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_znode.h>
52 #include "zfs_namecheck.h"
55 #include <sys/zfs_vfsops.h>
59 * Filesystem and Snapshot Limits
60 * ------------------------------
62 * These limits are used to restrict the number of filesystems and/or snapshots
63 * that can be created at a given level in the tree or below. A typical
64 * use-case is with a delegated dataset where the administrator wants to ensure
65 * that a user within the zone is not creating too many additional filesystems
66 * or snapshots, even though they're not exceeding their space quota.
68 * The filesystem and snapshot counts are stored as extensible properties. This
69 * capability is controlled by a feature flag and must be enabled to be used.
70 * Once enabled, the feature is not active until the first limit is set. At
71 * that point, future operations to create/destroy filesystems or snapshots
72 * will validate and update the counts.
74 * Because the count properties will not exist before the feature is active,
75 * the counts are updated when a limit is first set on an uninitialized
76 * dsl_dir node in the tree (The filesystem/snapshot count on a node includes
77 * all of the nested filesystems/snapshots. Thus, a new leaf node has a
78 * filesystem count of 0 and a snapshot count of 0. Non-existent filesystem and
79 * snapshot count properties on a node indicate uninitialized counts on that
80 * node.) When first setting a limit on an uninitialized node, the code starts
81 * at the filesystem with the new limit and descends into all sub-filesystems
82 * to add the count properties.
84 * In practice this is lightweight since a limit is typically set when the
85 * filesystem is created and thus has no children. Once valid, changing the
86 * limit value won't require a re-traversal since the counts are already valid.
87 * When recursively fixing the counts, if a node with a limit is encountered
88 * during the descent, the counts are known to be valid and there is no need to
89 * descend into that filesystem's children. The counts on filesystems above the
90 * one with the new limit will still be uninitialized, unless a limit is
91 * eventually set on one of those filesystems. The counts are always recursively
92 * updated when a limit is set on a dataset, unless there is already a limit.
93 * When a new limit value is set on a filesystem with an existing limit, it is
94 * possible for the new limit to be less than the current count at that level
95 * since a user who can change the limit is also allowed to exceed the limit.
97 * Once the feature is active, then whenever a filesystem or snapshot is
98 * created, the code recurses up the tree, validating the new count against the
99 * limit at each initialized level. In practice, most levels will not have a
100 * limit set. If there is a limit at any initialized level up the tree, the
101 * check must pass or the creation will fail. Likewise, when a filesystem or
102 * snapshot is destroyed, the counts are recursively adjusted all the way up
103 * the initialized nodes in the tree. Renaming a filesystem into different point
104 * in the tree will first validate, then update the counts on each branch up to
105 * the common ancestor. A receive will also validate the counts and then update
108 * An exception to the above behavior is that the limit is not enforced if the
109 * user has permission to modify the limit. This is primarily so that
110 * recursive snapshots in the global zone always work. We want to prevent a
111 * denial-of-service in which a lower level delegated dataset could max out its
112 * limit and thus block recursive snapshots from being taken in the global zone.
113 * Because of this, it is possible for the snapshot count to be over the limit
114 * and snapshots taken in the global zone could cause a lower level dataset to
115 * hit or exceed its limit. The administrator taking the global zone recursive
116 * snapshot should be aware of this side-effect and behave accordingly.
117 * For consistency, the filesystem limit is also not enforced if the user can
120 * The filesystem and snapshot limits are validated by dsl_fs_ss_limit_check()
121 * and updated by dsl_fs_ss_count_adjust(). A new limit value is setup in
122 * dsl_dir_activate_fs_ss_limit() and the counts are adjusted, if necessary, by
123 * dsl_dir_init_fs_ss_count().
125 * There is a special case when we receive a filesystem that already exists. In
126 * this case a temporary clone name of %X is created (see dmu_recv_begin). We
127 * never update the filesystem counts for temporary clones.
129 * Likewise, we do not update the snapshot counts for temporary snapshots,
130 * such as those created by zfs diff.
133 extern inline dsl_dir_phys_t *dsl_dir_phys(dsl_dir_t *dd);
135 static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
137 typedef struct ddulrt_arg {
138 dsl_dir_t *ddulrta_dd;
143 dsl_dir_evict_async(void *dbu)
147 dsl_pool_t *dp __maybe_unused = dd->dd_pool;
151 for (t = 0; t < TXG_SIZE; t++) {
152 ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t));
153 ASSERT(dd->dd_tempreserved[t] == 0);
154 ASSERT(dd->dd_space_towrite[t] == 0);
158 dsl_dir_async_rele(dd->dd_parent, dd);
160 spa_async_close(dd->dd_pool->dp_spa, dd);
162 if (dsl_deadlist_is_open(&dd->dd_livelist))
163 dsl_dir_livelist_close(dd);
166 cv_destroy(&dd->dd_activity_cv);
167 mutex_destroy(&dd->dd_activity_lock);
168 mutex_destroy(&dd->dd_lock);
169 kmem_free(dd, sizeof (dsl_dir_t));
173 dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj,
174 const char *tail, void *tag, dsl_dir_t **ddp)
178 dmu_object_info_t doi;
181 ASSERT(dsl_pool_config_held(dp));
183 err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
186 dd = dmu_buf_get_user(dbuf);
188 dmu_object_info_from_db(dbuf, &doi);
189 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_DSL_DIR);
190 ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
195 dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
196 dd->dd_object = ddobj;
200 if (dsl_dir_is_zapified(dd) &&
201 zap_contains(dp->dp_meta_objset, ddobj,
202 DD_FIELD_CRYPTO_KEY_OBJ) == 0) {
203 VERIFY0(zap_lookup(dp->dp_meta_objset,
204 ddobj, DD_FIELD_CRYPTO_KEY_OBJ,
205 sizeof (uint64_t), 1, &dd->dd_crypto_obj));
207 /* check for on-disk format errata */
208 if (dsl_dir_incompatible_encryption_version(dd)) {
209 dp->dp_spa->spa_errata =
210 ZPOOL_ERRATA_ZOL_6845_ENCRYPTION;
214 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
215 mutex_init(&dd->dd_activity_lock, NULL, MUTEX_DEFAULT, NULL);
216 cv_init(&dd->dd_activity_cv, NULL, CV_DEFAULT, NULL);
219 dsl_dir_snap_cmtime_update(dd);
221 if (dsl_dir_phys(dd)->dd_parent_obj) {
222 err = dsl_dir_hold_obj(dp,
223 dsl_dir_phys(dd)->dd_parent_obj, NULL, dd,
231 err = zap_lookup(dp->dp_meta_objset,
232 dsl_dir_phys(dd->dd_parent)->
233 dd_child_dir_zapobj, tail,
234 sizeof (foundobj), 1, &foundobj);
235 ASSERT(err || foundobj == ddobj);
237 (void) strlcpy(dd->dd_myname, tail,
238 sizeof (dd->dd_myname));
240 err = zap_value_search(dp->dp_meta_objset,
241 dsl_dir_phys(dd->dd_parent)->
243 ddobj, 0, dd->dd_myname);
248 (void) strlcpy(dd->dd_myname, spa_name(dp->dp_spa),
249 sizeof (dd->dd_myname));
252 if (dsl_dir_is_clone(dd)) {
253 dmu_buf_t *origin_bonus;
254 dsl_dataset_phys_t *origin_phys;
257 * We can't open the origin dataset, because
258 * that would require opening this dsl_dir.
259 * Just look at its phys directly instead.
261 err = dmu_bonus_hold(dp->dp_meta_objset,
262 dsl_dir_phys(dd)->dd_origin_obj, FTAG,
266 origin_phys = origin_bonus->db_data;
268 origin_phys->ds_creation_txg;
269 dmu_buf_rele(origin_bonus, FTAG);
270 if (dsl_dir_is_zapified(dd)) {
272 err = zap_lookup(dp->dp_meta_objset,
273 dd->dd_object, DD_FIELD_LIVELIST,
274 sizeof (uint64_t), 1, &obj);
276 dsl_dir_livelist_open(dd, obj);
277 else if (err != ENOENT)
282 dmu_buf_init_user(&dd->dd_dbu, NULL, dsl_dir_evict_async,
284 winner = dmu_buf_set_user_ie(dbuf, &dd->dd_dbu);
285 if (winner != NULL) {
287 dsl_dir_rele(dd->dd_parent, dd);
288 if (dsl_deadlist_is_open(&dd->dd_livelist))
289 dsl_dir_livelist_close(dd);
291 cv_destroy(&dd->dd_activity_cv);
292 mutex_destroy(&dd->dd_activity_lock);
293 mutex_destroy(&dd->dd_lock);
294 kmem_free(dd, sizeof (dsl_dir_t));
297 spa_open_ref(dp->dp_spa, dd);
302 * The dsl_dir_t has both open-to-close and instantiate-to-evict
303 * holds on the spa. We need the open-to-close holds because
304 * otherwise the spa_refcnt wouldn't change when we open a
305 * dir which the spa also has open, so we could incorrectly
306 * think it was OK to unload/export/destroy the pool. We need
307 * the instantiate-to-evict hold because the dsl_dir_t has a
308 * pointer to the dd_pool, which has a pointer to the spa_t.
