4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
26 #include <sys/types.h>
27 #include <sys/param.h>
29 #include <sys/systm.h>
30 #include <sys/sysmacros.h>
31 #include <sys/resource.h>
33 #include <sys/vnode.h>
38 #include <sys/pathname.h>
39 #include <sys/cmn_err.h>
40 #include <sys/errno.h>
42 #include <sys/unistd.h>
43 #include <sys/sunddi.h>
44 #include <sys/random.h>
45 #include <sys/policy.h>
46 #include <sys/zfs_dir.h>
47 #include <sys/zfs_acl.h>
48 #include <sys/fs/zfs.h>
49 #include "fs/fs_subr.h"
52 #include <sys/atomic.h>
53 #include <sys/zfs_fuid.h>
55 #include <sys/zfs_sa.h>
57 #include <sys/extdirent.h>
60 * zfs_match_find() is used by zfs_dirent_lock() to peform zap lookups
61 * of names after deciding which is the appropriate lookup interface.
64 zfs_match_find(zfs_sb_t *zsb, znode_t *dzp, char *name, boolean_t exact,
65 boolean_t update, int *deflags, pathname_t *rpnp, uint64_t *zoid)
67 boolean_t conflict = B_FALSE;
71 matchtype_t mt = MT_FIRST;
77 bufsz = rpnp->pn_bufsize;
82 * In the non-mixed case we only expect there would ever
83 * be one match, but we need to use the normalizing lookup.
85 error = zap_lookup_norm(zsb->z_os, dzp->z_id, name, 8, 1,
86 zoid, mt, buf, bufsz, &conflict);
88 error = zap_lookup(zsb->z_os, dzp->z_id, name, 8, 1, zoid);
92 * Allow multiple entries provided the first entry is
93 * the object id. Non-zpl consumers may safely make
94 * use of the additional space.
96 * XXX: This should be a feature flag for compatibility
98 if (error == EOVERFLOW)
101 if (zsb->z_norm && !error && deflags)
102 *deflags = conflict ? ED_CASE_CONFLICT : 0;
104 *zoid = ZFS_DIRENT_OBJ(*zoid);
107 if (error == ENOENT && update)
108 dnlc_update(ZTOI(dzp), name, DNLC_NO_VNODE);
109 #endif /* HAVE_DNLC */
115 * Lock a directory entry. A dirlock on <dzp, name> protects that name
116 * in dzp's directory zap object. As long as you hold a dirlock, you can
117 * assume two things: (1) dzp cannot be reaped, and (2) no other thread
118 * can change the zap entry for (i.e. link or unlink) this name.
121 * dzp - znode for directory
122 * name - name of entry to lock
123 * flag - ZNEW: if the entry already exists, fail with EEXIST.
124 * ZEXISTS: if the entry does not exist, fail with ENOENT.
125 * ZSHARED: allow concurrent access with other ZSHARED callers.
126 * ZXATTR: we want dzp's xattr directory
127 * ZCILOOK: On a mixed sensitivity file system,
128 * this lookup should be case-insensitive.
129 * ZCIEXACT: On a purely case-insensitive file system,
130 * this lookup should be case-sensitive.
131 * ZRENAMING: we are locking for renaming, force narrow locks
132 * ZHAVELOCK: Don't grab the z_name_lock for this call. The
133 * current thread already holds it.
136 * zpp - pointer to the znode for the entry (NULL if there isn't one)
137 * dlpp - pointer to the dirlock for this entry (NULL on error)
138 * direntflags - (case-insensitive lookup only)
139 * flags if multiple case-sensitive matches exist in directory
140 * realpnp - (case-insensitive lookup only)
141 * actual name matched within the directory
143 * Return value: 0 on success or errno on failure.
145 * NOTE: Always checks for, and rejects, '.' and '..'.
146 * NOTE: For case-insensitive file systems we take wide locks (see below),
147 * but return znode pointers to a single match.
