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, 2018 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>
52 #include "zfs_namecheck.h"
55 #include <sys/zfs_vfsops.h>
59 * Filesystem and Snapshot Limits
60 * ------------------------------
62 * These limits are used to restrict the number of filesystems and/or snapshots
63 * that can be created at a given level in the tree or below. A typical
64 * use-case is with a delegated dataset where the administrator wants to ensure
65 * that a user within the zone is not creating too many additional filesystems
66 * or snapshots, even though they're not exceeding their space quota.
68 * The filesystem and snapshot counts are stored as extensible properties. This
69 * capability is controlled by a feature flag and must be enabled to be used.
70 * Once enabled, the feature is not active until the first limit is set. At
71 * that point, future operations to create/destroy filesystems or snapshots
72 * will validate and update the counts.
74 * Because the count properties will not exist before the feature is active,
75 * the counts are updated when a limit is first set on an uninitialized
76 * dsl_dir node in the tree (The filesystem/snapshot count on a node includes
77 * all of the nested filesystems/snapshots. Thus, a new leaf node has a
78 * filesystem count of 0 and a snapshot count of 0. Non-existent filesystem and
79 * snapshot count properties on a node indicate uninitialized counts on that
80 * node.) When first setting a limit on an uninitialized node, the code starts
81 * at the filesystem with the new limit and descends into all sub-filesystems
82 * to add the count properties.
84 * In practice this is lightweight since a limit is typically set when the
85 * filesystem is created and thus has no children. Once valid, changing the
86 * limit value won't require a re-traversal since the counts are already valid.
87 * When recursively fixing the counts, if a node with a limit is encountered
88 * during the descent, the counts are known to be valid and there is no need to
89 * descend into that filesystem's children. The counts on filesystems above the
90 * one with the new limit will still be uninitialized, unless a limit is
91 * eventually set on one of those filesystems. The counts are always recursively
92 * updated when a limit is set on a dataset, unless there is already a limit.
93 * When a new limit value is set on a filesystem with an existing limit, it is
94 * possible for the new limit to be less than the current count at that level
95 * since a user who can change the limit is also allowed to exceed the limit.
97 * Once the feature is active, then whenever a filesystem or snapshot is
98 * created, the code recurses up the tree, validating the new count against the
99 * limit at each initialized level. In practice, most levels will not have a
100 * limit set. If there is a limit at any initialized level up the tree, the
101 * check must pass or the creation will fail. Likewise, when a filesystem or
102 * snapshot is destroyed, the counts are recursively adjusted all the way up
103 * the initialized nodes in the tree. Renaming a filesystem into different point
104 * in the tree will first validate, then update the counts on each branch up to
105 * the common ancestor. A receive will also validate the counts and then update
108 * An exception to the above behavior is that the limit is not enforced if the
109 * user has permission to modify the limit. This is primarily so that
110 * recursive snapshots in the global zone always work. We want to prevent a
111 * denial-of-service in which a lower level delegated dataset could max out its
112 * limit and thus block recursive snapshots from being taken in the global zone.
113 * Because of this, it is possible for the snapshot count to be over the limit
114 * and snapshots taken in the global zone could cause a lower level dataset to
115 * hit or exceed its limit. The administrator taking the global zone recursive
116 * snapshot should be aware of this side-effect and behave accordingly.
117 * For consistency, the filesystem limit is also not enforced if the user can
120 * The filesystem and snapshot limits are validated by dsl_fs_ss_limit_check()
121 * and updated by dsl_fs_ss_count_adjust(). A new limit value is setup in
122 * dsl_dir_activate_fs_ss_limit() and the counts are adjusted, if necessary, by
123 * dsl_dir_init_fs_ss_count().
125 * There is a special case when we receive a filesystem that already exists. In
126 * this case a temporary clone name of %X is created (see dmu_recv_begin). We
127 * never update the filesystem counts for temporary clones.
129 * Likewise, we do not update the snapshot counts for temporary snapshots,
130 * such as those created by zfs diff.
133 extern inline dsl_dir_phys_t *dsl_dir_phys(dsl_dir_t *dd);
135 static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
137 typedef struct ddulrt_arg {
138 dsl_dir_t *ddulrta_dd;
143 dsl_dir_evict_async(void *dbu)
147 dsl_pool_t *dp __maybe_unused = dd->dd_pool;
151 for (t = 0; t < TXG_SIZE; t++) {
152 ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t));
153 ASSERT(dd->dd_tempreserved[t] == 0);
154 ASSERT(dd->dd_space_towrite[t] == 0);
158 dsl_dir_async_rele(dd->dd_parent, dd);
160 spa_async_close(dd->dd_pool->dp_spa, dd);
162 if (dsl_deadlist_is_open(&dd->dd_livelist))
163 dsl_dir_livelist_close(dd);
166 cv_destroy(&dd->dd_activity_cv);
167 mutex_destroy(&dd->dd_activity_lock);
168 mutex_destroy(&dd->dd_lock);
169 kmem_free(dd, sizeof (dsl_dir_t));
173 dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj,
174 const char *tail, void *tag, dsl_dir_t **ddp)
178 dmu_object_info_t doi;
181 ASSERT(dsl_pool_config_held(dp));
183 err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
186 dd = dmu_buf_get_user(dbuf);
188 dmu_object_info_from_db(dbuf, &doi);
189 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_DSL_DIR);
190 ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
195 dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
196 dd->dd_object = ddobj;
200 if (dsl_dir_is_zapified(dd) &&
201 zap_contains(dp->dp_meta_objset, ddobj,
202 DD_FIELD_CRYPTO_KEY_OBJ) == 0) {
203 VERIFY0(zap_lookup(dp->dp_meta_objset,
204 ddobj, DD_FIELD_CRYPTO_KEY_OBJ,
205 sizeof (uint64_t), 1, &dd->dd_crypto_obj));
207 /* check for on-disk format errata */
208 if (dsl_dir_incompatible_encryption_version(dd)) {
209 dp->dp_spa->spa_errata =
210 ZPOOL_ERRATA_ZOL_6845_ENCRYPTION;
214 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
215 mutex_init(&dd->dd_activity_lock, NULL, MUTEX_DEFAULT, NULL);
216 cv_init(&dd->dd_activity_cv, NULL, CV_DEFAULT, NULL);
219 dsl_dir_snap_cmtime_update(dd);
221 if (dsl_dir_phys(dd)->dd_parent_obj) {
222 err = dsl_dir_hold_obj(dp,
223 dsl_dir_phys(dd)->dd_parent_obj, NULL, dd,
231 err = zap_lookup(dp->dp_meta_objset,
232 dsl_dir_phys(dd->dd_parent)->
233 dd_child_dir_zapobj, tail,
234 sizeof (foundobj), 1, &foundobj);
235 ASSERT(err || foundobj == ddobj);
237 (void) strlcpy(dd->dd_myname, tail,
238 sizeof (dd->dd_myname));
240 err = zap_value_search(dp->dp_meta_objset,
241 dsl_dir_phys(dd->dd_parent)->
243 ddobj, 0, dd->dd_myname);
248 (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
251 if (dsl_dir_is_clone(dd)) {
252 dmu_buf_t *origin_bonus;
253 dsl_dataset_phys_t *origin_phys;
256 * We can't open the origin dataset, because
257 * that would require opening this dsl_dir.
258 * Just look at its phys directly instead.
260 err = dmu_bonus_hold(dp->dp_meta_objset,
261 dsl_dir_phys(dd)->dd_origin_obj, FTAG,
265 origin_phys = origin_bonus->db_data;
267 origin_phys->ds_creation_txg;
268 dmu_buf_rele(origin_bonus, FTAG);
269 if (dsl_dir_is_zapified(dd)) {
271 err = zap_lookup(dp->dp_meta_objset,
272 dd->dd_object, DD_FIELD_LIVELIST,
273 sizeof (uint64_t), 1, &obj);
275 dsl_dir_livelist_open(dd, obj);
276 else if (err != ENOENT)
281 dmu_buf_init_user(&dd->dd_dbu, NULL, dsl_dir_evict_async,
283 winner = dmu_buf_set_user_ie(dbuf, &dd->dd_dbu);
284 if (winner != NULL) {
286 dsl_dir_rele(dd->dd_parent, dd);
287 if (dsl_deadlist_is_open(&dd->dd_livelist))
288 dsl_dir_livelist_close(dd);
290 cv_destroy(&dd->dd_activity_cv);
291 mutex_destroy(&dd->dd_activity_lock);
292 mutex_destroy(&dd->dd_lock);
293 kmem_free(dd, sizeof (dsl_dir_t));
296 spa_open_ref(dp->dp_spa, dd);
301 * The dsl_dir_t has both open-to-close and instantiate-to-evict
302 * holds on the spa. We need the open-to-close holds because
303 * otherwise the spa_refcnt wouldn't change when we open a
304 * dir which the spa also has open, so we could incorrectly
305 * think it was OK to unload/export/destroy the pool. We need
306 * the instantiate-to-evict hold because the dsl_dir_t has a
307 * pointer to the dd_pool, which has a pointer to the spa_t.
