4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2013 by Delphix. All rights reserved.
26 /* Portions Copyright 2007 Jeremy Teo */
27 /* Portions Copyright 2010 Robert Milkowski */
29 #include <sys/types.h>
30 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/sysmacros.h>
34 #include <sys/resource.h>
36 #include <sys/vnode.h>
40 #include <sys/taskq.h>
42 #include <sys/atomic.h>
43 #include <sys/namei.h>
45 #include <sys/cmn_err.h>
46 #include <sys/errno.h>
47 #include <sys/unistd.h>
48 #include <sys/zfs_dir.h>
49 #include <sys/zfs_ioctl.h>
50 #include <sys/fs/zfs.h>
52 #include <sys/dmu_objset.h>
58 #include <sys/dirent.h>
59 #include <sys/policy.h>
60 #include <sys/sunddi.h>
61 #include <sys/filio.h>
63 #include <sys/zfs_ctldir.h>
64 #include <sys/zfs_fuid.h>
65 #include <sys/zfs_sa.h>
67 #include <sys/zfs_rlock.h>
68 #include <sys/extdirent.h>
69 #include <sys/kidmap.h>
72 #include <sys/sf_buf.h>
73 #include <sys/sched.h>
75 #include <vm/vm_pageout.h>
80 * Each vnode op performs some logical unit of work. To do this, the ZPL must
81 * properly lock its in-core state, create a DMU transaction, do the work,
82 * record this work in the intent log (ZIL), commit the DMU transaction,
83 * and wait for the intent log to commit if it is a synchronous operation.
84 * Moreover, the vnode ops must work in both normal and log replay context.
85 * The ordering of events is important to avoid deadlocks and references
86 * to freed memory. The example below illustrates the following Big Rules:
88 * (1) A check must be made in each zfs thread for a mounted file system.
89 * This is done avoiding races using ZFS_ENTER(zfsvfs).
90 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
91 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
92 * can return EIO from the calling function.
94 * (2) VN_RELE() should always be the last thing except for zil_commit()
95 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
96 * First, if it's the last reference, the vnode/znode
97 * can be freed, so the zp may point to freed memory. Second, the last
98 * reference will call zfs_zinactive(), which may induce a lot of work --
99 * pushing cached pages (which acquires range locks) and syncing out
100 * cached atime changes. Third, zfs_zinactive() may require a new tx,
101 * which could deadlock the system if you were already holding one.
102 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
104 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
105 * as they can span dmu_tx_assign() calls.
107 * (4) Always pass TXG_NOWAIT as the second argument to dmu_tx_assign().
108 * This is critical because we don't want to block while holding locks.
109 * Note, in particular, that if a lock is sometimes acquired before
110 * the tx assigns, and sometimes after (e.g. z_lock), then failing to
111 * use a non-blocking assign can deadlock the system. The scenario:
113 * Thread A has grabbed a lock before calling dmu_tx_assign().
114 * Thread B is in an already-assigned tx, and blocks for this lock.
115 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
116 * forever, because the previous txg can't quiesce until B's tx commits.
118 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
119 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
120 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
121 * to indicate that this operation has already called dmu_tx_wait().
122 * This will ensure that we don't retry forever, waiting a short bit
125 * (5) If the operation succeeded, generate the intent log entry for it
126 * before dropping locks. This ensures that the ordering of events
127 * in the intent log matches the order in which they actually occurred.
128 * During ZIL replay the zfs_log_* functions will update the sequence
129 * number to indicate the zil transaction has replayed.
131 * (6) At the end of each vnode op, the DMU tx must always commit,
132 * regardless of whether there were any errors.
134 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
135 * to ensure that synchronous semantics are provided when necessary.
137 * In general, this is how things should be ordered in each vnode op:
139 * ZFS_ENTER(zfsvfs); // exit if unmounted
141 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD())
142 * rw_enter(...); // grab any other locks you need
143 * tx = dmu_tx_create(...); // get DMU tx
144 * dmu_tx_hold_*(); // hold each object you might modify
145 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
147 * rw_exit(...); // drop locks
148 * zfs_dirent_unlock(dl); // unlock directory entry
149 * VN_RELE(...); // release held vnodes
150 * if (error == ERESTART) {
156 * dmu_tx_abort(tx); // abort DMU tx
157 * ZFS_EXIT(zfsvfs); // finished in zfs
158 * return (error); // really out of space
160 * error = do_real_work(); // do whatever this VOP does
162 * zfs_log_*(...); // on success, make ZIL entry
163 * dmu_tx_commit(tx); // commit DMU tx -- error or not
164 * rw_exit(...); // drop locks
165 * zfs_dirent_unlock(dl); // unlock directory entry
166 * VN_RELE(...); // release held vnodes
167 * zil_commit(zilog, foid); // synchronous when necessary
168 * ZFS_EXIT(zfsvfs); // finished in zfs
169 * return (error); // done, report error
174 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
176 znode_t *zp = VTOZ(*vpp);
177 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
182 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
183 ((flag & FAPPEND) == 0)) {
185 return (SET_ERROR(EPERM));
188 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
189 ZTOV(zp)->v_type == VREG &&
190 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
191 if (fs_vscan(*vpp, cr, 0) != 0) {
193 return (SET_ERROR(EACCES));
197 /* Keep a count of the synchronous opens in the znode */
198 if (flag & (FSYNC | FDSYNC))
199 atomic_inc_32(&zp->z_sync_cnt);
207 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
208 caller_context_t *ct)
210 znode_t *zp = VTOZ(vp);
211 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
214 * Clean up any locks held by this process on the vp.
216 cleanlocks(vp, ddi_get_pid(), 0);
217 cleanshares(vp, ddi_get_pid());
222 /* Decrement the synchronous opens in the znode */
223 if ((flag & (FSYNC | FDSYNC)) && (count == 1))
224 atomic_dec_32(&zp->z_sync_cnt);
226 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
227 ZTOV(zp)->v_type == VREG &&
228 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
229 VERIFY(fs_vscan(vp, cr, 1) == 0);
236 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
237 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
240 zfs_holey(vnode_t *vp, u_long cmd, offset_t *off)
242 znode_t *zp = VTOZ(vp);
243 uint64_t noff = (uint64_t)*off; /* new offset */
248 file_sz = zp->z_size;
249 if (noff >= file_sz) {
250 return (SET_ERROR(ENXIO));
253 if (cmd == _FIO_SEEK_HOLE)
258 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
261 if ((error == ESRCH) || (noff > file_sz)) {
263 * Handle the virtual hole at the end of file.
269 return (SET_ERROR(ENXIO));
280 zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred,
281 int *rvalp, caller_context_t *ct)
293 * The following two ioctls are used by bfu. Faking out,
294 * necessary to avoid bfu errors.
303 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
304 return (SET_ERROR(EFAULT));
306 off = *(offset_t *)data;
309 zfsvfs = zp->z_zfsvfs;
313 /* offset parameter is in/out */
314 error = zfs_holey(vp, com, &off);
319 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
320 return (SET_ERROR(EFAULT));
322 *(offset_t *)data = off;
326 return (SET_ERROR(ENOTTY));
330 page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes)
337 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE
338 * aligned boundaries, if the range is not aligned. As a result a
339 * DEV_BSIZE subrange with partially dirty data may get marked as clean.
340 * It may happen that all DEV_BSIZE subranges are marked clean and thus
341 * the whole page would be considred clean despite have some dirty data.
342 * For this reason we should shrink the range to DEV_BSIZE aligned
343 * boundaries before calling vm_page_clear_dirty.
345 end = rounddown2(off + nbytes, DEV_BSIZE);
346 off = roundup2(off, DEV_BSIZE);
350 VM_OBJECT_LOCK_ASSERT(obj, MA_OWNED);
353 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
355 if ((pp->oflags & VPO_BUSY) != 0) {
357 * Reference the page before unlocking and
358 * sleeping so that the page daemon is less
359 * likely to reclaim it.
361 vm_page_reference(pp);
362 vm_page_sleep(pp, "zfsmwb");
365 } else if (pp == NULL) {
366 pp = vm_page_alloc(obj, OFF_TO_IDX(start),
367 VM_ALLOC_SYSTEM | VM_ALLOC_IFCACHED |
370 ASSERT(pp != NULL && !pp->valid);
375 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
376 vm_object_pip_add(obj, 1);
377 vm_page_io_start(pp);
378 pmap_remove_write(pp);
380 vm_page_clear_dirty(pp, off, nbytes);
388 page_unbusy(vm_page_t pp)
391 vm_page_io_finish(pp);
392 vm_object_pip_subtract(pp->object, 1);
396 page_hold(vnode_t *vp, int64_t start)
402 VM_OBJECT_LOCK_ASSERT(obj, MA_OWNED);
405 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
407 if ((pp->oflags & VPO_BUSY) != 0) {
409 * Reference the page before unlocking and
410 * sleeping so that the page daemon is less
411 * likely to reclaim it.
413 vm_page_reference(pp);
414 vm_page_sleep(pp, "zfsmwb");
418 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
431 page_unhold(vm_page_t pp)
440 zfs_map_page(vm_page_t pp, struct sf_buf **sfp)
443 *sfp = sf_buf_alloc(pp, 0);
444 return ((caddr_t)sf_buf_kva(*sfp));
448 zfs_unmap_page(struct sf_buf *sf)
455 * When a file is memory mapped, we must keep the IO data synchronized
456 * between the DMU cache and the memory mapped pages. What this means:
458 * On Write: If we find a memory mapped page, we write to *both*
459 * the page and the dmu buffer.
462 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid,
463 int segflg, dmu_tx_t *tx)
470 ASSERT(vp->v_mount != NULL);
474 off = start & PAGEOFFSET;
476 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
478 int nbytes = imin(PAGESIZE - off, len);
480 if (segflg == UIO_NOCOPY) {
481 pp = vm_page_lookup(obj, OFF_TO_IDX(start));
483 ("zfs update_pages: NULL page in putpages case"));
485 ("zfs update_pages: unaligned data in putpages case"));
486 KASSERT(pp->valid == VM_PAGE_BITS_ALL,
487 ("zfs update_pages: invalid page in putpages case"));
488 KASSERT(pp->busy > 0,
489 ("zfs update_pages: unbusy page in putpages case"));
490 KASSERT(!pmap_page_is_write_mapped(pp),
491 ("zfs update_pages: writable page in putpages case"));
492 VM_OBJECT_UNLOCK(obj);
494 va = zfs_map_page(pp, &sf);
495 (void) dmu_write(os, oid, start, nbytes, va, tx);
500 } else if ((pp = page_busy(vp, start, off, nbytes)) != NULL) {
501 VM_OBJECT_UNLOCK(obj);
503 va = zfs_map_page(pp, &sf);
504 (void) dmu_read(os, oid, start+off, nbytes,
505 va+off, DMU_READ_PREFETCH);;
514 if (segflg != UIO_NOCOPY)
515 vm_object_pip_wakeupn(obj, 0);
516 VM_OBJECT_UNLOCK(obj);
520 * Read with UIO_NOCOPY flag means that sendfile(2) requests
521 * ZFS to populate a range of page cache pages with data.
523 * NOTE: this function could be optimized to pre-allocate
524 * all pages in advance, drain VPO_BUSY on all of them,
525 * map them into contiguous KVA region and populate them
526 * in one single dmu_read() call.
529 mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio)
531 znode_t *zp = VTOZ(vp);
532 objset_t *os = zp->z_zfsvfs->z_os;
542 ASSERT(uio->uio_segflg == UIO_NOCOPY);
543 ASSERT(vp->v_mount != NULL);
546 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0);
549 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) {
550 int bytes = MIN(PAGESIZE, len);
552 pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_NOBUSY |
553 VM_ALLOC_NORMAL | VM_ALLOC_RETRY | VM_ALLOC_IGN_SBUSY);
554 if (pp->valid == 0) {
555 vm_page_io_start(pp);
556 VM_OBJECT_UNLOCK(obj);
557 va = zfs_map_page(pp, &sf);
558 error = dmu_read(os, zp->z_id, start, bytes, va,
560 if (bytes != PAGESIZE && error == 0)
561 bzero(va + bytes, PAGESIZE - bytes);
564 vm_page_io_finish(pp);
567 if (pp->wire_count == 0 && pp->valid == 0 &&
568 pp->busy == 0 && !(pp->oflags & VPO_BUSY))
571 pp->valid = VM_PAGE_BITS_ALL;
572 vm_page_activate(pp);
578 uio->uio_resid -= bytes;
579 uio->uio_offset += bytes;
582 VM_OBJECT_UNLOCK(obj);
587 * When a file is memory mapped, we must keep the IO data synchronized
588 * between the DMU cache and the memory mapped pages. What this means:
590 * On Read: We "read" preferentially from memory mapped pages,
591 * else we default from the dmu buffer.
593 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
594 * the file is memory mapped.
597 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
599 znode_t *zp = VTOZ(vp);
600 objset_t *os = zp->z_zfsvfs->z_os;
608 ASSERT(vp->v_mount != NULL);
612 start = uio->uio_loffset;
613 off = start & PAGEOFFSET;
615 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
617 uint64_t bytes = MIN(PAGESIZE - off, len);
619 if (pp = page_hold(vp, start)) {
623 VM_OBJECT_UNLOCK(obj);
624 va = zfs_map_page(pp, &sf);
625 error = uiomove(va + off, bytes, UIO_READ, uio);
630 VM_OBJECT_UNLOCK(obj);
631 error = dmu_read_uio(os, zp->z_id, uio, bytes);
639 VM_OBJECT_UNLOCK(obj);
643 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
646 * Read bytes from specified file into supplied buffer.
648 * IN: vp - vnode of file to be read from.
649 * uio - structure supplying read location, range info,
651 * ioflag - SYNC flags; used to provide FRSYNC semantics.
652 * cr - credentials of caller.
653 * ct - caller context
655 * OUT: uio - updated offset and range, buffer filled.
657 * RETURN: 0 on success, error code on failure.
660 * vp - atime updated if byte count > 0
664 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
666 znode_t *zp = VTOZ(vp);
667 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
678 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
680 return (SET_ERROR(EACCES));
684 * Validate file offset
686 if (uio->uio_loffset < (offset_t)0) {
688 return (SET_ERROR(EINVAL));
692 * Fasttrack empty reads
694 if (uio->uio_resid == 0) {
700 * Check for mandatory locks
702 if (MANDMODE(zp->z_mode)) {
703 if (error = chklock(vp, FREAD,
704 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
711 * If we're in FRSYNC mode, sync out this znode before reading it.
714 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
715 zil_commit(zfsvfs->z_log, zp->z_id);
718 * Lock the range against changes.
720 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
723 * If we are reading past end-of-file we can skip
724 * to the end; but we might still need to set atime.
726 if (uio->uio_loffset >= zp->z_size) {
731 ASSERT(uio->uio_loffset < zp->z_size);
732 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
735 if ((uio->uio_extflg == UIO_XUIO) &&
736 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
738 int blksz = zp->z_blksz;
739 uint64_t offset = uio->uio_loffset;
741 xuio = (xuio_t *)uio;
743 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
746 ASSERT(offset + n <= blksz);
749 (void) dmu_xuio_init(xuio, nblk);
751 if (vn_has_cached_data(vp)) {
753 * For simplicity, we always allocate a full buffer
754 * even if we only expect to read a portion of a block.
756 while (--nblk >= 0) {
757 (void) dmu_xuio_add(xuio,
758 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
766 nbytes = MIN(n, zfs_read_chunk_size -
767 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
770 if (uio->uio_segflg == UIO_NOCOPY)
771 error = mappedread_sf(vp, nbytes, uio);
773 #endif /* __FreeBSD__ */
774 if (vn_has_cached_data(vp))
775 error = mappedread(vp, nbytes, uio);
777 error = dmu_read_uio(os, zp->z_id, uio, nbytes);
779 /* convert checksum errors into IO errors */
781 error = SET_ERROR(EIO);
788 zfs_range_unlock(rl);
790 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
796 * Write the bytes to a file.
798 * IN: vp - vnode of file to be written to.
799 * uio - structure supplying write location, range info,
801 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is
802 * set if in append mode.
803 * cr - credentials of caller.
804 * ct - caller context (NFS/CIFS fem monitor only)
806 * OUT: uio - updated offset and range.
808 * RETURN: 0 on success, error code on failure.
