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) 2011 Pawel Jakub Dawidek <pawel@dawidek.net>.
24 * All rights reserved.
25 * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
26 * Copyright (c) 2014 Joyent, Inc. All rights reserved.
27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28 * Copyright 2015 Nexenta Systems, Inc. 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/metaslab.h>
45 #include <sys/sunddi.h>
48 #include <sys/zfs_vfsops.h>
50 #include <sys/zfeature.h>
51 #include <sys/policy.h>
52 #include <sys/zfs_znode.h>
53 #include "zfs_namecheck.h"
57 * Filesystem and Snapshot Limits
58 * ------------------------------
60 * These limits are used to restrict the number of filesystems and/or snapshots
61 * that can be created at a given level in the tree or below. A typical
62 * use-case is with a delegated dataset where the administrator wants to ensure
63 * that a user within the zone is not creating too many additional filesystems
64 * or snapshots, even though they're not exceeding their space quota.
66 * The filesystem and snapshot counts are stored as extensible properties. This
67 * capability is controlled by a feature flag and must be enabled to be used.
68 * Once enabled, the feature is not active until the first limit is set. At
69 * that point, future operations to create/destroy filesystems or snapshots
70 * will validate and update the counts.
72 * Because the count properties will not exist before the feature is active,
73 * the counts are updated when a limit is first set on an uninitialized
74 * dsl_dir node in the tree (The filesystem/snapshot count on a node includes
75 * all of the nested filesystems/snapshots. Thus, a new leaf node has a
76 * filesystem count of 0 and a snapshot count of 0. Non-existent filesystem and
77 * snapshot count properties on a node indicate uninitialized counts on that
78 * node.) When first setting a limit on an uninitialized node, the code starts
79 * at the filesystem with the new limit and descends into all sub-filesystems
80 * to add the count properties.
82 * In practice this is lightweight since a limit is typically set when the
83 * filesystem is created and thus has no children. Once valid, changing the
84 * limit value won't require a re-traversal since the counts are already valid.
85 * When recursively fixing the counts, if a node with a limit is encountered
86 * during the descent, the counts are known to be valid and there is no need to
87 * descend into that filesystem's children. The counts on filesystems above the
88 * one with the new limit will still be uninitialized, unless a limit is
89 * eventually set on one of those filesystems. The counts are always recursively
90 * updated when a limit is set on a dataset, unless there is already a limit.
91 * When a new limit value is set on a filesystem with an existing limit, it is
92 * possible for the new limit to be less than the current count at that level
93 * since a user who can change the limit is also allowed to exceed the limit.
95 * Once the feature is active, then whenever a filesystem or snapshot is
96 * created, the code recurses up the tree, validating the new count against the
97 * limit at each initialized level. In practice, most levels will not have a
98 * limit set. If there is a limit at any initialized level up the tree, the
99 * check must pass or the creation will fail. Likewise, when a filesystem or
100 * snapshot is destroyed, the counts are recursively adjusted all the way up
101 * the initizized nodes in the tree. Renaming a filesystem into different point
102 * in the tree will first validate, then update the counts on each branch up to
103 * the common ancestor. A receive will also validate the counts and then update
106 * An exception to the above behavior is that the limit is not enforced if the
107 * user has permission to modify the limit. This is primarily so that
108 * recursive snapshots in the global zone always work. We want to prevent a
109 * denial-of-service in which a lower level delegated dataset could max out its
110 * limit and thus block recursive snapshots from being taken in the global zone.
111 * Because of this, it is possible for the snapshot count to be over the limit
112 * and snapshots taken in the global zone could cause a lower level dataset to
113 * hit or exceed its limit. The administrator taking the global zone recursive
114 * snapshot should be aware of this side-effect and behave accordingly.
115 * For consistency, the filesystem limit is also not enforced if the user can
118 * The filesystem and snapshot limits are validated by dsl_fs_ss_limit_check()
119 * and updated by dsl_fs_ss_count_adjust(). A new limit value is setup in
120 * dsl_dir_activate_fs_ss_limit() and the counts are adjusted, if necessary, by
121 * dsl_dir_init_fs_ss_count().
123 * There is a special case when we receive a filesystem that already exists. In
124 * this case a temporary clone name of %X is created (see dmu_recv_begin). We
125 * never update the filesystem counts for temporary clones.
127 * Likewise, we do not update the snapshot counts for temporary snapshots,
128 * such as those created by zfs diff.
131 extern inline dsl_dir_phys_t *dsl_dir_phys(dsl_dir_t *dd);
133 static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
136 dsl_dir_evict(void *dbu)
139 dsl_pool_t *dp = dd->dd_pool;
144 for (t = 0; t < TXG_SIZE; t++) {
145 ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t));
146 ASSERT(dd->dd_tempreserved[t] == 0);
147 ASSERT(dd->dd_space_towrite[t] == 0);
151 dsl_dir_async_rele(dd->dd_parent, dd);
153 spa_async_close(dd->dd_pool->dp_spa, dd);
156 mutex_destroy(&dd->dd_lock);
157 kmem_free(dd, sizeof (dsl_dir_t));
161 dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj,
162 const char *tail, void *tag, dsl_dir_t **ddp)
168 ASSERT(dsl_pool_config_held(dp));
170 err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
173 dd = dmu_buf_get_user(dbuf);
176 dmu_object_info_t doi;
177 dmu_object_info_from_db(dbuf, &doi);
178 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_DSL_DIR);
179 ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
185 dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
186 dd->dd_object = ddobj;
189 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
192 dsl_dir_snap_cmtime_update(dd);
194 if (dsl_dir_phys(dd)->dd_parent_obj) {
195 err = dsl_dir_hold_obj(dp,
196 dsl_dir_phys(dd)->dd_parent_obj, NULL, dd,
204 err = zap_lookup(dp->dp_meta_objset,
205 dsl_dir_phys(dd->dd_parent)->
206 dd_child_dir_zapobj, tail,
207 sizeof (foundobj), 1, &foundobj);
208 ASSERT(err || foundobj == ddobj);
210 (void) strcpy(dd->dd_myname, tail);
212 err = zap_value_search(dp->dp_meta_objset,
213 dsl_dir_phys(dd->dd_parent)->
215 ddobj, 0, dd->dd_myname);
220 (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
223 if (dsl_dir_is_clone(dd)) {
224 dmu_buf_t *origin_bonus;
225 dsl_dataset_phys_t *origin_phys;
228 * We can't open the origin dataset, because
229 * that would require opening this dsl_dir.
230 * Just look at its phys directly instead.
