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) 2013, 2016 by Delphix. All rights reserved.
25 * Copyright 2017 Nexenta Systems, Inc.
28 #include <sys/types.h>
29 #include <sys/param.h>
31 #include <sys/sysmacros.h>
33 #include <sys/vnode.h>
37 #include <sys/pathname.h>
38 #include <sys/cmn_err.h>
39 #include <sys/errno.h>
41 #include <sys/sunddi.h>
42 #include <sys/random.h>
43 #include <sys/policy.h>
44 #include <sys/zfs_dir.h>
45 #include <sys/zfs_acl.h>
46 #include <sys/zfs_vnops.h>
47 #include <sys/fs/zfs.h>
50 #include <sys/atomic.h>
51 #include <sys/zfs_ctldir.h>
52 #include <sys/zfs_fuid.h>
54 #include <sys/zfs_sa.h>
55 #include <sys/dmu_objset.h>
56 #include <sys/dsl_dir.h>
59 * zfs_match_find() is used by zfs_dirent_lock() to perform zap lookups
60 * of names after deciding which is the appropriate lookup interface.
63 zfs_match_find(zfsvfs_t *zfsvfs, znode_t *dzp, char *name, matchtype_t mt,
64 boolean_t update, int *deflags, pathname_t *rpnp, uint64_t *zoid)
66 boolean_t conflict = B_FALSE;
75 bufsz = rpnp->pn_bufsize;
79 * In the non-mixed case we only expect there would ever
80 * be one match, but we need to use the normalizing lookup.
82 error = zap_lookup_norm(zfsvfs->z_os, dzp->z_id, name, 8, 1,
83 zoid, mt, buf, bufsz, &conflict);
85 error = zap_lookup(zfsvfs->z_os, dzp->z_id, name, 8, 1, zoid);
89 * Allow multiple entries provided the first entry is
90 * the object id. Non-zpl consumers may safely make
91 * use of the additional space.
93 * XXX: This should be a feature flag for compatibility
95 if (error == EOVERFLOW)
98 if (zfsvfs->z_norm && !error && deflags)
99 *deflags = conflict ? ED_CASE_CONFLICT : 0;
101 *zoid = ZFS_DIRENT_OBJ(*zoid);
107 * Lock a directory entry. A dirlock on <dzp, name> protects that name
108 * in dzp's directory zap object. As long as you hold a dirlock, you can
109 * assume two things: (1) dzp cannot be reaped, and (2) no other thread
110 * can change the zap entry for (i.e. link or unlink) this name.
113 * dzp - znode for directory
114 * name - name of entry to lock
115 * flag - ZNEW: if the entry already exists, fail with EEXIST.
116 * ZEXISTS: if the entry does not exist, fail with ENOENT.
117 * ZSHARED: allow concurrent access with other ZSHARED callers.
118 * ZXATTR: we want dzp's xattr directory
119 * ZCILOOK: On a mixed sensitivity file system,
120 * this lookup should be case-insensitive.
121 * ZCIEXACT: On a purely case-insensitive file system,
122 * this lookup should be case-sensitive.
123 * ZRENAMING: we are locking for renaming, force narrow locks
124 * ZHAVELOCK: Don't grab the z_name_lock for this call. The
125 * current thread already holds it.
128 * zpp - pointer to the znode for the entry (NULL if there isn't one)
129 * dlpp - pointer to the dirlock for this entry (NULL on error)
130 * direntflags - (case-insensitive lookup only)
131 * flags if multiple case-sensitive matches exist in directory
132 * realpnp - (case-insensitive lookup only)
133 * actual name matched within the directory
135 * Return value: 0 on success or errno on failure.
137 * NOTE: Always checks for, and rejects, '.' and '..'.
138 * NOTE: For case-insensitive file systems we take wide locks (see below),
139 * but return znode pointers to a single match.
142 zfs_dirent_lock(zfs_dirlock_t **dlpp, znode_t *dzp, char *name, znode_t **zpp,
143 int flag, int *direntflags, pathname_t *realpnp)
145 zfsvfs_t *zfsvfs = ZTOZSB(dzp);
157 * Verify that we are not trying to lock '.', '..', or '.zfs'
159 if ((name[0] == '.' &&
160 (name[1] == '\0' || (name[1] == '.' && name[2] == '\0'))) ||
161 (zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0))
162 return (SET_ERROR(EEXIST));
165 * Case sensitivity and normalization preferences are set when
166 * the file system is created. These are stored in the
167 * zfsvfs->z_case and zfsvfs->z_norm fields. These choices
168 * affect what vnodes can be cached in the DNLC, how we
169 * perform zap lookups, and the "width" of our dirlocks.