811 * vp - ctime|mtime updated if byte count > 0
816 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
818 znode_t *zp = VTOZ(vp);
819 rlim64_t limit = MAXOFFSET_T;
820 ssize_t start_resid = uio->uio_resid;
824 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
829 int max_blksz = zfsvfs->z_max_blksz;
832 iovec_t *aiov = NULL;
835 int iovcnt = uio->uio_iovcnt;
836 iovec_t *iovp = uio->uio_iov;
839 sa_bulk_attr_t bulk[4];
840 uint64_t mtime[2], ctime[2];
843 * Fasttrack empty write
849 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
855 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
856 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
857 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
859 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
863 * If immutable or not appending then return EPERM
865 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
866 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
867 (uio->uio_loffset < zp->z_size))) {
869 return (SET_ERROR(EPERM));
872 zilog = zfsvfs->z_log;
875 * Validate file offset
877 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
880 return (SET_ERROR(EINVAL));
884 * Check for mandatory locks before calling zfs_range_lock()
885 * in order to prevent a deadlock with locks set via fcntl().
887 if (MANDMODE((mode_t)zp->z_mode) &&
888 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
895 * Pre-fault the pages to ensure slow (eg NFS) pages
897 * Skip this if uio contains loaned arc_buf.
899 if ((uio->uio_extflg == UIO_XUIO) &&
900 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
901 xuio = (xuio_t *)uio;
903 uio_prefaultpages(MIN(n, max_blksz), uio);
907 * If in append mode, set the io offset pointer to eof.
909 if (ioflag & FAPPEND) {
911 * Obtain an appending range lock to guarantee file append
912 * semantics. We reset the write offset once we have the lock.
914 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
916 if (rl->r_len == UINT64_MAX) {
918 * We overlocked the file because this write will cause
919 * the file block size to increase.
920 * Note that zp_size cannot change with this lock held.
924 uio->uio_loffset = woff;
927 * Note that if the file block size will change as a result of
928 * this write, then this range lock will lock the entire file
929 * so that we can re-write the block safely.
931 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
934 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) {
935 zfs_range_unlock(rl);
941 zfs_range_unlock(rl);
943 return (SET_ERROR(EFBIG));
946 if ((woff + n) > limit || woff > (limit - n))
949 /* Will this write extend the file length? */
950 write_eof = (woff + n > zp->z_size);
952 end_size = MAX(zp->z_size, woff + n);
955 * Write the file in reasonable size chunks. Each chunk is written
956 * in a separate transaction; this keeps the intent log records small
957 * and allows us to do more fine-grained space accounting.
961 woff = uio->uio_loffset;
963 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
964 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
966 dmu_return_arcbuf(abuf);
967 error = SET_ERROR(EDQUOT);
971 if (xuio && abuf == NULL) {
972 ASSERT(i_iov < iovcnt);
974 abuf = dmu_xuio_arcbuf(xuio, i_iov);
975 dmu_xuio_clear(xuio, i_iov);
976 DTRACE_PROBE3(zfs_cp_write, int, i_iov,
977 iovec_t *, aiov, arc_buf_t *, abuf);
978 ASSERT((aiov->iov_base == abuf->b_data) ||
979 ((char *)aiov->iov_base - (char *)abuf->b_data +
980 aiov->iov_len == arc_buf_size(abuf)));
982 } else if (abuf == NULL && n >= max_blksz &&
983 woff >= zp->z_size &&
984 P2PHASE(woff, max_blksz) == 0 &&
985 zp->z_blksz == max_blksz) {
987 * This write covers a full block. "Borrow" a buffer
988 * from the dmu so that we can fill it before we enter
989 * a transaction. This avoids the possibility of
990 * holding up the transaction if the data copy hangs
991 * up on a pagefault (e.g., from an NFS server mapping).
995 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
997 ASSERT(abuf != NULL);
998 ASSERT(arc_buf_size(abuf) == max_blksz);
999 if (error = uiocopy(abuf->b_data, max_blksz,
1000 UIO_WRITE, uio, &cbytes)) {
1001 dmu_return_arcbuf(abuf);
1004 ASSERT(cbytes == max_blksz);
1008 * Start a transaction.
1010 tx = dmu_tx_create(zfsvfs->z_os);
1011 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1012 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
1013 zfs_sa_upgrade_txholds(tx, zp);
1014 error = dmu_tx_assign(tx, TXG_NOWAIT);
1016 if (error == ERESTART) {
1023 dmu_return_arcbuf(abuf);
1028 * If zfs_range_lock() over-locked we grow the blocksize
1029 * and then reduce the lock range. This will only happen
1030 * on the first iteration since zfs_range_reduce() will
1031 * shrink down r_len to the appropriate size.
1033 if (rl->r_len == UINT64_MAX) {
1036 if (zp->z_blksz > max_blksz) {
1037 ASSERT(!ISP2(zp->z_blksz));
1038 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
1040 new_blksz = MIN(end_size, max_blksz);
1042 zfs_grow_blocksize(zp, new_blksz, tx);
1043 zfs_range_reduce(rl, woff, n);
1047 * XXX - should we really limit each write to z_max_blksz?
1048 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
1050 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
1052 if (woff + nbytes > zp->z_size)
1053 vnode_pager_setsize(vp, woff + nbytes);
1056 tx_bytes = uio->uio_resid;
1057 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
1059 tx_bytes -= uio->uio_resid;
1062 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
1064 * If this is not a full block write, but we are
1065 * extending the file past EOF and this data starts
1066 * block-aligned, use assign_arcbuf(). Otherwise,
1067 * write via dmu_write().
1069 if (tx_bytes < max_blksz && (!write_eof ||
1070 aiov->iov_base != abuf->b_data)) {
1072 dmu_write(zfsvfs->z_os, zp->z_id, woff,
1073 aiov->iov_len, aiov->iov_base, tx);
1074 dmu_return_arcbuf(abuf);
1075 xuio_stat_wbuf_copied();
1077 ASSERT(xuio || tx_bytes == max_blksz);
1078 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
1081 ASSERT(tx_bytes <= uio->uio_resid);
1082 uioskip(uio, tx_bytes);
1084 if (tx_bytes && vn_has_cached_data(vp)) {
1085 update_pages(vp, woff, tx_bytes, zfsvfs->z_os,
1086 zp->z_id, uio->uio_segflg, tx);
1090 * If we made no progress, we're done. If we made even
1091 * partial progress, update the znode and ZIL accordingly.
1093 if (tx_bytes == 0) {
1094 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
1095 (void *)&zp->z_size, sizeof (uint64_t), tx);
1102 * Clear Set-UID/Set-GID bits on successful write if not
1103 * privileged and at least one of the excute bits is set.
1105 * It would be nice to to this after all writes have
1106 * been done, but that would still expose the ISUID/ISGID
1107 * to another app after the partial write is committed.
1109 * Note: we don't call zfs_fuid_map_id() here because
1110 * user 0 is not an ephemeral uid.
1112 mutex_enter(&zp->z_acl_lock);
1113 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
1114 (S_IXUSR >> 6))) != 0 &&
1115 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
1116 secpolicy_vnode_setid_retain(vp, cr,
1117 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
1119 zp->z_mode &= ~(S_ISUID | S_ISGID);
1120 newmode = zp->z_mode;
1121 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
1122 (void *)&newmode, sizeof (uint64_t), tx);
1124 mutex_exit(&zp->z_acl_lock);
1126 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
1130 * Update the file size (zp_size) if it has changed;
1131 * account for possible concurrent updates.
1133 while ((end_size = zp->z_size) < uio->uio_loffset) {
1134 (void) atomic_cas_64(&zp->z_size, end_size,
1139 * If we are replaying and eof is non zero then force
1140 * the file size to the specified eof. Note, there's no
1141 * concurrency during replay.
1143 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
1144 zp->z_size = zfsvfs->z_replay_eof;
1146 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1148 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
1153 ASSERT(tx_bytes == nbytes);
1158 uio_prefaultpages(MIN(n, max_blksz), uio);
1162 zfs_range_unlock(rl);
1165 * If we're in replay mode, or we made no progress, return error.
1166 * Otherwise, it's at least a partial write, so it's successful.
1168 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1173 if (ioflag & (FSYNC | FDSYNC) ||
1174 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1175 zil_commit(zilog, zp->z_id);
1182 zfs_get_done(zgd_t *zgd, int error)
1184 znode_t *zp = zgd->zgd_private;
1185 objset_t *os = zp->z_zfsvfs->z_os;
1189 dmu_buf_rele(zgd->zgd_db, zgd);
1191 zfs_range_unlock(zgd->zgd_rl);
1193 vfslocked = VFS_LOCK_GIANT(zp->z_zfsvfs->z_vfs);
1195 * Release the vnode asynchronously as we currently have the
1196 * txg stopped from syncing.
1198 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1200 if (error == 0 && zgd->zgd_bp)
1201 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1203 kmem_free(zgd, sizeof (zgd_t));
1204 VFS_UNLOCK_GIANT(vfslocked);
1208 static int zil_fault_io = 0;
1212 * Get data to generate a TX_WRITE intent log record.
1215 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1217 zfsvfs_t *zfsvfs = arg;
1218 objset_t *os = zfsvfs->z_os;
1220 uint64_t object = lr->lr_foid;
1221 uint64_t offset = lr->lr_offset;
1222 uint64_t size = lr->lr_length;
1223 blkptr_t *bp = &lr->lr_blkptr;
1228 ASSERT(zio != NULL);
1232 * Nothing to do if the file has been removed
1234 if (zfs_zget(zfsvfs, object, &zp) != 0)
1235 return (SET_ERROR(ENOENT));
1236 if (zp->z_unlinked) {
1238 * Release the vnode asynchronously as we currently have the
1239 * txg stopped from syncing.
1241 VN_RELE_ASYNC(ZTOV(zp),
1242 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1243 return (SET_ERROR(ENOENT));
1246 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1247 zgd->zgd_zilog = zfsvfs->z_log;
1248 zgd->zgd_private = zp;
1251 * Write records come in two flavors: immediate and indirect.
1252 * For small writes it's cheaper to store the data with the
1253 * log record (immediate); for large writes it's cheaper to
1254 * sync the data and get a pointer to it (indirect) so that
1255 * we don't have to write the data twice.
1257 if (buf != NULL) { /* immediate write */
1258 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1259 /* test for truncation needs to be done while range locked */
1260 if (offset >= zp->z_size) {
1261 error = SET_ERROR(ENOENT);
1263 error = dmu_read(os, object, offset, size, buf,
1264 DMU_READ_NO_PREFETCH);
1266 ASSERT(error == 0 || error == ENOENT);
1267 } else { /* indirect write */
1269 * Have to lock the whole block to ensure when it's
1270 * written out and it's checksum is being calculated
1271 * that no one can change the data. We need to re-check
1272 * blocksize after we get the lock in case it's changed!
1277 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1279 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1281 if (zp->z_blksz == size)
1284 zfs_range_unlock(zgd->zgd_rl);
1286 /* test for truncation needs to be done while range locked */
1287 if (lr->lr_offset >= zp->z_size)
1288 error = SET_ERROR(ENOENT);
1291 error = SET_ERROR(EIO);
1296 error = dmu_buf_hold(os, object, offset, zgd, &db,
1297 DMU_READ_NO_PREFETCH);
1300 blkptr_t *obp = dmu_buf_get_blkptr(db);
1302 ASSERT(BP_IS_HOLE(bp));
1309 ASSERT(db->db_offset == offset);
1310 ASSERT(db->db_size == size);
1312 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1314 ASSERT(error || lr->lr_length <= zp->z_blksz);
1317 * On success, we need to wait for the write I/O
1318 * initiated by dmu_sync() to complete before we can
1319 * release this dbuf. We will finish everything up
1320 * in the zfs_get_done() callback.
1325 if (error == EALREADY) {
1326 lr->lr_common.lrc_txtype = TX_WRITE2;
1332 zfs_get_done(zgd, error);
1339 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1340 caller_context_t *ct)
1342 znode_t *zp = VTOZ(vp);
1343 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1349 if (flag & V_ACE_MASK)
1350 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1352 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1359 * If vnode is for a device return a specfs vnode instead.
1362 specvp_check(vnode_t **vpp, cred_t *cr)
1366 if (IS_DEVVP(*vpp)) {
1369 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1372 error = SET_ERROR(ENOSYS);
1380 * Lookup an entry in a directory, or an extended attribute directory.
1381 * If it exists, return a held vnode reference for it.
1383 * IN: dvp - vnode of directory to search.
1384 * nm - name of entry to lookup.
1385 * pnp - full pathname to lookup [UNUSED].
1386 * flags - LOOKUP_XATTR set if looking for an attribute.
1387 * rdir - root directory vnode [UNUSED].
1388 * cr - credentials of caller.
1389 * ct - caller context
1390 * direntflags - directory lookup flags
1391 * realpnp - returned pathname.
1393 * OUT: vpp - vnode of located entry, NULL if not found.
1395 * RETURN: 0 on success, error code on failure.
1402 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp,
1403 int nameiop, cred_t *cr, kthread_t *td, int flags)
1405 znode_t *zdp = VTOZ(dvp);
1406 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1408 int *direntflags = NULL;
1409 void *realpnp = NULL;
1412 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1414 if (dvp->v_type != VDIR) {
1415 return (SET_ERROR(ENOTDIR));
1416 } else if (zdp->z_sa_hdl == NULL) {
1417 return (SET_ERROR(EIO));
1420 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1421 error = zfs_fastaccesschk_execute(zdp, cr);
1429 vnode_t *tvp = dnlc_lookup(dvp, nm);
1432 error = zfs_fastaccesschk_execute(zdp, cr);
1437 if (tvp == DNLC_NO_VNODE) {
1439 return (SET_ERROR(ENOENT));
1442 return (specvp_check(vpp, cr));
1448 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1455 if (flags & LOOKUP_XATTR) {
1458 * If the xattr property is off, refuse the lookup request.
1460 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1462 return (SET_ERROR(EINVAL));
1467 * We don't allow recursive attributes..
1468 * Maybe someday we will.
1470 if (zdp->z_pflags & ZFS_XATTR) {
1472 return (SET_ERROR(EINVAL));
1475 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1481 * Do we have permission to get into attribute directory?
1484 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1494 if (dvp->v_type != VDIR) {
1496 return (SET_ERROR(ENOTDIR));
1500 * Check accessibility of directory.
1503 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1508 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1509 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1511 return (SET_ERROR(EILSEQ));
1514 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1516 error = specvp_check(vpp, cr);
1518 /* Translate errors and add SAVENAME when needed. */
1519 if (cnp->cn_flags & ISLASTCN) {
1523 if (error == ENOENT) {
1524 error = EJUSTRETURN;
1525 cnp->cn_flags |= SAVENAME;
1531 cnp->cn_flags |= SAVENAME;
1535 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) {
1538 if (cnp->cn_flags & ISDOTDOT) {
1539 ltype = VOP_ISLOCKED(dvp);
1543 error = vn_lock(*vpp, cnp->cn_lkflags);
1544 if (cnp->cn_flags & ISDOTDOT)
1545 vn_lock(dvp, ltype | LK_RETRY);
1555 #ifdef FREEBSD_NAMECACHE
1557 * Insert name into cache (as non-existent) if appropriate.
1559 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
1560 cache_enter(dvp, *vpp, cnp);
1562 * Insert name into cache if appropriate.
1564 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) {
1565 if (!(cnp->cn_flags & ISLASTCN) ||
1566 (nameiop != DELETE && nameiop != RENAME)) {
1567 cache_enter(dvp, *vpp, cnp);
1576 * Attempt to create a new entry in a directory. If the entry
1577 * already exists, truncate the file if permissible, else return
1578 * an error. Return the vp of the created or trunc'd file.
1580 * IN: dvp - vnode of directory to put new file entry in.
1581 * name - name of new file entry.
1582 * vap - attributes of new file.
1583 * excl - flag indicating exclusive or non-exclusive mode.
1584 * mode - mode to open file with.
1585 * cr - credentials of caller.
1586 * flag - large file flag [UNUSED].
1587 * ct - caller context
1588 * vsecp - ACL to be set
1590 * OUT: vpp - vnode of created or trunc'd entry.
1592 * RETURN: 0 on success, error code on failure.
1595 * dvp - ctime|mtime updated if new entry created
1596 * vp - ctime|mtime always, atime if new
1601 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode,
1602 vnode_t **vpp, cred_t *cr, kthread_t *td)
1604 znode_t *zp, *dzp = VTOZ(dvp);
1605 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1613 gid_t gid = crgetgid(cr);
1614 zfs_acl_ids_t acl_ids;
1615 boolean_t fuid_dirtied;
1616 boolean_t have_acl = B_FALSE;
1617 boolean_t waited = B_FALSE;
1622 * If we have an ephemeral id, ACL, or XVATTR then
1623 * make sure file system is at proper version
1626 ksid = crgetsid(cr, KSID_OWNER);
1628 uid = ksid_getid(ksid);
1632 if (zfsvfs->z_use_fuids == B_FALSE &&
1633 (vsecp || (vap->va_mask & AT_XVATTR) ||
1634 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1635 return (SET_ERROR(EINVAL));
1640 zilog = zfsvfs->z_log;
1642 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1643 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1645 return (SET_ERROR(EILSEQ));
1648 if (vap->va_mask & AT_XVATTR) {
1649 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
1650 crgetuid(cr), cr, vap->va_type)) != 0) {
1658 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr))
1659 vap->va_mode &= ~S_ISVTX;
1661 if (*name == '\0') {
1663 * Null component name refers to the directory itself.
