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 https://opensource.org/licenses/CDDL-1.0.
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 static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
126 typedef struct ddulrt_arg {
127 dsl_dir_t *ddulrta_dd;
132 dsl_dir_evict_async(void *dbu)
136 dsl_pool_t *dp __maybe_unused = dd->dd_pool;
140 for (t = 0; t < TXG_SIZE; t++) {
141 ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t));
142 ASSERT(dd->dd_tempreserved[t] == 0);
143 ASSERT(dd->dd_space_towrite[t] == 0);
147 dsl_dir_async_rele(dd->dd_parent, dd);
149 spa_async_close(dd->dd_pool->dp_spa, dd);
151 if (dsl_deadlist_is_open(&dd->dd_livelist))
152 dsl_dir_livelist_close(dd);
155 cv_destroy(&dd->dd_activity_cv);
156 mutex_destroy(&dd->dd_activity_lock);
157 mutex_destroy(&dd->dd_lock);
158 kmem_free(dd, sizeof (dsl_dir_t));
162 dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj,
163 const char *tail, const void *tag, dsl_dir_t **ddp)
167 dmu_object_info_t doi;
170 ASSERT(dsl_pool_config_held(dp));
172 err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
175 dd = dmu_buf_get_user(dbuf);
177 dmu_object_info_from_db(dbuf, &doi);
178 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_DSL_DIR);
179 ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
184 dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
185 dd->dd_object = ddobj;
189 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
190 mutex_init(&dd->dd_activity_lock, NULL, MUTEX_DEFAULT, NULL);
191 cv_init(&dd->dd_activity_cv, NULL, CV_DEFAULT, NULL);
194 if (dsl_dir_is_zapified(dd)) {
195 err = zap_lookup(dp->dp_meta_objset,
196 ddobj, DD_FIELD_CRYPTO_KEY_OBJ,
197 sizeof (uint64_t), 1, &dd->dd_crypto_obj);
199 /* check for on-disk format errata */
200 if (dsl_dir_incompatible_encryption_version(
202 dp->dp_spa->spa_errata =
203 ZPOOL_ERRATA_ZOL_6845_ENCRYPTION;
205 } else if (err != ENOENT) {
210 if (dsl_dir_phys(dd)->dd_parent_obj) {
211 err = dsl_dir_hold_obj(dp,
212 dsl_dir_phys(dd)->dd_parent_obj, NULL, dd,
220 err = zap_lookup(dp->dp_meta_objset,
221 dsl_dir_phys(dd->dd_parent)->
222 dd_child_dir_zapobj, tail,
223 sizeof (foundobj), 1, &foundobj);
224 ASSERT(err || foundobj == ddobj);
226 (void) strlcpy(dd->dd_myname, tail,
227 sizeof (dd->dd_myname));
229 err = zap_value_search(dp->dp_meta_objset,
230 dsl_dir_phys(dd->dd_parent)->
232 ddobj, 0, dd->dd_myname);
237 (void) strlcpy(dd->dd_myname, spa_name(dp->dp_spa),
238 sizeof (dd->dd_myname));
241 if (dsl_dir_is_clone(dd)) {
242 dmu_buf_t *origin_bonus;
243 dsl_dataset_phys_t *origin_phys;
246 * We can't open the origin dataset, because
247 * that would require opening this dsl_dir.
248 * Just look at its phys directly instead.
250 err = dmu_bonus_hold(dp->dp_meta_objset,
251 dsl_dir_phys(dd)->dd_origin_obj, FTAG,
255 origin_phys = origin_bonus->db_data;
257 origin_phys->ds_creation_txg;
258 dmu_buf_rele(origin_bonus, FTAG);
259 if (dsl_dir_is_zapified(dd)) {
261 err = zap_lookup(dp->dp_meta_objset,
262 dd->dd_object, DD_FIELD_LIVELIST,
263 sizeof (uint64_t), 1, &obj);
265 dsl_dir_livelist_open(dd, obj);
266 else if (err != ENOENT)
271 if (dsl_dir_is_zapified(dd)) {
272 inode_timespec_t t = {0};
273 zap_lookup(dp->dp_meta_objset, ddobj,
274 DD_FIELD_SNAPSHOTS_CHANGED,
276 sizeof (inode_timespec_t) / sizeof (uint64_t),
278 dd->dd_snap_cmtime = t;
281 dmu_buf_init_user(&dd->dd_dbu, NULL, dsl_dir_evict_async,
283 winner = dmu_buf_set_user_ie(dbuf, &dd->dd_dbu);
284 if (winner != NULL) {
286 dsl_dir_rele(dd->dd_parent, dd);
287 if (dsl_deadlist_is_open(&dd->dd_livelist))
288 dsl_dir_livelist_close(dd);
290 cv_destroy(&dd->dd_activity_cv);
291 mutex_destroy(&dd->dd_activity_lock);
292 mutex_destroy(&dd->dd_lock);
293 kmem_free(dd, sizeof (dsl_dir_t));
296 spa_open_ref(dp->dp_spa, dd);
301 * The dsl_dir_t has both open-to-close and instantiate-to-evict
302 * holds on the spa. We need the open-to-close holds because
303 * otherwise the spa_refcnt wouldn't change when we open a
304 * dir which the spa also has open, so we could incorrectly
305 * think it was OK to unload/export/destroy the pool. We need
306 * the instantiate-to-evict hold because the dsl_dir_t has a
307 * pointer to the dd_pool, which has a pointer to the spa_t.
