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, 2016 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.
31 #include <sys/dmu_objset.h>
32 #include <sys/dmu_tx.h>
33 #include <sys/dsl_dataset.h>
34 #include <sys/dsl_dir.h>
35 #include <sys/dsl_prop.h>
36 #include <sys/dsl_synctask.h>
37 #include <sys/dsl_deleg.h>
38 #include <sys/dmu_impl.h>
40 #include <sys/metaslab.h>
44 #include <sys/sunddi.h>
45 #include <sys/zfeature.h>
46 #include <sys/policy.h>
47 #include <sys/zfs_znode.h>
49 #include "zfs_namecheck.h"
53 * Filesystem and Snapshot Limits
54 * ------------------------------
56 * These limits are used to restrict the number of filesystems and/or snapshots
57 * that can be created at a given level in the tree or below. A typical
58 * use-case is with a delegated dataset where the administrator wants to ensure
59 * that a user within the zone is not creating too many additional filesystems
60 * or snapshots, even though they're not exceeding their space quota.
62 * The filesystem and snapshot counts are stored as extensible properties. This
63 * capability is controlled by a feature flag and must be enabled to be used.
64 * Once enabled, the feature is not active until the first limit is set. At
65 * that point, future operations to create/destroy filesystems or snapshots
66 * will validate and update the counts.
68 * Because the count properties will not exist before the feature is active,
69 * the counts are updated when a limit is first set on an uninitialized
70 * dsl_dir node in the tree (The filesystem/snapshot count on a node includes
71 * all of the nested filesystems/snapshots. Thus, a new leaf node has a
72 * filesystem count of 0 and a snapshot count of 0. Non-existent filesystem and
73 * snapshot count properties on a node indicate uninitialized counts on that
74 * node.) When first setting a limit on an uninitialized node, the code starts
75 * at the filesystem with the new limit and descends into all sub-filesystems
76 * to add the count properties.
78 * In practice this is lightweight since a limit is typically set when the
79 * filesystem is created and thus has no children. Once valid, changing the
80 * limit value won't require a re-traversal since the counts are already valid.
81 * When recursively fixing the counts, if a node with a limit is encountered
82 * during the descent, the counts are known to be valid and there is no need to
83 * descend into that filesystem's children. The counts on filesystems above the
84 * one with the new limit will still be uninitialized, unless a limit is
85 * eventually set on one of those filesystems. The counts are always recursively
86 * updated when a limit is set on a dataset, unless there is already a limit.
87 * When a new limit value is set on a filesystem with an existing limit, it is
88 * possible for the new limit to be less than the current count at that level
89 * since a user who can change the limit is also allowed to exceed the limit.
91 * Once the feature is active, then whenever a filesystem or snapshot is
92 * created, the code recurses up the tree, validating the new count against the
93 * limit at each initialized level. In practice, most levels will not have a
94 * limit set. If there is a limit at any initialized level up the tree, the
95 * check must pass or the creation will fail. Likewise, when a filesystem or
96 * snapshot is destroyed, the counts are recursively adjusted all the way up
97 * the initizized nodes in the tree. Renaming a filesystem into different point
98 * in the tree will first validate, then update the counts on each branch up to
99 * the common ancestor. A receive will also validate the counts and then update
102 * An exception to the above behavior is that the limit is not enforced if the
103 * user has permission to modify the limit. This is primarily so that
104 * recursive snapshots in the global zone always work. We want to prevent a
105 * denial-of-service in which a lower level delegated dataset could max out its
106 * limit and thus block recursive snapshots from being taken in the global zone.
107 * Because of this, it is possible for the snapshot count to be over the limit
108 * and snapshots taken in the global zone could cause a lower level dataset to
109 * hit or exceed its limit. The administrator taking the global zone recursive
110 * snapshot should be aware of this side-effect and behave accordingly.
111 * For consistency, the filesystem limit is also not enforced if the user can
114 * The filesystem and snapshot limits are validated by dsl_fs_ss_limit_check()
115 * and updated by dsl_fs_ss_count_adjust(). A new limit value is setup in
116 * dsl_dir_activate_fs_ss_limit() and the counts are adjusted, if necessary, by
117 * dsl_dir_init_fs_ss_count().
119 * There is a special case when we receive a filesystem that already exists. In
120 * this case a temporary clone name of %X is created (see dmu_recv_begin). We
121 * never update the filesystem counts for temporary clones.
123 * Likewise, we do not update the snapshot counts for temporary snapshots,
124 * such as those created by zfs diff.
127 extern inline dsl_dir_phys_t *dsl_dir_phys(dsl_dir_t *dd);
129 static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
132 dsl_dir_evict_async(void *dbu)
136 ASSERTV(dsl_pool_t *dp = dd->dd_pool);
140 for (t = 0; t < TXG_SIZE; t++) {
141 ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t));
142 ASSERT(dd->dd_tempreserved[t] == 0);
143 ASSERT(dd->dd_space_towrite[t] == 0);
147 dsl_dir_async_rele(dd->dd_parent, dd);
149 spa_async_close(dd->dd_pool->dp_spa, dd);
152 mutex_destroy(&dd->dd_lock);
153 kmem_free(dd, sizeof (dsl_dir_t));
157 dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj,
158 const char *tail, void *tag, dsl_dir_t **ddp)
162 dmu_object_info_t doi;
165 ASSERT(dsl_pool_config_held(dp));
167 err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
170 dd = dmu_buf_get_user(dbuf);
172 dmu_object_info_from_db(dbuf, &doi);
173 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_DSL_DIR);
174 ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
179 dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
180 dd->dd_object = ddobj;
184 if (dsl_dir_is_zapified(dd) &&
185 zap_contains(dp->dp_meta_objset, ddobj,
186 DD_FIELD_CRYPTO_KEY_OBJ) == 0) {
187 VERIFY0(zap_lookup(dp->dp_meta_objset,
188 ddobj, DD_FIELD_CRYPTO_KEY_OBJ,
189 sizeof (uint64_t), 1, &dd->dd_crypto_obj));
192 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
195 dsl_dir_snap_cmtime_update(dd);
197 if (dsl_dir_phys(dd)->dd_parent_obj) {
198 err = dsl_dir_hold_obj(dp,
199 dsl_dir_phys(dd)->dd_parent_obj, NULL, dd,
207 err = zap_lookup(dp->dp_meta_objset,
208 dsl_dir_phys(dd->dd_parent)->
209 dd_child_dir_zapobj, tail,
210 sizeof (foundobj), 1, &foundobj);
211 ASSERT(err || foundobj == ddobj);
213 (void) strlcpy(dd->dd_myname, tail,
214 sizeof (dd->dd_myname));
216 err = zap_value_search(dp->dp_meta_objset,
217 dsl_dir_phys(dd->dd_parent)->
219 ddobj, 0, dd->dd_myname);
224 (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
227 if (dsl_dir_is_clone(dd)) {
228 dmu_buf_t *origin_bonus;
229 dsl_dataset_phys_t *origin_phys;
232 * We can't open the origin dataset, because
233 * that would require opening this dsl_dir.
234 * Just look at its phys directly instead.