1670 /* possible VN_HOLD(zp) */
1673 if (flag & FIGNORECASE)
1676 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1680 zfs_acl_ids_free(&acl_ids);
1681 if (strcmp(name, "..") == 0)
1682 error = SET_ERROR(EISDIR);
1692 * Create a new file object and update the directory
1695 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1697 zfs_acl_ids_free(&acl_ids);
1702 * We only support the creation of regular files in
1703 * extended attribute directories.
1706 if ((dzp->z_pflags & ZFS_XATTR) &&
1707 (vap->va_type != VREG)) {
1709 zfs_acl_ids_free(&acl_ids);
1710 error = SET_ERROR(EINVAL);
1714 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1715 cr, vsecp, &acl_ids)) != 0)
1719 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1720 zfs_acl_ids_free(&acl_ids);
1721 error = SET_ERROR(EDQUOT);
1725 tx = dmu_tx_create(os);
1727 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1728 ZFS_SA_BASE_ATTR_SIZE);
1730 fuid_dirtied = zfsvfs->z_fuid_dirty;
1732 zfs_fuid_txhold(zfsvfs, tx);
1733 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1734 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1735 if (!zfsvfs->z_use_sa &&
1736 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1737 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1738 0, acl_ids.z_aclp->z_acl_bytes);
1740 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1742 zfs_dirent_unlock(dl);
1743 if (error == ERESTART) {
1749 zfs_acl_ids_free(&acl_ids);
1754 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1757 zfs_fuid_sync(zfsvfs, tx);
1759 (void) zfs_link_create(dl, zp, tx, ZNEW);
1760 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1761 if (flag & FIGNORECASE)
1763 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1764 vsecp, acl_ids.z_fuidp, vap);
1765 zfs_acl_ids_free(&acl_ids);
1768 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1771 zfs_acl_ids_free(&acl_ids);
1775 * A directory entry already exists for this name.
1778 * Can't truncate an existing file if in exclusive mode.
1781 error = SET_ERROR(EEXIST);
1785 * Can't open a directory for writing.
1787 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1788 error = SET_ERROR(EISDIR);
1792 * Verify requested access to file.
1794 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1798 mutex_enter(&dzp->z_lock);
1800 mutex_exit(&dzp->z_lock);
1803 * Truncate regular files if requested.
1805 if ((ZTOV(zp)->v_type == VREG) &&
1806 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1807 /* we can't hold any locks when calling zfs_freesp() */
1808 zfs_dirent_unlock(dl);
1810 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1812 vnevent_create(ZTOV(zp), ct);
1818 zfs_dirent_unlock(dl);
1825 error = specvp_check(vpp, cr);
1828 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1829 zil_commit(zilog, 0);
1836 * Remove an entry from a directory.
1838 * IN: dvp - vnode of directory to remove entry from.
1839 * name - name of entry to remove.
1840 * cr - credentials of caller.
1841 * ct - caller context
1842 * flags - case flags
1844 * RETURN: 0 on success, error code on failure.
1848 * vp - ctime (if nlink > 0)
1851 uint64_t null_xattr = 0;
1855 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1858 znode_t *zp, *dzp = VTOZ(dvp);
1861 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1863 uint64_t acl_obj, xattr_obj;
1864 uint64_t xattr_obj_unlinked = 0;
1868 boolean_t may_delete_now, delete_now = FALSE;
1869 boolean_t unlinked, toobig = FALSE;
1871 pathname_t *realnmp = NULL;
1875 boolean_t waited = B_FALSE;
1879 zilog = zfsvfs->z_log;
1881 if (flags & FIGNORECASE) {
1891 * Attempt to lock directory; fail if entry doesn't exist.
1893 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1903 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1908 * Need to use rmdir for removing directories.
1910 if (vp->v_type == VDIR) {
1911 error = SET_ERROR(EPERM);
1915 vnevent_remove(vp, dvp, name, ct);
1918 dnlc_remove(dvp, realnmp->pn_buf);
1920 dnlc_remove(dvp, name);
1923 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1927 * We may delete the znode now, or we may put it in the unlinked set;
1928 * it depends on whether we're the last link, and on whether there are
1929 * other holds on the vnode. So we dmu_tx_hold() the right things to
1930 * allow for either case.
1933 tx = dmu_tx_create(zfsvfs->z_os);
1934 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1935 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1936 zfs_sa_upgrade_txholds(tx, zp);
1937 zfs_sa_upgrade_txholds(tx, dzp);
1938 if (may_delete_now) {
1940 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1941 /* if the file is too big, only hold_free a token amount */
1942 dmu_tx_hold_free(tx, zp->z_id, 0,
1943 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1946 /* are there any extended attributes? */
1947 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1948 &xattr_obj, sizeof (xattr_obj));
1949 if (error == 0 && xattr_obj) {
1950 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1952 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1953 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1956 mutex_enter(&zp->z_lock);
1957 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1958 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1959 mutex_exit(&zp->z_lock);
1961 /* charge as an update -- would be nice not to charge at all */
1962 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1964 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1966 zfs_dirent_unlock(dl);
1970 if (error == ERESTART) {
1984 * Remove the directory entry.
1986 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1996 * Hold z_lock so that we can make sure that the ACL obj
1997 * hasn't changed. Could have been deleted due to
2000 mutex_enter(&zp->z_lock);
2002 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
2003 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
2004 delete_now = may_delete_now && !toobig &&
2005 vp->v_count == 1 && !vn_has_cached_data(vp) &&
2006 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
2013 panic("zfs_remove: delete_now branch taken");
2015 if (xattr_obj_unlinked) {
2016 ASSERT3U(xzp->z_links, ==, 2);
2017 mutex_enter(&xzp->z_lock);
2018 xzp->z_unlinked = 1;
2020 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
2021 &xzp->z_links, sizeof (xzp->z_links), tx);
2022 ASSERT3U(error, ==, 0);
2023 mutex_exit(&xzp->z_lock);
2024 zfs_unlinked_add(xzp, tx);
2027 error = sa_remove(zp->z_sa_hdl,
2028 SA_ZPL_XATTR(zfsvfs), tx);
2030 error = sa_update(zp->z_sa_hdl,
2031 SA_ZPL_XATTR(zfsvfs), &null_xattr,
2032 sizeof (uint64_t), tx);
2037 ASSERT0(vp->v_count);
2039 mutex_exit(&zp->z_lock);
2040 zfs_znode_delete(zp, tx);
2041 } else if (unlinked) {
2042 mutex_exit(&zp->z_lock);
2043 zfs_unlinked_add(zp, tx);
2045 vp->v_vflag |= VV_NOSYNC;
2050 if (flags & FIGNORECASE)
2052 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
2059 zfs_dirent_unlock(dl);
2066 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2067 zil_commit(zilog, 0);
2074 * Create a new directory and insert it into dvp using the name
2075 * provided. Return a pointer to the inserted directory.
2077 * IN: dvp - vnode of directory to add subdir to.
2078 * dirname - name of new directory.
2079 * vap - attributes of new directory.
2080 * cr - credentials of caller.
2081 * ct - caller context
2082 * flags - case flags
2083 * vsecp - ACL to be set
2085 * OUT: vpp - vnode of created directory.
2087 * RETURN: 0 on success, error code on failure.
2090 * dvp - ctime|mtime updated
2091 * vp - ctime|mtime|atime updated
2095 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
2096 caller_context_t *ct, int flags, vsecattr_t *vsecp)
2098 znode_t *zp, *dzp = VTOZ(dvp);
2099 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2108 gid_t gid = crgetgid(cr);
2109 zfs_acl_ids_t acl_ids;
2110 boolean_t fuid_dirtied;
2111 boolean_t waited = B_FALSE;
2113 ASSERT(vap->va_type == VDIR);
2116 * If we have an ephemeral id, ACL, or XVATTR then
2117 * make sure file system is at proper version
2120 ksid = crgetsid(cr, KSID_OWNER);
2122 uid = ksid_getid(ksid);
2125 if (zfsvfs->z_use_fuids == B_FALSE &&
2126 (vsecp || (vap->va_mask & AT_XVATTR) ||
2127 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
2128 return (SET_ERROR(EINVAL));
2132 zilog = zfsvfs->z_log;
2134 if (dzp->z_pflags & ZFS_XATTR) {
2136 return (SET_ERROR(EINVAL));
2139 if (zfsvfs->z_utf8 && u8_validate(dirname,
2140 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2142 return (SET_ERROR(EILSEQ));
2144 if (flags & FIGNORECASE)
2147 if (vap->va_mask & AT_XVATTR) {
2148 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
2149 crgetuid(cr), cr, vap->va_type)) != 0) {
2155 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
2156 vsecp, &acl_ids)) != 0) {
2161 * First make sure the new directory doesn't exist.
2163 * Existence is checked first to make sure we don't return
2164 * EACCES instead of EEXIST which can cause some applications
2170 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
2172 zfs_acl_ids_free(&acl_ids);
2177 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
2178 zfs_acl_ids_free(&acl_ids);
2179 zfs_dirent_unlock(dl);
2184 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
2185 zfs_acl_ids_free(&acl_ids);
2186 zfs_dirent_unlock(dl);
2188 return (SET_ERROR(EDQUOT));
2192 * Add a new entry to the directory.
2194 tx = dmu_tx_create(zfsvfs->z_os);
2195 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2196 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2197 fuid_dirtied = zfsvfs->z_fuid_dirty;
2199 zfs_fuid_txhold(zfsvfs, tx);
2200 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2201 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2202 acl_ids.z_aclp->z_acl_bytes);
2205 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2206 ZFS_SA_BASE_ATTR_SIZE);
2208 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2210 zfs_dirent_unlock(dl);
2211 if (error == ERESTART) {
2217 zfs_acl_ids_free(&acl_ids);
2226 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2229 zfs_fuid_sync(zfsvfs, tx);
2232 * Now put new name in parent dir.
2234 (void) zfs_link_create(dl, zp, tx, ZNEW);
2238 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2239 if (flags & FIGNORECASE)
2241 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2242 acl_ids.z_fuidp, vap);
2244 zfs_acl_ids_free(&acl_ids);
2248 zfs_dirent_unlock(dl);
2250 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2251 zil_commit(zilog, 0);
2258 * Remove a directory subdir entry. If the current working
2259 * directory is the same as the subdir to be removed, the
2262 * IN: dvp - vnode of directory to remove from.
2263 * name - name of directory to be removed.
2264 * cwd - vnode of current working directory.
2265 * cr - credentials of caller.
2266 * ct - caller context
2267 * flags - case flags
2269 * RETURN: 0 on success, error code on failure.
2272 * dvp - ctime|mtime updated
2276 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2277 caller_context_t *ct, int flags)
2279 znode_t *dzp = VTOZ(dvp);
2282 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2288 boolean_t waited = B_FALSE;
2292 zilog = zfsvfs->z_log;
2294 if (flags & FIGNORECASE)
2300 * Attempt to lock directory; fail if entry doesn't exist.
2302 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2310 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2314 if (vp->v_type != VDIR) {
2315 error = SET_ERROR(ENOTDIR);
2320 error = SET_ERROR(EINVAL);
2324 vnevent_rmdir(vp, dvp, name, ct);
2327 * Grab a lock on the directory to make sure that noone is
2328 * trying to add (or lookup) entries while we are removing it.
2330 rw_enter(&zp->z_name_lock, RW_WRITER);
2333 * Grab a lock on the parent pointer to make sure we play well
2334 * with the treewalk and directory rename code.
2336 rw_enter(&zp->z_parent_lock, RW_WRITER);
2338 tx = dmu_tx_create(zfsvfs->z_os);
2339 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2340 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2341 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2342 zfs_sa_upgrade_txholds(tx, zp);
2343 zfs_sa_upgrade_txholds(tx, dzp);
2344 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2346 rw_exit(&zp->z_parent_lock);
2347 rw_exit(&zp->z_name_lock);
2348 zfs_dirent_unlock(dl);
2350 if (error == ERESTART) {
2361 #ifdef FREEBSD_NAMECACHE
2365 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2368 uint64_t txtype = TX_RMDIR;
2369 if (flags & FIGNORECASE)
2371 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2376 rw_exit(&zp->z_parent_lock);
2377 rw_exit(&zp->z_name_lock);
2378 #ifdef FREEBSD_NAMECACHE
2382 zfs_dirent_unlock(dl);
2386 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2387 zil_commit(zilog, 0);
2394 * Read as many directory entries as will fit into the provided
2395 * buffer from the given directory cursor position (specified in
2396 * the uio structure).
2398 * IN: vp - vnode of directory to read.
2399 * uio - structure supplying read location, range info,
2400 * and return buffer.
2401 * cr - credentials of caller.
2402 * ct - caller context
2403 * flags - case flags
2405 * OUT: uio - updated offset and range, buffer filled.
2406 * eofp - set to true if end-of-file detected.
2408 * RETURN: 0 on success, error code on failure.
2411 * vp - atime updated
2413 * Note that the low 4 bits of the cookie returned by zap is always zero.
2414 * This allows us to use the low range for "special" directory entries:
2415 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2416 * we use the offset 2 for the '.zfs' directory.
2420 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies)
2422 znode_t *zp = VTOZ(vp);
2426 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2431 zap_attribute_t zap;
2432 uint_t bytes_wanted;
2433 uint64_t offset; /* must be unsigned; checks for < 1 */
2439 boolean_t check_sysattrs;
2442 u_long *cooks = NULL;
2448 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2449 &parent, sizeof (parent))) != 0) {
2455 * If we are not given an eof variable,
2462 * Check for valid iov_len.
2464 if (uio->uio_iov->iov_len <= 0) {
2466 return (SET_ERROR(EINVAL));
2470 * Quit if directory has been removed (posix)
2472 if ((*eofp = zp->z_unlinked) != 0) {
2479 offset = uio->uio_loffset;
2480 prefetch = zp->z_zn_prefetch;
2483 * Initialize the iterator cursor.
2487 * Start iteration from the beginning of the directory.
2489 zap_cursor_init(&zc, os, zp->z_id);
2492 * The offset is a serialized cursor.
2494 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2498 * Get space to change directory entries into fs independent format.
2500 iovp = uio->uio_iov;
2501 bytes_wanted = iovp->iov_len;
2502 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2503 bufsize = bytes_wanted;
2504 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2505 odp = (struct dirent64 *)outbuf;
2507 bufsize = bytes_wanted;
2509 odp = (struct dirent64 *)iovp->iov_base;
2511 eodp = (struct edirent *)odp;
2513 if (ncookies != NULL) {
2515 * Minimum entry size is dirent size and 1 byte for a file name.
2517 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1);
2518 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK);
2523 * If this VFS supports the system attribute view interface; and
2524 * we're looking at an extended attribute directory; and we care
2525 * about normalization conflicts on this vfs; then we must check
2526 * for normalization conflicts with the sysattr name space.
2529 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2530 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2531 (flags & V_RDDIR_ENTFLAGS);
2537 * Transform to file-system independent format
2540 while (outcount < bytes_wanted) {
2543 off64_t *next = NULL;
2546 * Special case `.', `..', and `.zfs'.
2549 (void) strcpy(zap.za_name, ".");
2550 zap.za_normalization_conflict = 0;
2553 } else if (offset == 1) {
2554 (void) strcpy(zap.za_name, "..");
2555 zap.za_normalization_conflict = 0;
2558 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2559 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2560 zap.za_normalization_conflict = 0;
2561 objnum = ZFSCTL_INO_ROOT;
2567 if (error = zap_cursor_retrieve(&zc, &zap)) {
2568 if ((*eofp = (error == ENOENT)) != 0)
2574 if (zap.za_integer_length != 8 ||
2575 zap.za_num_integers != 1) {
2576 cmn_err(CE_WARN, "zap_readdir: bad directory "
2577 "entry, obj = %lld, offset = %lld\n",
2578 (u_longlong_t)zp->z_id,
2579 (u_longlong_t)offset);
2580 error = SET_ERROR(ENXIO);
2584 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2586 * MacOS X can extract the object type here such as:
2587 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2589 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2591 if (check_sysattrs && !zap.za_normalization_conflict) {
2593 zap.za_normalization_conflict =
2594 xattr_sysattr_casechk(zap.za_name);
2596 panic("%s:%u: TODO", __func__, __LINE__);
2601 if (flags & V_RDDIR_ACCFILTER) {
2603 * If we have no access at all, don't include
2604 * this entry in the returned information
2607 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2609 if (!zfs_has_access(ezp, cr)) {
2616 if (flags & V_RDDIR_ENTFLAGS)
2617 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2619 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2622 * Will this entry fit in the buffer?