309 spa_open_ref(dp->dp_spa, tag);
310 ASSERT3P(dd->dd_pool, ==, dp);
311 ASSERT3U(dd->dd_object, ==, ddobj);
312 ASSERT3P(dd->dd_dbuf, ==, dbuf);
318 dsl_dir_rele(dd->dd_parent, dd);
319 if (dsl_deadlist_is_open(&dd->dd_livelist))
320 dsl_dir_livelist_close(dd);
322 cv_destroy(&dd->dd_activity_cv);
323 mutex_destroy(&dd->dd_activity_lock);
324 mutex_destroy(&dd->dd_lock);
325 kmem_free(dd, sizeof (dsl_dir_t));
326 dmu_buf_rele(dbuf, tag);
331 dsl_dir_rele(dsl_dir_t *dd, const void *tag)
333 dprintf_dd(dd, "%s\n", "");
334 spa_close(dd->dd_pool->dp_spa, tag);
335 dmu_buf_rele(dd->dd_dbuf, tag);
339 * Remove a reference to the given dsl dir that is being asynchronously
340 * released. Async releases occur from a taskq performing eviction of
341 * dsl datasets and dirs. This process is identical to a normal release
342 * with the exception of using the async API for releasing the reference on
346 dsl_dir_async_rele(dsl_dir_t *dd, const void *tag)
348 dprintf_dd(dd, "%s\n", "");
349 spa_async_close(dd->dd_pool->dp_spa, tag);
350 dmu_buf_rele(dd->dd_dbuf, tag);
353 /* buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes */
355 dsl_dir_name(dsl_dir_t *dd, char *buf)
358 dsl_dir_name(dd->dd_parent, buf);
359 VERIFY3U(strlcat(buf, "/", ZFS_MAX_DATASET_NAME_LEN), <,
360 ZFS_MAX_DATASET_NAME_LEN);
364 if (!MUTEX_HELD(&dd->dd_lock)) {
366 * recursive mutex so that we can use
367 * dprintf_dd() with dd_lock held
369 mutex_enter(&dd->dd_lock);
370 VERIFY3U(strlcat(buf, dd->dd_myname, ZFS_MAX_DATASET_NAME_LEN),
371 <, ZFS_MAX_DATASET_NAME_LEN);
372 mutex_exit(&dd->dd_lock);
374 VERIFY3U(strlcat(buf, dd->dd_myname, ZFS_MAX_DATASET_NAME_LEN),
375 <, ZFS_MAX_DATASET_NAME_LEN);
379 /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
381 dsl_dir_namelen(dsl_dir_t *dd)
386 /* parent's name + 1 for the "/" */
387 result = dsl_dir_namelen(dd->dd_parent) + 1;
390 if (!MUTEX_HELD(&dd->dd_lock)) {
391 /* see dsl_dir_name */
392 mutex_enter(&dd->dd_lock);
393 result += strlen(dd->dd_myname);
394 mutex_exit(&dd->dd_lock);
396 result += strlen(dd->dd_myname);
403 getcomponent(const char *path, char *component, const char **nextp)
407 if ((path == NULL) || (path[0] == '\0'))
408 return (SET_ERROR(ENOENT));
409 /* This would be a good place to reserve some namespace... */
410 p = strpbrk(path, "/@");
411 if (p && (p[1] == '/' || p[1] == '@')) {
412 /* two separators in a row */
413 return (SET_ERROR(EINVAL));
415 if (p == NULL || p == path) {
417 * if the first thing is an @ or /, it had better be an
418 * @ and it had better not have any more ats or slashes,
419 * and it had better have something after the @.
422 (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
423 return (SET_ERROR(EINVAL));
424 if (strlen(path) >= ZFS_MAX_DATASET_NAME_LEN)
425 return (SET_ERROR(ENAMETOOLONG));
426 (void) strlcpy(component, path, ZFS_MAX_DATASET_NAME_LEN);
428 } else if (p[0] == '/') {
429 if (p - path >= ZFS_MAX_DATASET_NAME_LEN)
430 return (SET_ERROR(ENAMETOOLONG));
431 (void) strncpy(component, path, p - path);
432 component[p - path] = '\0';
434 } else if (p[0] == '@') {
436 * if the next separator is an @, there better not be
439 if (strchr(path, '/'))
440 return (SET_ERROR(EINVAL));
441 if (p - path >= ZFS_MAX_DATASET_NAME_LEN)
442 return (SET_ERROR(ENAMETOOLONG));
443 (void) strncpy(component, path, p - path);
444 component[p - path] = '\0';
446 panic("invalid p=%p", (void *)p);
453 * Return the dsl_dir_t, and possibly the last component which couldn't
454 * be found in *tail. The name must be in the specified dsl_pool_t. This
455 * thread must hold the dp_config_rwlock for the pool. Returns NULL if the
456 * path is bogus, or if tail==NULL and we couldn't parse the whole name.
457 * (*tail)[0] == '@' means that the last component is a snapshot.
460 dsl_dir_hold(dsl_pool_t *dp, const char *name, const void *tag,
461 dsl_dir_t **ddp, const char **tailp)
464 const char *spaname, *next, *nextnext = NULL;
469 buf = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
470 err = getcomponent(name, buf, &next);
474 /* Make sure the name is in the specified pool. */
475 spaname = spa_name(dp->dp_spa);
476 if (strcmp(buf, spaname) != 0) {
477 err = SET_ERROR(EXDEV);
481 ASSERT(dsl_pool_config_held(dp));
483 err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
488 while (next != NULL) {
490 err = getcomponent(next, buf, &nextnext);
493 ASSERT(next[0] != '\0');
496 dprintf("looking up %s in obj%lld\n",
497 buf, (longlong_t)dsl_dir_phys(dd)->dd_child_dir_zapobj);
499 err = zap_lookup(dp->dp_meta_objset,
500 dsl_dir_phys(dd)->dd_child_dir_zapobj,
501 buf, sizeof (ddobj), 1, &ddobj);
508 err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_dd);
511 dsl_dir_rele(dd, tag);
517 dsl_dir_rele(dd, tag);
522 * It's an error if there's more than one component left, or
523 * tailp==NULL and there's any component left.
526 (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
528 dsl_dir_rele(dd, tag);
529 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
530 err = SET_ERROR(ENOENT);
537 kmem_free(buf, ZFS_MAX_DATASET_NAME_LEN);
542 * If the counts are already initialized for this filesystem and its
543 * descendants then do nothing, otherwise initialize the counts.
545 * The counts on this filesystem, and those below, may be uninitialized due to
546 * either the use of a pre-existing pool which did not support the
547 * filesystem/snapshot limit feature, or one in which the feature had not yet
550 * Recursively descend the filesystem tree and update the filesystem/snapshot
551 * counts on each filesystem below, then update the cumulative count on the
552 * current filesystem. If the filesystem already has a count set on it,
553 * then we know that its counts, and the counts on the filesystems below it,
554 * are already correct, so we don't have to update this filesystem.
