4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2013 by Delphix. All rights reserved.
26 /* Portions Copyright 2007 Jeremy Teo */
27 /* Portions Copyright 2010 Robert Milkowski */
29 #include <sys/types.h>
30 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/sysmacros.h>
34 #include <sys/resource.h>
35 #include <sys/resourcevar.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_pageout.h>
81 * Each vnode op performs some logical unit of work. To do this, the ZPL must
82 * properly lock its in-core state, create a DMU transaction, do the work,
83 * record this work in the intent log (ZIL), commit the DMU transaction,
84 * and wait for the intent log to commit if it is a synchronous operation.
85 * Moreover, the vnode ops must work in both normal and log replay context.
86 * The ordering of events is important to avoid deadlocks and references
87 * to freed memory. The example below illustrates the following Big Rules:
89 * (1) A check must be made in each zfs thread for a mounted file system.
90 * This is done avoiding races using ZFS_ENTER(zfsvfs).
91 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
92 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
93 * can return EIO from the calling function.
95 * (2) VN_RELE() should always be the last thing except for zil_commit()
96 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
97 * First, if it's the last reference, the vnode/znode
98 * can be freed, so the zp may point to freed memory. Second, the last
99 * reference will call zfs_zinactive(), which may induce a lot of work --
100 * pushing cached pages (which acquires range locks) and syncing out
101 * cached atime changes. Third, zfs_zinactive() may require a new tx,
102 * which could deadlock the system if you were already holding one.
103 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
105 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
106 * as they can span dmu_tx_assign() calls.
108 * (4) Always pass TXG_NOWAIT as the second argument to dmu_tx_assign().
109 * This is critical because we don't want to block while holding locks.
110 * Note, in particular, that if a lock is sometimes acquired before
111 * the tx assigns, and sometimes after (e.g. z_lock), then failing to
112 * use a non-blocking assign can deadlock the system. The scenario:
114 * Thread A has grabbed a lock before calling dmu_tx_assign().
115 * Thread B is in an already-assigned tx, and blocks for this lock.
116 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
117 * forever, because the previous txg can't quiesce until B's tx commits.
119 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
120 * then drop all locks, call dmu_tx_wait(), and try again.
122 * (5) If the operation succeeded, generate the intent log entry for it
123 * before dropping locks. This ensures that the ordering of events
124 * in the intent log matches the order in which they actually occurred.
125 * During ZIL replay the zfs_log_* functions will update the sequence
126 * number to indicate the zil transaction has replayed.
128 * (6) At the end of each vnode op, the DMU tx must always commit,
129 * regardless of whether there were any errors.
131 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
132 * to ensure that synchronous semantics are provided when necessary.
134 * In general, this is how things should be ordered in each vnode op:
136 * ZFS_ENTER(zfsvfs); // exit if unmounted
138 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD())
139 * rw_enter(...); // grab any other locks you need
140 * tx = dmu_tx_create(...); // get DMU tx
141 * dmu_tx_hold_*(); // hold each object you might modify
142 * error = dmu_tx_assign(tx, TXG_NOWAIT); // try to assign
144 * rw_exit(...); // drop locks
145 * zfs_dirent_unlock(dl); // unlock directory entry
146 * VN_RELE(...); // release held vnodes
147 * if (error == ERESTART) {
152 * dmu_tx_abort(tx); // abort DMU tx
153 * ZFS_EXIT(zfsvfs); // finished in zfs
154 * return (error); // really out of space
156 * error = do_real_work(); // do whatever this VOP does
158 * zfs_log_*(...); // on success, make ZIL entry
159 * dmu_tx_commit(tx); // commit DMU tx -- error or not
160 * rw_exit(...); // drop locks
161 * zfs_dirent_unlock(dl); // unlock directory entry
162 * VN_RELE(...); // release held vnodes
163 * zil_commit(zilog, foid); // synchronous when necessary
164 * ZFS_EXIT(zfsvfs); // finished in zfs
165 * return (error); // done, report error
170 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
172 znode_t *zp = VTOZ(*vpp);
173 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
178 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
179 ((flag & FAPPEND) == 0)) {
181 return (SET_ERROR(EPERM));
184 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
185 ZTOV(zp)->v_type == VREG &&
186 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
187 if (fs_vscan(*vpp, cr, 0) != 0) {
189 return (SET_ERROR(EACCES));
193 /* Keep a count of the synchronous opens in the znode */
194 if (flag & (FSYNC | FDSYNC))
195 atomic_inc_32(&zp->z_sync_cnt);
203 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
204 caller_context_t *ct)
206 znode_t *zp = VTOZ(vp);
207 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
210 * Clean up any locks held by this process on the vp.
212 cleanlocks(vp, ddi_get_pid(), 0);
213 cleanshares(vp, ddi_get_pid());
218 /* Decrement the synchronous opens in the znode */
219 if ((flag & (FSYNC | FDSYNC)) && (count == 1))
220 atomic_dec_32(&zp->z_sync_cnt);
222 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
223 ZTOV(zp)->v_type == VREG &&
224 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
225 VERIFY(fs_vscan(vp, cr, 1) == 0);
232 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
233 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
236 zfs_holey(vnode_t *vp, u_long cmd, offset_t *off)
238 znode_t *zp = VTOZ(vp);
239 uint64_t noff = (uint64_t)*off; /* new offset */
244 file_sz = zp->z_size;
245 if (noff >= file_sz) {
246 return (SET_ERROR(ENXIO));
249 if (cmd == _FIO_SEEK_HOLE)
254 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
257 if ((error == ESRCH) || (noff > file_sz)) {
259 * Handle the virtual hole at the end of file.
265 return (SET_ERROR(ENXIO));
276 zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred,
277 int *rvalp, caller_context_t *ct)
289 * The following two ioctls are used by bfu. Faking out,
290 * necessary to avoid bfu errors.
299 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
300 return (SET_ERROR(EFAULT));
302 off = *(offset_t *)data;
305 zfsvfs = zp->z_zfsvfs;
309 /* offset parameter is in/out */
310 error = zfs_holey(vp, com, &off);
315 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
316 return (SET_ERROR(EFAULT));
318 *(offset_t *)data = off;
322 return (SET_ERROR(ENOTTY));
326 page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes)
332 VM_OBJECT_LOCK_ASSERT(obj, MA_OWNED);
335 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
337 if (vm_page_sleep_if_busy(pp, FALSE, "zfsmwb"))
340 pp = vm_page_alloc(obj, OFF_TO_IDX(start),
341 VM_ALLOC_SYSTEM | VM_ALLOC_IFCACHED |
346 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
347 vm_object_pip_add(obj, 1);
348 vm_page_io_start(pp);
349 vm_page_lock_queues();
350 pmap_remove_write(pp);
351 vm_page_clear_dirty(pp, off, nbytes);
352 vm_page_unlock_queues();
360 page_unbusy(vm_page_t pp)
363 vm_page_io_finish(pp);
364 vm_object_pip_subtract(pp->object, 1);
368 page_hold(vnode_t *vp, int64_t start)
374 VM_OBJECT_LOCK_ASSERT(obj, MA_OWNED);
377 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
379 if (vm_page_sleep_if_busy(pp, FALSE, "zfsmwb"))
381 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
382 vm_page_lock_queues();
384 vm_page_unlock_queues();
394 page_unhold(vm_page_t pp)
397 vm_page_lock_queues();
399 vm_page_unlock_queues();
403 zfs_map_page(vm_page_t pp, struct sf_buf **sfp)
406 *sfp = sf_buf_alloc(pp, 0);
407 return ((caddr_t)sf_buf_kva(*sfp));
411 zfs_unmap_page(struct sf_buf *sf)
418 * When a file is memory mapped, we must keep the IO data synchronized
419 * between the DMU cache and the memory mapped pages. What this means:
421 * On Write: If we find a memory mapped page, we write to *both*
422 * the page and the dmu buffer.
425 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid,
426 int segflg, dmu_tx_t *tx)
433 ASSERT(vp->v_mount != NULL);
437 off = start & PAGEOFFSET;
439 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
441 int nbytes = imin(PAGESIZE - off, len);
443 if (segflg == UIO_NOCOPY) {
444 pp = vm_page_lookup(obj, OFF_TO_IDX(start));
446 ("zfs update_pages: NULL page in putpages case"));
448 ("zfs update_pages: unaligned data in putpages case"));
449 KASSERT(pp->valid == VM_PAGE_BITS_ALL,
450 ("zfs update_pages: invalid page in putpages case"));
451 KASSERT(pp->busy > 0,
452 ("zfs update_pages: unbusy page in putpages case"));
453 KASSERT((pp->flags & PG_WRITEABLE) == 0,
454 ("zfs update_pages: writable page in putpages case"));
455 VM_OBJECT_UNLOCK(obj);
457 va = zfs_map_page(pp, &sf);
458 (void) dmu_write(os, oid, start, nbytes, va, tx);
463 } else if ((pp = page_busy(vp, start, off, nbytes)) != NULL) {
464 VM_OBJECT_UNLOCK(obj);
466 va = zfs_map_page(pp, &sf);
467 (void) dmu_read(os, oid, start+off, nbytes,
468 va+off, DMU_READ_PREFETCH);;
477 if (segflg != UIO_NOCOPY)
478 vm_object_pip_wakeupn(obj, 0);
479 VM_OBJECT_UNLOCK(obj);
483 * Read with UIO_NOCOPY flag means that sendfile(2) requests
484 * ZFS to populate a range of page cache pages with data.
486 * NOTE: this function could be optimized to pre-allocate
487 * all pages in advance, drain VPO_BUSY on all of them,
488 * map them into contiguous KVA region and populate them
489 * in one single dmu_read() call.
492 mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio)
494 znode_t *zp = VTOZ(vp);
495 objset_t *os = zp->z_zfsvfs->z_os;
505 ASSERT(uio->uio_segflg == UIO_NOCOPY);
506 ASSERT(vp->v_mount != NULL);
509 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0);
512 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) {
513 int bytes = MIN(PAGESIZE, len);
516 pp = vm_page_lookup(obj, OFF_TO_IDX(start));
517 if (pp != NULL && vm_page_sleep_if_busy(pp, FALSE,
521 pp = vm_page_alloc(obj, OFF_TO_IDX(start),
522 VM_ALLOC_NOBUSY | VM_ALLOC_NORMAL);
524 VM_OBJECT_UNLOCK(obj);
530 if (pp->valid == 0) {
531 vm_page_io_start(pp);
532 VM_OBJECT_UNLOCK(obj);
533 va = zfs_map_page(pp, &sf);
534 error = dmu_read(os, zp->z_id, start, bytes, va,
536 if (bytes != PAGESIZE && error == 0)
537 bzero(va + bytes, PAGESIZE - bytes);
540 vm_page_io_finish(pp);
541 vm_page_lock_queues();
543 if (pp->wire_count == 0 && pp->valid == 0 &&
544 pp->busy == 0 && !(pp->oflags & VPO_BUSY))
547 pp->valid = VM_PAGE_BITS_ALL;
548 vm_page_activate(pp);
550 vm_page_unlock_queues();
554 uio->uio_resid -= bytes;
555 uio->uio_offset += bytes;
558 VM_OBJECT_UNLOCK(obj);
563 * When a file is memory mapped, we must keep the IO data synchronized
564 * between the DMU cache and the memory mapped pages. What this means:
566 * On Read: We "read" preferentially from memory mapped pages,
567 * else we default from the dmu buffer.
569 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
570 * the file is memory mapped.
573 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
575 znode_t *zp = VTOZ(vp);
576 objset_t *os = zp->z_zfsvfs->z_os;
584 ASSERT(vp->v_mount != NULL);
588 start = uio->uio_loffset;
589 off = start & PAGEOFFSET;
591 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
593 uint64_t bytes = MIN(PAGESIZE - off, len);
595 if (pp = page_hold(vp, start)) {
598 VM_OBJECT_UNLOCK(obj);
599 va = zfs_map_page(pp, &sf);
600 error = uiomove(va + off, bytes, UIO_READ, uio);
605 VM_OBJECT_UNLOCK(obj);
606 error = dmu_read_uio(os, zp->z_id, uio, bytes);
614 VM_OBJECT_UNLOCK(obj);
618 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
621 * Read bytes from specified file into supplied buffer.
623 * IN: vp - vnode of file to be read from.
624 * uio - structure supplying read location, range info,
626 * ioflag - SYNC flags; used to provide FRSYNC semantics.
627 * cr - credentials of caller.
628 * ct - caller context
630 * OUT: uio - updated offset and range, buffer filled.
632 * RETURN: 0 on success, error code on failure.
635 * vp - atime updated if byte count > 0
639 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
641 znode_t *zp = VTOZ(vp);
642 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
653 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
655 return (SET_ERROR(EACCES));
659 * Validate file offset
661 if (uio->uio_loffset < (offset_t)0) {
663 return (SET_ERROR(EINVAL));
667 * Fasttrack empty reads
669 if (uio->uio_resid == 0) {
675 * Check for mandatory locks
677 if (MANDMODE(zp->z_mode)) {
678 if (error = chklock(vp, FREAD,
679 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
686 * If we're in FRSYNC mode, sync out this znode before reading it.
689 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
690 zil_commit(zfsvfs->z_log, zp->z_id);
693 * Lock the range against changes.
695 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
698 * If we are reading past end-of-file we can skip
699 * to the end; but we might still need to set atime.
701 if (uio->uio_loffset >= zp->z_size) {
706 ASSERT(uio->uio_loffset < zp->z_size);
707 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
710 if ((uio->uio_extflg == UIO_XUIO) &&
711 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
713 int blksz = zp->z_blksz;
714 uint64_t offset = uio->uio_loffset;
716 xuio = (xuio_t *)uio;
718 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
721 ASSERT(offset + n <= blksz);
724 (void) dmu_xuio_init(xuio, nblk);
726 if (vn_has_cached_data(vp)) {
728 * For simplicity, we always allocate a full buffer
729 * even if we only expect to read a portion of a block.
731 while (--nblk >= 0) {
732 (void) dmu_xuio_add(xuio,
733 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
741 nbytes = MIN(n, zfs_read_chunk_size -
742 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
745 if (uio->uio_segflg == UIO_NOCOPY)
746 error = mappedread_sf(vp, nbytes, uio);
748 #endif /* __FreeBSD__ */
749 if (vn_has_cached_data(vp))
750 error = mappedread(vp, nbytes, uio);
752 error = dmu_read_uio(os, zp->z_id, uio, nbytes);
754 /* convert checksum errors into IO errors */
756 error = SET_ERROR(EIO);
763 zfs_range_unlock(rl);
765 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
771 * Write the bytes to a file.
773 * IN: vp - vnode of file to be written to.
774 * uio - structure supplying write location, range info,
776 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is
777 * set if in append mode.
778 * cr - credentials of caller.
779 * ct - caller context (NFS/CIFS fem monitor only)
781 * OUT: uio - updated offset and range.
783 * RETURN: 0 on success, error code on failure.
786 * vp - ctime|mtime updated if byte count > 0
791 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
793 znode_t *zp = VTOZ(vp);
794 rlim64_t limit = MAXOFFSET_T;
795 ssize_t start_resid = uio->uio_resid;
799 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
804 int max_blksz = zfsvfs->z_max_blksz;
807 iovec_t *aiov = NULL;
810 int iovcnt = uio->uio_iovcnt;
811 iovec_t *iovp = uio->uio_iov;
814 sa_bulk_attr_t bulk[4];
815 uint64_t mtime[2], ctime[2];
818 * Fasttrack empty write
824 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
830 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
831 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
832 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
834 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
838 * If immutable or not appending then return EPERM
840 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
841 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
842 (uio->uio_loffset < zp->z_size))) {
844 return (SET_ERROR(EPERM));
847 zilog = zfsvfs->z_log;
850 * Validate file offset
852 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
855 return (SET_ERROR(EINVAL));
859 * Check for mandatory locks before calling zfs_range_lock()
860 * in order to prevent a deadlock with locks set via fcntl().
862 if (MANDMODE((mode_t)zp->z_mode) &&
863 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
870 * Pre-fault the pages to ensure slow (eg NFS) pages
872 * Skip this if uio contains loaned arc_buf.
874 if ((uio->uio_extflg == UIO_XUIO) &&
875 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
876 xuio = (xuio_t *)uio;
878 uio_prefaultpages(MIN(n, max_blksz), uio);
882 * If in append mode, set the io offset pointer to eof.
