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) 2013 by Delphix. All rights reserved.
26 * Copyright (c) 2014 Joyent, Inc. All rights reserved.
30 #include <sys/dmu_objset.h>
31 #include <sys/dmu_tx.h>
32 #include <sys/dsl_dataset.h>
33 #include <sys/dsl_dir.h>
34 #include <sys/dsl_prop.h>
35 #include <sys/dsl_synctask.h>
36 #include <sys/dsl_deleg.h>
37 #include <sys/dmu_impl.h>
39 #include <sys/metaslab.h>
43 #include <sys/sunddi.h>
46 #include <sys/zfs_vfsops.h>
48 #include <sys/zfeature.h>
49 #include <sys/policy.h>
50 #include <sys/zfs_znode.h>
51 #include "zfs_namecheck.h"
55 * Filesystem and Snapshot Limits
56 * ------------------------------
58 * These limits are used to restrict the number of filesystems and/or snapshots
59 * that can be created at a given level in the tree or below. A typical
60 * use-case is with a delegated dataset where the administrator wants to ensure
61 * that a user within the zone is not creating too many additional filesystems
62 * or snapshots, even though they're not exceeding their space quota.
64 * The filesystem and snapshot counts are stored as extensible properties. This
65 * capability is controlled by a feature flag and must be enabled to be used.
66 * Once enabled, the feature is not active until the first limit is set. At
67 * that point, future operations to create/destroy filesystems or snapshots
68 * will validate and update the counts.
70 * Because the count properties will not exist before the feature is active,
71 * the counts are updated when a limit is first set on an uninitialized
72 * dsl_dir node in the tree (The filesystem/snapshot count on a node includes
73 * all of the nested filesystems/snapshots. Thus, a new leaf node has a
74 * filesystem count of 0 and a snapshot count of 0. Non-existent filesystem and
75 * snapshot count properties on a node indicate uninitialized counts on that
76 * node.) When first setting a limit on an uninitialized node, the code starts
77 * at the filesystem with the new limit and descends into all sub-filesystems
78 * to add the count properties.
80 * In practice this is lightweight since a limit is typically set when the
81 * filesystem is created and thus has no children. Once valid, changing the
82 * limit value won't require a re-traversal since the counts are already valid.
83 * When recursively fixing the counts, if a node with a limit is encountered
84 * during the descent, the counts are known to be valid and there is no need to
85 * descend into that filesystem's children. The counts on filesystems above the
86 * one with the new limit will still be uninitialized, unless a limit is
87 * eventually set on one of those filesystems. The counts are always recursively
88 * updated when a limit is set on a dataset, unless there is already a limit.
89 * When a new limit value is set on a filesystem with an existing limit, it is
90 * possible for the new limit to be less than the current count at that level
91 * since a user who can change the limit is also allowed to exceed the limit.
93 * Once the feature is active, then whenever a filesystem or snapshot is
94 * created, the code recurses up the tree, validating the new count against the
95 * limit at each initialized level. In practice, most levels will not have a
96 * limit set. If there is a limit at any initialized level up the tree, the
97 * check must pass or the creation will fail. Likewise, when a filesystem or
98 * snapshot is destroyed, the counts are recursively adjusted all the way up
99 * the initizized nodes in the tree. Renaming a filesystem into different point
100 * in the tree will first validate, then update the counts on each branch up to
101 * the common ancestor. A receive will also validate the counts and then update
104 * An exception to the above behavior is that the limit is not enforced if the
105 * user has permission to modify the limit. This is primarily so that
106 * recursive snapshots in the global zone always work. We want to prevent a
107 * denial-of-service in which a lower level delegated dataset could max out its
108 * limit and thus block recursive snapshots from being taken in the global zone.
109 * Because of this, it is possible for the snapshot count to be over the limit
110 * and snapshots taken in the global zone could cause a lower level dataset to
111 * hit or exceed its limit. The administrator taking the global zone recursive
112 * snapshot should be aware of this side-effect and behave accordingly.
113 * For consistency, the filesystem limit is also not enforced if the user can
116 * The filesystem and snapshot limits are validated by dsl_fs_ss_limit_check()
117 * and updated by dsl_fs_ss_count_adjust(). A new limit value is setup in
118 * dsl_dir_activate_fs_ss_limit() and the counts are adjusted, if necessary, by
119 * dsl_dir_init_fs_ss_count().
121 * There is a special case when we receive a filesystem that already exists. In
122 * this case a temporary clone name of %X is created (see dmu_recv_begin). We
123 * never update the filesystem counts for temporary clones.
125 * Likewise, we do not update the snapshot counts for temporary snapshots,
126 * such as those created by zfs diff.
129 static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
133 dsl_dir_evict(dmu_buf_t *db, void *arg)
136 dsl_pool_t *dp = dd->dd_pool;
139 for (t = 0; t < TXG_SIZE; t++) {
140 ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t));
141 ASSERT(dd->dd_tempreserved[t] == 0);
142 ASSERT(dd->dd_space_towrite[t] == 0);
146 dsl_dir_rele(dd->dd_parent, dd);
148 spa_close(dd->dd_pool->dp_spa, dd);
151 * The props callback list should have been cleaned up by
154 list_destroy(&dd->dd_prop_cbs);
155 mutex_destroy(&dd->dd_lock);
156 kmem_free(dd, sizeof (dsl_dir_t));
160 dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj,
161 const char *tail, void *tag, dsl_dir_t **ddp)
167 ASSERT(dsl_pool_config_held(dp));
169 err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
172 dd = dmu_buf_get_user(dbuf);
175 dmu_object_info_t doi;
176 dmu_object_info_from_db(dbuf, &doi);
177 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_DSL_DIR);
178 ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
184 dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
185 dd->dd_object = ddobj;
188 dd->dd_phys = dbuf->db_data;
189 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
191 list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t),
192 offsetof(dsl_prop_cb_record_t, cbr_node));
194 dsl_dir_snap_cmtime_update(dd);
196 if (dd->dd_phys->dd_parent_obj) {
197 err = dsl_dir_hold_obj(dp, dd->dd_phys->dd_parent_obj,
198 NULL, dd, &dd->dd_parent);
205 err = zap_lookup(dp->dp_meta_objset,
206 dd->dd_parent->dd_phys->dd_child_dir_zapobj,
207 tail, 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 dd->dd_parent->dd_phys->dd_child_dir_zapobj,
214 ddobj, 0, dd->dd_myname);
219 (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
222 if (dsl_dir_is_clone(dd)) {
223 dmu_buf_t *origin_bonus;
224 dsl_dataset_phys_t *origin_phys;
227 * We can't open the origin dataset, because
228 * that would require opening this dsl_dir.
229 * Just look at its phys directly instead.