557 dsl_dir_init_fs_ss_count(dsl_dir_t *dd, dmu_tx_t *tx)
559 uint64_t my_fs_cnt = 0;
560 uint64_t my_ss_cnt = 0;
561 dsl_pool_t *dp = dd->dd_pool;
562 objset_t *os = dp->dp_meta_objset;
567 ASSERT(spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT));
568 ASSERT(dsl_pool_config_held(dp));
569 ASSERT(dmu_tx_is_syncing(tx));
571 dsl_dir_zapify(dd, tx);
574 * If the filesystem count has already been initialized then we
575 * don't need to recurse down any further.
577 if (zap_contains(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT) == 0)
580 zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP);
581 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
583 /* Iterate my child dirs */
584 for (zap_cursor_init(zc, os, dsl_dir_phys(dd)->dd_child_dir_zapobj);
585 zap_cursor_retrieve(zc, za) == 0; zap_cursor_advance(zc)) {
589 VERIFY0(dsl_dir_hold_obj(dp, za->za_first_integer, NULL, FTAG,
593 * Ignore hidden ($FREE, $MOS & $ORIGIN) objsets.
595 if (chld_dd->dd_myname[0] == '$') {
596 dsl_dir_rele(chld_dd, FTAG);
600 my_fs_cnt++; /* count this child */
602 dsl_dir_init_fs_ss_count(chld_dd, tx);
604 VERIFY0(zap_lookup(os, chld_dd->dd_object,
605 DD_FIELD_FILESYSTEM_COUNT, sizeof (count), 1, &count));
607 VERIFY0(zap_lookup(os, chld_dd->dd_object,
608 DD_FIELD_SNAPSHOT_COUNT, sizeof (count), 1, &count));
611 dsl_dir_rele(chld_dd, FTAG);
614 /* Count my snapshots (we counted children's snapshots above) */
615 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
616 dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds));
618 for (zap_cursor_init(zc, os, dsl_dataset_phys(ds)->ds_snapnames_zapobj);
619 zap_cursor_retrieve(zc, za) == 0;
620 zap_cursor_advance(zc)) {
621 /* Don't count temporary snapshots */
622 if (za->za_name[0] != '%')
627 dsl_dataset_rele(ds, FTAG);
629 kmem_free(zc, sizeof (zap_cursor_t));
630 kmem_free(za, sizeof (zap_attribute_t));
632 /* we're in a sync task, update counts */
633 dmu_buf_will_dirty(dd->dd_dbuf, tx);
634 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
635 sizeof (my_fs_cnt), 1, &my_fs_cnt, tx));
636 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
637 sizeof (my_ss_cnt), 1, &my_ss_cnt, tx));
641 dsl_dir_actv_fs_ss_limit_check(void *arg, dmu_tx_t *tx)
643 char *ddname = (char *)arg;
644 dsl_pool_t *dp = dmu_tx_pool(tx);
649 error = dsl_dataset_hold(dp, ddname, FTAG, &ds);
653 if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
654 dsl_dataset_rele(ds, FTAG);
655 return (SET_ERROR(ENOTSUP));
659 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT) &&
660 dsl_dir_is_zapified(dd) &&
661 zap_contains(dp->dp_meta_objset, dd->dd_object,
662 DD_FIELD_FILESYSTEM_COUNT) == 0) {
663 dsl_dataset_rele(ds, FTAG);
664 return (SET_ERROR(EALREADY));
667 dsl_dataset_rele(ds, FTAG);
672 dsl_dir_actv_fs_ss_limit_sync(void *arg, dmu_tx_t *tx)
674 char *ddname = (char *)arg;
675 dsl_pool_t *dp = dmu_tx_pool(tx);
679 VERIFY0(dsl_dataset_hold(dp, ddname, FTAG, &ds));
681 spa = dsl_dataset_get_spa(ds);
683 if (!spa_feature_is_active(spa, SPA_FEATURE_FS_SS_LIMIT)) {
685 * Since the feature was not active and we're now setting a
686 * limit, increment the feature-active counter so that the
687 * feature becomes active for the first time.
689 * We are already in a sync task so we can update the MOS.
691 spa_feature_incr(spa, SPA_FEATURE_FS_SS_LIMIT, tx);
695 * Since we are now setting a non-UINT64_MAX limit on the filesystem,
696 * we need to ensure the counts are correct. Descend down the tree from
697 * this point and update all of the counts to be accurate.
699 dsl_dir_init_fs_ss_count(ds->ds_dir, tx);
701 dsl_dataset_rele(ds, FTAG);
705 * Make sure the feature is enabled and activate it if necessary.
706 * Since we're setting a limit, ensure the on-disk counts are valid.
707 * This is only called by the ioctl path when setting a limit value.
709 * We do not need to validate the new limit, since users who can change the
710 * limit are also allowed to exceed the limit.
713 dsl_dir_activate_fs_ss_limit(const char *ddname)
717 error = dsl_sync_task(ddname, dsl_dir_actv_fs_ss_limit_check,
718 dsl_dir_actv_fs_ss_limit_sync, (void *)ddname, 0,
719 ZFS_SPACE_CHECK_RESERVED);
721 if (error == EALREADY)
728 * Used to determine if the filesystem_limit or snapshot_limit should be
729 * enforced. We allow the limit to be exceeded if the user has permission to
730 * write the property value. We pass in the creds that we got in the open
731 * context since we will always be the GZ root in syncing context. We also have
732 * to handle the case where we are allowed to change the limit on the current
733 * dataset, but there may be another limit in the tree above.
735 * We can never modify these two properties within a non-global zone. In
736 * addition, the other checks are modeled on zfs_secpolicy_write_perms. We
737 * can't use that function since we are already holding the dp_config_rwlock.