171 * A normal dirlock locks a single name. Note that with
172 * normalization a name can be composed multiple ways, but
173 * when normalized, these names all compare equal. A wide
174 * dirlock locks multiple names. We need these when the file
175 * system is supporting mixed-mode access. It is sometimes
176 * necessary to lock all case permutations of file name at
177 * once so that simultaneous case-insensitive/case-sensitive
178 * behaves as rationally as possible.
182 * When matching we may need to normalize & change case according to
185 * Note that a normalized match is necessary for a case insensitive
186 * filesystem when the lookup request is not exact because normalization
187 * can fold case independent of normalizing code point sequences.
189 * See the table above zfs_dropname().
191 if (zfsvfs->z_norm != 0) {
195 * Determine if the match needs to honor the case specified in
196 * lookup, and if so keep track of that so that during
197 * normalization we don't fold case.
199 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE &&
200 (flag & ZCIEXACT)) ||
201 (zfsvfs->z_case == ZFS_CASE_MIXED && !(flag & ZCILOOK))) {
207 * Only look in or update the DNLC if we are looking for the
208 * name on a file system that does not require normalization
209 * or case folding. We can also look there if we happen to be
210 * on a non-normalizing, mixed sensitivity file system IF we
211 * are looking for the exact name.
213 * Maybe can add TO-UPPERed version of name to dnlc in ci-only
214 * case for performance improvement?
216 update = !zfsvfs->z_norm ||
217 (zfsvfs->z_case == ZFS_CASE_MIXED &&
218 !(zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER) && !(flag & ZCILOOK));
221 * ZRENAMING indicates we are in a situation where we should
222 * take narrow locks regardless of the file system's
223 * preferences for normalizing and case folding. This will
224 * prevent us deadlocking trying to grab the same wide lock
225 * twice if the two names happen to be case-insensitive
228 if (flag & ZRENAMING)
231 cmpflags = zfsvfs->z_norm;
234 * Wait until there are no locks on this name.
236 * Don't grab the lock if it is already held. However, cannot
237 * have both ZSHARED and ZHAVELOCK together.
239 ASSERT(!(flag & ZSHARED) || !(flag & ZHAVELOCK));
240 if (!(flag & ZHAVELOCK))
241 rw_enter(&dzp->z_name_lock, RW_READER);
243 mutex_enter(&dzp->z_lock);
245 if (dzp->z_unlinked && !(flag & ZXATTR)) {
246 mutex_exit(&dzp->z_lock);
247 if (!(flag & ZHAVELOCK))
248 rw_exit(&dzp->z_name_lock);
249 return (SET_ERROR(ENOENT));
251 for (dl = dzp->z_dirlocks; dl != NULL; dl = dl->dl_next) {
252 if ((u8_strcmp(name, dl->dl_name, 0, cmpflags,
253 U8_UNICODE_LATEST, &error) == 0) || error != 0)
257 mutex_exit(&dzp->z_lock);
258 if (!(flag & ZHAVELOCK))
259 rw_exit(&dzp->z_name_lock);
260 return (SET_ERROR(ENOENT));
264 * Allocate a new dirlock and add it to the list.
266 dl = kmem_alloc(sizeof (zfs_dirlock_t), KM_SLEEP);
267 cv_init(&dl->dl_cv, NULL, CV_DEFAULT, NULL);
273 dl->dl_next = dzp->z_dirlocks;
274 dzp->z_dirlocks = dl;
277 if ((flag & ZSHARED) && dl->dl_sharecnt != 0)
279 cv_wait(&dl->dl_cv, &dzp->z_lock);
283 * If the z_name_lock was NOT held for this dirlock record it.
285 if (flag & ZHAVELOCK)
288 if ((flag & ZSHARED) && ++dl->dl_sharecnt > 1 && dl->dl_namesize == 0) {
290 * We're the second shared reference to dl. Make a copy of
291 * dl_name in case the first thread goes away before we do.
292 * Note that we initialize the new name before storing its
293 * pointer into dl_name, because the first thread may load
294 * dl->dl_name at any time. It'll either see the old value,
295 * which belongs to it, or the new shared copy; either is OK.
297 dl->dl_namesize = strlen(dl->dl_name) + 1;
298 name = kmem_alloc(dl->dl_namesize, KM_SLEEP);
299 bcopy(dl->dl_name, name, dl->dl_namesize);
303 mutex_exit(&dzp->z_lock);
306 * We have a dirlock on the name. (Note that it is the dirlock,
307 * not the dzp's z_lock, that protects the name in the zap object.)
