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 at least ZFS_MAX_DATASET_NAME_LEN bytes */
308 dsl_dir_name(dsl_dir_t *dd, char *buf)
311 dsl_dir_name(dd->dd_parent, buf);
312 VERIFY3U(strlcat(buf, "/", ZFS_MAX_DATASET_NAME_LEN), <,
313 ZFS_MAX_DATASET_NAME_LEN);
317 if (!MUTEX_HELD(&dd->dd_lock)) {
319 * recursive mutex so that we can use
320 * dprintf_dd() with dd_lock held
322 mutex_enter(&dd->dd_lock);
323 VERIFY3U(strlcat(buf, dd->dd_myname, ZFS_MAX_DATASET_NAME_LEN),
324 <, ZFS_MAX_DATASET_NAME_LEN);
325 mutex_exit(&dd->dd_lock);
327 VERIFY3U(strlcat(buf, dd->dd_myname, ZFS_MAX_DATASET_NAME_LEN),
328 <, ZFS_MAX_DATASET_NAME_LEN);
332 /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
334 dsl_dir_namelen(dsl_dir_t *dd)
339 /* parent's name + 1 for the "/" */
340 result = dsl_dir_namelen(dd->dd_parent) + 1;
343 if (!MUTEX_HELD(&dd->dd_lock)) {
344 /* see dsl_dir_name */
345 mutex_enter(&dd->dd_lock);
346 result += strlen(dd->dd_myname);
347 mutex_exit(&dd->dd_lock);
349 result += strlen(dd->dd_myname);
356 getcomponent(const char *path, char *component, const char **nextp)
360 if ((path == NULL) || (path[0] == '\0'))
361 return (SET_ERROR(ENOENT));
362 /* This would be a good place to reserve some namespace... */
363 p = strpbrk(path, "/@");
364 if (p && (p[1] == '/' || p[1] == '@')) {
365 /* two separators in a row */
366 return (SET_ERROR(EINVAL));
368 if (p == NULL || p == path) {
370 * if the first thing is an @ or /, it had better be an
371 * @ and it had better not have any more ats or slashes,
372 * and it had better have something after the @.
375 (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
376 return (SET_ERROR(EINVAL));
377 if (strlen(path) >= ZFS_MAX_DATASET_NAME_LEN)
378 return (SET_ERROR(ENAMETOOLONG));
379 (void) strcpy(component, path);
381 } else if (p[0] == '/') {
382 if (p - path >= ZFS_MAX_DATASET_NAME_LEN)
383 return (SET_ERROR(ENAMETOOLONG));
384 (void) strncpy(component, path, p - path);
385 component[p - path] = '\0';
387 } else if (p[0] == '@') {
389 * if the next separator is an @, there better not be
392 if (strchr(path, '/'))
393 return (SET_ERROR(EINVAL));
394 if (p - path >= ZFS_MAX_DATASET_NAME_LEN)
395 return (SET_ERROR(ENAMETOOLONG));
396 (void) strncpy(component, path, p - path);
397 component[p - path] = '\0';
399 panic("invalid p=%p", (void *)p);
406 * Return the dsl_dir_t, and possibly the last component which couldn't
407 * be found in *tail. The name must be in the specified dsl_pool_t. This
408 * thread must hold the dp_config_rwlock for the pool. Returns NULL if the
409 * path is bogus, or if tail==NULL and we couldn't parse the whole name.
410 * (*tail)[0] == '@' means that the last component is a snapshot.
413 dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag,
414 dsl_dir_t **ddp, const char **tailp)
416 char buf[ZFS_MAX_DATASET_NAME_LEN];
417 const char *spaname, *next, *nextnext = NULL;
422 err = getcomponent(name, buf, &next);
426 /* Make sure the name is in the specified pool. */
427 spaname = spa_name(dp->dp_spa);
428 if (strcmp(buf, spaname) != 0)
429 return (SET_ERROR(EXDEV));
431 ASSERT(dsl_pool_config_held(dp));
433 err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
438 while (next != NULL) {
440 err = getcomponent(next, buf, &nextnext);
443 ASSERT(next[0] != '\0');
446 dprintf("looking up %s in obj%lld\n",
447 buf, dsl_dir_phys(dd)->dd_child_dir_zapobj);
449 err = zap_lookup(dp->dp_meta_objset,
450 dsl_dir_phys(dd)->dd_child_dir_zapobj,
451 buf, sizeof (ddobj), 1, &ddobj);
458 err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_dd);
461 dsl_dir_rele(dd, tag);
467 dsl_dir_rele(dd, tag);
472 * It's an error if there's more than one component left, or
473 * tailp==NULL and there's any component left.
476 (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
478 dsl_dir_rele(dd, tag);
479 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
480 err = SET_ERROR(ENOENT);
489 * If the counts are already initialized for this filesystem and its
490 * descendants then do nothing, otherwise initialize the counts.
492 * The counts on this filesystem, and those below, may be uninitialized due to
493 * either the use of a pre-existing pool which did not support the
494 * filesystem/snapshot limit feature, or one in which the feature had not yet
497 * Recursively descend the filesystem tree and update the filesystem/snapshot
498 * counts on each filesystem below, then update the cumulative count on the
499 * current filesystem. If the filesystem already has a count set on it,
500 * then we know that its counts, and the counts on the filesystems below it,
501 * are already correct, so we don't have to update this filesystem.
504 dsl_dir_init_fs_ss_count(dsl_dir_t *dd, dmu_tx_t *tx)
506 uint64_t my_fs_cnt = 0;
507 uint64_t my_ss_cnt = 0;
508 dsl_pool_t *dp = dd->dd_pool;
509 objset_t *os = dp->dp_meta_objset;
514 ASSERT(spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT));
515 ASSERT(dsl_pool_config_held(dp));
516 ASSERT(dmu_tx_is_syncing(tx));
518 dsl_dir_zapify(dd, tx);
521 * If the filesystem count has already been initialized then we
522 * don't need to recurse down any further.