236 err = dmu_bonus_hold(dp->dp_meta_objset,
237 dsl_dir_phys(dd)->dd_origin_obj, FTAG,
241 origin_phys = origin_bonus->db_data;
243 origin_phys->ds_creation_txg;
244 dmu_buf_rele(origin_bonus, FTAG);
247 dmu_buf_init_user(&dd->dd_dbu, NULL, dsl_dir_evict_async,
249 winner = dmu_buf_set_user_ie(dbuf, &dd->dd_dbu);
250 if (winner != NULL) {
252 dsl_dir_rele(dd->dd_parent, dd);
254 mutex_destroy(&dd->dd_lock);
255 kmem_free(dd, sizeof (dsl_dir_t));
258 spa_open_ref(dp->dp_spa, dd);
263 * The dsl_dir_t has both open-to-close and instantiate-to-evict
264 * holds on the spa. We need the open-to-close holds because
265 * otherwise the spa_refcnt wouldn't change when we open a
266 * dir which the spa also has open, so we could incorrectly
267 * think it was OK to unload/export/destroy the pool. We need
268 * the instantiate-to-evict hold because the dsl_dir_t has a
269 * pointer to the dd_pool, which has a pointer to the spa_t.
271 spa_open_ref(dp->dp_spa, tag);
272 ASSERT3P(dd->dd_pool, ==, dp);
273 ASSERT3U(dd->dd_object, ==, ddobj);
274 ASSERT3P(dd->dd_dbuf, ==, dbuf);
280 dsl_dir_rele(dd->dd_parent, dd);
282 mutex_destroy(&dd->dd_lock);
283 kmem_free(dd, sizeof (dsl_dir_t));
284 dmu_buf_rele(dbuf, tag);
289 dsl_dir_rele(dsl_dir_t *dd, void *tag)
291 dprintf_dd(dd, "%s\n", "");
292 spa_close(dd->dd_pool->dp_spa, tag);
293 dmu_buf_rele(dd->dd_dbuf, tag);
297 * Remove a reference to the given dsl dir that is being asynchronously
298 * released. Async releases occur from a taskq performing eviction of
299 * dsl datasets and dirs. This process is identical to a normal release
300 * with the exception of using the async API for releasing the reference on
304 dsl_dir_async_rele(dsl_dir_t *dd, void *tag)
306 dprintf_dd(dd, "%s\n", "");
307 spa_async_close(dd->dd_pool->dp_spa, tag);
308 dmu_buf_rele(dd->dd_dbuf, tag);
311 /* buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes */
313 dsl_dir_name(dsl_dir_t *dd, char *buf)
316 dsl_dir_name(dd->dd_parent, buf);
317 VERIFY3U(strlcat(buf, "/", ZFS_MAX_DATASET_NAME_LEN), <,
318 ZFS_MAX_DATASET_NAME_LEN);
322 if (!MUTEX_HELD(&dd->dd_lock)) {
324 * recursive mutex so that we can use
325 * dprintf_dd() with dd_lock held
327 mutex_enter(&dd->dd_lock);
328 VERIFY3U(strlcat(buf, dd->dd_myname, ZFS_MAX_DATASET_NAME_LEN),
329 <, ZFS_MAX_DATASET_NAME_LEN);
330 mutex_exit(&dd->dd_lock);
332 VERIFY3U(strlcat(buf, dd->dd_myname, ZFS_MAX_DATASET_NAME_LEN),
333 <, ZFS_MAX_DATASET_NAME_LEN);
337 /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
339 dsl_dir_namelen(dsl_dir_t *dd)
344 /* parent's name + 1 for the "/" */
345 result = dsl_dir_namelen(dd->dd_parent) + 1;
348 if (!MUTEX_HELD(&dd->dd_lock)) {
349 /* see dsl_dir_name */
350 mutex_enter(&dd->dd_lock);
351 result += strlen(dd->dd_myname);
352 mutex_exit(&dd->dd_lock);
354 result += strlen(dd->dd_myname);
361 getcomponent(const char *path, char *component, const char **nextp)
365 if ((path == NULL) || (path[0] == '\0'))
366 return (SET_ERROR(ENOENT));
367 /* This would be a good place to reserve some namespace... */
368 p = strpbrk(path, "/@");
369 if (p && (p[1] == '/' || p[1] == '@')) {
370 /* two separators in a row */
371 return (SET_ERROR(EINVAL));
373 if (p == NULL || p == path) {
375 * if the first thing is an @ or /, it had better be an
376 * @ and it had better not have any more ats or slashes,
377 * and it had better have something after the @.
380 (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
381 return (SET_ERROR(EINVAL));
382 if (strlen(path) >= ZFS_MAX_DATASET_NAME_LEN)
383 return (SET_ERROR(ENAMETOOLONG));
384 (void) strcpy(component, path);
386 } else if (p[0] == '/') {
387 if (p - path >= ZFS_MAX_DATASET_NAME_LEN)
388 return (SET_ERROR(ENAMETOOLONG));
389 (void) strncpy(component, path, p - path);
390 component[p - path] = '\0';
392 } else if (p[0] == '@') {
394 * if the next separator is an @, there better not be
397 if (strchr(path, '/'))
398 return (SET_ERROR(EINVAL));
399 if (p - path >= ZFS_MAX_DATASET_NAME_LEN)
400 return (SET_ERROR(ENAMETOOLONG));
401 (void) strncpy(component, path, p - path);
402 component[p - path] = '\0';
404 panic("invalid p=%p", (void *)p);
411 * Return the dsl_dir_t, and possibly the last component which couldn't
412 * be found in *tail. The name must be in the specified dsl_pool_t. This
413 * thread must hold the dp_config_rwlock for the pool. Returns NULL if the
414 * path is bogus, or if tail==NULL and we couldn't parse the whole name.
415 * (*tail)[0] == '@' means that the last component is a snapshot.
418 dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag,
419 dsl_dir_t **ddp, const char **tailp)
422 const char *spaname, *next, *nextnext = NULL;
427 buf = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
428 err = getcomponent(name, buf, &next);
432 /* Make sure the name is in the specified pool. */
433 spaname = spa_name(dp->dp_spa);
434 if (strcmp(buf, spaname) != 0) {
435 err = SET_ERROR(EXDEV);
439 ASSERT(dsl_pool_config_held(dp));
441 err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
446 while (next != NULL) {
448 err = getcomponent(next, buf, &nextnext);
451 ASSERT(next[0] != '\0');
454 dprintf("looking up %s in obj%lld\n",
455 buf, dsl_dir_phys(dd)->dd_child_dir_zapobj);
457 err = zap_lookup(dp->dp_meta_objset,
458 dsl_dir_phys(dd)->dd_child_dir_zapobj,
459 buf, sizeof (ddobj), 1, &ddobj);
466 err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_dd);
469 dsl_dir_rele(dd, tag);
475 dsl_dir_rele(dd, tag);
480 * It's an error if there's more than one component left, or
481 * tailp==NULL and there's any component left.
484 (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
486 dsl_dir_rele(dd, tag);
487 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
488 err = SET_ERROR(ENOENT);
494 kmem_free(buf, ZFS_MAX_DATASET_NAME_LEN);
499 * If the counts are already initialized for this filesystem and its
500 * descendants then do nothing, otherwise initialize the counts.
