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)
70 matchtype_t mt = MT_FIRST;
71 boolean_t conflict = B_FALSE;
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);
87 if (!error && deflags)
88 *deflags = conflict ? ED_CASE_CONFLICT : 0;
90 error = zap_lookup(zsb->z_os, dzp->z_id, name, 8, 1, zoid);
92 *zoid = ZFS_DIRENT_OBJ(*zoid);
95 if (error == ENOENT && update)
96 dnlc_update(ZTOI(dzp), name, DNLC_NO_VNODE);
97 #endif /* HAVE_DNLC */
103 * Lock a directory entry. A dirlock on <dzp, name> protects that name
104 * in dzp's directory zap object. As long as you hold a dirlock, you can
105 * assume two things: (1) dzp cannot be reaped, and (2) no other thread
106 * can change the zap entry for (i.e. link or unlink) this name.
109 * dzp - znode for directory
110 * name - name of entry to lock
111 * flag - ZNEW: if the entry already exists, fail with EEXIST.
112 * ZEXISTS: if the entry does not exist, fail with ENOENT.
113 * ZSHARED: allow concurrent access with other ZSHARED callers.
114 * ZXATTR: we want dzp's xattr directory
115 * ZCILOOK: On a mixed sensitivity file system,
116 * this lookup should be case-insensitive.
117 * ZCIEXACT: On a purely case-insensitive file system,
118 * this lookup should be case-sensitive.
119 * ZRENAMING: we are locking for renaming, force narrow locks
120 * ZHAVELOCK: Don't grab the z_name_lock for this call. The
121 * current thread already holds it.
124 * zpp - pointer to the znode for the entry (NULL if there isn't one)
125 * dlpp - pointer to the dirlock for this entry (NULL on error)
126 * direntflags - (case-insensitive lookup only)
127 * flags if multiple case-sensitive matches exist in directory
128 * realpnp - (case-insensitive lookup only)
129 * actual name matched within the directory
131 * Return value: 0 on success or errno on failure.
133 * NOTE: Always checks for, and rejects, '.' and '..'.
134 * NOTE: For case-insensitive file systems we take wide locks (see below),
135 * but return znode pointers to a single match.
138 zfs_dirent_lock(zfs_dirlock_t **dlpp, znode_t *dzp, char *name, znode_t **zpp,
139 int flag, int *direntflags, pathname_t *realpnp)
141 zfs_sb_t *zsb = ZTOZSB(dzp);
148 #endif /* HAVE_DNLC */
156 * Verify that we are not trying to lock '.', '..', or '.zfs'
158 if ((name[0] == '.' &&
159 (name[1] == '\0' || (name[1] == '.' && name[2] == '\0'))) ||
160 (zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0))
164 * Case sensitivity and normalization preferences are set when
165 * the file system is created. These are stored in the
166 * zsb->z_case and zsb->z_norm fields. These choices
167 * affect what vnodes can be cached in the DNLC, how we
168 * perform zap lookups, and the "width" of our dirlocks.
170 * A normal dirlock locks a single name. Note that with
171 * normalization a name can be composed multiple ways, but
172 * when normalized, these names all compare equal. A wide
173 * dirlock locks multiple names. We need these when the file
174 * system is supporting mixed-mode access. It is sometimes
175 * necessary to lock all case permutations of file name at
176 * once so that simultaneous case-insensitive/case-sensitive
177 * behaves as rationally as possible.
181 * Decide if exact matches should be requested when performing
182 * a zap lookup on file systems supporting case-insensitive
186 ((zsb->z_case == ZFS_CASE_INSENSITIVE) && (flag & ZCIEXACT)) ||
187 ((zsb->z_case == ZFS_CASE_MIXED) && !(flag & ZCILOOK));
190 * Only look in or update the DNLC if we are looking for the
191 * name on a file system that does not require normalization
192 * or case folding. We can also look there if we happen to be
193 * on a non-normalizing, mixed sensitivity file system IF we
194 * are looking for the exact name.
196 * Maybe can add TO-UPPERed version of name to dnlc in ci-only
197 * case for performance improvement?
