4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
24 * Copyright (c) 2013 Martin Matuska. All rights reserved.
25 * Copyright (c) 2014 Joyent, Inc. All rights reserved.
26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 * Copyright (c) 2016 Actifio, Inc. All rights reserved.
28 * Copyright (c) 2018, loli10K <ezomori.nozomu@gmail.com>. All rights reserved.
32 #include <sys/dmu_objset.h>
33 #include <sys/dmu_tx.h>
34 #include <sys/dsl_dataset.h>
35 #include <sys/dsl_dir.h>
36 #include <sys/dsl_prop.h>
37 #include <sys/dsl_synctask.h>
38 #include <sys/dsl_deleg.h>
39 #include <sys/dmu_impl.h>
41 #include <sys/spa_impl.h>
42 #include <sys/metaslab.h>
46 #include <sys/sunddi.h>
47 #include <sys/zfeature.h>
48 #include <sys/policy.h>
49 #include <sys/zfs_znode.h>
51 #include "zfs_namecheck.h"
55 * Filesystem and Snapshot Limits
56 * ------------------------------
58 * These limits are used to restrict the number of filesystems and/or snapshots
59 * that can be created at a given level in the tree or below. A typical
60 * use-case is with a delegated dataset where the administrator wants to ensure
61 * that a user within the zone is not creating too many additional filesystems
62 * or snapshots, even though they're not exceeding their space quota.
64 * The filesystem and snapshot counts are stored as extensible properties. This
65 * capability is controlled by a feature flag and must be enabled to be used.
66 * Once enabled, the feature is not active until the first limit is set. At
67 * that point, future operations to create/destroy filesystems or snapshots
68 * will validate and update the counts.
70 * Because the count properties will not exist before the feature is active,
71 * the counts are updated when a limit is first set on an uninitialized
72 * dsl_dir node in the tree (The filesystem/snapshot count on a node includes
73 * all of the nested filesystems/snapshots. Thus, a new leaf node has a
74 * filesystem count of 0 and a snapshot count of 0. Non-existent filesystem and
75 * snapshot count properties on a node indicate uninitialized counts on that
76 * node.) When first setting a limit on an uninitialized node, the code starts
77 * at the filesystem with the new limit and descends into all sub-filesystems
78 * to add the count properties.
80 * In practice this is lightweight since a limit is typically set when the
81 * filesystem is created and thus has no children. Once valid, changing the
82 * limit value won't require a re-traversal since the counts are already valid.
83 * When recursively fixing the counts, if a node with a limit is encountered
84 * during the descent, the counts are known to be valid and there is no need to
85 * descend into that filesystem's children. The counts on filesystems above the
86 * one with the new limit will still be uninitialized, unless a limit is
87 * eventually set on one of those filesystems. The counts are always recursively
88 * updated when a limit is set on a dataset, unless there is already a limit.
89 * When a new limit value is set on a filesystem with an existing limit, it is
90 * possible for the new limit to be less than the current count at that level
91 * since a user who can change the limit is also allowed to exceed the limit.
93 * Once the feature is active, then whenever a filesystem or snapshot is
94 * created, the code recurses up the tree, validating the new count against the
95 * limit at each initialized level. In practice, most levels will not have a
96 * limit set. If there is a limit at any initialized level up the tree, the
97 * check must pass or the creation will fail. Likewise, when a filesystem or
98 * snapshot is destroyed, the counts are recursively adjusted all the way up
99 * the initizized nodes in the tree. Renaming a filesystem into different point
100 * in the tree will first validate, then update the counts on each branch up to
101 * the common ancestor. A receive will also validate the counts and then update
104 * An exception to the above behavior is that the limit is not enforced if the
105 * user has permission to modify the limit. This is primarily so that
106 * recursive snapshots in the global zone always work. We want to prevent a
107 * denial-of-service in which a lower level delegated dataset could max out its
108 * limit and thus block recursive snapshots from being taken in the global zone.
109 * Because of this, it is possible for the snapshot count to be over the limit
110 * and snapshots taken in the global zone could cause a lower level dataset to
111 * hit or exceed its limit. The administrator taking the global zone recursive
112 * snapshot should be aware of this side-effect and behave accordingly.
113 * For consistency, the filesystem limit is also not enforced if the user can
116 * The filesystem and snapshot limits are validated by dsl_fs_ss_limit_check()
117 * and updated by dsl_fs_ss_count_adjust(). A new limit value is setup in
118 * dsl_dir_activate_fs_ss_limit() and the counts are adjusted, if necessary, by
119 * dsl_dir_init_fs_ss_count().
121 * There is a special case when we receive a filesystem that already exists. In
122 * this case a temporary clone name of %X is created (see dmu_recv_begin). We
123 * never update the filesystem counts for temporary clones.
125 * Likewise, we do not update the snapshot counts for temporary snapshots,
126 * such as those created by zfs diff.
129 extern inline dsl_dir_phys_t
*dsl_dir_phys(dsl_dir_t
*dd
);
131 static uint64_t dsl_dir_space_towrite(dsl_dir_t
*dd
);
133 typedef struct ddulrt_arg
{
134 dsl_dir_t
*ddulrta_dd
;
139 dsl_dir_evict_async(void *dbu
)
143 ASSERTV(dsl_pool_t
*dp
= dd
->dd_pool
);
147 for (t
= 0; t
< TXG_SIZE
; t
++) {
148 ASSERT(!txg_list_member(&dp
->dp_dirty_dirs
, dd
, t
));
149 ASSERT(dd
->dd_tempreserved
[t
] == 0);
150 ASSERT(dd
->dd_space_towrite
[t
] == 0);
154 dsl_dir_async_rele(dd
->dd_parent
, dd
);
156 spa_async_close(dd
->dd_pool
->dp_spa
, dd
);
159 mutex_destroy(&dd
->dd_lock
);
160 kmem_free(dd
, sizeof (dsl_dir_t
));
164 dsl_dir_hold_obj(dsl_pool_t
*dp
, uint64_t ddobj
,
165 const char *tail
, void *tag
, dsl_dir_t
**ddp
)
169 dmu_object_info_t doi
;
172 ASSERT(dsl_pool_config_held(dp
));
174 err
= dmu_bonus_hold(dp
->dp_meta_objset
, ddobj
, tag
, &dbuf
);
177 dd
= dmu_buf_get_user(dbuf
);
179 dmu_object_info_from_db(dbuf
, &doi
);
180 ASSERT3U(doi
.doi_bonus_type
, ==, DMU_OT_DSL_DIR
);
181 ASSERT3U(doi
.doi_bonus_size
, >=, sizeof (dsl_dir_phys_t
));
186 dd
= kmem_zalloc(sizeof (dsl_dir_t
), KM_SLEEP
);
187 dd
->dd_object
= ddobj
;
191 if (dsl_dir_is_zapified(dd
) &&
192 zap_contains(dp
->dp_meta_objset
, ddobj
,
193 DD_FIELD_CRYPTO_KEY_OBJ
) == 0) {
194 VERIFY0(zap_lookup(dp
->dp_meta_objset
,
195 ddobj
, DD_FIELD_CRYPTO_KEY_OBJ
,
196 sizeof (uint64_t), 1, &dd
->dd_crypto_obj
));
198 /* check for on-disk format errata */
199 if (dsl_dir_incompatible_encryption_version(dd
)) {
200 dp
->dp_spa
->spa_errata
=
201 ZPOOL_ERRATA_ZOL_6845_ENCRYPTION
;
205 mutex_init(&dd
->dd_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
208 dsl_dir_snap_cmtime_update(dd
);
210 if (dsl_dir_phys(dd
)->dd_parent_obj
) {
211 err
= dsl_dir_hold_obj(dp
,
212 dsl_dir_phys(dd
)->dd_parent_obj
, NULL
, dd
,
220 err
= zap_lookup(dp
->dp_meta_objset
,
221 dsl_dir_phys(dd
->dd_parent
)->
222 dd_child_dir_zapobj
, tail
,
223 sizeof (foundobj
), 1, &foundobj
);
224 ASSERT(err
|| foundobj
== ddobj
);
226 (void) strlcpy(dd
->dd_myname
, tail
,
227 sizeof (dd
->dd_myname
));
229 err
= zap_value_search(dp
->dp_meta_objset
,
230 dsl_dir_phys(dd
->dd_parent
)->
232 ddobj
, 0, dd
->dd_myname
);
237 (void) strcpy(dd
->dd_myname
, spa_name(dp
->dp_spa
));
240 if (dsl_dir_is_clone(dd
)) {
241 dmu_buf_t
*origin_bonus
;
242 dsl_dataset_phys_t
*origin_phys
;
245 * We can't open the origin dataset, because
246 * that would require opening this dsl_dir.
247 * Just look at its phys directly instead.
249 err
= dmu_bonus_hold(dp
->dp_meta_objset
,
250 dsl_dir_phys(dd
)->dd_origin_obj
, FTAG
,
254 origin_phys
= origin_bonus
->db_data
;
256 origin_phys
->ds_creation_txg
;
257 dmu_buf_rele(origin_bonus
, FTAG
);
260 dmu_buf_init_user(&dd
->dd_dbu
, NULL
, dsl_dir_evict_async
,
262 winner
= dmu_buf_set_user_ie(dbuf
, &dd
->dd_dbu
);
263 if (winner
!= NULL
) {
265 dsl_dir_rele(dd
->dd_parent
, dd
);
267 mutex_destroy(&dd
->dd_lock
);
268 kmem_free(dd
, sizeof (dsl_dir_t
));
271 spa_open_ref(dp
->dp_spa
, dd
);
276 * The dsl_dir_t has both open-to-close and instantiate-to-evict
277 * holds on the spa. We need the open-to-close holds because
278 * otherwise the spa_refcnt wouldn't change when we open a
279 * dir which the spa also has open, so we could incorrectly
280 * think it was OK to unload/export/destroy the pool. We need
281 * the instantiate-to-evict hold because the dsl_dir_t has a
282 * pointer to the dd_pool, which has a pointer to the spa_t.
