 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
2067 panic("zfs_remove: delete_now branch taken");
2069 if (xattr_obj_unlinked) {
2070 ASSERT3U(xzp->z_links, ==, 2);
2071 mutex_enter(&xzp->z_lock);
2072 xzp->z_unlinked = 1;
2074 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
2075 &xzp->z_links, sizeof (xzp->z_links), tx);
2076 ASSERT3U(error, ==, 0);
2077 mutex_exit(&xzp->z_lock);
2078 zfs_unlinked_add(xzp, tx);
2081 error = sa_remove(zp->z_sa_hdl,
2082 SA_ZPL_XATTR(zfsvfs), tx);
2084 error = sa_update(zp->z_sa_hdl,
2085 SA_ZPL_XATTR(zfsvfs), &null_xattr,
2086 sizeof (uint64_t), tx);
2091 ASSERT0(vp->v_count);
2093 mutex_exit(&zp->z_lock);
2094 zfs_znode_delete(zp, tx);
2095 } else if (unlinked) {
2096 mutex_exit(&zp->z_lock);
2097 zfs_unlinked_add(zp, tx);
2099 vp->v_vflag |= VV_NOSYNC;
2104 if (flags & FIGNORECASE)
2106 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
2113 zfs_dirent_unlock(dl);
2120 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2121 zil_commit(zilog, 0);
2128 * Create a new directory and insert it into dvp using the name
2129 * provided. Return a pointer to the inserted directory.
2131 * IN: dvp - vnode of directory to add subdir to.
2132 * dirname - name of new directory.
2133 * vap - attributes of new directory.
2134 * cr - credentials of caller.
2135 * ct - caller context
2136 * flags - case flags
2137 * vsecp - ACL to be set
2139 * OUT: vpp - vnode of created directory.
2141 * RETURN: 0 on success, error code on failure.
2144 * dvp - ctime|mtime updated
2145 * vp - ctime|mtime|atime updated
2149 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
2150 caller_context_t *ct, int flags, vsecattr_t *vsecp)
2152 znode_t *zp, *dzp = VTOZ(dvp);
2153 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2162 gid_t gid = crgetgid(cr);
2163 zfs_acl_ids_t acl_ids;
2164 boolean_t fuid_dirtied;
2165 boolean_t waited = B_FALSE;
2167 ASSERT(vap->va_type == VDIR);
2170 * If we have an ephemeral id, ACL, or XVATTR then
2171 * make sure file system is at proper version
2174 ksid = crgetsid(cr, KSID_OWNER);
2176 uid = ksid_getid(ksid);
2179 if (zfsvfs->z_use_fuids == B_FALSE &&
2180 (vsecp || (vap->va_mask & AT_XVATTR) ||
2181 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
2182 return (SET_ERROR(EINVAL));
2186 zilog = zfsvfs->z_log;
2188 if (dzp->z_pflags & ZFS_XATTR) {
2190 return (SET_ERROR(EINVAL));
2193 if (zfsvfs->z_utf8 && u8_validate(dirname,
2194 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2196 return (SET_ERROR(EILSEQ));
2198 if (flags & FIGNORECASE)
2201 if (vap->va_mask & AT_XVATTR) {
2202 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
2203 crgetuid(cr), cr, vap->va_type)) != 0) {
2209 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
2210 vsecp, &acl_ids)) != 0) {
2215 getnewvnode_reserve(1);
2218 * First make sure the new directory doesn't exist.
2220 * Existence is checked first to make sure we don't return
2221 * EACCES instead of EEXIST which can cause some applications
2227 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
2229 zfs_acl_ids_free(&acl_ids);
2230 getnewvnode_drop_reserve();
2235 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
2236 zfs_acl_ids_free(&acl_ids);
2237 zfs_dirent_unlock(dl);
2238 getnewvnode_drop_reserve();
2243 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
2244 zfs_acl_ids_free(&acl_ids);
2245 zfs_dirent_unlock(dl);
2246 getnewvnode_drop_reserve();
2248 return (SET_ERROR(EDQUOT));
2252 * Add a new entry to the directory.
2254 tx = dmu_tx_create(zfsvfs->z_os);
2255 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2256 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2257 fuid_dirtied = zfsvfs->z_fuid_dirty;
2259 zfs_fuid_txhold(zfsvfs, tx);
2260 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2261 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2262 acl_ids.z_aclp->z_acl_bytes);
2265 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2266 ZFS_SA_BASE_ATTR_SIZE);
2268 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2270 zfs_dirent_unlock(dl);
2271 if (error == ERESTART) {
2277 zfs_acl_ids_free(&acl_ids);
2279 getnewvnode_drop_reserve();
2287 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2290 zfs_fuid_sync(zfsvfs, tx);
2293 * Now put new name in parent dir.
2295 (void) zfs_link_create(dl, zp, tx, ZNEW);
2299 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2300 if (flags & FIGNORECASE)
2302 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2303 acl_ids.z_fuidp, vap);
2305 zfs_acl_ids_free(&acl_ids);
2309 getnewvnode_drop_reserve();
2311 zfs_dirent_unlock(dl);
2313 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2314 zil_commit(zilog, 0);
2321 * Remove a directory subdir entry. If the current working
2322 * directory is the same as the subdir to be removed, the
2325 * IN: dvp - vnode of directory to remove from.
2326 * name - name of directory to be removed.
2327 * cwd - vnode of current working directory.
2328 * cr - credentials of caller.
2329 * ct - caller context
2330 * flags - case flags
2332 * RETURN: 0 on success, error code on failure.
2335 * dvp - ctime|mtime updated
2339 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2340 caller_context_t *ct, int flags)
2342 znode_t *dzp = VTOZ(dvp);
2345 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2351 boolean_t waited = B_FALSE;
2355 zilog = zfsvfs->z_log;
2357 if (flags & FIGNORECASE)
2363 * Attempt to lock directory; fail if entry doesn't exist.
2365 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2373 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2377 if (vp->v_type != VDIR) {
2378 error = SET_ERROR(ENOTDIR);
2383 error = SET_ERROR(EINVAL);
2387 vnevent_rmdir(vp, dvp, name, ct);
2390 * Grab a lock on the directory to make sure that noone is
2391 * trying to add (or lookup) entries while we are removing it.
2393 rw_enter(&zp->z_name_lock, RW_WRITER);
2396 * Grab a lock on the parent pointer to make sure we play well
2397 * with the treewalk and directory rename code.
2399 rw_enter(&zp->z_parent_lock, RW_WRITER);
2401 tx = dmu_tx_create(zfsvfs->z_os);
2402 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2403 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2404 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2405 zfs_sa_upgrade_txholds(tx, zp);
2406 zfs_sa_upgrade_txholds(tx, dzp);
2407 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2409 rw_exit(&zp->z_parent_lock);
2410 rw_exit(&zp->z_name_lock);
2411 zfs_dirent_unlock(dl);
2413 if (error == ERESTART) {
2424 #ifdef FREEBSD_NAMECACHE
2428 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2431 uint64_t txtype = TX_RMDIR;
2432 if (flags & FIGNORECASE)
2434 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2439 rw_exit(&zp->z_parent_lock);
2440 rw_exit(&zp->z_name_lock);
2441 #ifdef FREEBSD_NAMECACHE
2445 zfs_dirent_unlock(dl);
2449 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2450 zil_commit(zilog, 0);
2457 * Read as many directory entries as will fit into the provided
2458 * buffer from the given directory cursor position (specified in
2459 * the uio structure).
2461 * IN: vp - vnode of directory to read.
2462 * uio - structure supplying read location, range info,
2463 * and return buffer.
2464 * cr - credentials of caller.
2465 * ct - caller context
2466 * flags - case flags
2468 * OUT: uio - updated offset and range, buffer filled.
2469 * eofp - set to true if end-of-file detected.
2471 * RETURN: 0 on success, error code on failure.
2474 * vp - atime updated
2476 * Note that the low 4 bits of the cookie returned by zap is always zero.
2477 * This allows us to use the low range for "special" directory entries:
2478 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2479 * we use the offset 2 for the '.zfs' directory.
2483 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies)
2485 znode_t *zp = VTOZ(vp);
2489 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2494 zap_attribute_t zap;
2495 uint_t bytes_wanted;
2496 uint64_t offset; /* must be unsigned; checks for < 1 */
2502 boolean_t check_sysattrs;
2505 u_long *cooks = NULL;
2511 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2512 &parent, sizeof (parent))) != 0) {
2518 * If we are not given an eof variable,
2525 * Check for valid iov_len.
2527 if (uio->uio_iov->iov_len <= 0) {
2529 return (SET_ERROR(EINVAL));
2533 * Quit if directory has been removed (posix)
2535 if ((*eofp = zp->z_unlinked) != 0) {
2542 offset = uio->uio_loffset;
2543 prefetch = zp->z_zn_prefetch;
2546 * Initialize the iterator cursor.
2550 * Start iteration from the beginning of the directory.
2552 zap_cursor_init(&zc, os, zp->z_id);
2555 * The offset is a serialized cursor.
2557 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2561 * Get space to change directory entries into fs independent format.
2563 iovp = uio->uio_iov;
2564 bytes_wanted = iovp->iov_len;
2565 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2566 bufsize = bytes_wanted;
2567 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2568 odp = (struct dirent64 *)outbuf;
2570 bufsize = bytes_wanted;
2572 odp = (struct dirent64 *)iovp->iov_base;
2574 eodp = (struct edirent *)odp;
2576 if (ncookies != NULL) {
2578 * Minimum entry size is dirent size and 1 byte for a file name.
2580 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1);
2581 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK);
2586 * If this VFS supports the system attribute view interface; and
2587 * we're looking at an extended attribute directory; and we care
2588 * about normalization conflicts on this vfs; then we must check
2589 * for normalization conflicts with the sysattr name space.
2592 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2593 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2594 (flags & V_RDDIR_ENTFLAGS);
2600 * Transform to file-system independent format
2603 while (outcount < bytes_wanted) {
2606 off64_t *next = NULL;
2609 * Special case `.', `..', and `.zfs'.
2612 (void) strcpy(zap.za_name, ".");
2613 zap.za_normalization_conflict = 0;
2616 } else if (offset == 1) {
2617 (void) strcpy(zap.za_name, "..");
2618 zap.za_normalization_conflict = 0;
2621 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2622 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2623 zap.za_normalization_conflict = 0;
2624 objnum = ZFSCTL_INO_ROOT;
2630 if (error = zap_cursor_retrieve(&zc, &zap)) {
2631 if ((*eofp = (error == ENOENT)) != 0)
2637 if (zap.za_integer_length != 8 ||
2638 zap.za_num_integers != 1) {
2639 cmn_err(CE_WARN, "zap_readdir: bad directory "
2640 "entry, obj = %lld, offset = %lld\n",
2641 (u_longlong_t)zp->z_id,
2642 (u_longlong_t)offset);
2643 error = SET_ERROR(ENXIO);
2647 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2649 * MacOS X can extract the object type here such as:
2650 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2652 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2654 if (check_sysattrs && !zap.za_normalization_conflict) {
2656 zap.za_normalization_conflict =
2657 xattr_sysattr_casechk(zap.za_name);
2659 panic("%s:%u: TODO", __func__, __LINE__);
2664 if (flags & V_RDDIR_ACCFILTER) {
2666 * If we have no access at all, don't include
2667 * this entry in the returned information
2670 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2672 if (!zfs_has_access(ezp, cr)) {
2679 if (flags & V_RDDIR_ENTFLAGS)
2680 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2682 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2685 * Will this entry fit in the buffer?
2687 if (outcount + reclen > bufsize) {
2689 * Did we manage to fit anything in the buffer?
2692 error = SET_ERROR(EINVAL);
2697 if (flags & V_RDDIR_ENTFLAGS) {
2699 * Add extended flag entry:
2701 eodp->ed_ino = objnum;
2702 eodp->ed_reclen = reclen;
2703 /* NOTE: ed_off is the offset for the *next* entry */
2704 next = &(eodp->ed_off);
2705 eodp->ed_eflags = zap.za_normalization_conflict ?
2706 ED_CASE_CONFLICT : 0;
2707 (void) strncpy(eodp->ed_name, zap.za_name,
2708 EDIRENT_NAMELEN(reclen));
2709 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2714 odp->d_ino = objnum;
2715 odp->d_reclen = reclen;
2716 odp->d_namlen = strlen(zap.za_name);
2717 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1);
2719 odp = (dirent64_t *)((intptr_t)odp + reclen);
2723 ASSERT(outcount <= bufsize);
2725 /* Prefetch znode */
2727 dmu_prefetch(os, objnum, 0, 0, 0,
2728 ZIO_PRIORITY_SYNC_READ);
2732 * Move to the next entry, fill in the previous offset.