310 spa_open_ref(dp->dp_spa, tag);
311 ASSERT3P(dd->dd_pool, ==, dp);
312 ASSERT3U(dd->dd_object, ==, ddobj);
313 ASSERT3P(dd->dd_dbuf, ==, dbuf);
319 dsl_dir_rele(dd->dd_parent, dd);
320 if (dsl_deadlist_is_open(&dd->dd_livelist))
321 dsl_dir_livelist_close(dd);
323 cv_destroy(&dd->dd_activity_cv);
324 mutex_destroy(&dd->dd_activity_lock);
325 mutex_destroy(&dd->dd_lock);
326 kmem_free(dd, sizeof (dsl_dir_t));
327 dmu_buf_rele(dbuf, tag);
332 dsl_dir_rele(dsl_dir_t *dd, void *tag)
334 dprintf_dd(dd, "%s\n", "");
335 spa_close(dd->dd_pool->dp_spa, tag);
336 dmu_buf_rele(dd->dd_dbuf, tag);
340 * Remove a reference to the given dsl dir that is being asynchronously
341 * released. Async releases occur from a taskq performing eviction of
342 * dsl datasets and dirs. This process is identical to a normal release
343 * with the exception of using the async API for releasing the reference on
347 dsl_dir_async_rele(dsl_dir_t *dd, void *tag)
349 dprintf_dd(dd, "%s\n", "");
350 spa_async_close(dd->dd_pool->dp_spa, tag);
351 dmu_buf_rele(dd->dd_dbuf, tag);
354 /* buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes */
356 dsl_dir_name(dsl_dir_t *dd, char *buf)
359 dsl_dir_name(dd->dd_parent, buf);
360 VERIFY3U(strlcat(buf, "/", ZFS_MAX_DATASET_NAME_LEN), <,
361 ZFS_MAX_DATASET_NAME_LEN);
365 if (!MUTEX_HELD(&dd->dd_lock)) {
367 * recursive mutex so that we can use
368 * dprintf_dd() with dd_lock held
370 mutex_enter(&dd->dd_lock);
371 VERIFY3U(strlcat(buf, dd->dd_myname, ZFS_MAX_DATASET_NAME_LEN),
372 <, ZFS_MAX_DATASET_NAME_LEN);
373 mutex_exit(&dd->dd_lock);
375 VERIFY3U(strlcat(buf, dd->dd_myname, ZFS_MAX_DATASET_NAME_LEN),
376 <, ZFS_MAX_DATASET_NAME_LEN);
380 /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
382 dsl_dir_namelen(dsl_dir_t *dd)
387 /* parent's name + 1 for the "/" */
388 result = dsl_dir_namelen(dd->dd_parent) + 1;
391 if (!MUTEX_HELD(&dd->dd_lock)) {
392 /* see dsl_dir_name */
393 mutex_enter(&dd->dd_lock);
394 result += strlen(dd->dd_myname);
395 mutex_exit(&dd->dd_lock);
397 result += strlen(dd->dd_myname);
404 getcomponent(const char *path, char *component, const char **nextp)
408 if ((path == NULL) || (path[0] == '\0'))
409 return (SET_ERROR(ENOENT));
410 /* This would be a good place to reserve some namespace... */
411 p = strpbrk(path, "/@");
412 if (p && (p[1] == '/' || p[1] == '@')) {
413 /* two separators in a row */
414 return (SET_ERROR(EINVAL));
416 if (p == NULL || p == path) {
418 * if the first thing is an @ or /, it had better be an
419 * @ and it had better not have any more ats or slashes,
420 * and it had better have something after the @.
423 (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
424 return (SET_ERROR(EINVAL));
425 if (strlen(path) >= ZFS_MAX_DATASET_NAME_LEN)
426 return (SET_ERROR(ENAMETOOLONG));
427 (void) strlcpy(component, path, ZFS_MAX_DATASET_NAME_LEN);
429 } else if (p[0] == '/') {
430 if (p - path >= ZFS_MAX_DATASET_NAME_LEN)
431 return (SET_ERROR(ENAMETOOLONG));
432 (void) strncpy(component, path, p - path);
433 component[p - path] = '\0';
435 } else if (p[0] == '@') {
437 * if the next separator is an @, there better not be
440 if (strchr(path, '/'))
441 return (SET_ERROR(EINVAL));
442 if (p - path >= ZFS_MAX_DATASET_NAME_LEN)
443 return (SET_ERROR(ENAMETOOLONG));
444 (void) strncpy(component, path, p - path);
445 component[p - path] = '\0';
447 panic("invalid p=%p", (void *)p);
454 * Return the dsl_dir_t, and possibly the last component which couldn't
455 * be found in *tail. The name must be in the specified dsl_pool_t. This
456 * thread must hold the dp_config_rwlock for the pool. Returns NULL if the
457 * path is bogus, or if tail==NULL and we couldn't parse the whole name.
458 * (*tail)[0] == '@' means that the last component is a snapshot.
461 dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag,
462 dsl_dir_t **ddp, const char **tailp)
465 const char *spaname, *next, *nextnext = NULL;
470 buf = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
471 err = getcomponent(name, buf, &next);
475 /* Make sure the name is in the specified pool. */
476 spaname = spa_name(dp->dp_spa);
477 if (strcmp(buf, spaname) != 0) {
478 err = SET_ERROR(EXDEV);
482 ASSERT(dsl_pool_config_held(dp));
484 err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
489 while (next != NULL) {
491 err = getcomponent(next, buf, &nextnext);
494 ASSERT(next[0] != '\0');
497 dprintf("looking up %s in obj%lld\n",
498 buf, dsl_dir_phys(dd)->dd_child_dir_zapobj);
500 err = zap_lookup(dp->dp_meta_objset,
501 dsl_dir_phys(dd)->dd_child_dir_zapobj,
502 buf, sizeof (ddobj), 1, &ddobj);
509 err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_dd);
512 dsl_dir_rele(dd, tag);
518 dsl_dir_rele(dd, tag);
523 * It's an error if there's more than one component left, or
524 * tailp==NULL and there's any component left.
527 (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
529 dsl_dir_rele(dd, tag);
530 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
531 err = SET_ERROR(ENOENT);
538 kmem_free(buf, ZFS_MAX_DATASET_NAME_LEN);
543 * If the counts are already initialized for this filesystem and its
544 * descendants then do nothing, otherwise initialize the counts.
546 * The counts on this filesystem, and those below, may be uninitialized due to
547 * either the use of a pre-existing pool which did not support the
548 * filesystem/snapshot limit feature, or one in which the feature had not yet
551 * Recursively descend the filesystem tree and update the filesystem/snapshot
552 * counts on each filesystem below, then update the cumulative count on the
553 * current filesystem. If the filesystem already has a count set on it,
554 * then we know that its counts, and the counts on the filesystems below it,
555 * are already correct, so we don't have to update this filesystem.
558 dsl_dir_init_fs_ss_count(dsl_dir_t *dd, dmu_tx_t *tx)
560 uint64_t my_fs_cnt = 0;
561 uint64_t my_ss_cnt = 0;
562 dsl_pool_t *dp = dd->dd_pool;
563 objset_t *os = dp->dp_meta_objset;
568 ASSERT(spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT));
569 ASSERT(dsl_pool_config_held(dp));
570 ASSERT(dmu_tx_is_syncing(tx));
572 dsl_dir_zapify(dd, tx);
575 * If the filesystem count has already been initialized then we
576 * don't need to recurse down any further.
578 if (zap_contains(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT) == 0)
581 zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP);
582 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
584 /* Iterate my child dirs */
585 for (zap_cursor_init(zc, os, dsl_dir_phys(dd)->dd_child_dir_zapobj);
586 zap_cursor_retrieve(zc, za) == 0; zap_cursor_advance(zc)) {
590 VERIFY0(dsl_dir_hold_obj(dp, za->za_first_integer, NULL, FTAG,
594 * Ignore hidden ($FREE, $MOS & $ORIGIN) objsets and
595 * temporary datasets.
