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]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
25 * Copyright (c) 2014 Integros [integros.com]
26 * Copyright 2017 Nexenta Systems, Inc.
29 /* Portions Copyright 2007 Jeremy Teo */
30 /* Portions Copyright 2010 Robert Milkowski */
33 #include <sys/types.h>
34 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/sysmacros.h>
38 #include <sys/resource.h>
40 #include <sys/endian.h>
42 #include <sys/vnode.h>
43 #if __FreeBSD_version >= 1300102
46 #include <sys/dirent.h>
50 #include <sys/taskq.h>
52 #include <sys/atomic.h>
53 #include <sys/namei.h>
55 #include <sys/cmn_err.h>
57 #include <sys/sysproto.h>
58 #include <sys/errno.h>
59 #include <sys/unistd.h>
60 #include <sys/zfs_dir.h>
61 #include <sys/zfs_ioctl.h>
62 #include <sys/fs/zfs.h>
64 #include <sys/dmu_objset.h>
70 #include <sys/policy.h>
71 #include <sys/sunddi.h>
72 #include <sys/filio.h>
74 #include <sys/zfs_ctldir.h>
75 #include <sys/zfs_fuid.h>
76 #include <sys/zfs_quota.h>
77 #include <sys/zfs_sa.h>
78 #include <sys/zfs_rlock.h>
79 #include <sys/extdirent.h>
82 #include <sys/sched.h>
84 #include <sys/vmmeter.h>
85 #include <vm/vm_param.h>
87 #include <sys/zfs_vnops.h>
89 #include <vm/vm_object.h>
91 #include <sys/extattr.h>
95 #define VN_OPEN_INVFS 0x0
100 #if __FreeBSD_version >= 1300047
101 #define vm_page_wire_lock(pp)
102 #define vm_page_wire_unlock(pp)
104 #define vm_page_wire_lock(pp) vm_page_lock(pp)
105 #define vm_page_wire_unlock(pp) vm_page_unlock(pp)
108 #ifdef DEBUG_VFS_LOCKS
109 #define VNCHECKREF(vp) \
110 VNASSERT((vp)->v_holdcnt > 0 && (vp)->v_usecount > 0, vp, \
111 ("%s: wrong ref counts", __func__));
113 #define VNCHECKREF(vp)
119 * Each vnode op performs some logical unit of work. To do this, the ZPL must
120 * properly lock its in-core state, create a DMU transaction, do the work,
121 * record this work in the intent log (ZIL), commit the DMU transaction,
122 * and wait for the intent log to commit if it is a synchronous operation.
123 * Moreover, the vnode ops must work in both normal and log replay context.
124 * The ordering of events is important to avoid deadlocks and references
125 * to freed memory. The example below illustrates the following Big Rules:
127 * (1) A check must be made in each zfs thread for a mounted file system.
128 * This is done avoiding races using ZFS_ENTER(zfsvfs).
129 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
130 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
131 * can return EIO from the calling function.
133 * (2) VN_RELE() should always be the last thing except for zil_commit()
134 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
135 * First, if it's the last reference, the vnode/znode
136 * can be freed, so the zp may point to freed memory. Second, the last
137 * reference will call zfs_zinactive(), which may induce a lot of work --
138 * pushing cached pages (which acquires range locks) and syncing out
139 * cached atime changes. Third, zfs_zinactive() may require a new tx,
140 * which could deadlock the system if you were already holding one.
141 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
143 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
144 * as they can span dmu_tx_assign() calls.
146 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
147 * dmu_tx_assign(). This is critical because we don't want to block
148 * while holding locks.
150 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
151 * reduces lock contention and CPU usage when we must wait (note that if
152 * throughput is constrained by the storage, nearly every transaction
155 * Note, in particular, that if a lock is sometimes acquired before
156 * the tx assigns, and sometimes after (e.g. z_lock), then failing
157 * to use a non-blocking assign can deadlock the system. The scenario:
159 * Thread A has grabbed a lock before calling dmu_tx_assign().
160 * Thread B is in an already-assigned tx, and blocks for this lock.
161 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
162 * forever, because the previous txg can't quiesce until B's tx commits.
164 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
165 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
166 * calls to dmu_tx_assign(), pass TXG_NOTHROTTLE in addition to TXG_NOWAIT,
167 * to indicate that this operation has already called dmu_tx_wait().
168 * This will ensure that we don't retry forever, waiting a short bit
171 * (5) If the operation succeeded, generate the intent log entry for it
172 * before dropping locks. This ensures that the ordering of events
173 * in the intent log matches the order in which they actually occurred.
174 * During ZIL replay the zfs_log_* functions will update the sequence
175 * number to indicate the zil transaction has replayed.
177 * (6) At the end of each vnode op, the DMU tx must always commit,
178 * regardless of whether there were any errors.
180 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
181 * to ensure that synchronous semantics are provided when necessary.
183 * In general, this is how things should be ordered in each vnode op:
185 * ZFS_ENTER(zfsvfs); // exit if unmounted
187 * zfs_dirent_lookup(&dl, ...) // lock directory entry (may VN_HOLD())
188 * rw_enter(...); // grab any other locks you need
189 * tx = dmu_tx_create(...); // get DMU tx
190 * dmu_tx_hold_*(); // hold each object you might modify
191 * error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
193 * rw_exit(...); // drop locks
194 * zfs_dirent_unlock(dl); // unlock directory entry
195 * VN_RELE(...); // release held vnodes
196 * if (error == ERESTART) {
202 * dmu_tx_abort(tx); // abort DMU tx
203 * ZFS_EXIT(zfsvfs); // finished in zfs
204 * return (error); // really out of space
206 * error = do_real_work(); // do whatever this VOP does
208 * zfs_log_*(...); // on success, make ZIL entry
209 * dmu_tx_commit(tx); // commit DMU tx -- error or not
210 * rw_exit(...); // drop locks
211 * zfs_dirent_unlock(dl); // unlock directory entry
212 * VN_RELE(...); // release held vnodes
213 * zil_commit(zilog, foid); // synchronous when necessary
214 * ZFS_EXIT(zfsvfs); // finished in zfs
215 * return (error); // done, report error
220 zfs_open(vnode_t **vpp, int flag, cred_t *cr)
222 znode_t *zp = VTOZ(*vpp);
223 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
228 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
229 ((flag & FAPPEND) == 0)) {
231 return (SET_ERROR(EPERM));
234 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
235 ZTOV(zp)->v_type == VREG &&
236 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
237 if (fs_vscan(*vpp, cr, 0) != 0) {
239 return (SET_ERROR(EACCES));
243 /* Keep a count of the synchronous opens in the znode */
244 if (flag & (FSYNC | FDSYNC))
245 atomic_inc_32(&zp->z_sync_cnt);
253 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr)
255 znode_t *zp = VTOZ(vp);
256 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
261 /* Decrement the synchronous opens in the znode */
262 if ((flag & (FSYNC | FDSYNC)) && (count == 1))
263 atomic_dec_32(&zp->z_sync_cnt);
265 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
266 ZTOV(zp)->v_type == VREG &&
267 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
268 VERIFY0(fs_vscan(vp, cr, 1));
276 zfs_ioctl(vnode_t *vp, ulong_t com, intptr_t data, int flag, cred_t *cred,
288 * The following two ioctls are used by bfu. Faking out,
289 * necessary to avoid bfu errors.
301 off = *(offset_t *)data;
302 /* offset parameter is in/out */
303 error = zfs_holey(VTOZ(vp), com, &off);
306 *(offset_t *)data = off;
310 return (SET_ERROR(ENOTTY));
314 page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes)
321 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE
322 * aligned boundaries, if the range is not aligned. As a result a
323 * DEV_BSIZE subrange with partially dirty data may get marked as clean.
324 * It may happen that all DEV_BSIZE subranges are marked clean and thus
325 * the whole page would be considered clean despite have some
327 * For this reason we should shrink the range to DEV_BSIZE aligned
328 * boundaries before calling vm_page_clear_dirty.
330 end = rounddown2(off + nbytes, DEV_BSIZE);
331 off = roundup2(off, DEV_BSIZE);
335 zfs_vmobject_assert_wlocked_12(obj);
336 #if __FreeBSD_version < 1300050
338 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
340 if (vm_page_xbusied(pp)) {
342 * Reference the page before unlocking and
343 * sleeping so that the page daemon is less
344 * likely to reclaim it.
346 vm_page_reference(pp);
348 zfs_vmobject_wunlock(obj);
349 vm_page_busy_sleep(pp, "zfsmwb", true);
350 zfs_vmobject_wlock(obj);
354 } else if (pp != NULL) {
359 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
360 vm_object_pip_add(obj, 1);
361 pmap_remove_write(pp);
363 vm_page_clear_dirty(pp, off, nbytes);
368 vm_page_grab_valid_unlocked(&pp, obj, OFF_TO_IDX(start),
369 VM_ALLOC_NOCREAT | VM_ALLOC_SBUSY | VM_ALLOC_NORMAL |
372 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
373 vm_object_pip_add(obj, 1);
374 pmap_remove_write(pp);
376 vm_page_clear_dirty(pp, off, nbytes);
383 page_unbusy(vm_page_t pp)
387 #if __FreeBSD_version >= 1300041
388 vm_object_pip_wakeup(pp->object);
390 vm_object_pip_subtract(pp->object, 1);
394 #if __FreeBSD_version > 1300051
396 page_hold(vnode_t *vp, int64_t start)
402 vm_page_grab_valid_unlocked(&m, obj, OFF_TO_IDX(start),
403 VM_ALLOC_NOCREAT | VM_ALLOC_WIRED | VM_ALLOC_IGN_SBUSY |
409 page_hold(vnode_t *vp, int64_t start)
415 zfs_vmobject_assert_wlocked(obj);
418 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
420 if (vm_page_xbusied(pp)) {
422 * Reference the page before unlocking and
423 * sleeping so that the page daemon is less
424 * likely to reclaim it.
426 vm_page_reference(pp);
428 zfs_vmobject_wunlock(obj);
429 vm_page_busy_sleep(pp, "zfsmwb", true);
430 zfs_vmobject_wlock(obj);
434 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
435 vm_page_wire_lock(pp);
437 vm_page_wire_unlock(pp);
448 page_unhold(vm_page_t pp)
451 vm_page_wire_lock(pp);
452 #if __FreeBSD_version >= 1300035
453 vm_page_unwire(pp, PQ_ACTIVE);
457 vm_page_wire_unlock(pp);
461 * When a file is memory mapped, we must keep the IO data synchronized
462 * between the DMU cache and the memory mapped pages. What this means:
464 * On Write: If we find a memory mapped page, we write to *both*
465 * the page and the dmu buffer.
468 update_pages(znode_t *zp, int64_t start, int len, objset_t *os)
472 vnode_t *vp = ZTOV(zp);
476 ASSERT3P(vp->v_mount, !=, NULL);
478 ASSERT3P(obj, !=, NULL);
480 off = start & PAGEOFFSET;
481 zfs_vmobject_wlock_12(obj);
482 #if __FreeBSD_version >= 1300041
483 vm_object_pip_add(obj, 1);
485 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
487 int nbytes = imin(PAGESIZE - off, len);
489 if ((pp = page_busy(vp, start, off, nbytes)) != NULL) {
490 zfs_vmobject_wunlock_12(obj);
492 va = zfs_map_page(pp, &sf);
493 (void) dmu_read(os, zp->z_id, start + off, nbytes,
494 va + off, DMU_READ_PREFETCH);
497 zfs_vmobject_wlock_12(obj);
503 #if __FreeBSD_version >= 1300041
504 vm_object_pip_wakeup(obj);
506 vm_object_pip_wakeupn(obj, 0);
508 zfs_vmobject_wunlock_12(obj);
512 * Read with UIO_NOCOPY flag means that sendfile(2) requests
513 * ZFS to populate a range of page cache pages with data.
515 * NOTE: this function could be optimized to pre-allocate
516 * all pages in advance, drain exclusive busy on all of them,
517 * map them into contiguous KVA region and populate them
518 * in one single dmu_read() call.
521 mappedread_sf(znode_t *zp, int nbytes, zfs_uio_t *uio)
523 vnode_t *vp = ZTOV(zp);
524 objset_t *os = zp->z_zfsvfs->z_os;
533 ASSERT3U(zfs_uio_segflg(uio), ==, UIO_NOCOPY);
534 ASSERT3P(vp->v_mount, !=, NULL);
536 ASSERT3P(obj, !=, NULL);
537 ASSERT0(zfs_uio_offset(uio) & PAGEOFFSET);
539 zfs_vmobject_wlock_12(obj);
540 for (start = zfs_uio_offset(uio); len > 0; start += PAGESIZE) {
541 int bytes = MIN(PAGESIZE, len);
543 pp = vm_page_grab_unlocked(obj, OFF_TO_IDX(start),
544 VM_ALLOC_SBUSY | VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY);
545 if (vm_page_none_valid(pp)) {
546 zfs_vmobject_wunlock_12(obj);
547 va = zfs_map_page(pp, &sf);
548 error = dmu_read(os, zp->z_id, start, bytes, va,
550 if (bytes != PAGESIZE && error == 0)
551 bzero(va + bytes, PAGESIZE - bytes);
553 zfs_vmobject_wlock_12(obj);
554 #if __FreeBSD_version >= 1300081
557 vm_page_activate(pp);
558 vm_page_do_sunbusy(pp);
560 zfs_vmobject_wlock(obj);
561 if (!vm_page_wired(pp) && pp->valid == 0 &&
562 vm_page_busy_tryupgrade(pp))
566 zfs_vmobject_wunlock(obj);
569 vm_page_do_sunbusy(pp);
572 if (pp->wire_count == 0 && pp->valid == 0 &&
576 pp->valid = VM_PAGE_BITS_ALL;
577 vm_page_activate(pp);
582 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
583 vm_page_do_sunbusy(pp);
587 zfs_uio_advance(uio, bytes);
590 zfs_vmobject_wunlock_12(obj);
595 * When a file is memory mapped, we must keep the IO data synchronized
596 * between the DMU cache and the memory mapped pages. What this means:
598 * On Read: We "read" preferentially from memory mapped pages,
599 * else we default from the dmu buffer.
601 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
602 * the file is memory mapped.
605 mappedread(znode_t *zp, int nbytes, zfs_uio_t *uio)
607 vnode_t *vp = ZTOV(zp);
614 ASSERT3P(vp->v_mount, !=, NULL);
616 ASSERT3P(obj, !=, NULL);
618 start = zfs_uio_offset(uio);
619 off = start & PAGEOFFSET;
620 zfs_vmobject_wlock_12(obj);
621 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
623 uint64_t bytes = MIN(PAGESIZE - off, len);
625 if ((pp = page_hold(vp, start))) {
629 zfs_vmobject_wunlock_12(obj);
630 va = zfs_map_page(pp, &sf);
631 error = vn_io_fault_uiomove(va + off, bytes,
632 GET_UIO_STRUCT(uio));
634 zfs_vmobject_wlock_12(obj);
637 zfs_vmobject_wunlock_12(obj);
638 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
640 zfs_vmobject_wlock_12(obj);
647 zfs_vmobject_wunlock_12(obj);
652 zfs_write_simple(znode_t *zp, const void *data, size_t len,
653 loff_t pos, size_t *presid)
658 error = vn_rdwr(UIO_WRITE, ZTOV(zp), __DECONST(void *, data), len, pos,
659 UIO_SYSSPACE, IO_SYNC, kcred, NOCRED, &resid, curthread);
662 return (SET_ERROR(error));
663 } else if (presid == NULL) {
665 error = SET_ERROR(EIO);
674 zfs_zrele_async(znode_t *zp)
676 vnode_t *vp = ZTOV(zp);
677 objset_t *os = ITOZSB(vp)->z_os;
679 VN_RELE_ASYNC(vp, dsl_pool_zrele_taskq(dmu_objset_pool(os)));
683 zfs_dd_callback(struct mount *mp, void *arg, int lkflags, struct vnode **vpp)
688 error = vn_lock(*vpp, lkflags);
695 zfs_lookup_lock(vnode_t *dvp, vnode_t *vp, const char *name, int lkflags)
697 znode_t *zdp = VTOZ(dvp);
698 zfsvfs_t *zfsvfs __unused = zdp->z_zfsvfs;
702 if (zfsvfs->z_replay == B_FALSE)
703 ASSERT_VOP_LOCKED(dvp, __func__);
705 if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) {
706 ASSERT3P(dvp, ==, vp);
708 ltype = lkflags & LK_TYPE_MASK;
709 if (ltype != VOP_ISLOCKED(dvp)) {
710 if (ltype == LK_EXCLUSIVE)
711 vn_lock(dvp, LK_UPGRADE | LK_RETRY);
712 else /* if (ltype == LK_SHARED) */
713 vn_lock(dvp, LK_DOWNGRADE | LK_RETRY);
716 * Relock for the "." case could leave us with
719 if (VN_IS_DOOMED(dvp)) {
721 return (SET_ERROR(ENOENT));
725 } else if (name[0] == '.' && name[1] == '.' && name[2] == 0) {
727 * Note that in this case, dvp is the child vnode, and we
728 * are looking up the parent vnode - exactly reverse from
729 * normal operation. Unlocking dvp requires some rather
730 * tricky unlock/relock dance to prevent mp from being freed;
731 * use vn_vget_ino_gen() which takes care of all that.
