4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2013 by Delphix. All rights reserved.
27 #include <sys/types.h>
28 #include <sys/param.h>
30 #include <sys/systm.h>
31 #include <sys/sysmacros.h>
32 #include <sys/resource.h>
34 #include <sys/vnode.h>
39 #include <sys/pathname.h>
40 #include <sys/cmn_err.h>
41 #include <sys/errno.h>
43 #include <sys/unistd.h>
44 #include <sys/sunddi.h>
45 #include <sys/random.h>
46 #include <sys/policy.h>
47 #include <sys/zfs_dir.h>
48 #include <sys/zfs_acl.h>
49 #include <sys/fs/zfs.h>
50 #include "fs/fs_subr.h"
53 #include <sys/atomic.h>
54 #include <sys/zfs_ctldir.h>
55 #include <sys/zfs_fuid.h>
57 #include <sys/zfs_sa.h>
59 #include <sys/extdirent.h>
62 * zfs_match_find() is used by zfs_dirent_lock() to peform zap lookups
63 * of names after deciding which is the appropriate lookup interface.
66 zfs_match_find(zfs_sb_t *zsb, znode_t *dzp, char *name, boolean_t exact,
67 boolean_t update, int *deflags, pathname_t *rpnp, uint64_t *zoid)
69 boolean_t conflict = B_FALSE;
73 matchtype_t mt = MT_FIRST;
79 bufsz = rpnp->pn_bufsize;
84 * In the non-mixed case we only expect there would ever
85 * be one match, but we need to use the normalizing lookup.
87 error = zap_lookup_norm(zsb->z_os, dzp->z_id, name, 8, 1,
88 zoid, mt, buf, bufsz, &conflict);
90 error = zap_lookup(zsb->z_os, dzp->z_id, name, 8, 1, zoid);
94 * Allow multiple entries provided the first entry is
95 * the object id. Non-zpl consumers may safely make
96 * use of the additional space.
98 * XXX: This should be a feature flag for compatibility
100 if (error == EOVERFLOW)
103 if (zsb->z_norm && !error && deflags)
104 *deflags = conflict ? ED_CASE_CONFLICT : 0;
106 *zoid = ZFS_DIRENT_OBJ(*zoid);
109 if (error == ENOENT && update)
110 dnlc_update(ZTOI(dzp), name, DNLC_NO_VNODE);
111 #endif /* HAVE_DNLC */
117 * Lock a directory entry. A dirlock on <dzp, name> protects that name
118 * in dzp's directory zap object. As long as you hold a dirlock, you can
119 * assume two things: (1) dzp cannot be reaped, and (2) no other thread
120 * can change the zap entry for (i.e. link or unlink) this name.
123 * dzp - znode for directory
124 * name - name of entry to lock
125 * flag - ZNEW: if the entry already exists, fail with EEXIST.
126 * ZEXISTS: if the entry does not exist, fail with ENOENT.
127 * ZSHARED: allow concurrent access with other ZSHARED callers.
128 * ZXATTR: we want dzp's xattr directory
129 * ZCILOOK: On a mixed sensitivity file system,
130 * this lookup should be case-insensitive.
131 * ZCIEXACT: On a purely case-insensitive file system,
132 * this lookup should be case-sensitive.
133 * ZRENAMING: we are locking for renaming, force narrow locks
134 * ZHAVELOCK: Don't grab the z_name_lock for this call. The
135 * current thread already holds it.
138 * zpp - pointer to the znode for the entry (NULL if there isn't one)
139 * dlpp - pointer to the dirlock for this entry (NULL on error)
140 * direntflags - (case-insensitive lookup only)
141 * flags if multiple case-sensitive matches exist in directory
142 * realpnp - (case-insensitive lookup only)
143 * actual name matched within the directory
145 * Return value: 0 on success or errno on failure.
147 * NOTE: Always checks for, and rejects, '.' and '..'.
148 * NOTE: For case-insensitive file systems we take wide locks (see below),
149 * but return znode pointers to a single match.