199 update = !zsb->z_norm ||
200 ((zsb->z_case == ZFS_CASE_MIXED) &&
201 !(zsb->z_norm & ~U8_TEXTPREP_TOUPPER) && !(flag & ZCILOOK));
204 * ZRENAMING indicates we are in a situation where we should
205 * take narrow locks regardless of the file system's
206 * preferences for normalizing and case folding. This will
207 * prevent us deadlocking trying to grab the same wide lock
208 * twice if the two names happen to be case-insensitive
211 if (flag & ZRENAMING)
214 cmpflags = zsb->z_norm;
217 * Wait until there are no locks on this name.
219 * Don't grab the the lock if it is already held. However, cannot
220 * have both ZSHARED and ZHAVELOCK together.
222 ASSERT(!(flag & ZSHARED) || !(flag & ZHAVELOCK));
223 if (!(flag & ZHAVELOCK))
224 rw_enter(&dzp->z_name_lock, RW_READER);
226 mutex_enter(&dzp->z_lock);
228 if (dzp->z_unlinked) {
229 mutex_exit(&dzp->z_lock);
230 if (!(flag & ZHAVELOCK))
231 rw_exit(&dzp->z_name_lock);
234 for (dl = dzp->z_dirlocks; dl != NULL; dl = dl->dl_next) {
235 if ((u8_strcmp(name, dl->dl_name, 0, cmpflags,
236 U8_UNICODE_LATEST, &error) == 0) || error != 0)
240 mutex_exit(&dzp->z_lock);
241 if (!(flag & ZHAVELOCK))
242 rw_exit(&dzp->z_name_lock);
247 * Allocate a new dirlock and add it to the list.
249 dl = kmem_alloc(sizeof (zfs_dirlock_t), KM_SLEEP);
250 cv_init(&dl->dl_cv, NULL, CV_DEFAULT, NULL);
256 dl->dl_next = dzp->z_dirlocks;
257 dzp->z_dirlocks = dl;
260 if ((flag & ZSHARED) && dl->dl_sharecnt != 0)
262 cv_wait(&dl->dl_cv, &dzp->z_lock);
266 * If the z_name_lock was NOT held for this dirlock record it.
268 if (flag & ZHAVELOCK)
271 if ((flag & ZSHARED) && ++dl->dl_sharecnt > 1 && dl->dl_namesize == 0) {
273 * We're the second shared reference to dl. Make a copy of
274 * dl_name in case the first thread goes away before we do.
275 * Note that we initialize the new name before storing its
276 * pointer into dl_name, because the first thread may load
277 * dl->dl_name at any time. He'll either see the old value,
278 * which is his, or the new shared copy; either is OK.
280 dl->dl_namesize = strlen(dl->dl_name) + 1;
281 name = kmem_alloc(dl->dl_namesize, KM_SLEEP);
282 bcopy(dl->dl_name, name, dl->dl_namesize);
286 mutex_exit(&dzp->z_lock);
289 * We have a dirlock on the name. (Note that it is the dirlock,
290 * not the dzp's z_lock, that protects the name in the zap object.)
291 * See if there's an object by this name; if so, put a hold on it.
294 error = sa_lookup(dzp->z_sa_hdl, SA_ZPL_XATTR(zsb), &zoid,
297 error = (zoid == 0 ? ENOENT : 0);
301 vp = dnlc_lookup(ZTOI(dzp), name);
302 if (vp == DNLC_NO_VNODE) {
307 zfs_dirent_unlock(dl);
315 error = zfs_match_find(zsb, dzp, name, exact,
316 update, direntflags, realpnp, &zoid);
319 error = zfs_match_find(zsb, dzp, name, exact,
320 update, direntflags, realpnp, &zoid);
321 #endif /* HAVE_DNLC */
324 if (error != ENOENT || (flag & ZEXISTS)) {
325 zfs_dirent_unlock(dl);
330 zfs_dirent_unlock(dl);
333 error = zfs_zget(zsb, zoid, zpp);
335 zfs_dirent_unlock(dl);
339 if (!(flag & ZXATTR) && update)
340 dnlc_update(ZTOI(dzp), name, ZTOI(*zpp));
341 #endif /* HAVE_DNLC */
350 * Unlock this directory entry and wake anyone who was waiting for it.
