4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2018 by Delphix. All rights reserved.
24 * Copyright (c) 2013 Martin Matuska. All rights reserved.
25 * Copyright (c) 2014 Joyent, Inc. All rights reserved.
26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 * Copyright (c) 2016 Actifio, Inc. All rights reserved.
28 * Copyright (c) 2018, loli10K <ezomori.nozomu@gmail.com>. All rights reserved.
32 #include <sys/dmu_objset.h>
33 #include <sys/dmu_tx.h>
34 #include <sys/dsl_dataset.h>
35 #include <sys/dsl_dir.h>
36 #include <sys/dsl_prop.h>
37 #include <sys/dsl_synctask.h>
38 #include <sys/dsl_deleg.h>
39 #include <sys/dmu_impl.h>
41 #include <sys/spa_impl.h>
42 #include <sys/metaslab.h>
46 #include <sys/sunddi.h>
47 #include <sys/zfeature.h>
48 #include <sys/policy.h>
49 #include <sys/zfs_vfsops.h>
50 #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 initialized 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().
124 extern inline dsl_dir_phys_t *dsl_dir_phys(dsl_dir_t *dd);
126 static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
128 typedef struct ddulrt_arg {
129 dsl_dir_t *ddulrta_dd;
134 dsl_dir_evict_async(void *dbu)
138 dsl_pool_t *dp __maybe_unused = dd->dd_pool;
142 for (t = 0; t < TXG_SIZE; t++) {
143 ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t));
144 ASSERT(dd->dd_tempreserved[t] == 0);
145 ASSERT(dd->dd_space_towrite[t] == 0);
149 dsl_dir_async_rele(dd->dd_parent, dd);
151 spa_async_close(dd->dd_pool->dp_spa, dd);
153 if (dsl_deadlist_is_open(&dd->dd_livelist))
154 dsl_dir_livelist_close(dd);
157 cv_destroy(&dd->dd_activity_cv);
158 mutex_destroy(&dd->dd_activity_lock);
159 mutex_destroy(&dd->dd_lock);
160 kmem_free(dd, sizeof (dsl_dir_t));
164 dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj,
165 const char *tail, void *tag, dsl_dir_t **ddp)
169 dmu_object_info_t doi;
172 ASSERT(dsl_pool_config_held(dp));
174 err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
177 dd = dmu_buf_get_user(dbuf);
179 dmu_object_info_from_db(dbuf, &doi);
180 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_DSL_DIR);
181 ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
186 dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
187 dd->dd_object = ddobj;
191 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
192 mutex_init(&dd->dd_activity_lock, NULL, MUTEX_DEFAULT, NULL);
193 cv_init(&dd->dd_activity_cv, NULL, CV_DEFAULT, NULL);
196 if (dsl_dir_is_zapified(dd)) {
197 err = zap_lookup(dp->dp_meta_objset,
198 ddobj, DD_FIELD_CRYPTO_KEY_OBJ,
199 sizeof (uint64_t), 1, &dd->dd_crypto_obj);
201 /* check for on-disk format errata */
202 if (dsl_dir_incompatible_encryption_version(
204 dp->dp_spa->spa_errata =
205 ZPOOL_ERRATA_ZOL_6845_ENCRYPTION;
207 } else if (err != ENOENT) {
212 dsl_dir_snap_cmtime_update(dd);
214 if (dsl_dir_phys(dd)->dd_parent_obj) {
215 err = dsl_dir_hold_obj(dp,
216 dsl_dir_phys(dd)->dd_parent_obj, NULL, dd,
224 err = zap_lookup(dp->dp_meta_objset,
225 dsl_dir_phys(dd->dd_parent)->
226 dd_child_dir_zapobj, tail,
227 sizeof (foundobj), 1, &foundobj);
228 ASSERT(err || foundobj == ddobj);
230 (void) strlcpy(dd->dd_myname, tail,
231 sizeof (dd->dd_myname));
233 err = zap_value_search(dp->dp_meta_objset,
234 dsl_dir_phys(dd->dd_parent)->
236 ddobj, 0, dd->dd_myname);
241 (void) strlcpy(dd->dd_myname, spa_name(dp->dp_spa),
242 sizeof (dd->dd_myname));
245 if (dsl_dir_is_clone(dd)) {
246 dmu_buf_t *origin_bonus;
247 dsl_dataset_phys_t *origin_phys;
250 * We can't open the origin dataset, because
251 * that would require opening this dsl_dir.
252 * Just look at its phys directly instead.
254 err = dmu_bonus_hold(dp->dp_meta_objset,
255 dsl_dir_phys(dd)->dd_origin_obj, FTAG,
259 origin_phys = origin_bonus->db_data;
261 origin_phys->ds_creation_txg;
262 dmu_buf_rele(origin_bonus, FTAG);
263 if (dsl_dir_is_zapified(dd)) {
265 err = zap_lookup(dp->dp_meta_objset,
266 dd->dd_object, DD_FIELD_LIVELIST,
267 sizeof (uint64_t), 1, &obj);
269 dsl_dir_livelist_open(dd, obj);
270 else if (err != ENOENT)
275 dmu_buf_init_user(&dd->dd_dbu, NULL, dsl_dir_evict_async,
277 winner = dmu_buf_set_user_ie(dbuf, &dd->dd_dbu);
278 if (winner != NULL) {
280 dsl_dir_rele(dd->dd_parent, dd);
281 if (dsl_deadlist_is_open(&dd->dd_livelist))
282 dsl_dir_livelist_close(dd);
284 cv_destroy(&dd->dd_activity_cv);
285 mutex_destroy(&dd->dd_activity_lock);
286 mutex_destroy(&dd->dd_lock);
287 kmem_free(dd, sizeof (dsl_dir_t));
290 spa_open_ref(dp->dp_spa, dd);
295 * The dsl_dir_t has both open-to-close and instantiate-to-evict
296 * holds on the spa. We need the open-to-close holds because
297 * otherwise the spa_refcnt wouldn't change when we open a
298 * dir which the spa also has open, so we could incorrectly
299 * think it was OK to unload/export/destroy the pool. We need
300 * the instantiate-to-evict hold because the dsl_dir_t has a
301 * pointer to the dd_pool, which has a pointer to the spa_t.
303 spa_open_ref(dp->dp_spa, tag);
304 ASSERT3P(dd->dd_pool, ==, dp);
305 ASSERT3U(dd->dd_object, ==, ddobj);
306 ASSERT3P(dd->dd_dbuf, ==, dbuf);
312 dsl_dir_rele(dd->dd_parent, dd);
313 if (dsl_deadlist_is_open(&dd->dd_livelist))
314 dsl_dir_livelist_close(dd);
316 cv_destroy(&dd->dd_activity_cv);
317 mutex_destroy(&dd->dd_activity_lock);
318 mutex_destroy(&dd->dd_lock);
319 kmem_free(dd, sizeof (dsl_dir_t));
320 dmu_buf_rele(dbuf, tag);
325 dsl_dir_rele(dsl_dir_t *dd, void *tag)
327 dprintf_dd(dd, "%s\n", "");
328 spa_close(dd->dd_pool->dp_spa, tag);
329 dmu_buf_rele(dd->dd_dbuf, tag);
333 * Remove a reference to the given dsl dir that is being asynchronously
334 * released. Async releases occur from a taskq performing eviction of
335 * dsl datasets and dirs. This process is identical to a normal release
336 * with the exception of using the async API for releasing the reference on
340 dsl_dir_async_rele(dsl_dir_t *dd, void *tag)
342 dprintf_dd(dd, "%s\n", "");
343 spa_async_close(dd->dd_pool->dp_spa, tag);
344 dmu_buf_rele(dd->dd_dbuf, tag);
347 /* buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes */
349 dsl_dir_name(dsl_dir_t *dd, char *buf)
352 dsl_dir_name(dd->dd_parent, buf);
353 VERIFY3U(strlcat(buf, "/", ZFS_MAX_DATASET_NAME_LEN), <,
354 ZFS_MAX_DATASET_NAME_LEN);
358 if (!MUTEX_HELD(&dd->dd_lock)) {
360 * recursive mutex so that we can use
361 * dprintf_dd() with dd_lock held
363 mutex_enter(&dd->dd_lock);
364 VERIFY3U(strlcat(buf, dd->dd_myname, ZFS_MAX_DATASET_NAME_LEN),
365 <, ZFS_MAX_DATASET_NAME_LEN);
366 mutex_exit(&dd->dd_lock);
368 VERIFY3U(strlcat(buf, dd->dd_myname, ZFS_MAX_DATASET_NAME_LEN),
369 <, ZFS_MAX_DATASET_NAME_LEN);
373 /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
375 dsl_dir_namelen(dsl_dir_t *dd)
380 /* parent's name + 1 for the "/" */
381 result = dsl_dir_namelen(dd->dd_parent) + 1;
384 if (!MUTEX_HELD(&dd->dd_lock)) {
385 /* see dsl_dir_name */
386 mutex_enter(&dd->dd_lock);
387 result += strlen(dd->dd_myname);
388 mutex_exit(&dd->dd_lock);
390 result += strlen(dd->dd_myname);
397 getcomponent(const char *path, char *component, const char **nextp)
401 if ((path == NULL) || (path[0] == '\0'))
402 return (SET_ERROR(ENOENT));
403 /* This would be a good place to reserve some namespace... */
404 p = strpbrk(path, "/@");
405 if (p && (p[1] == '/' || p[1] == '@')) {
406 /* two separators in a row */
407 return (SET_ERROR(EINVAL));
409 if (p == NULL || p == path) {
411 * if the first thing is an @ or /, it had better be an
412 * @ and it had better not have any more ats or slashes,
413 * and it had better have something after the @.