733 * XXX Note that there is a time window when both vnodes are
734 * unlocked. It is possible, although highly unlikely, that
735 * during that window the parent-child relationship between
736 * the vnodes may change, for example, get reversed.
737 * In that case we would have a wrong lock order for the vnodes.
738 * All other filesystems seem to ignore this problem, so we
740 * A potential solution could be implemented as follows:
741 * - using LK_NOWAIT when locking the second vnode and retrying
743 * - checking that the parent-child relationship still holds
744 * after locking both vnodes and retrying if it doesn't
746 error = vn_vget_ino_gen(dvp, zfs_dd_callback, vp, lkflags, &vp);
749 error = vn_lock(vp, lkflags);
757 * Lookup an entry in a directory, or an extended attribute directory.
758 * If it exists, return a held vnode reference for it.
760 * IN: dvp - vnode of directory to search.
761 * nm - name of entry to lookup.
762 * pnp - full pathname to lookup [UNUSED].
763 * flags - LOOKUP_XATTR set if looking for an attribute.
764 * rdir - root directory vnode [UNUSED].
765 * cr - credentials of caller.
766 * ct - caller context
768 * OUT: vpp - vnode of located entry, NULL if not found.
770 * RETURN: 0 on success, error code on failure.
777 zfs_lookup(vnode_t *dvp, const char *nm, vnode_t **vpp,
778 struct componentname *cnp, int nameiop, cred_t *cr, kthread_t *td,
779 int flags, boolean_t cached)
781 znode_t *zdp = VTOZ(dvp);
783 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
784 #if __FreeBSD_version > 1300124
790 * Fast path lookup, however we must skip DNLC lookup
791 * for case folding or normalizing lookups because the
792 * DNLC code only stores the passed in name. This means
793 * creating 'a' and removing 'A' on a case insensitive
794 * file system would work, but DNLC still thinks 'a'
795 * exists and won't let you create it again on the next
796 * pass through fast path.
798 if (!(flags & LOOKUP_XATTR)) {
799 if (dvp->v_type != VDIR) {
800 return (SET_ERROR(ENOTDIR));
801 } else if (zdp->z_sa_hdl == NULL) {
802 return (SET_ERROR(EIO));
806 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp,
812 #if __FreeBSD_version > 1300124
813 dvp_seqc = vn_seqc_read_notmodify(dvp);
818 if (flags & LOOKUP_XATTR) {
820 * If the xattr property is off, refuse the lookup request.
822 if (!(zfsvfs->z_flags & ZSB_XATTR)) {
824 return (SET_ERROR(EOPNOTSUPP));
828 * We don't allow recursive attributes..
829 * Maybe someday we will.
831 if (zdp->z_pflags & ZFS_XATTR) {
833 return (SET_ERROR(EINVAL));
836 if ((error = zfs_get_xattrdir(VTOZ(dvp), &zp, cr, flags))) {
843 * Do we have permission to get into attribute directory?
845 error = zfs_zaccess(zp, ACE_EXECUTE, 0, B_FALSE, cr);
855 * Check accessibility of directory if we're not coming in via
860 if ((cnp->cn_flags & NOEXECCHECK) != 0) {
861 cnp->cn_flags &= ~NOEXECCHECK;
864 if ((error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr))) {
870 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
871 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
873 return (SET_ERROR(EILSEQ));
878 * First handle the special cases.
880 if ((cnp->cn_flags & ISDOTDOT) != 0) {
882 * If we are a snapshot mounted under .zfs, return
883 * the vp for the snapshot directory.
885 if (zdp->z_id == zfsvfs->z_root && zfsvfs->z_parent != zfsvfs) {
886 struct componentname cn;
891 ltype = VOP_ISLOCKED(dvp);
893 error = zfsctl_root(zfsvfs->z_parent, LK_SHARED,
896 cn.cn_nameptr = "snapshot";
897 cn.cn_namelen = strlen(cn.cn_nameptr);
898 cn.cn_nameiop = cnp->cn_nameiop;
899 cn.cn_flags = cnp->cn_flags & ~ISDOTDOT;
900 cn.cn_lkflags = cnp->cn_lkflags;
901 error = VOP_LOOKUP(zfsctl_vp, vpp, &cn);
904 vn_lock(dvp, ltype | LK_RETRY);
908 if (zfs_has_ctldir(zdp) && strcmp(nm, ZFS_CTLDIR_NAME) == 0) {
910 if ((cnp->cn_flags & ISLASTCN) != 0 && nameiop != LOOKUP)
911 return (SET_ERROR(ENOTSUP));
912 error = zfsctl_root(zfsvfs, cnp->cn_lkflags, vpp);
917 * The loop is retry the lookup if the parent-child relationship
918 * changes during the dot-dot locking complexities.
923 error = zfs_dirlook(zdp, nm, &zp);
931 error = zfs_lookup_lock(dvp, *vpp, nm, cnp->cn_lkflags);
934 * If we've got a locking error, then the vnode
935 * got reclaimed because of a force unmount.
936 * We never enter doomed vnodes into the name cache.
942 if ((cnp->cn_flags & ISDOTDOT) == 0)
946 if (zdp->z_sa_hdl == NULL) {
947 error = SET_ERROR(EIO);
949 error = sa_lookup(zdp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
950 &parent, sizeof (parent));
957 if (zp->z_id == parent) {
967 /* Translate errors and add SAVENAME when needed. */
968 if (cnp->cn_flags & ISLASTCN) {
972 if (error == ENOENT) {
974 cnp->cn_flags |= SAVENAME;
980 cnp->cn_flags |= SAVENAME;
985 #if __FreeBSD_version > 1300124
986 if ((cnp->cn_flags & ISDOTDOT) != 0) {
988 * FIXME: zfs_lookup_lock relocks vnodes and does nothing to
989 * handle races. In particular different callers may end up
990 * with different vnodes and will try to add conflicting
991 * entries to the namecache.
993 * While finding different result may be acceptable in face
994 * of concurrent modification, adding conflicting entries
995 * trips over an assert in the namecache.
997 * Ultimately let an entry through once everything settles.
999 if (!vn_seqc_consistent(dvp, dvp_seqc)) {
1000 cnp->cn_flags &= ~MAKEENTRY;
1005 /* Insert name into cache (as non-existent) if appropriate. */
1006 if (zfsvfs->z_use_namecache && !zfsvfs->z_replay &&
1007 error == ENOENT && (cnp->cn_flags & MAKEENTRY) != 0)
1008 cache_enter(dvp, NULL, cnp);
1010 /* Insert name into cache if appropriate. */
1011 if (zfsvfs->z_use_namecache && !zfsvfs->z_replay &&
1012 error == 0 && (cnp->cn_flags & MAKEENTRY)) {
1013 if (!(cnp->cn_flags & ISLASTCN) ||
1014 (nameiop != DELETE && nameiop != RENAME)) {
1015 cache_enter(dvp, *vpp, cnp);
1023 * Attempt to create a new entry in a directory. If the entry
1024 * already exists, truncate the file if permissible, else return
1025 * an error. Return the vp of the created or trunc'd file.
1027 * IN: dvp - vnode of directory to put new file entry in.
1028 * name - name of new file entry.
1029 * vap - attributes of new file.
1030 * excl - flag indicating exclusive or non-exclusive mode.
1031 * mode - mode to open file with.
1032 * cr - credentials of caller.
1033 * flag - large file flag [UNUSED].
1034 * ct - caller context
1035 * vsecp - ACL to be set
1037 * OUT: vpp - vnode of created or trunc'd entry.
1039 * RETURN: 0 on success, error code on failure.
1042 * dvp - ctime|mtime updated if new entry created
1043 * vp - ctime|mtime always, atime if new
1048 zfs_create(znode_t *dzp, const char *name, vattr_t *vap, int excl, int mode,
1049 znode_t **zpp, cred_t *cr, int flag, vsecattr_t *vsecp)
1052 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1059 gid_t gid = crgetgid(cr);
1060 uint64_t projid = ZFS_DEFAULT_PROJID;
1061 zfs_acl_ids_t acl_ids;
1062 boolean_t fuid_dirtied;
1064 #ifdef DEBUG_VFS_LOCKS
1065 vnode_t *dvp = ZTOV(dzp);
1069 * If we have an ephemeral id, ACL, or XVATTR then
1070 * make sure file system is at proper version
1073 ksid = crgetsid(cr, KSID_OWNER);
1075 uid = ksid_getid(ksid);
1079 if (zfsvfs->z_use_fuids == B_FALSE &&
1080 (vsecp || (vap->va_mask & AT_XVATTR) ||
1081 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1082 return (SET_ERROR(EINVAL));
1087 zilog = zfsvfs->z_log;
1089 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1090 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1092 return (SET_ERROR(EILSEQ));
1095 if (vap->va_mask & AT_XVATTR) {
1096 if ((error = secpolicy_xvattr(ZTOV(dzp), (xvattr_t *)vap,
1097 crgetuid(cr), cr, vap->va_type)) != 0) {
1105 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr))
1106 vap->va_mode &= ~S_ISVTX;
1108 error = zfs_dirent_lookup(dzp, name, &zp, ZNEW);
1113 ASSERT3P(zp, ==, NULL);
1116 * Create a new file object and update the directory
1119 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
1124 * We only support the creation of regular files in
1125 * extended attribute directories.
1128 if ((dzp->z_pflags & ZFS_XATTR) &&
1129 (vap->va_type != VREG)) {
1130 error = SET_ERROR(EINVAL);
1134 if ((error = zfs_acl_ids_create(dzp, 0, vap,
1135 cr, vsecp, &acl_ids)) != 0)
1138 if (S_ISREG(vap->va_mode) || S_ISDIR(vap->va_mode))
1139 projid = zfs_inherit_projid(dzp);
1140 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, projid)) {
1141 zfs_acl_ids_free(&acl_ids);
1142 error = SET_ERROR(EDQUOT);
1146 getnewvnode_reserve_();
1148 tx = dmu_tx_create(os);
1150 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1151 ZFS_SA_BASE_ATTR_SIZE);
1153 fuid_dirtied = zfsvfs->z_fuid_dirty;
1155 zfs_fuid_txhold(zfsvfs, tx);
1156 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1157 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1158 if (!zfsvfs->z_use_sa &&
1159 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1160 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1161 0, acl_ids.z_aclp->z_acl_bytes);
1163 error = dmu_tx_assign(tx, TXG_WAIT);
1165 zfs_acl_ids_free(&acl_ids);
1167 getnewvnode_drop_reserve();
1171 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1173 zfs_fuid_sync(zfsvfs, tx);
1175 (void) zfs_link_create(dzp, name, zp, tx, ZNEW);
1176 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1177 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1178 vsecp, acl_ids.z_fuidp, vap);
1179 zfs_acl_ids_free(&acl_ids);
1182 getnewvnode_drop_reserve();
1190 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1191 zil_commit(zilog, 0);
1198 * Remove an entry from a directory.
1200 * IN: dvp - vnode of directory to remove entry from.
1201 * name - name of entry to remove.
1202 * cr - credentials of caller.
1203 * ct - caller context
1204 * flags - case flags
1206 * RETURN: 0 on success, error code on failure.
1210 * vp - ctime (if nlink > 0)
1215 zfs_remove_(vnode_t *dvp, vnode_t *vp, const char *name, cred_t *cr)
1217 znode_t *dzp = VTOZ(dvp);
1220 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1234 zilog = zfsvfs->z_log;
1239 if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
1244 * Need to use rmdir for removing directories.
1246 if (vp->v_type == VDIR) {
1247 error = SET_ERROR(EPERM);
1251 vnevent_remove(vp, dvp, name, ct);
1255 /* are there any extended attributes? */
1256 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1257 &xattr_obj, sizeof (xattr_obj));
1258 if (error == 0 && xattr_obj) {
1259 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1264 * We may delete the znode now, or we may put it in the unlinked set;
1265 * it depends on whether we're the last link, and on whether there are
1266 * other holds on the vnode. So we dmu_tx_hold() the right things to
1267 * allow for either case.
1269 tx = dmu_tx_create(zfsvfs->z_os);
1270 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1271 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1272 zfs_sa_upgrade_txholds(tx, zp);
1273 zfs_sa_upgrade_txholds(tx, dzp);
1276 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1277 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1280 /* charge as an update -- would be nice not to charge at all */
1281 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1284 * Mark this transaction as typically resulting in a net free of space
1286 dmu_tx_mark_netfree(tx);
1288 error = dmu_tx_assign(tx, TXG_WAIT);
1296 * Remove the directory entry.
1298 error = zfs_link_destroy(dzp, name, zp, tx, ZEXISTS, &unlinked);
1306 zfs_unlinked_add(zp, tx);
1307 vp->v_vflag |= VV_NOSYNC;
1309 /* XXX check changes to linux vnops */
1311 zfs_log_remove(zilog, tx, txtype, dzp, name, obj, unlinked);
1319 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1320 zil_commit(zilog, 0);
1329 zfs_lookup_internal(znode_t *dzp, const char *name, vnode_t **vpp,
1330 struct componentname *cnp, int nameiop)
1332 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1335 cnp->cn_nameptr = __DECONST(char *, name);
1336 cnp->cn_namelen = strlen(name);
1337 cnp->cn_nameiop = nameiop;
1338 cnp->cn_flags = ISLASTCN | SAVENAME;
1339 cnp->cn_lkflags = LK_EXCLUSIVE | LK_RETRY;
1340 cnp->cn_cred = kcred;
1341 cnp->cn_thread = curthread;
1343 if (zfsvfs->z_use_namecache && !zfsvfs->z_replay) {
1344 struct vop_lookup_args a;
1346 a.a_gen.a_desc = &vop_lookup_desc;
1347 a.a_dvp = ZTOV(dzp);
1350 error = vfs_cache_lookup(&a);
1352 error = zfs_lookup(ZTOV(dzp), name, vpp, cnp, nameiop, kcred,
1353 curthread, 0, B_FALSE);
1357 printf("got error %d on name %s on op %d\n", error, name,
1366 zfs_remove(znode_t *dzp, const char *name, cred_t *cr, int flags)
1370 struct componentname cn;
1372 if ((error = zfs_lookup_internal(dzp, name, &vp, &cn, DELETE)))
1375 error = zfs_remove_(ZTOV(dzp), vp, name, cr);
1380 * Create a new directory and insert it into dvp using the name
1381 * provided. Return a pointer to the inserted directory.
1383 * IN: dvp - vnode of directory to add subdir to.
1384 * dirname - name of new directory.
1385 * vap - attributes of new directory.
1386 * cr - credentials of caller.
1387 * ct - caller context
1388 * flags - case flags
1389 * vsecp - ACL to be set
1391 * OUT: vpp - vnode of created directory.
1393 * RETURN: 0 on success, error code on failure.
1396 * dvp - ctime|mtime updated
1397 * vp - ctime|mtime|atime updated
1401 zfs_mkdir(znode_t *dzp, const char *dirname, vattr_t *vap, znode_t **zpp,
1402 cred_t *cr, int flags, vsecattr_t *vsecp)
1405 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1412 gid_t gid = crgetgid(cr);
1413 zfs_acl_ids_t acl_ids;
1414 boolean_t fuid_dirtied;
1416 ASSERT3U(vap->va_type, ==, VDIR);
1419 * If we have an ephemeral id, ACL, or XVATTR then
1420 * make sure file system is at proper version
1423 ksid = crgetsid(cr, KSID_OWNER);
1425 uid = ksid_getid(ksid);
1428 if (zfsvfs->z_use_fuids == B_FALSE &&
1429 ((vap->va_mask & AT_XVATTR) ||
1430 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1431 return (SET_ERROR(EINVAL));
1435 zilog = zfsvfs->z_log;
1437 if (dzp->z_pflags & ZFS_XATTR) {
1439 return (SET_ERROR(EINVAL));
1442 if (zfsvfs->z_utf8 && u8_validate(dirname,
1443 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1445 return (SET_ERROR(EILSEQ));
1448 if (vap->va_mask & AT_XVATTR) {
1449 if ((error = secpolicy_xvattr(ZTOV(dzp), (xvattr_t *)vap,
1450 crgetuid(cr), cr, vap->va_type)) != 0) {
1456 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1457 NULL, &acl_ids)) != 0) {
1463 * First make sure the new directory doesn't exist.
1465 * Existence is checked first to make sure we don't return
1466 * EACCES instead of EEXIST which can cause some applications
1471 if ((error = zfs_dirent_lookup(dzp, dirname, &zp, ZNEW))) {
1472 zfs_acl_ids_free(&acl_ids);
1476 ASSERT3P(zp, ==, NULL);
1478 if ((error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr))) {
1479 zfs_acl_ids_free(&acl_ids);
1484 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, zfs_inherit_projid(dzp))) {
1485 zfs_acl_ids_free(&acl_ids);
1487 return (SET_ERROR(EDQUOT));
1491 * Add a new entry to the directory.