2734 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2735 zap_cursor_advance(&zc);
2736 offset = zap_cursor_serialize(&zc);
2741 if (cooks != NULL) {
2744 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks));
2747 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2749 /* Subtract unused cookies */
2750 if (ncookies != NULL)
2751 *ncookies -= ncooks;
2753 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2754 iovp->iov_base += outcount;
2755 iovp->iov_len -= outcount;
2756 uio->uio_resid -= outcount;
2757 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2759 * Reset the pointer.
2761 offset = uio->uio_loffset;
2765 zap_cursor_fini(&zc);
2766 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2767 kmem_free(outbuf, bufsize);
2769 if (error == ENOENT)
2772 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2774 uio->uio_loffset = offset;
2776 if (error != 0 && cookies != NULL) {
2777 free(*cookies, M_TEMP);
2784 ulong_t zfs_fsync_sync_cnt = 4;
2787 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2789 znode_t *zp = VTOZ(vp);
2790 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2792 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2794 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2797 zil_commit(zfsvfs->z_log, zp->z_id);
2805 * Get the requested file attributes and place them in the provided
2808 * IN: vp - vnode of file.
2809 * vap - va_mask identifies requested attributes.
2810 * If AT_XVATTR set, then optional attrs are requested
2811 * flags - ATTR_NOACLCHECK (CIFS server context)
2812 * cr - credentials of caller.
2813 * ct - caller context
2815 * OUT: vap - attribute values.
2817 * RETURN: 0 (always succeeds).
2821 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2822 caller_context_t *ct)
2824 znode_t *zp = VTOZ(vp);
2825 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2828 u_longlong_t nblocks;
2830 uint64_t mtime[2], ctime[2], crtime[2], rdev;
2831 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2832 xoptattr_t *xoap = NULL;
2833 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2834 sa_bulk_attr_t bulk[4];
2840 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2842 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2843 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2844 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
2845 if (vp->v_type == VBLK || vp->v_type == VCHR)
2846 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
2849 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2855 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2856 * Also, if we are the owner don't bother, since owner should
2857 * always be allowed to read basic attributes of file.
2859 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2860 (vap->va_uid != crgetuid(cr))) {
2861 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2869 * Return all attributes. It's cheaper to provide the answer
2870 * than to determine whether we were asked the question.
2873 mutex_enter(&zp->z_lock);
2874 vap->va_type = IFTOVT(zp->z_mode);
2875 vap->va_mode = zp->z_mode & ~S_IFMT;
2877 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2879 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
2881 vap->va_nodeid = zp->z_id;
2882 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2883 links = zp->z_links + 1;
2885 links = zp->z_links;
2886 vap->va_nlink = MIN(links, LINK_MAX); /* nlink_t limit! */
2887 vap->va_size = zp->z_size;
2889 vap->va_rdev = vp->v_rdev;
2891 if (vp->v_type == VBLK || vp->v_type == VCHR)
2892 vap->va_rdev = zfs_cmpldev(rdev);
2894 vap->va_seq = zp->z_seq;
2895 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */
2896 vap->va_filerev = zp->z_seq;
2899 * Add in any requested optional attributes and the create time.
2900 * Also set the corresponding bits in the returned attribute bitmap.
2902 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2903 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2905 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2906 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2909 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2910 xoap->xoa_readonly =
2911 ((zp->z_pflags & ZFS_READONLY) != 0);
2912 XVA_SET_RTN(xvap, XAT_READONLY);
2915 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2917 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2918 XVA_SET_RTN(xvap, XAT_SYSTEM);
2921 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2923 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2924 XVA_SET_RTN(xvap, XAT_HIDDEN);
2927 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2928 xoap->xoa_nounlink =
2929 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2930 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2933 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2934 xoap->xoa_immutable =
2935 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2936 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2939 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2940 xoap->xoa_appendonly =
2941 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2942 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2945 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2947 ((zp->z_pflags & ZFS_NODUMP) != 0);
2948 XVA_SET_RTN(xvap, XAT_NODUMP);
2951 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2953 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2954 XVA_SET_RTN(xvap, XAT_OPAQUE);
2957 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2958 xoap->xoa_av_quarantined =
2959 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2960 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2963 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2964 xoap->xoa_av_modified =
2965 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2966 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2969 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2970 vp->v_type == VREG) {
2971 zfs_sa_get_scanstamp(zp, xvap);
2974 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2977 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2978 times, sizeof (times));
2979 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2980 XVA_SET_RTN(xvap, XAT_CREATETIME);
2983 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2984 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2985 XVA_SET_RTN(xvap, XAT_REPARSE);
2987 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2988 xoap->xoa_generation = zp->z_gen;
2989 XVA_SET_RTN(xvap, XAT_GEN);
2992 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2994 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2995 XVA_SET_RTN(xvap, XAT_OFFLINE);
2998 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
3000 ((zp->z_pflags & ZFS_SPARSE) != 0);
3001 XVA_SET_RTN(xvap, XAT_SPARSE);
3005 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
3006 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
3007 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
3008 ZFS_TIME_DECODE(&vap->va_birthtime, crtime);
3010 mutex_exit(&zp->z_lock);
3012 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
3013 vap->va_blksize = blksize;
3014 vap->va_bytes = nblocks << 9; /* nblocks * 512 */
3016 if (zp->z_blksz == 0) {
3018 * Block size hasn't been set; suggest maximal I/O transfers.
3020 vap->va_blksize = zfsvfs->z_max_blksz;
3028 * Set the file attributes to the values contained in the
3031 * IN: vp - vnode of file to be modified.
3032 * vap - new attribute values.
3033 * If AT_XVATTR set, then optional attrs are being set
3034 * flags - ATTR_UTIME set if non-default time values provided.
3035 * - ATTR_NOACLCHECK (CIFS context only).
3036 * cr - credentials of caller.
3037 * ct - caller context
3039 * RETURN: 0 on success, error code on failure.
3042 * vp - ctime updated, mtime updated if size changed.
3046 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
3047 caller_context_t *ct)
3049 znode_t *zp = VTOZ(vp);
3050 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3055 uint_t mask = vap->va_mask;
3056 uint_t saved_mask = 0;
3057 uint64_t saved_mode;
3060 uint64_t new_uid, new_gid;
3062 uint64_t mtime[2], ctime[2];
3064 int need_policy = FALSE;
3066 zfs_fuid_info_t *fuidp = NULL;
3067 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
3070 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
3071 boolean_t fuid_dirtied = B_FALSE;
3072 sa_bulk_attr_t bulk[7], xattr_bulk[7];
3073 int count = 0, xattr_count = 0;
3078 if (mask & AT_NOSET)
3079 return (SET_ERROR(EINVAL));
3084 zilog = zfsvfs->z_log;
3087 * Make sure that if we have ephemeral uid/gid or xvattr specified
3088 * that file system is at proper version level
3091 if (zfsvfs->z_use_fuids == B_FALSE &&
3092 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
3093 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
3094 (mask & AT_XVATTR))) {
3096 return (SET_ERROR(EINVAL));
3099 if (mask & AT_SIZE && vp->v_type == VDIR) {
3101 return (SET_ERROR(EISDIR));
3104 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
3106 return (SET_ERROR(EINVAL));
3110 * If this is an xvattr_t, then get a pointer to the structure of
3111 * optional attributes. If this is NULL, then we have a vattr_t.
3113 xoap = xva_getxoptattr(xvap);
3115 xva_init(&tmpxvattr);
3118 * Immutable files can only alter immutable bit and atime
3120 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
3121 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
3122 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
3124 return (SET_ERROR(EPERM));
3127 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
3129 return (SET_ERROR(EPERM));
3133 * Verify timestamps doesn't overflow 32 bits.
3134 * ZFS can handle large timestamps, but 32bit syscalls can't
3135 * handle times greater than 2039. This check should be removed
3136 * once large timestamps are fully supported.
3138 if (mask & (AT_ATIME | AT_MTIME)) {
3139 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
3140 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
3142 return (SET_ERROR(EOVERFLOW));
3150 /* Can this be moved to before the top label? */
3151 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
3153 return (SET_ERROR(EROFS));
3157 * First validate permissions
3160 if (mask & AT_SIZE) {
3162 * XXX - Note, we are not providing any open
3163 * mode flags here (like FNDELAY), so we may
3164 * block if there are locks present... this
3165 * should be addressed in openat().
3167 /* XXX - would it be OK to generate a log record here? */
3168 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
3175 if (mask & (AT_ATIME|AT_MTIME) ||
3176 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
3177 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
3178 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
3179 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
3180 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
3181 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
3182 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
3183 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
3187 if (mask & (AT_UID|AT_GID)) {
3188 int idmask = (mask & (AT_UID|AT_GID));
3193 * NOTE: even if a new mode is being set,
3194 * we may clear S_ISUID/S_ISGID bits.
3197 if (!(mask & AT_MODE))
3198 vap->va_mode = zp->z_mode;
3201 * Take ownership or chgrp to group we are a member of
3204 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
3205 take_group = (mask & AT_GID) &&
3206 zfs_groupmember(zfsvfs, vap->va_gid, cr);
3209 * If both AT_UID and AT_GID are set then take_owner and
3210 * take_group must both be set in order to allow taking
3213 * Otherwise, send the check through secpolicy_vnode_setattr()
3217 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
3218 ((idmask == AT_UID) && take_owner) ||
3219 ((idmask == AT_GID) && take_group)) {
3220 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
3221 skipaclchk, cr) == 0) {
3223 * Remove setuid/setgid for non-privileged users
3225 secpolicy_setid_clear(vap, vp, cr);
3226 trim_mask = (mask & (AT_UID|AT_GID));
3235 mutex_enter(&zp->z_lock);
3236 oldva.va_mode = zp->z_mode;
3237 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
3238 if (mask & AT_XVATTR) {
3240 * Update xvattr mask to include only those attributes
3241 * that are actually changing.
3243 * the bits will be restored prior to actually setting
3244 * the attributes so the caller thinks they were set.
3246 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
3247 if (xoap->xoa_appendonly !=
3248 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
3251 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
3252 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
3256 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
3257 if (xoap->xoa_nounlink !=
3258 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
3261 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
3262 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
3266 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
3267 if (xoap->xoa_immutable !=
3268 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
3271 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
3272 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
3276 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
3277 if (xoap->xoa_nodump !=
3278 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
3281 XVA_CLR_REQ(xvap, XAT_NODUMP);
3282 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
3286 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3287 if (xoap->xoa_av_modified !=
3288 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3291 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3292 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
3296 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3297 if ((vp->v_type != VREG &&
3298 xoap->xoa_av_quarantined) ||
3299 xoap->xoa_av_quarantined !=
3300 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3303 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3304 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
3308 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3309 mutex_exit(&zp->z_lock);
3311 return (SET_ERROR(EPERM));
3314 if (need_policy == FALSE &&
3315 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3316 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3321 mutex_exit(&zp->z_lock);
3323 if (mask & AT_MODE) {
3324 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3325 err = secpolicy_setid_setsticky_clear(vp, vap,
3331 trim_mask |= AT_MODE;
3339 * If trim_mask is set then take ownership
3340 * has been granted or write_acl is present and user
3341 * has the ability to modify mode. In that case remove
3342 * UID|GID and or MODE from mask so that
3343 * secpolicy_vnode_setattr() doesn't revoke it.
3347 saved_mask = vap->va_mask;
3348 vap->va_mask &= ~trim_mask;
3349 if (trim_mask & AT_MODE) {
3351 * Save the mode, as secpolicy_vnode_setattr()
3352 * will overwrite it with ova.va_mode.
3354 saved_mode = vap->va_mode;
3357 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3358 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3365 vap->va_mask |= saved_mask;
3366 if (trim_mask & AT_MODE) {
3368 * Recover the mode after
3369 * secpolicy_vnode_setattr().
3371 vap->va_mode = saved_mode;
3377 * secpolicy_vnode_setattr, or take ownership may have
3380 mask = vap->va_mask;
3382 if ((mask & (AT_UID | AT_GID))) {
3383 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3384 &xattr_obj, sizeof (xattr_obj));
3386 if (err == 0 && xattr_obj) {
3387 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3391 if (mask & AT_UID) {
3392 new_uid = zfs_fuid_create(zfsvfs,
3393 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3394 if (new_uid != zp->z_uid &&
3395 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3397 VN_RELE(ZTOV(attrzp));
3398 err = SET_ERROR(EDQUOT);
3403 if (mask & AT_GID) {
3404 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3405 cr, ZFS_GROUP, &fuidp);
3406 if (new_gid != zp->z_gid &&
3407 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3409 VN_RELE(ZTOV(attrzp));
3410 err = SET_ERROR(EDQUOT);
3415 tx = dmu_tx_create(zfsvfs->z_os);
3417 if (mask & AT_MODE) {
3418 uint64_t pmode = zp->z_mode;
3420 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3422 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3423 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3424 err = SET_ERROR(EPERM);
3428 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3431 mutex_enter(&zp->z_lock);
3432 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3434 * Are we upgrading ACL from old V0 format
3437 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3438 zfs_znode_acl_version(zp) ==
3439 ZFS_ACL_VERSION_INITIAL) {
3440 dmu_tx_hold_free(tx, acl_obj, 0,
3442 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3443 0, aclp->z_acl_bytes);
3445 dmu_tx_hold_write(tx, acl_obj, 0,
3448 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3449 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3450 0, aclp->z_acl_bytes);
3452 mutex_exit(&zp->z_lock);
3453 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3455 if ((mask & AT_XVATTR) &&
3456 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3457 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3459 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3463 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3466 fuid_dirtied = zfsvfs->z_fuid_dirty;
3468 zfs_fuid_txhold(zfsvfs, tx);
3470 zfs_sa_upgrade_txholds(tx, zp);
3472 err = dmu_tx_assign(tx, TXG_WAIT);
3478 * Set each attribute requested.
