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>
38 #include <sys/pathname.h>
39 #include <sys/cmn_err.h>
40 #include <sys/errno.h>
42 #include <sys/sunddi.h>
43 #include <sys/random.h>
44 #include <sys/policy.h>
45 #include <sys/zfs_dir.h>
46 #include <sys/zfs_acl.h>
47 #include <sys/zfs_vnops.h>
48 #include <sys/fs/zfs.h>
51 #include <sys/atomic.h>
52 #include <sys/zfs_ctldir.h>
53 #include <sys/zfs_fuid.h>
55 #include <sys/zfs_sa.h>
58 * zfs_match_find() is used by zfs_dirent_lock() to peform zap lookups
59 * of names after deciding which is the appropriate lookup interface.
62 zfs_match_find(zfsvfs_t
*zfsvfs
, znode_t
*dzp
, char *name
, matchtype_t mt
,
63 boolean_t update
, int *deflags
, pathname_t
*rpnp
, uint64_t *zoid
)
65 boolean_t conflict
= B_FALSE
;
74 bufsz
= rpnp
->pn_bufsize
;
78 * In the non-mixed case we only expect there would ever
79 * be one match, but we need to use the normalizing lookup.
81 error
= zap_lookup_norm(zfsvfs
->z_os
, dzp
->z_id
, name
, 8, 1,
82 zoid
, mt
, buf
, bufsz
, &conflict
);
84 error
= zap_lookup(zfsvfs
->z_os
, dzp
->z_id
, name
, 8, 1, zoid
);
88 * Allow multiple entries provided the first entry is
89 * the object id. Non-zpl consumers may safely make
90 * use of the additional space.
92 * XXX: This should be a feature flag for compatibility
94 if (error
== EOVERFLOW
)
97 if (zfsvfs
->z_norm
&& !error
&& deflags
)
98 *deflags
= conflict
? ED_CASE_CONFLICT
: 0;
100 *zoid
= ZFS_DIRENT_OBJ(*zoid
);
106 * Lock a directory entry. A dirlock on <dzp, name> protects that name
107 * in dzp's directory zap object. As long as you hold a dirlock, you can
108 * assume two things: (1) dzp cannot be reaped, and (2) no other thread
109 * can change the zap entry for (i.e. link or unlink) this name.
112 * dzp - znode for directory
113 * name - name of entry to lock
114 * flag - ZNEW: if the entry already exists, fail with EEXIST.
115 * ZEXISTS: if the entry does not exist, fail with ENOENT.
116 * ZSHARED: allow concurrent access with other ZSHARED callers.
117 * ZXATTR: we want dzp's xattr directory
118 * ZCILOOK: On a mixed sensitivity file system,
119 * this lookup should be case-insensitive.
120 * ZCIEXACT: On a purely case-insensitive file system,
121 * this lookup should be case-sensitive.
122 * ZRENAMING: we are locking for renaming, force narrow locks
123 * ZHAVELOCK: Don't grab the z_name_lock for this call. The
124 * current thread already holds it.
127 * zpp - pointer to the znode for the entry (NULL if there isn't one)
128 * dlpp - pointer to the dirlock for this entry (NULL on error)
129 * direntflags - (case-insensitive lookup only)
130 * flags if multiple case-sensitive matches exist in directory
131 * realpnp - (case-insensitive lookup only)
132 * actual name matched within the directory
134 * Return value: 0 on success or errno on failure.
136 * NOTE: Always checks for, and rejects, '.' and '..'.
137 * NOTE: For case-insensitive file systems we take wide locks (see below),
138 * but return znode pointers to a single match.
141 zfs_dirent_lock(zfs_dirlock_t
**dlpp
, znode_t
*dzp
, char *name
, znode_t
**zpp
,
142 int flag
, int *direntflags
, pathname_t
*realpnp
)
144 zfsvfs_t
*zfsvfs
= ZTOZSB(dzp
);
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))
161 return (SET_ERROR(EEXIST
));
164 * Case sensitivity and normalization preferences are set when
165 * the file system is created. These are stored in the
166 * zfsvfs->z_case and zfsvfs->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 * When matching we may need to normalize & change case according to
184 * Note that a normalized match is necessary for a case insensitive
185 * filesystem when the lookup request is not exact because normalization
186 * can fold case independent of normalizing code point sequences.
188 * See the table above zfs_dropname().
190 if (zfsvfs
->z_norm
!= 0) {
194 * Determine if the match needs to honor the case specified in
195 * lookup, and if so keep track of that so that during
196 * normalization we don't fold case.
198 if ((zfsvfs
->z_case
== ZFS_CASE_INSENSITIVE
&&
199 (flag
& ZCIEXACT
)) ||
200 (zfsvfs
->z_case
== ZFS_CASE_MIXED
&& !(flag
& ZCILOOK
))) {
206 * Only look in or update the DNLC if we are looking for the
207 * name on a file system that does not require normalization
208 * or case folding. We can also look there if we happen to be
209 * on a non-normalizing, mixed sensitivity file system IF we
210 * are looking for the exact name.