308 * See if there's an object by this name; if so, put a hold on it.
311 error = sa_lookup(dzp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &zoid,
314 error = (zoid == 0 ? SET_ERROR(ENOENT) : 0);
316 error = zfs_match_find(zfsvfs, dzp, name, mt,
317 update, direntflags, realpnp, &zoid);
320 if (error != ENOENT || (flag & ZEXISTS)) {
321 zfs_dirent_unlock(dl);
326 zfs_dirent_unlock(dl);
327 return (SET_ERROR(EEXIST));
329 error = zfs_zget(zfsvfs, zoid, zpp);
331 zfs_dirent_unlock(dl);
342 * Unlock this directory entry and wake anyone who was waiting for it.
345 zfs_dirent_unlock(zfs_dirlock_t *dl)
347 znode_t *dzp = dl->dl_dzp;
348 zfs_dirlock_t **prev_dl, *cur_dl;
350 mutex_enter(&dzp->z_lock);
352 if (!dl->dl_namelock)
353 rw_exit(&dzp->z_name_lock);
355 if (dl->dl_sharecnt > 1) {
357 mutex_exit(&dzp->z_lock);
360 prev_dl = &dzp->z_dirlocks;
361 while ((cur_dl = *prev_dl) != dl)
362 prev_dl = &cur_dl->dl_next;
363 *prev_dl = dl->dl_next;
364 cv_broadcast(&dl->dl_cv);
365 mutex_exit(&dzp->z_lock);
367 if (dl->dl_namesize != 0)
368 kmem_free(dl->dl_name, dl->dl_namesize);
369 cv_destroy(&dl->dl_cv);
370 kmem_free(dl, sizeof (*dl));
374 * Look up an entry in a directory.
376 * NOTE: '.' and '..' are handled as special cases because
377 * no directory entries are actually stored for them. If this is
378 * the root of a filesystem, then '.zfs' is also treated as a
379 * special pseudo-directory.
382 zfs_dirlook(znode_t *dzp, char *name, znode_t **zpp, int flags,
383 int *deflg, pathname_t *rpnp)
391 if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) {
394 } else if (name[0] == '.' && name[1] == '.' && name[2] == 0) {
395 zfsvfs_t *zfsvfs = ZTOZSB(dzp);
398 * If we are a snapshot mounted under .zfs, return
399 * the inode pointer for the snapshot directory.
401 if ((error = sa_lookup(dzp->z_sa_hdl,
402 SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0)
405 if (parent == dzp->z_id && zfsvfs->z_parent != zfsvfs) {
406 error = zfsctl_root_lookup(zfsvfs->z_parent->z_ctldir,
407 "snapshot", &ip, 0, kcred, NULL, NULL);
411 rw_enter(&dzp->z_parent_lock, RW_READER);
412 error = zfs_zget(zfsvfs, parent, &zp);
415 rw_exit(&dzp->z_parent_lock);
416 } else if (zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0) {
417 ip = zfsctl_root(dzp);
422 zf = ZEXISTS | ZSHARED;
423 if (flags & FIGNORECASE)
426 error = zfs_dirent_lock(&dl, dzp, name, &zp, zf, deflg, rpnp);
429 zfs_dirent_unlock(dl);
430 dzp->z_zn_prefetch = B_TRUE; /* enable prefetching */
435 if ((flags & FIGNORECASE) && rpnp && !error)
436 (void) strlcpy(rpnp->pn_buf, name, rpnp->pn_bufsize);
442 * unlinked Set (formerly known as the "delete queue") Error Handling
444 * When dealing with the unlinked set, we dmu_tx_hold_zap(), but we
445 * don't specify the name of the entry that we will be manipulating. We
446 * also fib and say that we won't be adding any new entries to the
447 * unlinked set, even though we might (this is to lower the minimum file
448 * size that can be deleted in a full filesystem). So on the small
449 * chance that the nlink list is using a fat zap (ie. has more than
450 * 2000 entries), we *may* not pre-read a block that's needed.
451 * Therefore it is remotely possible for some of the assertions
452 * regarding the unlinked set below to fail due to i/o error. On a
453 * nondebug system, this will result in the space being leaked.