597 if (chld_dd->dd_myname[0] == '$' ||
598 chld_dd->dd_myname[0] == '%') {
599 dsl_dir_rele(chld_dd, FTAG);
603 my_fs_cnt++; /* count this child */
605 dsl_dir_init_fs_ss_count(chld_dd, tx);
607 VERIFY0(zap_lookup(os, chld_dd->dd_object,
608 DD_FIELD_FILESYSTEM_COUNT, sizeof (count), 1, &count));
610 VERIFY0(zap_lookup(os, chld_dd->dd_object,
611 DD_FIELD_SNAPSHOT_COUNT, sizeof (count), 1, &count));
614 dsl_dir_rele(chld_dd, FTAG);
617 /* Count my snapshots (we counted children's snapshots above) */
618 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
619 dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds));
621 for (zap_cursor_init(zc, os, dsl_dataset_phys(ds)->ds_snapnames_zapobj);
622 zap_cursor_retrieve(zc, za) == 0;
623 zap_cursor_advance(zc)) {
624 /* Don't count temporary snapshots */
625 if (za->za_name[0] != '%')
630 dsl_dataset_rele(ds, FTAG);
632 kmem_free(zc, sizeof (zap_cursor_t));
633 kmem_free(za, sizeof (zap_attribute_t));
635 /* we're in a sync task, update counts */
636 dmu_buf_will_dirty(dd->dd_dbuf, tx);
637 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
638 sizeof (my_fs_cnt), 1, &my_fs_cnt, tx));
639 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
640 sizeof (my_ss_cnt), 1, &my_ss_cnt, tx));
644 dsl_dir_actv_fs_ss_limit_check(void *arg, dmu_tx_t *tx)
646 char *ddname = (char *)arg;
647 dsl_pool_t *dp = dmu_tx_pool(tx);
652 error = dsl_dataset_hold(dp, ddname, FTAG, &ds);
656 if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
657 dsl_dataset_rele(ds, FTAG);
658 return (SET_ERROR(ENOTSUP));
662 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT) &&
663 dsl_dir_is_zapified(dd) &&
664 zap_contains(dp->dp_meta_objset, dd->dd_object,
665 DD_FIELD_FILESYSTEM_COUNT) == 0) {
666 dsl_dataset_rele(ds, FTAG);
667 return (SET_ERROR(EALREADY));
670 dsl_dataset_rele(ds, FTAG);
675 dsl_dir_actv_fs_ss_limit_sync(void *arg, dmu_tx_t *tx)
677 char *ddname = (char *)arg;
678 dsl_pool_t *dp = dmu_tx_pool(tx);
682 VERIFY0(dsl_dataset_hold(dp, ddname, FTAG, &ds));
684 spa = dsl_dataset_get_spa(ds);
686 if (!spa_feature_is_active(spa, SPA_FEATURE_FS_SS_LIMIT)) {
688 * Since the feature was not active and we're now setting a
689 * limit, increment the feature-active counter so that the
690 * feature becomes active for the first time.
692 * We are already in a sync task so we can update the MOS.
694 spa_feature_incr(spa, SPA_FEATURE_FS_SS_LIMIT, tx);
698 * Since we are now setting a non-UINT64_MAX limit on the filesystem,
699 * we need to ensure the counts are correct. Descend down the tree from
700 * this point and update all of the counts to be accurate.
702 dsl_dir_init_fs_ss_count(ds->ds_dir, tx);
704 dsl_dataset_rele(ds, FTAG);
708 * Make sure the feature is enabled and activate it if necessary.
709 * Since we're setting a limit, ensure the on-disk counts are valid.
710 * This is only called by the ioctl path when setting a limit value.
712 * We do not need to validate the new limit, since users who can change the
713 * limit are also allowed to exceed the limit.
716 dsl_dir_activate_fs_ss_limit(const char *ddname)
720 error = dsl_sync_task(ddname, dsl_dir_actv_fs_ss_limit_check,
721 dsl_dir_actv_fs_ss_limit_sync, (void *)ddname, 0,
722 ZFS_SPACE_CHECK_RESERVED);
724 if (error == EALREADY)
731 * Used to determine if the filesystem_limit or snapshot_limit should be
732 * enforced. We allow the limit to be exceeded if the user has permission to
733 * write the property value. We pass in the creds that we got in the open
734 * context since we will always be the GZ root in syncing context. We also have
735 * to handle the case where we are allowed to change the limit on the current
736 * dataset, but there may be another limit in the tree above.
738 * We can never modify these two properties within a non-global zone. In
739 * addition, the other checks are modeled on zfs_secpolicy_write_perms. We
740 * can't use that function since we are already holding the dp_config_rwlock.
741 * In addition, we already have the dd and dealing with snapshots is simplified
752 dsl_enforce_ds_ss_limits(dsl_dir_t *dd, zfs_prop_t prop,
753 cred_t *cr, proc_t *proc)
755 enforce_res_t enforce = ENFORCE_ALWAYS;
759 const char *zonedstr;
761 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
762 prop == ZFS_PROP_SNAPSHOT_LIMIT);
765 if (crgetzoneid(cr) != GLOBAL_ZONEID)
766 return (ENFORCE_ALWAYS);
769 * We are checking the saved credentials of the user process, which is
770 * not the current process. Note that we can't use secpolicy_zfs(),
771 * because it only works if the cred is that of the current process (on
774 if (secpolicy_zfs_proc(cr, proc) == 0)
775 return (ENFORCE_NEVER);
778 if ((obj = dsl_dir_phys(dd)->dd_head_dataset_obj) == 0)
779 return (ENFORCE_ALWAYS);
781 ASSERT(dsl_pool_config_held(dd->dd_pool));
783 if (dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds) != 0)
784 return (ENFORCE_ALWAYS);
786 zonedstr = zfs_prop_to_name(ZFS_PROP_ZONED);
787 if (dsl_prop_get_ds(ds, zonedstr, 8, 1, &zoned, NULL) || zoned) {
788 /* Only root can access zoned fs's from the GZ */
789 enforce = ENFORCE_ALWAYS;
791 if (dsl_deleg_access_impl(ds, zfs_prop_to_name(prop), cr) == 0)
792 enforce = ENFORCE_ABOVE;
795 dsl_dataset_rele(ds, FTAG);
800 * Check if adding additional child filesystem(s) would exceed any filesystem
801 * limits or adding additional snapshot(s) would exceed any snapshot limits.
802 * The prop argument indicates which limit to check.
804 * Note that all filesystem limits up to the root (or the highest
805 * initialized) filesystem or the given ancestor must be satisfied.
808 dsl_fs_ss_limit_check(dsl_dir_t *dd, uint64_t delta, zfs_prop_t prop,
809 dsl_dir_t *ancestor, cred_t *cr, proc_t *proc)
811 objset_t *os = dd->dd_pool->dp_meta_objset;
812 uint64_t limit, count;
814 enforce_res_t enforce;
817 ASSERT(dsl_pool_config_held(dd->dd_pool));
818 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
819 prop == ZFS_PROP_SNAPSHOT_LIMIT);
822 * If we're allowed to change the limit, don't enforce the limit
823 * e.g. this can happen if a snapshot is taken by an administrative
824 * user in the global zone (i.e. a recursive snapshot by root).
825 * However, we must handle the case of delegated permissions where we
826 * are allowed to change the limit on the current dataset, but there
827 * is another limit in the tree above.
829 enforce = dsl_enforce_ds_ss_limits(dd, prop, cr, proc);
830 if (enforce == ENFORCE_NEVER)
834 * e.g. if renaming a dataset with no snapshots, count adjustment
840 if (prop == ZFS_PROP_SNAPSHOT_LIMIT) {
842 * We don't enforce the limit for temporary snapshots. This is
843 * indicated by a NULL cred_t argument.
848 count_prop = DD_FIELD_SNAPSHOT_COUNT;
850 count_prop = DD_FIELD_FILESYSTEM_COUNT;
854 * If an ancestor has been provided, stop checking the limit once we
855 * hit that dir. We need this during rename so that we don't overcount
856 * the check once we recurse up to the common ancestor.
862 * If we hit an uninitialized node while recursing up the tree, we can
863 * stop since we know there is no limit here (or above). The counts are
864 * not valid on this node and we know we won't touch this node's counts.
866 if (!dsl_dir_is_zapified(dd) || zap_lookup(os, dd->dd_object,
867 count_prop, sizeof (count), 1, &count) == ENOENT)
870 err = dsl_prop_get_dd(dd, zfs_prop_to_name(prop), 8, 1, &limit, NULL,
875 /* Is there a limit which we've hit? */
876 if (enforce == ENFORCE_ALWAYS && (count + delta) > limit)
877 return (SET_ERROR(EDQUOT));
879 if (dd->dd_parent != NULL)
880 err = dsl_fs_ss_limit_check(dd->dd_parent, delta, prop,
887 * Adjust the filesystem or snapshot count for the specified dsl_dir_t and all
888 * parents. When a new filesystem/snapshot is created, increment the count on
889 * all parents, and when a filesystem/snapshot is destroyed, decrement the
893 dsl_fs_ss_count_adjust(dsl_dir_t *dd, int64_t delta, const char *prop,
897 objset_t *os = dd->dd_pool->dp_meta_objset;
900 ASSERT(dsl_pool_config_held(dd->dd_pool));
901 ASSERT(dmu_tx_is_syncing(tx));
902 ASSERT(strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0 ||
903 strcmp(prop, DD_FIELD_SNAPSHOT_COUNT) == 0);
906 * When we receive an incremental stream into a filesystem that already
907 * exists, a temporary clone is created. We don't count this temporary
908 * clone, whose name begins with a '%'. We also ignore hidden ($FREE,
909 * $MOS & $ORIGIN) objsets.
