4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012 by Delphix. All rights reserved.
26 /* Portions Copyright 2007 Jeremy Teo */
27 /* Portions Copyright 2010 Robert Milkowski */
29 #include <sys/types.h>
30 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/sysmacros.h>
34 #include <sys/resource.h>
37 #include <sys/vnode.h>
41 #include <sys/taskq.h>
43 #include <sys/atomic.h>
44 #include <sys/namei.h>
46 #include <sys/cmn_err.h>
47 #include <sys/errno.h>
48 #include <sys/unistd.h>
49 #include <sys/zfs_dir.h>
50 #include <sys/zfs_ioctl.h>
51 #include <sys/fs/zfs.h>
53 #include <sys/dmu_objset.h>
59 #include <sys/dirent.h>
60 #include <sys/policy.h>
61 #include <sys/sunddi.h>
62 #include <sys/filio.h>
64 #include <sys/zfs_ctldir.h>
65 #include <sys/zfs_fuid.h>
66 #include <sys/zfs_sa.h>
68 #include <sys/zfs_rlock.h>
69 #include <sys/extdirent.h>
70 #include <sys/kidmap.h>
73 #include <sys/sf_buf.h>
74 #include <sys/sched.h>
76 #include <vm/vm_param.h>
77 #include <vm/vm_pageout.h>
82 * Each vnode op performs some logical unit of work. To do this, the ZPL must
83 * properly lock its in-core state, create a DMU transaction, do the work,
84 * record this work in the intent log (ZIL), commit the DMU transaction,
85 * and wait for the intent log to commit if it is a synchronous operation.
86 * Moreover, the vnode ops must work in both normal and log replay context.
87 * The ordering of events is important to avoid deadlocks and references
88 * to freed memory. The example below illustrates the following Big Rules:
90 * (1) A check must be made in each zfs thread for a mounted file system.
91 * This is done avoiding races using ZFS_ENTER(zfsvfs).
92 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
93 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
94 * can return EIO from the calling function.
96 * (2) VN_RELE() should always be the last thing except for zil_commit()
97 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
98 * First, if it's the last reference, the vnode/znode
99 * can be freed, so the zp may point to freed memory. Second, the last
100 * reference will call zfs_zinactive(), which may induce a lot of work --
101 * pushing cached pages (which acquires range locks) and syncing out
102 * cached atime changes. Third, zfs_zinactive() may require a new tx,
103 * which could deadlock the system if you were already holding one.
104 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
106 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
107 * as they can span dmu_tx_assign() calls.
109 * (4) Always pass TXG_NOWAIT as the second argument to dmu_tx_assign().
110 * This is critical because we don't want to block while holding locks.
111 * Note, in particular, that if a lock is sometimes acquired before
112 * the tx assigns, and sometimes after (e.g. z_lock), then failing to
113 * use a non-blocking assign can deadlock the system. The scenario:
115 * Thread A has grabbed a lock before calling dmu_tx_assign().
116 * Thread B is in an already-assigned tx, and blocks for this lock.
117 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
118 * forever, because the previous txg can't quiesce until B's tx commits.
120 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
121 * then drop all locks, call dmu_tx_wait(), and try again.
123 * (5) If the operation succeeded, generate the intent log entry for it
124 * before dropping locks. This ensures that the ordering of events
125 * in the intent log matches the order in which they actually occurred.
126 * During ZIL replay the zfs_log_* functions will update the sequence
127 * number to indicate the zil transaction has replayed.
129 * (6) At the end of each vnode op, the DMU tx must always commit,
130 * regardless of whether there were any errors.
132 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
133 * to ensure that synchronous semantics are provided when necessary.
135 * In general, this is how things should be ordered in each vnode op:
137 * ZFS_ENTER(zfsvfs); // exit if unmounted
139 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD())
140 * rw_enter(...); // grab any other locks you need
141 * tx = dmu_tx_create(...); // get DMU tx
142 * dmu_tx_hold_*(); // hold each object you might modify
143 * error = dmu_tx_assign(tx, TXG_NOWAIT); // try to assign
145 * rw_exit(...); // drop locks
146 * zfs_dirent_unlock(dl); // unlock directory entry
147 * VN_RELE(...); // release held vnodes
148 * if (error == ERESTART) {
153 * dmu_tx_abort(tx); // abort DMU tx
154 * ZFS_EXIT(zfsvfs); // finished in zfs
155 * return (error); // really out of space
157 * error = do_real_work(); // do whatever this VOP does
159 * zfs_log_*(...); // on success, make ZIL entry
160 * dmu_tx_commit(tx); // commit DMU tx -- error or not
161 * rw_exit(...); // drop locks
162 * zfs_dirent_unlock(dl); // unlock directory entry
163 * VN_RELE(...); // release held vnodes
164 * zil_commit(zilog, foid); // synchronous when necessary
165 * ZFS_EXIT(zfsvfs); // finished in zfs
166 * return (error); // done, report error
171 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
173 znode_t *zp = VTOZ(*vpp);
174 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
179 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
180 ((flag & FAPPEND) == 0)) {
185 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
186 ZTOV(zp)->v_type == VREG &&
187 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
188 if (fs_vscan(*vpp, cr, 0) != 0) {
194 /* Keep a count of the synchronous opens in the znode */
195 if (flag & (FSYNC | FDSYNC))
196 atomic_inc_32(&zp->z_sync_cnt);
204 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
205 caller_context_t *ct)
207 znode_t *zp = VTOZ(vp);
208 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
211 * Clean up any locks held by this process on the vp.
213 cleanlocks(vp, ddi_get_pid(), 0);
214 cleanshares(vp, ddi_get_pid());
219 /* Decrement the synchronous opens in the znode */
220 if ((flag & (FSYNC | FDSYNC)) && (count == 1))
221 atomic_dec_32(&zp->z_sync_cnt);
223 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
224 ZTOV(zp)->v_type == VREG &&
225 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
226 VERIFY(fs_vscan(vp, cr, 1) == 0);
233 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
234 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
237 zfs_holey(vnode_t *vp, u_long cmd, offset_t *off)
239 znode_t *zp = VTOZ(vp);
240 uint64_t noff = (uint64_t)*off; /* new offset */
245 file_sz = zp->z_size;
246 if (noff >= file_sz) {
250 if (cmd == _FIO_SEEK_HOLE)
255 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
258 if ((error == ESRCH) || (noff > file_sz)) {
260 * Handle the virtual hole at the end of file.
277 zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred,
278 int *rvalp, caller_context_t *ct)
290 * The following two ioctls are used by bfu. Faking out,
291 * necessary to avoid bfu errors.
300 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
303 off = *(offset_t *)data;
306 zfsvfs = zp->z_zfsvfs;
310 /* offset parameter is in/out */
311 error = zfs_holey(vp, com, &off);
316 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
319 *(offset_t *)data = off;
327 page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes)
333 zfs_vmobject_assert_wlocked(obj);
336 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
338 if ((pp->oflags & VPO_BUSY) != 0) {
340 * Reference the page before unlocking and
341 * sleeping so that the page daemon is less
342 * likely to reclaim it.
344 vm_page_reference(pp);
345 vm_page_sleep(pp, "zfsmwb");
349 pp = vm_page_alloc(obj, OFF_TO_IDX(start),
350 VM_ALLOC_SYSTEM | VM_ALLOC_IFCACHED |
355 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
356 vm_object_pip_add(obj, 1);
357 vm_page_io_start(pp);
358 pmap_remove_write(pp);
359 vm_page_clear_dirty(pp, off, nbytes);
367 page_unbusy(vm_page_t pp)
370 vm_page_io_finish(pp);
371 vm_object_pip_subtract(pp->object, 1);
375 page_hold(vnode_t *vp, int64_t start)
381 zfs_vmobject_assert_wlocked(obj);
384 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
386 if ((pp->oflags & VPO_BUSY) != 0) {
388 * Reference the page before unlocking and
389 * sleeping so that the page daemon is less
390 * likely to reclaim it.
392 vm_page_reference(pp);
393 vm_page_sleep(pp, "zfsmwb");
397 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
410 page_unhold(vm_page_t pp)
419 zfs_map_page(vm_page_t pp, struct sf_buf **sfp)
422 *sfp = sf_buf_alloc(pp, 0);
423 return ((caddr_t)sf_buf_kva(*sfp));
427 zfs_unmap_page(struct sf_buf *sf)
434 * When a file is memory mapped, we must keep the IO data synchronized
435 * between the DMU cache and the memory mapped pages. What this means:
437 * On Write: If we find a memory mapped page, we write to *both*
438 * the page and the dmu buffer.
441 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid,
442 int segflg, dmu_tx_t *tx)
449 ASSERT(vp->v_mount != NULL);
453 off = start & PAGEOFFSET;
454 zfs_vmobject_wlock(obj);
455 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
457 int nbytes = imin(PAGESIZE - off, len);
459 if (segflg == UIO_NOCOPY) {
460 pp = vm_page_lookup(obj, OFF_TO_IDX(start));
462 ("zfs update_pages: NULL page in putpages case"));
464 ("zfs update_pages: unaligned data in putpages case"));
465 KASSERT(pp->valid == VM_PAGE_BITS_ALL,
466 ("zfs update_pages: invalid page in putpages case"));
467 KASSERT(pp->busy > 0,
468 ("zfs update_pages: unbusy page in putpages case"));
469 KASSERT(!pmap_page_is_write_mapped(pp),
470 ("zfs update_pages: writable page in putpages case"));
471 zfs_vmobject_wunlock(obj);
473 va = zfs_map_page(pp, &sf);
474 (void) dmu_write(os, oid, start, nbytes, va, tx);
477 zfs_vmobject_wlock(obj);
479 } else if ((pp = page_busy(vp, start, off, nbytes)) != NULL) {
480 zfs_vmobject_wunlock(obj);
482 va = zfs_map_page(pp, &sf);
483 (void) dmu_read(os, oid, start+off, nbytes,
484 va+off, DMU_READ_PREFETCH);;
487 zfs_vmobject_wlock(obj);
493 if (segflg != UIO_NOCOPY)
494 vm_object_pip_wakeupn(obj, 0);
495 zfs_vmobject_wunlock(obj);
499 * Read with UIO_NOCOPY flag means that sendfile(2) requests
500 * ZFS to populate a range of page cache pages with data.
502 * NOTE: this function could be optimized to pre-allocate
503 * all pages in advance, drain VPO_BUSY on all of them,
504 * map them into contiguous KVA region and populate them
505 * in one single dmu_read() call.
508 mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio)
510 znode_t *zp = VTOZ(vp);
511 objset_t *os = zp->z_zfsvfs->z_os;
521 ASSERT(uio->uio_segflg == UIO_NOCOPY);
522 ASSERT(vp->v_mount != NULL);
525 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0);
527 zfs_vmobject_wlock(obj);
528 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) {
529 int bytes = MIN(PAGESIZE, len);
531 pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_NOBUSY |
532 VM_ALLOC_NORMAL | VM_ALLOC_RETRY | VM_ALLOC_IGN_SBUSY);
533 if (pp->valid == 0) {
534 vm_page_io_start(pp);
535 zfs_vmobject_wunlock(obj);
536 va = zfs_map_page(pp, &sf);
537 error = dmu_read(os, zp->z_id, start, bytes, va,
539 if (bytes != PAGESIZE && error == 0)
540 bzero(va + bytes, PAGESIZE - bytes);
542 zfs_vmobject_wlock(obj);
543 vm_page_io_finish(pp);
548 pp->valid = VM_PAGE_BITS_ALL;
549 vm_page_activate(pp);
555 uio->uio_resid -= bytes;
556 uio->uio_offset += bytes;
559 zfs_vmobject_wunlock(obj);
564 * When a file is memory mapped, we must keep the IO data synchronized
565 * between the DMU cache and the memory mapped pages. What this means:
567 * On Read: We "read" preferentially from memory mapped pages,
568 * else we default from the dmu buffer.
570 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
571 * the file is memory mapped.
574 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
576 znode_t *zp = VTOZ(vp);
577 objset_t *os = zp->z_zfsvfs->z_os;
585 ASSERT(vp->v_mount != NULL);
589 start = uio->uio_loffset;
590 off = start & PAGEOFFSET;
591 zfs_vmobject_wlock(obj);
592 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
594 uint64_t bytes = MIN(PAGESIZE - off, len);
596 if (pp = page_hold(vp, start)) {
600 zfs_vmobject_wunlock(obj);
601 va = zfs_map_page(pp, &sf);
602 error = uiomove(va + off, bytes, UIO_READ, uio);
604 zfs_vmobject_wlock(obj);
607 zfs_vmobject_wunlock(obj);
608 error = dmu_read_uio(os, zp->z_id, uio, bytes);
609 zfs_vmobject_wlock(obj);
616 zfs_vmobject_wunlock(obj);
620 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
623 * Read bytes from specified file into supplied buffer.
625 * IN: vp - vnode of file to be read from.
626 * uio - structure supplying read location, range info,
628 * ioflag - SYNC flags; used to provide FRSYNC semantics.
629 * cr - credentials of caller.
630 * ct - caller context
632 * OUT: uio - updated offset and range, buffer filled.
634 * RETURN: 0 if success
635 * error code if failure
638 * vp - atime updated if byte count > 0
642 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
644 znode_t *zp = VTOZ(vp);
645 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
656 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
662 * Validate file offset
664 if (uio->uio_loffset < (offset_t)0) {
670 * Fasttrack empty reads
672 if (uio->uio_resid == 0) {
678 * Check for mandatory locks
680 if (MANDMODE(zp->z_mode)) {
681 if (error = chklock(vp, FREAD,
682 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
689 * If we're in FRSYNC mode, sync out this znode before reading it.
692 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
693 zil_commit(zfsvfs->z_log, zp->z_id);
696 * Lock the range against changes.
698 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
701 * If we are reading past end-of-file we can skip
702 * to the end; but we might still need to set atime.
704 if (uio->uio_loffset >= zp->z_size) {
709 ASSERT(uio->uio_loffset < zp->z_size);
710 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
713 if ((uio->uio_extflg == UIO_XUIO) &&
714 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
716 int blksz = zp->z_blksz;
717 uint64_t offset = uio->uio_loffset;
719 xuio = (xuio_t *)uio;
721 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
724 ASSERT(offset + n <= blksz);
727 (void) dmu_xuio_init(xuio, nblk);
729 if (vn_has_cached_data(vp)) {
731 * For simplicity, we always allocate a full buffer
732 * even if we only expect to read a portion of a block.
734 while (--nblk >= 0) {
735 (void) dmu_xuio_add(xuio,
736 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
744 nbytes = MIN(n, zfs_read_chunk_size -
745 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
748 if (uio->uio_segflg == UIO_NOCOPY)
749 error = mappedread_sf(vp, nbytes, uio);
751 #endif /* __FreeBSD__ */
752 if (vn_has_cached_data(vp))
753 error = mappedread(vp, nbytes, uio);
755 error = dmu_read_uio(os, zp->z_id, uio, nbytes);
757 /* convert checksum errors into IO errors */
766 zfs_range_unlock(rl);
768 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
774 * Write the bytes to a file.
776 * IN: vp - vnode of file to be written to.
777 * uio - structure supplying write location, range info,
779 * ioflag - FAPPEND flag set if in append mode.
780 * cr - credentials of caller.
781 * ct - caller context (NFS/CIFS fem monitor only)
783 * OUT: uio - updated offset and range.
785 * RETURN: 0 if success
786 * error code if failure
789 * vp - ctime|mtime updated if byte count > 0
794 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
796 znode_t *zp = VTOZ(vp);
797 rlim64_t limit = MAXOFFSET_T;
798 ssize_t start_resid = uio->uio_resid;
802 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
807 int max_blksz = zfsvfs->z_max_blksz;
810 iovec_t *aiov = NULL;
813 int iovcnt = uio->uio_iovcnt;
814 iovec_t *iovp = uio->uio_iov;
817 sa_bulk_attr_t bulk[4];
818 uint64_t mtime[2], ctime[2];
821 * Fasttrack empty write
827 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
833 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
834 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
835 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
837 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
841 * If immutable or not appending then return EPERM
843 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
844 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
845 (uio->uio_loffset < zp->z_size))) {
850 zilog = zfsvfs->z_log;
853 * Validate file offset
855 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
862 * Check for mandatory locks before calling zfs_range_lock()
863 * in order to prevent a deadlock with locks set via fcntl().