212 * Maybe can add TO-UPPERed version of name to dnlc in ci-only
213 * case for performance improvement?
215 update
= !zfsvfs
->z_norm
||
216 (zfsvfs
->z_case
== ZFS_CASE_MIXED
&&
217 !(zfsvfs
->z_norm
& ~U8_TEXTPREP_TOUPPER
) && !(flag
& ZCILOOK
));
220 * ZRENAMING indicates we are in a situation where we should
221 * take narrow locks regardless of the file system's
222 * preferences for normalizing and case folding. This will
223 * prevent us deadlocking trying to grab the same wide lock
224 * twice if the two names happen to be case-insensitive
227 if (flag
& ZRENAMING
)
230 cmpflags
= zfsvfs
->z_norm
;
233 * Wait until there are no locks on this name.
235 * Don't grab the the lock if it is already held. However, cannot
236 * have both ZSHARED and ZHAVELOCK together.
238 ASSERT(!(flag
& ZSHARED
) || !(flag
& ZHAVELOCK
));
239 if (!(flag
& ZHAVELOCK
))
240 rw_enter(&dzp
->z_name_lock
, RW_READER
);
242 mutex_enter(&dzp
->z_lock
);
244 if (dzp
->z_unlinked
&& !(flag
& ZXATTR
)) {
245 mutex_exit(&dzp
->z_lock
);
246 if (!(flag
& ZHAVELOCK
))
247 rw_exit(&dzp
->z_name_lock
);
248 return (SET_ERROR(ENOENT
));
250 for (dl
= dzp
->z_dirlocks
; dl
!= NULL
; dl
= dl
->dl_next
) {
251 if ((u8_strcmp(name
, dl
->dl_name
, 0, cmpflags
,
252 U8_UNICODE_LATEST
, &error
) == 0) || error
!= 0)
256 mutex_exit(&dzp
->z_lock
);
257 if (!(flag
& ZHAVELOCK
))
258 rw_exit(&dzp
->z_name_lock
);
259 return (SET_ERROR(ENOENT
));
263 * Allocate a new dirlock and add it to the list.
265 dl
= kmem_alloc(sizeof (zfs_dirlock_t
), KM_SLEEP
);
266 cv_init(&dl
->dl_cv
, NULL
, CV_DEFAULT
, NULL
);
272 dl
->dl_next
= dzp
->z_dirlocks
;
273 dzp
->z_dirlocks
= dl
;
276 if ((flag
& ZSHARED
) && dl
->dl_sharecnt
!= 0)
278 cv_wait(&dl
->dl_cv
, &dzp
->z_lock
);
282 * If the z_name_lock was NOT held for this dirlock record it.
284 if (flag
& ZHAVELOCK
)
287 if ((flag
& ZSHARED
) && ++dl
->dl_sharecnt
> 1 && dl
->dl_namesize
== 0) {
289 * We're the second shared reference to dl. Make a copy of
290 * dl_name in case the first thread goes away before we do.
291 * Note that we initialize the new name before storing its
292 * pointer into dl_name, because the first thread may load
293 * dl->dl_name at any time. It'll either see the old value,
294 * which belongs to it, or the new shared copy; either is OK.
296 dl
->dl_namesize
= strlen(dl
->dl_name
) + 1;
297 name
= kmem_alloc(dl
->dl_namesize
, KM_SLEEP
);
298 bcopy(dl
->dl_name
, name
, dl
->dl_namesize
);
302 mutex_exit(&dzp
->z_lock
);
305 * We have a dirlock on the name. (Note that it is the dirlock,
306 * not the dzp's z_lock, that protects the name in the zap object.)
307 * See if there's an object by this name; if so, put a hold on it.
310 error
= sa_lookup(dzp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
), &zoid
,
313 error
= (zoid
== 0 ? SET_ERROR(ENOENT
) : 0);
315 error
= zfs_match_find(zfsvfs
, dzp
, name
, mt
,
316 update
, direntflags
, realpnp
, &zoid
);
319 if (error
!= ENOENT
|| (flag
& ZEXISTS
)) {
320 zfs_dirent_unlock(dl
);
325 zfs_dirent_unlock(dl
);
326 return (SET_ERROR(EEXIST
));
328 error
= zfs_zget(zfsvfs
, zoid
, zpp
);
330 zfs_dirent_unlock(dl
);
341 * Unlock this directory entry and wake anyone who was waiting for it.
