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, 2014 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.
30 #include <sys/dmu_objset.h>
31 #include <sys/dmu_tx.h>
32 #include <sys/dsl_dataset.h>
33 #include <sys/dsl_dir.h>
34 #include <sys/dsl_prop.h>
35 #include <sys/dsl_synctask.h>
36 #include <sys/dsl_deleg.h>
37 #include <sys/dmu_impl.h>
39 #include <sys/metaslab.h>
43 #include <sys/sunddi.h>
44 #include <sys/zfeature.h>
45 #include <sys/policy.h>
46 #include <sys/zfs_znode.h>
47 #include "zfs_namecheck.h"
51 * Filesystem and Snapshot Limits
52 * ------------------------------
54 * These limits are used to restrict the number of filesystems and/or snapshots
55 * that can be created at a given level in the tree or below. A typical
56 * use-case is with a delegated dataset where the administrator wants to ensure
57 * that a user within the zone is not creating too many additional filesystems
58 * or snapshots, even though they're not exceeding their space quota.
60 * The filesystem and snapshot counts are stored as extensible properties. This
61 * capability is controlled by a feature flag and must be enabled to be used.
62 * Once enabled, the feature is not active until the first limit is set. At
63 * that point, future operations to create/destroy filesystems or snapshots
64 * will validate and update the counts.
66 * Because the count properties will not exist before the feature is active,
67 * the counts are updated when a limit is first set on an uninitialized
68 * dsl_dir node in the tree (The filesystem/snapshot count on a node includes
69 * all of the nested filesystems/snapshots. Thus, a new leaf node has a
70 * filesystem count of 0 and a snapshot count of 0. Non-existent filesystem and
71 * snapshot count properties on a node indicate uninitialized counts on that
72 * node.) When first setting a limit on an uninitialized node, the code starts
73 * at the filesystem with the new limit and descends into all sub-filesystems
74 * to add the count properties.
76 * In practice this is lightweight since a limit is typically set when the
77 * filesystem is created and thus has no children. Once valid, changing the
78 * limit value won't require a re-traversal since the counts are already valid.
79 * When recursively fixing the counts, if a node with a limit is encountered
80 * during the descent, the counts are known to be valid and there is no need to
81 * descend into that filesystem's children. The counts on filesystems above the
82 * one with the new limit will still be uninitialized, unless a limit is
83 * eventually set on one of those filesystems. The counts are always recursively
84 * updated when a limit is set on a dataset, unless there is already a limit.
85 * When a new limit value is set on a filesystem with an existing limit, it is
86 * possible for the new limit to be less than the current count at that level
87 * since a user who can change the limit is also allowed to exceed the limit.
89 * Once the feature is active, then whenever a filesystem or snapshot is
90 * created, the code recurses up the tree, validating the new count against the
91 * limit at each initialized level. In practice, most levels will not have a
92 * limit set. If there is a limit at any initialized level up the tree, the
93 * check must pass or the creation will fail. Likewise, when a filesystem or
94 * snapshot is destroyed, the counts are recursively adjusted all the way up
95 * the initizized nodes in the tree. Renaming a filesystem into different point
96 * in the tree will first validate, then update the counts on each branch up to
97 * the common ancestor. A receive will also validate the counts and then update
100 * An exception to the above behavior is that the limit is not enforced if the
101 * user has permission to modify the limit. This is primarily so that
102 * recursive snapshots in the global zone always work. We want to prevent a
103 * denial-of-service in which a lower level delegated dataset could max out its
104 * limit and thus block recursive snapshots from being taken in the global zone.
105 * Because of this, it is possible for the snapshot count to be over the limit
106 * and snapshots taken in the global zone could cause a lower level dataset to
107 * hit or exceed its limit. The administrator taking the global zone recursive
108 * snapshot should be aware of this side-effect and behave accordingly.
109 * For consistency, the filesystem limit is also not enforced if the user can
112 * The filesystem and snapshot limits are validated by dsl_fs_ss_limit_check()
113 * and updated by dsl_fs_ss_count_adjust(). A new limit value is setup in
114 * dsl_dir_activate_fs_ss_limit() and the counts are adjusted, if necessary, by
115 * dsl_dir_init_fs_ss_count().
117 * There is a special case when we receive a filesystem that already exists. In
118 * this case a temporary clone name of %X is created (see dmu_recv_begin). We
119 * never update the filesystem counts for temporary clones.
121 * Likewise, we do not update the snapshot counts for temporary snapshots,
122 * such as those created by zfs diff.
125 extern inline dsl_dir_phys_t
*dsl_dir_phys(dsl_dir_t
*dd
);
127 static uint64_t dsl_dir_space_towrite(dsl_dir_t
*dd
);
130 dsl_dir_evict(void *dbu
)
133 dsl_pool_t
*dp
= dd
->dd_pool
;
138 for (t
= 0; t
< TXG_SIZE
; t
++) {
139 ASSERT(!txg_list_member(&dp
->dp_dirty_dirs
, dd
, t
));
140 ASSERT(dd
->dd_tempreserved
[t
] == 0);
141 ASSERT(dd
->dd_space_towrite
[t
] == 0);
145 dsl_dir_async_rele(dd
->dd_parent
, dd
);
147 spa_async_close(dd
->dd_pool
->dp_spa
, dd
);
150 * The props callback list should have been cleaned up by
153 list_destroy(&dd
->dd_prop_cbs
);
154 mutex_destroy(&dd
->dd_lock
);
155 kmem_free(dd
, sizeof (dsl_dir_t
));
159 dsl_dir_hold_obj(dsl_pool_t
*dp
, uint64_t ddobj
,
160 const char *tail
, void *tag
, dsl_dir_t
**ddp
)
166 ASSERT(dsl_pool_config_held(dp
));
168 err
= dmu_bonus_hold(dp
->dp_meta_objset
, ddobj
, tag
, &dbuf
);
171 dd
= dmu_buf_get_user(dbuf
);
174 dmu_object_info_t doi
;
175 dmu_object_info_from_db(dbuf
, &doi
);
176 ASSERT3U(doi
.doi_bonus_type
, ==, DMU_OT_DSL_DIR
);
177 ASSERT3U(doi
.doi_bonus_size
, >=, sizeof (dsl_dir_phys_t
));
183 dd
= kmem_zalloc(sizeof (dsl_dir_t
), KM_SLEEP
);
184 dd
->dd_object
= ddobj
;
187 mutex_init(&dd
->dd_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
189 list_create(&dd
->dd_prop_cbs
, sizeof (dsl_prop_cb_record_t
),
190 offsetof(dsl_prop_cb_record_t
, cbr_node
));
192 dsl_dir_snap_cmtime_update(dd
);
194 if (dsl_dir_phys(dd
)->dd_parent_obj
) {
195 err
= dsl_dir_hold_obj(dp
,
196 dsl_dir_phys(dd
)->dd_parent_obj
, NULL
, dd
,
204 err
= zap_lookup(dp
->dp_meta_objset
,
205 dsl_dir_phys(dd
->dd_parent
)->
206 dd_child_dir_zapobj
, tail
,
207 sizeof (foundobj
), 1, &foundobj
);
208 ASSERT(err
|| foundobj
== ddobj
);
210 (void) strcpy(dd
->dd_myname
, tail
);
212 err
= zap_value_search(dp
->dp_meta_objset
,
213 dsl_dir_phys(dd
->dd_parent
)->
215 ddobj
, 0, dd
->dd_myname
);
220 (void) strcpy(dd
->dd_myname
, spa_name(dp
->dp_spa
));
223 if (dsl_dir_is_clone(dd
)) {
224 dmu_buf_t
*origin_bonus
;
225 dsl_dataset_phys_t
*origin_phys
;
228 * We can't open the origin dataset, because
229 * that would require opening this dsl_dir.
230 * Just look at its phys directly instead.
232 err
= dmu_bonus_hold(dp
->dp_meta_objset
,
233 dsl_dir_phys(dd
)->dd_origin_obj
, FTAG
,
237 origin_phys
= origin_bonus
->db_data
;
239 origin_phys
->ds_creation_txg
;
240 dmu_buf_rele(origin_bonus
, FTAG
);
243 dmu_buf_init_user(&dd
->dd_dbu
, dsl_dir_evict
, &dd
->dd_dbuf
);
244 winner
= dmu_buf_set_user_ie(dbuf
, &dd
->dd_dbu
);
245 if (winner
!= NULL
) {
247 dsl_dir_rele(dd
->dd_parent
, dd
);
248 mutex_destroy(&dd
->dd_lock
);
249 kmem_free(dd
, sizeof (dsl_dir_t
));
252 spa_open_ref(dp
->dp_spa
, dd
);
257 * The dsl_dir_t has both open-to-close and instantiate-to-evict
258 * holds on the spa. We need the open-to-close holds because
259 * otherwise the spa_refcnt wouldn't change when we open a
260 * dir which the spa also has open, so we could incorrectly
261 * think it was OK to unload/export/destroy the pool. We need
262 * the instantiate-to-evict hold because the dsl_dir_t has a
263 * pointer to the dd_pool, which has a pointer to the spa_t.