865 if (MANDMODE((mode_t)zp->z_mode) &&
866 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
873 * Pre-fault the pages to ensure slow (eg NFS) pages
875 * Skip this if uio contains loaned arc_buf.
877 if ((uio->uio_extflg == UIO_XUIO) &&
878 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
879 xuio = (xuio_t *)uio;
881 uio_prefaultpages(MIN(n, max_blksz), uio);
885 * If in append mode, set the io offset pointer to eof.
887 if (ioflag & FAPPEND) {
889 * Obtain an appending range lock to guarantee file append
890 * semantics. We reset the write offset once we have the lock.
892 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
894 if (rl->r_len == UINT64_MAX) {
896 * We overlocked the file because this write will cause
897 * the file block size to increase.
898 * Note that zp_size cannot change with this lock held.
902 uio->uio_loffset = woff;
905 * Note that if the file block size will change as a result of
906 * this write, then this range lock will lock the entire file
907 * so that we can re-write the block safely.
909 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
912 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) {
913 zfs_range_unlock(rl);
919 zfs_range_unlock(rl);
924 if ((woff + n) > limit || woff > (limit - n))
927 /* Will this write extend the file length? */
928 write_eof = (woff + n > zp->z_size);
930 end_size = MAX(zp->z_size, woff + n);
933 * Write the file in reasonable size chunks. Each chunk is written
934 * in a separate transaction; this keeps the intent log records small
935 * and allows us to do more fine-grained space accounting.
939 woff = uio->uio_loffset;
941 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
942 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
944 dmu_return_arcbuf(abuf);
949 if (xuio && abuf == NULL) {
950 ASSERT(i_iov < iovcnt);
952 abuf = dmu_xuio_arcbuf(xuio, i_iov);
953 dmu_xuio_clear(xuio, i_iov);
954 DTRACE_PROBE3(zfs_cp_write, int, i_iov,
955 iovec_t *, aiov, arc_buf_t *, abuf);
956 ASSERT((aiov->iov_base == abuf->b_data) ||
957 ((char *)aiov->iov_base - (char *)abuf->b_data +
958 aiov->iov_len == arc_buf_size(abuf)));
960 } else if (abuf == NULL && n >= max_blksz &&
961 woff >= zp->z_size &&
962 P2PHASE(woff, max_blksz) == 0 &&
963 zp->z_blksz == max_blksz) {
965 * This write covers a full block. "Borrow" a buffer
966 * from the dmu so that we can fill it before we enter
967 * a transaction. This avoids the possibility of
968 * holding up the transaction if the data copy hangs
969 * up on a pagefault (e.g., from an NFS server mapping).
973 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
975 ASSERT(abuf != NULL);
976 ASSERT(arc_buf_size(abuf) == max_blksz);
977 if (error = uiocopy(abuf->b_data, max_blksz,
978 UIO_WRITE, uio, &cbytes)) {
979 dmu_return_arcbuf(abuf);
982 ASSERT(cbytes == max_blksz);
986 * Start a transaction.
988 tx = dmu_tx_create(zfsvfs->z_os);
989 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
990 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
991 zfs_sa_upgrade_txholds(tx, zp);
992 error = dmu_tx_assign(tx, TXG_NOWAIT);
994 if (error == ERESTART) {
1001 dmu_return_arcbuf(abuf);
1006 * If zfs_range_lock() over-locked we grow the blocksize
1007 * and then reduce the lock range. This will only happen
1008 * on the first iteration since zfs_range_reduce() will
1009 * shrink down r_len to the appropriate size.
1011 if (rl->r_len == UINT64_MAX) {
1014 if (zp->z_blksz > max_blksz) {
1015 ASSERT(!ISP2(zp->z_blksz));
1016 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
1018 new_blksz = MIN(end_size, max_blksz);
1020 zfs_grow_blocksize(zp, new_blksz, tx);
1021 zfs_range_reduce(rl, woff, n);
1025 * XXX - should we really limit each write to z_max_blksz?
1026 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
1028 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
1030 if (woff + nbytes > zp->z_size)
1031 vnode_pager_setsize(vp, woff + nbytes);
1034 tx_bytes = uio->uio_resid;
1035 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
1037 tx_bytes -= uio->uio_resid;
1040 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
1042 * If this is not a full block write, but we are
1043 * extending the file past EOF and this data starts
1044 * block-aligned, use assign_arcbuf(). Otherwise,
1045 * write via dmu_write().
1047 if (tx_bytes < max_blksz && (!write_eof ||
1048 aiov->iov_base != abuf->b_data)) {
1050 dmu_write(zfsvfs->z_os, zp->z_id, woff,
1051 aiov->iov_len, aiov->iov_base, tx);
1052 dmu_return_arcbuf(abuf);
1053 xuio_stat_wbuf_copied();
1055 ASSERT(xuio || tx_bytes == max_blksz);
1056 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
1059 ASSERT(tx_bytes <= uio->uio_resid);
1060 uioskip(uio, tx_bytes);
1062 if (tx_bytes && vn_has_cached_data(vp)) {
1063 update_pages(vp, woff, tx_bytes, zfsvfs->z_os,
1064 zp->z_id, uio->uio_segflg, tx);
1068 * If we made no progress, we're done. If we made even
1069 * partial progress, update the znode and ZIL accordingly.
1071 if (tx_bytes == 0) {
1072 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
1073 (void *)&zp->z_size, sizeof (uint64_t), tx);
1080 * Clear Set-UID/Set-GID bits on successful write if not
1081 * privileged and at least one of the excute bits is set.
1083 * It would be nice to to this after all writes have
1084 * been done, but that would still expose the ISUID/ISGID
1085 * to another app after the partial write is committed.
1087 * Note: we don't call zfs_fuid_map_id() here because
1088 * user 0 is not an ephemeral uid.
1090 mutex_enter(&zp->z_acl_lock);
1091 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
1092 (S_IXUSR >> 6))) != 0 &&
1093 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
1094 secpolicy_vnode_setid_retain(vp, cr,
1095 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
1097 zp->z_mode &= ~(S_ISUID | S_ISGID);
1098 newmode = zp->z_mode;
1099 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
1100 (void *)&newmode, sizeof (uint64_t), tx);
1102 mutex_exit(&zp->z_acl_lock);
1104 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
1108 * Update the file size (zp_size) if it has changed;
1109 * account for possible concurrent updates.
1111 while ((end_size = zp->z_size) < uio->uio_loffset) {
1112 (void) atomic_cas_64(&zp->z_size, end_size,
1117 * If we are replaying and eof is non zero then force
1118 * the file size to the specified eof. Note, there's no
1119 * concurrency during replay.
1121 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
1122 zp->z_size = zfsvfs->z_replay_eof;
1124 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1126 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
1131 ASSERT(tx_bytes == nbytes);
1136 uio_prefaultpages(MIN(n, max_blksz), uio);
1140 zfs_range_unlock(rl);
1143 * If we're in replay mode, or we made no progress, return error.
1144 * Otherwise, it's at least a partial write, so it's successful.
1146 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1151 if (ioflag & (FSYNC | FDSYNC) ||
1152 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1153 zil_commit(zilog, zp->z_id);
1160 zfs_get_done(zgd_t *zgd, int error)
1162 znode_t *zp = zgd->zgd_private;
1163 objset_t *os = zp->z_zfsvfs->z_os;
1166 dmu_buf_rele(zgd->zgd_db, zgd);
1168 zfs_range_unlock(zgd->zgd_rl);
1171 * Release the vnode asynchronously as we currently have the
1172 * txg stopped from syncing.
1174 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1176 if (error == 0 && zgd->zgd_bp)
1177 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1179 kmem_free(zgd, sizeof (zgd_t));
1183 static int zil_fault_io = 0;
1187 * Get data to generate a TX_WRITE intent log record.
1190 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1192 zfsvfs_t *zfsvfs = arg;
1193 objset_t *os = zfsvfs->z_os;
1195 uint64_t object = lr->lr_foid;
1196 uint64_t offset = lr->lr_offset;
1197 uint64_t size = lr->lr_length;
1198 blkptr_t *bp = &lr->lr_blkptr;
1203 ASSERT(zio != NULL);
1207 * Nothing to do if the file has been removed
1209 if (zfs_zget(zfsvfs, object, &zp) != 0)
1211 if (zp->z_unlinked) {
1213 * Release the vnode asynchronously as we currently have the
1214 * txg stopped from syncing.
1216 VN_RELE_ASYNC(ZTOV(zp),
1217 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1221 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1222 zgd->zgd_zilog = zfsvfs->z_log;
1223 zgd->zgd_private = zp;
1226 * Write records come in two flavors: immediate and indirect.
1227 * For small writes it's cheaper to store the data with the
1228 * log record (immediate); for large writes it's cheaper to
1229 * sync the data and get a pointer to it (indirect) so that
1230 * we don't have to write the data twice.
1232 if (buf != NULL) { /* immediate write */
1233 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1234 /* test for truncation needs to be done while range locked */
1235 if (offset >= zp->z_size) {
1238 error = dmu_read(os, object, offset, size, buf,
1239 DMU_READ_NO_PREFETCH);
1241 ASSERT(error == 0 || error == ENOENT);
1242 } else { /* indirect write */
1244 * Have to lock the whole block to ensure when it's
1245 * written out and it's checksum is being calculated
1246 * that no one can change the data. We need to re-check
1247 * blocksize after we get the lock in case it's changed!
1252 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1254 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1256 if (zp->z_blksz == size)
1259 zfs_range_unlock(zgd->zgd_rl);
1261 /* test for truncation needs to be done while range locked */
1262 if (lr->lr_offset >= zp->z_size)
1271 error = dmu_buf_hold(os, object, offset, zgd, &db,
1272 DMU_READ_NO_PREFETCH);
1275 blkptr_t *obp = dmu_buf_get_blkptr(db);
1277 ASSERT(BP_IS_HOLE(bp));
1284 ASSERT(db->db_offset == offset);
1285 ASSERT(db->db_size == size);
1287 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1289 ASSERT(error || lr->lr_length <= zp->z_blksz);
1292 * On success, we need to wait for the write I/O
1293 * initiated by dmu_sync() to complete before we can
1294 * release this dbuf. We will finish everything up
1295 * in the zfs_get_done() callback.
1300 if (error == EALREADY) {
1301 lr->lr_common.lrc_txtype = TX_WRITE2;
1307 zfs_get_done(zgd, error);
1314 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1315 caller_context_t *ct)
1317 znode_t *zp = VTOZ(vp);
1318 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1324 if (flag & V_ACE_MASK)
1325 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1327 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1334 * If vnode is for a device return a specfs vnode instead.
1337 specvp_check(vnode_t **vpp, cred_t *cr)
1341 if (IS_DEVVP(*vpp)) {
1344 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1355 * Lookup an entry in a directory, or an extended attribute directory.
1356 * If it exists, return a held vnode reference for it.
1358 * IN: dvp - vnode of directory to search.
1359 * nm - name of entry to lookup.
1360 * pnp - full pathname to lookup [UNUSED].
1361 * flags - LOOKUP_XATTR set if looking for an attribute.
1362 * rdir - root directory vnode [UNUSED].
1363 * cr - credentials of caller.
1364 * ct - caller context
1365 * direntflags - directory lookup flags
1366 * realpnp - returned pathname.
1368 * OUT: vpp - vnode of located entry, NULL if not found.
1370 * RETURN: 0 if success
1371 * error code if failure
1378 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp,
1379 int nameiop, cred_t *cr, kthread_t *td, int flags)
1381 znode_t *zdp = VTOZ(dvp);
1382 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1384 int *direntflags = NULL;
1385 void *realpnp = NULL;
1388 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1390 if (dvp->v_type != VDIR) {
1392 } else if (zdp->z_sa_hdl == NULL) {
1396 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1397 error = zfs_fastaccesschk_execute(zdp, cr);
1405 vnode_t *tvp = dnlc_lookup(dvp, nm);
1408 error = zfs_fastaccesschk_execute(zdp, cr);
1413 if (tvp == DNLC_NO_VNODE) {
1418 return (specvp_check(vpp, cr));
1424 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1431 if (flags & LOOKUP_XATTR) {
1434 * If the xattr property is off, refuse the lookup request.
1436 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1443 * We don't allow recursive attributes..
1444 * Maybe someday we will.
1446 if (zdp->z_pflags & ZFS_XATTR) {
1451 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1457 * Do we have permission to get into attribute directory?
1460 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1470 if (dvp->v_type != VDIR) {
1476 * Check accessibility of directory.
1479 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1484 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1485 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1490 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1492 error = specvp_check(vpp, cr);
1494 /* Translate errors and add SAVENAME when needed. */
1495 if (cnp->cn_flags & ISLASTCN) {
1499 if (error == ENOENT) {
1500 error = EJUSTRETURN;
1501 cnp->cn_flags |= SAVENAME;
1507 cnp->cn_flags |= SAVENAME;
1511 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) {
1514 if (cnp->cn_flags & ISDOTDOT) {
1515 ltype = VOP_ISLOCKED(dvp);
1519 error = zfs_vnode_lock(*vpp, cnp->cn_lkflags);
1520 if (cnp->cn_flags & ISDOTDOT)
1521 vn_lock(dvp, ltype | LK_RETRY);
1531 #ifdef FREEBSD_NAMECACHE
1533 * Insert name into cache (as non-existent) if appropriate.
1535 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
1536 cache_enter(dvp, *vpp, cnp);
1538 * Insert name into cache if appropriate.
1540 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) {
1541 if (!(cnp->cn_flags & ISLASTCN) ||
1542 (nameiop != DELETE && nameiop != RENAME)) {
1543 cache_enter(dvp, *vpp, cnp);
1552 * Attempt to create a new entry in a directory. If the entry
1553 * already exists, truncate the file if permissible, else return
1554 * an error. Return the vp of the created or trunc'd file.
1556 * IN: dvp - vnode of directory to put new file entry in.
1557 * name - name of new file entry.
1558 * vap - attributes of new file.
1559 * excl - flag indicating exclusive or non-exclusive mode.
1560 * mode - mode to open file with.
1561 * cr - credentials of caller.
1562 * flag - large file flag [UNUSED].
1563 * ct - caller context
1564 * vsecp - ACL to be set
1566 * OUT: vpp - vnode of created or trunc'd entry.
1568 * RETURN: 0 if success
1569 * error code if failure
1572 * dvp - ctime|mtime updated if new entry created
1573 * vp - ctime|mtime always, atime if new
1578 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode,
1579 vnode_t **vpp, cred_t *cr, kthread_t *td)
1581 znode_t *zp, *dzp = VTOZ(dvp);
1582 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1590 gid_t gid = crgetgid(cr);
1591 zfs_acl_ids_t acl_ids;
1592 boolean_t fuid_dirtied;
1593 boolean_t have_acl = B_FALSE;
1598 * If we have an ephemeral id, ACL, or XVATTR then
1599 * make sure file system is at proper version
1602 ksid = crgetsid(cr, KSID_OWNER);
1604 uid = ksid_getid(ksid);
1608 if (zfsvfs->z_use_fuids == B_FALSE &&
1609 (vsecp || (vap->va_mask & AT_XVATTR) ||
1610 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1616 zilog = zfsvfs->z_log;
1618 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1619 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1624 if (vap->va_mask & AT_XVATTR) {
1625 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
1626 crgetuid(cr), cr, vap->va_type)) != 0) {
1634 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr))
1635 vap->va_mode &= ~S_ISVTX;
1637 if (*name == '\0') {
1639 * Null component name refers to the directory itself.
1646 /* possible VN_HOLD(zp) */
1649 if (flag & FIGNORECASE)
1652 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1656 zfs_acl_ids_free(&acl_ids);
1657 if (strcmp(name, "..") == 0)
1668 * Create a new file object and update the directory
1671 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1673 zfs_acl_ids_free(&acl_ids);
1678 * We only support the creation of regular files in
1679 * extended attribute directories.