231 err = dmu_bonus_hold(dp->dp_meta_objset,
232 dd->dd_phys->dd_origin_obj, FTAG, &origin_bonus);
235 origin_phys = origin_bonus->db_data;
237 origin_phys->ds_creation_txg;
238 dmu_buf_rele(origin_bonus, FTAG);
241 winner = dmu_buf_set_user_ie(dbuf, dd, &dd->dd_phys,
245 dsl_dir_rele(dd->dd_parent, dd);
246 mutex_destroy(&dd->dd_lock);
247 kmem_free(dd, sizeof (dsl_dir_t));
250 spa_open_ref(dp->dp_spa, dd);
255 * The dsl_dir_t has both open-to-close and instantiate-to-evict
256 * holds on the spa. We need the open-to-close holds because
257 * otherwise the spa_refcnt wouldn't change when we open a
258 * dir which the spa also has open, so we could incorrectly
259 * think it was OK to unload/export/destroy the pool. We need
260 * the instantiate-to-evict hold because the dsl_dir_t has a
261 * pointer to the dd_pool, which has a pointer to the spa_t.
263 spa_open_ref(dp->dp_spa, tag);
264 ASSERT3P(dd->dd_pool, ==, dp);
265 ASSERT3U(dd->dd_object, ==, ddobj);
266 ASSERT3P(dd->dd_dbuf, ==, dbuf);
272 dsl_dir_rele(dd->dd_parent, dd);
273 mutex_destroy(&dd->dd_lock);
274 kmem_free(dd, sizeof (dsl_dir_t));
275 dmu_buf_rele(dbuf, tag);
280 dsl_dir_rele(dsl_dir_t *dd, void *tag)
282 dprintf_dd(dd, "%s\n", "");
283 spa_close(dd->dd_pool->dp_spa, tag);
284 dmu_buf_rele(dd->dd_dbuf, tag);
287 /* buf must be long enough (MAXNAMELEN + strlen(MOS_DIR_NAME) + 1 should do) */
289 dsl_dir_name(dsl_dir_t *dd, char *buf)
292 dsl_dir_name(dd->dd_parent, buf);
293 (void) strcat(buf, "/");
297 if (!MUTEX_HELD(&dd->dd_lock)) {
299 * recursive mutex so that we can use
300 * dprintf_dd() with dd_lock held
302 mutex_enter(&dd->dd_lock);
303 (void) strcat(buf, dd->dd_myname);
304 mutex_exit(&dd->dd_lock);
306 (void) strcat(buf, dd->dd_myname);
310 /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
312 dsl_dir_namelen(dsl_dir_t *dd)
317 /* parent's name + 1 for the "/" */
318 result = dsl_dir_namelen(dd->dd_parent) + 1;
321 if (!MUTEX_HELD(&dd->dd_lock)) {
322 /* see dsl_dir_name */
323 mutex_enter(&dd->dd_lock);
324 result += strlen(dd->dd_myname);
325 mutex_exit(&dd->dd_lock);
327 result += strlen(dd->dd_myname);
334 getcomponent(const char *path, char *component, const char **nextp)
338 if ((path == NULL) || (path[0] == '\0'))
339 return (SET_ERROR(ENOENT));
340 /* This would be a good place to reserve some namespace... */
341 p = strpbrk(path, "/@");
342 if (p && (p[1] == '/' || p[1] == '@')) {
343 /* two separators in a row */
344 return (SET_ERROR(EINVAL));
346 if (p == NULL || p == path) {
348 * if the first thing is an @ or /, it had better be an
349 * @ and it had better not have any more ats or slashes,
350 * and it had better have something after the @.
353 (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
354 return (SET_ERROR(EINVAL));
355 if (strlen(path) >= MAXNAMELEN)
356 return (SET_ERROR(ENAMETOOLONG));
357 (void) strcpy(component, path);
359 } else if (p[0] == '/') {
360 if (p - path >= MAXNAMELEN)
361 return (SET_ERROR(ENAMETOOLONG));
362 (void) strncpy(component, path, p - path);
363 component[p - path] = '\0';
365 } else if (p[0] == '@') {
367 * if the next separator is an @, there better not be
370 if (strchr(path, '/'))
371 return (SET_ERROR(EINVAL));
372 if (p - path >= MAXNAMELEN)
373 return (SET_ERROR(ENAMETOOLONG));
374 (void) strncpy(component, path, p - path);
375 component[p - path] = '\0';
377 panic("invalid p=%p", (void *)p);
384 * Return the dsl_dir_t, and possibly the last component which couldn't
385 * be found in *tail. The name must be in the specified dsl_pool_t. This
386 * thread must hold the dp_config_rwlock for the pool. Returns NULL if the
387 * path is bogus, or if tail==NULL and we couldn't parse the whole name.
388 * (*tail)[0] == '@' means that the last component is a snapshot.
391 dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag,
392 dsl_dir_t **ddp, const char **tailp)
394 char buf[MAXNAMELEN];
395 const char *spaname, *next, *nextnext = NULL;
400 err = getcomponent(name, buf, &next);
404 /* Make sure the name is in the specified pool. */
405 spaname = spa_name(dp->dp_spa);
406 if (strcmp(buf, spaname) != 0)
407 return (SET_ERROR(EINVAL));
409 ASSERT(dsl_pool_config_held(dp));
411 err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
416 while (next != NULL) {
418 err = getcomponent(next, buf, &nextnext);
421 ASSERT(next[0] != '\0');
424 dprintf("looking up %s in obj%lld\n",
425 buf, dd->dd_phys->dd_child_dir_zapobj);
427 err = zap_lookup(dp->dp_meta_objset,
428 dd->dd_phys->dd_child_dir_zapobj,
429 buf, sizeof (ddobj), 1, &ddobj);
436 err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_ds);
439 dsl_dir_rele(dd, tag);
445 dsl_dir_rele(dd, tag);
450 * It's an error if there's more than one component left, or
451 * tailp==NULL and there's any component left.
454 (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
456 dsl_dir_rele(dd, tag);
457 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
458 err = SET_ERROR(ENOENT);
467 * If the counts are already initialized for this filesystem and its
468 * descendants then do nothing, otherwise initialize the counts.
470 * The counts on this filesystem, and those below, may be uninitialized due to
471 * either the use of a pre-existing pool which did not support the
472 * filesystem/snapshot limit feature, or one in which the feature had not yet
475 * Recursively descend the filesystem tree and update the filesystem/snapshot
476 * counts on each filesystem below, then update the cumulative count on the
477 * current filesystem. If the filesystem already has a count set on it,
478 * then we know that its counts, and the counts on the filesystems below it,
479 * are already correct, so we don't have to update this filesystem.
482 dsl_dir_init_fs_ss_count(dsl_dir_t *dd, dmu_tx_t *tx)
484 uint64_t my_fs_cnt = 0;
485 uint64_t my_ss_cnt = 0;
486 dsl_pool_t *dp = dd->dd_pool;
487 objset_t *os = dp->dp_meta_objset;
492 ASSERT(spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT));
493 ASSERT(dsl_pool_config_held(dp));
494 ASSERT(dmu_tx_is_syncing(tx));
496 dsl_dir_zapify(dd, tx);
499 * If the filesystem count has already been initialized then we
500 * don't need to recurse down any further.