344 zfs_dirent_unlock(zfs_dirlock_t
*dl
)
346 znode_t
*dzp
= dl
->dl_dzp
;
347 zfs_dirlock_t
**prev_dl
, *cur_dl
;
349 mutex_enter(&dzp
->z_lock
);
351 if (!dl
->dl_namelock
)
352 rw_exit(&dzp
->z_name_lock
);
354 if (dl
->dl_sharecnt
> 1) {
356 mutex_exit(&dzp
->z_lock
);
359 prev_dl
= &dzp
->z_dirlocks
;
360 while ((cur_dl
= *prev_dl
) != dl
)
361 prev_dl
= &cur_dl
->dl_next
;
362 *prev_dl
= dl
->dl_next
;
363 cv_broadcast(&dl
->dl_cv
);
364 mutex_exit(&dzp
->z_lock
);
366 if (dl
->dl_namesize
!= 0)
367 kmem_free(dl
->dl_name
, dl
->dl_namesize
);
368 cv_destroy(&dl
->dl_cv
);
369 kmem_free(dl
, sizeof (*dl
));
373 * Look up an entry in a directory.
375 * NOTE: '.' and '..' are handled as special cases because
376 * no directory entries are actually stored for them. If this is
377 * the root of a filesystem, then '.zfs' is also treated as a
378 * special pseudo-directory.
381 zfs_dirlook(znode_t
*dzp
, char *name
, struct inode
**ipp
, int flags
,
382 int *deflg
, pathname_t
*rpnp
)
389 if (name
[0] == 0 || (name
[0] == '.' && name
[1] == 0)) {
392 } else if (name
[0] == '.' && name
[1] == '.' && name
[2] == 0) {
393 zfsvfs_t
*zfsvfs
= ZTOZSB(dzp
);
396 * If we are a snapshot mounted under .zfs, return
397 * the inode pointer for the snapshot directory.
399 if ((error
= sa_lookup(dzp
->z_sa_hdl
,
400 SA_ZPL_PARENT(zfsvfs
), &parent
, sizeof (parent
))) != 0)
403 if (parent
== dzp
->z_id
&& zfsvfs
->z_parent
!= zfsvfs
) {
404 error
= zfsctl_root_lookup(zfsvfs
->z_parent
->z_ctldir
,
405 "snapshot", ipp
, 0, kcred
, NULL
, NULL
);
408 rw_enter(&dzp
->z_parent_lock
, RW_READER
);
409 error
= zfs_zget(zfsvfs
, parent
, &zp
);
412 rw_exit(&dzp
->z_parent_lock
);
413 } else if (zfs_has_ctldir(dzp
) && strcmp(name
, ZFS_CTLDIR_NAME
) == 0) {
414 *ipp
= zfsctl_root(dzp
);
418 zf
= ZEXISTS
| ZSHARED
;
419 if (flags
& FIGNORECASE
)
422 error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zf
, deflg
, rpnp
);
425 zfs_dirent_unlock(dl
);
426 dzp
->z_zn_prefetch
= B_TRUE
; /* enable prefetching */
431 if ((flags
& FIGNORECASE
) && rpnp
&& !error
)
432 (void) strlcpy(rpnp
->pn_buf
, name
, rpnp
->pn_bufsize
);
438 * unlinked Set (formerly known as the "delete queue") Error Handling
440 * When dealing with the unlinked set, we dmu_tx_hold_zap(), but we
441 * don't specify the name of the entry that we will be manipulating. We
442 * also fib and say that we won't be adding any new entries to the
443 * unlinked set, even though we might (this is to lower the minimum file
444 * size that can be deleted in a full filesystem). So on the small
445 * chance that the nlink list is using a fat zap (ie. has more than
446 * 2000 entries), we *may* not pre-read a block that's needed.
447 * Therefore it is remotely possible for some of the assertions
448 * regarding the unlinked set below to fail due to i/o error. On a
449 * nondebug system, this will result in the space being leaked.
452 zfs_unlinked_add(znode_t
*zp
, dmu_tx_t
*tx
)
454 zfsvfs_t
*zfsvfs
= ZTOZSB(zp
);
456 ASSERT(zp
->z_unlinked
);
457 ASSERT(ZTOI(zp
)->i_nlink
== 0);
460 zap_add_int(zfsvfs
->z_os
, zfsvfs
->z_unlinkedobj
, zp
->z_id
, tx
));
462 dataset_kstats_update_nunlinks_kstat(&zfsvfs
->z_kstat
, 1);
466 * Clean up any znodes that had no links when we either crashed or
467 * (force) umounted the file system.
470 zfs_unlinked_drain_task(void *arg
)
472 zfsvfs_t
*zfsvfs
= arg
;
475 dmu_object_info_t doi
;
479 ASSERT3B(zfsvfs
->z_draining
, ==, B_TRUE
);
482 * Iterate over the contents of the unlinked set.
484 for (zap_cursor_init(&zc
, zfsvfs
->z_os
, zfsvfs
->z_unlinkedobj
);
485 zap_cursor_retrieve(&zc
, &zap
) == 0 && !zfsvfs
->z_drain_cancel
;
486 zap_cursor_advance(&zc
)) {
489 * See what kind of object we have in list
492 error
= dmu_object_info(zfsvfs
->z_os
,
493 zap
.za_first_integer
, &doi
);
497 ASSERT((doi
.doi_type
== DMU_OT_PLAIN_FILE_CONTENTS
) ||
498 (doi
.doi_type
== DMU_OT_DIRECTORY_CONTENTS
));
500 * We need to re-mark these list entries for deletion,
501 * so we pull them back into core and set zp->z_unlinked.