416 (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
417 return (SET_ERROR(EINVAL));
418 if (strlen(path) >= ZFS_MAX_DATASET_NAME_LEN)
419 return (SET_ERROR(ENAMETOOLONG));
420 (void) strlcpy(component, path, ZFS_MAX_DATASET_NAME_LEN);
422 } else if (p[0] == '/') {
423 if (p - path >= ZFS_MAX_DATASET_NAME_LEN)
424 return (SET_ERROR(ENAMETOOLONG));
425 (void) strncpy(component, path, p - path);
426 component[p - path] = '\0';
428 } else if (p[0] == '@') {
430 * if the next separator is an @, there better not be
433 if (strchr(path, '/'))
434 return (SET_ERROR(EINVAL));
435 if (p - path >= ZFS_MAX_DATASET_NAME_LEN)
436 return (SET_ERROR(ENAMETOOLONG));
437 (void) strncpy(component, path, p - path);
438 component[p - path] = '\0';
440 panic("invalid p=%p", (void *)p);
447 * Return the dsl_dir_t, and possibly the last component which couldn't
448 * be found in *tail. The name must be in the specified dsl_pool_t. This
449 * thread must hold the dp_config_rwlock for the pool. Returns NULL if the
450 * path is bogus, or if tail==NULL and we couldn't parse the whole name.
451 * (*tail)[0] == '@' means that the last component is a snapshot.
454 dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag,
455 dsl_dir_t **ddp, const char **tailp)
458 const char *spaname, *next, *nextnext = NULL;
463 buf = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
464 err = getcomponent(name, buf, &next);
468 /* Make sure the name is in the specified pool. */
469 spaname = spa_name(dp->dp_spa);
470 if (strcmp(buf, spaname) != 0) {
471 err = SET_ERROR(EXDEV);
475 ASSERT(dsl_pool_config_held(dp));
477 err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
482 while (next != NULL) {
484 err = getcomponent(next, buf, &nextnext);
487 ASSERT(next[0] != '\0');
490 dprintf("looking up %s in obj%lld\n",
491 buf, (longlong_t)dsl_dir_phys(dd)->dd_child_dir_zapobj);
493 err = zap_lookup(dp->dp_meta_objset,
494 dsl_dir_phys(dd)->dd_child_dir_zapobj,
495 buf, sizeof (ddobj), 1, &ddobj);
502 err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_dd);
505 dsl_dir_rele(dd, tag);
511 dsl_dir_rele(dd, tag);
516 * It's an error if there's more than one component left, or
517 * tailp==NULL and there's any component left.
520 (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
522 dsl_dir_rele(dd, tag);
523 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
524 err = SET_ERROR(ENOENT);
531 kmem_free(buf, ZFS_MAX_DATASET_NAME_LEN);
536 * If the counts are already initialized for this filesystem and its
537 * descendants then do nothing, otherwise initialize the counts.
539 * The counts on this filesystem, and those below, may be uninitialized due to
540 * either the use of a pre-existing pool which did not support the
541 * filesystem/snapshot limit feature, or one in which the feature had not yet
544 * Recursively descend the filesystem tree and update the filesystem/snapshot
545 * counts on each filesystem below, then update the cumulative count on the
546 * current filesystem. If the filesystem already has a count set on it,
547 * then we know that its counts, and the counts on the filesystems below it,
548 * are already correct, so we don't have to update this filesystem.
551 dsl_dir_init_fs_ss_count(dsl_dir_t *dd, dmu_tx_t *tx)
553 uint64_t my_fs_cnt = 0;
554 uint64_t my_ss_cnt = 0;
555 dsl_pool_t *dp = dd->dd_pool;
556 objset_t *os = dp->dp_meta_objset;
561 ASSERT(spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT));
562 ASSERT(dsl_pool_config_held(dp));
563 ASSERT(dmu_tx_is_syncing(tx));
565 dsl_dir_zapify(dd, tx);
568 * If the filesystem count has already been initialized then we
569 * don't need to recurse down any further.
571 if (zap_contains(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT) == 0)
574 zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP);
575 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
577 /* Iterate my child dirs */
578 for (zap_cursor_init(zc, os, dsl_dir_phys(dd)->dd_child_dir_zapobj);
579 zap_cursor_retrieve(zc, za) == 0; zap_cursor_advance(zc)) {
583 VERIFY0(dsl_dir_hold_obj(dp, za->za_first_integer, NULL, FTAG,
587 * Ignore hidden ($FREE, $MOS & $ORIGIN) objsets.
589 if (chld_dd->dd_myname[0] == '$') {
590 dsl_dir_rele(chld_dd, FTAG);
594 my_fs_cnt++; /* count this child */
596 dsl_dir_init_fs_ss_count(chld_dd, tx);
598 VERIFY0(zap_lookup(os, chld_dd->dd_object,
599 DD_FIELD_FILESYSTEM_COUNT, sizeof (count), 1, &count));
601 VERIFY0(zap_lookup(os, chld_dd->dd_object,
602 DD_FIELD_SNAPSHOT_COUNT, sizeof (count), 1, &count));
605 dsl_dir_rele(chld_dd, FTAG);
608 /* Count my snapshots (we counted children's snapshots above) */
609 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
610 dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds));
612 for (zap_cursor_init(zc, os, dsl_dataset_phys(ds)->ds_snapnames_zapobj);
613 zap_cursor_retrieve(zc, za) == 0;
614 zap_cursor_advance(zc)) {
615 /* Don't count temporary snapshots */
616 if (za->za_name[0] != '%')
621 dsl_dataset_rele(ds, FTAG);
623 kmem_free(zc, sizeof (zap_cursor_t));
624 kmem_free(za, sizeof (zap_attribute_t));
626 /* we're in a sync task, update counts */
627 dmu_buf_will_dirty(dd->dd_dbuf, tx);
628 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
629 sizeof (my_fs_cnt), 1, &my_fs_cnt, tx));
630 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
631 sizeof (my_ss_cnt), 1, &my_ss_cnt, tx));
635 dsl_dir_actv_fs_ss_limit_check(void *arg, dmu_tx_t *tx)
637 char *ddname = (char *)arg;
638 dsl_pool_t *dp = dmu_tx_pool(tx);
643 error = dsl_dataset_hold(dp, ddname, FTAG, &ds);
647 if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
648 dsl_dataset_rele(ds, FTAG);
649 return (SET_ERROR(ENOTSUP));
653 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT) &&
654 dsl_dir_is_zapified(dd) &&
655 zap_contains(dp->dp_meta_objset, dd->dd_object,
656 DD_FIELD_FILESYSTEM_COUNT) == 0) {
657 dsl_dataset_rele(ds, FTAG);
658 return (SET_ERROR(EALREADY));
661 dsl_dataset_rele(ds, FTAG);
666 dsl_dir_actv_fs_ss_limit_sync(void *arg, dmu_tx_t *tx)
668 char *ddname = (char *)arg;
669 dsl_pool_t *dp = dmu_tx_pool(tx);
673 VERIFY0(dsl_dataset_hold(dp, ddname, FTAG, &ds));
675 spa = dsl_dataset_get_spa(ds);
677 if (!spa_feature_is_active(spa, SPA_FEATURE_FS_SS_LIMIT)) {
679 * Since the feature was not active and we're now setting a
680 * limit, increment the feature-active counter so that the
681 * feature becomes active for the first time.
683 * We are already in a sync task so we can update the MOS.
685 spa_feature_incr(spa, SPA_FEATURE_FS_SS_LIMIT, tx);
689 * Since we are now setting a non-UINT64_MAX limit on the filesystem,
690 * we need to ensure the counts are correct. Descend down the tree from
691 * this point and update all of the counts to be accurate.