911 if ((dd->dd_myname[0] == '%' || dd->dd_myname[0] == '$') &&
912 strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0)
916 * e.g. if renaming a dataset with no snapshots, count adjustment is 0
922 * If we hit an uninitialized node while recursing up the tree, we can
923 * stop since we know the counts are not valid on this node and we
924 * know we shouldn't touch this node's counts. An uninitialized count
925 * on the node indicates that either the feature has not yet been
926 * activated or there are no limits on this part of the tree.
928 if (!dsl_dir_is_zapified(dd) || (err = zap_lookup(os, dd->dd_object,
929 prop, sizeof (count), 1, &count)) == ENOENT)
934 /* Use a signed verify to make sure we're not neg. */
935 VERIFY3S(count, >=, 0);
937 VERIFY0(zap_update(os, dd->dd_object, prop, sizeof (count), 1, &count,
940 /* Roll up this additional count into our ancestors */
941 if (dd->dd_parent != NULL)
942 dsl_fs_ss_count_adjust(dd->dd_parent, delta, prop, tx);
946 dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
949 objset_t *mos = dp->dp_meta_objset;
951 dsl_dir_phys_t *ddphys;
954 ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
955 DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
957 VERIFY0(zap_add(mos, dsl_dir_phys(pds)->dd_child_dir_zapobj,
958 name, sizeof (uint64_t), 1, &ddobj, tx));
960 /* it's the root dir */
961 VERIFY0(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
962 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
964 VERIFY0(dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
965 dmu_buf_will_dirty(dbuf, tx);
966 ddphys = dbuf->db_data;
968 ddphys->dd_creation_time = gethrestime_sec();
970 ddphys->dd_parent_obj = pds->dd_object;
972 /* update the filesystem counts */
973 dsl_fs_ss_count_adjust(pds, 1, DD_FIELD_FILESYSTEM_COUNT, tx);
975 ddphys->dd_props_zapobj = zap_create(mos,
976 DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
977 ddphys->dd_child_dir_zapobj = zap_create(mos,
978 DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
979 if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
980 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
982 dmu_buf_rele(dbuf, FTAG);
988 dsl_dir_is_clone(dsl_dir_t *dd)
990 return (dsl_dir_phys(dd)->dd_origin_obj &&
991 (dd->dd_pool->dp_origin_snap == NULL ||
992 dsl_dir_phys(dd)->dd_origin_obj !=
993 dd->dd_pool->dp_origin_snap->ds_object));
997 dsl_dir_get_used(dsl_dir_t *dd)
999 return (dsl_dir_phys(dd)->dd_used_bytes);
1003 dsl_dir_get_compressed(dsl_dir_t *dd)
1005 return (dsl_dir_phys(dd)->dd_compressed_bytes);
1009 dsl_dir_get_quota(dsl_dir_t *dd)
1011 return (dsl_dir_phys(dd)->dd_quota);
1015 dsl_dir_get_reservation(dsl_dir_t *dd)
1017 return (dsl_dir_phys(dd)->dd_reserved);
1021 dsl_dir_get_compressratio(dsl_dir_t *dd)
1023 /* a fixed point number, 100x the ratio */
1024 return (dsl_dir_phys(dd)->dd_compressed_bytes == 0 ? 100 :
1025 (dsl_dir_phys(dd)->dd_uncompressed_bytes * 100 /
1026 dsl_dir_phys(dd)->dd_compressed_bytes));
1030 dsl_dir_get_logicalused(dsl_dir_t *dd)
1032 return (dsl_dir_phys(dd)->dd_uncompressed_bytes);
1036 dsl_dir_get_usedsnap(dsl_dir_t *dd)
1038 return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]);
1042 dsl_dir_get_usedds(dsl_dir_t *dd)
1044 return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_HEAD]);
1048 dsl_dir_get_usedrefreserv(dsl_dir_t *dd)
1050 return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_REFRSRV]);
1054 dsl_dir_get_usedchild(dsl_dir_t *dd)
1056 return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD] +
1057 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD_RSRV]);
1061 dsl_dir_get_origin(dsl_dir_t *dd, char *buf)
1064 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
1065 dsl_dir_phys(dd)->dd_origin_obj, FTAG, &ds));
1067 dsl_dataset_name(ds, buf);
1069 dsl_dataset_rele(ds, FTAG);
1073 dsl_dir_get_filesystem_count(dsl_dir_t *dd, uint64_t *count)
1075 if (dsl_dir_is_zapified(dd)) {
1076 objset_t *os = dd->dd_pool->dp_meta_objset;
1077 return (zap_lookup(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
1078 sizeof (*count), 1, count));
1080 return (SET_ERROR(ENOENT));
1085 dsl_dir_get_snapshot_count(dsl_dir_t *dd, uint64_t *count)
1087 if (dsl_dir_is_zapified(dd)) {
1088 objset_t *os = dd->dd_pool->dp_meta_objset;
1089 return (zap_lookup(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
1090 sizeof (*count), 1, count));
1092 return (SET_ERROR(ENOENT));
1097 dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
1099 mutex_enter(&dd->dd_lock);
1100 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA,
1101 dsl_dir_get_quota(dd));
1102 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
1103 dsl_dir_get_reservation(dd));
1104 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED,
1105 dsl_dir_get_logicalused(dd));
1106 if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1107 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
1108 dsl_dir_get_usedsnap(dd));
1109 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
1110 dsl_dir_get_usedds(dd));
1111 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
1112 dsl_dir_get_usedrefreserv(dd));
1113 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
1114 dsl_dir_get_usedchild(dd));
1116 mutex_exit(&dd->dd_lock);
1119 if (dsl_dir_get_filesystem_count(dd, &count) == 0) {
1120 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_FILESYSTEM_COUNT,
1123 if (dsl_dir_get_snapshot_count(dd, &count) == 0) {
1124 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_SNAPSHOT_COUNT,
1128 if (dsl_dir_is_clone(dd)) {
1129 char buf[ZFS_MAX_DATASET_NAME_LEN];
1130 dsl_dir_get_origin(dd, buf);
1131 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
1137 dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
1139 dsl_pool_t *dp = dd->dd_pool;
1141 ASSERT(dsl_dir_phys(dd));
1143 if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) {
1144 /* up the hold count until we can be written out */
1145 dmu_buf_add_ref(dd->dd_dbuf, dd);
1150 parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
1152 uint64_t old_accounted = MAX(used, dsl_dir_phys(dd)->dd_reserved);
1153 uint64_t new_accounted =
1154 MAX(used + delta, dsl_dir_phys(dd)->dd_reserved);
1155 return (new_accounted - old_accounted);
1159 dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
1161 ASSERT(dmu_tx_is_syncing(tx));
1163 mutex_enter(&dd->dd_lock);
1164 ASSERT0(dd->dd_tempreserved[tx->tx_txg & TXG_MASK]);
1165 dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
1166 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] / 1024);
1167 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] = 0;
1168 mutex_exit(&dd->dd_lock);
1170 /* release the hold from dsl_dir_dirty */
1171 dmu_buf_rele(dd->dd_dbuf, dd);
1175 dsl_dir_space_towrite(dsl_dir_t *dd)
1179 ASSERT(MUTEX_HELD(&dd->dd_lock));
1181 for (int i = 0; i < TXG_SIZE; i++) {
1182 space += dd->dd_space_towrite[i & TXG_MASK];
1183 ASSERT3U(dd->dd_space_towrite[i & TXG_MASK], >=, 0);
1189 * How much space would dd have available if ancestor had delta applied
1190 * to it? If ondiskonly is set, we're only interested in what's
1191 * on-disk, not estimated pending changes.
1194 dsl_dir_space_available(dsl_dir_t *dd,
1195 dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
1197 uint64_t parentspace, myspace, quota, used;
1200 * If there are no restrictions otherwise, assume we have
1201 * unlimited space available.