502 * The counts on this filesystem, and those below, may be uninitialized due to
503 * either the use of a pre-existing pool which did not support the
504 * filesystem/snapshot limit feature, or one in which the feature had not yet
507 * Recursively descend the filesystem tree and update the filesystem/snapshot
508 * counts on each filesystem below, then update the cumulative count on the
509 * current filesystem. If the filesystem already has a count set on it,
510 * then we know that its counts, and the counts on the filesystems below it,
511 * are already correct, so we don't have to update this filesystem.
514 dsl_dir_init_fs_ss_count(dsl_dir_t *dd, dmu_tx_t *tx)
516 uint64_t my_fs_cnt = 0;
517 uint64_t my_ss_cnt = 0;
518 dsl_pool_t *dp = dd->dd_pool;
519 objset_t *os = dp->dp_meta_objset;
524 ASSERT(spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT));
525 ASSERT(dsl_pool_config_held(dp));
526 ASSERT(dmu_tx_is_syncing(tx));
528 dsl_dir_zapify(dd, tx);
531 * If the filesystem count has already been initialized then we
532 * don't need to recurse down any further.
534 if (zap_contains(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT) == 0)
537 zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP);
538 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
540 /* Iterate my child dirs */
541 for (zap_cursor_init(zc, os, dsl_dir_phys(dd)->dd_child_dir_zapobj);
542 zap_cursor_retrieve(zc, za) == 0; zap_cursor_advance(zc)) {
546 VERIFY0(dsl_dir_hold_obj(dp, za->za_first_integer, NULL, FTAG,
550 * Ignore hidden ($FREE, $MOS & $ORIGIN) objsets and
551 * temporary datasets.
553 if (chld_dd->dd_myname[0] == '$' ||
554 chld_dd->dd_myname[0] == '%') {
555 dsl_dir_rele(chld_dd, FTAG);
559 my_fs_cnt++; /* count this child */
561 dsl_dir_init_fs_ss_count(chld_dd, tx);
563 VERIFY0(zap_lookup(os, chld_dd->dd_object,
564 DD_FIELD_FILESYSTEM_COUNT, sizeof (count), 1, &count));
566 VERIFY0(zap_lookup(os, chld_dd->dd_object,
567 DD_FIELD_SNAPSHOT_COUNT, sizeof (count), 1, &count));
570 dsl_dir_rele(chld_dd, FTAG);
573 /* Count my snapshots (we counted children's snapshots above) */
574 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
575 dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds));
577 for (zap_cursor_init(zc, os, dsl_dataset_phys(ds)->ds_snapnames_zapobj);
578 zap_cursor_retrieve(zc, za) == 0;
579 zap_cursor_advance(zc)) {
580 /* Don't count temporary snapshots */
581 if (za->za_name[0] != '%')
586 dsl_dataset_rele(ds, FTAG);
588 kmem_free(zc, sizeof (zap_cursor_t));
589 kmem_free(za, sizeof (zap_attribute_t));
591 /* we're in a sync task, update counts */
592 dmu_buf_will_dirty(dd->dd_dbuf, tx);
593 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
594 sizeof (my_fs_cnt), 1, &my_fs_cnt, tx));
595 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
596 sizeof (my_ss_cnt), 1, &my_ss_cnt, tx));
600 dsl_dir_actv_fs_ss_limit_check(void *arg, dmu_tx_t *tx)
602 char *ddname = (char *)arg;
603 dsl_pool_t *dp = dmu_tx_pool(tx);
608 error = dsl_dataset_hold(dp, ddname, FTAG, &ds);
612 if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
613 dsl_dataset_rele(ds, FTAG);
614 return (SET_ERROR(ENOTSUP));
618 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT) &&
619 dsl_dir_is_zapified(dd) &&
620 zap_contains(dp->dp_meta_objset, dd->dd_object,
621 DD_FIELD_FILESYSTEM_COUNT) == 0) {
622 dsl_dataset_rele(ds, FTAG);
623 return (SET_ERROR(EALREADY));
626 dsl_dataset_rele(ds, FTAG);
631 dsl_dir_actv_fs_ss_limit_sync(void *arg, dmu_tx_t *tx)
633 char *ddname = (char *)arg;
634 dsl_pool_t *dp = dmu_tx_pool(tx);
638 VERIFY0(dsl_dataset_hold(dp, ddname, FTAG, &ds));
640 spa = dsl_dataset_get_spa(ds);
642 if (!spa_feature_is_active(spa, SPA_FEATURE_FS_SS_LIMIT)) {
644 * Since the feature was not active and we're now setting a
645 * limit, increment the feature-active counter so that the
646 * feature becomes active for the first time.
648 * We are already in a sync task so we can update the MOS.
650 spa_feature_incr(spa, SPA_FEATURE_FS_SS_LIMIT, tx);
654 * Since we are now setting a non-UINT64_MAX limit on the filesystem,
655 * we need to ensure the counts are correct. Descend down the tree from
656 * this point and update all of the counts to be accurate.
658 dsl_dir_init_fs_ss_count(ds->ds_dir, tx);
660 dsl_dataset_rele(ds, FTAG);
664 * Make sure the feature is enabled and activate it if necessary.
665 * Since we're setting a limit, ensure the on-disk counts are valid.
666 * This is only called by the ioctl path when setting a limit value.
668 * We do not need to validate the new limit, since users who can change the
669 * limit are also allowed to exceed the limit.
672 dsl_dir_activate_fs_ss_limit(const char *ddname)
676 error = dsl_sync_task(ddname, dsl_dir_actv_fs_ss_limit_check,
677 dsl_dir_actv_fs_ss_limit_sync, (void *)ddname, 0,
678 ZFS_SPACE_CHECK_RESERVED);
680 if (error == EALREADY)
687 * Used to determine if the filesystem_limit or snapshot_limit should be
688 * enforced. We allow the limit to be exceeded if the user has permission to
689 * write the property value. We pass in the creds that we got in the open
690 * context since we will always be the GZ root in syncing context. We also have
691 * to handle the case where we are allowed to change the limit on the current
692 * dataset, but there may be another limit in the tree above.
694 * We can never modify these two properties within a non-global zone. In
695 * addition, the other checks are modeled on zfs_secpolicy_write_perms. We
696 * can't use that function since we are already holding the dp_config_rwlock.
697 * In addition, we already have the dd and dealing with snapshots is simplified
708 dsl_enforce_ds_ss_limits(dsl_dir_t *dd, zfs_prop_t prop, cred_t *cr)
710 enforce_res_t enforce = ENFORCE_ALWAYS;
715 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
716 prop == ZFS_PROP_SNAPSHOT_LIMIT);
719 if (crgetzoneid(cr) != GLOBAL_ZONEID)
720 return (ENFORCE_ALWAYS);
722 if (secpolicy_zfs(cr) == 0)
723 return (ENFORCE_NEVER);
726 if ((obj = dsl_dir_phys(dd)->dd_head_dataset_obj) == 0)
727 return (ENFORCE_ALWAYS);
729 ASSERT(dsl_pool_config_held(dd->dd_pool));
731 if (dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds) != 0)
732 return (ENFORCE_ALWAYS);
734 if (dsl_prop_get_ds(ds, "zoned", 8, 1, &zoned, NULL) || zoned) {
735 /* Only root can access zoned fs's from the GZ */
736 enforce = ENFORCE_ALWAYS;
738 if (dsl_deleg_access_impl(ds, zfs_prop_to_name(prop), cr) == 0)
739 enforce = ENFORCE_ABOVE;
742 dsl_dataset_rele(ds, FTAG);
747 * Check if adding additional child filesystem(s) would exceed any filesystem
748 * limits or adding additional snapshot(s) would exceed any snapshot limits.