503 error
= zfs_zget(zfsvfs
, zap
.za_first_integer
, &zp
);
506 * We may pick up znodes that are already marked for deletion.
507 * This could happen during the purge of an extended attribute
508 * directory. All we need to do is skip over them, since they
509 * are already in the system marked z_unlinked.
514 zp
->z_unlinked
= B_TRUE
;
517 * iput() is Linux's equivalent to illumos' VN_RELE(). It will
518 * decrement the inode's ref count and may cause the inode to be
519 * synchronously freed. We interrupt freeing of this inode, by
520 * checking the return value of dmu_objset_zfs_unmounting() in
521 * dmu_free_long_range(), when an unmount is requested.
524 ASSERT3B(zfsvfs
->z_unmounted
, ==, B_FALSE
);
526 zap_cursor_fini(&zc
);
528 zfsvfs
->z_draining
= B_FALSE
;
529 zfsvfs
->z_drain_task
= TASKQID_INVALID
;
533 * Sets z_draining then tries to dispatch async unlinked drain.
534 * If that fails executes synchronous unlinked drain.
537 zfs_unlinked_drain(zfsvfs_t
*zfsvfs
)
539 ASSERT3B(zfsvfs
->z_unmounted
, ==, B_FALSE
);
540 ASSERT3B(zfsvfs
->z_draining
, ==, B_FALSE
);
542 zfsvfs
->z_draining
= B_TRUE
;
543 zfsvfs
->z_drain_cancel
= B_FALSE
;
545 zfsvfs
->z_drain_task
= taskq_dispatch(
546 dsl_pool_unlinked_drain_taskq(dmu_objset_pool(zfsvfs
->z_os
)),
547 zfs_unlinked_drain_task
, zfsvfs
, TQ_SLEEP
);
548 if (zfsvfs
->z_drain_task
== TASKQID_INVALID
) {
549 zfs_dbgmsg("async zfs_unlinked_drain dispatch failed");
550 zfs_unlinked_drain_task(zfsvfs
);
555 * Wait for the unlinked drain taskq task to stop. This will interrupt the
556 * unlinked set processing if it is in progress.
559 zfs_unlinked_drain_stop_wait(zfsvfs_t
*zfsvfs
)
561 ASSERT3B(zfsvfs
->z_unmounted
, ==, B_FALSE
);
563 if (zfsvfs
->z_draining
) {
564 zfsvfs
->z_drain_cancel
= B_TRUE
;
565 taskq_cancel_id(dsl_pool_unlinked_drain_taskq(
566 dmu_objset_pool(zfsvfs
->z_os
)), zfsvfs
->z_drain_task
);
567 zfsvfs
->z_drain_task
= TASKQID_INVALID
;
568 zfsvfs
->z_draining
= B_FALSE
;
573 * Delete the entire contents of a directory. Return a count
574 * of the number of entries that could not be deleted. If we encounter
575 * an error, return a count of at least one so that the directory stays
576 * in the unlinked set.
578 * NOTE: this function assumes that the directory is inactive,
579 * so there is no need to lock its entries before deletion.
580 * Also, it assumes the directory contents is *only* regular
584 zfs_purgedir(znode_t
*dzp
)
590 zfsvfs_t
*zfsvfs
= ZTOZSB(dzp
);
595 for (zap_cursor_init(&zc
, zfsvfs
->z_os
, dzp
->z_id
);
596 (error
= zap_cursor_retrieve(&zc
, &zap
)) == 0;
597 zap_cursor_advance(&zc
)) {
598 error
= zfs_zget(zfsvfs
,
599 ZFS_DIRENT_OBJ(zap
.za_first_integer
), &xzp
);
605 ASSERT(S_ISREG(ZTOI(xzp
)->i_mode
) ||
606 S_ISLNK(ZTOI(xzp
)->i_mode
));
608 tx
= dmu_tx_create(zfsvfs
->z_os
);
609 dmu_tx_hold_sa(tx
, dzp
->z_sa_hdl
, B_FALSE
);
610 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, zap
.za_name
);
611 dmu_tx_hold_sa(tx
, xzp
->z_sa_hdl
, B_FALSE
);
612 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
613 /* Is this really needed ? */
614 zfs_sa_upgrade_txholds(tx
, xzp
);
615 dmu_tx_mark_netfree(tx
);
616 error
= dmu_tx_assign(tx
, TXG_WAIT
);
619 zfs_iput_async(ZTOI(xzp
));
623 bzero(&dl
, sizeof (dl
));
625 dl
.dl_name
= zap
.za_name
;
627 error
= zfs_link_destroy(&dl
, xzp
, tx
, 0, NULL
);
632 zfs_iput_async(ZTOI(xzp
));
634 zap_cursor_fini(&zc
);
641 zfs_rmnode(znode_t
*zp
)
643 zfsvfs_t
*zfsvfs
= ZTOZSB(zp
);
644 objset_t
*os
= zfsvfs
->z_os
;
652 ASSERT(ZTOI(zp
)->i_nlink
== 0);
653 ASSERT(atomic_read(&ZTOI(zp
)->i_count
) == 0);
656 * If this is an attribute directory, purge its contents.