456 zfs_unlinked_add(znode_t *zp, dmu_tx_t *tx)
458 zfsvfs_t *zfsvfs = ZTOZSB(zp);
460 ASSERT(zp->z_unlinked);
461 ASSERT(ZTOI(zp)->i_nlink == 0);
464 zap_add_int(zfsvfs->z_os, zfsvfs->z_unlinkedobj, zp->z_id, tx));
466 dataset_kstats_update_nunlinks_kstat(&zfsvfs->z_kstat, 1);
470 * Clean up any znodes that had no links when we either crashed or
471 * (force) umounted the file system.
474 zfs_unlinked_drain_task(void *arg)
476 zfsvfs_t *zfsvfs = arg;
479 dmu_object_info_t doi;
483 ASSERT3B(zfsvfs->z_draining, ==, B_TRUE);
486 * Iterate over the contents of the unlinked set.
488 for (zap_cursor_init(&zc, zfsvfs->z_os, zfsvfs->z_unlinkedobj);
489 zap_cursor_retrieve(&zc, &zap) == 0 && !zfsvfs->z_drain_cancel;
490 zap_cursor_advance(&zc)) {
493 * See what kind of object we have in list
496 error = dmu_object_info(zfsvfs->z_os,
497 zap.za_first_integer, &doi);
501 ASSERT((doi.doi_type == DMU_OT_PLAIN_FILE_CONTENTS) ||
502 (doi.doi_type == DMU_OT_DIRECTORY_CONTENTS));
504 * We need to re-mark these list entries for deletion,
505 * so we pull them back into core and set zp->z_unlinked.
507 error = zfs_zget(zfsvfs, zap.za_first_integer, &zp);
510 * We may pick up znodes that are already marked for deletion.
511 * This could happen during the purge of an extended attribute
512 * directory. All we need to do is skip over them, since they
513 * are already in the system marked z_unlinked.
518 zp->z_unlinked = B_TRUE;
521 * zrele() decrements the znode's ref count and may cause
522 * it to be synchronously freed. We interrupt freeing
523 * of this znode by checking the return value of
524 * dmu_objset_zfs_unmounting() in dmu_free_long_range()
525 * when an unmount is requested.
528 ASSERT3B(zfsvfs->z_unmounted, ==, B_FALSE);
530 zap_cursor_fini(&zc);
532 zfsvfs->z_draining = B_FALSE;
533 zfsvfs->z_drain_task = TASKQID_INVALID;
537 * Sets z_draining then tries to dispatch async unlinked drain.
538 * If that fails executes synchronous unlinked drain.
541 zfs_unlinked_drain(zfsvfs_t *zfsvfs)
543 ASSERT3B(zfsvfs->z_unmounted, ==, B_FALSE);
544 ASSERT3B(zfsvfs->z_draining, ==, B_FALSE);
546 zfsvfs->z_draining = B_TRUE;
547 zfsvfs->z_drain_cancel = B_FALSE;
549 zfsvfs->z_drain_task = taskq_dispatch(
550 dsl_pool_unlinked_drain_taskq(dmu_objset_pool(zfsvfs->z_os)),
551 zfs_unlinked_drain_task, zfsvfs, TQ_SLEEP);
552 if (zfsvfs->z_drain_task == TASKQID_INVALID) {
553 zfs_dbgmsg("async zfs_unlinked_drain dispatch failed");
554 zfs_unlinked_drain_task(zfsvfs);
559 * Wait for the unlinked drain taskq task to stop. This will interrupt the
560 * unlinked set processing if it is in progress.
563 zfs_unlinked_drain_stop_wait(zfsvfs_t *zfsvfs)
565 ASSERT3B(zfsvfs->z_unmounted, ==, B_FALSE);
567 if (zfsvfs->z_draining) {
568 zfsvfs->z_drain_cancel = B_TRUE;
569 taskq_cancel_id(dsl_pool_unlinked_drain_taskq(
570 dmu_objset_pool(zfsvfs->z_os)), zfsvfs->z_drain_task);
571 zfsvfs->z_drain_task = TASKQID_INVALID;
572 zfsvfs->z_draining = B_FALSE;
577 * Delete the entire contents of a directory. Return a count
578 * of the number of entries that could not be deleted. If we encounter
579 * an error, return a count of at least one so that the directory stays
580 * in the unlinked set.
582 * NOTE: this function assumes that the directory is inactive,
583 * so there is no need to lock its entries before deletion.