2624 if (outcount + reclen > bufsize) {
2626 * Did we manage to fit anything in the buffer?
2629 error = SET_ERROR(EINVAL);
2634 if (flags & V_RDDIR_ENTFLAGS) {
2636 * Add extended flag entry:
2638 eodp->ed_ino = objnum;
2639 eodp->ed_reclen = reclen;
2640 /* NOTE: ed_off is the offset for the *next* entry */
2641 next = &(eodp->ed_off);
2642 eodp->ed_eflags = zap.za_normalization_conflict ?
2643 ED_CASE_CONFLICT : 0;
2644 (void) strncpy(eodp->ed_name, zap.za_name,
2645 EDIRENT_NAMELEN(reclen));
2646 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2651 odp->d_ino = objnum;
2652 odp->d_reclen = reclen;
2653 odp->d_namlen = strlen(zap.za_name);
2654 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1);
2656 odp = (dirent64_t *)((intptr_t)odp + reclen);
2660 ASSERT(outcount <= bufsize);
2662 /* Prefetch znode */
2664 dmu_prefetch(os, objnum, 0, 0);
2668 * Move to the next entry, fill in the previous offset.
2670 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2671 zap_cursor_advance(&zc);
2672 offset = zap_cursor_serialize(&zc);
2677 if (cooks != NULL) {
2680 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks));
2683 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2685 /* Subtract unused cookies */
2686 if (ncookies != NULL)
2687 *ncookies -= ncooks;
2689 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2690 iovp->iov_base += outcount;
2691 iovp->iov_len -= outcount;
2692 uio->uio_resid -= outcount;
2693 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2695 * Reset the pointer.
2697 offset = uio->uio_loffset;
2701 zap_cursor_fini(&zc);
2702 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2703 kmem_free(outbuf, bufsize);
2705 if (error == ENOENT)
2708 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2710 uio->uio_loffset = offset;
2712 if (error != 0 && cookies != NULL) {
2713 free(*cookies, M_TEMP);
2720 ulong_t zfs_fsync_sync_cnt = 4;
2723 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2725 znode_t *zp = VTOZ(vp);
2726 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2728 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2730 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2733 zil_commit(zfsvfs->z_log, zp->z_id);
2741 * Get the requested file attributes and place them in the provided
2744 * IN: vp - vnode of file.
2745 * vap - va_mask identifies requested attributes.
2746 * If AT_XVATTR set, then optional attrs are requested
2747 * flags - ATTR_NOACLCHECK (CIFS server context)
2748 * cr - credentials of caller.
2749 * ct - caller context
2751 * OUT: vap - attribute values.
2753 * RETURN: 0 (always succeeds).
2757 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2758 caller_context_t *ct)
2760 znode_t *zp = VTOZ(vp);
2761 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2764 u_longlong_t nblocks;
2766 uint64_t mtime[2], ctime[2], crtime[2], rdev;
2767 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2768 xoptattr_t *xoap = NULL;
2769 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2770 sa_bulk_attr_t bulk[4];
2776 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2778 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2779 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2780 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
2781 if (vp->v_type == VBLK || vp->v_type == VCHR)
2782 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
2785 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2791 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2792 * Also, if we are the owner don't bother, since owner should
2793 * always be allowed to read basic attributes of file.
2795 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2796 (vap->va_uid != crgetuid(cr))) {
2797 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2805 * Return all attributes. It's cheaper to provide the answer
2806 * than to determine whether we were asked the question.
2809 mutex_enter(&zp->z_lock);
2810 vap->va_type = IFTOVT(zp->z_mode);
2811 vap->va_mode = zp->z_mode & ~S_IFMT;
2813 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2815 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
2817 vap->va_nodeid = zp->z_id;
2818 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2819 links = zp->z_links + 1;
2821 links = zp->z_links;
2822 vap->va_nlink = MIN(links, LINK_MAX); /* nlink_t limit! */
2823 vap->va_size = zp->z_size;
2825 vap->va_rdev = vp->v_rdev;
2827 if (vp->v_type == VBLK || vp->v_type == VCHR)
2828 vap->va_rdev = zfs_cmpldev(rdev);
2830 vap->va_seq = zp->z_seq;
2831 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */
2834 * Add in any requested optional attributes and the create time.
2835 * Also set the corresponding bits in the returned attribute bitmap.
2837 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2838 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2840 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2841 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2844 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2845 xoap->xoa_readonly =
2846 ((zp->z_pflags & ZFS_READONLY) != 0);
2847 XVA_SET_RTN(xvap, XAT_READONLY);
2850 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2852 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2853 XVA_SET_RTN(xvap, XAT_SYSTEM);
2856 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2858 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2859 XVA_SET_RTN(xvap, XAT_HIDDEN);
2862 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2863 xoap->xoa_nounlink =
2864 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2865 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2868 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2869 xoap->xoa_immutable =
2870 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2871 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2874 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2875 xoap->xoa_appendonly =
2876 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2877 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2880 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2882 ((zp->z_pflags & ZFS_NODUMP) != 0);
2883 XVA_SET_RTN(xvap, XAT_NODUMP);
2886 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2888 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2889 XVA_SET_RTN(xvap, XAT_OPAQUE);
2892 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2893 xoap->xoa_av_quarantined =
2894 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2895 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2898 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2899 xoap->xoa_av_modified =
2900 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2901 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2904 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2905 vp->v_type == VREG) {
2906 zfs_sa_get_scanstamp(zp, xvap);
2909 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2912 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2913 times, sizeof (times));
2914 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2915 XVA_SET_RTN(xvap, XAT_CREATETIME);
2918 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2919 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2920 XVA_SET_RTN(xvap, XAT_REPARSE);
2922 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2923 xoap->xoa_generation = zp->z_gen;
2924 XVA_SET_RTN(xvap, XAT_GEN);
2927 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2929 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2930 XVA_SET_RTN(xvap, XAT_OFFLINE);
2933 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2935 ((zp->z_pflags & ZFS_SPARSE) != 0);
2936 XVA_SET_RTN(xvap, XAT_SPARSE);
2940 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2941 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2942 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2943 ZFS_TIME_DECODE(&vap->va_birthtime, crtime);
2945 mutex_exit(&zp->z_lock);
2947 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2948 vap->va_blksize = blksize;
2949 vap->va_bytes = nblocks << 9; /* nblocks * 512 */
2951 if (zp->z_blksz == 0) {
2953 * Block size hasn't been set; suggest maximal I/O transfers.
2955 vap->va_blksize = zfsvfs->z_max_blksz;
2963 * Set the file attributes to the values contained in the
2966 * IN: vp - vnode of file to be modified.
2967 * vap - new attribute values.
2968 * If AT_XVATTR set, then optional attrs are being set
2969 * flags - ATTR_UTIME set if non-default time values provided.
2970 * - ATTR_NOACLCHECK (CIFS context only).
2971 * cr - credentials of caller.
2972 * ct - caller context
2974 * RETURN: 0 on success, error code on failure.
2977 * vp - ctime updated, mtime updated if size changed.
2981 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2982 caller_context_t *ct)
2984 znode_t *zp = VTOZ(vp);
2985 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2990 uint_t mask = vap->va_mask;
2991 uint_t saved_mask = 0;
2992 uint64_t saved_mode;
2995 uint64_t new_uid, new_gid;
2997 uint64_t mtime[2], ctime[2];
2999 int need_policy = FALSE;
3001 zfs_fuid_info_t *fuidp = NULL;
3002 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
3005 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
3006 boolean_t fuid_dirtied = B_FALSE;
3007 sa_bulk_attr_t bulk[7], xattr_bulk[7];
3008 int count = 0, xattr_count = 0;
3013 if (mask & AT_NOSET)
3014 return (SET_ERROR(EINVAL));
3019 zilog = zfsvfs->z_log;
3022 * Make sure that if we have ephemeral uid/gid or xvattr specified
3023 * that file system is at proper version level
3026 if (zfsvfs->z_use_fuids == B_FALSE &&
3027 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
3028 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
3029 (mask & AT_XVATTR))) {
3031 return (SET_ERROR(EINVAL));
3034 if (mask & AT_SIZE && vp->v_type == VDIR) {
3036 return (SET_ERROR(EISDIR));
3039 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
3041 return (SET_ERROR(EINVAL));
3045 * If this is an xvattr_t, then get a pointer to the structure of
3046 * optional attributes. If this is NULL, then we have a vattr_t.
3048 xoap = xva_getxoptattr(xvap);
3050 xva_init(&tmpxvattr);
3053 * Immutable files can only alter immutable bit and atime
3055 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
3056 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
3057 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
3059 return (SET_ERROR(EPERM));
3062 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
3064 return (SET_ERROR(EPERM));
3068 * Verify timestamps doesn't overflow 32 bits.
3069 * ZFS can handle large timestamps, but 32bit syscalls can't
3070 * handle times greater than 2039. This check should be removed
3071 * once large timestamps are fully supported.
3073 if (mask & (AT_ATIME | AT_MTIME)) {
3074 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
3075 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
3077 return (SET_ERROR(EOVERFLOW));
3085 /* Can this be moved to before the top label? */
3086 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
3088 return (SET_ERROR(EROFS));
3092 * First validate permissions
3095 if (mask & AT_SIZE) {
3097 * XXX - Note, we are not providing any open
3098 * mode flags here (like FNDELAY), so we may
3099 * block if there are locks present... this
3100 * should be addressed in openat().
3102 /* XXX - would it be OK to generate a log record here? */
3103 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
3110 if (mask & (AT_ATIME|AT_MTIME) ||
3111 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
3112 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
3113 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
3114 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
3115 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
3116 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
3117 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
3118 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
3122 if (mask & (AT_UID|AT_GID)) {
3123 int idmask = (mask & (AT_UID|AT_GID));
3128 * NOTE: even if a new mode is being set,
3129 * we may clear S_ISUID/S_ISGID bits.
3132 if (!(mask & AT_MODE))
3133 vap->va_mode = zp->z_mode;
3136 * Take ownership or chgrp to group we are a member of
3139 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
3140 take_group = (mask & AT_GID) &&
3141 zfs_groupmember(zfsvfs, vap->va_gid, cr);
3144 * If both AT_UID and AT_GID are set then take_owner and
3145 * take_group must both be set in order to allow taking
3148 * Otherwise, send the check through secpolicy_vnode_setattr()
3152 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
3153 ((idmask == AT_UID) && take_owner) ||
3154 ((idmask == AT_GID) && take_group)) {
3155 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
3156 skipaclchk, cr) == 0) {
3158 * Remove setuid/setgid for non-privileged users
3160 secpolicy_setid_clear(vap, vp, cr);
3161 trim_mask = (mask & (AT_UID|AT_GID));
3170 mutex_enter(&zp->z_lock);
3171 oldva.va_mode = zp->z_mode;
3172 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
3173 if (mask & AT_XVATTR) {
3175 * Update xvattr mask to include only those attributes
3176 * that are actually changing.
3178 * the bits will be restored prior to actually setting
3179 * the attributes so the caller thinks they were set.
3181 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
3182 if (xoap->xoa_appendonly !=
3183 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
3186 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
3187 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
3191 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
3192 if (xoap->xoa_nounlink !=
3193 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
3196 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
3197 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
3201 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
3202 if (xoap->xoa_immutable !=
3203 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
3206 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
3207 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
3211 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
3212 if (xoap->xoa_nodump !=
3213 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
3216 XVA_CLR_REQ(xvap, XAT_NODUMP);
3217 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
3221 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3222 if (xoap->xoa_av_modified !=
3223 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3226 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3227 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
3231 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3232 if ((vp->v_type != VREG &&
3233 xoap->xoa_av_quarantined) ||
3234 xoap->xoa_av_quarantined !=
3235 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3238 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3239 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
3243 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3244 mutex_exit(&zp->z_lock);
3246 return (SET_ERROR(EPERM));
3249 if (need_policy == FALSE &&
3250 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3251 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3256 mutex_exit(&zp->z_lock);
3258 if (mask & AT_MODE) {
3259 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3260 err = secpolicy_setid_setsticky_clear(vp, vap,
3266 trim_mask |= AT_MODE;
3274 * If trim_mask is set then take ownership
3275 * has been granted or write_acl is present and user
3276 * has the ability to modify mode. In that case remove
3277 * UID|GID and or MODE from mask so that
3278 * secpolicy_vnode_setattr() doesn't revoke it.
3282 saved_mask = vap->va_mask;
3283 vap->va_mask &= ~trim_mask;
3284 if (trim_mask & AT_MODE) {
3286 * Save the mode, as secpolicy_vnode_setattr()
3287 * will overwrite it with ova.va_mode.
3289 saved_mode = vap->va_mode;
3292 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3293 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3300 vap->va_mask |= saved_mask;
3301 if (trim_mask & AT_MODE) {
3303 * Recover the mode after
3304 * secpolicy_vnode_setattr().
3306 vap->va_mode = saved_mode;
3312 * secpolicy_vnode_setattr, or take ownership may have
3315 mask = vap->va_mask;
3317 if ((mask & (AT_UID | AT_GID))) {
3318 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3319 &xattr_obj, sizeof (xattr_obj));
3321 if (err == 0 && xattr_obj) {
3322 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3326 if (mask & AT_UID) {
3327 new_uid = zfs_fuid_create(zfsvfs,
3328 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3329 if (new_uid != zp->z_uid &&
3330 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3332 VN_RELE(ZTOV(attrzp));
3333 err = SET_ERROR(EDQUOT);
3338 if (mask & AT_GID) {
3339 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3340 cr, ZFS_GROUP, &fuidp);
3341 if (new_gid != zp->z_gid &&
3342 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3344 VN_RELE(ZTOV(attrzp));
3345 err = SET_ERROR(EDQUOT);
3350 tx = dmu_tx_create(zfsvfs->z_os);
3352 if (mask & AT_MODE) {
3353 uint64_t pmode = zp->z_mode;
3355 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3357 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3358 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3359 err = SET_ERROR(EPERM);
3363 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3366 mutex_enter(&zp->z_lock);
3367 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3369 * Are we upgrading ACL from old V0 format
3372 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3373 zfs_znode_acl_version(zp) ==
3374 ZFS_ACL_VERSION_INITIAL) {
3375 dmu_tx_hold_free(tx, acl_obj, 0,
3377 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3378 0, aclp->z_acl_bytes);
3380 dmu_tx_hold_write(tx, acl_obj, 0,
3383 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3384 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3385 0, aclp->z_acl_bytes);
3387 mutex_exit(&zp->z_lock);
3388 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3390 if ((mask & AT_XVATTR) &&
3391 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3392 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3394 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3398 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3401 fuid_dirtied = zfsvfs->z_fuid_dirty;
3403 zfs_fuid_txhold(zfsvfs, tx);
3405 zfs_sa_upgrade_txholds(tx, zp);
3407 err = dmu_tx_assign(tx, TXG_NOWAIT);
3409 if (err == ERESTART)
3416 * Set each attribute requested.
3417 * We group settings according to the locks they need to acquire.
3419 * Note: you cannot set ctime directly, although it will be
3420 * updated as a side-effect of calling this function.
3424 if (mask & (AT_UID|AT_GID|AT_MODE))
3425 mutex_enter(&zp->z_acl_lock);
3426 mutex_enter(&zp->z_lock);
3428 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3429 &zp->z_pflags, sizeof (zp->z_pflags));
3432 if (mask & (AT_UID|AT_GID|AT_MODE))
3433 mutex_enter(&attrzp->z_acl_lock);
3434 mutex_enter(&attrzp->z_lock);
3435 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3436 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3437 sizeof (attrzp->z_pflags));
3440 if (mask & (AT_UID|AT_GID)) {
3442 if (mask & AT_UID) {
3443 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3444 &new_uid, sizeof (new_uid));
3445 zp->z_uid = new_uid;
3447 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3448 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3450 attrzp->z_uid = new_uid;
3454 if (mask & AT_GID) {
3455 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3456 NULL, &new_gid, sizeof (new_gid));
3457 zp->z_gid = new_gid;
3459 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3460 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3462 attrzp->z_gid = new_gid;
3465 if (!(mask & AT_MODE)) {
3466 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3467 NULL, &new_mode, sizeof (new_mode));
3468 new_mode = zp->z_mode;
3470 err = zfs_acl_chown_setattr(zp);
3473 err = zfs_acl_chown_setattr(attrzp);
3478 if (mask & AT_MODE) {
3479 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3480 &new_mode, sizeof (new_mode));
3481 zp->z_mode = new_mode;
3482 ASSERT3U((uintptr_t)aclp, !=, 0);
3483 err = zfs_aclset_common(zp, aclp, cr, tx);
3485 if (zp->z_acl_cached)
3486 zfs_acl_free(zp->z_acl_cached);
3487 zp->z_acl_cached = aclp;
3492 if (mask & AT_ATIME) {
3493 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3494 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3495 &zp->z_atime, sizeof (zp->z_atime));
3498 if (mask & AT_MTIME) {
3499 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3500 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3501 mtime, sizeof (mtime));
3504 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3505 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3506 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3507 NULL, mtime, sizeof (mtime));
3508 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3509 &ctime, sizeof (ctime));
3510 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3512 } else if (mask != 0) {
3513 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3514 &ctime, sizeof (ctime));
3515 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3518 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3519 SA_ZPL_CTIME(zfsvfs), NULL,
3520 &ctime, sizeof (ctime));
3521 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3522 mtime, ctime, B_TRUE);
3526 * Do this after setting timestamps to prevent timestamp
3527 * update from toggling bit
3530 if (xoap && (mask & AT_XVATTR)) {
3533 * restore trimmed off masks
3534 * so that return masks can be set for caller.