1493 getnewvnode_reserve_();
1494 tx = dmu_tx_create(zfsvfs->z_os);
1495 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1496 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1497 fuid_dirtied = zfsvfs->z_fuid_dirty;
1499 zfs_fuid_txhold(zfsvfs, tx);
1500 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1501 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1502 acl_ids.z_aclp->z_acl_bytes);
1505 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1506 ZFS_SA_BASE_ATTR_SIZE);
1508 error = dmu_tx_assign(tx, TXG_WAIT);
1510 zfs_acl_ids_free(&acl_ids);
1512 getnewvnode_drop_reserve();
1520 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1523 zfs_fuid_sync(zfsvfs, tx);
1526 * Now put new name in parent dir.
1528 (void) zfs_link_create(dzp, dirname, zp, tx, ZNEW);
1532 txtype = zfs_log_create_txtype(Z_DIR, NULL, vap);
1533 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, NULL,
1534 acl_ids.z_fuidp, vap);
1536 zfs_acl_ids_free(&acl_ids);
1540 getnewvnode_drop_reserve();
1542 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1543 zil_commit(zilog, 0);
1549 #if __FreeBSD_version < 1300124
1551 cache_vop_rmdir(struct vnode *dvp, struct vnode *vp)
1560 * Remove a directory subdir entry. If the current working
1561 * directory is the same as the subdir to be removed, the
1564 * IN: dvp - vnode of directory to remove from.
1565 * name - name of directory to be removed.
1566 * cwd - vnode of current working directory.
1567 * cr - credentials of caller.
1568 * ct - caller context
1569 * flags - case flags
1571 * RETURN: 0 on success, error code on failure.
1574 * dvp - ctime|mtime updated
1578 zfs_rmdir_(vnode_t *dvp, vnode_t *vp, const char *name, cred_t *cr)
1580 znode_t *dzp = VTOZ(dvp);
1581 znode_t *zp = VTOZ(vp);
1582 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1590 zilog = zfsvfs->z_log;
1593 if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
1597 if (vp->v_type != VDIR) {
1598 error = SET_ERROR(ENOTDIR);
1602 vnevent_rmdir(vp, dvp, name, ct);
1604 tx = dmu_tx_create(zfsvfs->z_os);
1605 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1606 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1607 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1608 zfs_sa_upgrade_txholds(tx, zp);
1609 zfs_sa_upgrade_txholds(tx, dzp);
1610 dmu_tx_mark_netfree(tx);
1611 error = dmu_tx_assign(tx, TXG_WAIT);
1618 error = zfs_link_destroy(dzp, name, zp, tx, ZEXISTS, NULL);
1621 uint64_t txtype = TX_RMDIR;
1622 zfs_log_remove(zilog, tx, txtype, dzp, name,
1623 ZFS_NO_OBJECT, B_FALSE);
1628 cache_vop_rmdir(dvp, vp);
1630 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1631 zil_commit(zilog, 0);
1638 zfs_rmdir(znode_t *dzp, const char *name, znode_t *cwd, cred_t *cr, int flags)
1640 struct componentname cn;
1644 if ((error = zfs_lookup_internal(dzp, name, &vp, &cn, DELETE)))
1647 error = zfs_rmdir_(ZTOV(dzp), vp, name, cr);
1653 * Read as many directory entries as will fit into the provided
1654 * buffer from the given directory cursor position (specified in
1655 * the uio structure).
1657 * IN: vp - vnode of directory to read.
1658 * uio - structure supplying read location, range info,
1659 * and return buffer.
1660 * cr - credentials of caller.
1661 * ct - caller context
1662 * flags - case flags
1664 * OUT: uio - updated offset and range, buffer filled.
1665 * eofp - set to true if end-of-file detected.
1667 * RETURN: 0 on success, error code on failure.
1670 * vp - atime updated
1672 * Note that the low 4 bits of the cookie returned by zap is always zero.
1673 * This allows us to use the low range for "special" directory entries:
1674 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
1675 * we use the offset 2 for the '.zfs' directory.
1679 zfs_readdir(vnode_t *vp, zfs_uio_t *uio, cred_t *cr, int *eofp,
1680 int *ncookies, ulong_t **cookies)
1682 znode_t *zp = VTOZ(vp);
1686 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1691 zap_attribute_t zap;
1692 uint_t bytes_wanted;
1693 uint64_t offset; /* must be unsigned; checks for < 1 */
1699 boolean_t check_sysattrs;
1702 ulong_t *cooks = NULL;
1708 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
1709 &parent, sizeof (parent))) != 0) {
1715 * If we are not given an eof variable,
1722 * Check for valid iov_len.
1724 if (GET_UIO_STRUCT(uio)->uio_iov->iov_len <= 0) {
1726 return (SET_ERROR(EINVAL));
1730 * Quit if directory has been removed (posix)
1732 if ((*eofp = zp->z_unlinked) != 0) {
1739 offset = zfs_uio_offset(uio);
1740 prefetch = zp->z_zn_prefetch;
1743 * Initialize the iterator cursor.
1747 * Start iteration from the beginning of the directory.
1749 zap_cursor_init(&zc, os, zp->z_id);
1752 * The offset is a serialized cursor.
1754 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
1758 * Get space to change directory entries into fs independent format.
1760 iovp = GET_UIO_STRUCT(uio)->uio_iov;
1761 bytes_wanted = iovp->iov_len;
1762 if (zfs_uio_segflg(uio) != UIO_SYSSPACE || zfs_uio_iovcnt(uio) != 1) {
1763 bufsize = bytes_wanted;
1764 outbuf = kmem_alloc(bufsize, KM_SLEEP);
1765 odp = (struct dirent64 *)outbuf;
1767 bufsize = bytes_wanted;
1769 odp = (struct dirent64 *)iovp->iov_base;
1771 eodp = (struct edirent *)odp;
1773 if (ncookies != NULL) {
1775 * Minimum entry size is dirent size and 1 byte for a file name.
1777 ncooks = zfs_uio_resid(uio) / (sizeof (struct dirent) -
1778 sizeof (((struct dirent *)NULL)->d_name) + 1);
1779 cooks = malloc(ncooks * sizeof (ulong_t), M_TEMP, M_WAITOK);
1784 * If this VFS supports the system attribute view interface; and
1785 * we're looking at an extended attribute directory; and we care
1786 * about normalization conflicts on this vfs; then we must check
1787 * for normalization conflicts with the sysattr name space.
1790 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
1791 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
1792 (flags & V_RDDIR_ENTFLAGS);
1798 * Transform to file-system independent format
1801 while (outcount < bytes_wanted) {
1804 off64_t *next = NULL;
1807 * Special case `.', `..', and `.zfs'.
1810 (void) strcpy(zap.za_name, ".");
1811 zap.za_normalization_conflict = 0;
1814 } else if (offset == 1) {
1815 (void) strcpy(zap.za_name, "..");
1816 zap.za_normalization_conflict = 0;
1819 } else if (offset == 2 && zfs_show_ctldir(zp)) {
1820 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
1821 zap.za_normalization_conflict = 0;
1822 objnum = ZFSCTL_INO_ROOT;
1828 if ((error = zap_cursor_retrieve(&zc, &zap))) {
1829 if ((*eofp = (error == ENOENT)) != 0)
1835 if (zap.za_integer_length != 8 ||
1836 zap.za_num_integers != 1) {
1837 cmn_err(CE_WARN, "zap_readdir: bad directory "
1838 "entry, obj = %lld, offset = %lld\n",
1839 (u_longlong_t)zp->z_id,
1840 (u_longlong_t)offset);
1841 error = SET_ERROR(ENXIO);
1845 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
1847 * MacOS X can extract the object type here such as:
1848 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
1850 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
1852 if (check_sysattrs && !zap.za_normalization_conflict) {
1854 zap.za_normalization_conflict =
1855 xattr_sysattr_casechk(zap.za_name);
1857 panic("%s:%u: TODO", __func__, __LINE__);
1862 if (flags & V_RDDIR_ACCFILTER) {
1864 * If we have no access at all, don't include
1865 * this entry in the returned information
1868 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
1870 if (!zfs_has_access(ezp, cr)) {
1877 if (flags & V_RDDIR_ENTFLAGS)
1878 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
1880 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
1883 * Will this entry fit in the buffer?
1885 if (outcount + reclen > bufsize) {
1887 * Did we manage to fit anything in the buffer?
1890 error = SET_ERROR(EINVAL);
1895 if (flags & V_RDDIR_ENTFLAGS) {
1897 * Add extended flag entry:
1899 eodp->ed_ino = objnum;
1900 eodp->ed_reclen = reclen;
1901 /* NOTE: ed_off is the offset for the *next* entry */
1902 next = &(eodp->ed_off);
1903 eodp->ed_eflags = zap.za_normalization_conflict ?
1904 ED_CASE_CONFLICT : 0;
1905 (void) strncpy(eodp->ed_name, zap.za_name,
1906 EDIRENT_NAMELEN(reclen));
1907 eodp = (edirent_t *)((intptr_t)eodp + reclen);
1912 odp->d_ino = objnum;
1913 odp->d_reclen = reclen;
1914 odp->d_namlen = strlen(zap.za_name);
1915 /* NOTE: d_off is the offset for the *next* entry. */
1917 strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1);
1919 dirent_terminate(odp);
1920 odp = (dirent64_t *)((intptr_t)odp + reclen);
1924 ASSERT3S(outcount, <=, bufsize);
1926 /* Prefetch znode */
1928 dmu_prefetch(os, objnum, 0, 0, 0,
1929 ZIO_PRIORITY_SYNC_READ);
1933 * Move to the next entry, fill in the previous offset.
1935 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
1936 zap_cursor_advance(&zc);
1937 offset = zap_cursor_serialize(&zc);
1942 /* Fill the offset right after advancing the cursor. */
1945 if (cooks != NULL) {
1948 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks));
1951 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
1953 /* Subtract unused cookies */
1954 if (ncookies != NULL)
1955 *ncookies -= ncooks;
1957 if (zfs_uio_segflg(uio) == UIO_SYSSPACE && zfs_uio_iovcnt(uio) == 1) {
1958 iovp->iov_base += outcount;
1959 iovp->iov_len -= outcount;
1960 zfs_uio_resid(uio) -= outcount;
1962 zfs_uiomove(outbuf, (long)outcount, UIO_READ, uio))) {
1964 * Reset the pointer.
1966 offset = zfs_uio_offset(uio);
1970 zap_cursor_fini(&zc);
1971 if (zfs_uio_segflg(uio) != UIO_SYSSPACE || zfs_uio_iovcnt(uio) != 1)
1972 kmem_free(outbuf, bufsize);
1974 if (error == ENOENT)
1977 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
1979 zfs_uio_setoffset(uio, offset);
1981 if (error != 0 && cookies != NULL) {
1982 free(*cookies, M_TEMP);
1990 * Get the requested file attributes and place them in the provided
1993 * IN: vp - vnode of file.
1994 * vap - va_mask identifies requested attributes.
1995 * If AT_XVATTR set, then optional attrs are requested
1996 * flags - ATTR_NOACLCHECK (CIFS server context)
1997 * cr - credentials of caller.
1999 * OUT: vap - attribute values.
2001 * RETURN: 0 (always succeeds).
2005 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr)
2007 znode_t *zp = VTOZ(vp);
2008 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2011 u_longlong_t nblocks;
2012 uint64_t mtime[2], ctime[2], crtime[2], rdev;
2013 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2014 xoptattr_t *xoap = NULL;
2015 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2016 sa_bulk_attr_t bulk[4];
2022 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2024 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2025 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2026 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
2027 if (vp->v_type == VBLK || vp->v_type == VCHR)
2028 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
2031 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2037 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2038 * Also, if we are the owner don't bother, since owner should
2039 * always be allowed to read basic attributes of file.
2041 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2042 (vap->va_uid != crgetuid(cr))) {
2043 if ((error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2051 * Return all attributes. It's cheaper to provide the answer
2052 * than to determine whether we were asked the question.
2055 vap->va_type = IFTOVT(zp->z_mode);
2056 vap->va_mode = zp->z_mode & ~S_IFMT;
2058 vap->va_nodeid = zp->z_id;
2059 vap->va_nlink = zp->z_links;
2060 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp) &&
2061 zp->z_links < ZFS_LINK_MAX)
2063 vap->va_size = zp->z_size;
2064 if (vp->v_type == VBLK || vp->v_type == VCHR)
2065 vap->va_rdev = zfs_cmpldev(rdev);
2066 vap->va_seq = zp->z_seq;
2067 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */
2068 vap->va_filerev = zp->z_seq;
2071 * Add in any requested optional attributes and the create time.
2072 * Also set the corresponding bits in the returned attribute bitmap.
2074 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2075 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2077 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2078 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2081 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2082 xoap->xoa_readonly =
2083 ((zp->z_pflags & ZFS_READONLY) != 0);
2084 XVA_SET_RTN(xvap, XAT_READONLY);
2087 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2089 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2090 XVA_SET_RTN(xvap, XAT_SYSTEM);
2093 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2095 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2096 XVA_SET_RTN(xvap, XAT_HIDDEN);
2099 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2100 xoap->xoa_nounlink =
2101 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2102 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2105 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2106 xoap->xoa_immutable =
2107 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2108 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2111 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2112 xoap->xoa_appendonly =
2113 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2114 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2117 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2119 ((zp->z_pflags & ZFS_NODUMP) != 0);
2120 XVA_SET_RTN(xvap, XAT_NODUMP);
2123 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2125 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2126 XVA_SET_RTN(xvap, XAT_OPAQUE);
2129 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2130 xoap->xoa_av_quarantined =
2131 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2132 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2135 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2136 xoap->xoa_av_modified =
2137 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2138 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2141 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2142 vp->v_type == VREG) {
2143 zfs_sa_get_scanstamp(zp, xvap);
2146 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2147 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2148 XVA_SET_RTN(xvap, XAT_REPARSE);
2150 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2151 xoap->xoa_generation = zp->z_gen;
2152 XVA_SET_RTN(xvap, XAT_GEN);
2155 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2157 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2158 XVA_SET_RTN(xvap, XAT_OFFLINE);
2161 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2163 ((zp->z_pflags & ZFS_SPARSE) != 0);
2164 XVA_SET_RTN(xvap, XAT_SPARSE);
2167 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) {
2168 xoap->xoa_projinherit =
2169 ((zp->z_pflags & ZFS_PROJINHERIT) != 0);
2170 XVA_SET_RTN(xvap, XAT_PROJINHERIT);
2173 if (XVA_ISSET_REQ(xvap, XAT_PROJID)) {
2174 xoap->xoa_projid = zp->z_projid;
2175 XVA_SET_RTN(xvap, XAT_PROJID);
2179 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2180 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2181 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2182 ZFS_TIME_DECODE(&vap->va_birthtime, crtime);
2185 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2186 vap->va_blksize = blksize;
2187 vap->va_bytes = nblocks << 9; /* nblocks * 512 */
2189 if (zp->z_blksz == 0) {
2191 * Block size hasn't been set; suggest maximal I/O transfers.
2193 vap->va_blksize = zfsvfs->z_max_blksz;
2201 * Set the file attributes to the values contained in the
2204 * IN: zp - znode of file to be modified.
2205 * vap - new attribute values.
2206 * If AT_XVATTR set, then optional attrs are being set
2207 * flags - ATTR_UTIME set if non-default time values provided.
2208 * - ATTR_NOACLCHECK (CIFS context only).
2209 * cr - credentials of caller.
2210 * ct - caller context
2212 * RETURN: 0 on success, error code on failure.
2215 * vp - ctime updated, mtime updated if size changed.
2219 zfs_setattr(znode_t *zp, vattr_t *vap, int flags, cred_t *cr)
2221 vnode_t *vp = ZTOV(zp);
2222 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2223 objset_t *os = zfsvfs->z_os;
2228 uint_t mask = vap->va_mask;
2229 uint_t saved_mask = 0;
2230 uint64_t saved_mode;
2233 uint64_t new_uid, new_gid;
2235 uint64_t mtime[2], ctime[2];
2236 uint64_t projid = ZFS_INVALID_PROJID;
2238 int need_policy = FALSE;
2240 zfs_fuid_info_t *fuidp = NULL;
2241 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2244 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2245 boolean_t fuid_dirtied = B_FALSE;
2246 sa_bulk_attr_t bulk[7], xattr_bulk[7];
2247 int count = 0, xattr_count = 0;
2252 if (mask & AT_NOSET)
2253 return (SET_ERROR(EINVAL));
2258 zilog = zfsvfs->z_log;
2261 * Make sure that if we have ephemeral uid/gid or xvattr specified
2262 * that file system is at proper version level
2265 if (zfsvfs->z_use_fuids == B_FALSE &&
2266 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2267 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2268 (mask & AT_XVATTR))) {
2270 return (SET_ERROR(EINVAL));
2273 if (mask & AT_SIZE && vp->v_type == VDIR) {
2275 return (SET_ERROR(EISDIR));
2278 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2280 return (SET_ERROR(EINVAL));
2284 * If this is an xvattr_t, then get a pointer to the structure of
2285 * optional attributes. If this is NULL, then we have a vattr_t.