502 if (zap_contains(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT) == 0)
505 zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP);
506 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
508 /* Iterate my child dirs */
509 for (zap_cursor_init(zc, os, dd->dd_phys->dd_child_dir_zapobj);
510 zap_cursor_retrieve(zc, za) == 0; zap_cursor_advance(zc)) {
514 VERIFY0(dsl_dir_hold_obj(dp, za->za_first_integer, NULL, FTAG,
518 * Ignore hidden ($FREE, $MOS & $ORIGIN) objsets and
519 * temporary datasets.
521 if (chld_dd->dd_myname[0] == '$' ||
522 chld_dd->dd_myname[0] == '%') {
523 dsl_dir_rele(chld_dd, FTAG);
527 my_fs_cnt++; /* count this child */
529 dsl_dir_init_fs_ss_count(chld_dd, tx);
531 VERIFY0(zap_lookup(os, chld_dd->dd_object,
532 DD_FIELD_FILESYSTEM_COUNT, sizeof (count), 1, &count));
534 VERIFY0(zap_lookup(os, chld_dd->dd_object,
535 DD_FIELD_SNAPSHOT_COUNT, sizeof (count), 1, &count));
538 dsl_dir_rele(chld_dd, FTAG);
541 /* Count my snapshots (we counted children's snapshots above) */
542 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
543 dd->dd_phys->dd_head_dataset_obj, FTAG, &ds));
545 for (zap_cursor_init(zc, os, ds->ds_phys->ds_snapnames_zapobj);
546 zap_cursor_retrieve(zc, za) == 0;
547 zap_cursor_advance(zc)) {
548 /* Don't count temporary snapshots */
549 if (za->za_name[0] != '%')
553 dsl_dataset_rele(ds, FTAG);
555 kmem_free(zc, sizeof (zap_cursor_t));
556 kmem_free(za, sizeof (zap_attribute_t));
558 /* we're in a sync task, update counts */
559 dmu_buf_will_dirty(dd->dd_dbuf, tx);
560 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
561 sizeof (my_fs_cnt), 1, &my_fs_cnt, tx));
562 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
563 sizeof (my_ss_cnt), 1, &my_ss_cnt, tx));
567 dsl_dir_actv_fs_ss_limit_check(void *arg, dmu_tx_t *tx)
569 char *ddname = (char *)arg;
570 dsl_pool_t *dp = dmu_tx_pool(tx);
575 error = dsl_dataset_hold(dp, ddname, FTAG, &ds);
579 if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
580 dsl_dataset_rele(ds, FTAG);
581 return (SET_ERROR(ENOTSUP));
585 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT) &&
586 dsl_dir_is_zapified(dd) &&
587 zap_contains(dp->dp_meta_objset, dd->dd_object,
588 DD_FIELD_FILESYSTEM_COUNT) == 0) {
589 dsl_dataset_rele(ds, FTAG);
590 return (SET_ERROR(EALREADY));
593 dsl_dataset_rele(ds, FTAG);
598 dsl_dir_actv_fs_ss_limit_sync(void *arg, dmu_tx_t *tx)
600 char *ddname = (char *)arg;
601 dsl_pool_t *dp = dmu_tx_pool(tx);
605 VERIFY0(dsl_dataset_hold(dp, ddname, FTAG, &ds));
607 spa = dsl_dataset_get_spa(ds);
609 if (!spa_feature_is_active(spa, SPA_FEATURE_FS_SS_LIMIT)) {
611 * Since the feature was not active and we're now setting a
612 * limit, increment the feature-active counter so that the
613 * feature becomes active for the first time.
615 * We are already in a sync task so we can update the MOS.
617 spa_feature_incr(spa, SPA_FEATURE_FS_SS_LIMIT, tx);
621 * Since we are now setting a non-UINT64_MAX limit on the filesystem,
622 * we need to ensure the counts are correct. Descend down the tree from
623 * this point and update all of the counts to be accurate.
625 dsl_dir_init_fs_ss_count(ds->ds_dir, tx);
627 dsl_dataset_rele(ds, FTAG);
631 * Make sure the feature is enabled and activate it if necessary.
632 * Since we're setting a limit, ensure the on-disk counts are valid.
633 * This is only called by the ioctl path when setting a limit value.
635 * We do not need to validate the new limit, since users who can change the
636 * limit are also allowed to exceed the limit.
639 dsl_dir_activate_fs_ss_limit(const char *ddname)
643 error = dsl_sync_task(ddname, dsl_dir_actv_fs_ss_limit_check,
644 dsl_dir_actv_fs_ss_limit_sync, (void *)ddname, 0);
646 if (error == EALREADY)
653 * Used to determine if the filesystem_limit or snapshot_limit should be
654 * enforced. We allow the limit to be exceeded if the user has permission to
655 * write the property value. We pass in the creds that we got in the open
656 * context since we will always be the GZ root in syncing context. We also have
657 * to handle the case where we are allowed to change the limit on the current
658 * dataset, but there may be another limit in the tree above.
660 * We can never modify these two properties within a non-global zone. In
661 * addition, the other checks are modeled on zfs_secpolicy_write_perms. We
662 * can't use that function since we are already holding the dp_config_rwlock.
663 * In addition, we already have the dd and dealing with snapshots is simplified
674 dsl_enforce_ds_ss_limits(dsl_dir_t *dd, zfs_prop_t prop, cred_t *cr)
676 enforce_res_t enforce = ENFORCE_ALWAYS;
681 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
682 prop == ZFS_PROP_SNAPSHOT_LIMIT);
688 if (crgetzoneid(cr) != GLOBAL_ZONEID)
690 return (ENFORCE_ALWAYS);
692 if (secpolicy_zfs(cr) == 0)
693 return (ENFORCE_NEVER);
696 if ((obj = dd->dd_phys->dd_head_dataset_obj) == 0)
697 return (ENFORCE_ALWAYS);
699 ASSERT(dsl_pool_config_held(dd->dd_pool));
701 if (dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds) != 0)
702 return (ENFORCE_ALWAYS);
704 if (dsl_prop_get_ds(ds, "zoned", 8, 1, &zoned, NULL) || zoned) {
705 /* Only root can access zoned fs's from the GZ */
706 enforce = ENFORCE_ALWAYS;
708 if (dsl_deleg_access_impl(ds, zfs_prop_to_name(prop), cr) == 0)
709 enforce = ENFORCE_ABOVE;
712 dsl_dataset_rele(ds, FTAG);
717 * Check if adding additional child filesystem(s) would exceed any filesystem
718 * limits or adding additional snapshot(s) would exceed any snapshot limits.