265 spa_open_ref(dp
->dp_spa
, tag
);
266 ASSERT3P(dd
->dd_pool
, ==, dp
);
267 ASSERT3U(dd
->dd_object
, ==, ddobj
);
268 ASSERT3P(dd
->dd_dbuf
, ==, dbuf
);
274 dsl_dir_rele(dd
->dd_parent
, dd
);
275 mutex_destroy(&dd
->dd_lock
);
276 kmem_free(dd
, sizeof (dsl_dir_t
));
277 dmu_buf_rele(dbuf
, tag
);
282 dsl_dir_rele(dsl_dir_t
*dd
, void *tag
)
284 dprintf_dd(dd
, "%s\n", "");
285 spa_close(dd
->dd_pool
->dp_spa
, tag
);
286 dmu_buf_rele(dd
->dd_dbuf
, tag
);
290 * Remove a reference to the given dsl dir that is being asynchronously
291 * released. Async releases occur from a taskq performing eviction of
292 * dsl datasets and dirs. This process is identical to a normal release
293 * with the exception of using the async API for releasing the reference on
297 dsl_dir_async_rele(dsl_dir_t
*dd
, void *tag
)
299 dprintf_dd(dd
, "%s\n", "");
300 spa_async_close(dd
->dd_pool
->dp_spa
, tag
);
301 dmu_buf_rele(dd
->dd_dbuf
, tag
);
304 /* buf must be long enough (MAXNAMELEN + strlen(MOS_DIR_NAME) + 1 should do) */
306 dsl_dir_name(dsl_dir_t
*dd
, char *buf
)
309 dsl_dir_name(dd
->dd_parent
, buf
);
310 (void) strcat(buf
, "/");
314 if (!MUTEX_HELD(&dd
->dd_lock
)) {
316 * recursive mutex so that we can use
317 * dprintf_dd() with dd_lock held
319 mutex_enter(&dd
->dd_lock
);
320 (void) strcat(buf
, dd
->dd_myname
);
321 mutex_exit(&dd
->dd_lock
);
323 (void) strcat(buf
, dd
->dd_myname
);
327 /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
329 dsl_dir_namelen(dsl_dir_t
*dd
)
334 /* parent's name + 1 for the "/" */
335 result
= dsl_dir_namelen(dd
->dd_parent
) + 1;
338 if (!MUTEX_HELD(&dd
->dd_lock
)) {
339 /* see dsl_dir_name */
340 mutex_enter(&dd
->dd_lock
);
341 result
+= strlen(dd
->dd_myname
);
342 mutex_exit(&dd
->dd_lock
);
344 result
+= strlen(dd
->dd_myname
);
351 getcomponent(const char *path
, char *component
, const char **nextp
)
355 if ((path
== NULL
) || (path
[0] == '\0'))
356 return (SET_ERROR(ENOENT
));
357 /* This would be a good place to reserve some namespace... */
358 p
= strpbrk(path
, "/@");
359 if (p
&& (p
[1] == '/' || p
[1] == '@')) {
360 /* two separators in a row */
361 return (SET_ERROR(EINVAL
));
363 if (p
== NULL
|| p
== path
) {
365 * if the first thing is an @ or /, it had better be an
366 * @ and it had better not have any more ats or slashes,
367 * and it had better have something after the @.
370 (p
[0] != '@' || strpbrk(path
+1, "/@") || p
[1] == '\0'))
371 return (SET_ERROR(EINVAL
));
372 if (strlen(path
) >= MAXNAMELEN
)
373 return (SET_ERROR(ENAMETOOLONG
));
374 (void) strcpy(component
, path
);
376 } else if (p
[0] == '/') {
377 if (p
- path
>= MAXNAMELEN
)
378 return (SET_ERROR(ENAMETOOLONG
));
379 (void) strncpy(component
, path
, p
- path
);
380 component
[p
- path
] = '\0';
382 } else if (p
[0] == '@') {
384 * if the next separator is an @, there better not be
387 if (strchr(path
, '/'))
388 return (SET_ERROR(EINVAL
));
389 if (p
- path
>= MAXNAMELEN
)
390 return (SET_ERROR(ENAMETOOLONG
));
391 (void) strncpy(component
, path
, p
- path
);
392 component
[p
- path
] = '\0';
394 panic("invalid p=%p", (void *)p
);
401 * Return the dsl_dir_t, and possibly the last component which couldn't
402 * be found in *tail. The name must be in the specified dsl_pool_t. This
403 * thread must hold the dp_config_rwlock for the pool. Returns NULL if the
404 * path is bogus, or if tail==NULL and we couldn't parse the whole name.
405 * (*tail)[0] == '@' means that the last component is a snapshot.
408 dsl_dir_hold(dsl_pool_t
*dp
, const char *name
, void *tag
,
409 dsl_dir_t
**ddp
, const char **tailp
)
411 char buf
[MAXNAMELEN
];
412 const char *spaname
, *next
, *nextnext
= NULL
;
417 err
= getcomponent(name
, buf
, &next
);
421 /* Make sure the name is in the specified pool. */
422 spaname
= spa_name(dp
->dp_spa
);
423 if (strcmp(buf
, spaname
) != 0)
424 return (SET_ERROR(EINVAL
));
426 ASSERT(dsl_pool_config_held(dp
));
428 err
= dsl_dir_hold_obj(dp
, dp
->dp_root_dir_obj
, NULL
, tag
, &dd
);
433 while (next
!= NULL
) {
435 err
= getcomponent(next
, buf
, &nextnext
);
438 ASSERT(next
[0] != '\0');
441 dprintf("looking up %s in obj%lld\n",
442 buf
, dsl_dir_phys(dd
)->dd_child_dir_zapobj
);
444 err
= zap_lookup(dp
->dp_meta_objset
,
445 dsl_dir_phys(dd
)->dd_child_dir_zapobj
,
446 buf
, sizeof (ddobj
), 1, &ddobj
);
453 err
= dsl_dir_hold_obj(dp
, ddobj
, buf
, tag
, &child_dd
);
456 dsl_dir_rele(dd
, tag
);
462 dsl_dir_rele(dd
, tag
);
467 * It's an error if there's more than one component left, or
468 * tailp==NULL and there's any component left.
471 (tailp
== NULL
|| (nextnext
&& nextnext
[0] != '\0'))) {
473 dsl_dir_rele(dd
, tag
);
474 dprintf("next=%p (%s) tail=%p\n", next
, next
?next
:"", tailp
);
475 err
= SET_ERROR(ENOENT
);
484 * If the counts are already initialized for this filesystem and its
485 * descendants then do nothing, otherwise initialize the counts.
487 * The counts on this filesystem, and those below, may be uninitialized due to
488 * either the use of a pre-existing pool which did not support the
489 * filesystem/snapshot limit feature, or one in which the feature had not yet
492 * Recursively descend the filesystem tree and update the filesystem/snapshot
493 * counts on each filesystem below, then update the cumulative count on the
494 * current filesystem. If the filesystem already has a count set on it,
495 * then we know that its counts, and the counts on the filesystems below it,
496 * are already correct, so we don't have to update this filesystem.
499 dsl_dir_init_fs_ss_count(dsl_dir_t
*dd
, dmu_tx_t
*tx
)
501 uint64_t my_fs_cnt
= 0;
502 uint64_t my_ss_cnt
= 0;
503 dsl_pool_t
*dp
= dd
->dd_pool
;
504 objset_t
*os
= dp
->dp_meta_objset
;
509 ASSERT(spa_feature_is_active(dp
->dp_spa
, SPA_FEATURE_FS_SS_LIMIT
));
510 ASSERT(dsl_pool_config_held(dp
));
511 ASSERT(dmu_tx_is_syncing(tx
));
513 dsl_dir_zapify(dd
, tx
);
516 * If the filesystem count has already been initialized then we
517 * don't need to recurse down any further.
519 if (zap_contains(os
, dd
->dd_object
, DD_FIELD_FILESYSTEM_COUNT
) == 0)
522 zc
= kmem_alloc(sizeof (zap_cursor_t
), KM_SLEEP
);
523 za
= kmem_alloc(sizeof (zap_attribute_t
), KM_SLEEP
);
525 /* Iterate my child dirs */
526 for (zap_cursor_init(zc
, os
, dsl_dir_phys(dd
)->dd_child_dir_zapobj
);
527 zap_cursor_retrieve(zc
, za
) == 0; zap_cursor_advance(zc
)) {
531 VERIFY0(dsl_dir_hold_obj(dp
, za
->za_first_integer
, NULL
, FTAG
,
535 * Ignore hidden ($FREE, $MOS & $ORIGIN) objsets and
536 * temporary datasets.