284 spa_open_ref(dp
->dp_spa
, tag
);
285 ASSERT3P(dd
->dd_pool
, ==, dp
);
286 ASSERT3U(dd
->dd_object
, ==, ddobj
);
287 ASSERT3P(dd
->dd_dbuf
, ==, dbuf
);
293 dsl_dir_rele(dd
->dd_parent
, dd
);
295 mutex_destroy(&dd
->dd_lock
);
296 kmem_free(dd
, sizeof (dsl_dir_t
));
297 dmu_buf_rele(dbuf
, tag
);
302 dsl_dir_rele(dsl_dir_t
*dd
, void *tag
)
304 dprintf_dd(dd
, "%s\n", "");
305 spa_close(dd
->dd_pool
->dp_spa
, tag
);
306 dmu_buf_rele(dd
->dd_dbuf
, tag
);
310 * Remove a reference to the given dsl dir that is being asynchronously
311 * released. Async releases occur from a taskq performing eviction of
312 * dsl datasets and dirs. This process is identical to a normal release
313 * with the exception of using the async API for releasing the reference on
317 dsl_dir_async_rele(dsl_dir_t
*dd
, void *tag
)
319 dprintf_dd(dd
, "%s\n", "");
320 spa_async_close(dd
->dd_pool
->dp_spa
, tag
);
321 dmu_buf_rele(dd
->dd_dbuf
, tag
);
324 /* buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes */
326 dsl_dir_name(dsl_dir_t
*dd
, char *buf
)
329 dsl_dir_name(dd
->dd_parent
, buf
);
330 VERIFY3U(strlcat(buf
, "/", ZFS_MAX_DATASET_NAME_LEN
), <,
331 ZFS_MAX_DATASET_NAME_LEN
);
335 if (!MUTEX_HELD(&dd
->dd_lock
)) {
337 * recursive mutex so that we can use
338 * dprintf_dd() with dd_lock held
340 mutex_enter(&dd
->dd_lock
);
341 VERIFY3U(strlcat(buf
, dd
->dd_myname
, ZFS_MAX_DATASET_NAME_LEN
),
342 <, ZFS_MAX_DATASET_NAME_LEN
);
343 mutex_exit(&dd
->dd_lock
);
345 VERIFY3U(strlcat(buf
, dd
->dd_myname
, ZFS_MAX_DATASET_NAME_LEN
),
346 <, ZFS_MAX_DATASET_NAME_LEN
);
350 /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
352 dsl_dir_namelen(dsl_dir_t
*dd
)
357 /* parent's name + 1 for the "/" */
358 result
= dsl_dir_namelen(dd
->dd_parent
) + 1;
361 if (!MUTEX_HELD(&dd
->dd_lock
)) {
362 /* see dsl_dir_name */
363 mutex_enter(&dd
->dd_lock
);
364 result
+= strlen(dd
->dd_myname
);
365 mutex_exit(&dd
->dd_lock
);
367 result
+= strlen(dd
->dd_myname
);
374 getcomponent(const char *path
, char *component
, const char **nextp
)
378 if ((path
== NULL
) || (path
[0] == '\0'))
379 return (SET_ERROR(ENOENT
));
380 /* This would be a good place to reserve some namespace... */
381 p
= strpbrk(path
, "/@");
382 if (p
&& (p
[1] == '/' || p
[1] == '@')) {
383 /* two separators in a row */
384 return (SET_ERROR(EINVAL
));
386 if (p
== NULL
|| p
== path
) {
388 * if the first thing is an @ or /, it had better be an
389 * @ and it had better not have any more ats or slashes,
390 * and it had better have something after the @.
393 (p
[0] != '@' || strpbrk(path
+1, "/@") || p
[1] == '\0'))
394 return (SET_ERROR(EINVAL
));
395 if (strlen(path
) >= ZFS_MAX_DATASET_NAME_LEN
)
396 return (SET_ERROR(ENAMETOOLONG
));
397 (void) strcpy(component
, path
);
399 } else if (p
[0] == '/') {
400 if (p
- path
>= ZFS_MAX_DATASET_NAME_LEN
)
401 return (SET_ERROR(ENAMETOOLONG
));
402 (void) strncpy(component
, path
, p
- path
);
403 component
[p
- path
] = '\0';
405 } else if (p
[0] == '@') {
407 * if the next separator is an @, there better not be
410 if (strchr(path
, '/'))
411 return (SET_ERROR(EINVAL
));
412 if (p
- path
>= ZFS_MAX_DATASET_NAME_LEN
)
413 return (SET_ERROR(ENAMETOOLONG
));
414 (void) strncpy(component
, path
, p
- path
);
415 component
[p
- path
] = '\0';
417 panic("invalid p=%p", (void *)p
);
424 * Return the dsl_dir_t, and possibly the last component which couldn't
425 * be found in *tail. The name must be in the specified dsl_pool_t. This
426 * thread must hold the dp_config_rwlock for the pool. Returns NULL if the
427 * path is bogus, or if tail==NULL and we couldn't parse the whole name.
428 * (*tail)[0] == '@' means that the last component is a snapshot.
431 dsl_dir_hold(dsl_pool_t
*dp
, const char *name
, void *tag
,
432 dsl_dir_t
**ddp
, const char **tailp
)
435 const char *spaname
, *next
, *nextnext
= NULL
;
440 buf
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
441 err
= getcomponent(name
, buf
, &next
);
445 /* Make sure the name is in the specified pool. */
446 spaname
= spa_name(dp
->dp_spa
);
447 if (strcmp(buf
, spaname
) != 0) {
448 err
= SET_ERROR(EXDEV
);
452 ASSERT(dsl_pool_config_held(dp
));
454 err
= dsl_dir_hold_obj(dp
, dp
->dp_root_dir_obj
, NULL
, tag
, &dd
);
459 while (next
!= NULL
) {
461 err
= getcomponent(next
, buf
, &nextnext
);
464 ASSERT(next
[0] != '\0');
467 dprintf("looking up %s in obj%lld\n",
468 buf
, dsl_dir_phys(dd
)->dd_child_dir_zapobj
);
470 err
= zap_lookup(dp
->dp_meta_objset
,
471 dsl_dir_phys(dd
)->dd_child_dir_zapobj
,
472 buf
, sizeof (ddobj
), 1, &ddobj
);
479 err
= dsl_dir_hold_obj(dp
, ddobj
, buf
, tag
, &child_dd
);
482 dsl_dir_rele(dd
, tag
);
488 dsl_dir_rele(dd
, tag
);
493 * It's an error if there's more than one component left, or
494 * tailp==NULL and there's any component left.
497 (tailp
== NULL
|| (nextnext
&& nextnext
[0] != '\0'))) {
499 dsl_dir_rele(dd
, tag
);
500 dprintf("next=%p (%s) tail=%p\n", next
, next
?next
:"", tailp
);
501 err
= SET_ERROR(ENOENT
);
508 kmem_free(buf
, ZFS_MAX_DATASET_NAME_LEN
);
513 * If the counts are already initialized for this filesystem and its
514 * descendants then do nothing, otherwise initialize the counts.
516 * The counts on this filesystem, and those below, may be uninitialized due to
517 * either the use of a pre-existing pool which did not support the
518 * filesystem/snapshot limit feature, or one in which the feature had not yet
521 * Recursively descend the filesystem tree and update the filesystem/snapshot
522 * counts on each filesystem below, then update the cumulative count on the
523 * current filesystem. If the filesystem already has a count set on it,
524 * then we know that its counts, and the counts on the filesystems below it,
525 * are already correct, so we don't have to update this filesystem.
528 dsl_dir_init_fs_ss_count(dsl_dir_t
*dd
, dmu_tx_t
*tx
)
530 uint64_t my_fs_cnt
= 0;
531 uint64_t my_ss_cnt
= 0;
532 dsl_pool_t
*dp
= dd
->dd_pool
;
533 objset_t
*os
= dp
->dp_meta_objset
;
538 ASSERT(spa_feature_is_active(dp
->dp_spa
, SPA_FEATURE_FS_SS_LIMIT
));
539 ASSERT(dsl_pool_config_held(dp
));
540 ASSERT(dmu_tx_is_syncing(tx
));
542 dsl_dir_zapify(dd
, tx
);
545 * If the filesystem count has already been initialized then we
546 * don't need to recurse down any further.
548 if (zap_contains(os
, dd
->dd_object
, DD_FIELD_FILESYSTEM_COUNT
) == 0)
551 zc
= kmem_alloc(sizeof (zap_cursor_t
), KM_SLEEP
);
552 za
= kmem_alloc(sizeof (zap_attribute_t
), KM_SLEEP
);
554 /* Iterate my child dirs */
555 for (zap_cursor_init(zc
, os
, dsl_dir_phys(dd
)->dd_child_dir_zapobj
);
556 zap_cursor_retrieve(zc
, za
) == 0; zap_cursor_advance(zc
)) {
560 VERIFY0(dsl_dir_hold_obj(dp
, za
->za_first_integer
, NULL
, FTAG
,
564 * Ignore hidden ($FREE, $MOS & $ORIGIN) objsets and
565 * temporary datasets.