1682 if ((dzp->z_pflags & ZFS_XATTR) &&
1683 (vap->va_type != VREG)) {
1685 zfs_acl_ids_free(&acl_ids);
1690 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1691 cr, vsecp, &acl_ids)) != 0)
1695 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1696 zfs_acl_ids_free(&acl_ids);
1701 tx = dmu_tx_create(os);
1703 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1704 ZFS_SA_BASE_ATTR_SIZE);
1706 fuid_dirtied = zfsvfs->z_fuid_dirty;
1708 zfs_fuid_txhold(zfsvfs, tx);
1709 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1710 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1711 if (!zfsvfs->z_use_sa &&
1712 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1713 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1714 0, acl_ids.z_aclp->z_acl_bytes);
1716 error = dmu_tx_assign(tx, TXG_NOWAIT);
1718 zfs_dirent_unlock(dl);
1719 if (error == ERESTART) {
1724 zfs_acl_ids_free(&acl_ids);
1729 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1732 zfs_fuid_sync(zfsvfs, tx);
1734 (void) zfs_link_create(dl, zp, tx, ZNEW);
1735 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1736 if (flag & FIGNORECASE)
1738 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1739 vsecp, acl_ids.z_fuidp, vap);
1740 zfs_acl_ids_free(&acl_ids);
1743 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1746 zfs_acl_ids_free(&acl_ids);
1750 * A directory entry already exists for this name.
1753 * Can't truncate an existing file if in exclusive mode.
1760 * Can't open a directory for writing.
1762 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1767 * Verify requested access to file.
1769 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1773 mutex_enter(&dzp->z_lock);
1775 mutex_exit(&dzp->z_lock);
1778 * Truncate regular files if requested.
1780 if ((ZTOV(zp)->v_type == VREG) &&
1781 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1782 /* we can't hold any locks when calling zfs_freesp() */
1783 zfs_dirent_unlock(dl);
1785 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1787 vnevent_create(ZTOV(zp), ct);
1793 zfs_dirent_unlock(dl);
1800 error = specvp_check(vpp, cr);
1803 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1804 zil_commit(zilog, 0);
1811 * Remove an entry from a directory.
1813 * IN: dvp - vnode of directory to remove entry from.
1814 * name - name of entry to remove.
1815 * cr - credentials of caller.
1816 * ct - caller context
1817 * flags - case flags
1819 * RETURN: 0 if success
1820 * error code if failure
1824 * vp - ctime (if nlink > 0)
1827 uint64_t null_xattr = 0;
1831 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1834 znode_t *zp, *dzp = VTOZ(dvp);
1837 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1839 uint64_t acl_obj, xattr_obj;
1840 uint64_t xattr_obj_unlinked = 0;
1844 boolean_t may_delete_now, delete_now = FALSE;
1845 boolean_t unlinked, toobig = FALSE;
1847 pathname_t *realnmp = NULL;
1854 zilog = zfsvfs->z_log;
1856 if (flags & FIGNORECASE) {
1866 * Attempt to lock directory; fail if entry doesn't exist.
1868 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1878 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1883 * Need to use rmdir for removing directories.
1885 if (vp->v_type == VDIR) {
1890 vnevent_remove(vp, dvp, name, ct);
1893 dnlc_remove(dvp, realnmp->pn_buf);
1895 dnlc_remove(dvp, name);
1898 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1902 * We may delete the znode now, or we may put it in the unlinked set;
1903 * it depends on whether we're the last link, and on whether there are
1904 * other holds on the vnode. So we dmu_tx_hold() the right things to
1905 * allow for either case.
1908 tx = dmu_tx_create(zfsvfs->z_os);
1909 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1910 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1911 zfs_sa_upgrade_txholds(tx, zp);
1912 zfs_sa_upgrade_txholds(tx, dzp);
1913 if (may_delete_now) {
1915 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1916 /* if the file is too big, only hold_free a token amount */
1917 dmu_tx_hold_free(tx, zp->z_id, 0,
1918 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1921 /* are there any extended attributes? */
1922 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1923 &xattr_obj, sizeof (xattr_obj));
1924 if (error == 0 && xattr_obj) {
1925 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1927 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1928 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1931 mutex_enter(&zp->z_lock);
1932 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1933 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1934 mutex_exit(&zp->z_lock);
1936 /* charge as an update -- would be nice not to charge at all */
1937 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1939 error = dmu_tx_assign(tx, TXG_NOWAIT);
1941 zfs_dirent_unlock(dl);
1945 if (error == ERESTART) {
1958 * Remove the directory entry.
1960 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1970 * Hold z_lock so that we can make sure that the ACL obj
1971 * hasn't changed. Could have been deleted due to
1974 mutex_enter(&zp->z_lock);
1976 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1977 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1978 delete_now = may_delete_now && !toobig &&
1979 vp->v_count == 1 && !vn_has_cached_data(vp) &&
1980 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1987 panic("zfs_remove: delete_now branch taken");
1989 if (xattr_obj_unlinked) {
1990 ASSERT3U(xzp->z_links, ==, 2);
1991 mutex_enter(&xzp->z_lock);
1992 xzp->z_unlinked = 1;
1994 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1995 &xzp->z_links, sizeof (xzp->z_links), tx);
1996 ASSERT3U(error, ==, 0);
1997 mutex_exit(&xzp->z_lock);
1998 zfs_unlinked_add(xzp, tx);
2001 error = sa_remove(zp->z_sa_hdl,
2002 SA_ZPL_XATTR(zfsvfs), tx);
2004 error = sa_update(zp->z_sa_hdl,
2005 SA_ZPL_XATTR(zfsvfs), &null_xattr,
2006 sizeof (uint64_t), tx);
2011 ASSERT0(vp->v_count);
2013 mutex_exit(&zp->z_lock);
2014 zfs_znode_delete(zp, tx);
2015 } else if (unlinked) {
2016 mutex_exit(&zp->z_lock);
2017 zfs_unlinked_add(zp, tx);
2019 vp->v_vflag |= VV_NOSYNC;
2024 if (flags & FIGNORECASE)
2026 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
2033 zfs_dirent_unlock(dl);
2040 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2041 zil_commit(zilog, 0);
2048 * Create a new directory and insert it into dvp using the name
2049 * provided. Return a pointer to the inserted directory.
2051 * IN: dvp - vnode of directory to add subdir to.
2052 * dirname - name of new directory.
2053 * vap - attributes of new directory.
2054 * cr - credentials of caller.
2055 * ct - caller context
2056 * vsecp - ACL to be set
2058 * OUT: vpp - vnode of created directory.
2060 * RETURN: 0 if success
2061 * error code if failure
2064 * dvp - ctime|mtime updated
2065 * vp - ctime|mtime|atime updated
2069 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
2070 caller_context_t *ct, int flags, vsecattr_t *vsecp)
2072 znode_t *zp, *dzp = VTOZ(dvp);
2073 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2082 gid_t gid = crgetgid(cr);
2083 zfs_acl_ids_t acl_ids;
2084 boolean_t fuid_dirtied;
2086 ASSERT(vap->va_type == VDIR);
2089 * If we have an ephemeral id, ACL, or XVATTR then
2090 * make sure file system is at proper version
2093 ksid = crgetsid(cr, KSID_OWNER);
2095 uid = ksid_getid(ksid);
2098 if (zfsvfs->z_use_fuids == B_FALSE &&
2099 (vsecp || (vap->va_mask & AT_XVATTR) ||
2100 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
2105 zilog = zfsvfs->z_log;
2107 if (dzp->z_pflags & ZFS_XATTR) {
2112 if (zfsvfs->z_utf8 && u8_validate(dirname,
2113 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2117 if (flags & FIGNORECASE)
2120 if (vap->va_mask & AT_XVATTR) {
2121 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
2122 crgetuid(cr), cr, vap->va_type)) != 0) {
2128 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
2129 vsecp, &acl_ids)) != 0) {
2134 * First make sure the new directory doesn't exist.
2136 * Existence is checked first to make sure we don't return
2137 * EACCES instead of EEXIST which can cause some applications
2143 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
2145 zfs_acl_ids_free(&acl_ids);
2150 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
2151 zfs_acl_ids_free(&acl_ids);
2152 zfs_dirent_unlock(dl);
2157 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
2158 zfs_acl_ids_free(&acl_ids);
2159 zfs_dirent_unlock(dl);
2165 * Add a new entry to the directory.
2167 tx = dmu_tx_create(zfsvfs->z_os);
2168 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2169 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2170 fuid_dirtied = zfsvfs->z_fuid_dirty;
2172 zfs_fuid_txhold(zfsvfs, tx);
2173 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2174 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2175 acl_ids.z_aclp->z_acl_bytes);
2178 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2179 ZFS_SA_BASE_ATTR_SIZE);
2181 error = dmu_tx_assign(tx, TXG_NOWAIT);
2183 zfs_dirent_unlock(dl);
2184 if (error == ERESTART) {
2189 zfs_acl_ids_free(&acl_ids);
2198 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2201 zfs_fuid_sync(zfsvfs, tx);
2204 * Now put new name in parent dir.
2206 (void) zfs_link_create(dl, zp, tx, ZNEW);
2210 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2211 if (flags & FIGNORECASE)
2213 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2214 acl_ids.z_fuidp, vap);
2216 zfs_acl_ids_free(&acl_ids);
2220 zfs_dirent_unlock(dl);
2222 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2223 zil_commit(zilog, 0);
2230 * Remove a directory subdir entry. If the current working
2231 * directory is the same as the subdir to be removed, the
2234 * IN: dvp - vnode of directory to remove from.
2235 * name - name of directory to be removed.
2236 * cwd - vnode of current working directory.
2237 * cr - credentials of caller.
2238 * ct - caller context
2239 * flags - case flags
2241 * RETURN: 0 if success
2242 * error code if failure
2245 * dvp - ctime|mtime updated
2249 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2250 caller_context_t *ct, int flags)
2252 znode_t *dzp = VTOZ(dvp);
2255 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2264 zilog = zfsvfs->z_log;
2266 if (flags & FIGNORECASE)
2272 * Attempt to lock directory; fail if entry doesn't exist.
2274 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2282 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2286 if (vp->v_type != VDIR) {
2296 vnevent_rmdir(vp, dvp, name, ct);
2299 * Grab a lock on the directory to make sure that noone is
2300 * trying to add (or lookup) entries while we are removing it.
2302 rw_enter(&zp->z_name_lock, RW_WRITER);
2305 * Grab a lock on the parent pointer to make sure we play well
2306 * with the treewalk and directory rename code.
2308 rw_enter(&zp->z_parent_lock, RW_WRITER);
2310 tx = dmu_tx_create(zfsvfs->z_os);
2311 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2312 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2313 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2314 zfs_sa_upgrade_txholds(tx, zp);
2315 zfs_sa_upgrade_txholds(tx, dzp);
2316 error = dmu_tx_assign(tx, TXG_NOWAIT);
2318 rw_exit(&zp->z_parent_lock);
2319 rw_exit(&zp->z_name_lock);
2320 zfs_dirent_unlock(dl);
2322 if (error == ERESTART) {
2332 #ifdef FREEBSD_NAMECACHE
2336 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2339 uint64_t txtype = TX_RMDIR;
2340 if (flags & FIGNORECASE)
2342 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2347 rw_exit(&zp->z_parent_lock);
2348 rw_exit(&zp->z_name_lock);
2349 #ifdef FREEBSD_NAMECACHE
2353 zfs_dirent_unlock(dl);
2357 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2358 zil_commit(zilog, 0);
2365 * Read as many directory entries as will fit into the provided
2366 * buffer from the given directory cursor position (specified in
2367 * the uio structure.
2369 * IN: vp - vnode of directory to read.
2370 * uio - structure supplying read location, range info,
2371 * and return buffer.
2372 * cr - credentials of caller.
2373 * ct - caller context
2374 * flags - case flags
2376 * OUT: uio - updated offset and range, buffer filled.
2377 * eofp - set to true if end-of-file detected.
2379 * RETURN: 0 if success
2380 * error code if failure
2383 * vp - atime updated
2385 * Note that the low 4 bits of the cookie returned by zap is always zero.
2386 * This allows us to use the low range for "special" directory entries:
2387 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2388 * we use the offset 2 for the '.zfs' directory.
2392 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies)
2394 znode_t *zp = VTOZ(vp);
2398 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2403 zap_attribute_t zap;
2404 uint_t bytes_wanted;
2405 uint64_t offset; /* must be unsigned; checks for < 1 */
2411 boolean_t check_sysattrs;
2414 u_long *cooks = NULL;
2420 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2421 &parent, sizeof (parent))) != 0) {
2427 * If we are not given an eof variable,
2434 * Check for valid iov_len.
2436 if (uio->uio_iov->iov_len <= 0) {
2442 * Quit if directory has been removed (posix)
2444 if ((*eofp = zp->z_unlinked) != 0) {
2451 offset = uio->uio_loffset;
2452 prefetch = zp->z_zn_prefetch;
2455 * Initialize the iterator cursor.
2459 * Start iteration from the beginning of the directory.
2461 zap_cursor_init(&zc, os, zp->z_id);
2464 * The offset is a serialized cursor.
2466 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2470 * Get space to change directory entries into fs independent format.
2472 iovp = uio->uio_iov;
2473 bytes_wanted = iovp->iov_len;
2474 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2475 bufsize = bytes_wanted;
2476 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2477 odp = (struct dirent64 *)outbuf;
2479 bufsize = bytes_wanted;
2481 odp = (struct dirent64 *)iovp->iov_base;
2483 eodp = (struct edirent *)odp;
2485 if (ncookies != NULL) {
2487 * Minimum entry size is dirent size and 1 byte for a file name.
2489 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1);
2490 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK);
2495 * If this VFS supports the system attribute view interface; and
2496 * we're looking at an extended attribute directory; and we care
2497 * about normalization conflicts on this vfs; then we must check
2498 * for normalization conflicts with the sysattr name space.
2501 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2502 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2503 (flags & V_RDDIR_ENTFLAGS);
2509 * Transform to file-system independent format
2512 while (outcount < bytes_wanted) {
2515 off64_t *next = NULL;
2518 * Special case `.', `..', and `.zfs'.
2521 (void) strcpy(zap.za_name, ".");
2522 zap.za_normalization_conflict = 0;
2525 } else if (offset == 1) {
2526 (void) strcpy(zap.za_name, "..");
2527 zap.za_normalization_conflict = 0;
2530 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2531 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2532 zap.za_normalization_conflict = 0;
2533 objnum = ZFSCTL_INO_ROOT;
2539 if (error = zap_cursor_retrieve(&zc, &zap)) {
2540 if ((*eofp = (error == ENOENT)) != 0)
2546 if (zap.za_integer_length != 8 ||
2547 zap.za_num_integers != 1) {
2548 cmn_err(CE_WARN, "zap_readdir: bad directory "
2549 "entry, obj = %lld, offset = %lld\n",
2550 (u_longlong_t)zp->z_id,
2551 (u_longlong_t)offset);
2556 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2558 * MacOS X can extract the object type here such as:
2559 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2561 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2563 if (check_sysattrs && !zap.za_normalization_conflict) {
2565 zap.za_normalization_conflict =
2566 xattr_sysattr_casechk(zap.za_name);
2568 panic("%s:%u: TODO", __func__, __LINE__);
2573 if (flags & V_RDDIR_ACCFILTER) {
2575 * If we have no access at all, don't include
2576 * this entry in the returned information
2579 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2581 if (!zfs_has_access(ezp, cr)) {
2588 if (flags & V_RDDIR_ENTFLAGS)
2589 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2591 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2594 * Will this entry fit in the buffer?
2596 if (outcount + reclen > bufsize) {
2598 * Did we manage to fit anything in the buffer?
2606 if (flags & V_RDDIR_ENTFLAGS) {
2608 * Add extended flag entry:
2610 eodp->ed_ino = objnum;
2611 eodp->ed_reclen = reclen;
2612 /* NOTE: ed_off is the offset for the *next* entry */
2613 next = &(eodp->ed_off);
2614 eodp->ed_eflags = zap.za_normalization_conflict ?