884 if (ioflag & FAPPEND) {
886 * Obtain an appending range lock to guarantee file append
887 * semantics. We reset the write offset once we have the lock.
889 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
891 if (rl->r_len == UINT64_MAX) {
893 * We overlocked the file because this write will cause
894 * the file block size to increase.
895 * Note that zp_size cannot change with this lock held.
899 uio->uio_loffset = woff;
902 * Note that if the file block size will change as a result of
903 * this write, then this range lock will lock the entire file
904 * so that we can re-write the block safely.
906 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
910 zfs_range_unlock(rl);
912 return (SET_ERROR(EFBIG));
915 if ((woff + n) > limit || woff > (limit - n))
918 /* Will this write extend the file length? */
919 write_eof = (woff + n > zp->z_size);
921 end_size = MAX(zp->z_size, woff + n);
924 * Write the file in reasonable size chunks. Each chunk is written
925 * in a separate transaction; this keeps the intent log records small
926 * and allows us to do more fine-grained space accounting.
930 woff = uio->uio_loffset;
932 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
933 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
935 dmu_return_arcbuf(abuf);
936 error = SET_ERROR(EDQUOT);
940 if (xuio && abuf == NULL) {
941 ASSERT(i_iov < iovcnt);
943 abuf = dmu_xuio_arcbuf(xuio, i_iov);
944 dmu_xuio_clear(xuio, i_iov);
945 DTRACE_PROBE3(zfs_cp_write, int, i_iov,
946 iovec_t *, aiov, arc_buf_t *, abuf);
947 ASSERT((aiov->iov_base == abuf->b_data) ||
948 ((char *)aiov->iov_base - (char *)abuf->b_data +
949 aiov->iov_len == arc_buf_size(abuf)));
951 } else if (abuf == NULL && n >= max_blksz &&
952 woff >= zp->z_size &&
953 P2PHASE(woff, max_blksz) == 0 &&
954 zp->z_blksz == max_blksz) {
956 * This write covers a full block. "Borrow" a buffer
957 * from the dmu so that we can fill it before we enter
958 * a transaction. This avoids the possibility of
959 * holding up the transaction if the data copy hangs
960 * up on a pagefault (e.g., from an NFS server mapping).
964 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
966 ASSERT(abuf != NULL);
967 ASSERT(arc_buf_size(abuf) == max_blksz);
968 if (error = uiocopy(abuf->b_data, max_blksz,
969 UIO_WRITE, uio, &cbytes)) {
970 dmu_return_arcbuf(abuf);
973 ASSERT(cbytes == max_blksz);
977 * Start a transaction.
979 tx = dmu_tx_create(zfsvfs->z_os);
980 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
981 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
982 zfs_sa_upgrade_txholds(tx, zp);
983 error = dmu_tx_assign(tx, TXG_NOWAIT);
985 if (error == ERESTART) {
992 dmu_return_arcbuf(abuf);
997 * If zfs_range_lock() over-locked we grow the blocksize
998 * and then reduce the lock range. This will only happen
999 * on the first iteration since zfs_range_reduce() will
1000 * shrink down r_len to the appropriate size.
1002 if (rl->r_len == UINT64_MAX) {
1005 if (zp->z_blksz > max_blksz) {
1006 ASSERT(!ISP2(zp->z_blksz));
1007 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
1009 new_blksz = MIN(end_size, max_blksz);
1011 zfs_grow_blocksize(zp, new_blksz, tx);
1012 zfs_range_reduce(rl, woff, n);
1016 * XXX - should we really limit each write to z_max_blksz?
1017 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
1019 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
1021 if (woff + nbytes > zp->z_size)
1022 vnode_pager_setsize(vp, woff + nbytes);
1025 tx_bytes = uio->uio_resid;
1026 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
1028 tx_bytes -= uio->uio_resid;
1031 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
1033 * If this is not a full block write, but we are
1034 * extending the file past EOF and this data starts
1035 * block-aligned, use assign_arcbuf(). Otherwise,
1036 * write via dmu_write().
1038 if (tx_bytes < max_blksz && (!write_eof ||
1039 aiov->iov_base != abuf->b_data)) {
1041 dmu_write(zfsvfs->z_os, zp->z_id, woff,
1042 aiov->iov_len, aiov->iov_base, tx);
1043 dmu_return_arcbuf(abuf);
1044 xuio_stat_wbuf_copied();
1046 ASSERT(xuio || tx_bytes == max_blksz);
1047 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
1050 ASSERT(tx_bytes <= uio->uio_resid);
1051 uioskip(uio, tx_bytes);
1053 if (tx_bytes && vn_has_cached_data(vp)) {
1054 update_pages(vp, woff, tx_bytes, zfsvfs->z_os,
1055 zp->z_id, uio->uio_segflg, tx);
1059 * If we made no progress, we're done. If we made even
1060 * partial progress, update the znode and ZIL accordingly.
1062 if (tx_bytes == 0) {
1063 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
1064 (void *)&zp->z_size, sizeof (uint64_t), tx);
1071 * Clear Set-UID/Set-GID bits on successful write if not
1072 * privileged and at least one of the excute bits is set.
1074 * It would be nice to to this after all writes have
1075 * been done, but that would still expose the ISUID/ISGID
1076 * to another app after the partial write is committed.
1078 * Note: we don't call zfs_fuid_map_id() here because
1079 * user 0 is not an ephemeral uid.
1081 mutex_enter(&zp->z_acl_lock);
1082 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
1083 (S_IXUSR >> 6))) != 0 &&
1084 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
1085 secpolicy_vnode_setid_retain(vp, cr,
1086 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
1088 zp->z_mode &= ~(S_ISUID | S_ISGID);
1089 newmode = zp->z_mode;
1090 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
1091 (void *)&newmode, sizeof (uint64_t), tx);
1093 mutex_exit(&zp->z_acl_lock);
1095 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
1099 * Update the file size (zp_size) if it has changed;
1100 * account for possible concurrent updates.
1102 while ((end_size = zp->z_size) < uio->uio_loffset) {
1103 (void) atomic_cas_64(&zp->z_size, end_size,
1108 * If we are replaying and eof is non zero then force
1109 * the file size to the specified eof. Note, there's no
1110 * concurrency during replay.
1112 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
1113 zp->z_size = zfsvfs->z_replay_eof;
1115 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1117 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
1122 ASSERT(tx_bytes == nbytes);
1127 uio_prefaultpages(MIN(n, max_blksz), uio);
1131 zfs_range_unlock(rl);
1134 * If we're in replay mode, or we made no progress, return error.
1135 * Otherwise, it's at least a partial write, so it's successful.
1137 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1142 if (ioflag & (FSYNC | FDSYNC) ||
1143 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1144 zil_commit(zilog, zp->z_id);
1151 zfs_get_done(zgd_t *zgd, int error)
1153 znode_t *zp = zgd->zgd_private;
1154 objset_t *os = zp->z_zfsvfs->z_os;
1158 dmu_buf_rele(zgd->zgd_db, zgd);
1160 zfs_range_unlock(zgd->zgd_rl);
1162 vfslocked = VFS_LOCK_GIANT(zp->z_zfsvfs->z_vfs);
1164 * Release the vnode asynchronously as we currently have the
1165 * txg stopped from syncing.
1167 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1169 if (error == 0 && zgd->zgd_bp)
1170 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1172 kmem_free(zgd, sizeof (zgd_t));
1173 VFS_UNLOCK_GIANT(vfslocked);
1177 static int zil_fault_io = 0;
1181 * Get data to generate a TX_WRITE intent log record.
1184 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1186 zfsvfs_t *zfsvfs = arg;
1187 objset_t *os = zfsvfs->z_os;
1189 uint64_t object = lr->lr_foid;
1190 uint64_t offset = lr->lr_offset;
1191 uint64_t size = lr->lr_length;
1192 blkptr_t *bp = &lr->lr_blkptr;
1197 ASSERT(zio != NULL);
1201 * Nothing to do if the file has been removed
1203 if (zfs_zget(zfsvfs, object, &zp) != 0)
1204 return (SET_ERROR(ENOENT));
1205 if (zp->z_unlinked) {
1207 * Release the vnode asynchronously as we currently have the
1208 * txg stopped from syncing.
1210 VN_RELE_ASYNC(ZTOV(zp),
1211 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1212 return (SET_ERROR(ENOENT));
1215 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1216 zgd->zgd_zilog = zfsvfs->z_log;
1217 zgd->zgd_private = zp;
1220 * Write records come in two flavors: immediate and indirect.
1221 * For small writes it's cheaper to store the data with the
1222 * log record (immediate); for large writes it's cheaper to
1223 * sync the data and get a pointer to it (indirect) so that
1224 * we don't have to write the data twice.
1226 if (buf != NULL) { /* immediate write */
1227 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1228 /* test for truncation needs to be done while range locked */
1229 if (offset >= zp->z_size) {
1230 error = SET_ERROR(ENOENT);
1232 error = dmu_read(os, object, offset, size, buf,
1233 DMU_READ_NO_PREFETCH);
1235 ASSERT(error == 0 || error == ENOENT);
1236 } else { /* indirect write */
1238 * Have to lock the whole block to ensure when it's
1239 * written out and it's checksum is being calculated
1240 * that no one can change the data. We need to re-check
1241 * blocksize after we get the lock in case it's changed!
1246 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1248 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1250 if (zp->z_blksz == size)
1253 zfs_range_unlock(zgd->zgd_rl);
1255 /* test for truncation needs to be done while range locked */
1256 if (lr->lr_offset >= zp->z_size)
1257 error = SET_ERROR(ENOENT);
1260 error = SET_ERROR(EIO);
1265 error = dmu_buf_hold(os, object, offset, zgd, &db,
1266 DMU_READ_NO_PREFETCH);
1269 blkptr_t *obp = dmu_buf_get_blkptr(db);
1271 ASSERT(BP_IS_HOLE(bp));
1278 ASSERT(db->db_offset == offset);
1279 ASSERT(db->db_size == size);
1281 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1283 ASSERT(error || lr->lr_length <= zp->z_blksz);
1286 * On success, we need to wait for the write I/O
1287 * initiated by dmu_sync() to complete before we can
1288 * release this dbuf. We will finish everything up
1289 * in the zfs_get_done() callback.
1294 if (error == EALREADY) {
1295 lr->lr_common.lrc_txtype = TX_WRITE2;
1301 zfs_get_done(zgd, error);
1308 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1309 caller_context_t *ct)
1311 znode_t *zp = VTOZ(vp);
1312 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1318 if (flag & V_ACE_MASK)
1319 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1321 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1328 * If vnode is for a device return a specfs vnode instead.
1331 specvp_check(vnode_t **vpp, cred_t *cr)
1335 if (IS_DEVVP(*vpp)) {
1338 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1341 error = SET_ERROR(ENOSYS);
1349 * Lookup an entry in a directory, or an extended attribute directory.
1350 * If it exists, return a held vnode reference for it.
1352 * IN: dvp - vnode of directory to search.
1353 * nm - name of entry to lookup.
1354 * pnp - full pathname to lookup [UNUSED].
1355 * flags - LOOKUP_XATTR set if looking for an attribute.
1356 * rdir - root directory vnode [UNUSED].
1357 * cr - credentials of caller.
1358 * ct - caller context
1359 * direntflags - directory lookup flags
1360 * realpnp - returned pathname.
1362 * OUT: vpp - vnode of located entry, NULL if not found.
1364 * RETURN: 0 on success, error code on failure.
1371 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp,
1372 int nameiop, cred_t *cr, kthread_t *td, int flags)
1374 znode_t *zdp = VTOZ(dvp);
1375 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1377 int *direntflags = NULL;
1378 void *realpnp = NULL;
1381 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1383 if (dvp->v_type != VDIR) {
1384 return (SET_ERROR(ENOTDIR));
1385 } else if (zdp->z_sa_hdl == NULL) {
1386 return (SET_ERROR(EIO));
1389 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1390 error = zfs_fastaccesschk_execute(zdp, cr);
1398 vnode_t *tvp = dnlc_lookup(dvp, nm);
1401 error = zfs_fastaccesschk_execute(zdp, cr);
1406 if (tvp == DNLC_NO_VNODE) {
1408 return (SET_ERROR(ENOENT));
1411 return (specvp_check(vpp, cr));
1417 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1424 if (flags & LOOKUP_XATTR) {
1427 * If the xattr property is off, refuse the lookup request.
1429 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1431 return (SET_ERROR(EINVAL));
1436 * We don't allow recursive attributes..
1437 * Maybe someday we will.
1439 if (zdp->z_pflags & ZFS_XATTR) {
1441 return (SET_ERROR(EINVAL));
1444 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1450 * Do we have permission to get into attribute directory?
1453 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1463 if (dvp->v_type != VDIR) {
1465 return (SET_ERROR(ENOTDIR));
1469 * Check accessibility of directory.
1472 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1477 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1478 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1480 return (SET_ERROR(EILSEQ));
1483 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1485 error = specvp_check(vpp, cr);
1487 /* Translate errors and add SAVENAME when needed. */
1488 if (cnp->cn_flags & ISLASTCN) {
1492 if (error == ENOENT) {
1493 error = EJUSTRETURN;
1494 cnp->cn_flags |= SAVENAME;
1500 cnp->cn_flags |= SAVENAME;
1504 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) {
1507 if (cnp->cn_flags & ISDOTDOT) {
1508 ltype = VOP_ISLOCKED(dvp);
1512 error = vn_lock(*vpp, cnp->cn_lkflags);
1513 if (cnp->cn_flags & ISDOTDOT)
1514 vn_lock(dvp, ltype | LK_RETRY);
1524 #ifdef FREEBSD_NAMECACHE
1526 * Insert name into cache (as non-existent) if appropriate.
1528 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
1529 cache_enter(dvp, *vpp, cnp);
1531 * Insert name into cache if appropriate.
1533 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) {
1534 if (!(cnp->cn_flags & ISLASTCN) ||
1535 (nameiop != DELETE && nameiop != RENAME)) {
1536 cache_enter(dvp, *vpp, cnp);
1545 * Attempt to create a new entry in a directory. If the entry
1546 * already exists, truncate the file if permissible, else return
1547 * an error. Return the vp of the created or trunc'd file.
1549 * IN: dvp - vnode of directory to put new file entry in.
1550 * name - name of new file entry.
1551 * vap - attributes of new file.
1552 * excl - flag indicating exclusive or non-exclusive mode.
1553 * mode - mode to open file with.
1554 * cr - credentials of caller.
1555 * flag - large file flag [UNUSED].
1556 * ct - caller context
1557 * vsecp - ACL to be set
1559 * OUT: vpp - vnode of created or trunc'd entry.
1561 * RETURN: 0 on success, error code on failure.
1564 * dvp - ctime|mtime updated if new entry created
1565 * vp - ctime|mtime always, atime if new
1570 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode,
1571 vnode_t **vpp, cred_t *cr, kthread_t *td)
1573 znode_t *zp, *dzp = VTOZ(dvp);
1574 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1582 gid_t gid = crgetgid(cr);
1583 zfs_acl_ids_t acl_ids;
1584 boolean_t fuid_dirtied;
1585 boolean_t have_acl = B_FALSE;
1590 * If we have an ephemeral id, ACL, or XVATTR then
1591 * make sure file system is at proper version
1594 ksid = crgetsid(cr, KSID_OWNER);
1596 uid = ksid_getid(ksid);
1600 if (zfsvfs->z_use_fuids == B_FALSE &&
1601 (vsecp || (vap->va_mask & AT_XVATTR) ||
1602 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1603 return (SET_ERROR(EINVAL));
1608 zilog = zfsvfs->z_log;
1610 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1611 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1613 return (SET_ERROR(EILSEQ));
1616 if (vap->va_mask & AT_XVATTR) {
1617 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
1618 crgetuid(cr), cr, vap->va_type)) != 0) {
1626 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr))
1627 vap->va_mode &= ~S_ISVTX;
1629 if (*name == '\0') {
1631 * Null component name refers to the directory itself.
1638 /* possible VN_HOLD(zp) */
1641 if (flag & FIGNORECASE)
1644 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1648 zfs_acl_ids_free(&acl_ids);
1649 if (strcmp(name, "..") == 0)
1650 error = SET_ERROR(EISDIR);
1660 * Create a new file object and update the directory
1663 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1665 zfs_acl_ids_free(&acl_ids);
1670 * We only support the creation of regular files in
1671 * extended attribute directories.