584 * Also, it assumes the directory contents is *only* regular
588 zfs_purgedir(znode_t *dzp)
594 zfsvfs_t *zfsvfs = ZTOZSB(dzp);
599 for (zap_cursor_init(&zc, zfsvfs->z_os, dzp->z_id);
600 (error = zap_cursor_retrieve(&zc, &zap)) == 0;
601 zap_cursor_advance(&zc)) {
602 error = zfs_zget(zfsvfs,
603 ZFS_DIRENT_OBJ(zap.za_first_integer), &xzp);
609 ASSERT(S_ISREG(ZTOI(xzp)->i_mode) ||
610 S_ISLNK(ZTOI(xzp)->i_mode));
612 tx = dmu_tx_create(zfsvfs->z_os);
613 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
614 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, zap.za_name);
615 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
616 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
617 /* Is this really needed ? */
618 zfs_sa_upgrade_txholds(tx, xzp);
619 dmu_tx_mark_netfree(tx);
620 error = dmu_tx_assign(tx, TXG_WAIT);
623 zfs_zrele_async(xzp);
627 bzero(&dl, sizeof (dl));
629 dl.dl_name = zap.za_name;
631 error = zfs_link_destroy(&dl, xzp, tx, 0, NULL);
636 zfs_zrele_async(xzp);
638 zap_cursor_fini(&zc);
645 zfs_rmnode(znode_t *zp)
647 zfsvfs_t *zfsvfs = ZTOZSB(zp);
648 objset_t *os = zfsvfs->z_os;
656 ASSERT(ZTOI(zp)->i_nlink == 0);
657 ASSERT(atomic_read(&ZTOI(zp)->i_count) == 0);
660 * If this is an attribute directory, purge its contents.
662 if (S_ISDIR(ZTOI(zp)->i_mode) && (zp->z_pflags & ZFS_XATTR)) {
663 if (zfs_purgedir(zp) != 0) {
665 * Not enough space to delete some xattrs.
666 * Leave it in the unlinked set.
668 zfs_znode_dmu_fini(zp);
675 * Free up all the data in the file. We don't do this for directories
676 * because we need truncate and remove to be in the same tx, like in
677 * zfs_znode_delete(). Otherwise, if we crash here we'll end up with
678 * an inconsistent truncated zap object in the delete queue. Note a
679 * truncated file is harmless since it only contains user data.
681 if (S_ISREG(ZTOI(zp)->i_mode)) {
682 error = dmu_free_long_range(os, zp->z_id, 0, DMU_OBJECT_END);
685 * Not enough space or we were interrupted by unmount.
686 * Leave the file in the unlinked set.
688 zfs_znode_dmu_fini(zp);
694 * If the file has extended attributes, we're going to unlink
697 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
698 &xattr_obj, sizeof (xattr_obj));
699 if (error == 0 && xattr_obj) {
700 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
704 acl_obj = zfs_external_acl(zp);
707 * Set up the final transaction.
709 tx = dmu_tx_create(os);
710 dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END);
711 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
713 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, TRUE, NULL);
714 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
717 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
719 zfs_sa_upgrade_txholds(tx, zp);
720 error = dmu_tx_assign(tx, TXG_WAIT);
723 * Not enough space to delete the file. Leave it in the
724 * unlinked set, leaking it until the fs is remounted (at
725 * which point we'll call zfs_unlinked_drain() to process it).
728 zfs_znode_dmu_fini(zp);
734 mutex_enter(&xzp->z_lock);
735 xzp->z_unlinked = B_TRUE; /* mark xzp for deletion */
736 clear_nlink(ZTOI(xzp)); /* no more links to it */
738 VERIFY(0 == sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
739 &links, sizeof (links), tx));
740 mutex_exit(&xzp->z_lock);
741 zfs_unlinked_add(xzp, tx);
744 mutex_enter(&os->os_dsl_dataset->ds_dir->dd_activity_lock);
747 * Remove this znode from the unlinked set. If a has rollback has
748 * occurred while a file is open and unlinked. Then when the file
749 * is closed post rollback it will not exist in the rolled back
750 * version of the unlinked object.
752 error = zap_remove_int(zfsvfs->z_os, zfsvfs->z_unlinkedobj,
754 VERIFY(error == 0 || error == ENOENT);
757 if (zap_count(os, zfsvfs->z_unlinkedobj, &count) == 0 && count == 0) {
758 cv_broadcast(&os->os_dsl_dataset->ds_dir->dd_activity_cv);
761 mutex_exit(&os->os_dsl_dataset->ds_dir->dd_activity_lock);
763 dataset_kstats_update_nunlinked_kstat(&zfsvfs->z_kstat, 1);
765 zfs_znode_delete(zp, tx);
770 zfs_zrele_async(xzp);
774 zfs_dirent(znode_t *zp, uint64_t mode)
776 uint64_t de = zp->z_id;
778 if (ZTOZSB(zp)->z_version >= ZPL_VERSION_DIRENT_TYPE)
779 de |= IFTODT(mode) << 60;
784 * Link zp into dl. Can fail in the following cases :
785 * - if zp has been unlinked.