3479 * We group settings according to the locks they need to acquire.
3481 * Note: you cannot set ctime directly, although it will be
3482 * updated as a side-effect of calling this function.
3486 if (mask & (AT_UID|AT_GID|AT_MODE))
3487 mutex_enter(&zp->z_acl_lock);
3488 mutex_enter(&zp->z_lock);
3490 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3491 &zp->z_pflags, sizeof (zp->z_pflags));
3494 if (mask & (AT_UID|AT_GID|AT_MODE))
3495 mutex_enter(&attrzp->z_acl_lock);
3496 mutex_enter(&attrzp->z_lock);
3497 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3498 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3499 sizeof (attrzp->z_pflags));
3502 if (mask & (AT_UID|AT_GID)) {
3504 if (mask & AT_UID) {
3505 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3506 &new_uid, sizeof (new_uid));
3507 zp->z_uid = new_uid;
3509 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3510 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3512 attrzp->z_uid = new_uid;
3516 if (mask & AT_GID) {
3517 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3518 NULL, &new_gid, sizeof (new_gid));
3519 zp->z_gid = new_gid;
3521 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3522 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3524 attrzp->z_gid = new_gid;
3527 if (!(mask & AT_MODE)) {
3528 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3529 NULL, &new_mode, sizeof (new_mode));
3530 new_mode = zp->z_mode;
3532 err = zfs_acl_chown_setattr(zp);
3535 err = zfs_acl_chown_setattr(attrzp);
3540 if (mask & AT_MODE) {
3541 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3542 &new_mode, sizeof (new_mode));
3543 zp->z_mode = new_mode;
3544 ASSERT3U((uintptr_t)aclp, !=, 0);
3545 err = zfs_aclset_common(zp, aclp, cr, tx);
3547 if (zp->z_acl_cached)
3548 zfs_acl_free(zp->z_acl_cached);
3549 zp->z_acl_cached = aclp;
3554 if (mask & AT_ATIME) {
3555 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3556 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3557 &zp->z_atime, sizeof (zp->z_atime));
3560 if (mask & AT_MTIME) {
3561 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3562 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3563 mtime, sizeof (mtime));
3566 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3567 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3568 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3569 NULL, mtime, sizeof (mtime));
3570 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3571 &ctime, sizeof (ctime));
3572 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3574 } else if (mask != 0) {
3575 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3576 &ctime, sizeof (ctime));
3577 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3580 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3581 SA_ZPL_CTIME(zfsvfs), NULL,
3582 &ctime, sizeof (ctime));
3583 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3584 mtime, ctime, B_TRUE);
3588 * Do this after setting timestamps to prevent timestamp
3589 * update from toggling bit
3592 if (xoap && (mask & AT_XVATTR)) {
3595 * restore trimmed off masks
3596 * so that return masks can be set for caller.
3599 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3600 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3602 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3603 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3605 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3606 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3608 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3609 XVA_SET_REQ(xvap, XAT_NODUMP);
3611 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3612 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3614 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3615 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3618 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3619 ASSERT(vp->v_type == VREG);
3621 zfs_xvattr_set(zp, xvap, tx);
3625 zfs_fuid_sync(zfsvfs, tx);
3628 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3630 mutex_exit(&zp->z_lock);
3631 if (mask & (AT_UID|AT_GID|AT_MODE))
3632 mutex_exit(&zp->z_acl_lock);
3635 if (mask & (AT_UID|AT_GID|AT_MODE))
3636 mutex_exit(&attrzp->z_acl_lock);
3637 mutex_exit(&attrzp->z_lock);
3640 if (err == 0 && attrzp) {
3641 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3647 VN_RELE(ZTOV(attrzp));
3653 zfs_fuid_info_free(fuidp);
3659 if (err == ERESTART)
3662 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3667 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3668 zil_commit(zilog, 0);
3674 typedef struct zfs_zlock {
3675 krwlock_t *zl_rwlock; /* lock we acquired */
3676 znode_t *zl_znode; /* znode we held */
3677 struct zfs_zlock *zl_next; /* next in list */
3681 * Drop locks and release vnodes that were held by zfs_rename_lock().
3684 zfs_rename_unlock(zfs_zlock_t **zlpp)
3688 while ((zl = *zlpp) != NULL) {
3689 if (zl->zl_znode != NULL)
3690 VN_RELE(ZTOV(zl->zl_znode));
3691 rw_exit(zl->zl_rwlock);
3692 *zlpp = zl->zl_next;
3693 kmem_free(zl, sizeof (*zl));
3698 * Search back through the directory tree, using the ".." entries.
3699 * Lock each directory in the chain to prevent concurrent renames.
3700 * Fail any attempt to move a directory into one of its own descendants.
3701 * XXX - z_parent_lock can overlap with map or grow locks
3704 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3708 uint64_t rootid = zp->z_zfsvfs->z_root;
3709 uint64_t oidp = zp->z_id;
3710 krwlock_t *rwlp = &szp->z_parent_lock;
3711 krw_t rw = RW_WRITER;
3714 * First pass write-locks szp and compares to zp->z_id.
3715 * Later passes read-lock zp and compare to zp->z_parent.
3718 if (!rw_tryenter(rwlp, rw)) {
3720 * Another thread is renaming in this path.
3721 * Note that if we are a WRITER, we don't have any
3722 * parent_locks held yet.
3724 if (rw == RW_READER && zp->z_id > szp->z_id) {
3726 * Drop our locks and restart
3728 zfs_rename_unlock(&zl);
3732 rwlp = &szp->z_parent_lock;
3737 * Wait for other thread to drop its locks
3743 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3744 zl->zl_rwlock = rwlp;
3745 zl->zl_znode = NULL;
3746 zl->zl_next = *zlpp;
3749 if (oidp == szp->z_id) /* We're a descendant of szp */
3750 return (SET_ERROR(EINVAL));
3752 if (oidp == rootid) /* We've hit the top */
3755 if (rw == RW_READER) { /* i.e. not the first pass */
3756 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3761 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3762 &oidp, sizeof (oidp));
3763 rwlp = &zp->z_parent_lock;
3766 } while (zp->z_id != sdzp->z_id);
3772 * Move an entry from the provided source directory to the target
3773 * directory. Change the entry name as indicated.
3775 * IN: sdvp - Source directory containing the "old entry".
3776 * snm - Old entry name.
3777 * tdvp - Target directory to contain the "new entry".
3778 * tnm - New entry name.
3779 * cr - credentials of caller.
3780 * ct - caller context
3781 * flags - case flags
3783 * RETURN: 0 on success, error code on failure.
3786 * sdvp,tdvp - ctime|mtime updated
3790 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3791 caller_context_t *ct, int flags)
3793 znode_t *tdzp, *sdzp, *szp, *tzp;
3797 zfs_dirlock_t *sdl, *tdl;
3800 int cmp, serr, terr;
3803 boolean_t waited = B_FALSE;
3806 ZFS_VERIFY_ZP(tdzp);
3807 zfsvfs = tdzp->z_zfsvfs;
3809 zilog = zfsvfs->z_log;
3813 * In case sdzp is not valid, let's be sure to exit from the right
3816 if (sdzp->z_sa_hdl == NULL) {
3818 return (SET_ERROR(EIO));
3822 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3823 * ctldir appear to have the same v_vfsp.
3825 if (sdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) {
3827 return (SET_ERROR(EXDEV));
3830 if (zfsvfs->z_utf8 && u8_validate(tnm,
3831 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3833 return (SET_ERROR(EILSEQ));
3836 if (flags & FIGNORECASE)
3845 * This is to prevent the creation of links into attribute space
3846 * by renaming a linked file into/outof an attribute directory.
3847 * See the comment in zfs_link() for why this is considered bad.
3849 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3851 return (SET_ERROR(EINVAL));
3855 * Lock source and target directory entries. To prevent deadlock,
3856 * a lock ordering must be defined. We lock the directory with
3857 * the smallest object id first, or if it's a tie, the one with
3858 * the lexically first name.
3860 if (sdzp->z_id < tdzp->z_id) {
3862 } else if (sdzp->z_id > tdzp->z_id) {
3866 * First compare the two name arguments without
3867 * considering any case folding.
3869 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3871 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3872 ASSERT(error == 0 || !zfsvfs->z_utf8);
3875 * POSIX: "If the old argument and the new argument
3876 * both refer to links to the same existing file,
3877 * the rename() function shall return successfully
3878 * and perform no other action."
3884 * If the file system is case-folding, then we may
3885 * have some more checking to do. A case-folding file
3886 * system is either supporting mixed case sensitivity
3887 * access or is completely case-insensitive. Note
3888 * that the file system is always case preserving.
3890 * In mixed sensitivity mode case sensitive behavior
3891 * is the default. FIGNORECASE must be used to
3892 * explicitly request case insensitive behavior.
3894 * If the source and target names provided differ only
3895 * by case (e.g., a request to rename 'tim' to 'Tim'),
3896 * we will treat this as a special case in the
3897 * case-insensitive mode: as long as the source name
3898 * is an exact match, we will allow this to proceed as
3899 * a name-change request.
3901 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3902 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3903 flags & FIGNORECASE)) &&
3904 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3907 * case preserving rename request, require exact
3916 * If the source and destination directories are the same, we should
3917 * grab the z_name_lock of that directory only once.
3921 rw_enter(&sdzp->z_name_lock, RW_READER);
3925 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3926 ZEXISTS | zflg, NULL, NULL);
3927 terr = zfs_dirent_lock(&tdl,
3928 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3930 terr = zfs_dirent_lock(&tdl,
3931 tdzp, tnm, &tzp, zflg, NULL, NULL);
3932 serr = zfs_dirent_lock(&sdl,
3933 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3939 * Source entry invalid or not there.
3942 zfs_dirent_unlock(tdl);
3948 rw_exit(&sdzp->z_name_lock);
3951 * FreeBSD: In OpenSolaris they only check if rename source is
3952 * ".." here, because "." is handled in their lookup. This is
3953 * not the case for FreeBSD, so we check for "." explicitly.
3955 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0)
3956 serr = SET_ERROR(EINVAL);
3961 zfs_dirent_unlock(sdl);
3965 rw_exit(&sdzp->z_name_lock);
3967 if (strcmp(tnm, "..") == 0)
3968 terr = SET_ERROR(EINVAL);
3974 * Must have write access at the source to remove the old entry
3975 * and write access at the target to create the new entry.
3976 * Note that if target and source are the same, this can be
3977 * done in a single check.
3980 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3983 if (ZTOV(szp)->v_type == VDIR) {
3985 * Check to make sure rename is valid.
3986 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3988 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3993 * Does target exist?
3997 * Source and target must be the same type.
3999 if (ZTOV(szp)->v_type == VDIR) {
4000 if (ZTOV(tzp)->v_type != VDIR) {
4001 error = SET_ERROR(ENOTDIR);
4005 if (ZTOV(tzp)->v_type == VDIR) {
4006 error = SET_ERROR(EISDIR);
4011 * POSIX dictates that when the source and target
4012 * entries refer to the same file object, rename
4013 * must do nothing and exit without error.
4015 if (szp->z_id == tzp->z_id) {
4021 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
4023 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
4026 * notify the target directory if it is not the same
4027 * as source directory.