524 if (zap_contains(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT) == 0)
527 zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP);
528 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
530 /* Iterate my child dirs */
531 for (zap_cursor_init(zc, os, dsl_dir_phys(dd)->dd_child_dir_zapobj);
532 zap_cursor_retrieve(zc, za) == 0; zap_cursor_advance(zc)) {
536 VERIFY0(dsl_dir_hold_obj(dp, za->za_first_integer, NULL, FTAG,
540 * Ignore hidden ($FREE, $MOS & $ORIGIN) objsets and
541 * temporary datasets.
543 if (chld_dd->dd_myname[0] == '$' ||
544 chld_dd->dd_myname[0] == '%') {
545 dsl_dir_rele(chld_dd, FTAG);
549 my_fs_cnt++; /* count this child */
551 dsl_dir_init_fs_ss_count(chld_dd, tx);
553 VERIFY0(zap_lookup(os, chld_dd->dd_object,
554 DD_FIELD_FILESYSTEM_COUNT, sizeof (count), 1, &count));
556 VERIFY0(zap_lookup(os, chld_dd->dd_object,
557 DD_FIELD_SNAPSHOT_COUNT, sizeof (count), 1, &count));
560 dsl_dir_rele(chld_dd, FTAG);
563 /* Count my snapshots (we counted children's snapshots above) */
564 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
565 dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds));
567 for (zap_cursor_init(zc, os, dsl_dataset_phys(ds)->ds_snapnames_zapobj);
568 zap_cursor_retrieve(zc, za) == 0;
569 zap_cursor_advance(zc)) {
570 /* Don't count temporary snapshots */
571 if (za->za_name[0] != '%')
576 dsl_dataset_rele(ds, FTAG);
578 kmem_free(zc, sizeof (zap_cursor_t));
579 kmem_free(za, sizeof (zap_attribute_t));
581 /* we're in a sync task, update counts */
582 dmu_buf_will_dirty(dd->dd_dbuf, tx);
583 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
584 sizeof (my_fs_cnt), 1, &my_fs_cnt, tx));
585 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
586 sizeof (my_ss_cnt), 1, &my_ss_cnt, tx));
590 dsl_dir_actv_fs_ss_limit_check(void *arg, dmu_tx_t *tx)
592 char *ddname = (char *)arg;
593 dsl_pool_t *dp = dmu_tx_pool(tx);
598 error = dsl_dataset_hold(dp, ddname, FTAG, &ds);
602 if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
603 dsl_dataset_rele(ds, FTAG);
604 return (SET_ERROR(ENOTSUP));
608 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT) &&
609 dsl_dir_is_zapified(dd) &&
610 zap_contains(dp->dp_meta_objset, dd->dd_object,
611 DD_FIELD_FILESYSTEM_COUNT) == 0) {
612 dsl_dataset_rele(ds, FTAG);
613 return (SET_ERROR(EALREADY));
616 dsl_dataset_rele(ds, FTAG);
621 dsl_dir_actv_fs_ss_limit_sync(void *arg, dmu_tx_t *tx)
623 char *ddname = (char *)arg;
624 dsl_pool_t *dp = dmu_tx_pool(tx);
628 VERIFY0(dsl_dataset_hold(dp, ddname, FTAG, &ds));
630 spa = dsl_dataset_get_spa(ds);
632 if (!spa_feature_is_active(spa, SPA_FEATURE_FS_SS_LIMIT)) {
634 * Since the feature was not active and we're now setting a
635 * limit, increment the feature-active counter so that the
636 * feature becomes active for the first time.
638 * We are already in a sync task so we can update the MOS.
640 spa_feature_incr(spa, SPA_FEATURE_FS_SS_LIMIT, tx);
644 * Since we are now setting a non-UINT64_MAX limit on the filesystem,
645 * we need to ensure the counts are correct. Descend down the tree from
646 * this point and update all of the counts to be accurate.
648 dsl_dir_init_fs_ss_count(ds->ds_dir, tx);
650 dsl_dataset_rele(ds, FTAG);
654 * Make sure the feature is enabled and activate it if necessary.
655 * Since we're setting a limit, ensure the on-disk counts are valid.
656 * This is only called by the ioctl path when setting a limit value.
658 * We do not need to validate the new limit, since users who can change the
659 * limit are also allowed to exceed the limit.
662 dsl_dir_activate_fs_ss_limit(const char *ddname)
666 error = dsl_sync_task(ddname, dsl_dir_actv_fs_ss_limit_check,
667 dsl_dir_actv_fs_ss_limit_sync, (void *)ddname, 0,
668 ZFS_SPACE_CHECK_RESERVED);
670 if (error == EALREADY)
677 * Used to determine if the filesystem_limit or snapshot_limit should be
678 * enforced. We allow the limit to be exceeded if the user has permission to
679 * write the property value. We pass in the creds that we got in the open
680 * context since we will always be the GZ root in syncing context. We also have
681 * to handle the case where we are allowed to change the limit on the current
682 * dataset, but there may be another limit in the tree above.
684 * We can never modify these two properties within a non-global zone. In
685 * addition, the other checks are modeled on zfs_secpolicy_write_perms. We
686 * can't use that function since we are already holding the dp_config_rwlock.
687 * In addition, we already have the dd and dealing with snapshots is simplified
698 dsl_enforce_ds_ss_limits(dsl_dir_t *dd, zfs_prop_t prop, cred_t *cr)
700 enforce_res_t enforce = ENFORCE_ALWAYS;
705 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
706 prop == ZFS_PROP_SNAPSHOT_LIMIT);
712 if (crgetzoneid(cr) != GLOBAL_ZONEID)
714 return (ENFORCE_ALWAYS);
716 if (secpolicy_zfs(cr) == 0)
717 return (ENFORCE_NEVER);
720 if ((obj = dsl_dir_phys(dd)->dd_head_dataset_obj) == 0)
721 return (ENFORCE_ALWAYS);
723 ASSERT(dsl_pool_config_held(dd->dd_pool));
725 if (dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds) != 0)
726 return (ENFORCE_ALWAYS);
728 if (dsl_prop_get_ds(ds, "zoned", 8, 1, &zoned, NULL) || zoned) {
729 /* Only root can access zoned fs's from the GZ */
730 enforce = ENFORCE_ALWAYS;
732 if (dsl_deleg_access_impl(ds, zfs_prop_to_name(prop), cr) == 0)
733 enforce = ENFORCE_ABOVE;
736 dsl_dataset_rele(ds, FTAG);
741 * Check if adding additional child filesystem(s) would exceed any filesystem
742 * limits or adding additional snapshot(s) would exceed any snapshot limits.