567 if (chld_dd
->dd_myname
[0] == '$' ||
568 chld_dd
->dd_myname
[0] == '%') {
569 dsl_dir_rele(chld_dd
, FTAG
);
573 my_fs_cnt
++; /* count this child */
575 dsl_dir_init_fs_ss_count(chld_dd
, tx
);
577 VERIFY0(zap_lookup(os
, chld_dd
->dd_object
,
578 DD_FIELD_FILESYSTEM_COUNT
, sizeof (count
), 1, &count
));
580 VERIFY0(zap_lookup(os
, chld_dd
->dd_object
,
581 DD_FIELD_SNAPSHOT_COUNT
, sizeof (count
), 1, &count
));
584 dsl_dir_rele(chld_dd
, FTAG
);
587 /* Count my snapshots (we counted children's snapshots above) */
588 VERIFY0(dsl_dataset_hold_obj(dd
->dd_pool
,
589 dsl_dir_phys(dd
)->dd_head_dataset_obj
, FTAG
, &ds
));
591 for (zap_cursor_init(zc
, os
, dsl_dataset_phys(ds
)->ds_snapnames_zapobj
);
592 zap_cursor_retrieve(zc
, za
) == 0;
593 zap_cursor_advance(zc
)) {
594 /* Don't count temporary snapshots */
595 if (za
->za_name
[0] != '%')
600 dsl_dataset_rele(ds
, FTAG
);
602 kmem_free(zc
, sizeof (zap_cursor_t
));
603 kmem_free(za
, sizeof (zap_attribute_t
));
605 /* we're in a sync task, update counts */
606 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
607 VERIFY0(zap_add(os
, dd
->dd_object
, DD_FIELD_FILESYSTEM_COUNT
,
608 sizeof (my_fs_cnt
), 1, &my_fs_cnt
, tx
));
609 VERIFY0(zap_add(os
, dd
->dd_object
, DD_FIELD_SNAPSHOT_COUNT
,
610 sizeof (my_ss_cnt
), 1, &my_ss_cnt
, tx
));
614 dsl_dir_actv_fs_ss_limit_check(void *arg
, dmu_tx_t
*tx
)
616 char *ddname
= (char *)arg
;
617 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
622 error
= dsl_dataset_hold(dp
, ddname
, FTAG
, &ds
);
626 if (!spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_FS_SS_LIMIT
)) {
627 dsl_dataset_rele(ds
, FTAG
);
628 return (SET_ERROR(ENOTSUP
));
632 if (spa_feature_is_active(dp
->dp_spa
, SPA_FEATURE_FS_SS_LIMIT
) &&
633 dsl_dir_is_zapified(dd
) &&
634 zap_contains(dp
->dp_meta_objset
, dd
->dd_object
,
635 DD_FIELD_FILESYSTEM_COUNT
) == 0) {
636 dsl_dataset_rele(ds
, FTAG
);
637 return (SET_ERROR(EALREADY
));
640 dsl_dataset_rele(ds
, FTAG
);
645 dsl_dir_actv_fs_ss_limit_sync(void *arg
, dmu_tx_t
*tx
)
647 char *ddname
= (char *)arg
;
648 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
652 VERIFY0(dsl_dataset_hold(dp
, ddname
, FTAG
, &ds
));
654 spa
= dsl_dataset_get_spa(ds
);
656 if (!spa_feature_is_active(spa
, SPA_FEATURE_FS_SS_LIMIT
)) {
658 * Since the feature was not active and we're now setting a
659 * limit, increment the feature-active counter so that the
660 * feature becomes active for the first time.
662 * We are already in a sync task so we can update the MOS.
664 spa_feature_incr(spa
, SPA_FEATURE_FS_SS_LIMIT
, tx
);
668 * Since we are now setting a non-UINT64_MAX limit on the filesystem,
669 * we need to ensure the counts are correct. Descend down the tree from
670 * this point and update all of the counts to be accurate.
672 dsl_dir_init_fs_ss_count(ds
->ds_dir
, tx
);
674 dsl_dataset_rele(ds
, FTAG
);
678 * Make sure the feature is enabled and activate it if necessary.
679 * Since we're setting a limit, ensure the on-disk counts are valid.
680 * This is only called by the ioctl path when setting a limit value.
682 * We do not need to validate the new limit, since users who can change the
683 * limit are also allowed to exceed the limit.
686 dsl_dir_activate_fs_ss_limit(const char *ddname
)
690 error
= dsl_sync_task(ddname
, dsl_dir_actv_fs_ss_limit_check
,
691 dsl_dir_actv_fs_ss_limit_sync
, (void *)ddname
, 0,
692 ZFS_SPACE_CHECK_RESERVED
);
694 if (error
== EALREADY
)
701 * Used to determine if the filesystem_limit or snapshot_limit should be
702 * enforced. We allow the limit to be exceeded if the user has permission to
703 * write the property value. We pass in the creds that we got in the open
704 * context since we will always be the GZ root in syncing context. We also have
705 * to handle the case where we are allowed to change the limit on the current
706 * dataset, but there may be another limit in the tree above.
708 * We can never modify these two properties within a non-global zone. In
709 * addition, the other checks are modeled on zfs_secpolicy_write_perms. We
710 * can't use that function since we are already holding the dp_config_rwlock.
711 * In addition, we already have the dd and dealing with snapshots is simplified
722 dsl_enforce_ds_ss_limits(dsl_dir_t
*dd
, zfs_prop_t prop
, cred_t
*cr
)
724 enforce_res_t enforce
= ENFORCE_ALWAYS
;
729 ASSERT(prop
== ZFS_PROP_FILESYSTEM_LIMIT
||
730 prop
== ZFS_PROP_SNAPSHOT_LIMIT
);
733 if (crgetzoneid(cr
) != GLOBAL_ZONEID
)
734 return (ENFORCE_ALWAYS
);
736 if (secpolicy_zfs(cr
) == 0)
737 return (ENFORCE_NEVER
);
740 if ((obj
= dsl_dir_phys(dd
)->dd_head_dataset_obj
) == 0)
741 return (ENFORCE_ALWAYS
);
743 ASSERT(dsl_pool_config_held(dd
->dd_pool
));
745 if (dsl_dataset_hold_obj(dd
->dd_pool
, obj
, FTAG
, &ds
) != 0)
746 return (ENFORCE_ALWAYS
);
748 if (dsl_prop_get_ds(ds
, "zoned", 8, 1, &zoned
, NULL
) || zoned
) {
749 /* Only root can access zoned fs's from the GZ */
750 enforce
= ENFORCE_ALWAYS
;
752 if (dsl_deleg_access_impl(ds
, zfs_prop_to_name(prop
), cr
) == 0)
753 enforce
= ENFORCE_ABOVE
;
756 dsl_dataset_rele(ds
, FTAG
);
761 dsl_dir_update_last_remap_txg_sync(void *varg
, dmu_tx_t
*tx
)
763 ddulrt_arg_t
*arg
= varg
;
764 uint64_t last_remap_txg
;
765 dsl_dir_t
*dd
= arg
->ddulrta_dd
;
766 objset_t
*mos
= dd
->dd_pool
->dp_meta_objset
;
768 dsl_dir_zapify(dd
, tx
);
769 if (zap_lookup(mos
, dd
->dd_object
, DD_FIELD_LAST_REMAP_TXG
,
770 sizeof (last_remap_txg
), 1, &last_remap_txg
) != 0 ||
771 last_remap_txg
< arg
->ddlrta_txg
) {
772 VERIFY0(zap_update(mos
, dd
->dd_object
, DD_FIELD_LAST_REMAP_TXG
,
773 sizeof (arg
->ddlrta_txg
), 1, &arg
->ddlrta_txg
, tx
));
778 dsl_dir_update_last_remap_txg(dsl_dir_t
*dd
, uint64_t txg
)
782 arg
.ddlrta_txg
= txg
;
784 return (dsl_sync_task(spa_name(dd
->dd_pool
->dp_spa
),
785 NULL
, dsl_dir_update_last_remap_txg_sync
, &arg
,
786 1, ZFS_SPACE_CHECK_RESERVED
));
790 * Check if adding additional child filesystem(s) would exceed any filesystem
791 * limits or adding additional snapshot(s) would exceed any snapshot limits.
792 * The prop argument indicates which limit to check.
794 * Note that all filesystem limits up to the root (or the highest
795 * initialized) filesystem or the given ancestor must be satisfied.
798 dsl_fs_ss_limit_check(dsl_dir_t
*dd
, uint64_t delta
, zfs_prop_t prop
,
799 dsl_dir_t
*ancestor
, cred_t
*cr
)
801 objset_t
*os
= dd
->dd_pool
->dp_meta_objset
;
802 uint64_t limit
, count
;
804 enforce_res_t enforce
;
807 ASSERT(dsl_pool_config_held(dd
->dd_pool
));
808 ASSERT(prop
== ZFS_PROP_FILESYSTEM_LIMIT
||
809 prop
== ZFS_PROP_SNAPSHOT_LIMIT
);
812 * If we're allowed to change the limit, don't enforce the limit
813 * e.g. this can happen if a snapshot is taken by an administrative
814 * user in the global zone (i.e. a recursive snapshot by root).
815 * However, we must handle the case of delegated permissions where we
816 * are allowed to change the limit on the current dataset, but there
817 * is another limit in the tree above.
819 enforce
= dsl_enforce_ds_ss_limits(dd
, prop
, cr
);
820 if (enforce
== ENFORCE_NEVER
)
824 * e.g. if renaming a dataset with no snapshots, count adjustment
830 if (prop
== ZFS_PROP_SNAPSHOT_LIMIT
) {
832 * We don't enforce the limit for temporary snapshots. This is
833 * indicated by a NULL cred_t argument.
838 count_prop
= DD_FIELD_SNAPSHOT_COUNT
;
840 count_prop
= DD_FIELD_FILESYSTEM_COUNT
;
844 * If an ancestor has been provided, stop checking the limit once we
845 * hit that dir. We need this during rename so that we don't overcount
846 * the check once we recurse up to the common ancestor.
852 * If we hit an uninitialized node while recursing up the tree, we can
853 * stop since we know there is no limit here (or above). The counts are
854 * not valid on this node and we know we won't touch this node's counts.