693 dsl_dir_init_fs_ss_count(ds->ds_dir, tx);
695 dsl_dataset_rele(ds, FTAG);
699 * Make sure the feature is enabled and activate it if necessary.
700 * Since we're setting a limit, ensure the on-disk counts are valid.
701 * This is only called by the ioctl path when setting a limit value.
703 * We do not need to validate the new limit, since users who can change the
704 * limit are also allowed to exceed the limit.
707 dsl_dir_activate_fs_ss_limit(const char *ddname)
711 error = dsl_sync_task(ddname, dsl_dir_actv_fs_ss_limit_check,
712 dsl_dir_actv_fs_ss_limit_sync, (void *)ddname, 0,
713 ZFS_SPACE_CHECK_RESERVED);
715 if (error == EALREADY)
722 * Used to determine if the filesystem_limit or snapshot_limit should be
723 * enforced. We allow the limit to be exceeded if the user has permission to
724 * write the property value. We pass in the creds that we got in the open
725 * context since we will always be the GZ root in syncing context. We also have
726 * to handle the case where we are allowed to change the limit on the current
727 * dataset, but there may be another limit in the tree above.
729 * We can never modify these two properties within a non-global zone. In
730 * addition, the other checks are modeled on zfs_secpolicy_write_perms. We
731 * can't use that function since we are already holding the dp_config_rwlock.
732 * In addition, we already have the dd and dealing with snapshots is simplified
743 dsl_enforce_ds_ss_limits(dsl_dir_t *dd, zfs_prop_t prop,
744 cred_t *cr, proc_t *proc)
746 enforce_res_t enforce = ENFORCE_ALWAYS;
750 const char *zonedstr;
752 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
753 prop == ZFS_PROP_SNAPSHOT_LIMIT);
756 if (crgetzoneid(cr) != GLOBAL_ZONEID)
757 return (ENFORCE_ALWAYS);
760 * We are checking the saved credentials of the user process, which is
761 * not the current process. Note that we can't use secpolicy_zfs(),
762 * because it only works if the cred is that of the current process (on
765 if (secpolicy_zfs_proc(cr, proc) == 0)
766 return (ENFORCE_NEVER);
771 if ((obj = dsl_dir_phys(dd)->dd_head_dataset_obj) == 0)
772 return (ENFORCE_ALWAYS);
774 ASSERT(dsl_pool_config_held(dd->dd_pool));
776 if (dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds) != 0)
777 return (ENFORCE_ALWAYS);
779 zonedstr = zfs_prop_to_name(ZFS_PROP_ZONED);
780 if (dsl_prop_get_ds(ds, zonedstr, 8, 1, &zoned, NULL) || zoned) {
781 /* Only root can access zoned fs's from the GZ */
782 enforce = ENFORCE_ALWAYS;
784 if (dsl_deleg_access_impl(ds, zfs_prop_to_name(prop), cr) == 0)
785 enforce = ENFORCE_ABOVE;
788 dsl_dataset_rele(ds, FTAG);
793 * Check if adding additional child filesystem(s) would exceed any filesystem
794 * limits or adding additional snapshot(s) would exceed any snapshot limits.
795 * The prop argument indicates which limit to check.
797 * Note that all filesystem limits up to the root (or the highest
798 * initialized) filesystem or the given ancestor must be satisfied.
801 dsl_fs_ss_limit_check(dsl_dir_t *dd, uint64_t delta, zfs_prop_t prop,
802 dsl_dir_t *ancestor, cred_t *cr, proc_t *proc)
804 objset_t *os = dd->dd_pool->dp_meta_objset;
805 uint64_t limit, count;
807 enforce_res_t enforce;
810 ASSERT(dsl_pool_config_held(dd->dd_pool));
811 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
812 prop == ZFS_PROP_SNAPSHOT_LIMIT);
815 * If we're allowed to change the limit, don't enforce the limit
816 * e.g. this can happen if a snapshot is taken by an administrative
817 * user in the global zone (i.e. a recursive snapshot by root).
818 * However, we must handle the case of delegated permissions where we
819 * are allowed to change the limit on the current dataset, but there
820 * is another limit in the tree above.
822 enforce = dsl_enforce_ds_ss_limits(dd, prop, cr, proc);
823 if (enforce == ENFORCE_NEVER)
827 * e.g. if renaming a dataset with no snapshots, count adjustment
833 if (prop == ZFS_PROP_SNAPSHOT_LIMIT) {
835 * We don't enforce the limit for temporary snapshots. This is
836 * indicated by a NULL cred_t argument.
841 count_prop = DD_FIELD_SNAPSHOT_COUNT;
843 count_prop = DD_FIELD_FILESYSTEM_COUNT;
847 * If an ancestor has been provided, stop checking the limit once we
848 * hit that dir. We need this during rename so that we don't overcount
849 * the check once we recurse up to the common ancestor.
855 * If we hit an uninitialized node while recursing up the tree, we can
856 * stop since we know there is no limit here (or above). The counts are
857 * not valid on this node and we know we won't touch this node's counts.
859 if (!dsl_dir_is_zapified(dd))
861 err = zap_lookup(os, dd->dd_object,
862 count_prop, sizeof (count), 1, &count);
868 err = dsl_prop_get_dd(dd, zfs_prop_to_name(prop), 8, 1, &limit, NULL,
873 /* Is there a limit which we've hit? */
874 if (enforce == ENFORCE_ALWAYS && (count + delta) > limit)
875 return (SET_ERROR(EDQUOT));
877 if (dd->dd_parent != NULL)
878 err = dsl_fs_ss_limit_check(dd->dd_parent, delta, prop,
885 * Adjust the filesystem or snapshot count for the specified dsl_dir_t and all
886 * parents. When a new filesystem/snapshot is created, increment the count on
887 * all parents, and when a filesystem/snapshot is destroyed, decrement the
891 dsl_fs_ss_count_adjust(dsl_dir_t *dd, int64_t delta, const char *prop,
895 objset_t *os = dd->dd_pool->dp_meta_objset;
898 ASSERT(dsl_pool_config_held(dd->dd_pool));
899 ASSERT(dmu_tx_is_syncing(tx));
900 ASSERT(strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0 ||
901 strcmp(prop, DD_FIELD_SNAPSHOT_COUNT) == 0);
904 * We don't do accounting for hidden ($FREE, $MOS & $ORIGIN) objsets.
906 if (dd->dd_myname[0] == '$' && strcmp(prop,
907 DD_FIELD_FILESYSTEM_COUNT) == 0) {
912 * e.g. if renaming a dataset with no snapshots, count adjustment is 0
918 * If we hit an uninitialized node while recursing up the tree, we can
919 * stop since we know the counts are not valid on this node and we
920 * know we shouldn't touch this node's counts. An uninitialized count
921 * on the node indicates that either the feature has not yet been
922 * activated or there are no limits on this part of the tree.