232 err = dmu_bonus_hold(dp->dp_meta_objset,
233 dsl_dir_phys(dd)->dd_origin_obj, FTAG,
237 origin_phys = origin_bonus->db_data;
239 origin_phys->ds_creation_txg;
240 dmu_buf_rele(origin_bonus, FTAG);
243 dmu_buf_init_user(&dd->dd_dbu, dsl_dir_evict, &dd->dd_dbuf);
244 winner = dmu_buf_set_user_ie(dbuf, &dd->dd_dbu);
245 if (winner != NULL) {
247 dsl_dir_rele(dd->dd_parent, dd);
249 mutex_destroy(&dd->dd_lock);
250 kmem_free(dd, sizeof (dsl_dir_t));
253 spa_open_ref(dp->dp_spa, dd);
258 * The dsl_dir_t has both open-to-close and instantiate-to-evict
259 * holds on the spa. We need the open-to-close holds because
260 * otherwise the spa_refcnt wouldn't change when we open a
261 * dir which the spa also has open, so we could incorrectly
262 * think it was OK to unload/export/destroy the pool. We need
263 * the instantiate-to-evict hold because the dsl_dir_t has a
264 * pointer to the dd_pool, which has a pointer to the spa_t.
266 spa_open_ref(dp->dp_spa, tag);
267 ASSERT3P(dd->dd_pool, ==, dp);
268 ASSERT3U(dd->dd_object, ==, ddobj);
269 ASSERT3P(dd->dd_dbuf, ==, dbuf);
275 dsl_dir_rele(dd->dd_parent, dd);
277 mutex_destroy(&dd->dd_lock);
278 kmem_free(dd, sizeof (dsl_dir_t));
279 dmu_buf_rele(dbuf, tag);
284 dsl_dir_rele(dsl_dir_t *dd, void *tag)
286 dprintf_dd(dd, "%s\n", "");
287 spa_close(dd->dd_pool->dp_spa, tag);
288 dmu_buf_rele(dd->dd_dbuf, tag);
292 * Remove a reference to the given dsl dir that is being asynchronously
293 * released. Async releases occur from a taskq performing eviction of
294 * dsl datasets and dirs. This process is identical to a normal release
295 * with the exception of using the async API for releasing the reference on
299 dsl_dir_async_rele(dsl_dir_t *dd, void *tag)
301 dprintf_dd(dd, "%s\n", "");
302 spa_async_close(dd->dd_pool->dp_spa, tag);
303 dmu_buf_rele(dd->dd_dbuf, tag);
306 /* buf must be long enough (MAXNAMELEN + strlen(MOS_DIR_NAME) + 1 should do) */
308 dsl_dir_name(dsl_dir_t *dd, char *buf)
311 dsl_dir_name(dd->dd_parent, buf);
312 (void) strcat(buf, "/");
316 if (!MUTEX_HELD(&dd->dd_lock)) {
318 * recursive mutex so that we can use
319 * dprintf_dd() with dd_lock held
321 mutex_enter(&dd->dd_lock);
322 (void) strcat(buf, dd->dd_myname);
323 mutex_exit(&dd->dd_lock);
325 (void) strcat(buf, dd->dd_myname);
329 /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
331 dsl_dir_namelen(dsl_dir_t *dd)
336 /* parent's name + 1 for the "/" */
337 result = dsl_dir_namelen(dd->dd_parent) + 1;
340 if (!MUTEX_HELD(&dd->dd_lock)) {
341 /* see dsl_dir_name */
342 mutex_enter(&dd->dd_lock);
343 result += strlen(dd->dd_myname);
344 mutex_exit(&dd->dd_lock);
346 result += strlen(dd->dd_myname);
353 getcomponent(const char *path, char *component, const char **nextp)
357 if ((path == NULL) || (path[0] == '\0'))
358 return (SET_ERROR(ENOENT));
359 /* This would be a good place to reserve some namespace... */
360 p = strpbrk(path, "/@");
361 if (p && (p[1] == '/' || p[1] == '@')) {
362 /* two separators in a row */
363 return (SET_ERROR(EINVAL));
365 if (p == NULL || p == path) {
367 * if the first thing is an @ or /, it had better be an
368 * @ and it had better not have any more ats or slashes,
369 * and it had better have something after the @.
372 (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
373 return (SET_ERROR(EINVAL));
374 if (strlen(path) >= MAXNAMELEN)
375 return (SET_ERROR(ENAMETOOLONG));
376 (void) strcpy(component, path);
378 } else if (p[0] == '/') {
379 if (p - path >= MAXNAMELEN)
380 return (SET_ERROR(ENAMETOOLONG));
381 (void) strncpy(component, path, p - path);
382 component[p - path] = '\0';
384 } else if (p[0] == '@') {
386 * if the next separator is an @, there better not be
389 if (strchr(path, '/'))
390 return (SET_ERROR(EINVAL));
391 if (p - path >= MAXNAMELEN)
392 return (SET_ERROR(ENAMETOOLONG));
393 (void) strncpy(component, path, p - path);
394 component[p - path] = '\0';
396 panic("invalid p=%p", (void *)p);
403 * Return the dsl_dir_t, and possibly the last component which couldn't
404 * be found in *tail. The name must be in the specified dsl_pool_t. This
405 * thread must hold the dp_config_rwlock for the pool. Returns NULL if the
406 * path is bogus, or if tail==NULL and we couldn't parse the whole name.
407 * (*tail)[0] == '@' means that the last component is a snapshot.
410 dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag,
411 dsl_dir_t **ddp, const char **tailp)
413 char buf[MAXNAMELEN];
414 const char *spaname, *next, *nextnext = NULL;
419 err = getcomponent(name, buf, &next);
423 /* Make sure the name is in the specified pool. */
424 spaname = spa_name(dp->dp_spa);
425 if (strcmp(buf, spaname) != 0)
426 return (SET_ERROR(EXDEV));
428 ASSERT(dsl_pool_config_held(dp));
430 err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
435 while (next != NULL) {
437 err = getcomponent(next, buf, &nextnext);
440 ASSERT(next[0] != '\0');
443 dprintf("looking up %s in obj%lld\n",
444 buf, dsl_dir_phys(dd)->dd_child_dir_zapobj);
446 err = zap_lookup(dp->dp_meta_objset,
447 dsl_dir_phys(dd)->dd_child_dir_zapobj,
448 buf, sizeof (ddobj), 1, &ddobj);
455 err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_dd);
458 dsl_dir_rele(dd, tag);
464 dsl_dir_rele(dd, tag);
469 * It's an error if there's more than one component left, or
470 * tailp==NULL and there's any component left.
473 (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
475 dsl_dir_rele(dd, tag);
476 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
477 err = SET_ERROR(ENOENT);
486 * If the counts are already initialized for this filesystem and its
487 * descendants then do nothing, otherwise initialize the counts.
489 * The counts on this filesystem, and those below, may be uninitialized due to
490 * either the use of a pre-existing pool which did not support the
491 * filesystem/snapshot limit feature, or one in which the feature had not yet
494 * Recursively descend the filesystem tree and update the filesystem/snapshot
495 * counts on each filesystem below, then update the cumulative count on the
496 * current filesystem. If the filesystem already has a count set on it,
497 * then we know that its counts, and the counts on the filesystems below it,
498 * are already correct, so we don't have to update this filesystem.