749 * The prop argument indicates which limit to check.
751 * Note that all filesystem limits up to the root (or the highest
752 * initialized) filesystem or the given ancestor must be satisfied.
755 dsl_fs_ss_limit_check(dsl_dir_t *dd, uint64_t delta, zfs_prop_t prop,
756 dsl_dir_t *ancestor, cred_t *cr)
758 objset_t *os = dd->dd_pool->dp_meta_objset;
759 uint64_t limit, count;
761 enforce_res_t enforce;
764 ASSERT(dsl_pool_config_held(dd->dd_pool));
765 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
766 prop == ZFS_PROP_SNAPSHOT_LIMIT);
769 * If we're allowed to change the limit, don't enforce the limit
770 * e.g. this can happen if a snapshot is taken by an administrative
771 * user in the global zone (i.e. a recursive snapshot by root).
772 * However, we must handle the case of delegated permissions where we
773 * are allowed to change the limit on the current dataset, but there
774 * is another limit in the tree above.
776 enforce = dsl_enforce_ds_ss_limits(dd, prop, cr);
777 if (enforce == ENFORCE_NEVER)
781 * e.g. if renaming a dataset with no snapshots, count adjustment
787 if (prop == ZFS_PROP_SNAPSHOT_LIMIT) {
789 * We don't enforce the limit for temporary snapshots. This is
790 * indicated by a NULL cred_t argument.
795 count_prop = DD_FIELD_SNAPSHOT_COUNT;
797 count_prop = DD_FIELD_FILESYSTEM_COUNT;
801 * If an ancestor has been provided, stop checking the limit once we
802 * hit that dir. We need this during rename so that we don't overcount
803 * the check once we recurse up to the common ancestor.
809 * If we hit an uninitialized node while recursing up the tree, we can
810 * stop since we know there is no limit here (or above). The counts are
811 * not valid on this node and we know we won't touch this node's counts.
813 if (!dsl_dir_is_zapified(dd) || zap_lookup(os, dd->dd_object,
814 count_prop, sizeof (count), 1, &count) == ENOENT)
817 err = dsl_prop_get_dd(dd, zfs_prop_to_name(prop), 8, 1, &limit, NULL,
822 /* Is there a limit which we've hit? */
823 if (enforce == ENFORCE_ALWAYS && (count + delta) > limit)
824 return (SET_ERROR(EDQUOT));
826 if (dd->dd_parent != NULL)
827 err = dsl_fs_ss_limit_check(dd->dd_parent, delta, prop,
834 * Adjust the filesystem or snapshot count for the specified dsl_dir_t and all
835 * parents. When a new filesystem/snapshot is created, increment the count on
836 * all parents, and when a filesystem/snapshot is destroyed, decrement the
840 dsl_fs_ss_count_adjust(dsl_dir_t *dd, int64_t delta, const char *prop,
844 objset_t *os = dd->dd_pool->dp_meta_objset;
847 ASSERT(dsl_pool_config_held(dd->dd_pool));
848 ASSERT(dmu_tx_is_syncing(tx));
849 ASSERT(strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0 ||
850 strcmp(prop, DD_FIELD_SNAPSHOT_COUNT) == 0);
853 * When we receive an incremental stream into a filesystem that already
854 * exists, a temporary clone is created. We don't count this temporary
855 * clone, whose name begins with a '%'. We also ignore hidden ($FREE,
856 * $MOS & $ORIGIN) objsets.
858 if ((dd->dd_myname[0] == '%' || dd->dd_myname[0] == '$') &&
859 strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0)
863 * e.g. if renaming a dataset with no snapshots, count adjustment is 0
869 * If we hit an uninitialized node while recursing up the tree, we can
870 * stop since we know the counts are not valid on this node and we
871 * know we shouldn't touch this node's counts. An uninitialized count
872 * on the node indicates that either the feature has not yet been
873 * activated or there are no limits on this part of the tree.
875 if (!dsl_dir_is_zapified(dd) || (err = zap_lookup(os, dd->dd_object,
876 prop, sizeof (count), 1, &count)) == ENOENT)
881 /* Use a signed verify to make sure we're not neg. */
882 VERIFY3S(count, >=, 0);
884 VERIFY0(zap_update(os, dd->dd_object, prop, sizeof (count), 1, &count,
887 /* Roll up this additional count into our ancestors */
888 if (dd->dd_parent != NULL)
889 dsl_fs_ss_count_adjust(dd->dd_parent, delta, prop, tx);
893 dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
896 objset_t *mos = dp->dp_meta_objset;
898 dsl_dir_phys_t *ddphys;
901 ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
902 DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
904 VERIFY(0 == zap_add(mos, dsl_dir_phys(pds)->dd_child_dir_zapobj,
905 name, sizeof (uint64_t), 1, &ddobj, tx));
907 /* it's the root dir */
908 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
909 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
911 VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
912 dmu_buf_will_dirty(dbuf, tx);
913 ddphys = dbuf->db_data;
915 ddphys->dd_creation_time = gethrestime_sec();
917 ddphys->dd_parent_obj = pds->dd_object;
919 /* update the filesystem counts */
920 dsl_fs_ss_count_adjust(pds, 1, DD_FIELD_FILESYSTEM_COUNT, tx);
922 ddphys->dd_props_zapobj = zap_create(mos,
923 DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
924 ddphys->dd_child_dir_zapobj = zap_create(mos,
925 DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
926 if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
927 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
929 dmu_buf_rele(dbuf, FTAG);
935 dsl_dir_is_clone(dsl_dir_t *dd)
937 return (dsl_dir_phys(dd)->dd_origin_obj &&
938 (dd->dd_pool->dp_origin_snap == NULL ||
939 dsl_dir_phys(dd)->dd_origin_obj !=
940 dd->dd_pool->dp_origin_snap->ds_object));
944 dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
948 mutex_enter(&dd->dd_lock);
949 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
950 dsl_dir_phys(dd)->dd_used_bytes);
951 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA,
952 dsl_dir_phys(dd)->dd_quota);
953 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
954 dsl_dir_phys(dd)->dd_reserved);
955 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
956 dsl_dir_phys(dd)->dd_compressed_bytes == 0 ? 100 :
957 (dsl_dir_phys(dd)->dd_uncompressed_bytes * 100 /
958 dsl_dir_phys(dd)->dd_compressed_bytes));
959 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED,
960 dsl_dir_phys(dd)->dd_uncompressed_bytes);
961 if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
962 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
963 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]);
964 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
965 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_HEAD]);
966 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
967 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_REFRSRV]);
968 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
969 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD] +
970 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD_RSRV]);
972 mutex_exit(&dd->dd_lock);
974 if (dsl_dir_is_zapified(dd)) {
975 objset_t *os = dd->dd_pool->dp_meta_objset;
977 if (zap_lookup(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
978 sizeof (intval), 1, &intval) == 0) {
979 dsl_prop_nvlist_add_uint64(nv,
980 ZFS_PROP_FILESYSTEM_COUNT, intval);
982 if (zap_lookup(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
983 sizeof (intval), 1, &intval) == 0) {
984 dsl_prop_nvlist_add_uint64(nv,
985 ZFS_PROP_SNAPSHOT_COUNT, intval);
989 if (dsl_dir_is_clone(dd)) {
991 char buf[ZFS_MAX_DATASET_NAME_LEN];
993 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
994 dsl_dir_phys(dd)->dd_origin_obj, FTAG, &ds));
995 dsl_dataset_name(ds, buf);
996 dsl_dataset_rele(ds, FTAG);
997 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
1002 dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
1004 dsl_pool_t *dp = dd->dd_pool;
1006 ASSERT(dsl_dir_phys(dd));
1008 if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) {
1009 /* up the hold count until we can be written out */
1010 dmu_buf_add_ref(dd->dd_dbuf, dd);
1015 parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
1017 uint64_t old_accounted = MAX(used, dsl_dir_phys(dd)->dd_reserved);
1018 uint64_t new_accounted =
1019 MAX(used + delta, dsl_dir_phys(dd)->dd_reserved);
1020 return (new_accounted - old_accounted);
1024 dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
1026 ASSERT(dmu_tx_is_syncing(tx));
1028 mutex_enter(&dd->dd_lock);
1029 ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]);
1030 dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
1031 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
1032 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
1033 mutex_exit(&dd->dd_lock);
1035 /* release the hold from dsl_dir_dirty */
1036 dmu_buf_rele(dd->dd_dbuf, dd);
1040 dsl_dir_space_towrite(dsl_dir_t *dd)
1044 ASSERT(MUTEX_HELD(&dd->dd_lock));
1046 for (int i = 0; i < TXG_SIZE; i++) {
1047 space += dd->dd_space_towrite[i & TXG_MASK];
1048 ASSERT3U(dd->dd_space_towrite[i & TXG_MASK], >=, 0);
1054 * How much space would dd have available if ancestor had delta applied
1055 * to it? If ondiskonly is set, we're only interested in what's
1056 * on-disk, not estimated pending changes.