743 * The prop argument indicates which limit to check.
745 * Note that all filesystem limits up to the root (or the highest
746 * initialized) filesystem or the given ancestor must be satisfied.
749 dsl_fs_ss_limit_check(dsl_dir_t *dd, uint64_t delta, zfs_prop_t prop,
750 dsl_dir_t *ancestor, cred_t *cr)
752 objset_t *os = dd->dd_pool->dp_meta_objset;
753 uint64_t limit, count;
755 enforce_res_t enforce;
758 ASSERT(dsl_pool_config_held(dd->dd_pool));
759 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
760 prop == ZFS_PROP_SNAPSHOT_LIMIT);
763 * If we're allowed to change the limit, don't enforce the limit
764 * e.g. this can happen if a snapshot is taken by an administrative
765 * user in the global zone (i.e. a recursive snapshot by root).
766 * However, we must handle the case of delegated permissions where we
767 * are allowed to change the limit on the current dataset, but there
768 * is another limit in the tree above.
770 enforce = dsl_enforce_ds_ss_limits(dd, prop, cr);
771 if (enforce == ENFORCE_NEVER)
775 * e.g. if renaming a dataset with no snapshots, count adjustment
781 if (prop == ZFS_PROP_SNAPSHOT_LIMIT) {
783 * We don't enforce the limit for temporary snapshots. This is
784 * indicated by a NULL cred_t argument.
789 count_prop = DD_FIELD_SNAPSHOT_COUNT;
791 count_prop = DD_FIELD_FILESYSTEM_COUNT;
795 * If an ancestor has been provided, stop checking the limit once we
796 * hit that dir. We need this during rename so that we don't overcount
797 * the check once we recurse up to the common ancestor.
803 * If we hit an uninitialized node while recursing up the tree, we can
804 * stop since we know there is no limit here (or above). The counts are
805 * not valid on this node and we know we won't touch this node's counts.
807 if (!dsl_dir_is_zapified(dd) || zap_lookup(os, dd->dd_object,
808 count_prop, sizeof (count), 1, &count) == ENOENT)
811 err = dsl_prop_get_dd(dd, zfs_prop_to_name(prop), 8, 1, &limit, NULL,
816 /* Is there a limit which we've hit? */
817 if (enforce == ENFORCE_ALWAYS && (count + delta) > limit)
818 return (SET_ERROR(EDQUOT));
820 if (dd->dd_parent != NULL)
821 err = dsl_fs_ss_limit_check(dd->dd_parent, delta, prop,
828 * Adjust the filesystem or snapshot count for the specified dsl_dir_t and all
829 * parents. When a new filesystem/snapshot is created, increment the count on
830 * all parents, and when a filesystem/snapshot is destroyed, decrement the
834 dsl_fs_ss_count_adjust(dsl_dir_t *dd, int64_t delta, const char *prop,
838 objset_t *os = dd->dd_pool->dp_meta_objset;
841 ASSERT(dsl_pool_config_held(dd->dd_pool));
842 ASSERT(dmu_tx_is_syncing(tx));
843 ASSERT(strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0 ||
844 strcmp(prop, DD_FIELD_SNAPSHOT_COUNT) == 0);
847 * When we receive an incremental stream into a filesystem that already
848 * exists, a temporary clone is created. We don't count this temporary
849 * clone, whose name begins with a '%'. We also ignore hidden ($FREE,
850 * $MOS & $ORIGIN) objsets.
852 if ((dd->dd_myname[0] == '%' || dd->dd_myname[0] == '$') &&
853 strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0)
857 * e.g. if renaming a dataset with no snapshots, count adjustment is 0
863 * If we hit an uninitialized node while recursing up the tree, we can
864 * stop since we know the counts are not valid on this node and we
865 * know we shouldn't touch this node's counts. An uninitialized count
866 * on the node indicates that either the feature has not yet been
867 * activated or there are no limits on this part of the tree.