3537 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3538 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3540 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3541 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3543 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3544 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3546 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3547 XVA_SET_REQ(xvap, XAT_NODUMP);
3549 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3550 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3552 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3553 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3556 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3557 ASSERT(vp->v_type == VREG);
3559 zfs_xvattr_set(zp, xvap, tx);
3563 zfs_fuid_sync(zfsvfs, tx);
3566 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3568 mutex_exit(&zp->z_lock);
3569 if (mask & (AT_UID|AT_GID|AT_MODE))
3570 mutex_exit(&zp->z_acl_lock);
3573 if (mask & (AT_UID|AT_GID|AT_MODE))
3574 mutex_exit(&attrzp->z_acl_lock);
3575 mutex_exit(&attrzp->z_lock);
3578 if (err == 0 && attrzp) {
3579 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3585 VN_RELE(ZTOV(attrzp));
3591 zfs_fuid_info_free(fuidp);
3597 if (err == ERESTART)
3600 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3605 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3606 zil_commit(zilog, 0);
3612 typedef struct zfs_zlock {
3613 krwlock_t *zl_rwlock; /* lock we acquired */
3614 znode_t *zl_znode; /* znode we held */
3615 struct zfs_zlock *zl_next; /* next in list */
3619 * Drop locks and release vnodes that were held by zfs_rename_lock().
3622 zfs_rename_unlock(zfs_zlock_t **zlpp)
3626 while ((zl = *zlpp) != NULL) {
3627 if (zl->zl_znode != NULL)
3628 VN_RELE(ZTOV(zl->zl_znode));
3629 rw_exit(zl->zl_rwlock);
3630 *zlpp = zl->zl_next;
3631 kmem_free(zl, sizeof (*zl));
3636 * Search back through the directory tree, using the ".." entries.
3637 * Lock each directory in the chain to prevent concurrent renames.
3638 * Fail any attempt to move a directory into one of its own descendants.
3639 * XXX - z_parent_lock can overlap with map or grow locks
3642 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3646 uint64_t rootid = zp->z_zfsvfs->z_root;
3647 uint64_t oidp = zp->z_id;
3648 krwlock_t *rwlp = &szp->z_parent_lock;
3649 krw_t rw = RW_WRITER;
3652 * First pass write-locks szp and compares to zp->z_id.
3653 * Later passes read-lock zp and compare to zp->z_parent.
3656 if (!rw_tryenter(rwlp, rw)) {
3658 * Another thread is renaming in this path.
3659 * Note that if we are a WRITER, we don't have any
3660 * parent_locks held yet.
3662 if (rw == RW_READER && zp->z_id > szp->z_id) {
3664 * Drop our locks and restart
3666 zfs_rename_unlock(&zl);
3670 rwlp = &szp->z_parent_lock;
3675 * Wait for other thread to drop its locks
3681 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3682 zl->zl_rwlock = rwlp;
3683 zl->zl_znode = NULL;
3684 zl->zl_next = *zlpp;
3687 if (oidp == szp->z_id) /* We're a descendant of szp */
3688 return (SET_ERROR(EINVAL));
3690 if (oidp == rootid) /* We've hit the top */
3693 if (rw == RW_READER) { /* i.e. not the first pass */
3694 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3699 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3700 &oidp, sizeof (oidp));
3701 rwlp = &zp->z_parent_lock;
3704 } while (zp->z_id != sdzp->z_id);
3710 * Move an entry from the provided source directory to the target
3711 * directory. Change the entry name as indicated.
3713 * IN: sdvp - Source directory containing the "old entry".
3714 * snm - Old entry name.
3715 * tdvp - Target directory to contain the "new entry".
3716 * tnm - New entry name.
3717 * cr - credentials of caller.
3718 * ct - caller context
3719 * flags - case flags
3721 * RETURN: 0 on success, error code on failure.
3724 * sdvp,tdvp - ctime|mtime updated
3728 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3729 caller_context_t *ct, int flags)
3731 znode_t *tdzp, *szp, *tzp;
3732 znode_t *sdzp = VTOZ(sdvp);
3733 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs;
3736 zfs_dirlock_t *sdl, *tdl;
3739 int cmp, serr, terr;
3742 boolean_t waited = B_FALSE;
3745 ZFS_VERIFY_ZP(sdzp);
3746 zilog = zfsvfs->z_log;
3749 * Make sure we have the real vp for the target directory.
3751 if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3754 if (tdvp->v_vfsp != sdvp->v_vfsp || zfsctl_is_node(tdvp)) {
3756 return (SET_ERROR(EXDEV));
3760 ZFS_VERIFY_ZP(tdzp);
3761 if (zfsvfs->z_utf8 && u8_validate(tnm,
3762 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3764 return (SET_ERROR(EILSEQ));
3767 if (flags & FIGNORECASE)
3776 * This is to prevent the creation of links into attribute space
3777 * by renaming a linked file into/outof an attribute directory.
3778 * See the comment in zfs_link() for why this is considered bad.
3780 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3782 return (SET_ERROR(EINVAL));
3786 * Lock source and target directory entries. To prevent deadlock,
3787 * a lock ordering must be defined. We lock the directory with
3788 * the smallest object id first, or if it's a tie, the one with
3789 * the lexically first name.
3791 if (sdzp->z_id < tdzp->z_id) {
3793 } else if (sdzp->z_id > tdzp->z_id) {
3797 * First compare the two name arguments without
3798 * considering any case folding.
3800 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3802 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3803 ASSERT(error == 0 || !zfsvfs->z_utf8);
3806 * POSIX: "If the old argument and the new argument
3807 * both refer to links to the same existing file,
3808 * the rename() function shall return successfully
3809 * and perform no other action."
3815 * If the file system is case-folding, then we may
3816 * have some more checking to do. A case-folding file
3817 * system is either supporting mixed case sensitivity
3818 * access or is completely case-insensitive. Note
3819 * that the file system is always case preserving.
3821 * In mixed sensitivity mode case sensitive behavior
3822 * is the default. FIGNORECASE must be used to
3823 * explicitly request case insensitive behavior.
3825 * If the source and target names provided differ only
3826 * by case (e.g., a request to rename 'tim' to 'Tim'),
3827 * we will treat this as a special case in the
3828 * case-insensitive mode: as long as the source name
3829 * is an exact match, we will allow this to proceed as
3830 * a name-change request.
3832 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3833 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3834 flags & FIGNORECASE)) &&
3835 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3838 * case preserving rename request, require exact
3847 * If the source and destination directories are the same, we should
3848 * grab the z_name_lock of that directory only once.
3852 rw_enter(&sdzp->z_name_lock, RW_READER);
3856 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3857 ZEXISTS | zflg, NULL, NULL);
3858 terr = zfs_dirent_lock(&tdl,
3859 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3861 terr = zfs_dirent_lock(&tdl,
3862 tdzp, tnm, &tzp, zflg, NULL, NULL);
3863 serr = zfs_dirent_lock(&sdl,
3864 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3870 * Source entry invalid or not there.
3873 zfs_dirent_unlock(tdl);
3879 rw_exit(&sdzp->z_name_lock);
3882 * FreeBSD: In OpenSolaris they only check if rename source is
3883 * ".." here, because "." is handled in their lookup. This is
3884 * not the case for FreeBSD, so we check for "." explicitly.
3886 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0)
3887 serr = SET_ERROR(EINVAL);
3892 zfs_dirent_unlock(sdl);
3896 rw_exit(&sdzp->z_name_lock);
3898 if (strcmp(tnm, "..") == 0)
3899 terr = SET_ERROR(EINVAL);
3905 * Must have write access at the source to remove the old entry
3906 * and write access at the target to create the new entry.
3907 * Note that if target and source are the same, this can be
3908 * done in a single check.
3911 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3914 if (ZTOV(szp)->v_type == VDIR) {
3916 * Check to make sure rename is valid.
3917 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3919 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3924 * Does target exist?
3928 * Source and target must be the same type.
3930 if (ZTOV(szp)->v_type == VDIR) {
3931 if (ZTOV(tzp)->v_type != VDIR) {
3932 error = SET_ERROR(ENOTDIR);
3936 if (ZTOV(tzp)->v_type == VDIR) {
3937 error = SET_ERROR(EISDIR);
3942 * POSIX dictates that when the source and target
3943 * entries refer to the same file object, rename
3944 * must do nothing and exit without error.
3946 if (szp->z_id == tzp->z_id) {
3952 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3954 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3957 * notify the target directory if it is not the same
3958 * as source directory.
3961 vnevent_rename_dest_dir(tdvp, ct);
3964 tx = dmu_tx_create(zfsvfs->z_os);
3965 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3966 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3967 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3968 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3970 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3971 zfs_sa_upgrade_txholds(tx, tdzp);
3974 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3975 zfs_sa_upgrade_txholds(tx, tzp);
3978 zfs_sa_upgrade_txholds(tx, szp);
3979 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3980 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3983 zfs_rename_unlock(&zl);
3984 zfs_dirent_unlock(sdl);
3985 zfs_dirent_unlock(tdl);
3988 rw_exit(&sdzp->z_name_lock);
3993 if (error == ERESTART) {
4004 if (tzp) /* Attempt to remove the existing target */
4005 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
4008 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
4010 szp->z_pflags |= ZFS_AV_MODIFIED;
4012 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
4013 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
4016 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
4018 zfs_log_rename(zilog, tx, TX_RENAME |
4019 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
4020 sdl->dl_name, tdzp, tdl->dl_name, szp);
4023 * Update path information for the target vnode
4025 vn_renamepath(tdvp, ZTOV(szp), tnm,
4029 * At this point, we have successfully created
4030 * the target name, but have failed to remove
4031 * the source name. Since the create was done
4032 * with the ZRENAMING flag, there are
4033 * complications; for one, the link count is
4034 * wrong. The easiest way to deal with this
4035 * is to remove the newly created target, and
4036 * return the original error. This must
4037 * succeed; fortunately, it is very unlikely to
4038 * fail, since we just created it.
4040 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
4041 ZRENAMING, NULL), ==, 0);
4044 #ifdef FREEBSD_NAMECACHE
4055 zfs_rename_unlock(&zl);
4057 zfs_dirent_unlock(sdl);
4058 zfs_dirent_unlock(tdl);
4061 rw_exit(&sdzp->z_name_lock);
4068 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4069 zil_commit(zilog, 0);
4077 * Insert the indicated symbolic reference entry into the directory.
4079 * IN: dvp - Directory to contain new symbolic link.
4080 * link - Name for new symlink entry.
4081 * vap - Attributes of new entry.
4082 * cr - credentials of caller.
4083 * ct - caller context
4084 * flags - case flags
4086 * RETURN: 0 on success, error code on failure.
4089 * dvp - ctime|mtime updated
4093 zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link,
4094 cred_t *cr, kthread_t *td)
4096 znode_t *zp, *dzp = VTOZ(dvp);
4099 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4101 uint64_t len = strlen(link);
4104 zfs_acl_ids_t acl_ids;
4105 boolean_t fuid_dirtied;
4106 uint64_t txtype = TX_SYMLINK;
4107 boolean_t waited = B_FALSE;
4110 ASSERT(vap->va_type == VLNK);
4114 zilog = zfsvfs->z_log;
4116 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
4117 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4119 return (SET_ERROR(EILSEQ));
4121 if (flags & FIGNORECASE)
4124 if (len > MAXPATHLEN) {
4126 return (SET_ERROR(ENAMETOOLONG));
4129 if ((error = zfs_acl_ids_create(dzp, 0,
4130 vap, cr, NULL, &acl_ids)) != 0) {
4136 * Attempt to lock directory; fail if entry already exists.
4138 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
4140 zfs_acl_ids_free(&acl_ids);
4145 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4146 zfs_acl_ids_free(&acl_ids);
4147 zfs_dirent_unlock(dl);
4152 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
4153 zfs_acl_ids_free(&acl_ids);
4154 zfs_dirent_unlock(dl);
4156 return (SET_ERROR(EDQUOT));
4158 tx = dmu_tx_create(zfsvfs->z_os);
4159 fuid_dirtied = zfsvfs->z_fuid_dirty;
4160 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
4161 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4162 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
4163 ZFS_SA_BASE_ATTR_SIZE + len);
4164 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
4165 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
4166 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
4167 acl_ids.z_aclp->z_acl_bytes);
4170 zfs_fuid_txhold(zfsvfs, tx);
4171 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4173 zfs_dirent_unlock(dl);
4174 if (error == ERESTART) {
4180 zfs_acl_ids_free(&acl_ids);
4187 * Create a new object for the symlink.
4188 * for version 4 ZPL datsets the symlink will be an SA attribute
4190 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
4193 zfs_fuid_sync(zfsvfs, tx);
4195 mutex_enter(&zp->z_lock);
4197 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
4200 zfs_sa_symlink(zp, link, len, tx);
4201 mutex_exit(&zp->z_lock);
4204 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
4205 &zp->z_size, sizeof (zp->z_size), tx);
4207 * Insert the new object into the directory.
4209 (void) zfs_link_create(dl, zp, tx, ZNEW);
4211 if (flags & FIGNORECASE)
4213 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
4216 zfs_acl_ids_free(&acl_ids);
4220 zfs_dirent_unlock(dl);
4222 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4223 zil_commit(zilog, 0);
4230 * Return, in the buffer contained in the provided uio structure,
4231 * the symbolic path referred to by vp.
4233 * IN: vp - vnode of symbolic link.
4234 * uio - structure to contain the link path.
4235 * cr - credentials of caller.
4236 * ct - caller context
4238 * OUT: uio - structure containing the link path.
4240 * RETURN: 0 on success, error code on failure.
4243 * vp - atime updated
4247 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
4249 znode_t *zp = VTOZ(vp);
4250 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4256 mutex_enter(&zp->z_lock);
4258 error = sa_lookup_uio(zp->z_sa_hdl,
4259 SA_ZPL_SYMLINK(zfsvfs), uio);
4261 error = zfs_sa_readlink(zp, uio);
4262 mutex_exit(&zp->z_lock);
4264 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4271 * Insert a new entry into directory tdvp referencing svp.
4273 * IN: tdvp - Directory to contain new entry.
4274 * svp - vnode of new entry.
4275 * name - name of new entry.
4276 * cr - credentials of caller.
4277 * ct - caller context
4279 * RETURN: 0 on success, error code on failure.
4282 * tdvp - ctime|mtime updated
4283 * svp - ctime updated
4287 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
4288 caller_context_t *ct, int flags)
4290 znode_t *dzp = VTOZ(tdvp);
4292 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4301 boolean_t waited = B_FALSE;
4303 ASSERT(tdvp->v_type == VDIR);
4307 zilog = zfsvfs->z_log;
4309 if (VOP_REALVP(svp, &realvp, ct) == 0)
4313 * POSIX dictates that we return EPERM here.
4314 * Better choices include ENOTSUP or EISDIR.
4316 if (svp->v_type == VDIR) {
4318 return (SET_ERROR(EPERM));
4321 if (svp->v_vfsp != tdvp->v_vfsp || zfsctl_is_node(svp)) {
4323 return (SET_ERROR(EXDEV));
4329 /* Prevent links to .zfs/shares files */
4331 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4332 &parent, sizeof (uint64_t))) != 0) {
4336 if (parent == zfsvfs->z_shares_dir) {
4338 return (SET_ERROR(EPERM));
4341 if (zfsvfs->z_utf8 && u8_validate(name,
4342 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4344 return (SET_ERROR(EILSEQ));
4346 if (flags & FIGNORECASE)
4350 * We do not support links between attributes and non-attributes
4351 * because of the potential security risk of creating links
4352 * into "normal" file space in order to circumvent restrictions
4353 * imposed in attribute space.
4355 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4357 return (SET_ERROR(EINVAL));
4361 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4362 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) {
4364 return (SET_ERROR(EPERM));
4367 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4374 * Attempt to lock directory; fail if entry already exists.