538 if (chld_dd
->dd_myname
[0] == '$' ||
539 chld_dd
->dd_myname
[0] == '%') {
540 dsl_dir_rele(chld_dd
, FTAG
);
544 my_fs_cnt
++; /* count this child */
546 dsl_dir_init_fs_ss_count(chld_dd
, tx
);
548 VERIFY0(zap_lookup(os
, chld_dd
->dd_object
,
549 DD_FIELD_FILESYSTEM_COUNT
, sizeof (count
), 1, &count
));
551 VERIFY0(zap_lookup(os
, chld_dd
->dd_object
,
552 DD_FIELD_SNAPSHOT_COUNT
, sizeof (count
), 1, &count
));
555 dsl_dir_rele(chld_dd
, FTAG
);
558 /* Count my snapshots (we counted children's snapshots above) */
559 VERIFY0(dsl_dataset_hold_obj(dd
->dd_pool
,
560 dsl_dir_phys(dd
)->dd_head_dataset_obj
, FTAG
, &ds
));
562 for (zap_cursor_init(zc
, os
, dsl_dataset_phys(ds
)->ds_snapnames_zapobj
);
563 zap_cursor_retrieve(zc
, za
) == 0;
564 zap_cursor_advance(zc
)) {
565 /* Don't count temporary snapshots */
566 if (za
->za_name
[0] != '%')
571 dsl_dataset_rele(ds
, FTAG
);
573 kmem_free(zc
, sizeof (zap_cursor_t
));
574 kmem_free(za
, sizeof (zap_attribute_t
));
576 /* we're in a sync task, update counts */
577 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
578 VERIFY0(zap_add(os
, dd
->dd_object
, DD_FIELD_FILESYSTEM_COUNT
,
579 sizeof (my_fs_cnt
), 1, &my_fs_cnt
, tx
));
580 VERIFY0(zap_add(os
, dd
->dd_object
, DD_FIELD_SNAPSHOT_COUNT
,
581 sizeof (my_ss_cnt
), 1, &my_ss_cnt
, tx
));
585 dsl_dir_actv_fs_ss_limit_check(void *arg
, dmu_tx_t
*tx
)
587 char *ddname
= (char *)arg
;
588 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
593 error
= dsl_dataset_hold(dp
, ddname
, FTAG
, &ds
);
597 if (!spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_FS_SS_LIMIT
)) {
598 dsl_dataset_rele(ds
, FTAG
);
599 return (SET_ERROR(ENOTSUP
));
603 if (spa_feature_is_active(dp
->dp_spa
, SPA_FEATURE_FS_SS_LIMIT
) &&
604 dsl_dir_is_zapified(dd
) &&
605 zap_contains(dp
->dp_meta_objset
, dd
->dd_object
,
606 DD_FIELD_FILESYSTEM_COUNT
) == 0) {
607 dsl_dataset_rele(ds
, FTAG
);
608 return (SET_ERROR(EALREADY
));
611 dsl_dataset_rele(ds
, FTAG
);
616 dsl_dir_actv_fs_ss_limit_sync(void *arg
, dmu_tx_t
*tx
)
618 char *ddname
= (char *)arg
;
619 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
623 VERIFY0(dsl_dataset_hold(dp
, ddname
, FTAG
, &ds
));
625 spa
= dsl_dataset_get_spa(ds
);
627 if (!spa_feature_is_active(spa
, SPA_FEATURE_FS_SS_LIMIT
)) {
629 * Since the feature was not active and we're now setting a
630 * limit, increment the feature-active counter so that the
631 * feature becomes active for the first time.
633 * We are already in a sync task so we can update the MOS.
635 spa_feature_incr(spa
, SPA_FEATURE_FS_SS_LIMIT
, tx
);
639 * Since we are now setting a non-UINT64_MAX limit on the filesystem,
640 * we need to ensure the counts are correct. Descend down the tree from
641 * this point and update all of the counts to be accurate.
643 dsl_dir_init_fs_ss_count(ds
->ds_dir
, tx
);
645 dsl_dataset_rele(ds
, FTAG
);
649 * Make sure the feature is enabled and activate it if necessary.
650 * Since we're setting a limit, ensure the on-disk counts are valid.
651 * This is only called by the ioctl path when setting a limit value.
653 * We do not need to validate the new limit, since users who can change the
654 * limit are also allowed to exceed the limit.
657 dsl_dir_activate_fs_ss_limit(const char *ddname
)
661 error
= dsl_sync_task(ddname
, dsl_dir_actv_fs_ss_limit_check
,
662 dsl_dir_actv_fs_ss_limit_sync
, (void *)ddname
, 0,
663 ZFS_SPACE_CHECK_RESERVED
);
665 if (error
== EALREADY
)
672 * Used to determine if the filesystem_limit or snapshot_limit should be
673 * enforced. We allow the limit to be exceeded if the user has permission to
674 * write the property value. We pass in the creds that we got in the open
675 * context since we will always be the GZ root in syncing context. We also have
676 * to handle the case where we are allowed to change the limit on the current
677 * dataset, but there may be another limit in the tree above.
679 * We can never modify these two properties within a non-global zone. In
680 * addition, the other checks are modeled on zfs_secpolicy_write_perms. We
681 * can't use that function since we are already holding the dp_config_rwlock.
682 * In addition, we already have the dd and dealing with snapshots is simplified
693 dsl_enforce_ds_ss_limits(dsl_dir_t
*dd
, zfs_prop_t prop
, cred_t
*cr
)
695 enforce_res_t enforce
= ENFORCE_ALWAYS
;
700 ASSERT(prop
== ZFS_PROP_FILESYSTEM_LIMIT
||
701 prop
== ZFS_PROP_SNAPSHOT_LIMIT
);
704 if (crgetzoneid(cr
) != GLOBAL_ZONEID
)
705 return (ENFORCE_ALWAYS
);
707 if (secpolicy_zfs(cr
) == 0)
708 return (ENFORCE_NEVER
);
711 if ((obj
= dsl_dir_phys(dd
)->dd_head_dataset_obj
) == 0)
712 return (ENFORCE_ALWAYS
);
714 ASSERT(dsl_pool_config_held(dd
->dd_pool
));
716 if (dsl_dataset_hold_obj(dd
->dd_pool
, obj
, FTAG
, &ds
) != 0)
717 return (ENFORCE_ALWAYS
);
719 if (dsl_prop_get_ds(ds
, "zoned", 8, 1, &zoned
, NULL
) || zoned
) {
720 /* Only root can access zoned fs's from the GZ */
721 enforce
= ENFORCE_ALWAYS
;
723 if (dsl_deleg_access_impl(ds
, zfs_prop_to_name(prop
), cr
) == 0)
724 enforce
= ENFORCE_ABOVE
;
727 dsl_dataset_rele(ds
, FTAG
);
732 * Check if adding additional child filesystem(s) would exceed any filesystem
733 * limits or adding additional snapshot(s) would exceed any snapshot limits.
734 * The prop argument indicates which limit to check.
736 * Note that all filesystem limits up to the root (or the highest
737 * initialized) filesystem or the given ancestor must be satisfied.
740 dsl_fs_ss_limit_check(dsl_dir_t
*dd
, uint64_t delta
, zfs_prop_t prop
,
741 dsl_dir_t
*ancestor
, cred_t
*cr
)
743 objset_t
*os
= dd
->dd_pool
->dp_meta_objset
;
744 uint64_t limit
, count
;
746 enforce_res_t enforce
;
749 ASSERT(dsl_pool_config_held(dd
->dd_pool
));
750 ASSERT(prop
== ZFS_PROP_FILESYSTEM_LIMIT
||
751 prop
== ZFS_PROP_SNAPSHOT_LIMIT
);
754 * If we're allowed to change the limit, don't enforce the limit
755 * e.g. this can happen if a snapshot is taken by an administrative
756 * user in the global zone (i.e. a recursive snapshot by root).
757 * However, we must handle the case of delegated permissions where we
758 * are allowed to change the limit on the current dataset, but there
759 * is another limit in the tree above.
761 enforce
= dsl_enforce_ds_ss_limits(dd
, prop
, cr
);
762 if (enforce
== ENFORCE_NEVER
)
766 * e.g. if renaming a dataset with no snapshots, count adjustment
772 if (prop
== ZFS_PROP_SNAPSHOT_LIMIT
) {
774 * We don't enforce the limit for temporary snapshots. This is
775 * indicated by a NULL cred_t argument.
780 count_prop
= DD_FIELD_SNAPSHOT_COUNT
;
782 count_prop
= DD_FIELD_FILESYSTEM_COUNT
;
786 * If an ancestor has been provided, stop checking the limit once we
787 * hit that dir. We need this during rename so that we don't overcount
788 * the check once we recurse up to the common ancestor.
794 * If we hit an uninitialized node while recursing up the tree, we can
795 * stop since we know there is no limit here (or above). The counts are
796 * not valid on this node and we know we won't touch this node's counts.