152 zfs_dirent_lock(zfs_dirlock_t **dlpp, znode_t *dzp, char *name, znode_t **zpp,
153 int flag, int *direntflags, pathname_t *realpnp)
155 zfs_sb_t *zsb = ZTOZSB(dzp);
162 #endif /* HAVE_DNLC */
170 * Verify that we are not trying to lock '.', '..', or '.zfs'
172 if ((name[0] == '.' &&
173 (name[1] == '\0' || (name[1] == '.' && name[2] == '\0'))) ||
174 (zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0))
175 return (SET_ERROR(EEXIST));
178 * Case sensitivity and normalization preferences are set when
179 * the file system is created. These are stored in the
180 * zsb->z_case and zsb->z_norm fields. These choices
181 * affect what vnodes can be cached in the DNLC, how we
182 * perform zap lookups, and the "width" of our dirlocks.
184 * A normal dirlock locks a single name. Note that with
185 * normalization a name can be composed multiple ways, but
186 * when normalized, these names all compare equal. A wide
187 * dirlock locks multiple names. We need these when the file
188 * system is supporting mixed-mode access. It is sometimes
189 * necessary to lock all case permutations of file name at
190 * once so that simultaneous case-insensitive/case-sensitive
191 * behaves as rationally as possible.
195 * Decide if exact matches should be requested when performing
196 * a zap lookup on file systems supporting case-insensitive
200 ((zsb->z_case == ZFS_CASE_INSENSITIVE) && (flag & ZCIEXACT)) ||
201 ((zsb->z_case == ZFS_CASE_MIXED) && !(flag & ZCILOOK));
204 * Only look in or update the DNLC if we are looking for the
205 * name on a file system that does not require normalization
206 * or case folding. We can also look there if we happen to be
207 * on a non-normalizing, mixed sensitivity file system IF we
208 * are looking for the exact name.
210 * Maybe can add TO-UPPERed version of name to dnlc in ci-only
211 * case for performance improvement?
213 update = !zsb->z_norm ||
214 ((zsb->z_case == ZFS_CASE_MIXED) &&
215 !(zsb->z_norm & ~U8_TEXTPREP_TOUPPER) && !(flag & ZCILOOK));
218 * ZRENAMING indicates we are in a situation where we should
219 * take narrow locks regardless of the file system's
220 * preferences for normalizing and case folding. This will
221 * prevent us deadlocking trying to grab the same wide lock
222 * twice if the two names happen to be case-insensitive
225 if (flag & ZRENAMING)
228 cmpflags = zsb->z_norm;
231 * Wait until there are no locks on this name.
233 * Don't grab the the lock if it is already held. However, cannot
234 * have both ZSHARED and ZHAVELOCK together.
236 ASSERT(!(flag & ZSHARED) || !(flag & ZHAVELOCK));
237 if (!(flag & ZHAVELOCK))
238 rw_enter(&dzp->z_name_lock, RW_READER);
240 mutex_enter(&dzp->z_lock);
242 if (dzp->z_unlinked) {
243 mutex_exit(&dzp->z_lock);
244 if (!(flag & ZHAVELOCK))
245 rw_exit(&dzp->z_name_lock);
246 return (SET_ERROR(ENOENT));
248 for (dl = dzp->z_dirlocks; dl != NULL; dl = dl->dl_next) {
249 if ((u8_strcmp(name, dl->dl_name, 0, cmpflags,
250 U8_UNICODE_LATEST, &error) == 0) || error != 0)
254 mutex_exit(&dzp->z_lock);
255 if (!(flag & ZHAVELOCK))
256 rw_exit(&dzp->z_name_lock);
257 return (SET_ERROR(ENOENT));
261 * Allocate a new dirlock and add it to the list.