501 dsl_dir_init_fs_ss_count(dsl_dir_t *dd, dmu_tx_t *tx)
503 uint64_t my_fs_cnt = 0;
504 uint64_t my_ss_cnt = 0;
505 dsl_pool_t *dp = dd->dd_pool;
506 objset_t *os = dp->dp_meta_objset;
511 ASSERT(spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT));
512 ASSERT(dsl_pool_config_held(dp));
513 ASSERT(dmu_tx_is_syncing(tx));
515 dsl_dir_zapify(dd, tx);
518 * If the filesystem count has already been initialized then we
519 * don't need to recurse down any further.
521 if (zap_contains(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT) == 0)
524 zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP);
525 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
527 /* Iterate my child dirs */
528 for (zap_cursor_init(zc, os, dsl_dir_phys(dd)->dd_child_dir_zapobj);
529 zap_cursor_retrieve(zc, za) == 0; zap_cursor_advance(zc)) {
533 VERIFY0(dsl_dir_hold_obj(dp, za->za_first_integer, NULL, FTAG,
537 * Ignore hidden ($FREE, $MOS & $ORIGIN) objsets and
538 * temporary datasets.
540 if (chld_dd->dd_myname[0] == '$' ||
541 chld_dd->dd_myname[0] == '%') {
542 dsl_dir_rele(chld_dd, FTAG);
546 my_fs_cnt++; /* count this child */
548 dsl_dir_init_fs_ss_count(chld_dd, tx);
550 VERIFY0(zap_lookup(os, chld_dd->dd_object,
551 DD_FIELD_FILESYSTEM_COUNT, sizeof (count), 1, &count));
553 VERIFY0(zap_lookup(os, chld_dd->dd_object,
554 DD_FIELD_SNAPSHOT_COUNT, sizeof (count), 1, &count));
557 dsl_dir_rele(chld_dd, FTAG);
560 /* Count my snapshots (we counted children's snapshots above) */
561 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
562 dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds));
564 for (zap_cursor_init(zc, os, dsl_dataset_phys(ds)->ds_snapnames_zapobj);
565 zap_cursor_retrieve(zc, za) == 0;
566 zap_cursor_advance(zc)) {
567 /* Don't count temporary snapshots */
568 if (za->za_name[0] != '%')
573 dsl_dataset_rele(ds, FTAG);
575 kmem_free(zc, sizeof (zap_cursor_t));
576 kmem_free(za, sizeof (zap_attribute_t));
578 /* we're in a sync task, update counts */
579 dmu_buf_will_dirty(dd->dd_dbuf, tx);
580 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
581 sizeof (my_fs_cnt), 1, &my_fs_cnt, tx));
582 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
583 sizeof (my_ss_cnt), 1, &my_ss_cnt, tx));
587 dsl_dir_actv_fs_ss_limit_check(void *arg, dmu_tx_t *tx)
589 char *ddname = (char *)arg;
590 dsl_pool_t *dp = dmu_tx_pool(tx);
595 error = dsl_dataset_hold(dp, ddname, FTAG, &ds);
599 if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
600 dsl_dataset_rele(ds, FTAG);
601 return (SET_ERROR(ENOTSUP));
605 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT) &&
606 dsl_dir_is_zapified(dd) &&
607 zap_contains(dp->dp_meta_objset, dd->dd_object,
608 DD_FIELD_FILESYSTEM_COUNT) == 0) {
609 dsl_dataset_rele(ds, FTAG);
610 return (SET_ERROR(EALREADY));
613 dsl_dataset_rele(ds, FTAG);
618 dsl_dir_actv_fs_ss_limit_sync(void *arg, dmu_tx_t *tx)
620 char *ddname = (char *)arg;
621 dsl_pool_t *dp = dmu_tx_pool(tx);
625 VERIFY0(dsl_dataset_hold(dp, ddname, FTAG, &ds));
627 spa = dsl_dataset_get_spa(ds);
629 if (!spa_feature_is_active(spa, SPA_FEATURE_FS_SS_LIMIT)) {
631 * Since the feature was not active and we're now setting a
632 * limit, increment the feature-active counter so that the
633 * feature becomes active for the first time.
635 * We are already in a sync task so we can update the MOS.
637 spa_feature_incr(spa, SPA_FEATURE_FS_SS_LIMIT, tx);
641 * Since we are now setting a non-UINT64_MAX limit on the filesystem,
642 * we need to ensure the counts are correct. Descend down the tree from
643 * this point and update all of the counts to be accurate.
645 dsl_dir_init_fs_ss_count(ds->ds_dir, tx);
647 dsl_dataset_rele(ds, FTAG);
651 * Make sure the feature is enabled and activate it if necessary.
652 * Since we're setting a limit, ensure the on-disk counts are valid.
653 * This is only called by the ioctl path when setting a limit value.
655 * We do not need to validate the new limit, since users who can change the
656 * limit are also allowed to exceed the limit.
659 dsl_dir_activate_fs_ss_limit(const char *ddname)
663 error = dsl_sync_task(ddname, dsl_dir_actv_fs_ss_limit_check,
664 dsl_dir_actv_fs_ss_limit_sync, (void *)ddname, 0,
665 ZFS_SPACE_CHECK_RESERVED);
667 if (error == EALREADY)
674 * Used to determine if the filesystem_limit or snapshot_limit should be
675 * enforced. We allow the limit to be exceeded if the user has permission to
676 * write the property value. We pass in the creds that we got in the open
677 * context since we will always be the GZ root in syncing context. We also have
678 * to handle the case where we are allowed to change the limit on the current
679 * dataset, but there may be another limit in the tree above.
681 * We can never modify these two properties within a non-global zone. In
682 * addition, the other checks are modeled on zfs_secpolicy_write_perms. We
683 * can't use that function since we are already holding the dp_config_rwlock.
684 * In addition, we already have the dd and dealing with snapshots is simplified
695 dsl_enforce_ds_ss_limits(dsl_dir_t *dd, zfs_prop_t prop, cred_t *cr)
697 enforce_res_t enforce = ENFORCE_ALWAYS;
702 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
703 prop == ZFS_PROP_SNAPSHOT_LIMIT);
709 if (crgetzoneid(cr) != GLOBAL_ZONEID)
711 return (ENFORCE_ALWAYS);
713 if (secpolicy_zfs(cr) == 0)
714 return (ENFORCE_NEVER);
717 if ((obj = dsl_dir_phys(dd)->dd_head_dataset_obj) == 0)
718 return (ENFORCE_ALWAYS);
720 ASSERT(dsl_pool_config_held(dd->dd_pool));
722 if (dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds) != 0)
723 return (ENFORCE_ALWAYS);
725 if (dsl_prop_get_ds(ds, "zoned", 8, 1, &zoned, NULL) || zoned) {
726 /* Only root can access zoned fs's from the GZ */
727 enforce = ENFORCE_ALWAYS;
729 if (dsl_deleg_access_impl(ds, zfs_prop_to_name(prop), cr) == 0)
730 enforce = ENFORCE_ABOVE;
733 dsl_dataset_rele(ds, FTAG);
738 * Check if adding additional child filesystem(s) would exceed any filesystem
739 * limits or adding additional snapshot(s) would exceed any snapshot limits.