309 spa_open_ref(dp->dp_spa, tag);
310 ASSERT3P(dd->dd_pool, ==, dp);
311 ASSERT3U(dd->dd_object, ==, ddobj);
312 ASSERT3P(dd->dd_dbuf, ==, dbuf);
318 dsl_dir_rele(dd->dd_parent, dd);
319 if (dsl_deadlist_is_open(&dd->dd_livelist))
320 dsl_dir_livelist_close(dd);
322 cv_destroy(&dd->dd_activity_cv);
323 mutex_destroy(&dd->dd_activity_lock);
324 mutex_destroy(&dd->dd_lock);
325 kmem_free(dd, sizeof (dsl_dir_t));
326 dmu_buf_rele(dbuf, tag);
331 dsl_dir_rele(dsl_dir_t *dd, void *tag)
333 dprintf_dd(dd, "%s\n", "");
334 spa_close(dd->dd_pool->dp_spa, tag);
335 dmu_buf_rele(dd->dd_dbuf, tag);
339 * Remove a reference to the given dsl dir that is being asynchronously
340 * released. Async releases occur from a taskq performing eviction of
341 * dsl datasets and dirs. This process is identical to a normal release
342 * with the exception of using the async API for releasing the reference on
346 dsl_dir_async_rele(dsl_dir_t *dd, void *tag)
348 dprintf_dd(dd, "%s\n", "");
349 spa_async_close(dd->dd_pool->dp_spa, tag);
350 dmu_buf_rele(dd->dd_dbuf, tag);
353 /* buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes */
355 dsl_dir_name(dsl_dir_t *dd, char *buf)
358 dsl_dir_name(dd->dd_parent, buf);
359 VERIFY3U(strlcat(buf, "/", ZFS_MAX_DATASET_NAME_LEN), <,
360 ZFS_MAX_DATASET_NAME_LEN);
364 if (!MUTEX_HELD(&dd->dd_lock)) {
366 * recursive mutex so that we can use
367 * dprintf_dd() with dd_lock held
369 mutex_enter(&dd->dd_lock);
370 VERIFY3U(strlcat(buf, dd->dd_myname, ZFS_MAX_DATASET_NAME_LEN),
371 <, ZFS_MAX_DATASET_NAME_LEN);
372 mutex_exit(&dd->dd_lock);
374 VERIFY3U(strlcat(buf, dd->dd_myname, ZFS_MAX_DATASET_NAME_LEN),
375 <, ZFS_MAX_DATASET_NAME_LEN);
379 /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
381 dsl_dir_namelen(dsl_dir_t *dd)
386 /* parent's name + 1 for the "/" */
387 result = dsl_dir_namelen(dd->dd_parent) + 1;
390 if (!MUTEX_HELD(&dd->dd_lock)) {
391 /* see dsl_dir_name */
392 mutex_enter(&dd->dd_lock);
393 result += strlen(dd->dd_myname);
394 mutex_exit(&dd->dd_lock);
396 result += strlen(dd->dd_myname);
403 getcomponent(const char *path, char *component, const char **nextp)
407 if ((path == NULL) || (path[0] == '\0'))
408 return (SET_ERROR(ENOENT));
409 /* This would be a good place to reserve some namespace... */
410 p = strpbrk(path, "/@");
411 if (p && (p[1] == '/' || p[1] == '@')) {
412 /* two separators in a row */
413 return (SET_ERROR(EINVAL));
415 if (p == NULL || p == path) {
417 * if the first thing is an @ or /, it had better be an
418 * @ and it had better not have any more ats or slashes,
419 * and it had better have something after the @.
422 (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
423 return (SET_ERROR(EINVAL));
424 if (strlen(path) >= ZFS_MAX_DATASET_NAME_LEN)
425 return (SET_ERROR(ENAMETOOLONG));
426 (void) strcpy(component, path);
428 } else if (p[0] == '/') {
429 if (p - path >= ZFS_MAX_DATASET_NAME_LEN)
430 return (SET_ERROR(ENAMETOOLONG));
431 (void) strncpy(component, path, p - path);
432 component[p - path] = '\0';
434 } else if (p[0] == '@') {
436 * if the next separator is an @, there better not be
439 if (strchr(path, '/'))
440 return (SET_ERROR(EINVAL));
441 if (p - path >= ZFS_MAX_DATASET_NAME_LEN)
442 return (SET_ERROR(ENAMETOOLONG));
443 (void) strncpy(component, path, p - path);
444 component[p - path] = '\0';
446 panic("invalid p=%p", (void *)p);
453 * Return the dsl_dir_t, and possibly the last component which couldn't
454 * be found in *tail. The name must be in the specified dsl_pool_t. This
455 * thread must hold the dp_config_rwlock for the pool. Returns NULL if the
456 * path is bogus, or if tail==NULL and we couldn't parse the whole name.
457 * (*tail)[0] == '@' means that the last component is a snapshot.
460 dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag,
461 dsl_dir_t **ddp, const char **tailp)
464 const char *spaname, *next, *nextnext = NULL;
469 buf = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
470 err = getcomponent(name, buf, &next);
474 /* Make sure the name is in the specified pool. */
475 spaname = spa_name(dp->dp_spa);
476 if (strcmp(buf, spaname) != 0) {
477 err = SET_ERROR(EXDEV);
481 ASSERT(dsl_pool_config_held(dp));
483 err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
488 while (next != NULL) {
490 err = getcomponent(next, buf, &nextnext);
493 ASSERT(next[0] != '\0');
496 dprintf("looking up %s in obj%lld\n",
497 buf, dsl_dir_phys(dd)->dd_child_dir_zapobj);
499 err = zap_lookup(dp->dp_meta_objset,
500 dsl_dir_phys(dd)->dd_child_dir_zapobj,
501 buf, sizeof (ddobj), 1, &ddobj);
508 err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_dd);
511 dsl_dir_rele(dd, tag);
517 dsl_dir_rele(dd, tag);
522 * It's an error if there's more than one component left, or
523 * tailp==NULL and there's any component left.
526 (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
528 dsl_dir_rele(dd, tag);
529 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
530 err = SET_ERROR(ENOENT);
537 kmem_free(buf, ZFS_MAX_DATASET_NAME_LEN);
542 * If the counts are already initialized for this filesystem and its
543 * descendants then do nothing, otherwise initialize the counts.
545 * The counts on this filesystem, and those below, may be uninitialized due to
546 * either the use of a pre-existing pool which did not support the
547 * filesystem/snapshot limit feature, or one in which the feature had not yet
550 * Recursively descend the filesystem tree and update the filesystem/snapshot
551 * counts on each filesystem below, then update the cumulative count on the
552 * current filesystem. If the filesystem already has a count set on it,
553 * then we know that its counts, and the counts on the filesystems below it,
554 * are already correct, so we don't have to update this filesystem.
557 dsl_dir_init_fs_ss_count(dsl_dir_t *dd, dmu_tx_t *tx)
559 uint64_t my_fs_cnt = 0;
560 uint64_t my_ss_cnt = 0;
561 dsl_pool_t *dp = dd->dd_pool;
562 objset_t *os = dp->dp_meta_objset;
567 ASSERT(spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT));
568 ASSERT(dsl_pool_config_held(dp));
569 ASSERT(dmu_tx_is_syncing(tx));
571 dsl_dir_zapify(dd, tx);
574 * If the filesystem count has already been initialized then we
575 * don't need to recurse down any further.
577 if (zap_contains(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT) == 0)
580 zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP);
581 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
583 /* Iterate my child dirs */
584 for (zap_cursor_init(zc, os, dsl_dir_phys(dd)->dd_child_dir_zapobj);
585 zap_cursor_retrieve(zc, za) == 0; zap_cursor_advance(zc)) {
589 VERIFY0(dsl_dir_hold_obj(dp, za->za_first_integer, NULL, FTAG,
593 * Ignore hidden ($FREE, $MOS & $ORIGIN) objsets and
594 * temporary datasets.