1674 if ((dzp->z_pflags & ZFS_XATTR) &&
1675 (vap->va_type != VREG)) {
1677 zfs_acl_ids_free(&acl_ids);
1678 error = SET_ERROR(EINVAL);
1682 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1683 cr, vsecp, &acl_ids)) != 0)
1687 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1688 zfs_acl_ids_free(&acl_ids);
1689 error = SET_ERROR(EDQUOT);
1693 tx = dmu_tx_create(os);
1695 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1696 ZFS_SA_BASE_ATTR_SIZE);
1698 fuid_dirtied = zfsvfs->z_fuid_dirty;
1700 zfs_fuid_txhold(zfsvfs, tx);
1701 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1702 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1703 if (!zfsvfs->z_use_sa &&
1704 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1705 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1706 0, acl_ids.z_aclp->z_acl_bytes);
1708 error = dmu_tx_assign(tx, TXG_NOWAIT);
1710 zfs_dirent_unlock(dl);
1711 if (error == ERESTART) {
1716 zfs_acl_ids_free(&acl_ids);
1721 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1724 zfs_fuid_sync(zfsvfs, tx);
1726 (void) zfs_link_create(dl, zp, tx, ZNEW);
1727 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1728 if (flag & FIGNORECASE)
1730 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1731 vsecp, acl_ids.z_fuidp, vap);
1732 zfs_acl_ids_free(&acl_ids);
1735 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1738 zfs_acl_ids_free(&acl_ids);
1742 * A directory entry already exists for this name.
1745 * Can't truncate an existing file if in exclusive mode.
1748 error = SET_ERROR(EEXIST);
1752 * Can't open a directory for writing.
1754 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1755 error = SET_ERROR(EISDIR);
1759 * Verify requested access to file.
1761 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1765 mutex_enter(&dzp->z_lock);
1767 mutex_exit(&dzp->z_lock);
1770 * Truncate regular files if requested.
1772 if ((ZTOV(zp)->v_type == VREG) &&
1773 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1774 /* we can't hold any locks when calling zfs_freesp() */
1775 zfs_dirent_unlock(dl);
1777 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1779 vnevent_create(ZTOV(zp), ct);
1785 zfs_dirent_unlock(dl);
1792 error = specvp_check(vpp, cr);
1795 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1796 zil_commit(zilog, 0);
1803 * Remove an entry from a directory.
1805 * IN: dvp - vnode of directory to remove entry from.
1806 * name - name of entry to remove.
1807 * cr - credentials of caller.
1808 * ct - caller context
1809 * flags - case flags
1811 * RETURN: 0 on success, error code on failure.
1815 * vp - ctime (if nlink > 0)
1818 uint64_t null_xattr = 0;
1822 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1825 znode_t *zp, *dzp = VTOZ(dvp);
1828 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1830 uint64_t acl_obj, xattr_obj;
1831 uint64_t xattr_obj_unlinked = 0;
1835 boolean_t may_delete_now, delete_now = FALSE;
1836 boolean_t unlinked, toobig = FALSE;
1838 pathname_t *realnmp = NULL;
1845 zilog = zfsvfs->z_log;
1847 if (flags & FIGNORECASE) {
1857 * Attempt to lock directory; fail if entry doesn't exist.
1859 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1869 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1874 * Need to use rmdir for removing directories.
1876 if (vp->v_type == VDIR) {
1877 error = SET_ERROR(EPERM);
1881 vnevent_remove(vp, dvp, name, ct);
1884 dnlc_remove(dvp, realnmp->pn_buf);
1886 dnlc_remove(dvp, name);
1889 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1893 * We may delete the znode now, or we may put it in the unlinked set;
1894 * it depends on whether we're the last link, and on whether there are
1895 * other holds on the vnode. So we dmu_tx_hold() the right things to
1896 * allow for either case.
1899 tx = dmu_tx_create(zfsvfs->z_os);
1900 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1901 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1902 zfs_sa_upgrade_txholds(tx, zp);
1903 zfs_sa_upgrade_txholds(tx, dzp);
1904 if (may_delete_now) {
1906 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1907 /* if the file is too big, only hold_free a token amount */
1908 dmu_tx_hold_free(tx, zp->z_id, 0,
1909 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1912 /* are there any extended attributes? */
1913 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1914 &xattr_obj, sizeof (xattr_obj));
1915 if (error == 0 && xattr_obj) {
1916 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1918 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1919 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1922 mutex_enter(&zp->z_lock);
1923 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1924 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1925 mutex_exit(&zp->z_lock);
1927 /* charge as an update -- would be nice not to charge at all */
1928 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1930 error = dmu_tx_assign(tx, TXG_NOWAIT);
1932 zfs_dirent_unlock(dl);
1936 if (error == ERESTART) {
1949 * Remove the directory entry.
1951 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1961 * Hold z_lock so that we can make sure that the ACL obj
1962 * hasn't changed. Could have been deleted due to
1965 mutex_enter(&zp->z_lock);
1967 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1968 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1969 delete_now = may_delete_now && !toobig &&
1970 vp->v_count == 1 && !vn_has_cached_data(vp) &&
1971 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1978 panic("zfs_remove: delete_now branch taken");
1980 if (xattr_obj_unlinked) {
1981 ASSERT3U(xzp->z_links, ==, 2);
1982 mutex_enter(&xzp->z_lock);
1983 xzp->z_unlinked = 1;
1985 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1986 &xzp->z_links, sizeof (xzp->z_links), tx);
1987 ASSERT3U(error, ==, 0);
1988 mutex_exit(&xzp->z_lock);
1989 zfs_unlinked_add(xzp, tx);
1992 error = sa_remove(zp->z_sa_hdl,
1993 SA_ZPL_XATTR(zfsvfs), tx);
1995 error = sa_update(zp->z_sa_hdl,
1996 SA_ZPL_XATTR(zfsvfs), &null_xattr,
1997 sizeof (uint64_t), tx);
2002 ASSERT0(vp->v_count);
2004 mutex_exit(&zp->z_lock);
2005 zfs_znode_delete(zp, tx);
2006 } else if (unlinked) {
2007 mutex_exit(&zp->z_lock);
2008 zfs_unlinked_add(zp, tx);
2010 vp->v_vflag |= VV_NOSYNC;
2015 if (flags & FIGNORECASE)
2017 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
2024 zfs_dirent_unlock(dl);
2031 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2032 zil_commit(zilog, 0);
2039 * Create a new directory and insert it into dvp using the name
2040 * provided. Return a pointer to the inserted directory.
2042 * IN: dvp - vnode of directory to add subdir to.
2043 * dirname - name of new directory.
2044 * vap - attributes of new directory.
2045 * cr - credentials of caller.
2046 * ct - caller context
2047 * flags - case flags
2048 * vsecp - ACL to be set
2050 * OUT: vpp - vnode of created directory.
2052 * RETURN: 0 on success, error code on failure.
2055 * dvp - ctime|mtime updated
2056 * vp - ctime|mtime|atime updated
2060 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
2061 caller_context_t *ct, int flags, vsecattr_t *vsecp)
2063 znode_t *zp, *dzp = VTOZ(dvp);
2064 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2073 gid_t gid = crgetgid(cr);
2074 zfs_acl_ids_t acl_ids;
2075 boolean_t fuid_dirtied;
2077 ASSERT(vap->va_type == VDIR);
2080 * If we have an ephemeral id, ACL, or XVATTR then
2081 * make sure file system is at proper version
2084 ksid = crgetsid(cr, KSID_OWNER);
2086 uid = ksid_getid(ksid);
2089 if (zfsvfs->z_use_fuids == B_FALSE &&
2090 (vsecp || (vap->va_mask & AT_XVATTR) ||
2091 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
2092 return (SET_ERROR(EINVAL));
2096 zilog = zfsvfs->z_log;
2098 if (dzp->z_pflags & ZFS_XATTR) {
2100 return (SET_ERROR(EINVAL));
2103 if (zfsvfs->z_utf8 && u8_validate(dirname,
2104 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2106 return (SET_ERROR(EILSEQ));
2108 if (flags & FIGNORECASE)
2111 if (vap->va_mask & AT_XVATTR) {
2112 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
2113 crgetuid(cr), cr, vap->va_type)) != 0) {
2119 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
2120 vsecp, &acl_ids)) != 0) {
2125 * First make sure the new directory doesn't exist.
2127 * Existence is checked first to make sure we don't return
2128 * EACCES instead of EEXIST which can cause some applications
2134 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
2136 zfs_acl_ids_free(&acl_ids);
2141 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
2142 zfs_acl_ids_free(&acl_ids);
2143 zfs_dirent_unlock(dl);
2148 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
2149 zfs_acl_ids_free(&acl_ids);
2150 zfs_dirent_unlock(dl);
2152 return (SET_ERROR(EDQUOT));
2156 * Add a new entry to the directory.
2158 tx = dmu_tx_create(zfsvfs->z_os);
2159 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2160 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2161 fuid_dirtied = zfsvfs->z_fuid_dirty;
2163 zfs_fuid_txhold(zfsvfs, tx);
2164 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2165 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2166 acl_ids.z_aclp->z_acl_bytes);
2169 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2170 ZFS_SA_BASE_ATTR_SIZE);
2172 error = dmu_tx_assign(tx, TXG_NOWAIT);
2174 zfs_dirent_unlock(dl);
2175 if (error == ERESTART) {
2180 zfs_acl_ids_free(&acl_ids);
2189 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2192 zfs_fuid_sync(zfsvfs, tx);
2195 * Now put new name in parent dir.
2197 (void) zfs_link_create(dl, zp, tx, ZNEW);
2201 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2202 if (flags & FIGNORECASE)
2204 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2205 acl_ids.z_fuidp, vap);
2207 zfs_acl_ids_free(&acl_ids);
2211 zfs_dirent_unlock(dl);
2213 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2214 zil_commit(zilog, 0);
2221 * Remove a directory subdir entry. If the current working
2222 * directory is the same as the subdir to be removed, the
2225 * IN: dvp - vnode of directory to remove from.
2226 * name - name of directory to be removed.
2227 * cwd - vnode of current working directory.
2228 * cr - credentials of caller.
2229 * ct - caller context
2230 * flags - case flags
2232 * RETURN: 0 on success, error code on failure.
2235 * dvp - ctime|mtime updated
2239 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2240 caller_context_t *ct, int flags)
2242 znode_t *dzp = VTOZ(dvp);
2245 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2254 zilog = zfsvfs->z_log;
2256 if (flags & FIGNORECASE)
2262 * Attempt to lock directory; fail if entry doesn't exist.
2264 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2272 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2276 if (vp->v_type != VDIR) {
2277 error = SET_ERROR(ENOTDIR);
2282 error = SET_ERROR(EINVAL);
2286 vnevent_rmdir(vp, dvp, name, ct);
2289 * Grab a lock on the directory to make sure that noone is
2290 * trying to add (or lookup) entries while we are removing it.
2292 rw_enter(&zp->z_name_lock, RW_WRITER);
2295 * Grab a lock on the parent pointer to make sure we play well
2296 * with the treewalk and directory rename code.
2298 rw_enter(&zp->z_parent_lock, RW_WRITER);
2300 tx = dmu_tx_create(zfsvfs->z_os);
2301 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2302 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2303 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2304 zfs_sa_upgrade_txholds(tx, zp);
2305 zfs_sa_upgrade_txholds(tx, dzp);
2306 error = dmu_tx_assign(tx, TXG_NOWAIT);
2308 rw_exit(&zp->z_parent_lock);
2309 rw_exit(&zp->z_name_lock);
2310 zfs_dirent_unlock(dl);
2312 if (error == ERESTART) {
2322 #ifdef FREEBSD_NAMECACHE
2326 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2329 uint64_t txtype = TX_RMDIR;
2330 if (flags & FIGNORECASE)
2332 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2337 rw_exit(&zp->z_parent_lock);
2338 rw_exit(&zp->z_name_lock);
2339 #ifdef FREEBSD_NAMECACHE
2343 zfs_dirent_unlock(dl);
2347 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2348 zil_commit(zilog, 0);
2355 * Read as many directory entries as will fit into the provided
2356 * buffer from the given directory cursor position (specified in
2357 * the uio structure).
2359 * IN: vp - vnode of directory to read.
2360 * uio - structure supplying read location, range info,
2361 * and return buffer.
2362 * cr - credentials of caller.
2363 * ct - caller context
2364 * flags - case flags
2366 * OUT: uio - updated offset and range, buffer filled.
2367 * eofp - set to true if end-of-file detected.
2369 * RETURN: 0 on success, error code on failure.
2372 * vp - atime updated
2374 * Note that the low 4 bits of the cookie returned by zap is always zero.
2375 * This allows us to use the low range for "special" directory entries:
2376 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2377 * we use the offset 2 for the '.zfs' directory.
2381 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies)
2383 znode_t *zp = VTOZ(vp);
2387 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2392 zap_attribute_t zap;
2393 uint_t bytes_wanted;
2394 uint64_t offset; /* must be unsigned; checks for < 1 */
2400 boolean_t check_sysattrs;
2403 u_long *cooks = NULL;
2409 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2410 &parent, sizeof (parent))) != 0) {
2416 * If we are not given an eof variable,
2423 * Check for valid iov_len.
2425 if (uio->uio_iov->iov_len <= 0) {
2427 return (SET_ERROR(EINVAL));
2431 * Quit if directory has been removed (posix)
2433 if ((*eofp = zp->z_unlinked) != 0) {
2440 offset = uio->uio_loffset;
2441 prefetch = zp->z_zn_prefetch;
2444 * Initialize the iterator cursor.
2448 * Start iteration from the beginning of the directory.
2450 zap_cursor_init(&zc, os, zp->z_id);
2453 * The offset is a serialized cursor.
2455 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2459 * Get space to change directory entries into fs independent format.
2461 iovp = uio->uio_iov;
2462 bytes_wanted = iovp->iov_len;
2463 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2464 bufsize = bytes_wanted;
2465 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2466 odp = (struct dirent64 *)outbuf;
2468 bufsize = bytes_wanted;
2470 odp = (struct dirent64 *)iovp->iov_base;
2472 eodp = (struct edirent *)odp;
2474 if (ncookies != NULL) {
2476 * Minimum entry size is dirent size and 1 byte for a file name.
2478 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1);
2479 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK);
2484 * If this VFS supports the system attribute view interface; and
2485 * we're looking at an extended attribute directory; and we care
2486 * about normalization conflicts on this vfs; then we must check
2487 * for normalization conflicts with the sysattr name space.
2490 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2491 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2492 (flags & V_RDDIR_ENTFLAGS);
2498 * Transform to file-system independent format
2501 while (outcount < bytes_wanted) {
2504 off64_t *next = NULL;
2507 * Special case `.', `..', and `.zfs'.
2510 (void) strcpy(zap.za_name, ".");
2511 zap.za_normalization_conflict = 0;
2514 } else if (offset == 1) {
2515 (void) strcpy(zap.za_name, "..");
2516 zap.za_normalization_conflict = 0;
2519 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2520 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2521 zap.za_normalization_conflict = 0;
2522 objnum = ZFSCTL_INO_ROOT;
2528 if (error = zap_cursor_retrieve(&zc, &zap)) {
2529 if ((*eofp = (error == ENOENT)) != 0)
2535 if (zap.za_integer_length != 8 ||
2536 zap.za_num_integers != 1) {
2537 cmn_err(CE_WARN, "zap_readdir: bad directory "
2538 "entry, obj = %lld, offset = %lld\n",
2539 (u_longlong_t)zp->z_id,
2540 (u_longlong_t)offset);
2541 error = SET_ERROR(ENXIO);
2545 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2547 * MacOS X can extract the object type here such as:
2548 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2550 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2552 if (check_sysattrs && !zap.za_normalization_conflict) {
2554 zap.za_normalization_conflict =
2555 xattr_sysattr_casechk(zap.za_name);
2557 panic("%s:%u: TODO", __func__, __LINE__);
2562 if (flags & V_RDDIR_ACCFILTER) {
2564 * If we have no access at all, don't include
2565 * this entry in the returned information
2568 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2570 if (!zfs_has_access(ezp, cr)) {
2577 if (flags & V_RDDIR_ENTFLAGS)
2578 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2580 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2583 * Will this entry fit in the buffer?