924 if (!dsl_dir_is_zapified(dd) || (err = zap_lookup(os, dd->dd_object,
925 prop, sizeof (count), 1, &count)) == ENOENT)
930 /* Use a signed verify to make sure we're not neg. */
931 VERIFY3S(count, >=, 0);
933 VERIFY0(zap_update(os, dd->dd_object, prop, sizeof (count), 1, &count,
936 /* Roll up this additional count into our ancestors */
937 if (dd->dd_parent != NULL)
938 dsl_fs_ss_count_adjust(dd->dd_parent, delta, prop, tx);
942 dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
945 objset_t *mos = dp->dp_meta_objset;
947 dsl_dir_phys_t *ddphys;
950 ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
951 DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
953 VERIFY0(zap_add(mos, dsl_dir_phys(pds)->dd_child_dir_zapobj,
954 name, sizeof (uint64_t), 1, &ddobj, tx));
956 /* it's the root dir */
957 VERIFY0(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
958 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
960 VERIFY0(dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
961 dmu_buf_will_dirty(dbuf, tx);
962 ddphys = dbuf->db_data;
964 ddphys->dd_creation_time = gethrestime_sec();
966 ddphys->dd_parent_obj = pds->dd_object;
968 /* update the filesystem counts */
969 dsl_fs_ss_count_adjust(pds, 1, DD_FIELD_FILESYSTEM_COUNT, tx);
971 ddphys->dd_props_zapobj = zap_create(mos,
972 DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
973 ddphys->dd_child_dir_zapobj = zap_create(mos,
974 DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
975 if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
976 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
978 dmu_buf_rele(dbuf, FTAG);
984 dsl_dir_is_clone(dsl_dir_t *dd)
986 return (dsl_dir_phys(dd)->dd_origin_obj &&
987 (dd->dd_pool->dp_origin_snap == NULL ||
988 dsl_dir_phys(dd)->dd_origin_obj !=
989 dd->dd_pool->dp_origin_snap->ds_object));
993 dsl_dir_get_used(dsl_dir_t *dd)
995 return (dsl_dir_phys(dd)->dd_used_bytes);
999 dsl_dir_get_compressed(dsl_dir_t *dd)
1001 return (dsl_dir_phys(dd)->dd_compressed_bytes);
1005 dsl_dir_get_quota(dsl_dir_t *dd)
1007 return (dsl_dir_phys(dd)->dd_quota);
1011 dsl_dir_get_reservation(dsl_dir_t *dd)
1013 return (dsl_dir_phys(dd)->dd_reserved);
1017 dsl_dir_get_compressratio(dsl_dir_t *dd)
1019 /* a fixed point number, 100x the ratio */
1020 return (dsl_dir_phys(dd)->dd_compressed_bytes == 0 ? 100 :
1021 (dsl_dir_phys(dd)->dd_uncompressed_bytes * 100 /
1022 dsl_dir_phys(dd)->dd_compressed_bytes));
1026 dsl_dir_get_logicalused(dsl_dir_t *dd)
1028 return (dsl_dir_phys(dd)->dd_uncompressed_bytes);
1032 dsl_dir_get_usedsnap(dsl_dir_t *dd)
1034 return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]);
1038 dsl_dir_get_usedds(dsl_dir_t *dd)
1040 return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_HEAD]);
1044 dsl_dir_get_usedrefreserv(dsl_dir_t *dd)
1046 return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_REFRSRV]);
1050 dsl_dir_get_usedchild(dsl_dir_t *dd)
1052 return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD] +
1053 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD_RSRV]);
1057 dsl_dir_get_origin(dsl_dir_t *dd, char *buf)
1060 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
1061 dsl_dir_phys(dd)->dd_origin_obj, FTAG, &ds));
1063 dsl_dataset_name(ds, buf);
1065 dsl_dataset_rele(ds, FTAG);
1069 dsl_dir_get_filesystem_count(dsl_dir_t *dd, uint64_t *count)
1071 if (dsl_dir_is_zapified(dd)) {
1072 objset_t *os = dd->dd_pool->dp_meta_objset;
1073 return (zap_lookup(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
1074 sizeof (*count), 1, count));
1076 return (SET_ERROR(ENOENT));
1081 dsl_dir_get_snapshot_count(dsl_dir_t *dd, uint64_t *count)
1083 if (dsl_dir_is_zapified(dd)) {
1084 objset_t *os = dd->dd_pool->dp_meta_objset;
1085 return (zap_lookup(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
1086 sizeof (*count), 1, count));
1088 return (SET_ERROR(ENOENT));
1093 dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
1095 mutex_enter(&dd->dd_lock);
1096 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA,
1097 dsl_dir_get_quota(dd));
1098 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
1099 dsl_dir_get_reservation(dd));
1100 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED,
1101 dsl_dir_get_logicalused(dd));
1102 if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1103 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
1104 dsl_dir_get_usedsnap(dd));
1105 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
1106 dsl_dir_get_usedds(dd));
1107 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
1108 dsl_dir_get_usedrefreserv(dd));
1109 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
1110 dsl_dir_get_usedchild(dd));
1112 mutex_exit(&dd->dd_lock);
1115 if (dsl_dir_get_filesystem_count(dd, &count) == 0) {
1116 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_FILESYSTEM_COUNT,
1119 if (dsl_dir_get_snapshot_count(dd, &count) == 0) {
1120 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_SNAPSHOT_COUNT,
1124 if (dsl_dir_is_clone(dd)) {
1125 char buf[ZFS_MAX_DATASET_NAME_LEN];
1126 dsl_dir_get_origin(dd, buf);
1127 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
1133 dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
1135 dsl_pool_t *dp = dd->dd_pool;
1137 ASSERT(dsl_dir_phys(dd));
1139 if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) {
1140 /* up the hold count until we can be written out */
1141 dmu_buf_add_ref(dd->dd_dbuf, dd);
1146 parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
1148 uint64_t old_accounted = MAX(used, dsl_dir_phys(dd)->dd_reserved);
1149 uint64_t new_accounted =
1150 MAX(used + delta, dsl_dir_phys(dd)->dd_reserved);
1151 return (new_accounted - old_accounted);
1155 dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
1157 ASSERT(dmu_tx_is_syncing(tx));
1159 mutex_enter(&dd->dd_lock);
1160 ASSERT0(dd->dd_tempreserved[tx->tx_txg & TXG_MASK]);
1161 dprintf_dd(dd, "txg=%llu towrite=%lluK\n", (u_longlong_t)tx->tx_txg,
1162 (u_longlong_t)dd->dd_space_towrite[tx->tx_txg & TXG_MASK] / 1024);
1163 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] = 0;
1164 mutex_exit(&dd->dd_lock);
1166 /* release the hold from dsl_dir_dirty */
1167 dmu_buf_rele(dd->dd_dbuf, dd);
1171 dsl_dir_space_towrite(dsl_dir_t *dd)
1175 ASSERT(MUTEX_HELD(&dd->dd_lock));
1177 for (int i = 0; i < TXG_SIZE; i++) {
1178 space += dd->dd_space_towrite[i & TXG_MASK];
1179 ASSERT3U(dd->dd_space_towrite[i & TXG_MASK], >=, 0);
1185 * How much space would dd have available if ancestor had delta applied
1186 * to it? If ondiskonly is set, we're only interested in what's
1187 * on-disk, not estimated pending changes.
1190 dsl_dir_space_available(dsl_dir_t *dd,
1191 dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
1193 uint64_t parentspace, myspace, quota, used;
1196 * If there are no restrictions otherwise, assume we have
1197 * unlimited space available.
1200 parentspace = UINT64_MAX;
1202 if (dd->dd_parent != NULL) {
1203 parentspace = dsl_dir_space_available(dd->dd_parent,
1204 ancestor, delta, ondiskonly);
1207 mutex_enter(&dd->dd_lock);
1208 if (dsl_dir_phys(dd)->dd_quota != 0)
1209 quota = dsl_dir_phys(dd)->dd_quota;
1210 used = dsl_dir_phys(dd)->dd_used_bytes;
1212 used += dsl_dir_space_towrite(dd);
1214 if (dd->dd_parent == NULL) {
1215 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool,
1216 ZFS_SPACE_CHECK_NORMAL);
1217 quota = MIN(quota, poolsize);
1220 if (dsl_dir_phys(dd)->dd_reserved > used && parentspace != UINT64_MAX) {
1222 * We have some space reserved, in addition to what our
1225 parentspace += dsl_dir_phys(dd)->dd_reserved - used;
1228 if (dd == ancestor) {
1230 ASSERT(used >= -delta);
1232 if (parentspace != UINT64_MAX)
1233 parentspace -= delta;
1241 * the lesser of the space provided by our parent and
1242 * the space left in our quota
1244 myspace = MIN(parentspace, quota - used);
1247 mutex_exit(&dd->dd_lock);
1252 struct tempreserve {
1253 list_node_t tr_node;
1259 dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
1260 boolean_t ignorequota, list_t *tr_list,
1261 dmu_tx_t *tx, boolean_t first)
1265 struct tempreserve *tr;
1274 ASSERT3U(txg, !=, 0);
1275 ASSERT3S(asize, >, 0);
1277 mutex_enter(&dd->dd_lock);
1280 * Check against the dsl_dir's quota. We don't add in the delta
1281 * when checking for over-quota because they get one free hit.
1283 uint64_t est_inflight = dsl_dir_space_towrite(dd);
1284 for (int i = 0; i < TXG_SIZE; i++)
1285 est_inflight += dd->dd_tempreserved[i];
1286 uint64_t used_on_disk = dsl_dir_phys(dd)->dd_used_bytes;
1289 * On the first iteration, fetch the dataset's used-on-disk and
1290 * refreservation values. Also, if checkrefquota is set, test if
1291 * allocating this space would exceed the dataset's refquota.
1293 if (first && tx->tx_objset) {
1295 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
1297 error = dsl_dataset_check_quota(ds, !netfree,
1298 asize, est_inflight, &used_on_disk, &ref_rsrv);
1300 mutex_exit(&dd->dd_lock);
1301 DMU_TX_STAT_BUMP(dmu_tx_quota);
1307 * If this transaction will result in a net free of space,
1308 * we want to let it through.
1310 if (ignorequota || netfree || dsl_dir_phys(dd)->dd_quota == 0)
1313 quota = dsl_dir_phys(dd)->dd_quota;
1316 * Adjust the quota against the actual pool size at the root
1317 * minus any outstanding deferred frees.