4376 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4382 tx = dmu_tx_create(zfsvfs->z_os);
4383 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4384 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4385 zfs_sa_upgrade_txholds(tx, szp);
4386 zfs_sa_upgrade_txholds(tx, dzp);
4387 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4389 zfs_dirent_unlock(dl);
4390 if (error == ERESTART) {
4401 error = zfs_link_create(dl, szp, tx, 0);
4404 uint64_t txtype = TX_LINK;
4405 if (flags & FIGNORECASE)
4407 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4412 zfs_dirent_unlock(dl);
4415 vnevent_link(svp, ct);
4418 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4419 zil_commit(zilog, 0);
4427 * zfs_null_putapage() is used when the file system has been force
4428 * unmounted. It just drops the pages.
4432 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4433 size_t *lenp, int flags, cred_t *cr)
4435 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4440 * Push a page out to disk, klustering if possible.
4442 * IN: vp - file to push page to.
4443 * pp - page to push.
4444 * flags - additional flags.
4445 * cr - credentials of caller.
4447 * OUT: offp - start of range pushed.
4448 * lenp - len of range pushed.
4450 * RETURN: 0 on success, error code on failure.
4452 * NOTE: callers must have locked the page to be pushed. On
4453 * exit, the page (and all other pages in the kluster) must be
4458 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4459 size_t *lenp, int flags, cred_t *cr)
4461 znode_t *zp = VTOZ(vp);
4462 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4464 u_offset_t off, koff;
4471 * If our blocksize is bigger than the page size, try to kluster
4472 * multiple pages so that we write a full block (thus avoiding
4473 * a read-modify-write).
4475 if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4476 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4477 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4478 ASSERT(koff <= zp->z_size);
4479 if (koff + klen > zp->z_size)
4480 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4481 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4483 ASSERT3U(btop(len), ==, btopr(len));
4486 * Can't push pages past end-of-file.
4488 if (off >= zp->z_size) {
4489 /* ignore all pages */
4492 } else if (off + len > zp->z_size) {
4493 int npages = btopr(zp->z_size - off);
4496 page_list_break(&pp, &trunc, npages);
4497 /* ignore pages past end of file */
4499 pvn_write_done(trunc, flags);
4500 len = zp->z_size - off;
4503 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4504 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4505 err = SET_ERROR(EDQUOT);
4509 tx = dmu_tx_create(zfsvfs->z_os);
4510 dmu_tx_hold_write(tx, zp->z_id, off, len);
4512 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4513 zfs_sa_upgrade_txholds(tx, zp);
4514 err = dmu_tx_assign(tx, TXG_NOWAIT);
4516 if (err == ERESTART) {
4525 if (zp->z_blksz <= PAGESIZE) {
4526 caddr_t va = zfs_map_page(pp, S_READ);
4527 ASSERT3U(len, <=, PAGESIZE);
4528 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4529 zfs_unmap_page(pp, va);
4531 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4535 uint64_t mtime[2], ctime[2];
4536 sa_bulk_attr_t bulk[3];
4539 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4541 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4543 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4545 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4547 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4552 pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4562 * Copy the portion of the file indicated from pages into the file.
4563 * The pages are stored in a page list attached to the files vnode.
4565 * IN: vp - vnode of file to push page data to.
4566 * off - position in file to put data.
4567 * len - amount of data to write.
4568 * flags - flags to control the operation.
4569 * cr - credentials of caller.
4570 * ct - caller context.
4572 * RETURN: 0 on success, error code on failure.
4575 * vp - ctime|mtime updated
4579 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4580 caller_context_t *ct)
4582 znode_t *zp = VTOZ(vp);
4583 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4595 * Align this request to the file block size in case we kluster.
4596 * XXX - this can result in pretty aggresive locking, which can
4597 * impact simultanious read/write access. One option might be
4598 * to break up long requests (len == 0) into block-by-block
4599 * operations to get narrower locking.
4601 blksz = zp->z_blksz;
4603 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4606 if (len > 0 && ISP2(blksz))
4607 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4613 * Search the entire vp list for pages >= io_off.
4615 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4616 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4619 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4621 if (off > zp->z_size) {
4622 /* past end of file */
4623 zfs_range_unlock(rl);
4628 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4630 for (off = io_off; io_off < off + len; io_off += io_len) {
4631 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4632 pp = page_lookup(vp, io_off,
4633 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4635 pp = page_lookup_nowait(vp, io_off,
4636 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4639 if (pp != NULL && pvn_getdirty(pp, flags)) {
4643 * Found a dirty page to push
4645 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4653 zfs_range_unlock(rl);
4654 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4655 zil_commit(zfsvfs->z_log, zp->z_id);
4663 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4665 znode_t *zp = VTOZ(vp);
4666 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4669 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4670 if (zp->z_sa_hdl == NULL) {
4672 * The fs has been unmounted, or we did a
4673 * suspend/resume and this file no longer exists.
4675 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4676 vrecycle(vp, curthread);
4680 mutex_enter(&zp->z_lock);
4681 if (zp->z_unlinked) {
4683 * Fast path to recycle a vnode of a removed file.
4685 mutex_exit(&zp->z_lock);
4686 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4687 vrecycle(vp, curthread);
4690 mutex_exit(&zp->z_lock);
4692 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4693 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4695 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4696 zfs_sa_upgrade_txholds(tx, zp);
4697 error = dmu_tx_assign(tx, TXG_WAIT);
4701 mutex_enter(&zp->z_lock);
4702 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4703 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4704 zp->z_atime_dirty = 0;
4705 mutex_exit(&zp->z_lock);
4709 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4714 * Bounds-check the seek operation.
4716 * IN: vp - vnode seeking within
4717 * ooff - old file offset
4718 * noffp - pointer to new file offset
4719 * ct - caller context
4721 * RETURN: 0 on success, EINVAL if new offset invalid.
4725 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4726 caller_context_t *ct)
4728 if (vp->v_type == VDIR)
4730 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4734 * Pre-filter the generic locking function to trap attempts to place
4735 * a mandatory lock on a memory mapped file.
4738 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4739 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4741 znode_t *zp = VTOZ(vp);
4742 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4748 * We are following the UFS semantics with respect to mapcnt
4749 * here: If we see that the file is mapped already, then we will
4750 * return an error, but we don't worry about races between this
4751 * function and zfs_map().
4753 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4755 return (SET_ERROR(EAGAIN));
4758 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4762 * If we can't find a page in the cache, we will create a new page
4763 * and fill it with file data. For efficiency, we may try to fill
4764 * multiple pages at once (klustering) to fill up the supplied page
4765 * list. Note that the pages to be filled are held with an exclusive
4766 * lock to prevent access by other threads while they are being filled.
4769 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4770 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4772 znode_t *zp = VTOZ(vp);
4773 page_t *pp, *cur_pp;
4774 objset_t *os = zp->z_zfsvfs->z_os;
4775 u_offset_t io_off, total;
4779 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4781 * We only have a single page, don't bother klustering
4785 pp = page_create_va(vp, io_off, io_len,
4786 PG_EXCL | PG_WAIT, seg, addr);
4789 * Try to find enough pages to fill the page list
4791 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4792 &io_len, off, plsz, 0);
4796 * The page already exists, nothing to do here.
4803 * Fill the pages in the kluster.
4806 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4809 ASSERT3U(io_off, ==, cur_pp->p_offset);
4810 va = zfs_map_page(cur_pp, S_WRITE);
4811 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4813 zfs_unmap_page(cur_pp, va);
4815 /* On error, toss the entire kluster */
4816 pvn_read_done(pp, B_ERROR);
4817 /* convert checksum errors into IO errors */
4819 err = SET_ERROR(EIO);
4822 cur_pp = cur_pp->p_next;
4826 * Fill in the page list array from the kluster starting
4827 * from the desired offset `off'.
4828 * NOTE: the page list will always be null terminated.
4830 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4831 ASSERT(pl == NULL || (*pl)->p_offset == off);
4837 * Return pointers to the pages for the file region [off, off + len]
4838 * in the pl array. If plsz is greater than len, this function may
4839 * also return page pointers from after the specified region
4840 * (i.e. the region [off, off + plsz]). These additional pages are
4841 * only returned if they are already in the cache, or were created as
4842 * part of a klustered read.
4844 * IN: vp - vnode of file to get data from.
4845 * off - position in file to get data from.
4846 * len - amount of data to retrieve.
4847 * plsz - length of provided page list.
4848 * seg - segment to obtain pages for.
4849 * addr - virtual address of fault.
4850 * rw - mode of created pages.
4851 * cr - credentials of caller.
4852 * ct - caller context.
4854 * OUT: protp - protection mode of created pages.
4855 * pl - list of pages created.
4857 * RETURN: 0 on success, error code on failure.
4860 * vp - atime updated
4864 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4865 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4866 enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4868 znode_t *zp = VTOZ(vp);
4869 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4873 /* we do our own caching, faultahead is unnecessary */
4876 else if (len > plsz)
4879 len = P2ROUNDUP(len, PAGESIZE);
4880 ASSERT(plsz >= len);
4889 * Loop through the requested range [off, off + len) looking
4890 * for pages. If we don't find a page, we will need to create
4891 * a new page and fill it with data from the file.
4894 if (*pl = page_lookup(vp, off, SE_SHARED))
4896 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4899 ASSERT3U((*pl)->p_offset, ==, off);
4903 ASSERT3U(len, >=, PAGESIZE);
4906 ASSERT3U(plsz, >=, PAGESIZE);
4913 * Fill out the page array with any pages already in the cache.
4916 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4923 * Release any pages we have previously locked.
4928 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4938 * Request a memory map for a section of a file. This code interacts
4939 * with common code and the VM system as follows:
4941 * - common code calls mmap(), which ends up in smmap_common()
4942 * - this calls VOP_MAP(), which takes you into (say) zfs
4943 * - zfs_map() calls as_map(), passing segvn_create() as the callback
4944 * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4945 * - zfs_addmap() updates z_mapcnt
4949 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4950 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4951 caller_context_t *ct)
4953 znode_t *zp = VTOZ(vp);
4954 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4955 segvn_crargs_t vn_a;
4961 if ((prot & PROT_WRITE) && (zp->z_pflags &
4962 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4964 return (SET_ERROR(EPERM));
4967 if ((prot & (PROT_READ | PROT_EXEC)) &&
4968 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4970 return (SET_ERROR(EACCES));
4973 if (vp->v_flag & VNOMAP) {
4975 return (SET_ERROR(ENOSYS));
4978 if (off < 0 || len > MAXOFFSET_T - off) {
4980 return (SET_ERROR(ENXIO));
4983 if (vp->v_type != VREG) {
4985 return (SET_ERROR(ENODEV));
4989 * If file is locked, disallow mapping.
4991 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
4993 return (SET_ERROR(EAGAIN));
4997 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
5005 vn_a.offset = (u_offset_t)off;
5006 vn_a.type = flags & MAP_TYPE;
5008 vn_a.maxprot = maxprot;
5011 vn_a.flags = flags & ~MAP_TYPE;
5013 vn_a.lgrp_mem_policy_flags = 0;
5015 error = as_map(as, *addrp, len, segvn_create, &vn_a);
5024 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5025 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
5026 caller_context_t *ct)
5028 uint64_t pages = btopr(len);
5030 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
5035 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
5036 * more accurate mtime for the associated file. Since we don't have a way of
5037 * detecting when the data was actually modified, we have to resort to
5038 * heuristics. If an explicit msync() is done, then we mark the mtime when the
5039 * last page is pushed. The problem occurs when the msync() call is omitted,
5040 * which by far the most common case:
5048 * putpage() via fsflush
5050 * If we wait until fsflush to come along, we can have a modification time that
5051 * is some arbitrary point in the future. In order to prevent this in the
5052 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
5057 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5058 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
5059 caller_context_t *ct)
5061 uint64_t pages = btopr(len);
5063 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
5064 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
5066 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
5067 vn_has_cached_data(vp))
5068 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
5074 * Free or allocate space in a file. Currently, this function only
5075 * supports the `F_FREESP' command. However, this command is somewhat
5076 * misnamed, as its functionality includes the ability to allocate as
5077 * well as free space.
5079 * IN: vp - vnode of file to free data in.
5080 * cmd - action to take (only F_FREESP supported).
5081 * bfp - section of file to free/alloc.
5082 * flag - current file open mode flags.
5083 * offset - current file offset.
5084 * cr - credentials of caller [UNUSED].
5085 * ct - caller context.
5087 * RETURN: 0 on success, error code on failure.
5090 * vp - ctime|mtime updated
5094 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
5095 offset_t offset, cred_t *cr, caller_context_t *ct)
5097 znode_t *zp = VTOZ(vp);
5098 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5105 if (cmd != F_FREESP) {
5107 return (SET_ERROR(EINVAL));
5110 if (error = convoff(vp, bfp, 0, offset)) {
5115 if (bfp->l_len < 0) {
5117 return (SET_ERROR(EINVAL));
5121 len = bfp->l_len; /* 0 means from off to end of file */
5123 error = zfs_freesp(zp, off, len, flag, TRUE);
5130 CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid));
5131 CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid));
5135 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
5137 znode_t *zp = VTOZ(vp);
5138 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5141 uint64_t object = zp->z_id;
5148 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
5149 &gen64, sizeof (uint64_t))) != 0) {
5154 gen = (uint32_t)gen64;
5156 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
5159 if (fidp->fid_len < size) {
5160 fidp->fid_len = size;
5162 return (SET_ERROR(ENOSPC));
5165 fidp->fid_len = size;
5168 zfid = (zfid_short_t *)fidp;
5170 zfid->zf_len = size;
5172 for (i = 0; i < sizeof (zfid->zf_object); i++)
5173 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
5175 /* Must have a non-zero generation number to distinguish from .zfs */
5178 for (i = 0; i < sizeof (zfid->zf_gen); i++)
5179 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
5181 if (size == LONG_FID_LEN) {
5182 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
5185 zlfid = (zfid_long_t *)fidp;
5187 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
5188 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
5190 /* XXX - this should be the generation number for the objset */
5191 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
5192 zlfid->zf_setgen[i] = 0;
5200 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
5201 caller_context_t *ct)
5213 case _PC_FILESIZEBITS:
5217 case _PC_XATTR_EXISTS:
5219 zfsvfs = zp->z_zfsvfs;
5223 error = zfs_dirent_lock(&dl, zp, "", &xzp,
5224 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
5226 zfs_dirent_unlock(dl);
5227 if (!zfs_dirempty(xzp))
5230 } else if (error == ENOENT) {
5232 * If there aren't extended attributes, it's the
5233 * same as having zero of them.
5240 case _PC_SATTR_ENABLED:
5241 case _PC_SATTR_EXISTS:
5242 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
5243 (vp->v_type == VREG || vp->v_type == VDIR);
5246 case _PC_ACCESS_FILTERING:
5247 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
5251 case _PC_ACL_ENABLED:
5252 *valp = _ACL_ACE_ENABLED;
5255 case _PC_MIN_HOLE_SIZE:
5256 *valp = (int)SPA_MINBLOCKSIZE;
5259 case _PC_TIMESTAMP_RESOLUTION:
5260 /* nanosecond timestamp resolution */
5264 case _PC_ACL_EXTENDED:
5272 case _PC_ACL_PATH_MAX:
5273 *valp = ACL_MAX_ENTRIES;
5277 return (EOPNOTSUPP);
5283 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5284 caller_context_t *ct)
5286 znode_t *zp = VTOZ(vp);
5287 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5289 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5293 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5301 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5302 caller_context_t *ct)
5304 znode_t *zp = VTOZ(vp);
5305 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5307 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5308 zilog_t *zilog = zfsvfs->z_log;
5313 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5315 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5316 zil_commit(zilog, 0);
5324 * The smallest read we may consider to loan out an arcbuf.
5325 * This must be a power of 2.
5327 int zcr_blksz_min = (1 << 10); /* 1K */
5329 * If set to less than the file block size, allow loaning out of an
5330 * arcbuf for a partial block read. This must be a power of 2.
5332 int zcr_blksz_max = (1 << 17); /* 128K */
5336 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5337 caller_context_t *ct)
5339 znode_t *zp = VTOZ(vp);
5340 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5341 int max_blksz = zfsvfs->z_max_blksz;
5342 uio_t *uio = &xuio->xu_uio;
5343 ssize_t size = uio->uio_resid;
5344 offset_t offset = uio->uio_loffset;
5349 int preamble, postamble;
5351 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5352 return (SET_ERROR(EINVAL));
5359 * Loan out an arc_buf for write if write size is bigger than
5360 * max_blksz, and the file's block size is also max_blksz.