353 zfs_dirent_unlock(zfs_dirlock_t *dl)
355 znode_t *dzp = dl->dl_dzp;
356 zfs_dirlock_t **prev_dl, *cur_dl;
358 mutex_enter(&dzp->z_lock);
360 if (!dl->dl_namelock)
361 rw_exit(&dzp->z_name_lock);
363 if (dl->dl_sharecnt > 1) {
365 mutex_exit(&dzp->z_lock);
368 prev_dl = &dzp->z_dirlocks;
369 while ((cur_dl = *prev_dl) != dl)
370 prev_dl = &cur_dl->dl_next;
371 *prev_dl = dl->dl_next;
372 cv_broadcast(&dl->dl_cv);
373 mutex_exit(&dzp->z_lock);
375 if (dl->dl_namesize != 0)
376 kmem_free(dl->dl_name, dl->dl_namesize);
377 cv_destroy(&dl->dl_cv);
378 kmem_free(dl, sizeof (*dl));
382 * Look up an entry in a directory.
384 * NOTE: '.' and '..' are handled as special cases because
385 * no directory entries are actually stored for them. If this is
386 * the root of a filesystem, then '.zfs' is also treated as a
387 * special pseudo-directory.
390 zfs_dirlook(znode_t *dzp, char *name, struct inode **ipp, int flags,
391 int *deflg, pathname_t *rpnp)
398 if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) {
401 } else if (name[0] == '.' && name[1] == '.' && name[2] == 0) {
402 zfs_sb_t *zsb = ZTOZSB(dzp);
405 * If we are a snapshot mounted under .zfs, return
406 * the vp for the snapshot directory.
408 if ((error = sa_lookup(dzp->z_sa_hdl,
409 SA_ZPL_PARENT(zsb), &parent, sizeof (parent))) != 0)
412 if (parent == dzp->z_id && zsb->z_parent != zsb) {
413 error = zfsctl_root_lookup(zsb->z_parent->z_ctldir,
414 "snapshot", ipp, NULL, 0, NULL, kcred,
418 #endif /* HAVE_SNAPSHOT */
419 rw_enter(&dzp->z_parent_lock, RW_READER);
420 error = zfs_zget(zsb, parent, &zp);
423 rw_exit(&dzp->z_parent_lock);
425 } else if (zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0) {
426 *ipp = zfsctl_root(dzp);
427 #endif /* HAVE_SNAPSHOT */
431 zf = ZEXISTS | ZSHARED;
432 if (flags & FIGNORECASE)
435 error = zfs_dirent_lock(&dl, dzp, name, &zp, zf, deflg, rpnp);
438 zfs_dirent_unlock(dl);
439 dzp->z_zn_prefetch = B_TRUE; /* enable prefetching */
444 if ((flags & FIGNORECASE) && rpnp && !error)
445 (void) strlcpy(rpnp->pn_buf, name, rpnp->pn_bufsize);
451 * unlinked Set (formerly known as the "delete queue") Error Handling
453 * When dealing with the unlinked set, we dmu_tx_hold_zap(), but we
454 * don't specify the name of the entry that we will be manipulating. We
455 * also fib and say that we won't be adding any new entries to the
456 * unlinked set, even though we might (this is to lower the minimum file
457 * size that can be deleted in a full filesystem). So on the small
458 * chance that the nlink list is using a fat zap (ie. has more than
459 * 2000 entries), we *may* not pre-read a block that's needed.
460 * Therefore it is remotely possible for some of the assertions
461 * regarding the unlinked set below to fail due to i/o error. On a
462 * nondebug system, this will result in the space being leaked.
465 zfs_unlinked_add(znode_t *zp, dmu_tx_t *tx)
467 zfs_sb_t *zsb = ZTOZSB(zp);
469 ASSERT(zp->z_unlinked);
470 ASSERT(zp->z_links == 0);
473 zap_add_int(zsb->z_os, zsb->z_unlinkedobj, zp->z_id, tx));
477 * Clean up any znodes that had no links when we either crashed or
478 * (force) umounted the file system.