2615 ED_CASE_CONFLICT : 0;
2616 (void) strncpy(eodp->ed_name, zap.za_name,
2617 EDIRENT_NAMELEN(reclen));
2618 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2623 odp->d_ino = objnum;
2624 odp->d_reclen = reclen;
2625 odp->d_namlen = strlen(zap.za_name);
2626 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1);
2628 odp = (dirent64_t *)((intptr_t)odp + reclen);
2632 ASSERT(outcount <= bufsize);
2634 /* Prefetch znode */
2636 dmu_prefetch(os, objnum, 0, 0);
2640 * Move to the next entry, fill in the previous offset.
2642 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2643 zap_cursor_advance(&zc);
2644 offset = zap_cursor_serialize(&zc);
2649 if (cooks != NULL) {
2652 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks));
2655 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2657 /* Subtract unused cookies */
2658 if (ncookies != NULL)
2659 *ncookies -= ncooks;
2661 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2662 iovp->iov_base += outcount;
2663 iovp->iov_len -= outcount;
2664 uio->uio_resid -= outcount;
2665 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2667 * Reset the pointer.
2669 offset = uio->uio_loffset;
2673 zap_cursor_fini(&zc);
2674 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2675 kmem_free(outbuf, bufsize);
2677 if (error == ENOENT)
2680 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2682 uio->uio_loffset = offset;
2684 if (error != 0 && cookies != NULL) {
2685 free(*cookies, M_TEMP);
2692 ulong_t zfs_fsync_sync_cnt = 4;
2695 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2697 znode_t *zp = VTOZ(vp);
2698 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2700 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2702 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2705 zil_commit(zfsvfs->z_log, zp->z_id);
2713 * Get the requested file attributes and place them in the provided
2716 * IN: vp - vnode of file.
2717 * vap - va_mask identifies requested attributes.
2718 * If AT_XVATTR set, then optional attrs are requested
2719 * flags - ATTR_NOACLCHECK (CIFS server context)
2720 * cr - credentials of caller.
2721 * ct - caller context
2723 * OUT: vap - attribute values.
2725 * RETURN: 0 (always succeeds)
2729 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2730 caller_context_t *ct)
2732 znode_t *zp = VTOZ(vp);
2733 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2736 u_longlong_t nblocks;
2738 uint64_t mtime[2], ctime[2], crtime[2], rdev;
2739 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2740 xoptattr_t *xoap = NULL;
2741 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2742 sa_bulk_attr_t bulk[4];
2748 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2750 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2751 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2752 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
2753 if (vp->v_type == VBLK || vp->v_type == VCHR)
2754 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
2757 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2763 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2764 * Also, if we are the owner don't bother, since owner should
2765 * always be allowed to read basic attributes of file.
2767 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2768 (vap->va_uid != crgetuid(cr))) {
2769 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2777 * Return all attributes. It's cheaper to provide the answer
2778 * than to determine whether we were asked the question.
2781 mutex_enter(&zp->z_lock);
2782 vap->va_type = IFTOVT(zp->z_mode);
2783 vap->va_mode = zp->z_mode & ~S_IFMT;
2785 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2787 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
2789 vap->va_nodeid = zp->z_id;
2790 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2791 links = zp->z_links + 1;
2793 links = zp->z_links;
2794 vap->va_nlink = MIN(links, LINK_MAX); /* nlink_t limit! */
2795 vap->va_size = zp->z_size;
2797 vap->va_rdev = vp->v_rdev;
2799 if (vp->v_type == VBLK || vp->v_type == VCHR)
2800 vap->va_rdev = zfs_cmpldev(rdev);
2802 vap->va_seq = zp->z_seq;
2803 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */
2806 * Add in any requested optional attributes and the create time.
2807 * Also set the corresponding bits in the returned attribute bitmap.
2809 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2810 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2812 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2813 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2816 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2817 xoap->xoa_readonly =
2818 ((zp->z_pflags & ZFS_READONLY) != 0);
2819 XVA_SET_RTN(xvap, XAT_READONLY);
2822 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2824 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2825 XVA_SET_RTN(xvap, XAT_SYSTEM);
2828 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2830 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2831 XVA_SET_RTN(xvap, XAT_HIDDEN);
2834 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2835 xoap->xoa_nounlink =
2836 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2837 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2840 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2841 xoap->xoa_immutable =
2842 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2843 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2846 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2847 xoap->xoa_appendonly =
2848 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2849 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2852 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2854 ((zp->z_pflags & ZFS_NODUMP) != 0);
2855 XVA_SET_RTN(xvap, XAT_NODUMP);
2858 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2860 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2861 XVA_SET_RTN(xvap, XAT_OPAQUE);
2864 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2865 xoap->xoa_av_quarantined =
2866 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2867 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2870 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2871 xoap->xoa_av_modified =
2872 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2873 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2876 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2877 vp->v_type == VREG) {
2878 zfs_sa_get_scanstamp(zp, xvap);
2881 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2884 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2885 times, sizeof (times));
2886 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2887 XVA_SET_RTN(xvap, XAT_CREATETIME);
2890 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2891 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2892 XVA_SET_RTN(xvap, XAT_REPARSE);
2894 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2895 xoap->xoa_generation = zp->z_gen;
2896 XVA_SET_RTN(xvap, XAT_GEN);
2899 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2901 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2902 XVA_SET_RTN(xvap, XAT_OFFLINE);
2905 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2907 ((zp->z_pflags & ZFS_SPARSE) != 0);
2908 XVA_SET_RTN(xvap, XAT_SPARSE);
2912 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2913 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2914 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2915 ZFS_TIME_DECODE(&vap->va_birthtime, crtime);
2917 mutex_exit(&zp->z_lock);
2919 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2920 vap->va_blksize = blksize;
2921 vap->va_bytes = nblocks << 9; /* nblocks * 512 */
2923 if (zp->z_blksz == 0) {
2925 * Block size hasn't been set; suggest maximal I/O transfers.
2927 vap->va_blksize = zfsvfs->z_max_blksz;
2935 * Set the file attributes to the values contained in the
2938 * IN: vp - vnode of file to be modified.
2939 * vap - new attribute values.
2940 * If AT_XVATTR set, then optional attrs are being set
2941 * flags - ATTR_UTIME set if non-default time values provided.
2942 * - ATTR_NOACLCHECK (CIFS context only).
2943 * cr - credentials of caller.
2944 * ct - caller context
2946 * RETURN: 0 if success
2947 * error code if failure
2950 * vp - ctime updated, mtime updated if size changed.
2954 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2955 caller_context_t *ct)
2957 znode_t *zp = VTOZ(vp);
2958 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2963 uint_t mask = vap->va_mask;
2964 uint_t saved_mask = 0;
2965 uint64_t saved_mode;
2968 uint64_t new_uid, new_gid;
2970 uint64_t mtime[2], ctime[2];
2972 int need_policy = FALSE;
2974 zfs_fuid_info_t *fuidp = NULL;
2975 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2978 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2979 boolean_t fuid_dirtied = B_FALSE;
2980 sa_bulk_attr_t bulk[7], xattr_bulk[7];
2981 int count = 0, xattr_count = 0;
2986 if (mask & AT_NOSET)
2992 zilog = zfsvfs->z_log;
2995 * Make sure that if we have ephemeral uid/gid or xvattr specified
2996 * that file system is at proper version level
2999 if (zfsvfs->z_use_fuids == B_FALSE &&
3000 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
3001 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
3002 (mask & AT_XVATTR))) {
3007 if (mask & AT_SIZE && vp->v_type == VDIR) {
3012 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
3018 * If this is an xvattr_t, then get a pointer to the structure of
3019 * optional attributes. If this is NULL, then we have a vattr_t.
3021 xoap = xva_getxoptattr(xvap);
3023 xva_init(&tmpxvattr);
3026 * Immutable files can only alter immutable bit and atime
3028 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
3029 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
3030 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
3035 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
3041 * Verify timestamps doesn't overflow 32 bits.
3042 * ZFS can handle large timestamps, but 32bit syscalls can't
3043 * handle times greater than 2039. This check should be removed
3044 * once large timestamps are fully supported.
3046 if (mask & (AT_ATIME | AT_MTIME)) {
3047 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
3048 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
3058 /* Can this be moved to before the top label? */
3059 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
3065 * First validate permissions
3068 if (mask & AT_SIZE) {
3070 * XXX - Note, we are not providing any open
3071 * mode flags here (like FNDELAY), so we may
3072 * block if there are locks present... this
3073 * should be addressed in openat().
3075 /* XXX - would it be OK to generate a log record here? */
3076 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
3083 if (mask & (AT_ATIME|AT_MTIME) ||
3084 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
3085 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
3086 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
3087 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
3088 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
3089 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
3090 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
3091 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
3095 if (mask & (AT_UID|AT_GID)) {
3096 int idmask = (mask & (AT_UID|AT_GID));
3101 * NOTE: even if a new mode is being set,
3102 * we may clear S_ISUID/S_ISGID bits.
3105 if (!(mask & AT_MODE))
3106 vap->va_mode = zp->z_mode;
3109 * Take ownership or chgrp to group we are a member of
3112 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
3113 take_group = (mask & AT_GID) &&
3114 zfs_groupmember(zfsvfs, vap->va_gid, cr);
3117 * If both AT_UID and AT_GID are set then take_owner and
3118 * take_group must both be set in order to allow taking
3121 * Otherwise, send the check through secpolicy_vnode_setattr()
3125 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
3126 ((idmask == AT_UID) && take_owner) ||
3127 ((idmask == AT_GID) && take_group)) {
3128 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
3129 skipaclchk, cr) == 0) {
3131 * Remove setuid/setgid for non-privileged users
3133 secpolicy_setid_clear(vap, vp, cr);
3134 trim_mask = (mask & (AT_UID|AT_GID));
3143 mutex_enter(&zp->z_lock);
3144 oldva.va_mode = zp->z_mode;
3145 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
3146 if (mask & AT_XVATTR) {
3148 * Update xvattr mask to include only those attributes
3149 * that are actually changing.
3151 * the bits will be restored prior to actually setting
3152 * the attributes so the caller thinks they were set.
3154 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
3155 if (xoap->xoa_appendonly !=
3156 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
3159 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
3160 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
3164 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
3165 if (xoap->xoa_nounlink !=
3166 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
3169 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
3170 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
3174 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
3175 if (xoap->xoa_immutable !=
3176 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
3179 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
3180 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
3184 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
3185 if (xoap->xoa_nodump !=
3186 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
3189 XVA_CLR_REQ(xvap, XAT_NODUMP);
3190 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
3194 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3195 if (xoap->xoa_av_modified !=
3196 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3199 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3200 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
3204 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3205 if ((vp->v_type != VREG &&
3206 xoap->xoa_av_quarantined) ||
3207 xoap->xoa_av_quarantined !=
3208 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3211 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3212 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
3216 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3217 mutex_exit(&zp->z_lock);
3222 if (need_policy == FALSE &&
3223 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3224 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3229 mutex_exit(&zp->z_lock);
3231 if (mask & AT_MODE) {
3232 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3233 err = secpolicy_setid_setsticky_clear(vp, vap,
3239 trim_mask |= AT_MODE;
3247 * If trim_mask is set then take ownership
3248 * has been granted or write_acl is present and user
3249 * has the ability to modify mode. In that case remove
3250 * UID|GID and or MODE from mask so that
3251 * secpolicy_vnode_setattr() doesn't revoke it.
3255 saved_mask = vap->va_mask;
3256 vap->va_mask &= ~trim_mask;
3257 if (trim_mask & AT_MODE) {
3259 * Save the mode, as secpolicy_vnode_setattr()
3260 * will overwrite it with ova.va_mode.
3262 saved_mode = vap->va_mode;
3265 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3266 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3273 vap->va_mask |= saved_mask;
3274 if (trim_mask & AT_MODE) {
3276 * Recover the mode after
3277 * secpolicy_vnode_setattr().
3279 vap->va_mode = saved_mode;
3285 * secpolicy_vnode_setattr, or take ownership may have
3288 mask = vap->va_mask;
3290 if ((mask & (AT_UID | AT_GID))) {
3291 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3292 &xattr_obj, sizeof (xattr_obj));
3294 if (err == 0 && xattr_obj) {
3295 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3299 if (mask & AT_UID) {
3300 new_uid = zfs_fuid_create(zfsvfs,
3301 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3302 if (new_uid != zp->z_uid &&
3303 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3305 VN_RELE(ZTOV(attrzp));
3311 if (mask & AT_GID) {
3312 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3313 cr, ZFS_GROUP, &fuidp);
3314 if (new_gid != zp->z_gid &&
3315 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3317 VN_RELE(ZTOV(attrzp));
3323 tx = dmu_tx_create(zfsvfs->z_os);
3325 if (mask & AT_MODE) {
3326 uint64_t pmode = zp->z_mode;
3328 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3330 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3331 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3336 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3339 mutex_enter(&zp->z_lock);
3340 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3342 * Are we upgrading ACL from old V0 format
3345 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3346 zfs_znode_acl_version(zp) ==
3347 ZFS_ACL_VERSION_INITIAL) {
3348 dmu_tx_hold_free(tx, acl_obj, 0,
3350 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3351 0, aclp->z_acl_bytes);
3353 dmu_tx_hold_write(tx, acl_obj, 0,
3356 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3357 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3358 0, aclp->z_acl_bytes);
3360 mutex_exit(&zp->z_lock);
3361 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3363 if ((mask & AT_XVATTR) &&
3364 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3365 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3367 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3371 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3374 fuid_dirtied = zfsvfs->z_fuid_dirty;
3376 zfs_fuid_txhold(zfsvfs, tx);
3378 zfs_sa_upgrade_txholds(tx, zp);
3380 err = dmu_tx_assign(tx, TXG_NOWAIT);
3382 if (err == ERESTART)
3389 * Set each attribute requested.
3390 * We group settings according to the locks they need to acquire.
3392 * Note: you cannot set ctime directly, although it will be
3393 * updated as a side-effect of calling this function.
3397 if (mask & (AT_UID|AT_GID|AT_MODE))
3398 mutex_enter(&zp->z_acl_lock);
3399 mutex_enter(&zp->z_lock);
3401 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3402 &zp->z_pflags, sizeof (zp->z_pflags));
3405 if (mask & (AT_UID|AT_GID|AT_MODE))
3406 mutex_enter(&attrzp->z_acl_lock);
3407 mutex_enter(&attrzp->z_lock);
3408 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3409 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3410 sizeof (attrzp->z_pflags));
3413 if (mask & (AT_UID|AT_GID)) {
3415 if (mask & AT_UID) {
3416 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3417 &new_uid, sizeof (new_uid));
3418 zp->z_uid = new_uid;
3420 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3421 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3423 attrzp->z_uid = new_uid;
3427 if (mask & AT_GID) {
3428 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3429 NULL, &new_gid, sizeof (new_gid));
3430 zp->z_gid = new_gid;
3432 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3433 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3435 attrzp->z_gid = new_gid;
3438 if (!(mask & AT_MODE)) {
3439 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3440 NULL, &new_mode, sizeof (new_mode));
3441 new_mode = zp->z_mode;
3443 err = zfs_acl_chown_setattr(zp);
3446 err = zfs_acl_chown_setattr(attrzp);
3451 if (mask & AT_MODE) {
3452 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3453 &new_mode, sizeof (new_mode));
3454 zp->z_mode = new_mode;
3455 ASSERT3U((uintptr_t)aclp, !=, 0);
3456 err = zfs_aclset_common(zp, aclp, cr, tx);
3458 if (zp->z_acl_cached)
3459 zfs_acl_free(zp->z_acl_cached);
3460 zp->z_acl_cached = aclp;
3465 if (mask & AT_ATIME) {
3466 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3467 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3468 &zp->z_atime, sizeof (zp->z_atime));
3471 if (mask & AT_MTIME) {
3472 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3473 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3474 mtime, sizeof (mtime));
3477 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3478 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3479 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3480 NULL, mtime, sizeof (mtime));
3481 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3482 &ctime, sizeof (ctime));
3483 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3485 } else if (mask != 0) {
3486 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3487 &ctime, sizeof (ctime));
3488 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3491 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3492 SA_ZPL_CTIME(zfsvfs), NULL,
3493 &ctime, sizeof (ctime));
3494 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3495 mtime, ctime, B_TRUE);
3499 * Do this after setting timestamps to prevent timestamp
3500 * update from toggling bit
3503 if (xoap && (mask & AT_XVATTR)) {
3506 * restore trimmed off masks
3507 * so that return masks can be set for caller.