150 zfs_dirent_lock(zfs_dirlock_t **dlpp, znode_t *dzp, char *name, znode_t **zpp,
151 int flag, int *direntflags, pathname_t *realpnp)
153 zfs_sb_t *zsb = ZTOZSB(dzp);
160 #endif /* HAVE_DNLC */
168 * Verify that we are not trying to lock '.', '..', or '.zfs'
170 if ((name[0] == '.' &&
171 (name[1] == '\0' || (name[1] == '.' && name[2] == '\0'))) ||
172 (zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0))
176 * Case sensitivity and normalization preferences are set when
177 * the file system is created. These are stored in the
178 * zsb->z_case and zsb->z_norm fields. These choices
179 * affect what vnodes can be cached in the DNLC, how we
180 * perform zap lookups, and the "width" of our dirlocks.
182 * A normal dirlock locks a single name. Note that with
183 * normalization a name can be composed multiple ways, but
184 * when normalized, these names all compare equal. A wide
185 * dirlock locks multiple names. We need these when the file
186 * system is supporting mixed-mode access. It is sometimes
187 * necessary to lock all case permutations of file name at
188 * once so that simultaneous case-insensitive/case-sensitive
189 * behaves as rationally as possible.
193 * Decide if exact matches should be requested when performing
194 * a zap lookup on file systems supporting case-insensitive
198 ((zsb->z_case == ZFS_CASE_INSENSITIVE) && (flag & ZCIEXACT)) ||
199 ((zsb->z_case == ZFS_CASE_MIXED) && !(flag & ZCILOOK));
202 * Only look in or update the DNLC if we are looking for the
203 * name on a file system that does not require normalization
204 * or case folding. We can also look there if we happen to be
205 * on a non-normalizing, mixed sensitivity file system IF we
206 * are looking for the exact name.
208 * Maybe can add TO-UPPERed version of name to dnlc in ci-only
209 * case for performance improvement?
211 update = !zsb->z_norm ||
212 ((zsb->z_case == ZFS_CASE_MIXED) &&
213 !(zsb->z_norm & ~U8_TEXTPREP_TOUPPER) && !(flag & ZCILOOK));
216 * ZRENAMING indicates we are in a situation where we should
217 * take narrow locks regardless of the file system's
218 * preferences for normalizing and case folding. This will
219 * prevent us deadlocking trying to grab the same wide lock
220 * twice if the two names happen to be case-insensitive
223 if (flag & ZRENAMING)
226 cmpflags = zsb->z_norm;
229 * Wait until there are no locks on this name.
231 * Don't grab the the lock if it is already held. However, cannot
232 * have both ZSHARED and ZHAVELOCK together.
234 ASSERT(!(flag & ZSHARED) || !(flag & ZHAVELOCK));
235 if (!(flag & ZHAVELOCK))
236 rw_enter(&dzp->z_name_lock, RW_READER);
238 mutex_enter(&dzp->z_lock);
240 if (dzp->z_unlinked) {
241 mutex_exit(&dzp->z_lock);
242 if (!(flag & ZHAVELOCK))
243 rw_exit(&dzp->z_name_lock);
246 for (dl = dzp->z_dirlocks; dl != NULL; dl = dl->dl_next) {
247 if ((u8_strcmp(name, dl->dl_name, 0, cmpflags,
248 U8_UNICODE_LATEST, &error) == 0) || error != 0)
252 mutex_exit(&dzp->z_lock);
253 if (!(flag & ZHAVELOCK))
254 rw_exit(&dzp->z_name_lock);
259 * Allocate a new dirlock and add it to the list.
261 dl = kmem_alloc(sizeof (zfs_dirlock_t), KM_SLEEP);
262 cv_init(&dl->dl_cv, NULL, CV_DEFAULT, NULL);
268 dl->dl_next = dzp->z_dirlocks;
269 dzp->z_dirlocks = dl;
272 if ((flag & ZSHARED) && dl->dl_sharecnt != 0)
274 cv_wait(&dl->dl_cv, &dzp->z_lock);
278 * If the z_name_lock was NOT held for this dirlock record it.