1204 parentspace = UINT64_MAX;
1206 if (dd->dd_parent != NULL) {
1207 parentspace = dsl_dir_space_available(dd->dd_parent,
1208 ancestor, delta, ondiskonly);
1211 mutex_enter(&dd->dd_lock);
1212 if (dsl_dir_phys(dd)->dd_quota != 0)
1213 quota = dsl_dir_phys(dd)->dd_quota;
1214 used = dsl_dir_phys(dd)->dd_used_bytes;
1216 used += dsl_dir_space_towrite(dd);
1218 if (dd->dd_parent == NULL) {
1219 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool,
1220 ZFS_SPACE_CHECK_NORMAL);
1221 quota = MIN(quota, poolsize);
1224 if (dsl_dir_phys(dd)->dd_reserved > used && parentspace != UINT64_MAX) {
1226 * We have some space reserved, in addition to what our
1229 parentspace += dsl_dir_phys(dd)->dd_reserved - used;
1232 if (dd == ancestor) {
1234 ASSERT(used >= -delta);
1236 if (parentspace != UINT64_MAX)
1237 parentspace -= delta;
1245 * the lesser of the space provided by our parent and
1246 * the space left in our quota
1248 myspace = MIN(parentspace, quota - used);
1251 mutex_exit(&dd->dd_lock);
1256 struct tempreserve {
1257 list_node_t tr_node;
1263 dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
1264 boolean_t ignorequota, list_t *tr_list,
1265 dmu_tx_t *tx, boolean_t first)
1269 struct tempreserve *tr;
1278 ASSERT3U(txg, !=, 0);
1279 ASSERT3S(asize, >, 0);
1281 mutex_enter(&dd->dd_lock);
1284 * Check against the dsl_dir's quota. We don't add in the delta
1285 * when checking for over-quota because they get one free hit.
1287 uint64_t est_inflight = dsl_dir_space_towrite(dd);
1288 for (int i = 0; i < TXG_SIZE; i++)
1289 est_inflight += dd->dd_tempreserved[i];
1290 uint64_t used_on_disk = dsl_dir_phys(dd)->dd_used_bytes;
1293 * On the first iteration, fetch the dataset's used-on-disk and
1294 * refreservation values. Also, if checkrefquota is set, test if
1295 * allocating this space would exceed the dataset's refquota.
1297 if (first && tx->tx_objset) {
1299 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
1301 error = dsl_dataset_check_quota(ds, !netfree,
1302 asize, est_inflight, &used_on_disk, &ref_rsrv);
1304 mutex_exit(&dd->dd_lock);
1305 DMU_TX_STAT_BUMP(dmu_tx_quota);
1311 * If this transaction will result in a net free of space,
1312 * we want to let it through.
1314 if (ignorequota || netfree || dsl_dir_phys(dd)->dd_quota == 0)
1317 quota = dsl_dir_phys(dd)->dd_quota;
1320 * Adjust the quota against the actual pool size at the root
1321 * minus any outstanding deferred frees.
1322 * To ensure that it's possible to remove files from a full
1323 * pool without inducing transient overcommits, we throttle
1324 * netfree transactions against a quota that is slightly larger,
1325 * but still within the pool's allocation slop. In cases where
1326 * we're very close to full, this will allow a steady trickle of
1327 * removes to get through.
1329 uint64_t deferred = 0;
1330 if (dd->dd_parent == NULL) {
1331 uint64_t avail = dsl_pool_unreserved_space(dd->dd_pool,
1333 ZFS_SPACE_CHECK_RESERVED : ZFS_SPACE_CHECK_NORMAL);
1335 if (avail < quota) {
1337 retval = SET_ERROR(ENOSPC);
1342 * If they are requesting more space, and our current estimate
1343 * is over quota, they get to try again unless the actual
1344 * on-disk is over quota and there are no pending changes (which
1345 * may free up space for us).
1347 if (used_on_disk + est_inflight >= quota) {
1348 if (est_inflight > 0 || used_on_disk < quota ||
1349 (retval == ENOSPC && used_on_disk < quota + deferred))
1351 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
1352 "quota=%lluK tr=%lluK err=%d\n",
1353 used_on_disk>>10, est_inflight>>10,
1354 quota>>10, asize>>10, retval);
1355 mutex_exit(&dd->dd_lock);
1356 DMU_TX_STAT_BUMP(dmu_tx_quota);
1357 return (SET_ERROR(retval));
1360 /* We need to up our estimated delta before dropping dd_lock */
1361 dd->dd_tempreserved[txg & TXG_MASK] += asize;
1363 uint64_t parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
1365 mutex_exit(&dd->dd_lock);
1367 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1369 tr->tr_size = asize;
1370 list_insert_tail(tr_list, tr);
1372 /* see if it's OK with our parent */
1373 if (dd->dd_parent != NULL && parent_rsrv != 0) {
1375 * Recurse on our parent without recursion. This has been
1376 * observed to be potentially large stack usage even within
1377 * the test suite. Largest seen stack was 7632 bytes on linux.
1381 asize = parent_rsrv;
1382 ignorequota = (dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
1384 goto top_of_function;
1392 * Reserve space in this dsl_dir, to be used in this tx's txg.
1393 * After the space has been dirtied (and dsl_dir_willuse_space()
1394 * has been called), the reservation should be canceled, using
1395 * dsl_dir_tempreserve_clear().
1398 dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
1399 boolean_t netfree, void **tr_cookiep, dmu_tx_t *tx)
1409 tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
1410 list_create(tr_list, sizeof (struct tempreserve),
1411 offsetof(struct tempreserve, tr_node));
1412 ASSERT3S(asize, >, 0);
1414 err = arc_tempreserve_space(dd->dd_pool->dp_spa, lsize, tx->tx_txg);
1416 struct tempreserve *tr;
1418 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1419 tr->tr_size = lsize;
1420 list_insert_tail(tr_list, tr);
1422 if (err == EAGAIN) {
1424 * If arc_memory_throttle() detected that pageout
1425 * is running and we are low on memory, we delay new
1426 * non-pageout transactions to give pageout an
1429 * It is unfortunate to be delaying while the caller's
1432 txg_delay(dd->dd_pool, tx->tx_txg,
1433 MSEC2NSEC(10), MSEC2NSEC(10));
1434 err = SET_ERROR(ERESTART);
1439 err = dsl_dir_tempreserve_impl(dd, asize, netfree,
1440 B_FALSE, tr_list, tx, B_TRUE);
1444 dsl_dir_tempreserve_clear(tr_list, tx);
1446 *tr_cookiep = tr_list;
1452 * Clear a temporary reservation that we previously made with
1453 * dsl_dir_tempreserve_space().
1456 dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
1458 int txgidx = tx->tx_txg & TXG_MASK;
1459 list_t *tr_list = tr_cookie;
1460 struct tempreserve *tr;
1462 ASSERT3U(tx->tx_txg, !=, 0);
1464 if (tr_cookie == NULL)
1467 while ((tr = list_head(tr_list)) != NULL) {
1469 mutex_enter(&tr->tr_ds->dd_lock);
1470 ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
1472 tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
1473 mutex_exit(&tr->tr_ds->dd_lock);
1475 arc_tempreserve_clear(tr->tr_size);
1477 list_remove(tr_list, tr);
1478 kmem_free(tr, sizeof (struct tempreserve));
1481 kmem_free(tr_list, sizeof (list_t));
1485 * This should be called from open context when we think we're going to write
1486 * or free space, for example when dirtying data. Be conservative; it's okay
1487 * to write less space or free more, but we don't want to write more or free
1488 * less than the amount specified.
1490 * NOTE: The behavior of this function is identical to the Illumos / FreeBSD
1491 * version however it has been adjusted to use an iterative rather than
1492 * recursive algorithm to minimize stack usage.