3510 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3511 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3513 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3514 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3516 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3517 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3519 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3520 XVA_SET_REQ(xvap, XAT_NODUMP);
3522 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3523 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3525 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3526 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3529 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3530 ASSERT(vp->v_type == VREG);
3532 zfs_xvattr_set(zp, xvap, tx);
3536 zfs_fuid_sync(zfsvfs, tx);
3539 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3541 mutex_exit(&zp->z_lock);
3542 if (mask & (AT_UID|AT_GID|AT_MODE))
3543 mutex_exit(&zp->z_acl_lock);
3546 if (mask & (AT_UID|AT_GID|AT_MODE))
3547 mutex_exit(&attrzp->z_acl_lock);
3548 mutex_exit(&attrzp->z_lock);
3551 if (err == 0 && attrzp) {
3552 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3558 VN_RELE(ZTOV(attrzp));
3563 zfs_fuid_info_free(fuidp);
3569 if (err == ERESTART)
3572 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3577 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3578 zil_commit(zilog, 0);
3584 typedef struct zfs_zlock {
3585 krwlock_t *zl_rwlock; /* lock we acquired */
3586 znode_t *zl_znode; /* znode we held */
3587 struct zfs_zlock *zl_next; /* next in list */
3591 * Drop locks and release vnodes that were held by zfs_rename_lock().
3594 zfs_rename_unlock(zfs_zlock_t **zlpp)
3598 while ((zl = *zlpp) != NULL) {
3599 if (zl->zl_znode != NULL)
3600 VN_RELE(ZTOV(zl->zl_znode));
3601 rw_exit(zl->zl_rwlock);
3602 *zlpp = zl->zl_next;
3603 kmem_free(zl, sizeof (*zl));
3608 * Search back through the directory tree, using the ".." entries.
3609 * Lock each directory in the chain to prevent concurrent renames.
3610 * Fail any attempt to move a directory into one of its own descendants.
3611 * XXX - z_parent_lock can overlap with map or grow locks
3614 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3618 uint64_t rootid = zp->z_zfsvfs->z_root;
3619 uint64_t oidp = zp->z_id;
3620 krwlock_t *rwlp = &szp->z_parent_lock;
3621 krw_t rw = RW_WRITER;
3624 * First pass write-locks szp and compares to zp->z_id.
3625 * Later passes read-lock zp and compare to zp->z_parent.
3628 if (!rw_tryenter(rwlp, rw)) {
3630 * Another thread is renaming in this path.
3631 * Note that if we are a WRITER, we don't have any
3632 * parent_locks held yet.
3634 if (rw == RW_READER && zp->z_id > szp->z_id) {
3636 * Drop our locks and restart
3638 zfs_rename_unlock(&zl);
3642 rwlp = &szp->z_parent_lock;
3647 * Wait for other thread to drop its locks
3653 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3654 zl->zl_rwlock = rwlp;
3655 zl->zl_znode = NULL;
3656 zl->zl_next = *zlpp;
3659 if (oidp == szp->z_id) /* We're a descendant of szp */
3662 if (oidp == rootid) /* We've hit the top */
3665 if (rw == RW_READER) { /* i.e. not the first pass */
3666 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3671 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3672 &oidp, sizeof (oidp));
3673 rwlp = &zp->z_parent_lock;
3676 } while (zp->z_id != sdzp->z_id);
3682 * Move an entry from the provided source directory to the target
3683 * directory. Change the entry name as indicated.
3685 * IN: sdvp - Source directory containing the "old entry".
3686 * snm - Old entry name.
3687 * tdvp - Target directory to contain the "new entry".
3688 * tnm - New entry name.
3689 * cr - credentials of caller.
3690 * ct - caller context
3691 * flags - case flags
3693 * RETURN: 0 if success
3694 * error code if failure
3697 * sdvp,tdvp - ctime|mtime updated
3701 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3702 caller_context_t *ct, int flags)
3704 znode_t *tdzp, *szp, *tzp;
3705 znode_t *sdzp = VTOZ(sdvp);
3706 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs;
3709 zfs_dirlock_t *sdl, *tdl;
3712 int cmp, serr, terr;
3717 ZFS_VERIFY_ZP(sdzp);
3718 zilog = zfsvfs->z_log;
3721 * Make sure we have the real vp for the target directory.
3723 if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3726 if (tdvp->v_vfsp != sdvp->v_vfsp || zfsctl_is_node(tdvp)) {
3732 ZFS_VERIFY_ZP(tdzp);
3733 if (zfsvfs->z_utf8 && u8_validate(tnm,
3734 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3739 if (flags & FIGNORECASE)
3748 * This is to prevent the creation of links into attribute space
3749 * by renaming a linked file into/outof an attribute directory.
3750 * See the comment in zfs_link() for why this is considered bad.
3752 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3758 * Lock source and target directory entries. To prevent deadlock,
3759 * a lock ordering must be defined. We lock the directory with
3760 * the smallest object id first, or if it's a tie, the one with
3761 * the lexically first name.
3763 if (sdzp->z_id < tdzp->z_id) {
3765 } else if (sdzp->z_id > tdzp->z_id) {
3769 * First compare the two name arguments without
3770 * considering any case folding.
3772 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3774 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3775 ASSERT(error == 0 || !zfsvfs->z_utf8);
3778 * POSIX: "If the old argument and the new argument
3779 * both refer to links to the same existing file,
3780 * the rename() function shall return successfully
3781 * and perform no other action."
3787 * If the file system is case-folding, then we may
3788 * have some more checking to do. A case-folding file
3789 * system is either supporting mixed case sensitivity
3790 * access or is completely case-insensitive. Note
3791 * that the file system is always case preserving.
3793 * In mixed sensitivity mode case sensitive behavior
3794 * is the default. FIGNORECASE must be used to
3795 * explicitly request case insensitive behavior.
3797 * If the source and target names provided differ only
3798 * by case (e.g., a request to rename 'tim' to 'Tim'),
3799 * we will treat this as a special case in the
3800 * case-insensitive mode: as long as the source name
3801 * is an exact match, we will allow this to proceed as
3802 * a name-change request.
3804 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3805 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3806 flags & FIGNORECASE)) &&
3807 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3810 * case preserving rename request, require exact
3819 * If the source and destination directories are the same, we should
3820 * grab the z_name_lock of that directory only once.
3824 rw_enter(&sdzp->z_name_lock, RW_READER);
3828 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3829 ZEXISTS | zflg, NULL, NULL);
3830 terr = zfs_dirent_lock(&tdl,
3831 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3833 terr = zfs_dirent_lock(&tdl,
3834 tdzp, tnm, &tzp, zflg, NULL, NULL);
3835 serr = zfs_dirent_lock(&sdl,
3836 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3842 * Source entry invalid or not there.
3845 zfs_dirent_unlock(tdl);
3851 rw_exit(&sdzp->z_name_lock);
3854 * FreeBSD: In OpenSolaris they only check if rename source is
3855 * ".." here, because "." is handled in their lookup. This is
3856 * not the case for FreeBSD, so we check for "." explicitly.
3858 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0)
3864 zfs_dirent_unlock(sdl);
3868 rw_exit(&sdzp->z_name_lock);
3870 if (strcmp(tnm, "..") == 0)
3877 * Must have write access at the source to remove the old entry
3878 * and write access at the target to create the new entry.
3879 * Note that if target and source are the same, this can be
3880 * done in a single check.
3883 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3886 if (ZTOV(szp)->v_type == VDIR) {
3888 * Check to make sure rename is valid.
3889 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3891 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3896 * Does target exist?
3900 * Source and target must be the same type.
3902 if (ZTOV(szp)->v_type == VDIR) {
3903 if (ZTOV(tzp)->v_type != VDIR) {
3908 if (ZTOV(tzp)->v_type == VDIR) {
3914 * POSIX dictates that when the source and target
3915 * entries refer to the same file object, rename
3916 * must do nothing and exit without error.
3918 if (szp->z_id == tzp->z_id) {
3924 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3926 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3929 * notify the target directory if it is not the same
3930 * as source directory.
3933 vnevent_rename_dest_dir(tdvp, ct);
3936 tx = dmu_tx_create(zfsvfs->z_os);
3937 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3938 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3939 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3940 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3942 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3943 zfs_sa_upgrade_txholds(tx, tdzp);
3946 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3947 zfs_sa_upgrade_txholds(tx, tzp);
3950 zfs_sa_upgrade_txholds(tx, szp);
3951 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3952 error = dmu_tx_assign(tx, TXG_NOWAIT);
3955 zfs_rename_unlock(&zl);
3956 zfs_dirent_unlock(sdl);
3957 zfs_dirent_unlock(tdl);
3960 rw_exit(&sdzp->z_name_lock);
3965 if (error == ERESTART) {
3975 if (tzp) /* Attempt to remove the existing target */
3976 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3979 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3981 szp->z_pflags |= ZFS_AV_MODIFIED;
3983 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3984 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3987 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3989 zfs_log_rename(zilog, tx, TX_RENAME |
3990 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3991 sdl->dl_name, tdzp, tdl->dl_name, szp);
3994 * Update path information for the target vnode
3996 vn_renamepath(tdvp, ZTOV(szp), tnm,
4000 * At this point, we have successfully created
4001 * the target name, but have failed to remove
4002 * the source name. Since the create was done
4003 * with the ZRENAMING flag, there are
4004 * complications; for one, the link count is
4005 * wrong. The easiest way to deal with this
4006 * is to remove the newly created target, and
4007 * return the original error. This must
4008 * succeed; fortunately, it is very unlikely to
4009 * fail, since we just created it.
4011 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
4012 ZRENAMING, NULL), ==, 0);
4015 #ifdef FREEBSD_NAMECACHE
4019 cache_purge(ZTOV(szp));
4021 cache_purge(ZTOV(tzp));
4029 zfs_rename_unlock(&zl);
4031 zfs_dirent_unlock(sdl);
4032 zfs_dirent_unlock(tdl);
4035 rw_exit(&sdzp->z_name_lock);
4042 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4043 zil_commit(zilog, 0);
4051 * Insert the indicated symbolic reference entry into the directory.
4053 * IN: dvp - Directory to contain new symbolic link.
4054 * link - Name for new symlink entry.
4055 * vap - Attributes of new entry.
4056 * target - Target path of new symlink.
4057 * cr - credentials of caller.
4058 * ct - caller context
4059 * flags - case flags
4061 * RETURN: 0 if success
4062 * error code if failure
4065 * dvp - ctime|mtime updated
4069 zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link,
4070 cred_t *cr, kthread_t *td)
4072 znode_t *zp, *dzp = VTOZ(dvp);
4075 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4077 uint64_t len = strlen(link);
4080 zfs_acl_ids_t acl_ids;
4081 boolean_t fuid_dirtied;
4082 uint64_t txtype = TX_SYMLINK;
4085 ASSERT(vap->va_type == VLNK);
4089 zilog = zfsvfs->z_log;
4091 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
4092 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4096 if (flags & FIGNORECASE)
4099 if (len > MAXPATHLEN) {
4101 return (ENAMETOOLONG);
4104 if ((error = zfs_acl_ids_create(dzp, 0,
4105 vap, cr, NULL, &acl_ids)) != 0) {
4111 * Attempt to lock directory; fail if entry already exists.
4113 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
4115 zfs_acl_ids_free(&acl_ids);
4120 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4121 zfs_acl_ids_free(&acl_ids);
4122 zfs_dirent_unlock(dl);
4127 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
4128 zfs_acl_ids_free(&acl_ids);
4129 zfs_dirent_unlock(dl);
4133 tx = dmu_tx_create(zfsvfs->z_os);
4134 fuid_dirtied = zfsvfs->z_fuid_dirty;
4135 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
4136 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4137 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
4138 ZFS_SA_BASE_ATTR_SIZE + len);
4139 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
4140 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
4141 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
4142 acl_ids.z_aclp->z_acl_bytes);
4145 zfs_fuid_txhold(zfsvfs, tx);
4146 error = dmu_tx_assign(tx, TXG_NOWAIT);
4148 zfs_dirent_unlock(dl);
4149 if (error == ERESTART) {
4154 zfs_acl_ids_free(&acl_ids);
4161 * Create a new object for the symlink.
4162 * for version 4 ZPL datsets the symlink will be an SA attribute
4164 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
4167 zfs_fuid_sync(zfsvfs, tx);
4169 mutex_enter(&zp->z_lock);
4171 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
4174 zfs_sa_symlink(zp, link, len, tx);
4175 mutex_exit(&zp->z_lock);
4178 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
4179 &zp->z_size, sizeof (zp->z_size), tx);
4181 * Insert the new object into the directory.
4183 (void) zfs_link_create(dl, zp, tx, ZNEW);
4185 if (flags & FIGNORECASE)
4187 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
4190 zfs_acl_ids_free(&acl_ids);
4194 zfs_dirent_unlock(dl);
4196 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4197 zil_commit(zilog, 0);
4204 * Return, in the buffer contained in the provided uio structure,
4205 * the symbolic path referred to by vp.
4207 * IN: vp - vnode of symbolic link.
4208 * uoip - structure to contain the link path.
4209 * cr - credentials of caller.
4210 * ct - caller context
4212 * OUT: uio - structure to contain the link path.
4214 * RETURN: 0 if success
4215 * error code if failure
4218 * vp - atime updated
4222 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
4224 znode_t *zp = VTOZ(vp);
4225 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4231 mutex_enter(&zp->z_lock);
4233 error = sa_lookup_uio(zp->z_sa_hdl,
4234 SA_ZPL_SYMLINK(zfsvfs), uio);
4236 error = zfs_sa_readlink(zp, uio);
4237 mutex_exit(&zp->z_lock);
4239 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4246 * Insert a new entry into directory tdvp referencing svp.
4248 * IN: tdvp - Directory to contain new entry.
4249 * svp - vnode of new entry.
4250 * name - name of new entry.
4251 * cr - credentials of caller.
4252 * ct - caller context
4254 * RETURN: 0 if success
4255 * error code if failure
4258 * tdvp - ctime|mtime updated
4259 * svp - ctime updated
4263 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
4264 caller_context_t *ct, int flags)
4266 znode_t *dzp = VTOZ(tdvp);
4268 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4278 ASSERT(tdvp->v_type == VDIR);
4282 zilog = zfsvfs->z_log;
4284 if (VOP_REALVP(svp, &realvp, ct) == 0)
4288 * POSIX dictates that we return EPERM here.
4289 * Better choices include ENOTSUP or EISDIR.
4291 if (svp->v_type == VDIR) {
4296 if (svp->v_vfsp != tdvp->v_vfsp || zfsctl_is_node(svp)) {
4304 /* Prevent links to .zfs/shares files */
4306 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4307 &parent, sizeof (uint64_t))) != 0) {
4311 if (parent == zfsvfs->z_shares_dir) {
4316 if (zfsvfs->z_utf8 && u8_validate(name,
4317 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4321 if (flags & FIGNORECASE)
4325 * We do not support links between attributes and non-attributes
4326 * because of the potential security risk of creating links
4327 * into "normal" file space in order to circumvent restrictions
4328 * imposed in attribute space.
4330 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4336 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4337 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) {
4342 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4349 * Attempt to lock directory; fail if entry already exists.
4351 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4357 tx = dmu_tx_create(zfsvfs->z_os);
4358 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4359 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4360 zfs_sa_upgrade_txholds(tx, szp);
4361 zfs_sa_upgrade_txholds(tx, dzp);
4362 error = dmu_tx_assign(tx, TXG_NOWAIT);
4364 zfs_dirent_unlock(dl);
4365 if (error == ERESTART) {
4375 error = zfs_link_create(dl, szp, tx, 0);
4378 uint64_t txtype = TX_LINK;
4379 if (flags & FIGNORECASE)
4381 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4386 zfs_dirent_unlock(dl);
4389 vnevent_link(svp, ct);
4392 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4393 zil_commit(zilog, 0);
4401 * zfs_null_putapage() is used when the file system has been force
4402 * unmounted. It just drops the pages.