869 if (!dsl_dir_is_zapified(dd) || (err = zap_lookup(os, dd->dd_object,
870 prop, sizeof (count), 1, &count)) == ENOENT)
875 /* Use a signed verify to make sure we're not neg. */
876 VERIFY3S(count, >=, 0);
878 VERIFY0(zap_update(os, dd->dd_object, prop, sizeof (count), 1, &count,
881 /* Roll up this additional count into our ancestors */
882 if (dd->dd_parent != NULL)
883 dsl_fs_ss_count_adjust(dd->dd_parent, delta, prop, tx);
887 dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
890 objset_t *mos = dp->dp_meta_objset;
892 dsl_dir_phys_t *ddphys;
895 ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
896 DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
898 VERIFY(0 == zap_add(mos, dsl_dir_phys(pds)->dd_child_dir_zapobj,
899 name, sizeof (uint64_t), 1, &ddobj, tx));
901 /* it's the root dir */
902 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
903 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
905 VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
906 dmu_buf_will_dirty(dbuf, tx);
907 ddphys = dbuf->db_data;
909 ddphys->dd_creation_time = gethrestime_sec();
911 ddphys->dd_parent_obj = pds->dd_object;
913 /* update the filesystem counts */
914 dsl_fs_ss_count_adjust(pds, 1, DD_FIELD_FILESYSTEM_COUNT, tx);
916 ddphys->dd_props_zapobj = zap_create(mos,
917 DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
918 ddphys->dd_child_dir_zapobj = zap_create(mos,
919 DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
920 if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
921 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
922 dmu_buf_rele(dbuf, FTAG);
928 dsl_dir_is_clone(dsl_dir_t *dd)
930 return (dsl_dir_phys(dd)->dd_origin_obj &&
931 (dd->dd_pool->dp_origin_snap == NULL ||
932 dsl_dir_phys(dd)->dd_origin_obj !=
933 dd->dd_pool->dp_origin_snap->ds_object));
937 dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
939 mutex_enter(&dd->dd_lock);
940 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
941 dsl_dir_phys(dd)->dd_used_bytes);
942 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA,
943 dsl_dir_phys(dd)->dd_quota);
944 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
945 dsl_dir_phys(dd)->dd_reserved);
946 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
947 dsl_dir_phys(dd)->dd_compressed_bytes == 0 ? 100 :
948 (dsl_dir_phys(dd)->dd_uncompressed_bytes * 100 /
949 dsl_dir_phys(dd)->dd_compressed_bytes));
950 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED,
951 dsl_dir_phys(dd)->dd_uncompressed_bytes);
952 if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
953 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
954 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]);
955 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
956 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_HEAD]);
957 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
958 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_REFRSRV]);
959 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
960 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD] +
961 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD_RSRV]);
963 mutex_exit(&dd->dd_lock);
965 if (dsl_dir_is_zapified(dd)) {
967 objset_t *os = dd->dd_pool->dp_meta_objset;
969 if (zap_lookup(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
970 sizeof (count), 1, &count) == 0) {
971 dsl_prop_nvlist_add_uint64(nv,
972 ZFS_PROP_FILESYSTEM_COUNT, count);
974 if (zap_lookup(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
975 sizeof (count), 1, &count) == 0) {
976 dsl_prop_nvlist_add_uint64(nv,
977 ZFS_PROP_SNAPSHOT_COUNT, count);
981 if (dsl_dir_is_clone(dd)) {
983 char buf[ZFS_MAX_DATASET_NAME_LEN];
985 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
986 dsl_dir_phys(dd)->dd_origin_obj, FTAG, &ds));
987 dsl_dataset_name(ds, buf);
988 dsl_dataset_rele(ds, FTAG);
989 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
994 dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
996 dsl_pool_t *dp = dd->dd_pool;
998 ASSERT(dsl_dir_phys(dd));
1000 if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) {
1001 /* up the hold count until we can be written out */
1002 dmu_buf_add_ref(dd->dd_dbuf, dd);
1007 parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
1009 uint64_t old_accounted = MAX(used, dsl_dir_phys(dd)->dd_reserved);
1010 uint64_t new_accounted =
1011 MAX(used + delta, dsl_dir_phys(dd)->dd_reserved);
1012 return (new_accounted - old_accounted);
1016 dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
1018 ASSERT(dmu_tx_is_syncing(tx));
1020 mutex_enter(&dd->dd_lock);
1021 ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]);
1022 dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
1023 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
1024 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
1025 mutex_exit(&dd->dd_lock);
1027 /* release the hold from dsl_dir_dirty */
1028 dmu_buf_rele(dd->dd_dbuf, dd);
1032 dsl_dir_space_towrite(dsl_dir_t *dd)
1037 ASSERT(MUTEX_HELD(&dd->dd_lock));
1039 for (i = 0; i < TXG_SIZE; i++) {
1040 space += dd->dd_space_towrite[i&TXG_MASK];
1041 ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0);
1047 * How much space would dd have available if ancestor had delta applied
1048 * to it? If ondiskonly is set, we're only interested in what's
1049 * on-disk, not estimated pending changes.
1052 dsl_dir_space_available(dsl_dir_t *dd,
1053 dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
1055 uint64_t parentspace, myspace, quota, used;
1058 * If there are no restrictions otherwise, assume we have
1059 * unlimited space available.
1062 parentspace = UINT64_MAX;
1064 if (dd->dd_parent != NULL) {
1065 parentspace = dsl_dir_space_available(dd->dd_parent,
1066 ancestor, delta, ondiskonly);
1069 mutex_enter(&dd->dd_lock);
1070 if (dsl_dir_phys(dd)->dd_quota != 0)
1071 quota = dsl_dir_phys(dd)->dd_quota;
1072 used = dsl_dir_phys(dd)->dd_used_bytes;
1074 used += dsl_dir_space_towrite(dd);
1076 if (dd->dd_parent == NULL) {
1077 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
1078 quota = MIN(quota, poolsize);
1081 if (dsl_dir_phys(dd)->dd_reserved > used && parentspace != UINT64_MAX) {
1083 * We have some space reserved, in addition to what our
1086 parentspace += dsl_dir_phys(dd)->dd_reserved - used;
1089 if (dd == ancestor) {
1091 ASSERT(used >= -delta);
1093 if (parentspace != UINT64_MAX)
1094 parentspace -= delta;
1102 * the lesser of the space provided by our parent and
1103 * the space left in our quota
1105 myspace = MIN(parentspace, quota - used);
1108 mutex_exit(&dd->dd_lock);
1113 struct tempreserve {
1114 list_node_t tr_node;
1120 dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
1121 boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list,
1122 dmu_tx_t *tx, boolean_t first)
1124 uint64_t txg = tx->tx_txg;
1125 uint64_t est_inflight, used_on_disk, quota, parent_rsrv;
1126 uint64_t deferred = 0;
1127 struct tempreserve *tr;
1128 int retval = EDQUOT;
1129 int txgidx = txg & TXG_MASK;
1131 uint64_t ref_rsrv = 0;
1133 ASSERT3U(txg, !=, 0);
1134 ASSERT3S(asize, >, 0);
1136 mutex_enter(&dd->dd_lock);
1139 * Check against the dsl_dir's quota. We don't add in the delta
1140 * when checking for over-quota because they get one free hit.