596 if (chld_dd->dd_myname[0] == '$' ||
597 chld_dd->dd_myname[0] == '%') {
598 dsl_dir_rele(chld_dd, FTAG);
602 my_fs_cnt++; /* count this child */
604 dsl_dir_init_fs_ss_count(chld_dd, tx);
606 VERIFY0(zap_lookup(os, chld_dd->dd_object,
607 DD_FIELD_FILESYSTEM_COUNT, sizeof (count), 1, &count));
609 VERIFY0(zap_lookup(os, chld_dd->dd_object,
610 DD_FIELD_SNAPSHOT_COUNT, sizeof (count), 1, &count));
613 dsl_dir_rele(chld_dd, FTAG);
616 /* Count my snapshots (we counted children's snapshots above) */
617 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
618 dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds));
620 for (zap_cursor_init(zc, os, dsl_dataset_phys(ds)->ds_snapnames_zapobj);
621 zap_cursor_retrieve(zc, za) == 0;
622 zap_cursor_advance(zc)) {
623 /* Don't count temporary snapshots */
624 if (za->za_name[0] != '%')
629 dsl_dataset_rele(ds, FTAG);
631 kmem_free(zc, sizeof (zap_cursor_t));
632 kmem_free(za, sizeof (zap_attribute_t));
634 /* we're in a sync task, update counts */
635 dmu_buf_will_dirty(dd->dd_dbuf, tx);
636 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
637 sizeof (my_fs_cnt), 1, &my_fs_cnt, tx));
638 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
639 sizeof (my_ss_cnt), 1, &my_ss_cnt, tx));
643 dsl_dir_actv_fs_ss_limit_check(void *arg, dmu_tx_t *tx)
645 char *ddname = (char *)arg;
646 dsl_pool_t *dp = dmu_tx_pool(tx);
651 error = dsl_dataset_hold(dp, ddname, FTAG, &ds);
655 if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
656 dsl_dataset_rele(ds, FTAG);
657 return (SET_ERROR(ENOTSUP));
661 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT) &&
662 dsl_dir_is_zapified(dd) &&
663 zap_contains(dp->dp_meta_objset, dd->dd_object,
664 DD_FIELD_FILESYSTEM_COUNT) == 0) {
665 dsl_dataset_rele(ds, FTAG);
666 return (SET_ERROR(EALREADY));
669 dsl_dataset_rele(ds, FTAG);
674 dsl_dir_actv_fs_ss_limit_sync(void *arg, dmu_tx_t *tx)
676 char *ddname = (char *)arg;
677 dsl_pool_t *dp = dmu_tx_pool(tx);
681 VERIFY0(dsl_dataset_hold(dp, ddname, FTAG, &ds));
683 spa = dsl_dataset_get_spa(ds);
685 if (!spa_feature_is_active(spa, SPA_FEATURE_FS_SS_LIMIT)) {
687 * Since the feature was not active and we're now setting a
688 * limit, increment the feature-active counter so that the
689 * feature becomes active for the first time.
691 * We are already in a sync task so we can update the MOS.
693 spa_feature_incr(spa, SPA_FEATURE_FS_SS_LIMIT, tx);
697 * Since we are now setting a non-UINT64_MAX limit on the filesystem,
698 * we need to ensure the counts are correct. Descend down the tree from
699 * this point and update all of the counts to be accurate.
701 dsl_dir_init_fs_ss_count(ds->ds_dir, tx);
703 dsl_dataset_rele(ds, FTAG);
707 * Make sure the feature is enabled and activate it if necessary.
708 * Since we're setting a limit, ensure the on-disk counts are valid.
709 * This is only called by the ioctl path when setting a limit value.
711 * We do not need to validate the new limit, since users who can change the
712 * limit are also allowed to exceed the limit.
715 dsl_dir_activate_fs_ss_limit(const char *ddname)
719 error = dsl_sync_task(ddname, dsl_dir_actv_fs_ss_limit_check,
720 dsl_dir_actv_fs_ss_limit_sync, (void *)ddname, 0,
721 ZFS_SPACE_CHECK_RESERVED);
723 if (error == EALREADY)
730 * Used to determine if the filesystem_limit or snapshot_limit should be
731 * enforced. We allow the limit to be exceeded if the user has permission to
732 * write the property value. We pass in the creds that we got in the open
733 * context since we will always be the GZ root in syncing context. We also have
734 * to handle the case where we are allowed to change the limit on the current
735 * dataset, but there may be another limit in the tree above.
737 * We can never modify these two properties within a non-global zone. In
738 * addition, the other checks are modeled on zfs_secpolicy_write_perms. We
739 * can't use that function since we are already holding the dp_config_rwlock.
740 * In addition, we already have the dd and dealing with snapshots is simplified
751 dsl_enforce_ds_ss_limits(dsl_dir_t *dd, zfs_prop_t prop, cred_t *cr)
753 enforce_res_t enforce = ENFORCE_ALWAYS;
757 const char *zonedstr;
759 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
760 prop == ZFS_PROP_SNAPSHOT_LIMIT);
763 if (crgetzoneid(cr) != GLOBAL_ZONEID)
764 return (ENFORCE_ALWAYS);
766 if (secpolicy_zfs(cr) == 0)
767 return (ENFORCE_NEVER);
770 if ((obj = dsl_dir_phys(dd)->dd_head_dataset_obj) == 0)
771 return (ENFORCE_ALWAYS);
773 ASSERT(dsl_pool_config_held(dd->dd_pool));
775 if (dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds) != 0)
776 return (ENFORCE_ALWAYS);
778 zonedstr = zfs_prop_to_name(ZFS_PROP_ZONED);
779 if (dsl_prop_get_ds(ds, zonedstr, 8, 1, &zoned, NULL) || zoned) {
780 /* Only root can access zoned fs's from the GZ */
781 enforce = ENFORCE_ALWAYS;
783 if (dsl_deleg_access_impl(ds, zfs_prop_to_name(prop), cr) == 0)
784 enforce = ENFORCE_ABOVE;
787 dsl_dataset_rele(ds, FTAG);
792 * Check if adding additional child filesystem(s) would exceed any filesystem
793 * limits or adding additional snapshot(s) would exceed any snapshot limits.
794 * The prop argument indicates which limit to check.
796 * Note that all filesystem limits up to the root (or the highest
797 * initialized) filesystem or the given ancestor must be satisfied.
800 dsl_fs_ss_limit_check(dsl_dir_t *dd, uint64_t delta, zfs_prop_t prop,
801 dsl_dir_t *ancestor, cred_t *cr)
803 objset_t *os = dd->dd_pool->dp_meta_objset;
804 uint64_t limit, count;
806 enforce_res_t enforce;
809 ASSERT(dsl_pool_config_held(dd->dd_pool));
810 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
811 prop == ZFS_PROP_SNAPSHOT_LIMIT);
814 * If we're allowed to change the limit, don't enforce the limit
815 * e.g. this can happen if a snapshot is taken by an administrative
816 * user in the global zone (i.e. a recursive snapshot by root).
817 * However, we must handle the case of delegated permissions where we
818 * are allowed to change the limit on the current dataset, but there
819 * is another limit in the tree above.
821 enforce = dsl_enforce_ds_ss_limits(dd, prop, cr);
822 if (enforce == ENFORCE_NEVER)
826 * e.g. if renaming a dataset with no snapshots, count adjustment
832 if (prop == ZFS_PROP_SNAPSHOT_LIMIT) {
834 * We don't enforce the limit for temporary snapshots. This is
835 * indicated by a NULL cred_t argument.
840 count_prop = DD_FIELD_SNAPSHOT_COUNT;
842 count_prop = DD_FIELD_FILESYSTEM_COUNT;
846 * If an ancestor has been provided, stop checking the limit once we
847 * hit that dir. We need this during rename so that we don't overcount
848 * the check once we recurse up to the common ancestor.
854 * If we hit an uninitialized node while recursing up the tree, we can
855 * stop since we know there is no limit here (or above). The counts are
856 * not valid on this node and we know we won't touch this node's counts.
858 if (!dsl_dir_is_zapified(dd) || zap_lookup(os, dd->dd_object,
859 count_prop, sizeof (count), 1, &count) == ENOENT)
862 err = dsl_prop_get_dd(dd, zfs_prop_to_name(prop), 8, 1, &limit, NULL,
867 /* Is there a limit which we've hit? */
868 if (enforce == ENFORCE_ALWAYS && (count + delta) > limit)
869 return (SET_ERROR(EDQUOT));
871 if (dd->dd_parent != NULL)
872 err = dsl_fs_ss_limit_check(dd->dd_parent, delta, prop,
879 * Adjust the filesystem or snapshot count for the specified dsl_dir_t and all
880 * parents. When a new filesystem/snapshot is created, increment the count on
881 * all parents, and when a filesystem/snapshot is destroyed, decrement the
885 dsl_fs_ss_count_adjust(dsl_dir_t *dd, int64_t delta, const char *prop,
889 objset_t *os = dd->dd_pool->dp_meta_objset;
892 ASSERT(dsl_pool_config_held(dd->dd_pool));
893 ASSERT(dmu_tx_is_syncing(tx));
894 ASSERT(strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0 ||
895 strcmp(prop, DD_FIELD_SNAPSHOT_COUNT) == 0);
898 * When we receive an incremental stream into a filesystem that already
899 * exists, a temporary clone is created. We don't count this temporary
900 * clone, whose name begins with a '%'. We also ignore hidden ($FREE,
901 * $MOS & $ORIGIN) objsets.