263 dl = kmem_alloc(sizeof (zfs_dirlock_t), KM_SLEEP);
264 cv_init(&dl->dl_cv, NULL, CV_DEFAULT, NULL);
270 dl->dl_next = dzp->z_dirlocks;
271 dzp->z_dirlocks = dl;
274 if ((flag & ZSHARED) && dl->dl_sharecnt != 0)
276 cv_wait(&dl->dl_cv, &dzp->z_lock);
280 * If the z_name_lock was NOT held for this dirlock record it.
282 if (flag & ZHAVELOCK)
285 if ((flag & ZSHARED) && ++dl->dl_sharecnt > 1 && dl->dl_namesize == 0) {
287 * We're the second shared reference to dl. Make a copy of
288 * dl_name in case the first thread goes away before we do.
289 * Note that we initialize the new name before storing its
290 * pointer into dl_name, because the first thread may load
291 * dl->dl_name at any time. He'll either see the old value,
292 * which is his, or the new shared copy; either is OK.
294 dl->dl_namesize = strlen(dl->dl_name) + 1;
295 name = kmem_alloc(dl->dl_namesize, KM_SLEEP);
296 bcopy(dl->dl_name, name, dl->dl_namesize);
300 mutex_exit(&dzp->z_lock);
303 * We have a dirlock on the name. (Note that it is the dirlock,
304 * not the dzp's z_lock, that protects the name in the zap object.)
305 * See if there's an object by this name; if so, put a hold on it.
308 error = sa_lookup(dzp->z_sa_hdl, SA_ZPL_XATTR(zsb), &zoid,
311 error = (zoid == 0 ? SET_ERROR(ENOENT) : 0);
315 vp = dnlc_lookup(ZTOI(dzp), name);
316 if (vp == DNLC_NO_VNODE) {
318 error = SET_ERROR(ENOENT);
321 zfs_dirent_unlock(dl);
323 return (SET_ERROR(EEXIST));
329 error = zfs_match_find(zsb, dzp, name, exact,
330 update, direntflags, realpnp, &zoid);
333 error = zfs_match_find(zsb, dzp, name, exact,
334 update, direntflags, realpnp, &zoid);
335 #endif /* HAVE_DNLC */
338 if (error != ENOENT || (flag & ZEXISTS)) {
339 zfs_dirent_unlock(dl);
344 zfs_dirent_unlock(dl);
345 return (SET_ERROR(EEXIST));
347 error = zfs_zget(zsb, zoid, zpp);
349 zfs_dirent_unlock(dl);
353 if (!(flag & ZXATTR) && update)
354 dnlc_update(ZTOI(dzp), name, ZTOI(*zpp));
355 #endif /* HAVE_DNLC */
364 * Unlock this directory entry and wake anyone who was waiting for it.
367 zfs_dirent_unlock(zfs_dirlock_t *dl)
369 znode_t *dzp = dl->dl_dzp;
370 zfs_dirlock_t **prev_dl, *cur_dl;
372 mutex_enter(&dzp->z_lock);
374 if (!dl->dl_namelock)
375 rw_exit(&dzp->z_name_lock);
377 if (dl->dl_sharecnt > 1) {
379 mutex_exit(&dzp->z_lock);
382 prev_dl = &dzp->z_dirlocks;
383 while ((cur_dl = *prev_dl) != dl)
384 prev_dl = &cur_dl->dl_next;
385 *prev_dl = dl->dl_next;
386 cv_broadcast(&dl->dl_cv);
387 mutex_exit(&dzp->z_lock);
389 if (dl->dl_namesize != 0)
390 kmem_free(dl->dl_name, dl->dl_namesize);
391 cv_destroy(&dl->dl_cv);
392 kmem_free(dl, sizeof (*dl));
396 * Look up an entry in a directory.
398 * NOTE: '.' and '..' are handled as special cases because
399 * no directory entries are actually stored for them. If this is
400 * the root of a filesystem, then '.zfs' is also treated as a
401 * special pseudo-directory.