856 if (!dsl_dir_is_zapified(dd
) || zap_lookup(os
, dd
->dd_object
,
857 count_prop
, sizeof (count
), 1, &count
) == ENOENT
)
860 err
= dsl_prop_get_dd(dd
, zfs_prop_to_name(prop
), 8, 1, &limit
, NULL
,
865 /* Is there a limit which we've hit? */
866 if (enforce
== ENFORCE_ALWAYS
&& (count
+ delta
) > limit
)
867 return (SET_ERROR(EDQUOT
));
869 if (dd
->dd_parent
!= NULL
)
870 err
= dsl_fs_ss_limit_check(dd
->dd_parent
, delta
, prop
,
877 * Adjust the filesystem or snapshot count for the specified dsl_dir_t and all
878 * parents. When a new filesystem/snapshot is created, increment the count on
879 * all parents, and when a filesystem/snapshot is destroyed, decrement the
883 dsl_fs_ss_count_adjust(dsl_dir_t
*dd
, int64_t delta
, const char *prop
,
887 objset_t
*os
= dd
->dd_pool
->dp_meta_objset
;
890 ASSERT(dsl_pool_config_held(dd
->dd_pool
));
891 ASSERT(dmu_tx_is_syncing(tx
));
892 ASSERT(strcmp(prop
, DD_FIELD_FILESYSTEM_COUNT
) == 0 ||
893 strcmp(prop
, DD_FIELD_SNAPSHOT_COUNT
) == 0);
896 * When we receive an incremental stream into a filesystem that already
897 * exists, a temporary clone is created. We don't count this temporary
898 * clone, whose name begins with a '%'. We also ignore hidden ($FREE,
899 * $MOS & $ORIGIN) objsets.
901 if ((dd
->dd_myname
[0] == '%' || dd
->dd_myname
[0] == '$') &&
902 strcmp(prop
, DD_FIELD_FILESYSTEM_COUNT
) == 0)
906 * e.g. if renaming a dataset with no snapshots, count adjustment is 0
912 * If we hit an uninitialized node while recursing up the tree, we can
913 * stop since we know the counts are not valid on this node and we
914 * know we shouldn't touch this node's counts. An uninitialized count
915 * on the node indicates that either the feature has not yet been
916 * activated or there are no limits on this part of the tree.
918 if (!dsl_dir_is_zapified(dd
) || (err
= zap_lookup(os
, dd
->dd_object
,
919 prop
, sizeof (count
), 1, &count
)) == ENOENT
)
924 /* Use a signed verify to make sure we're not neg. */
925 VERIFY3S(count
, >=, 0);
927 VERIFY0(zap_update(os
, dd
->dd_object
, prop
, sizeof (count
), 1, &count
,
930 /* Roll up this additional count into our ancestors */
931 if (dd
->dd_parent
!= NULL
)
932 dsl_fs_ss_count_adjust(dd
->dd_parent
, delta
, prop
, tx
);
936 dsl_dir_create_sync(dsl_pool_t
*dp
, dsl_dir_t
*pds
, const char *name
,
939 objset_t
*mos
= dp
->dp_meta_objset
;
941 dsl_dir_phys_t
*ddphys
;
944 ddobj
= dmu_object_alloc(mos
, DMU_OT_DSL_DIR
, 0,
945 DMU_OT_DSL_DIR
, sizeof (dsl_dir_phys_t
), tx
);
947 VERIFY0(zap_add(mos
, dsl_dir_phys(pds
)->dd_child_dir_zapobj
,
948 name
, sizeof (uint64_t), 1, &ddobj
, tx
));
950 /* it's the root dir */
951 VERIFY0(zap_add(mos
, DMU_POOL_DIRECTORY_OBJECT
,
952 DMU_POOL_ROOT_DATASET
, sizeof (uint64_t), 1, &ddobj
, tx
));
954 VERIFY0(dmu_bonus_hold(mos
, ddobj
, FTAG
, &dbuf
));
955 dmu_buf_will_dirty(dbuf
, tx
);
956 ddphys
= dbuf
->db_data
;
958 ddphys
->dd_creation_time
= gethrestime_sec();
960 ddphys
->dd_parent_obj
= pds
->dd_object
;
962 /* update the filesystem counts */
963 dsl_fs_ss_count_adjust(pds
, 1, DD_FIELD_FILESYSTEM_COUNT
, tx
);
965 ddphys
->dd_props_zapobj
= zap_create(mos
,
966 DMU_OT_DSL_PROPS
, DMU_OT_NONE
, 0, tx
);
967 ddphys
->dd_child_dir_zapobj
= zap_create(mos
,
968 DMU_OT_DSL_DIR_CHILD_MAP
, DMU_OT_NONE
, 0, tx
);
969 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_USED_BREAKDOWN
)
970 ddphys
->dd_flags
|= DD_FLAG_USED_BREAKDOWN
;
972 dmu_buf_rele(dbuf
, FTAG
);
978 dsl_dir_is_clone(dsl_dir_t
*dd
)
980 return (dsl_dir_phys(dd
)->dd_origin_obj
&&
981 (dd
->dd_pool
->dp_origin_snap
== NULL
||
982 dsl_dir_phys(dd
)->dd_origin_obj
!=
983 dd
->dd_pool
->dp_origin_snap
->ds_object
));
987 dsl_dir_get_used(dsl_dir_t
*dd
)
989 return (dsl_dir_phys(dd
)->dd_used_bytes
);
993 dsl_dir_get_compressed(dsl_dir_t
*dd
)
995 return (dsl_dir_phys(dd
)->dd_compressed_bytes
);
999 dsl_dir_get_quota(dsl_dir_t
*dd
)
1001 return (dsl_dir_phys(dd
)->dd_quota
);
1005 dsl_dir_get_reservation(dsl_dir_t
*dd
)
1007 return (dsl_dir_phys(dd
)->dd_reserved
);
1011 dsl_dir_get_compressratio(dsl_dir_t
*dd
)
1013 /* a fixed point number, 100x the ratio */
1014 return (dsl_dir_phys(dd
)->dd_compressed_bytes
== 0 ? 100 :
1015 (dsl_dir_phys(dd
)->dd_uncompressed_bytes
* 100 /
1016 dsl_dir_phys(dd
)->dd_compressed_bytes
));
1020 dsl_dir_get_logicalused(dsl_dir_t
*dd
)
1022 return (dsl_dir_phys(dd
)->dd_uncompressed_bytes
);
1026 dsl_dir_get_usedsnap(dsl_dir_t
*dd
)
1028 return (dsl_dir_phys(dd
)->dd_used_breakdown
[DD_USED_SNAP
]);
1032 dsl_dir_get_usedds(dsl_dir_t
*dd
)
1034 return (dsl_dir_phys(dd
)->dd_used_breakdown
[DD_USED_HEAD
]);
1038 dsl_dir_get_usedrefreserv(dsl_dir_t
*dd
)
1040 return (dsl_dir_phys(dd
)->dd_used_breakdown
[DD_USED_REFRSRV
]);
1044 dsl_dir_get_usedchild(dsl_dir_t
*dd
)
1046 return (dsl_dir_phys(dd
)->dd_used_breakdown
[DD_USED_CHILD
] +
1047 dsl_dir_phys(dd
)->dd_used_breakdown
[DD_USED_CHILD_RSRV
]);
1051 dsl_dir_get_origin(dsl_dir_t
*dd
, char *buf
)
1054 VERIFY0(dsl_dataset_hold_obj(dd
->dd_pool
,
1055 dsl_dir_phys(dd
)->dd_origin_obj
, FTAG
, &ds
));
1057 dsl_dataset_name(ds
, buf
);
1059 dsl_dataset_rele(ds
, FTAG
);
1063 dsl_dir_get_filesystem_count(dsl_dir_t
*dd
, uint64_t *count
)
1065 if (dsl_dir_is_zapified(dd
)) {
1066 objset_t
*os
= dd
->dd_pool
->dp_meta_objset
;
1067 return (zap_lookup(os
, dd
->dd_object
, DD_FIELD_FILESYSTEM_COUNT
,
1068 sizeof (*count
), 1, count
));
1075 dsl_dir_get_snapshot_count(dsl_dir_t
*dd
, uint64_t *count
)
1077 if (dsl_dir_is_zapified(dd
)) {
1078 objset_t
*os
= dd
->dd_pool
->dp_meta_objset
;
1079 return (zap_lookup(os
, dd
->dd_object
, DD_FIELD_SNAPSHOT_COUNT
,
1080 sizeof (*count
), 1, count
));
1087 dsl_dir_get_remaptxg(dsl_dir_t
*dd
, uint64_t *count
)
1089 if (dsl_dir_is_zapified(dd
)) {
1090 objset_t
*os
= dd
->dd_pool
->dp_meta_objset
;
1091 return (zap_lookup(os
, dd
->dd_object
, DD_FIELD_LAST_REMAP_TXG
,
1092 sizeof (*count
), 1, count
));
1100 dsl_dir_stats(dsl_dir_t
*dd
, nvlist_t
*nv
)
1102 mutex_enter(&dd
->dd_lock
);
1103 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_QUOTA
,
1104 dsl_dir_get_quota(dd
));
1105 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_RESERVATION
,
1106 dsl_dir_get_reservation(dd
));
1107 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_LOGICALUSED
,
1108 dsl_dir_get_logicalused(dd
));
1109 if (dsl_dir_phys(dd
)->dd_flags
& DD_FLAG_USED_BREAKDOWN
) {
1110 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USEDSNAP
,
1111 dsl_dir_get_usedsnap(dd
));
1112 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USEDDS
,
1113 dsl_dir_get_usedds(dd
));
1114 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USEDREFRESERV
,
1115 dsl_dir_get_usedrefreserv(dd
));
1116 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USEDCHILD
,
1117 dsl_dir_get_usedchild(dd
));
1119 mutex_exit(&dd
->dd_lock
);
1122 if (dsl_dir_get_filesystem_count(dd
, &count
) == 0) {
1123 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_FILESYSTEM_COUNT
,
1126 if (dsl_dir_get_snapshot_count(dd
, &count
) == 0) {
1127 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_SNAPSHOT_COUNT
,
1130 if (dsl_dir_get_remaptxg(dd
, &count
) == 0) {
1131 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REMAPTXG
,
1135 if (dsl_dir_is_clone(dd
)) {
1136 char buf
[ZFS_MAX_DATASET_NAME_LEN
];
1137 dsl_dir_get_origin(dd
, buf
);
1138 dsl_prop_nvlist_add_string(nv
, ZFS_PROP_ORIGIN
, buf
);
1144 dsl_dir_dirty(dsl_dir_t
*dd
, dmu_tx_t
*tx
)
1146 dsl_pool_t
*dp
= dd
->dd_pool
;
1148 ASSERT(dsl_dir_phys(dd
));
1150 if (txg_list_add(&dp
->dp_dirty_dirs
, dd
, tx
->tx_txg
)) {
1151 /* up the hold count until we can be written out */
1152 dmu_buf_add_ref(dd
->dd_dbuf
, dd
);
1157 parent_delta(dsl_dir_t
*dd
, uint64_t used
, int64_t delta
)
1159 uint64_t old_accounted
= MAX(used
, dsl_dir_phys(dd
)->dd_reserved
);
1160 uint64_t new_accounted
=
1161 MAX(used
+ delta
, dsl_dir_phys(dd
)->dd_reserved
);
1162 return (new_accounted
- old_accounted
);
1166 dsl_dir_sync(dsl_dir_t
*dd
, dmu_tx_t
*tx
)
1168 ASSERT(dmu_tx_is_syncing(tx
));
1170 mutex_enter(&dd
->dd_lock
);
1171 ASSERT0(dd
->dd_tempreserved
[tx
->tx_txg
&TXG_MASK
]);
1172 dprintf_dd(dd
, "txg=%llu towrite=%lluK\n", tx
->tx_txg
,
1173 dd
->dd_space_towrite
[tx
->tx_txg
&TXG_MASK
] / 1024);
1174 dd
->dd_space_towrite
[tx
->tx_txg
&TXG_MASK
] = 0;
1175 mutex_exit(&dd
->dd_lock
);
1177 /* release the hold from dsl_dir_dirty */
1178 dmu_buf_rele(dd
->dd_dbuf
, dd
);
1182 dsl_dir_space_towrite(dsl_dir_t
*dd
)
1186 ASSERT(MUTEX_HELD(&dd
->dd_lock
));
1188 for (int i
= 0; i
< TXG_SIZE
; i
++) {
1189 space
+= dd
->dd_space_towrite
[i
& TXG_MASK
];
1190 ASSERT3U(dd
->dd_space_towrite
[i
& TXG_MASK
], >=, 0);
1196 * How much space would dd have available if ancestor had delta applied
1197 * to it? If ondiskonly is set, we're only interested in what's
1198 * on-disk, not estimated pending changes.