740 * The prop argument indicates which limit to check.
742 * Note that all filesystem limits up to the root (or the highest
743 * initialized) filesystem or the given ancestor must be satisfied.
746 dsl_fs_ss_limit_check(dsl_dir_t *dd, uint64_t delta, zfs_prop_t prop,
747 dsl_dir_t *ancestor, cred_t *cr)
749 objset_t *os = dd->dd_pool->dp_meta_objset;
750 uint64_t limit, count;
752 enforce_res_t enforce;
755 ASSERT(dsl_pool_config_held(dd->dd_pool));
756 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
757 prop == ZFS_PROP_SNAPSHOT_LIMIT);
760 * If we're allowed to change the limit, don't enforce the limit
761 * e.g. this can happen if a snapshot is taken by an administrative
762 * user in the global zone (i.e. a recursive snapshot by root).
763 * However, we must handle the case of delegated permissions where we
764 * are allowed to change the limit on the current dataset, but there
765 * is another limit in the tree above.
767 enforce = dsl_enforce_ds_ss_limits(dd, prop, cr);
768 if (enforce == ENFORCE_NEVER)
772 * e.g. if renaming a dataset with no snapshots, count adjustment
778 if (prop == ZFS_PROP_SNAPSHOT_LIMIT) {
780 * We don't enforce the limit for temporary snapshots. This is
781 * indicated by a NULL cred_t argument.
786 count_prop = DD_FIELD_SNAPSHOT_COUNT;
788 count_prop = DD_FIELD_FILESYSTEM_COUNT;
792 * If an ancestor has been provided, stop checking the limit once we
793 * hit that dir. We need this during rename so that we don't overcount
794 * the check once we recurse up to the common ancestor.
800 * If we hit an uninitialized node while recursing up the tree, we can
801 * stop since we know there is no limit here (or above). The counts are
802 * not valid on this node and we know we won't touch this node's counts.
804 if (!dsl_dir_is_zapified(dd) || zap_lookup(os, dd->dd_object,
805 count_prop, sizeof (count), 1, &count) == ENOENT)
808 err = dsl_prop_get_dd(dd, zfs_prop_to_name(prop), 8, 1, &limit, NULL,
813 /* Is there a limit which we've hit? */
814 if (enforce == ENFORCE_ALWAYS && (count + delta) > limit)
815 return (SET_ERROR(EDQUOT));
817 if (dd->dd_parent != NULL)
818 err = dsl_fs_ss_limit_check(dd->dd_parent, delta, prop,
825 * Adjust the filesystem or snapshot count for the specified dsl_dir_t and all
826 * parents. When a new filesystem/snapshot is created, increment the count on
827 * all parents, and when a filesystem/snapshot is destroyed, decrement the
831 dsl_fs_ss_count_adjust(dsl_dir_t *dd, int64_t delta, const char *prop,
835 objset_t *os = dd->dd_pool->dp_meta_objset;
838 ASSERT(dsl_pool_config_held(dd->dd_pool));
839 ASSERT(dmu_tx_is_syncing(tx));
840 ASSERT(strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0 ||
841 strcmp(prop, DD_FIELD_SNAPSHOT_COUNT) == 0);
844 * When we receive an incremental stream into a filesystem that already
845 * exists, a temporary clone is created. We don't count this temporary
846 * clone, whose name begins with a '%'. We also ignore hidden ($FREE,
847 * $MOS & $ORIGIN) objsets.
849 if ((dd->dd_myname[0] == '%' || dd->dd_myname[0] == '$') &&
850 strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0)
854 * e.g. if renaming a dataset with no snapshots, count adjustment is 0
860 * If we hit an uninitialized node while recursing up the tree, we can
861 * stop since we know the counts are not valid on this node and we
862 * know we shouldn't touch this node's counts. An uninitialized count
863 * on the node indicates that either the feature has not yet been
864 * activated or there are no limits on this part of the tree.
866 if (!dsl_dir_is_zapified(dd) || (err = zap_lookup(os, dd->dd_object,
867 prop, sizeof (count), 1, &count)) == ENOENT)
872 /* Use a signed verify to make sure we're not neg. */
873 VERIFY3S(count, >=, 0);
875 VERIFY0(zap_update(os, dd->dd_object, prop, sizeof (count), 1, &count,
878 /* Roll up this additional count into our ancestors */
879 if (dd->dd_parent != NULL)
880 dsl_fs_ss_count_adjust(dd->dd_parent, delta, prop, tx);
884 dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
887 objset_t *mos = dp->dp_meta_objset;
889 dsl_dir_phys_t *ddphys;
892 ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
893 DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
895 VERIFY(0 == zap_add(mos, dsl_dir_phys(pds)->dd_child_dir_zapobj,
896 name, sizeof (uint64_t), 1, &ddobj, tx));
898 /* it's the root dir */
899 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
900 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
902 VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
903 dmu_buf_will_dirty(dbuf, tx);
904 ddphys = dbuf->db_data;
906 ddphys->dd_creation_time = gethrestime_sec();
908 ddphys->dd_parent_obj = pds->dd_object;
910 /* update the filesystem counts */
911 dsl_fs_ss_count_adjust(pds, 1, DD_FIELD_FILESYSTEM_COUNT, tx);
913 ddphys->dd_props_zapobj = zap_create(mos,
914 DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
915 ddphys->dd_child_dir_zapobj = zap_create(mos,
916 DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
917 if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
918 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
919 dmu_buf_rele(dbuf, FTAG);
925 dsl_dir_is_clone(dsl_dir_t *dd)
927 return (dsl_dir_phys(dd)->dd_origin_obj &&
928 (dd->dd_pool->dp_origin_snap == NULL ||
929 dsl_dir_phys(dd)->dd_origin_obj !=
930 dd->dd_pool->dp_origin_snap->ds_object));
934 dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
936 mutex_enter(&dd->dd_lock);
937 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
938 dsl_dir_phys(dd)->dd_used_bytes);
939 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA,
940 dsl_dir_phys(dd)->dd_quota);
941 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
942 dsl_dir_phys(dd)->dd_reserved);
943 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
944 dsl_dir_phys(dd)->dd_compressed_bytes == 0 ? 100 :
945 (dsl_dir_phys(dd)->dd_uncompressed_bytes * 100 /
946 dsl_dir_phys(dd)->dd_compressed_bytes));
947 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED,
948 dsl_dir_phys(dd)->dd_uncompressed_bytes);
949 if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
950 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
951 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]);
952 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
953 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_HEAD]);
954 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
955 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_REFRSRV]);
956 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
957 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD] +
958 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD_RSRV]);
960 mutex_exit(&dd->dd_lock);
962 if (dsl_dir_is_zapified(dd)) {
964 objset_t *os = dd->dd_pool->dp_meta_objset;
966 if (zap_lookup(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
967 sizeof (count), 1, &count) == 0) {
968 dsl_prop_nvlist_add_uint64(nv,
969 ZFS_PROP_FILESYSTEM_COUNT, count);
971 if (zap_lookup(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
972 sizeof (count), 1, &count) == 0) {
973 dsl_prop_nvlist_add_uint64(nv,
974 ZFS_PROP_SNAPSHOT_COUNT, count);
978 if (dsl_dir_is_clone(dd)) {
980 char buf[MAXNAMELEN];
982 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
983 dsl_dir_phys(dd)->dd_origin_obj, FTAG, &ds));
984 dsl_dataset_name(ds, buf);
985 dsl_dataset_rele(ds, FTAG);
986 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
991 dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
993 dsl_pool_t *dp = dd->dd_pool;
995 ASSERT(dsl_dir_phys(dd));
997 if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) {
998 /* up the hold count until we can be written out */
999 dmu_buf_add_ref(dd->dd_dbuf, dd);
1004 parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
1006 uint64_t old_accounted = MAX(used, dsl_dir_phys(dd)->dd_reserved);
1007 uint64_t new_accounted =
1008 MAX(used + delta, dsl_dir_phys(dd)->dd_reserved);
1009 return (new_accounted - old_accounted);
1013 dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
1015 ASSERT(dmu_tx_is_syncing(tx));
1017 mutex_enter(&dd->dd_lock);
1018 ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]);
1019 dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
1020 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
1021 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
1022 mutex_exit(&dd->dd_lock);
1024 /* release the hold from dsl_dir_dirty */
1025 dmu_buf_rele(dd->dd_dbuf, dd);
1029 dsl_dir_space_towrite(dsl_dir_t *dd)
1034 ASSERT(MUTEX_HELD(&dd->dd_lock));
1036 for (i = 0; i < TXG_SIZE; i++) {
1037 space += dd->dd_space_towrite[i&TXG_MASK];
1038 ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0);
1044 * How much space would dd have available if ancestor had delta applied
1045 * to it? If ondiskonly is set, we're only interested in what's
1046 * on-disk, not estimated pending changes.