481 zfs_unlinked_drain(zfs_sb_t *zsb)
485 dmu_object_info_t doi;
490 * Interate over the contents of the unlinked set.
492 for (zap_cursor_init(&zc, zsb->z_os, zsb->z_unlinkedobj);
493 zap_cursor_retrieve(&zc, &zap) == 0;
494 zap_cursor_advance(&zc)) {
497 * See what kind of object we have in list
500 error = dmu_object_info(zsb->z_os, zap.za_first_integer, &doi);
504 ASSERT((doi.doi_type == DMU_OT_PLAIN_FILE_CONTENTS) ||
505 (doi.doi_type == DMU_OT_DIRECTORY_CONTENTS));
507 * We need to re-mark these list entries for deletion,
508 * so we pull them back into core and set zp->z_unlinked.
510 error = zfs_zget(zsb, zap.za_first_integer, &zp);
513 * We may pick up znodes that are already marked for deletion.
514 * This could happen during the purge of an extended attribute
515 * directory. All we need to do is skip over them, since they
516 * are already in the system marked z_unlinked.
521 zp->z_unlinked = B_TRUE;
524 zap_cursor_fini(&zc);
528 * Delete the entire contents of a directory. Return a count
529 * of the number of entries that could not be deleted. If we encounter
530 * an error, return a count of at least one so that the directory stays
531 * in the unlinked set.
533 * NOTE: this function assumes that the directory is inactive,
534 * so there is no need to lock its entries before deletion.
535 * Also, it assumes the directory contents is *only* regular
539 zfs_purgedir(znode_t *dzp)
545 zfs_sb_t *zsb = ZTOZSB(dzp);
550 for (zap_cursor_init(&zc, zsb->z_os, dzp->z_id);
551 (error = zap_cursor_retrieve(&zc, &zap)) == 0;
552 zap_cursor_advance(&zc)) {
553 error = zfs_zget(zsb,
554 ZFS_DIRENT_OBJ(zap.za_first_integer), &xzp);
560 ASSERT(S_ISREG(ZTOI(xzp)->i_mode)||S_ISLNK(ZTOI(xzp)->i_mode));
562 tx = dmu_tx_create(zsb->z_os);
563 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
564 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, zap.za_name);
565 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
566 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
567 /* Is this really needed ? */
568 zfs_sa_upgrade_txholds(tx, xzp);
569 error = dmu_tx_assign(tx, TXG_WAIT);
576 bzero(&dl, sizeof (dl));
578 dl.dl_name = zap.za_name;
580 error = zfs_link_destroy(&dl, xzp, tx, 0, NULL);
587 zap_cursor_fini(&zc);
594 zfs_rmnode(znode_t *zp)
596 zfs_sb_t *zsb = ZTOZSB(zp);
597 objset_t *os = zsb->z_os;
604 ASSERT(zp->z_links == 0);
605 ASSERT(atomic_read(&ZTOI(zp)->i_count) == 0);
608 * If this is an attribute directory, purge its contents.
610 if (S_ISDIR(ZTOI(zp)->i_mode) && (zp->z_pflags & ZFS_XATTR)) {
611 if (zfs_purgedir(zp) != 0) {
613 * Not enough space to delete some xattrs.
614 * Leave it in the unlinked set.
616 zfs_znode_dmu_fini(zp);
617 zfs_inode_destroy(ZTOI(zp));
624 * Free up all the data in the file.
626 error = dmu_free_long_range(os, zp->z_id, 0, DMU_OBJECT_END);
629 * Not enough space. Leave the file in the unlinked set.
631 zfs_znode_dmu_fini(zp);
632 zfs_inode_destroy(ZTOI(zp));
637 * If the file has extended attributes, we're going to unlink
640 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zsb),
641 &xattr_obj, sizeof (xattr_obj));
642 if (error == 0 && xattr_obj) {
643 error = zfs_zget(zsb, xattr_obj, &xzp);
647 acl_obj = zfs_external_acl(zp);
650 * Set up the final transaction.