4406 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4407 size_t *lenp, int flags, cred_t *cr)
4409 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4414 * Push a page out to disk, klustering if possible.
4416 * IN: vp - file to push page to.
4417 * pp - page to push.
4418 * flags - additional flags.
4419 * cr - credentials of caller.
4421 * OUT: offp - start of range pushed.
4422 * lenp - len of range pushed.
4424 * RETURN: 0 if success
4425 * error code if failure
4427 * NOTE: callers must have locked the page to be pushed. On
4428 * exit, the page (and all other pages in the kluster) must be
4433 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4434 size_t *lenp, int flags, cred_t *cr)
4436 znode_t *zp = VTOZ(vp);
4437 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4439 u_offset_t off, koff;
4446 * If our blocksize is bigger than the page size, try to kluster
4447 * multiple pages so that we write a full block (thus avoiding
4448 * a read-modify-write).
4450 if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4451 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4452 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4453 ASSERT(koff <= zp->z_size);
4454 if (koff + klen > zp->z_size)
4455 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4456 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4458 ASSERT3U(btop(len), ==, btopr(len));
4461 * Can't push pages past end-of-file.
4463 if (off >= zp->z_size) {
4464 /* ignore all pages */
4467 } else if (off + len > zp->z_size) {
4468 int npages = btopr(zp->z_size - off);
4471 page_list_break(&pp, &trunc, npages);
4472 /* ignore pages past end of file */
4474 pvn_write_done(trunc, flags);
4475 len = zp->z_size - off;
4478 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4479 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4484 tx = dmu_tx_create(zfsvfs->z_os);
4485 dmu_tx_hold_write(tx, zp->z_id, off, len);
4487 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4488 zfs_sa_upgrade_txholds(tx, zp);
4489 err = dmu_tx_assign(tx, TXG_NOWAIT);
4491 if (err == ERESTART) {
4500 if (zp->z_blksz <= PAGESIZE) {
4501 caddr_t va = zfs_map_page(pp, S_READ);
4502 ASSERT3U(len, <=, PAGESIZE);
4503 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4504 zfs_unmap_page(pp, va);
4506 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4510 uint64_t mtime[2], ctime[2];
4511 sa_bulk_attr_t bulk[3];
4514 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4516 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4518 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4520 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4522 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4527 pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4537 * Copy the portion of the file indicated from pages into the file.
4538 * The pages are stored in a page list attached to the files vnode.
4540 * IN: vp - vnode of file to push page data to.
4541 * off - position in file to put data.
4542 * len - amount of data to write.
4543 * flags - flags to control the operation.
4544 * cr - credentials of caller.
4545 * ct - caller context.
4547 * RETURN: 0 if success
4548 * error code if failure
4551 * vp - ctime|mtime updated
4555 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4556 caller_context_t *ct)
4558 znode_t *zp = VTOZ(vp);
4559 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4571 * Align this request to the file block size in case we kluster.
4572 * XXX - this can result in pretty aggresive locking, which can
4573 * impact simultanious read/write access. One option might be
4574 * to break up long requests (len == 0) into block-by-block
4575 * operations to get narrower locking.
4577 blksz = zp->z_blksz;
4579 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4582 if (len > 0 && ISP2(blksz))
4583 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4589 * Search the entire vp list for pages >= io_off.
4591 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4592 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4595 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4597 if (off > zp->z_size) {
4598 /* past end of file */
4599 zfs_range_unlock(rl);
4604 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4606 for (off = io_off; io_off < off + len; io_off += io_len) {
4607 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4608 pp = page_lookup(vp, io_off,
4609 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4611 pp = page_lookup_nowait(vp, io_off,
4612 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4615 if (pp != NULL && pvn_getdirty(pp, flags)) {
4619 * Found a dirty page to push
4621 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4629 zfs_range_unlock(rl);
4630 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4631 zil_commit(zfsvfs->z_log, zp->z_id);
4639 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4641 znode_t *zp = VTOZ(vp);
4642 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4645 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4646 if (zp->z_sa_hdl == NULL) {
4648 * The fs has been unmounted, or we did a
4649 * suspend/resume and this file no longer exists.
4651 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4656 mutex_enter(&zp->z_lock);
4657 if (zp->z_unlinked) {
4659 * Fast path to recycle a vnode of a removed file.
4661 mutex_exit(&zp->z_lock);
4662 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4666 mutex_exit(&zp->z_lock);
4668 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4669 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4671 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4672 zfs_sa_upgrade_txholds(tx, zp);
4673 error = dmu_tx_assign(tx, TXG_WAIT);
4677 mutex_enter(&zp->z_lock);
4678 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4679 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4680 zp->z_atime_dirty = 0;
4681 mutex_exit(&zp->z_lock);
4685 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4690 * Bounds-check the seek operation.
4692 * IN: vp - vnode seeking within
4693 * ooff - old file offset
4694 * noffp - pointer to new file offset
4695 * ct - caller context
4697 * RETURN: 0 if success
4698 * EINVAL if new offset invalid
4702 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4703 caller_context_t *ct)
4705 if (vp->v_type == VDIR)
4707 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4711 * Pre-filter the generic locking function to trap attempts to place
4712 * a mandatory lock on a memory mapped file.
4715 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4716 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4718 znode_t *zp = VTOZ(vp);
4719 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4725 * We are following the UFS semantics with respect to mapcnt
4726 * here: If we see that the file is mapped already, then we will
4727 * return an error, but we don't worry about races between this
4728 * function and zfs_map().
4730 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4735 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4739 * If we can't find a page in the cache, we will create a new page
4740 * and fill it with file data. For efficiency, we may try to fill
4741 * multiple pages at once (klustering) to fill up the supplied page
4742 * list. Note that the pages to be filled are held with an exclusive
4743 * lock to prevent access by other threads while they are being filled.
4746 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4747 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4749 znode_t *zp = VTOZ(vp);
4750 page_t *pp, *cur_pp;
4751 objset_t *os = zp->z_zfsvfs->z_os;
4752 u_offset_t io_off, total;
4756 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4758 * We only have a single page, don't bother klustering
4762 pp = page_create_va(vp, io_off, io_len,
4763 PG_EXCL | PG_WAIT, seg, addr);
4766 * Try to find enough pages to fill the page list
4768 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4769 &io_len, off, plsz, 0);
4773 * The page already exists, nothing to do here.
4780 * Fill the pages in the kluster.
4783 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4786 ASSERT3U(io_off, ==, cur_pp->p_offset);
4787 va = zfs_map_page(cur_pp, S_WRITE);
4788 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4790 zfs_unmap_page(cur_pp, va);
4792 /* On error, toss the entire kluster */
4793 pvn_read_done(pp, B_ERROR);
4794 /* convert checksum errors into IO errors */
4799 cur_pp = cur_pp->p_next;
4803 * Fill in the page list array from the kluster starting
4804 * from the desired offset `off'.
4805 * NOTE: the page list will always be null terminated.
4807 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4808 ASSERT(pl == NULL || (*pl)->p_offset == off);
4814 * Return pointers to the pages for the file region [off, off + len]
4815 * in the pl array. If plsz is greater than len, this function may
4816 * also return page pointers from after the specified region
4817 * (i.e. the region [off, off + plsz]). These additional pages are
4818 * only returned if they are already in the cache, or were created as
4819 * part of a klustered read.
4821 * IN: vp - vnode of file to get data from.
4822 * off - position in file to get data from.
4823 * len - amount of data to retrieve.
4824 * plsz - length of provided page list.
4825 * seg - segment to obtain pages for.
4826 * addr - virtual address of fault.
4827 * rw - mode of created pages.
4828 * cr - credentials of caller.
4829 * ct - caller context.
4831 * OUT: protp - protection mode of created pages.
4832 * pl - list of pages created.
4834 * RETURN: 0 if success
4835 * error code if failure
4838 * vp - atime updated
4842 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4843 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4844 enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4846 znode_t *zp = VTOZ(vp);
4847 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4851 /* we do our own caching, faultahead is unnecessary */
4854 else if (len > plsz)
4857 len = P2ROUNDUP(len, PAGESIZE);
4858 ASSERT(plsz >= len);
4867 * Loop through the requested range [off, off + len) looking
4868 * for pages. If we don't find a page, we will need to create
4869 * a new page and fill it with data from the file.
4872 if (*pl = page_lookup(vp, off, SE_SHARED))
4874 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4877 ASSERT3U((*pl)->p_offset, ==, off);
4881 ASSERT3U(len, >=, PAGESIZE);
4884 ASSERT3U(plsz, >=, PAGESIZE);
4891 * Fill out the page array with any pages already in the cache.
4894 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4901 * Release any pages we have previously locked.
4906 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4916 * Request a memory map for a section of a file. This code interacts
4917 * with common code and the VM system as follows:
4919 * common code calls mmap(), which ends up in smmap_common()
4921 * this calls VOP_MAP(), which takes you into (say) zfs
4923 * zfs_map() calls as_map(), passing segvn_create() as the callback
4925 * segvn_create() creates the new segment and calls VOP_ADDMAP()
4927 * zfs_addmap() updates z_mapcnt
4931 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4932 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4933 caller_context_t *ct)
4935 znode_t *zp = VTOZ(vp);
4936 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4937 segvn_crargs_t vn_a;
4943 if ((prot & PROT_WRITE) && (zp->z_pflags &
4944 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4949 if ((prot & (PROT_READ | PROT_EXEC)) &&
4950 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4955 if (vp->v_flag & VNOMAP) {
4960 if (off < 0 || len > MAXOFFSET_T - off) {
4965 if (vp->v_type != VREG) {
4971 * If file is locked, disallow mapping.
4973 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
4979 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4987 vn_a.offset = (u_offset_t)off;
4988 vn_a.type = flags & MAP_TYPE;
4990 vn_a.maxprot = maxprot;
4993 vn_a.flags = flags & ~MAP_TYPE;
4995 vn_a.lgrp_mem_policy_flags = 0;
4997 error = as_map(as, *addrp, len, segvn_create, &vn_a);
5006 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5007 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
5008 caller_context_t *ct)
5010 uint64_t pages = btopr(len);
5012 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
5017 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
5018 * more accurate mtime for the associated file. Since we don't have a way of
5019 * detecting when the data was actually modified, we have to resort to
5020 * heuristics. If an explicit msync() is done, then we mark the mtime when the
5021 * last page is pushed. The problem occurs when the msync() call is omitted,
5022 * which by far the most common case:
5030 * putpage() via fsflush
5032 * If we wait until fsflush to come along, we can have a modification time that
5033 * is some arbitrary point in the future. In order to prevent this in the
5034 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
5039 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5040 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
5041 caller_context_t *ct)
5043 uint64_t pages = btopr(len);
5045 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
5046 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
5048 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
5049 vn_has_cached_data(vp))
5050 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
5056 * Free or allocate space in a file. Currently, this function only
5057 * supports the `F_FREESP' command. However, this command is somewhat
5058 * misnamed, as its functionality includes the ability to allocate as
5059 * well as free space.
5061 * IN: vp - vnode of file to free data in.
5062 * cmd - action to take (only F_FREESP supported).
5063 * bfp - section of file to free/alloc.
5064 * flag - current file open mode flags.
5065 * offset - current file offset.
5066 * cr - credentials of caller [UNUSED].
5067 * ct - caller context.
5069 * RETURN: 0 if success
5070 * error code if failure
5073 * vp - ctime|mtime updated
5077 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
5078 offset_t offset, cred_t *cr, caller_context_t *ct)
5080 znode_t *zp = VTOZ(vp);
5081 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5088 if (cmd != F_FREESP) {
5093 if (error = convoff(vp, bfp, 0, offset)) {
5098 if (bfp->l_len < 0) {
5104 len = bfp->l_len; /* 0 means from off to end of file */
5106 error = zfs_freesp(zp, off, len, flag, TRUE);
5113 CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid));
5114 CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid));
5118 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
5120 znode_t *zp = VTOZ(vp);
5121 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5124 uint64_t object = zp->z_id;
5131 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
5132 &gen64, sizeof (uint64_t))) != 0) {
5137 gen = (uint32_t)gen64;
5139 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
5140 fidp->fid_len = size;
5142 zfid = (zfid_short_t *)fidp;
5144 zfid->zf_len = size;
5146 for (i = 0; i < sizeof (zfid->zf_object); i++)
5147 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
5149 /* Must have a non-zero generation number to distinguish from .zfs */
5152 for (i = 0; i < sizeof (zfid->zf_gen); i++)
5153 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
5155 if (size == LONG_FID_LEN) {
5156 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
5159 zlfid = (zfid_long_t *)fidp;
5161 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
5162 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
5164 /* XXX - this should be the generation number for the objset */
5165 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
5166 zlfid->zf_setgen[i] = 0;
5174 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
5175 caller_context_t *ct)
5187 case _PC_FILESIZEBITS:
5191 case _PC_XATTR_EXISTS:
5193 zfsvfs = zp->z_zfsvfs;
5197 error = zfs_dirent_lock(&dl, zp, "", &xzp,
5198 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
5200 zfs_dirent_unlock(dl);
5201 if (!zfs_dirempty(xzp))
5204 } else if (error == ENOENT) {
5206 * If there aren't extended attributes, it's the
5207 * same as having zero of them.
5214 case _PC_SATTR_ENABLED:
5215 case _PC_SATTR_EXISTS:
5216 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
5217 (vp->v_type == VREG || vp->v_type == VDIR);
5220 case _PC_ACCESS_FILTERING:
5221 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
5225 case _PC_ACL_ENABLED:
5226 *valp = _ACL_ACE_ENABLED;
5229 case _PC_MIN_HOLE_SIZE:
5230 *valp = (int)SPA_MINBLOCKSIZE;
5233 case _PC_TIMESTAMP_RESOLUTION:
5234 /* nanosecond timestamp resolution */
5238 case _PC_ACL_EXTENDED:
5246 case _PC_ACL_PATH_MAX:
5247 *valp = ACL_MAX_ENTRIES;
5251 return (EOPNOTSUPP);
5257 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5258 caller_context_t *ct)
5260 znode_t *zp = VTOZ(vp);
5261 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5263 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5267 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5275 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5276 caller_context_t *ct)
5278 znode_t *zp = VTOZ(vp);
5279 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5281 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5282 zilog_t *zilog = zfsvfs->z_log;
5287 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5289 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5290 zil_commit(zilog, 0);
5298 * Tunable, both must be a power of 2.
5300 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
5301 * zcr_blksz_max: if set to less than the file block size, allow loaning out of
5302 * an arcbuf for a partial block read
5304 int zcr_blksz_min = (1 << 10); /* 1K */
5305 int zcr_blksz_max = (1 << 17); /* 128K */
5309 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5310 caller_context_t *ct)
5312 znode_t *zp = VTOZ(vp);
5313 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5314 int max_blksz = zfsvfs->z_max_blksz;
5315 uio_t *uio = &xuio->xu_uio;
5316 ssize_t size = uio->uio_resid;
5317 offset_t offset = uio->uio_loffset;
5322 int preamble, postamble;
5324 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5332 * Loan out an arc_buf for write if write size is bigger than
5333 * max_blksz, and the file's block size is also max_blksz.
5336 if (size < blksz || zp->z_blksz != blksz) {
5341 * Caller requests buffers for write before knowing where the
5342 * write offset might be (e.g. NFS TCP write).
5347 preamble = P2PHASE(offset, blksz);
5349 preamble = blksz - preamble;
5354 postamble = P2PHASE(size, blksz);
5357 fullblk = size / blksz;
5358 (void) dmu_xuio_init(xuio,
5359 (preamble != 0) + fullblk + (postamble != 0));
5360 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5361 int, postamble, int,
5362 (preamble != 0) + fullblk + (postamble != 0));
5365 * Have to fix iov base/len for partial buffers. They
5366 * currently represent full arc_buf's.
5369 /* data begins in the middle of the arc_buf */
5370 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5373 (void) dmu_xuio_add(xuio, abuf,
5374 blksz - preamble, preamble);
5377 for (i = 0; i < fullblk; i++) {
5378 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5381 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5385 /* data ends in the middle of the arc_buf */
5386 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5389 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5394 * Loan out an arc_buf for read if the read size is larger than
5395 * the current file block size. Block alignment is not
5396 * considered. Partial arc_buf will be loaned out for read.
