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
);
103 if (error
== ENOENT
&& update
)
104 dnlc_update(ZTOI(dzp
), name
, DNLC_NO_VNODE
);
105 #endif /* HAVE_DNLC */
111 * Lock a directory entry. A dirlock on <dzp, name> protects that name
112 * in dzp's directory zap object. As long as you hold a dirlock, you can
113 * assume two things: (1) dzp cannot be reaped, and (2) no other thread
114 * can change the zap entry for (i.e. link or unlink) this name.
117 * dzp - znode for directory
118 * name - name of entry to lock
119 * flag - ZNEW: if the entry already exists, fail with EEXIST.
120 * ZEXISTS: if the entry does not exist, fail with ENOENT.
121 * ZSHARED: allow concurrent access with other ZSHARED callers.
122 * ZXATTR: we want dzp's xattr directory
123 * ZCILOOK: On a mixed sensitivity file system,
124 * this lookup should be case-insensitive.
125 * ZCIEXACT: On a purely case-insensitive file system,
126 * this lookup should be case-sensitive.
127 * ZRENAMING: we are locking for renaming, force narrow locks
128 * ZHAVELOCK: Don't grab the z_name_lock for this call. The
129 * current thread already holds it.
132 * zpp - pointer to the znode for the entry (NULL if there isn't one)
133 * dlpp - pointer to the dirlock for this entry (NULL on error)
134 * direntflags - (case-insensitive lookup only)
135 * flags if multiple case-sensitive matches exist in directory
136 * realpnp - (case-insensitive lookup only)
137 * actual name matched within the directory
139 * Return value: 0 on success or errno on failure.
141 * NOTE: Always checks for, and rejects, '.' and '..'.
142 * NOTE: For case-insensitive file systems we take wide locks (see below),
143 * but return znode pointers to a single match.
146 zfs_dirent_lock(zfs_dirlock_t
**dlpp
, znode_t
*dzp
, char *name
, znode_t
**zpp
,
147 int flag
, int *direntflags
, pathname_t
*realpnp
)
149 zfsvfs_t
*zfsvfs
= ZTOZSB(dzp
);
156 #endif /* HAVE_DNLC */
164 * Verify that we are not trying to lock '.', '..', or '.zfs'
166 if ((name
[0] == '.' &&
167 (name
[1] == '\0' || (name
[1] == '.' && name
[2] == '\0'))) ||
168 (zfs_has_ctldir(dzp
) && strcmp(name
, ZFS_CTLDIR_NAME
) == 0))
169 return (SET_ERROR(EEXIST
));
172 * Case sensitivity and normalization preferences are set when
173 * the file system is created. These are stored in the
174 * zfsvfs->z_case and zfsvfs->z_norm fields. These choices
175 * affect what vnodes can be cached in the DNLC, how we
176 * perform zap lookups, and the "width" of our dirlocks.
178 * A normal dirlock locks a single name. Note that with
179 * normalization a name can be composed multiple ways, but
180 * when normalized, these names all compare equal. A wide
181 * dirlock locks multiple names. We need these when the file
182 * system is supporting mixed-mode access. It is sometimes
183 * necessary to lock all case permutations of file name at
184 * once so that simultaneous case-insensitive/case-sensitive
185 * behaves as rationally as possible.
189 * When matching we may need to normalize & change case according to
192 * Note that a normalized match is necessary for a case insensitive
193 * filesystem when the lookup request is not exact because normalization
194 * can fold case independent of normalizing code point sequences.
196 * See the table above zfs_dropname().
198 if (zfsvfs
->z_norm
!= 0) {
202 * Determine if the match needs to honor the case specified in
203 * lookup, and if so keep track of that so that during
204 * normalization we don't fold case.
206 if ((zfsvfs
->z_case
== ZFS_CASE_INSENSITIVE
&&
207 (flag
& ZCIEXACT
)) ||
208 (zfsvfs
->z_case
== ZFS_CASE_MIXED
&& !(flag
& ZCILOOK
))) {
214 * Only look in or update the DNLC if we are looking for the
215 * name on a file system that does not require normalization
216 * or case folding. We can also look there if we happen to be
217 * on a non-normalizing, mixed sensitivity file system IF we
218 * are looking for the exact name.
220 * Maybe can add TO-UPPERed version of name to dnlc in ci-only
221 * case for performance improvement?