738 * In addition, we already have the dd and dealing with snapshots is simplified
749 dsl_enforce_ds_ss_limits(dsl_dir_t *dd, zfs_prop_t prop,
750 cred_t *cr, proc_t *proc)
752 enforce_res_t enforce = ENFORCE_ALWAYS;
756 const char *zonedstr;
758 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
759 prop == ZFS_PROP_SNAPSHOT_LIMIT);
762 if (crgetzoneid(cr) != GLOBAL_ZONEID)
763 return (ENFORCE_ALWAYS);
766 * We are checking the saved credentials of the user process, which is
767 * not the current process. Note that we can't use secpolicy_zfs(),
768 * because it only works if the cred is that of the current process (on
771 if (secpolicy_zfs_proc(cr, proc) == 0)
772 return (ENFORCE_NEVER);
777 if ((obj = dsl_dir_phys(dd)->dd_head_dataset_obj) == 0)
778 return (ENFORCE_ALWAYS);
780 ASSERT(dsl_pool_config_held(dd->dd_pool));
782 if (dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds) != 0)
783 return (ENFORCE_ALWAYS);
785 zonedstr = zfs_prop_to_name(ZFS_PROP_ZONED);
786 if (dsl_prop_get_ds(ds, zonedstr, 8, 1, &zoned, NULL) || zoned) {
787 /* Only root can access zoned fs's from the GZ */
788 enforce = ENFORCE_ALWAYS;
790 if (dsl_deleg_access_impl(ds, zfs_prop_to_name(prop), cr) == 0)
791 enforce = ENFORCE_ABOVE;
794 dsl_dataset_rele(ds, FTAG);
799 * Check if adding additional child filesystem(s) would exceed any filesystem
800 * limits or adding additional snapshot(s) would exceed any snapshot limits.
801 * The prop argument indicates which limit to check.
803 * Note that all filesystem limits up to the root (or the highest
804 * initialized) filesystem or the given ancestor must be satisfied.
807 dsl_fs_ss_limit_check(dsl_dir_t *dd, uint64_t delta, zfs_prop_t prop,
808 dsl_dir_t *ancestor, cred_t *cr, proc_t *proc)
810 objset_t *os = dd->dd_pool->dp_meta_objset;
811 uint64_t limit, count;
812 const char *count_prop;
813 enforce_res_t enforce;
816 ASSERT(dsl_pool_config_held(dd->dd_pool));
817 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
818 prop == ZFS_PROP_SNAPSHOT_LIMIT);
821 * If we're allowed to change the limit, don't enforce the limit
822 * e.g. this can happen if a snapshot is taken by an administrative
823 * user in the global zone (i.e. a recursive snapshot by root).
824 * However, we must handle the case of delegated permissions where we
825 * are allowed to change the limit on the current dataset, but there
826 * is another limit in the tree above.
828 enforce = dsl_enforce_ds_ss_limits(dd, prop, cr, proc);
829 if (enforce == ENFORCE_NEVER)
833 * e.g. if renaming a dataset with no snapshots, count adjustment
839 if (prop == ZFS_PROP_SNAPSHOT_LIMIT) {
841 * We don't enforce the limit for temporary snapshots. This is
842 * indicated by a NULL cred_t argument.
847 count_prop = DD_FIELD_SNAPSHOT_COUNT;
849 count_prop = DD_FIELD_FILESYSTEM_COUNT;
853 * If an ancestor has been provided, stop checking the limit once we
854 * hit that dir. We need this during rename so that we don't overcount
855 * the check once we recurse up to the common ancestor.
861 * If we hit an uninitialized node while recursing up the tree, we can
862 * stop since we know there is no limit here (or above). The counts are
863 * not valid on this node and we know we won't touch this node's counts.
865 if (!dsl_dir_is_zapified(dd))
867 err = zap_lookup(os, dd->dd_object,
868 count_prop, sizeof (count), 1, &count);
874 err = dsl_prop_get_dd(dd, zfs_prop_to_name(prop), 8, 1, &limit, NULL,
879 /* Is there a limit which we've hit? */
880 if (enforce == ENFORCE_ALWAYS && (count + delta) > limit)
881 return (SET_ERROR(EDQUOT));
883 if (dd->dd_parent != NULL)
884 err = dsl_fs_ss_limit_check(dd->dd_parent, delta, prop,
891 * Adjust the filesystem or snapshot count for the specified dsl_dir_t and all
892 * parents. When a new filesystem/snapshot is created, increment the count on
893 * all parents, and when a filesystem/snapshot is destroyed, decrement the
897 dsl_fs_ss_count_adjust(dsl_dir_t *dd, int64_t delta, const char *prop,
901 objset_t *os = dd->dd_pool->dp_meta_objset;
904 ASSERT(dsl_pool_config_held(dd->dd_pool));
905 ASSERT(dmu_tx_is_syncing(tx));
906 ASSERT(strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0 ||
907 strcmp(prop, DD_FIELD_SNAPSHOT_COUNT) == 0);
910 * We don't do accounting for hidden ($FREE, $MOS & $ORIGIN) objsets.
912 if (dd->dd_myname[0] == '$' && strcmp(prop,
913 DD_FIELD_FILESYSTEM_COUNT) == 0) {
918 * e.g. if renaming a dataset with no snapshots, count adjustment is 0
924 * If we hit an uninitialized node while recursing up the tree, we can
925 * stop since we know the counts are not valid on this node and we
926 * know we shouldn't touch this node's counts. An uninitialized count
927 * on the node indicates that either the feature has not yet been
928 * activated or there are no limits on this part of the tree.