719 * The prop argument indicates which limit to check.
721 * Note that all filesystem limits up to the root (or the highest
722 * initialized) filesystem or the given ancestor must be satisfied.
725 dsl_fs_ss_limit_check(dsl_dir_t *dd, uint64_t delta, zfs_prop_t prop,
726 dsl_dir_t *ancestor, cred_t *cr)
728 objset_t *os = dd->dd_pool->dp_meta_objset;
729 uint64_t limit, count;
731 enforce_res_t enforce;
734 ASSERT(dsl_pool_config_held(dd->dd_pool));
735 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
736 prop == ZFS_PROP_SNAPSHOT_LIMIT);
739 * If we're allowed to change the limit, don't enforce the limit
740 * e.g. this can happen if a snapshot is taken by an administrative
741 * user in the global zone (i.e. a recursive snapshot by root).
742 * However, we must handle the case of delegated permissions where we
743 * are allowed to change the limit on the current dataset, but there
744 * is another limit in the tree above.
746 enforce = dsl_enforce_ds_ss_limits(dd, prop, cr);
747 if (enforce == ENFORCE_NEVER)
751 * e.g. if renaming a dataset with no snapshots, count adjustment
757 if (prop == ZFS_PROP_SNAPSHOT_LIMIT) {
759 * We don't enforce the limit for temporary snapshots. This is
760 * indicated by a NULL cred_t argument.
765 count_prop = DD_FIELD_SNAPSHOT_COUNT;
767 count_prop = DD_FIELD_FILESYSTEM_COUNT;
771 * If an ancestor has been provided, stop checking the limit once we
772 * hit that dir. We need this during rename so that we don't overcount
773 * the check once we recurse up to the common ancestor.
779 * If we hit an uninitialized node while recursing up the tree, we can
780 * stop since we know there is no limit here (or above). The counts are
781 * not valid on this node and we know we won't touch this node's counts.
783 if (!dsl_dir_is_zapified(dd) || zap_lookup(os, dd->dd_object,
784 count_prop, sizeof (count), 1, &count) == ENOENT)
787 err = dsl_prop_get_dd(dd, zfs_prop_to_name(prop), 8, 1, &limit, NULL,
792 /* Is there a limit which we've hit? */
793 if (enforce == ENFORCE_ALWAYS && (count + delta) > limit)
794 return (SET_ERROR(EDQUOT));
796 if (dd->dd_parent != NULL)
797 err = dsl_fs_ss_limit_check(dd->dd_parent, delta, prop,
804 * Adjust the filesystem or snapshot count for the specified dsl_dir_t and all
805 * parents. When a new filesystem/snapshot is created, increment the count on
806 * all parents, and when a filesystem/snapshot is destroyed, decrement the
810 dsl_fs_ss_count_adjust(dsl_dir_t *dd, int64_t delta, const char *prop,
814 objset_t *os = dd->dd_pool->dp_meta_objset;
817 ASSERT(dsl_pool_config_held(dd->dd_pool));
818 ASSERT(dmu_tx_is_syncing(tx));
819 ASSERT(strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0 ||
820 strcmp(prop, DD_FIELD_SNAPSHOT_COUNT) == 0);
823 * When we receive an incremental stream into a filesystem that already
824 * exists, a temporary clone is created. We don't count this temporary
825 * clone, whose name begins with a '%'. We also ignore hidden ($FREE,
826 * $MOS & $ORIGIN) objsets.
828 if ((dd->dd_myname[0] == '%' || dd->dd_myname[0] == '$') &&
829 strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0)
833 * e.g. if renaming a dataset with no snapshots, count adjustment is 0
839 * If we hit an uninitialized node while recursing up the tree, we can
840 * stop since we know the counts are not valid on this node and we
841 * know we shouldn't touch this node's counts. An uninitialized count
842 * on the node indicates that either the feature has not yet been
843 * activated or there are no limits on this part of the tree.
845 if (!dsl_dir_is_zapified(dd) || (err = zap_lookup(os, dd->dd_object,
846 prop, sizeof (count), 1, &count)) == ENOENT)
851 /* Use a signed verify to make sure we're not neg. */
852 VERIFY3S(count, >=, 0);
854 VERIFY0(zap_update(os, dd->dd_object, prop, sizeof (count), 1, &count,
857 /* Roll up this additional count into our ancestors */
858 if (dd->dd_parent != NULL)
859 dsl_fs_ss_count_adjust(dd->dd_parent, delta, prop, tx);
863 dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
866 objset_t *mos = dp->dp_meta_objset;
868 dsl_dir_phys_t *ddphys;
871 ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
872 DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
874 VERIFY(0 == zap_add(mos, pds->dd_phys->dd_child_dir_zapobj,
875 name, sizeof (uint64_t), 1, &ddobj, tx));
877 /* it's the root dir */
878 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
879 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
881 VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
882 dmu_buf_will_dirty(dbuf, tx);
883 ddphys = dbuf->db_data;
885 ddphys->dd_creation_time = gethrestime_sec();
887 ddphys->dd_parent_obj = pds->dd_object;
889 /* update the filesystem counts */
890 dsl_fs_ss_count_adjust(pds, 1, DD_FIELD_FILESYSTEM_COUNT, tx);
892 ddphys->dd_props_zapobj = zap_create(mos,
893 DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
894 ddphys->dd_child_dir_zapobj = zap_create(mos,
895 DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
896 if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
897 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
898 dmu_buf_rele(dbuf, FTAG);
904 dsl_dir_is_clone(dsl_dir_t *dd)
906 return (dd->dd_phys->dd_origin_obj &&
907 (dd->dd_pool->dp_origin_snap == NULL ||
908 dd->dd_phys->dd_origin_obj !=
909 dd->dd_pool->dp_origin_snap->ds_object));
913 dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
915 mutex_enter(&dd->dd_lock);
916 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
917 dd->dd_phys->dd_used_bytes);
918 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, dd->dd_phys->dd_quota);
919 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
920 dd->dd_phys->dd_reserved);
921 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
922 dd->dd_phys->dd_compressed_bytes == 0 ? 100 :
923 (dd->dd_phys->dd_uncompressed_bytes * 100 /
924 dd->dd_phys->dd_compressed_bytes));
925 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED,
926 dd->dd_phys->dd_uncompressed_bytes);
927 if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
928 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
929 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]);
930 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
931 dd->dd_phys->dd_used_breakdown[DD_USED_HEAD]);
932 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
933 dd->dd_phys->dd_used_breakdown[DD_USED_REFRSRV]);
934 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
935 dd->dd_phys->dd_used_breakdown[DD_USED_CHILD] +
936 dd->dd_phys->dd_used_breakdown[DD_USED_CHILD_RSRV]);
938 mutex_exit(&dd->dd_lock);
940 if (dsl_dir_is_zapified(dd)) {
942 objset_t *os = dd->dd_pool->dp_meta_objset;
944 if (zap_lookup(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
945 sizeof (count), 1, &count) == 0) {
946 dsl_prop_nvlist_add_uint64(nv,
947 ZFS_PROP_FILESYSTEM_COUNT, count);
949 if (zap_lookup(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
950 sizeof (count), 1, &count) == 0) {
951 dsl_prop_nvlist_add_uint64(nv,
952 ZFS_PROP_SNAPSHOT_COUNT, count);
956 if (dsl_dir_is_clone(dd)) {
958 char buf[MAXNAMELEN];
960 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
961 dd->dd_phys->dd_origin_obj, FTAG, &ds));
962 dsl_dataset_name(ds, buf);
963 dsl_dataset_rele(ds, FTAG);
964 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
969 dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
971 dsl_pool_t *dp = dd->dd_pool;
975 if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) {
976 /* up the hold count until we can be written out */
977 dmu_buf_add_ref(dd->dd_dbuf, dd);
982 parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
984 uint64_t old_accounted = MAX(used, dd->dd_phys->dd_reserved);
985 uint64_t new_accounted = MAX(used + delta, dd->dd_phys->dd_reserved);
986 return (new_accounted - old_accounted);
990 dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
992 ASSERT(dmu_tx_is_syncing(tx));
994 mutex_enter(&dd->dd_lock);
995 ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]);
996 dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
997 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
998 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
999 mutex_exit(&dd->dd_lock);
1001 /* release the hold from dsl_dir_dirty */
1002 dmu_buf_rele(dd->dd_dbuf, dd);
1006 dsl_dir_space_towrite(dsl_dir_t *dd)
1011 ASSERT(MUTEX_HELD(&dd->dd_lock));
1013 for (i = 0; i < TXG_SIZE; i++) {
1014 space += dd->dd_space_towrite[i&TXG_MASK];
1015 ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0);
1021 * How much space would dd have available if ancestor had delta applied
1022 * to it? If ondiskonly is set, we're only interested in what's
1023 * on-disk, not estimated pending changes.