903 if ((dd->dd_myname[0] == '%' || dd->dd_myname[0] == '$') &&
904 strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0)
908 * e.g. if renaming a dataset with no snapshots, count adjustment is 0
914 * If we hit an uninitialized node while recursing up the tree, we can
915 * stop since we know the counts are not valid on this node and we
916 * know we shouldn't touch this node's counts. An uninitialized count
917 * on the node indicates that either the feature has not yet been
918 * activated or there are no limits on this part of the tree.
920 if (!dsl_dir_is_zapified(dd) || (err = zap_lookup(os, dd->dd_object,
921 prop, sizeof (count), 1, &count)) == ENOENT)
926 /* Use a signed verify to make sure we're not neg. */
927 VERIFY3S(count, >=, 0);
929 VERIFY0(zap_update(os, dd->dd_object, prop, sizeof (count), 1, &count,
932 /* Roll up this additional count into our ancestors */
933 if (dd->dd_parent != NULL)
934 dsl_fs_ss_count_adjust(dd->dd_parent, delta, prop, tx);
938 dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
941 objset_t *mos = dp->dp_meta_objset;
943 dsl_dir_phys_t *ddphys;
946 ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
947 DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
949 VERIFY0(zap_add(mos, dsl_dir_phys(pds)->dd_child_dir_zapobj,
950 name, sizeof (uint64_t), 1, &ddobj, tx));
952 /* it's the root dir */
953 VERIFY0(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
954 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
956 VERIFY0(dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
957 dmu_buf_will_dirty(dbuf, tx);
958 ddphys = dbuf->db_data;
960 ddphys->dd_creation_time = gethrestime_sec();
962 ddphys->dd_parent_obj = pds->dd_object;
964 /* update the filesystem counts */
965 dsl_fs_ss_count_adjust(pds, 1, DD_FIELD_FILESYSTEM_COUNT, tx);
967 ddphys->dd_props_zapobj = zap_create(mos,
968 DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
969 ddphys->dd_child_dir_zapobj = zap_create(mos,
970 DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
971 if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
972 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
974 dmu_buf_rele(dbuf, FTAG);
980 dsl_dir_is_clone(dsl_dir_t *dd)
982 return (dsl_dir_phys(dd)->dd_origin_obj &&
983 (dd->dd_pool->dp_origin_snap == NULL ||
984 dsl_dir_phys(dd)->dd_origin_obj !=
985 dd->dd_pool->dp_origin_snap->ds_object));
989 dsl_dir_get_used(dsl_dir_t *dd)
991 return (dsl_dir_phys(dd)->dd_used_bytes);
995 dsl_dir_get_compressed(dsl_dir_t *dd)
997 return (dsl_dir_phys(dd)->dd_compressed_bytes);
1001 dsl_dir_get_quota(dsl_dir_t *dd)
1003 return (dsl_dir_phys(dd)->dd_quota);
1007 dsl_dir_get_reservation(dsl_dir_t *dd)
1009 return (dsl_dir_phys(dd)->dd_reserved);
1013 dsl_dir_get_compressratio(dsl_dir_t *dd)
1015 /* a fixed point number, 100x the ratio */
1016 return (dsl_dir_phys(dd)->dd_compressed_bytes == 0 ? 100 :
1017 (dsl_dir_phys(dd)->dd_uncompressed_bytes * 100 /
1018 dsl_dir_phys(dd)->dd_compressed_bytes));
1022 dsl_dir_get_logicalused(dsl_dir_t *dd)
1024 return (dsl_dir_phys(dd)->dd_uncompressed_bytes);
1028 dsl_dir_get_usedsnap(dsl_dir_t *dd)
1030 return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]);
1034 dsl_dir_get_usedds(dsl_dir_t *dd)
1036 return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_HEAD]);
1040 dsl_dir_get_usedrefreserv(dsl_dir_t *dd)
1042 return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_REFRSRV]);
1046 dsl_dir_get_usedchild(dsl_dir_t *dd)
1048 return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD] +
1049 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD_RSRV]);
1053 dsl_dir_get_origin(dsl_dir_t *dd, char *buf)
1056 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
1057 dsl_dir_phys(dd)->dd_origin_obj, FTAG, &ds));
1059 dsl_dataset_name(ds, buf);
1061 dsl_dataset_rele(ds, FTAG);
1065 dsl_dir_get_filesystem_count(dsl_dir_t *dd, uint64_t *count)
1067 if (dsl_dir_is_zapified(dd)) {
1068 objset_t *os = dd->dd_pool->dp_meta_objset;
1069 return (zap_lookup(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
1070 sizeof (*count), 1, count));
1072 return (SET_ERROR(ENOENT));
1077 dsl_dir_get_snapshot_count(dsl_dir_t *dd, uint64_t *count)
1079 if (dsl_dir_is_zapified(dd)) {
1080 objset_t *os = dd->dd_pool->dp_meta_objset;
1081 return (zap_lookup(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
1082 sizeof (*count), 1, count));
1084 return (SET_ERROR(ENOENT));
1089 dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
1091 mutex_enter(&dd->dd_lock);
1092 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA,
1093 dsl_dir_get_quota(dd));
1094 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
1095 dsl_dir_get_reservation(dd));
1096 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED,
1097 dsl_dir_get_logicalused(dd));
1098 if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1099 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
1100 dsl_dir_get_usedsnap(dd));
1101 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
1102 dsl_dir_get_usedds(dd));
1103 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
1104 dsl_dir_get_usedrefreserv(dd));
1105 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
1106 dsl_dir_get_usedchild(dd));
1108 mutex_exit(&dd->dd_lock);
1111 if (dsl_dir_get_filesystem_count(dd, &count) == 0) {
1112 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_FILESYSTEM_COUNT,
1115 if (dsl_dir_get_snapshot_count(dd, &count) == 0) {
1116 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_SNAPSHOT_COUNT,
1120 if (dsl_dir_is_clone(dd)) {
1121 char buf[ZFS_MAX_DATASET_NAME_LEN];
1122 dsl_dir_get_origin(dd, buf);
1123 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
1129 dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
1131 dsl_pool_t *dp = dd->dd_pool;
1133 ASSERT(dsl_dir_phys(dd));
1135 if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) {
1136 /* up the hold count until we can be written out */
1137 dmu_buf_add_ref(dd->dd_dbuf, dd);
1142 parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
1144 uint64_t old_accounted = MAX(used, dsl_dir_phys(dd)->dd_reserved);
1145 uint64_t new_accounted =
1146 MAX(used + delta, dsl_dir_phys(dd)->dd_reserved);
1147 return (new_accounted - old_accounted);
1151 dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
1153 ASSERT(dmu_tx_is_syncing(tx));
1155 mutex_enter(&dd->dd_lock);
1156 ASSERT0(dd->dd_tempreserved[tx->tx_txg & TXG_MASK]);
1157 dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
1158 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] / 1024);
1159 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] = 0;
1160 mutex_exit(&dd->dd_lock);
1162 /* release the hold from dsl_dir_dirty */
1163 dmu_buf_rele(dd->dd_dbuf, dd);
1167 dsl_dir_space_towrite(dsl_dir_t *dd)
1171 ASSERT(MUTEX_HELD(&dd->dd_lock));
1173 for (int i = 0; i < TXG_SIZE; i++) {
1174 space += dd->dd_space_towrite[i & TXG_MASK];
1175 ASSERT3U(dd->dd_space_towrite[i & TXG_MASK], >=, 0);
1181 * How much space would dd have available if ancestor had delta applied
1182 * to it? If ondiskonly is set, we're only interested in what's
1183 * on-disk, not estimated pending changes.
1186 dsl_dir_space_available(dsl_dir_t *dd,
1187 dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
1189 uint64_t parentspace, myspace, quota, used;
1192 * If there are no restrictions otherwise, assume we have
1193 * unlimited space available.