930 if (!dsl_dir_is_zapified(dd) || (err = zap_lookup(os, dd->dd_object,
931 prop, sizeof (count), 1, &count)) == ENOENT)
936 /* Use a signed verify to make sure we're not neg. */
937 VERIFY3S(count, >=, 0);
939 VERIFY0(zap_update(os, dd->dd_object, prop, sizeof (count), 1, &count,
942 /* Roll up this additional count into our ancestors */
943 if (dd->dd_parent != NULL)
944 dsl_fs_ss_count_adjust(dd->dd_parent, delta, prop, tx);
948 dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
951 objset_t *mos = dp->dp_meta_objset;
953 dsl_dir_phys_t *ddphys;
956 ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
957 DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
959 VERIFY0(zap_add(mos, dsl_dir_phys(pds)->dd_child_dir_zapobj,
960 name, sizeof (uint64_t), 1, &ddobj, tx));
962 /* it's the root dir */
963 VERIFY0(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
964 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
966 VERIFY0(dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
967 dmu_buf_will_dirty(dbuf, tx);
968 ddphys = dbuf->db_data;
970 ddphys->dd_creation_time = gethrestime_sec();
972 ddphys->dd_parent_obj = pds->dd_object;
974 /* update the filesystem counts */
975 dsl_fs_ss_count_adjust(pds, 1, DD_FIELD_FILESYSTEM_COUNT, tx);
977 ddphys->dd_props_zapobj = zap_create(mos,
978 DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
979 ddphys->dd_child_dir_zapobj = zap_create(mos,
980 DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
981 if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
982 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
984 dmu_buf_rele(dbuf, FTAG);
990 dsl_dir_is_clone(dsl_dir_t *dd)
992 return (dsl_dir_phys(dd)->dd_origin_obj &&
993 (dd->dd_pool->dp_origin_snap == NULL ||
994 dsl_dir_phys(dd)->dd_origin_obj !=
995 dd->dd_pool->dp_origin_snap->ds_object));
999 dsl_dir_get_used(dsl_dir_t *dd)
1001 return (dsl_dir_phys(dd)->dd_used_bytes);
1005 dsl_dir_get_compressed(dsl_dir_t *dd)
1007 return (dsl_dir_phys(dd)->dd_compressed_bytes);
1011 dsl_dir_get_quota(dsl_dir_t *dd)
1013 return (dsl_dir_phys(dd)->dd_quota);
1017 dsl_dir_get_reservation(dsl_dir_t *dd)
1019 return (dsl_dir_phys(dd)->dd_reserved);
1023 dsl_dir_get_compressratio(dsl_dir_t *dd)
1025 /* a fixed point number, 100x the ratio */
1026 return (dsl_dir_phys(dd)->dd_compressed_bytes == 0 ? 100 :
1027 (dsl_dir_phys(dd)->dd_uncompressed_bytes * 100 /
1028 dsl_dir_phys(dd)->dd_compressed_bytes));
1032 dsl_dir_get_logicalused(dsl_dir_t *dd)
1034 return (dsl_dir_phys(dd)->dd_uncompressed_bytes);
1038 dsl_dir_get_usedsnap(dsl_dir_t *dd)
1040 return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]);
1044 dsl_dir_get_usedds(dsl_dir_t *dd)
1046 return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_HEAD]);
1050 dsl_dir_get_usedrefreserv(dsl_dir_t *dd)
1052 return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_REFRSRV]);
1056 dsl_dir_get_usedchild(dsl_dir_t *dd)
1058 return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD] +
1059 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD_RSRV]);
1063 dsl_dir_get_origin(dsl_dir_t *dd, char *buf)
1066 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
1067 dsl_dir_phys(dd)->dd_origin_obj, FTAG, &ds));
1069 dsl_dataset_name(ds, buf);
1071 dsl_dataset_rele(ds, FTAG);
1075 dsl_dir_get_filesystem_count(dsl_dir_t *dd, uint64_t *count)
1077 if (dsl_dir_is_zapified(dd)) {
1078 objset_t *os = dd->dd_pool->dp_meta_objset;
1079 return (zap_lookup(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
1080 sizeof (*count), 1, count));
1082 return (SET_ERROR(ENOENT));
1087 dsl_dir_get_snapshot_count(dsl_dir_t *dd, uint64_t *count)
1089 if (dsl_dir_is_zapified(dd)) {
1090 objset_t *os = dd->dd_pool->dp_meta_objset;
1091 return (zap_lookup(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
1092 sizeof (*count), 1, count));
1094 return (SET_ERROR(ENOENT));
1099 dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
1101 mutex_enter(&dd->dd_lock);
1102 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA,
1103 dsl_dir_get_quota(dd));
1104 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
1105 dsl_dir_get_reservation(dd));
1106 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED,
1107 dsl_dir_get_logicalused(dd));
1108 if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1109 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
1110 dsl_dir_get_usedsnap(dd));
1111 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
1112 dsl_dir_get_usedds(dd));
1113 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
1114 dsl_dir_get_usedrefreserv(dd));
1115 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
1116 dsl_dir_get_usedchild(dd));
1118 mutex_exit(&dd->dd_lock);
1121 if (dsl_dir_get_filesystem_count(dd, &count) == 0) {
1122 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_FILESYSTEM_COUNT,
1125 if (dsl_dir_get_snapshot_count(dd, &count) == 0) {
1126 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_SNAPSHOT_COUNT,
1130 if (dsl_dir_is_clone(dd)) {
1131 char buf[ZFS_MAX_DATASET_NAME_LEN];
1132 dsl_dir_get_origin(dd, buf);
1133 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
1139 dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
1141 dsl_pool_t *dp = dd->dd_pool;
1143 ASSERT(dsl_dir_phys(dd));
1145 if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) {
1146 /* up the hold count until we can be written out */
1147 dmu_buf_add_ref(dd->dd_dbuf, dd);
1152 parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
1154 uint64_t old_accounted = MAX(used, dsl_dir_phys(dd)->dd_reserved);
1155 uint64_t new_accounted =
1156 MAX(used + delta, dsl_dir_phys(dd)->dd_reserved);
1157 return (new_accounted - old_accounted);
1161 dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
1163 ASSERT(dmu_tx_is_syncing(tx));
1165 mutex_enter(&dd->dd_lock);
1166 ASSERT0(dd->dd_tempreserved[tx->tx_txg & TXG_MASK]);
1167 dprintf_dd(dd, "txg=%llu towrite=%lluK\n", (u_longlong_t)tx->tx_txg,
1168 (u_longlong_t)dd->dd_space_towrite[tx->tx_txg & TXG_MASK] / 1024);
1169 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] = 0;
1170 mutex_exit(&dd->dd_lock);
1172 /* release the hold from dsl_dir_dirty */
1173 dmu_buf_rele(dd->dd_dbuf, dd);
1177 dsl_dir_space_towrite(dsl_dir_t *dd)
1181 ASSERT(MUTEX_HELD(&dd->dd_lock));
1183 for (int i = 0; i < TXG_SIZE; i++) {
1184 space += dd->dd_space_towrite[i & TXG_MASK];
1185 ASSERT3U(dd->dd_space_towrite[i & TXG_MASK], >=, 0);
1191 * How much space would dd have available if ancestor had delta applied
1192 * to it? If ondiskonly is set, we're only interested in what's
1193 * on-disk, not estimated pending changes.
1196 dsl_dir_space_available(dsl_dir_t *dd,
1197 dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
1199 uint64_t parentspace, myspace, quota, used;
1202 * If there are no restrictions otherwise, assume we have
1203 * unlimited space available.