223 update
= !zfsvfs
->z_norm
||
224 (zfsvfs
->z_case
== ZFS_CASE_MIXED
&&
225 !(zfsvfs
->z_norm
& ~U8_TEXTPREP_TOUPPER
) && !(flag
& ZCILOOK
));
228 * ZRENAMING indicates we are in a situation where we should
229 * take narrow locks regardless of the file system's
230 * preferences for normalizing and case folding. This will
231 * prevent us deadlocking trying to grab the same wide lock
232 * twice if the two names happen to be case-insensitive
235 if (flag
& ZRENAMING
)
238 cmpflags
= zfsvfs
->z_norm
;
241 * Wait until there are no locks on this name.
243 * Don't grab the the lock if it is already held. However, cannot
244 * have both ZSHARED and ZHAVELOCK together.
246 ASSERT(!(flag
& ZSHARED
) || !(flag
& ZHAVELOCK
));
247 if (!(flag
& ZHAVELOCK
))
248 rw_enter(&dzp
->z_name_lock
, RW_READER
);
250 mutex_enter(&dzp
->z_lock
);
252 if (dzp
->z_unlinked
&& !(flag
& ZXATTR
)) {
253 mutex_exit(&dzp
->z_lock
);
254 if (!(flag
& ZHAVELOCK
))
255 rw_exit(&dzp
->z_name_lock
);
256 return (SET_ERROR(ENOENT
));
258 for (dl
= dzp
->z_dirlocks
; dl
!= NULL
; dl
= dl
->dl_next
) {
259 if ((u8_strcmp(name
, dl
->dl_name
, 0, cmpflags
,
260 U8_UNICODE_LATEST
, &error
) == 0) || error
!= 0)
264 mutex_exit(&dzp
->z_lock
);
265 if (!(flag
& ZHAVELOCK
))
266 rw_exit(&dzp
->z_name_lock
);
267 return (SET_ERROR(ENOENT
));
271 * Allocate a new dirlock and add it to the list.
273 dl
= kmem_alloc(sizeof (zfs_dirlock_t
), KM_SLEEP
);
274 cv_init(&dl
->dl_cv
, NULL
, CV_DEFAULT
, NULL
);
280 dl
->dl_next
= dzp
->z_dirlocks
;
281 dzp
->z_dirlocks
= dl
;
284 if ((flag
& ZSHARED
) && dl
->dl_sharecnt
!= 0)
286 cv_wait(&dl
->dl_cv
, &dzp
->z_lock
);
290 * If the z_name_lock was NOT held for this dirlock record it.
292 if (flag
& ZHAVELOCK
)
295 if ((flag
& ZSHARED
) && ++dl
->dl_sharecnt
> 1 && dl
->dl_namesize
== 0) {
297 * We're the second shared reference to dl. Make a copy of
298 * dl_name in case the first thread goes away before we do.
299 * Note that we initialize the new name before storing its
300 * pointer into dl_name, because the first thread may load
301 * dl->dl_name at any time. It'll either see the old value,
302 * which belongs to it, or the new shared copy; either is OK.
304 dl
->dl_namesize
= strlen(dl
->dl_name
) + 1;
305 name
= kmem_alloc(dl
->dl_namesize
, KM_SLEEP
);
306 bcopy(dl
->dl_name
, name
, dl
->dl_namesize
);
310 mutex_exit(&dzp
->z_lock
);
313 * We have a dirlock on the name. (Note that it is the dirlock,
314 * not the dzp's z_lock, that protects the name in the zap object.)
315 * See if there's an object by this name; if so, put a hold on it.
318 error
= sa_lookup(dzp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
), &zoid
,
321 error
= (zoid
== 0 ? SET_ERROR(ENOENT
) : 0);
325 vp
= dnlc_lookup(ZTOI(dzp
), name
);
326 if (vp
== DNLC_NO_VNODE
) {
328 error
= SET_ERROR(ENOENT
);
331 zfs_dirent_unlock(dl
);
333 return (SET_ERROR(EEXIST
));
339 error
= zfs_match_find(zfsvfs
, dzp
, name
, mt
,
340 update
, direntflags
, realpnp
, &zoid
);
343 error
= zfs_match_find(zfsvfs
, dzp
, name
, mt
,
344 update
, direntflags
, realpnp
, &zoid
);
345 #endif /* HAVE_DNLC */
348 if (error
!= ENOENT
|| (flag
& ZEXISTS
)) {
349 zfs_dirent_unlock(dl
);
354 zfs_dirent_unlock(dl
);
355 return (SET_ERROR(EEXIST
));
357 error
= zfs_zget(zfsvfs
, zoid
, zpp
);
359 zfs_dirent_unlock(dl
);
363 if (!(flag
& ZXATTR
) && update
)
364 dnlc_update(ZTOI(dzp
), name
, ZTOI(*zpp
));
365 #endif /* HAVE_DNLC */
374 * Unlock this directory entry and wake anyone who was waiting for it.