1201 dsl_dir_space_available(dsl_dir_t
*dd
,
1202 dsl_dir_t
*ancestor
, int64_t delta
, int ondiskonly
)
1204 uint64_t parentspace
, myspace
, quota
, used
;
1207 * If there are no restrictions otherwise, assume we have
1208 * unlimited space available.
1211 parentspace
= UINT64_MAX
;
1213 if (dd
->dd_parent
!= NULL
) {
1214 parentspace
= dsl_dir_space_available(dd
->dd_parent
,
1215 ancestor
, delta
, ondiskonly
);
1218 mutex_enter(&dd
->dd_lock
);
1219 if (dsl_dir_phys(dd
)->dd_quota
!= 0)
1220 quota
= dsl_dir_phys(dd
)->dd_quota
;
1221 used
= dsl_dir_phys(dd
)->dd_used_bytes
;
1223 used
+= dsl_dir_space_towrite(dd
);
1225 if (dd
->dd_parent
== NULL
) {
1226 uint64_t poolsize
= dsl_pool_adjustedsize(dd
->dd_pool
,
1227 ZFS_SPACE_CHECK_NORMAL
);
1228 quota
= MIN(quota
, poolsize
);
1231 if (dsl_dir_phys(dd
)->dd_reserved
> used
&& parentspace
!= UINT64_MAX
) {
1233 * We have some space reserved, in addition to what our
1236 parentspace
+= dsl_dir_phys(dd
)->dd_reserved
- used
;
1239 if (dd
== ancestor
) {
1241 ASSERT(used
>= -delta
);
1243 if (parentspace
!= UINT64_MAX
)
1244 parentspace
-= delta
;
1252 * the lesser of the space provided by our parent and
1253 * the space left in our quota
1255 myspace
= MIN(parentspace
, quota
- used
);
1258 mutex_exit(&dd
->dd_lock
);
1263 struct tempreserve
{
1264 list_node_t tr_node
;
1270 dsl_dir_tempreserve_impl(dsl_dir_t
*dd
, uint64_t asize
, boolean_t netfree
,
1271 boolean_t ignorequota
, list_t
*tr_list
,
1272 dmu_tx_t
*tx
, boolean_t first
)
1276 struct tempreserve
*tr
;
1285 ASSERT3U(txg
, !=, 0);
1286 ASSERT3S(asize
, >, 0);
1288 mutex_enter(&dd
->dd_lock
);
1291 * Check against the dsl_dir's quota. We don't add in the delta
1292 * when checking for over-quota because they get one free hit.
1294 uint64_t est_inflight
= dsl_dir_space_towrite(dd
);
1295 for (int i
= 0; i
< TXG_SIZE
; i
++)
1296 est_inflight
+= dd
->dd_tempreserved
[i
];
1297 uint64_t used_on_disk
= dsl_dir_phys(dd
)->dd_used_bytes
;
1300 * On the first iteration, fetch the dataset's used-on-disk and
1301 * refreservation values. Also, if checkrefquota is set, test if
1302 * allocating this space would exceed the dataset's refquota.
1304 if (first
&& tx
->tx_objset
) {
1306 dsl_dataset_t
*ds
= tx
->tx_objset
->os_dsl_dataset
;
1308 error
= dsl_dataset_check_quota(ds
, !netfree
,
1309 asize
, est_inflight
, &used_on_disk
, &ref_rsrv
);
1311 mutex_exit(&dd
->dd_lock
);
1312 DMU_TX_STAT_BUMP(dmu_tx_quota
);
1318 * If this transaction will result in a net free of space,
1319 * we want to let it through.
1321 if (ignorequota
|| netfree
|| dsl_dir_phys(dd
)->dd_quota
== 0)
1324 quota
= dsl_dir_phys(dd
)->dd_quota
;
1327 * Adjust the quota against the actual pool size at the root
1328 * minus any outstanding deferred frees.
1329 * To ensure that it's possible to remove files from a full
1330 * pool without inducing transient overcommits, we throttle
1331 * netfree transactions against a quota that is slightly larger,
1332 * but still within the pool's allocation slop. In cases where
1333 * we're very close to full, this will allow a steady trickle of
1334 * removes to get through.
1336 uint64_t deferred
= 0;
1337 if (dd
->dd_parent
== NULL
) {
1338 uint64_t avail
= dsl_pool_unreserved_space(dd
->dd_pool
,
1340 ZFS_SPACE_CHECK_RESERVED
: ZFS_SPACE_CHECK_NORMAL
);
1342 if (avail
< quota
) {
1349 * If they are requesting more space, and our current estimate
1350 * is over quota, they get to try again unless the actual
1351 * on-disk is over quota and there are no pending changes (which
1352 * may free up space for us).
1354 if (used_on_disk
+ est_inflight
>= quota
) {
1355 if (est_inflight
> 0 || used_on_disk
< quota
||
1356 (retval
== ENOSPC
&& used_on_disk
< quota
+ deferred
))
1358 dprintf_dd(dd
, "failing: used=%lluK inflight = %lluK "
1359 "quota=%lluK tr=%lluK err=%d\n",
1360 used_on_disk
>>10, est_inflight
>>10,
1361 quota
>>10, asize
>>10, retval
);
1362 mutex_exit(&dd
->dd_lock
);
1363 DMU_TX_STAT_BUMP(dmu_tx_quota
);
1364 return (SET_ERROR(retval
));
1367 /* We need to up our estimated delta before dropping dd_lock */
1368 dd
->dd_tempreserved
[txg
& TXG_MASK
] += asize
;
1370 uint64_t parent_rsrv
= parent_delta(dd
, used_on_disk
+ est_inflight
,
1372 mutex_exit(&dd
->dd_lock
);
1374 tr
= kmem_zalloc(sizeof (struct tempreserve
), KM_SLEEP
);
1376 tr
->tr_size
= asize
;
1377 list_insert_tail(tr_list
, tr
);
1379 /* see if it's OK with our parent */
1380 if (dd
->dd_parent
!= NULL
&& parent_rsrv
!= 0) {
1382 * Recurse on our parent without recursion. This has been
1383 * observed to be potentially large stack usage even within
1384 * the test suite. Largest seen stack was 7632 bytes on linux.
1388 asize
= parent_rsrv
;
1389 ignorequota
= (dsl_dir_phys(dd
)->dd_head_dataset_obj
== 0);
1391 goto top_of_function
;
1399 * Reserve space in this dsl_dir, to be used in this tx's txg.
1400 * After the space has been dirtied (and dsl_dir_willuse_space()
1401 * has been called), the reservation should be canceled, using
1402 * dsl_dir_tempreserve_clear().
1405 dsl_dir_tempreserve_space(dsl_dir_t
*dd
, uint64_t lsize
, uint64_t asize
,
1406 boolean_t netfree
, void **tr_cookiep
, dmu_tx_t
*tx
)
1416 tr_list
= kmem_alloc(sizeof (list_t
), KM_SLEEP
);
1417 list_create(tr_list
, sizeof (struct tempreserve
),
1418 offsetof(struct tempreserve
, tr_node
));
1419 ASSERT3S(asize
, >, 0);
1421 err
= arc_tempreserve_space(dd
->dd_pool
->dp_spa
, lsize
, tx
->tx_txg
);
1423 struct tempreserve
*tr
;
1425 tr
= kmem_zalloc(sizeof (struct tempreserve
), KM_SLEEP
);
1426 tr
->tr_size
= lsize
;
1427 list_insert_tail(tr_list
, tr
);
1429 if (err
== EAGAIN
) {
1431 * If arc_memory_throttle() detected that pageout
1432 * is running and we are low on memory, we delay new
1433 * non-pageout transactions to give pageout an
1436 * It is unfortunate to be delaying while the caller's
1439 txg_delay(dd
->dd_pool
, tx
->tx_txg
,
1440 MSEC2NSEC(10), MSEC2NSEC(10));
1441 err
= SET_ERROR(ERESTART
);
1446 err
= dsl_dir_tempreserve_impl(dd
, asize
, netfree
,
1447 B_FALSE
, tr_list
, tx
, B_TRUE
);
1451 dsl_dir_tempreserve_clear(tr_list
, tx
);
1453 *tr_cookiep
= tr_list
;
1459 * Clear a temporary reservation that we previously made with
1460 * dsl_dir_tempreserve_space().