5363 if (size < blksz || zp->z_blksz != blksz) {
5365 return (SET_ERROR(EINVAL));
5368 * Caller requests buffers for write before knowing where the
5369 * write offset might be (e.g. NFS TCP write).
5374 preamble = P2PHASE(offset, blksz);
5376 preamble = blksz - preamble;
5381 postamble = P2PHASE(size, blksz);
5384 fullblk = size / blksz;
5385 (void) dmu_xuio_init(xuio,
5386 (preamble != 0) + fullblk + (postamble != 0));
5387 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5388 int, postamble, int,
5389 (preamble != 0) + fullblk + (postamble != 0));
5392 * Have to fix iov base/len for partial buffers. They
5393 * currently represent full arc_buf's.
5396 /* data begins in the middle of the arc_buf */
5397 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5400 (void) dmu_xuio_add(xuio, abuf,
5401 blksz - preamble, preamble);
5404 for (i = 0; i < fullblk; i++) {
5405 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5408 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5412 /* data ends in the middle of the arc_buf */
5413 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5416 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5421 * Loan out an arc_buf for read if the read size is larger than
5422 * the current file block size. Block alignment is not
5423 * considered. Partial arc_buf will be loaned out for read.
5425 blksz = zp->z_blksz;
5426 if (blksz < zcr_blksz_min)
5427 blksz = zcr_blksz_min;
5428 if (blksz > zcr_blksz_max)
5429 blksz = zcr_blksz_max;
5430 /* avoid potential complexity of dealing with it */
5431 if (blksz > max_blksz) {
5433 return (SET_ERROR(EINVAL));
5436 maxsize = zp->z_size - uio->uio_loffset;
5440 if (size < blksz || vn_has_cached_data(vp)) {
5442 return (SET_ERROR(EINVAL));
5447 return (SET_ERROR(EINVAL));
5450 uio->uio_extflg = UIO_XUIO;
5451 XUIO_XUZC_RW(xuio) = ioflag;
5458 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5462 int ioflag = XUIO_XUZC_RW(xuio);
5464 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5466 i = dmu_xuio_cnt(xuio);
5468 abuf = dmu_xuio_arcbuf(xuio, i);
5470 * if abuf == NULL, it must be a write buffer
5471 * that has been returned in zfs_write().
5474 dmu_return_arcbuf(abuf);
5475 ASSERT(abuf || ioflag == UIO_WRITE);
5478 dmu_xuio_fini(xuio);
5483 * Predeclare these here so that the compiler assumes that
5484 * this is an "old style" function declaration that does
5485 * not include arguments => we won't get type mismatch errors
5486 * in the initializations that follow.
5488 static int zfs_inval();
5489 static int zfs_isdir();
5494 return (SET_ERROR(EINVAL));
5500 return (SET_ERROR(EISDIR));
5503 * Directory vnode operations template
5505 vnodeops_t *zfs_dvnodeops;
5506 const fs_operation_def_t zfs_dvnodeops_template[] = {
5507 VOPNAME_OPEN, { .vop_open = zfs_open },
5508 VOPNAME_CLOSE, { .vop_close = zfs_close },
5509 VOPNAME_READ, { .error = zfs_isdir },
5510 VOPNAME_WRITE, { .error = zfs_isdir },
5511 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5512 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5513 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5514 VOPNAME_ACCESS, { .vop_access = zfs_access },
5515 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5516 VOPNAME_CREATE, { .vop_create = zfs_create },
5517 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5518 VOPNAME_LINK, { .vop_link = zfs_link },
5519 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5520 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir },
5521 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5522 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5523 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink },
5524 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5525 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5526 VOPNAME_FID, { .vop_fid = zfs_fid },
5527 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5528 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5529 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5530 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5531 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5536 * Regular file vnode operations template
5538 vnodeops_t *zfs_fvnodeops;
5539 const fs_operation_def_t zfs_fvnodeops_template[] = {
5540 VOPNAME_OPEN, { .vop_open = zfs_open },
5541 VOPNAME_CLOSE, { .vop_close = zfs_close },
5542 VOPNAME_READ, { .vop_read = zfs_read },
5543 VOPNAME_WRITE, { .vop_write = zfs_write },
5544 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5545 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5546 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5547 VOPNAME_ACCESS, { .vop_access = zfs_access },
5548 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5549 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5550 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5551 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5552 VOPNAME_FID, { .vop_fid = zfs_fid },
5553 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5554 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock },
5555 VOPNAME_SPACE, { .vop_space = zfs_space },
5556 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage },
5557 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage },
5558 VOPNAME_MAP, { .vop_map = zfs_map },
5559 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap },
5560 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap },
5561 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5562 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5563 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5564 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5565 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf },
5566 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf },
5571 * Symbolic link vnode operations template
5573 vnodeops_t *zfs_symvnodeops;
5574 const fs_operation_def_t zfs_symvnodeops_template[] = {
5575 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5576 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5577 VOPNAME_ACCESS, { .vop_access = zfs_access },
5578 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5579 VOPNAME_READLINK, { .vop_readlink = zfs_readlink },
5580 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5581 VOPNAME_FID, { .vop_fid = zfs_fid },
5582 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5583 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5588 * special share hidden files vnode operations template
5590 vnodeops_t *zfs_sharevnodeops;
5591 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5592 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5593 VOPNAME_ACCESS, { .vop_access = zfs_access },
5594 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5595 VOPNAME_FID, { .vop_fid = zfs_fid },
5596 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5597 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5598 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5599 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5604 * Extended attribute directory vnode operations template
5606 * This template is identical to the directory vnodes
5607 * operation template except for restricted operations:
5611 * Note that there are other restrictions embedded in:
5612 * zfs_create() - restrict type to VREG
5613 * zfs_link() - no links into/out of attribute space
5614 * zfs_rename() - no moves into/out of attribute space
5616 vnodeops_t *zfs_xdvnodeops;
5617 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5618 VOPNAME_OPEN, { .vop_open = zfs_open },
5619 VOPNAME_CLOSE, { .vop_close = zfs_close },
5620 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5621 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5622 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5623 VOPNAME_ACCESS, { .vop_access = zfs_access },
5624 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5625 VOPNAME_CREATE, { .vop_create = zfs_create },
5626 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5627 VOPNAME_LINK, { .vop_link = zfs_link },
5628 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5629 VOPNAME_MKDIR, { .error = zfs_inval },
5630 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5631 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5632 VOPNAME_SYMLINK, { .error = zfs_inval },
5633 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5634 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5635 VOPNAME_FID, { .vop_fid = zfs_fid },
5636 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5637 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5638 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5639 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5640 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5645 * Error vnode operations template
5647 vnodeops_t *zfs_evnodeops;
5648 const fs_operation_def_t zfs_evnodeops_template[] = {
5649 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5650 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5656 ioflags(int ioflags)
5660 if (ioflags & IO_APPEND)
5662 if (ioflags & IO_NDELAY)
5664 if (ioflags & IO_SYNC)
5665 flags |= (FSYNC | FDSYNC | FRSYNC);
5671 zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage)
5673 znode_t *zp = VTOZ(vp);
5674 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5675 objset_t *os = zp->z_zfsvfs->z_os;
5676 vm_page_t mfirst, mlast, mreq;
5680 off_t startoff, endoff;
5682 vm_pindex_t reqstart, reqend;
5683 int pcount, lsize, reqsize, size;
5688 pcount = OFF_TO_IDX(round_page(count));
5690 object = mreq->object;
5693 KASSERT(vp->v_object == object, ("mismatching object"));
5695 if (pcount > 1 && zp->z_blksz > PAGESIZE) {
5696 startoff = rounddown(IDX_TO_OFF(mreq->pindex), zp->z_blksz);
5697 reqstart = OFF_TO_IDX(round_page(startoff));
5698 if (reqstart < m[0]->pindex)
5701 reqstart = reqstart - m[0]->pindex;
5702 endoff = roundup(IDX_TO_OFF(mreq->pindex) + PAGE_SIZE,
5704 reqend = OFF_TO_IDX(trunc_page(endoff)) - 1;
5705 if (reqend > m[pcount - 1]->pindex)
5706 reqend = m[pcount - 1]->pindex;
5707 reqsize = reqend - m[reqstart]->pindex + 1;
5708 KASSERT(reqstart <= reqpage && reqpage < reqstart + reqsize,
5709 ("reqpage beyond [reqstart, reqstart + reqsize[ bounds"));
5714 mfirst = m[reqstart];
5715 mlast = m[reqstart + reqsize - 1];
5717 VM_OBJECT_LOCK(object);
5719 for (i = 0; i < reqstart; i++) {
5722 vm_page_unlock(m[i]);
5724 for (i = reqstart + reqsize; i < pcount; i++) {
5727 vm_page_unlock(m[i]);
5730 if (mreq->valid && reqsize == 1) {
5731 if (mreq->valid != VM_PAGE_BITS_ALL)
5732 vm_page_zero_invalid(mreq, TRUE);
5733 VM_OBJECT_UNLOCK(object);
5735 return (VM_PAGER_OK);
5738 PCPU_INC(cnt.v_vnodein);
5739 PCPU_ADD(cnt.v_vnodepgsin, reqsize);
5741 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) {
5742 for (i = reqstart; i < reqstart + reqsize; i++) {
5746 vm_page_unlock(m[i]);
5749 VM_OBJECT_UNLOCK(object);
5751 return (VM_PAGER_BAD);
5755 if (IDX_TO_OFF(mlast->pindex) + lsize > object->un_pager.vnp.vnp_size)
5756 lsize = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mlast->pindex);
5758 VM_OBJECT_UNLOCK(object);
5760 for (i = reqstart; i < reqstart + reqsize; i++) {
5762 if (i == (reqstart + reqsize - 1))
5764 va = zfs_map_page(m[i], &sf);
5765 error = dmu_read(os, zp->z_id, IDX_TO_OFF(m[i]->pindex),
5766 size, va, DMU_READ_PREFETCH);
5767 if (size != PAGE_SIZE)
5768 bzero(va + size, PAGE_SIZE - size);
5774 VM_OBJECT_LOCK(object);
5776 for (i = reqstart; i < reqstart + reqsize; i++) {
5778 m[i]->valid = VM_PAGE_BITS_ALL;
5779 KASSERT(m[i]->dirty == 0, ("zfs_getpages: page %p is dirty", m[i]));
5781 vm_page_readahead_finish(m[i]);
5784 VM_OBJECT_UNLOCK(object);
5786 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
5788 return (error ? VM_PAGER_ERROR : VM_PAGER_OK);
5792 zfs_freebsd_getpages(ap)
5793 struct vop_getpages_args /* {
5798 vm_ooffset_t a_offset;
5802 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage));
5806 zfs_freebsd_bmap(ap)
5807 struct vop_bmap_args /* {
5810 struct bufobj **a_bop;
5817 if (ap->a_bop != NULL)
5818 *ap->a_bop = &ap->a_vp->v_bufobj;
5819 if (ap->a_bnp != NULL)
5820 *ap->a_bnp = ap->a_bn;
5821 if (ap->a_runp != NULL)
5823 if (ap->a_runb != NULL)
5830 zfs_freebsd_open(ap)
5831 struct vop_open_args /* {
5834 struct ucred *a_cred;
5835 struct thread *a_td;
5838 vnode_t *vp = ap->a_vp;
5839 znode_t *zp = VTOZ(vp);
5842 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL);
5844 vnode_create_vobject(vp, zp->z_size, ap->a_td);
5849 zfs_freebsd_close(ap)
5850 struct vop_close_args /* {
5853 struct ucred *a_cred;
5854 struct thread *a_td;
5858 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL));
5862 zfs_freebsd_ioctl(ap)
5863 struct vop_ioctl_args /* {
5873 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data,
5874 ap->a_fflag, ap->a_cred, NULL, NULL));
5878 zfs_freebsd_read(ap)
5879 struct vop_read_args /* {
5883 struct ucred *a_cred;
5887 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5892 zfs_freebsd_write(ap)
5893 struct vop_write_args /* {
5897 struct ucred *a_cred;
5901 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5906 zfs_freebsd_access(ap)
5907 struct vop_access_args /* {
5909 accmode_t a_accmode;
5910 struct ucred *a_cred;
5911 struct thread *a_td;
5914 vnode_t *vp = ap->a_vp;
5915 znode_t *zp = VTOZ(vp);
5920 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND,
5922 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND);
5924 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL);
5927 * VADMIN has to be handled by vaccess().
5930 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND);
5932 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
5933 zp->z_gid, accmode, ap->a_cred, NULL);
5938 * For VEXEC, ensure that at least one execute bit is set for
5941 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR &&
5942 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
5950 zfs_freebsd_lookup(ap)
5951 struct vop_lookup_args /* {
5952 struct vnode *a_dvp;
5953 struct vnode **a_vpp;
5954 struct componentname *a_cnp;
5957 struct componentname *cnp = ap->a_cnp;
5958 char nm[NAME_MAX + 1];
5960 ASSERT(cnp->cn_namelen < sizeof(nm));
5961 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm)));
5963 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop,
5964 cnp->cn_cred, cnp->cn_thread, 0));
5968 zfs_freebsd_create(ap)
5969 struct vop_create_args /* {
5970 struct vnode *a_dvp;
5971 struct vnode **a_vpp;
5972 struct componentname *a_cnp;
5973 struct vattr *a_vap;
5976 struct componentname *cnp = ap->a_cnp;
5977 vattr_t *vap = ap->a_vap;
5980 ASSERT(cnp->cn_flags & SAVENAME);
5982 vattr_init_mask(vap);
5983 mode = vap->va_mode & ALLPERMS;
5985 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode,
5986 ap->a_vpp, cnp->cn_cred, cnp->cn_thread));
5990 zfs_freebsd_remove(ap)
5991 struct vop_remove_args /* {
5992 struct vnode *a_dvp;
5994 struct componentname *a_cnp;
5998 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6000 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr,
6001 ap->a_cnp->cn_cred, NULL, 0));
6005 zfs_freebsd_mkdir(ap)
6006 struct vop_mkdir_args /* {
6007 struct vnode *a_dvp;
6008 struct vnode **a_vpp;
6009 struct componentname *a_cnp;
6010 struct vattr *a_vap;
6013 vattr_t *vap = ap->a_vap;
6015 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6017 vattr_init_mask(vap);
6019 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp,
6020 ap->a_cnp->cn_cred, NULL, 0, NULL));
6024 zfs_freebsd_rmdir(ap)
6025 struct vop_rmdir_args /* {
6026 struct vnode *a_dvp;
6028 struct componentname *a_cnp;
6031 struct componentname *cnp = ap->a_cnp;
6033 ASSERT(cnp->cn_flags & SAVENAME);
6035 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0));
6039 zfs_freebsd_readdir(ap)
6040 struct vop_readdir_args /* {
6043 struct ucred *a_cred;
6050 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag,
6051 ap->a_ncookies, ap->a_cookies));
6055 zfs_freebsd_fsync(ap)
6056 struct vop_fsync_args /* {
6059 struct thread *a_td;
6064 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL));
6068 zfs_freebsd_getattr(ap)
6069 struct vop_getattr_args /* {
6071 struct vattr *a_vap;
6072 struct ucred *a_cred;
6075 vattr_t *vap = ap->a_vap;
6081 xvap.xva_vattr = *vap;
6082 xvap.xva_vattr.va_mask |= AT_XVATTR;
6084 /* Convert chflags into ZFS-type flags. */
6085 /* XXX: what about SF_SETTABLE?. */
6086 XVA_SET_REQ(&xvap, XAT_IMMUTABLE);
6087 XVA_SET_REQ(&xvap, XAT_APPENDONLY);
6088 XVA_SET_REQ(&xvap, XAT_NOUNLINK);
6089 XVA_SET_REQ(&xvap, XAT_NODUMP);
6090 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL);
6094 /* Convert ZFS xattr into chflags. */
6095 #define FLAG_CHECK(fflag, xflag, xfield) do { \
6096 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \
6097 fflags |= (fflag); \
6099 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE,
6100 xvap.xva_xoptattrs.xoa_immutable);
6101 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY,
6102 xvap.xva_xoptattrs.xoa_appendonly);
6103 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK,
6104 xvap.xva_xoptattrs.xoa_nounlink);
6105 FLAG_CHECK(UF_NODUMP, XAT_NODUMP,
6106 xvap.xva_xoptattrs.xoa_nodump);
6108 *vap = xvap.xva_vattr;
6109 vap->va_flags = fflags;
6114 zfs_freebsd_setattr(ap)
6115 struct vop_setattr_args /* {
6117 struct vattr *a_vap;
6118 struct ucred *a_cred;
6121 vnode_t *vp = ap->a_vp;
6122 vattr_t *vap = ap->a_vap;
6123 cred_t *cred = ap->a_cred;
6128 vattr_init_mask(vap);
6129 vap->va_mask &= ~AT_NOSET;
6132 xvap.xva_vattr = *vap;
6134 zflags = VTOZ(vp)->z_pflags;
6136 if (vap->va_flags != VNOVAL) {
6137 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs;
6140 if (zfsvfs->z_use_fuids == B_FALSE)
6141 return (EOPNOTSUPP);
6143 fflags = vap->va_flags;
6144 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_NODUMP)) != 0)
6145 return (EOPNOTSUPP);
6147 * Unprivileged processes are not permitted to unset system
6148 * flags, or modify flags if any system flags are set.