377 zfs_dirent_unlock(zfs_dirlock_t
*dl
)
379 znode_t
*dzp
= dl
->dl_dzp
;
380 zfs_dirlock_t
**prev_dl
, *cur_dl
;
382 mutex_enter(&dzp
->z_lock
);
384 if (!dl
->dl_namelock
)
385 rw_exit(&dzp
->z_name_lock
);
387 if (dl
->dl_sharecnt
> 1) {
389 mutex_exit(&dzp
->z_lock
);
392 prev_dl
= &dzp
->z_dirlocks
;
393 while ((cur_dl
= *prev_dl
) != dl
)
394 prev_dl
= &cur_dl
->dl_next
;
395 *prev_dl
= dl
->dl_next
;
396 cv_broadcast(&dl
->dl_cv
);
397 mutex_exit(&dzp
->z_lock
);
399 if (dl
->dl_namesize
!= 0)
400 kmem_free(dl
->dl_name
, dl
->dl_namesize
);
401 cv_destroy(&dl
->dl_cv
);
402 kmem_free(dl
, sizeof (*dl
));
406 * Look up an entry in a directory.
408 * NOTE: '.' and '..' are handled as special cases because
409 * no directory entries are actually stored for them. If this is
410 * the root of a filesystem, then '.zfs' is also treated as a
411 * special pseudo-directory.
414 zfs_dirlook(znode_t
*dzp
, char *name
, struct inode
**ipp
, int flags
,
415 int *deflg
, pathname_t
*rpnp
)
422 if (name
[0] == 0 || (name
[0] == '.' && name
[1] == 0)) {
425 } else if (name
[0] == '.' && name
[1] == '.' && name
[2] == 0) {
426 zfsvfs_t
*zfsvfs
= ZTOZSB(dzp
);
429 * If we are a snapshot mounted under .zfs, return
430 * the inode pointer for the snapshot directory.
432 if ((error
= sa_lookup(dzp
->z_sa_hdl
,
433 SA_ZPL_PARENT(zfsvfs
), &parent
, sizeof (parent
))) != 0)
436 if (parent
== dzp
->z_id
&& zfsvfs
->z_parent
!= zfsvfs
) {
437 error
= zfsctl_root_lookup(zfsvfs
->z_parent
->z_ctldir
,
438 "snapshot", ipp
, 0, kcred
, NULL
, NULL
);
441 rw_enter(&dzp
->z_parent_lock
, RW_READER
);
442 error
= zfs_zget(zfsvfs
, parent
, &zp
);
445 rw_exit(&dzp
->z_parent_lock
);
446 } else if (zfs_has_ctldir(dzp
) && strcmp(name
, ZFS_CTLDIR_NAME
) == 0) {
447 *ipp
= zfsctl_root(dzp
);
451 zf
= ZEXISTS
| ZSHARED
;
452 if (flags
& FIGNORECASE
)
455 error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zf
, deflg
, rpnp
);
458 zfs_dirent_unlock(dl
);
459 dzp
->z_zn_prefetch
= B_TRUE
; /* enable prefetching */
464 if ((flags
& FIGNORECASE
) && rpnp
&& !error
)
465 (void) strlcpy(rpnp
->pn_buf
, name
, rpnp
->pn_bufsize
);
471 * unlinked Set (formerly known as the "delete queue") Error Handling
473 * When dealing with the unlinked set, we dmu_tx_hold_zap(), but we
474 * don't specify the name of the entry that we will be manipulating. We
475 * also fib and say that we won't be adding any new entries to the
476 * unlinked set, even though we might (this is to lower the minimum file
477 * size that can be deleted in a full filesystem). So on the small
478 * chance that the nlink list is using a fat zap (ie. has more than
479 * 2000 entries), we *may* not pre-read a block that's needed.
480 * Therefore it is remotely possible for some of the assertions
481 * regarding the unlinked set below to fail due to i/o error. On a
482 * nondebug system, this will result in the space being leaked.
485 zfs_unlinked_add(znode_t
*zp
, dmu_tx_t
*tx
)
487 zfsvfs_t
*zfsvfs
= ZTOZSB(zp
);
489 ASSERT(zp
->z_unlinked
);
490 ASSERT(ZTOI(zp
)->i_nlink
== 0);
493 zap_add_int(zfsvfs
->z_os
, zfsvfs
->z_unlinkedobj
, zp
->z_id
, tx
));
497 * Clean up any znodes that had no links when we either crashed or
498 * (force) umounted the file system.