798 if (!dsl_dir_is_zapified(dd
) || zap_lookup(os
, dd
->dd_object
,
799 count_prop
, sizeof (count
), 1, &count
) == ENOENT
)
802 err
= dsl_prop_get_dd(dd
, zfs_prop_to_name(prop
), 8, 1, &limit
, NULL
,
807 /* Is there a limit which we've hit? */
808 if (enforce
== ENFORCE_ALWAYS
&& (count
+ delta
) > limit
)
809 return (SET_ERROR(EDQUOT
));
811 if (dd
->dd_parent
!= NULL
)
812 err
= dsl_fs_ss_limit_check(dd
->dd_parent
, delta
, prop
,
819 * Adjust the filesystem or snapshot count for the specified dsl_dir_t and all
820 * parents. When a new filesystem/snapshot is created, increment the count on
821 * all parents, and when a filesystem/snapshot is destroyed, decrement the
825 dsl_fs_ss_count_adjust(dsl_dir_t
*dd
, int64_t delta
, const char *prop
,
829 objset_t
*os
= dd
->dd_pool
->dp_meta_objset
;
832 ASSERT(dsl_pool_config_held(dd
->dd_pool
));
833 ASSERT(dmu_tx_is_syncing(tx
));
834 ASSERT(strcmp(prop
, DD_FIELD_FILESYSTEM_COUNT
) == 0 ||
835 strcmp(prop
, DD_FIELD_SNAPSHOT_COUNT
) == 0);
838 * When we receive an incremental stream into a filesystem that already
839 * exists, a temporary clone is created. We don't count this temporary
840 * clone, whose name begins with a '%'. We also ignore hidden ($FREE,
841 * $MOS & $ORIGIN) objsets.
843 if ((dd
->dd_myname
[0] == '%' || dd
->dd_myname
[0] == '$') &&
844 strcmp(prop
, DD_FIELD_FILESYSTEM_COUNT
) == 0)
848 * e.g. if renaming a dataset with no snapshots, count adjustment is 0
854 * If we hit an uninitialized node while recursing up the tree, we can
855 * stop since we know the counts are not valid on this node and we
856 * know we shouldn't touch this node's counts. An uninitialized count
857 * on the node indicates that either the feature has not yet been
858 * activated or there are no limits on this part of the tree.
860 if (!dsl_dir_is_zapified(dd
) || (err
= zap_lookup(os
, dd
->dd_object
,
861 prop
, sizeof (count
), 1, &count
)) == ENOENT
)
866 /* Use a signed verify to make sure we're not neg. */
867 VERIFY3S(count
, >=, 0);
869 VERIFY0(zap_update(os
, dd
->dd_object
, prop
, sizeof (count
), 1, &count
,
872 /* Roll up this additional count into our ancestors */
873 if (dd
->dd_parent
!= NULL
)
874 dsl_fs_ss_count_adjust(dd
->dd_parent
, delta
, prop
, tx
);
878 dsl_dir_create_sync(dsl_pool_t
*dp
, dsl_dir_t
*pds
, const char *name
,
881 objset_t
*mos
= dp
->dp_meta_objset
;
883 dsl_dir_phys_t
*ddphys
;
886 ddobj
= dmu_object_alloc(mos
, DMU_OT_DSL_DIR
, 0,
887 DMU_OT_DSL_DIR
, sizeof (dsl_dir_phys_t
), tx
);
889 VERIFY(0 == zap_add(mos
, dsl_dir_phys(pds
)->dd_child_dir_zapobj
,
890 name
, sizeof (uint64_t), 1, &ddobj
, tx
));
892 /* it's the root dir */
893 VERIFY(0 == zap_add(mos
, DMU_POOL_DIRECTORY_OBJECT
,
894 DMU_POOL_ROOT_DATASET
, sizeof (uint64_t), 1, &ddobj
, tx
));
896 VERIFY(0 == dmu_bonus_hold(mos
, ddobj
, FTAG
, &dbuf
));
897 dmu_buf_will_dirty(dbuf
, tx
);
898 ddphys
= dbuf
->db_data
;
900 ddphys
->dd_creation_time
= gethrestime_sec();
902 ddphys
->dd_parent_obj
= pds
->dd_object
;
904 /* update the filesystem counts */
905 dsl_fs_ss_count_adjust(pds
, 1, DD_FIELD_FILESYSTEM_COUNT
, tx
);
907 ddphys
->dd_props_zapobj
= zap_create(mos
,
908 DMU_OT_DSL_PROPS
, DMU_OT_NONE
, 0, tx
);
909 ddphys
->dd_child_dir_zapobj
= zap_create(mos
,
910 DMU_OT_DSL_DIR_CHILD_MAP
, DMU_OT_NONE
, 0, tx
);
911 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_USED_BREAKDOWN
)
912 ddphys
->dd_flags
|= DD_FLAG_USED_BREAKDOWN
;
913 dmu_buf_rele(dbuf
, FTAG
);
919 dsl_dir_is_clone(dsl_dir_t
*dd
)
921 return (dsl_dir_phys(dd
)->dd_origin_obj
&&
922 (dd
->dd_pool
->dp_origin_snap
== NULL
||
923 dsl_dir_phys(dd
)->dd_origin_obj
!=
924 dd
->dd_pool
->dp_origin_snap
->ds_object
));
928 dsl_dir_stats(dsl_dir_t
*dd
, nvlist_t
*nv
)
930 mutex_enter(&dd
->dd_lock
);
931 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USED
,
932 dsl_dir_phys(dd
)->dd_used_bytes
);
933 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_QUOTA
,
934 dsl_dir_phys(dd
)->dd_quota
);
935 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_RESERVATION
,
936 dsl_dir_phys(dd
)->dd_reserved
);
937 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_COMPRESSRATIO
,
938 dsl_dir_phys(dd
)->dd_compressed_bytes
== 0 ? 100 :
939 (dsl_dir_phys(dd
)->dd_uncompressed_bytes
* 100 /
940 dsl_dir_phys(dd
)->dd_compressed_bytes
));
941 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_LOGICALUSED
,
942 dsl_dir_phys(dd
)->dd_uncompressed_bytes
);
943 if (dsl_dir_phys(dd
)->dd_flags
& DD_FLAG_USED_BREAKDOWN
) {
944 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USEDSNAP
,
945 dsl_dir_phys(dd
)->dd_used_breakdown
[DD_USED_SNAP
]);
946 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USEDDS
,
947 dsl_dir_phys(dd
)->dd_used_breakdown
[DD_USED_HEAD
]);
948 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USEDREFRESERV
,
949 dsl_dir_phys(dd
)->dd_used_breakdown
[DD_USED_REFRSRV
]);
950 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USEDCHILD
,
951 dsl_dir_phys(dd
)->dd_used_breakdown
[DD_USED_CHILD
] +
952 dsl_dir_phys(dd
)->dd_used_breakdown
[DD_USED_CHILD_RSRV
]);
954 mutex_exit(&dd
->dd_lock
);
956 if (dsl_dir_is_zapified(dd
)) {
958 objset_t
*os
= dd
->dd_pool
->dp_meta_objset
;
960 if (zap_lookup(os
, dd
->dd_object
, DD_FIELD_FILESYSTEM_COUNT
,
961 sizeof (count
), 1, &count
) == 0) {
962 dsl_prop_nvlist_add_uint64(nv
,
963 ZFS_PROP_FILESYSTEM_COUNT
, count
);
965 if (zap_lookup(os
, dd
->dd_object
, DD_FIELD_SNAPSHOT_COUNT
,
966 sizeof (count
), 1, &count
) == 0) {
967 dsl_prop_nvlist_add_uint64(nv
,
968 ZFS_PROP_SNAPSHOT_COUNT
, count
);
972 if (dsl_dir_is_clone(dd
)) {
974 char buf
[MAXNAMELEN
];
976 VERIFY0(dsl_dataset_hold_obj(dd
->dd_pool
,
977 dsl_dir_phys(dd
)->dd_origin_obj
, FTAG
, &ds
));
978 dsl_dataset_name(ds
, buf
);
979 dsl_dataset_rele(ds
, FTAG
);
980 dsl_prop_nvlist_add_string(nv
, ZFS_PROP_ORIGIN
, buf
);
985 dsl_dir_dirty(dsl_dir_t
*dd
, dmu_tx_t
*tx
)
987 dsl_pool_t
*dp
= dd
->dd_pool
;
989 ASSERT(dsl_dir_phys(dd
));
991 if (txg_list_add(&dp
->dp_dirty_dirs
, dd
, tx
->tx_txg
)) {
992 /* up the hold count until we can be written out */
993 dmu_buf_add_ref(dd
->dd_dbuf
, dd
);
998 parent_delta(dsl_dir_t
*dd
, uint64_t used
, int64_t delta
)
1000 uint64_t old_accounted
= MAX(used
, dsl_dir_phys(dd
)->dd_reserved
);
1001 uint64_t new_accounted
=
1002 MAX(used
+ delta
, dsl_dir_phys(dd
)->dd_reserved
);
1003 return (new_accounted
- old_accounted
);
1007 dsl_dir_sync(dsl_dir_t
*dd
, dmu_tx_t
*tx
)
1009 ASSERT(dmu_tx_is_syncing(tx
));
1011 mutex_enter(&dd
->dd_lock
);
1012 ASSERT0(dd
->dd_tempreserved
[tx
->tx_txg
&TXG_MASK
]);
1013 dprintf_dd(dd
, "txg=%llu towrite=%lluK\n", tx
->tx_txg
,
1014 dd
->dd_space_towrite
[tx
->tx_txg
&TXG_MASK
] / 1024);
1015 dd
->dd_space_towrite
[tx
->tx_txg
&TXG_MASK
] = 0;
1016 mutex_exit(&dd
->dd_lock
);
1018 /* release the hold from dsl_dir_dirty */
1019 dmu_buf_rele(dd
->dd_dbuf
, dd
);
1023 dsl_dir_space_towrite(dsl_dir_t
*dd
)
1028 ASSERT(MUTEX_HELD(&dd
->dd_lock
));
1030 for (i
= 0; i
< TXG_SIZE
; i
++) {
1031 space
+= dd
->dd_space_towrite
[i
&TXG_MASK
];
1032 ASSERT3U(dd
->dd_space_towrite
[i
&TXG_MASK
], >=, 0);
1038 * How much space would dd have available if ancestor had delta applied
1039 * to it? If ondiskonly is set, we're only interested in what's
1040 * on-disk, not estimated pending changes.