4030 vnevent_rename_dest_dir(tdvp, ct);
4033 tx = dmu_tx_create(zfsvfs->z_os);
4034 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4035 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
4036 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
4037 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
4039 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
4040 zfs_sa_upgrade_txholds(tx, tdzp);
4043 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
4044 zfs_sa_upgrade_txholds(tx, tzp);
4047 zfs_sa_upgrade_txholds(tx, szp);
4048 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
4049 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4052 zfs_rename_unlock(&zl);
4053 zfs_dirent_unlock(sdl);
4054 zfs_dirent_unlock(tdl);
4057 rw_exit(&sdzp->z_name_lock);
4062 if (error == ERESTART) {
4073 if (tzp) /* Attempt to remove the existing target */
4074 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
4077 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
4079 szp->z_pflags |= ZFS_AV_MODIFIED;
4081 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
4082 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
4085 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
4087 zfs_log_rename(zilog, tx, TX_RENAME |
4088 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
4089 sdl->dl_name, tdzp, tdl->dl_name, szp);
4092 * Update path information for the target vnode
4094 vn_renamepath(tdvp, ZTOV(szp), tnm,
4098 * At this point, we have successfully created
4099 * the target name, but have failed to remove
4100 * the source name. Since the create was done
4101 * with the ZRENAMING flag, there are
4102 * complications; for one, the link count is
4103 * wrong. The easiest way to deal with this
4104 * is to remove the newly created target, and
4105 * return the original error. This must
4106 * succeed; fortunately, it is very unlikely to
4107 * fail, since we just created it.
4109 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
4110 ZRENAMING, NULL), ==, 0);
4113 #ifdef FREEBSD_NAMECACHE
4117 cache_purge(ZTOV(szp));
4119 cache_purge(ZTOV(tzp));
4127 zfs_rename_unlock(&zl);
4129 zfs_dirent_unlock(sdl);
4130 zfs_dirent_unlock(tdl);
4133 rw_exit(&sdzp->z_name_lock);
4140 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4141 zil_commit(zilog, 0);
4149 * Insert the indicated symbolic reference entry into the directory.
4151 * IN: dvp - Directory to contain new symbolic link.
4152 * link - Name for new symlink entry.
4153 * vap - Attributes of new entry.
4154 * cr - credentials of caller.
4155 * ct - caller context
4156 * flags - case flags
4158 * RETURN: 0 on success, error code on failure.
4161 * dvp - ctime|mtime updated
4165 zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link,
4166 cred_t *cr, kthread_t *td)
4168 znode_t *zp, *dzp = VTOZ(dvp);
4171 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4173 uint64_t len = strlen(link);
4176 zfs_acl_ids_t acl_ids;
4177 boolean_t fuid_dirtied;
4178 uint64_t txtype = TX_SYMLINK;
4179 boolean_t waited = B_FALSE;
4182 ASSERT(vap->va_type == VLNK);
4186 zilog = zfsvfs->z_log;
4188 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
4189 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4191 return (SET_ERROR(EILSEQ));
4193 if (flags & FIGNORECASE)
4196 if (len > MAXPATHLEN) {
4198 return (SET_ERROR(ENAMETOOLONG));
4201 if ((error = zfs_acl_ids_create(dzp, 0,
4202 vap, cr, NULL, &acl_ids)) != 0) {
4207 getnewvnode_reserve(1);
4211 * Attempt to lock directory; fail if entry already exists.
4213 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
4215 zfs_acl_ids_free(&acl_ids);
4216 getnewvnode_drop_reserve();
4221 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4222 zfs_acl_ids_free(&acl_ids);
4223 zfs_dirent_unlock(dl);
4224 getnewvnode_drop_reserve();
4229 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
4230 zfs_acl_ids_free(&acl_ids);
4231 zfs_dirent_unlock(dl);
4232 getnewvnode_drop_reserve();
4234 return (SET_ERROR(EDQUOT));
4236 tx = dmu_tx_create(zfsvfs->z_os);
4237 fuid_dirtied = zfsvfs->z_fuid_dirty;
4238 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
4239 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4240 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
4241 ZFS_SA_BASE_ATTR_SIZE + len);
4242 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
4243 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
4244 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
4245 acl_ids.z_aclp->z_acl_bytes);
4248 zfs_fuid_txhold(zfsvfs, tx);
4249 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4251 zfs_dirent_unlock(dl);
4252 if (error == ERESTART) {
4258 zfs_acl_ids_free(&acl_ids);
4260 getnewvnode_drop_reserve();
4266 * Create a new object for the symlink.
4267 * for version 4 ZPL datsets the symlink will be an SA attribute
4269 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
4272 zfs_fuid_sync(zfsvfs, tx);
4274 mutex_enter(&zp->z_lock);
4276 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
4279 zfs_sa_symlink(zp, link, len, tx);
4280 mutex_exit(&zp->z_lock);
4283 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
4284 &zp->z_size, sizeof (zp->z_size), tx);
4286 * Insert the new object into the directory.
4288 (void) zfs_link_create(dl, zp, tx, ZNEW);
4290 if (flags & FIGNORECASE)
4292 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
4295 zfs_acl_ids_free(&acl_ids);
4299 getnewvnode_drop_reserve();
4301 zfs_dirent_unlock(dl);
4303 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4304 zil_commit(zilog, 0);
4311 * Return, in the buffer contained in the provided uio structure,
4312 * the symbolic path referred to by vp.
4314 * IN: vp - vnode of symbolic link.
4315 * uio - structure to contain the link path.
4316 * cr - credentials of caller.
4317 * ct - caller context
4319 * OUT: uio - structure containing the link path.
4321 * RETURN: 0 on success, error code on failure.
4324 * vp - atime updated
4328 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
4330 znode_t *zp = VTOZ(vp);
4331 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4337 mutex_enter(&zp->z_lock);
4339 error = sa_lookup_uio(zp->z_sa_hdl,
4340 SA_ZPL_SYMLINK(zfsvfs), uio);
4342 error = zfs_sa_readlink(zp, uio);
4343 mutex_exit(&zp->z_lock);
4345 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4352 * Insert a new entry into directory tdvp referencing svp.
4354 * IN: tdvp - Directory to contain new entry.
4355 * svp - vnode of new entry.
4356 * name - name of new entry.
4357 * cr - credentials of caller.
4358 * ct - caller context
4360 * RETURN: 0 on success, error code on failure.
4363 * tdvp - ctime|mtime updated
4364 * svp - ctime updated
4368 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
4369 caller_context_t *ct, int flags)
4371 znode_t *dzp = VTOZ(tdvp);
4373 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4382 boolean_t waited = B_FALSE;
4384 ASSERT(tdvp->v_type == VDIR);
4388 zilog = zfsvfs->z_log;
4390 if (VOP_REALVP(svp, &realvp, ct) == 0)
4394 * POSIX dictates that we return EPERM here.
4395 * Better choices include ENOTSUP or EISDIR.
4397 if (svp->v_type == VDIR) {
4399 return (SET_ERROR(EPERM));
4406 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
4407 * ctldir appear to have the same v_vfsp.
4409 if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) {
4411 return (SET_ERROR(EXDEV));
4414 /* Prevent links to .zfs/shares files */
4416 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4417 &parent, sizeof (uint64_t))) != 0) {
4421 if (parent == zfsvfs->z_shares_dir) {
4423 return (SET_ERROR(EPERM));
4426 if (zfsvfs->z_utf8 && u8_validate(name,
4427 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4429 return (SET_ERROR(EILSEQ));
4431 if (flags & FIGNORECASE)
4435 * We do not support links between attributes and non-attributes
4436 * because of the potential security risk of creating links
4437 * into "normal" file space in order to circumvent restrictions
4438 * imposed in attribute space.
4440 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4442 return (SET_ERROR(EINVAL));
4446 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4447 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) {
4449 return (SET_ERROR(EPERM));
4452 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4459 * Attempt to lock directory; fail if entry already exists.
4461 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4467 tx = dmu_tx_create(zfsvfs->z_os);
4468 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4469 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4470 zfs_sa_upgrade_txholds(tx, szp);
4471 zfs_sa_upgrade_txholds(tx, dzp);
4472 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4474 zfs_dirent_unlock(dl);
4475 if (error == ERESTART) {
4486 error = zfs_link_create(dl, szp, tx, 0);
4489 uint64_t txtype = TX_LINK;
4490 if (flags & FIGNORECASE)
4492 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4497 zfs_dirent_unlock(dl);
4500 vnevent_link(svp, ct);
4503 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4504 zil_commit(zilog, 0);
4512 * zfs_null_putapage() is used when the file system has been force
4513 * unmounted. It just drops the pages.
4517 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4518 size_t *lenp, int flags, cred_t *cr)
4520 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4525 * Push a page out to disk, klustering if possible.
4527 * IN: vp - file to push page to.
4528 * pp - page to push.
4529 * flags - additional flags.
4530 * cr - credentials of caller.
4532 * OUT: offp - start of range pushed.
4533 * lenp - len of range pushed.
4535 * RETURN: 0 on success, error code on failure.
4537 * NOTE: callers must have locked the page to be pushed. On
4538 * exit, the page (and all other pages in the kluster) must be
4543 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4544 size_t *lenp, int flags, cred_t *cr)
4546 znode_t *zp = VTOZ(vp);
4547 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4549 u_offset_t off, koff;
4556 * If our blocksize is bigger than the page size, try to kluster
4557 * multiple pages so that we write a full block (thus avoiding
4558 * a read-modify-write).
4560 if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4561 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4562 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4563 ASSERT(koff <= zp->z_size);
4564 if (koff + klen > zp->z_size)
4565 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4566 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4568 ASSERT3U(btop(len), ==, btopr(len));
4571 * Can't push pages past end-of-file.
4573 if (off >= zp->z_size) {
4574 /* ignore all pages */
4577 } else if (off + len > zp->z_size) {
4578 int npages = btopr(zp->z_size - off);
4581 page_list_break(&pp, &trunc, npages);
4582 /* ignore pages past end of file */
4584 pvn_write_done(trunc, flags);
4585 len = zp->z_size - off;
4588 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4589 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4590 err = SET_ERROR(EDQUOT);
4593 tx = dmu_tx_create(zfsvfs->z_os);
4594 dmu_tx_hold_write(tx, zp->z_id, off, len);
4596 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4597 zfs_sa_upgrade_txholds(tx, zp);
4598 err = dmu_tx_assign(tx, TXG_WAIT);
4604 if (zp->z_blksz <= PAGESIZE) {
4605 caddr_t va = zfs_map_page(pp, S_READ);
4606 ASSERT3U(len, <=, PAGESIZE);
4607 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4608 zfs_unmap_page(pp, va);
4610 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4614 uint64_t mtime[2], ctime[2];
4615 sa_bulk_attr_t bulk[3];
4618 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4620 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4622 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4624 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4626 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4631 pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4641 * Copy the portion of the file indicated from pages into the file.
4642 * The pages are stored in a page list attached to the files vnode.
4644 * IN: vp - vnode of file to push page data to.
4645 * off - position in file to put data.
4646 * len - amount of data to write.
4647 * flags - flags to control the operation.
4648 * cr - credentials of caller.
4649 * ct - caller context.
4651 * RETURN: 0 on success, error code on failure.
4654 * vp - ctime|mtime updated
4658 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4659 caller_context_t *ct)
4661 znode_t *zp = VTOZ(vp);
4662 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4674 * Align this request to the file block size in case we kluster.
4675 * XXX - this can result in pretty aggresive locking, which can
4676 * impact simultanious read/write access. One option might be
4677 * to break up long requests (len == 0) into block-by-block
4678 * operations to get narrower locking.
4680 blksz = zp->z_blksz;
4682 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4685 if (len > 0 && ISP2(blksz))
4686 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4692 * Search the entire vp list for pages >= io_off.
4694 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4695 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4698 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4700 if (off > zp->z_size) {
4701 /* past end of file */
4702 zfs_range_unlock(rl);
4707 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4709 for (off = io_off; io_off < off + len; io_off += io_len) {
4710 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4711 pp = page_lookup(vp, io_off,
4712 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4714 pp = page_lookup_nowait(vp, io_off,
4715 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4718 if (pp != NULL && pvn_getdirty(pp, flags)) {
4722 * Found a dirty page to push
4724 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4732 zfs_range_unlock(rl);
4733 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4734 zil_commit(zfsvfs->z_log, zp->z_id);
4738 #endif /* illumos */
4742 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4744 znode_t *zp = VTOZ(vp);
4745 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4748 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4749 if (zp->z_sa_hdl == NULL) {
4751 * The fs has been unmounted, or we did a
4752 * suspend/resume and this file no longer exists.
4754 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4759 mutex_enter(&zp->z_lock);
4760 if (zp->z_unlinked) {
4762 * Fast path to recycle a vnode of a removed file.
4764 mutex_exit(&zp->z_lock);
4765 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4769 mutex_exit(&zp->z_lock);
4771 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4772 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4774 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4775 zfs_sa_upgrade_txholds(tx, zp);
4776 error = dmu_tx_assign(tx, TXG_WAIT);
4780 mutex_enter(&zp->z_lock);
4781 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4782 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4783 zp->z_atime_dirty = 0;
4784 mutex_exit(&zp->z_lock);
4788 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4793 * Bounds-check the seek operation.
4795 * IN: vp - vnode seeking within
4796 * ooff - old file offset
4797 * noffp - pointer to new file offset
4798 * ct - caller context
4800 * RETURN: 0 on success, EINVAL if new offset invalid.