2585 if (outcount + reclen > bufsize) {
2587 * Did we manage to fit anything in the buffer?
2590 error = SET_ERROR(EINVAL);
2595 if (flags & V_RDDIR_ENTFLAGS) {
2597 * Add extended flag entry:
2599 eodp->ed_ino = objnum;
2600 eodp->ed_reclen = reclen;
2601 /* NOTE: ed_off is the offset for the *next* entry */
2602 next = &(eodp->ed_off);
2603 eodp->ed_eflags = zap.za_normalization_conflict ?
2604 ED_CASE_CONFLICT : 0;
2605 (void) strncpy(eodp->ed_name, zap.za_name,
2606 EDIRENT_NAMELEN(reclen));
2607 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2612 odp->d_ino = objnum;
2613 odp->d_reclen = reclen;
2614 odp->d_namlen = strlen(zap.za_name);
2615 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1);
2617 odp = (dirent64_t *)((intptr_t)odp + reclen);
2621 ASSERT(outcount <= bufsize);
2623 /* Prefetch znode */
2625 dmu_prefetch(os, objnum, 0, 0);
2629 * Move to the next entry, fill in the previous offset.
2631 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2632 zap_cursor_advance(&zc);
2633 offset = zap_cursor_serialize(&zc);
2638 if (cooks != NULL) {
2641 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks));
2644 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2646 /* Subtract unused cookies */
2647 if (ncookies != NULL)
2648 *ncookies -= ncooks;
2650 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2651 iovp->iov_base += outcount;
2652 iovp->iov_len -= outcount;
2653 uio->uio_resid -= outcount;
2654 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2656 * Reset the pointer.
2658 offset = uio->uio_loffset;
2662 zap_cursor_fini(&zc);
2663 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2664 kmem_free(outbuf, bufsize);
2666 if (error == ENOENT)
2669 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2671 uio->uio_loffset = offset;
2673 if (error != 0 && cookies != NULL) {
2674 free(*cookies, M_TEMP);
2681 ulong_t zfs_fsync_sync_cnt = 4;
2684 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2686 znode_t *zp = VTOZ(vp);
2687 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2689 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2691 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2694 zil_commit(zfsvfs->z_log, zp->z_id);
2702 * Get the requested file attributes and place them in the provided
2705 * IN: vp - vnode of file.
2706 * vap - va_mask identifies requested attributes.
2707 * If AT_XVATTR set, then optional attrs are requested
2708 * flags - ATTR_NOACLCHECK (CIFS server context)
2709 * cr - credentials of caller.
2710 * ct - caller context
2712 * OUT: vap - attribute values.
2714 * RETURN: 0 (always succeeds).
2718 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2719 caller_context_t *ct)
2721 znode_t *zp = VTOZ(vp);
2722 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2725 u_longlong_t nblocks;
2727 uint64_t mtime[2], ctime[2], crtime[2], rdev;
2728 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2729 xoptattr_t *xoap = NULL;
2730 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2731 sa_bulk_attr_t bulk[4];
2737 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2739 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2740 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2741 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
2742 if (vp->v_type == VBLK || vp->v_type == VCHR)
2743 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
2746 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2752 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2753 * Also, if we are the owner don't bother, since owner should
2754 * always be allowed to read basic attributes of file.
2756 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2757 (vap->va_uid != crgetuid(cr))) {
2758 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2766 * Return all attributes. It's cheaper to provide the answer
2767 * than to determine whether we were asked the question.
2770 mutex_enter(&zp->z_lock);
2771 vap->va_type = IFTOVT(zp->z_mode);
2772 vap->va_mode = zp->z_mode & ~S_IFMT;
2774 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2776 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
2778 vap->va_nodeid = zp->z_id;
2779 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2780 links = zp->z_links + 1;
2782 links = zp->z_links;
2783 vap->va_nlink = MIN(links, UINT32_MAX); /* nlink_t limit! */
2784 vap->va_size = zp->z_size;
2786 vap->va_rdev = vp->v_rdev;
2788 if (vp->v_type == VBLK || vp->v_type == VCHR)
2789 vap->va_rdev = zfs_cmpldev(rdev);
2791 vap->va_seq = zp->z_seq;
2792 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */
2795 * Add in any requested optional attributes and the create time.
2796 * Also set the corresponding bits in the returned attribute bitmap.
2798 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2799 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2801 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2802 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2805 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2806 xoap->xoa_readonly =
2807 ((zp->z_pflags & ZFS_READONLY) != 0);
2808 XVA_SET_RTN(xvap, XAT_READONLY);
2811 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2813 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2814 XVA_SET_RTN(xvap, XAT_SYSTEM);
2817 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2819 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2820 XVA_SET_RTN(xvap, XAT_HIDDEN);
2823 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2824 xoap->xoa_nounlink =
2825 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2826 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2829 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2830 xoap->xoa_immutable =
2831 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2832 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2835 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2836 xoap->xoa_appendonly =
2837 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2838 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2841 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2843 ((zp->z_pflags & ZFS_NODUMP) != 0);
2844 XVA_SET_RTN(xvap, XAT_NODUMP);
2847 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2849 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2850 XVA_SET_RTN(xvap, XAT_OPAQUE);
2853 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2854 xoap->xoa_av_quarantined =
2855 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2856 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2859 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2860 xoap->xoa_av_modified =
2861 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2862 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2865 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2866 vp->v_type == VREG) {
2867 zfs_sa_get_scanstamp(zp, xvap);
2870 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2873 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2874 times, sizeof (times));
2875 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2876 XVA_SET_RTN(xvap, XAT_CREATETIME);
2879 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2880 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2881 XVA_SET_RTN(xvap, XAT_REPARSE);
2883 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2884 xoap->xoa_generation = zp->z_gen;
2885 XVA_SET_RTN(xvap, XAT_GEN);
2888 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2890 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2891 XVA_SET_RTN(xvap, XAT_OFFLINE);
2894 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2896 ((zp->z_pflags & ZFS_SPARSE) != 0);
2897 XVA_SET_RTN(xvap, XAT_SPARSE);
2901 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2902 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2903 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2904 ZFS_TIME_DECODE(&vap->va_birthtime, crtime);
2906 mutex_exit(&zp->z_lock);
2908 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2909 vap->va_blksize = blksize;
2910 vap->va_bytes = nblocks << 9; /* nblocks * 512 */
2912 if (zp->z_blksz == 0) {
2914 * Block size hasn't been set; suggest maximal I/O transfers.
2916 vap->va_blksize = zfsvfs->z_max_blksz;
2924 * Set the file attributes to the values contained in the
2927 * IN: vp - vnode of file to be modified.
2928 * vap - new attribute values.
2929 * If AT_XVATTR set, then optional attrs are being set
2930 * flags - ATTR_UTIME set if non-default time values provided.
2931 * - ATTR_NOACLCHECK (CIFS context only).
2932 * cr - credentials of caller.
2933 * ct - caller context
2935 * RETURN: 0 on success, error code on failure.
2938 * vp - ctime updated, mtime updated if size changed.
2942 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2943 caller_context_t *ct)
2945 znode_t *zp = VTOZ(vp);
2946 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2951 uint_t mask = vap->va_mask;
2952 uint_t saved_mask = 0;
2953 uint64_t saved_mode;
2956 uint64_t new_uid, new_gid;
2958 uint64_t mtime[2], ctime[2];
2960 int need_policy = FALSE;
2962 zfs_fuid_info_t *fuidp = NULL;
2963 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2966 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2967 boolean_t fuid_dirtied = B_FALSE;
2968 sa_bulk_attr_t bulk[7], xattr_bulk[7];
2969 int count = 0, xattr_count = 0;
2974 if (mask & AT_NOSET)
2975 return (SET_ERROR(EINVAL));
2980 zilog = zfsvfs->z_log;
2983 * Make sure that if we have ephemeral uid/gid or xvattr specified
2984 * that file system is at proper version level
2987 if (zfsvfs->z_use_fuids == B_FALSE &&
2988 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2989 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2990 (mask & AT_XVATTR))) {
2992 return (SET_ERROR(EINVAL));
2995 if (mask & AT_SIZE && vp->v_type == VDIR) {
2997 return (SET_ERROR(EISDIR));
3000 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
3002 return (SET_ERROR(EINVAL));
3006 * If this is an xvattr_t, then get a pointer to the structure of
3007 * optional attributes. If this is NULL, then we have a vattr_t.
3009 xoap = xva_getxoptattr(xvap);
3011 xva_init(&tmpxvattr);
3014 * Immutable files can only alter immutable bit and atime
3016 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
3017 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
3018 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
3020 return (SET_ERROR(EPERM));
3023 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
3025 return (SET_ERROR(EPERM));
3029 * Verify timestamps doesn't overflow 32 bits.
3030 * ZFS can handle large timestamps, but 32bit syscalls can't
3031 * handle times greater than 2039. This check should be removed
3032 * once large timestamps are fully supported.
3034 if (mask & (AT_ATIME | AT_MTIME)) {
3035 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
3036 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
3038 return (SET_ERROR(EOVERFLOW));
3046 /* Can this be moved to before the top label? */
3047 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
3049 return (SET_ERROR(EROFS));
3053 * First validate permissions
3056 if (mask & AT_SIZE) {
3058 * XXX - Note, we are not providing any open
3059 * mode flags here (like FNDELAY), so we may
3060 * block if there are locks present... this
3061 * should be addressed in openat().
3063 /* XXX - would it be OK to generate a log record here? */
3064 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
3071 if (mask & (AT_ATIME|AT_MTIME) ||
3072 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
3073 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
3074 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
3075 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
3076 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
3077 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
3078 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
3079 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
3083 if (mask & (AT_UID|AT_GID)) {
3084 int idmask = (mask & (AT_UID|AT_GID));
3089 * NOTE: even if a new mode is being set,
3090 * we may clear S_ISUID/S_ISGID bits.
3093 if (!(mask & AT_MODE))
3094 vap->va_mode = zp->z_mode;
3097 * Take ownership or chgrp to group we are a member of
3100 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
3101 take_group = (mask & AT_GID) &&
3102 zfs_groupmember(zfsvfs, vap->va_gid, cr);
3105 * If both AT_UID and AT_GID are set then take_owner and
3106 * take_group must both be set in order to allow taking
3109 * Otherwise, send the check through secpolicy_vnode_setattr()
3113 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
3114 ((idmask == AT_UID) && take_owner) ||
3115 ((idmask == AT_GID) && take_group)) {
3116 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
3117 skipaclchk, cr) == 0) {
3119 * Remove setuid/setgid for non-privileged users
3121 secpolicy_setid_clear(vap, vp, cr);
3122 trim_mask = (mask & (AT_UID|AT_GID));
3131 mutex_enter(&zp->z_lock);
3132 oldva.va_mode = zp->z_mode;
3133 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
3134 if (mask & AT_XVATTR) {
3136 * Update xvattr mask to include only those attributes
3137 * that are actually changing.
3139 * the bits will be restored prior to actually setting
3140 * the attributes so the caller thinks they were set.
3142 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
3143 if (xoap->xoa_appendonly !=
3144 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
3147 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
3148 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
3152 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
3153 if (xoap->xoa_nounlink !=
3154 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
3157 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
3158 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
3162 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
3163 if (xoap->xoa_immutable !=
3164 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
3167 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
3168 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
3172 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
3173 if (xoap->xoa_nodump !=
3174 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
3177 XVA_CLR_REQ(xvap, XAT_NODUMP);
3178 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
3182 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3183 if (xoap->xoa_av_modified !=
3184 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3187 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3188 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
3192 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3193 if ((vp->v_type != VREG &&
3194 xoap->xoa_av_quarantined) ||
3195 xoap->xoa_av_quarantined !=
3196 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3199 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3200 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
3204 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3205 mutex_exit(&zp->z_lock);
3207 return (SET_ERROR(EPERM));
3210 if (need_policy == FALSE &&
3211 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3212 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3217 mutex_exit(&zp->z_lock);
3219 if (mask & AT_MODE) {
3220 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3221 err = secpolicy_setid_setsticky_clear(vp, vap,
3227 trim_mask |= AT_MODE;
3235 * If trim_mask is set then take ownership
3236 * has been granted or write_acl is present and user
3237 * has the ability to modify mode. In that case remove
3238 * UID|GID and or MODE from mask so that
3239 * secpolicy_vnode_setattr() doesn't revoke it.
3243 saved_mask = vap->va_mask;
3244 vap->va_mask &= ~trim_mask;
3245 if (trim_mask & AT_MODE) {
3247 * Save the mode, as secpolicy_vnode_setattr()
3248 * will overwrite it with ova.va_mode.
3250 saved_mode = vap->va_mode;
3253 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3254 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3261 vap->va_mask |= saved_mask;
3262 if (trim_mask & AT_MODE) {
3264 * Recover the mode after
3265 * secpolicy_vnode_setattr().
3267 vap->va_mode = saved_mode;
3273 * secpolicy_vnode_setattr, or take ownership may have
3276 mask = vap->va_mask;
3278 if ((mask & (AT_UID | AT_GID))) {
3279 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3280 &xattr_obj, sizeof (xattr_obj));
3282 if (err == 0 && xattr_obj) {
3283 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3287 if (mask & AT_UID) {
3288 new_uid = zfs_fuid_create(zfsvfs,
3289 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3290 if (new_uid != zp->z_uid &&
3291 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3293 VN_RELE(ZTOV(attrzp));
3294 err = SET_ERROR(EDQUOT);
3299 if (mask & AT_GID) {
3300 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3301 cr, ZFS_GROUP, &fuidp);
3302 if (new_gid != zp->z_gid &&
3303 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3305 VN_RELE(ZTOV(attrzp));
3306 err = SET_ERROR(EDQUOT);
3311 tx = dmu_tx_create(zfsvfs->z_os);
3313 if (mask & AT_MODE) {
3314 uint64_t pmode = zp->z_mode;
3316 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3318 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3319 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3320 err = SET_ERROR(EPERM);
3324 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3327 mutex_enter(&zp->z_lock);
3328 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3330 * Are we upgrading ACL from old V0 format
3333 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3334 zfs_znode_acl_version(zp) ==
3335 ZFS_ACL_VERSION_INITIAL) {
3336 dmu_tx_hold_free(tx, acl_obj, 0,
3338 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3339 0, aclp->z_acl_bytes);
3341 dmu_tx_hold_write(tx, acl_obj, 0,
3344 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3345 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3346 0, aclp->z_acl_bytes);
3348 mutex_exit(&zp->z_lock);
3349 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3351 if ((mask & AT_XVATTR) &&
3352 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3353 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3355 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3359 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3362 fuid_dirtied = zfsvfs->z_fuid_dirty;
3364 zfs_fuid_txhold(zfsvfs, tx);
3366 zfs_sa_upgrade_txholds(tx, zp);
3368 err = dmu_tx_assign(tx, TXG_NOWAIT);
3370 if (err == ERESTART)
3377 * Set each attribute requested.
3378 * We group settings according to the locks they need to acquire.
3380 * Note: you cannot set ctime directly, although it will be
3381 * updated as a side-effect of calling this function.