1142 est_inflight = dsl_dir_space_towrite(dd);
1143 for (i = 0; i < TXG_SIZE; i++)
1144 est_inflight += dd->dd_tempreserved[i];
1145 used_on_disk = dsl_dir_phys(dd)->dd_used_bytes;
1148 * On the first iteration, fetch the dataset's used-on-disk and
1149 * refreservation values. Also, if checkrefquota is set, test if
1150 * allocating this space would exceed the dataset's refquota.
1152 if (first && tx->tx_objset) {
1154 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
1156 error = dsl_dataset_check_quota(ds, checkrefquota,
1157 asize, est_inflight, &used_on_disk, &ref_rsrv);
1159 mutex_exit(&dd->dd_lock);
1165 * If this transaction will result in a net free of space,
1166 * we want to let it through.
1168 if (ignorequota || netfree || dsl_dir_phys(dd)->dd_quota == 0)
1171 quota = dsl_dir_phys(dd)->dd_quota;
1174 * Adjust the quota against the actual pool size at the root
1175 * minus any outstanding deferred frees.
1176 * To ensure that it's possible to remove files from a full
1177 * pool without inducing transient overcommits, we throttle
1178 * netfree transactions against a quota that is slightly larger,
1179 * but still within the pool's allocation slop. In cases where
1180 * we're very close to full, this will allow a steady trickle of
1181 * removes to get through.
1183 if (dd->dd_parent == NULL) {
1184 spa_t *spa = dd->dd_pool->dp_spa;
1185 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
1186 deferred = metaslab_class_get_deferred(spa_normal_class(spa));
1187 if (poolsize - deferred < quota) {
1188 quota = poolsize - deferred;
1194 * If they are requesting more space, and our current estimate
1195 * is over quota, they get to try again unless the actual
1196 * on-disk is over quota and there are no pending changes (which
1197 * may free up space for us).
1199 if (used_on_disk + est_inflight >= quota) {
1200 if (est_inflight > 0 || used_on_disk < quota ||
1201 (retval == ENOSPC && used_on_disk < quota + deferred))
1203 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
1204 "quota=%lluK tr=%lluK err=%d\n",
1205 used_on_disk>>10, est_inflight>>10,
1206 quota>>10, asize>>10, retval);
1207 mutex_exit(&dd->dd_lock);
1208 return (SET_ERROR(retval));
1211 /* We need to up our estimated delta before dropping dd_lock */
1212 dd->dd_tempreserved[txgidx] += asize;
1214 parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
1216 mutex_exit(&dd->dd_lock);
1218 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1220 tr->tr_size = asize;
1221 list_insert_tail(tr_list, tr);
1223 /* see if it's OK with our parent */
1224 if (dd->dd_parent && parent_rsrv) {
1225 boolean_t ismos = (dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
1227 return (dsl_dir_tempreserve_impl(dd->dd_parent,
1228 parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE));
1235 * Reserve space in this dsl_dir, to be used in this tx's txg.
1236 * After the space has been dirtied (and dsl_dir_willuse_space()
1237 * has been called), the reservation should be canceled, using
1238 * dsl_dir_tempreserve_clear().
1241 dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
1242 uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx)
1252 tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
1253 list_create(tr_list, sizeof (struct tempreserve),
1254 offsetof(struct tempreserve, tr_node));
1255 ASSERT3S(asize, >, 0);
1256 ASSERT3S(fsize, >=, 0);
1258 err = arc_tempreserve_space(lsize, tx->tx_txg);
1260 struct tempreserve *tr;
1262 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1263 tr->tr_size = lsize;
1264 list_insert_tail(tr_list, tr);
1266 if (err == EAGAIN) {
1268 * If arc_memory_throttle() detected that pageout
1269 * is running and we are low on memory, we delay new
1270 * non-pageout transactions to give pageout an
1273 * It is unfortunate to be delaying while the caller's
1276 txg_delay(dd->dd_pool, tx->tx_txg,
1277 MSEC2NSEC(10), MSEC2NSEC(10));
1278 err = SET_ERROR(ERESTART);
1283 err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize,
1284 FALSE, asize > usize, tr_list, tx, TRUE);
1288 dsl_dir_tempreserve_clear(tr_list, tx);
1290 *tr_cookiep = tr_list;
1296 * Clear a temporary reservation that we previously made with
1297 * dsl_dir_tempreserve_space().
1300 dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
1302 int txgidx = tx->tx_txg & TXG_MASK;
1303 list_t *tr_list = tr_cookie;
1304 struct tempreserve *tr;
1306 ASSERT3U(tx->tx_txg, !=, 0);
1308 if (tr_cookie == NULL)
1311 while ((tr = list_head(tr_list)) != NULL) {
1313 mutex_enter(&tr->tr_ds->dd_lock);
1314 ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
1316 tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
1317 mutex_exit(&tr->tr_ds->dd_lock);
1319 arc_tempreserve_clear(tr->tr_size);
1321 list_remove(tr_list, tr);
1322 kmem_free(tr, sizeof (struct tempreserve));
1325 kmem_free(tr_list, sizeof (list_t));
1329 * This should be called from open context when we think we're going to write
1330 * or free space, for example when dirtying data. Be conservative; it's okay
1331 * to write less space or free more, but we don't want to write more or free
1332 * less than the amount specified.