658 if (S_ISDIR(ZTOI(zp
)->i_mode
) && (zp
->z_pflags
& ZFS_XATTR
)) {
659 if (zfs_purgedir(zp
) != 0) {
661 * Not enough space to delete some xattrs.
662 * Leave it in the unlinked set.
664 zfs_znode_dmu_fini(zp
);
671 * Free up all the data in the file. We don't do this for directories
672 * because we need truncate and remove to be in the same tx, like in
673 * zfs_znode_delete(). Otherwise, if we crash here we'll end up with
674 * an inconsistent truncated zap object in the delete queue. Note a
675 * truncated file is harmless since it only contains user data.
677 if (S_ISREG(ZTOI(zp
)->i_mode
)) {
678 error
= dmu_free_long_range(os
, zp
->z_id
, 0, DMU_OBJECT_END
);
681 * Not enough space or we were interrupted by unmount.
682 * Leave the file in the unlinked set.
684 zfs_znode_dmu_fini(zp
);
690 * If the file has extended attributes, we're going to unlink
693 error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
),
694 &xattr_obj
, sizeof (xattr_obj
));
695 if (error
== 0 && xattr_obj
) {
696 error
= zfs_zget(zfsvfs
, xattr_obj
, &xzp
);
700 acl_obj
= zfs_external_acl(zp
);
703 * Set up the final transaction.
705 tx
= dmu_tx_create(os
);
706 dmu_tx_hold_free(tx
, zp
->z_id
, 0, DMU_OBJECT_END
);
707 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
709 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, TRUE
, NULL
);
710 dmu_tx_hold_sa(tx
, xzp
->z_sa_hdl
, B_FALSE
);
713 dmu_tx_hold_free(tx
, acl_obj
, 0, DMU_OBJECT_END
);
715 zfs_sa_upgrade_txholds(tx
, zp
);
716 error
= dmu_tx_assign(tx
, TXG_WAIT
);
719 * Not enough space to delete the file. Leave it in the
720 * unlinked set, leaking it until the fs is remounted (at
721 * which point we'll call zfs_unlinked_drain() to process it).
724 zfs_znode_dmu_fini(zp
);
730 mutex_enter(&xzp
->z_lock
);
731 xzp
->z_unlinked
= B_TRUE
; /* mark xzp for deletion */
732 clear_nlink(ZTOI(xzp
)); /* no more links to it */
734 VERIFY(0 == sa_update(xzp
->z_sa_hdl
, SA_ZPL_LINKS(zfsvfs
),
735 &links
, sizeof (links
), tx
));
736 mutex_exit(&xzp
->z_lock
);
737 zfs_unlinked_add(xzp
, tx
);
740 /* Remove this znode from the unlinked set */
742 zap_remove_int(zfsvfs
->z_os
, zfsvfs
->z_unlinkedobj
, zp
->z_id
, tx
));
744 dataset_kstats_update_nunlinked_kstat(&zfsvfs
->z_kstat
, 1);
746 zfs_znode_delete(zp
, tx
);
751 zfs_iput_async(ZTOI(xzp
));
755 zfs_dirent(znode_t
*zp
, uint64_t mode
)
757 uint64_t de
= zp
->z_id
;
759 if (ZTOZSB(zp
)->z_version
>= ZPL_VERSION_DIRENT_TYPE
)
760 de
|= IFTODT(mode
) << 60;
765 * Link zp into dl. Can fail in the following cases :
766 * - if zp has been unlinked.
767 * - if the number of entries with the same hash (aka. colliding entries)
768 * exceed the capacity of a leaf-block of fatzap and splitting of the
769 * leaf-block does not help.
772 zfs_link_create(zfs_dirlock_t
*dl
, znode_t
*zp
, dmu_tx_t
*tx
, int flag
)
774 znode_t
*dzp
= dl
->dl_dzp
;
775 zfsvfs_t
*zfsvfs
= ZTOZSB(zp
);
777 int zp_is_dir
= S_ISDIR(ZTOI(zp
)->i_mode
);
778 sa_bulk_attr_t bulk
[5];
779 uint64_t mtime
[2], ctime
[2];
784 mutex_enter(&zp
->z_lock
);
786 if (!(flag
& ZRENAMING
)) {
787 if (zp
->z_unlinked
) { /* no new links to unlinked zp */
788 ASSERT(!(flag
& (ZNEW
| ZEXISTS
)));
789 mutex_exit(&zp
->z_lock
);
790 return (SET_ERROR(ENOENT
));
792 if (!(flag
& ZNEW
)) {
794 * ZNEW nodes come from zfs_mknode() where the link
795 * count has already been initialised
798 links
= ZTOI(zp
)->i_nlink
;
799 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_LINKS(zfsvfs
),
800 NULL
, &links
, sizeof (links
));
804 value
= zfs_dirent(zp
, zp
->z_mode
);
805 error
= zap_add(ZTOZSB(zp
)->z_os
, dzp
->z_id
, dl
->dl_name
, 8, 1,
809 * zap_add could fail to add the entry if it exceeds the capacity of the
810 * leaf-block and zap_leaf_split() failed to help.