1043 dsl_dir_space_available(dsl_dir_t
*dd
,
1044 dsl_dir_t
*ancestor
, int64_t delta
, int ondiskonly
)
1046 uint64_t parentspace
, myspace
, quota
, used
;
1049 * If there are no restrictions otherwise, assume we have
1050 * unlimited space available.
1053 parentspace
= UINT64_MAX
;
1055 if (dd
->dd_parent
!= NULL
) {
1056 parentspace
= dsl_dir_space_available(dd
->dd_parent
,
1057 ancestor
, delta
, ondiskonly
);
1060 mutex_enter(&dd
->dd_lock
);
1061 if (dsl_dir_phys(dd
)->dd_quota
!= 0)
1062 quota
= dsl_dir_phys(dd
)->dd_quota
;
1063 used
= dsl_dir_phys(dd
)->dd_used_bytes
;
1065 used
+= dsl_dir_space_towrite(dd
);
1067 if (dd
->dd_parent
== NULL
) {
1068 uint64_t poolsize
= dsl_pool_adjustedsize(dd
->dd_pool
, FALSE
);
1069 quota
= MIN(quota
, poolsize
);
1072 if (dsl_dir_phys(dd
)->dd_reserved
> used
&& parentspace
!= UINT64_MAX
) {
1074 * We have some space reserved, in addition to what our
1077 parentspace
+= dsl_dir_phys(dd
)->dd_reserved
- used
;
1080 if (dd
== ancestor
) {
1082 ASSERT(used
>= -delta
);
1084 if (parentspace
!= UINT64_MAX
)
1085 parentspace
-= delta
;
1093 * the lesser of the space provided by our parent and
1094 * the space left in our quota
1096 myspace
= MIN(parentspace
, quota
- used
);
1099 mutex_exit(&dd
->dd_lock
);
1104 struct tempreserve
{
1105 list_node_t tr_node
;
1111 dsl_dir_tempreserve_impl(dsl_dir_t
*dd
, uint64_t asize
, boolean_t netfree
,
1112 boolean_t ignorequota
, boolean_t checkrefquota
, list_t
*tr_list
,
1113 dmu_tx_t
*tx
, boolean_t first
)
1115 uint64_t txg
= tx
->tx_txg
;
1116 uint64_t est_inflight
, used_on_disk
, quota
, parent_rsrv
;
1117 uint64_t deferred
= 0;
1118 struct tempreserve
*tr
;
1119 int retval
= EDQUOT
;
1120 int txgidx
= txg
& TXG_MASK
;
1122 uint64_t ref_rsrv
= 0;
1124 ASSERT3U(txg
, !=, 0);
1125 ASSERT3S(asize
, >, 0);
1127 mutex_enter(&dd
->dd_lock
);
1130 * Check against the dsl_dir's quota. We don't add in the delta
1131 * when checking for over-quota because they get one free hit.
1133 est_inflight
= dsl_dir_space_towrite(dd
);
1134 for (i
= 0; i
< TXG_SIZE
; i
++)
1135 est_inflight
+= dd
->dd_tempreserved
[i
];
1136 used_on_disk
= dsl_dir_phys(dd
)->dd_used_bytes
;
1139 * On the first iteration, fetch the dataset's used-on-disk and
1140 * refreservation values. Also, if checkrefquota is set, test if
1141 * allocating this space would exceed the dataset's refquota.
1143 if (first
&& tx
->tx_objset
) {
1145 dsl_dataset_t
*ds
= tx
->tx_objset
->os_dsl_dataset
;
1147 error
= dsl_dataset_check_quota(ds
, checkrefquota
,
1148 asize
, est_inflight
, &used_on_disk
, &ref_rsrv
);
1150 mutex_exit(&dd
->dd_lock
);
1156 * If this transaction will result in a net free of space,
1157 * we want to let it through.
1159 if (ignorequota
|| netfree
|| dsl_dir_phys(dd
)->dd_quota
== 0)
1162 quota
= dsl_dir_phys(dd
)->dd_quota
;
1165 * Adjust the quota against the actual pool size at the root
1166 * minus any outstanding deferred frees.
1167 * To ensure that it's possible to remove files from a full
1168 * pool without inducing transient overcommits, we throttle
1169 * netfree transactions against a quota that is slightly larger,
1170 * but still within the pool's allocation slop. In cases where
1171 * we're very close to full, this will allow a steady trickle of
1172 * removes to get through.
1174 if (dd
->dd_parent
== NULL
) {
1175 spa_t
*spa
= dd
->dd_pool
->dp_spa
;
1176 uint64_t poolsize
= dsl_pool_adjustedsize(dd
->dd_pool
, netfree
);
1177 deferred
= metaslab_class_get_deferred(spa_normal_class(spa
));
1178 if (poolsize
- deferred
< quota
) {
1179 quota
= poolsize
- deferred
;
1185 * If they are requesting more space, and our current estimate
1186 * is over quota, they get to try again unless the actual
1187 * on-disk is over quota and there are no pending changes (which
1188 * may free up space for us).
1190 if (used_on_disk
+ est_inflight
>= quota
) {
1191 if (est_inflight
> 0 || used_on_disk
< quota
||
1192 (retval
== ENOSPC
&& used_on_disk
< quota
+ deferred
))
1194 dprintf_dd(dd
, "failing: used=%lluK inflight = %lluK "
1195 "quota=%lluK tr=%lluK err=%d\n",
1196 used_on_disk
>>10, est_inflight
>>10,
1197 quota
>>10, asize
>>10, retval
);
1198 mutex_exit(&dd
->dd_lock
);
1199 return (SET_ERROR(retval
));
1202 /* We need to up our estimated delta before dropping dd_lock */
1203 dd
->dd_tempreserved
[txgidx
] += asize
;
1205 parent_rsrv
= parent_delta(dd
, used_on_disk
+ est_inflight
,
1207 mutex_exit(&dd
->dd_lock
);
1209 tr
= kmem_zalloc(sizeof (struct tempreserve
), KM_SLEEP
);
1211 tr
->tr_size
= asize
;
1212 list_insert_tail(tr_list
, tr
);
1214 /* see if it's OK with our parent */
1215 if (dd
->dd_parent
&& parent_rsrv
) {
1216 boolean_t ismos
= (dsl_dir_phys(dd
)->dd_head_dataset_obj
== 0);
1218 return (dsl_dir_tempreserve_impl(dd
->dd_parent
,
1219 parent_rsrv
, netfree
, ismos
, TRUE
, tr_list
, tx
, FALSE
));
1226 * Reserve space in this dsl_dir, to be used in this tx's txg.
1227 * After the space has been dirtied (and dsl_dir_willuse_space()
1228 * has been called), the reservation should be canceled, using
1229 * dsl_dir_tempreserve_clear().
1232 dsl_dir_tempreserve_space(dsl_dir_t
*dd
, uint64_t lsize
, uint64_t asize
,
1233 uint64_t fsize
, uint64_t usize
, void **tr_cookiep
, dmu_tx_t
*tx
)
1243 tr_list
= kmem_alloc(sizeof (list_t
), KM_SLEEP
);
1244 list_create(tr_list
, sizeof (struct tempreserve
),
1245 offsetof(struct tempreserve
, tr_node
));
1246 ASSERT3S(asize
, >, 0);
1247 ASSERT3S(fsize
, >=, 0);
1249 err
= arc_tempreserve_space(lsize
, tx
->tx_txg
);
1251 struct tempreserve
*tr
;
1253 tr
= kmem_zalloc(sizeof (struct tempreserve
), KM_SLEEP
);
1254 tr
->tr_size
= lsize
;
1255 list_insert_tail(tr_list
, tr
);
1257 if (err
== EAGAIN
) {
1259 * If arc_memory_throttle() detected that pageout
1260 * is running and we are low on memory, we delay new
1261 * non-pageout transactions to give pageout an
1264 * It is unfortunate to be delaying while the caller's
1267 txg_delay(dd
->dd_pool
, tx
->tx_txg
,
1268 MSEC2NSEC(10), MSEC2NSEC(10));
1269 err
= SET_ERROR(ERESTART
);
1274 err
= dsl_dir_tempreserve_impl(dd
, asize
, fsize
>= asize
,
1275 FALSE
, asize
> usize
, tr_list
, tx
, TRUE
);
1279 dsl_dir_tempreserve_clear(tr_list
, tx
);
1281 *tr_cookiep
= tr_list
;
1287 * Clear a temporary reservation that we previously made with
1288 * dsl_dir_tempreserve_space().