786 * - if the number of entries with the same hash (aka. colliding entries)
787 * exceed the capacity of a leaf-block of fatzap and splitting of the
788 * leaf-block does not help.
791 zfs_link_create(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag)
793 znode_t *dzp = dl->dl_dzp;
794 zfsvfs_t *zfsvfs = ZTOZSB(zp);
796 int zp_is_dir = S_ISDIR(ZTOI(zp)->i_mode);
797 sa_bulk_attr_t bulk[5];
798 uint64_t mtime[2], ctime[2];
803 mutex_enter(&zp->z_lock);
805 if (!(flag & ZRENAMING)) {
806 if (zp->z_unlinked) { /* no new links to unlinked zp */
807 ASSERT(!(flag & (ZNEW | ZEXISTS)));
808 mutex_exit(&zp->z_lock);
809 return (SET_ERROR(ENOENT));
811 if (!(flag & ZNEW)) {
813 * ZNEW nodes come from zfs_mknode() where the link
814 * count has already been initialised
817 links = ZTOI(zp)->i_nlink;
818 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs),
819 NULL, &links, sizeof (links));
823 value = zfs_dirent(zp, zp->z_mode);
824 error = zap_add(ZTOZSB(zp)->z_os, dzp->z_id, dl->dl_name, 8, 1,
828 * zap_add could fail to add the entry if it exceeds the capacity of the
829 * leaf-block and zap_leaf_split() failed to help.
830 * The caller of this routine is responsible for failing the transaction
831 * which will rollback the SA updates done above.
834 if (!(flag & ZRENAMING) && !(flag & ZNEW))
835 drop_nlink(ZTOI(zp));
836 mutex_exit(&zp->z_lock);
840 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL,
841 &dzp->z_id, sizeof (dzp->z_id));
842 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
843 &zp->z_pflags, sizeof (zp->z_pflags));
845 if (!(flag & ZNEW)) {
846 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
847 ctime, sizeof (ctime));
848 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime,
851 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
854 mutex_exit(&zp->z_lock);
856 mutex_enter(&dzp->z_lock);
859 inc_nlink(ZTOI(dzp));
860 links = ZTOI(dzp)->i_nlink;
862 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
863 &dzp->z_size, sizeof (dzp->z_size));
864 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
865 &links, sizeof (links));
866 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
867 mtime, sizeof (mtime));
868 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
869 ctime, sizeof (ctime));
870 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
871 &dzp->z_pflags, sizeof (dzp->z_pflags));
872 zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime);
873 error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx);
875 mutex_exit(&dzp->z_lock);
881 * The match type in the code for this function should conform to:
883 * ------------------------------------------------------------------------
884 * fs type | z_norm | lookup type | match type
885 * ---------|-------------|-------------|----------------------------------
886 * CS !norm | 0 | 0 | 0 (exact)
887 * CS norm | formX | 0 | MT_NORMALIZE
888 * CI !norm | upper | !ZCIEXACT | MT_NORMALIZE
889 * CI !norm | upper | ZCIEXACT | MT_NORMALIZE | MT_MATCH_CASE
890 * CI norm | upper|formX | !ZCIEXACT | MT_NORMALIZE
891 * CI norm | upper|formX | ZCIEXACT | MT_NORMALIZE | MT_MATCH_CASE
892 * CM !norm | upper | !ZCILOOK | MT_NORMALIZE | MT_MATCH_CASE
893 * CM !norm | upper | ZCILOOK | MT_NORMALIZE
894 * CM norm | upper|formX | !ZCILOOK | MT_NORMALIZE | MT_MATCH_CASE
895 * CM norm | upper|formX | ZCILOOK | MT_NORMALIZE
898 * CS = Case Sensitive, CI = Case Insensitive, CM = Case Mixed
899 * upper = case folding set by fs type on creation (U8_TEXTPREP_TOUPPER)
900 * formX = unicode normalization form set on fs creation
903 zfs_dropname(zfs_dirlock_t *dl, znode_t *zp, znode_t *dzp, dmu_tx_t *tx,
908 if (ZTOZSB(zp)->z_norm) {
909 matchtype_t mt = MT_NORMALIZE;
911 if ((ZTOZSB(zp)->z_case == ZFS_CASE_INSENSITIVE &&
912 (flag & ZCIEXACT)) ||
913 (ZTOZSB(zp)->z_case == ZFS_CASE_MIXED &&
914 !(flag & ZCILOOK))) {
918 error = zap_remove_norm(ZTOZSB(zp)->z_os, dzp->z_id,
919 dl->dl_name, mt, tx);
921 error = zap_remove(ZTOZSB(zp)->z_os, dzp->z_id, dl->dl_name,
929 * Unlink zp from dl, and mark zp for deletion if this was the last link. Can
930 * fail if zp is a mount point (EBUSY) or a non-empty directory (ENOTEMPTY).