280 if (flag & ZHAVELOCK)
283 if ((flag & ZSHARED) && ++dl->dl_sharecnt > 1 && dl->dl_namesize == 0) {
285 * We're the second shared reference to dl. Make a copy of
286 * dl_name in case the first thread goes away before we do.
287 * Note that we initialize the new name before storing its
288 * pointer into dl_name, because the first thread may load
289 * dl->dl_name at any time. He'll either see the old value,
290 * which is his, or the new shared copy; either is OK.
292 dl->dl_namesize = strlen(dl->dl_name) + 1;
293 name = kmem_alloc(dl->dl_namesize, KM_SLEEP);
294 bcopy(dl->dl_name, name, dl->dl_namesize);
298 mutex_exit(&dzp->z_lock);
301 * We have a dirlock on the name. (Note that it is the dirlock,
302 * not the dzp's z_lock, that protects the name in the zap object.)
303 * See if there's an object by this name; if so, put a hold on it.
306 error = sa_lookup(dzp->z_sa_hdl, SA_ZPL_XATTR(zsb), &zoid,
309 error = (zoid == 0 ? ENOENT : 0);
313 vp = dnlc_lookup(ZTOI(dzp), name);
314 if (vp == DNLC_NO_VNODE) {
319 zfs_dirent_unlock(dl);
327 error = zfs_match_find(zsb, dzp, name, exact,
328 update, direntflags, realpnp, &zoid);
331 error = zfs_match_find(zsb, dzp, name, exact,
332 update, direntflags, realpnp, &zoid);
333 #endif /* HAVE_DNLC */
336 if (error != ENOENT || (flag & ZEXISTS)) {
337 zfs_dirent_unlock(dl);
342 zfs_dirent_unlock(dl);
345 error = zfs_zget(zsb, zoid, zpp);
347 zfs_dirent_unlock(dl);
351 if (!(flag & ZXATTR) && update)
352 dnlc_update(ZTOI(dzp), name, ZTOI(*zpp));
353 #endif /* HAVE_DNLC */
362 * Unlock this directory entry and wake anyone who was waiting for it.
365 zfs_dirent_unlock(zfs_dirlock_t *dl)
367 znode_t *dzp = dl->dl_dzp;
368 zfs_dirlock_t **prev_dl, *cur_dl;
370 mutex_enter(&dzp->z_lock);
372 if (!dl->dl_namelock)
373 rw_exit(&dzp->z_name_lock);
375 if (dl->dl_sharecnt > 1) {
377 mutex_exit(&dzp->z_lock);
380 prev_dl = &dzp->z_dirlocks;
381 while ((cur_dl = *prev_dl) != dl)
382 prev_dl = &cur_dl->dl_next;
383 *prev_dl = dl->dl_next;
384 cv_broadcast(&dl->dl_cv);
385 mutex_exit(&dzp->z_lock);
387 if (dl->dl_namesize != 0)
388 kmem_free(dl->dl_name, dl->dl_namesize);
389 cv_destroy(&dl->dl_cv);
390 kmem_free(dl, sizeof (*dl));
394 * Look up an entry in a directory.
396 * NOTE: '.' and '..' are handled as special cases because
397 * no directory entries are actually stored for them. If this is
398 * the root of a filesystem, then '.zfs' is also treated as a
399 * special pseudo-directory.
402 zfs_dirlook(znode_t *dzp, char *name, struct inode **ipp, int flags,
403 int *deflg, pathname_t *rpnp)
410 if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) {
413 } else if (name[0] == '.' && name[1] == '.' && name[2] == 0) {
414 zfs_sb_t *zsb = ZTOZSB(dzp);
417 * If we are a snapshot mounted under .zfs, return
418 * the vp for the snapshot directory.