1026 dsl_dir_space_available(dsl_dir_t *dd,
1027 dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
1029 uint64_t parentspace, myspace, quota, used;
1032 * If there are no restrictions otherwise, assume we have
1033 * unlimited space available.
1036 parentspace = UINT64_MAX;
1038 if (dd->dd_parent != NULL) {
1039 parentspace = dsl_dir_space_available(dd->dd_parent,
1040 ancestor, delta, ondiskonly);
1043 mutex_enter(&dd->dd_lock);
1044 if (dd->dd_phys->dd_quota != 0)
1045 quota = dd->dd_phys->dd_quota;
1046 used = dd->dd_phys->dd_used_bytes;
1048 used += dsl_dir_space_towrite(dd);
1050 if (dd->dd_parent == NULL) {
1051 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
1052 quota = MIN(quota, poolsize);
1055 if (dd->dd_phys->dd_reserved > used && parentspace != UINT64_MAX) {
1057 * We have some space reserved, in addition to what our
1060 parentspace += dd->dd_phys->dd_reserved - used;
1063 if (dd == ancestor) {
1065 ASSERT(used >= -delta);
1067 if (parentspace != UINT64_MAX)
1068 parentspace -= delta;
1076 * the lesser of the space provided by our parent and
1077 * the space left in our quota
1079 myspace = MIN(parentspace, quota - used);
1082 mutex_exit(&dd->dd_lock);
1087 struct tempreserve {
1088 list_node_t tr_node;
1094 dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
1095 boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list,
1096 dmu_tx_t *tx, boolean_t first)
1098 uint64_t txg = tx->tx_txg;
1099 uint64_t est_inflight, used_on_disk, quota, parent_rsrv;
1100 uint64_t deferred = 0;
1101 struct tempreserve *tr;
1102 int retval = EDQUOT;
1103 int txgidx = txg & TXG_MASK;
1105 uint64_t ref_rsrv = 0;
1107 ASSERT3U(txg, !=, 0);
1108 ASSERT3S(asize, >, 0);
1110 mutex_enter(&dd->dd_lock);
1113 * Check against the dsl_dir's quota. We don't add in the delta
1114 * when checking for over-quota because they get one free hit.
1116 est_inflight = dsl_dir_space_towrite(dd);
1117 for (i = 0; i < TXG_SIZE; i++)
1118 est_inflight += dd->dd_tempreserved[i];
1119 used_on_disk = dd->dd_phys->dd_used_bytes;
1122 * On the first iteration, fetch the dataset's used-on-disk and
1123 * refreservation values. Also, if checkrefquota is set, test if
1124 * allocating this space would exceed the dataset's refquota.
1126 if (first && tx->tx_objset) {
1128 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
1130 error = dsl_dataset_check_quota(ds, checkrefquota,
1131 asize, est_inflight, &used_on_disk, &ref_rsrv);
1133 mutex_exit(&dd->dd_lock);
1139 * If this transaction will result in a net free of space,
1140 * we want to let it through.
1142 if (ignorequota || netfree || dd->dd_phys->dd_quota == 0)
1145 quota = dd->dd_phys->dd_quota;
1148 * Adjust the quota against the actual pool size at the root
1149 * minus any outstanding deferred frees.
1150 * To ensure that it's possible to remove files from a full
1151 * pool without inducing transient overcommits, we throttle
1152 * netfree transactions against a quota that is slightly larger,
1153 * but still within the pool's allocation slop. In cases where
1154 * we're very close to full, this will allow a steady trickle of
1155 * removes to get through.
1157 if (dd->dd_parent == NULL) {
1158 spa_t *spa = dd->dd_pool->dp_spa;
1159 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
1160 deferred = metaslab_class_get_deferred(spa_normal_class(spa));
1161 if (poolsize - deferred < quota) {
1162 quota = poolsize - deferred;
1168 * If they are requesting more space, and our current estimate
1169 * is over quota, they get to try again unless the actual
1170 * on-disk is over quota and there are no pending changes (which
1171 * may free up space for us).
1173 if (used_on_disk + est_inflight >= quota) {
1174 if (est_inflight > 0 || used_on_disk < quota ||
1175 (retval == ENOSPC && used_on_disk < quota + deferred))
1177 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
1178 "quota=%lluK tr=%lluK err=%d\n",
1179 used_on_disk>>10, est_inflight>>10,
1180 quota>>10, asize>>10, retval);
1181 mutex_exit(&dd->dd_lock);
1182 return (SET_ERROR(retval));
1185 /* We need to up our estimated delta before dropping dd_lock */
1186 dd->dd_tempreserved[txgidx] += asize;
1188 parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
1190 mutex_exit(&dd->dd_lock);
1192 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1194 tr->tr_size = asize;
1195 list_insert_tail(tr_list, tr);
1197 /* see if it's OK with our parent */
1198 if (dd->dd_parent && parent_rsrv) {
1199 boolean_t ismos = (dd->dd_phys->dd_head_dataset_obj == 0);
1201 return (dsl_dir_tempreserve_impl(dd->dd_parent,
1202 parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE));
1209 * Reserve space in this dsl_dir, to be used in this tx's txg.