1291 dsl_dir_tempreserve_clear(void *tr_cookie
, dmu_tx_t
*tx
)
1293 int txgidx
= tx
->tx_txg
& TXG_MASK
;
1294 list_t
*tr_list
= tr_cookie
;
1295 struct tempreserve
*tr
;
1297 ASSERT3U(tx
->tx_txg
, !=, 0);
1299 if (tr_cookie
== NULL
)
1302 while ((tr
= list_head(tr_list
)) != NULL
) {
1304 mutex_enter(&tr
->tr_ds
->dd_lock
);
1305 ASSERT3U(tr
->tr_ds
->dd_tempreserved
[txgidx
], >=,
1307 tr
->tr_ds
->dd_tempreserved
[txgidx
] -= tr
->tr_size
;
1308 mutex_exit(&tr
->tr_ds
->dd_lock
);
1310 arc_tempreserve_clear(tr
->tr_size
);
1312 list_remove(tr_list
, tr
);
1313 kmem_free(tr
, sizeof (struct tempreserve
));
1316 kmem_free(tr_list
, sizeof (list_t
));
1320 * This should be called from open context when we think we're going to write
1321 * or free space, for example when dirtying data. Be conservative; it's okay
1322 * to write less space or free more, but we don't want to write more or free
1323 * less than the amount specified.
1326 dsl_dir_willuse_space(dsl_dir_t
*dd
, int64_t space
, dmu_tx_t
*tx
)
1328 int64_t parent_space
;
1331 mutex_enter(&dd
->dd_lock
);
1333 dd
->dd_space_towrite
[tx
->tx_txg
& TXG_MASK
] += space
;
1335 est_used
= dsl_dir_space_towrite(dd
) + dsl_dir_phys(dd
)->dd_used_bytes
;
1336 parent_space
= parent_delta(dd
, est_used
, space
);
1337 mutex_exit(&dd
->dd_lock
);
1339 /* Make sure that we clean up dd_space_to* */
1340 dsl_dir_dirty(dd
, tx
);
1342 /* XXX this is potentially expensive and unnecessary... */
1343 if (parent_space
&& dd
->dd_parent
)
1344 dsl_dir_willuse_space(dd
->dd_parent
, parent_space
, tx
);
1347 /* call from syncing context when we actually write/free space for this dd */
1349 dsl_dir_diduse_space(dsl_dir_t
*dd
, dd_used_t type
,
1350 int64_t used
, int64_t compressed
, int64_t uncompressed
, dmu_tx_t
*tx
)
1352 int64_t accounted_delta
;
1355 * dsl_dataset_set_refreservation_sync_impl() calls this with
1356 * dd_lock held, so that it can atomically update
1357 * ds->ds_reserved and the dsl_dir accounting, so that
1358 * dsl_dataset_check_quota() can see dataset and dir accounting
1361 boolean_t needlock
= !MUTEX_HELD(&dd
->dd_lock
);
1363 ASSERT(dmu_tx_is_syncing(tx
));
1364 ASSERT(type
< DD_USED_NUM
);
1366 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
1369 mutex_enter(&dd
->dd_lock
);
1371 parent_delta(dd
, dsl_dir_phys(dd
)->dd_used_bytes
, used
);
1372 ASSERT(used
>= 0 || dsl_dir_phys(dd
)->dd_used_bytes
>= -used
);
1373 ASSERT(compressed
>= 0 ||
1374 dsl_dir_phys(dd
)->dd_compressed_bytes
>= -compressed
);
1375 ASSERT(uncompressed
>= 0 ||
1376 dsl_dir_phys(dd
)->dd_uncompressed_bytes
>= -uncompressed
);
1377 dsl_dir_phys(dd
)->dd_used_bytes
+= used
;
1378 dsl_dir_phys(dd
)->dd_uncompressed_bytes
+= uncompressed
;
1379 dsl_dir_phys(dd
)->dd_compressed_bytes
+= compressed
;
1381 if (dsl_dir_phys(dd
)->dd_flags
& DD_FLAG_USED_BREAKDOWN
) {
1383 dsl_dir_phys(dd
)->dd_used_breakdown
[type
] >= -used
);
1384 dsl_dir_phys(dd
)->dd_used_breakdown
[type
] += used
;
1388 for (t
= 0; t
< DD_USED_NUM
; t
++)
1389 u
+= dsl_dir_phys(dd
)->dd_used_breakdown
[t
];
1390 ASSERT3U(u
, ==, dsl_dir_phys(dd
)->dd_used_bytes
);
1394 mutex_exit(&dd
->dd_lock
);
1396 if (dd
->dd_parent
!= NULL
) {
1397 dsl_dir_diduse_space(dd
->dd_parent
, DD_USED_CHILD
,
1398 accounted_delta
, compressed
, uncompressed
, tx
);
1399 dsl_dir_transfer_space(dd
->dd_parent
,
1400 used
- accounted_delta
,
1401 DD_USED_CHILD_RSRV
, DD_USED_CHILD
, tx
);
1406 dsl_dir_transfer_space(dsl_dir_t
*dd
, int64_t delta
,
1407 dd_used_t oldtype
, dd_used_t newtype
, dmu_tx_t
*tx
)
1409 ASSERT(dmu_tx_is_syncing(tx
));
1410 ASSERT(oldtype
< DD_USED_NUM
);
1411 ASSERT(newtype
< DD_USED_NUM
);
1414 !(dsl_dir_phys(dd
)->dd_flags
& DD_FLAG_USED_BREAKDOWN
))
1417 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
1418 mutex_enter(&dd
->dd_lock
);
1420 dsl_dir_phys(dd
)->dd_used_breakdown
[oldtype
] >= delta
:
1421 dsl_dir_phys(dd
)->dd_used_breakdown
[newtype
] >= -delta
);
1422 ASSERT(dsl_dir_phys(dd
)->dd_used_bytes
>= ABS(delta
));
1423 dsl_dir_phys(dd
)->dd_used_breakdown
[oldtype
] -= delta
;
1424 dsl_dir_phys(dd
)->dd_used_breakdown
[newtype
] += delta
;
1425 mutex_exit(&dd
->dd_lock
);
1428 typedef struct dsl_dir_set_qr_arg
{
1429 const char *ddsqra_name
;
1430 zprop_source_t ddsqra_source
;
1431 uint64_t ddsqra_value
;
1432 } dsl_dir_set_qr_arg_t
;
1435 dsl_dir_set_quota_check(void *arg
, dmu_tx_t
*tx
)
1437 dsl_dir_set_qr_arg_t
*ddsqra
= arg
;
1438 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1441 uint64_t towrite
, newval
;
1443 error
= dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
);
1447 error
= dsl_prop_predict(ds
->ds_dir
, "quota",
1448 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, &newval
);
1450 dsl_dataset_rele(ds
, FTAG
);
1455 dsl_dataset_rele(ds
, FTAG
);
1459 mutex_enter(&ds
->ds_dir
->dd_lock
);
1461 * If we are doing the preliminary check in open context, and
1462 * there are pending changes, then don't fail it, since the
1463 * pending changes could under-estimate the amount of space to be
1466 towrite
= dsl_dir_space_towrite(ds
->ds_dir
);
1467 if ((dmu_tx_is_syncing(tx
) || towrite
== 0) &&
1468 (newval
< dsl_dir_phys(ds
->ds_dir
)->dd_reserved
||
1469 newval
< dsl_dir_phys(ds
->ds_dir
)->dd_used_bytes
+ towrite
)) {
1470 error
= SET_ERROR(ENOSPC
);
1472 mutex_exit(&ds
->ds_dir
->dd_lock
);
1473 dsl_dataset_rele(ds
, FTAG
);
1478 dsl_dir_set_quota_sync(void *arg
, dmu_tx_t
*tx
)
1480 dsl_dir_set_qr_arg_t
*ddsqra
= arg
;
1481 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1485 VERIFY0(dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
));
1487 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_RECVD_PROPS
) {
1488 dsl_prop_set_sync_impl(ds
, zfs_prop_to_name(ZFS_PROP_QUOTA
),
1489 ddsqra
->ddsqra_source
, sizeof (ddsqra
->ddsqra_value
), 1,
1490 &ddsqra
->ddsqra_value
, tx
);
1492 VERIFY0(dsl_prop_get_int_ds(ds
,
1493 zfs_prop_to_name(ZFS_PROP_QUOTA
), &newval
));
1495 newval
= ddsqra
->ddsqra_value
;
1496 spa_history_log_internal_ds(ds
, "set", tx
, "%s=%lld",
1497 zfs_prop_to_name(ZFS_PROP_QUOTA
), (longlong_t
)newval
);
1500 dmu_buf_will_dirty(ds
->ds_dir
->dd_dbuf
, tx
);
1501 mutex_enter(&ds
->ds_dir
->dd_lock
);
1502 dsl_dir_phys(ds
->ds_dir
)->dd_quota
= newval
;
1503 mutex_exit(&ds
->ds_dir
->dd_lock
);
1504 dsl_dataset_rele(ds
, FTAG
);
1508 dsl_dir_set_quota(const char *ddname
, zprop_source_t source
, uint64_t quota
)
1510 dsl_dir_set_qr_arg_t ddsqra
;
1512 ddsqra
.ddsqra_name
= ddname
;
1513 ddsqra
.ddsqra_source
= source
;
1514 ddsqra
.ddsqra_value
= quota
;
1516 return (dsl_sync_task(ddname
, dsl_dir_set_quota_check
,
1517 dsl_dir_set_quota_sync
, &ddsqra
, 0, ZFS_SPACE_CHECK_NONE
));
1521 dsl_dir_set_reservation_check(void *arg
, dmu_tx_t
*tx
)
1523 dsl_dir_set_qr_arg_t
*ddsqra
= arg
;
1524 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1527 uint64_t newval
, used
, avail
;
1530 error
= dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
);
1536 * If we are doing the preliminary check in open context, the
1537 * space estimates may be inaccurate.