420 if ((error = sa_lookup(dzp->z_sa_hdl,
421 SA_ZPL_PARENT(zsb), &parent, sizeof (parent))) != 0)
424 if (parent == dzp->z_id && zsb->z_parent != zsb) {
425 error = zfsctl_root_lookup(zsb->z_parent->z_ctldir,
426 "snapshot", ipp, NULL, 0, NULL, kcred,
430 #endif /* HAVE_SNAPSHOT */
431 rw_enter(&dzp->z_parent_lock, RW_READER);
432 error = zfs_zget(zsb, parent, &zp);
435 rw_exit(&dzp->z_parent_lock);
437 } else if (zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0) {
438 *ipp = zfsctl_root(dzp);
439 #endif /* HAVE_SNAPSHOT */
443 zf = ZEXISTS | ZSHARED;
444 if (flags & FIGNORECASE)
447 error = zfs_dirent_lock(&dl, dzp, name, &zp, zf, deflg, rpnp);
450 zfs_dirent_unlock(dl);
451 dzp->z_zn_prefetch = B_TRUE; /* enable prefetching */
456 if ((flags & FIGNORECASE) && rpnp && !error)
457 (void) strlcpy(rpnp->pn_buf, name, rpnp->pn_bufsize);
463 * unlinked Set (formerly known as the "delete queue") Error Handling
465 * When dealing with the unlinked set, we dmu_tx_hold_zap(), but we
466 * don't specify the name of the entry that we will be manipulating. We
467 * also fib and say that we won't be adding any new entries to the
468 * unlinked set, even though we might (this is to lower the minimum file
469 * size that can be deleted in a full filesystem). So on the small
470 * chance that the nlink list is using a fat zap (ie. has more than
471 * 2000 entries), we *may* not pre-read a block that's needed.
472 * Therefore it is remotely possible for some of the assertions
473 * regarding the unlinked set below to fail due to i/o error. On a
474 * nondebug system, this will result in the space being leaked.
477 zfs_unlinked_add(znode_t *zp, dmu_tx_t *tx)
479 zfs_sb_t *zsb = ZTOZSB(zp);
481 ASSERT(zp->z_unlinked);
482 ASSERT(zp->z_links == 0);
485 zap_add_int(zsb->z_os, zsb->z_unlinkedobj, zp->z_id, tx));
489 * Delete the entire contents of a directory. Return a count
490 * of the number of entries that could not be deleted. If we encounter
491 * an error, return a count of at least one so that the directory stays
492 * in the unlinked set.
494 * NOTE: this function assumes that the directory is inactive,
495 * so there is no need to lock its entries before deletion.
496 * Also, it assumes the directory contents is *only* regular
500 zfs_purgedir(znode_t *dzp)
506 zfs_sb_t *zsb = ZTOZSB(dzp);
511 for (zap_cursor_init(&zc, zsb->z_os, dzp->z_id);
512 (error = zap_cursor_retrieve(&zc, &zap)) == 0;
513 zap_cursor_advance(&zc)) {
514 error = zfs_zget(zsb,
515 ZFS_DIRENT_OBJ(zap.za_first_integer), &xzp);
521 ASSERT(S_ISREG(ZTOI(xzp)->i_mode)||S_ISLNK(ZTOI(xzp)->i_mode));
523 tx = dmu_tx_create(zsb->z_os);
524 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
525 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, zap.za_name);
526 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
527 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
528 /* Is this really needed ? */
529 zfs_sa_upgrade_txholds(tx, xzp);
530 error = dmu_tx_assign(tx, TXG_WAIT);
537 bzero(&dl, sizeof (dl));
539 dl.dl_name = zap.za_name;
541 error = zfs_link_destroy(&dl, xzp, tx, 0, NULL);
548 zap_cursor_fini(&zc);
555 * Clean up any znodes that had no links when we either crashed or
556 * (force) umounted the file system.