1059 dsl_dir_space_available(dsl_dir_t *dd,
1060 dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
1062 uint64_t parentspace, myspace, quota, used;
1065 * If there are no restrictions otherwise, assume we have
1066 * unlimited space available.
1069 parentspace = UINT64_MAX;
1071 if (dd->dd_parent != NULL) {
1072 parentspace = dsl_dir_space_available(dd->dd_parent,
1073 ancestor, delta, ondiskonly);
1076 mutex_enter(&dd->dd_lock);
1077 if (dsl_dir_phys(dd)->dd_quota != 0)
1078 quota = dsl_dir_phys(dd)->dd_quota;
1079 used = dsl_dir_phys(dd)->dd_used_bytes;
1081 used += dsl_dir_space_towrite(dd);
1083 if (dd->dd_parent == NULL) {
1084 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
1085 quota = MIN(quota, poolsize);
1088 if (dsl_dir_phys(dd)->dd_reserved > used && parentspace != UINT64_MAX) {
1090 * We have some space reserved, in addition to what our
1093 parentspace += dsl_dir_phys(dd)->dd_reserved - used;
1096 if (dd == ancestor) {
1098 ASSERT(used >= -delta);
1100 if (parentspace != UINT64_MAX)
1101 parentspace -= delta;
1109 * the lesser of the space provided by our parent and
1110 * the space left in our quota
1112 myspace = MIN(parentspace, quota - used);
1115 mutex_exit(&dd->dd_lock);
1120 struct tempreserve {
1121 list_node_t tr_node;
1127 dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
1128 boolean_t ignorequota, list_t *tr_list,
1129 dmu_tx_t *tx, boolean_t first)
1133 struct tempreserve *tr;
1142 ASSERT3U(txg, !=, 0);
1143 ASSERT3S(asize, >, 0);
1145 mutex_enter(&dd->dd_lock);
1148 * Check against the dsl_dir's quota. We don't add in the delta
1149 * when checking for over-quota because they get one free hit.
1151 uint64_t est_inflight = dsl_dir_space_towrite(dd);
1152 for (int i = 0; i < TXG_SIZE; i++)
1153 est_inflight += dd->dd_tempreserved[i];
1154 uint64_t used_on_disk = dsl_dir_phys(dd)->dd_used_bytes;
1157 * On the first iteration, fetch the dataset's used-on-disk and
1158 * refreservation values. Also, if checkrefquota is set, test if
1159 * allocating this space would exceed the dataset's refquota.
1161 if (first && tx->tx_objset) {
1163 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
1165 error = dsl_dataset_check_quota(ds, !netfree,
1166 asize, est_inflight, &used_on_disk, &ref_rsrv);
1168 mutex_exit(&dd->dd_lock);
1169 DMU_TX_STAT_BUMP(dmu_tx_quota);
1175 * If this transaction will result in a net free of space,
1176 * we want to let it through.
1178 if (ignorequota || netfree || dsl_dir_phys(dd)->dd_quota == 0)
1181 quota = dsl_dir_phys(dd)->dd_quota;
1184 * Adjust the quota against the actual pool size at the root
1185 * minus any outstanding deferred frees.
1186 * To ensure that it's possible to remove files from a full
1187 * pool without inducing transient overcommits, we throttle
1188 * netfree transactions against a quota that is slightly larger,
1189 * but still within the pool's allocation slop. In cases where
1190 * we're very close to full, this will allow a steady trickle of
1191 * removes to get through.
1193 uint64_t deferred = 0;
1194 if (dd->dd_parent == NULL) {
1195 spa_t *spa = dd->dd_pool->dp_spa;
1196 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
1197 deferred = metaslab_class_get_deferred(spa_normal_class(spa));
1198 if (poolsize - deferred < quota) {
1199 quota = poolsize - deferred;
1205 * If they are requesting more space, and our current estimate
1206 * is over quota, they get to try again unless the actual
1207 * on-disk is over quota and there are no pending changes (which
1208 * may free up space for us).
1210 if (used_on_disk + est_inflight >= quota) {
1211 if (est_inflight > 0 || used_on_disk < quota ||
1212 (retval == ENOSPC && used_on_disk < quota + deferred))
1214 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
1215 "quota=%lluK tr=%lluK err=%d\n",
1216 used_on_disk>>10, est_inflight>>10,
1217 quota>>10, asize>>10, retval);
1218 mutex_exit(&dd->dd_lock);
1219 DMU_TX_STAT_BUMP(dmu_tx_quota);
1220 return (SET_ERROR(retval));
1223 /* We need to up our estimated delta before dropping dd_lock */
1224 dd->dd_tempreserved[txg & TXG_MASK] += asize;
1226 uint64_t parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
1228 mutex_exit(&dd->dd_lock);
1230 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1232 tr->tr_size = asize;
1233 list_insert_tail(tr_list, tr);
1235 /* see if it's OK with our parent */
1236 if (dd->dd_parent != NULL && parent_rsrv != 0) {
1238 * Recurse on our parent without recursion. This has been
1239 * observed to be potentially large stack usage even within
1240 * the test suite. Largest seen stack was 7632 bytes on linux.
1244 asize = parent_rsrv;
1245 ignorequota = (dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
1247 goto top_of_function;
1255 * Reserve space in this dsl_dir, to be used in this tx's txg.