931 * If 'unlinkedp' is NULL, we put unlinked znodes on the unlinked list.
932 * If it's non-NULL, we use it to indicate whether the znode needs deletion,
933 * and it's the caller's job to do it.
936 zfs_link_destroy(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag,
937 boolean_t *unlinkedp)
939 znode_t *dzp = dl->dl_dzp;
940 zfsvfs_t *zfsvfs = ZTOZSB(dzp);
941 int zp_is_dir = S_ISDIR(ZTOI(zp)->i_mode);
942 boolean_t unlinked = B_FALSE;
943 sa_bulk_attr_t bulk[5];
944 uint64_t mtime[2], ctime[2];
949 if (!(flag & ZRENAMING)) {
950 mutex_enter(&zp->z_lock);
952 if (zp_is_dir && !zfs_dirempty(zp)) {
953 mutex_exit(&zp->z_lock);
954 return (SET_ERROR(ENOTEMPTY));
958 * If we get here, we are going to try to remove the object.
959 * First try removing the name from the directory; if that
960 * fails, return the error.
962 error = zfs_dropname(dl, zp, dzp, tx, flag);
964 mutex_exit(&zp->z_lock);
968 if (ZTOI(zp)->i_nlink <= zp_is_dir) {
969 zfs_panic_recover("zfs: link count on %lu is %u, "
970 "should be at least %u", zp->z_id,
971 (int)ZTOI(zp)->i_nlink, zp_is_dir + 1);
972 set_nlink(ZTOI(zp), zp_is_dir + 1);
974 drop_nlink(ZTOI(zp));
975 if (ZTOI(zp)->i_nlink == zp_is_dir) {
976 zp->z_unlinked = B_TRUE;
977 clear_nlink(ZTOI(zp));
980 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs),
981 NULL, &ctime, sizeof (ctime));
982 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
983 NULL, &zp->z_pflags, sizeof (zp->z_pflags));
984 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime,
987 links = ZTOI(zp)->i_nlink;
988 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs),
989 NULL, &links, sizeof (links));
990 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
993 mutex_exit(&zp->z_lock);
995 error = zfs_dropname(dl, zp, dzp, tx, flag);
1000 mutex_enter(&dzp->z_lock);
1001 dzp->z_size--; /* one dirent removed */
1003 drop_nlink(ZTOI(dzp)); /* ".." link from zp */
1004 links = ZTOI(dzp)->i_nlink;
1005 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs),
1006 NULL, &links, sizeof (links));
1007 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
1008 NULL, &dzp->z_size, sizeof (dzp->z_size));
1009 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs),
1010 NULL, ctime, sizeof (ctime));
1011 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
1012 NULL, mtime, sizeof (mtime));
1013 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1014 NULL, &dzp->z_pflags, sizeof (dzp->z_pflags));
1015 zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime);
1016 error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx);
1018 mutex_exit(&dzp->z_lock);
1020 if (unlinkedp != NULL)
1021 *unlinkedp = unlinked;
1023 zfs_unlinked_add(zp, tx);
1029 * Indicate whether the directory is empty. Works with or without z_lock
1030 * held, but can only be consider a hint in the latter case. Returns true
1031 * if only "." and ".." remain and there's no work in progress.
1033 * The internal ZAP size, rather than zp->z_size, needs to be checked since
1034 * some consumers (Lustre) do not strictly maintain an accurate SA_ZPL_SIZE.