559 zfs_unlinked_drain(zfs_sb_t *zsb)
563 dmu_object_info_t doi;
568 * Interate over the contents of the unlinked set.
570 for (zap_cursor_init(&zc, zsb->z_os, zsb->z_unlinkedobj);
571 zap_cursor_retrieve(&zc, &zap) == 0;
572 zap_cursor_advance(&zc)) {
575 * See what kind of object we have in list
578 error = dmu_object_info(zsb->z_os, zap.za_first_integer, &doi);
582 ASSERT((doi.doi_type == DMU_OT_PLAIN_FILE_CONTENTS) ||
583 (doi.doi_type == DMU_OT_DIRECTORY_CONTENTS));
585 * We need to re-mark these list entries for deletion,
586 * so we pull them back into core and set zp->z_unlinked.
588 error = zfs_zget(zsb, zap.za_first_integer, &zp);
591 * We may pick up znodes that are already marked for deletion.
592 * This could happen during the purge of an extended attribute
593 * directory. All we need to do is skip over them, since they
594 * are already in the system marked z_unlinked.
599 zp->z_unlinked = B_TRUE;
602 * If this is an attribute directory, purge its contents.
604 if (S_ISDIR(ZTOI(zp)->i_mode) && (zp->z_pflags & ZFS_XATTR)) {
606 * We don't need to check the return value of
607 * zfs_purgedir here, because zfs_rmnode will just
608 * return this xattr directory to the unlinked set
609 * until all of its xattrs are gone.
611 (void) zfs_purgedir(zp);
616 zap_cursor_fini(&zc);
620 zfs_rmnode(znode_t *zp)
622 zfs_sb_t *zsb = ZTOZSB(zp);
623 objset_t *os = zsb->z_os;
631 ASSERT(zp->z_links == 0);
632 ASSERT(atomic_read(&ZTOI(zp)->i_count) == 0);
635 * If this is an attribute directory, purge its contents.
637 if (S_ISDIR(ZTOI(zp)->i_mode) && (zp->z_pflags & ZFS_XATTR)) {
638 error = zap_count(os, zp->z_id, &count);
640 zfs_znode_dmu_fini(zp);
648 * There are still directory entries in this xattr
649 * directory. Let zfs_unlinked_drain() deal with
650 * them to avoid deadlocking this process in the
651 * zfs_purgedir()->zfs_zget()->ilookup() callpath
652 * on the xattr inode's I_FREEING bit.
654 taskq = dsl_pool_iput_taskq(dmu_objset_pool(os));
655 taskq_dispatch(taskq, (task_func_t *)
656 zfs_unlinked_drain, zsb, TQ_SLEEP);
658 zfs_znode_dmu_fini(zp);
664 * Free up all the data in the file.
666 error = dmu_free_long_range(os, zp->z_id, 0, DMU_OBJECT_END);
669 * Not enough space. Leave the file in the unlinked set.
671 zfs_znode_dmu_fini(zp);
676 * If the file has extended attributes, we're going to unlink
679 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zsb),
680 &xattr_obj, sizeof (xattr_obj));
681 if (error == 0 && xattr_obj) {
682 error = zfs_zget(zsb, xattr_obj, &xzp);
686 acl_obj = zfs_external_acl(zp);
689 * Set up the final transaction.
691 tx = dmu_tx_create(os);
692 dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END);
693 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
695 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, TRUE, NULL);
696 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
699 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
701 zfs_sa_upgrade_txholds(tx, zp);
702 error = dmu_tx_assign(tx, TXG_WAIT);
705 * Not enough space to delete the file. Leave it in the
706 * unlinked set, leaking it until the fs is remounted (at
707 * which point we'll call zfs_unlinked_drain() to process it).