1196 parentspace = UINT64_MAX;
1198 if (dd->dd_parent != NULL) {
1199 parentspace = dsl_dir_space_available(dd->dd_parent,
1200 ancestor, delta, ondiskonly);
1203 mutex_enter(&dd->dd_lock);
1204 if (dsl_dir_phys(dd)->dd_quota != 0)
1205 quota = dsl_dir_phys(dd)->dd_quota;
1206 used = dsl_dir_phys(dd)->dd_used_bytes;
1208 used += dsl_dir_space_towrite(dd);
1210 if (dd->dd_parent == NULL) {
1211 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool,
1212 ZFS_SPACE_CHECK_NORMAL);
1213 quota = MIN(quota, poolsize);
1216 if (dsl_dir_phys(dd)->dd_reserved > used && parentspace != UINT64_MAX) {
1218 * We have some space reserved, in addition to what our
1221 parentspace += dsl_dir_phys(dd)->dd_reserved - used;
1224 if (dd == ancestor) {
1226 ASSERT(used >= -delta);
1228 if (parentspace != UINT64_MAX)
1229 parentspace -= delta;
1237 * the lesser of the space provided by our parent and
1238 * the space left in our quota
1240 myspace = MIN(parentspace, quota - used);
1243 mutex_exit(&dd->dd_lock);
1248 struct tempreserve {
1249 list_node_t tr_node;
1255 dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
1256 boolean_t ignorequota, list_t *tr_list,
1257 dmu_tx_t *tx, boolean_t first)
1261 struct tempreserve *tr;
1270 ASSERT3U(txg, !=, 0);
1271 ASSERT3S(asize, >, 0);
1273 mutex_enter(&dd->dd_lock);
1276 * Check against the dsl_dir's quota. We don't add in the delta
1277 * when checking for over-quota because they get one free hit.
1279 uint64_t est_inflight = dsl_dir_space_towrite(dd);
1280 for (int i = 0; i < TXG_SIZE; i++)
1281 est_inflight += dd->dd_tempreserved[i];
1282 uint64_t used_on_disk = dsl_dir_phys(dd)->dd_used_bytes;
1285 * On the first iteration, fetch the dataset's used-on-disk and
1286 * refreservation values. Also, if checkrefquota is set, test if
1287 * allocating this space would exceed the dataset's refquota.
1289 if (first && tx->tx_objset) {
1291 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
1293 error = dsl_dataset_check_quota(ds, !netfree,
1294 asize, est_inflight, &used_on_disk, &ref_rsrv);
1296 mutex_exit(&dd->dd_lock);
1297 DMU_TX_STAT_BUMP(dmu_tx_quota);
1303 * If this transaction will result in a net free of space,
1304 * we want to let it through.
1306 if (ignorequota || netfree || dsl_dir_phys(dd)->dd_quota == 0)
1309 quota = dsl_dir_phys(dd)->dd_quota;
1312 * Adjust the quota against the actual pool size at the root
1313 * minus any outstanding deferred frees.
1314 * To ensure that it's possible to remove files from a full
1315 * pool without inducing transient overcommits, we throttle
1316 * netfree transactions against a quota that is slightly larger,
1317 * but still within the pool's allocation slop. In cases where
1318 * we're very close to full, this will allow a steady trickle of
1319 * removes to get through.
1321 uint64_t deferred = 0;
1322 if (dd->dd_parent == NULL) {
1323 uint64_t avail = dsl_pool_unreserved_space(dd->dd_pool,
1325 ZFS_SPACE_CHECK_RESERVED : ZFS_SPACE_CHECK_NORMAL);
1327 if (avail < quota) {
1329 retval = SET_ERROR(ENOSPC);
1334 * If they are requesting more space, and our current estimate
1335 * is over quota, they get to try again unless the actual
1336 * on-disk is over quota and there are no pending changes (which
1337 * may free up space for us).
1339 if (used_on_disk + est_inflight >= quota) {
1340 if (est_inflight > 0 || used_on_disk < quota ||
1341 (retval == ENOSPC && used_on_disk < quota + deferred))
1343 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
1344 "quota=%lluK tr=%lluK err=%d\n",
1345 used_on_disk>>10, est_inflight>>10,
1346 quota>>10, asize>>10, retval);
1347 mutex_exit(&dd->dd_lock);
1348 DMU_TX_STAT_BUMP(dmu_tx_quota);
1349 return (SET_ERROR(retval));
1352 /* We need to up our estimated delta before dropping dd_lock */
1353 dd->dd_tempreserved[txg & TXG_MASK] += asize;
1355 uint64_t parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
1357 mutex_exit(&dd->dd_lock);
1359 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1361 tr->tr_size = asize;
1362 list_insert_tail(tr_list, tr);
1364 /* see if it's OK with our parent */
1365 if (dd->dd_parent != NULL && parent_rsrv != 0) {
1367 * Recurse on our parent without recursion. This has been
1368 * observed to be potentially large stack usage even within
1369 * the test suite. Largest seen stack was 7632 bytes on linux.
1373 asize = parent_rsrv;
1374 ignorequota = (dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
1376 goto top_of_function;
1384 * Reserve space in this dsl_dir, to be used in this tx's txg.
1385 * After the space has been dirtied (and dsl_dir_willuse_space()
1386 * has been called), the reservation should be canceled, using
1387 * dsl_dir_tempreserve_clear().
1390 dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
1391 boolean_t netfree, void **tr_cookiep, dmu_tx_t *tx)
1401 tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
1402 list_create(tr_list, sizeof (struct tempreserve),
1403 offsetof(struct tempreserve, tr_node));
1404 ASSERT3S(asize, >, 0);
1406 err = arc_tempreserve_space(dd->dd_pool->dp_spa, lsize, tx->tx_txg);
1408 struct tempreserve *tr;
1410 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1411 tr->tr_size = lsize;
1412 list_insert_tail(tr_list, tr);
1414 if (err == EAGAIN) {
1416 * If arc_memory_throttle() detected that pageout
1417 * is running and we are low on memory, we delay new
1418 * non-pageout transactions to give pageout an
1421 * It is unfortunate to be delaying while the caller's
1424 txg_delay(dd->dd_pool, tx->tx_txg,
1425 MSEC2NSEC(10), MSEC2NSEC(10));
1426 err = SET_ERROR(ERESTART);
1431 err = dsl_dir_tempreserve_impl(dd, asize, netfree,
1432 B_FALSE, tr_list, tx, B_TRUE);
1436 dsl_dir_tempreserve_clear(tr_list, tx);
1438 *tr_cookiep = tr_list;
1444 * Clear a temporary reservation that we previously made with
1445 * dsl_dir_tempreserve_space().
1448 dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
1450 int txgidx = tx->tx_txg & TXG_MASK;
1451 list_t *tr_list = tr_cookie;
1452 struct tempreserve *tr;
1454 ASSERT3U(tx->tx_txg, !=, 0);
1456 if (tr_cookie == NULL)
1459 while ((tr = list_head(tr_list)) != NULL) {
1461 mutex_enter(&tr->tr_ds->dd_lock);
1462 ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
1464 tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
1465 mutex_exit(&tr->tr_ds->dd_lock);
1467 arc_tempreserve_clear(tr->tr_size);
1469 list_remove(tr_list, tr);
1470 kmem_free(tr, sizeof (struct tempreserve));
1473 kmem_free(tr_list, sizeof (list_t));
1477 * This should be called from open context when we think we're going to write
1478 * or free space, for example when dirtying data. Be conservative; it's okay
1479 * to write less space or free more, but we don't want to write more or free
1480 * less than the amount specified.
1482 * NOTE: The behavior of this function is identical to the Illumos / FreeBSD
1483 * version however it has been adjusted to use an iterative rather than
1484 * recursive algorithm to minimize stack usage.