1210 * After the space has been dirtied (and dsl_dir_willuse_space()
1211 * has been called), the reservation should be canceled, using
1212 * dsl_dir_tempreserve_clear().
1215 dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
1216 uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx)
1226 tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
1227 list_create(tr_list, sizeof (struct tempreserve),
1228 offsetof(struct tempreserve, tr_node));
1229 ASSERT3S(asize, >, 0);
1230 ASSERT3S(fsize, >=, 0);
1232 err = arc_tempreserve_space(lsize, tx->tx_txg);
1234 struct tempreserve *tr;
1236 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1237 tr->tr_size = lsize;
1238 list_insert_tail(tr_list, tr);
1240 if (err == EAGAIN) {
1242 * If arc_memory_throttle() detected that pageout
1243 * is running and we are low on memory, we delay new
1244 * non-pageout transactions to give pageout an
1247 * It is unfortunate to be delaying while the caller's
1250 txg_delay(dd->dd_pool, tx->tx_txg,
1251 MSEC2NSEC(10), MSEC2NSEC(10));
1252 err = SET_ERROR(ERESTART);
1257 err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize,
1258 FALSE, asize > usize, tr_list, tx, TRUE);
1262 dsl_dir_tempreserve_clear(tr_list, tx);
1264 *tr_cookiep = tr_list;
1270 * Clear a temporary reservation that we previously made with
1271 * dsl_dir_tempreserve_space().
1274 dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
1276 int txgidx = tx->tx_txg & TXG_MASK;
1277 list_t *tr_list = tr_cookie;
1278 struct tempreserve *tr;
1280 ASSERT3U(tx->tx_txg, !=, 0);
1282 if (tr_cookie == NULL)
1285 while ((tr = list_head(tr_list)) != NULL) {
1287 mutex_enter(&tr->tr_ds->dd_lock);
1288 ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
1290 tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
1291 mutex_exit(&tr->tr_ds->dd_lock);
1293 arc_tempreserve_clear(tr->tr_size);
1295 list_remove(tr_list, tr);
1296 kmem_free(tr, sizeof (struct tempreserve));
1299 kmem_free(tr_list, sizeof (list_t));
1303 * This should be called from open context when we think we're going to write
1304 * or free space, for example when dirtying data. Be conservative; it's okay
1305 * to write less space or free more, but we don't want to write more or free
1306 * less than the amount specified.
1309 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
1311 int64_t parent_space;
1314 mutex_enter(&dd->dd_lock);
1316 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
1318 est_used = dsl_dir_space_towrite(dd) + dd->dd_phys->dd_used_bytes;
1319 parent_space = parent_delta(dd, est_used, space);
1320 mutex_exit(&dd->dd_lock);
1322 /* Make sure that we clean up dd_space_to* */
1323 dsl_dir_dirty(dd, tx);
1325 /* XXX this is potentially expensive and unnecessary... */
1326 if (parent_space && dd->dd_parent)
1327 dsl_dir_willuse_space(dd->dd_parent, parent_space, tx);
1330 /* call from syncing context when we actually write/free space for this dd */
1332 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
1333 int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
1335 int64_t accounted_delta;
1338 * dsl_dataset_set_refreservation_sync_impl() calls this with
1339 * dd_lock held, so that it can atomically update
1340 * ds->ds_reserved and the dsl_dir accounting, so that
1341 * dsl_dataset_check_quota() can see dataset and dir accounting
1344 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
1346 ASSERT(dmu_tx_is_syncing(tx));
1347 ASSERT(type < DD_USED_NUM);
1349 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1352 mutex_enter(&dd->dd_lock);
1353 accounted_delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, used);
1354 ASSERT(used >= 0 || dd->dd_phys->dd_used_bytes >= -used);
1355 ASSERT(compressed >= 0 ||
1356 dd->dd_phys->dd_compressed_bytes >= -compressed);
1357 ASSERT(uncompressed >= 0 ||
1358 dd->dd_phys->dd_uncompressed_bytes >= -uncompressed);
1359 dd->dd_phys->dd_used_bytes += used;
1360 dd->dd_phys->dd_uncompressed_bytes += uncompressed;
1361 dd->dd_phys->dd_compressed_bytes += compressed;
1363 if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1365 dd->dd_phys->dd_used_breakdown[type] >= -used);
1366 dd->dd_phys->dd_used_breakdown[type] += used;
1370 for (t = 0; t < DD_USED_NUM; t++)
1371 u += dd->dd_phys->dd_used_breakdown[t];
1372 ASSERT3U(u, ==, dd->dd_phys->dd_used_bytes);
1376 mutex_exit(&dd->dd_lock);
1378 if (dd->dd_parent != NULL) {
1379 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1380 accounted_delta, compressed, uncompressed, tx);
1381 dsl_dir_transfer_space(dd->dd_parent,
1382 used - accounted_delta,
1383 DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
1388 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
1389 dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
1391 ASSERT(dmu_tx_is_syncing(tx));
1392 ASSERT(oldtype < DD_USED_NUM);
1393 ASSERT(newtype < DD_USED_NUM);
1395 if (delta == 0 || !(dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN))
1398 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1399 mutex_enter(&dd->dd_lock);
1401 dd->dd_phys->dd_used_breakdown[oldtype] >= delta :
1402 dd->dd_phys->dd_used_breakdown[newtype] >= -delta);
1403 ASSERT(dd->dd_phys->dd_used_bytes >= ABS(delta));
1404 dd->dd_phys->dd_used_breakdown[oldtype] -= delta;
1405 dd->dd_phys->dd_used_breakdown[newtype] += delta;
1406 mutex_exit(&dd->dd_lock);
1409 typedef struct dsl_dir_set_qr_arg {
1410 const char *ddsqra_name;
1411 zprop_source_t ddsqra_source;
1412 uint64_t ddsqra_value;
1413 } dsl_dir_set_qr_arg_t;
1416 dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx)
1418 dsl_dir_set_qr_arg_t *ddsqra = arg;
1419 dsl_pool_t *dp = dmu_tx_pool(tx);
1422 uint64_t towrite, newval;
1424 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1428 error = dsl_prop_predict(ds->ds_dir, "quota",
1429 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1431 dsl_dataset_rele(ds, FTAG);
1436 dsl_dataset_rele(ds, FTAG);
1440 mutex_enter(&ds->ds_dir->dd_lock);
1442 * If we are doing the preliminary check in open context, and
1443 * there are pending changes, then don't fail it, since the
1444 * pending changes could under-estimate the amount of space to be
1447 towrite = dsl_dir_space_towrite(ds->ds_dir);
1448 if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
1449 (newval < ds->ds_dir->dd_phys->dd_reserved ||
1450 newval < ds->ds_dir->dd_phys->dd_used_bytes + towrite)) {
1451 error = SET_ERROR(ENOSPC);
1453 mutex_exit(&ds->ds_dir->dd_lock);
1454 dsl_dataset_rele(ds, FTAG);
1459 dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx)
1461 dsl_dir_set_qr_arg_t *ddsqra = arg;
1462 dsl_pool_t *dp = dmu_tx_pool(tx);
1466 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1468 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1469 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA),
1470 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1471 &ddsqra->ddsqra_value, tx);
1473 VERIFY0(dsl_prop_get_int_ds(ds,
1474 zfs_prop_to_name(ZFS_PROP_QUOTA), &newval));
1476 newval = ddsqra->ddsqra_value;
1477 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1478 zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval);
1481 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1482 mutex_enter(&ds->ds_dir->dd_lock);
1483 ds->ds_dir->dd_phys->dd_quota = newval;
1484 mutex_exit(&ds->ds_dir->dd_lock);
1485 dsl_dataset_rele(ds, FTAG);
1489 dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
1491 dsl_dir_set_qr_arg_t ddsqra;
1493 ddsqra.ddsqra_name = ddname;
1494 ddsqra.ddsqra_source = source;
1495 ddsqra.ddsqra_value = quota;
1497 return (dsl_sync_task(ddname, dsl_dir_set_quota_check,
1498 dsl_dir_set_quota_sync, &ddsqra, 0));
1502 dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx)
1504 dsl_dir_set_qr_arg_t *ddsqra = arg;
1505 dsl_pool_t *dp = dmu_tx_pool(tx);
1508 uint64_t newval, used, avail;
1511 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1517 * If we are doing the preliminary check in open context, the
1518 * space estimates may be inaccurate.