404 zfs_dirlook(znode_t *dzp, char *name, struct inode **ipp, int flags,
405 int *deflg, pathname_t *rpnp)
412 if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) {
415 } else if (name[0] == '.' && name[1] == '.' && name[2] == 0) {
416 zfs_sb_t *zsb = ZTOZSB(dzp);
419 * If we are a snapshot mounted under .zfs, return
420 * the inode pointer for the snapshot directory.
422 if ((error = sa_lookup(dzp->z_sa_hdl,
423 SA_ZPL_PARENT(zsb), &parent, sizeof (parent))) != 0)
426 if (parent == dzp->z_id && zsb->z_parent != zsb) {
427 error = zfsctl_root_lookup(zsb->z_parent->z_ctldir,
428 "snapshot", ipp, 0, kcred, NULL, NULL);
431 rw_enter(&dzp->z_parent_lock, RW_READER);
432 error = zfs_zget(zsb, parent, &zp);
435 rw_exit(&dzp->z_parent_lock);
436 } else if (zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0) {
437 *ipp = zfsctl_root(dzp);
441 zf = ZEXISTS | ZSHARED;
442 if (flags & FIGNORECASE)
445 error = zfs_dirent_lock(&dl, dzp, name, &zp, zf, deflg, rpnp);
448 zfs_dirent_unlock(dl);
449 dzp->z_zn_prefetch = B_TRUE; /* enable prefetching */
454 if ((flags & FIGNORECASE) && rpnp && !error)
455 (void) strlcpy(rpnp->pn_buf, name, rpnp->pn_bufsize);
461 * unlinked Set (formerly known as the "delete queue") Error Handling
463 * When dealing with the unlinked set, we dmu_tx_hold_zap(), but we
464 * don't specify the name of the entry that we will be manipulating. We
465 * also fib and say that we won't be adding any new entries to the
466 * unlinked set, even though we might (this is to lower the minimum file
467 * size that can be deleted in a full filesystem). So on the small
468 * chance that the nlink list is using a fat zap (ie. has more than
469 * 2000 entries), we *may* not pre-read a block that's needed.
470 * Therefore it is remotely possible for some of the assertions
471 * regarding the unlinked set below to fail due to i/o error. On a
472 * nondebug system, this will result in the space being leaked.
475 zfs_unlinked_add(znode_t *zp, dmu_tx_t *tx)
477 zfs_sb_t *zsb = ZTOZSB(zp);
479 ASSERT(zp->z_unlinked);
480 ASSERT(zp->z_links == 0);
483 zap_add_int(zsb->z_os, zsb->z_unlinkedobj, zp->z_id, tx));
487 * Delete the entire contents of a directory. Return a count
488 * of the number of entries that could not be deleted. If we encounter
489 * an error, return a count of at least one so that the directory stays
490 * in the unlinked set.
492 * NOTE: this function assumes that the directory is inactive,
493 * so there is no need to lock its entries before deletion.
494 * Also, it assumes the directory contents is *only* regular
498 zfs_purgedir(znode_t *dzp)
504 zfs_sb_t *zsb = ZTOZSB(dzp);
509 for (zap_cursor_init(&zc, zsb->z_os, dzp->z_id);
510 (error = zap_cursor_retrieve(&zc, &zap)) == 0;
511 zap_cursor_advance(&zc)) {
512 error = zfs_zget(zsb,
513 ZFS_DIRENT_OBJ(zap.za_first_integer), &xzp);
519 ASSERT(S_ISREG(ZTOI(xzp)->i_mode)||S_ISLNK(ZTOI(xzp)->i_mode));
521 tx = dmu_tx_create(zsb->z_os);
522 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
523 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, zap.za_name);
524 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
525 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
526 /* Is this really needed ? */
527 zfs_sa_upgrade_txholds(tx, xzp);
528 error = dmu_tx_assign(tx, TXG_WAIT);
535 bzero(&dl, sizeof (dl));
537 dl.dl_name = zap.za_name;
539 error = zfs_link_destroy(&dl, xzp, tx, 0, NULL);
546 zap_cursor_fini(&zc);
553 * Clean up any znodes that had no links when we either crashed or
554 * (force) umounted the file system.