1049 dsl_dir_space_available(dsl_dir_t *dd,
1050 dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
1052 uint64_t parentspace, myspace, quota, used;
1055 * If there are no restrictions otherwise, assume we have
1056 * unlimited space available.
1059 parentspace = UINT64_MAX;
1061 if (dd->dd_parent != NULL) {
1062 parentspace = dsl_dir_space_available(dd->dd_parent,
1063 ancestor, delta, ondiskonly);
1066 mutex_enter(&dd->dd_lock);
1067 if (dsl_dir_phys(dd)->dd_quota != 0)
1068 quota = dsl_dir_phys(dd)->dd_quota;
1069 used = dsl_dir_phys(dd)->dd_used_bytes;
1071 used += dsl_dir_space_towrite(dd);
1073 if (dd->dd_parent == NULL) {
1074 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
1075 quota = MIN(quota, poolsize);
1078 if (dsl_dir_phys(dd)->dd_reserved > used && parentspace != UINT64_MAX) {
1080 * We have some space reserved, in addition to what our
1083 parentspace += dsl_dir_phys(dd)->dd_reserved - used;
1086 if (dd == ancestor) {
1088 ASSERT(used >= -delta);
1090 if (parentspace != UINT64_MAX)
1091 parentspace -= delta;
1099 * the lesser of the space provided by our parent and
1100 * the space left in our quota
1102 myspace = MIN(parentspace, quota - used);
1105 mutex_exit(&dd->dd_lock);
1110 struct tempreserve {
1111 list_node_t tr_node;
1117 dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
1118 boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list,
1119 dmu_tx_t *tx, boolean_t first)
1121 uint64_t txg = tx->tx_txg;
1122 uint64_t est_inflight, used_on_disk, quota, parent_rsrv;
1123 uint64_t deferred = 0;
1124 struct tempreserve *tr;
1125 int retval = EDQUOT;
1126 int txgidx = txg & TXG_MASK;
1128 uint64_t ref_rsrv = 0;
1130 ASSERT3U(txg, !=, 0);
1131 ASSERT3S(asize, >, 0);
1133 mutex_enter(&dd->dd_lock);
1136 * Check against the dsl_dir's quota. We don't add in the delta
1137 * when checking for over-quota because they get one free hit.
1139 est_inflight = dsl_dir_space_towrite(dd);
1140 for (i = 0; i < TXG_SIZE; i++)
1141 est_inflight += dd->dd_tempreserved[i];
1142 used_on_disk = dsl_dir_phys(dd)->dd_used_bytes;
1145 * On the first iteration, fetch the dataset's used-on-disk and
1146 * refreservation values. Also, if checkrefquota is set, test if
1147 * allocating this space would exceed the dataset's refquota.
1149 if (first && tx->tx_objset) {
1151 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
1153 error = dsl_dataset_check_quota(ds, checkrefquota,
1154 asize, est_inflight, &used_on_disk, &ref_rsrv);
1156 mutex_exit(&dd->dd_lock);
1162 * If this transaction will result in a net free of space,
1163 * we want to let it through.
1165 if (ignorequota || netfree || dsl_dir_phys(dd)->dd_quota == 0)
1168 quota = dsl_dir_phys(dd)->dd_quota;
1171 * Adjust the quota against the actual pool size at the root
1172 * minus any outstanding deferred frees.
1173 * To ensure that it's possible to remove files from a full
1174 * pool without inducing transient overcommits, we throttle
1175 * netfree transactions against a quota that is slightly larger,
1176 * but still within the pool's allocation slop. In cases where
1177 * we're very close to full, this will allow a steady trickle of
1178 * removes to get through.
1180 if (dd->dd_parent == NULL) {
1181 spa_t *spa = dd->dd_pool->dp_spa;
1182 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
1183 deferred = metaslab_class_get_deferred(spa_normal_class(spa));
1184 if (poolsize - deferred < quota) {
1185 quota = poolsize - deferred;
1191 * If they are requesting more space, and our current estimate
1192 * is over quota, they get to try again unless the actual
1193 * on-disk is over quota and there are no pending changes (which
1194 * may free up space for us).
1196 if (used_on_disk + est_inflight >= quota) {
1197 if (est_inflight > 0 || used_on_disk < quota ||
1198 (retval == ENOSPC && used_on_disk < quota + deferred))
1200 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
1201 "quota=%lluK tr=%lluK err=%d\n",
1202 used_on_disk>>10, est_inflight>>10,
1203 quota>>10, asize>>10, retval);
1204 mutex_exit(&dd->dd_lock);
1205 return (SET_ERROR(retval));
1208 /* We need to up our estimated delta before dropping dd_lock */
1209 dd->dd_tempreserved[txgidx] += asize;
1211 parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
1213 mutex_exit(&dd->dd_lock);
1215 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1217 tr->tr_size = asize;
1218 list_insert_tail(tr_list, tr);
1220 /* see if it's OK with our parent */
1221 if (dd->dd_parent && parent_rsrv) {
1222 boolean_t ismos = (dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
1224 return (dsl_dir_tempreserve_impl(dd->dd_parent,
1225 parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE));
1232 * Reserve space in this dsl_dir, to be used in this tx's txg.