4804 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4805 caller_context_t *ct)
4807 if (vp->v_type == VDIR)
4809 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4813 * Pre-filter the generic locking function to trap attempts to place
4814 * a mandatory lock on a memory mapped file.
4817 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4818 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4820 znode_t *zp = VTOZ(vp);
4821 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4827 * We are following the UFS semantics with respect to mapcnt
4828 * here: If we see that the file is mapped already, then we will
4829 * return an error, but we don't worry about races between this
4830 * function and zfs_map().
4832 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4834 return (SET_ERROR(EAGAIN));
4837 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4841 * If we can't find a page in the cache, we will create a new page
4842 * and fill it with file data. For efficiency, we may try to fill
4843 * multiple pages at once (klustering) to fill up the supplied page
4844 * list. Note that the pages to be filled are held with an exclusive
4845 * lock to prevent access by other threads while they are being filled.
4848 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4849 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4851 znode_t *zp = VTOZ(vp);
4852 page_t *pp, *cur_pp;
4853 objset_t *os = zp->z_zfsvfs->z_os;
4854 u_offset_t io_off, total;
4858 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4860 * We only have a single page, don't bother klustering
4864 pp = page_create_va(vp, io_off, io_len,
4865 PG_EXCL | PG_WAIT, seg, addr);
4868 * Try to find enough pages to fill the page list
4870 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4871 &io_len, off, plsz, 0);
4875 * The page already exists, nothing to do here.
4882 * Fill the pages in the kluster.
4885 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4888 ASSERT3U(io_off, ==, cur_pp->p_offset);
4889 va = zfs_map_page(cur_pp, S_WRITE);
4890 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4892 zfs_unmap_page(cur_pp, va);
4894 /* On error, toss the entire kluster */
4895 pvn_read_done(pp, B_ERROR);
4896 /* convert checksum errors into IO errors */
4898 err = SET_ERROR(EIO);
4901 cur_pp = cur_pp->p_next;
4905 * Fill in the page list array from the kluster starting
4906 * from the desired offset `off'.
4907 * NOTE: the page list will always be null terminated.
4909 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4910 ASSERT(pl == NULL || (*pl)->p_offset == off);
4916 * Return pointers to the pages for the file region [off, off + len]
4917 * in the pl array. If plsz is greater than len, this function may
4918 * also return page pointers from after the specified region
4919 * (i.e. the region [off, off + plsz]). These additional pages are
4920 * only returned if they are already in the cache, or were created as
4921 * part of a klustered read.
4923 * IN: vp - vnode of file to get data from.
4924 * off - position in file to get data from.
4925 * len - amount of data to retrieve.
4926 * plsz - length of provided page list.
4927 * seg - segment to obtain pages for.
4928 * addr - virtual address of fault.
4929 * rw - mode of created pages.
4930 * cr - credentials of caller.
4931 * ct - caller context.
4933 * OUT: protp - protection mode of created pages.
4934 * pl - list of pages created.
4936 * RETURN: 0 on success, error code on failure.
4939 * vp - atime updated
4943 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4944 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4945 enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4947 znode_t *zp = VTOZ(vp);
4948 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4952 /* we do our own caching, faultahead is unnecessary */
4955 else if (len > plsz)
4958 len = P2ROUNDUP(len, PAGESIZE);
4959 ASSERT(plsz >= len);
4968 * Loop through the requested range [off, off + len) looking
4969 * for pages. If we don't find a page, we will need to create
4970 * a new page and fill it with data from the file.
4973 if (*pl = page_lookup(vp, off, SE_SHARED))
4975 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4978 ASSERT3U((*pl)->p_offset, ==, off);
4982 ASSERT3U(len, >=, PAGESIZE);
4985 ASSERT3U(plsz, >=, PAGESIZE);
4992 * Fill out the page array with any pages already in the cache.
4995 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
5002 * Release any pages we have previously locked.
5007 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
5017 * Request a memory map for a section of a file. This code interacts
5018 * with common code and the VM system as follows:
5020 * - common code calls mmap(), which ends up in smmap_common()
5021 * - this calls VOP_MAP(), which takes you into (say) zfs
5022 * - zfs_map() calls as_map(), passing segvn_create() as the callback
5023 * - segvn_create() creates the new segment and calls VOP_ADDMAP()
5024 * - zfs_addmap() updates z_mapcnt
5028 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
5029 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
5030 caller_context_t *ct)
5032 znode_t *zp = VTOZ(vp);
5033 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5034 segvn_crargs_t vn_a;
5040 if ((prot & PROT_WRITE) && (zp->z_pflags &
5041 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
5043 return (SET_ERROR(EPERM));
5046 if ((prot & (PROT_READ | PROT_EXEC)) &&
5047 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
5049 return (SET_ERROR(EACCES));
5052 if (vp->v_flag & VNOMAP) {
5054 return (SET_ERROR(ENOSYS));
5057 if (off < 0 || len > MAXOFFSET_T - off) {
5059 return (SET_ERROR(ENXIO));
5062 if (vp->v_type != VREG) {
5064 return (SET_ERROR(ENODEV));
5068 * If file is locked, disallow mapping.
5070 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
5072 return (SET_ERROR(EAGAIN));
5076 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
5084 vn_a.offset = (u_offset_t)off;
5085 vn_a.type = flags & MAP_TYPE;
5087 vn_a.maxprot = maxprot;
5090 vn_a.flags = flags & ~MAP_TYPE;
5092 vn_a.lgrp_mem_policy_flags = 0;
5094 error = as_map(as, *addrp, len, segvn_create, &vn_a);
5103 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5104 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
5105 caller_context_t *ct)
5107 uint64_t pages = btopr(len);
5109 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
5114 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
5115 * more accurate mtime for the associated file. Since we don't have a way of
5116 * detecting when the data was actually modified, we have to resort to
5117 * heuristics. If an explicit msync() is done, then we mark the mtime when the
5118 * last page is pushed. The problem occurs when the msync() call is omitted,
5119 * which by far the most common case:
5127 * putpage() via fsflush
5129 * If we wait until fsflush to come along, we can have a modification time that
5130 * is some arbitrary point in the future. In order to prevent this in the
5131 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
5136 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5137 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
5138 caller_context_t *ct)
5140 uint64_t pages = btopr(len);
5142 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
5143 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
5145 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
5146 vn_has_cached_data(vp))
5147 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
5153 * Free or allocate space in a file. Currently, this function only
5154 * supports the `F_FREESP' command. However, this command is somewhat
5155 * misnamed, as its functionality includes the ability to allocate as
5156 * well as free space.
5158 * IN: vp - vnode of file to free data in.
5159 * cmd - action to take (only F_FREESP supported).
5160 * bfp - section of file to free/alloc.
5161 * flag - current file open mode flags.
5162 * offset - current file offset.
5163 * cr - credentials of caller [UNUSED].
5164 * ct - caller context.
5166 * RETURN: 0 on success, error code on failure.
5169 * vp - ctime|mtime updated
5173 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
5174 offset_t offset, cred_t *cr, caller_context_t *ct)
5176 znode_t *zp = VTOZ(vp);
5177 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5184 if (cmd != F_FREESP) {
5186 return (SET_ERROR(EINVAL));
5190 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
5191 * callers might not be able to detect properly that we are read-only,
5192 * so check it explicitly here.
5194 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
5196 return (SET_ERROR(EROFS));
5199 if (error = convoff(vp, bfp, 0, offset)) {
5204 if (bfp->l_len < 0) {
5206 return (SET_ERROR(EINVAL));
5210 len = bfp->l_len; /* 0 means from off to end of file */
5212 error = zfs_freesp(zp, off, len, flag, TRUE);
5217 #endif /* illumos */
5219 CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid));
5220 CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid));
5224 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
5226 znode_t *zp = VTOZ(vp);
5227 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5230 uint64_t object = zp->z_id;
5237 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
5238 &gen64, sizeof (uint64_t))) != 0) {
5243 gen = (uint32_t)gen64;
5245 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
5248 if (fidp->fid_len < size) {
5249 fidp->fid_len = size;
5251 return (SET_ERROR(ENOSPC));
5254 fidp->fid_len = size;
5257 zfid = (zfid_short_t *)fidp;
5259 zfid->zf_len = size;
5261 for (i = 0; i < sizeof (zfid->zf_object); i++)
5262 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
5264 /* Must have a non-zero generation number to distinguish from .zfs */
5267 for (i = 0; i < sizeof (zfid->zf_gen); i++)
5268 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
5270 if (size == LONG_FID_LEN) {
5271 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
5274 zlfid = (zfid_long_t *)fidp;
5276 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
5277 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
5279 /* XXX - this should be the generation number for the objset */
5280 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
5281 zlfid->zf_setgen[i] = 0;
5289 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
5290 caller_context_t *ct)
5302 case _PC_FILESIZEBITS:
5306 case _PC_XATTR_EXISTS:
5308 zfsvfs = zp->z_zfsvfs;
5312 error = zfs_dirent_lock(&dl, zp, "", &xzp,
5313 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
5315 zfs_dirent_unlock(dl);
5316 if (!zfs_dirempty(xzp))
5319 } else if (error == ENOENT) {
5321 * If there aren't extended attributes, it's the
5322 * same as having zero of them.
5329 case _PC_SATTR_ENABLED:
5330 case _PC_SATTR_EXISTS:
5331 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
5332 (vp->v_type == VREG || vp->v_type == VDIR);
5335 case _PC_ACCESS_FILTERING:
5336 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
5340 case _PC_ACL_ENABLED:
5341 *valp = _ACL_ACE_ENABLED;
5343 #endif /* illumos */
5344 case _PC_MIN_HOLE_SIZE:
5345 *valp = (int)SPA_MINBLOCKSIZE;
5348 case _PC_TIMESTAMP_RESOLUTION:
5349 /* nanosecond timestamp resolution */
5353 case _PC_ACL_EXTENDED:
5361 case _PC_ACL_PATH_MAX:
5362 *valp = ACL_MAX_ENTRIES;
5366 return (EOPNOTSUPP);
5372 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5373 caller_context_t *ct)
5375 znode_t *zp = VTOZ(vp);
5376 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5378 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5382 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5390 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5391 caller_context_t *ct)
5393 znode_t *zp = VTOZ(vp);
5394 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5396 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5397 zilog_t *zilog = zfsvfs->z_log;
5402 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5404 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5405 zil_commit(zilog, 0);
5413 * The smallest read we may consider to loan out an arcbuf.
5414 * This must be a power of 2.
5416 int zcr_blksz_min = (1 << 10); /* 1K */
5418 * If set to less than the file block size, allow loaning out of an
5419 * arcbuf for a partial block read. This must be a power of 2.
5421 int zcr_blksz_max = (1 << 17); /* 128K */
5425 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5426 caller_context_t *ct)
5428 znode_t *zp = VTOZ(vp);
5429 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5430 int max_blksz = zfsvfs->z_max_blksz;
5431 uio_t *uio = &xuio->xu_uio;
5432 ssize_t size = uio->uio_resid;
5433 offset_t offset = uio->uio_loffset;
5438 int preamble, postamble;
5440 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5441 return (SET_ERROR(EINVAL));
5448 * Loan out an arc_buf for write if write size is bigger than
5449 * max_blksz, and the file's block size is also max_blksz.
5452 if (size < blksz || zp->z_blksz != blksz) {
5454 return (SET_ERROR(EINVAL));
5457 * Caller requests buffers for write before knowing where the
5458 * write offset might be (e.g. NFS TCP write).
5463 preamble = P2PHASE(offset, blksz);
5465 preamble = blksz - preamble;
5470 postamble = P2PHASE(size, blksz);
5473 fullblk = size / blksz;
5474 (void) dmu_xuio_init(xuio,
5475 (preamble != 0) + fullblk + (postamble != 0));
5476 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5477 int, postamble, int,
5478 (preamble != 0) + fullblk + (postamble != 0));
5481 * Have to fix iov base/len for partial buffers. They
5482 * currently represent full arc_buf's.
5485 /* data begins in the middle of the arc_buf */
5486 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5489 (void) dmu_xuio_add(xuio, abuf,
5490 blksz - preamble, preamble);
5493 for (i = 0; i < fullblk; i++) {
5494 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5497 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5501 /* data ends in the middle of the arc_buf */
5502 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5505 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5510 * Loan out an arc_buf for read if the read size is larger than
5511 * the current file block size. Block alignment is not
5512 * considered. Partial arc_buf will be loaned out for read.