1463 dsl_dir_tempreserve_clear(void *tr_cookie
, dmu_tx_t
*tx
)
1465 int txgidx
= tx
->tx_txg
& TXG_MASK
;
1466 list_t
*tr_list
= tr_cookie
;
1467 struct tempreserve
*tr
;
1469 ASSERT3U(tx
->tx_txg
, !=, 0);
1471 if (tr_cookie
== NULL
)
1474 while ((tr
= list_head(tr_list
)) != NULL
) {
1476 mutex_enter(&tr
->tr_ds
->dd_lock
);
1477 ASSERT3U(tr
->tr_ds
->dd_tempreserved
[txgidx
], >=,
1479 tr
->tr_ds
->dd_tempreserved
[txgidx
] -= tr
->tr_size
;
1480 mutex_exit(&tr
->tr_ds
->dd_lock
);
1482 arc_tempreserve_clear(tr
->tr_size
);
1484 list_remove(tr_list
, tr
);
1485 kmem_free(tr
, sizeof (struct tempreserve
));
1488 kmem_free(tr_list
, sizeof (list_t
));
1492 * This should be called from open context when we think we're going to write
1493 * or free space, for example when dirtying data. Be conservative; it's okay
1494 * to write less space or free more, but we don't want to write more or free
1495 * less than the amount specified.
1497 * NOTE: The behavior of this function is identical to the Illumos / FreeBSD
1498 * version however it has been adjusted to use an iterative rather then
1499 * recursive algorithm to minimize stack usage.
1502 dsl_dir_willuse_space(dsl_dir_t
*dd
, int64_t space
, dmu_tx_t
*tx
)
1504 int64_t parent_space
;
1508 mutex_enter(&dd
->dd_lock
);
1510 dd
->dd_space_towrite
[tx
->tx_txg
& TXG_MASK
] += space
;
1512 est_used
= dsl_dir_space_towrite(dd
) +
1513 dsl_dir_phys(dd
)->dd_used_bytes
;
1514 parent_space
= parent_delta(dd
, est_used
, space
);
1515 mutex_exit(&dd
->dd_lock
);
1517 /* Make sure that we clean up dd_space_to* */
1518 dsl_dir_dirty(dd
, tx
);
1521 space
= parent_space
;
1522 } while (space
&& dd
);
1525 /* call from syncing context when we actually write/free space for this dd */
1527 dsl_dir_diduse_space(dsl_dir_t
*dd
, dd_used_t type
,
1528 int64_t used
, int64_t compressed
, int64_t uncompressed
, dmu_tx_t
*tx
)
1530 int64_t accounted_delta
;
1533 * dsl_dataset_set_refreservation_sync_impl() calls this with
1534 * dd_lock held, so that it can atomically update
1535 * ds->ds_reserved and the dsl_dir accounting, so that
1536 * dsl_dataset_check_quota() can see dataset and dir accounting
1539 boolean_t needlock
= !MUTEX_HELD(&dd
->dd_lock
);
1541 ASSERT(dmu_tx_is_syncing(tx
));
1542 ASSERT(type
< DD_USED_NUM
);
1544 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
1547 mutex_enter(&dd
->dd_lock
);
1549 parent_delta(dd
, dsl_dir_phys(dd
)->dd_used_bytes
, used
);
1550 ASSERT(used
>= 0 || dsl_dir_phys(dd
)->dd_used_bytes
>= -used
);
1551 ASSERT(compressed
>= 0 ||
1552 dsl_dir_phys(dd
)->dd_compressed_bytes
>= -compressed
);
1553 ASSERT(uncompressed
>= 0 ||
1554 dsl_dir_phys(dd
)->dd_uncompressed_bytes
>= -uncompressed
);
1555 dsl_dir_phys(dd
)->dd_used_bytes
+= used
;
1556 dsl_dir_phys(dd
)->dd_uncompressed_bytes
+= uncompressed
;
1557 dsl_dir_phys(dd
)->dd_compressed_bytes
+= compressed
;
1559 if (dsl_dir_phys(dd
)->dd_flags
& DD_FLAG_USED_BREAKDOWN
) {
1561 dsl_dir_phys(dd
)->dd_used_breakdown
[type
] >= -used
);
1562 dsl_dir_phys(dd
)->dd_used_breakdown
[type
] += used
;
1567 for (t
= 0; t
< DD_USED_NUM
; t
++)
1568 u
+= dsl_dir_phys(dd
)->dd_used_breakdown
[t
];
1569 ASSERT3U(u
, ==, dsl_dir_phys(dd
)->dd_used_bytes
);
1574 mutex_exit(&dd
->dd_lock
);
1576 if (dd
->dd_parent
!= NULL
) {
1577 dsl_dir_diduse_space(dd
->dd_parent
, DD_USED_CHILD
,
1578 accounted_delta
, compressed
, uncompressed
, tx
);
1579 dsl_dir_transfer_space(dd
->dd_parent
,
1580 used
- accounted_delta
,
1581 DD_USED_CHILD_RSRV
, DD_USED_CHILD
, tx
);
1586 dsl_dir_transfer_space(dsl_dir_t
*dd
, int64_t delta
,
1587 dd_used_t oldtype
, dd_used_t newtype
, dmu_tx_t
*tx
)
1589 ASSERT(dmu_tx_is_syncing(tx
));
1590 ASSERT(oldtype
< DD_USED_NUM
);
1591 ASSERT(newtype
< DD_USED_NUM
);
1594 !(dsl_dir_phys(dd
)->dd_flags
& DD_FLAG_USED_BREAKDOWN
))
1597 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
1598 mutex_enter(&dd
->dd_lock
);
1600 dsl_dir_phys(dd
)->dd_used_breakdown
[oldtype
] >= delta
:
1601 dsl_dir_phys(dd
)->dd_used_breakdown
[newtype
] >= -delta
);
1602 ASSERT(dsl_dir_phys(dd
)->dd_used_bytes
>= ABS(delta
));
1603 dsl_dir_phys(dd
)->dd_used_breakdown
[oldtype
] -= delta
;
1604 dsl_dir_phys(dd
)->dd_used_breakdown
[newtype
] += delta
;
1605 mutex_exit(&dd
->dd_lock
);
1608 typedef struct dsl_dir_set_qr_arg
{
1609 const char *ddsqra_name
;
1610 zprop_source_t ddsqra_source
;
1611 uint64_t ddsqra_value
;
1612 } dsl_dir_set_qr_arg_t
;
1615 dsl_dir_set_quota_check(void *arg
, dmu_tx_t
*tx
)
1617 dsl_dir_set_qr_arg_t
*ddsqra
= arg
;
1618 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1621 uint64_t towrite
, newval
;
1623 error
= dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
);
1627 error
= dsl_prop_predict(ds
->ds_dir
, "quota",
1628 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, &newval
);
1630 dsl_dataset_rele(ds
, FTAG
);
1635 dsl_dataset_rele(ds
, FTAG
);
1639 mutex_enter(&ds
->ds_dir
->dd_lock
);
1641 * If we are doing the preliminary check in open context, and
1642 * there are pending changes, then don't fail it, since the
1643 * pending changes could under-estimate the amount of space to be
1646 towrite
= dsl_dir_space_towrite(ds
->ds_dir
);
1647 if ((dmu_tx_is_syncing(tx
) || towrite
== 0) &&
1648 (newval
< dsl_dir_phys(ds
->ds_dir
)->dd_reserved
||
1649 newval
< dsl_dir_phys(ds
->ds_dir
)->dd_used_bytes
+ towrite
)) {
1650 error
= SET_ERROR(ENOSPC
);
1652 mutex_exit(&ds
->ds_dir
->dd_lock
);
1653 dsl_dataset_rele(ds
, FTAG
);
1658 dsl_dir_set_quota_sync(void *arg
, dmu_tx_t
*tx
)
1660 dsl_dir_set_qr_arg_t
*ddsqra
= arg
;
1661 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1665 VERIFY0(dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
));
1667 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_RECVD_PROPS
) {
1668 dsl_prop_set_sync_impl(ds
, zfs_prop_to_name(ZFS_PROP_QUOTA
),
1669 ddsqra
->ddsqra_source
, sizeof (ddsqra
->ddsqra_value
), 1,
1670 &ddsqra
->ddsqra_value
, tx
);
1672 VERIFY0(dsl_prop_get_int_ds(ds
,
1673 zfs_prop_to_name(ZFS_PROP_QUOTA
), &newval
));
1675 newval
= ddsqra
->ddsqra_value
;
1676 spa_history_log_internal_ds(ds
, "set", tx
, "%s=%lld",
1677 zfs_prop_to_name(ZFS_PROP_QUOTA
), (longlong_t
)newval
);
1680 dmu_buf_will_dirty(ds
->ds_dir
->dd_dbuf
, tx
);
1681 mutex_enter(&ds
->ds_dir
->dd_lock
);
1682 dsl_dir_phys(ds
->ds_dir
)->dd_quota
= newval
;
1683 mutex_exit(&ds
->ds_dir
->dd_lock
);
1684 dsl_dataset_rele(ds
, FTAG
);
1688 dsl_dir_set_quota(const char *ddname
, zprop_source_t source
, uint64_t quota
)
1690 dsl_dir_set_qr_arg_t ddsqra
;
1692 ddsqra
.ddsqra_name
= ddname
;
1693 ddsqra
.ddsqra_source
= source
;
1694 ddsqra
.ddsqra_value
= quota
;
1696 return (dsl_sync_task(ddname
, dsl_dir_set_quota_check
,
1697 dsl_dir_set_quota_sync
, &ddsqra
, 0,
1698 ZFS_SPACE_CHECK_EXTRA_RESERVED
));
1702 dsl_dir_set_reservation_check(void *arg
, dmu_tx_t
*tx
)
1704 dsl_dir_set_qr_arg_t
*ddsqra
= arg
;
1705 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1708 uint64_t newval
, used
, avail
;
1711 error
= dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
);
1717 * If we are doing the preliminary check in open context, the
1718 * space estimates may be inaccurate.