501 zfs_unlinked_drain(zfsvfs_t
*zfsvfs
)
505 dmu_object_info_t doi
;
510 * Iterate over the contents of the unlinked set.
512 for (zap_cursor_init(&zc
, zfsvfs
->z_os
, zfsvfs
->z_unlinkedobj
);
513 zap_cursor_retrieve(&zc
, &zap
) == 0;
514 zap_cursor_advance(&zc
)) {
517 * See what kind of object we have in list
520 error
= dmu_object_info(zfsvfs
->z_os
,
521 zap
.za_first_integer
, &doi
);
525 ASSERT((doi
.doi_type
== DMU_OT_PLAIN_FILE_CONTENTS
) ||
526 (doi
.doi_type
== DMU_OT_DIRECTORY_CONTENTS
));
528 * We need to re-mark these list entries for deletion,
529 * so we pull them back into core and set zp->z_unlinked.
531 error
= zfs_zget(zfsvfs
, zap
.za_first_integer
, &zp
);
534 * We may pick up znodes that are already marked for deletion.
535 * This could happen during the purge of an extended attribute
536 * directory. All we need to do is skip over them, since they
537 * are already in the system marked z_unlinked.
542 zp
->z_unlinked
= B_TRUE
;
545 zap_cursor_fini(&zc
);
549 * Delete the entire contents of a directory. Return a count
550 * of the number of entries that could not be deleted. If we encounter
551 * an error, return a count of at least one so that the directory stays
552 * in the unlinked set.
554 * NOTE: this function assumes that the directory is inactive,
555 * so there is no need to lock its entries before deletion.
556 * Also, it assumes the directory contents is *only* regular
560 zfs_purgedir(znode_t
*dzp
)
566 zfsvfs_t
*zfsvfs
= ZTOZSB(dzp
);
571 for (zap_cursor_init(&zc
, zfsvfs
->z_os
, dzp
->z_id
);
572 (error
= zap_cursor_retrieve(&zc
, &zap
)) == 0;
573 zap_cursor_advance(&zc
)) {
574 error
= zfs_zget(zfsvfs
,
575 ZFS_DIRENT_OBJ(zap
.za_first_integer
), &xzp
);
581 ASSERT(S_ISREG(ZTOI(xzp
)->i_mode
) ||
582 S_ISLNK(ZTOI(xzp
)->i_mode
));
584 tx
= dmu_tx_create(zfsvfs
->z_os
);
585 dmu_tx_hold_sa(tx
, dzp
->z_sa_hdl
, B_FALSE
);
586 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, zap
.za_name
);
587 dmu_tx_hold_sa(tx
, xzp
->z_sa_hdl
, B_FALSE
);
588 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
589 /* Is this really needed ? */
590 zfs_sa_upgrade_txholds(tx
, xzp
);
591 dmu_tx_mark_netfree(tx
);
592 error
= dmu_tx_assign(tx
, TXG_WAIT
);
595 zfs_iput_async(ZTOI(xzp
));
599 bzero(&dl
, sizeof (dl
));
601 dl
.dl_name
= zap
.za_name
;
603 error
= zfs_link_destroy(&dl
, xzp
, tx
, 0, NULL
);
608 zfs_iput_async(ZTOI(xzp
));
610 zap_cursor_fini(&zc
);
617 zfs_rmnode(znode_t
*zp
)
619 zfsvfs_t
*zfsvfs
= ZTOZSB(zp
);
620 objset_t
*os
= zfsvfs
->z_os
;
628 ASSERT(ZTOI(zp
)->i_nlink
== 0);
629 ASSERT(atomic_read(&ZTOI(zp
)->i_count
) == 0);
632 * If this is an attribute directory, purge its contents.
634 if (S_ISDIR(ZTOI(zp
)->i_mode
) && (zp
->z_pflags
& ZFS_XATTR
)) {
635 if (zfs_purgedir(zp
) != 0) {
637 * Not enough space to delete some xattrs.
638 * Leave it in the unlinked set.
640 zfs_znode_dmu_fini(zp
);
647 * Free up all the data in the file. We don't do this for directories
648 * because we need truncate and remove to be in the same tx, like in
649 * zfs_znode_delete(). Otherwise, if we crash here we'll end up with
650 * an inconsistent truncated zap object in the delete queue. Note a
651 * truncated file is harmless since it only contains user data.
653 if (S_ISREG(ZTOI(zp
)->i_mode
)) {
654 error
= dmu_free_long_range(os
, zp
->z_id
, 0, DMU_OBJECT_END
);
657 * Not enough space or we were interrupted by unmount.