1256 * After the space has been dirtied (and dsl_dir_willuse_space()
1257 * has been called), the reservation should be canceled, using
1258 * dsl_dir_tempreserve_clear().
1261 dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
1262 boolean_t netfree, void **tr_cookiep, dmu_tx_t *tx)
1272 tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
1273 list_create(tr_list, sizeof (struct tempreserve),
1274 offsetof(struct tempreserve, tr_node));
1275 ASSERT3S(asize, >, 0);
1277 err = arc_tempreserve_space(lsize, tx->tx_txg);
1279 struct tempreserve *tr;
1281 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1282 tr->tr_size = lsize;
1283 list_insert_tail(tr_list, tr);
1285 if (err == EAGAIN) {
1287 * If arc_memory_throttle() detected that pageout
1288 * is running and we are low on memory, we delay new
1289 * non-pageout transactions to give pageout an
1292 * It is unfortunate to be delaying while the caller's
1295 txg_delay(dd->dd_pool, tx->tx_txg,
1296 MSEC2NSEC(10), MSEC2NSEC(10));
1297 err = SET_ERROR(ERESTART);
1302 err = dsl_dir_tempreserve_impl(dd, asize, netfree,
1303 B_FALSE, tr_list, tx, B_TRUE);
1307 dsl_dir_tempreserve_clear(tr_list, tx);
1309 *tr_cookiep = tr_list;
1315 * Clear a temporary reservation that we previously made with
1316 * dsl_dir_tempreserve_space().
1319 dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
1321 int txgidx = tx->tx_txg & TXG_MASK;
1322 list_t *tr_list = tr_cookie;
1323 struct tempreserve *tr;
1325 ASSERT3U(tx->tx_txg, !=, 0);
1327 if (tr_cookie == NULL)
1330 while ((tr = list_head(tr_list)) != NULL) {
1332 mutex_enter(&tr->tr_ds->dd_lock);
1333 ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
1335 tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
1336 mutex_exit(&tr->tr_ds->dd_lock);
1338 arc_tempreserve_clear(tr->tr_size);
1340 list_remove(tr_list, tr);
1341 kmem_free(tr, sizeof (struct tempreserve));
1344 kmem_free(tr_list, sizeof (list_t));
1348 * This should be called from open context when we think we're going to write
1349 * or free space, for example when dirtying data. Be conservative; it's okay
1350 * to write less space or free more, but we don't want to write more or free
1351 * less than the amount specified.
1353 * NOTE: The behavior of this function is identical to the Illumos / FreeBSD
1354 * version however it has been adjusted to use an iterative rather then
1355 * recursive algorithm to minimize stack usage.
1358 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
1360 int64_t parent_space;
1364 mutex_enter(&dd->dd_lock);
1366 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
1368 est_used = dsl_dir_space_towrite(dd) +
1369 dsl_dir_phys(dd)->dd_used_bytes;
1370 parent_space = parent_delta(dd, est_used, space);
1371 mutex_exit(&dd->dd_lock);
1373 /* Make sure that we clean up dd_space_to* */
1374 dsl_dir_dirty(dd, tx);
1377 space = parent_space;
1378 } while (space && dd);
1381 /* call from syncing context when we actually write/free space for this dd */
1383 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
1384 int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
1386 int64_t accounted_delta;
1389 * dsl_dataset_set_refreservation_sync_impl() calls this with
1390 * dd_lock held, so that it can atomically update
1391 * ds->ds_reserved and the dsl_dir accounting, so that
1392 * dsl_dataset_check_quota() can see dataset and dir accounting
1395 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
1397 ASSERT(dmu_tx_is_syncing(tx));
1398 ASSERT(type < DD_USED_NUM);
1400 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1403 mutex_enter(&dd->dd_lock);
1405 parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, used);
1406 ASSERT(used >= 0 || dsl_dir_phys(dd)->dd_used_bytes >= -used);
1407 ASSERT(compressed >= 0 ||
1408 dsl_dir_phys(dd)->dd_compressed_bytes >= -compressed);
1409 ASSERT(uncompressed >= 0 ||
1410 dsl_dir_phys(dd)->dd_uncompressed_bytes >= -uncompressed);
1411 dsl_dir_phys(dd)->dd_used_bytes += used;
1412 dsl_dir_phys(dd)->dd_uncompressed_bytes += uncompressed;
1413 dsl_dir_phys(dd)->dd_compressed_bytes += compressed;
1415 if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1417 dsl_dir_phys(dd)->dd_used_breakdown[type] >= -used);
1418 dsl_dir_phys(dd)->dd_used_breakdown[type] += used;
1423 for (t = 0; t < DD_USED_NUM; t++)
1424 u += dsl_dir_phys(dd)->dd_used_breakdown[t];
1425 ASSERT3U(u, ==, dsl_dir_phys(dd)->dd_used_bytes);
1430 mutex_exit(&dd->dd_lock);
1432 if (dd->dd_parent != NULL) {
1433 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1434 accounted_delta, compressed, uncompressed, tx);
1435 dsl_dir_transfer_space(dd->dd_parent,
1436 used - accounted_delta,
1437 DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
1442 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
1443 dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
1445 ASSERT(dmu_tx_is_syncing(tx));
1446 ASSERT(oldtype < DD_USED_NUM);
1447 ASSERT(newtype < DD_USED_NUM);
1450 !(dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN))
1453 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1454 mutex_enter(&dd->dd_lock);
1456 dsl_dir_phys(dd)->dd_used_breakdown[oldtype] >= delta :
1457 dsl_dir_phys(dd)->dd_used_breakdown[newtype] >= -delta);
1458 ASSERT(dsl_dir_phys(dd)->dd_used_bytes >= ABS(delta));
1459 dsl_dir_phys(dd)->dd_used_breakdown[oldtype] -= delta;
1460 dsl_dir_phys(dd)->dd_used_breakdown[newtype] += delta;
1461 mutex_exit(&dd->dd_lock);
1464 typedef struct dsl_dir_set_qr_arg {
1465 const char *ddsqra_name;
1466 zprop_source_t ddsqra_source;
1467 uint64_t ddsqra_value;
1468 } dsl_dir_set_qr_arg_t;
1471 dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx)
1473 dsl_dir_set_qr_arg_t *ddsqra = arg;
1474 dsl_pool_t *dp = dmu_tx_pool(tx);
1477 uint64_t towrite, newval;
1479 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1483 error = dsl_prop_predict(ds->ds_dir, "quota",
1484 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1486 dsl_dataset_rele(ds, FTAG);
1491 dsl_dataset_rele(ds, FTAG);
1495 mutex_enter(&ds->ds_dir->dd_lock);
1497 * If we are doing the preliminary check in open context, and
1498 * there are pending changes, then don't fail it, since the
1499 * pending changes could under-estimate the amount of space to be
1502 towrite = dsl_dir_space_towrite(ds->ds_dir);
1503 if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
1504 (newval < dsl_dir_phys(ds->ds_dir)->dd_reserved ||
1505 newval < dsl_dir_phys(ds->ds_dir)->dd_used_bytes + towrite)) {
1506 error = SET_ERROR(ENOSPC);
1508 mutex_exit(&ds->ds_dir->dd_lock);
1509 dsl_dataset_rele(ds, FTAG);
1514 dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx)
1516 dsl_dir_set_qr_arg_t *ddsqra = arg;
1517 dsl_pool_t *dp = dmu_tx_pool(tx);
1521 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1523 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1524 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA),
1525 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1526 &ddsqra->ddsqra_value, tx);
1528 VERIFY0(dsl_prop_get_int_ds(ds,
1529 zfs_prop_to_name(ZFS_PROP_QUOTA), &newval));
1531 newval = ddsqra->ddsqra_value;
1532 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1533 zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval);
1536 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1537 mutex_enter(&ds->ds_dir->dd_lock);
1538 dsl_dir_phys(ds->ds_dir)->dd_quota = newval;
1539 mutex_exit(&ds->ds_dir->dd_lock);
1540 dsl_dataset_rele(ds, FTAG);
1544 dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
1546 dsl_dir_set_qr_arg_t ddsqra;
1548 ddsqra.ddsqra_name = ddname;
1549 ddsqra.ddsqra_source = source;
1550 ddsqra.ddsqra_value = quota;
1552 return (dsl_sync_task(ddname, dsl_dir_set_quota_check,
1553 dsl_dir_set_quota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
1557 dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx)
1559 dsl_dir_set_qr_arg_t *ddsqra = arg;
1560 dsl_pool_t *dp = dmu_tx_pool(tx);
1563 uint64_t newval, used, avail;
1566 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1572 * If we are doing the preliminary check in open context, the
1573 * space estimates may be inaccurate.