3385 if (mask & (AT_UID|AT_GID|AT_MODE))
3386 mutex_enter(&zp->z_acl_lock);
3387 mutex_enter(&zp->z_lock);
3389 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3390 &zp->z_pflags, sizeof (zp->z_pflags));
3393 if (mask & (AT_UID|AT_GID|AT_MODE))
3394 mutex_enter(&attrzp->z_acl_lock);
3395 mutex_enter(&attrzp->z_lock);
3396 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3397 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3398 sizeof (attrzp->z_pflags));
3401 if (mask & (AT_UID|AT_GID)) {
3403 if (mask & AT_UID) {
3404 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3405 &new_uid, sizeof (new_uid));
3406 zp->z_uid = new_uid;
3408 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3409 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3411 attrzp->z_uid = new_uid;
3415 if (mask & AT_GID) {
3416 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3417 NULL, &new_gid, sizeof (new_gid));
3418 zp->z_gid = new_gid;
3420 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3421 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3423 attrzp->z_gid = new_gid;
3426 if (!(mask & AT_MODE)) {
3427 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3428 NULL, &new_mode, sizeof (new_mode));
3429 new_mode = zp->z_mode;
3431 err = zfs_acl_chown_setattr(zp);
3434 err = zfs_acl_chown_setattr(attrzp);
3439 if (mask & AT_MODE) {
3440 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3441 &new_mode, sizeof (new_mode));
3442 zp->z_mode = new_mode;
3443 ASSERT3U((uintptr_t)aclp, !=, 0);
3444 err = zfs_aclset_common(zp, aclp, cr, tx);
3446 if (zp->z_acl_cached)
3447 zfs_acl_free(zp->z_acl_cached);
3448 zp->z_acl_cached = aclp;
3453 if (mask & AT_ATIME) {
3454 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3455 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3456 &zp->z_atime, sizeof (zp->z_atime));
3459 if (mask & AT_MTIME) {
3460 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3461 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3462 mtime, sizeof (mtime));
3465 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3466 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3467 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3468 NULL, mtime, sizeof (mtime));
3469 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3470 &ctime, sizeof (ctime));
3471 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3473 } else if (mask != 0) {
3474 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3475 &ctime, sizeof (ctime));
3476 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3479 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3480 SA_ZPL_CTIME(zfsvfs), NULL,
3481 &ctime, sizeof (ctime));
3482 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3483 mtime, ctime, B_TRUE);
3487 * Do this after setting timestamps to prevent timestamp
3488 * update from toggling bit
3491 if (xoap && (mask & AT_XVATTR)) {
3494 * restore trimmed off masks
3495 * so that return masks can be set for caller.
3498 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3499 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3501 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3502 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3504 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3505 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3507 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3508 XVA_SET_REQ(xvap, XAT_NODUMP);
3510 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3511 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3513 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3514 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3517 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3518 ASSERT(vp->v_type == VREG);
3520 zfs_xvattr_set(zp, xvap, tx);
3524 zfs_fuid_sync(zfsvfs, tx);
3527 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3529 mutex_exit(&zp->z_lock);
3530 if (mask & (AT_UID|AT_GID|AT_MODE))
3531 mutex_exit(&zp->z_acl_lock);
3534 if (mask & (AT_UID|AT_GID|AT_MODE))
3535 mutex_exit(&attrzp->z_acl_lock);
3536 mutex_exit(&attrzp->z_lock);
3539 if (err == 0 && attrzp) {
3540 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3546 VN_RELE(ZTOV(attrzp));
3552 zfs_fuid_info_free(fuidp);
3558 if (err == ERESTART)
3561 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3566 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3567 zil_commit(zilog, 0);
3573 typedef struct zfs_zlock {
3574 krwlock_t *zl_rwlock; /* lock we acquired */
3575 znode_t *zl_znode; /* znode we held */
3576 struct zfs_zlock *zl_next; /* next in list */
3580 * Drop locks and release vnodes that were held by zfs_rename_lock().
3583 zfs_rename_unlock(zfs_zlock_t **zlpp)
3587 while ((zl = *zlpp) != NULL) {
3588 if (zl->zl_znode != NULL)
3589 VN_RELE(ZTOV(zl->zl_znode));
3590 rw_exit(zl->zl_rwlock);
3591 *zlpp = zl->zl_next;
3592 kmem_free(zl, sizeof (*zl));
3597 * Search back through the directory tree, using the ".." entries.
3598 * Lock each directory in the chain to prevent concurrent renames.
3599 * Fail any attempt to move a directory into one of its own descendants.
3600 * XXX - z_parent_lock can overlap with map or grow locks
3603 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3607 uint64_t rootid = zp->z_zfsvfs->z_root;
3608 uint64_t oidp = zp->z_id;
3609 krwlock_t *rwlp = &szp->z_parent_lock;
3610 krw_t rw = RW_WRITER;
3613 * First pass write-locks szp and compares to zp->z_id.
3614 * Later passes read-lock zp and compare to zp->z_parent.
3617 if (!rw_tryenter(rwlp, rw)) {
3619 * Another thread is renaming in this path.
3620 * Note that if we are a WRITER, we don't have any
3621 * parent_locks held yet.
3623 if (rw == RW_READER && zp->z_id > szp->z_id) {
3625 * Drop our locks and restart
3627 zfs_rename_unlock(&zl);
3631 rwlp = &szp->z_parent_lock;
3636 * Wait for other thread to drop its locks
3642 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3643 zl->zl_rwlock = rwlp;
3644 zl->zl_znode = NULL;
3645 zl->zl_next = *zlpp;
3648 if (oidp == szp->z_id) /* We're a descendant of szp */
3649 return (SET_ERROR(EINVAL));
3651 if (oidp == rootid) /* We've hit the top */
3654 if (rw == RW_READER) { /* i.e. not the first pass */
3655 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3660 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3661 &oidp, sizeof (oidp));
3662 rwlp = &zp->z_parent_lock;
3665 } while (zp->z_id != sdzp->z_id);
3671 * Move an entry from the provided source directory to the target
3672 * directory. Change the entry name as indicated.
3674 * IN: sdvp - Source directory containing the "old entry".
3675 * snm - Old entry name.
3676 * tdvp - Target directory to contain the "new entry".
3677 * tnm - New entry name.
3678 * cr - credentials of caller.
3679 * ct - caller context
3680 * flags - case flags
3682 * RETURN: 0 on success, error code on failure.
3685 * sdvp,tdvp - ctime|mtime updated
3689 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3690 caller_context_t *ct, int flags)
3692 znode_t *tdzp, *szp, *tzp;
3693 znode_t *sdzp = VTOZ(sdvp);
3694 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs;
3697 zfs_dirlock_t *sdl, *tdl;
3700 int cmp, serr, terr;
3705 ZFS_VERIFY_ZP(sdzp);
3706 zilog = zfsvfs->z_log;
3709 * Make sure we have the real vp for the target directory.
3711 if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3714 if (tdvp->v_vfsp != sdvp->v_vfsp || zfsctl_is_node(tdvp)) {
3716 return (SET_ERROR(EXDEV));
3720 ZFS_VERIFY_ZP(tdzp);
3721 if (zfsvfs->z_utf8 && u8_validate(tnm,
3722 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3724 return (SET_ERROR(EILSEQ));
3727 if (flags & FIGNORECASE)
3736 * This is to prevent the creation of links into attribute space
3737 * by renaming a linked file into/outof an attribute directory.
3738 * See the comment in zfs_link() for why this is considered bad.
3740 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3742 return (SET_ERROR(EINVAL));
3746 * Lock source and target directory entries. To prevent deadlock,
3747 * a lock ordering must be defined. We lock the directory with
3748 * the smallest object id first, or if it's a tie, the one with
3749 * the lexically first name.
3751 if (sdzp->z_id < tdzp->z_id) {
3753 } else if (sdzp->z_id > tdzp->z_id) {
3757 * First compare the two name arguments without
3758 * considering any case folding.
3760 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3762 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3763 ASSERT(error == 0 || !zfsvfs->z_utf8);
3766 * POSIX: "If the old argument and the new argument
3767 * both refer to links to the same existing file,
3768 * the rename() function shall return successfully
3769 * and perform no other action."
3775 * If the file system is case-folding, then we may
3776 * have some more checking to do. A case-folding file
3777 * system is either supporting mixed case sensitivity
3778 * access or is completely case-insensitive. Note
3779 * that the file system is always case preserving.
3781 * In mixed sensitivity mode case sensitive behavior
3782 * is the default. FIGNORECASE must be used to
3783 * explicitly request case insensitive behavior.
3785 * If the source and target names provided differ only
3786 * by case (e.g., a request to rename 'tim' to 'Tim'),
3787 * we will treat this as a special case in the
3788 * case-insensitive mode: as long as the source name
3789 * is an exact match, we will allow this to proceed as
3790 * a name-change request.
3792 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3793 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3794 flags & FIGNORECASE)) &&
3795 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3798 * case preserving rename request, require exact
3807 * If the source and destination directories are the same, we should
3808 * grab the z_name_lock of that directory only once.
3812 rw_enter(&sdzp->z_name_lock, RW_READER);
3816 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3817 ZEXISTS | zflg, NULL, NULL);
3818 terr = zfs_dirent_lock(&tdl,
3819 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3821 terr = zfs_dirent_lock(&tdl,
3822 tdzp, tnm, &tzp, zflg, NULL, NULL);
3823 serr = zfs_dirent_lock(&sdl,
3824 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3830 * Source entry invalid or not there.
3833 zfs_dirent_unlock(tdl);
3839 rw_exit(&sdzp->z_name_lock);
3842 * FreeBSD: In OpenSolaris they only check if rename source is
3843 * ".." here, because "." is handled in their lookup. This is
3844 * not the case for FreeBSD, so we check for "." explicitly.
3846 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0)
3847 serr = SET_ERROR(EINVAL);
3852 zfs_dirent_unlock(sdl);
3856 rw_exit(&sdzp->z_name_lock);
3858 if (strcmp(tnm, "..") == 0)
3859 terr = SET_ERROR(EINVAL);
3865 * Must have write access at the source to remove the old entry
3866 * and write access at the target to create the new entry.
3867 * Note that if target and source are the same, this can be
3868 * done in a single check.
3871 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3874 if (ZTOV(szp)->v_type == VDIR) {
3876 * Check to make sure rename is valid.
3877 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3879 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3884 * Does target exist?
3888 * Source and target must be the same type.
3890 if (ZTOV(szp)->v_type == VDIR) {
3891 if (ZTOV(tzp)->v_type != VDIR) {
3892 error = SET_ERROR(ENOTDIR);
3896 if (ZTOV(tzp)->v_type == VDIR) {
3897 error = SET_ERROR(EISDIR);
3902 * POSIX dictates that when the source and target
3903 * entries refer to the same file object, rename
3904 * must do nothing and exit without error.
3906 if (szp->z_id == tzp->z_id) {
3912 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3914 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3917 * notify the target directory if it is not the same
3918 * as source directory.
3921 vnevent_rename_dest_dir(tdvp, ct);
3924 tx = dmu_tx_create(zfsvfs->z_os);
3925 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3926 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3927 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3928 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3930 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3931 zfs_sa_upgrade_txholds(tx, tdzp);
3934 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3935 zfs_sa_upgrade_txholds(tx, tzp);
3938 zfs_sa_upgrade_txholds(tx, szp);
3939 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3940 error = dmu_tx_assign(tx, TXG_NOWAIT);
3943 zfs_rename_unlock(&zl);
3944 zfs_dirent_unlock(sdl);
3945 zfs_dirent_unlock(tdl);
3948 rw_exit(&sdzp->z_name_lock);
3953 if (error == ERESTART) {
3963 if (tzp) /* Attempt to remove the existing target */
3964 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3967 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3969 szp->z_pflags |= ZFS_AV_MODIFIED;
3971 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3972 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3975 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3977 zfs_log_rename(zilog, tx, TX_RENAME |
3978 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3979 sdl->dl_name, tdzp, tdl->dl_name, szp);
3982 * Update path information for the target vnode
3984 vn_renamepath(tdvp, ZTOV(szp), tnm,
3988 * At this point, we have successfully created
3989 * the target name, but have failed to remove
3990 * the source name. Since the create was done
3991 * with the ZRENAMING flag, there are
3992 * complications; for one, the link count is
3993 * wrong. The easiest way to deal with this
3994 * is to remove the newly created target, and
3995 * return the original error. This must
3996 * succeed; fortunately, it is very unlikely to
3997 * fail, since we just created it.
3999 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
4000 ZRENAMING, NULL), ==, 0);
4003 #ifdef FREEBSD_NAMECACHE
4014 zfs_rename_unlock(&zl);
4016 zfs_dirent_unlock(sdl);
4017 zfs_dirent_unlock(tdl);
4020 rw_exit(&sdzp->z_name_lock);
4027 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4028 zil_commit(zilog, 0);
4036 * Insert the indicated symbolic reference entry into the directory.
4038 * IN: dvp - Directory to contain new symbolic link.
4039 * link - Name for new symlink entry.
4040 * vap - Attributes of new entry.
4041 * cr - credentials of caller.
4042 * ct - caller context
4043 * flags - case flags
4045 * RETURN: 0 on success, error code on failure.
4048 * dvp - ctime|mtime updated
4052 zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link,
4053 cred_t *cr, kthread_t *td)
4055 znode_t *zp, *dzp = VTOZ(dvp);
4058 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4060 uint64_t len = strlen(link);
4063 zfs_acl_ids_t acl_ids;
4064 boolean_t fuid_dirtied;
4065 uint64_t txtype = TX_SYMLINK;
4068 ASSERT(vap->va_type == VLNK);
4072 zilog = zfsvfs->z_log;
4074 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
4075 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4077 return (SET_ERROR(EILSEQ));
4079 if (flags & FIGNORECASE)
4082 if (len > MAXPATHLEN) {
4084 return (SET_ERROR(ENAMETOOLONG));
4087 if ((error = zfs_acl_ids_create(dzp, 0,
4088 vap, cr, NULL, &acl_ids)) != 0) {
4094 * Attempt to lock directory; fail if entry already exists.
4096 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
4098 zfs_acl_ids_free(&acl_ids);
4103 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4104 zfs_acl_ids_free(&acl_ids);
4105 zfs_dirent_unlock(dl);
4110 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
4111 zfs_acl_ids_free(&acl_ids);
4112 zfs_dirent_unlock(dl);
4114 return (SET_ERROR(EDQUOT));
4116 tx = dmu_tx_create(zfsvfs->z_os);
4117 fuid_dirtied = zfsvfs->z_fuid_dirty;
4118 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
4119 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4120 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
4121 ZFS_SA_BASE_ATTR_SIZE + len);
4122 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
4123 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
4124 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
4125 acl_ids.z_aclp->z_acl_bytes);
4128 zfs_fuid_txhold(zfsvfs, tx);
4129 error = dmu_tx_assign(tx, TXG_NOWAIT);
4131 zfs_dirent_unlock(dl);
4132 if (error == ERESTART) {
4137 zfs_acl_ids_free(&acl_ids);
4144 * Create a new object for the symlink.
4145 * for version 4 ZPL datsets the symlink will be an SA attribute
4147 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
4150 zfs_fuid_sync(zfsvfs, tx);
4152 mutex_enter(&zp->z_lock);
4154 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
4157 zfs_sa_symlink(zp, link, len, tx);
4158 mutex_exit(&zp->z_lock);
4161 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
4162 &zp->z_size, sizeof (zp->z_size), tx);
4164 * Insert the new object into the directory.
4166 (void) zfs_link_create(dl, zp, tx, ZNEW);
4168 if (flags & FIGNORECASE)
4170 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
4173 zfs_acl_ids_free(&acl_ids);
4177 zfs_dirent_unlock(dl);
4179 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4180 zil_commit(zilog, 0);
4187 * Return, in the buffer contained in the provided uio structure,
4188 * the symbolic path referred to by vp.
4190 * IN: vp - vnode of symbolic link.
4191 * uio - structure to contain the link path.
4192 * cr - credentials of caller.
4193 * ct - caller context
4195 * OUT: uio - structure containing the link path.
4197 * RETURN: 0 on success, error code on failure.
4200 * vp - atime updated
4204 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
4206 znode_t *zp = VTOZ(vp);
4207 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4213 mutex_enter(&zp->z_lock);
4215 error = sa_lookup_uio(zp->z_sa_hdl,
4216 SA_ZPL_SYMLINK(zfsvfs), uio);
4218 error = zfs_sa_readlink(zp, uio);
4219 mutex_exit(&zp->z_lock);
4221 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4228 * Insert a new entry into directory tdvp referencing svp.
4230 * IN: tdvp - Directory to contain new entry.
4231 * svp - vnode of new entry.
4232 * name - name of new entry.
4233 * cr - credentials of caller.
4234 * ct - caller context
4236 * RETURN: 0 on success, error code on failure.
4239 * tdvp - ctime|mtime updated
4240 * svp - ctime updated
4244 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
4245 caller_context_t *ct, int flags)
4247 znode_t *dzp = VTOZ(tdvp);
4249 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4259 ASSERT(tdvp->v_type == VDIR);
4263 zilog = zfsvfs->z_log;
4265 if (VOP_REALVP(svp, &realvp, ct) == 0)
4269 * POSIX dictates that we return EPERM here.
4270 * Better choices include ENOTSUP or EISDIR.