1520 if (!dmu_tx_is_syncing(tx)) {
1521 dsl_dataset_rele(ds, FTAG);
1525 error = dsl_prop_predict(ds->ds_dir,
1526 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1527 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1529 dsl_dataset_rele(ds, FTAG);
1533 mutex_enter(&dd->dd_lock);
1534 used = dd->dd_phys->dd_used_bytes;
1535 mutex_exit(&dd->dd_lock);
1537 if (dd->dd_parent) {
1538 avail = dsl_dir_space_available(dd->dd_parent,
1541 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
1544 if (MAX(used, newval) > MAX(used, dd->dd_phys->dd_reserved)) {
1545 uint64_t delta = MAX(used, newval) -
1546 MAX(used, dd->dd_phys->dd_reserved);
1548 if (delta > avail ||
1549 (dd->dd_phys->dd_quota > 0 &&
1550 newval > dd->dd_phys->dd_quota))
1551 error = SET_ERROR(ENOSPC);
1554 dsl_dataset_rele(ds, FTAG);
1559 dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
1564 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1566 mutex_enter(&dd->dd_lock);
1567 used = dd->dd_phys->dd_used_bytes;
1568 delta = MAX(used, value) - MAX(used, dd->dd_phys->dd_reserved);
1569 dd->dd_phys->dd_reserved = value;
1571 if (dd->dd_parent != NULL) {
1572 /* Roll up this additional usage into our ancestors */
1573 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1576 mutex_exit(&dd->dd_lock);
1580 dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx)
1582 dsl_dir_set_qr_arg_t *ddsqra = arg;
1583 dsl_pool_t *dp = dmu_tx_pool(tx);
1587 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1589 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1590 dsl_prop_set_sync_impl(ds,
1591 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1592 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1593 &ddsqra->ddsqra_value, tx);
1595 VERIFY0(dsl_prop_get_int_ds(ds,
1596 zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval));
1598 newval = ddsqra->ddsqra_value;
1599 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1600 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1601 (longlong_t)newval);
1604 dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx);
1605 dsl_dataset_rele(ds, FTAG);
1609 dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1610 uint64_t reservation)
1612 dsl_dir_set_qr_arg_t ddsqra;
1614 ddsqra.ddsqra_name = ddname;
1615 ddsqra.ddsqra_source = source;
1616 ddsqra.ddsqra_value = reservation;
1618 return (dsl_sync_task(ddname, dsl_dir_set_reservation_check,
1619 dsl_dir_set_reservation_sync, &ddsqra, 0));
1623 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1625 for (; ds1; ds1 = ds1->dd_parent) {
1627 for (dd = ds2; dd; dd = dd->dd_parent) {
1636 * If delta is applied to dd, how much of that delta would be applied to
1637 * ancestor? Syncing context only.
1640 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1645 mutex_enter(&dd->dd_lock);
1646 delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, delta);
1647 mutex_exit(&dd->dd_lock);
1648 return (would_change(dd->dd_parent, delta, ancestor));
1651 typedef struct dsl_dir_rename_arg {
1652 const char *ddra_oldname;
1653 const char *ddra_newname;
1655 } dsl_dir_rename_arg_t;
1659 dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1662 char namebuf[MAXNAMELEN];
1664 dsl_dataset_name(ds, namebuf);
1666 if (strlen(namebuf) + *deltap >= MAXNAMELEN)
1667 return (SET_ERROR(ENAMETOOLONG));
1672 dsl_dir_rename_check(void *arg, dmu_tx_t *tx)
1674 dsl_dir_rename_arg_t *ddra = arg;
1675 dsl_pool_t *dp = dmu_tx_pool(tx);
1676 dsl_dir_t *dd, *newparent;
1677 const char *mynewname;
1679 int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname);
1681 /* target dir should exist */
1682 error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL);
1686 /* new parent should exist */
1687 error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG,
1688 &newparent, &mynewname);
1690 dsl_dir_rele(dd, FTAG);
1694 /* can't rename to different pool */
1695 if (dd->dd_pool != newparent->dd_pool) {
1696 dsl_dir_rele(newparent, FTAG);
1697 dsl_dir_rele(dd, FTAG);
1698 return (SET_ERROR(ENXIO));
1701 /* new name should not already exist */
1702 if (mynewname == NULL) {
1703 dsl_dir_rele(newparent, FTAG);
1704 dsl_dir_rele(dd, FTAG);
1705 return (SET_ERROR(EEXIST));
1708 /* if the name length is growing, validate child name lengths */
1710 error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename,
1711 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1713 dsl_dir_rele(newparent, FTAG);
1714 dsl_dir_rele(dd, FTAG);
1719 if (dmu_tx_is_syncing(tx)) {
1720 if (spa_feature_is_enabled(dp->dp_spa,
1721 SPA_FEATURE_FS_SS_LIMIT)) {
1723 * Although this is the check function and we don't
1724 * normally make on-disk changes in check functions,
1725 * we need to do that here.
1727 * Ensure this portion of the tree's counts have been
1728 * initialized in case the new parent has limits set.