1233 * After the space has been dirtied (and dsl_dir_willuse_space()
1234 * has been called), the reservation should be canceled, using
1235 * dsl_dir_tempreserve_clear().
1238 dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
1239 uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx)
1249 tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
1250 list_create(tr_list, sizeof (struct tempreserve),
1251 offsetof(struct tempreserve, tr_node));
1252 ASSERT3S(asize, >, 0);
1253 ASSERT3S(fsize, >=, 0);
1255 err = arc_tempreserve_space(lsize, tx->tx_txg);
1257 struct tempreserve *tr;
1259 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1260 tr->tr_size = lsize;
1261 list_insert_tail(tr_list, tr);
1263 if (err == EAGAIN) {
1265 * If arc_memory_throttle() detected that pageout
1266 * is running and we are low on memory, we delay new
1267 * non-pageout transactions to give pageout an
1270 * It is unfortunate to be delaying while the caller's
1273 txg_delay(dd->dd_pool, tx->tx_txg,
1274 MSEC2NSEC(10), MSEC2NSEC(10));
1275 err = SET_ERROR(ERESTART);
1280 err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize,
1281 FALSE, asize > usize, tr_list, tx, TRUE);
1285 dsl_dir_tempreserve_clear(tr_list, tx);
1287 *tr_cookiep = tr_list;
1293 * Clear a temporary reservation that we previously made with
1294 * dsl_dir_tempreserve_space().
1297 dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
1299 int txgidx = tx->tx_txg & TXG_MASK;
1300 list_t *tr_list = tr_cookie;
1301 struct tempreserve *tr;
1303 ASSERT3U(tx->tx_txg, !=, 0);
1305 if (tr_cookie == NULL)
1308 while ((tr = list_head(tr_list)) != NULL) {
1310 mutex_enter(&tr->tr_ds->dd_lock);
1311 ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
1313 tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
1314 mutex_exit(&tr->tr_ds->dd_lock);
1316 arc_tempreserve_clear(tr->tr_size);
1318 list_remove(tr_list, tr);
1319 kmem_free(tr, sizeof (struct tempreserve));
1322 kmem_free(tr_list, sizeof (list_t));
1326 * This should be called from open context when we think we're going to write
1327 * or free space, for example when dirtying data. Be conservative; it's okay
1328 * to write less space or free more, but we don't want to write more or free
1329 * less than the amount specified.
1332 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
1334 int64_t parent_space;
1337 mutex_enter(&dd->dd_lock);
1339 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
1341 est_used = dsl_dir_space_towrite(dd) + dsl_dir_phys(dd)->dd_used_bytes;
1342 parent_space = parent_delta(dd, est_used, space);
1343 mutex_exit(&dd->dd_lock);
1345 /* Make sure that we clean up dd_space_to* */
1346 dsl_dir_dirty(dd, tx);
1348 /* XXX this is potentially expensive and unnecessary... */
1349 if (parent_space && dd->dd_parent)
1350 dsl_dir_willuse_space(dd->dd_parent, parent_space, tx);
1353 /* call from syncing context when we actually write/free space for this dd */
1355 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
1356 int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
1358 int64_t accounted_delta;
1361 * dsl_dataset_set_refreservation_sync_impl() calls this with
1362 * dd_lock held, so that it can atomically update
1363 * ds->ds_reserved and the dsl_dir accounting, so that
1364 * dsl_dataset_check_quota() can see dataset and dir accounting
1367 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
1369 ASSERT(dmu_tx_is_syncing(tx));
1370 ASSERT(type < DD_USED_NUM);
1372 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1375 mutex_enter(&dd->dd_lock);
1377 parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, used);
1378 ASSERT(used >= 0 || dsl_dir_phys(dd)->dd_used_bytes >= -used);
1379 ASSERT(compressed >= 0 ||
1380 dsl_dir_phys(dd)->dd_compressed_bytes >= -compressed);
1381 ASSERT(uncompressed >= 0 ||
1382 dsl_dir_phys(dd)->dd_uncompressed_bytes >= -uncompressed);
1383 dsl_dir_phys(dd)->dd_used_bytes += used;
1384 dsl_dir_phys(dd)->dd_uncompressed_bytes += uncompressed;
1385 dsl_dir_phys(dd)->dd_compressed_bytes += compressed;
1387 if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1389 dsl_dir_phys(dd)->dd_used_breakdown[type] >= -used);
1390 dsl_dir_phys(dd)->dd_used_breakdown[type] += used;
1394 for (t = 0; t < DD_USED_NUM; t++)
1395 u += dsl_dir_phys(dd)->dd_used_breakdown[t];
1396 ASSERT3U(u, ==, dsl_dir_phys(dd)->dd_used_bytes);
1400 mutex_exit(&dd->dd_lock);
1402 if (dd->dd_parent != NULL) {
1403 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1404 accounted_delta, compressed, uncompressed, tx);
1405 dsl_dir_transfer_space(dd->dd_parent,
1406 used - accounted_delta,
1407 DD_USED_CHILD_RSRV, DD_USED_CHILD, NULL);
1412 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
1413 dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
1415 ASSERT(tx == NULL || dmu_tx_is_syncing(tx));
1416 ASSERT(oldtype < DD_USED_NUM);
1417 ASSERT(newtype < DD_USED_NUM);
1420 !(dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN))
1424 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1425 mutex_enter(&dd->dd_lock);
1427 dsl_dir_phys(dd)->dd_used_breakdown[oldtype] >= delta :
1428 dsl_dir_phys(dd)->dd_used_breakdown[newtype] >= -delta);
1429 ASSERT(dsl_dir_phys(dd)->dd_used_bytes >= ABS(delta));
1430 dsl_dir_phys(dd)->dd_used_breakdown[oldtype] -= delta;
1431 dsl_dir_phys(dd)->dd_used_breakdown[newtype] += delta;
1432 mutex_exit(&dd->dd_lock);
1435 typedef struct dsl_dir_set_qr_arg {
1436 const char *ddsqra_name;
1437 zprop_source_t ddsqra_source;
1438 uint64_t ddsqra_value;
1439 } dsl_dir_set_qr_arg_t;
1442 dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx)
1444 dsl_dir_set_qr_arg_t *ddsqra = arg;
1445 dsl_pool_t *dp = dmu_tx_pool(tx);
1448 uint64_t towrite, newval;
1450 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1454 error = dsl_prop_predict(ds->ds_dir, "quota",
1455 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1457 dsl_dataset_rele(ds, FTAG);
1462 dsl_dataset_rele(ds, FTAG);
1466 mutex_enter(&ds->ds_dir->dd_lock);
1468 * If we are doing the preliminary check in open context, and
1469 * there are pending changes, then don't fail it, since the
1470 * pending changes could under-estimate the amount of space to be
1473 towrite = dsl_dir_space_towrite(ds->ds_dir);
1474 if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
1475 (newval < dsl_dir_phys(ds->ds_dir)->dd_reserved ||
1476 newval < dsl_dir_phys(ds->ds_dir)->dd_used_bytes + towrite)) {
1477 error = SET_ERROR(ENOSPC);
1479 mutex_exit(&ds->ds_dir->dd_lock);
1480 dsl_dataset_rele(ds, FTAG);
1485 dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx)
1487 dsl_dir_set_qr_arg_t *ddsqra = arg;
1488 dsl_pool_t *dp = dmu_tx_pool(tx);
1492 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1494 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1495 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA),
1496 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1497 &ddsqra->ddsqra_value, tx);
1499 VERIFY0(dsl_prop_get_int_ds(ds,
1500 zfs_prop_to_name(ZFS_PROP_QUOTA), &newval));
1502 newval = ddsqra->ddsqra_value;
1503 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1504 zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval);
1507 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1508 mutex_enter(&ds->ds_dir->dd_lock);
1509 dsl_dir_phys(ds->ds_dir)->dd_quota = newval;
1510 mutex_exit(&ds->ds_dir->dd_lock);
1511 dsl_dataset_rele(ds, FTAG);
1515 dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
1517 dsl_dir_set_qr_arg_t ddsqra;
1519 ddsqra.ddsqra_name = ddname;
1520 ddsqra.ddsqra_source = source;
1521 ddsqra.ddsqra_value = quota;
1523 return (dsl_sync_task(ddname, dsl_dir_set_quota_check,
1524 dsl_dir_set_quota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
1528 dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx)
1530 dsl_dir_set_qr_arg_t *ddsqra = arg;
1531 dsl_pool_t *dp = dmu_tx_pool(tx);
1534 uint64_t newval, used, avail;
1537 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1543 * If we are doing the preliminary check in open context, the
1544 * space estimates may be inaccurate.