710 zfs_znode_dmu_fini(zp);
716 mutex_enter(&xzp->z_lock);
717 xzp->z_unlinked = B_TRUE; /* mark xzp for deletion */
718 xzp->z_links = 0; /* no more links to it */
719 VERIFY(0 == sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zsb),
720 &xzp->z_links, sizeof (xzp->z_links), tx));
721 mutex_exit(&xzp->z_lock);
722 zfs_unlinked_add(xzp, tx);
725 /* Remove this znode from the unlinked set */
727 zap_remove_int(zsb->z_os, zsb->z_unlinkedobj, zp->z_id, tx));
729 zfs_znode_delete(zp, tx);
738 zfs_dirent(znode_t *zp, uint64_t mode)
740 uint64_t de = zp->z_id;
742 if (ZTOZSB(zp)->z_version >= ZPL_VERSION_DIRENT_TYPE)
743 de |= IFTODT(mode) << 60;
748 * Link zp into dl. Can only fail if zp has been unlinked.
751 zfs_link_create(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag)
753 znode_t *dzp = dl->dl_dzp;
754 zfs_sb_t *zsb = ZTOZSB(zp);
756 int zp_is_dir = S_ISDIR(ZTOI(zp)->i_mode);
757 sa_bulk_attr_t bulk[5];
758 uint64_t mtime[2], ctime[2];
762 mutex_enter(&zp->z_lock);
764 if (!(flag & ZRENAMING)) {
765 if (zp->z_unlinked) { /* no new links to unlinked zp */
766 ASSERT(!(flag & (ZNEW | ZEXISTS)));
767 mutex_exit(&zp->z_lock);
771 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zsb), NULL,
772 &zp->z_links, sizeof (zp->z_links));
775 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zsb), NULL,
776 &dzp->z_id, sizeof (dzp->z_id));
777 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb), NULL,
778 &zp->z_pflags, sizeof (zp->z_pflags));
780 if (!(flag & ZNEW)) {
781 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL,
782 ctime, sizeof (ctime));
783 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime,
786 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
789 mutex_exit(&zp->z_lock);
791 mutex_enter(&dzp->z_lock);
793 dzp->z_links += zp_is_dir;
795 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zsb), NULL,
796 &dzp->z_size, sizeof (dzp->z_size));
797 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zsb), NULL,
798 &dzp->z_links, sizeof (dzp->z_links));
799 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL,
800 mtime, sizeof (mtime));
801 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL,
802 ctime, sizeof (ctime));
803 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb), NULL,
804 &dzp->z_pflags, sizeof (dzp->z_pflags));
805 zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
806 error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx);
808 mutex_exit(&dzp->z_lock);
810 value = zfs_dirent(zp, zp->z_mode);
811 error = zap_add(ZTOZSB(zp)->z_os, dzp->z_id, dl->dl_name,
819 zfs_dropname(zfs_dirlock_t *dl, znode_t *zp, znode_t *dzp, dmu_tx_t *tx,
824 if (ZTOZSB(zp)->z_norm) {
825 if (((ZTOZSB(zp)->z_case == ZFS_CASE_INSENSITIVE) &&
826 (flag & ZCIEXACT)) ||
827 ((ZTOZSB(zp)->z_case == ZFS_CASE_MIXED) &&
829 error = zap_remove_norm(ZTOZSB(zp)->z_os,
830 dzp->z_id, dl->dl_name, MT_EXACT, tx);
832 error = zap_remove_norm(ZTOZSB(zp)->z_os,
833 dzp->z_id, dl->dl_name, MT_FIRST, tx);
835 error = zap_remove(ZTOZSB(zp)->z_os,
836 dzp->z_id, dl->dl_name, tx);
843 * Unlink zp from dl, and mark zp for deletion if this was the last link.
844 * Can fail if zp is a mount point (EBUSY) or a non-empty directory (EEXIST).
845 * If 'unlinkedp' is NULL, we put unlinked znodes on the unlinked list.