1335 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
1337 int64_t parent_space;
1340 mutex_enter(&dd->dd_lock);
1342 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
1344 est_used = dsl_dir_space_towrite(dd) + dsl_dir_phys(dd)->dd_used_bytes;
1345 parent_space = parent_delta(dd, est_used, space);
1346 mutex_exit(&dd->dd_lock);
1348 /* Make sure that we clean up dd_space_to* */
1349 dsl_dir_dirty(dd, tx);
1351 /* XXX this is potentially expensive and unnecessary... */
1352 if (parent_space && dd->dd_parent)
1353 dsl_dir_willuse_space(dd->dd_parent, parent_space, tx);
1356 /* call from syncing context when we actually write/free space for this dd */
1358 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
1359 int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
1361 int64_t accounted_delta;
1364 * dsl_dataset_set_refreservation_sync_impl() calls this with
1365 * dd_lock held, so that it can atomically update
1366 * ds->ds_reserved and the dsl_dir accounting, so that
1367 * dsl_dataset_check_quota() can see dataset and dir accounting
1370 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
1372 ASSERT(dmu_tx_is_syncing(tx));
1373 ASSERT(type < DD_USED_NUM);
1375 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1378 mutex_enter(&dd->dd_lock);
1380 parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, used);
1381 ASSERT(used >= 0 || dsl_dir_phys(dd)->dd_used_bytes >= -used);
1382 ASSERT(compressed >= 0 ||
1383 dsl_dir_phys(dd)->dd_compressed_bytes >= -compressed);
1384 ASSERT(uncompressed >= 0 ||
1385 dsl_dir_phys(dd)->dd_uncompressed_bytes >= -uncompressed);
1386 dsl_dir_phys(dd)->dd_used_bytes += used;
1387 dsl_dir_phys(dd)->dd_uncompressed_bytes += uncompressed;
1388 dsl_dir_phys(dd)->dd_compressed_bytes += compressed;
1390 if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1392 dsl_dir_phys(dd)->dd_used_breakdown[type] >= -used);
1393 dsl_dir_phys(dd)->dd_used_breakdown[type] += used;
1397 for (t = 0; t < DD_USED_NUM; t++)
1398 u += dsl_dir_phys(dd)->dd_used_breakdown[t];
1399 ASSERT3U(u, ==, dsl_dir_phys(dd)->dd_used_bytes);
1403 mutex_exit(&dd->dd_lock);
1405 if (dd->dd_parent != NULL) {
1406 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1407 accounted_delta, compressed, uncompressed, tx);
1408 dsl_dir_transfer_space(dd->dd_parent,
1409 used - accounted_delta,
1410 DD_USED_CHILD_RSRV, DD_USED_CHILD, NULL);
1415 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
1416 dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
1418 ASSERT(tx == NULL || dmu_tx_is_syncing(tx));
1419 ASSERT(oldtype < DD_USED_NUM);
1420 ASSERT(newtype < DD_USED_NUM);
1423 !(dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN))
1427 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1428 mutex_enter(&dd->dd_lock);
1430 dsl_dir_phys(dd)->dd_used_breakdown[oldtype] >= delta :
1431 dsl_dir_phys(dd)->dd_used_breakdown[newtype] >= -delta);
1432 ASSERT(dsl_dir_phys(dd)->dd_used_bytes >= ABS(delta));
1433 dsl_dir_phys(dd)->dd_used_breakdown[oldtype] -= delta;
1434 dsl_dir_phys(dd)->dd_used_breakdown[newtype] += delta;
1435 mutex_exit(&dd->dd_lock);
1438 typedef struct dsl_dir_set_qr_arg {
1439 const char *ddsqra_name;
1440 zprop_source_t ddsqra_source;
1441 uint64_t ddsqra_value;
1442 } dsl_dir_set_qr_arg_t;
1445 dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx)
1447 dsl_dir_set_qr_arg_t *ddsqra = arg;
1448 dsl_pool_t *dp = dmu_tx_pool(tx);
1451 uint64_t towrite, newval;
1453 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1457 error = dsl_prop_predict(ds->ds_dir, "quota",
1458 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1460 dsl_dataset_rele(ds, FTAG);
1465 dsl_dataset_rele(ds, FTAG);
1469 mutex_enter(&ds->ds_dir->dd_lock);
1471 * If we are doing the preliminary check in open context, and
1472 * there are pending changes, then don't fail it, since the
1473 * pending changes could under-estimate the amount of space to be
1476 towrite = dsl_dir_space_towrite(ds->ds_dir);
1477 if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
1478 (newval < dsl_dir_phys(ds->ds_dir)->dd_reserved ||
1479 newval < dsl_dir_phys(ds->ds_dir)->dd_used_bytes + towrite)) {
1480 error = SET_ERROR(ENOSPC);
1482 mutex_exit(&ds->ds_dir->dd_lock);
1483 dsl_dataset_rele(ds, FTAG);
1488 dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx)
1490 dsl_dir_set_qr_arg_t *ddsqra = arg;
1491 dsl_pool_t *dp = dmu_tx_pool(tx);
1495 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1497 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1498 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA),
1499 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1500 &ddsqra->ddsqra_value, tx);
1502 VERIFY0(dsl_prop_get_int_ds(ds,
1503 zfs_prop_to_name(ZFS_PROP_QUOTA), &newval));
1505 newval = ddsqra->ddsqra_value;
1506 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1507 zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval);
1510 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1511 mutex_enter(&ds->ds_dir->dd_lock);
1512 dsl_dir_phys(ds->ds_dir)->dd_quota = newval;
1513 mutex_exit(&ds->ds_dir->dd_lock);
1514 dsl_dataset_rele(ds, FTAG);
1518 dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
1520 dsl_dir_set_qr_arg_t ddsqra;
1522 ddsqra.ddsqra_name = ddname;
1523 ddsqra.ddsqra_source = source;
1524 ddsqra.ddsqra_value = quota;
1526 return (dsl_sync_task(ddname, dsl_dir_set_quota_check,
1527 dsl_dir_set_quota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
1531 dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx)
1533 dsl_dir_set_qr_arg_t *ddsqra = arg;
1534 dsl_pool_t *dp = dmu_tx_pool(tx);
1537 uint64_t newval, used, avail;
1540 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1546 * If we are doing the preliminary check in open context, the
1547 * space estimates may be inaccurate.