557 zfs_unlinked_drain(zfs_sb_t *zsb)
561 dmu_object_info_t doi;
566 * Iterate over the contents of the unlinked set.
568 for (zap_cursor_init(&zc, zsb->z_os, zsb->z_unlinkedobj);
569 zap_cursor_retrieve(&zc, &zap) == 0;
570 zap_cursor_advance(&zc)) {
573 * See what kind of object we have in list
576 error = dmu_object_info(zsb->z_os, zap.za_first_integer, &doi);
580 ASSERT((doi.doi_type == DMU_OT_PLAIN_FILE_CONTENTS) ||
581 (doi.doi_type == DMU_OT_DIRECTORY_CONTENTS));
583 * We need to re-mark these list entries for deletion,
584 * so we pull them back into core and set zp->z_unlinked.
586 error = zfs_zget(zsb, zap.za_first_integer, &zp);
589 * We may pick up znodes that are already marked for deletion.
590 * This could happen during the purge of an extended attribute
591 * directory. All we need to do is skip over them, since they
592 * are already in the system marked z_unlinked.
597 zp->z_unlinked = B_TRUE;
600 * If this is an attribute directory, purge its contents.
602 if (S_ISDIR(ZTOI(zp)->i_mode) && (zp->z_pflags & ZFS_XATTR)) {
604 * We don't need to check the return value of
605 * zfs_purgedir here, because zfs_rmnode will just
606 * return this xattr directory to the unlinked set
607 * until all of its xattrs are gone.
609 (void) zfs_purgedir(zp);
614 zap_cursor_fini(&zc);
618 zfs_rmnode(znode_t *zp)
620 zfs_sb_t *zsb = ZTOZSB(zp);
621 objset_t *os = zsb->z_os;
629 ASSERT(zp->z_links == 0);
630 ASSERT(atomic_read(&ZTOI(zp)->i_count) == 0);
633 * If this is an attribute directory, purge its contents.
635 if (S_ISDIR(ZTOI(zp)->i_mode) && (zp->z_pflags & ZFS_XATTR)) {
636 error = zap_count(os, zp->z_id, &count);
638 zfs_znode_dmu_fini(zp);
646 * There are still directory entries in this xattr
647 * directory. Let zfs_unlinked_drain() deal with
648 * them to avoid deadlocking this process in the
649 * zfs_purgedir()->zfs_zget()->ilookup() callpath
650 * on the xattr inode's I_FREEING bit.
652 taskq = dsl_pool_iput_taskq(dmu_objset_pool(os));
653 taskq_dispatch(taskq, (task_func_t *)
654 zfs_unlinked_drain, zsb, TQ_SLEEP);
656 zfs_znode_dmu_fini(zp);
662 * Free up all the data in the file.
664 error = dmu_free_long_range(os, zp->z_id, 0, DMU_OBJECT_END);
667 * Not enough space. Leave the file in the unlinked set.
669 zfs_znode_dmu_fini(zp);
674 * If the file has extended attributes, we're going to unlink
677 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zsb),
678 &xattr_obj, sizeof (xattr_obj));
679 if (error == 0 && xattr_obj) {
680 error = zfs_zget(zsb, xattr_obj, &xzp);
684 acl_obj = zfs_external_acl(zp);
687 * Set up the final transaction.
689 tx = dmu_tx_create(os);
690 dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END);
691 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
693 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, TRUE, NULL);
694 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
697 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
699 zfs_sa_upgrade_txholds(tx, zp);
700 error = dmu_tx_assign(tx, TXG_WAIT);
703 * Not enough space to delete the file. Leave it in the
704 * unlinked set, leaking it until the fs is remounted (at
705 * which point we'll call zfs_unlinked_drain() to process it).