846 * If it's non-NULL, we use it to indicate whether the znode needs deletion,
847 * and it's the caller's job to do it.
850 zfs_link_destroy(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag,
851 boolean_t *unlinkedp)
853 znode_t *dzp = dl->dl_dzp;
854 zfs_sb_t *zsb = ZTOZSB(dzp);
855 int zp_is_dir = S_ISDIR(ZTOI(zp)->i_mode);
856 boolean_t unlinked = B_FALSE;
857 sa_bulk_attr_t bulk[5];
858 uint64_t mtime[2], ctime[2];
863 dnlc_remove(ZTOI(dzp), dl->dl_name);
864 #endif /* HAVE_DNLC */
866 if (!(flag & ZRENAMING)) {
867 mutex_enter(&zp->z_lock);
869 if (zp_is_dir && !zfs_dirempty(zp)) {
870 mutex_exit(&zp->z_lock);
875 * If we get here, we are going to try to remove the object.
876 * First try removing the name from the directory; if that
877 * fails, return the error.
879 error = zfs_dropname(dl, zp, dzp, tx, flag);
881 mutex_exit(&zp->z_lock);
885 if (zp->z_links <= zp_is_dir) {
886 zfs_panic_recover("zfs: link count on %lu is %u, "
887 "should be at least %u", zp->z_id,
888 (int)zp->z_links, zp_is_dir + 1);
889 zp->z_links = zp_is_dir + 1;
891 if (--zp->z_links == zp_is_dir) {
892 zp->z_unlinked = B_TRUE;
896 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb),
897 NULL, &ctime, sizeof (ctime));
898 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb),
899 NULL, &zp->z_pflags, sizeof (zp->z_pflags));
900 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
903 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zsb),
904 NULL, &zp->z_links, sizeof (zp->z_links));
905 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
908 mutex_exit(&zp->z_lock);
910 error = zfs_dropname(dl, zp, dzp, tx, flag);
915 mutex_enter(&dzp->z_lock);
916 dzp->z_size--; /* one dirent removed */
917 dzp->z_links -= zp_is_dir; /* ".." link from zp */
918 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zsb),
919 NULL, &dzp->z_links, sizeof (dzp->z_links));
920 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zsb),
921 NULL, &dzp->z_size, sizeof (dzp->z_size));
922 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb),
923 NULL, ctime, sizeof (ctime));
924 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb),
925 NULL, mtime, sizeof (mtime));
926 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb),
927 NULL, &dzp->z_pflags, sizeof (dzp->z_pflags));
928 zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
929 error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx);
931 mutex_exit(&dzp->z_lock);
933 if (unlinkedp != NULL)
934 *unlinkedp = unlinked;
936 zfs_unlinked_add(zp, tx);
942 * Indicate whether the directory is empty. Works with or without z_lock
943 * held, but can only be consider a hint in the latter case. Returns true
944 * if only "." and ".." remain and there's no work in progress.