1549 if (!dmu_tx_is_syncing(tx)) {
1550 dsl_dataset_rele(ds, FTAG);
1554 error = dsl_prop_predict(ds->ds_dir,
1555 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1556 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1558 dsl_dataset_rele(ds, FTAG);
1562 mutex_enter(&dd->dd_lock);
1563 used = dsl_dir_phys(dd)->dd_used_bytes;
1564 mutex_exit(&dd->dd_lock);
1566 if (dd->dd_parent) {
1567 avail = dsl_dir_space_available(dd->dd_parent,
1570 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
1573 if (MAX(used, newval) > MAX(used, dsl_dir_phys(dd)->dd_reserved)) {
1574 uint64_t delta = MAX(used, newval) -
1575 MAX(used, dsl_dir_phys(dd)->dd_reserved);
1577 if (delta > avail ||
1578 (dsl_dir_phys(dd)->dd_quota > 0 &&
1579 newval > dsl_dir_phys(dd)->dd_quota))
1580 error = SET_ERROR(ENOSPC);
1583 dsl_dataset_rele(ds, FTAG);
1588 dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
1593 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1595 mutex_enter(&dd->dd_lock);
1596 used = dsl_dir_phys(dd)->dd_used_bytes;
1597 delta = MAX(used, value) - MAX(used, dsl_dir_phys(dd)->dd_reserved);
1598 dsl_dir_phys(dd)->dd_reserved = value;
1600 if (dd->dd_parent != NULL) {
1601 /* Roll up this additional usage into our ancestors */
1602 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1605 mutex_exit(&dd->dd_lock);
1609 dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx)
1611 dsl_dir_set_qr_arg_t *ddsqra = arg;
1612 dsl_pool_t *dp = dmu_tx_pool(tx);
1616 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1618 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1619 dsl_prop_set_sync_impl(ds,
1620 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1621 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1622 &ddsqra->ddsqra_value, tx);
1624 VERIFY0(dsl_prop_get_int_ds(ds,
1625 zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval));
1627 newval = ddsqra->ddsqra_value;
1628 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1629 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1630 (longlong_t)newval);
1633 dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx);
1634 dsl_dataset_rele(ds, FTAG);
1638 dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1639 uint64_t reservation)
1641 dsl_dir_set_qr_arg_t ddsqra;
1643 ddsqra.ddsqra_name = ddname;
1644 ddsqra.ddsqra_source = source;
1645 ddsqra.ddsqra_value = reservation;
1647 return (dsl_sync_task(ddname, dsl_dir_set_reservation_check,
1648 dsl_dir_set_reservation_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
1652 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1654 for (; ds1; ds1 = ds1->dd_parent) {
1656 for (dd = ds2; dd; dd = dd->dd_parent) {
1665 * If delta is applied to dd, how much of that delta would be applied to
1666 * ancestor? Syncing context only.
1669 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1674 mutex_enter(&dd->dd_lock);
1675 delta = parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, delta);
1676 mutex_exit(&dd->dd_lock);
1677 return (would_change(dd->dd_parent, delta, ancestor));
1680 typedef struct dsl_dir_rename_arg {
1681 const char *ddra_oldname;
1682 const char *ddra_newname;
1684 } dsl_dir_rename_arg_t;
1688 dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1691 char namebuf[ZFS_MAX_DATASET_NAME_LEN];
1693 dsl_dataset_name(ds, namebuf);
1695 if (strlen(namebuf) + *deltap >= ZFS_MAX_DATASET_NAME_LEN)
1696 return (SET_ERROR(ENAMETOOLONG));
1701 dsl_dir_rename_check(void *arg, dmu_tx_t *tx)
1703 dsl_dir_rename_arg_t *ddra = arg;
1704 dsl_pool_t *dp = dmu_tx_pool(tx);
1705 dsl_dir_t *dd, *newparent;
1706 const char *mynewname;
1708 int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname);
1710 /* target dir should exist */
1711 error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL);
1715 /* new parent should exist */
1716 error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG,
1717 &newparent, &mynewname);
1719 dsl_dir_rele(dd, FTAG);
1723 /* can't rename to different pool */
1724 if (dd->dd_pool != newparent->dd_pool) {
1725 dsl_dir_rele(newparent, FTAG);
1726 dsl_dir_rele(dd, FTAG);
1727 return (SET_ERROR(EXDEV));
1730 /* new name should not already exist */
1731 if (mynewname == NULL) {
1732 dsl_dir_rele(newparent, FTAG);
1733 dsl_dir_rele(dd, FTAG);
1734 return (SET_ERROR(EEXIST));
1737 /* if the name length is growing, validate child name lengths */
1739 error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename,
1740 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1742 dsl_dir_rele(newparent, FTAG);
1743 dsl_dir_rele(dd, FTAG);
1748 if (dmu_tx_is_syncing(tx)) {
1749 if (spa_feature_is_active(dp->dp_spa,
1750 SPA_FEATURE_FS_SS_LIMIT)) {
1752 * Although this is the check function and we don't
1753 * normally make on-disk changes in check functions,
1754 * we need to do that here.
1756 * Ensure this portion of the tree's counts have been
1757 * initialized in case the new parent has limits set.