658 * Leave the file in the unlinked set.
660 zfs_znode_dmu_fini(zp
);
666 * If the file has extended attributes, we're going to unlink
669 error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
),
670 &xattr_obj
, sizeof (xattr_obj
));
671 if (error
== 0 && xattr_obj
) {
672 error
= zfs_zget(zfsvfs
, xattr_obj
, &xzp
);
676 acl_obj
= zfs_external_acl(zp
);
679 * Set up the final transaction.
681 tx
= dmu_tx_create(os
);
682 dmu_tx_hold_free(tx
, zp
->z_id
, 0, DMU_OBJECT_END
);
683 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
685 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, TRUE
, NULL
);
686 dmu_tx_hold_sa(tx
, xzp
->z_sa_hdl
, B_FALSE
);
689 dmu_tx_hold_free(tx
, acl_obj
, 0, DMU_OBJECT_END
);
691 zfs_sa_upgrade_txholds(tx
, zp
);
692 error
= dmu_tx_assign(tx
, TXG_WAIT
);
695 * Not enough space to delete the file. Leave it in the
696 * unlinked set, leaking it until the fs is remounted (at
697 * which point we'll call zfs_unlinked_drain() to process it).
700 zfs_znode_dmu_fini(zp
);
706 mutex_enter(&xzp
->z_lock
);
707 xzp
->z_unlinked
= B_TRUE
; /* mark xzp for deletion */
708 clear_nlink(ZTOI(xzp
)); /* no more links to it */
710 VERIFY(0 == sa_update(xzp
->z_sa_hdl
, SA_ZPL_LINKS(zfsvfs
),
711 &links
, sizeof (links
), tx
));
712 mutex_exit(&xzp
->z_lock
);
713 zfs_unlinked_add(xzp
, tx
);
716 /* Remove this znode from the unlinked set */
718 zap_remove_int(zfsvfs
->z_os
, zfsvfs
->z_unlinkedobj
, zp
->z_id
, tx
));
720 zfs_znode_delete(zp
, tx
);
725 zfs_iput_async(ZTOI(xzp
));
729 zfs_dirent(znode_t
*zp
, uint64_t mode
)
731 uint64_t de
= zp
->z_id
;
733 if (ZTOZSB(zp
)->z_version
>= ZPL_VERSION_DIRENT_TYPE
)
734 de
|= IFTODT(mode
) << 60;
739 * Link zp into dl. Can fail in the following cases :
740 * - if zp has been unlinked.
741 * - if the number of entries with the same hash (aka. colliding entries)
742 * exceed the capacity of a leaf-block of fatzap and splitting of the
743 * leaf-block does not help.
746 zfs_link_create(zfs_dirlock_t
*dl
, znode_t
*zp
, dmu_tx_t
*tx
, int flag
)
748 znode_t
*dzp
= dl
->dl_dzp
;
749 zfsvfs_t
*zfsvfs
= ZTOZSB(zp
);
751 int zp_is_dir
= S_ISDIR(ZTOI(zp
)->i_mode
);
752 sa_bulk_attr_t bulk
[5];
753 uint64_t mtime
[2], ctime
[2];
758 mutex_enter(&zp
->z_lock
);
760 if (!(flag
& ZRENAMING
)) {
761 if (zp
->z_unlinked
) { /* no new links to unlinked zp */
762 ASSERT(!(flag
& (ZNEW
| ZEXISTS
)));
763 mutex_exit(&zp
->z_lock
);
764 return (SET_ERROR(ENOENT
));
766 if (!(flag
& ZNEW
)) {
768 * ZNEW nodes come from zfs_mknode() where the link
769 * count has already been initialised
772 links
= ZTOI(zp
)->i_nlink
;
773 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_LINKS(zfsvfs
),
774 NULL
, &links
, sizeof (links
));
778 value
= zfs_dirent(zp
, zp
->z_mode
);
779 error
= zap_add(ZTOZSB(zp
)->z_os
, dzp
->z_id
, dl
->dl_name
, 8, 1,
783 * zap_add could fail to add the entry if it exceeds the capacity of the
784 * leaf-block and zap_leaf_split() failed to help.
785 * The caller of this routine is responsible for failing the transaction
786 * which will rollback the SA updates done above.