1487 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
1489 int64_t parent_space;
1493 mutex_enter(&dd->dd_lock);
1495 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
1497 est_used = dsl_dir_space_towrite(dd) +
1498 dsl_dir_phys(dd)->dd_used_bytes;
1499 parent_space = parent_delta(dd, est_used, space);
1500 mutex_exit(&dd->dd_lock);
1502 /* Make sure that we clean up dd_space_to* */
1503 dsl_dir_dirty(dd, tx);
1506 space = parent_space;
1507 } while (space && dd);
1510 /* call from syncing context when we actually write/free space for this dd */
1512 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
1513 int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
1515 int64_t accounted_delta;
1518 * dsl_dataset_set_refreservation_sync_impl() calls this with
1519 * dd_lock held, so that it can atomically update
1520 * ds->ds_reserved and the dsl_dir accounting, so that
1521 * dsl_dataset_check_quota() can see dataset and dir accounting
1524 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
1526 ASSERT(dmu_tx_is_syncing(tx));
1527 ASSERT(type < DD_USED_NUM);
1529 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1532 mutex_enter(&dd->dd_lock);
1534 parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, used);
1535 ASSERT(used >= 0 || dsl_dir_phys(dd)->dd_used_bytes >= -used);
1536 ASSERT(compressed >= 0 ||
1537 dsl_dir_phys(dd)->dd_compressed_bytes >= -compressed);
1538 ASSERT(uncompressed >= 0 ||
1539 dsl_dir_phys(dd)->dd_uncompressed_bytes >= -uncompressed);
1540 dsl_dir_phys(dd)->dd_used_bytes += used;
1541 dsl_dir_phys(dd)->dd_uncompressed_bytes += uncompressed;
1542 dsl_dir_phys(dd)->dd_compressed_bytes += compressed;
1544 if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1546 dsl_dir_phys(dd)->dd_used_breakdown[type] >= -used);
1547 dsl_dir_phys(dd)->dd_used_breakdown[type] += used;
1552 for (t = 0; t < DD_USED_NUM; t++)
1553 u += dsl_dir_phys(dd)->dd_used_breakdown[t];
1554 ASSERT3U(u, ==, dsl_dir_phys(dd)->dd_used_bytes);
1559 mutex_exit(&dd->dd_lock);
1561 if (dd->dd_parent != NULL) {
1562 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1563 accounted_delta, compressed, uncompressed, tx);
1564 dsl_dir_transfer_space(dd->dd_parent,
1565 used - accounted_delta,
1566 DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
1571 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
1572 dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
1574 ASSERT(dmu_tx_is_syncing(tx));
1575 ASSERT(oldtype < DD_USED_NUM);
1576 ASSERT(newtype < DD_USED_NUM);
1579 !(dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN))
1582 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1583 mutex_enter(&dd->dd_lock);
1585 dsl_dir_phys(dd)->dd_used_breakdown[oldtype] >= delta :
1586 dsl_dir_phys(dd)->dd_used_breakdown[newtype] >= -delta);
1587 ASSERT(dsl_dir_phys(dd)->dd_used_bytes >= ABS(delta));
1588 dsl_dir_phys(dd)->dd_used_breakdown[oldtype] -= delta;
1589 dsl_dir_phys(dd)->dd_used_breakdown[newtype] += delta;
1590 mutex_exit(&dd->dd_lock);
1593 typedef struct dsl_dir_set_qr_arg {
1594 const char *ddsqra_name;
1595 zprop_source_t ddsqra_source;
1596 uint64_t ddsqra_value;
1597 } dsl_dir_set_qr_arg_t;
1600 dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx)
1602 dsl_dir_set_qr_arg_t *ddsqra = arg;
1603 dsl_pool_t *dp = dmu_tx_pool(tx);
1606 uint64_t towrite, newval;
1608 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1612 error = dsl_prop_predict(ds->ds_dir, "quota",
1613 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1615 dsl_dataset_rele(ds, FTAG);
1620 dsl_dataset_rele(ds, FTAG);
1624 mutex_enter(&ds->ds_dir->dd_lock);
1626 * If we are doing the preliminary check in open context, and
1627 * there are pending changes, then don't fail it, since the
1628 * pending changes could under-estimate the amount of space to be
1631 towrite = dsl_dir_space_towrite(ds->ds_dir);
1632 if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
1633 (newval < dsl_dir_phys(ds->ds_dir)->dd_reserved ||
1634 newval < dsl_dir_phys(ds->ds_dir)->dd_used_bytes + towrite)) {
1635 error = SET_ERROR(ENOSPC);
1637 mutex_exit(&ds->ds_dir->dd_lock);
1638 dsl_dataset_rele(ds, FTAG);
1643 dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx)
1645 dsl_dir_set_qr_arg_t *ddsqra = arg;
1646 dsl_pool_t *dp = dmu_tx_pool(tx);
1650 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1652 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1653 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA),
1654 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1655 &ddsqra->ddsqra_value, tx);
1657 VERIFY0(dsl_prop_get_int_ds(ds,
1658 zfs_prop_to_name(ZFS_PROP_QUOTA), &newval));
1660 newval = ddsqra->ddsqra_value;
1661 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1662 zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval);
1665 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1666 mutex_enter(&ds->ds_dir->dd_lock);
1667 dsl_dir_phys(ds->ds_dir)->dd_quota = newval;
1668 mutex_exit(&ds->ds_dir->dd_lock);
1669 dsl_dataset_rele(ds, FTAG);
1673 dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
1675 dsl_dir_set_qr_arg_t ddsqra;
1677 ddsqra.ddsqra_name = ddname;
1678 ddsqra.ddsqra_source = source;
1679 ddsqra.ddsqra_value = quota;
1681 return (dsl_sync_task(ddname, dsl_dir_set_quota_check,
1682 dsl_dir_set_quota_sync, &ddsqra, 0,
1683 ZFS_SPACE_CHECK_EXTRA_RESERVED));
1687 dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx)
1689 dsl_dir_set_qr_arg_t *ddsqra = arg;
1690 dsl_pool_t *dp = dmu_tx_pool(tx);
1693 uint64_t newval, used, avail;
1696 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1702 * If we are doing the preliminary check in open context, the
1703 * space estimates may be inaccurate.
1705 if (!dmu_tx_is_syncing(tx)) {
1706 dsl_dataset_rele(ds, FTAG);
1710 error = dsl_prop_predict(ds->ds_dir,
1711 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1712 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1714 dsl_dataset_rele(ds, FTAG);
1718 mutex_enter(&dd->dd_lock);
1719 used = dsl_dir_phys(dd)->dd_used_bytes;
1720 mutex_exit(&dd->dd_lock);
1722 if (dd->dd_parent) {
1723 avail = dsl_dir_space_available(dd->dd_parent,
1726 avail = dsl_pool_adjustedsize(dd->dd_pool,
1727 ZFS_SPACE_CHECK_NORMAL) - used;
1730 if (MAX(used, newval) > MAX(used, dsl_dir_phys(dd)->dd_reserved)) {
1731 uint64_t delta = MAX(used, newval) -
1732 MAX(used, dsl_dir_phys(dd)->dd_reserved);
1734 if (delta > avail ||
1735 (dsl_dir_phys(dd)->dd_quota > 0 &&
1736 newval > dsl_dir_phys(dd)->dd_quota))
1737 error = SET_ERROR(ENOSPC);
1740 dsl_dataset_rele(ds, FTAG);
1745 dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
1750 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1752 mutex_enter(&dd->dd_lock);
1753 used = dsl_dir_phys(dd)->dd_used_bytes;
1754 delta = MAX(used, value) - MAX(used, dsl_dir_phys(dd)->dd_reserved);
1755 dsl_dir_phys(dd)->dd_reserved = value;
1757 if (dd->dd_parent != NULL) {
1758 /* Roll up this additional usage into our ancestors */
1759 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1762 mutex_exit(&dd->dd_lock);
1766 dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx)
1768 dsl_dir_set_qr_arg_t *ddsqra = arg;
1769 dsl_pool_t *dp = dmu_tx_pool(tx);
1773 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1775 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1776 dsl_prop_set_sync_impl(ds,
1777 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1778 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1779 &ddsqra->ddsqra_value, tx);
1781 VERIFY0(dsl_prop_get_int_ds(ds,
1782 zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval));
1784 newval = ddsqra->ddsqra_value;
1785 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1786 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1787 (longlong_t)newval);
1790 dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx);
1791 dsl_dataset_rele(ds, FTAG);
1795 dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1796 uint64_t reservation)
1798 dsl_dir_set_qr_arg_t ddsqra;
1800 ddsqra.ddsqra_name = ddname;
1801 ddsqra.ddsqra_source = source;
1802 ddsqra.ddsqra_value = reservation;
1804 return (dsl_sync_task(ddname, dsl_dir_set_reservation_check,
1805 dsl_dir_set_reservation_sync, &ddsqra, 0,
1806 ZFS_SPACE_CHECK_EXTRA_RESERVED));
1810 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1812 for (; ds1; ds1 = ds1->dd_parent) {
1814 for (dd = ds2; dd; dd = dd->dd_parent) {
1823 * If delta is applied to dd, how much of that delta would be applied to
1824 * ancestor? Syncing context only.