2287 xoap = xva_getxoptattr(xvap);
2289 xva_init(&tmpxvattr);
2292 * Immutable files can only alter immutable bit and atime
2294 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2295 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2296 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2298 return (SET_ERROR(EPERM));
2302 * Note: ZFS_READONLY is handled in zfs_zaccess_common.
2306 * Verify timestamps doesn't overflow 32 bits.
2307 * ZFS can handle large timestamps, but 32bit syscalls can't
2308 * handle times greater than 2039. This check should be removed
2309 * once large timestamps are fully supported.
2311 if (mask & (AT_ATIME | AT_MTIME)) {
2312 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2313 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2315 return (SET_ERROR(EOVERFLOW));
2318 if (xoap != NULL && (mask & AT_XVATTR)) {
2319 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME) &&
2320 TIMESPEC_OVERFLOW(&vap->va_birthtime)) {
2322 return (SET_ERROR(EOVERFLOW));
2325 if (XVA_ISSET_REQ(xvap, XAT_PROJID)) {
2326 if (!dmu_objset_projectquota_enabled(os) ||
2327 (!S_ISREG(zp->z_mode) && !S_ISDIR(zp->z_mode))) {
2329 return (SET_ERROR(EOPNOTSUPP));
2332 projid = xoap->xoa_projid;
2333 if (unlikely(projid == ZFS_INVALID_PROJID)) {
2335 return (SET_ERROR(EINVAL));
2338 if (projid == zp->z_projid && zp->z_pflags & ZFS_PROJID)
2339 projid = ZFS_INVALID_PROJID;
2344 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT) &&
2345 (xoap->xoa_projinherit !=
2346 ((zp->z_pflags & ZFS_PROJINHERIT) != 0)) &&
2347 (!dmu_objset_projectquota_enabled(os) ||
2348 (!S_ISREG(zp->z_mode) && !S_ISDIR(zp->z_mode)))) {
2350 return (SET_ERROR(EOPNOTSUPP));
2357 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2359 return (SET_ERROR(EROFS));
2363 * First validate permissions
2366 if (mask & AT_SIZE) {
2368 * XXX - Note, we are not providing any open
2369 * mode flags here (like FNDELAY), so we may
2370 * block if there are locks present... this
2371 * should be addressed in openat().
2373 /* XXX - would it be OK to generate a log record here? */
2374 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2381 if (mask & (AT_ATIME|AT_MTIME) ||
2382 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2383 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2384 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2385 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2386 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2387 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2388 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2389 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2393 if (mask & (AT_UID|AT_GID)) {
2394 int idmask = (mask & (AT_UID|AT_GID));
2399 * NOTE: even if a new mode is being set,
2400 * we may clear S_ISUID/S_ISGID bits.
2403 if (!(mask & AT_MODE))
2404 vap->va_mode = zp->z_mode;
2407 * Take ownership or chgrp to group we are a member of
2410 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2411 take_group = (mask & AT_GID) &&
2412 zfs_groupmember(zfsvfs, vap->va_gid, cr);
2415 * If both AT_UID and AT_GID are set then take_owner and
2416 * take_group must both be set in order to allow taking
2419 * Otherwise, send the check through secpolicy_vnode_setattr()
2423 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2424 ((idmask == AT_UID) && take_owner) ||
2425 ((idmask == AT_GID) && take_group)) {
2426 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2427 skipaclchk, cr) == 0) {
2429 * Remove setuid/setgid for non-privileged users
2431 secpolicy_setid_clear(vap, vp, cr);
2432 trim_mask = (mask & (AT_UID|AT_GID));
2441 oldva.va_mode = zp->z_mode;
2442 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2443 if (mask & AT_XVATTR) {
2445 * Update xvattr mask to include only those attributes
2446 * that are actually changing.
2448 * the bits will be restored prior to actually setting
2449 * the attributes so the caller thinks they were set.
2451 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2452 if (xoap->xoa_appendonly !=
2453 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2456 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2457 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
2461 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) {
2462 if (xoap->xoa_projinherit !=
2463 ((zp->z_pflags & ZFS_PROJINHERIT) != 0)) {
2466 XVA_CLR_REQ(xvap, XAT_PROJINHERIT);
2467 XVA_SET_REQ(&tmpxvattr, XAT_PROJINHERIT);
2471 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2472 if (xoap->xoa_nounlink !=
2473 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2476 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2477 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
2481 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2482 if (xoap->xoa_immutable !=
2483 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2486 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2487 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
2491 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2492 if (xoap->xoa_nodump !=
2493 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2496 XVA_CLR_REQ(xvap, XAT_NODUMP);
2497 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
2501 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2502 if (xoap->xoa_av_modified !=
2503 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2506 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2507 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
2511 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2512 if ((vp->v_type != VREG &&
2513 xoap->xoa_av_quarantined) ||
2514 xoap->xoa_av_quarantined !=
2515 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2518 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2519 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
2523 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2525 return (SET_ERROR(EPERM));
2528 if (need_policy == FALSE &&
2529 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2530 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2535 if (mask & AT_MODE) {
2536 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2537 err = secpolicy_setid_setsticky_clear(vp, vap,
2543 trim_mask |= AT_MODE;
2551 * If trim_mask is set then take ownership
2552 * has been granted or write_acl is present and user
2553 * has the ability to modify mode. In that case remove
2554 * UID|GID and or MODE from mask so that
2555 * secpolicy_vnode_setattr() doesn't revoke it.
2559 saved_mask = vap->va_mask;
2560 vap->va_mask &= ~trim_mask;
2561 if (trim_mask & AT_MODE) {
2563 * Save the mode, as secpolicy_vnode_setattr()
2564 * will overwrite it with ova.va_mode.
2566 saved_mode = vap->va_mode;
2569 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2570 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2577 vap->va_mask |= saved_mask;
2578 if (trim_mask & AT_MODE) {
2580 * Recover the mode after
2581 * secpolicy_vnode_setattr().
2583 vap->va_mode = saved_mode;
2589 * secpolicy_vnode_setattr, or take ownership may have
2592 mask = vap->va_mask;
2594 if ((mask & (AT_UID | AT_GID)) || projid != ZFS_INVALID_PROJID) {
2595 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
2596 &xattr_obj, sizeof (xattr_obj));
2598 if (err == 0 && xattr_obj) {
2599 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
2601 err = vn_lock(ZTOV(attrzp), LK_EXCLUSIVE);
2603 vrele(ZTOV(attrzp));
2608 if (mask & AT_UID) {
2609 new_uid = zfs_fuid_create(zfsvfs,
2610 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
2611 if (new_uid != zp->z_uid &&
2612 zfs_id_overquota(zfsvfs, DMU_USERUSED_OBJECT,
2616 err = SET_ERROR(EDQUOT);
2621 if (mask & AT_GID) {
2622 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
2623 cr, ZFS_GROUP, &fuidp);
2624 if (new_gid != zp->z_gid &&
2625 zfs_id_overquota(zfsvfs, DMU_GROUPUSED_OBJECT,
2629 err = SET_ERROR(EDQUOT);
2634 if (projid != ZFS_INVALID_PROJID &&
2635 zfs_id_overquota(zfsvfs, DMU_PROJECTUSED_OBJECT, projid)) {
2638 err = SET_ERROR(EDQUOT);
2642 tx = dmu_tx_create(os);
2644 if (mask & AT_MODE) {
2645 uint64_t pmode = zp->z_mode;
2647 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2649 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
2650 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
2651 err = SET_ERROR(EPERM);
2655 if ((err = zfs_acl_chmod_setattr(zp, &aclp, new_mode)))
2658 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
2660 * Are we upgrading ACL from old V0 format
2663 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
2664 zfs_znode_acl_version(zp) ==
2665 ZFS_ACL_VERSION_INITIAL) {
2666 dmu_tx_hold_free(tx, acl_obj, 0,
2668 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2669 0, aclp->z_acl_bytes);
2671 dmu_tx_hold_write(tx, acl_obj, 0,
2674 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2675 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2676 0, aclp->z_acl_bytes);
2678 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2680 if (((mask & AT_XVATTR) &&
2681 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) ||
2682 (projid != ZFS_INVALID_PROJID &&
2683 !(zp->z_pflags & ZFS_PROJID)))
2684 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2686 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2690 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
2693 fuid_dirtied = zfsvfs->z_fuid_dirty;
2695 zfs_fuid_txhold(zfsvfs, tx);
2697 zfs_sa_upgrade_txholds(tx, zp);
2699 err = dmu_tx_assign(tx, TXG_WAIT);
2705 * Set each attribute requested.
2706 * We group settings according to the locks they need to acquire.
2708 * Note: you cannot set ctime directly, although it will be
2709 * updated as a side-effect of calling this function.
2712 if (projid != ZFS_INVALID_PROJID && !(zp->z_pflags & ZFS_PROJID)) {
2714 * For the existed object that is upgraded from old system,
2715 * its on-disk layout has no slot for the project ID attribute.
2716 * But quota accounting logic needs to access related slots by
2717 * offset directly. So we need to adjust old objects' layout
2718 * to make the project ID to some unified and fixed offset.
2721 err = sa_add_projid(attrzp->z_sa_hdl, tx, projid);
2723 err = sa_add_projid(zp->z_sa_hdl, tx, projid);
2725 if (unlikely(err == EEXIST))
2730 projid = ZFS_INVALID_PROJID;
2733 if (mask & (AT_UID|AT_GID|AT_MODE))
2734 mutex_enter(&zp->z_acl_lock);
2736 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
2737 &zp->z_pflags, sizeof (zp->z_pflags));
2740 if (mask & (AT_UID|AT_GID|AT_MODE))
2741 mutex_enter(&attrzp->z_acl_lock);
2742 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2743 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
2744 sizeof (attrzp->z_pflags));
2745 if (projid != ZFS_INVALID_PROJID) {
2746 attrzp->z_projid = projid;
2747 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2748 SA_ZPL_PROJID(zfsvfs), NULL, &attrzp->z_projid,
2749 sizeof (attrzp->z_projid));
2753 if (mask & (AT_UID|AT_GID)) {
2755 if (mask & AT_UID) {
2756 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
2757 &new_uid, sizeof (new_uid));
2758 zp->z_uid = new_uid;
2760 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2761 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
2763 attrzp->z_uid = new_uid;
2767 if (mask & AT_GID) {
2768 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
2769 NULL, &new_gid, sizeof (new_gid));
2770 zp->z_gid = new_gid;
2772 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2773 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
2775 attrzp->z_gid = new_gid;
2778 if (!(mask & AT_MODE)) {
2779 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
2780 NULL, &new_mode, sizeof (new_mode));
2781 new_mode = zp->z_mode;
2783 err = zfs_acl_chown_setattr(zp);
2786 vn_seqc_write_begin(ZTOV(attrzp));
2787 err = zfs_acl_chown_setattr(attrzp);
2788 vn_seqc_write_end(ZTOV(attrzp));
2793 if (mask & AT_MODE) {
2794 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
2795 &new_mode, sizeof (new_mode));
2796 zp->z_mode = new_mode;
2797 ASSERT3P(aclp, !=, NULL);
2798 err = zfs_aclset_common(zp, aclp, cr, tx);
2800 if (zp->z_acl_cached)
2801 zfs_acl_free(zp->z_acl_cached);
2802 zp->z_acl_cached = aclp;
2807 if (mask & AT_ATIME) {
2808 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
2809 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
2810 &zp->z_atime, sizeof (zp->z_atime));
2813 if (mask & AT_MTIME) {
2814 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
2815 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
2816 mtime, sizeof (mtime));
2819 if (projid != ZFS_INVALID_PROJID) {
2820 zp->z_projid = projid;
2821 SA_ADD_BULK_ATTR(bulk, count,
2822 SA_ZPL_PROJID(zfsvfs), NULL, &zp->z_projid,
2823 sizeof (zp->z_projid));
2826 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
2827 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
2828 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
2829 NULL, mtime, sizeof (mtime));
2830 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
2831 &ctime, sizeof (ctime));
2832 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime);
2833 } else if (mask != 0) {
2834 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
2835 &ctime, sizeof (ctime));
2836 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime);
2838 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2839 SA_ZPL_CTIME(zfsvfs), NULL,
2840 &ctime, sizeof (ctime));
2841 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
2847 * Do this after setting timestamps to prevent timestamp
2848 * update from toggling bit
2851 if (xoap && (mask & AT_XVATTR)) {
2853 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME))
2854 xoap->xoa_createtime = vap->va_birthtime;
2856 * restore trimmed off masks
2857 * so that return masks can be set for caller.
2860 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
2861 XVA_SET_REQ(xvap, XAT_APPENDONLY);
2863 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
2864 XVA_SET_REQ(xvap, XAT_NOUNLINK);
2866 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
2867 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
2869 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
2870 XVA_SET_REQ(xvap, XAT_NODUMP);
2872 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
2873 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
2875 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
2876 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
2878 if (XVA_ISSET_REQ(&tmpxvattr, XAT_PROJINHERIT)) {
2879 XVA_SET_REQ(xvap, XAT_PROJINHERIT);
2882 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
2883 ASSERT3S(vp->v_type, ==, VREG);
2885 zfs_xvattr_set(zp, xvap, tx);
2889 zfs_fuid_sync(zfsvfs, tx);
2892 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
2894 if (mask & (AT_UID|AT_GID|AT_MODE))
2895 mutex_exit(&zp->z_acl_lock);
2898 if (mask & (AT_UID|AT_GID|AT_MODE))
2899 mutex_exit(&attrzp->z_acl_lock);
2902 if (err == 0 && attrzp) {
2903 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
2915 zfs_fuid_info_free(fuidp);
2922 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
2927 if (os->os_sync == ZFS_SYNC_ALWAYS)
2928 zil_commit(zilog, 0);
2935 * We acquire all but fdvp locks using non-blocking acquisitions. If we
2936 * fail to acquire any lock in the path we will drop all held locks,
2937 * acquire the new lock in a blocking fashion, and then release it and
2938 * restart the rename. This acquire/release step ensures that we do not
2939 * spin on a lock waiting for release. On error release all vnode locks
2940 * and decrement references the way tmpfs_rename() would do.
2943 zfs_rename_relock(struct vnode *sdvp, struct vnode **svpp,
2944 struct vnode *tdvp, struct vnode **tvpp,
2945 const struct componentname *scnp, const struct componentname *tcnp)
2948 struct vnode *nvp, *svp, *tvp;
2949 znode_t *sdzp, *tdzp, *szp, *tzp;
2950 const char *snm = scnp->cn_nameptr;
2951 const char *tnm = tcnp->cn_nameptr;
2955 if (*tvpp != NULL && *tvpp != tdvp)
2959 error = vn_lock(sdvp, LK_EXCLUSIVE);
2964 error = vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT);
2969 error = vn_lock(tdvp, LK_EXCLUSIVE);
2978 * Before using sdzp and tdzp we must ensure that they are live.
2979 * As a porting legacy from illumos we have two things to worry
2980 * about. One is typical for FreeBSD and it is that the vnode is
2981 * not reclaimed (doomed). The other is that the znode is live.
2982 * The current code can invalidate the znode without acquiring the
2983 * corresponding vnode lock if the object represented by the znode
2984 * and vnode is no longer valid after a rollback or receive operation.
2985 * z_teardown_lock hidden behind ZFS_ENTER and ZFS_EXIT is the lock
2986 * that protects the znodes from the invalidation.
2988 zfsvfs = sdzp->z_zfsvfs;
2989 ASSERT3P(zfsvfs, ==, tdzp->z_zfsvfs);
2993 * We can not use ZFS_VERIFY_ZP() here because it could directly return
2994 * bypassing the cleanup code in the case of an error.
2996 if (tdzp->z_sa_hdl == NULL || sdzp->z_sa_hdl == NULL) {
3000 error = SET_ERROR(EIO);
3005 * Re-resolve svp to be certain it still exists and fetch the
3008 error = zfs_dirent_lookup(sdzp, snm, &szp, ZEXISTS);
3010 /* Source entry invalid or not there. */
3014 if ((scnp->cn_flags & ISDOTDOT) != 0 ||
3015 (scnp->cn_namelen == 1 && scnp->cn_nameptr[0] == '.'))
3016 error = SET_ERROR(EINVAL);
3022 * Re-resolve tvp, if it disappeared we just carry on.
3024 error = zfs_dirent_lookup(tdzp, tnm, &tzp, 0);
3030 if ((tcnp->cn_flags & ISDOTDOT) != 0)
3031 error = SET_ERROR(EINVAL);
3040 * At present the vnode locks must be acquired before z_teardown_lock,
3041 * although it would be more logical to use the opposite order.
3046 * Now try acquire locks on svp and tvp.
3049 error = vn_lock(nvp, LK_EXCLUSIVE | LK_NOWAIT);
3055 if (error != EBUSY) {
3059 error = vn_lock(nvp, LK_EXCLUSIVE);
3066 * Concurrent rename race.