708 zfs_znode_dmu_fini(zp);
714 mutex_enter(&xzp->z_lock);
715 xzp->z_unlinked = B_TRUE; /* mark xzp for deletion */
716 xzp->z_links = 0; /* no more links to it */
717 VERIFY(0 == sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zsb),
718 &xzp->z_links, sizeof (xzp->z_links), tx));
719 mutex_exit(&xzp->z_lock);
720 zfs_unlinked_add(xzp, tx);
723 /* Remove this znode from the unlinked set */
725 zap_remove_int(zsb->z_os, zsb->z_unlinkedobj, zp->z_id, tx));
727 zfs_znode_delete(zp, tx);
736 zfs_dirent(znode_t *zp, uint64_t mode)
738 uint64_t de = zp->z_id;
740 if (ZTOZSB(zp)->z_version >= ZPL_VERSION_DIRENT_TYPE)
741 de |= IFTODT(mode) << 60;
746 * Link zp into dl. Can only fail if zp has been unlinked.
749 zfs_link_create(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag)
751 znode_t *dzp = dl->dl_dzp;
752 zfs_sb_t *zsb = ZTOZSB(zp);
754 int zp_is_dir = S_ISDIR(ZTOI(zp)->i_mode);
755 sa_bulk_attr_t bulk[5];
756 uint64_t mtime[2], ctime[2];
760 mutex_enter(&zp->z_lock);
762 if (!(flag & ZRENAMING)) {
763 if (zp->z_unlinked) { /* no new links to unlinked zp */
764 ASSERT(!(flag & (ZNEW | ZEXISTS)));
765 mutex_exit(&zp->z_lock);
766 return (SET_ERROR(ENOENT));
769 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zsb), NULL,
770 &zp->z_links, sizeof (zp->z_links));
773 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zsb), NULL,
774 &dzp->z_id, sizeof (dzp->z_id));
775 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb), NULL,
776 &zp->z_pflags, sizeof (zp->z_pflags));
778 if (!(flag & ZNEW)) {
779 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL,
780 ctime, sizeof (ctime));
781 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime,
784 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
787 mutex_exit(&zp->z_lock);
789 mutex_enter(&dzp->z_lock);
791 dzp->z_links += zp_is_dir;
793 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zsb), NULL,
794 &dzp->z_size, sizeof (dzp->z_size));
795 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zsb), NULL,
796 &dzp->z_links, sizeof (dzp->z_links));
797 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL,
798 mtime, sizeof (mtime));
799 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL,
800 ctime, sizeof (ctime));
801 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb), NULL,
802 &dzp->z_pflags, sizeof (dzp->z_pflags));
803 zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
804 error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx);
806 mutex_exit(&dzp->z_lock);
808 value = zfs_dirent(zp, zp->z_mode);
809 error = zap_add(ZTOZSB(zp)->z_os, dzp->z_id, dl->dl_name,
817 zfs_dropname(zfs_dirlock_t *dl, znode_t *zp, znode_t *dzp, dmu_tx_t *tx,
822 if (ZTOZSB(zp)->z_norm) {
823 if (((ZTOZSB(zp)->z_case == ZFS_CASE_INSENSITIVE) &&
824 (flag & ZCIEXACT)) ||
825 ((ZTOZSB(zp)->z_case == ZFS_CASE_MIXED) &&
827 error = zap_remove_norm(ZTOZSB(zp)->z_os,
828 dzp->z_id, dl->dl_name, MT_EXACT, tx);
830 error = zap_remove_norm(ZTOZSB(zp)->z_os,
831 dzp->z_id, dl->dl_name, MT_FIRST, tx);
833 error = zap_remove(ZTOZSB(zp)->z_os,
834 dzp->z_id, dl->dl_name, tx);
841 * Unlink zp from dl, and mark zp for deletion if this was the last link. Can
842 * fail if zp is a mount point (EBUSY) or a non-empty directory (ENOTEMPTY).
843 * If 'unlinkedp' is NULL, we put unlinked znodes on the unlinked list.