789 if (!(flag
& ZRENAMING
) && !(flag
& ZNEW
))
790 drop_nlink(ZTOI(zp
));
791 mutex_exit(&zp
->z_lock
);
795 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_PARENT(zfsvfs
), NULL
,
796 &dzp
->z_id
, sizeof (dzp
->z_id
));
797 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
), NULL
,
798 &zp
->z_pflags
, sizeof (zp
->z_pflags
));
800 if (!(flag
& ZNEW
)) {
801 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
,
802 ctime
, sizeof (ctime
));
803 zfs_tstamp_update_setup(zp
, STATE_CHANGED
, mtime
,
806 error
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
809 mutex_exit(&zp
->z_lock
);
811 mutex_enter(&dzp
->z_lock
);
814 inc_nlink(ZTOI(dzp
));
815 links
= ZTOI(dzp
)->i_nlink
;
817 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_SIZE(zfsvfs
), NULL
,
818 &dzp
->z_size
, sizeof (dzp
->z_size
));
819 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_LINKS(zfsvfs
), NULL
,
820 &links
, sizeof (links
));
821 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
,
822 mtime
, sizeof (mtime
));
823 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
,
824 ctime
, sizeof (ctime
));
825 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
), NULL
,
826 &dzp
->z_pflags
, sizeof (dzp
->z_pflags
));
827 zfs_tstamp_update_setup(dzp
, CONTENT_MODIFIED
, mtime
, ctime
);
828 error
= sa_bulk_update(dzp
->z_sa_hdl
, bulk
, count
, tx
);
830 mutex_exit(&dzp
->z_lock
);
836 * The match type in the code for this function should conform to:
838 * ------------------------------------------------------------------------
839 * fs type | z_norm | lookup type | match type
840 * ---------|-------------|-------------|----------------------------------
841 * CS !norm | 0 | 0 | 0 (exact)
842 * CS norm | formX | 0 | MT_NORMALIZE
843 * CI !norm | upper | !ZCIEXACT | MT_NORMALIZE
844 * CI !norm | upper | ZCIEXACT | MT_NORMALIZE | MT_MATCH_CASE
845 * CI norm | upper|formX | !ZCIEXACT | MT_NORMALIZE
846 * CI norm | upper|formX | ZCIEXACT | MT_NORMALIZE | MT_MATCH_CASE
847 * CM !norm | upper | !ZCILOOK | MT_NORMALIZE | MT_MATCH_CASE
848 * CM !norm | upper | ZCILOOK | MT_NORMALIZE
849 * CM norm | upper|formX | !ZCILOOK | MT_NORMALIZE | MT_MATCH_CASE
850 * CM norm | upper|formX | ZCILOOK | MT_NORMALIZE
853 * CS = Case Sensitive, CI = Case Insensitive, CM = Case Mixed
854 * upper = case folding set by fs type on creation (U8_TEXTPREP_TOUPPER)
855 * formX = unicode normalization form set on fs creation
858 zfs_dropname(zfs_dirlock_t
*dl
, znode_t
*zp
, znode_t
*dzp
, dmu_tx_t
*tx
,
863 if (ZTOZSB(zp
)->z_norm
) {
864 matchtype_t mt
= MT_NORMALIZE
;
866 if ((ZTOZSB(zp
)->z_case
== ZFS_CASE_INSENSITIVE
&&
867 (flag
& ZCIEXACT
)) ||
868 (ZTOZSB(zp
)->z_case
== ZFS_CASE_MIXED
&&
869 !(flag
& ZCILOOK
))) {
873 error
= zap_remove_norm(ZTOZSB(zp
)->z_os
, dzp
->z_id
,
874 dl
->dl_name
, mt
, tx
);
876 error
= zap_remove(ZTOZSB(zp
)->z_os
, dzp
->z_id
, dl
->dl_name
,
884 * Unlink zp from dl, and mark zp for deletion if this was the last link. Can
885 * fail if zp is a mount point (EBUSY) or a non-empty directory (ENOTEMPTY).
886 * If 'unlinkedp' is NULL, we put unlinked znodes on the unlinked list.
887 * If it's non-NULL, we use it to indicate whether the znode needs deletion,
888 * and it's the caller's job to do it.
891 zfs_link_destroy(zfs_dirlock_t
*dl
, znode_t
*zp
, dmu_tx_t
*tx
, int flag
,
892 boolean_t
*unlinkedp
)
894 znode_t
*dzp
= dl
->dl_dzp
;
895 zfsvfs_t
*zfsvfs
= ZTOZSB(dzp
);
896 int zp_is_dir
= S_ISDIR(ZTOI(zp
)->i_mode
);
897 boolean_t unlinked
= B_FALSE
;
898 sa_bulk_attr_t bulk
[5];
899 uint64_t mtime
[2], ctime
[2];
905 dnlc_remove(ZTOI(dzp
), dl
->dl_name
);
906 #endif /* HAVE_DNLC */
908 if (!(flag
& ZRENAMING
)) {
909 mutex_enter(&zp
->z_lock
);
911 if (zp_is_dir
&& !zfs_dirempty(zp
)) {
912 mutex_exit(&zp
->z_lock
);
913 return (SET_ERROR(ENOTEMPTY
));
917 * If we get here, we are going to try to remove the object.