947 zfs_dirempty(znode_t *dzp)
949 return (dzp->z_size == 2 && dzp->z_dirlocks == 0);
953 zfs_make_xattrdir(znode_t *zp, vattr_t *vap, struct inode **xipp, cred_t *cr)
955 zfs_sb_t *zsb = ZTOZSB(zp);
959 zfs_acl_ids_t acl_ids;
960 boolean_t fuid_dirtied;
967 if ((error = zfs_zaccess(zp, ACE_WRITE_NAMED_ATTRS, 0, B_FALSE, cr)))
970 if ((error = zfs_acl_ids_create(zp, IS_XATTR, vap, cr, NULL,
973 if (zfs_acl_ids_overquota(zsb, &acl_ids)) {
974 zfs_acl_ids_free(&acl_ids);
979 tx = dmu_tx_create(zsb->z_os);
980 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
981 ZFS_SA_BASE_ATTR_SIZE);
982 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
983 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
984 fuid_dirtied = zsb->z_fuid_dirty;
986 zfs_fuid_txhold(zsb, tx);
987 error = dmu_tx_assign(tx, TXG_NOWAIT);
989 if (error == ERESTART) {
994 zfs_acl_ids_free(&acl_ids);
998 zfs_mknode(zp, vap, tx, cr, IS_XATTR, &xzp, &acl_ids);
1001 zfs_fuid_sync(zsb, tx);
1004 error = sa_lookup(xzp->z_sa_hdl, SA_ZPL_PARENT(zsb),
1005 &parent, sizeof (parent));
1006 ASSERT(error == 0 && parent == zp->z_id);
1009 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_XATTR(zsb), &xzp->z_id,
1010 sizeof (xzp->z_id), tx));
1012 (void) zfs_log_create(zsb->z_log, tx, TX_MKXATTR, zp,
1013 xzp, "", NULL, acl_ids.z_fuidp, vap);
1015 zfs_acl_ids_free(&acl_ids);
1024 * Return a znode for the extended attribute directory for zp.
1025 * ** If the directory does not already exist, it is created **
1027 * IN: zp - znode to obtain attribute directory from
1028 * cr - credentials of caller
1029 * flags - flags from the VOP_LOOKUP call
1031 * OUT: xipp - pointer to extended attribute znode
1033 * RETURN: 0 on success
1034 * error number on failure
1037 zfs_get_xattrdir(znode_t *zp, struct inode **xipp, cred_t *cr, int flags)
1039 zfs_sb_t *zsb = ZTOZSB(zp);
1045 error = zfs_dirent_lock(&dl, zp, "", &xzp, ZXATTR, NULL, NULL);
1051 zfs_dirent_unlock(dl);
1055 if (!(flags & CREATE_XATTR_DIR)) {
1056 zfs_dirent_unlock(dl);
1060 if (zfs_is_readonly(zsb)) {
1061 zfs_dirent_unlock(dl);
1066 * The ability to 'create' files in an attribute
1067 * directory comes from the write_xattr permission on the base file.
1069 * The ability to 'search' an attribute directory requires
1070 * read_xattr permission on the base file.
1072 * Once in a directory the ability to read/write attributes
1073 * is controlled by the permissions on the attribute file.
1075 va.va_mask = ATTR_MODE | ATTR_UID | ATTR_GID;
1076 va.va_mode = S_IFDIR | S_ISVTX | 0777;
1077 zfs_fuid_map_ids(zp, cr, &va.va_uid, &va.va_gid);
1079 va.va_dentry = NULL;
1080 error = zfs_make_xattrdir(zp, &va, xipp, cr);
1081 zfs_dirent_unlock(dl);
1083 if (error == ERESTART) {
1084 /* NB: we already did dmu_tx_wait() if necessary */
1092 * Decide whether it is okay to remove within a sticky directory.
1094 * In sticky directories, write access is not sufficient;
1095 * you can remove entries from a directory only if:
1097 * you own the directory,
1098 * you own the entry,
1099 * the entry is a plain file and you have write access,
1100 * or you are privileged (checked in secpolicy...).
1102 * The function returns 0 if remove access is granted.
1105 zfs_sticky_remove_access(znode_t *zdp, znode_t *zp, cred_t *cr)
1110 zfs_sb_t *zsb = ZTOZSB(zdp);
1115 if ((zdp->z_mode & S_ISVTX) == 0)
1118 downer = zfs_fuid_map_id(zsb, zdp->z_uid, cr, ZFS_OWNER);
1119 fowner = zfs_fuid_map_id(zsb, zp->z_uid, cr, ZFS_OWNER);
1121 if ((uid = crgetuid(cr)) == downer || uid == fowner ||
1122 (S_ISDIR(ZTOI(zp)->i_mode) &&
1123 zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr) == 0))
1126 return (secpolicy_vnode_remove(cr));