844 * If it's non-NULL, we use it to indicate whether the znode needs deletion,
845 * and it's the caller's job to do it.
848 zfs_link_destroy(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag,
849 boolean_t *unlinkedp)
851 znode_t *dzp = dl->dl_dzp;
852 zfs_sb_t *zsb = ZTOZSB(dzp);
853 int zp_is_dir = S_ISDIR(ZTOI(zp)->i_mode);
854 boolean_t unlinked = B_FALSE;
855 sa_bulk_attr_t bulk[5];
856 uint64_t mtime[2], ctime[2];
861 dnlc_remove(ZTOI(dzp), dl->dl_name);
862 #endif /* HAVE_DNLC */
864 if (!(flag & ZRENAMING)) {
865 mutex_enter(&zp->z_lock);
867 if (zp_is_dir && !zfs_dirempty(zp)) {
868 mutex_exit(&zp->z_lock);
869 return (SET_ERROR(ENOTEMPTY));
873 * If we get here, we are going to try to remove the object.
874 * First try removing the name from the directory; if that
875 * fails, return the error.
877 error = zfs_dropname(dl, zp, dzp, tx, flag);
879 mutex_exit(&zp->z_lock);
883 if (zp->z_links <= zp_is_dir) {
884 zfs_panic_recover("zfs: link count on %lu is %u, "
885 "should be at least %u", zp->z_id,
886 (int)zp->z_links, zp_is_dir + 1);
887 zp->z_links = zp_is_dir + 1;
889 if (--zp->z_links == zp_is_dir) {
890 zp->z_unlinked = B_TRUE;
894 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb),
895 NULL, &ctime, sizeof (ctime));
896 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb),
897 NULL, &zp->z_pflags, sizeof (zp->z_pflags));
898 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
901 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zsb),
902 NULL, &zp->z_links, sizeof (zp->z_links));
903 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
906 mutex_exit(&zp->z_lock);
908 error = zfs_dropname(dl, zp, dzp, tx, flag);
913 mutex_enter(&dzp->z_lock);
914 dzp->z_size--; /* one dirent removed */
915 dzp->z_links -= zp_is_dir; /* ".." link from zp */
916 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zsb),
917 NULL, &dzp->z_links, sizeof (dzp->z_links));
918 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zsb),
919 NULL, &dzp->z_size, sizeof (dzp->z_size));
920 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb),
921 NULL, ctime, sizeof (ctime));
922 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb),
923 NULL, mtime, sizeof (mtime));
924 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb),
925 NULL, &dzp->z_pflags, sizeof (dzp->z_pflags));
926 zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
927 error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx);
929 mutex_exit(&dzp->z_lock);
931 if (unlinkedp != NULL)
932 *unlinkedp = unlinked;
934 zfs_unlinked_add(zp, tx);
940 * Indicate whether the directory is empty. Works with or without z_lock
941 * held, but can only be consider a hint in the latter case. Returns true
942 * if only "." and ".." remain and there's no work in progress.