1495 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
1497 int64_t parent_space;
1501 mutex_enter(&dd->dd_lock);
1503 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
1505 est_used = dsl_dir_space_towrite(dd) +
1506 dsl_dir_phys(dd)->dd_used_bytes;
1507 parent_space = parent_delta(dd, est_used, space);
1508 mutex_exit(&dd->dd_lock);
1510 /* Make sure that we clean up dd_space_to* */
1511 dsl_dir_dirty(dd, tx);
1514 space = parent_space;
1515 } while (space && dd);
1518 /* call from syncing context when we actually write/free space for this dd */
1520 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
1521 int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
1523 int64_t accounted_delta;
1526 * dsl_dataset_set_refreservation_sync_impl() calls this with
1527 * dd_lock held, so that it can atomically update
1528 * ds->ds_reserved and the dsl_dir accounting, so that
1529 * dsl_dataset_check_quota() can see dataset and dir accounting
1532 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
1534 ASSERT(dmu_tx_is_syncing(tx));
1535 ASSERT(type < DD_USED_NUM);
1537 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1540 mutex_enter(&dd->dd_lock);
1542 parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, used);
1543 ASSERT(used >= 0 || dsl_dir_phys(dd)->dd_used_bytes >= -used);
1544 ASSERT(compressed >= 0 ||
1545 dsl_dir_phys(dd)->dd_compressed_bytes >= -compressed);
1546 ASSERT(uncompressed >= 0 ||
1547 dsl_dir_phys(dd)->dd_uncompressed_bytes >= -uncompressed);
1548 dsl_dir_phys(dd)->dd_used_bytes += used;
1549 dsl_dir_phys(dd)->dd_uncompressed_bytes += uncompressed;
1550 dsl_dir_phys(dd)->dd_compressed_bytes += compressed;
1552 if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1554 dsl_dir_phys(dd)->dd_used_breakdown[type] >= -used);
1555 dsl_dir_phys(dd)->dd_used_breakdown[type] += used;
1560 for (t = 0; t < DD_USED_NUM; t++)
1561 u += dsl_dir_phys(dd)->dd_used_breakdown[t];
1562 ASSERT3U(u, ==, dsl_dir_phys(dd)->dd_used_bytes);
1567 mutex_exit(&dd->dd_lock);
1569 if (dd->dd_parent != NULL) {
1570 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1571 accounted_delta, compressed, uncompressed, tx);
1572 dsl_dir_transfer_space(dd->dd_parent,
1573 used - accounted_delta,
1574 DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
1579 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
1580 dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
1582 ASSERT(dmu_tx_is_syncing(tx));
1583 ASSERT(oldtype < DD_USED_NUM);
1584 ASSERT(newtype < DD_USED_NUM);
1587 !(dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN))
1590 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1591 mutex_enter(&dd->dd_lock);
1593 dsl_dir_phys(dd)->dd_used_breakdown[oldtype] >= delta :
1594 dsl_dir_phys(dd)->dd_used_breakdown[newtype] >= -delta);
1595 ASSERT(dsl_dir_phys(dd)->dd_used_bytes >= ABS(delta));
1596 dsl_dir_phys(dd)->dd_used_breakdown[oldtype] -= delta;
1597 dsl_dir_phys(dd)->dd_used_breakdown[newtype] += delta;
1598 mutex_exit(&dd->dd_lock);
1601 typedef struct dsl_dir_set_qr_arg {
1602 const char *ddsqra_name;
1603 zprop_source_t ddsqra_source;
1604 uint64_t ddsqra_value;
1605 } dsl_dir_set_qr_arg_t;
1608 dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx)
1610 dsl_dir_set_qr_arg_t *ddsqra = arg;
1611 dsl_pool_t *dp = dmu_tx_pool(tx);
1614 uint64_t towrite, newval;
1616 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1620 error = dsl_prop_predict(ds->ds_dir, "quota",
1621 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1623 dsl_dataset_rele(ds, FTAG);
1628 dsl_dataset_rele(ds, FTAG);
1632 mutex_enter(&ds->ds_dir->dd_lock);
1634 * If we are doing the preliminary check in open context, and
1635 * there are pending changes, then don't fail it, since the
1636 * pending changes could under-estimate the amount of space to be
1639 towrite = dsl_dir_space_towrite(ds->ds_dir);
1640 if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
1641 (newval < dsl_dir_phys(ds->ds_dir)->dd_reserved ||
1642 newval < dsl_dir_phys(ds->ds_dir)->dd_used_bytes + towrite)) {
1643 error = SET_ERROR(ENOSPC);
1645 mutex_exit(&ds->ds_dir->dd_lock);
1646 dsl_dataset_rele(ds, FTAG);
1651 dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx)
1653 dsl_dir_set_qr_arg_t *ddsqra = arg;
1654 dsl_pool_t *dp = dmu_tx_pool(tx);
1658 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1660 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1661 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA),
1662 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1663 &ddsqra->ddsqra_value, tx);
1665 VERIFY0(dsl_prop_get_int_ds(ds,
1666 zfs_prop_to_name(ZFS_PROP_QUOTA), &newval));
1668 newval = ddsqra->ddsqra_value;
1669 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1670 zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval);
1673 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1674 mutex_enter(&ds->ds_dir->dd_lock);
1675 dsl_dir_phys(ds->ds_dir)->dd_quota = newval;
1676 mutex_exit(&ds->ds_dir->dd_lock);
1677 dsl_dataset_rele(ds, FTAG);
1681 dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
1683 dsl_dir_set_qr_arg_t ddsqra;
1685 ddsqra.ddsqra_name = ddname;
1686 ddsqra.ddsqra_source = source;
1687 ddsqra.ddsqra_value = quota;
1689 return (dsl_sync_task(ddname, dsl_dir_set_quota_check,
1690 dsl_dir_set_quota_sync, &ddsqra, 0,
1691 ZFS_SPACE_CHECK_EXTRA_RESERVED));
1695 dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx)
1697 dsl_dir_set_qr_arg_t *ddsqra = arg;
1698 dsl_pool_t *dp = dmu_tx_pool(tx);
1701 uint64_t newval, used, avail;
1704 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1710 * If we are doing the preliminary check in open context, the
1711 * space estimates may be inaccurate.
1713 if (!dmu_tx_is_syncing(tx)) {
1714 dsl_dataset_rele(ds, FTAG);
1718 error = dsl_prop_predict(ds->ds_dir,
1719 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1720 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1722 dsl_dataset_rele(ds, FTAG);
1726 mutex_enter(&dd->dd_lock);
1727 used = dsl_dir_phys(dd)->dd_used_bytes;
1728 mutex_exit(&dd->dd_lock);
1730 if (dd->dd_parent) {
1731 avail = dsl_dir_space_available(dd->dd_parent,
1734 avail = dsl_pool_adjustedsize(dd->dd_pool,
1735 ZFS_SPACE_CHECK_NORMAL) - used;
1738 if (MAX(used, newval) > MAX(used, dsl_dir_phys(dd)->dd_reserved)) {
1739 uint64_t delta = MAX(used, newval) -
1740 MAX(used, dsl_dir_phys(dd)->dd_reserved);
1742 if (delta > avail ||
1743 (dsl_dir_phys(dd)->dd_quota > 0 &&
1744 newval > dsl_dir_phys(dd)->dd_quota))
1745 error = SET_ERROR(ENOSPC);
1748 dsl_dataset_rele(ds, FTAG);
1753 dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
1758 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1760 mutex_enter(&dd->dd_lock);
1761 used = dsl_dir_phys(dd)->dd_used_bytes;
1762 delta = MAX(used, value) - MAX(used, dsl_dir_phys(dd)->dd_reserved);
1763 dsl_dir_phys(dd)->dd_reserved = value;
1765 if (dd->dd_parent != NULL) {
1766 /* Roll up this additional usage into our ancestors */
1767 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1770 mutex_exit(&dd->dd_lock);
1774 dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx)
1776 dsl_dir_set_qr_arg_t *ddsqra = arg;
1777 dsl_pool_t *dp = dmu_tx_pool(tx);
1781 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1783 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1784 dsl_prop_set_sync_impl(ds,
1785 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1786 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1787 &ddsqra->ddsqra_value, tx);
1789 VERIFY0(dsl_prop_get_int_ds(ds,
1790 zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval));
1792 newval = ddsqra->ddsqra_value;
1793 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1794 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1795 (longlong_t)newval);
1798 dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx);
1799 dsl_dataset_rele(ds, FTAG);
1803 dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1804 uint64_t reservation)
1806 dsl_dir_set_qr_arg_t ddsqra;
1808 ddsqra.ddsqra_name = ddname;
1809 ddsqra.ddsqra_source = source;
1810 ddsqra.ddsqra_value = reservation;
1812 return (dsl_sync_task(ddname, dsl_dir_set_reservation_check,
1813 dsl_dir_set_reservation_sync, &ddsqra, 0,
1814 ZFS_SPACE_CHECK_EXTRA_RESERVED));
1818 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1820 for (; ds1; ds1 = ds1->dd_parent) {
1822 for (dd = ds2; dd; dd = dd->dd_parent) {
1831 * If delta is applied to dd, how much of that delta would be applied to
1832 * ancestor? Syncing context only.