1720 if (!dmu_tx_is_syncing(tx
)) {
1721 dsl_dataset_rele(ds
, FTAG
);
1725 error
= dsl_prop_predict(ds
->ds_dir
,
1726 zfs_prop_to_name(ZFS_PROP_RESERVATION
),
1727 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, &newval
);
1729 dsl_dataset_rele(ds
, FTAG
);
1733 mutex_enter(&dd
->dd_lock
);
1734 used
= dsl_dir_phys(dd
)->dd_used_bytes
;
1735 mutex_exit(&dd
->dd_lock
);
1737 if (dd
->dd_parent
) {
1738 avail
= dsl_dir_space_available(dd
->dd_parent
,
1741 avail
= dsl_pool_adjustedsize(dd
->dd_pool
,
1742 ZFS_SPACE_CHECK_NORMAL
) - used
;
1745 if (MAX(used
, newval
) > MAX(used
, dsl_dir_phys(dd
)->dd_reserved
)) {
1746 uint64_t delta
= MAX(used
, newval
) -
1747 MAX(used
, dsl_dir_phys(dd
)->dd_reserved
);
1749 if (delta
> avail
||
1750 (dsl_dir_phys(dd
)->dd_quota
> 0 &&
1751 newval
> dsl_dir_phys(dd
)->dd_quota
))
1752 error
= SET_ERROR(ENOSPC
);
1755 dsl_dataset_rele(ds
, FTAG
);
1760 dsl_dir_set_reservation_sync_impl(dsl_dir_t
*dd
, uint64_t value
, dmu_tx_t
*tx
)
1765 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
1767 mutex_enter(&dd
->dd_lock
);
1768 used
= dsl_dir_phys(dd
)->dd_used_bytes
;
1769 delta
= MAX(used
, value
) - MAX(used
, dsl_dir_phys(dd
)->dd_reserved
);
1770 dsl_dir_phys(dd
)->dd_reserved
= value
;
1772 if (dd
->dd_parent
!= NULL
) {
1773 /* Roll up this additional usage into our ancestors */
1774 dsl_dir_diduse_space(dd
->dd_parent
, DD_USED_CHILD_RSRV
,
1777 mutex_exit(&dd
->dd_lock
);
1781 dsl_dir_set_reservation_sync(void *arg
, dmu_tx_t
*tx
)
1783 dsl_dir_set_qr_arg_t
*ddsqra
= arg
;
1784 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1788 VERIFY0(dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
));
1790 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_RECVD_PROPS
) {
1791 dsl_prop_set_sync_impl(ds
,
1792 zfs_prop_to_name(ZFS_PROP_RESERVATION
),
1793 ddsqra
->ddsqra_source
, sizeof (ddsqra
->ddsqra_value
), 1,
1794 &ddsqra
->ddsqra_value
, tx
);
1796 VERIFY0(dsl_prop_get_int_ds(ds
,
1797 zfs_prop_to_name(ZFS_PROP_RESERVATION
), &newval
));
1799 newval
= ddsqra
->ddsqra_value
;
1800 spa_history_log_internal_ds(ds
, "set", tx
, "%s=%lld",
1801 zfs_prop_to_name(ZFS_PROP_RESERVATION
),
1802 (longlong_t
)newval
);
1805 dsl_dir_set_reservation_sync_impl(ds
->ds_dir
, newval
, tx
);
1806 dsl_dataset_rele(ds
, FTAG
);
1810 dsl_dir_set_reservation(const char *ddname
, zprop_source_t source
,
1811 uint64_t reservation
)
1813 dsl_dir_set_qr_arg_t ddsqra
;
1815 ddsqra
.ddsqra_name
= ddname
;
1816 ddsqra
.ddsqra_source
= source
;
1817 ddsqra
.ddsqra_value
= reservation
;
1819 return (dsl_sync_task(ddname
, dsl_dir_set_reservation_check
,
1820 dsl_dir_set_reservation_sync
, &ddsqra
, 0,
1821 ZFS_SPACE_CHECK_EXTRA_RESERVED
));
1825 closest_common_ancestor(dsl_dir_t
*ds1
, dsl_dir_t
*ds2
)
1827 for (; ds1
; ds1
= ds1
->dd_parent
) {
1829 for (dd
= ds2
; dd
; dd
= dd
->dd_parent
) {
1838 * If delta is applied to dd, how much of that delta would be applied to
1839 * ancestor? Syncing context only.
1842 would_change(dsl_dir_t
*dd
, int64_t delta
, dsl_dir_t
*ancestor
)
1847 mutex_enter(&dd
->dd_lock
);
1848 delta
= parent_delta(dd
, dsl_dir_phys(dd
)->dd_used_bytes
, delta
);
1849 mutex_exit(&dd
->dd_lock
);
1850 return (would_change(dd
->dd_parent
, delta
, ancestor
));
1853 typedef struct dsl_dir_rename_arg
{
1854 const char *ddra_oldname
;
1855 const char *ddra_newname
;
1857 } dsl_dir_rename_arg_t
;
1859 typedef struct dsl_valid_rename_arg
{
1862 } dsl_valid_rename_arg_t
;
1866 dsl_valid_rename(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *arg
)
1868 dsl_valid_rename_arg_t
*dvra
= arg
;
1869 char namebuf
[ZFS_MAX_DATASET_NAME_LEN
];
1871 dsl_dataset_name(ds
, namebuf
);
1873 ASSERT3U(strnlen(namebuf
, ZFS_MAX_DATASET_NAME_LEN
),
1874 <, ZFS_MAX_DATASET_NAME_LEN
);
1875 int namelen
= strlen(namebuf
) + dvra
->char_delta
;
1876 int depth
= get_dataset_depth(namebuf
) + dvra
->nest_delta
;
1878 if (namelen
>= ZFS_MAX_DATASET_NAME_LEN
)
1879 return (SET_ERROR(ENAMETOOLONG
));
1880 if (dvra
->nest_delta
> 0 && depth
>= zfs_max_dataset_nesting
)
1881 return (SET_ERROR(ENAMETOOLONG
));
1886 dsl_dir_rename_check(void *arg
, dmu_tx_t
*tx
)
1888 dsl_dir_rename_arg_t
*ddra
= arg
;
1889 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1890 dsl_dir_t
*dd
, *newparent
;
1891 dsl_valid_rename_arg_t dvra
;
1892 dsl_dataset_t
*parentds
;
1894 const char *mynewname
;
1897 /* target dir should exist */
1898 error
= dsl_dir_hold(dp
, ddra
->ddra_oldname
, FTAG
, &dd
, NULL
);
1902 /* new parent should exist */
1903 error
= dsl_dir_hold(dp
, ddra
->ddra_newname
, FTAG
,
1904 &newparent
, &mynewname
);
1906 dsl_dir_rele(dd
, FTAG
);
1910 /* can't rename to different pool */
1911 if (dd
->dd_pool
!= newparent
->dd_pool
) {
1912 dsl_dir_rele(newparent
, FTAG
);
1913 dsl_dir_rele(dd
, FTAG
);
1914 return (SET_ERROR(EXDEV
));
1917 /* new name should not already exist */
1918 if (mynewname
== NULL
) {
1919 dsl_dir_rele(newparent
, FTAG
);
1920 dsl_dir_rele(dd
, FTAG
);
1921 return (SET_ERROR(EEXIST
));
1924 /* can't rename below anything but filesystems (eg. no ZVOLs) */
1925 error
= dsl_dataset_hold_obj(newparent
->dd_pool
,
1926 dsl_dir_phys(newparent
)->dd_head_dataset_obj
, FTAG
, &parentds
);
1928 dsl_dir_rele(newparent
, FTAG
);
1929 dsl_dir_rele(dd
, FTAG
);
1932 error
= dmu_objset_from_ds(parentds
, &parentos
);
1934 dsl_dataset_rele(parentds
, FTAG
);
1935 dsl_dir_rele(newparent
, FTAG
);
1936 dsl_dir_rele(dd
, FTAG
);
1939 if (dmu_objset_type(parentos
) != DMU_OST_ZFS
) {
1940 dsl_dataset_rele(parentds
, FTAG
);
1941 dsl_dir_rele(newparent
, FTAG
);
1942 dsl_dir_rele(dd
, FTAG
);
1943 return (SET_ERROR(ZFS_ERR_WRONG_PARENT
));
1945 dsl_dataset_rele(parentds
, FTAG
);
1947 ASSERT3U(strnlen(ddra
->ddra_newname
, ZFS_MAX_DATASET_NAME_LEN
),
1948 <, ZFS_MAX_DATASET_NAME_LEN
);
1949 ASSERT3U(strnlen(ddra
->ddra_oldname
, ZFS_MAX_DATASET_NAME_LEN
),
1950 <, ZFS_MAX_DATASET_NAME_LEN
);
1951 dvra
.char_delta
= strlen(ddra
->ddra_newname
)
1952 - strlen(ddra
->ddra_oldname
);
1953 dvra
.nest_delta
= get_dataset_depth(ddra
->ddra_newname
)
1954 - get_dataset_depth(ddra
->ddra_oldname
);
1956 /* if the name length is growing, validate child name lengths */
1957 if (dvra
.char_delta
> 0 || dvra
.nest_delta
> 0) {
1958 error
= dmu_objset_find_dp(dp
, dd
->dd_object
, dsl_valid_rename
,
1959 &dvra
, DS_FIND_CHILDREN
| DS_FIND_SNAPSHOTS
);
1961 dsl_dir_rele(newparent
, FTAG
);
1962 dsl_dir_rele(dd
, FTAG
);
1967 if (dmu_tx_is_syncing(tx
)) {
1968 if (spa_feature_is_active(dp
->dp_spa
,
1969 SPA_FEATURE_FS_SS_LIMIT
)) {
1971 * Although this is the check function and we don't
1972 * normally make on-disk changes in check functions,
1973 * we need to do that here.