3071 error = SET_ERROR(EINVAL);
3086 error = vn_lock(nvp, LK_EXCLUSIVE | LK_NOWAIT);
3091 if (error != EBUSY) {
3095 error = vn_lock(nvp, LK_EXCLUSIVE);
3113 * Note that we must use VRELE_ASYNC in this function as it walks
3114 * up the directory tree and vrele may need to acquire an exclusive
3115 * lock if a last reference to a vnode is dropped.
3118 zfs_rename_check(znode_t *szp, znode_t *sdzp, znode_t *tdzp)
3125 zfsvfs = tdzp->z_zfsvfs;
3127 return (SET_ERROR(EINVAL));
3130 if (tdzp->z_id == zfsvfs->z_root)
3134 ASSERT(!zp->z_unlinked);
3135 if ((error = sa_lookup(zp->z_sa_hdl,
3136 SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0)
3139 if (parent == szp->z_id) {
3140 error = SET_ERROR(EINVAL);
3143 if (parent == zfsvfs->z_root)
3145 if (parent == sdzp->z_id)
3148 error = zfs_zget(zfsvfs, parent, &zp1);
3153 VN_RELE_ASYNC(ZTOV(zp),
3154 dsl_pool_zrele_taskq(
3155 dmu_objset_pool(zfsvfs->z_os)));
3159 if (error == ENOTDIR)
3160 panic("checkpath: .. not a directory\n");
3162 VN_RELE_ASYNC(ZTOV(zp),
3163 dsl_pool_zrele_taskq(dmu_objset_pool(zfsvfs->z_os)));
3167 #if __FreeBSD_version < 1300124
3169 cache_vop_rename(struct vnode *fdvp, struct vnode *fvp, struct vnode *tdvp,
3170 struct vnode *tvp, struct componentname *fcnp, struct componentname *tcnp)
3176 cache_purge_negative(tdvp);
3181 * Move an entry from the provided source directory to the target
3182 * directory. Change the entry name as indicated.
3184 * IN: sdvp - Source directory containing the "old entry".
3185 * snm - Old entry name.
3186 * tdvp - Target directory to contain the "new entry".
3187 * tnm - New entry name.
3188 * cr - credentials of caller.
3189 * ct - caller context
3190 * flags - case flags
3192 * RETURN: 0 on success, error code on failure.
3195 * sdvp,tdvp - ctime|mtime updated
3199 zfs_rename_(vnode_t *sdvp, vnode_t **svpp, struct componentname *scnp,
3200 vnode_t *tdvp, vnode_t **tvpp, struct componentname *tcnp,
3201 cred_t *cr, int log)
3204 znode_t *sdzp, *tdzp, *szp, *tzp;
3205 zilog_t *zilog = NULL;
3207 const char *snm = scnp->cn_nameptr;
3208 const char *tnm = tcnp->cn_nameptr;
3210 bool want_seqc_end __maybe_unused = false;
3212 /* Reject renames across filesystems. */
3213 if ((*svpp)->v_mount != tdvp->v_mount ||
3214 ((*tvpp) != NULL && (*svpp)->v_mount != (*tvpp)->v_mount)) {
3215 error = SET_ERROR(EXDEV);
3219 if (zfsctl_is_node(tdvp)) {
3220 error = SET_ERROR(EXDEV);
3225 * Lock all four vnodes to ensure safety and semantics of renaming.
3227 error = zfs_rename_relock(sdvp, svpp, tdvp, tvpp, scnp, tcnp);
3229 /* no vnodes are locked in the case of error here */
3235 zfsvfs = tdzp->z_zfsvfs;
3236 zilog = zfsvfs->z_log;
3239 * After we re-enter ZFS_ENTER() we will have to revalidate all
3244 if (zfsvfs->z_utf8 && u8_validate(tnm,
3245 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3246 error = SET_ERROR(EILSEQ);
3250 /* If source and target are the same file, there is nothing to do. */
3251 if ((*svpp) == (*tvpp)) {
3256 if (((*svpp)->v_type == VDIR && (*svpp)->v_mountedhere != NULL) ||
3257 ((*tvpp) != NULL && (*tvpp)->v_type == VDIR &&
3258 (*tvpp)->v_mountedhere != NULL)) {
3259 error = SET_ERROR(EXDEV);
3264 * We can not use ZFS_VERIFY_ZP() here because it could directly return
3265 * bypassing the cleanup code in the case of an error.
3267 if (tdzp->z_sa_hdl == NULL || sdzp->z_sa_hdl == NULL) {
3268 error = SET_ERROR(EIO);
3273 tzp = *tvpp == NULL ? NULL : VTOZ(*tvpp);
3274 if (szp->z_sa_hdl == NULL || (tzp != NULL && tzp->z_sa_hdl == NULL)) {
3275 error = SET_ERROR(EIO);
3280 * This is to prevent the creation of links into attribute space
3281 * by renaming a linked file into/outof an attribute directory.
3282 * See the comment in zfs_link() for why this is considered bad.
3284 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3285 error = SET_ERROR(EINVAL);
3290 * If we are using project inheritance, means if the directory has
3291 * ZFS_PROJINHERIT set, then its descendant directories will inherit
3292 * not only the project ID, but also the ZFS_PROJINHERIT flag. Under
3293 * such case, we only allow renames into our tree when the project
3296 if (tdzp->z_pflags & ZFS_PROJINHERIT &&
3297 tdzp->z_projid != szp->z_projid) {
3298 error = SET_ERROR(EXDEV);
3303 * Must have write access at the source to remove the old entry
3304 * and write access at the target to create the new entry.
3305 * Note that if target and source are the same, this can be
3306 * done in a single check.
3308 if ((error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)))
3311 if ((*svpp)->v_type == VDIR) {
3313 * Avoid ".", "..", and aliases of "." for obvious reasons.
3315 if ((scnp->cn_namelen == 1 && scnp->cn_nameptr[0] == '.') ||
3317 (scnp->cn_flags | tcnp->cn_flags) & ISDOTDOT) {
3323 * Check to make sure rename is valid.
3324 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3326 if ((error = zfs_rename_check(szp, sdzp, tdzp)))
3331 * Does target exist?
3335 * Source and target must be the same type.
3337 if ((*svpp)->v_type == VDIR) {
3338 if ((*tvpp)->v_type != VDIR) {
3339 error = SET_ERROR(ENOTDIR);
3347 if ((*tvpp)->v_type == VDIR) {
3348 error = SET_ERROR(EISDIR);
3354 vn_seqc_write_begin(*svpp);
3355 vn_seqc_write_begin(sdvp);
3357 vn_seqc_write_begin(*tvpp);
3359 vn_seqc_write_begin(tdvp);
3360 #if __FreeBSD_version >= 1300102
3361 want_seqc_end = true;
3363 vnevent_rename_src(*svpp, sdvp, scnp->cn_nameptr, ct);
3365 vnevent_rename_dest(*tvpp, tdvp, tnm, ct);
3368 * notify the target directory if it is not the same
3369 * as source directory.
3372 vnevent_rename_dest_dir(tdvp, ct);
3375 tx = dmu_tx_create(zfsvfs->z_os);
3376 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3377 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3378 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3379 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3381 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3382 zfs_sa_upgrade_txholds(tx, tdzp);
3385 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3386 zfs_sa_upgrade_txholds(tx, tzp);
3389 zfs_sa_upgrade_txholds(tx, szp);
3390 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3391 error = dmu_tx_assign(tx, TXG_WAIT);
3398 if (tzp) /* Attempt to remove the existing target */
3399 error = zfs_link_destroy(tdzp, tnm, tzp, tx, 0, NULL);
3402 error = zfs_link_create(tdzp, tnm, szp, tx, ZRENAMING);
3404 szp->z_pflags |= ZFS_AV_MODIFIED;
3406 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3407 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3410 error = zfs_link_destroy(sdzp, snm, szp, tx, ZRENAMING,
3413 zfs_log_rename(zilog, tx, TX_RENAME, sdzp,
3414 snm, tdzp, tnm, szp);
3417 * Update path information for the target vnode
3419 vn_renamepath(tdvp, *svpp, tnm, strlen(tnm));
3422 * At this point, we have successfully created
3423 * the target name, but have failed to remove
3424 * the source name. Since the create was done
3425 * with the ZRENAMING flag, there are
3426 * complications; for one, the link count is
3427 * wrong. The easiest way to deal with this
3428 * is to remove the newly created target, and
3429 * return the original error. This must
3430 * succeed; fortunately, it is very unlikely to
3431 * fail, since we just created it.
3433 VERIFY0(zfs_link_destroy(tdzp, tnm, szp, tx,
3438 cache_vop_rename(sdvp, *svpp, tdvp, *tvpp, scnp, tcnp);
3444 unlockout: /* all 4 vnodes are locked, ZFS_ENTER called */
3446 if (want_seqc_end) {
3447 vn_seqc_write_end(*svpp);
3448 vn_seqc_write_end(sdvp);
3450 vn_seqc_write_end(*tvpp);
3452 vn_seqc_write_end(tdvp);
3453 want_seqc_end = false;
3458 out: /* original two vnodes are locked */
3459 MPASS(!want_seqc_end);
3460 if (error == 0 && zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3461 zil_commit(zilog, 0);
3471 zfs_rename(znode_t *sdzp, const char *sname, znode_t *tdzp, const char *tname,
3472 cred_t *cr, int flags)
3474 struct componentname scn, tcn;
3475 vnode_t *sdvp, *tdvp;
3482 error = zfs_lookup_internal(sdzp, sname, &svp, &scn, DELETE);
3483 if (sdzp->z_zfsvfs->z_replay == B_FALSE)
3489 vn_lock(tdvp, LK_EXCLUSIVE | LK_RETRY);
3490 error = zfs_lookup_internal(tdzp, tname, &tvp, &tcn, RENAME);
3491 if (error == EJUSTRETURN)
3493 else if (error != 0) {
3498 error = zfs_rename_(sdvp, &svp, &scn, tdvp, &tvp, &tcn, cr, 0);
3509 * Insert the indicated symbolic reference entry into the directory.
3511 * IN: dvp - Directory to contain new symbolic link.
3512 * link - Name for new symlink entry.
3513 * vap - Attributes of new entry.
3514 * cr - credentials of caller.
3515 * ct - caller context
3516 * flags - case flags
3518 * RETURN: 0 on success, error code on failure.
3521 * dvp - ctime|mtime updated
3525 zfs_symlink(znode_t *dzp, const char *name, vattr_t *vap,
3526 const char *link, znode_t **zpp, cred_t *cr, int flags)
3530 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
3532 uint64_t len = strlen(link);
3534 zfs_acl_ids_t acl_ids;
3535 boolean_t fuid_dirtied;
3536 uint64_t txtype = TX_SYMLINK;
3538 ASSERT3S(vap->va_type, ==, VLNK);
3542 zilog = zfsvfs->z_log;
3544 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3545 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3547 return (SET_ERROR(EILSEQ));
3550 if (len > MAXPATHLEN) {
3552 return (SET_ERROR(ENAMETOOLONG));
3555 if ((error = zfs_acl_ids_create(dzp, 0,
3556 vap, cr, NULL, &acl_ids)) != 0) {
3562 * Attempt to lock directory; fail if entry already exists.
3564 error = zfs_dirent_lookup(dzp, name, &zp, ZNEW);
3566 zfs_acl_ids_free(&acl_ids);
3571 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
3572 zfs_acl_ids_free(&acl_ids);
3577 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids,
3579 zfs_acl_ids_free(&acl_ids);
3581 return (SET_ERROR(EDQUOT));
3584 getnewvnode_reserve_();
3585 tx = dmu_tx_create(zfsvfs->z_os);
3586 fuid_dirtied = zfsvfs->z_fuid_dirty;
3587 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3588 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3589 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3590 ZFS_SA_BASE_ATTR_SIZE + len);
3591 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3592 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3593 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3594 acl_ids.z_aclp->z_acl_bytes);
3597 zfs_fuid_txhold(zfsvfs, tx);
3598 error = dmu_tx_assign(tx, TXG_WAIT);
3600 zfs_acl_ids_free(&acl_ids);
3602 getnewvnode_drop_reserve();
3608 * Create a new object for the symlink.
3609 * for version 4 ZPL datasets the symlink will be an SA attribute
3611 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3614 zfs_fuid_sync(zfsvfs, tx);
3617 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
3618 __DECONST(void *, link), len, tx);
3620 zfs_sa_symlink(zp, __DECONST(char *, link), len, tx);
3623 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
3624 &zp->z_size, sizeof (zp->z_size), tx);
3626 * Insert the new object into the directory.
3628 (void) zfs_link_create(dzp, name, zp, tx, ZNEW);
3630 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3633 zfs_acl_ids_free(&acl_ids);
3637 getnewvnode_drop_reserve();
3639 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3640 zil_commit(zilog, 0);
3647 * Return, in the buffer contained in the provided uio structure,
3648 * the symbolic path referred to by vp.
3650 * IN: vp - vnode of symbolic link.
3651 * uio - structure to contain the link path.
3652 * cr - credentials of caller.
3653 * ct - caller context
3655 * OUT: uio - structure containing the link path.
3657 * RETURN: 0 on success, error code on failure.
3660 * vp - atime updated
3664 zfs_readlink(vnode_t *vp, zfs_uio_t *uio, cred_t *cr, caller_context_t *ct)
3666 znode_t *zp = VTOZ(vp);
3667 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3674 error = sa_lookup_uio(zp->z_sa_hdl,
3675 SA_ZPL_SYMLINK(zfsvfs), uio);
3677 error = zfs_sa_readlink(zp, uio);
3679 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3686 * Insert a new entry into directory tdvp referencing svp.
3688 * IN: tdvp - Directory to contain new entry.
3689 * svp - vnode of new entry.
3690 * name - name of new entry.
3691 * cr - credentials of caller.
3693 * RETURN: 0 on success, error code on failure.
3696 * tdvp - ctime|mtime updated
3697 * svp - ctime updated
3701 zfs_link(znode_t *tdzp, znode_t *szp, const char *name, cred_t *cr,
3705 zfsvfs_t *zfsvfs = tdzp->z_zfsvfs;
3712 ASSERT3S(ZTOV(tdzp)->v_type, ==, VDIR);
3715 ZFS_VERIFY_ZP(tdzp);
3716 zilog = zfsvfs->z_log;
3719 * POSIX dictates that we return EPERM here.
3720 * Better choices include ENOTSUP or EISDIR.
3722 if (ZTOV(szp)->v_type == VDIR) {
3724 return (SET_ERROR(EPERM));
3730 * If we are using project inheritance, means if the directory has
3731 * ZFS_PROJINHERIT set, then its descendant directories will inherit
3732 * not only the project ID, but also the ZFS_PROJINHERIT flag. Under
3733 * such case, we only allow hard link creation in our tree when the
3734 * project IDs are the same.
3736 if (tdzp->z_pflags & ZFS_PROJINHERIT &&
3737 tdzp->z_projid != szp->z_projid) {
3739 return (SET_ERROR(EXDEV));
3742 if (szp->z_pflags & (ZFS_APPENDONLY |
3743 ZFS_IMMUTABLE | ZFS_READONLY)) {
3745 return (SET_ERROR(EPERM));
3748 /* Prevent links to .zfs/shares files */
3750 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
3751 &parent, sizeof (uint64_t))) != 0) {
3755 if (parent == zfsvfs->z_shares_dir) {
3757 return (SET_ERROR(EPERM));
3760 if (zfsvfs->z_utf8 && u8_validate(name,
3761 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3763 return (SET_ERROR(EILSEQ));
3767 * We do not support links between attributes and non-attributes
3768 * because of the potential security risk of creating links
3769 * into "normal" file space in order to circumvent restrictions
3770 * imposed in attribute space.
3772 if ((szp->z_pflags & ZFS_XATTR) != (tdzp->z_pflags & ZFS_XATTR)) {
3774 return (SET_ERROR(EINVAL));
3778 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
3779 if (owner != crgetuid(cr) && secpolicy_basic_link(ZTOV(szp), cr) != 0) {
3781 return (SET_ERROR(EPERM));
3784 if ((error = zfs_zaccess(tdzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
3790 * Attempt to lock directory; fail if entry already exists.
3792 error = zfs_dirent_lookup(tdzp, name, &tzp, ZNEW);
3798 tx = dmu_tx_create(zfsvfs->z_os);
3799 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3800 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, name);
3801 zfs_sa_upgrade_txholds(tx, szp);
3802 zfs_sa_upgrade_txholds(tx, tdzp);
3803 error = dmu_tx_assign(tx, TXG_WAIT);
3810 error = zfs_link_create(tdzp, name, szp, tx, 0);
3813 uint64_t txtype = TX_LINK;
3814 zfs_log_link(zilog, tx, txtype, tdzp, szp, name);
3820 vnevent_link(ZTOV(szp), ct);
3823 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3824 zil_commit(zilog, 0);
3831 * Free or allocate space in a file. Currently, this function only
3832 * supports the `F_FREESP' command. However, this command is somewhat
3833 * misnamed, as its functionality includes the ability to allocate as
3834 * well as free space.
3836 * IN: ip - inode of file to free data in.
3837 * cmd - action to take (only F_FREESP supported).