1835 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1840 mutex_enter(&dd->dd_lock);
1841 delta = parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, delta);
1842 mutex_exit(&dd->dd_lock);
1843 return (would_change(dd->dd_parent, delta, ancestor));
1846 typedef struct dsl_dir_rename_arg {
1847 const char *ddra_oldname;
1848 const char *ddra_newname;
1851 } dsl_dir_rename_arg_t;
1853 typedef struct dsl_valid_rename_arg {
1856 } dsl_valid_rename_arg_t;
1860 dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1862 dsl_valid_rename_arg_t *dvra = arg;
1863 char namebuf[ZFS_MAX_DATASET_NAME_LEN];
1865 dsl_dataset_name(ds, namebuf);
1867 ASSERT3U(strnlen(namebuf, ZFS_MAX_DATASET_NAME_LEN),
1868 <, ZFS_MAX_DATASET_NAME_LEN);
1869 int namelen = strlen(namebuf) + dvra->char_delta;
1870 int depth = get_dataset_depth(namebuf) + dvra->nest_delta;
1872 if (namelen >= ZFS_MAX_DATASET_NAME_LEN)
1873 return (SET_ERROR(ENAMETOOLONG));
1874 if (dvra->nest_delta > 0 && depth >= zfs_max_dataset_nesting)
1875 return (SET_ERROR(ENAMETOOLONG));
1880 dsl_dir_rename_check(void *arg, dmu_tx_t *tx)
1882 dsl_dir_rename_arg_t *ddra = arg;
1883 dsl_pool_t *dp = dmu_tx_pool(tx);
1884 dsl_dir_t *dd, *newparent;
1885 dsl_valid_rename_arg_t dvra;
1886 dsl_dataset_t *parentds;
1888 const char *mynewname;
1891 /* target dir should exist */
1892 error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL);
1896 /* new parent should exist */
1897 error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG,
1898 &newparent, &mynewname);
1900 dsl_dir_rele(dd, FTAG);
1904 /* can't rename to different pool */
1905 if (dd->dd_pool != newparent->dd_pool) {
1906 dsl_dir_rele(newparent, FTAG);
1907 dsl_dir_rele(dd, FTAG);
1908 return (SET_ERROR(EXDEV));
1911 /* new name should not already exist */
1912 if (mynewname == NULL) {
1913 dsl_dir_rele(newparent, FTAG);
1914 dsl_dir_rele(dd, FTAG);
1915 return (SET_ERROR(EEXIST));
1918 /* can't rename below anything but filesystems (eg. no ZVOLs) */
1919 error = dsl_dataset_hold_obj(newparent->dd_pool,
1920 dsl_dir_phys(newparent)->dd_head_dataset_obj, FTAG, &parentds);
1922 dsl_dir_rele(newparent, FTAG);
1923 dsl_dir_rele(dd, FTAG);
1926 error = dmu_objset_from_ds(parentds, &parentos);
1928 dsl_dataset_rele(parentds, FTAG);
1929 dsl_dir_rele(newparent, FTAG);
1930 dsl_dir_rele(dd, FTAG);
1933 if (dmu_objset_type(parentos) != DMU_OST_ZFS) {
1934 dsl_dataset_rele(parentds, FTAG);
1935 dsl_dir_rele(newparent, FTAG);
1936 dsl_dir_rele(dd, FTAG);
1937 return (SET_ERROR(ZFS_ERR_WRONG_PARENT));
1939 dsl_dataset_rele(parentds, FTAG);
1941 ASSERT3U(strnlen(ddra->ddra_newname, ZFS_MAX_DATASET_NAME_LEN),
1942 <, ZFS_MAX_DATASET_NAME_LEN);
1943 ASSERT3U(strnlen(ddra->ddra_oldname, ZFS_MAX_DATASET_NAME_LEN),
1944 <, ZFS_MAX_DATASET_NAME_LEN);
1945 dvra.char_delta = strlen(ddra->ddra_newname)
1946 - strlen(ddra->ddra_oldname);
1947 dvra.nest_delta = get_dataset_depth(ddra->ddra_newname)
1948 - get_dataset_depth(ddra->ddra_oldname);
1950 /* if the name length is growing, validate child name lengths */
1951 if (dvra.char_delta > 0 || dvra.nest_delta > 0) {
1952 error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename,
1953 &dvra, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1955 dsl_dir_rele(newparent, FTAG);
1956 dsl_dir_rele(dd, FTAG);
1961 if (dmu_tx_is_syncing(tx)) {
1962 if (spa_feature_is_active(dp->dp_spa,
1963 SPA_FEATURE_FS_SS_LIMIT)) {
1965 * Although this is the check function and we don't
1966 * normally make on-disk changes in check functions,
1967 * we need to do that here.
1969 * Ensure this portion of the tree's counts have been
1970 * initialized in case the new parent has limits set.
1972 dsl_dir_init_fs_ss_count(dd, tx);
1976 if (newparent != dd->dd_parent) {
1977 /* is there enough space? */
1979 MAX(dsl_dir_phys(dd)->dd_used_bytes,
1980 dsl_dir_phys(dd)->dd_reserved);
1981 objset_t *os = dd->dd_pool->dp_meta_objset;
1982 uint64_t fs_cnt = 0;
1983 uint64_t ss_cnt = 0;
1985 if (dsl_dir_is_zapified(dd)) {
1988 err = zap_lookup(os, dd->dd_object,
1989 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1991 if (err != ENOENT && err != 0) {
1992 dsl_dir_rele(newparent, FTAG);
1993 dsl_dir_rele(dd, FTAG);
1998 * have to add 1 for the filesystem itself that we're
2003 err = zap_lookup(os, dd->dd_object,
2004 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
2006 if (err != ENOENT && err != 0) {
2007 dsl_dir_rele(newparent, FTAG);
2008 dsl_dir_rele(dd, FTAG);
2013 /* check for encryption errors */
2014 error = dsl_dir_rename_crypt_check(dd, newparent);
2016 dsl_dir_rele(newparent, FTAG);
2017 dsl_dir_rele(dd, FTAG);
2018 return (SET_ERROR(EACCES));
2021 /* no rename into our descendant */
2022 if (closest_common_ancestor(dd, newparent) == dd) {
2023 dsl_dir_rele(newparent, FTAG);
2024 dsl_dir_rele(dd, FTAG);
2025 return (SET_ERROR(EINVAL));
2028 error = dsl_dir_transfer_possible(dd->dd_parent,
2029 newparent, fs_cnt, ss_cnt, myspace,
2030 ddra->ddra_cred, ddra->ddra_proc);
2032 dsl_dir_rele(newparent, FTAG);
2033 dsl_dir_rele(dd, FTAG);
2038 dsl_dir_rele(newparent, FTAG);
2039 dsl_dir_rele(dd, FTAG);
2044 dsl_dir_rename_sync(void *arg, dmu_tx_t *tx)
2046 dsl_dir_rename_arg_t *ddra = arg;
2047 dsl_pool_t *dp = dmu_tx_pool(tx);
2048 dsl_dir_t *dd, *newparent;
2049 const char *mynewname;
2051 objset_t *mos = dp->dp_meta_objset;
2053 VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL));
2054 VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
2057 /* Log this before we change the name. */
2058 spa_history_log_internal_dd(dd, "rename", tx,
2059 "-> %s", ddra->ddra_newname);
2061 if (newparent != dd->dd_parent) {
2062 objset_t *os = dd->dd_pool->dp_meta_objset;
2063 uint64_t fs_cnt = 0;
2064 uint64_t ss_cnt = 0;
2067 * We already made sure the dd counts were initialized in the
2070 if (spa_feature_is_active(dp->dp_spa,
2071 SPA_FEATURE_FS_SS_LIMIT)) {
2072 VERIFY0(zap_lookup(os, dd->dd_object,
2073 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
2075 /* add 1 for the filesystem itself that we're moving */
2078 VERIFY0(zap_lookup(os, dd->dd_object,
2079 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
2083 dsl_fs_ss_count_adjust(dd->dd_parent, -fs_cnt,
2084 DD_FIELD_FILESYSTEM_COUNT, tx);
2085 dsl_fs_ss_count_adjust(newparent, fs_cnt,
2086 DD_FIELD_FILESYSTEM_COUNT, tx);
2088 dsl_fs_ss_count_adjust(dd->dd_parent, -ss_cnt,
2089 DD_FIELD_SNAPSHOT_COUNT, tx);
2090 dsl_fs_ss_count_adjust(newparent, ss_cnt,
2091 DD_FIELD_SNAPSHOT_COUNT, tx);
2093 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
2094 -dsl_dir_phys(dd)->dd_used_bytes,
2095 -dsl_dir_phys(dd)->dd_compressed_bytes,
2096 -dsl_dir_phys(dd)->dd_uncompressed_bytes, tx);
2097 dsl_dir_diduse_space(newparent, DD_USED_CHILD,
2098 dsl_dir_phys(dd)->dd_used_bytes,
2099 dsl_dir_phys(dd)->dd_compressed_bytes,
2100 dsl_dir_phys(dd)->dd_uncompressed_bytes, tx);
2102 if (dsl_dir_phys(dd)->dd_reserved >
2103 dsl_dir_phys(dd)->dd_used_bytes) {
2104 uint64_t unused_rsrv = dsl_dir_phys(dd)->dd_reserved -
2105 dsl_dir_phys(dd)->dd_used_bytes;