1318 * To ensure that it's possible to remove files from a full
1319 * pool without inducing transient overcommits, we throttle
1320 * netfree transactions against a quota that is slightly larger,
1321 * but still within the pool's allocation slop. In cases where
1322 * we're very close to full, this will allow a steady trickle of
1323 * removes to get through.
1325 if (dd->dd_parent == NULL) {
1326 uint64_t avail = dsl_pool_unreserved_space(dd->dd_pool,
1328 ZFS_SPACE_CHECK_RESERVED : ZFS_SPACE_CHECK_NORMAL);
1330 if (avail < quota) {
1332 retval = SET_ERROR(ENOSPC);
1337 * If they are requesting more space, and our current estimate
1338 * is over quota, they get to try again unless the actual
1339 * on-disk is over quota and there are no pending changes (which
1340 * may free up space for us).
1342 if (used_on_disk + est_inflight >= quota) {
1343 if (est_inflight > 0 || used_on_disk < quota ||
1344 (retval == ENOSPC && used_on_disk < quota))
1346 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
1347 "quota=%lluK tr=%lluK err=%d\n",
1348 (u_longlong_t)used_on_disk>>10,
1349 (u_longlong_t)est_inflight>>10,
1350 (u_longlong_t)quota>>10, (u_longlong_t)asize>>10, retval);
1351 mutex_exit(&dd->dd_lock);
1352 DMU_TX_STAT_BUMP(dmu_tx_quota);
1353 return (SET_ERROR(retval));
1356 /* We need to up our estimated delta before dropping dd_lock */
1357 dd->dd_tempreserved[txg & TXG_MASK] += asize;
1359 uint64_t parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
1361 mutex_exit(&dd->dd_lock);
1363 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1365 tr->tr_size = asize;
1366 list_insert_tail(tr_list, tr);
1368 /* see if it's OK with our parent */
1369 if (dd->dd_parent != NULL && parent_rsrv != 0) {
1371 * Recurse on our parent without recursion. This has been
1372 * observed to be potentially large stack usage even within
1373 * the test suite. Largest seen stack was 7632 bytes on linux.
1377 asize = parent_rsrv;
1378 ignorequota = (dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
1380 goto top_of_function;
1388 * Reserve space in this dsl_dir, to be used in this tx's txg.
1389 * After the space has been dirtied (and dsl_dir_willuse_space()
1390 * has been called), the reservation should be canceled, using
1391 * dsl_dir_tempreserve_clear().
1394 dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
1395 boolean_t netfree, void **tr_cookiep, dmu_tx_t *tx)
1405 tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
1406 list_create(tr_list, sizeof (struct tempreserve),
1407 offsetof(struct tempreserve, tr_node));
1408 ASSERT3S(asize, >, 0);
1410 err = arc_tempreserve_space(dd->dd_pool->dp_spa, lsize, tx->tx_txg);
1412 struct tempreserve *tr;
1414 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1415 tr->tr_size = lsize;
1416 list_insert_tail(tr_list, tr);
1418 if (err == EAGAIN) {
1420 * If arc_memory_throttle() detected that pageout
1421 * is running and we are low on memory, we delay new
1422 * non-pageout transactions to give pageout an
1425 * It is unfortunate to be delaying while the caller's
1428 txg_delay(dd->dd_pool, tx->tx_txg,
1429 MSEC2NSEC(10), MSEC2NSEC(10));
1430 err = SET_ERROR(ERESTART);
1435 err = dsl_dir_tempreserve_impl(dd, asize, netfree,
1436 B_FALSE, tr_list, tx, B_TRUE);
1440 dsl_dir_tempreserve_clear(tr_list, tx);
1442 *tr_cookiep = tr_list;
1448 * Clear a temporary reservation that we previously made with
1449 * dsl_dir_tempreserve_space().
1452 dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
1454 int txgidx = tx->tx_txg & TXG_MASK;
1455 list_t *tr_list = tr_cookie;
1456 struct tempreserve *tr;
1458 ASSERT3U(tx->tx_txg, !=, 0);
1460 if (tr_cookie == NULL)
1463 while ((tr = list_head(tr_list)) != NULL) {
1465 mutex_enter(&tr->tr_ds->dd_lock);
1466 ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
1468 tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
1469 mutex_exit(&tr->tr_ds->dd_lock);
1471 arc_tempreserve_clear(tr->tr_size);
1473 list_remove(tr_list, tr);
1474 kmem_free(tr, sizeof (struct tempreserve));
1477 kmem_free(tr_list, sizeof (list_t));
1481 * This should be called from open context when we think we're going to write
1482 * or free space, for example when dirtying data. Be conservative; it's okay
1483 * to write less space or free more, but we don't want to write more or free
1484 * less than the amount specified.
1486 * NOTE: The behavior of this function is identical to the Illumos / FreeBSD
1487 * version however it has been adjusted to use an iterative rather than
1488 * recursive algorithm to minimize stack usage.
1491 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
1493 int64_t parent_space;
1497 mutex_enter(&dd->dd_lock);
1499 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
1501 est_used = dsl_dir_space_towrite(dd) +
1502 dsl_dir_phys(dd)->dd_used_bytes;
1503 parent_space = parent_delta(dd, est_used, space);
1504 mutex_exit(&dd->dd_lock);
1506 /* Make sure that we clean up dd_space_to* */
1507 dsl_dir_dirty(dd, tx);
1510 space = parent_space;
1511 } while (space && dd);
1514 /* call from syncing context when we actually write/free space for this dd */
1516 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
1517 int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
1519 int64_t accounted_delta;
1521 ASSERT(dmu_tx_is_syncing(tx));
1522 ASSERT(type < DD_USED_NUM);
1524 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1527 * dsl_dataset_set_refreservation_sync_impl() calls this with
1528 * dd_lock held, so that it can atomically update
1529 * ds->ds_reserved and the dsl_dir accounting, so that
1530 * dsl_dataset_check_quota() can see dataset and dir accounting
1533 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
1535 mutex_enter(&dd->dd_lock);
1536 dsl_dir_phys_t *ddp = dsl_dir_phys(dd);
1537 accounted_delta = parent_delta(dd, ddp->dd_used_bytes, used);
1538 ASSERT(used >= 0 || ddp->dd_used_bytes >= -used);
1539 ASSERT(compressed >= 0 || ddp->dd_compressed_bytes >= -compressed);
1540 ASSERT(uncompressed >= 0 ||
1541 ddp->dd_uncompressed_bytes >= -uncompressed);
1542 ddp->dd_used_bytes += used;
1543 ddp->dd_uncompressed_bytes += uncompressed;
1544 ddp->dd_compressed_bytes += compressed;
1546 if (ddp->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1547 ASSERT(used >= 0 || ddp->dd_used_breakdown[type] >= -used);
1548 ddp->dd_used_breakdown[type] += used;
1553 for (t = 0; t < DD_USED_NUM; t++)
1554 u += ddp->dd_used_breakdown[t];
1555 ASSERT3U(u, ==, ddp->dd_used_bytes);
1560 mutex_exit(&dd->dd_lock);
1562 if (dd->dd_parent != NULL) {
1563 dsl_dir_diduse_transfer_space(dd->dd_parent,
1564 accounted_delta, compressed, uncompressed,
1565 used, DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
1570 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
1571 dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
1573 ASSERT(dmu_tx_is_syncing(tx));
1574 ASSERT(oldtype < DD_USED_NUM);
1575 ASSERT(newtype < DD_USED_NUM);
1577 dsl_dir_phys_t *ddp = dsl_dir_phys(dd);
1579 !(ddp->dd_flags & DD_FLAG_USED_BREAKDOWN))
1582 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1583 mutex_enter(&dd->dd_lock);
1585 ddp->dd_used_breakdown[oldtype] >= delta :
1586 ddp->dd_used_breakdown[newtype] >= -delta);