1730 dsl_dir_init_fs_ss_count(dd, tx);
1734 if (newparent != dd->dd_parent) {
1735 /* is there enough space? */
1737 MAX(dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_reserved);
1738 objset_t *os = dd->dd_pool->dp_meta_objset;
1739 uint64_t fs_cnt = 0;
1740 uint64_t ss_cnt = 0;
1742 if (dsl_dir_is_zapified(dd)) {
1745 err = zap_lookup(os, dd->dd_object,
1746 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1748 if (err != ENOENT && err != 0)
1752 * have to add 1 for the filesystem itself that we're
1757 err = zap_lookup(os, dd->dd_object,
1758 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1760 if (err != ENOENT && err != 0)
1764 /* no rename into our descendant */
1765 if (closest_common_ancestor(dd, newparent) == dd) {
1766 dsl_dir_rele(newparent, FTAG);
1767 dsl_dir_rele(dd, FTAG);
1768 return (SET_ERROR(EINVAL));
1771 error = dsl_dir_transfer_possible(dd->dd_parent,
1772 newparent, fs_cnt, ss_cnt, myspace, ddra->ddra_cred);
1774 dsl_dir_rele(newparent, FTAG);
1775 dsl_dir_rele(dd, FTAG);
1780 dsl_dir_rele(newparent, FTAG);
1781 dsl_dir_rele(dd, FTAG);
1786 dsl_dir_rename_sync(void *arg, dmu_tx_t *tx)
1788 dsl_dir_rename_arg_t *ddra = arg;
1789 dsl_pool_t *dp = dmu_tx_pool(tx);
1790 dsl_dir_t *dd, *newparent;
1791 const char *mynewname;
1793 objset_t *mos = dp->dp_meta_objset;
1795 VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL));
1796 VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
1799 /* Log this before we change the name. */
1800 spa_history_log_internal_dd(dd, "rename", tx,
1801 "-> %s", ddra->ddra_newname);
1803 if (newparent != dd->dd_parent) {
1804 objset_t *os = dd->dd_pool->dp_meta_objset;
1805 uint64_t fs_cnt = 0;
1806 uint64_t ss_cnt = 0;
1809 * We already made sure the dd counts were initialized in the
1812 if (spa_feature_is_enabled(dp->dp_spa,
1813 SPA_FEATURE_FS_SS_LIMIT)) {
1814 VERIFY0(zap_lookup(os, dd->dd_object,
1815 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1817 /* add 1 for the filesystem itself that we're moving */
1820 VERIFY0(zap_lookup(os, dd->dd_object,
1821 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1825 dsl_fs_ss_count_adjust(dd->dd_parent, -fs_cnt,
1826 DD_FIELD_FILESYSTEM_COUNT, tx);
1827 dsl_fs_ss_count_adjust(newparent, fs_cnt,
1828 DD_FIELD_FILESYSTEM_COUNT, tx);
1830 dsl_fs_ss_count_adjust(dd->dd_parent, -ss_cnt,
1831 DD_FIELD_SNAPSHOT_COUNT, tx);
1832 dsl_fs_ss_count_adjust(newparent, ss_cnt,
1833 DD_FIELD_SNAPSHOT_COUNT, tx);
1835 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1836 -dd->dd_phys->dd_used_bytes,
1837 -dd->dd_phys->dd_compressed_bytes,
1838 -dd->dd_phys->dd_uncompressed_bytes, tx);
1839 dsl_dir_diduse_space(newparent, DD_USED_CHILD,
1840 dd->dd_phys->dd_used_bytes,
1841 dd->dd_phys->dd_compressed_bytes,
1842 dd->dd_phys->dd_uncompressed_bytes, tx);
1844 if (dd->dd_phys->dd_reserved > dd->dd_phys->dd_used_bytes) {
1845 uint64_t unused_rsrv = dd->dd_phys->dd_reserved -
1846 dd->dd_phys->dd_used_bytes;
1848 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1849 -unused_rsrv, 0, 0, tx);
1850 dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV,
1851 unused_rsrv, 0, 0, tx);
1855 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1857 /* remove from old parent zapobj */
1858 error = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj,
1862 (void) strcpy(dd->dd_myname, mynewname);
1863 dsl_dir_rele(dd->dd_parent, dd);
1864 dd->dd_phys->dd_parent_obj = newparent->dd_object;
1865 VERIFY0(dsl_dir_hold_obj(dp,
1866 newparent->dd_object, NULL, dd, &dd->dd_parent));
1868 /* add to new parent zapobj */
1869 VERIFY0(zap_add(mos, newparent->dd_phys->dd_child_dir_zapobj,
1870 dd->dd_myname, 8, 1, &dd->dd_object, tx));
1874 zfsvfs_update_fromname(ddra->ddra_oldname, ddra->ddra_newname);
1875 zvol_rename_minors(ddra->ddra_oldname, ddra->ddra_newname);
1879 dsl_prop_notify_all(dd);
1881 dsl_dir_rele(newparent, FTAG);
1882 dsl_dir_rele(dd, FTAG);
1886 dsl_dir_rename(const char *oldname, const char *newname)
1888 dsl_dir_rename_arg_t ddra;
1890 ddra.ddra_oldname = oldname;
1891 ddra.ddra_newname = newname;
1892 ddra.ddra_cred = CRED();
1894 return (dsl_sync_task(oldname,
1895 dsl_dir_rename_check, dsl_dir_rename_sync, &ddra, 3));
1899 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd,
1900 uint64_t fs_cnt, uint64_t ss_cnt, uint64_t space, cred_t *cr)
1902 dsl_dir_t *ancestor;
1907 ancestor = closest_common_ancestor(sdd, tdd);
1908 adelta = would_change(sdd, -space, ancestor);
1909 avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1911 return (SET_ERROR(ENOSPC));
1913 err = dsl_fs_ss_limit_check(tdd, fs_cnt, ZFS_PROP_FILESYSTEM_LIMIT,
1917 err = dsl_fs_ss_limit_check(tdd, ss_cnt, ZFS_PROP_SNAPSHOT_LIMIT,
1926 dsl_dir_snap_cmtime(dsl_dir_t *dd)
1930 mutex_enter(&dd->dd_lock);
1931 t = dd->dd_snap_cmtime;
1932 mutex_exit(&dd->dd_lock);
1938 dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
1943 mutex_enter(&dd->dd_lock);
1944 dd->dd_snap_cmtime = t;
1945 mutex_exit(&dd->dd_lock);
1949 dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx)
1951 objset_t *mos = dd->dd_pool->dp_meta_objset;
1952 dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx);
1956 dsl_dir_is_zapified(dsl_dir_t *dd)
1958 dmu_object_info_t doi;
1960 dmu_object_info_from_db(dd->dd_dbuf, &doi);
1961 return (doi.doi_type == DMU_OTN_ZAP_METADATA);