2107 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
2108 -unused_rsrv, 0, 0, tx);
2109 dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV,
2110 unused_rsrv, 0, 0, tx);
2114 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2116 /* remove from old parent zapobj */
2117 error = zap_remove(mos,
2118 dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj,
2122 (void) strlcpy(dd->dd_myname, mynewname,
2123 sizeof (dd->dd_myname));
2124 dsl_dir_rele(dd->dd_parent, dd);
2125 dsl_dir_phys(dd)->dd_parent_obj = newparent->dd_object;
2126 VERIFY0(dsl_dir_hold_obj(dp,
2127 newparent->dd_object, NULL, dd, &dd->dd_parent));
2129 /* add to new parent zapobj */
2130 VERIFY0(zap_add(mos, dsl_dir_phys(newparent)->dd_child_dir_zapobj,
2131 dd->dd_myname, 8, 1, &dd->dd_object, tx));
2133 zvol_rename_minors(dp->dp_spa, ddra->ddra_oldname,
2134 ddra->ddra_newname, B_TRUE);
2136 dsl_prop_notify_all(dd);
2138 dsl_dir_rele(newparent, FTAG);
2139 dsl_dir_rele(dd, FTAG);
2143 dsl_dir_rename(const char *oldname, const char *newname)
2145 dsl_dir_rename_arg_t ddra;
2147 ddra.ddra_oldname = oldname;
2148 ddra.ddra_newname = newname;
2149 ddra.ddra_cred = CRED();
2150 ddra.ddra_proc = curproc;
2152 return (dsl_sync_task(oldname,
2153 dsl_dir_rename_check, dsl_dir_rename_sync, &ddra,
2154 3, ZFS_SPACE_CHECK_RESERVED));
2158 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd,
2159 uint64_t fs_cnt, uint64_t ss_cnt, uint64_t space,
2160 cred_t *cr, proc_t *proc)
2162 dsl_dir_t *ancestor;
2167 ancestor = closest_common_ancestor(sdd, tdd);
2168 adelta = would_change(sdd, -space, ancestor);
2169 avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
2171 return (SET_ERROR(ENOSPC));
2173 err = dsl_fs_ss_limit_check(tdd, fs_cnt, ZFS_PROP_FILESYSTEM_LIMIT,
2174 ancestor, cr, proc);
2177 err = dsl_fs_ss_limit_check(tdd, ss_cnt, ZFS_PROP_SNAPSHOT_LIMIT,
2178 ancestor, cr, proc);
2186 dsl_dir_snap_cmtime(dsl_dir_t *dd)
2190 mutex_enter(&dd->dd_lock);
2191 t = dd->dd_snap_cmtime;
2192 mutex_exit(&dd->dd_lock);
2198 dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
2203 mutex_enter(&dd->dd_lock);
2204 dd->dd_snap_cmtime = t;
2205 mutex_exit(&dd->dd_lock);
2209 dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx)
2211 objset_t *mos = dd->dd_pool->dp_meta_objset;
2212 dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx);
2216 dsl_dir_is_zapified(dsl_dir_t *dd)
2218 dmu_object_info_t doi;
2220 dmu_object_info_from_db(dd->dd_dbuf, &doi);
2221 return (doi.doi_type == DMU_OTN_ZAP_METADATA);
2225 dsl_dir_livelist_open(dsl_dir_t *dd, uint64_t obj)
2227 objset_t *mos = dd->dd_pool->dp_meta_objset;
2228 ASSERT(spa_feature_is_active(dd->dd_pool->dp_spa,
2229 SPA_FEATURE_LIVELIST));
2230 dsl_deadlist_open(&dd->dd_livelist, mos, obj);
2231 bplist_create(&dd->dd_pending_allocs);
2232 bplist_create(&dd->dd_pending_frees);
2236 dsl_dir_livelist_close(dsl_dir_t *dd)
2238 dsl_deadlist_close(&dd->dd_livelist);
2239 bplist_destroy(&dd->dd_pending_allocs);
2240 bplist_destroy(&dd->dd_pending_frees);
2244 dsl_dir_remove_livelist(dsl_dir_t *dd, dmu_tx_t *tx, boolean_t total)
2247 dsl_pool_t *dp = dmu_tx_pool(tx);
2248 spa_t *spa = dp->dp_spa;
2249 livelist_condense_entry_t to_condense = spa->spa_to_condense;
2251 if (!dsl_deadlist_is_open(&dd->dd_livelist))
2255 * If the livelist being removed is set to be condensed, stop the
2256 * condense zthr and indicate the cancellation in the spa_to_condense
2257 * struct in case the condense no-wait synctask has already started
2259 zthr_t *ll_condense_thread = spa->spa_livelist_condense_zthr;
2260 if (ll_condense_thread != NULL &&
2261 (to_condense.ds != NULL) && (to_condense.ds->ds_dir == dd)) {
2263 * We use zthr_wait_cycle_done instead of zthr_cancel
2264 * because we don't want to destroy the zthr, just have
2265 * it skip its current task.
2267 spa->spa_to_condense.cancelled = B_TRUE;
2268 zthr_wait_cycle_done(ll_condense_thread);
2270 * If we've returned from zthr_wait_cycle_done without
2271 * clearing the to_condense data structure it's either
2272 * because the no-wait synctask has started (which is
2273 * indicated by 'syncing' field of to_condense) and we
2274 * can expect it to clear to_condense on its own.
2275 * Otherwise, we returned before the zthr ran. The
2276 * checkfunc will now fail as cancelled == B_TRUE so we
2277 * can safely NULL out ds, allowing a different dir's
2278 * livelist to be condensed.
2280 * We can be sure that the to_condense struct will not
2281 * be repopulated at this stage because both this
2282 * function and dsl_livelist_try_condense execute in
2285 if ((spa->spa_to_condense.ds != NULL) &&
2286 !spa->spa_to_condense.syncing) {
2287 dmu_buf_rele(spa->spa_to_condense.ds->ds_dbuf,
2289 spa->spa_to_condense.ds = NULL;
2293 dsl_dir_livelist_close(dd);
2294 int err = zap_lookup(dp->dp_meta_objset, dd->dd_object,
2295 DD_FIELD_LIVELIST, sizeof (uint64_t), 1, &obj);
2297 VERIFY0(zap_remove(dp->dp_meta_objset, dd->dd_object,
2298 DD_FIELD_LIVELIST, tx));
2300 dsl_deadlist_free(dp->dp_meta_objset, obj, tx);
2301 spa_feature_decr(spa, SPA_FEATURE_LIVELIST, tx);
2304 ASSERT3U(err, !=, ENOENT);
2309 dsl_dir_activity_in_progress(dsl_dir_t *dd, dsl_dataset_t *ds,
2310 zfs_wait_activity_t activity, boolean_t *in_progress)
2314 ASSERT(MUTEX_HELD(&dd->dd_activity_lock));
2317 case ZFS_WAIT_DELETEQ: {
2320 error = dmu_objset_from_ds(ds, &os);
2324 mutex_enter(&os->os_user_ptr_lock);
2325 void *user = dmu_objset_get_user(os);
2326 mutex_exit(&os->os_user_ptr_lock);
2327 if (dmu_objset_type(os) != DMU_OST_ZFS ||
2328 user == NULL || zfs_get_vfs_flag_unmounted(os)) {
2329 *in_progress = B_FALSE;
2333 uint64_t readonly = B_FALSE;
2334 error = zfs_get_temporary_prop(ds, ZFS_PROP_READONLY, &readonly,
2340 if (readonly || !spa_writeable(dd->dd_pool->dp_spa)) {
2341 *in_progress = B_FALSE;
2345 uint64_t count, unlinked_obj;
2346 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
2349 dsl_dataset_rele(ds, FTAG);
2352 error = zap_count(os, unlinked_obj, &count);
2355 *in_progress = (count != 0);
2359 * The delete queue is ZPL specific, and libzpool doesn't have
2360 * it. It doesn't make sense to wait for it.
2362 *in_progress = B_FALSE;
2367 panic("unrecognized value for activity %d", activity);
2374 dsl_dir_wait(dsl_dir_t *dd, dsl_dataset_t *ds, zfs_wait_activity_t activity,
2378 boolean_t in_progress;
2379 dsl_pool_t *dp = dd->dd_pool;
2381 dsl_pool_config_enter(dp, FTAG);
2382 error = dsl_dir_activity_in_progress(dd, ds, activity,
2384 dsl_pool_config_exit(dp, FTAG);
2385 if (error != 0 || !in_progress)
2390 if (cv_wait_sig(&dd->dd_activity_cv, &dd->dd_activity_lock) ==
2391 0 || dd->dd_activity_cancelled) {
2392 error = SET_ERROR(EINTR);
2400 dsl_dir_cancel_waiters(dsl_dir_t *dd)
2402 mutex_enter(&dd->dd_activity_lock);
2403 dd->dd_activity_cancelled = B_TRUE;
2404 cv_broadcast(&dd->dd_activity_cv);
2405 while (dd->dd_activity_waiters > 0)
2406 cv_wait(&dd->dd_activity_cv, &dd->dd_activity_lock);
2407 mutex_exit(&dd->dd_activity_lock);
2410 #if defined(_KERNEL)
2411 EXPORT_SYMBOL(dsl_dir_set_quota);
2412 EXPORT_SYMBOL(dsl_dir_set_reservation);