1975 * Ensure this portion of the tree's counts have been
1976 * initialized in case the new parent has limits set.
1978 dsl_dir_init_fs_ss_count(dd
, tx
);
1982 if (newparent
!= dd
->dd_parent
) {
1983 /* is there enough space? */
1985 MAX(dsl_dir_phys(dd
)->dd_used_bytes
,
1986 dsl_dir_phys(dd
)->dd_reserved
);
1987 objset_t
*os
= dd
->dd_pool
->dp_meta_objset
;
1988 uint64_t fs_cnt
= 0;
1989 uint64_t ss_cnt
= 0;
1991 if (dsl_dir_is_zapified(dd
)) {
1994 err
= zap_lookup(os
, dd
->dd_object
,
1995 DD_FIELD_FILESYSTEM_COUNT
, sizeof (fs_cnt
), 1,
1997 if (err
!= ENOENT
&& err
!= 0) {
1998 dsl_dir_rele(newparent
, FTAG
);
1999 dsl_dir_rele(dd
, FTAG
);
2004 * have to add 1 for the filesystem itself that we're
2009 err
= zap_lookup(os
, dd
->dd_object
,
2010 DD_FIELD_SNAPSHOT_COUNT
, sizeof (ss_cnt
), 1,
2012 if (err
!= ENOENT
&& err
!= 0) {
2013 dsl_dir_rele(newparent
, FTAG
);
2014 dsl_dir_rele(dd
, FTAG
);
2019 /* check for encryption errors */
2020 error
= dsl_dir_rename_crypt_check(dd
, newparent
);
2022 dsl_dir_rele(newparent
, FTAG
);
2023 dsl_dir_rele(dd
, FTAG
);
2024 return (SET_ERROR(EACCES
));
2027 /* no rename into our descendant */
2028 if (closest_common_ancestor(dd
, newparent
) == dd
) {
2029 dsl_dir_rele(newparent
, FTAG
);
2030 dsl_dir_rele(dd
, FTAG
);
2031 return (SET_ERROR(EINVAL
));
2034 error
= dsl_dir_transfer_possible(dd
->dd_parent
,
2035 newparent
, fs_cnt
, ss_cnt
, myspace
, ddra
->ddra_cred
);
2037 dsl_dir_rele(newparent
, FTAG
);
2038 dsl_dir_rele(dd
, FTAG
);
2043 dsl_dir_rele(newparent
, FTAG
);
2044 dsl_dir_rele(dd
, FTAG
);
2049 dsl_dir_rename_sync(void *arg
, dmu_tx_t
*tx
)
2051 dsl_dir_rename_arg_t
*ddra
= arg
;
2052 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2053 dsl_dir_t
*dd
, *newparent
;
2054 const char *mynewname
;
2056 objset_t
*mos
= dp
->dp_meta_objset
;
2058 VERIFY0(dsl_dir_hold(dp
, ddra
->ddra_oldname
, FTAG
, &dd
, NULL
));
2059 VERIFY0(dsl_dir_hold(dp
, ddra
->ddra_newname
, FTAG
, &newparent
,
2062 /* Log this before we change the name. */
2063 spa_history_log_internal_dd(dd
, "rename", tx
,
2064 "-> %s", ddra
->ddra_newname
);
2066 if (newparent
!= dd
->dd_parent
) {
2067 objset_t
*os
= dd
->dd_pool
->dp_meta_objset
;
2068 uint64_t fs_cnt
= 0;
2069 uint64_t ss_cnt
= 0;
2072 * We already made sure the dd counts were initialized in the
2075 if (spa_feature_is_active(dp
->dp_spa
,
2076 SPA_FEATURE_FS_SS_LIMIT
)) {
2077 VERIFY0(zap_lookup(os
, dd
->dd_object
,
2078 DD_FIELD_FILESYSTEM_COUNT
, sizeof (fs_cnt
), 1,
2080 /* add 1 for the filesystem itself that we're moving */
2083 VERIFY0(zap_lookup(os
, dd
->dd_object
,
2084 DD_FIELD_SNAPSHOT_COUNT
, sizeof (ss_cnt
), 1,
2088 dsl_fs_ss_count_adjust(dd
->dd_parent
, -fs_cnt
,
2089 DD_FIELD_FILESYSTEM_COUNT
, tx
);
2090 dsl_fs_ss_count_adjust(newparent
, fs_cnt
,
2091 DD_FIELD_FILESYSTEM_COUNT
, tx
);
2093 dsl_fs_ss_count_adjust(dd
->dd_parent
, -ss_cnt
,
2094 DD_FIELD_SNAPSHOT_COUNT
, tx
);
2095 dsl_fs_ss_count_adjust(newparent
, ss_cnt
,
2096 DD_FIELD_SNAPSHOT_COUNT
, tx
);
2098 dsl_dir_diduse_space(dd
->dd_parent
, DD_USED_CHILD
,
2099 -dsl_dir_phys(dd
)->dd_used_bytes
,
2100 -dsl_dir_phys(dd
)->dd_compressed_bytes
,
2101 -dsl_dir_phys(dd
)->dd_uncompressed_bytes
, tx
);
2102 dsl_dir_diduse_space(newparent
, DD_USED_CHILD
,
2103 dsl_dir_phys(dd
)->dd_used_bytes
,
2104 dsl_dir_phys(dd
)->dd_compressed_bytes
,
2105 dsl_dir_phys(dd
)->dd_uncompressed_bytes
, tx
);
2107 if (dsl_dir_phys(dd
)->dd_reserved
>
2108 dsl_dir_phys(dd
)->dd_used_bytes
) {
2109 uint64_t unused_rsrv
= dsl_dir_phys(dd
)->dd_reserved
-
2110 dsl_dir_phys(dd
)->dd_used_bytes
;
2112 dsl_dir_diduse_space(dd
->dd_parent
, DD_USED_CHILD_RSRV
,
2113 -unused_rsrv
, 0, 0, tx
);
2114 dsl_dir_diduse_space(newparent
, DD_USED_CHILD_RSRV
,
2115 unused_rsrv
, 0, 0, tx
);
2119 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
2121 /* remove from old parent zapobj */
2122 error
= zap_remove(mos
,
2123 dsl_dir_phys(dd
->dd_parent
)->dd_child_dir_zapobj
,
2127 (void) strlcpy(dd
->dd_myname
, mynewname
,
2128 sizeof (dd
->dd_myname
));
2129 dsl_dir_rele(dd
->dd_parent
, dd
);
2130 dsl_dir_phys(dd
)->dd_parent_obj
= newparent
->dd_object
;
2131 VERIFY0(dsl_dir_hold_obj(dp
,
2132 newparent
->dd_object
, NULL
, dd
, &dd
->dd_parent
));
2134 /* add to new parent zapobj */
2135 VERIFY0(zap_add(mos
, dsl_dir_phys(newparent
)->dd_child_dir_zapobj
,
2136 dd
->dd_myname
, 8, 1, &dd
->dd_object
, tx
));
2138 zvol_rename_minors(dp
->dp_spa
, ddra
->ddra_oldname
,
2139 ddra
->ddra_newname
, B_TRUE
);
2141 dsl_prop_notify_all(dd
);
2143 dsl_dir_rele(newparent
, FTAG
);
2144 dsl_dir_rele(dd
, FTAG
);
2148 dsl_dir_rename(const char *oldname
, const char *newname
)
2150 dsl_dir_rename_arg_t ddra
;
2152 ddra
.ddra_oldname
= oldname
;
2153 ddra
.ddra_newname
= newname
;
2154 ddra
.ddra_cred
= CRED();
2156 return (dsl_sync_task(oldname
,
2157 dsl_dir_rename_check
, dsl_dir_rename_sync
, &ddra
,
2158 3, ZFS_SPACE_CHECK_RESERVED
));
2162 dsl_dir_transfer_possible(dsl_dir_t
*sdd
, dsl_dir_t
*tdd
,
2163 uint64_t fs_cnt
, uint64_t ss_cnt
, uint64_t space
, cred_t
*cr
)
2165 dsl_dir_t
*ancestor
;
2170 ancestor
= closest_common_ancestor(sdd
, tdd
);
2171 adelta
= would_change(sdd
, -space
, ancestor
);
2172 avail
= dsl_dir_space_available(tdd
, ancestor
, adelta
, FALSE
);
2174 return (SET_ERROR(ENOSPC
));
2176 err
= dsl_fs_ss_limit_check(tdd
, fs_cnt
, ZFS_PROP_FILESYSTEM_LIMIT
,
2180 err
= dsl_fs_ss_limit_check(tdd
, ss_cnt
, ZFS_PROP_SNAPSHOT_LIMIT
,
2189 dsl_dir_snap_cmtime(dsl_dir_t
*dd
)
2193 mutex_enter(&dd
->dd_lock
);
2194 t
= dd
->dd_snap_cmtime
;
2195 mutex_exit(&dd
->dd_lock
);
2201 dsl_dir_snap_cmtime_update(dsl_dir_t
*dd
)
2206 mutex_enter(&dd
->dd_lock
);
2207 dd
->dd_snap_cmtime
= t
;
2208 mutex_exit(&dd
->dd_lock
);
2212 dsl_dir_zapify(dsl_dir_t
*dd
, dmu_tx_t
*tx
)
2214 objset_t
*mos
= dd
->dd_pool
->dp_meta_objset
;
2215 dmu_object_zapify(mos
, dd
->dd_object
, DMU_OT_DSL_DIR
, tx
);
2219 dsl_dir_is_zapified(dsl_dir_t
*dd
)
2221 dmu_object_info_t doi
;
2223 dmu_object_info_from_db(dd
->dd_dbuf
, &doi
);
2224 return (doi
.doi_type
== DMU_OTN_ZAP_METADATA
);
2227 #if defined(_KERNEL)
2228 EXPORT_SYMBOL(dsl_dir_set_quota
);
2229 EXPORT_SYMBOL(dsl_dir_set_reservation
);