918 * First try removing the name from the directory; if that
919 * fails, return the error.
921 error
= zfs_dropname(dl
, zp
, dzp
, tx
, flag
);
923 mutex_exit(&zp
->z_lock
);
927 if (ZTOI(zp
)->i_nlink
<= zp_is_dir
) {
928 zfs_panic_recover("zfs: link count on %lu is %u, "
929 "should be at least %u", zp
->z_id
,
930 (int)ZTOI(zp
)->i_nlink
, zp_is_dir
+ 1);
931 set_nlink(ZTOI(zp
), zp_is_dir
+ 1);
933 drop_nlink(ZTOI(zp
));
934 if (ZTOI(zp
)->i_nlink
== zp_is_dir
) {
935 zp
->z_unlinked
= B_TRUE
;
936 clear_nlink(ZTOI(zp
));
939 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
),
940 NULL
, &ctime
, sizeof (ctime
));
941 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
),
942 NULL
, &zp
->z_pflags
, sizeof (zp
->z_pflags
));
943 zfs_tstamp_update_setup(zp
, STATE_CHANGED
, mtime
,
946 links
= ZTOI(zp
)->i_nlink
;
947 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_LINKS(zfsvfs
),
948 NULL
, &links
, sizeof (links
));
949 error
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
952 mutex_exit(&zp
->z_lock
);
954 error
= zfs_dropname(dl
, zp
, dzp
, tx
, flag
);
959 mutex_enter(&dzp
->z_lock
);
960 dzp
->z_size
--; /* one dirent removed */
962 drop_nlink(ZTOI(dzp
)); /* ".." link from zp */
963 links
= ZTOI(dzp
)->i_nlink
;
964 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_LINKS(zfsvfs
),
965 NULL
, &links
, sizeof (links
));
966 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_SIZE(zfsvfs
),
967 NULL
, &dzp
->z_size
, sizeof (dzp
->z_size
));
968 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
),
969 NULL
, ctime
, sizeof (ctime
));
970 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
),
971 NULL
, mtime
, sizeof (mtime
));
972 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
),
973 NULL
, &dzp
->z_pflags
, sizeof (dzp
->z_pflags
));
974 zfs_tstamp_update_setup(dzp
, CONTENT_MODIFIED
, mtime
, ctime
);
975 error
= sa_bulk_update(dzp
->z_sa_hdl
, bulk
, count
, tx
);
977 mutex_exit(&dzp
->z_lock
);
979 if (unlinkedp
!= NULL
)
980 *unlinkedp
= unlinked
;
982 zfs_unlinked_add(zp
, tx
);
988 * Indicate whether the directory is empty. Works with or without z_lock
989 * held, but can only be consider a hint in the latter case. Returns true
990 * if only "." and ".." remain and there's no work in progress.
992 * The internal ZAP size, rather than zp->z_size, needs to be checked since
993 * some consumers (Lustre) do not strictly maintain an accurate SA_ZPL_SIZE.