3838 * bfp - section of file to free/alloc.
3839 * flag - current file open mode flags.
3840 * offset - current file offset.
3841 * cr - credentials of caller.
3843 * RETURN: 0 on success, error code on failure.
3846 * ip - ctime|mtime updated
3850 zfs_space(znode_t *zp, int cmd, flock64_t *bfp, int flag,
3851 offset_t offset, cred_t *cr)
3853 zfsvfs_t *zfsvfs = ZTOZSB(zp);
3860 if (cmd != F_FREESP) {
3862 return (SET_ERROR(EINVAL));
3866 * Callers might not be able to detect properly that we are read-only,
3867 * so check it explicitly here.
3869 if (zfs_is_readonly(zfsvfs)) {
3871 return (SET_ERROR(EROFS));
3874 if (bfp->l_len < 0) {
3876 return (SET_ERROR(EINVAL));
3880 * Permissions aren't checked on Solaris because on this OS
3881 * zfs_space() can only be called with an opened file handle.
3882 * On Linux we can get here through truncate_range() which
3883 * operates directly on inodes, so we need to check access rights.
3885 if ((error = zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr))) {
3891 len = bfp->l_len; /* 0 means from off to end of file */
3893 error = zfs_freesp(zp, off, len, flag, TRUE);
3901 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
3903 znode_t *zp = VTOZ(vp);
3904 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3907 ZFS_TEARDOWN_INACTIVE_ENTER_READ(zfsvfs);
3908 if (zp->z_sa_hdl == NULL) {
3910 * The fs has been unmounted, or we did a
3911 * suspend/resume and this file no longer exists.
3913 ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs);
3918 if (zp->z_unlinked) {
3920 * Fast path to recycle a vnode of a removed file.
3922 ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs);
3927 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
3928 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
3930 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3931 zfs_sa_upgrade_txholds(tx, zp);
3932 error = dmu_tx_assign(tx, TXG_WAIT);
3936 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
3937 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
3938 zp->z_atime_dirty = 0;
3942 ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs);
3946 CTASSERT(sizeof (struct zfid_short) <= sizeof (struct fid));
3947 CTASSERT(sizeof (struct zfid_long) <= sizeof (struct fid));
3951 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
3953 znode_t *zp = VTOZ(vp);
3954 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3957 uint64_t object = zp->z_id;
3964 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
3965 &gen64, sizeof (uint64_t))) != 0) {
3970 gen = (uint32_t)gen64;
3972 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
3973 fidp->fid_len = size;
3975 zfid = (zfid_short_t *)fidp;
3977 zfid->zf_len = size;
3979 for (i = 0; i < sizeof (zfid->zf_object); i++)
3980 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
3982 /* Must have a non-zero generation number to distinguish from .zfs */
3985 for (i = 0; i < sizeof (zfid->zf_gen); i++)
3986 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
3988 if (size == LONG_FID_LEN) {
3989 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
3992 zlfid = (zfid_long_t *)fidp;
3994 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
3995 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
3997 /* XXX - this should be the generation number for the objset */
3998 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
3999 zlfid->zf_setgen[i] = 0;
4007 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4008 caller_context_t *ct)
4015 *valp = MIN(LONG_MAX, ZFS_LINK_MAX);
4018 case _PC_FILESIZEBITS:
4021 case _PC_MIN_HOLE_SIZE:
4022 *valp = (int)SPA_MINBLOCKSIZE;
4024 case _PC_ACL_EXTENDED:
4025 #if 0 /* POSIX ACLs are not implemented for ZFS on FreeBSD yet. */
4027 zfsvfs = zp->z_zfsvfs;
4030 *valp = zfsvfs->z_acl_type == ZFSACLTYPE_POSIX ? 1 : 0;
4039 zfsvfs = zp->z_zfsvfs;
4042 *valp = zfsvfs->z_acl_type == ZFS_ACLTYPE_NFSV4 ? 1 : 0;
4046 case _PC_ACL_PATH_MAX:
4047 *valp = ACL_MAX_ENTRIES;
4051 return (EOPNOTSUPP);
4056 zfs_getpages(struct vnode *vp, vm_page_t *ma, int count, int *rbehind,
4059 znode_t *zp = VTOZ(vp);
4060 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4061 objset_t *os = zp->z_zfsvfs->z_os;
4062 zfs_locked_range_t *lr;
4064 off_t start, end, obj_size;
4066 int pgsin_b, pgsin_a;
4072 start = IDX_TO_OFF(ma[0]->pindex);
4073 end = IDX_TO_OFF(ma[count - 1]->pindex + 1);
4076 * Lock a range covering all required and optional pages.
4077 * Note that we need to handle the case of the block size growing.
4080 blksz = zp->z_blksz;
4081 lr = zfs_rangelock_tryenter(&zp->z_rangelock,
4082 rounddown(start, blksz),
4083 roundup(end, blksz) - rounddown(start, blksz), RL_READER);
4085 if (rahead != NULL) {
4089 if (rbehind != NULL) {
4095 if (blksz == zp->z_blksz)
4097 zfs_rangelock_exit(lr);
4100 object = ma[0]->object;
4101 zfs_vmobject_wlock(object);
4102 obj_size = object->un_pager.vnp.vnp_size;
4103 zfs_vmobject_wunlock(object);
4104 if (IDX_TO_OFF(ma[count - 1]->pindex) >= obj_size) {
4106 zfs_rangelock_exit(lr);
4108 return (zfs_vm_pagerret_bad);
4112 if (rbehind != NULL) {
4113 pgsin_b = OFF_TO_IDX(start - rounddown(start, blksz));
4114 pgsin_b = MIN(*rbehind, pgsin_b);
4118 if (rahead != NULL) {
4119 pgsin_a = OFF_TO_IDX(roundup(end, blksz) - end);
4120 if (end + IDX_TO_OFF(pgsin_a) >= obj_size)
4121 pgsin_a = OFF_TO_IDX(round_page(obj_size) - end);
4122 pgsin_a = MIN(*rahead, pgsin_a);
4126 * NB: we need to pass the exact byte size of the data that we expect
4127 * to read after accounting for the file size. This is required because
4128 * ZFS will panic if we request DMU to read beyond the end of the last
4131 error = dmu_read_pages(os, zp->z_id, ma, count, &pgsin_b, &pgsin_a,
4132 MIN(end, obj_size) - (end - PAGE_SIZE));
4135 zfs_rangelock_exit(lr);
4136 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4140 return (zfs_vm_pagerret_error);
4142 VM_CNT_INC(v_vnodein);
4143 VM_CNT_ADD(v_vnodepgsin, count + pgsin_b + pgsin_a);
4144 if (rbehind != NULL)
4148 return (zfs_vm_pagerret_ok);
4151 #ifndef _SYS_SYSPROTO_H_
4152 struct vop_getpages_args {
4162 zfs_freebsd_getpages(struct vop_getpages_args *ap)
4165 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_rbehind,
4170 zfs_putpages(struct vnode *vp, vm_page_t *ma, size_t len, int flags,
4173 znode_t *zp = VTOZ(vp);
4174 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4175 zfs_locked_range_t *lr;
4183 vm_ooffset_t lo_off;
4194 object = vp->v_object;
4198 KASSERT(ma[0]->object == object, ("mismatching object"));
4199 KASSERT(len > 0 && (len & PAGE_MASK) == 0, ("unexpected length"));
4201 for (i = 0; i < pcount; i++)
4202 rtvals[i] = zfs_vm_pagerret_error;
4204 off = IDX_TO_OFF(ma[0]->pindex);
4205 blksz = zp->z_blksz;
4206 lo_off = rounddown(off, blksz);
4207 lo_len = roundup(len + (off - lo_off), blksz);
4208 lr = zfs_rangelock_enter(&zp->z_rangelock, lo_off, lo_len, RL_WRITER);
4210 zfs_vmobject_wlock(object);
4211 if (len + off > object->un_pager.vnp.vnp_size) {
4212 if (object->un_pager.vnp.vnp_size > off) {
4215 len = object->un_pager.vnp.vnp_size - off;
4217 if ((pgoff = (int)len & PAGE_MASK) != 0) {
4219 * If the object is locked and the following
4220 * conditions hold, then the page's dirty
4221 * field cannot be concurrently changed by a
4225 vm_page_assert_sbusied(m);
4226 KASSERT(!pmap_page_is_write_mapped(m),
4227 ("zfs_putpages: page %p is not read-only",
4229 vm_page_clear_dirty(m, pgoff, PAGE_SIZE -
4236 if (ncount < pcount) {
4237 for (i = ncount; i < pcount; i++) {
4238 rtvals[i] = zfs_vm_pagerret_bad;
4242 zfs_vmobject_wunlock(object);
4247 if (zfs_id_overblockquota(zfsvfs, DMU_USERUSED_OBJECT, zp->z_uid) ||
4248 zfs_id_overblockquota(zfsvfs, DMU_GROUPUSED_OBJECT, zp->z_gid) ||
4249 (zp->z_projid != ZFS_DEFAULT_PROJID &&
4250 zfs_id_overblockquota(zfsvfs, DMU_PROJECTUSED_OBJECT,
4255 tx = dmu_tx_create(zfsvfs->z_os);
4256 dmu_tx_hold_write(tx, zp->z_id, off, len);
4258 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4259 zfs_sa_upgrade_txholds(tx, zp);
4260 err = dmu_tx_assign(tx, TXG_WAIT);
4266 if (zp->z_blksz < PAGE_SIZE) {
4267 for (i = 0; len > 0; off += tocopy, len -= tocopy, i++) {
4268 tocopy = len > PAGE_SIZE ? PAGE_SIZE : len;
4269 va = zfs_map_page(ma[i], &sf);
4270 dmu_write(zfsvfs->z_os, zp->z_id, off, tocopy, va, tx);
4274 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, ma, tx);
4278 uint64_t mtime[2], ctime[2];
4279 sa_bulk_attr_t bulk[3];
4282 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4284 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4286 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4288 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime);
4289 err = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
4292 * XXX we should be passing a callback to undirty
4293 * but that would make the locking messier
4295 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off,
4296 len, 0, NULL, NULL);
4298 zfs_vmobject_wlock(object);
4299 for (i = 0; i < ncount; i++) {
4300 rtvals[i] = zfs_vm_pagerret_ok;
4301 vm_page_undirty(ma[i]);
4303 zfs_vmobject_wunlock(object);
4304 VM_CNT_INC(v_vnodeout);
4305 VM_CNT_ADD(v_vnodepgsout, ncount);
4310 zfs_rangelock_exit(lr);
4311 if ((flags & (zfs_vm_pagerput_sync | zfs_vm_pagerput_inval)) != 0 ||
4312 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4313 zil_commit(zfsvfs->z_log, zp->z_id);
4318 #ifndef _SYS_SYSPROTO_H_
4319 struct vop_putpages_args {
4329 zfs_freebsd_putpages(struct vop_putpages_args *ap)
4332 return (zfs_putpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_sync,
4336 #ifndef _SYS_SYSPROTO_H_
4337 struct vop_bmap_args {
4340 struct bufobj **a_bop;
4348 zfs_freebsd_bmap(struct vop_bmap_args *ap)
4351 if (ap->a_bop != NULL)
4352 *ap->a_bop = &ap->a_vp->v_bufobj;
4353 if (ap->a_bnp != NULL)
4354 *ap->a_bnp = ap->a_bn;
4355 if (ap->a_runp != NULL)
4357 if (ap->a_runb != NULL)
4363 #ifndef _SYS_SYSPROTO_H_
4364 struct vop_open_args {
4367 struct ucred *a_cred;
4368 struct thread *a_td;
4373 zfs_freebsd_open(struct vop_open_args *ap)
4375 vnode_t *vp = ap->a_vp;
4376 znode_t *zp = VTOZ(vp);
4379 error = zfs_open(&vp, ap->a_mode, ap->a_cred);
4381 vnode_create_vobject(vp, zp->z_size, ap->a_td);
4385 #ifndef _SYS_SYSPROTO_H_
4386 struct vop_close_args {
4389 struct ucred *a_cred;
4390 struct thread *a_td;
4395 zfs_freebsd_close(struct vop_close_args *ap)
4398 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred));
4401 #ifndef _SYS_SYSPROTO_H_
4402 struct vop_ioctl_args {
4413 zfs_freebsd_ioctl(struct vop_ioctl_args *ap)
4416 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data,
4417 ap->a_fflag, ap->a_cred, NULL));
4421 ioflags(int ioflags)
4425 if (ioflags & IO_APPEND)
4427 if (ioflags & IO_NDELAY)
4429 if (ioflags & IO_SYNC)
4430 flags |= (FSYNC | FDSYNC | FRSYNC);
4435 #ifndef _SYS_SYSPROTO_H_
4436 struct vop_read_args {
4440 struct ucred *a_cred;
4445 zfs_freebsd_read(struct vop_read_args *ap)
4448 zfs_uio_init(&uio, ap->a_uio);
4449 return (zfs_read(VTOZ(ap->a_vp), &uio, ioflags(ap->a_ioflag),
4453 #ifndef _SYS_SYSPROTO_H_
4454 struct vop_write_args {
4458 struct ucred *a_cred;
4463 zfs_freebsd_write(struct vop_write_args *ap)
4466 zfs_uio_init(&uio, ap->a_uio);
4467 return (zfs_write(VTOZ(ap->a_vp), &uio, ioflags(ap->a_ioflag),
4471 #if __FreeBSD_version >= 1300102
4473 * VOP_FPLOOKUP_VEXEC routines are subject to special circumstances, see
4474 * the comment above cache_fplookup for details.
4477 zfs_freebsd_fplookup_vexec(struct vop_fplookup_vexec_args *v)
4485 if (__predict_false(zp == NULL))
4487 pflags = atomic_load_64(&zp->z_pflags);
4488 if (pflags & ZFS_AV_QUARANTINED)
4490 if (pflags & ZFS_XATTR)
4492 if ((pflags & ZFS_NO_EXECS_DENIED) == 0)
4498 #if __FreeBSD_version >= 1300139
4500 zfs_freebsd_fplookup_symlink(struct vop_fplookup_symlink_args *v)
4508 if (__predict_false(zp == NULL)) {
4512 target = atomic_load_consume_ptr(&zp->z_cached_symlink);
4513 if (target == NULL) {
4516 return (cache_symlink_resolve(v->a_fpl, target, strlen(target)));
4520 #ifndef _SYS_SYSPROTO_H_
4521 struct vop_access_args {
4523 accmode_t a_accmode;
4524 struct ucred *a_cred;
4525 struct thread *a_td;
4530 zfs_freebsd_access(struct vop_access_args *ap)
4532 vnode_t *vp = ap->a_vp;
4533 znode_t *zp = VTOZ(vp);
4538 if (ap->a_accmode == VEXEC) {
4539 if (zfs_fastaccesschk_execute(zp, ap->a_cred) == 0)
4544 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND,
4546 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND);
4548 error = zfs_access(zp, accmode, 0, ap->a_cred);
4551 * VADMIN has to be handled by vaccess().