5398 blksz = zp->z_blksz;
5399 if (blksz < zcr_blksz_min)
5400 blksz = zcr_blksz_min;
5401 if (blksz > zcr_blksz_max)
5402 blksz = zcr_blksz_max;
5403 /* avoid potential complexity of dealing with it */
5404 if (blksz > max_blksz) {
5409 maxsize = zp->z_size - uio->uio_loffset;
5413 if (size < blksz || vn_has_cached_data(vp)) {
5423 uio->uio_extflg = UIO_XUIO;
5424 XUIO_XUZC_RW(xuio) = ioflag;
5431 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5435 int ioflag = XUIO_XUZC_RW(xuio);
5437 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5439 i = dmu_xuio_cnt(xuio);
5441 abuf = dmu_xuio_arcbuf(xuio, i);
5443 * if abuf == NULL, it must be a write buffer
5444 * that has been returned in zfs_write().
5447 dmu_return_arcbuf(abuf);
5448 ASSERT(abuf || ioflag == UIO_WRITE);
5451 dmu_xuio_fini(xuio);
5456 * Predeclare these here so that the compiler assumes that
5457 * this is an "old style" function declaration that does
5458 * not include arguments => we won't get type mismatch errors
5459 * in the initializations that follow.
5461 static int zfs_inval();
5462 static int zfs_isdir();
5476 * Directory vnode operations template
5478 vnodeops_t *zfs_dvnodeops;
5479 const fs_operation_def_t zfs_dvnodeops_template[] = {
5480 VOPNAME_OPEN, { .vop_open = zfs_open },
5481 VOPNAME_CLOSE, { .vop_close = zfs_close },
5482 VOPNAME_READ, { .error = zfs_isdir },
5483 VOPNAME_WRITE, { .error = zfs_isdir },
5484 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5485 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5486 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5487 VOPNAME_ACCESS, { .vop_access = zfs_access },
5488 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5489 VOPNAME_CREATE, { .vop_create = zfs_create },
5490 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5491 VOPNAME_LINK, { .vop_link = zfs_link },
5492 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5493 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir },
5494 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5495 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5496 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink },
5497 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5498 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5499 VOPNAME_FID, { .vop_fid = zfs_fid },
5500 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5501 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5502 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5503 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5504 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5509 * Regular file vnode operations template
5511 vnodeops_t *zfs_fvnodeops;
5512 const fs_operation_def_t zfs_fvnodeops_template[] = {
5513 VOPNAME_OPEN, { .vop_open = zfs_open },
5514 VOPNAME_CLOSE, { .vop_close = zfs_close },
5515 VOPNAME_READ, { .vop_read = zfs_read },
5516 VOPNAME_WRITE, { .vop_write = zfs_write },
5517 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5518 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5519 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5520 VOPNAME_ACCESS, { .vop_access = zfs_access },
5521 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5522 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5523 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5524 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5525 VOPNAME_FID, { .vop_fid = zfs_fid },
5526 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5527 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock },
5528 VOPNAME_SPACE, { .vop_space = zfs_space },
5529 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage },
5530 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage },
5531 VOPNAME_MAP, { .vop_map = zfs_map },
5532 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap },
5533 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap },
5534 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5535 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5536 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5537 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5538 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf },
5539 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf },
5544 * Symbolic link vnode operations template
5546 vnodeops_t *zfs_symvnodeops;
5547 const fs_operation_def_t zfs_symvnodeops_template[] = {
5548 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5549 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5550 VOPNAME_ACCESS, { .vop_access = zfs_access },
5551 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5552 VOPNAME_READLINK, { .vop_readlink = zfs_readlink },
5553 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5554 VOPNAME_FID, { .vop_fid = zfs_fid },
5555 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5556 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5561 * special share hidden files vnode operations template
5563 vnodeops_t *zfs_sharevnodeops;
5564 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5565 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5566 VOPNAME_ACCESS, { .vop_access = zfs_access },
5567 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5568 VOPNAME_FID, { .vop_fid = zfs_fid },
5569 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5570 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5571 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5572 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5577 * Extended attribute directory vnode operations template
5578 * This template is identical to the directory vnodes
5579 * operation template except for restricted operations:
5582 * Note that there are other restrictions embedded in:
5583 * zfs_create() - restrict type to VREG
5584 * zfs_link() - no links into/out of attribute space
5585 * zfs_rename() - no moves into/out of attribute space
5587 vnodeops_t *zfs_xdvnodeops;
5588 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5589 VOPNAME_OPEN, { .vop_open = zfs_open },
5590 VOPNAME_CLOSE, { .vop_close = zfs_close },
5591 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5592 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5593 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5594 VOPNAME_ACCESS, { .vop_access = zfs_access },
5595 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5596 VOPNAME_CREATE, { .vop_create = zfs_create },
5597 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5598 VOPNAME_LINK, { .vop_link = zfs_link },
5599 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5600 VOPNAME_MKDIR, { .error = zfs_inval },
5601 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5602 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5603 VOPNAME_SYMLINK, { .error = zfs_inval },
5604 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5605 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5606 VOPNAME_FID, { .vop_fid = zfs_fid },
5607 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5608 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5609 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5610 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5611 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5616 * Error vnode operations template
5618 vnodeops_t *zfs_evnodeops;
5619 const fs_operation_def_t zfs_evnodeops_template[] = {
5620 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5621 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5627 ioflags(int ioflags)
5631 if (ioflags & IO_APPEND)
5633 if (ioflags & IO_NDELAY)
5635 if (ioflags & IO_SYNC)
5636 flags |= (FSYNC | FDSYNC | FRSYNC);
5642 zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage)
5644 znode_t *zp = VTOZ(vp);
5645 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5646 objset_t *os = zp->z_zfsvfs->z_os;
5647 vm_page_t mfirst, mlast, mreq;
5651 off_t startoff, endoff;
5653 vm_pindex_t reqstart, reqend;
5654 int pcount, lsize, reqsize, size;
5659 pcount = OFF_TO_IDX(round_page(count));
5661 object = mreq->object;
5664 KASSERT(vp->v_object == object, ("mismatching object"));
5666 if (pcount > 1 && zp->z_blksz > PAGESIZE) {
5667 startoff = rounddown(IDX_TO_OFF(mreq->pindex), zp->z_blksz);
5668 reqstart = OFF_TO_IDX(round_page(startoff));
5669 if (reqstart < m[0]->pindex)
5672 reqstart = reqstart - m[0]->pindex;
5673 endoff = roundup(IDX_TO_OFF(mreq->pindex) + PAGE_SIZE,
5675 reqend = OFF_TO_IDX(trunc_page(endoff)) - 1;
5676 if (reqend > m[pcount - 1]->pindex)
5677 reqend = m[pcount - 1]->pindex;
5678 reqsize = reqend - m[reqstart]->pindex + 1;
5679 KASSERT(reqstart <= reqpage && reqpage < reqstart + reqsize,
5680 ("reqpage beyond [reqstart, reqstart + reqsize[ bounds"));
5685 mfirst = m[reqstart];
5686 mlast = m[reqstart + reqsize - 1];
5688 zfs_vmobject_wlock(object);
5690 for (i = 0; i < reqstart; i++) {
5693 vm_page_unlock(m[i]);
5695 for (i = reqstart + reqsize; i < pcount; i++) {
5698 vm_page_unlock(m[i]);
5701 if (mreq->valid && reqsize == 1) {
5702 if (mreq->valid != VM_PAGE_BITS_ALL)
5703 vm_page_zero_invalid(mreq, TRUE);
5704 zfs_vmobject_wunlock(object);
5706 return (zfs_vm_pagerret_ok);
5709 PCPU_INC(cnt.v_vnodein);
5710 PCPU_ADD(cnt.v_vnodepgsin, reqsize);
5712 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) {
5713 for (i = reqstart; i < reqstart + reqsize; i++) {
5717 vm_page_unlock(m[i]);
5720 zfs_vmobject_wunlock(object);
5722 return (zfs_vm_pagerret_bad);
5726 if (IDX_TO_OFF(mlast->pindex) + lsize > object->un_pager.vnp.vnp_size)
5727 lsize = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mlast->pindex);
5729 zfs_vmobject_wunlock(object);
5731 for (i = reqstart; i < reqstart + reqsize; i++) {
5733 if (i == (reqstart + reqsize - 1))
5735 va = zfs_map_page(m[i], &sf);
5736 error = dmu_read(os, zp->z_id, IDX_TO_OFF(m[i]->pindex),
5737 size, va, DMU_READ_PREFETCH);
5738 if (size != PAGE_SIZE)
5739 bzero(va + size, PAGE_SIZE - size);
5745 zfs_vmobject_wlock(object);
5747 for (i = reqstart; i < reqstart + reqsize; i++) {
5749 m[i]->valid = VM_PAGE_BITS_ALL;
5750 KASSERT(m[i]->dirty == 0, ("zfs_getpages: page %p is dirty", m[i]));
5752 vm_page_readahead_finish(m[i]);
5755 zfs_vmobject_wunlock(object);
5757 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
5759 return (error ? zfs_vm_pagerret_error : zfs_vm_pagerret_ok);
5763 zfs_freebsd_getpages(ap)
5764 struct vop_getpages_args /* {
5769 vm_ooffset_t a_offset;
5773 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage));
5777 zfs_freebsd_bmap(ap)
5778 struct vop_bmap_args /* {
5781 struct bufobj **a_bop;
5788 if (ap->a_bop != NULL)
5789 *ap->a_bop = &ap->a_vp->v_bufobj;
5790 if (ap->a_bnp != NULL)
5791 *ap->a_bnp = ap->a_bn;
5792 if (ap->a_runp != NULL)
5794 if (ap->a_runb != NULL)
5801 zfs_freebsd_open(ap)
5802 struct vop_open_args /* {
5805 struct ucred *a_cred;
5806 struct thread *a_td;
5809 vnode_t *vp = ap->a_vp;
5810 znode_t *zp = VTOZ(vp);
5813 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL);
5815 vnode_create_vobject(vp, zp->z_size, ap->a_td);
5820 zfs_freebsd_close(ap)
5821 struct vop_close_args /* {
5824 struct ucred *a_cred;
5825 struct thread *a_td;
5829 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL));
5833 zfs_freebsd_ioctl(ap)
5834 struct vop_ioctl_args /* {
5844 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data,
5845 ap->a_fflag, ap->a_cred, NULL, NULL));
5849 zfs_freebsd_read(ap)
5850 struct vop_read_args /* {
5854 struct ucred *a_cred;
5858 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5863 zfs_freebsd_write(ap)
5864 struct vop_write_args /* {
5868 struct ucred *a_cred;
5872 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5877 zfs_freebsd_access(ap)
5878 struct vop_access_args /* {
5880 accmode_t a_accmode;
5881 struct ucred *a_cred;
5882 struct thread *a_td;
5885 vnode_t *vp = ap->a_vp;
5886 znode_t *zp = VTOZ(vp);
5891 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND,
5893 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND);
5895 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL);
5898 * VADMIN has to be handled by vaccess().
5901 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND);
5903 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
5904 zp->z_gid, accmode, ap->a_cred, NULL);
5909 * For VEXEC, ensure that at least one execute bit is set for
5912 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR &&
5913 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
5921 zfs_freebsd_lookup(ap)
5922 struct vop_lookup_args /* {
5923 struct vnode *a_dvp;
5924 struct vnode **a_vpp;
5925 struct componentname *a_cnp;
5928 struct componentname *cnp = ap->a_cnp;
5929 char nm[NAME_MAX + 1];
5931 ASSERT(cnp->cn_namelen < sizeof(nm));
5932 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm)));
5934 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop,
5935 cnp->cn_cred, cnp->cn_thread, 0));
5939 zfs_freebsd_create(ap)
5940 struct vop_create_args /* {
5941 struct vnode *a_dvp;
5942 struct vnode **a_vpp;
5943 struct componentname *a_cnp;
5944 struct vattr *a_vap;
5947 struct componentname *cnp = ap->a_cnp;
5948 vattr_t *vap = ap->a_vap;
5951 ASSERT(cnp->cn_flags & SAVENAME);
5953 vattr_init_mask(vap);
5954 mode = vap->va_mode & ALLPERMS;
5956 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode,
5957 ap->a_vpp, cnp->cn_cred, cnp->cn_thread));
5961 zfs_freebsd_remove(ap)
5962 struct vop_remove_args /* {
5963 struct vnode *a_dvp;
5965 struct componentname *a_cnp;
5969 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
5971 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr,
5972 ap->a_cnp->cn_cred, NULL, 0));
5976 zfs_freebsd_mkdir(ap)
5977 struct vop_mkdir_args /* {
5978 struct vnode *a_dvp;
5979 struct vnode **a_vpp;
5980 struct componentname *a_cnp;
5981 struct vattr *a_vap;
5984 vattr_t *vap = ap->a_vap;
5986 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
5988 vattr_init_mask(vap);
5990 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp,
5991 ap->a_cnp->cn_cred, NULL, 0, NULL));
5995 zfs_freebsd_rmdir(ap)
5996 struct vop_rmdir_args /* {
5997 struct vnode *a_dvp;
5999 struct componentname *a_cnp;
6002 struct componentname *cnp = ap->a_cnp;
6004 ASSERT(cnp->cn_flags & SAVENAME);
6006 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0));
6010 zfs_freebsd_readdir(ap)
6011 struct vop_readdir_args /* {
6014 struct ucred *a_cred;
6021 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag,
6022 ap->a_ncookies, ap->a_cookies));
6026 zfs_freebsd_fsync(ap)
6027 struct vop_fsync_args /* {
6030 struct thread *a_td;
6035 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL));
6039 zfs_freebsd_getattr(ap)
6040 struct vop_getattr_args /* {
6042 struct vattr *a_vap;
6043 struct ucred *a_cred;
6046 vattr_t *vap = ap->a_vap;
6052 xvap.xva_vattr = *vap;
6053 xvap.xva_vattr.va_mask |= AT_XVATTR;
6055 /* Convert chflags into ZFS-type flags. */
6056 /* XXX: what about SF_SETTABLE?. */
6057 XVA_SET_REQ(&xvap, XAT_IMMUTABLE);
6058 XVA_SET_REQ(&xvap, XAT_APPENDONLY);
6059 XVA_SET_REQ(&xvap, XAT_NOUNLINK);
6060 XVA_SET_REQ(&xvap, XAT_NODUMP);
6061 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL);
6065 /* Convert ZFS xattr into chflags. */
6066 #define FLAG_CHECK(fflag, xflag, xfield) do { \
6067 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \
6068 fflags |= (fflag); \
6070 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE,
6071 xvap.xva_xoptattrs.xoa_immutable);
6072 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY,
6073 xvap.xva_xoptattrs.xoa_appendonly);
6074 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK,
6075 xvap.xva_xoptattrs.xoa_nounlink);
6076 FLAG_CHECK(UF_NODUMP, XAT_NODUMP,
6077 xvap.xva_xoptattrs.xoa_nodump);
6079 *vap = xvap.xva_vattr;
6080 vap->va_flags = fflags;
6085 zfs_freebsd_setattr(ap)
6086 struct vop_setattr_args /* {
6088 struct vattr *a_vap;
6089 struct ucred *a_cred;
6092 vnode_t *vp = ap->a_vp;
6093 vattr_t *vap = ap->a_vap;
6094 cred_t *cred = ap->a_cred;
6099 vattr_init_mask(vap);
6100 vap->va_mask &= ~AT_NOSET;
6103 xvap.xva_vattr = *vap;
6105 zflags = VTOZ(vp)->z_pflags;
6107 if (vap->va_flags != VNOVAL) {
6108 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs;
6111 if (zfsvfs->z_use_fuids == B_FALSE)
6112 return (EOPNOTSUPP);
6114 fflags = vap->va_flags;
6115 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_NODUMP)) != 0)
6116 return (EOPNOTSUPP);
6118 * Unprivileged processes are not permitted to unset system
6119 * flags, or modify flags if any system flags are set.