1206 parentspace = UINT64_MAX;
1208 if (dd->dd_parent != NULL) {
1209 parentspace = dsl_dir_space_available(dd->dd_parent,
1210 ancestor, delta, ondiskonly);
1213 mutex_enter(&dd->dd_lock);
1214 if (dsl_dir_phys(dd)->dd_quota != 0)
1215 quota = dsl_dir_phys(dd)->dd_quota;
1216 used = dsl_dir_phys(dd)->dd_used_bytes;
1218 used += dsl_dir_space_towrite(dd);
1220 if (dd->dd_parent == NULL) {
1221 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool,
1222 ZFS_SPACE_CHECK_NORMAL);
1223 quota = MIN(quota, poolsize);
1226 if (dsl_dir_phys(dd)->dd_reserved > used && parentspace != UINT64_MAX) {
1228 * We have some space reserved, in addition to what our
1231 parentspace += dsl_dir_phys(dd)->dd_reserved - used;
1234 if (dd == ancestor) {
1236 ASSERT(used >= -delta);
1238 if (parentspace != UINT64_MAX)
1239 parentspace -= delta;
1247 * the lesser of the space provided by our parent and
1248 * the space left in our quota
1250 myspace = MIN(parentspace, quota - used);
1253 mutex_exit(&dd->dd_lock);
1258 struct tempreserve {
1259 list_node_t tr_node;
1265 dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
1266 boolean_t ignorequota, list_t *tr_list,
1267 dmu_tx_t *tx, boolean_t first)
1271 struct tempreserve *tr;
1280 ASSERT3U(txg, !=, 0);
1281 ASSERT3S(asize, >, 0);
1283 mutex_enter(&dd->dd_lock);
1286 * Check against the dsl_dir's quota. We don't add in the delta
1287 * when checking for over-quota because they get one free hit.
1289 uint64_t est_inflight = dsl_dir_space_towrite(dd);
1290 for (int i = 0; i < TXG_SIZE; i++)
1291 est_inflight += dd->dd_tempreserved[i];
1292 uint64_t used_on_disk = dsl_dir_phys(dd)->dd_used_bytes;
1295 * On the first iteration, fetch the dataset's used-on-disk and
1296 * refreservation values. Also, if checkrefquota is set, test if
1297 * allocating this space would exceed the dataset's refquota.
1299 if (first && tx->tx_objset) {
1301 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
1303 error = dsl_dataset_check_quota(ds, !netfree,
1304 asize, est_inflight, &used_on_disk, &ref_rsrv);
1306 mutex_exit(&dd->dd_lock);
1307 DMU_TX_STAT_BUMP(dmu_tx_quota);
1313 * If this transaction will result in a net free of space,
1314 * we want to let it through.
1316 if (ignorequota || netfree || dsl_dir_phys(dd)->dd_quota == 0)
1319 quota = dsl_dir_phys(dd)->dd_quota;
1322 * Adjust the quota against the actual pool size at the root
1323 * minus any outstanding deferred frees.
1324 * To ensure that it's possible to remove files from a full
1325 * pool without inducing transient overcommits, we throttle
1326 * netfree transactions against a quota that is slightly larger,
1327 * but still within the pool's allocation slop. In cases where
1328 * we're very close to full, this will allow a steady trickle of
1329 * removes to get through.
1331 if (dd->dd_parent == NULL) {
1332 uint64_t avail = dsl_pool_unreserved_space(dd->dd_pool,
1334 ZFS_SPACE_CHECK_RESERVED : ZFS_SPACE_CHECK_NORMAL);
1336 if (avail < quota) {
1338 retval = SET_ERROR(ENOSPC);
1343 * If they are requesting more space, and our current estimate
1344 * is over quota, they get to try again unless the actual
1345 * on-disk is over quota and there are no pending changes
1346 * or deferred frees (which may free up space for us).
1348 if (used_on_disk + est_inflight >= quota) {
1349 if (est_inflight > 0 || used_on_disk < quota) {
1350 retval = SET_ERROR(ERESTART);
1352 ASSERT3U(used_on_disk, >=, quota);
1354 if (retval == ENOSPC && (used_on_disk - quota) <
1355 dsl_pool_deferred_space(dd->dd_pool)) {
1356 retval = SET_ERROR(ERESTART);
1360 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
1361 "quota=%lluK tr=%lluK err=%d\n",
1362 (u_longlong_t)used_on_disk>>10,
1363 (u_longlong_t)est_inflight>>10,
1364 (u_longlong_t)quota>>10, (u_longlong_t)asize>>10, retval);
1365 mutex_exit(&dd->dd_lock);
1366 DMU_TX_STAT_BUMP(dmu_tx_quota);
1370 /* We need to up our estimated delta before dropping dd_lock */
1371 dd->dd_tempreserved[txg & TXG_MASK] += asize;
1373 uint64_t parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
1375 mutex_exit(&dd->dd_lock);
1377 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1379 tr->tr_size = asize;
1380 list_insert_tail(tr_list, tr);
1382 /* see if it's OK with our parent */
1383 if (dd->dd_parent != NULL && parent_rsrv != 0) {
1385 * Recurse on our parent without recursion. This has been
1386 * observed to be potentially large stack usage even within
1387 * the test suite. Largest seen stack was 7632 bytes on linux.
1391 asize = parent_rsrv;
1392 ignorequota = (dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
1394 goto top_of_function;
1402 * Reserve space in this dsl_dir, to be used in this tx's txg.
1403 * After the space has been dirtied (and dsl_dir_willuse_space()
1404 * has been called), the reservation should be canceled, using
1405 * dsl_dir_tempreserve_clear().
1408 dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
1409 boolean_t netfree, void **tr_cookiep, dmu_tx_t *tx)
1419 tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
1420 list_create(tr_list, sizeof (struct tempreserve),
1421 offsetof(struct tempreserve, tr_node));
1422 ASSERT3S(asize, >, 0);
1424 err = arc_tempreserve_space(dd->dd_pool->dp_spa, lsize, tx->tx_txg);
1426 struct tempreserve *tr;
1428 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1429 tr->tr_size = lsize;
1430 list_insert_tail(tr_list, tr);
1432 if (err == EAGAIN) {
1434 * If arc_memory_throttle() detected that pageout
1435 * is running and we are low on memory, we delay new
1436 * non-pageout transactions to give pageout an
1439 * It is unfortunate to be delaying while the caller's
1442 txg_delay(dd->dd_pool, tx->tx_txg,
1443 MSEC2NSEC(10), MSEC2NSEC(10));
1444 err = SET_ERROR(ERESTART);
1449 err = dsl_dir_tempreserve_impl(dd, asize, netfree,
1450 B_FALSE, tr_list, tx, B_TRUE);
1454 dsl_dir_tempreserve_clear(tr_list, tx);
1456 *tr_cookiep = tr_list;
1462 * Clear a temporary reservation that we previously made with
1463 * dsl_dir_tempreserve_space().