1546 if (!dmu_tx_is_syncing(tx)) {
1547 dsl_dataset_rele(ds, FTAG);
1551 error = dsl_prop_predict(ds->ds_dir,
1552 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1553 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1555 dsl_dataset_rele(ds, FTAG);
1559 mutex_enter(&dd->dd_lock);
1560 used = dsl_dir_phys(dd)->dd_used_bytes;
1561 mutex_exit(&dd->dd_lock);
1563 if (dd->dd_parent) {
1564 avail = dsl_dir_space_available(dd->dd_parent,
1567 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
1570 if (MAX(used, newval) > MAX(used, dsl_dir_phys(dd)->dd_reserved)) {
1571 uint64_t delta = MAX(used, newval) -
1572 MAX(used, dsl_dir_phys(dd)->dd_reserved);
1574 if (delta > avail ||
1575 (dsl_dir_phys(dd)->dd_quota > 0 &&
1576 newval > dsl_dir_phys(dd)->dd_quota))
1577 error = SET_ERROR(ENOSPC);
1580 dsl_dataset_rele(ds, FTAG);
1585 dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
1590 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1592 mutex_enter(&dd->dd_lock);
1593 used = dsl_dir_phys(dd)->dd_used_bytes;
1594 delta = MAX(used, value) - MAX(used, dsl_dir_phys(dd)->dd_reserved);
1595 dsl_dir_phys(dd)->dd_reserved = value;
1597 if (dd->dd_parent != NULL) {
1598 /* Roll up this additional usage into our ancestors */
1599 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1602 mutex_exit(&dd->dd_lock);
1606 dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx)
1608 dsl_dir_set_qr_arg_t *ddsqra = arg;
1609 dsl_pool_t *dp = dmu_tx_pool(tx);
1613 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1615 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1616 dsl_prop_set_sync_impl(ds,
1617 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1618 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1619 &ddsqra->ddsqra_value, tx);
1621 VERIFY0(dsl_prop_get_int_ds(ds,
1622 zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval));
1624 newval = ddsqra->ddsqra_value;
1625 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1626 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1627 (longlong_t)newval);
1630 dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx);
1631 dsl_dataset_rele(ds, FTAG);
1635 dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1636 uint64_t reservation)
1638 dsl_dir_set_qr_arg_t ddsqra;
1640 ddsqra.ddsqra_name = ddname;
1641 ddsqra.ddsqra_source = source;
1642 ddsqra.ddsqra_value = reservation;
1644 return (dsl_sync_task(ddname, dsl_dir_set_reservation_check,
1645 dsl_dir_set_reservation_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
1649 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1651 for (; ds1; ds1 = ds1->dd_parent) {
1653 for (dd = ds2; dd; dd = dd->dd_parent) {
1662 * If delta is applied to dd, how much of that delta would be applied to
1663 * ancestor? Syncing context only.
1666 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1671 mutex_enter(&dd->dd_lock);
1672 delta = parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, delta);
1673 mutex_exit(&dd->dd_lock);
1674 return (would_change(dd->dd_parent, delta, ancestor));
1677 typedef struct dsl_dir_rename_arg {
1678 const char *ddra_oldname;
1679 const char *ddra_newname;
1681 } dsl_dir_rename_arg_t;
1685 dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1688 char namebuf[MAXNAMELEN];
1690 dsl_dataset_name(ds, namebuf);
1692 if (strlen(namebuf) + *deltap >= MAXNAMELEN)
1693 return (SET_ERROR(ENAMETOOLONG));
1698 dsl_dir_rename_check(void *arg, dmu_tx_t *tx)
1700 dsl_dir_rename_arg_t *ddra = arg;
1701 dsl_pool_t *dp = dmu_tx_pool(tx);
1702 dsl_dir_t *dd, *newparent;
1703 const char *mynewname;
1705 int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname);
1707 /* target dir should exist */
1708 error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL);
1712 /* new parent should exist */
1713 error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG,
1714 &newparent, &mynewname);
1716 dsl_dir_rele(dd, FTAG);
1720 /* can't rename to different pool */
1721 if (dd->dd_pool != newparent->dd_pool) {
1722 dsl_dir_rele(newparent, FTAG);
1723 dsl_dir_rele(dd, FTAG);
1724 return (SET_ERROR(EXDEV));
1727 /* new name should not already exist */
1728 if (mynewname == NULL) {
1729 dsl_dir_rele(newparent, FTAG);
1730 dsl_dir_rele(dd, FTAG);
1731 return (SET_ERROR(EEXIST));
1734 /* if the name length is growing, validate child name lengths */
1736 error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename,
1737 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1739 dsl_dir_rele(newparent, FTAG);
1740 dsl_dir_rele(dd, FTAG);
1745 if (dmu_tx_is_syncing(tx)) {
1746 if (spa_feature_is_active(dp->dp_spa,
1747 SPA_FEATURE_FS_SS_LIMIT)) {
1749 * Although this is the check function and we don't
1750 * normally make on-disk changes in check functions,
1751 * we need to do that here.
1753 * Ensure this portion of the tree's counts have been
1754 * initialized in case the new parent has limits set.