1575 if (!dmu_tx_is_syncing(tx)) {
1576 dsl_dataset_rele(ds, FTAG);
1580 error = dsl_prop_predict(ds->ds_dir,
1581 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1582 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1584 dsl_dataset_rele(ds, FTAG);
1588 mutex_enter(&dd->dd_lock);
1589 used = dsl_dir_phys(dd)->dd_used_bytes;
1590 mutex_exit(&dd->dd_lock);
1592 if (dd->dd_parent) {
1593 avail = dsl_dir_space_available(dd->dd_parent,
1596 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
1599 if (MAX(used, newval) > MAX(used, dsl_dir_phys(dd)->dd_reserved)) {
1600 uint64_t delta = MAX(used, newval) -
1601 MAX(used, dsl_dir_phys(dd)->dd_reserved);
1603 if (delta > avail ||
1604 (dsl_dir_phys(dd)->dd_quota > 0 &&
1605 newval > dsl_dir_phys(dd)->dd_quota))
1606 error = SET_ERROR(ENOSPC);
1609 dsl_dataset_rele(ds, FTAG);
1614 dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
1619 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1621 mutex_enter(&dd->dd_lock);
1622 used = dsl_dir_phys(dd)->dd_used_bytes;
1623 delta = MAX(used, value) - MAX(used, dsl_dir_phys(dd)->dd_reserved);
1624 dsl_dir_phys(dd)->dd_reserved = value;
1626 if (dd->dd_parent != NULL) {
1627 /* Roll up this additional usage into our ancestors */
1628 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1631 mutex_exit(&dd->dd_lock);
1635 dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx)
1637 dsl_dir_set_qr_arg_t *ddsqra = arg;
1638 dsl_pool_t *dp = dmu_tx_pool(tx);
1642 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1644 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1645 dsl_prop_set_sync_impl(ds,
1646 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1647 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1648 &ddsqra->ddsqra_value, tx);
1650 VERIFY0(dsl_prop_get_int_ds(ds,
1651 zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval));
1653 newval = ddsqra->ddsqra_value;
1654 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1655 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1656 (longlong_t)newval);
1659 dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx);
1660 dsl_dataset_rele(ds, FTAG);
1664 dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1665 uint64_t reservation)
1667 dsl_dir_set_qr_arg_t ddsqra;
1669 ddsqra.ddsqra_name = ddname;
1670 ddsqra.ddsqra_source = source;
1671 ddsqra.ddsqra_value = reservation;
1673 return (dsl_sync_task(ddname, dsl_dir_set_reservation_check,
1674 dsl_dir_set_reservation_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
1678 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1680 for (; ds1; ds1 = ds1->dd_parent) {
1682 for (dd = ds2; dd; dd = dd->dd_parent) {
1691 * If delta is applied to dd, how much of that delta would be applied to
1692 * ancestor? Syncing context only.
1695 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1700 mutex_enter(&dd->dd_lock);
1701 delta = parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, delta);
1702 mutex_exit(&dd->dd_lock);
1703 return (would_change(dd->dd_parent, delta, ancestor));
1706 typedef struct dsl_dir_rename_arg {
1707 const char *ddra_oldname;
1708 const char *ddra_newname;
1710 } dsl_dir_rename_arg_t;
1714 dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1717 char namebuf[ZFS_MAX_DATASET_NAME_LEN];
1719 dsl_dataset_name(ds, namebuf);
1721 if (strlen(namebuf) + *deltap >= ZFS_MAX_DATASET_NAME_LEN)
1722 return (SET_ERROR(ENAMETOOLONG));
1727 dsl_dir_rename_check(void *arg, dmu_tx_t *tx)
1729 dsl_dir_rename_arg_t *ddra = arg;
1730 dsl_pool_t *dp = dmu_tx_pool(tx);
1731 dsl_dir_t *dd, *newparent;
1732 const char *mynewname;
1734 int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname);
1736 /* target dir should exist */
1737 error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL);
1741 /* new parent should exist */
1742 error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG,
1743 &newparent, &mynewname);
1745 dsl_dir_rele(dd, FTAG);
1749 /* can't rename to different pool */
1750 if (dd->dd_pool != newparent->dd_pool) {
1751 dsl_dir_rele(newparent, FTAG);
1752 dsl_dir_rele(dd, FTAG);
1753 return (SET_ERROR(EXDEV));
1756 /* new name should not already exist */
1757 if (mynewname == NULL) {
1758 dsl_dir_rele(newparent, FTAG);
1759 dsl_dir_rele(dd, FTAG);
1760 return (SET_ERROR(EEXIST));
1763 /* if the name length is growing, validate child name lengths */
1765 error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename,
1766 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1768 dsl_dir_rele(newparent, FTAG);
1769 dsl_dir_rele(dd, FTAG);
1774 if (dmu_tx_is_syncing(tx)) {
1775 if (spa_feature_is_active(dp->dp_spa,
1776 SPA_FEATURE_FS_SS_LIMIT)) {
1778 * Although this is the check function and we don't
1779 * normally make on-disk changes in check functions,
1780 * we need to do that here.
1782 * Ensure this portion of the tree's counts have been
1783 * initialized in case the new parent has limits set.