652 tx = dmu_tx_create(os);
653 dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END);
654 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
656 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, TRUE, NULL);
657 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
660 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
662 zfs_sa_upgrade_txholds(tx, zp);
663 error = dmu_tx_assign(tx, TXG_WAIT);
666 * Not enough space to delete the file. Leave it in the
667 * unlinked set, leaking it until the fs is remounted (at
668 * which point we'll call zfs_unlinked_drain() to process it).
671 zfs_znode_dmu_fini(zp);
672 zfs_inode_destroy(ZTOI(zp));
678 mutex_enter(&xzp->z_lock);
679 xzp->z_unlinked = B_TRUE; /* mark xzp for deletion */
680 xzp->z_links = 0; /* no more links to it */
681 VERIFY(0 == sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zsb),
682 &xzp->z_links, sizeof (xzp->z_links), tx));
683 mutex_exit(&xzp->z_lock);
684 zfs_unlinked_add(xzp, tx);
687 /* Remove this znode from the unlinked set */
689 zap_remove_int(zsb->z_os, zsb->z_unlinkedobj, zp->z_id, tx));
691 zfs_znode_delete(zp, tx);
700 zfs_dirent(znode_t *zp, uint64_t mode)
702 uint64_t de = zp->z_id;
704 if (ZTOZSB(zp)->z_version >= ZPL_VERSION_DIRENT_TYPE)
705 de |= IFTODT(mode) << 60;
710 * Link zp into dl. Can only fail if zp has been unlinked.
713 zfs_link_create(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag)
715 znode_t *dzp = dl->dl_dzp;
716 zfs_sb_t *zsb = ZTOZSB(zp);
718 int zp_is_dir = S_ISDIR(ZTOI(zp)->i_mode);
719 sa_bulk_attr_t bulk[5];
720 uint64_t mtime[2], ctime[2];
724 mutex_enter(&zp->z_lock);
726 if (!(flag & ZRENAMING)) {
727 if (zp->z_unlinked) { /* no new links to unlinked zp */
728 ASSERT(!(flag & (ZNEW | ZEXISTS)));
729 mutex_exit(&zp->z_lock);
733 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zsb), NULL,
734 &zp->z_links, sizeof (zp->z_links));
737 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zsb), NULL,
738 &dzp->z_id, sizeof (dzp->z_id));
739 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb), NULL,
740 &zp->z_pflags, sizeof (zp->z_pflags));
742 if (!(flag & ZNEW)) {
743 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL,
744 ctime, sizeof (ctime));
745 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime,
748 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
751 mutex_exit(&zp->z_lock);
753 mutex_enter(&dzp->z_lock);
755 dzp->z_links += zp_is_dir;
757 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zsb), NULL,
758 &dzp->z_size, sizeof (dzp->z_size));
759 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zsb), NULL,
760 &dzp->z_links, sizeof (dzp->z_links));
761 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL,
762 mtime, sizeof (mtime));
763 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL,
764 ctime, sizeof (ctime));
765 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb), NULL,
766 &dzp->z_pflags, sizeof (dzp->z_pflags));
767 zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
768 error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx);
770 mutex_exit(&dzp->z_lock);
772 value = zfs_dirent(zp, zp->z_mode);
773 error = zap_add(ZTOZSB(zp)->z_os, dzp->z_id, dl->dl_name,
778 dnlc_update(ZTOI(dzp), dl->dl_name, vp);
779 #endif /* HAVE_DNLC */
785 zfs_dropname(zfs_dirlock_t *dl, znode_t *zp, znode_t *dzp, dmu_tx_t *tx,
790 if (ZTOZSB(zp)->z_norm) {
791 if (((ZTOZSB(zp)->z_case == ZFS_CASE_INSENSITIVE) &&
792 (flag & ZCIEXACT)) ||
793 ((ZTOZSB(zp)->z_case == ZFS_CASE_MIXED) &&
795 error = zap_remove_norm(ZTOZSB(zp)->z_os,
796 dzp->z_id, dl->dl_name, MT_EXACT, tx);
798 error = zap_remove_norm(ZTOZSB(zp)->z_os,
799 dzp->z_id, dl->dl_name, MT_FIRST, tx);
801 error = zap_remove(ZTOZSB(zp)->z_os,
802 dzp->z_id, dl->dl_name, tx);
809 * Unlink zp from dl, and mark zp for deletion if this was the last link.