4554 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND);
4556 #if __FreeBSD_version >= 1300105
4557 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
4558 zp->z_gid, accmode, ap->a_cred);
4560 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
4561 zp->z_gid, accmode, ap->a_cred, NULL);
4567 * For VEXEC, ensure that at least one execute bit is set for
4570 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR &&
4571 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
4578 #ifndef _SYS_SYSPROTO_H_
4579 struct vop_lookup_args {
4580 struct vnode *a_dvp;
4581 struct vnode **a_vpp;
4582 struct componentname *a_cnp;
4587 zfs_freebsd_lookup(struct vop_lookup_args *ap, boolean_t cached)
4589 struct componentname *cnp = ap->a_cnp;
4590 char nm[NAME_MAX + 1];
4592 ASSERT3U(cnp->cn_namelen, <, sizeof (nm));
4593 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof (nm)));
4595 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop,
4596 cnp->cn_cred, cnp->cn_thread, 0, cached));
4600 zfs_freebsd_cachedlookup(struct vop_cachedlookup_args *ap)
4603 return (zfs_freebsd_lookup((struct vop_lookup_args *)ap, B_TRUE));
4606 #ifndef _SYS_SYSPROTO_H_
4607 struct vop_lookup_args {
4608 struct vnode *a_dvp;
4609 struct vnode **a_vpp;
4610 struct componentname *a_cnp;
4615 zfs_cache_lookup(struct vop_lookup_args *ap)
4619 zfsvfs = ap->a_dvp->v_mount->mnt_data;
4620 if (zfsvfs->z_use_namecache)
4621 return (vfs_cache_lookup(ap));
4623 return (zfs_freebsd_lookup(ap, B_FALSE));
4626 #ifndef _SYS_SYSPROTO_H_
4627 struct vop_create_args {
4628 struct vnode *a_dvp;
4629 struct vnode **a_vpp;
4630 struct componentname *a_cnp;
4631 struct vattr *a_vap;
4636 zfs_freebsd_create(struct vop_create_args *ap)
4639 struct componentname *cnp = ap->a_cnp;
4640 vattr_t *vap = ap->a_vap;
4644 ASSERT(cnp->cn_flags & SAVENAME);
4646 vattr_init_mask(vap);
4647 mode = vap->va_mode & ALLPERMS;
4648 zfsvfs = ap->a_dvp->v_mount->mnt_data;
4651 rc = zfs_create(VTOZ(ap->a_dvp), cnp->cn_nameptr, vap, !EXCL, mode,
4652 &zp, cnp->cn_cred, 0 /* flag */, NULL /* vsecattr */);
4654 *ap->a_vpp = ZTOV(zp);
4655 if (zfsvfs->z_use_namecache &&
4656 rc == 0 && (cnp->cn_flags & MAKEENTRY) != 0)
4657 cache_enter(ap->a_dvp, *ap->a_vpp, cnp);
4662 #ifndef _SYS_SYSPROTO_H_
4663 struct vop_remove_args {
4664 struct vnode *a_dvp;
4666 struct componentname *a_cnp;
4671 zfs_freebsd_remove(struct vop_remove_args *ap)
4674 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
4676 return (zfs_remove_(ap->a_dvp, ap->a_vp, ap->a_cnp->cn_nameptr,
4677 ap->a_cnp->cn_cred));
4680 #ifndef _SYS_SYSPROTO_H_
4681 struct vop_mkdir_args {
4682 struct vnode *a_dvp;
4683 struct vnode **a_vpp;
4684 struct componentname *a_cnp;
4685 struct vattr *a_vap;
4690 zfs_freebsd_mkdir(struct vop_mkdir_args *ap)
4692 vattr_t *vap = ap->a_vap;
4696 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
4698 vattr_init_mask(vap);
4701 rc = zfs_mkdir(VTOZ(ap->a_dvp), ap->a_cnp->cn_nameptr, vap, &zp,
4702 ap->a_cnp->cn_cred, 0, NULL);
4705 *ap->a_vpp = ZTOV(zp);
4709 #ifndef _SYS_SYSPROTO_H_
4710 struct vop_rmdir_args {
4711 struct vnode *a_dvp;
4713 struct componentname *a_cnp;
4718 zfs_freebsd_rmdir(struct vop_rmdir_args *ap)
4720 struct componentname *cnp = ap->a_cnp;
4722 ASSERT(cnp->cn_flags & SAVENAME);
4724 return (zfs_rmdir_(ap->a_dvp, ap->a_vp, cnp->cn_nameptr, cnp->cn_cred));
4727 #ifndef _SYS_SYSPROTO_H_
4728 struct vop_readdir_args {
4731 struct ucred *a_cred;
4734 ulong_t **a_cookies;
4739 zfs_freebsd_readdir(struct vop_readdir_args *ap)
4742 zfs_uio_init(&uio, ap->a_uio);
4743 return (zfs_readdir(ap->a_vp, &uio, ap->a_cred, ap->a_eofflag,
4744 ap->a_ncookies, ap->a_cookies));
4747 #ifndef _SYS_SYSPROTO_H_
4748 struct vop_fsync_args {
4751 struct thread *a_td;
4756 zfs_freebsd_fsync(struct vop_fsync_args *ap)
4760 return (zfs_fsync(VTOZ(ap->a_vp), 0, ap->a_td->td_ucred));
4763 #ifndef _SYS_SYSPROTO_H_
4764 struct vop_getattr_args {
4766 struct vattr *a_vap;
4767 struct ucred *a_cred;
4772 zfs_freebsd_getattr(struct vop_getattr_args *ap)
4774 vattr_t *vap = ap->a_vap;
4780 xvap.xva_vattr = *vap;
4781 xvap.xva_vattr.va_mask |= AT_XVATTR;
4783 /* Convert chflags into ZFS-type flags. */
4784 /* XXX: what about SF_SETTABLE?. */
4785 XVA_SET_REQ(&xvap, XAT_IMMUTABLE);
4786 XVA_SET_REQ(&xvap, XAT_APPENDONLY);
4787 XVA_SET_REQ(&xvap, XAT_NOUNLINK);
4788 XVA_SET_REQ(&xvap, XAT_NODUMP);
4789 XVA_SET_REQ(&xvap, XAT_READONLY);
4790 XVA_SET_REQ(&xvap, XAT_ARCHIVE);
4791 XVA_SET_REQ(&xvap, XAT_SYSTEM);
4792 XVA_SET_REQ(&xvap, XAT_HIDDEN);
4793 XVA_SET_REQ(&xvap, XAT_REPARSE);
4794 XVA_SET_REQ(&xvap, XAT_OFFLINE);
4795 XVA_SET_REQ(&xvap, XAT_SPARSE);
4797 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred);
4801 /* Convert ZFS xattr into chflags. */
4802 #define FLAG_CHECK(fflag, xflag, xfield) do { \
4803 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \
4804 fflags |= (fflag); \
4806 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE,
4807 xvap.xva_xoptattrs.xoa_immutable);
4808 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY,
4809 xvap.xva_xoptattrs.xoa_appendonly);
4810 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK,
4811 xvap.xva_xoptattrs.xoa_nounlink);
4812 FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE,
4813 xvap.xva_xoptattrs.xoa_archive);
4814 FLAG_CHECK(UF_NODUMP, XAT_NODUMP,
4815 xvap.xva_xoptattrs.xoa_nodump);
4816 FLAG_CHECK(UF_READONLY, XAT_READONLY,
4817 xvap.xva_xoptattrs.xoa_readonly);
4818 FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM,
4819 xvap.xva_xoptattrs.xoa_system);
4820 FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN,
4821 xvap.xva_xoptattrs.xoa_hidden);
4822 FLAG_CHECK(UF_REPARSE, XAT_REPARSE,
4823 xvap.xva_xoptattrs.xoa_reparse);
4824 FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE,
4825 xvap.xva_xoptattrs.xoa_offline);
4826 FLAG_CHECK(UF_SPARSE, XAT_SPARSE,
4827 xvap.xva_xoptattrs.xoa_sparse);
4830 *vap = xvap.xva_vattr;
4831 vap->va_flags = fflags;
4835 #ifndef _SYS_SYSPROTO_H_
4836 struct vop_setattr_args {
4838 struct vattr *a_vap;
4839 struct ucred *a_cred;
4844 zfs_freebsd_setattr(struct vop_setattr_args *ap)
4846 vnode_t *vp = ap->a_vp;
4847 vattr_t *vap = ap->a_vap;
4848 cred_t *cred = ap->a_cred;
4853 vattr_init_mask(vap);
4854 vap->va_mask &= ~AT_NOSET;
4857 xvap.xva_vattr = *vap;
4859 zflags = VTOZ(vp)->z_pflags;
4861 if (vap->va_flags != VNOVAL) {
4862 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs;
4865 if (zfsvfs->z_use_fuids == B_FALSE)
4866 return (EOPNOTSUPP);
4868 fflags = vap->va_flags;
4871 * We need to figure out whether it makes sense to allow
4872 * UF_REPARSE through, since we don't really have other
4873 * facilities to handle reparse points and zfs_setattr()
4874 * doesn't currently allow setting that attribute anyway.
4876 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE|
4877 UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE|
4878 UF_OFFLINE|UF_SPARSE)) != 0)
4879 return (EOPNOTSUPP);
4881 * Unprivileged processes are not permitted to unset system
4882 * flags, or modify flags if any system flags are set.
4883 * Privileged non-jail processes may not modify system flags
4884 * if securelevel > 0 and any existing system flags are set.
4885 * Privileged jail processes behave like privileged non-jail
4886 * processes if the PR_ALLOW_CHFLAGS permission bit is set;
4887 * otherwise, they behave like unprivileged processes.
4889 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 ||
4890 spl_priv_check_cred(cred, PRIV_VFS_SYSFLAGS) == 0) {
4892 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
4893 error = securelevel_gt(cred, 0);
4899 * Callers may only modify the file flags on
4900 * objects they have VADMIN rights for.
4902 if ((error = VOP_ACCESS(vp, VADMIN, cred,
4906 (ZFS_IMMUTABLE | ZFS_APPENDONLY |
4911 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) {
4916 #define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \
4917 if (((fflags & (fflag)) && !(zflags & (zflag))) || \
4918 ((zflags & (zflag)) && !(fflags & (fflag)))) { \
4919 XVA_SET_REQ(&xvap, (xflag)); \
4920 (xfield) = ((fflags & (fflag)) != 0); \
4923 /* Convert chflags into ZFS-type flags. */
4924 /* XXX: what about SF_SETTABLE?. */
4925 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE,
4926 xvap.xva_xoptattrs.xoa_immutable);
4927 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY,
4928 xvap.xva_xoptattrs.xoa_appendonly);
4929 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK,
4930 xvap.xva_xoptattrs.xoa_nounlink);
4931 FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE,
4932 xvap.xva_xoptattrs.xoa_archive);
4933 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP,
4934 xvap.xva_xoptattrs.xoa_nodump);
4935 FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY,
4936 xvap.xva_xoptattrs.xoa_readonly);
4937 FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM,
4938 xvap.xva_xoptattrs.xoa_system);
4939 FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN,
4940 xvap.xva_xoptattrs.xoa_hidden);
4941 FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE,
4942 xvap.xva_xoptattrs.xoa_reparse);
4943 FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE,
4944 xvap.xva_xoptattrs.xoa_offline);
4945 FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE,
4946 xvap.xva_xoptattrs.xoa_sparse);
4949 if (vap->va_birthtime.tv_sec != VNOVAL) {
4950 xvap.xva_vattr.va_mask |= AT_XVATTR;
4951 XVA_SET_REQ(&xvap, XAT_CREATETIME);
4953 return (zfs_setattr(VTOZ(vp), (vattr_t *)&xvap, 0, cred));
4956 #ifndef _SYS_SYSPROTO_H_
4957 struct vop_rename_args {
4958 struct vnode *a_fdvp;
4959 struct vnode *a_fvp;
4960 struct componentname *a_fcnp;
4961 struct vnode *a_tdvp;
4962 struct vnode *a_tvp;
4963 struct componentname *a_tcnp;
4968 zfs_freebsd_rename(struct vop_rename_args *ap)
4970 vnode_t *fdvp = ap->a_fdvp;
4971 vnode_t *fvp = ap->a_fvp;
4972 vnode_t *tdvp = ap->a_tdvp;
4973 vnode_t *tvp = ap->a_tvp;
4976 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART));
4977 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART));
4979 error = zfs_rename_(fdvp, &fvp, ap->a_fcnp, tdvp, &tvp,
4980 ap->a_tcnp, ap->a_fcnp->cn_cred, 1);
4991 #ifndef _SYS_SYSPROTO_H_
4992 struct vop_symlink_args {
4993 struct vnode *a_dvp;
4994 struct vnode **a_vpp;
4995 struct componentname *a_cnp;
4996 struct vattr *a_vap;
5002 zfs_freebsd_symlink(struct vop_symlink_args *ap)
5004 struct componentname *cnp = ap->a_cnp;
5005 vattr_t *vap = ap->a_vap;
5007 #if __FreeBSD_version >= 1300139
5013 ASSERT(cnp->cn_flags & SAVENAME);
5015 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */
5016 vattr_init_mask(vap);
5019 rc = zfs_symlink(VTOZ(ap->a_dvp), cnp->cn_nameptr, vap,
5020 ap->a_target, &zp, cnp->cn_cred, 0 /* flags */);
5022 *ap->a_vpp = ZTOV(zp);
5023 ASSERT_VOP_ELOCKED(ZTOV(zp), __func__);
5024 #if __FreeBSD_version >= 1300139
5025 MPASS(zp->z_cached_symlink == NULL);
5026 symlink_len = strlen(ap->a_target);
5027 symlink = cache_symlink_alloc(symlink_len + 1, M_WAITOK);
5028 if (symlink != NULL) {
5029 memcpy(symlink, ap->a_target, symlink_len);
5030 symlink[symlink_len] = '\0';
5031 atomic_store_rel_ptr((uintptr_t *)&zp->z_cached_symlink,
5032 (uintptr_t)symlink);
5039 #ifndef _SYS_SYSPROTO_H_
5040 struct vop_readlink_args {
5043 struct ucred *a_cred;
5048 zfs_freebsd_readlink(struct vop_readlink_args *ap)
5052 #if __FreeBSD_version >= 1300139
5053 znode_t *zp = VTOZ(ap->a_vp);
5054 char *symlink, *base;
5059 zfs_uio_init(&uio, ap->a_uio);
5060 #if __FreeBSD_version >= 1300139
5062 if (zfs_uio_segflg(&uio) == UIO_SYSSPACE &&
5063 zfs_uio_iovcnt(&uio) == 1) {
5064 base = zfs_uio_iovbase(&uio, 0);
5065 symlink_len = zfs_uio_iovlen(&uio, 0);
5069 error = zfs_readlink(ap->a_vp, &uio, ap->a_cred, NULL);
5070 #if __FreeBSD_version >= 1300139
5071 if (atomic_load_ptr(&zp->z_cached_symlink) != NULL ||
5072 error != 0 || !trycache) {
5075 symlink_len -= zfs_uio_resid(&uio);
5076 symlink = cache_symlink_alloc(symlink_len + 1, M_WAITOK);
5077 if (symlink != NULL) {
5078 memcpy(symlink, base, symlink_len);
5079 symlink[symlink_len] = '\0';
5080 if (!atomic_cmpset_rel_ptr((uintptr_t *)&zp->z_cached_symlink,
5081 (uintptr_t)NULL, (uintptr_t)symlink)) {
5082 cache_symlink_free(symlink, symlink_len + 1);
5089 #ifndef _SYS_SYSPROTO_H_
5090 struct vop_link_args {
5091 struct vnode *a_tdvp;
5093 struct componentname *a_cnp;
5098 zfs_freebsd_link(struct vop_link_args *ap)
5100 struct componentname *cnp = ap->a_cnp;
5101 vnode_t *vp = ap->a_vp;
5102 vnode_t *tdvp = ap->a_tdvp;
5104 if (tdvp->v_mount != vp->v_mount)
5107 ASSERT(cnp->cn_flags & SAVENAME);
5109 return (zfs_link(VTOZ(tdvp), VTOZ(vp),
5110 cnp->cn_nameptr, cnp->cn_cred, 0));
5113 #ifndef _SYS_SYSPROTO_H_
5114 struct vop_inactive_args {
5116 struct thread *a_td;
5121 zfs_freebsd_inactive(struct vop_inactive_args *ap)
5123 vnode_t *vp = ap->a_vp;
5125 #if __FreeBSD_version >= 1300123
5126 zfs_inactive(vp, curthread->td_ucred, NULL);
5128 zfs_inactive(vp, ap->a_td->td_ucred, NULL);
5133 #if __FreeBSD_version >= 1300042
5134 #ifndef _SYS_SYSPROTO_H_
5135 struct vop_need_inactive_args {
5137 struct thread *a_td;
5142 zfs_freebsd_need_inactive(struct vop_need_inactive_args *ap)
5144 vnode_t *vp = ap->a_vp;
5145 znode_t *zp = VTOZ(vp);
5146 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5149 if (vn_need_pageq_flush(vp))
5152 if (!ZFS_TEARDOWN_INACTIVE_TRY_ENTER_READ(zfsvfs))
5154 need = (zp->z_sa_hdl == NULL || zp->z_unlinked || zp->z_atime_dirty);
5155 ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs);
5161 #ifndef _SYS_SYSPROTO_H_
5162 struct vop_reclaim_args {
5164 struct thread *a_td;
5169 zfs_freebsd_reclaim(struct vop_reclaim_args *ap)
5171 vnode_t *vp = ap->a_vp;
5172 znode_t *zp = VTOZ(vp);
5173 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5175 ASSERT3P(zp, !=, NULL);
5177 #if __FreeBSD_version < 1300042
5178 /* Destroy the vm object and flush associated pages. */
5179 vnode_destroy_vobject(vp);
5182 * z_teardown_inactive_lock protects from a race with
5183 * zfs_znode_dmu_fini in zfsvfs_teardown during
5186 ZFS_TEARDOWN_INACTIVE_ENTER_READ(zfsvfs);
5187 if (zp->z_sa_hdl == NULL)
5191 ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs);
5197 #ifndef _SYS_SYSPROTO_H_
5198 struct vop_fid_args {
5205 zfs_freebsd_fid(struct vop_fid_args *ap)
5208 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL));
5212 #ifndef _SYS_SYSPROTO_H_
5213 struct vop_pathconf_args {
5216 register_t *a_retval;
5221 zfs_freebsd_pathconf(struct vop_pathconf_args *ap)
5226 error = zfs_pathconf(ap->a_vp, ap->a_name, &val,
5227 curthread->td_ucred, NULL);
5229 *ap->a_retval = val;
5232 if (error != EOPNOTSUPP)
5235 switch (ap->a_name) {
5237 *ap->a_retval = NAME_MAX;
5240 if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO) {
5241 *ap->a_retval = PIPE_BUF;
5246 return (vop_stdpathconf(ap));
5251 * FreeBSD's extended attributes namespace defines file name prefix for ZFS'
5252 * extended attribute name:
5255 * system freebsd:system:
5256 * user (none, can be used to access ZFS fsattr(5) attributes
5257 * created on Solaris)
5260 zfs_create_attrname(int attrnamespace, const char *name, char *attrname,
5263 const char *namespace, *prefix, *suffix;
5265 /* We don't allow '/' character in attribute name. */
5266 if (strchr(name, '/') != NULL)
5267 return (SET_ERROR(EINVAL));
5268 /* We don't allow attribute names that start with "freebsd:" string. */
5269 if (strncmp(name, "freebsd:", 8) == 0)
5270 return (SET_ERROR(EINVAL));
5272 bzero(attrname, size);
5274 switch (attrnamespace) {
5275 case EXTATTR_NAMESPACE_USER:
5277 prefix = "freebsd:";
5278 namespace = EXTATTR_NAMESPACE_USER_STRING;
5282 * This is the default namespace by which we can access all
5283 * attributes created on Solaris.