5514 blksz = zp->z_blksz;
5515 if (blksz < zcr_blksz_min)
5516 blksz = zcr_blksz_min;
5517 if (blksz > zcr_blksz_max)
5518 blksz = zcr_blksz_max;
5519 /* avoid potential complexity of dealing with it */
5520 if (blksz > max_blksz) {
5522 return (SET_ERROR(EINVAL));
5525 maxsize = zp->z_size - uio->uio_loffset;
5529 if (size < blksz || vn_has_cached_data(vp)) {
5531 return (SET_ERROR(EINVAL));
5536 return (SET_ERROR(EINVAL));
5539 uio->uio_extflg = UIO_XUIO;
5540 XUIO_XUZC_RW(xuio) = ioflag;
5547 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5551 int ioflag = XUIO_XUZC_RW(xuio);
5553 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5555 i = dmu_xuio_cnt(xuio);
5557 abuf = dmu_xuio_arcbuf(xuio, i);
5559 * if abuf == NULL, it must be a write buffer
5560 * that has been returned in zfs_write().
5563 dmu_return_arcbuf(abuf);
5564 ASSERT(abuf || ioflag == UIO_WRITE);
5567 dmu_xuio_fini(xuio);
5572 * Predeclare these here so that the compiler assumes that
5573 * this is an "old style" function declaration that does
5574 * not include arguments => we won't get type mismatch errors
5575 * in the initializations that follow.
5577 static int zfs_inval();
5578 static int zfs_isdir();
5583 return (SET_ERROR(EINVAL));
5589 return (SET_ERROR(EISDIR));
5592 * Directory vnode operations template
5594 vnodeops_t *zfs_dvnodeops;
5595 const fs_operation_def_t zfs_dvnodeops_template[] = {
5596 VOPNAME_OPEN, { .vop_open = zfs_open },
5597 VOPNAME_CLOSE, { .vop_close = zfs_close },
5598 VOPNAME_READ, { .error = zfs_isdir },
5599 VOPNAME_WRITE, { .error = zfs_isdir },
5600 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5601 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5602 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5603 VOPNAME_ACCESS, { .vop_access = zfs_access },
5604 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5605 VOPNAME_CREATE, { .vop_create = zfs_create },
5606 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5607 VOPNAME_LINK, { .vop_link = zfs_link },
5608 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5609 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir },
5610 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5611 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5612 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink },
5613 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5614 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5615 VOPNAME_FID, { .vop_fid = zfs_fid },
5616 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5617 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5618 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5619 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5620 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5625 * Regular file vnode operations template
5627 vnodeops_t *zfs_fvnodeops;
5628 const fs_operation_def_t zfs_fvnodeops_template[] = {
5629 VOPNAME_OPEN, { .vop_open = zfs_open },
5630 VOPNAME_CLOSE, { .vop_close = zfs_close },
5631 VOPNAME_READ, { .vop_read = zfs_read },
5632 VOPNAME_WRITE, { .vop_write = zfs_write },
5633 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5634 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5635 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5636 VOPNAME_ACCESS, { .vop_access = zfs_access },
5637 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5638 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5639 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5640 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5641 VOPNAME_FID, { .vop_fid = zfs_fid },
5642 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5643 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock },
5644 VOPNAME_SPACE, { .vop_space = zfs_space },
5645 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage },
5646 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage },
5647 VOPNAME_MAP, { .vop_map = zfs_map },
5648 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap },
5649 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap },
5650 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5651 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5652 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5653 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5654 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf },
5655 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf },
5660 * Symbolic link vnode operations template
5662 vnodeops_t *zfs_symvnodeops;
5663 const fs_operation_def_t zfs_symvnodeops_template[] = {
5664 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5665 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5666 VOPNAME_ACCESS, { .vop_access = zfs_access },
5667 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5668 VOPNAME_READLINK, { .vop_readlink = zfs_readlink },
5669 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5670 VOPNAME_FID, { .vop_fid = zfs_fid },
5671 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5672 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5677 * special share hidden files vnode operations template
5679 vnodeops_t *zfs_sharevnodeops;
5680 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5681 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5682 VOPNAME_ACCESS, { .vop_access = zfs_access },
5683 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5684 VOPNAME_FID, { .vop_fid = zfs_fid },
5685 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5686 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5687 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5688 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5693 * Extended attribute directory vnode operations template
5695 * This template is identical to the directory vnodes
5696 * operation template except for restricted operations:
5700 * Note that there are other restrictions embedded in:
5701 * zfs_create() - restrict type to VREG
5702 * zfs_link() - no links into/out of attribute space
5703 * zfs_rename() - no moves into/out of attribute space
5705 vnodeops_t *zfs_xdvnodeops;
5706 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5707 VOPNAME_OPEN, { .vop_open = zfs_open },
5708 VOPNAME_CLOSE, { .vop_close = zfs_close },
5709 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5710 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5711 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5712 VOPNAME_ACCESS, { .vop_access = zfs_access },
5713 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5714 VOPNAME_CREATE, { .vop_create = zfs_create },
5715 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5716 VOPNAME_LINK, { .vop_link = zfs_link },
5717 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5718 VOPNAME_MKDIR, { .error = zfs_inval },
5719 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5720 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5721 VOPNAME_SYMLINK, { .error = zfs_inval },
5722 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5723 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5724 VOPNAME_FID, { .vop_fid = zfs_fid },
5725 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5726 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5727 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5728 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5729 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5734 * Error vnode operations template
5736 vnodeops_t *zfs_evnodeops;
5737 const fs_operation_def_t zfs_evnodeops_template[] = {
5738 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5739 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5742 #endif /* illumos */
5745 ioflags(int ioflags)
5749 if (ioflags & IO_APPEND)
5751 if (ioflags & IO_NDELAY)
5753 if (ioflags & IO_SYNC)
5754 flags |= (FSYNC | FDSYNC | FRSYNC);
5760 zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage)
5762 znode_t *zp = VTOZ(vp);
5763 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5764 objset_t *os = zp->z_zfsvfs->z_os;
5765 vm_page_t mfirst, mlast, mreq;
5769 off_t startoff, endoff;
5771 vm_pindex_t reqstart, reqend;
5772 int pcount, lsize, reqsize, size;
5777 pcount = OFF_TO_IDX(round_page(count));
5779 object = mreq->object;
5782 KASSERT(vp->v_object == object, ("mismatching object"));
5784 if (pcount > 1 && zp->z_blksz > PAGESIZE) {
5785 startoff = rounddown(IDX_TO_OFF(mreq->pindex), zp->z_blksz);
5786 reqstart = OFF_TO_IDX(round_page(startoff));
5787 if (reqstart < m[0]->pindex)
5790 reqstart = reqstart - m[0]->pindex;
5791 endoff = roundup(IDX_TO_OFF(mreq->pindex) + PAGE_SIZE,
5793 reqend = OFF_TO_IDX(trunc_page(endoff)) - 1;
5794 if (reqend > m[pcount - 1]->pindex)
5795 reqend = m[pcount - 1]->pindex;
5796 reqsize = reqend - m[reqstart]->pindex + 1;
5797 KASSERT(reqstart <= reqpage && reqpage < reqstart + reqsize,
5798 ("reqpage beyond [reqstart, reqstart + reqsize[ bounds"));
5803 mfirst = m[reqstart];
5804 mlast = m[reqstart + reqsize - 1];
5806 zfs_vmobject_wlock(object);
5808 for (i = 0; i < reqstart; i++) {
5811 vm_page_unlock(m[i]);
5813 for (i = reqstart + reqsize; i < pcount; i++) {
5816 vm_page_unlock(m[i]);
5819 if (mreq->valid && reqsize == 1) {
5820 if (mreq->valid != VM_PAGE_BITS_ALL)
5821 vm_page_zero_invalid(mreq, TRUE);
5822 zfs_vmobject_wunlock(object);
5824 return (zfs_vm_pagerret_ok);
5827 PCPU_INC(cnt.v_vnodein);
5828 PCPU_ADD(cnt.v_vnodepgsin, reqsize);
5830 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) {
5831 for (i = reqstart; i < reqstart + reqsize; i++) {
5835 vm_page_unlock(m[i]);
5838 zfs_vmobject_wunlock(object);
5840 return (zfs_vm_pagerret_bad);
5844 if (IDX_TO_OFF(mlast->pindex) + lsize > object->un_pager.vnp.vnp_size)
5845 lsize = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mlast->pindex);
5847 zfs_vmobject_wunlock(object);
5849 for (i = reqstart; i < reqstart + reqsize; i++) {
5851 if (i == (reqstart + reqsize - 1))
5853 va = zfs_map_page(m[i], &sf);
5854 error = dmu_read(os, zp->z_id, IDX_TO_OFF(m[i]->pindex),
5855 size, va, DMU_READ_PREFETCH);
5856 if (size != PAGE_SIZE)
5857 bzero(va + size, PAGE_SIZE - size);
5863 zfs_vmobject_wlock(object);
5865 for (i = reqstart; i < reqstart + reqsize; i++) {
5867 m[i]->valid = VM_PAGE_BITS_ALL;
5868 KASSERT(m[i]->dirty == 0, ("zfs_getpages: page %p is dirty", m[i]));
5870 vm_page_readahead_finish(m[i]);
5873 zfs_vmobject_wunlock(object);
5875 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
5877 return (error ? zfs_vm_pagerret_error : zfs_vm_pagerret_ok);
5881 zfs_freebsd_getpages(ap)
5882 struct vop_getpages_args /* {
5887 vm_ooffset_t a_offset;
5891 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage));
5895 zfs_putpages(struct vnode *vp, vm_page_t *ma, size_t len, int flags,
5898 znode_t *zp = VTOZ(vp);
5899 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5908 vm_ooffset_t lo_off;
5919 object = vp->v_object;
5923 KASSERT(ma[0]->object == object, ("mismatching object"));
5924 KASSERT(len > 0 && (len & PAGE_MASK) == 0, ("unexpected length"));
5926 for (i = 0; i < pcount; i++)
5927 rtvals[i] = zfs_vm_pagerret_error;
5929 off = IDX_TO_OFF(ma[0]->pindex);
5930 blksz = zp->z_blksz;
5931 lo_off = rounddown(off, blksz);
5932 lo_len = roundup(len + (off - lo_off), blksz);
5933 rl = zfs_range_lock(zp, lo_off, lo_len, RL_WRITER);
5935 zfs_vmobject_wlock(object);
5936 if (len + off > object->un_pager.vnp.vnp_size) {
5937 if (object->un_pager.vnp.vnp_size > off) {
5940 len = object->un_pager.vnp.vnp_size - off;
5942 if ((pgoff = (int)len & PAGE_MASK) != 0) {
5944 * If the object is locked and the following
5945 * conditions hold, then the page's dirty
5946 * field cannot be concurrently changed by a
5950 vm_page_assert_sbusied(m);
5951 KASSERT(!pmap_page_is_write_mapped(m),
5952 ("zfs_putpages: page %p is not read-only", m));
5953 vm_page_clear_dirty(m, pgoff, PAGE_SIZE -
5960 if (ncount < pcount) {
5961 for (i = ncount; i < pcount; i++) {
5962 rtvals[i] = zfs_vm_pagerret_bad;
5966 zfs_vmobject_wunlock(object);
5971 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
5972 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
5977 tx = dmu_tx_create(zfsvfs->z_os);
5978 dmu_tx_hold_write(tx, zp->z_id, off, len);
5980 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
5981 zfs_sa_upgrade_txholds(tx, zp);
5982 err = dmu_tx_assign(tx, TXG_NOWAIT);
5984 if (err == ERESTART) {
5993 if (zp->z_blksz < PAGE_SIZE) {
5995 for (i = 0; len > 0; off += tocopy, len -= tocopy, i++) {
5996 tocopy = len > PAGE_SIZE ? PAGE_SIZE : len;
5997 va = zfs_map_page(ma[i], &sf);
5998 dmu_write(zfsvfs->z_os, zp->z_id, off, tocopy, va, tx);
6002 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, ma, tx);
6006 uint64_t mtime[2], ctime[2];
6007 sa_bulk_attr_t bulk[3];
6010 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
6012 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
6014 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
6016 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
6018 (void)sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
6019 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
6021 zfs_vmobject_wlock(object);
6022 for (i = 0; i < ncount; i++) {
6023 rtvals[i] = zfs_vm_pagerret_ok;
6024 vm_page_undirty(ma[i]);
6026 zfs_vmobject_wunlock(object);
6027 PCPU_INC(cnt.v_vnodeout);
6028 PCPU_ADD(cnt.v_vnodepgsout, ncount);
6033 zfs_range_unlock(rl);
6034 if ((flags & (zfs_vm_pagerput_sync | zfs_vm_pagerput_inval)) != 0 ||
6035 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
6036 zil_commit(zfsvfs->z_log, zp->z_id);
6042 zfs_freebsd_putpages(ap)
6043 struct vop_putpages_args /* {
6049 vm_ooffset_t a_offset;
6053 return (zfs_putpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_sync,
6058 zfs_freebsd_bmap(ap)
6059 struct vop_bmap_args /* {
6062 struct bufobj **a_bop;
6069 if (ap->a_bop != NULL)
6070 *ap->a_bop = &ap->a_vp->v_bufobj;
6071 if (ap->a_bnp != NULL)
6072 *ap->a_bnp = ap->a_bn;
6073 if (ap->a_runp != NULL)
6075 if (ap->a_runb != NULL)
6082 zfs_freebsd_open(ap)
6083 struct vop_open_args /* {
6086 struct ucred *a_cred;
6087 struct thread *a_td;
6090 vnode_t *vp = ap->a_vp;
6091 znode_t *zp = VTOZ(vp);
6094 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL);
6096 vnode_create_vobject(vp, zp->z_size, ap->a_td);
6101 zfs_freebsd_close(ap)
6102 struct vop_close_args /* {
6105 struct ucred *a_cred;
6106 struct thread *a_td;
6110 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL));
6114 zfs_freebsd_ioctl(ap)
6115 struct vop_ioctl_args /* {
6125 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data,
6126 ap->a_fflag, ap->a_cred, NULL, NULL));
6130 zfs_freebsd_read(ap)
6131 struct vop_read_args /* {
6135 struct ucred *a_cred;
6139 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
6144 zfs_freebsd_write(ap)
6145 struct vop_write_args /* {
6149 struct ucred *a_cred;
6153 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
6158 zfs_freebsd_access(ap)
6159 struct vop_access_args /* {
6161 accmode_t a_accmode;
6162 struct ucred *a_cred;
6163 struct thread *a_td;
6166 vnode_t *vp = ap->a_vp;
6167 znode_t *zp = VTOZ(vp);
6172 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND,
6174 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND);
6176 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL);
6179 * VADMIN has to be handled by vaccess().