996 zfs_dirempty(znode_t
*dzp
)
998 zfsvfs_t
*zfsvfs
= ZTOZSB(dzp
);
1002 if (dzp
->z_dirlocks
!= NULL
)
1005 error
= zap_count(zfsvfs
->z_os
, dzp
->z_id
, &count
);
1006 if (error
!= 0 || count
!= 0)
1013 zfs_make_xattrdir(znode_t
*zp
, vattr_t
*vap
, struct inode
**xipp
, cred_t
*cr
)
1015 zfsvfs_t
*zfsvfs
= ZTOZSB(zp
);
1019 zfs_acl_ids_t acl_ids
;
1020 boolean_t fuid_dirtied
;
1027 if ((error
= zfs_zaccess(zp
, ACE_WRITE_NAMED_ATTRS
, 0, B_FALSE
, cr
)))
1030 if ((error
= zfs_acl_ids_create(zp
, IS_XATTR
, vap
, cr
, NULL
,
1033 if (zfs_acl_ids_overquota(zfsvfs
, &acl_ids
, zp
->z_projid
)) {
1034 zfs_acl_ids_free(&acl_ids
);
1035 return (SET_ERROR(EDQUOT
));
1038 tx
= dmu_tx_create(zfsvfs
->z_os
);
1039 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
1040 ZFS_SA_BASE_ATTR_SIZE
);
1041 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
1042 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, FALSE
, NULL
);
1043 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
1045 zfs_fuid_txhold(zfsvfs
, tx
);
1046 error
= dmu_tx_assign(tx
, TXG_WAIT
);
1048 zfs_acl_ids_free(&acl_ids
);
1052 zfs_mknode(zp
, vap
, tx
, cr
, IS_XATTR
, &xzp
, &acl_ids
);
1055 zfs_fuid_sync(zfsvfs
, tx
);
1058 error
= sa_lookup(xzp
->z_sa_hdl
, SA_ZPL_PARENT(zfsvfs
),
1059 &parent
, sizeof (parent
));
1060 ASSERT(error
== 0 && parent
== zp
->z_id
);
1063 VERIFY(0 == sa_update(zp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
), &xzp
->z_id
,
1064 sizeof (xzp
->z_id
), tx
));
1066 if (!zp
->z_unlinked
)
1067 (void) zfs_log_create(zfsvfs
->z_log
, tx
, TX_MKXATTR
, zp
,
1068 xzp
, "", NULL
, acl_ids
.z_fuidp
, vap
);
1070 zfs_acl_ids_free(&acl_ids
);
1079 * Return a znode for the extended attribute directory for zp.
1080 * ** If the directory does not already exist, it is created **
1082 * IN: zp - znode to obtain attribute directory from
1083 * cr - credentials of caller
1084 * flags - flags from the VOP_LOOKUP call
1086 * OUT: xipp - pointer to extended attribute znode
1088 * RETURN: 0 on success
1089 * error number on failure
1092 zfs_get_xattrdir(znode_t
*zp
, struct inode
**xipp
, cred_t
*cr
, int flags
)
1094 zfsvfs_t
*zfsvfs
= ZTOZSB(zp
);
1100 error
= zfs_dirent_lock(&dl
, zp
, "", &xzp
, ZXATTR
, NULL
, NULL
);
1106 zfs_dirent_unlock(dl
);
1110 if (!(flags
& CREATE_XATTR_DIR
)) {
1111 zfs_dirent_unlock(dl
);
1112 return (SET_ERROR(ENOENT
));
1115 if (zfs_is_readonly(zfsvfs
)) {
1116 zfs_dirent_unlock(dl
);
1117 return (SET_ERROR(EROFS
));
1121 * The ability to 'create' files in an attribute
1122 * directory comes from the write_xattr permission on the base file.
1124 * The ability to 'search' an attribute directory requires
1125 * read_xattr permission on the base file.
1127 * Once in a directory the ability to read/write attributes
1128 * is controlled by the permissions on the attribute file.
1130 va
.va_mask
= ATTR_MODE
| ATTR_UID
| ATTR_GID
;
1131 va
.va_mode
= S_IFDIR
| S_ISVTX
| 0777;
1132 zfs_fuid_map_ids(zp
, cr
, &va
.va_uid
, &va
.va_gid
);
1134 va
.va_dentry
= NULL
;
1135 error
= zfs_make_xattrdir(zp
, &va
, xipp
, cr
);
1136 zfs_dirent_unlock(dl
);
1138 if (error
== ERESTART
) {
1139 /* NB: we already did dmu_tx_wait() if necessary */
1147 * Decide whether it is okay to remove within a sticky directory.
1149 * In sticky directories, write access is not sufficient;
1150 * you can remove entries from a directory only if:
1152 * you own the directory,
1153 * you own the entry,
1154 * you have write access to the entry,
1155 * or you are privileged (checked in secpolicy...).
1157 * The function returns 0 if remove access is granted.
1160 zfs_sticky_remove_access(znode_t
*zdp
, znode_t
*zp
, cred_t
*cr
)
1165 zfsvfs_t
*zfsvfs
= ZTOZSB(zdp
);
1167 if (zfsvfs
->z_replay
)
1170 if ((zdp
->z_mode
& S_ISVTX
) == 0)
1173 downer
= zfs_fuid_map_id(zfsvfs
, KUID_TO_SUID(ZTOI(zdp
)->i_uid
),
1175 fowner
= zfs_fuid_map_id(zfsvfs
, KUID_TO_SUID(ZTOI(zp
)->i_uid
),
1178 if ((uid
= crgetuid(cr
)) == downer
|| uid
== fowner
||
1179 zfs_zaccess(zp
, ACE_WRITE_DATA
, 0, B_FALSE
, cr
) == 0)
1182 return (secpolicy_vnode_remove(cr
));