4272 if (svp->v_type == VDIR) {
4274 return (SET_ERROR(EPERM));
4277 if (svp->v_vfsp != tdvp->v_vfsp || zfsctl_is_node(svp)) {
4279 return (SET_ERROR(EXDEV));
4285 /* Prevent links to .zfs/shares files */
4287 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4288 &parent, sizeof (uint64_t))) != 0) {
4292 if (parent == zfsvfs->z_shares_dir) {
4294 return (SET_ERROR(EPERM));
4297 if (zfsvfs->z_utf8 && u8_validate(name,
4298 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4300 return (SET_ERROR(EILSEQ));
4302 if (flags & FIGNORECASE)
4306 * We do not support links between attributes and non-attributes
4307 * because of the potential security risk of creating links
4308 * into "normal" file space in order to circumvent restrictions
4309 * imposed in attribute space.
4311 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4313 return (SET_ERROR(EINVAL));
4317 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4318 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) {
4320 return (SET_ERROR(EPERM));
4323 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4330 * Attempt to lock directory; fail if entry already exists.
4332 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4338 tx = dmu_tx_create(zfsvfs->z_os);
4339 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4340 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4341 zfs_sa_upgrade_txholds(tx, szp);
4342 zfs_sa_upgrade_txholds(tx, dzp);
4343 error = dmu_tx_assign(tx, TXG_NOWAIT);
4345 zfs_dirent_unlock(dl);
4346 if (error == ERESTART) {
4356 error = zfs_link_create(dl, szp, tx, 0);
4359 uint64_t txtype = TX_LINK;
4360 if (flags & FIGNORECASE)
4362 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4367 zfs_dirent_unlock(dl);
4370 vnevent_link(svp, ct);
4373 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4374 zil_commit(zilog, 0);
4382 * zfs_null_putapage() is used when the file system has been force
4383 * unmounted. It just drops the pages.
4387 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4388 size_t *lenp, int flags, cred_t *cr)
4390 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4395 * Push a page out to disk, klustering if possible.
4397 * IN: vp - file to push page to.
4398 * pp - page to push.
4399 * flags - additional flags.
4400 * cr - credentials of caller.
4402 * OUT: offp - start of range pushed.
4403 * lenp - len of range pushed.
4405 * RETURN: 0 on success, error code on failure.
4407 * NOTE: callers must have locked the page to be pushed. On
4408 * exit, the page (and all other pages in the kluster) must be
4413 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4414 size_t *lenp, int flags, cred_t *cr)
4416 znode_t *zp = VTOZ(vp);
4417 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4419 u_offset_t off, koff;
4426 * If our blocksize is bigger than the page size, try to kluster
4427 * multiple pages so that we write a full block (thus avoiding
4428 * a read-modify-write).
4430 if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4431 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4432 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4433 ASSERT(koff <= zp->z_size);
4434 if (koff + klen > zp->z_size)
4435 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4436 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4438 ASSERT3U(btop(len), ==, btopr(len));
4441 * Can't push pages past end-of-file.
4443 if (off >= zp->z_size) {
4444 /* ignore all pages */
4447 } else if (off + len > zp->z_size) {
4448 int npages = btopr(zp->z_size - off);
4451 page_list_break(&pp, &trunc, npages);
4452 /* ignore pages past end of file */
4454 pvn_write_done(trunc, flags);
4455 len = zp->z_size - off;
4458 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4459 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4460 err = SET_ERROR(EDQUOT);
4464 tx = dmu_tx_create(zfsvfs->z_os);
4465 dmu_tx_hold_write(tx, zp->z_id, off, len);
4467 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4468 zfs_sa_upgrade_txholds(tx, zp);
4469 err = dmu_tx_assign(tx, TXG_NOWAIT);
4471 if (err == ERESTART) {
4480 if (zp->z_blksz <= PAGESIZE) {
4481 caddr_t va = zfs_map_page(pp, S_READ);
4482 ASSERT3U(len, <=, PAGESIZE);
4483 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4484 zfs_unmap_page(pp, va);
4486 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4490 uint64_t mtime[2], ctime[2];
4491 sa_bulk_attr_t bulk[3];
4494 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4496 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4498 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4500 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4502 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4507 pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4517 * Copy the portion of the file indicated from pages into the file.
4518 * The pages are stored in a page list attached to the files vnode.
4520 * IN: vp - vnode of file to push page data to.
4521 * off - position in file to put data.
4522 * len - amount of data to write.
4523 * flags - flags to control the operation.
4524 * cr - credentials of caller.
4525 * ct - caller context.
4527 * RETURN: 0 on success, error code on failure.
4530 * vp - ctime|mtime updated
4534 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4535 caller_context_t *ct)
4537 znode_t *zp = VTOZ(vp);
4538 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4550 * Align this request to the file block size in case we kluster.
4551 * XXX - this can result in pretty aggresive locking, which can
4552 * impact simultanious read/write access. One option might be
4553 * to break up long requests (len == 0) into block-by-block
4554 * operations to get narrower locking.
4556 blksz = zp->z_blksz;
4558 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4561 if (len > 0 && ISP2(blksz))
4562 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4568 * Search the entire vp list for pages >= io_off.
4570 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4571 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4574 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4576 if (off > zp->z_size) {
4577 /* past end of file */
4578 zfs_range_unlock(rl);
4583 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4585 for (off = io_off; io_off < off + len; io_off += io_len) {
4586 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4587 pp = page_lookup(vp, io_off,
4588 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4590 pp = page_lookup_nowait(vp, io_off,
4591 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4594 if (pp != NULL && pvn_getdirty(pp, flags)) {
4598 * Found a dirty page to push
4600 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4608 zfs_range_unlock(rl);
4609 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4610 zil_commit(zfsvfs->z_log, zp->z_id);
4618 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4620 znode_t *zp = VTOZ(vp);
4621 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4624 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4625 if (zp->z_sa_hdl == NULL) {
4627 * The fs has been unmounted, or we did a
4628 * suspend/resume and this file no longer exists.
4630 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4631 vrecycle(vp, curthread);
4635 mutex_enter(&zp->z_lock);
4636 if (zp->z_unlinked) {
4638 * Fast path to recycle a vnode of a removed file.
4640 mutex_exit(&zp->z_lock);
4641 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4642 vrecycle(vp, curthread);
4645 mutex_exit(&zp->z_lock);
4647 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4648 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4650 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4651 zfs_sa_upgrade_txholds(tx, zp);
4652 error = dmu_tx_assign(tx, TXG_WAIT);
4656 mutex_enter(&zp->z_lock);
4657 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4658 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4659 zp->z_atime_dirty = 0;
4660 mutex_exit(&zp->z_lock);
4664 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4669 * Bounds-check the seek operation.
4671 * IN: vp - vnode seeking within
4672 * ooff - old file offset
4673 * noffp - pointer to new file offset
4674 * ct - caller context
4676 * RETURN: 0 on success, EINVAL if new offset invalid.
4680 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4681 caller_context_t *ct)
4683 if (vp->v_type == VDIR)
4685 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4689 * Pre-filter the generic locking function to trap attempts to place
4690 * a mandatory lock on a memory mapped file.
4693 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4694 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4696 znode_t *zp = VTOZ(vp);
4697 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4703 * We are following the UFS semantics with respect to mapcnt
4704 * here: If we see that the file is mapped already, then we will
4705 * return an error, but we don't worry about races between this
4706 * function and zfs_map().
4708 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4710 return (SET_ERROR(EAGAIN));
4713 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4717 * If we can't find a page in the cache, we will create a new page
4718 * and fill it with file data. For efficiency, we may try to fill
4719 * multiple pages at once (klustering) to fill up the supplied page
4720 * list. Note that the pages to be filled are held with an exclusive
4721 * lock to prevent access by other threads while they are being filled.
4724 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4725 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4727 znode_t *zp = VTOZ(vp);
4728 page_t *pp, *cur_pp;
4729 objset_t *os = zp->z_zfsvfs->z_os;
4730 u_offset_t io_off, total;
4734 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4736 * We only have a single page, don't bother klustering
4740 pp = page_create_va(vp, io_off, io_len,
4741 PG_EXCL | PG_WAIT, seg, addr);
4744 * Try to find enough pages to fill the page list
4746 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4747 &io_len, off, plsz, 0);
4751 * The page already exists, nothing to do here.
4758 * Fill the pages in the kluster.
4761 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4764 ASSERT3U(io_off, ==, cur_pp->p_offset);
4765 va = zfs_map_page(cur_pp, S_WRITE);
4766 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4768 zfs_unmap_page(cur_pp, va);
4770 /* On error, toss the entire kluster */
4771 pvn_read_done(pp, B_ERROR);
4772 /* convert checksum errors into IO errors */
4774 err = SET_ERROR(EIO);
4777 cur_pp = cur_pp->p_next;
4781 * Fill in the page list array from the kluster starting
4782 * from the desired offset `off'.
4783 * NOTE: the page list will always be null terminated.
4785 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4786 ASSERT(pl == NULL || (*pl)->p_offset == off);
4792 * Return pointers to the pages for the file region [off, off + len]
4793 * in the pl array. If plsz is greater than len, this function may
4794 * also return page pointers from after the specified region
4795 * (i.e. the region [off, off + plsz]). These additional pages are
4796 * only returned if they are already in the cache, or were created as
4797 * part of a klustered read.
4799 * IN: vp - vnode of file to get data from.
4800 * off - position in file to get data from.
4801 * len - amount of data to retrieve.
4802 * plsz - length of provided page list.
4803 * seg - segment to obtain pages for.
4804 * addr - virtual address of fault.
4805 * rw - mode of created pages.
4806 * cr - credentials of caller.
4807 * ct - caller context.
4809 * OUT: protp - protection mode of created pages.
4810 * pl - list of pages created.
4812 * RETURN: 0 on success, error code on failure.
4815 * vp - atime updated
4819 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4820 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4821 enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4823 znode_t *zp = VTOZ(vp);
4824 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4828 /* we do our own caching, faultahead is unnecessary */
4831 else if (len > plsz)
4834 len = P2ROUNDUP(len, PAGESIZE);
4835 ASSERT(plsz >= len);
4844 * Loop through the requested range [off, off + len) looking
4845 * for pages. If we don't find a page, we will need to create
4846 * a new page and fill it with data from the file.
4849 if (*pl = page_lookup(vp, off, SE_SHARED))
4851 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4854 ASSERT3U((*pl)->p_offset, ==, off);
4858 ASSERT3U(len, >=, PAGESIZE);
4861 ASSERT3U(plsz, >=, PAGESIZE);
4868 * Fill out the page array with any pages already in the cache.
4871 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4878 * Release any pages we have previously locked.
4883 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4893 * Request a memory map for a section of a file. This code interacts
4894 * with common code and the VM system as follows:
4896 * - common code calls mmap(), which ends up in smmap_common()
4897 * - this calls VOP_MAP(), which takes you into (say) zfs
4898 * - zfs_map() calls as_map(), passing segvn_create() as the callback
4899 * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4900 * - zfs_addmap() updates z_mapcnt
4904 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4905 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4906 caller_context_t *ct)
4908 znode_t *zp = VTOZ(vp);
4909 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4910 segvn_crargs_t vn_a;
4916 if ((prot & PROT_WRITE) && (zp->z_pflags &
4917 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4919 return (SET_ERROR(EPERM));
4922 if ((prot & (PROT_READ | PROT_EXEC)) &&
4923 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4925 return (SET_ERROR(EACCES));
4928 if (vp->v_flag & VNOMAP) {
4930 return (SET_ERROR(ENOSYS));
4933 if (off < 0 || len > MAXOFFSET_T - off) {
4935 return (SET_ERROR(ENXIO));
4938 if (vp->v_type != VREG) {
4940 return (SET_ERROR(ENODEV));
4944 * If file is locked, disallow mapping.
4946 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
4948 return (SET_ERROR(EAGAIN));
4952 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4960 vn_a.offset = (u_offset_t)off;
4961 vn_a.type = flags & MAP_TYPE;
4963 vn_a.maxprot = maxprot;
4966 vn_a.flags = flags & ~MAP_TYPE;
4968 vn_a.lgrp_mem_policy_flags = 0;
4970 error = as_map(as, *addrp, len, segvn_create, &vn_a);
4979 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4980 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4981 caller_context_t *ct)
4983 uint64_t pages = btopr(len);
4985 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
4990 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4991 * more accurate mtime for the associated file. Since we don't have a way of
4992 * detecting when the data was actually modified, we have to resort to
4993 * heuristics. If an explicit msync() is done, then we mark the mtime when the
4994 * last page is pushed. The problem occurs when the msync() call is omitted,
4995 * which by far the most common case:
5003 * putpage() via fsflush
5005 * If we wait until fsflush to come along, we can have a modification time that
5006 * is some arbitrary point in the future. In order to prevent this in the
5007 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
5012 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5013 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
5014 caller_context_t *ct)
5016 uint64_t pages = btopr(len);
5018 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
5019 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
5021 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
5022 vn_has_cached_data(vp))
5023 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
5029 * Free or allocate space in a file. Currently, this function only
5030 * supports the `F_FREESP' command. However, this command is somewhat
5031 * misnamed, as its functionality includes the ability to allocate as
5032 * well as free space.
5034 * IN: vp - vnode of file to free data in.
5035 * cmd - action to take (only F_FREESP supported).
5036 * bfp - section of file to free/alloc.
5037 * flag - current file open mode flags.
5038 * offset - current file offset.
5039 * cr - credentials of caller [UNUSED].
5040 * ct - caller context.
5042 * RETURN: 0 on success, error code on failure.
5045 * vp - ctime|mtime updated
5049 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
5050 offset_t offset, cred_t *cr, caller_context_t *ct)
5052 znode_t *zp = VTOZ(vp);
5053 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5060 if (cmd != F_FREESP) {
5062 return (SET_ERROR(EINVAL));
5065 if (error = convoff(vp, bfp, 0, offset)) {
5070 if (bfp->l_len < 0) {
5072 return (SET_ERROR(EINVAL));
5076 len = bfp->l_len; /* 0 means from off to end of file */
5078 error = zfs_freesp(zp, off, len, flag, TRUE);
5085 CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid));
5086 CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid));
5090 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
5092 znode_t *zp = VTOZ(vp);
5093 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5096 uint64_t object = zp->z_id;
5103 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
5104 &gen64, sizeof (uint64_t))) != 0) {
5109 gen = (uint32_t)gen64;
5111 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
5114 if (fidp->fid_len < size) {
5115 fidp->fid_len = size;
5117 return (SET_ERROR(ENOSPC));
5120 fidp->fid_len = size;
5123 zfid = (zfid_short_t *)fidp;
5125 zfid->zf_len = size;
5127 for (i = 0; i < sizeof (zfid->zf_object); i++)
5128 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
5130 /* Must have a non-zero generation number to distinguish from .zfs */
5133 for (i = 0; i < sizeof (zfid->zf_gen); i++)
5134 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
5136 if (size == LONG_FID_LEN) {
5137 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
5140 zlfid = (zfid_long_t *)fidp;
5142 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
5143 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
5145 /* XXX - this should be the generation number for the objset */
5146 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
5147 zlfid->zf_setgen[i] = 0;
5155 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
5156 caller_context_t *ct)
5168 case _PC_FILESIZEBITS:
5172 case _PC_XATTR_EXISTS:
5174 zfsvfs = zp->z_zfsvfs;
5178 error = zfs_dirent_lock(&dl, zp, "", &xzp,
5179 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
5181 zfs_dirent_unlock(dl);
5182 if (!zfs_dirempty(xzp))
5185 } else if (error == ENOENT) {
5187 * If there aren't extended attributes, it's the
5188 * same as having zero of them.
5195 case _PC_SATTR_ENABLED:
5196 case _PC_SATTR_EXISTS:
5197 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
5198 (vp->v_type == VREG || vp->v_type == VDIR);
5201 case _PC_ACCESS_FILTERING:
5202 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
5206 case _PC_ACL_ENABLED:
5207 *valp = _ACL_ACE_ENABLED;
5210 case _PC_MIN_HOLE_SIZE:
5211 *valp = (int)SPA_MINBLOCKSIZE;
5214 case _PC_TIMESTAMP_RESOLUTION:
5215 /* nanosecond timestamp resolution */
5219 case _PC_ACL_EXTENDED:
5227 case _PC_ACL_PATH_MAX:
5228 *valp = ACL_MAX_ENTRIES;
5232 return (EOPNOTSUPP);
5238 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5239 caller_context_t *ct)
5241 znode_t *zp = VTOZ(vp);
5242 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5244 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5248 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5256 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5257 caller_context_t *ct)
5259 znode_t *zp = VTOZ(vp);
5260 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5262 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5263 zilog_t *zilog = zfsvfs->z_log;
5268 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5270 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5271 zil_commit(zilog, 0);
5279 * The smallest read we may consider to loan out an arcbuf.