945 zfs_dirempty(znode_t *dzp)
947 return (dzp->z_size == 2 && dzp->z_dirlocks == 0);
951 zfs_make_xattrdir(znode_t *zp, vattr_t *vap, struct inode **xipp, cred_t *cr)
953 zfs_sb_t *zsb = ZTOZSB(zp);
957 zfs_acl_ids_t acl_ids;
958 boolean_t fuid_dirtied;
965 if ((error = zfs_zaccess(zp, ACE_WRITE_NAMED_ATTRS, 0, B_FALSE, cr)))
968 if ((error = zfs_acl_ids_create(zp, IS_XATTR, vap, cr, NULL,
971 if (zfs_acl_ids_overquota(zsb, &acl_ids)) {
972 zfs_acl_ids_free(&acl_ids);
973 return (SET_ERROR(EDQUOT));
976 tx = dmu_tx_create(zsb->z_os);
977 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
978 ZFS_SA_BASE_ATTR_SIZE);
979 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
980 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
981 fuid_dirtied = zsb->z_fuid_dirty;
983 zfs_fuid_txhold(zsb, tx);
984 error = dmu_tx_assign(tx, TXG_WAIT);
986 zfs_acl_ids_free(&acl_ids);
990 zfs_mknode(zp, vap, tx, cr, IS_XATTR, &xzp, &acl_ids);
993 zfs_fuid_sync(zsb, tx);
996 error = sa_lookup(xzp->z_sa_hdl, SA_ZPL_PARENT(zsb),
997 &parent, sizeof (parent));
998 ASSERT(error == 0 && parent == zp->z_id);
1001 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_XATTR(zsb), &xzp->z_id,
1002 sizeof (xzp->z_id), tx));
1004 (void) zfs_log_create(zsb->z_log, tx, TX_MKXATTR, zp,
1005 xzp, "", NULL, acl_ids.z_fuidp, vap);
1007 zfs_acl_ids_free(&acl_ids);
1016 * Return a znode for the extended attribute directory for zp.
1017 * ** If the directory does not already exist, it is created **
1019 * IN: zp - znode to obtain attribute directory from
1020 * cr - credentials of caller
1021 * flags - flags from the VOP_LOOKUP call
1023 * OUT: xipp - pointer to extended attribute znode
1025 * RETURN: 0 on success
1026 * error number on failure
1029 zfs_get_xattrdir(znode_t *zp, struct inode **xipp, cred_t *cr, int flags)
1031 zfs_sb_t *zsb = ZTOZSB(zp);
1037 error = zfs_dirent_lock(&dl, zp, "", &xzp, ZXATTR, NULL, NULL);
1043 zfs_dirent_unlock(dl);
1047 if (!(flags & CREATE_XATTR_DIR)) {
1048 zfs_dirent_unlock(dl);
1049 return (SET_ERROR(ENOENT));
1052 if (zfs_is_readonly(zsb)) {
1053 zfs_dirent_unlock(dl);
1054 return (SET_ERROR(EROFS));
1058 * The ability to 'create' files in an attribute
1059 * directory comes from the write_xattr permission on the base file.
1061 * The ability to 'search' an attribute directory requires
1062 * read_xattr permission on the base file.
1064 * Once in a directory the ability to read/write attributes
1065 * is controlled by the permissions on the attribute file.
1067 va.va_mask = ATTR_MODE | ATTR_UID | ATTR_GID;
1068 va.va_mode = S_IFDIR | S_ISVTX | 0777;
1069 zfs_fuid_map_ids(zp, cr, &va.va_uid, &va.va_gid);
1071 va.va_dentry = NULL;
1072 error = zfs_make_xattrdir(zp, &va, xipp, cr);
1073 zfs_dirent_unlock(dl);
1075 if (error == ERESTART) {
1076 /* NB: we already did dmu_tx_wait() if necessary */
1084 * Decide whether it is okay to remove within a sticky directory.
1086 * In sticky directories, write access is not sufficient;
1087 * you can remove entries from a directory only if:
1089 * you own the directory,
1090 * you own the entry,
1091 * the entry is a plain file and you have write access,
1092 * or you are privileged (checked in secpolicy...).
1094 * The function returns 0 if remove access is granted.
1097 zfs_sticky_remove_access(znode_t *zdp, znode_t *zp, cred_t *cr)
1102 zfs_sb_t *zsb = ZTOZSB(zdp);
1107 if ((zdp->z_mode & S_ISVTX) == 0)
1110 downer = zfs_fuid_map_id(zsb, zdp->z_uid, cr, ZFS_OWNER);
1111 fowner = zfs_fuid_map_id(zsb, zp->z_uid, cr, ZFS_OWNER);
1113 if ((uid = crgetuid(cr)) == downer || uid == fowner ||
1114 (S_ISDIR(ZTOI(zp)->i_mode) &&
1115 zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr) == 0))
1118 return (secpolicy_vnode_remove(cr));