1037 zfs_dirempty(znode_t *dzp)
1039 zfsvfs_t *zfsvfs = ZTOZSB(dzp);
1043 if (dzp->z_dirlocks != NULL)
1046 error = zap_count(zfsvfs->z_os, dzp->z_id, &count);
1047 if (error != 0 || count != 0)
1054 zfs_make_xattrdir(znode_t *zp, vattr_t *vap, znode_t **xzpp, cred_t *cr)
1056 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1060 zfs_acl_ids_t acl_ids;
1061 boolean_t fuid_dirtied;
1068 if ((error = zfs_zaccess(zp, ACE_WRITE_NAMED_ATTRS, 0, B_FALSE, cr)))
1071 if ((error = zfs_acl_ids_create(zp, IS_XATTR, vap, cr, NULL,
1074 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, zp->z_projid)) {
1075 zfs_acl_ids_free(&acl_ids);
1076 return (SET_ERROR(EDQUOT));
1079 tx = dmu_tx_create(zfsvfs->z_os);
1080 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1081 ZFS_SA_BASE_ATTR_SIZE);
1082 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1083 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1084 fuid_dirtied = zfsvfs->z_fuid_dirty;
1086 zfs_fuid_txhold(zfsvfs, tx);
1087 error = dmu_tx_assign(tx, TXG_WAIT);
1089 zfs_acl_ids_free(&acl_ids);
1093 zfs_mknode(zp, vap, tx, cr, IS_XATTR, &xzp, &acl_ids);
1096 zfs_fuid_sync(zfsvfs, tx);
1099 error = sa_lookup(xzp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
1100 &parent, sizeof (parent));
1101 ASSERT(error == 0 && parent == zp->z_id);
1104 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &xzp->z_id,
1105 sizeof (xzp->z_id), tx));
1107 if (!zp->z_unlinked)
1108 (void) zfs_log_create(zfsvfs->z_log, tx, TX_MKXATTR, zp,
1109 xzp, "", NULL, acl_ids.z_fuidp, vap);
1111 zfs_acl_ids_free(&acl_ids);
1120 * Return a znode for the extended attribute directory for zp.
1121 * ** If the directory does not already exist, it is created **
1123 * IN: zp - znode to obtain attribute directory from
1124 * cr - credentials of caller
1125 * flags - flags from the VOP_LOOKUP call
1127 * OUT: xipp - pointer to extended attribute znode
1129 * RETURN: 0 on success
1130 * error number on failure
1133 zfs_get_xattrdir(znode_t *zp, znode_t **xzpp, cred_t *cr, int flags)
1135 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1141 error = zfs_dirent_lock(&dl, zp, "", &xzp, ZXATTR, NULL, NULL);
1147 zfs_dirent_unlock(dl);
1151 if (!(flags & CREATE_XATTR_DIR)) {
1152 zfs_dirent_unlock(dl);
1153 return (SET_ERROR(ENOENT));
1156 if (zfs_is_readonly(zfsvfs)) {
1157 zfs_dirent_unlock(dl);
1158 return (SET_ERROR(EROFS));
1162 * The ability to 'create' files in an attribute
1163 * directory comes from the write_xattr permission on the base file.
1165 * The ability to 'search' an attribute directory requires
1166 * read_xattr permission on the base file.
1168 * Once in a directory the ability to read/write attributes
1169 * is controlled by the permissions on the attribute file.
1171 va.va_mask = ATTR_MODE | ATTR_UID | ATTR_GID;
1172 va.va_mode = S_IFDIR | S_ISVTX | 0777;
1173 zfs_fuid_map_ids(zp, cr, &va.va_uid, &va.va_gid);
1175 va.va_dentry = NULL;
1176 error = zfs_make_xattrdir(zp, &va, xzpp, cr);
1177 zfs_dirent_unlock(dl);
1179 if (error == ERESTART) {
1180 /* NB: we already did dmu_tx_wait() if necessary */
1188 * Decide whether it is okay to remove within a sticky directory.
1190 * In sticky directories, write access is not sufficient;
1191 * you can remove entries from a directory only if:
1193 * you own the directory,
1194 * you own the entry,
1195 * you have write access to the entry,
1196 * or you are privileged (checked in secpolicy...).
1198 * The function returns 0 if remove access is granted.
1201 zfs_sticky_remove_access(znode_t *zdp, znode_t *zp, cred_t *cr)
1206 zfsvfs_t *zfsvfs = ZTOZSB(zdp);
1208 if (zfsvfs->z_replay)
1211 if ((zdp->z_mode & S_ISVTX) == 0)
1214 downer = zfs_fuid_map_id(zfsvfs, KUID_TO_SUID(ZTOI(zdp)->i_uid),
1216 fowner = zfs_fuid_map_id(zfsvfs, KUID_TO_SUID(ZTOI(zp)->i_uid),
1219 if ((uid = crgetuid(cr)) == downer || uid == fowner ||
1220 zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr) == 0)
1223 return (secpolicy_vnode_remove(cr));