5280 * This must be a power of 2.
5282 int zcr_blksz_min = (1 << 10); /* 1K */
5284 * If set to less than the file block size, allow loaning out of an
5285 * arcbuf for a partial block read. This must be a power of 2.
5287 int zcr_blksz_max = (1 << 17); /* 128K */
5291 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5292 caller_context_t *ct)
5294 znode_t *zp = VTOZ(vp);
5295 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5296 int max_blksz = zfsvfs->z_max_blksz;
5297 uio_t *uio = &xuio->xu_uio;
5298 ssize_t size = uio->uio_resid;
5299 offset_t offset = uio->uio_loffset;
5304 int preamble, postamble;
5306 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5307 return (SET_ERROR(EINVAL));
5314 * Loan out an arc_buf for write if write size is bigger than
5315 * max_blksz, and the file's block size is also max_blksz.
5318 if (size < blksz || zp->z_blksz != blksz) {
5320 return (SET_ERROR(EINVAL));
5323 * Caller requests buffers for write before knowing where the
5324 * write offset might be (e.g. NFS TCP write).
5329 preamble = P2PHASE(offset, blksz);
5331 preamble = blksz - preamble;
5336 postamble = P2PHASE(size, blksz);
5339 fullblk = size / blksz;
5340 (void) dmu_xuio_init(xuio,
5341 (preamble != 0) + fullblk + (postamble != 0));
5342 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5343 int, postamble, int,
5344 (preamble != 0) + fullblk + (postamble != 0));
5347 * Have to fix iov base/len for partial buffers. They
5348 * currently represent full arc_buf's.
5351 /* data begins in the middle of the arc_buf */
5352 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5355 (void) dmu_xuio_add(xuio, abuf,
5356 blksz - preamble, preamble);
5359 for (i = 0; i < fullblk; i++) {
5360 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5363 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5367 /* data ends in the middle of the arc_buf */
5368 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5371 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5376 * Loan out an arc_buf for read if the read size is larger than
5377 * the current file block size. Block alignment is not
5378 * considered. Partial arc_buf will be loaned out for read.
5380 blksz = zp->z_blksz;
5381 if (blksz < zcr_blksz_min)
5382 blksz = zcr_blksz_min;
5383 if (blksz > zcr_blksz_max)
5384 blksz = zcr_blksz_max;
5385 /* avoid potential complexity of dealing with it */
5386 if (blksz > max_blksz) {
5388 return (SET_ERROR(EINVAL));
5391 maxsize = zp->z_size - uio->uio_loffset;
5395 if (size < blksz || vn_has_cached_data(vp)) {
5397 return (SET_ERROR(EINVAL));
5402 return (SET_ERROR(EINVAL));
5405 uio->uio_extflg = UIO_XUIO;
5406 XUIO_XUZC_RW(xuio) = ioflag;
5413 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5417 int ioflag = XUIO_XUZC_RW(xuio);
5419 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5421 i = dmu_xuio_cnt(xuio);
5423 abuf = dmu_xuio_arcbuf(xuio, i);
5425 * if abuf == NULL, it must be a write buffer
5426 * that has been returned in zfs_write().
5429 dmu_return_arcbuf(abuf);
5430 ASSERT(abuf || ioflag == UIO_WRITE);
5433 dmu_xuio_fini(xuio);
5438 * Predeclare these here so that the compiler assumes that
5439 * this is an "old style" function declaration that does
5440 * not include arguments => we won't get type mismatch errors
5441 * in the initializations that follow.
5443 static int zfs_inval();
5444 static int zfs_isdir();
5449 return (SET_ERROR(EINVAL));
5455 return (SET_ERROR(EISDIR));
5458 * Directory vnode operations template
5460 vnodeops_t *zfs_dvnodeops;
5461 const fs_operation_def_t zfs_dvnodeops_template[] = {
5462 VOPNAME_OPEN, { .vop_open = zfs_open },
5463 VOPNAME_CLOSE, { .vop_close = zfs_close },
5464 VOPNAME_READ, { .error = zfs_isdir },
5465 VOPNAME_WRITE, { .error = zfs_isdir },
5466 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5467 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5468 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5469 VOPNAME_ACCESS, { .vop_access = zfs_access },
5470 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5471 VOPNAME_CREATE, { .vop_create = zfs_create },
5472 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5473 VOPNAME_LINK, { .vop_link = zfs_link },
5474 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5475 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir },
5476 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5477 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5478 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink },
5479 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5480 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5481 VOPNAME_FID, { .vop_fid = zfs_fid },
5482 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5483 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5484 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5485 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5486 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5491 * Regular file vnode operations template
5493 vnodeops_t *zfs_fvnodeops;
5494 const fs_operation_def_t zfs_fvnodeops_template[] = {
5495 VOPNAME_OPEN, { .vop_open = zfs_open },
5496 VOPNAME_CLOSE, { .vop_close = zfs_close },
5497 VOPNAME_READ, { .vop_read = zfs_read },
5498 VOPNAME_WRITE, { .vop_write = zfs_write },
5499 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5500 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5501 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5502 VOPNAME_ACCESS, { .vop_access = zfs_access },
5503 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5504 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5505 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5506 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5507 VOPNAME_FID, { .vop_fid = zfs_fid },
5508 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5509 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock },
5510 VOPNAME_SPACE, { .vop_space = zfs_space },
5511 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage },
5512 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage },
5513 VOPNAME_MAP, { .vop_map = zfs_map },
5514 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap },
5515 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap },
5516 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5517 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5518 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5519 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5520 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf },
5521 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf },
5526 * Symbolic link vnode operations template
5528 vnodeops_t *zfs_symvnodeops;
5529 const fs_operation_def_t zfs_symvnodeops_template[] = {
5530 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5531 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5532 VOPNAME_ACCESS, { .vop_access = zfs_access },
5533 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5534 VOPNAME_READLINK, { .vop_readlink = zfs_readlink },
5535 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5536 VOPNAME_FID, { .vop_fid = zfs_fid },
5537 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5538 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5543 * special share hidden files vnode operations template
5545 vnodeops_t *zfs_sharevnodeops;
5546 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5547 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5548 VOPNAME_ACCESS, { .vop_access = zfs_access },
5549 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5550 VOPNAME_FID, { .vop_fid = zfs_fid },
5551 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5552 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5553 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5554 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5559 * Extended attribute directory vnode operations template
5561 * This template is identical to the directory vnodes
5562 * operation template except for restricted operations:
5566 * Note that there are other restrictions embedded in:
5567 * zfs_create() - restrict type to VREG
5568 * zfs_link() - no links into/out of attribute space
5569 * zfs_rename() - no moves into/out of attribute space
5571 vnodeops_t *zfs_xdvnodeops;
5572 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5573 VOPNAME_OPEN, { .vop_open = zfs_open },
5574 VOPNAME_CLOSE, { .vop_close = zfs_close },
5575 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5576 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5577 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5578 VOPNAME_ACCESS, { .vop_access = zfs_access },
5579 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5580 VOPNAME_CREATE, { .vop_create = zfs_create },
5581 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5582 VOPNAME_LINK, { .vop_link = zfs_link },
5583 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5584 VOPNAME_MKDIR, { .error = zfs_inval },
5585 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5586 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5587 VOPNAME_SYMLINK, { .error = zfs_inval },
5588 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5589 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5590 VOPNAME_FID, { .vop_fid = zfs_fid },
5591 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5592 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5593 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5594 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5595 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5600 * Error vnode operations template
5602 vnodeops_t *zfs_evnodeops;
5603 const fs_operation_def_t zfs_evnodeops_template[] = {
5604 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5605 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5611 ioflags(int ioflags)
5615 if (ioflags & IO_APPEND)
5617 if (ioflags & IO_NDELAY)
5619 if (ioflags & IO_SYNC)
5620 flags |= (FSYNC | FDSYNC | FRSYNC);
5626 zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage)
5628 znode_t *zp = VTOZ(vp);
5629 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5630 objset_t *os = zp->z_zfsvfs->z_os;
5641 pcount = round_page(count) / PAGE_SIZE;
5643 object = mreq->object;
5646 KASSERT(vp->v_object == object, ("mismatching object"));
5648 VM_OBJECT_LOCK(object);
5649 vm_page_lock_queues();
5650 for (i = 0; i < pcount; i++) {
5655 vm_page_unlock_queues();
5658 if (mreq->valid != VM_PAGE_BITS_ALL)
5659 vm_page_zero_invalid(mreq, TRUE);
5660 VM_OBJECT_UNLOCK(object);
5662 return (VM_PAGER_OK);
5665 PCPU_INC(cnt.v_vnodein);
5666 PCPU_INC(cnt.v_vnodepgsin);
5668 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) {
5669 VM_OBJECT_UNLOCK(object);
5671 return (VM_PAGER_BAD);
5675 if (IDX_TO_OFF(mreq->pindex) + size > object->un_pager.vnp.vnp_size)
5676 size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mreq->pindex);
5678 VM_OBJECT_UNLOCK(object);
5679 va = zfs_map_page(mreq, &sf);
5680 error = dmu_read(os, zp->z_id, IDX_TO_OFF(mreq->pindex),
5681 size, va, DMU_READ_PREFETCH);
5682 if (size != PAGE_SIZE)
5683 bzero(va + size, PAGE_SIZE - size);
5685 VM_OBJECT_LOCK(object);
5688 mreq->valid = VM_PAGE_BITS_ALL;
5689 KASSERT(mreq->dirty == 0, ("zfs_getpages: page %p is dirty", mreq));
5691 VM_OBJECT_UNLOCK(object);
5693 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
5695 return (error ? VM_PAGER_ERROR : VM_PAGER_OK);
5699 zfs_freebsd_getpages(ap)
5700 struct vop_getpages_args /* {
5705 vm_ooffset_t a_offset;
5709 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage));
5713 zfs_freebsd_open(ap)
5714 struct vop_open_args /* {
5717 struct ucred *a_cred;
5718 struct thread *a_td;
5721 vnode_t *vp = ap->a_vp;
5722 znode_t *zp = VTOZ(vp);
5725 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL);
5727 vnode_create_vobject(vp, zp->z_size, ap->a_td);
5732 zfs_freebsd_close(ap)
5733 struct vop_close_args /* {
5736 struct ucred *a_cred;
5737 struct thread *a_td;
5741 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL));
5745 zfs_freebsd_ioctl(ap)
5746 struct vop_ioctl_args /* {
5756 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data,
5757 ap->a_fflag, ap->a_cred, NULL, NULL));
5761 zfs_freebsd_read(ap)
5762 struct vop_read_args /* {
5766 struct ucred *a_cred;
5770 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5775 zfs_freebsd_write(ap)
5776 struct vop_write_args /* {
5780 struct ucred *a_cred;
5784 if (vn_rlimit_fsize(ap->a_vp, ap->a_uio, ap->a_uio->uio_td))
5787 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5792 zfs_freebsd_access(ap)
5793 struct vop_access_args /* {
5795 accmode_t a_accmode;
5796 struct ucred *a_cred;
5797 struct thread *a_td;
5804 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND,
5806 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND);
5808 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL);
5811 * VADMIN has to be handled by vaccess().
5814 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND);
5816 vnode_t *vp = ap->a_vp;
5817 znode_t *zp = VTOZ(vp);
5819 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
5820 zp->z_gid, accmode, ap->a_cred, NULL);
5828 zfs_freebsd_lookup(ap)
5829 struct vop_lookup_args /* {
5830 struct vnode *a_dvp;
5831 struct vnode **a_vpp;
5832 struct componentname *a_cnp;
5835 struct componentname *cnp = ap->a_cnp;
5836 char nm[NAME_MAX + 1];
5838 ASSERT(cnp->cn_namelen < sizeof(nm));
5839 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm)));
5841 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop,
5842 cnp->cn_cred, cnp->cn_thread, 0));
5846 zfs_freebsd_create(ap)
5847 struct vop_create_args /* {
5848 struct vnode *a_dvp;
5849 struct vnode **a_vpp;
5850 struct componentname *a_cnp;
5851 struct vattr *a_vap;
5854 struct componentname *cnp = ap->a_cnp;
5855 vattr_t *vap = ap->a_vap;
5858 ASSERT(cnp->cn_flags & SAVENAME);
5860 vattr_init_mask(vap);
5861 mode = vap->va_mode & ALLPERMS;
5863 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode,
5864 ap->a_vpp, cnp->cn_cred, cnp->cn_thread));
5868 zfs_freebsd_remove(ap)
5869 struct vop_remove_args /* {
5870 struct vnode *a_dvp;
5872 struct componentname *a_cnp;
5876 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
5878 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr,
5879 ap->a_cnp->cn_cred, NULL, 0));
5883 zfs_freebsd_mkdir(ap)
5884 struct vop_mkdir_args /* {
5885 struct vnode *a_dvp;
5886 struct vnode **a_vpp;
5887 struct componentname *a_cnp;
5888 struct vattr *a_vap;
5891 vattr_t *vap = ap->a_vap;
5893 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
5895 vattr_init_mask(vap);
5897 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp,
5898 ap->a_cnp->cn_cred, NULL, 0, NULL));
5902 zfs_freebsd_rmdir(ap)
5903 struct vop_rmdir_args /* {
5904 struct vnode *a_dvp;
5906 struct componentname *a_cnp;
5909 struct componentname *cnp = ap->a_cnp;
5911 ASSERT(cnp->cn_flags & SAVENAME);
5913 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0));
5917 zfs_freebsd_readdir(ap)
5918 struct vop_readdir_args /* {
5921 struct ucred *a_cred;
5928 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag,
5929 ap->a_ncookies, ap->a_cookies));
5933 zfs_freebsd_fsync(ap)
5934 struct vop_fsync_args /* {
5937 struct thread *a_td;
5942 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL));
5946 zfs_freebsd_getattr(ap)
5947 struct vop_getattr_args /* {
5949 struct vattr *a_vap;
5950 struct ucred *a_cred;
5953 vattr_t *vap = ap->a_vap;
5959 xvap.xva_vattr = *vap;
5960 xvap.xva_vattr.va_mask |= AT_XVATTR;
5962 /* Convert chflags into ZFS-type flags. */
5963 /* XXX: what about SF_SETTABLE?. */
5964 XVA_SET_REQ(&xvap, XAT_IMMUTABLE);
5965 XVA_SET_REQ(&xvap, XAT_APPENDONLY);
5966 XVA_SET_REQ(&xvap, XAT_NOUNLINK);
5967 XVA_SET_REQ(&xvap, XAT_NODUMP);
5968 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL);
5972 /* Convert ZFS xattr into chflags. */
5973 #define FLAG_CHECK(fflag, xflag, xfield) do { \
5974 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \
5975 fflags |= (fflag); \
5977 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE,
5978 xvap.xva_xoptattrs.xoa_immutable);
5979 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY,
5980 xvap.xva_xoptattrs.xoa_appendonly);
5981 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK,
5982 xvap.xva_xoptattrs.xoa_nounlink);
5983 FLAG_CHECK(UF_NODUMP, XAT_NODUMP,
5984 xvap.xva_xoptattrs.xoa_nodump);
5986 *vap = xvap.xva_vattr;
5987 vap->va_flags = fflags;
5992 zfs_freebsd_setattr(ap)
5993 struct vop_setattr_args /* {
5995 struct vattr *a_vap;
5996 struct ucred *a_cred;
5999 vnode_t *vp = ap->a_vp;
6000 vattr_t *vap = ap->a_vap;
6001 cred_t *cred = ap->a_cred;
6006 vattr_init_mask(vap);
6007 vap->va_mask &= ~AT_NOSET;
6010 xvap.xva_vattr = *vap;
6012 zflags = VTOZ(vp)->z_pflags;
6014 if (vap->va_flags != VNOVAL) {
6015 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs;
6018 if (zfsvfs->z_use_fuids == B_FALSE)
6019 return (EOPNOTSUPP);
6021 fflags = vap->va_flags;
6022 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_NODUMP)) != 0)
6023 return (EOPNOTSUPP);
6025 * Unprivileged processes are not permitted to unset system
6026 * flags, or modify flags if any system flags are set.