811 * The caller of this routine is responsible for failing the transaction
812 * which will rollback the SA updates done above.
815 if (!(flag
& ZRENAMING
) && !(flag
& ZNEW
))
816 drop_nlink(ZTOI(zp
));
817 mutex_exit(&zp
->z_lock
);
821 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_PARENT(zfsvfs
), NULL
,
822 &dzp
->z_id
, sizeof (dzp
->z_id
));
823 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
), NULL
,
824 &zp
->z_pflags
, sizeof (zp
->z_pflags
));
826 if (!(flag
& ZNEW
)) {
827 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
,
828 ctime
, sizeof (ctime
));
829 zfs_tstamp_update_setup(zp
, STATE_CHANGED
, mtime
,
832 error
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
835 mutex_exit(&zp
->z_lock
);
837 mutex_enter(&dzp
->z_lock
);
840 inc_nlink(ZTOI(dzp
));
841 links
= ZTOI(dzp
)->i_nlink
;
843 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_SIZE(zfsvfs
), NULL
,
844 &dzp
->z_size
, sizeof (dzp
->z_size
));
845 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_LINKS(zfsvfs
), NULL
,
846 &links
, sizeof (links
));
847 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
,
848 mtime
, sizeof (mtime
));
849 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
,
850 ctime
, sizeof (ctime
));
851 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
), NULL
,
852 &dzp
->z_pflags
, sizeof (dzp
->z_pflags
));
853 zfs_tstamp_update_setup(dzp
, CONTENT_MODIFIED
, mtime
, ctime
);
854 error
= sa_bulk_update(dzp
->z_sa_hdl
, bulk
, count
, tx
);
856 mutex_exit(&dzp
->z_lock
);
862 * The match type in the code for this function should conform to:
864 * ------------------------------------------------------------------------
865 * fs type | z_norm | lookup type | match type
866 * ---------|-------------|-------------|----------------------------------
867 * CS !norm | 0 | 0 | 0 (exact)
868 * CS norm | formX | 0 | MT_NORMALIZE
869 * CI !norm | upper | !ZCIEXACT | MT_NORMALIZE
870 * CI !norm | upper | ZCIEXACT | MT_NORMALIZE | MT_MATCH_CASE
871 * CI norm | upper|formX | !ZCIEXACT | MT_NORMALIZE
872 * CI norm | upper|formX | ZCIEXACT | MT_NORMALIZE | MT_MATCH_CASE
873 * CM !norm | upper | !ZCILOOK | MT_NORMALIZE | MT_MATCH_CASE
874 * CM !norm | upper | ZCILOOK | MT_NORMALIZE
875 * CM norm | upper|formX | !ZCILOOK | MT_NORMALIZE | MT_MATCH_CASE
876 * CM norm | upper|formX | ZCILOOK | MT_NORMALIZE
879 * CS = Case Sensitive, CI = Case Insensitive, CM = Case Mixed
880 * upper = case folding set by fs type on creation (U8_TEXTPREP_TOUPPER)
881 * formX = unicode normalization form set on fs creation
884 zfs_dropname(zfs_dirlock_t
*dl
, znode_t
*zp
, znode_t
*dzp
, dmu_tx_t
*tx
,
889 if (ZTOZSB(zp
)->z_norm
) {
890 matchtype_t mt
= MT_NORMALIZE
;
892 if ((ZTOZSB(zp
)->z_case
== ZFS_CASE_INSENSITIVE
&&
893 (flag
& ZCIEXACT
)) ||
894 (ZTOZSB(zp
)->z_case
== ZFS_CASE_MIXED
&&
895 !(flag
& ZCILOOK
))) {
899 error
= zap_remove_norm(ZTOZSB(zp
)->z_os
, dzp
->z_id
,
900 dl
->dl_name
, mt
, tx
);
902 error
= zap_remove(ZTOZSB(zp
)->z_os
, dzp
->z_id
, dl
->dl_name
,
910 * Unlink zp from dl, and mark zp for deletion if this was the last link. Can
911 * fail if zp is a mount point (EBUSY) or a non-empty directory (ENOTEMPTY).
912 * If 'unlinkedp' is NULL, we put unlinked znodes on the unlinked list.