810 * Can fail if zp is a mount point (EBUSY) or a non-empty directory (EEXIST).
811 * If 'unlinkedp' is NULL, we put unlinked znodes on the unlinked list.
812 * If it's non-NULL, we use it to indicate whether the znode needs deletion,
813 * and it's the caller's job to do it.
816 zfs_link_destroy(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag,
817 boolean_t *unlinkedp)
819 znode_t *dzp = dl->dl_dzp;
820 zfs_sb_t *zsb = ZTOZSB(dzp);
821 int zp_is_dir = S_ISDIR(ZTOI(zp)->i_mode);
822 boolean_t unlinked = B_FALSE;
823 sa_bulk_attr_t bulk[5];
824 uint64_t mtime[2], ctime[2];
829 dnlc_remove(ZTOI(dzp), dl->dl_name);
830 #endif /* HAVE_DNLC */
832 if (!(flag & ZRENAMING)) {
833 mutex_enter(&zp->z_lock);
835 if (zp_is_dir && !zfs_dirempty(zp)) {
836 mutex_exit(&zp->z_lock);
841 * If we get here, we are going to try to remove the object.
842 * First try removing the name from the directory; if that
843 * fails, return the error.
845 error = zfs_dropname(dl, zp, dzp, tx, flag);
847 mutex_exit(&zp->z_lock);
851 if (zp->z_links <= zp_is_dir) {
852 zfs_panic_recover("zfs: link count on %lu is %u, "
853 "should be at least %u", zp->z_id,
854 (int)zp->z_links, zp_is_dir + 1);
855 zp->z_links = zp_is_dir + 1;
857 if (--zp->z_links == zp_is_dir) {
858 zp->z_unlinked = B_TRUE;
862 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb),
863 NULL, &ctime, sizeof (ctime));
864 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb),
865 NULL, &zp->z_pflags, sizeof (zp->z_pflags));
866 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
869 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zsb),
870 NULL, &zp->z_links, sizeof (zp->z_links));
871 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
874 mutex_exit(&zp->z_lock);
876 error = zfs_dropname(dl, zp, dzp, tx, flag);
881 mutex_enter(&dzp->z_lock);
882 dzp->z_size--; /* one dirent removed */
883 dzp->z_links -= zp_is_dir; /* ".." link from zp */
884 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zsb),
885 NULL, &dzp->z_links, sizeof (dzp->z_links));
886 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zsb),
887 NULL, &dzp->z_size, sizeof (dzp->z_size));
888 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb),
889 NULL, ctime, sizeof (ctime));
890 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb),
891 NULL, mtime, sizeof (mtime));
892 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb),
893 NULL, &dzp->z_pflags, sizeof (dzp->z_pflags));
894 zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
895 error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx);
897 mutex_exit(&dzp->z_lock);
899 if (unlinkedp != NULL)
900 *unlinkedp = unlinked;
902 zfs_unlinked_add(zp, tx);
908 * Indicate whether the directory is empty. Works with or without z_lock
909 * held, but can only be consider a hint in the latter case. Returns true
910 * if only "." and ".." remain and there's no work in progress.