1756 dsl_dir_init_fs_ss_count(dd, tx);
1760 if (newparent != dd->dd_parent) {
1761 /* is there enough space? */
1763 MAX(dsl_dir_phys(dd)->dd_used_bytes,
1764 dsl_dir_phys(dd)->dd_reserved);
1765 objset_t *os = dd->dd_pool->dp_meta_objset;
1766 uint64_t fs_cnt = 0;
1767 uint64_t ss_cnt = 0;
1769 if (dsl_dir_is_zapified(dd)) {
1772 err = zap_lookup(os, dd->dd_object,
1773 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1775 if (err != ENOENT && err != 0) {
1776 dsl_dir_rele(newparent, FTAG);
1777 dsl_dir_rele(dd, FTAG);
1782 * have to add 1 for the filesystem itself that we're
1787 err = zap_lookup(os, dd->dd_object,
1788 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1790 if (err != ENOENT && err != 0) {
1791 dsl_dir_rele(newparent, FTAG);
1792 dsl_dir_rele(dd, FTAG);
1797 /* no rename into our descendant */
1798 if (closest_common_ancestor(dd, newparent) == dd) {
1799 dsl_dir_rele(newparent, FTAG);
1800 dsl_dir_rele(dd, FTAG);
1801 return (SET_ERROR(EINVAL));
1804 error = dsl_dir_transfer_possible(dd->dd_parent,
1805 newparent, fs_cnt, ss_cnt, myspace, ddra->ddra_cred);
1807 dsl_dir_rele(newparent, FTAG);
1808 dsl_dir_rele(dd, FTAG);
1813 dsl_dir_rele(newparent, FTAG);
1814 dsl_dir_rele(dd, FTAG);
1819 dsl_dir_rename_sync(void *arg, dmu_tx_t *tx)
1821 dsl_dir_rename_arg_t *ddra = arg;
1822 dsl_pool_t *dp = dmu_tx_pool(tx);
1823 dsl_dir_t *dd, *newparent;
1824 const char *mynewname;
1826 objset_t *mos = dp->dp_meta_objset;
1828 VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL));
1829 VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
1832 /* Log this before we change the name. */
1833 spa_history_log_internal_dd(dd, "rename", tx,
1834 "-> %s", ddra->ddra_newname);
1836 if (newparent != dd->dd_parent) {
1837 objset_t *os = dd->dd_pool->dp_meta_objset;
1838 uint64_t fs_cnt = 0;
1839 uint64_t ss_cnt = 0;
1842 * We already made sure the dd counts were initialized in the
1845 if (spa_feature_is_active(dp->dp_spa,
1846 SPA_FEATURE_FS_SS_LIMIT)) {
1847 VERIFY0(zap_lookup(os, dd->dd_object,
1848 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1850 /* add 1 for the filesystem itself that we're moving */
1853 VERIFY0(zap_lookup(os, dd->dd_object,
1854 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1858 dsl_fs_ss_count_adjust(dd->dd_parent, -fs_cnt,
1859 DD_FIELD_FILESYSTEM_COUNT, tx);
1860 dsl_fs_ss_count_adjust(newparent, fs_cnt,
1861 DD_FIELD_FILESYSTEM_COUNT, tx);
1863 dsl_fs_ss_count_adjust(dd->dd_parent, -ss_cnt,
1864 DD_FIELD_SNAPSHOT_COUNT, tx);
1865 dsl_fs_ss_count_adjust(newparent, ss_cnt,
1866 DD_FIELD_SNAPSHOT_COUNT, tx);
1868 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1869 -dsl_dir_phys(dd)->dd_used_bytes,
1870 -dsl_dir_phys(dd)->dd_compressed_bytes,
1871 -dsl_dir_phys(dd)->dd_uncompressed_bytes, tx);
1872 dsl_dir_diduse_space(newparent, DD_USED_CHILD,
1873 dsl_dir_phys(dd)->dd_used_bytes,
1874 dsl_dir_phys(dd)->dd_compressed_bytes,
1875 dsl_dir_phys(dd)->dd_uncompressed_bytes, tx);
1877 if (dsl_dir_phys(dd)->dd_reserved >
1878 dsl_dir_phys(dd)->dd_used_bytes) {
1879 uint64_t unused_rsrv = dsl_dir_phys(dd)->dd_reserved -
1880 dsl_dir_phys(dd)->dd_used_bytes;
1882 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1883 -unused_rsrv, 0, 0, tx);
1884 dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV,
1885 unused_rsrv, 0, 0, tx);
1889 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1891 /* remove from old parent zapobj */
1892 error = zap_remove(mos,
1893 dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj,
1897 (void) strcpy(dd->dd_myname, mynewname);
1898 dsl_dir_rele(dd->dd_parent, dd);
1899 dsl_dir_phys(dd)->dd_parent_obj = newparent->dd_object;
1900 VERIFY0(dsl_dir_hold_obj(dp,
1901 newparent->dd_object, NULL, dd, &dd->dd_parent));
1903 /* add to new parent zapobj */
1904 VERIFY0(zap_add(mos, dsl_dir_phys(newparent)->dd_child_dir_zapobj,
1905 dd->dd_myname, 8, 1, &dd->dd_object, tx));
1909 zfsvfs_update_fromname(ddra->ddra_oldname, ddra->ddra_newname);
1910 zvol_rename_minors(ddra->ddra_oldname, ddra->ddra_newname);
1914 dsl_prop_notify_all(dd);
1916 dsl_dir_rele(newparent, FTAG);
1917 dsl_dir_rele(dd, FTAG);
1921 dsl_dir_rename(const char *oldname, const char *newname)
1923 dsl_dir_rename_arg_t ddra;
1925 ddra.ddra_oldname = oldname;
1926 ddra.ddra_newname = newname;
1927 ddra.ddra_cred = CRED();
1929 return (dsl_sync_task(oldname,
1930 dsl_dir_rename_check, dsl_dir_rename_sync, &ddra,
1931 3, ZFS_SPACE_CHECK_RESERVED));
1935 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd,
1936 uint64_t fs_cnt, uint64_t ss_cnt, uint64_t space, cred_t *cr)
1938 dsl_dir_t *ancestor;
1943 ancestor = closest_common_ancestor(sdd, tdd);
1944 adelta = would_change(sdd, -space, ancestor);
1945 avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1947 return (SET_ERROR(ENOSPC));
1949 err = dsl_fs_ss_limit_check(tdd, fs_cnt, ZFS_PROP_FILESYSTEM_LIMIT,
1953 err = dsl_fs_ss_limit_check(tdd, ss_cnt, ZFS_PROP_SNAPSHOT_LIMIT,
1962 dsl_dir_snap_cmtime(dsl_dir_t *dd)
1966 mutex_enter(&dd->dd_lock);
1967 t = dd->dd_snap_cmtime;
1968 mutex_exit(&dd->dd_lock);
1974 dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
1979 mutex_enter(&dd->dd_lock);
1980 dd->dd_snap_cmtime = t;
1981 mutex_exit(&dd->dd_lock);
1985 dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx)
1987 objset_t *mos = dd->dd_pool->dp_meta_objset;
1988 dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx);
1992 dsl_dir_is_zapified(dsl_dir_t *dd)
1994 dmu_object_info_t doi;
1996 dmu_object_info_from_db(dd->dd_dbuf, &doi);
1997 return (doi.doi_type == DMU_OTN_ZAP_METADATA);