1759 dsl_dir_init_fs_ss_count(dd, tx);
1763 if (newparent != dd->dd_parent) {
1764 /* is there enough space? */
1766 MAX(dsl_dir_phys(dd)->dd_used_bytes,
1767 dsl_dir_phys(dd)->dd_reserved);
1768 objset_t *os = dd->dd_pool->dp_meta_objset;
1769 uint64_t fs_cnt = 0;
1770 uint64_t ss_cnt = 0;
1772 if (dsl_dir_is_zapified(dd)) {
1775 err = zap_lookup(os, dd->dd_object,
1776 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1778 if (err != ENOENT && err != 0) {
1779 dsl_dir_rele(newparent, FTAG);
1780 dsl_dir_rele(dd, FTAG);
1785 * have to add 1 for the filesystem itself that we're
1790 err = zap_lookup(os, dd->dd_object,
1791 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1793 if (err != ENOENT && err != 0) {
1794 dsl_dir_rele(newparent, FTAG);
1795 dsl_dir_rele(dd, FTAG);
1800 /* no rename into our descendant */
1801 if (closest_common_ancestor(dd, newparent) == dd) {
1802 dsl_dir_rele(newparent, FTAG);
1803 dsl_dir_rele(dd, FTAG);
1804 return (SET_ERROR(EINVAL));
1807 error = dsl_dir_transfer_possible(dd->dd_parent,
1808 newparent, fs_cnt, ss_cnt, myspace, ddra->ddra_cred);
1810 dsl_dir_rele(newparent, FTAG);
1811 dsl_dir_rele(dd, FTAG);
1816 dsl_dir_rele(newparent, FTAG);
1817 dsl_dir_rele(dd, FTAG);
1822 dsl_dir_rename_sync(void *arg, dmu_tx_t *tx)
1824 dsl_dir_rename_arg_t *ddra = arg;
1825 dsl_pool_t *dp = dmu_tx_pool(tx);
1826 dsl_dir_t *dd, *newparent;
1827 const char *mynewname;
1829 objset_t *mos = dp->dp_meta_objset;
1831 VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL));
1832 VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
1835 /* Log this before we change the name. */
1836 spa_history_log_internal_dd(dd, "rename", tx,
1837 "-> %s", ddra->ddra_newname);
1839 if (newparent != dd->dd_parent) {
1840 objset_t *os = dd->dd_pool->dp_meta_objset;
1841 uint64_t fs_cnt = 0;
1842 uint64_t ss_cnt = 0;
1845 * We already made sure the dd counts were initialized in the
1848 if (spa_feature_is_active(dp->dp_spa,
1849 SPA_FEATURE_FS_SS_LIMIT)) {
1850 VERIFY0(zap_lookup(os, dd->dd_object,
1851 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1853 /* add 1 for the filesystem itself that we're moving */
1856 VERIFY0(zap_lookup(os, dd->dd_object,
1857 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1861 dsl_fs_ss_count_adjust(dd->dd_parent, -fs_cnt,
1862 DD_FIELD_FILESYSTEM_COUNT, tx);
1863 dsl_fs_ss_count_adjust(newparent, fs_cnt,
1864 DD_FIELD_FILESYSTEM_COUNT, tx);
1866 dsl_fs_ss_count_adjust(dd->dd_parent, -ss_cnt,
1867 DD_FIELD_SNAPSHOT_COUNT, tx);
1868 dsl_fs_ss_count_adjust(newparent, ss_cnt,
1869 DD_FIELD_SNAPSHOT_COUNT, tx);
1871 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1872 -dsl_dir_phys(dd)->dd_used_bytes,
1873 -dsl_dir_phys(dd)->dd_compressed_bytes,
1874 -dsl_dir_phys(dd)->dd_uncompressed_bytes, tx);
1875 dsl_dir_diduse_space(newparent, DD_USED_CHILD,
1876 dsl_dir_phys(dd)->dd_used_bytes,
1877 dsl_dir_phys(dd)->dd_compressed_bytes,
1878 dsl_dir_phys(dd)->dd_uncompressed_bytes, tx);
1880 if (dsl_dir_phys(dd)->dd_reserved >
1881 dsl_dir_phys(dd)->dd_used_bytes) {
1882 uint64_t unused_rsrv = dsl_dir_phys(dd)->dd_reserved -
1883 dsl_dir_phys(dd)->dd_used_bytes;
1885 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1886 -unused_rsrv, 0, 0, tx);
1887 dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV,
1888 unused_rsrv, 0, 0, tx);
1892 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1894 /* remove from old parent zapobj */
1895 error = zap_remove(mos,
1896 dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj,
1900 (void) strcpy(dd->dd_myname, mynewname);
1901 dsl_dir_rele(dd->dd_parent, dd);
1902 dsl_dir_phys(dd)->dd_parent_obj = newparent->dd_object;
1903 VERIFY0(dsl_dir_hold_obj(dp,
1904 newparent->dd_object, NULL, dd, &dd->dd_parent));
1906 /* add to new parent zapobj */
1907 VERIFY0(zap_add(mos, dsl_dir_phys(newparent)->dd_child_dir_zapobj,
1908 dd->dd_myname, 8, 1, &dd->dd_object, tx));
1912 zfsvfs_update_fromname(ddra->ddra_oldname, ddra->ddra_newname);
1913 zvol_rename_minors(ddra->ddra_oldname, ddra->ddra_newname);
1917 dsl_prop_notify_all(dd);
1919 dsl_dir_rele(newparent, FTAG);
1920 dsl_dir_rele(dd, FTAG);
1924 dsl_dir_rename(const char *oldname, const char *newname)
1926 dsl_dir_rename_arg_t ddra;
1928 ddra.ddra_oldname = oldname;
1929 ddra.ddra_newname = newname;
1930 ddra.ddra_cred = CRED();
1932 return (dsl_sync_task(oldname,
1933 dsl_dir_rename_check, dsl_dir_rename_sync, &ddra,
1934 3, ZFS_SPACE_CHECK_RESERVED));
1938 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd,
1939 uint64_t fs_cnt, uint64_t ss_cnt, uint64_t space, cred_t *cr)
1941 dsl_dir_t *ancestor;
1946 ancestor = closest_common_ancestor(sdd, tdd);
1947 adelta = would_change(sdd, -space, ancestor);
1948 avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1950 return (SET_ERROR(ENOSPC));
1952 err = dsl_fs_ss_limit_check(tdd, fs_cnt, ZFS_PROP_FILESYSTEM_LIMIT,
1956 err = dsl_fs_ss_limit_check(tdd, ss_cnt, ZFS_PROP_SNAPSHOT_LIMIT,
1965 dsl_dir_snap_cmtime(dsl_dir_t *dd)
1969 mutex_enter(&dd->dd_lock);
1970 t = dd->dd_snap_cmtime;
1971 mutex_exit(&dd->dd_lock);
1977 dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
1982 mutex_enter(&dd->dd_lock);
1983 dd->dd_snap_cmtime = t;
1984 mutex_exit(&dd->dd_lock);
1988 dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx)
1990 objset_t *mos = dd->dd_pool->dp_meta_objset;
1991 dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx);
1995 dsl_dir_is_zapified(dsl_dir_t *dd)
1997 dmu_object_info_t doi;
1999 dmu_object_info_from_db(dd->dd_dbuf, &doi);
2000 return (doi.doi_type == DMU_OTN_ZAP_METADATA);