913 zfs_dirempty(znode_t *dzp)
915 return (dzp->z_size == 2 && dzp->z_dirlocks == 0);
919 zfs_make_xattrdir(znode_t *zp, vattr_t *vap, struct inode **xipp, cred_t *cr)
921 zfs_sb_t *zsb = ZTOZSB(zp);
925 zfs_acl_ids_t acl_ids;
926 boolean_t fuid_dirtied;
933 if ((error = zfs_zaccess(zp, ACE_WRITE_NAMED_ATTRS, 0, B_FALSE, cr)))
936 if ((error = zfs_acl_ids_create(zp, IS_XATTR, vap, cr, NULL,
939 if (zfs_acl_ids_overquota(zsb, &acl_ids)) {
940 zfs_acl_ids_free(&acl_ids);
945 tx = dmu_tx_create(zsb->z_os);
946 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
947 ZFS_SA_BASE_ATTR_SIZE);
948 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
949 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
950 fuid_dirtied = zsb->z_fuid_dirty;
952 zfs_fuid_txhold(zsb, tx);
953 error = dmu_tx_assign(tx, TXG_NOWAIT);
955 if (error == ERESTART) {
960 zfs_acl_ids_free(&acl_ids);
964 zfs_mknode(zp, vap, tx, cr, IS_XATTR, &xzp, &acl_ids);
967 zfs_fuid_sync(zsb, tx);
970 error = sa_lookup(xzp->z_sa_hdl, SA_ZPL_PARENT(zsb),
971 &parent, sizeof (parent));
972 ASSERT(error == 0 && parent == zp->z_id);
975 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_XATTR(zsb), &xzp->z_id,
976 sizeof (xzp->z_id), tx));
978 (void) zfs_log_create(zsb->z_log, tx, TX_MKXATTR, zp,
979 xzp, "", NULL, acl_ids.z_fuidp, vap);
981 zfs_acl_ids_free(&acl_ids);
990 * Return a znode for the extended attribute directory for zp.
991 * ** If the directory does not already exist, it is created **
993 * IN: zp - znode to obtain attribute directory from
994 * cr - credentials of caller
995 * flags - flags from the VOP_LOOKUP call
997 * OUT: xipp - pointer to extended attribute znode
999 * RETURN: 0 on success
1000 * error number on failure
1003 zfs_get_xattrdir(znode_t *zp, struct inode **xipp, cred_t *cr, int flags)
1005 zfs_sb_t *zsb = ZTOZSB(zp);
1011 error = zfs_dirent_lock(&dl, zp, "", &xzp, ZXATTR, NULL, NULL);
1017 zfs_dirent_unlock(dl);
1021 if (!(flags & CREATE_XATTR_DIR)) {
1022 zfs_dirent_unlock(dl);
1026 if (zsb->z_vfs->mnt_flags & MNT_READONLY) {
1027 zfs_dirent_unlock(dl);
1032 * The ability to 'create' files in an attribute
1033 * directory comes from the write_xattr permission on the base file.
1035 * The ability to 'search' an attribute directory requires
1036 * read_xattr permission on the base file.
1038 * Once in a directory the ability to read/write attributes
1039 * is controlled by the permissions on the attribute file.
1041 va.va_mask = ATTR_MODE | ATTR_UID | ATTR_GID;
1042 va.va_mode = S_IFDIR | S_ISVTX | 0777;
1043 zfs_fuid_map_ids(zp, cr, &va.va_uid, &va.va_gid);
1045 error = zfs_make_xattrdir(zp, &va, xipp, cr);
1046 zfs_dirent_unlock(dl);
1048 if (error == ERESTART) {
1049 /* NB: we already did dmu_tx_wait() if necessary */
1057 * Decide whether it is okay to remove within a sticky directory.
1059 * In sticky directories, write access is not sufficient;
1060 * you can remove entries from a directory only if:
1062 * you own the directory,
1063 * you own the entry,
1064 * the entry is a plain file and you have write access,
1065 * or you are privileged (checked in secpolicy...).
1067 * The function returns 0 if remove access is granted.
1070 zfs_sticky_remove_access(znode_t *zdp, znode_t *zp, cred_t *cr)
1075 zfs_sb_t *zsb = ZTOZSB(zdp);
1080 if ((zdp->z_mode & S_ISVTX) == 0)
1083 downer = zfs_fuid_map_id(zsb, zdp->z_uid, cr, ZFS_OWNER);
1084 fowner = zfs_fuid_map_id(zsb, zp->z_uid, cr, ZFS_OWNER);
1086 if ((uid = crgetuid(cr)) == downer || uid == fowner ||
1087 (S_ISDIR(ZTOI(zp)->i_mode) &&
1088 zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr) == 0))
1091 return (secpolicy_vnode_remove(cr));