913 * If it's non-NULL, we use it to indicate whether the znode needs deletion,
914 * and it's the caller's job to do it.
917 zfs_link_destroy(zfs_dirlock_t
*dl
, znode_t
*zp
, dmu_tx_t
*tx
, int flag
,
918 boolean_t
*unlinkedp
)
920 znode_t
*dzp
= dl
->dl_dzp
;
921 zfsvfs_t
*zfsvfs
= ZTOZSB(dzp
);
922 int zp_is_dir
= S_ISDIR(ZTOI(zp
)->i_mode
);
923 boolean_t unlinked
= B_FALSE
;
924 sa_bulk_attr_t bulk
[5];
925 uint64_t mtime
[2], ctime
[2];
930 if (!(flag
& ZRENAMING
)) {
931 mutex_enter(&zp
->z_lock
);
933 if (zp_is_dir
&& !zfs_dirempty(zp
)) {
934 mutex_exit(&zp
->z_lock
);
935 return (SET_ERROR(ENOTEMPTY
));
939 * If we get here, we are going to try to remove the object.
940 * First try removing the name from the directory; if that
941 * fails, return the error.
943 error
= zfs_dropname(dl
, zp
, dzp
, tx
, flag
);
945 mutex_exit(&zp
->z_lock
);
949 if (ZTOI(zp
)->i_nlink
<= zp_is_dir
) {
950 zfs_panic_recover("zfs: link count on %lu is %u, "
951 "should be at least %u", zp
->z_id
,
952 (int)ZTOI(zp
)->i_nlink
, zp_is_dir
+ 1);
953 set_nlink(ZTOI(zp
), zp_is_dir
+ 1);
955 drop_nlink(ZTOI(zp
));
956 if (ZTOI(zp
)->i_nlink
== zp_is_dir
) {
957 zp
->z_unlinked
= B_TRUE
;
958 clear_nlink(ZTOI(zp
));
961 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
),
962 NULL
, &ctime
, sizeof (ctime
));
963 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
),
964 NULL
, &zp
->z_pflags
, sizeof (zp
->z_pflags
));
965 zfs_tstamp_update_setup(zp
, STATE_CHANGED
, mtime
,
968 links
= ZTOI(zp
)->i_nlink
;
969 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_LINKS(zfsvfs
),
970 NULL
, &links
, sizeof (links
));
971 error
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
974 mutex_exit(&zp
->z_lock
);
976 error
= zfs_dropname(dl
, zp
, dzp
, tx
, flag
);
981 mutex_enter(&dzp
->z_lock
);
982 dzp
->z_size
--; /* one dirent removed */
984 drop_nlink(ZTOI(dzp
)); /* ".." link from zp */
985 links
= ZTOI(dzp
)->i_nlink
;
986 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_LINKS(zfsvfs
),
987 NULL
, &links
, sizeof (links
));
988 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_SIZE(zfsvfs
),
989 NULL
, &dzp
->z_size
, sizeof (dzp
->z_size
));
990 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
),
991 NULL
, ctime
, sizeof (ctime
));
992 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
),
993 NULL
, mtime
, sizeof (mtime
));
994 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
),
995 NULL
, &dzp
->z_pflags
, sizeof (dzp
->z_pflags
));
996 zfs_tstamp_update_setup(dzp
, CONTENT_MODIFIED
, mtime
, ctime
);
997 error
= sa_bulk_update(dzp
->z_sa_hdl
, bulk
, count
, tx
);
999 mutex_exit(&dzp
->z_lock
);
1001 if (unlinkedp
!= NULL
)
1002 *unlinkedp
= unlinked
;
1004 zfs_unlinked_add(zp
, tx
);
1010 * Indicate whether the directory is empty. Works with or without z_lock
1011 * held, but can only be consider a hint in the latter case. Returns true
1012 * if only "." and ".." remain and there's no work in progress.
1014 * The internal ZAP size, rather than zp->z_size, needs to be checked since
1015 * some consumers (Lustre) do not strictly maintain an accurate SA_ZPL_SIZE.