1587 ASSERT(ddp->dd_used_bytes >= ABS(delta));
1588 ddp->dd_used_breakdown[oldtype] -= delta;
1589 ddp->dd_used_breakdown[newtype] += delta;
1590 mutex_exit(&dd->dd_lock);
1594 dsl_dir_diduse_transfer_space(dsl_dir_t *dd, int64_t used,
1595 int64_t compressed, int64_t uncompressed, int64_t tonew,
1596 dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
1598 int64_t accounted_delta;
1600 ASSERT(dmu_tx_is_syncing(tx));
1601 ASSERT(oldtype < DD_USED_NUM);
1602 ASSERT(newtype < DD_USED_NUM);
1604 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1606 mutex_enter(&dd->dd_lock);
1607 dsl_dir_phys_t *ddp = dsl_dir_phys(dd);
1608 accounted_delta = parent_delta(dd, ddp->dd_used_bytes, used);
1609 ASSERT(used >= 0 || ddp->dd_used_bytes >= -used);
1610 ASSERT(compressed >= 0 || ddp->dd_compressed_bytes >= -compressed);
1611 ASSERT(uncompressed >= 0 ||
1612 ddp->dd_uncompressed_bytes >= -uncompressed);
1613 ddp->dd_used_bytes += used;
1614 ddp->dd_uncompressed_bytes += uncompressed;
1615 ddp->dd_compressed_bytes += compressed;
1617 if (ddp->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1618 ASSERT(tonew - used <= 0 ||
1619 ddp->dd_used_breakdown[oldtype] >= tonew - used);
1620 ASSERT(tonew >= 0 ||
1621 ddp->dd_used_breakdown[newtype] >= -tonew);
1622 ddp->dd_used_breakdown[oldtype] -= tonew - used;
1623 ddp->dd_used_breakdown[newtype] += tonew;
1628 for (t = 0; t < DD_USED_NUM; t++)
1629 u += ddp->dd_used_breakdown[t];
1630 ASSERT3U(u, ==, ddp->dd_used_bytes);
1634 mutex_exit(&dd->dd_lock);
1636 if (dd->dd_parent != NULL) {
1637 dsl_dir_diduse_transfer_space(dd->dd_parent,
1638 accounted_delta, compressed, uncompressed,
1639 used, DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
1643 typedef struct dsl_dir_set_qr_arg {
1644 const char *ddsqra_name;
1645 zprop_source_t ddsqra_source;
1646 uint64_t ddsqra_value;
1647 } dsl_dir_set_qr_arg_t;
1650 dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx)
1652 dsl_dir_set_qr_arg_t *ddsqra = arg;
1653 dsl_pool_t *dp = dmu_tx_pool(tx);
1656 uint64_t towrite, newval;
1658 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1662 error = dsl_prop_predict(ds->ds_dir, "quota",
1663 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1665 dsl_dataset_rele(ds, FTAG);
1670 dsl_dataset_rele(ds, FTAG);
1674 mutex_enter(&ds->ds_dir->dd_lock);
1676 * If we are doing the preliminary check in open context, and
1677 * there are pending changes, then don't fail it, since the
1678 * pending changes could under-estimate the amount of space to be
1681 towrite = dsl_dir_space_towrite(ds->ds_dir);
1682 if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
1683 (newval < dsl_dir_phys(ds->ds_dir)->dd_reserved ||
1684 newval < dsl_dir_phys(ds->ds_dir)->dd_used_bytes + towrite)) {
1685 error = SET_ERROR(ENOSPC);
1687 mutex_exit(&ds->ds_dir->dd_lock);
1688 dsl_dataset_rele(ds, FTAG);
1693 dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx)
1695 dsl_dir_set_qr_arg_t *ddsqra = arg;
1696 dsl_pool_t *dp = dmu_tx_pool(tx);
1700 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1702 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1703 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA),
1704 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1705 &ddsqra->ddsqra_value, tx);
1707 VERIFY0(dsl_prop_get_int_ds(ds,
1708 zfs_prop_to_name(ZFS_PROP_QUOTA), &newval));
1710 newval = ddsqra->ddsqra_value;
1711 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1712 zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval);
1715 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1716 mutex_enter(&ds->ds_dir->dd_lock);
1717 dsl_dir_phys(ds->ds_dir)->dd_quota = newval;
1718 mutex_exit(&ds->ds_dir->dd_lock);
1719 dsl_dataset_rele(ds, FTAG);
1723 dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
1725 dsl_dir_set_qr_arg_t ddsqra;
1727 ddsqra.ddsqra_name = ddname;
1728 ddsqra.ddsqra_source = source;
1729 ddsqra.ddsqra_value = quota;
1731 return (dsl_sync_task(ddname, dsl_dir_set_quota_check,
1732 dsl_dir_set_quota_sync, &ddsqra, 0,
1733 ZFS_SPACE_CHECK_EXTRA_RESERVED));
1737 dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx)
1739 dsl_dir_set_qr_arg_t *ddsqra = arg;
1740 dsl_pool_t *dp = dmu_tx_pool(tx);
1743 uint64_t newval, used, avail;
1746 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1752 * If we are doing the preliminary check in open context, the
1753 * space estimates may be inaccurate.
1755 if (!dmu_tx_is_syncing(tx)) {
1756 dsl_dataset_rele(ds, FTAG);
1760 error = dsl_prop_predict(ds->ds_dir,
1761 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1762 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1764 dsl_dataset_rele(ds, FTAG);
1768 mutex_enter(&dd->dd_lock);
1769 used = dsl_dir_phys(dd)->dd_used_bytes;
1770 mutex_exit(&dd->dd_lock);
1772 if (dd->dd_parent) {
1773 avail = dsl_dir_space_available(dd->dd_parent,
1776 avail = dsl_pool_adjustedsize(dd->dd_pool,
1777 ZFS_SPACE_CHECK_NORMAL) - used;
1780 if (MAX(used, newval) > MAX(used, dsl_dir_phys(dd)->dd_reserved)) {
1781 uint64_t delta = MAX(used, newval) -
1782 MAX(used, dsl_dir_phys(dd)->dd_reserved);
1784 if (delta > avail ||
1785 (dsl_dir_phys(dd)->dd_quota > 0 &&
1786 newval > dsl_dir_phys(dd)->dd_quota))
1787 error = SET_ERROR(ENOSPC);
1790 dsl_dataset_rele(ds, FTAG);
1795 dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
1800 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1802 mutex_enter(&dd->dd_lock);
1803 used = dsl_dir_phys(dd)->dd_used_bytes;
1804 delta = MAX(used, value) - MAX(used, dsl_dir_phys(dd)->dd_reserved);
1805 dsl_dir_phys(dd)->dd_reserved = value;
1807 if (dd->dd_parent != NULL) {
1808 /* Roll up this additional usage into our ancestors */
1809 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1812 mutex_exit(&dd->dd_lock);
1816 dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx)
1818 dsl_dir_set_qr_arg_t *ddsqra = arg;
1819 dsl_pool_t *dp = dmu_tx_pool(tx);
1823 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1825 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1826 dsl_prop_set_sync_impl(ds,
1827 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1828 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1829 &ddsqra->ddsqra_value, tx);
1831 VERIFY0(dsl_prop_get_int_ds(ds,
1832 zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval));
1834 newval = ddsqra->ddsqra_value;
1835 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1836 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1837 (longlong_t)newval);
1840 dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx);
1841 dsl_dataset_rele(ds, FTAG);
1845 dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1846 uint64_t reservation)
1848 dsl_dir_set_qr_arg_t ddsqra;
1850 ddsqra.ddsqra_name = ddname;
1851 ddsqra.ddsqra_source = source;
1852 ddsqra.ddsqra_value = reservation;
1854 return (dsl_sync_task(ddname, dsl_dir_set_reservation_check,
1855 dsl_dir_set_reservation_sync, &ddsqra, 0,
1856 ZFS_SPACE_CHECK_EXTRA_RESERVED));
1860 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1862 for (; ds1; ds1 = ds1->dd_parent) {
1864 for (dd = ds2; dd; dd = dd->dd_parent) {
1873 * If delta is applied to dd, how much of that delta would be applied to
1874 * ancestor? Syncing context only.