6027 * Privileged non-jail processes may not modify system flags
6028 * if securelevel > 0 and any existing system flags are set.
6029 * Privileged jail processes behave like privileged non-jail
6030 * processes if the security.jail.chflags_allowed sysctl is
6031 * is non-zero; otherwise, they behave like unprivileged
6034 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 ||
6035 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) {
6037 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6038 error = securelevel_gt(cred, 0);
6044 * Callers may only modify the file flags on objects they
6045 * have VADMIN rights for.
6047 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0)
6050 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6054 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) {
6059 #define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \
6060 if (((fflags & (fflag)) && !(zflags & (zflag))) || \
6061 ((zflags & (zflag)) && !(fflags & (fflag)))) { \
6062 XVA_SET_REQ(&xvap, (xflag)); \
6063 (xfield) = ((fflags & (fflag)) != 0); \
6066 /* Convert chflags into ZFS-type flags. */
6067 /* XXX: what about SF_SETTABLE?. */
6068 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE,
6069 xvap.xva_xoptattrs.xoa_immutable);
6070 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY,
6071 xvap.xva_xoptattrs.xoa_appendonly);
6072 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK,
6073 xvap.xva_xoptattrs.xoa_nounlink);
6074 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP,
6075 xvap.xva_xoptattrs.xoa_nodump);
6078 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL));
6082 zfs_freebsd_rename(ap)
6083 struct vop_rename_args /* {
6084 struct vnode *a_fdvp;
6085 struct vnode *a_fvp;
6086 struct componentname *a_fcnp;
6087 struct vnode *a_tdvp;
6088 struct vnode *a_tvp;
6089 struct componentname *a_tcnp;
6092 vnode_t *fdvp = ap->a_fdvp;
6093 vnode_t *fvp = ap->a_fvp;
6094 vnode_t *tdvp = ap->a_tdvp;
6095 vnode_t *tvp = ap->a_tvp;
6098 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART));
6099 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART));
6101 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp,
6102 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0);
6117 zfs_freebsd_symlink(ap)
6118 struct vop_symlink_args /* {
6119 struct vnode *a_dvp;
6120 struct vnode **a_vpp;
6121 struct componentname *a_cnp;
6122 struct vattr *a_vap;
6126 struct componentname *cnp = ap->a_cnp;
6127 vattr_t *vap = ap->a_vap;
6129 ASSERT(cnp->cn_flags & SAVENAME);
6131 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */
6132 vattr_init_mask(vap);
6134 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap,
6135 ap->a_target, cnp->cn_cred, cnp->cn_thread));
6139 zfs_freebsd_readlink(ap)
6140 struct vop_readlink_args /* {
6143 struct ucred *a_cred;
6147 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL));
6151 zfs_freebsd_link(ap)
6152 struct vop_link_args /* {
6153 struct vnode *a_tdvp;
6155 struct componentname *a_cnp;
6158 struct componentname *cnp = ap->a_cnp;
6160 ASSERT(cnp->cn_flags & SAVENAME);
6162 return (zfs_link(ap->a_tdvp, ap->a_vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0));
6166 zfs_freebsd_inactive(ap)
6167 struct vop_inactive_args /* {
6169 struct thread *a_td;
6172 vnode_t *vp = ap->a_vp;
6174 zfs_inactive(vp, ap->a_td->td_ucred, NULL);
6179 zfs_freebsd_reclaim(ap)
6180 struct vop_reclaim_args /* {
6182 struct thread *a_td;
6185 vnode_t *vp = ap->a_vp;
6186 znode_t *zp = VTOZ(vp);
6187 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
6191 /* Destroy the vm object and flush associated pages. */
6192 vnode_destroy_vobject(vp);
6195 * z_teardown_inactive_lock protects from a race with
6196 * zfs_znode_dmu_fini in zfsvfs_teardown during
6199 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
6200 if (zp->z_sa_hdl == NULL)
6204 rw_exit(&zfsvfs->z_teardown_inactive_lock);
6212 struct vop_fid_args /* {
6218 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL));
6222 zfs_freebsd_pathconf(ap)
6223 struct vop_pathconf_args /* {
6226 register_t *a_retval;
6232 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL);
6234 *ap->a_retval = val;
6235 else if (error == EOPNOTSUPP)
6236 error = vop_stdpathconf(ap);
6241 zfs_freebsd_fifo_pathconf(ap)
6242 struct vop_pathconf_args /* {
6245 register_t *a_retval;
6249 switch (ap->a_name) {
6250 case _PC_ACL_EXTENDED:
6252 case _PC_ACL_PATH_MAX:
6253 case _PC_MAC_PRESENT:
6254 return (zfs_freebsd_pathconf(ap));
6256 return (fifo_specops.vop_pathconf(ap));
6261 * FreeBSD's extended attributes namespace defines file name prefix for ZFS'
6262 * extended attribute name:
6265 * system freebsd:system:
6266 * user (none, can be used to access ZFS fsattr(5) attributes
6267 * created on Solaris)
6270 zfs_create_attrname(int attrnamespace, const char *name, char *attrname,
6273 const char *namespace, *prefix, *suffix;
6275 /* We don't allow '/' character in attribute name. */
6276 if (strchr(name, '/') != NULL)
6278 /* We don't allow attribute names that start with "freebsd:" string. */
6279 if (strncmp(name, "freebsd:", 8) == 0)
6282 bzero(attrname, size);
6284 switch (attrnamespace) {
6285 case EXTATTR_NAMESPACE_USER:
6287 prefix = "freebsd:";
6288 namespace = EXTATTR_NAMESPACE_USER_STRING;
6292 * This is the default namespace by which we can access all
6293 * attributes created on Solaris.
6295 prefix = namespace = suffix = "";
6298 case EXTATTR_NAMESPACE_SYSTEM:
6299 prefix = "freebsd:";
6300 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING;
6303 case EXTATTR_NAMESPACE_EMPTY:
6307 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix,
6309 return (ENAMETOOLONG);
6315 * Vnode operating to retrieve a named extended attribute.
6318 zfs_getextattr(struct vop_getextattr_args *ap)
6321 IN struct vnode *a_vp;
6322 IN int a_attrnamespace;
6323 IN const char *a_name;
6324 INOUT struct uio *a_uio;
6326 IN struct ucred *a_cred;
6327 IN struct thread *a_td;
6331 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6332 struct thread *td = ap->a_td;
6333 struct nameidata nd;
6336 vnode_t *xvp = NULL, *vp;
6339 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6340 ap->a_cred, ap->a_td, VREAD);
6344 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6351 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6359 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_SYSSPACE, attrname,
6361 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL);
6363 NDFREE(&nd, NDF_ONLY_PNBUF);
6366 if (error == ENOENT)
6371 if (ap->a_size != NULL) {
6372 error = VOP_GETATTR(vp, &va, ap->a_cred);
6374 *ap->a_size = (size_t)va.va_size;
6375 } else if (ap->a_uio != NULL)
6376 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6379 vn_close(vp, flags, ap->a_cred, td);
6386 * Vnode operation to remove a named attribute.
6389 zfs_deleteextattr(struct vop_deleteextattr_args *ap)
6392 IN struct vnode *a_vp;
6393 IN int a_attrnamespace;
6394 IN const char *a_name;
6395 IN struct ucred *a_cred;
6396 IN struct thread *a_td;
6400 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6401 struct thread *td = ap->a_td;
6402 struct nameidata nd;
6405 vnode_t *xvp = NULL, *vp;
6408 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6409 ap->a_cred, ap->a_td, VWRITE);
6413 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6420 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6427 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF | MPSAFE,
6428 UIO_SYSSPACE, attrname, xvp, td);
6431 NDFREE(&nd, NDF_ONLY_PNBUF);
6434 if (error == ENOENT)
6438 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
6441 if (vp == nd.ni_dvp)
6451 * Vnode operation to set a named attribute.
6454 zfs_setextattr(struct vop_setextattr_args *ap)
6457 IN struct vnode *a_vp;
6458 IN int a_attrnamespace;
6459 IN const char *a_name;
6460 INOUT struct uio *a_uio;
6461 IN struct ucred *a_cred;
6462 IN struct thread *a_td;
6466 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6467 struct thread *td = ap->a_td;
6468 struct nameidata nd;
6471 vnode_t *xvp = NULL, *vp;
6474 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6475 ap->a_cred, ap->a_td, VWRITE);
6479 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6486 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6487 LOOKUP_XATTR | CREATE_XATTR_DIR);
6493 flags = FFLAGS(O_WRONLY | O_CREAT);
6494 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_SYSSPACE, attrname,
6496 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL);
6498 NDFREE(&nd, NDF_ONLY_PNBUF);
6506 error = VOP_SETATTR(vp, &va, ap->a_cred);
6508 VOP_WRITE(vp, ap->a_uio, IO_UNIT | IO_SYNC, ap->a_cred);
6511 vn_close(vp, flags, ap->a_cred, td);
6518 * Vnode operation to retrieve extended attributes on a vnode.
6521 zfs_listextattr(struct vop_listextattr_args *ap)
6524 IN struct vnode *a_vp;
6525 IN int a_attrnamespace;
6526 INOUT struct uio *a_uio;
6528 IN struct ucred *a_cred;
6529 IN struct thread *a_td;
6533 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6534 struct thread *td = ap->a_td;
6535 struct nameidata nd;
6536 char attrprefix[16];
6537 u_char dirbuf[sizeof(struct dirent)];
6540 struct uio auio, *uio = ap->a_uio;
6541 size_t *sizep = ap->a_size;
6543 vnode_t *xvp = NULL, *vp;
6544 int done, error, eof, pos;
6546 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6547 ap->a_cred, ap->a_td, VREAD);
6551 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix,
6552 sizeof(attrprefix));
6555 plen = strlen(attrprefix);
6562 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6567 * ENOATTR means that the EA directory does not yet exist,
6568 * i.e. there are no extended attributes there.
6570 if (error == ENOATTR)
6575 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED | MPSAFE,
6576 UIO_SYSSPACE, ".", xvp, td);
6579 NDFREE(&nd, NDF_ONLY_PNBUF);
6585 auio.uio_iov = &aiov;
6586 auio.uio_iovcnt = 1;
6587 auio.uio_segflg = UIO_SYSSPACE;
6589 auio.uio_rw = UIO_READ;
6590 auio.uio_offset = 0;
6595 aiov.iov_base = (void *)dirbuf;
6596 aiov.iov_len = sizeof(dirbuf);
6597 auio.uio_resid = sizeof(dirbuf);
6598 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL);
6599 done = sizeof(dirbuf) - auio.uio_resid;
6602 for (pos = 0; pos < done;) {
6603 dp = (struct dirent *)(dirbuf + pos);
6604 pos += dp->d_reclen;
6606 * XXX: Temporarily we also accept DT_UNKNOWN, as this
6607 * is what we get when attribute was created on Solaris.
6609 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN)
6611 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0)
6613 else if (strncmp(dp->d_name, attrprefix, plen) != 0)
6615 nlen = dp->d_namlen - plen;
6618 else if (uio != NULL) {
6620 * Format of extattr name entry is one byte for
6621 * length and the rest for name.
6623 error = uiomove(&nlen, 1, uio->uio_rw, uio);
6625 error = uiomove(dp->d_name + plen, nlen,
6632 } while (!eof && error == 0);
6641 zfs_freebsd_getacl(ap)
6642 struct vop_getacl_args /* {
6651 vsecattr_t vsecattr;
6653 if (ap->a_type != ACL_TYPE_NFS4)
6656 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
6657 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL))
6660 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt);
6661 if (vsecattr.vsa_aclentp != NULL)
6662 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz);
6668 zfs_freebsd_setacl(ap)
6669 struct vop_setacl_args /* {
6678 vsecattr_t vsecattr;
6679 int aclbsize; /* size of acl list in bytes */
6682 if (ap->a_type != ACL_TYPE_NFS4)
6685 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES)
6689 * With NFSv4 ACLs, chmod(2) may need to add additional entries,
6690 * splitting every entry into two and appending "canonical six"
6691 * entries at the end. Don't allow for setting an ACL that would
6692 * cause chmod(2) to run out of ACL entries.
6694 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES)
6697 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR);
6701 vsecattr.vsa_mask = VSA_ACE;
6702 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t);
6703 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP);
6704 aaclp = vsecattr.vsa_aclentp;
6705 vsecattr.vsa_aclentsz = aclbsize;
6707 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp);
6708 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL);
6709 kmem_free(aaclp, aclbsize);
6715 zfs_freebsd_aclcheck(ap)
6716 struct vop_aclcheck_args /* {
6725 return (EOPNOTSUPP);
6728 struct vop_vector zfs_vnodeops;
6729 struct vop_vector zfs_fifoops;
6730 struct vop_vector zfs_shareops;
6732 struct vop_vector zfs_vnodeops = {
6733 .vop_default = &default_vnodeops,
6734 .vop_inactive = zfs_freebsd_inactive,
6735 .vop_reclaim = zfs_freebsd_reclaim,
6736 .vop_access = zfs_freebsd_access,
6737 #ifdef FREEBSD_NAMECACHE
6738 .vop_lookup = vfs_cache_lookup,
6739 .vop_cachedlookup = zfs_freebsd_lookup,
6741 .vop_lookup = zfs_freebsd_lookup,
6743 .vop_getattr = zfs_freebsd_getattr,
6744 .vop_setattr = zfs_freebsd_setattr,
6745 .vop_create = zfs_freebsd_create,
6746 .vop_mknod = zfs_freebsd_create,
6747 .vop_mkdir = zfs_freebsd_mkdir,
6748 .vop_readdir = zfs_freebsd_readdir,
6749 .vop_fsync = zfs_freebsd_fsync,
6750 .vop_open = zfs_freebsd_open,
6751 .vop_close = zfs_freebsd_close,
6752 .vop_rmdir = zfs_freebsd_rmdir,
6753 .vop_ioctl = zfs_freebsd_ioctl,
6754 .vop_link = zfs_freebsd_link,
6755 .vop_symlink = zfs_freebsd_symlink,
6756 .vop_readlink = zfs_freebsd_readlink,
6757 .vop_read = zfs_freebsd_read,
6758 .vop_write = zfs_freebsd_write,
6759 .vop_remove = zfs_freebsd_remove,
6760 .vop_rename = zfs_freebsd_rename,
6761 .vop_pathconf = zfs_freebsd_pathconf,
6762 .vop_bmap = VOP_EOPNOTSUPP,
6763 .vop_fid = zfs_freebsd_fid,
6764 .vop_getextattr = zfs_getextattr,
6765 .vop_deleteextattr = zfs_deleteextattr,
6766 .vop_setextattr = zfs_setextattr,
6767 .vop_listextattr = zfs_listextattr,
6768 .vop_getacl = zfs_freebsd_getacl,
6769 .vop_setacl = zfs_freebsd_setacl,
6770 .vop_aclcheck = zfs_freebsd_aclcheck,
6771 .vop_getpages = zfs_freebsd_getpages,
6774 struct vop_vector zfs_fifoops = {
6775 .vop_default = &fifo_specops,
6776 .vop_fsync = zfs_freebsd_fsync,
6777 .vop_access = zfs_freebsd_access,
6778 .vop_getattr = zfs_freebsd_getattr,
6779 .vop_inactive = zfs_freebsd_inactive,
6780 .vop_read = VOP_PANIC,
6781 .vop_reclaim = zfs_freebsd_reclaim,
6782 .vop_setattr = zfs_freebsd_setattr,
6783 .vop_write = VOP_PANIC,
6784 .vop_pathconf = zfs_freebsd_fifo_pathconf,
6785 .vop_fid = zfs_freebsd_fid,
6786 .vop_getacl = zfs_freebsd_getacl,
6787 .vop_setacl = zfs_freebsd_setacl,
6788 .vop_aclcheck = zfs_freebsd_aclcheck,
6792 * special share hidden files vnode operations template
6794 struct vop_vector zfs_shareops = {
6795 .vop_default = &default_vnodeops,
6796 .vop_access = zfs_freebsd_access,
6797 .vop_inactive = zfs_freebsd_inactive,
6798 .vop_reclaim = zfs_freebsd_reclaim,
6799 .vop_fid = zfs_freebsd_fid,
6800 .vop_pathconf = zfs_freebsd_pathconf,