1018 zfs_dirempty(znode_t
*dzp
)
1020 zfsvfs_t
*zfsvfs
= ZTOZSB(dzp
);
1024 if (dzp
->z_dirlocks
!= NULL
)
1027 error
= zap_count(zfsvfs
->z_os
, dzp
->z_id
, &count
);
1028 if (error
!= 0 || count
!= 0)
1035 zfs_make_xattrdir(znode_t
*zp
, vattr_t
*vap
, struct inode
**xipp
, cred_t
*cr
)
1037 zfsvfs_t
*zfsvfs
= ZTOZSB(zp
);
1041 zfs_acl_ids_t acl_ids
;
1042 boolean_t fuid_dirtied
;
1049 if ((error
= zfs_zaccess(zp
, ACE_WRITE_NAMED_ATTRS
, 0, B_FALSE
, cr
)))
1052 if ((error
= zfs_acl_ids_create(zp
, IS_XATTR
, vap
, cr
, NULL
,
1055 if (zfs_acl_ids_overquota(zfsvfs
, &acl_ids
, zp
->z_projid
)) {
1056 zfs_acl_ids_free(&acl_ids
);
1057 return (SET_ERROR(EDQUOT
));
1060 tx
= dmu_tx_create(zfsvfs
->z_os
);
1061 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
1062 ZFS_SA_BASE_ATTR_SIZE
);
1063 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
1064 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, FALSE
, NULL
);
1065 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
1067 zfs_fuid_txhold(zfsvfs
, tx
);
1068 error
= dmu_tx_assign(tx
, TXG_WAIT
);
1070 zfs_acl_ids_free(&acl_ids
);
1074 zfs_mknode(zp
, vap
, tx
, cr
, IS_XATTR
, &xzp
, &acl_ids
);
1077 zfs_fuid_sync(zfsvfs
, tx
);
1080 error
= sa_lookup(xzp
->z_sa_hdl
, SA_ZPL_PARENT(zfsvfs
),
1081 &parent
, sizeof (parent
));
1082 ASSERT(error
== 0 && parent
== zp
->z_id
);
1085 VERIFY(0 == sa_update(zp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
), &xzp
->z_id
,
1086 sizeof (xzp
->z_id
), tx
));
1088 if (!zp
->z_unlinked
)
1089 (void) zfs_log_create(zfsvfs
->z_log
, tx
, TX_MKXATTR
, zp
,
1090 xzp
, "", NULL
, acl_ids
.z_fuidp
, vap
);
1092 zfs_acl_ids_free(&acl_ids
);
1101 * Return a znode for the extended attribute directory for zp.
1102 * ** If the directory does not already exist, it is created **
1104 * IN: zp - znode to obtain attribute directory from
1105 * cr - credentials of caller
1106 * flags - flags from the VOP_LOOKUP call
1108 * OUT: xipp - pointer to extended attribute znode
1110 * RETURN: 0 on success
1111 * error number on failure
1114 zfs_get_xattrdir(znode_t
*zp
, struct inode
**xipp
, cred_t
*cr
, int flags
)
1116 zfsvfs_t
*zfsvfs
= ZTOZSB(zp
);
1122 error
= zfs_dirent_lock(&dl
, zp
, "", &xzp
, ZXATTR
, NULL
, NULL
);
1128 zfs_dirent_unlock(dl
);
1132 if (!(flags
& CREATE_XATTR_DIR
)) {
1133 zfs_dirent_unlock(dl
);
1134 return (SET_ERROR(ENOENT
));
1137 if (zfs_is_readonly(zfsvfs
)) {
1138 zfs_dirent_unlock(dl
);
1139 return (SET_ERROR(EROFS
));
1143 * The ability to 'create' files in an attribute
1144 * directory comes from the write_xattr permission on the base file.
1146 * The ability to 'search' an attribute directory requires
1147 * read_xattr permission on the base file.
1149 * Once in a directory the ability to read/write attributes
1150 * is controlled by the permissions on the attribute file.
1152 va
.va_mask
= ATTR_MODE
| ATTR_UID
| ATTR_GID
;
1153 va
.va_mode
= S_IFDIR
| S_ISVTX
| 0777;
1154 zfs_fuid_map_ids(zp
, cr
, &va
.va_uid
, &va
.va_gid
);
1156 va
.va_dentry
= NULL
;
1157 error
= zfs_make_xattrdir(zp
, &va
, xipp
, cr
);
1158 zfs_dirent_unlock(dl
);
1160 if (error
== ERESTART
) {
1161 /* NB: we already did dmu_tx_wait() if necessary */
1169 * Decide whether it is okay to remove within a sticky directory.
1171 * In sticky directories, write access is not sufficient;
1172 * you can remove entries from a directory only if:
1174 * you own the directory,
1175 * you own the entry,
1176 * you have write access to the entry,
1177 * or you are privileged (checked in secpolicy...).
1179 * The function returns 0 if remove access is granted.
1182 zfs_sticky_remove_access(znode_t
*zdp
, znode_t
*zp
, cred_t
*cr
)
1187 zfsvfs_t
*zfsvfs
= ZTOZSB(zdp
);
1189 if (zfsvfs
->z_replay
)
1192 if ((zdp
->z_mode
& S_ISVTX
) == 0)
1195 downer
= zfs_fuid_map_id(zfsvfs
, KUID_TO_SUID(ZTOI(zdp
)->i_uid
),
1197 fowner
= zfs_fuid_map_id(zfsvfs
, KUID_TO_SUID(ZTOI(zp
)->i_uid
),
1200 if ((uid
= crgetuid(cr
)) == downer
|| uid
== fowner
||
1201 zfs_zaccess(zp
, ACE_WRITE_DATA
, 0, B_FALSE
, cr
) == 0)
1204 return (secpolicy_vnode_remove(cr
));