1827 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1832 mutex_enter(&dd->dd_lock);
1833 delta = parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, delta);
1834 mutex_exit(&dd->dd_lock);
1835 return (would_change(dd->dd_parent, delta, ancestor));
1838 typedef struct dsl_dir_rename_arg {
1839 const char *ddra_oldname;
1840 const char *ddra_newname;
1842 } dsl_dir_rename_arg_t;
1844 typedef struct dsl_valid_rename_arg {
1847 } dsl_valid_rename_arg_t;
1851 dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1853 dsl_valid_rename_arg_t *dvra = arg;
1854 char namebuf[ZFS_MAX_DATASET_NAME_LEN];
1856 dsl_dataset_name(ds, namebuf);
1858 ASSERT3U(strnlen(namebuf, ZFS_MAX_DATASET_NAME_LEN),
1859 <, ZFS_MAX_DATASET_NAME_LEN);
1860 int namelen = strlen(namebuf) + dvra->char_delta;
1861 int depth = get_dataset_depth(namebuf) + dvra->nest_delta;
1863 if (namelen >= ZFS_MAX_DATASET_NAME_LEN)
1864 return (SET_ERROR(ENAMETOOLONG));
1865 if (dvra->nest_delta > 0 && depth >= zfs_max_dataset_nesting)
1866 return (SET_ERROR(ENAMETOOLONG));
1871 dsl_dir_rename_check(void *arg, dmu_tx_t *tx)
1873 dsl_dir_rename_arg_t *ddra = arg;
1874 dsl_pool_t *dp = dmu_tx_pool(tx);
1875 dsl_dir_t *dd, *newparent;
1876 dsl_valid_rename_arg_t dvra;
1877 dsl_dataset_t *parentds;
1879 const char *mynewname;
1882 /* target dir should exist */
1883 error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL);
1887 /* new parent should exist */
1888 error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG,
1889 &newparent, &mynewname);
1891 dsl_dir_rele(dd, FTAG);
1895 /* can't rename to different pool */
1896 if (dd->dd_pool != newparent->dd_pool) {
1897 dsl_dir_rele(newparent, FTAG);
1898 dsl_dir_rele(dd, FTAG);
1899 return (SET_ERROR(EXDEV));
1902 /* new name should not already exist */
1903 if (mynewname == NULL) {
1904 dsl_dir_rele(newparent, FTAG);
1905 dsl_dir_rele(dd, FTAG);
1906 return (SET_ERROR(EEXIST));
1909 /* can't rename below anything but filesystems (eg. no ZVOLs) */
1910 error = dsl_dataset_hold_obj(newparent->dd_pool,
1911 dsl_dir_phys(newparent)->dd_head_dataset_obj, FTAG, &parentds);
1913 dsl_dir_rele(newparent, FTAG);
1914 dsl_dir_rele(dd, FTAG);
1917 error = dmu_objset_from_ds(parentds, &parentos);
1919 dsl_dataset_rele(parentds, FTAG);
1920 dsl_dir_rele(newparent, FTAG);
1921 dsl_dir_rele(dd, FTAG);
1924 if (dmu_objset_type(parentos) != DMU_OST_ZFS) {
1925 dsl_dataset_rele(parentds, FTAG);
1926 dsl_dir_rele(newparent, FTAG);
1927 dsl_dir_rele(dd, FTAG);
1928 return (SET_ERROR(ZFS_ERR_WRONG_PARENT));
1930 dsl_dataset_rele(parentds, FTAG);
1932 ASSERT3U(strnlen(ddra->ddra_newname, ZFS_MAX_DATASET_NAME_LEN),
1933 <, ZFS_MAX_DATASET_NAME_LEN);
1934 ASSERT3U(strnlen(ddra->ddra_oldname, ZFS_MAX_DATASET_NAME_LEN),
1935 <, ZFS_MAX_DATASET_NAME_LEN);
1936 dvra.char_delta = strlen(ddra->ddra_newname)
1937 - strlen(ddra->ddra_oldname);
1938 dvra.nest_delta = get_dataset_depth(ddra->ddra_newname)
1939 - get_dataset_depth(ddra->ddra_oldname);
1941 /* if the name length is growing, validate child name lengths */
1942 if (dvra.char_delta > 0 || dvra.nest_delta > 0) {
1943 error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename,
1944 &dvra, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1946 dsl_dir_rele(newparent, FTAG);
1947 dsl_dir_rele(dd, FTAG);
1952 if (dmu_tx_is_syncing(tx)) {
1953 if (spa_feature_is_active(dp->dp_spa,
1954 SPA_FEATURE_FS_SS_LIMIT)) {
1956 * Although this is the check function and we don't
1957 * normally make on-disk changes in check functions,
1958 * we need to do that here.
1960 * Ensure this portion of the tree's counts have been
1961 * initialized in case the new parent has limits set.
1963 dsl_dir_init_fs_ss_count(dd, tx);
1967 if (newparent != dd->dd_parent) {
1968 /* is there enough space? */
1970 MAX(dsl_dir_phys(dd)->dd_used_bytes,
1971 dsl_dir_phys(dd)->dd_reserved);
1972 objset_t *os = dd->dd_pool->dp_meta_objset;
1973 uint64_t fs_cnt = 0;
1974 uint64_t ss_cnt = 0;
1976 if (dsl_dir_is_zapified(dd)) {
1979 err = zap_lookup(os, dd->dd_object,
1980 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1982 if (err != ENOENT && err != 0) {
1983 dsl_dir_rele(newparent, FTAG);
1984 dsl_dir_rele(dd, FTAG);
1989 * have to add 1 for the filesystem itself that we're
1994 err = zap_lookup(os, dd->dd_object,
1995 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1997 if (err != ENOENT && err != 0) {
1998 dsl_dir_rele(newparent, FTAG);
1999 dsl_dir_rele(dd, FTAG);
2004 /* check for encryption errors */
2005 error = dsl_dir_rename_crypt_check(dd, newparent);
2007 dsl_dir_rele(newparent, FTAG);
2008 dsl_dir_rele(dd, FTAG);
2009 return (SET_ERROR(EACCES));
2012 /* no rename into our descendant */
2013 if (closest_common_ancestor(dd, newparent) == dd) {
2014 dsl_dir_rele(newparent, FTAG);
2015 dsl_dir_rele(dd, FTAG);
2016 return (SET_ERROR(EINVAL));
2019 error = dsl_dir_transfer_possible(dd->dd_parent,
2020 newparent, fs_cnt, ss_cnt, myspace, ddra->ddra_cred);
2022 dsl_dir_rele(newparent, FTAG);
2023 dsl_dir_rele(dd, FTAG);
2028 dsl_dir_rele(newparent, FTAG);
2029 dsl_dir_rele(dd, FTAG);
2034 dsl_dir_rename_sync(void *arg, dmu_tx_t *tx)
2036 dsl_dir_rename_arg_t *ddra = arg;
2037 dsl_pool_t *dp = dmu_tx_pool(tx);
2038 dsl_dir_t *dd, *newparent;
2039 const char *mynewname;
2041 objset_t *mos = dp->dp_meta_objset;
2043 VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL));
2044 VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
2047 /* Log this before we change the name. */
2048 spa_history_log_internal_dd(dd, "rename", tx,
2049 "-> %s", ddra->ddra_newname);
2051 if (newparent != dd->dd_parent) {
2052 objset_t *os = dd->dd_pool->dp_meta_objset;
2053 uint64_t fs_cnt = 0;
2054 uint64_t ss_cnt = 0;
2057 * We already made sure the dd counts were initialized in the
2060 if (spa_feature_is_active(dp->dp_spa,
2061 SPA_FEATURE_FS_SS_LIMIT)) {
2062 VERIFY0(zap_lookup(os, dd->dd_object,
2063 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
2065 /* add 1 for the filesystem itself that we're moving */
2068 VERIFY0(zap_lookup(os, dd->dd_object,
2069 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
2073 dsl_fs_ss_count_adjust(dd->dd_parent, -fs_cnt,
2074 DD_FIELD_FILESYSTEM_COUNT, tx);
2075 dsl_fs_ss_count_adjust(newparent, fs_cnt,
2076 DD_FIELD_FILESYSTEM_COUNT, tx);
2078 dsl_fs_ss_count_adjust(dd->dd_parent, -ss_cnt,
2079 DD_FIELD_SNAPSHOT_COUNT, tx);
2080 dsl_fs_ss_count_adjust(newparent, ss_cnt,
2081 DD_FIELD_SNAPSHOT_COUNT, tx);
2083 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
2084 -dsl_dir_phys(dd)->dd_used_bytes,
2085 -dsl_dir_phys(dd)->dd_compressed_bytes,
2086 -dsl_dir_phys(dd)->dd_uncompressed_bytes, tx);
2087 dsl_dir_diduse_space(newparent, DD_USED_CHILD,
2088 dsl_dir_phys(dd)->dd_used_bytes,
2089 dsl_dir_phys(dd)->dd_compressed_bytes,
2090 dsl_dir_phys(dd)->dd_uncompressed_bytes, tx);
2092 if (dsl_dir_phys(dd)->dd_reserved >
2093 dsl_dir_phys(dd)->dd_used_bytes) {
2094 uint64_t unused_rsrv = dsl_dir_phys(dd)->dd_reserved -
2095 dsl_dir_phys(dd)->dd_used_bytes;