6149 * Privileged non-jail processes may not modify system flags
6150 * if securelevel > 0 and any existing system flags are set.
6151 * Privileged jail processes behave like privileged non-jail
6152 * processes if the security.jail.chflags_allowed sysctl is
6153 * is non-zero; otherwise, they behave like unprivileged
6156 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 ||
6157 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) {
6159 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6160 error = securelevel_gt(cred, 0);
6166 * Callers may only modify the file flags on objects they
6167 * have VADMIN rights for.
6169 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0)
6172 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6176 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) {
6181 #define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \
6182 if (((fflags & (fflag)) && !(zflags & (zflag))) || \
6183 ((zflags & (zflag)) && !(fflags & (fflag)))) { \
6184 XVA_SET_REQ(&xvap, (xflag)); \
6185 (xfield) = ((fflags & (fflag)) != 0); \
6188 /* Convert chflags into ZFS-type flags. */
6189 /* XXX: what about SF_SETTABLE?. */
6190 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE,
6191 xvap.xva_xoptattrs.xoa_immutable);
6192 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY,
6193 xvap.xva_xoptattrs.xoa_appendonly);
6194 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK,
6195 xvap.xva_xoptattrs.xoa_nounlink);
6196 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP,
6197 xvap.xva_xoptattrs.xoa_nodump);
6200 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL));
6204 zfs_freebsd_rename(ap)
6205 struct vop_rename_args /* {
6206 struct vnode *a_fdvp;
6207 struct vnode *a_fvp;
6208 struct componentname *a_fcnp;
6209 struct vnode *a_tdvp;
6210 struct vnode *a_tvp;
6211 struct componentname *a_tcnp;
6214 vnode_t *fdvp = ap->a_fdvp;
6215 vnode_t *fvp = ap->a_fvp;
6216 vnode_t *tdvp = ap->a_tdvp;
6217 vnode_t *tvp = ap->a_tvp;
6220 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART));
6221 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART));
6223 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp,
6224 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0);
6239 zfs_freebsd_symlink(ap)
6240 struct vop_symlink_args /* {
6241 struct vnode *a_dvp;
6242 struct vnode **a_vpp;
6243 struct componentname *a_cnp;
6244 struct vattr *a_vap;
6248 struct componentname *cnp = ap->a_cnp;
6249 vattr_t *vap = ap->a_vap;
6251 ASSERT(cnp->cn_flags & SAVENAME);
6253 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */
6254 vattr_init_mask(vap);
6256 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap,
6257 ap->a_target, cnp->cn_cred, cnp->cn_thread));
6261 zfs_freebsd_readlink(ap)
6262 struct vop_readlink_args /* {
6265 struct ucred *a_cred;
6269 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL));
6273 zfs_freebsd_link(ap)
6274 struct vop_link_args /* {
6275 struct vnode *a_tdvp;
6277 struct componentname *a_cnp;
6280 struct componentname *cnp = ap->a_cnp;
6282 ASSERT(cnp->cn_flags & SAVENAME);
6284 return (zfs_link(ap->a_tdvp, ap->a_vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0));
6288 zfs_freebsd_inactive(ap)
6289 struct vop_inactive_args /* {
6291 struct thread *a_td;
6294 vnode_t *vp = ap->a_vp;
6296 zfs_inactive(vp, ap->a_td->td_ucred, NULL);
6301 zfs_freebsd_reclaim(ap)
6302 struct vop_reclaim_args /* {
6304 struct thread *a_td;
6307 vnode_t *vp = ap->a_vp;
6308 znode_t *zp = VTOZ(vp);
6309 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
6313 /* Destroy the vm object and flush associated pages. */
6314 vnode_destroy_vobject(vp);
6317 * z_teardown_inactive_lock protects from a race with
6318 * zfs_znode_dmu_fini in zfsvfs_teardown during
6321 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
6322 if (zp->z_sa_hdl == NULL)
6326 rw_exit(&zfsvfs->z_teardown_inactive_lock);
6334 struct vop_fid_args /* {
6340 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL));
6344 zfs_freebsd_pathconf(ap)
6345 struct vop_pathconf_args /* {
6348 register_t *a_retval;
6354 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL);
6356 *ap->a_retval = val;
6357 else if (error == EOPNOTSUPP)
6358 error = vop_stdpathconf(ap);
6363 zfs_freebsd_fifo_pathconf(ap)
6364 struct vop_pathconf_args /* {
6367 register_t *a_retval;
6371 switch (ap->a_name) {
6372 case _PC_ACL_EXTENDED:
6374 case _PC_ACL_PATH_MAX:
6375 case _PC_MAC_PRESENT:
6376 return (zfs_freebsd_pathconf(ap));
6378 return (fifo_specops.vop_pathconf(ap));
6383 * FreeBSD's extended attributes namespace defines file name prefix for ZFS'
6384 * extended attribute name:
6387 * system freebsd:system:
6388 * user (none, can be used to access ZFS fsattr(5) attributes
6389 * created on Solaris)
6392 zfs_create_attrname(int attrnamespace, const char *name, char *attrname,
6395 const char *namespace, *prefix, *suffix;
6397 /* We don't allow '/' character in attribute name. */
6398 if (strchr(name, '/') != NULL)
6400 /* We don't allow attribute names that start with "freebsd:" string. */
6401 if (strncmp(name, "freebsd:", 8) == 0)
6404 bzero(attrname, size);
6406 switch (attrnamespace) {
6407 case EXTATTR_NAMESPACE_USER:
6409 prefix = "freebsd:";
6410 namespace = EXTATTR_NAMESPACE_USER_STRING;
6414 * This is the default namespace by which we can access all
6415 * attributes created on Solaris.
6417 prefix = namespace = suffix = "";
6420 case EXTATTR_NAMESPACE_SYSTEM:
6421 prefix = "freebsd:";
6422 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING;
6425 case EXTATTR_NAMESPACE_EMPTY:
6429 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix,
6431 return (ENAMETOOLONG);
6437 * Vnode operating to retrieve a named extended attribute.
6440 zfs_getextattr(struct vop_getextattr_args *ap)
6443 IN struct vnode *a_vp;
6444 IN int a_attrnamespace;
6445 IN const char *a_name;
6446 INOUT struct uio *a_uio;
6448 IN struct ucred *a_cred;
6449 IN struct thread *a_td;
6453 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6454 struct thread *td = ap->a_td;
6455 struct nameidata nd;
6458 vnode_t *xvp = NULL, *vp;
6461 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6462 ap->a_cred, ap->a_td, VREAD);
6466 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6473 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6481 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_SYSSPACE, attrname,
6483 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL);
6485 NDFREE(&nd, NDF_ONLY_PNBUF);
6488 if (error == ENOENT)
6493 if (ap->a_size != NULL) {
6494 error = VOP_GETATTR(vp, &va, ap->a_cred);
6496 *ap->a_size = (size_t)va.va_size;
6497 } else if (ap->a_uio != NULL)
6498 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6501 vn_close(vp, flags, ap->a_cred, td);
6508 * Vnode operation to remove a named attribute.
6511 zfs_deleteextattr(struct vop_deleteextattr_args *ap)
6514 IN struct vnode *a_vp;
6515 IN int a_attrnamespace;
6516 IN const char *a_name;
6517 IN struct ucred *a_cred;
6518 IN struct thread *a_td;
6522 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6523 struct thread *td = ap->a_td;
6524 struct nameidata nd;
6527 vnode_t *xvp = NULL, *vp;
6530 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6531 ap->a_cred, ap->a_td, VWRITE);
6535 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6542 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6549 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF | MPSAFE,
6550 UIO_SYSSPACE, attrname, xvp, td);
6553 NDFREE(&nd, NDF_ONLY_PNBUF);
6556 if (error == ENOENT)
6560 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
6563 if (vp == nd.ni_dvp)
6573 * Vnode operation to set a named attribute.
6576 zfs_setextattr(struct vop_setextattr_args *ap)
6579 IN struct vnode *a_vp;
6580 IN int a_attrnamespace;
6581 IN const char *a_name;
6582 INOUT struct uio *a_uio;
6583 IN struct ucred *a_cred;
6584 IN struct thread *a_td;
6588 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6589 struct thread *td = ap->a_td;
6590 struct nameidata nd;
6593 vnode_t *xvp = NULL, *vp;
6596 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6597 ap->a_cred, ap->a_td, VWRITE);
6601 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6608 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6609 LOOKUP_XATTR | CREATE_XATTR_DIR);
6615 flags = FFLAGS(O_WRONLY | O_CREAT);
6616 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_SYSSPACE, attrname,
6618 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL);
6620 NDFREE(&nd, NDF_ONLY_PNBUF);
6628 error = VOP_SETATTR(vp, &va, ap->a_cred);
6630 VOP_WRITE(vp, ap->a_uio, IO_UNIT | IO_SYNC, ap->a_cred);
6633 vn_close(vp, flags, ap->a_cred, td);
6640 * Vnode operation to retrieve extended attributes on a vnode.
6643 zfs_listextattr(struct vop_listextattr_args *ap)
6646 IN struct vnode *a_vp;
6647 IN int a_attrnamespace;
6648 INOUT struct uio *a_uio;
6650 IN struct ucred *a_cred;
6651 IN struct thread *a_td;
6655 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6656 struct thread *td = ap->a_td;
6657 struct nameidata nd;
6658 char attrprefix[16];
6659 u_char dirbuf[sizeof(struct dirent)];
6662 struct uio auio, *uio = ap->a_uio;
6663 size_t *sizep = ap->a_size;
6665 vnode_t *xvp = NULL, *vp;
6666 int done, error, eof, pos;
6668 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6669 ap->a_cred, ap->a_td, VREAD);
6673 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix,
6674 sizeof(attrprefix));
6677 plen = strlen(attrprefix);
6684 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6689 * ENOATTR means that the EA directory does not yet exist,
6690 * i.e. there are no extended attributes there.
6692 if (error == ENOATTR)
6697 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED | MPSAFE,
6698 UIO_SYSSPACE, ".", xvp, td);
6701 NDFREE(&nd, NDF_ONLY_PNBUF);
6707 auio.uio_iov = &aiov;
6708 auio.uio_iovcnt = 1;
6709 auio.uio_segflg = UIO_SYSSPACE;
6711 auio.uio_rw = UIO_READ;
6712 auio.uio_offset = 0;
6717 aiov.iov_base = (void *)dirbuf;
6718 aiov.iov_len = sizeof(dirbuf);
6719 auio.uio_resid = sizeof(dirbuf);
6720 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL);
6721 done = sizeof(dirbuf) - auio.uio_resid;
6724 for (pos = 0; pos < done;) {
6725 dp = (struct dirent *)(dirbuf + pos);
6726 pos += dp->d_reclen;
6728 * XXX: Temporarily we also accept DT_UNKNOWN, as this
6729 * is what we get when attribute was created on Solaris.
6731 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN)
6733 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0)
6735 else if (strncmp(dp->d_name, attrprefix, plen) != 0)
6737 nlen = dp->d_namlen - plen;
6740 else if (uio != NULL) {
6742 * Format of extattr name entry is one byte for
6743 * length and the rest for name.
6745 error = uiomove(&nlen, 1, uio->uio_rw, uio);
6747 error = uiomove(dp->d_name + plen, nlen,
6754 } while (!eof && error == 0);
6763 zfs_freebsd_getacl(ap)
6764 struct vop_getacl_args /* {
6773 vsecattr_t vsecattr;
6775 if (ap->a_type != ACL_TYPE_NFS4)
6778 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
6779 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL))
6782 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt);
6783 if (vsecattr.vsa_aclentp != NULL)
6784 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz);
6790 zfs_freebsd_setacl(ap)
6791 struct vop_setacl_args /* {
6800 vsecattr_t vsecattr;
6801 int aclbsize; /* size of acl list in bytes */
6804 if (ap->a_type != ACL_TYPE_NFS4)
6807 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES)
6811 * With NFSv4 ACLs, chmod(2) may need to add additional entries,
6812 * splitting every entry into two and appending "canonical six"
6813 * entries at the end. Don't allow for setting an ACL that would
6814 * cause chmod(2) to run out of ACL entries.
6816 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES)
6819 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR);
6823 vsecattr.vsa_mask = VSA_ACE;
6824 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t);
6825 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP);
6826 aaclp = vsecattr.vsa_aclentp;
6827 vsecattr.vsa_aclentsz = aclbsize;
6829 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp);
6830 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL);
6831 kmem_free(aaclp, aclbsize);
6837 zfs_freebsd_aclcheck(ap)
6838 struct vop_aclcheck_args /* {
6847 return (EOPNOTSUPP);
6850 struct vop_vector zfs_vnodeops;
6851 struct vop_vector zfs_fifoops;
6852 struct vop_vector zfs_shareops;
6854 struct vop_vector zfs_vnodeops = {
6855 .vop_default = &default_vnodeops,
6856 .vop_inactive = zfs_freebsd_inactive,
6857 .vop_reclaim = zfs_freebsd_reclaim,
6858 .vop_access = zfs_freebsd_access,
6859 #ifdef FREEBSD_NAMECACHE
6860 .vop_lookup = vfs_cache_lookup,
6861 .vop_cachedlookup = zfs_freebsd_lookup,
6863 .vop_lookup = zfs_freebsd_lookup,
6865 .vop_getattr = zfs_freebsd_getattr,
6866 .vop_setattr = zfs_freebsd_setattr,
6867 .vop_create = zfs_freebsd_create,
6868 .vop_mknod = zfs_freebsd_create,
6869 .vop_mkdir = zfs_freebsd_mkdir,
6870 .vop_readdir = zfs_freebsd_readdir,
6871 .vop_fsync = zfs_freebsd_fsync,
6872 .vop_open = zfs_freebsd_open,
6873 .vop_close = zfs_freebsd_close,
6874 .vop_rmdir = zfs_freebsd_rmdir,
6875 .vop_ioctl = zfs_freebsd_ioctl,
6876 .vop_link = zfs_freebsd_link,
6877 .vop_symlink = zfs_freebsd_symlink,
6878 .vop_readlink = zfs_freebsd_readlink,
6879 .vop_read = zfs_freebsd_read,
6880 .vop_write = zfs_freebsd_write,
6881 .vop_remove = zfs_freebsd_remove,
6882 .vop_rename = zfs_freebsd_rename,
6883 .vop_pathconf = zfs_freebsd_pathconf,
6884 .vop_bmap = zfs_freebsd_bmap,
6885 .vop_fid = zfs_freebsd_fid,
6886 .vop_getextattr = zfs_getextattr,
6887 .vop_deleteextattr = zfs_deleteextattr,
6888 .vop_setextattr = zfs_setextattr,
6889 .vop_listextattr = zfs_listextattr,
6890 .vop_getacl = zfs_freebsd_getacl,
6891 .vop_setacl = zfs_freebsd_setacl,
6892 .vop_aclcheck = zfs_freebsd_aclcheck,
6893 .vop_getpages = zfs_freebsd_getpages,
6896 struct vop_vector zfs_fifoops = {
6897 .vop_default = &fifo_specops,
6898 .vop_fsync = zfs_freebsd_fsync,
6899 .vop_access = zfs_freebsd_access,
6900 .vop_getattr = zfs_freebsd_getattr,
6901 .vop_inactive = zfs_freebsd_inactive,
6902 .vop_read = VOP_PANIC,
6903 .vop_reclaim = zfs_freebsd_reclaim,
6904 .vop_setattr = zfs_freebsd_setattr,
6905 .vop_write = VOP_PANIC,
6906 .vop_pathconf = zfs_freebsd_fifo_pathconf,
6907 .vop_fid = zfs_freebsd_fid,
6908 .vop_getacl = zfs_freebsd_getacl,
6909 .vop_setacl = zfs_freebsd_setacl,
6910 .vop_aclcheck = zfs_freebsd_aclcheck,
6914 * special share hidden files vnode operations template
6916 struct vop_vector zfs_shareops = {
6917 .vop_default = &default_vnodeops,
6918 .vop_access = zfs_freebsd_access,
6919 .vop_inactive = zfs_freebsd_inactive,
6920 .vop_reclaim = zfs_freebsd_reclaim,
6921 .vop_fid = zfs_freebsd_fid,
6922 .vop_pathconf = zfs_freebsd_pathconf,