1539 if (!dmu_tx_is_syncing(tx
)) {
1540 dsl_dataset_rele(ds
, FTAG
);
1544 error
= dsl_prop_predict(ds
->ds_dir
,
1545 zfs_prop_to_name(ZFS_PROP_RESERVATION
),
1546 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, &newval
);
1548 dsl_dataset_rele(ds
, FTAG
);
1552 mutex_enter(&dd
->dd_lock
);
1553 used
= dsl_dir_phys(dd
)->dd_used_bytes
;
1554 mutex_exit(&dd
->dd_lock
);
1556 if (dd
->dd_parent
) {
1557 avail
= dsl_dir_space_available(dd
->dd_parent
,
1560 avail
= dsl_pool_adjustedsize(dd
->dd_pool
, B_FALSE
) - used
;
1563 if (MAX(used
, newval
) > MAX(used
, dsl_dir_phys(dd
)->dd_reserved
)) {
1564 uint64_t delta
= MAX(used
, newval
) -
1565 MAX(used
, dsl_dir_phys(dd
)->dd_reserved
);
1567 if (delta
> avail
||
1568 (dsl_dir_phys(dd
)->dd_quota
> 0 &&
1569 newval
> dsl_dir_phys(dd
)->dd_quota
))
1570 error
= SET_ERROR(ENOSPC
);
1573 dsl_dataset_rele(ds
, FTAG
);
1578 dsl_dir_set_reservation_sync_impl(dsl_dir_t
*dd
, uint64_t value
, dmu_tx_t
*tx
)
1583 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
1585 mutex_enter(&dd
->dd_lock
);
1586 used
= dsl_dir_phys(dd
)->dd_used_bytes
;
1587 delta
= MAX(used
, value
) - MAX(used
, dsl_dir_phys(dd
)->dd_reserved
);
1588 dsl_dir_phys(dd
)->dd_reserved
= value
;
1590 if (dd
->dd_parent
!= NULL
) {
1591 /* Roll up this additional usage into our ancestors */
1592 dsl_dir_diduse_space(dd
->dd_parent
, DD_USED_CHILD_RSRV
,
1595 mutex_exit(&dd
->dd_lock
);
1600 dsl_dir_set_reservation_sync(void *arg
, dmu_tx_t
*tx
)
1602 dsl_dir_set_qr_arg_t
*ddsqra
= arg
;
1603 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1607 VERIFY0(dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
));
1609 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_RECVD_PROPS
) {
1610 dsl_prop_set_sync_impl(ds
,
1611 zfs_prop_to_name(ZFS_PROP_RESERVATION
),
1612 ddsqra
->ddsqra_source
, sizeof (ddsqra
->ddsqra_value
), 1,
1613 &ddsqra
->ddsqra_value
, tx
);
1615 VERIFY0(dsl_prop_get_int_ds(ds
,
1616 zfs_prop_to_name(ZFS_PROP_RESERVATION
), &newval
));
1618 newval
= ddsqra
->ddsqra_value
;
1619 spa_history_log_internal_ds(ds
, "set", tx
, "%s=%lld",
1620 zfs_prop_to_name(ZFS_PROP_RESERVATION
),
1621 (longlong_t
)newval
);
1624 dsl_dir_set_reservation_sync_impl(ds
->ds_dir
, newval
, tx
);
1625 dsl_dataset_rele(ds
, FTAG
);
1629 dsl_dir_set_reservation(const char *ddname
, zprop_source_t source
,
1630 uint64_t reservation
)
1632 dsl_dir_set_qr_arg_t ddsqra
;
1634 ddsqra
.ddsqra_name
= ddname
;
1635 ddsqra
.ddsqra_source
= source
;
1636 ddsqra
.ddsqra_value
= reservation
;
1638 return (dsl_sync_task(ddname
, dsl_dir_set_reservation_check
,
1639 dsl_dir_set_reservation_sync
, &ddsqra
, 0, ZFS_SPACE_CHECK_NONE
));
1643 closest_common_ancestor(dsl_dir_t
*ds1
, dsl_dir_t
*ds2
)
1645 for (; ds1
; ds1
= ds1
->dd_parent
) {
1647 for (dd
= ds2
; dd
; dd
= dd
->dd_parent
) {
1656 * If delta is applied to dd, how much of that delta would be applied to
1657 * ancestor? Syncing context only.
1660 would_change(dsl_dir_t
*dd
, int64_t delta
, dsl_dir_t
*ancestor
)
1665 mutex_enter(&dd
->dd_lock
);
1666 delta
= parent_delta(dd
, dsl_dir_phys(dd
)->dd_used_bytes
, delta
);
1667 mutex_exit(&dd
->dd_lock
);
1668 return (would_change(dd
->dd_parent
, delta
, ancestor
));
1671 typedef struct dsl_dir_rename_arg
{
1672 const char *ddra_oldname
;
1673 const char *ddra_newname
;
1675 } dsl_dir_rename_arg_t
;
1679 dsl_valid_rename(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *arg
)
1682 char namebuf
[MAXNAMELEN
];
1684 dsl_dataset_name(ds
, namebuf
);
1686 if (strlen(namebuf
) + *deltap
>= MAXNAMELEN
)
1687 return (SET_ERROR(ENAMETOOLONG
));
1692 dsl_dir_rename_check(void *arg
, dmu_tx_t
*tx
)
1694 dsl_dir_rename_arg_t
*ddra
= arg
;
1695 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1696 dsl_dir_t
*dd
, *newparent
;
1697 const char *mynewname
;
1699 int delta
= strlen(ddra
->ddra_newname
) - strlen(ddra
->ddra_oldname
);
1701 /* target dir should exist */
1702 error
= dsl_dir_hold(dp
, ddra
->ddra_oldname
, FTAG
, &dd
, NULL
);
1706 /* new parent should exist */
1707 error
= dsl_dir_hold(dp
, ddra
->ddra_newname
, FTAG
,
1708 &newparent
, &mynewname
);
1710 dsl_dir_rele(dd
, FTAG
);
1714 /* can't rename to different pool */
1715 if (dd
->dd_pool
!= newparent
->dd_pool
) {
1716 dsl_dir_rele(newparent
, FTAG
);
1717 dsl_dir_rele(dd
, FTAG
);
1718 return (SET_ERROR(ENXIO
));
1721 /* new name should not already exist */
1722 if (mynewname
== NULL
) {
1723 dsl_dir_rele(newparent
, FTAG
);
1724 dsl_dir_rele(dd
, FTAG
);
1725 return (SET_ERROR(EEXIST
));
1728 /* if the name length is growing, validate child name lengths */
1730 error
= dmu_objset_find_dp(dp
, dd
->dd_object
, dsl_valid_rename
,
1731 &delta
, DS_FIND_CHILDREN
| DS_FIND_SNAPSHOTS
);
1733 dsl_dir_rele(newparent
, FTAG
);
1734 dsl_dir_rele(dd
, FTAG
);
1739 if (dmu_tx_is_syncing(tx
)) {
1740 if (spa_feature_is_active(dp
->dp_spa
,
1741 SPA_FEATURE_FS_SS_LIMIT
)) {
1743 * Although this is the check function and we don't
1744 * normally make on-disk changes in check functions,
1745 * we need to do that here.
1747 * Ensure this portion of the tree's counts have been
1748 * initialized in case the new parent has limits set.