1466 dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
1468 int txgidx = tx->tx_txg & TXG_MASK;
1469 list_t *tr_list = tr_cookie;
1470 struct tempreserve *tr;
1472 ASSERT3U(tx->tx_txg, !=, 0);
1474 if (tr_cookie == NULL)
1477 while ((tr = list_head(tr_list)) != NULL) {
1479 mutex_enter(&tr->tr_ds->dd_lock);
1480 ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
1482 tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
1483 mutex_exit(&tr->tr_ds->dd_lock);
1485 arc_tempreserve_clear(tr->tr_size);
1487 list_remove(tr_list, tr);
1488 kmem_free(tr, sizeof (struct tempreserve));
1491 kmem_free(tr_list, sizeof (list_t));
1495 * This should be called from open context when we think we're going to write
1496 * or free space, for example when dirtying data. Be conservative; it's okay
1497 * to write less space or free more, but we don't want to write more or free
1498 * less than the amount specified.
1500 * NOTE: The behavior of this function is identical to the Illumos / FreeBSD
1501 * version however it has been adjusted to use an iterative rather than
1502 * recursive algorithm to minimize stack usage.
1505 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
1507 int64_t parent_space;
1511 mutex_enter(&dd->dd_lock);
1513 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
1515 est_used = dsl_dir_space_towrite(dd) +
1516 dsl_dir_phys(dd)->dd_used_bytes;
1517 parent_space = parent_delta(dd, est_used, space);
1518 mutex_exit(&dd->dd_lock);
1520 /* Make sure that we clean up dd_space_to* */
1521 dsl_dir_dirty(dd, tx);
1524 space = parent_space;
1525 } while (space && dd);
1528 /* call from syncing context when we actually write/free space for this dd */
1530 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
1531 int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
1533 int64_t accounted_delta;
1535 ASSERT(dmu_tx_is_syncing(tx));
1536 ASSERT(type < DD_USED_NUM);
1538 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1541 * dsl_dataset_set_refreservation_sync_impl() calls this with
1542 * dd_lock held, so that it can atomically update
1543 * ds->ds_reserved and the dsl_dir accounting, so that
1544 * dsl_dataset_check_quota() can see dataset and dir accounting
1547 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
1549 mutex_enter(&dd->dd_lock);
1550 dsl_dir_phys_t *ddp = dsl_dir_phys(dd);
1551 accounted_delta = parent_delta(dd, ddp->dd_used_bytes, used);
1552 ASSERT(used >= 0 || ddp->dd_used_bytes >= -used);
1553 ASSERT(compressed >= 0 || ddp->dd_compressed_bytes >= -compressed);
1554 ASSERT(uncompressed >= 0 ||
1555 ddp->dd_uncompressed_bytes >= -uncompressed);
1556 ddp->dd_used_bytes += used;
1557 ddp->dd_uncompressed_bytes += uncompressed;
1558 ddp->dd_compressed_bytes += compressed;
1560 if (ddp->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1561 ASSERT(used >= 0 || ddp->dd_used_breakdown[type] >= -used);
1562 ddp->dd_used_breakdown[type] += used;
1567 for (t = 0; t < DD_USED_NUM; t++)
1568 u += ddp->dd_used_breakdown[t];
1569 ASSERT3U(u, ==, ddp->dd_used_bytes);
1574 mutex_exit(&dd->dd_lock);
1576 if (dd->dd_parent != NULL) {
1577 dsl_dir_diduse_transfer_space(dd->dd_parent,
1578 accounted_delta, compressed, uncompressed,
1579 used, DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
1584 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
1585 dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
1587 ASSERT(dmu_tx_is_syncing(tx));
1588 ASSERT(oldtype < DD_USED_NUM);
1589 ASSERT(newtype < DD_USED_NUM);
1591 dsl_dir_phys_t *ddp = dsl_dir_phys(dd);
1593 !(ddp->dd_flags & DD_FLAG_USED_BREAKDOWN))
1596 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1597 mutex_enter(&dd->dd_lock);
1599 ddp->dd_used_breakdown[oldtype] >= delta :
1600 ddp->dd_used_breakdown[newtype] >= -delta);
1601 ASSERT(ddp->dd_used_bytes >= ABS(delta));
1602 ddp->dd_used_breakdown[oldtype] -= delta;
1603 ddp->dd_used_breakdown[newtype] += delta;
1604 mutex_exit(&dd->dd_lock);
1608 dsl_dir_diduse_transfer_space(dsl_dir_t *dd, int64_t used,
1609 int64_t compressed, int64_t uncompressed, int64_t tonew,
1610 dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
1612 int64_t accounted_delta;
1614 ASSERT(dmu_tx_is_syncing(tx));
1615 ASSERT(oldtype < DD_USED_NUM);
1616 ASSERT(newtype < DD_USED_NUM);
1618 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1620 mutex_enter(&dd->dd_lock);
1621 dsl_dir_phys_t *ddp = dsl_dir_phys(dd);
1622 accounted_delta = parent_delta(dd, ddp->dd_used_bytes, used);
1623 ASSERT(used >= 0 || ddp->dd_used_bytes >= -used);
1624 ASSERT(compressed >= 0 || ddp->dd_compressed_bytes >= -compressed);
1625 ASSERT(uncompressed >= 0 ||
1626 ddp->dd_uncompressed_bytes >= -uncompressed);
1627 ddp->dd_used_bytes += used;
1628 ddp->dd_uncompressed_bytes += uncompressed;
1629 ddp->dd_compressed_bytes += compressed;
1631 if (ddp->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1632 ASSERT(tonew - used <= 0 ||
1633 ddp->dd_used_breakdown[oldtype] >= tonew - used);
1634 ASSERT(tonew >= 0 ||
1635 ddp->dd_used_breakdown[newtype] >= -tonew);
1636 ddp->dd_used_breakdown[oldtype] -= tonew - used;
1637 ddp->dd_used_breakdown[newtype] += tonew;
1642 for (t = 0; t < DD_USED_NUM; t++)
1643 u += ddp->dd_used_breakdown[t];
1644 ASSERT3U(u, ==, ddp->dd_used_bytes);
1648 mutex_exit(&dd->dd_lock);
1650 if (dd->dd_parent != NULL) {