1877 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1882 mutex_enter(&dd->dd_lock);
1883 delta = parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, delta);
1884 mutex_exit(&dd->dd_lock);
1885 return (would_change(dd->dd_parent, delta, ancestor));
1888 typedef struct dsl_dir_rename_arg {
1889 const char *ddra_oldname;
1890 const char *ddra_newname;
1893 } dsl_dir_rename_arg_t;
1895 typedef struct dsl_valid_rename_arg {
1898 } dsl_valid_rename_arg_t;
1901 dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1904 dsl_valid_rename_arg_t *dvra = arg;
1905 char namebuf[ZFS_MAX_DATASET_NAME_LEN];
1907 dsl_dataset_name(ds, namebuf);
1909 ASSERT3U(strnlen(namebuf, ZFS_MAX_DATASET_NAME_LEN),
1910 <, ZFS_MAX_DATASET_NAME_LEN);
1911 int namelen = strlen(namebuf) + dvra->char_delta;
1912 int depth = get_dataset_depth(namebuf) + dvra->nest_delta;
1914 if (namelen >= ZFS_MAX_DATASET_NAME_LEN)
1915 return (SET_ERROR(ENAMETOOLONG));
1916 if (dvra->nest_delta > 0 && depth >= zfs_max_dataset_nesting)
1917 return (SET_ERROR(ENAMETOOLONG));
1922 dsl_dir_rename_check(void *arg, dmu_tx_t *tx)
1924 dsl_dir_rename_arg_t *ddra = arg;
1925 dsl_pool_t *dp = dmu_tx_pool(tx);
1926 dsl_dir_t *dd, *newparent;
1927 dsl_valid_rename_arg_t dvra;
1928 dsl_dataset_t *parentds;
1930 const char *mynewname;
1933 /* target dir should exist */
1934 error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL);
1938 /* new parent should exist */
1939 error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG,
1940 &newparent, &mynewname);
1942 dsl_dir_rele(dd, FTAG);
1946 /* can't rename to different pool */
1947 if (dd->dd_pool != newparent->dd_pool) {
1948 dsl_dir_rele(newparent, FTAG);
1949 dsl_dir_rele(dd, FTAG);
1950 return (SET_ERROR(EXDEV));
1953 /* new name should not already exist */
1954 if (mynewname == NULL) {
1955 dsl_dir_rele(newparent, FTAG);
1956 dsl_dir_rele(dd, FTAG);
1957 return (SET_ERROR(EEXIST));
1960 /* can't rename below anything but filesystems (eg. no ZVOLs) */
1961 error = dsl_dataset_hold_obj(newparent->dd_pool,
1962 dsl_dir_phys(newparent)->dd_head_dataset_obj, FTAG, &parentds);
1964 dsl_dir_rele(newparent, FTAG);
1965 dsl_dir_rele(dd, FTAG);
1968 error = dmu_objset_from_ds(parentds, &parentos);
1970 dsl_dataset_rele(parentds, FTAG);
1971 dsl_dir_rele(newparent, FTAG);
1972 dsl_dir_rele(dd, FTAG);
1975 if (dmu_objset_type(parentos) != DMU_OST_ZFS) {
1976 dsl_dataset_rele(parentds, FTAG);
1977 dsl_dir_rele(newparent, FTAG);
1978 dsl_dir_rele(dd, FTAG);
1979 return (SET_ERROR(ZFS_ERR_WRONG_PARENT));
1981 dsl_dataset_rele(parentds, FTAG);
1983 ASSERT3U(strnlen(ddra->ddra_newname, ZFS_MAX_DATASET_NAME_LEN),
1984 <, ZFS_MAX_DATASET_NAME_LEN);
1985 ASSERT3U(strnlen(ddra->ddra_oldname, ZFS_MAX_DATASET_NAME_LEN),
1986 <, ZFS_MAX_DATASET_NAME_LEN);
1987 dvra.char_delta = strlen(ddra->ddra_newname)
1988 - strlen(ddra->ddra_oldname);
1989 dvra.nest_delta = get_dataset_depth(ddra->ddra_newname)
1990 - get_dataset_depth(ddra->ddra_oldname);
1992 /* if the name length is growing, validate child name lengths */
1993 if (dvra.char_delta > 0 || dvra.nest_delta > 0) {
1994 error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename,
1995 &dvra, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1997 dsl_dir_rele(newparent, FTAG);
1998 dsl_dir_rele(dd, FTAG);
2003 if (dmu_tx_is_syncing(tx)) {
2004 if (spa_feature_is_active(dp->dp_spa,
2005 SPA_FEATURE_FS_SS_LIMIT)) {
2007 * Although this is the check function and we don't
2008 * normally make on-disk changes in check functions,
2009 * we need to do that here.
2011 * Ensure this portion of the tree's counts have been
2012 * initialized in case the new parent has limits set.
2014 dsl_dir_init_fs_ss_count(dd, tx);
2018 if (newparent != dd->dd_parent) {
2019 /* is there enough space? */
2021 MAX(dsl_dir_phys(dd)->dd_used_bytes,
2022 dsl_dir_phys(dd)->dd_reserved);
2023 objset_t *os = dd->dd_pool->dp_meta_objset;
2024 uint64_t fs_cnt = 0;
2025 uint64_t ss_cnt = 0;
2027 if (dsl_dir_is_zapified(dd)) {
2030 err = zap_lookup(os, dd->dd_object,
2031 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
2033 if (err != ENOENT && err != 0) {
2034 dsl_dir_rele(newparent, FTAG);
2035 dsl_dir_rele(dd, FTAG);
2040 * have to add 1 for the filesystem itself that we're
2045 err = zap_lookup(os, dd->dd_object,
2046 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
2048 if (err != ENOENT && err != 0) {
2049 dsl_dir_rele(newparent, FTAG);
2050 dsl_dir_rele(dd, FTAG);
2055 /* check for encryption errors */
2056 error = dsl_dir_rename_crypt_check(dd, newparent);
2058 dsl_dir_rele(newparent, FTAG);
2059 dsl_dir_rele(dd, FTAG);
2060 return (SET_ERROR(EACCES));
2063 /* no rename into our descendant */
2064 if (closest_common_ancestor(dd, newparent) == dd) {
2065 dsl_dir_rele(newparent, FTAG);
2066 dsl_dir_rele(dd, FTAG);
2067 return (SET_ERROR(EINVAL));
2070 error = dsl_dir_transfer_possible(dd->dd_parent,
2071 newparent, fs_cnt, ss_cnt, myspace,
2072 ddra->ddra_cred, ddra->ddra_proc);
2074 dsl_dir_rele(newparent, FTAG);
2075 dsl_dir_rele(dd, FTAG);
2080 dsl_dir_rele(newparent, FTAG);
2081 dsl_dir_rele(dd, FTAG);
2086 dsl_dir_rename_sync(void *arg, dmu_tx_t *tx)
2088 dsl_dir_rename_arg_t *ddra = arg;
2089 dsl_pool_t *dp = dmu_tx_pool(tx);
2090 dsl_dir_t *dd, *newparent;
2091 const char *mynewname;
2092 objset_t *mos = dp->dp_meta_objset;
2094 VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL));
2095 VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
2098 /* Log this before we change the name. */
2099 spa_history_log_internal_dd(dd, "rename", tx,
2100 "-> %s", ddra->ddra_newname);
2102 if (newparent != dd->dd_parent) {
2103 objset_t *os = dd->dd_pool->dp_meta_objset;
2104 uint64_t fs_cnt = 0;
2105 uint64_t ss_cnt = 0;
2108 * We already made sure the dd counts were initialized in the
2111 if (spa_feature_is_active(dp->dp_spa,
2112 SPA_FEATURE_FS_SS_LIMIT)) {
2113 VERIFY0(zap_lookup(os, dd->dd_object,
2114 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
2116 /* add 1 for the filesystem itself that we're moving */
2119 VERIFY0(zap_lookup(os, dd->dd_object,
2120 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
2124 dsl_fs_ss_count_adjust(dd->dd_parent, -fs_cnt,
2125 DD_FIELD_FILESYSTEM_COUNT, tx);
2126 dsl_fs_ss_count_adjust(newparent, fs_cnt,
2127 DD_FIELD_FILESYSTEM_COUNT, tx);
2129 dsl_fs_ss_count_adjust(dd->dd_parent, -ss_cnt,
2130 DD_FIELD_SNAPSHOT_COUNT, tx);
2131 dsl_fs_ss_count_adjust(newparent, ss_cnt,
2132 DD_FIELD_SNAPSHOT_COUNT, tx);
2134 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
2135 -dsl_dir_phys(dd)->dd_used_bytes,
2136 -dsl_dir_phys(dd)->dd_compressed_bytes,
2137 -dsl_dir_phys(dd)->dd_uncompressed_bytes, tx);
2138 dsl_dir_diduse_space(newparent, DD_USED_CHILD,
2139 dsl_dir_phys(dd)->dd_used_bytes,
2140 dsl_dir_phys(dd)->dd_compressed_bytes,
2141 dsl_dir_phys(dd)->dd_uncompressed_bytes, tx);
2143 if (dsl_dir_phys(dd)->dd_reserved >
2144 dsl_dir_phys(dd)->dd_used_bytes) {
2145 uint64_t unused_rsrv = dsl_dir_phys(dd)->dd_reserved -
2146 dsl_dir_phys(dd)->dd_used_bytes;
2148 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
2149 -unused_rsrv, 0, 0, tx);