1750 dsl_dir_init_fs_ss_count(dd
, tx
);
1754 if (newparent
!= dd
->dd_parent
) {
1755 /* is there enough space? */
1757 MAX(dsl_dir_phys(dd
)->dd_used_bytes
,
1758 dsl_dir_phys(dd
)->dd_reserved
);
1759 objset_t
*os
= dd
->dd_pool
->dp_meta_objset
;
1760 uint64_t fs_cnt
= 0;
1761 uint64_t ss_cnt
= 0;
1763 if (dsl_dir_is_zapified(dd
)) {
1766 err
= zap_lookup(os
, dd
->dd_object
,
1767 DD_FIELD_FILESYSTEM_COUNT
, sizeof (fs_cnt
), 1,
1769 if (err
!= ENOENT
&& err
!= 0) {
1770 dsl_dir_rele(newparent
, FTAG
);
1771 dsl_dir_rele(dd
, FTAG
);
1776 * have to add 1 for the filesystem itself that we're
1781 err
= zap_lookup(os
, dd
->dd_object
,
1782 DD_FIELD_SNAPSHOT_COUNT
, sizeof (ss_cnt
), 1,
1784 if (err
!= ENOENT
&& err
!= 0) {
1785 dsl_dir_rele(newparent
, FTAG
);
1786 dsl_dir_rele(dd
, FTAG
);
1791 /* no rename into our descendant */
1792 if (closest_common_ancestor(dd
, newparent
) == dd
) {
1793 dsl_dir_rele(newparent
, FTAG
);
1794 dsl_dir_rele(dd
, FTAG
);
1795 return (SET_ERROR(EINVAL
));
1798 error
= dsl_dir_transfer_possible(dd
->dd_parent
,
1799 newparent
, fs_cnt
, ss_cnt
, myspace
, ddra
->ddra_cred
);
1801 dsl_dir_rele(newparent
, FTAG
);
1802 dsl_dir_rele(dd
, FTAG
);
1807 dsl_dir_rele(newparent
, FTAG
);
1808 dsl_dir_rele(dd
, FTAG
);
1813 dsl_dir_rename_sync(void *arg
, dmu_tx_t
*tx
)
1815 dsl_dir_rename_arg_t
*ddra
= arg
;
1816 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1817 dsl_dir_t
*dd
, *newparent
;
1818 const char *mynewname
;
1820 objset_t
*mos
= dp
->dp_meta_objset
;
1822 VERIFY0(dsl_dir_hold(dp
, ddra
->ddra_oldname
, FTAG
, &dd
, NULL
));
1823 VERIFY0(dsl_dir_hold(dp
, ddra
->ddra_newname
, FTAG
, &newparent
,
1826 /* Log this before we change the name. */
1827 spa_history_log_internal_dd(dd
, "rename", tx
,
1828 "-> %s", ddra
->ddra_newname
);
1830 if (newparent
!= dd
->dd_parent
) {
1831 objset_t
*os
= dd
->dd_pool
->dp_meta_objset
;
1832 uint64_t fs_cnt
= 0;
1833 uint64_t ss_cnt
= 0;
1836 * We already made sure the dd counts were initialized in the
1839 if (spa_feature_is_active(dp
->dp_spa
,
1840 SPA_FEATURE_FS_SS_LIMIT
)) {
1841 VERIFY0(zap_lookup(os
, dd
->dd_object
,
1842 DD_FIELD_FILESYSTEM_COUNT
, sizeof (fs_cnt
), 1,
1844 /* add 1 for the filesystem itself that we're moving */
1847 VERIFY0(zap_lookup(os
, dd
->dd_object
,
1848 DD_FIELD_SNAPSHOT_COUNT
, sizeof (ss_cnt
), 1,
1852 dsl_fs_ss_count_adjust(dd
->dd_parent
, -fs_cnt
,
1853 DD_FIELD_FILESYSTEM_COUNT
, tx
);
1854 dsl_fs_ss_count_adjust(newparent
, fs_cnt
,
1855 DD_FIELD_FILESYSTEM_COUNT
, tx
);
1857 dsl_fs_ss_count_adjust(dd
->dd_parent
, -ss_cnt
,
1858 DD_FIELD_SNAPSHOT_COUNT
, tx
);
1859 dsl_fs_ss_count_adjust(newparent
, ss_cnt
,
1860 DD_FIELD_SNAPSHOT_COUNT
, tx
);
1862 dsl_dir_diduse_space(dd
->dd_parent
, DD_USED_CHILD
,
1863 -dsl_dir_phys(dd
)->dd_used_bytes
,
1864 -dsl_dir_phys(dd
)->dd_compressed_bytes
,
1865 -dsl_dir_phys(dd
)->dd_uncompressed_bytes
, tx
);
1866 dsl_dir_diduse_space(newparent
, DD_USED_CHILD
,
1867 dsl_dir_phys(dd
)->dd_used_bytes
,
1868 dsl_dir_phys(dd
)->dd_compressed_bytes
,
1869 dsl_dir_phys(dd
)->dd_uncompressed_bytes
, tx
);
1871 if (dsl_dir_phys(dd
)->dd_reserved
>
1872 dsl_dir_phys(dd
)->dd_used_bytes
) {
1873 uint64_t unused_rsrv
= dsl_dir_phys(dd
)->dd_reserved
-
1874 dsl_dir_phys(dd
)->dd_used_bytes
;
1876 dsl_dir_diduse_space(dd
->dd_parent
, DD_USED_CHILD_RSRV
,
1877 -unused_rsrv
, 0, 0, tx
);
1878 dsl_dir_diduse_space(newparent
, DD_USED_CHILD_RSRV
,
1879 unused_rsrv
, 0, 0, tx
);
1883 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
1885 /* remove from old parent zapobj */
1886 error
= zap_remove(mos
,
1887 dsl_dir_phys(dd
->dd_parent
)->dd_child_dir_zapobj
,
1891 (void) strcpy(dd
->dd_myname
, mynewname
);
1892 dsl_dir_rele(dd
->dd_parent
, dd
);
1893 dsl_dir_phys(dd
)->dd_parent_obj
= newparent
->dd_object
;
1894 VERIFY0(dsl_dir_hold_obj(dp
,
1895 newparent
->dd_object
, NULL
, dd
, &dd
->dd_parent
));
1897 /* add to new parent zapobj */
1898 VERIFY0(zap_add(mos
, dsl_dir_phys(newparent
)->dd_child_dir_zapobj
,
1899 dd
->dd_myname
, 8, 1, &dd
->dd_object
, tx
));
1901 dsl_prop_notify_all(dd
);
1903 dsl_dir_rele(newparent
, FTAG
);
1904 dsl_dir_rele(dd
, FTAG
);
1908 dsl_dir_rename(const char *oldname
, const char *newname
)
1910 dsl_dir_rename_arg_t ddra
;
1912 ddra
.ddra_oldname
= oldname
;
1913 ddra
.ddra_newname
= newname
;
1914 ddra
.ddra_cred
= CRED();
1916 return (dsl_sync_task(oldname
,
1917 dsl_dir_rename_check
, dsl_dir_rename_sync
, &ddra
,
1918 3, ZFS_SPACE_CHECK_RESERVED
));
1922 dsl_dir_transfer_possible(dsl_dir_t
*sdd
, dsl_dir_t
*tdd
,
1923 uint64_t fs_cnt
, uint64_t ss_cnt
, uint64_t space
, cred_t
*cr
)
1925 dsl_dir_t
*ancestor
;
1930 ancestor
= closest_common_ancestor(sdd
, tdd
);
1931 adelta
= would_change(sdd
, -space
, ancestor
);
1932 avail
= dsl_dir_space_available(tdd
, ancestor
, adelta
, FALSE
);
1934 return (SET_ERROR(ENOSPC
));
1936 err
= dsl_fs_ss_limit_check(tdd
, fs_cnt
, ZFS_PROP_FILESYSTEM_LIMIT
,
1940 err
= dsl_fs_ss_limit_check(tdd
, ss_cnt
, ZFS_PROP_SNAPSHOT_LIMIT
,
1949 dsl_dir_snap_cmtime(dsl_dir_t
*dd
)
1953 mutex_enter(&dd
->dd_lock
);
1954 t
= dd
->dd_snap_cmtime
;
1955 mutex_exit(&dd
->dd_lock
);
1961 dsl_dir_snap_cmtime_update(dsl_dir_t
*dd
)
1966 mutex_enter(&dd
->dd_lock
);
1967 dd
->dd_snap_cmtime
= t
;
1968 mutex_exit(&dd
->dd_lock
);
1972 dsl_dir_zapify(dsl_dir_t
*dd
, dmu_tx_t
*tx
)
1974 objset_t
*mos
= dd
->dd_pool
->dp_meta_objset
;
1975 dmu_object_zapify(mos
, dd
->dd_object
, DMU_OT_DSL_DIR
, tx
);
1979 dsl_dir_is_zapified(dsl_dir_t
*dd
)
1981 dmu_object_info_t doi
;
1983 dmu_object_info_from_db(dd
->dd_dbuf
, &doi
);
1984 return (doi
.doi_type
== DMU_OTN_ZAP_METADATA
);