6182 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND);
6184 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
6185 zp->z_gid, accmode, ap->a_cred, NULL);
6190 * For VEXEC, ensure that at least one execute bit is set for
6193 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR &&
6194 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
6202 zfs_freebsd_lookup(ap)
6203 struct vop_lookup_args /* {
6204 struct vnode *a_dvp;
6205 struct vnode **a_vpp;
6206 struct componentname *a_cnp;
6209 struct componentname *cnp = ap->a_cnp;
6210 char nm[NAME_MAX + 1];
6212 ASSERT(cnp->cn_namelen < sizeof(nm));
6213 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm)));
6215 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop,
6216 cnp->cn_cred, cnp->cn_thread, 0));
6220 zfs_freebsd_create(ap)
6221 struct vop_create_args /* {
6222 struct vnode *a_dvp;
6223 struct vnode **a_vpp;
6224 struct componentname *a_cnp;
6225 struct vattr *a_vap;
6228 struct componentname *cnp = ap->a_cnp;
6229 vattr_t *vap = ap->a_vap;
6232 ASSERT(cnp->cn_flags & SAVENAME);
6234 vattr_init_mask(vap);
6235 mode = vap->va_mode & ALLPERMS;
6237 error = zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode,
6238 ap->a_vpp, cnp->cn_cred, cnp->cn_thread);
6239 #ifdef FREEBSD_NAMECACHE
6240 if (error == 0 && (cnp->cn_flags & MAKEENTRY) != 0)
6241 cache_enter(ap->a_dvp, *ap->a_vpp, cnp);
6247 zfs_freebsd_remove(ap)
6248 struct vop_remove_args /* {
6249 struct vnode *a_dvp;
6251 struct componentname *a_cnp;
6255 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6257 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr,
6258 ap->a_cnp->cn_cred, NULL, 0));
6262 zfs_freebsd_mkdir(ap)
6263 struct vop_mkdir_args /* {
6264 struct vnode *a_dvp;
6265 struct vnode **a_vpp;
6266 struct componentname *a_cnp;
6267 struct vattr *a_vap;
6270 vattr_t *vap = ap->a_vap;
6272 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6274 vattr_init_mask(vap);
6276 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp,
6277 ap->a_cnp->cn_cred, NULL, 0, NULL));
6281 zfs_freebsd_rmdir(ap)
6282 struct vop_rmdir_args /* {
6283 struct vnode *a_dvp;
6285 struct componentname *a_cnp;
6288 struct componentname *cnp = ap->a_cnp;
6290 ASSERT(cnp->cn_flags & SAVENAME);
6292 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0));
6296 zfs_freebsd_readdir(ap)
6297 struct vop_readdir_args /* {
6300 struct ucred *a_cred;
6307 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag,
6308 ap->a_ncookies, ap->a_cookies));
6312 zfs_freebsd_fsync(ap)
6313 struct vop_fsync_args /* {
6316 struct thread *a_td;
6321 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL));
6325 zfs_freebsd_getattr(ap)
6326 struct vop_getattr_args /* {
6328 struct vattr *a_vap;
6329 struct ucred *a_cred;
6332 vattr_t *vap = ap->a_vap;
6338 xvap.xva_vattr = *vap;
6339 xvap.xva_vattr.va_mask |= AT_XVATTR;
6341 /* Convert chflags into ZFS-type flags. */
6342 /* XXX: what about SF_SETTABLE?. */
6343 XVA_SET_REQ(&xvap, XAT_IMMUTABLE);
6344 XVA_SET_REQ(&xvap, XAT_APPENDONLY);
6345 XVA_SET_REQ(&xvap, XAT_NOUNLINK);
6346 XVA_SET_REQ(&xvap, XAT_NODUMP);
6347 XVA_SET_REQ(&xvap, XAT_READONLY);
6348 XVA_SET_REQ(&xvap, XAT_ARCHIVE);
6349 XVA_SET_REQ(&xvap, XAT_SYSTEM);
6350 XVA_SET_REQ(&xvap, XAT_HIDDEN);
6351 XVA_SET_REQ(&xvap, XAT_REPARSE);
6352 XVA_SET_REQ(&xvap, XAT_OFFLINE);
6353 XVA_SET_REQ(&xvap, XAT_SPARSE);
6355 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL);
6359 /* Convert ZFS xattr into chflags. */
6360 #define FLAG_CHECK(fflag, xflag, xfield) do { \
6361 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \
6362 fflags |= (fflag); \
6364 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE,
6365 xvap.xva_xoptattrs.xoa_immutable);
6366 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY,
6367 xvap.xva_xoptattrs.xoa_appendonly);
6368 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK,
6369 xvap.xva_xoptattrs.xoa_nounlink);
6370 FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE,
6371 xvap.xva_xoptattrs.xoa_archive);
6372 FLAG_CHECK(UF_NODUMP, XAT_NODUMP,
6373 xvap.xva_xoptattrs.xoa_nodump);
6374 FLAG_CHECK(UF_READONLY, XAT_READONLY,
6375 xvap.xva_xoptattrs.xoa_readonly);
6376 FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM,
6377 xvap.xva_xoptattrs.xoa_system);
6378 FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN,
6379 xvap.xva_xoptattrs.xoa_hidden);
6380 FLAG_CHECK(UF_REPARSE, XAT_REPARSE,
6381 xvap.xva_xoptattrs.xoa_reparse);
6382 FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE,
6383 xvap.xva_xoptattrs.xoa_offline);
6384 FLAG_CHECK(UF_SPARSE, XAT_SPARSE,
6385 xvap.xva_xoptattrs.xoa_sparse);
6388 *vap = xvap.xva_vattr;
6389 vap->va_flags = fflags;
6394 zfs_freebsd_setattr(ap)
6395 struct vop_setattr_args /* {
6397 struct vattr *a_vap;
6398 struct ucred *a_cred;
6401 vnode_t *vp = ap->a_vp;
6402 vattr_t *vap = ap->a_vap;
6403 cred_t *cred = ap->a_cred;
6408 vattr_init_mask(vap);
6409 vap->va_mask &= ~AT_NOSET;
6412 xvap.xva_vattr = *vap;
6414 zflags = VTOZ(vp)->z_pflags;
6416 if (vap->va_flags != VNOVAL) {
6417 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs;
6420 if (zfsvfs->z_use_fuids == B_FALSE)
6421 return (EOPNOTSUPP);
6423 fflags = vap->va_flags;
6426 * We need to figure out whether it makes sense to allow
6427 * UF_REPARSE through, since we don't really have other
6428 * facilities to handle reparse points and zfs_setattr()
6429 * doesn't currently allow setting that attribute anyway.
6431 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE|
6432 UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE|
6433 UF_OFFLINE|UF_SPARSE)) != 0)
6434 return (EOPNOTSUPP);
6436 * Unprivileged processes are not permitted to unset system
6437 * flags, or modify flags if any system flags are set.
6438 * Privileged non-jail processes may not modify system flags
6439 * if securelevel > 0 and any existing system flags are set.
6440 * Privileged jail processes behave like privileged non-jail
6441 * processes if the security.jail.chflags_allowed sysctl is
6442 * is non-zero; otherwise, they behave like unprivileged
6445 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 ||
6446 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) {
6448 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6449 error = securelevel_gt(cred, 0);
6455 * Callers may only modify the file flags on objects they
6456 * have VADMIN rights for.
6458 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0)
6461 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6465 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) {
6470 #define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \
6471 if (((fflags & (fflag)) && !(zflags & (zflag))) || \
6472 ((zflags & (zflag)) && !(fflags & (fflag)))) { \
6473 XVA_SET_REQ(&xvap, (xflag)); \
6474 (xfield) = ((fflags & (fflag)) != 0); \
6477 /* Convert chflags into ZFS-type flags. */
6478 /* XXX: what about SF_SETTABLE?. */
6479 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE,
6480 xvap.xva_xoptattrs.xoa_immutable);
6481 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY,
6482 xvap.xva_xoptattrs.xoa_appendonly);
6483 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK,
6484 xvap.xva_xoptattrs.xoa_nounlink);
6485 FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE,
6486 xvap.xva_xoptattrs.xoa_archive);
6487 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP,
6488 xvap.xva_xoptattrs.xoa_nodump);
6489 FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY,
6490 xvap.xva_xoptattrs.xoa_readonly);
6491 FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM,
6492 xvap.xva_xoptattrs.xoa_system);
6493 FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN,
6494 xvap.xva_xoptattrs.xoa_hidden);
6495 FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE,
6496 xvap.xva_xoptattrs.xoa_hidden);
6497 FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE,
6498 xvap.xva_xoptattrs.xoa_offline);
6499 FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE,
6500 xvap.xva_xoptattrs.xoa_sparse);
6503 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL));
6507 zfs_freebsd_rename(ap)
6508 struct vop_rename_args /* {
6509 struct vnode *a_fdvp;
6510 struct vnode *a_fvp;
6511 struct componentname *a_fcnp;
6512 struct vnode *a_tdvp;
6513 struct vnode *a_tvp;
6514 struct componentname *a_tcnp;
6517 vnode_t *fdvp = ap->a_fdvp;
6518 vnode_t *fvp = ap->a_fvp;
6519 vnode_t *tdvp = ap->a_tdvp;
6520 vnode_t *tvp = ap->a_tvp;
6523 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART));
6524 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART));
6527 * Check for cross-device rename.