5285 prefix = namespace = suffix = "";
5288 case EXTATTR_NAMESPACE_SYSTEM:
5289 prefix = "freebsd:";
5290 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING;
5293 case EXTATTR_NAMESPACE_EMPTY:
5295 return (SET_ERROR(EINVAL));
5297 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix,
5299 return (SET_ERROR(ENAMETOOLONG));
5304 #ifndef _SYS_SYSPROTO_H_
5305 struct vop_getextattr {
5306 IN struct vnode *a_vp;
5307 IN int a_attrnamespace;
5308 IN const char *a_name;
5309 INOUT struct uio *a_uio;
5311 IN struct ucred *a_cred;
5312 IN struct thread *a_td;
5317 * Vnode operating to retrieve a named extended attribute.
5320 zfs_getextattr(struct vop_getextattr_args *ap)
5322 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
5323 struct thread *td = ap->a_td;
5324 struct nameidata nd;
5327 vnode_t *xvp = NULL, *vp;
5331 * If the xattr property is off, refuse the request.
5333 if (!(zfsvfs->z_flags & ZSB_XATTR)) {
5334 return (SET_ERROR(EOPNOTSUPP));
5337 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
5338 ap->a_cred, ap->a_td, VREAD);
5342 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
5349 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
5350 LOOKUP_XATTR, B_FALSE);
5357 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
5359 error = vn_open_cred(&nd, &flags, 0, VN_OPEN_INVFS, ap->a_cred, NULL);
5361 NDFREE(&nd, NDF_ONLY_PNBUF);
5364 if (error == ENOENT)
5369 if (ap->a_size != NULL) {
5370 error = VOP_GETATTR(vp, &va, ap->a_cred);
5372 *ap->a_size = (size_t)va.va_size;
5373 } else if (ap->a_uio != NULL)
5374 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred);
5377 vn_close(vp, flags, ap->a_cred, td);
5382 #ifndef _SYS_SYSPROTO_H_
5383 struct vop_deleteextattr {
5384 IN struct vnode *a_vp;
5385 IN int a_attrnamespace;
5386 IN const char *a_name;
5387 IN struct ucred *a_cred;
5388 IN struct thread *a_td;
5393 * Vnode operation to remove a named attribute.
5396 zfs_deleteextattr(struct vop_deleteextattr_args *ap)
5398 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
5399 struct thread *td = ap->a_td;
5400 struct nameidata nd;
5402 vnode_t *xvp = NULL, *vp;
5406 * If the xattr property is off, refuse the request.
5408 if (!(zfsvfs->z_flags & ZSB_XATTR)) {
5409 return (SET_ERROR(EOPNOTSUPP));
5412 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
5413 ap->a_cred, ap->a_td, VWRITE);
5417 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
5424 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
5425 LOOKUP_XATTR, B_FALSE);
5431 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF,
5432 UIO_SYSSPACE, attrname, xvp, td);
5437 NDFREE(&nd, NDF_ONLY_PNBUF);
5438 if (error == ENOENT)
5443 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
5444 NDFREE(&nd, NDF_ONLY_PNBUF);
5447 if (vp == nd.ni_dvp)
5456 #ifndef _SYS_SYSPROTO_H_
5457 struct vop_setextattr {
5458 IN struct vnode *a_vp;
5459 IN int a_attrnamespace;
5460 IN const char *a_name;
5461 INOUT struct uio *a_uio;
5462 IN struct ucred *a_cred;
5463 IN struct thread *a_td;
5468 * Vnode operation to set a named attribute.
5471 zfs_setextattr(struct vop_setextattr_args *ap)
5473 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
5474 struct thread *td = ap->a_td;
5475 struct nameidata nd;
5478 vnode_t *xvp = NULL, *vp;
5482 * If the xattr property is off, refuse the request.
5484 if (!(zfsvfs->z_flags & ZSB_XATTR)) {
5485 return (SET_ERROR(EOPNOTSUPP));
5488 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
5489 ap->a_cred, ap->a_td, VWRITE);
5492 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
5499 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
5500 LOOKUP_XATTR | CREATE_XATTR_DIR, B_FALSE);
5506 flags = FFLAGS(O_WRONLY | O_CREAT);
5507 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
5509 error = vn_open_cred(&nd, &flags, 0600, VN_OPEN_INVFS, ap->a_cred,
5512 NDFREE(&nd, NDF_ONLY_PNBUF);
5520 error = VOP_SETATTR(vp, &va, ap->a_cred);
5522 VOP_WRITE(vp, ap->a_uio, IO_UNIT, ap->a_cred);
5525 vn_close(vp, flags, ap->a_cred, td);
5530 #ifndef _SYS_SYSPROTO_H_
5531 struct vop_listextattr {
5532 IN struct vnode *a_vp;
5533 IN int a_attrnamespace;
5534 INOUT struct uio *a_uio;
5536 IN struct ucred *a_cred;
5537 IN struct thread *a_td;
5542 * Vnode operation to retrieve extended attributes on a vnode.
5545 zfs_listextattr(struct vop_listextattr_args *ap)
5547 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
5548 struct thread *td = ap->a_td;
5549 struct nameidata nd;
5550 char attrprefix[16];
5551 uint8_t dirbuf[sizeof (struct dirent)];
5555 size_t *sizep = ap->a_size;
5557 vnode_t *xvp = NULL, *vp;
5558 int done, error, eof, pos;
5561 zfs_uio_init(&uio, ap->a_uio);
5564 * If the xattr property is off, refuse the request.
5566 if (!(zfsvfs->z_flags & ZSB_XATTR)) {
5567 return (SET_ERROR(EOPNOTSUPP));
5570 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
5571 ap->a_cred, ap->a_td, VREAD);
5575 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix,
5576 sizeof (attrprefix));
5579 plen = strlen(attrprefix);
5586 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
5587 LOOKUP_XATTR, B_FALSE);
5591 * ENOATTR means that the EA directory does not yet exist,
5592 * i.e. there are no extended attributes there.
5594 if (error == ENOATTR)
5599 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED,
5600 UIO_SYSSPACE, ".", xvp, td);
5603 NDFREE(&nd, NDF_ONLY_PNBUF);
5609 auio.uio_iov = &aiov;
5610 auio.uio_iovcnt = 1;
5611 auio.uio_segflg = UIO_SYSSPACE;
5613 auio.uio_rw = UIO_READ;
5614 auio.uio_offset = 0;
5619 aiov.iov_base = (void *)dirbuf;
5620 aiov.iov_len = sizeof (dirbuf);
5621 auio.uio_resid = sizeof (dirbuf);
5622 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL);
5623 done = sizeof (dirbuf) - auio.uio_resid;
5626 for (pos = 0; pos < done; ) {
5627 dp = (struct dirent *)(dirbuf + pos);
5628 pos += dp->d_reclen;
5630 * XXX: Temporarily we also accept DT_UNKNOWN, as this
5631 * is what we get when attribute was created on Solaris.
5633 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN)
5636 strncmp(dp->d_name, "freebsd:", 8) == 0)
5638 else if (strncmp(dp->d_name, attrprefix, plen) != 0)
5640 nlen = dp->d_namlen - plen;
5643 else if (GET_UIO_STRUCT(&uio) != NULL) {
5645 * Format of extattr name entry is one byte for
5646 * length and the rest for name.
5648 error = zfs_uiomove(&nlen, 1, zfs_uio_rw(&uio),
5651 error = zfs_uiomove(dp->d_name + plen,
5652 nlen, zfs_uio_rw(&uio), &uio);
5658 } while (!eof && error == 0);
5666 #ifndef _SYS_SYSPROTO_H_
5667 struct vop_getacl_args {
5677 zfs_freebsd_getacl(struct vop_getacl_args *ap)
5680 vsecattr_t vsecattr;
5682 if (ap->a_type != ACL_TYPE_NFS4)
5685 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
5686 if ((error = zfs_getsecattr(VTOZ(ap->a_vp),
5687 &vsecattr, 0, ap->a_cred)))
5690 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp,
5691 vsecattr.vsa_aclcnt);
5692 if (vsecattr.vsa_aclentp != NULL)
5693 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz);
5698 #ifndef _SYS_SYSPROTO_H_
5699 struct vop_setacl_args {
5709 zfs_freebsd_setacl(struct vop_setacl_args *ap)
5712 vsecattr_t vsecattr;
5713 int aclbsize; /* size of acl list in bytes */
5716 if (ap->a_type != ACL_TYPE_NFS4)
5719 if (ap->a_aclp == NULL)
5722 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES)
5726 * With NFSv4 ACLs, chmod(2) may need to add additional entries,
5727 * splitting every entry into two and appending "canonical six"
5728 * entries at the end. Don't allow for setting an ACL that would
5729 * cause chmod(2) to run out of ACL entries.
5731 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES)
5734 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR);
5738 vsecattr.vsa_mask = VSA_ACE;
5739 aclbsize = ap->a_aclp->acl_cnt * sizeof (ace_t);
5740 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP);
5741 aaclp = vsecattr.vsa_aclentp;
5742 vsecattr.vsa_aclentsz = aclbsize;
5744 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp);
5745 error = zfs_setsecattr(VTOZ(ap->a_vp), &vsecattr, 0, ap->a_cred);
5746 kmem_free(aaclp, aclbsize);
5751 #ifndef _SYS_SYSPROTO_H_
5752 struct vop_aclcheck_args {
5762 zfs_freebsd_aclcheck(struct vop_aclcheck_args *ap)
5765 return (EOPNOTSUPP);
5769 zfs_vptocnp(struct vop_vptocnp_args *ap)
5771 vnode_t *covered_vp;
5772 vnode_t *vp = ap->a_vp;
5773 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
5774 znode_t *zp = VTOZ(vp);
5782 * If we are a snapshot mounted under .zfs, run the operation
5783 * on the covered vnode.
5785 if (zp->z_id != zfsvfs->z_root || zfsvfs->z_parent == zfsvfs) {
5786 char name[MAXNAMLEN + 1];
5790 error = zfs_znode_parent_and_name(zp, &dzp, name);
5793 if (*ap->a_buflen < len)
5794 error = SET_ERROR(ENOMEM);
5797 *ap->a_buflen -= len;
5798 bcopy(name, ap->a_buf + *ap->a_buflen, len);
5799 *ap->a_vpp = ZTOV(dzp);
5806 covered_vp = vp->v_mount->mnt_vnodecovered;
5807 #if __FreeBSD_version >= 1300045
5808 enum vgetstate vs = vget_prep(covered_vp);
5812 ltype = VOP_ISLOCKED(vp);
5814 #if __FreeBSD_version >= 1300045
5815 error = vget_finish(covered_vp, LK_SHARED, vs);
5817 error = vget(covered_vp, LK_SHARED | LK_VNHELD, curthread);
5820 #if __FreeBSD_version >= 1300123
5821 error = VOP_VPTOCNP(covered_vp, ap->a_vpp, ap->a_buf,
5824 error = VOP_VPTOCNP(covered_vp, ap->a_vpp, ap->a_cred,
5825 ap->a_buf, ap->a_buflen);
5829 vn_lock(vp, ltype | LK_RETRY);
5830 if (VN_IS_DOOMED(vp))
5831 error = SET_ERROR(ENOENT);
5835 struct vop_vector zfs_vnodeops;
5836 struct vop_vector zfs_fifoops;
5837 struct vop_vector zfs_shareops;
5839 struct vop_vector zfs_vnodeops = {
5840 .vop_default = &default_vnodeops,
5841 .vop_inactive = zfs_freebsd_inactive,
5842 #if __FreeBSD_version >= 1300042
5843 .vop_need_inactive = zfs_freebsd_need_inactive,
5845 .vop_reclaim = zfs_freebsd_reclaim,
5846 #if __FreeBSD_version >= 1300102
5847 .vop_fplookup_vexec = zfs_freebsd_fplookup_vexec,
5849 #if __FreeBSD_version >= 1300139
5850 .vop_fplookup_symlink = zfs_freebsd_fplookup_symlink,
5852 .vop_access = zfs_freebsd_access,
5853 .vop_allocate = VOP_EINVAL,
5854 .vop_lookup = zfs_cache_lookup,
5855 .vop_cachedlookup = zfs_freebsd_cachedlookup,
5856 .vop_getattr = zfs_freebsd_getattr,
5857 .vop_setattr = zfs_freebsd_setattr,
5858 .vop_create = zfs_freebsd_create,
5859 .vop_mknod = (vop_mknod_t *)zfs_freebsd_create,
5860 .vop_mkdir = zfs_freebsd_mkdir,
5861 .vop_readdir = zfs_freebsd_readdir,
5862 .vop_fsync = zfs_freebsd_fsync,
5863 .vop_open = zfs_freebsd_open,
5864 .vop_close = zfs_freebsd_close,
5865 .vop_rmdir = zfs_freebsd_rmdir,
5866 .vop_ioctl = zfs_freebsd_ioctl,
5867 .vop_link = zfs_freebsd_link,
5868 .vop_symlink = zfs_freebsd_symlink,
5869 .vop_readlink = zfs_freebsd_readlink,
5870 .vop_read = zfs_freebsd_read,
5871 .vop_write = zfs_freebsd_write,
5872 .vop_remove = zfs_freebsd_remove,
5873 .vop_rename = zfs_freebsd_rename,
5874 .vop_pathconf = zfs_freebsd_pathconf,
5875 .vop_bmap = zfs_freebsd_bmap,
5876 .vop_fid = zfs_freebsd_fid,
5877 .vop_getextattr = zfs_getextattr,
5878 .vop_deleteextattr = zfs_deleteextattr,
5879 .vop_setextattr = zfs_setextattr,
5880 .vop_listextattr = zfs_listextattr,
5881 .vop_getacl = zfs_freebsd_getacl,
5882 .vop_setacl = zfs_freebsd_setacl,
5883 .vop_aclcheck = zfs_freebsd_aclcheck,
5884 .vop_getpages = zfs_freebsd_getpages,
5885 .vop_putpages = zfs_freebsd_putpages,
5886 .vop_vptocnp = zfs_vptocnp,
5887 #if __FreeBSD_version >= 1300064
5888 .vop_lock1 = vop_lock,
5889 .vop_unlock = vop_unlock,
5890 .vop_islocked = vop_islocked,
5893 VFS_VOP_VECTOR_REGISTER(zfs_vnodeops);
5895 struct vop_vector zfs_fifoops = {
5896 .vop_default = &fifo_specops,
5897 .vop_fsync = zfs_freebsd_fsync,
5898 #if __FreeBSD_version >= 1300102
5899 .vop_fplookup_vexec = zfs_freebsd_fplookup_vexec,
5901 #if __FreeBSD_version >= 1300139
5902 .vop_fplookup_symlink = zfs_freebsd_fplookup_symlink,
5904 .vop_access = zfs_freebsd_access,
5905 .vop_getattr = zfs_freebsd_getattr,
5906 .vop_inactive = zfs_freebsd_inactive,
5907 .vop_read = VOP_PANIC,
5908 .vop_reclaim = zfs_freebsd_reclaim,
5909 .vop_setattr = zfs_freebsd_setattr,
5910 .vop_write = VOP_PANIC,
5911 .vop_pathconf = zfs_freebsd_pathconf,
5912 .vop_fid = zfs_freebsd_fid,
5913 .vop_getacl = zfs_freebsd_getacl,
5914 .vop_setacl = zfs_freebsd_setacl,
5915 .vop_aclcheck = zfs_freebsd_aclcheck,
5917 VFS_VOP_VECTOR_REGISTER(zfs_fifoops);
5920 * special share hidden files vnode operations template
5922 struct vop_vector zfs_shareops = {
5923 .vop_default = &default_vnodeops,
5924 #if __FreeBSD_version >= 1300121
5925 .vop_fplookup_vexec = VOP_EAGAIN,
5927 #if __FreeBSD_version >= 1300139
5928 .vop_fplookup_symlink = VOP_EAGAIN,
5930 .vop_access = zfs_freebsd_access,
5931 .vop_inactive = zfs_freebsd_inactive,
5932 .vop_reclaim = zfs_freebsd_reclaim,
5933 .vop_fid = zfs_freebsd_fid,
5934 .vop_pathconf = zfs_freebsd_pathconf,
5936 VFS_VOP_VECTOR_REGISTER(zfs_shareops);