1785 dsl_dir_init_fs_ss_count(dd, tx);
1789 if (newparent != dd->dd_parent) {
1790 /* is there enough space? */
1792 MAX(dsl_dir_phys(dd)->dd_used_bytes,
1793 dsl_dir_phys(dd)->dd_reserved);
1794 objset_t *os = dd->dd_pool->dp_meta_objset;
1795 uint64_t fs_cnt = 0;
1796 uint64_t ss_cnt = 0;
1798 if (dsl_dir_is_zapified(dd)) {
1801 err = zap_lookup(os, dd->dd_object,
1802 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1804 if (err != ENOENT && err != 0) {
1805 dsl_dir_rele(newparent, FTAG);
1806 dsl_dir_rele(dd, FTAG);
1811 * have to add 1 for the filesystem itself that we're
1816 err = zap_lookup(os, dd->dd_object,
1817 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1819 if (err != ENOENT && err != 0) {
1820 dsl_dir_rele(newparent, FTAG);
1821 dsl_dir_rele(dd, FTAG);
1826 /* check for encryption errors */
1827 error = dsl_dir_rename_crypt_check(dd, newparent);
1829 dsl_dir_rele(newparent, FTAG);
1830 dsl_dir_rele(dd, FTAG);
1831 return (SET_ERROR(EACCES));
1834 /* no rename into our descendant */
1835 if (closest_common_ancestor(dd, newparent) == dd) {
1836 dsl_dir_rele(newparent, FTAG);
1837 dsl_dir_rele(dd, FTAG);
1838 return (SET_ERROR(EINVAL));
1841 error = dsl_dir_transfer_possible(dd->dd_parent,
1842 newparent, fs_cnt, ss_cnt, myspace, ddra->ddra_cred);
1844 dsl_dir_rele(newparent, FTAG);
1845 dsl_dir_rele(dd, FTAG);
1850 dsl_dir_rele(newparent, FTAG);
1851 dsl_dir_rele(dd, FTAG);
1856 dsl_dir_rename_sync(void *arg, dmu_tx_t *tx)
1858 dsl_dir_rename_arg_t *ddra = arg;
1859 dsl_pool_t *dp = dmu_tx_pool(tx);
1860 dsl_dir_t *dd, *newparent;
1861 const char *mynewname;
1863 objset_t *mos = dp->dp_meta_objset;
1865 VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL));
1866 VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
1869 /* Log this before we change the name. */
1870 spa_history_log_internal_dd(dd, "rename", tx,
1871 "-> %s", ddra->ddra_newname);
1873 if (newparent != dd->dd_parent) {
1874 objset_t *os = dd->dd_pool->dp_meta_objset;
1875 uint64_t fs_cnt = 0;
1876 uint64_t ss_cnt = 0;
1879 * We already made sure the dd counts were initialized in the
1882 if (spa_feature_is_active(dp->dp_spa,
1883 SPA_FEATURE_FS_SS_LIMIT)) {
1884 VERIFY0(zap_lookup(os, dd->dd_object,
1885 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1887 /* add 1 for the filesystem itself that we're moving */
1890 VERIFY0(zap_lookup(os, dd->dd_object,
1891 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1895 dsl_fs_ss_count_adjust(dd->dd_parent, -fs_cnt,
1896 DD_FIELD_FILESYSTEM_COUNT, tx);
1897 dsl_fs_ss_count_adjust(newparent, fs_cnt,
1898 DD_FIELD_FILESYSTEM_COUNT, tx);
1900 dsl_fs_ss_count_adjust(dd->dd_parent, -ss_cnt,
1901 DD_FIELD_SNAPSHOT_COUNT, tx);
1902 dsl_fs_ss_count_adjust(newparent, ss_cnt,
1903 DD_FIELD_SNAPSHOT_COUNT, tx);
1905 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1906 -dsl_dir_phys(dd)->dd_used_bytes,
1907 -dsl_dir_phys(dd)->dd_compressed_bytes,
1908 -dsl_dir_phys(dd)->dd_uncompressed_bytes, tx);
1909 dsl_dir_diduse_space(newparent, DD_USED_CHILD,
1910 dsl_dir_phys(dd)->dd_used_bytes,
1911 dsl_dir_phys(dd)->dd_compressed_bytes,
1912 dsl_dir_phys(dd)->dd_uncompressed_bytes, tx);
1914 if (dsl_dir_phys(dd)->dd_reserved >
1915 dsl_dir_phys(dd)->dd_used_bytes) {
1916 uint64_t unused_rsrv = dsl_dir_phys(dd)->dd_reserved -
1917 dsl_dir_phys(dd)->dd_used_bytes;
1919 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1920 -unused_rsrv, 0, 0, tx);
1921 dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV,
1922 unused_rsrv, 0, 0, tx);
1926 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1928 /* remove from old parent zapobj */
1929 error = zap_remove(mos,
1930 dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj,
1934 (void) strlcpy(dd->dd_myname, mynewname,
1935 sizeof (dd->dd_myname));
1936 dsl_dir_rele(dd->dd_parent, dd);
1937 dsl_dir_phys(dd)->dd_parent_obj = newparent->dd_object;
1938 VERIFY0(dsl_dir_hold_obj(dp,
1939 newparent->dd_object, NULL, dd, &dd->dd_parent));
1941 /* add to new parent zapobj */
1942 VERIFY0(zap_add(mos, dsl_dir_phys(newparent)->dd_child_dir_zapobj,
1943 dd->dd_myname, 8, 1, &dd->dd_object, tx));
1945 zvol_rename_minors(dp->dp_spa, ddra->ddra_oldname,
1946 ddra->ddra_newname, B_TRUE);
1948 dsl_prop_notify_all(dd);
1950 dsl_dir_rele(newparent, FTAG);
1951 dsl_dir_rele(dd, FTAG);
1955 dsl_dir_rename(const char *oldname, const char *newname)
1957 dsl_dir_rename_arg_t ddra;
1959 ddra.ddra_oldname = oldname;
1960 ddra.ddra_newname = newname;
1961 ddra.ddra_cred = CRED();
1963 return (dsl_sync_task(oldname,
1964 dsl_dir_rename_check, dsl_dir_rename_sync, &ddra,
1965 3, ZFS_SPACE_CHECK_RESERVED));
1969 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd,
1970 uint64_t fs_cnt, uint64_t ss_cnt, uint64_t space, cred_t *cr)
1972 dsl_dir_t *ancestor;
1977 ancestor = closest_common_ancestor(sdd, tdd);
1978 adelta = would_change(sdd, -space, ancestor);
1979 avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1981 return (SET_ERROR(ENOSPC));
1983 err = dsl_fs_ss_limit_check(tdd, fs_cnt, ZFS_PROP_FILESYSTEM_LIMIT,
1987 err = dsl_fs_ss_limit_check(tdd, ss_cnt, ZFS_PROP_SNAPSHOT_LIMIT,
1996 dsl_dir_snap_cmtime(dsl_dir_t *dd)
2000 mutex_enter(&dd->dd_lock);
2001 t = dd->dd_snap_cmtime;
2002 mutex_exit(&dd->dd_lock);
2008 dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
2013 mutex_enter(&dd->dd_lock);
2014 dd->dd_snap_cmtime = t;
2015 mutex_exit(&dd->dd_lock);
2019 dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx)
2021 objset_t *mos = dd->dd_pool->dp_meta_objset;
2022 dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx);
2026 dsl_dir_is_zapified(dsl_dir_t *dd)
2028 dmu_object_info_t doi;
2030 dmu_object_info_from_db(dd->dd_dbuf, &doi);
2031 return (doi.doi_type == DMU_OTN_ZAP_METADATA);
2034 #if defined(_KERNEL) && defined(HAVE_SPL)
2035 EXPORT_SYMBOL(dsl_dir_set_quota);
2036 EXPORT_SYMBOL(dsl_dir_set_reservation);