6529 if ((fdvp->v_mount != tdvp->v_mount) ||
6530 (tvp && (fdvp->v_mount != tvp->v_mount)))
6533 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp,
6534 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0);
6548 zfs_freebsd_symlink(ap)
6549 struct vop_symlink_args /* {
6550 struct vnode *a_dvp;
6551 struct vnode **a_vpp;
6552 struct componentname *a_cnp;
6553 struct vattr *a_vap;
6557 struct componentname *cnp = ap->a_cnp;
6558 vattr_t *vap = ap->a_vap;
6560 ASSERT(cnp->cn_flags & SAVENAME);
6562 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */
6563 vattr_init_mask(vap);
6565 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap,
6566 ap->a_target, cnp->cn_cred, cnp->cn_thread));
6570 zfs_freebsd_readlink(ap)
6571 struct vop_readlink_args /* {
6574 struct ucred *a_cred;
6578 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL));
6582 zfs_freebsd_link(ap)
6583 struct vop_link_args /* {
6584 struct vnode *a_tdvp;
6586 struct componentname *a_cnp;
6589 struct componentname *cnp = ap->a_cnp;
6590 vnode_t *vp = ap->a_vp;
6591 vnode_t *tdvp = ap->a_tdvp;
6593 if (tdvp->v_mount != vp->v_mount)
6596 ASSERT(cnp->cn_flags & SAVENAME);
6598 return (zfs_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0));
6602 zfs_freebsd_inactive(ap)
6603 struct vop_inactive_args /* {
6605 struct thread *a_td;
6608 vnode_t *vp = ap->a_vp;
6610 zfs_inactive(vp, ap->a_td->td_ucred, NULL);
6615 zfs_freebsd_reclaim(ap)
6616 struct vop_reclaim_args /* {
6618 struct thread *a_td;
6621 vnode_t *vp = ap->a_vp;
6622 znode_t *zp = VTOZ(vp);
6623 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
6627 /* Destroy the vm object and flush associated pages. */
6628 vnode_destroy_vobject(vp);
6631 * z_teardown_inactive_lock protects from a race with
6632 * zfs_znode_dmu_fini in zfsvfs_teardown during
6635 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
6636 if (zp->z_sa_hdl == NULL)
6640 rw_exit(&zfsvfs->z_teardown_inactive_lock);
6648 struct vop_fid_args /* {
6654 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL));
6658 zfs_freebsd_pathconf(ap)
6659 struct vop_pathconf_args /* {
6662 register_t *a_retval;
6668 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL);
6670 *ap->a_retval = val;
6671 else if (error == EOPNOTSUPP)
6672 error = vop_stdpathconf(ap);
6677 zfs_freebsd_fifo_pathconf(ap)
6678 struct vop_pathconf_args /* {
6681 register_t *a_retval;
6685 switch (ap->a_name) {
6686 case _PC_ACL_EXTENDED:
6688 case _PC_ACL_PATH_MAX:
6689 case _PC_MAC_PRESENT:
6690 return (zfs_freebsd_pathconf(ap));
6692 return (fifo_specops.vop_pathconf(ap));
6697 * FreeBSD's extended attributes namespace defines file name prefix for ZFS'
6698 * extended attribute name:
6701 * system freebsd:system:
6702 * user (none, can be used to access ZFS fsattr(5) attributes
6703 * created on Solaris)
6706 zfs_create_attrname(int attrnamespace, const char *name, char *attrname,
6709 const char *namespace, *prefix, *suffix;
6711 /* We don't allow '/' character in attribute name. */
6712 if (strchr(name, '/') != NULL)
6714 /* We don't allow attribute names that start with "freebsd:" string. */
6715 if (strncmp(name, "freebsd:", 8) == 0)
6718 bzero(attrname, size);
6720 switch (attrnamespace) {
6721 case EXTATTR_NAMESPACE_USER:
6723 prefix = "freebsd:";
6724 namespace = EXTATTR_NAMESPACE_USER_STRING;
6728 * This is the default namespace by which we can access all
6729 * attributes created on Solaris.
6731 prefix = namespace = suffix = "";
6734 case EXTATTR_NAMESPACE_SYSTEM:
6735 prefix = "freebsd:";
6736 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING;
6739 case EXTATTR_NAMESPACE_EMPTY:
6743 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix,
6745 return (ENAMETOOLONG);
6751 * Vnode operating to retrieve a named extended attribute.
6754 zfs_getextattr(struct vop_getextattr_args *ap)
6757 IN struct vnode *a_vp;
6758 IN int a_attrnamespace;
6759 IN const char *a_name;
6760 INOUT struct uio *a_uio;
6762 IN struct ucred *a_cred;
6763 IN struct thread *a_td;
6767 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6768 struct thread *td = ap->a_td;
6769 struct nameidata nd;
6772 vnode_t *xvp = NULL, *vp;
6775 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6776 ap->a_cred, ap->a_td, VREAD);
6780 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6787 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6795 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6797 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL);
6799 NDFREE(&nd, NDF_ONLY_PNBUF);
6802 if (error == ENOENT)
6807 if (ap->a_size != NULL) {
6808 error = VOP_GETATTR(vp, &va, ap->a_cred);
6810 *ap->a_size = (size_t)va.va_size;
6811 } else if (ap->a_uio != NULL)
6812 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6815 vn_close(vp, flags, ap->a_cred, td);
6822 * Vnode operation to remove a named attribute.
6825 zfs_deleteextattr(struct vop_deleteextattr_args *ap)
6828 IN struct vnode *a_vp;
6829 IN int a_attrnamespace;
6830 IN const char *a_name;
6831 IN struct ucred *a_cred;
6832 IN struct thread *a_td;
6836 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6837 struct thread *td = ap->a_td;
6838 struct nameidata nd;
6841 vnode_t *xvp = NULL, *vp;
6844 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6845 ap->a_cred, ap->a_td, VWRITE);
6849 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6856 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6863 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF,
6864 UIO_SYSSPACE, attrname, xvp, td);
6869 NDFREE(&nd, NDF_ONLY_PNBUF);
6870 if (error == ENOENT)
6875 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
6876 NDFREE(&nd, NDF_ONLY_PNBUF);
6879 if (vp == nd.ni_dvp)
6889 * Vnode operation to set a named attribute.
6892 zfs_setextattr(struct vop_setextattr_args *ap)
6895 IN struct vnode *a_vp;
6896 IN int a_attrnamespace;
6897 IN const char *a_name;
6898 INOUT struct uio *a_uio;
6899 IN struct ucred *a_cred;
6900 IN struct thread *a_td;
6904 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6905 struct thread *td = ap->a_td;
6906 struct nameidata nd;
6909 vnode_t *xvp = NULL, *vp;
6912 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6913 ap->a_cred, ap->a_td, VWRITE);
6917 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6924 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6925 LOOKUP_XATTR | CREATE_XATTR_DIR);
6931 flags = FFLAGS(O_WRONLY | O_CREAT);
6932 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6934 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL);
6936 NDFREE(&nd, NDF_ONLY_PNBUF);
6944 error = VOP_SETATTR(vp, &va, ap->a_cred);
6946 VOP_WRITE(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6949 vn_close(vp, flags, ap->a_cred, td);
6956 * Vnode operation to retrieve extended attributes on a vnode.
6959 zfs_listextattr(struct vop_listextattr_args *ap)
6962 IN struct vnode *a_vp;
6963 IN int a_attrnamespace;
6964 INOUT struct uio *a_uio;
6966 IN struct ucred *a_cred;
6967 IN struct thread *a_td;
6971 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6972 struct thread *td = ap->a_td;
6973 struct nameidata nd;
6974 char attrprefix[16];
6975 u_char dirbuf[sizeof(struct dirent)];
6978 struct uio auio, *uio = ap->a_uio;
6979 size_t *sizep = ap->a_size;
6981 vnode_t *xvp = NULL, *vp;
6982 int done, error, eof, pos;
6984 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6985 ap->a_cred, ap->a_td, VREAD);
6989 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix,
6990 sizeof(attrprefix));
6993 plen = strlen(attrprefix);
7000 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
7005 * ENOATTR means that the EA directory does not yet exist,
7006 * i.e. there are no extended attributes there.
7008 if (error == ENOATTR)
7013 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED,
7014 UIO_SYSSPACE, ".", xvp, td);
7017 NDFREE(&nd, NDF_ONLY_PNBUF);
7023 auio.uio_iov = &aiov;
7024 auio.uio_iovcnt = 1;
7025 auio.uio_segflg = UIO_SYSSPACE;
7027 auio.uio_rw = UIO_READ;
7028 auio.uio_offset = 0;
7033 aiov.iov_base = (void *)dirbuf;
7034 aiov.iov_len = sizeof(dirbuf);
7035 auio.uio_resid = sizeof(dirbuf);
7036 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL);
7037 done = sizeof(dirbuf) - auio.uio_resid;
7040 for (pos = 0; pos < done;) {
7041 dp = (struct dirent *)(dirbuf + pos);
7042 pos += dp->d_reclen;
7044 * XXX: Temporarily we also accept DT_UNKNOWN, as this
7045 * is what we get when attribute was created on Solaris.
7047 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN)
7049 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0)
7051 else if (strncmp(dp->d_name, attrprefix, plen) != 0)
7053 nlen = dp->d_namlen - plen;
7056 else if (uio != NULL) {
7058 * Format of extattr name entry is one byte for
7059 * length and the rest for name.
7061 error = uiomove(&nlen, 1, uio->uio_rw, uio);
7063 error = uiomove(dp->d_name + plen, nlen,
7070 } while (!eof && error == 0);
7079 zfs_freebsd_getacl(ap)
7080 struct vop_getacl_args /* {
7089 vsecattr_t vsecattr;
7091 if (ap->a_type != ACL_TYPE_NFS4)
7094 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
7095 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL))
7098 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt);
7099 if (vsecattr.vsa_aclentp != NULL)
7100 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz);
7106 zfs_freebsd_setacl(ap)
7107 struct vop_setacl_args /* {
7116 vsecattr_t vsecattr;
7117 int aclbsize; /* size of acl list in bytes */
7120 if (ap->a_type != ACL_TYPE_NFS4)
7123 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES)
7127 * With NFSv4 ACLs, chmod(2) may need to add additional entries,
7128 * splitting every entry into two and appending "canonical six"
7129 * entries at the end. Don't allow for setting an ACL that would
7130 * cause chmod(2) to run out of ACL entries.
7132 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES)
7135 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR);
7139 vsecattr.vsa_mask = VSA_ACE;
7140 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t);
7141 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP);
7142 aaclp = vsecattr.vsa_aclentp;
7143 vsecattr.vsa_aclentsz = aclbsize;
7145 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp);
7146 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL);
7147 kmem_free(aaclp, aclbsize);
7153 zfs_freebsd_aclcheck(ap)
7154 struct vop_aclcheck_args /* {
7163 return (EOPNOTSUPP);
7166 struct vop_vector zfs_vnodeops;
7167 struct vop_vector zfs_fifoops;
7168 struct vop_vector zfs_shareops;
7170 struct vop_vector zfs_vnodeops = {
7171 .vop_default = &default_vnodeops,
7172 .vop_inactive = zfs_freebsd_inactive,
7173 .vop_reclaim = zfs_freebsd_reclaim,
7174 .vop_access = zfs_freebsd_access,
7175 #ifdef FREEBSD_NAMECACHE
7176 .vop_lookup = vfs_cache_lookup,
7177 .vop_cachedlookup = zfs_freebsd_lookup,
7179 .vop_lookup = zfs_freebsd_lookup,
7181 .vop_getattr = zfs_freebsd_getattr,
7182 .vop_setattr = zfs_freebsd_setattr,
7183 .vop_create = zfs_freebsd_create,
7184 .vop_mknod = zfs_freebsd_create,
7185 .vop_mkdir = zfs_freebsd_mkdir,
7186 .vop_readdir = zfs_freebsd_readdir,
7187 .vop_fsync = zfs_freebsd_fsync,
7188 .vop_open = zfs_freebsd_open,
7189 .vop_close = zfs_freebsd_close,
7190 .vop_rmdir = zfs_freebsd_rmdir,
7191 .vop_ioctl = zfs_freebsd_ioctl,
7192 .vop_link = zfs_freebsd_link,
7193 .vop_symlink = zfs_freebsd_symlink,
7194 .vop_readlink = zfs_freebsd_readlink,
7195 .vop_read = zfs_freebsd_read,
7196 .vop_write = zfs_freebsd_write,
7197 .vop_remove = zfs_freebsd_remove,
7198 .vop_rename = zfs_freebsd_rename,
7199 .vop_pathconf = zfs_freebsd_pathconf,
7200 .vop_bmap = zfs_freebsd_bmap,
7201 .vop_fid = zfs_freebsd_fid,
7202 .vop_getextattr = zfs_getextattr,
7203 .vop_deleteextattr = zfs_deleteextattr,
7204 .vop_setextattr = zfs_setextattr,
7205 .vop_listextattr = zfs_listextattr,
7206 .vop_getacl = zfs_freebsd_getacl,
7207 .vop_setacl = zfs_freebsd_setacl,
7208 .vop_aclcheck = zfs_freebsd_aclcheck,
7209 .vop_getpages = zfs_freebsd_getpages,
7210 .vop_putpages = zfs_freebsd_putpages,
7213 struct vop_vector zfs_fifoops = {
7214 .vop_default = &fifo_specops,
7215 .vop_fsync = zfs_freebsd_fsync,
7216 .vop_access = zfs_freebsd_access,
7217 .vop_getattr = zfs_freebsd_getattr,
7218 .vop_inactive = zfs_freebsd_inactive,
7219 .vop_read = VOP_PANIC,
7220 .vop_reclaim = zfs_freebsd_reclaim,
7221 .vop_setattr = zfs_freebsd_setattr,
7222 .vop_write = VOP_PANIC,
7223 .vop_pathconf = zfs_freebsd_fifo_pathconf,
7224 .vop_fid = zfs_freebsd_fid,
7225 .vop_getacl = zfs_freebsd_getacl,
7226 .vop_setacl = zfs_freebsd_setacl,
7227 .vop_aclcheck = zfs_freebsd_aclcheck,
7231 * special share hidden files vnode operations template
7233 struct vop_vector zfs_shareops = {
7234 .vop_default = &default_vnodeops,
7235 .vop_access = zfs_freebsd_access,
7236 .vop_inactive = zfs_freebsd_inactive,
7237 .vop_reclaim = zfs_freebsd_reclaim,
7238 .vop_fid = zfs_freebsd_fid,
7239 .vop_pathconf = zfs_freebsd_pathconf,