2150 dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV,
2151 unused_rsrv, 0, 0, tx);
2155 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2157 /* remove from old parent zapobj */
2158 VERIFY0(zap_remove(mos,
2159 dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj,
2160 dd->dd_myname, tx));
2162 (void) strlcpy(dd->dd_myname, mynewname,
2163 sizeof (dd->dd_myname));
2164 dsl_dir_rele(dd->dd_parent, dd);
2165 dsl_dir_phys(dd)->dd_parent_obj = newparent->dd_object;
2166 VERIFY0(dsl_dir_hold_obj(dp,
2167 newparent->dd_object, NULL, dd, &dd->dd_parent));
2169 /* add to new parent zapobj */
2170 VERIFY0(zap_add(mos, dsl_dir_phys(newparent)->dd_child_dir_zapobj,
2171 dd->dd_myname, 8, 1, &dd->dd_object, tx));
2173 /* TODO: A rename callback to avoid these layering violations. */
2174 zfsvfs_update_fromname(ddra->ddra_oldname, ddra->ddra_newname);
2175 zvol_rename_minors(dp->dp_spa, ddra->ddra_oldname,
2176 ddra->ddra_newname, B_TRUE);
2178 dsl_prop_notify_all(dd);
2180 dsl_dir_rele(newparent, FTAG);
2181 dsl_dir_rele(dd, FTAG);
2185 dsl_dir_rename(const char *oldname, const char *newname)
2187 dsl_dir_rename_arg_t ddra;
2189 ddra.ddra_oldname = oldname;
2190 ddra.ddra_newname = newname;
2191 ddra.ddra_cred = CRED();
2192 ddra.ddra_proc = curproc;
2194 return (dsl_sync_task(oldname,
2195 dsl_dir_rename_check, dsl_dir_rename_sync, &ddra,
2196 3, ZFS_SPACE_CHECK_RESERVED));
2200 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd,
2201 uint64_t fs_cnt, uint64_t ss_cnt, uint64_t space,
2202 cred_t *cr, proc_t *proc)
2204 dsl_dir_t *ancestor;
2209 ancestor = closest_common_ancestor(sdd, tdd);
2210 adelta = would_change(sdd, -space, ancestor);
2211 avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
2213 return (SET_ERROR(ENOSPC));
2215 err = dsl_fs_ss_limit_check(tdd, fs_cnt, ZFS_PROP_FILESYSTEM_LIMIT,
2216 ancestor, cr, proc);
2219 err = dsl_fs_ss_limit_check(tdd, ss_cnt, ZFS_PROP_SNAPSHOT_LIMIT,
2220 ancestor, cr, proc);
2228 dsl_dir_snap_cmtime(dsl_dir_t *dd)
2232 mutex_enter(&dd->dd_lock);
2233 t = dd->dd_snap_cmtime;
2234 mutex_exit(&dd->dd_lock);
2240 dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
2245 mutex_enter(&dd->dd_lock);
2246 dd->dd_snap_cmtime = t;
2247 mutex_exit(&dd->dd_lock);
2251 dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx)
2253 objset_t *mos = dd->dd_pool->dp_meta_objset;
2254 dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx);
2258 dsl_dir_is_zapified(dsl_dir_t *dd)
2260 dmu_object_info_t doi;
2262 dmu_object_info_from_db(dd->dd_dbuf, &doi);
2263 return (doi.doi_type == DMU_OTN_ZAP_METADATA);
2267 dsl_dir_livelist_open(dsl_dir_t *dd, uint64_t obj)
2269 objset_t *mos = dd->dd_pool->dp_meta_objset;
2270 ASSERT(spa_feature_is_active(dd->dd_pool->dp_spa,
2271 SPA_FEATURE_LIVELIST));
2272 dsl_deadlist_open(&dd->dd_livelist, mos, obj);
2273 bplist_create(&dd->dd_pending_allocs);
2274 bplist_create(&dd->dd_pending_frees);
2278 dsl_dir_livelist_close(dsl_dir_t *dd)
2280 dsl_deadlist_close(&dd->dd_livelist);
2281 bplist_destroy(&dd->dd_pending_allocs);
2282 bplist_destroy(&dd->dd_pending_frees);
2286 dsl_dir_remove_livelist(dsl_dir_t *dd, dmu_tx_t *tx, boolean_t total)
2289 dsl_pool_t *dp = dmu_tx_pool(tx);
2290 spa_t *spa = dp->dp_spa;
2291 livelist_condense_entry_t to_condense = spa->spa_to_condense;
2293 if (!dsl_deadlist_is_open(&dd->dd_livelist))
2297 * If the livelist being removed is set to be condensed, stop the
2298 * condense zthr and indicate the cancellation in the spa_to_condense
2299 * struct in case the condense no-wait synctask has already started
2301 zthr_t *ll_condense_thread = spa->spa_livelist_condense_zthr;
2302 if (ll_condense_thread != NULL &&
2303 (to_condense.ds != NULL) && (to_condense.ds->ds_dir == dd)) {
2305 * We use zthr_wait_cycle_done instead of zthr_cancel
2306 * because we don't want to destroy the zthr, just have
2307 * it skip its current task.
2309 spa->spa_to_condense.cancelled = B_TRUE;
2310 zthr_wait_cycle_done(ll_condense_thread);
2312 * If we've returned from zthr_wait_cycle_done without
2313 * clearing the to_condense data structure it's either
2314 * because the no-wait synctask has started (which is
2315 * indicated by 'syncing' field of to_condense) and we
2316 * can expect it to clear to_condense on its own.
2317 * Otherwise, we returned before the zthr ran. The
2318 * checkfunc will now fail as cancelled == B_TRUE so we
2319 * can safely NULL out ds, allowing a different dir's
2320 * livelist to be condensed.
2322 * We can be sure that the to_condense struct will not
2323 * be repopulated at this stage because both this
2324 * function and dsl_livelist_try_condense execute in
2327 if ((spa->spa_to_condense.ds != NULL) &&
2328 !spa->spa_to_condense.syncing) {
2329 dmu_buf_rele(spa->spa_to_condense.ds->ds_dbuf,
2331 spa->spa_to_condense.ds = NULL;
2335 dsl_dir_livelist_close(dd);
2336 VERIFY0(zap_lookup(dp->dp_meta_objset, dd->dd_object,
2337 DD_FIELD_LIVELIST, sizeof (uint64_t), 1, &obj));
2338 VERIFY0(zap_remove(dp->dp_meta_objset, dd->dd_object,
2339 DD_FIELD_LIVELIST, tx));
2341 dsl_deadlist_free(dp->dp_meta_objset, obj, tx);
2342 spa_feature_decr(spa, SPA_FEATURE_LIVELIST, tx);
2347 dsl_dir_activity_in_progress(dsl_dir_t *dd, dsl_dataset_t *ds,
2348 zfs_wait_activity_t activity, boolean_t *in_progress)
2352 ASSERT(MUTEX_HELD(&dd->dd_activity_lock));
2355 case ZFS_WAIT_DELETEQ: {
2358 error = dmu_objset_from_ds(ds, &os);
2362 mutex_enter(&os->os_user_ptr_lock);
2363 void *user = dmu_objset_get_user(os);
2364 mutex_exit(&os->os_user_ptr_lock);
2365 if (dmu_objset_type(os) != DMU_OST_ZFS ||
2366 user == NULL || zfs_get_vfs_flag_unmounted(os)) {
2367 *in_progress = B_FALSE;
2371 uint64_t readonly = B_FALSE;
2372 error = zfs_get_temporary_prop(ds, ZFS_PROP_READONLY, &readonly,
2378 if (readonly || !spa_writeable(dd->dd_pool->dp_spa)) {
2379 *in_progress = B_FALSE;
2383 uint64_t count, unlinked_obj;
2384 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
2387 dsl_dataset_rele(ds, FTAG);
2390 error = zap_count(os, unlinked_obj, &count);
2393 *in_progress = (count != 0);
2397 * The delete queue is ZPL specific, and libzpool doesn't have
2398 * it. It doesn't make sense to wait for it.
2401 *in_progress = B_FALSE;
2406 panic("unrecognized value for activity %d", activity);
2413 dsl_dir_wait(dsl_dir_t *dd, dsl_dataset_t *ds, zfs_wait_activity_t activity,
2417 boolean_t in_progress;
2418 dsl_pool_t *dp = dd->dd_pool;
2420 dsl_pool_config_enter(dp, FTAG);
2421 error = dsl_dir_activity_in_progress(dd, ds, activity,
2423 dsl_pool_config_exit(dp, FTAG);
2424 if (error != 0 || !in_progress)
2429 if (cv_wait_sig(&dd->dd_activity_cv, &dd->dd_activity_lock) ==
2430 0 || dd->dd_activity_cancelled) {
2431 error = SET_ERROR(EINTR);
2439 dsl_dir_cancel_waiters(dsl_dir_t *dd)
2441 mutex_enter(&dd->dd_activity_lock);
2442 dd->dd_activity_cancelled = B_TRUE;
2443 cv_broadcast(&dd->dd_activity_cv);
2444 while (dd->dd_activity_waiters > 0)
2445 cv_wait(&dd->dd_activity_cv, &dd->dd_activity_lock);
2446 mutex_exit(&dd->dd_activity_lock);
2449 #if defined(_KERNEL)
2450 EXPORT_SYMBOL(dsl_dir_set_quota);
2451 EXPORT_SYMBOL(dsl_dir_set_reservation);