2097 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
2098 -unused_rsrv, 0, 0, tx);
2099 dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV,
2100 unused_rsrv, 0, 0, tx);
2104 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2106 /* remove from old parent zapobj */
2107 error = zap_remove(mos,
2108 dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj,
2112 (void) strlcpy(dd->dd_myname, mynewname,
2113 sizeof (dd->dd_myname));
2114 dsl_dir_rele(dd->dd_parent, dd);
2115 dsl_dir_phys(dd)->dd_parent_obj = newparent->dd_object;
2116 VERIFY0(dsl_dir_hold_obj(dp,
2117 newparent->dd_object, NULL, dd, &dd->dd_parent));
2119 /* add to new parent zapobj */
2120 VERIFY0(zap_add(mos, dsl_dir_phys(newparent)->dd_child_dir_zapobj,
2121 dd->dd_myname, 8, 1, &dd->dd_object, tx));
2123 zvol_rename_minors(dp->dp_spa, ddra->ddra_oldname,
2124 ddra->ddra_newname, B_TRUE);
2126 dsl_prop_notify_all(dd);
2128 dsl_dir_rele(newparent, FTAG);
2129 dsl_dir_rele(dd, FTAG);
2133 dsl_dir_rename(const char *oldname, const char *newname)
2135 dsl_dir_rename_arg_t ddra;
2137 ddra.ddra_oldname = oldname;
2138 ddra.ddra_newname = newname;
2139 ddra.ddra_cred = CRED();
2141 return (dsl_sync_task(oldname,
2142 dsl_dir_rename_check, dsl_dir_rename_sync, &ddra,
2143 3, ZFS_SPACE_CHECK_RESERVED));
2147 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd,
2148 uint64_t fs_cnt, uint64_t ss_cnt, uint64_t space, cred_t *cr)
2150 dsl_dir_t *ancestor;
2155 ancestor = closest_common_ancestor(sdd, tdd);
2156 adelta = would_change(sdd, -space, ancestor);
2157 avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
2159 return (SET_ERROR(ENOSPC));
2161 err = dsl_fs_ss_limit_check(tdd, fs_cnt, ZFS_PROP_FILESYSTEM_LIMIT,
2165 err = dsl_fs_ss_limit_check(tdd, ss_cnt, ZFS_PROP_SNAPSHOT_LIMIT,
2174 dsl_dir_snap_cmtime(dsl_dir_t *dd)
2178 mutex_enter(&dd->dd_lock);
2179 t = dd->dd_snap_cmtime;
2180 mutex_exit(&dd->dd_lock);
2186 dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
2191 mutex_enter(&dd->dd_lock);
2192 dd->dd_snap_cmtime = t;
2193 mutex_exit(&dd->dd_lock);
2197 dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx)
2199 objset_t *mos = dd->dd_pool->dp_meta_objset;
2200 dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx);
2204 dsl_dir_is_zapified(dsl_dir_t *dd)
2206 dmu_object_info_t doi;
2208 dmu_object_info_from_db(dd->dd_dbuf, &doi);
2209 return (doi.doi_type == DMU_OTN_ZAP_METADATA);
2213 dsl_dir_livelist_open(dsl_dir_t *dd, uint64_t obj)
2215 objset_t *mos = dd->dd_pool->dp_meta_objset;
2216 ASSERT(spa_feature_is_active(dd->dd_pool->dp_spa,
2217 SPA_FEATURE_LIVELIST));
2218 dsl_deadlist_open(&dd->dd_livelist, mos, obj);
2219 bplist_create(&dd->dd_pending_allocs);
2220 bplist_create(&dd->dd_pending_frees);
2224 dsl_dir_livelist_close(dsl_dir_t *dd)
2226 dsl_deadlist_close(&dd->dd_livelist);
2227 bplist_destroy(&dd->dd_pending_allocs);
2228 bplist_destroy(&dd->dd_pending_frees);
2232 dsl_dir_remove_livelist(dsl_dir_t *dd, dmu_tx_t *tx, boolean_t total)
2235 dsl_pool_t *dp = dmu_tx_pool(tx);
2236 spa_t *spa = dp->dp_spa;
2237 livelist_condense_entry_t to_condense = spa->spa_to_condense;
2239 if (!dsl_deadlist_is_open(&dd->dd_livelist))
2243 * If the livelist being removed is set to be condensed, stop the
2244 * condense zthr and indicate the cancellation in the spa_to_condense
2245 * struct in case the condense no-wait synctask has already started
2247 zthr_t *ll_condense_thread = spa->spa_livelist_condense_zthr;
2248 if (ll_condense_thread != NULL &&
2249 (to_condense.ds != NULL) && (to_condense.ds->ds_dir == dd)) {
2251 * We use zthr_wait_cycle_done instead of zthr_cancel
2252 * because we don't want to destroy the zthr, just have
2253 * it skip its current task.
2255 spa->spa_to_condense.cancelled = B_TRUE;
2256 zthr_wait_cycle_done(ll_condense_thread);
2258 * If we've returned from zthr_wait_cycle_done without
2259 * clearing the to_condense data structure it's either
2260 * because the no-wait synctask has started (which is
2261 * indicated by 'syncing' field of to_condense) and we
2262 * can expect it to clear to_condense on its own.
2263 * Otherwise, we returned before the zthr ran. The
2264 * checkfunc will now fail as cancelled == B_TRUE so we
2265 * can safely NULL out ds, allowing a different dir's
2266 * livelist to be condensed.
2268 * We can be sure that the to_condense struct will not
2269 * be repopulated at this stage because both this
2270 * function and dsl_livelist_try_condense execute in
2273 if ((spa->spa_to_condense.ds != NULL) &&
2274 !spa->spa_to_condense.syncing) {
2275 dmu_buf_rele(spa->spa_to_condense.ds->ds_dbuf,
2277 spa->spa_to_condense.ds = NULL;
2281 dsl_dir_livelist_close(dd);
2282 int err = zap_lookup(dp->dp_meta_objset, dd->dd_object,
2283 DD_FIELD_LIVELIST, sizeof (uint64_t), 1, &obj);
2285 VERIFY0(zap_remove(dp->dp_meta_objset, dd->dd_object,
2286 DD_FIELD_LIVELIST, tx));
2288 dsl_deadlist_free(dp->dp_meta_objset, obj, tx);
2289 spa_feature_decr(spa, SPA_FEATURE_LIVELIST, tx);
2292 ASSERT3U(err, !=, ENOENT);
2297 dsl_dir_activity_in_progress(dsl_dir_t *dd, dsl_dataset_t *ds,
2298 zfs_wait_activity_t activity, boolean_t *in_progress)
2302 ASSERT(MUTEX_HELD(&dd->dd_activity_lock));
2305 case ZFS_WAIT_DELETEQ: {
2308 error = dmu_objset_from_ds(ds, &os);
2312 mutex_enter(&os->os_user_ptr_lock);
2313 void *user = dmu_objset_get_user(os);
2314 mutex_exit(&os->os_user_ptr_lock);
2315 if (dmu_objset_type(os) != DMU_OST_ZFS ||
2316 user == NULL || zfs_get_vfs_flag_unmounted(os)) {
2317 *in_progress = B_FALSE;
2321 uint64_t readonly = B_FALSE;
2322 error = zfs_get_temporary_prop(ds, ZFS_PROP_READONLY, &readonly,
2328 if (readonly || !spa_writeable(dd->dd_pool->dp_spa)) {
2329 *in_progress = B_FALSE;
2333 uint64_t count, unlinked_obj;
2334 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
2337 dsl_dataset_rele(ds, FTAG);
2340 error = zap_count(os, unlinked_obj, &count);
2343 *in_progress = (count != 0);
2347 * The delete queue is ZPL specific, and libzpool doesn't have
2348 * it. It doesn't make sense to wait for it.
2350 *in_progress = B_FALSE;
2355 panic("unrecognized value for activity %d", activity);
2362 dsl_dir_wait(dsl_dir_t *dd, dsl_dataset_t *ds, zfs_wait_activity_t activity,
2366 boolean_t in_progress;
2367 dsl_pool_t *dp = dd->dd_pool;
2369 dsl_pool_config_enter(dp, FTAG);
2370 error = dsl_dir_activity_in_progress(dd, ds, activity,
2372 dsl_pool_config_exit(dp, FTAG);
2373 if (error != 0 || !in_progress)
2378 if (cv_wait_sig(&dd->dd_activity_cv, &dd->dd_activity_lock) ==
2379 0 || dd->dd_activity_cancelled) {
2380 error = SET_ERROR(EINTR);
2388 dsl_dir_cancel_waiters(dsl_dir_t *dd)
2390 mutex_enter(&dd->dd_activity_lock);
2391 dd->dd_activity_cancelled = B_TRUE;
2392 cv_broadcast(&dd->dd_activity_cv);
2393 while (dd->dd_activity_waiters > 0)
2394 cv_wait(&dd->dd_activity_cv, &dd->dd_activity_lock);
2395 mutex_exit(&dd->dd_activity_lock);
2398 #if defined(_KERNEL)
2399 EXPORT_SYMBOL(dsl_dir_set_quota);
2400 EXPORT_SYMBOL(dsl_dir_set_reservation);