6120 * Privileged non-jail processes may not modify system flags
6121 * if securelevel > 0 and any existing system flags are set.
6122 * Privileged jail processes behave like privileged non-jail
6123 * processes if the security.jail.chflags_allowed sysctl is
6124 * is non-zero; otherwise, they behave like unprivileged
6127 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 ||
6128 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) {
6130 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6131 error = securelevel_gt(cred, 0);
6137 * Callers may only modify the file flags on objects they
6138 * have VADMIN rights for.
6140 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0)
6143 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6147 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) {
6152 #define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \
6153 if (((fflags & (fflag)) && !(zflags & (zflag))) || \
6154 ((zflags & (zflag)) && !(fflags & (fflag)))) { \
6155 XVA_SET_REQ(&xvap, (xflag)); \
6156 (xfield) = ((fflags & (fflag)) != 0); \
6159 /* Convert chflags into ZFS-type flags. */
6160 /* XXX: what about SF_SETTABLE?. */
6161 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE,
6162 xvap.xva_xoptattrs.xoa_immutable);
6163 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY,
6164 xvap.xva_xoptattrs.xoa_appendonly);
6165 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK,
6166 xvap.xva_xoptattrs.xoa_nounlink);
6167 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP,
6168 xvap.xva_xoptattrs.xoa_nodump);
6171 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL));
6175 zfs_freebsd_rename(ap)
6176 struct vop_rename_args /* {
6177 struct vnode *a_fdvp;
6178 struct vnode *a_fvp;
6179 struct componentname *a_fcnp;
6180 struct vnode *a_tdvp;
6181 struct vnode *a_tvp;
6182 struct componentname *a_tcnp;
6185 vnode_t *fdvp = ap->a_fdvp;
6186 vnode_t *fvp = ap->a_fvp;
6187 vnode_t *tdvp = ap->a_tdvp;
6188 vnode_t *tvp = ap->a_tvp;
6191 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART));
6192 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART));
6194 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp,
6195 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0);
6210 zfs_freebsd_symlink(ap)
6211 struct vop_symlink_args /* {
6212 struct vnode *a_dvp;
6213 struct vnode **a_vpp;
6214 struct componentname *a_cnp;
6215 struct vattr *a_vap;
6219 struct componentname *cnp = ap->a_cnp;
6220 vattr_t *vap = ap->a_vap;
6222 ASSERT(cnp->cn_flags & SAVENAME);
6224 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */
6225 vattr_init_mask(vap);
6227 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap,
6228 ap->a_target, cnp->cn_cred, cnp->cn_thread));
6232 zfs_freebsd_readlink(ap)
6233 struct vop_readlink_args /* {
6236 struct ucred *a_cred;
6240 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL));
6244 zfs_freebsd_link(ap)
6245 struct vop_link_args /* {
6246 struct vnode *a_tdvp;
6248 struct componentname *a_cnp;
6251 struct componentname *cnp = ap->a_cnp;
6253 ASSERT(cnp->cn_flags & SAVENAME);
6255 return (zfs_link(ap->a_tdvp, ap->a_vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0));
6259 zfs_freebsd_inactive(ap)
6260 struct vop_inactive_args /* {
6262 struct thread *a_td;
6265 vnode_t *vp = ap->a_vp;
6267 zfs_inactive(vp, ap->a_td->td_ucred, NULL);
6272 zfs_freebsd_reclaim(ap)
6273 struct vop_reclaim_args /* {
6275 struct thread *a_td;
6278 vnode_t *vp = ap->a_vp;
6279 znode_t *zp = VTOZ(vp);
6280 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
6284 /* Destroy the vm object and flush associated pages. */
6285 vnode_destroy_vobject(vp);
6288 * z_teardown_inactive_lock protects from a race with
6289 * zfs_znode_dmu_fini in zfsvfs_teardown during
6292 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
6293 if (zp->z_sa_hdl == NULL)
6297 rw_exit(&zfsvfs->z_teardown_inactive_lock);
6305 struct vop_fid_args /* {
6311 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL));
6315 zfs_freebsd_pathconf(ap)
6316 struct vop_pathconf_args /* {
6319 register_t *a_retval;
6325 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL);
6327 *ap->a_retval = val;
6328 else if (error == EOPNOTSUPP)
6329 error = vop_stdpathconf(ap);
6334 zfs_freebsd_fifo_pathconf(ap)
6335 struct vop_pathconf_args /* {
6338 register_t *a_retval;
6342 switch (ap->a_name) {
6343 case _PC_ACL_EXTENDED:
6345 case _PC_ACL_PATH_MAX:
6346 case _PC_MAC_PRESENT:
6347 return (zfs_freebsd_pathconf(ap));
6349 return (fifo_specops.vop_pathconf(ap));
6354 * FreeBSD's extended attributes namespace defines file name prefix for ZFS'
6355 * extended attribute name:
6358 * system freebsd:system:
6359 * user (none, can be used to access ZFS fsattr(5) attributes
6360 * created on Solaris)
6363 zfs_create_attrname(int attrnamespace, const char *name, char *attrname,
6366 const char *namespace, *prefix, *suffix;
6368 /* We don't allow '/' character in attribute name. */
6369 if (strchr(name, '/') != NULL)
6371 /* We don't allow attribute names that start with "freebsd:" string. */
6372 if (strncmp(name, "freebsd:", 8) == 0)
6375 bzero(attrname, size);
6377 switch (attrnamespace) {
6378 case EXTATTR_NAMESPACE_USER:
6380 prefix = "freebsd:";
6381 namespace = EXTATTR_NAMESPACE_USER_STRING;
6385 * This is the default namespace by which we can access all
6386 * attributes created on Solaris.
6388 prefix = namespace = suffix = "";
6391 case EXTATTR_NAMESPACE_SYSTEM:
6392 prefix = "freebsd:";
6393 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING;
6396 case EXTATTR_NAMESPACE_EMPTY:
6400 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix,
6402 return (ENAMETOOLONG);
6408 * Vnode operating to retrieve a named extended attribute.
6411 zfs_getextattr(struct vop_getextattr_args *ap)
6414 IN struct vnode *a_vp;
6415 IN int a_attrnamespace;
6416 IN const char *a_name;
6417 INOUT struct uio *a_uio;
6419 IN struct ucred *a_cred;
6420 IN struct thread *a_td;
6424 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6425 struct thread *td = ap->a_td;
6426 struct nameidata nd;
6429 vnode_t *xvp = NULL, *vp;
6432 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6433 ap->a_cred, ap->a_td, VREAD);
6437 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6444 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6452 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6454 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL);
6456 NDFREE(&nd, NDF_ONLY_PNBUF);
6459 if (error == ENOENT)
6464 if (ap->a_size != NULL) {
6465 error = VOP_GETATTR(vp, &va, ap->a_cred);
6467 *ap->a_size = (size_t)va.va_size;
6468 } else if (ap->a_uio != NULL)
6469 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6472 vn_close(vp, flags, ap->a_cred, td);
6479 * Vnode operation to remove a named attribute.
6482 zfs_deleteextattr(struct vop_deleteextattr_args *ap)
6485 IN struct vnode *a_vp;
6486 IN int a_attrnamespace;
6487 IN const char *a_name;
6488 IN struct ucred *a_cred;
6489 IN struct thread *a_td;
6493 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6494 struct thread *td = ap->a_td;
6495 struct nameidata nd;
6498 vnode_t *xvp = NULL, *vp;
6501 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6502 ap->a_cred, ap->a_td, VWRITE);
6506 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6513 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6520 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF,
6521 UIO_SYSSPACE, attrname, xvp, td);
6524 NDFREE(&nd, NDF_ONLY_PNBUF);
6527 if (error == ENOENT)
6531 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
6534 if (vp == nd.ni_dvp)
6544 * Vnode operation to set a named attribute.
6547 zfs_setextattr(struct vop_setextattr_args *ap)
6550 IN struct vnode *a_vp;
6551 IN int a_attrnamespace;
6552 IN const char *a_name;
6553 INOUT struct uio *a_uio;
6554 IN struct ucred *a_cred;
6555 IN struct thread *a_td;
6559 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6560 struct thread *td = ap->a_td;
6561 struct nameidata nd;
6564 vnode_t *xvp = NULL, *vp;
6567 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6568 ap->a_cred, ap->a_td, VWRITE);
6572 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6579 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6580 LOOKUP_XATTR | CREATE_XATTR_DIR);
6586 flags = FFLAGS(O_WRONLY | O_CREAT);
6587 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6589 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL);
6591 NDFREE(&nd, NDF_ONLY_PNBUF);
6599 error = VOP_SETATTR(vp, &va, ap->a_cred);
6601 VOP_WRITE(vp, ap->a_uio, IO_UNIT | IO_SYNC, ap->a_cred);
6604 vn_close(vp, flags, ap->a_cred, td);
6611 * Vnode operation to retrieve extended attributes on a vnode.
6614 zfs_listextattr(struct vop_listextattr_args *ap)
6617 IN struct vnode *a_vp;
6618 IN int a_attrnamespace;
6619 INOUT struct uio *a_uio;
6621 IN struct ucred *a_cred;
6622 IN struct thread *a_td;
6626 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6627 struct thread *td = ap->a_td;
6628 struct nameidata nd;
6629 char attrprefix[16];
6630 u_char dirbuf[sizeof(struct dirent)];
6633 struct uio auio, *uio = ap->a_uio;
6634 size_t *sizep = ap->a_size;
6636 vnode_t *xvp = NULL, *vp;
6637 int done, error, eof, pos;
6639 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6640 ap->a_cred, ap->a_td, VREAD);
6644 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix,
6645 sizeof(attrprefix));
6648 plen = strlen(attrprefix);
6655 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6660 * ENOATTR means that the EA directory does not yet exist,
6661 * i.e. there are no extended attributes there.
6663 if (error == ENOATTR)
6668 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED,
6669 UIO_SYSSPACE, ".", xvp, td);
6672 NDFREE(&nd, NDF_ONLY_PNBUF);
6678 auio.uio_iov = &aiov;
6679 auio.uio_iovcnt = 1;
6680 auio.uio_segflg = UIO_SYSSPACE;
6682 auio.uio_rw = UIO_READ;
6683 auio.uio_offset = 0;
6688 aiov.iov_base = (void *)dirbuf;
6689 aiov.iov_len = sizeof(dirbuf);
6690 auio.uio_resid = sizeof(dirbuf);
6691 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL);
6692 done = sizeof(dirbuf) - auio.uio_resid;
6695 for (pos = 0; pos < done;) {
6696 dp = (struct dirent *)(dirbuf + pos);
6697 pos += dp->d_reclen;
6699 * XXX: Temporarily we also accept DT_UNKNOWN, as this
6700 * is what we get when attribute was created on Solaris.
6702 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN)
6704 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0)
6706 else if (strncmp(dp->d_name, attrprefix, plen) != 0)
6708 nlen = dp->d_namlen - plen;
6711 else if (uio != NULL) {
6713 * Format of extattr name entry is one byte for
6714 * length and the rest for name.
6716 error = uiomove(&nlen, 1, uio->uio_rw, uio);
6718 error = uiomove(dp->d_name + plen, nlen,
6725 } while (!eof && error == 0);
6734 zfs_freebsd_getacl(ap)
6735 struct vop_getacl_args /* {
6744 vsecattr_t vsecattr;
6746 if (ap->a_type != ACL_TYPE_NFS4)
6749 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
6750 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL))
6753 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt);
6754 if (vsecattr.vsa_aclentp != NULL)
6755 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz);
6761 zfs_freebsd_setacl(ap)
6762 struct vop_setacl_args /* {
6771 vsecattr_t vsecattr;
6772 int aclbsize; /* size of acl list in bytes */
6775 if (ap->a_type != ACL_TYPE_NFS4)
6778 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES)
6782 * With NFSv4 ACLs, chmod(2) may need to add additional entries,
6783 * splitting every entry into two and appending "canonical six"
6784 * entries at the end. Don't allow for setting an ACL that would
6785 * cause chmod(2) to run out of ACL entries.
6787 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES)
6790 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR);
6794 vsecattr.vsa_mask = VSA_ACE;
6795 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t);
6796 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP);
6797 aaclp = vsecattr.vsa_aclentp;
6798 vsecattr.vsa_aclentsz = aclbsize;
6800 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp);
6801 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL);
6802 kmem_free(aaclp, aclbsize);
6808 zfs_freebsd_aclcheck(ap)
6809 struct vop_aclcheck_args /* {
6818 return (EOPNOTSUPP);
6821 struct vop_vector zfs_vnodeops;
6822 struct vop_vector zfs_fifoops;
6823 struct vop_vector zfs_shareops;
6825 struct vop_vector zfs_vnodeops = {
6826 .vop_default = &default_vnodeops,
6827 .vop_inactive = zfs_freebsd_inactive,
6828 .vop_reclaim = zfs_freebsd_reclaim,
6829 .vop_access = zfs_freebsd_access,
6830 #ifdef FREEBSD_NAMECACHE
6831 .vop_lookup = vfs_cache_lookup,
6832 .vop_cachedlookup = zfs_freebsd_lookup,
6834 .vop_lookup = zfs_freebsd_lookup,
6836 .vop_getattr = zfs_freebsd_getattr,
6837 .vop_setattr = zfs_freebsd_setattr,
6838 .vop_create = zfs_freebsd_create,
6839 .vop_mknod = zfs_freebsd_create,
6840 .vop_mkdir = zfs_freebsd_mkdir,
6841 .vop_readdir = zfs_freebsd_readdir,
6842 .vop_fsync = zfs_freebsd_fsync,
6843 .vop_open = zfs_freebsd_open,
6844 .vop_close = zfs_freebsd_close,
6845 .vop_rmdir = zfs_freebsd_rmdir,
6846 .vop_ioctl = zfs_freebsd_ioctl,
6847 .vop_link = zfs_freebsd_link,
6848 .vop_symlink = zfs_freebsd_symlink,
6849 .vop_readlink = zfs_freebsd_readlink,
6850 .vop_read = zfs_freebsd_read,
6851 .vop_write = zfs_freebsd_write,
6852 .vop_remove = zfs_freebsd_remove,
6853 .vop_rename = zfs_freebsd_rename,
6854 .vop_pathconf = zfs_freebsd_pathconf,
6855 .vop_bmap = zfs_freebsd_bmap,
6856 .vop_fid = zfs_freebsd_fid,
6857 .vop_getextattr = zfs_getextattr,
6858 .vop_deleteextattr = zfs_deleteextattr,
6859 .vop_setextattr = zfs_setextattr,
6860 .vop_listextattr = zfs_listextattr,
6861 .vop_getacl = zfs_freebsd_getacl,
6862 .vop_setacl = zfs_freebsd_setacl,
6863 .vop_aclcheck = zfs_freebsd_aclcheck,
6864 .vop_getpages = zfs_freebsd_getpages,
6867 struct vop_vector zfs_fifoops = {
6868 .vop_default = &fifo_specops,
6869 .vop_fsync = zfs_freebsd_fsync,
6870 .vop_access = zfs_freebsd_access,
6871 .vop_getattr = zfs_freebsd_getattr,
6872 .vop_inactive = zfs_freebsd_inactive,
6873 .vop_read = VOP_PANIC,
6874 .vop_reclaim = zfs_freebsd_reclaim,
6875 .vop_setattr = zfs_freebsd_setattr,
6876 .vop_write = VOP_PANIC,
6877 .vop_pathconf = zfs_freebsd_fifo_pathconf,
6878 .vop_fid = zfs_freebsd_fid,
6879 .vop_getacl = zfs_freebsd_getacl,
6880 .vop_setacl = zfs_freebsd_setacl,
6881 .vop_aclcheck = zfs_freebsd_aclcheck,
6885 * special share hidden files vnode operations template
6887 struct vop_vector zfs_shareops = {
6888 .vop_default = &default_vnodeops,
6889 .vop_access = zfs_freebsd_access,
6890 .vop_inactive = zfs_freebsd_inactive,
6891 .vop_reclaim = zfs_freebsd_reclaim,
6892 .vop_fid = zfs_freebsd_fid,
6893 .vop_pathconf = zfs_freebsd_pathconf,