1651 dsl_dir_diduse_transfer_space(dd->dd_parent,
1652 accounted_delta, compressed, uncompressed,
1653 used, DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
1657 typedef struct dsl_dir_set_qr_arg {
1658 const char *ddsqra_name;
1659 zprop_source_t ddsqra_source;
1660 uint64_t ddsqra_value;
1661 } dsl_dir_set_qr_arg_t;
1664 dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx)
1666 dsl_dir_set_qr_arg_t *ddsqra = arg;
1667 dsl_pool_t *dp = dmu_tx_pool(tx);
1670 uint64_t towrite, newval;
1672 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1676 error = dsl_prop_predict(ds->ds_dir, "quota",
1677 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1679 dsl_dataset_rele(ds, FTAG);
1684 dsl_dataset_rele(ds, FTAG);
1688 mutex_enter(&ds->ds_dir->dd_lock);
1690 * If we are doing the preliminary check in open context, and
1691 * there are pending changes, then don't fail it, since the
1692 * pending changes could under-estimate the amount of space to be
1695 towrite = dsl_dir_space_towrite(ds->ds_dir);
1696 if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
1697 (newval < dsl_dir_phys(ds->ds_dir)->dd_reserved ||
1698 newval < dsl_dir_phys(ds->ds_dir)->dd_used_bytes + towrite)) {
1699 error = SET_ERROR(ENOSPC);
1701 mutex_exit(&ds->ds_dir->dd_lock);
1702 dsl_dataset_rele(ds, FTAG);
1707 dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx)
1709 dsl_dir_set_qr_arg_t *ddsqra = arg;
1710 dsl_pool_t *dp = dmu_tx_pool(tx);
1714 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1716 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1717 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA),
1718 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1719 &ddsqra->ddsqra_value, tx);
1721 VERIFY0(dsl_prop_get_int_ds(ds,
1722 zfs_prop_to_name(ZFS_PROP_QUOTA), &newval));
1724 newval = ddsqra->ddsqra_value;
1725 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1726 zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval);
1729 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1730 mutex_enter(&ds->ds_dir->dd_lock);
1731 dsl_dir_phys(ds->ds_dir)->dd_quota = newval;
1732 mutex_exit(&ds->ds_dir->dd_lock);
1733 dsl_dataset_rele(ds, FTAG);
1737 dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
1739 dsl_dir_set_qr_arg_t ddsqra;
1741 ddsqra.ddsqra_name = ddname;
1742 ddsqra.ddsqra_source = source;
1743 ddsqra.ddsqra_value = quota;
1745 return (dsl_sync_task(ddname, dsl_dir_set_quota_check,
1746 dsl_dir_set_quota_sync, &ddsqra, 0,
1747 ZFS_SPACE_CHECK_EXTRA_RESERVED));
1751 dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx)
1753 dsl_dir_set_qr_arg_t *ddsqra = arg;
1754 dsl_pool_t *dp = dmu_tx_pool(tx);
1757 uint64_t newval, used, avail;
1760 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1766 * If we are doing the preliminary check in open context, the
1767 * space estimates may be inaccurate.
1769 if (!dmu_tx_is_syncing(tx)) {
1770 dsl_dataset_rele(ds, FTAG);
1774 error = dsl_prop_predict(ds->ds_dir,
1775 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1776 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1778 dsl_dataset_rele(ds, FTAG);
1782 mutex_enter(&dd->dd_lock);
1783 used = dsl_dir_phys(dd)->dd_used_bytes;
1784 mutex_exit(&dd->dd_lock);
1786 if (dd->dd_parent) {
1787 avail = dsl_dir_space_available(dd->dd_parent,
1790 avail = dsl_pool_adjustedsize(dd->dd_pool,
1791 ZFS_SPACE_CHECK_NORMAL) - used;
1794 if (MAX(used, newval) > MAX(used, dsl_dir_phys(dd)->dd_reserved)) {
1795 uint64_t delta = MAX(used, newval) -
1796 MAX(used, dsl_dir_phys(dd)->dd_reserved);
1798 if (delta > avail ||
1799 (dsl_dir_phys(dd)->dd_quota > 0 &&
1800 newval > dsl_dir_phys(dd)->dd_quota))
1801 error = SET_ERROR(ENOSPC);
1804 dsl_dataset_rele(ds, FTAG);
1809 dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
1814 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1816 mutex_enter(&dd->dd_lock);
1817 used = dsl_dir_phys(dd)->dd_used_bytes;
1818 delta = MAX(used, value) - MAX(used, dsl_dir_phys(dd)->dd_reserved);
1819 dsl_dir_phys(dd)->dd_reserved = value;
1821 if (dd->dd_parent != NULL) {
1822 /* Roll up this additional usage into our ancestors */
1823 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1826 mutex_exit(&dd->dd_lock);
1830 dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx)
1832 dsl_dir_set_qr_arg_t *ddsqra = arg;
1833 dsl_pool_t *dp = dmu_tx_pool(tx);
1837 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1839 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1840 dsl_prop_set_sync_impl(ds,
1841 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1842 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1843 &ddsqra->ddsqra_value, tx);
1845 VERIFY0(dsl_prop_get_int_ds(ds,
1846 zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval));
1848 newval = ddsqra->ddsqra_value;
1849 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1850 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1851 (longlong_t)newval);
1854 dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx);
1855 dsl_dataset_rele(ds, FTAG);
1859 dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1860 uint64_t reservation)
1862 dsl_dir_set_qr_arg_t ddsqra;
1864 ddsqra.ddsqra_name = ddname;
1865 ddsqra.ddsqra_source = source;
1866 ddsqra.ddsqra_value = reservation;
1868 return (dsl_sync_task(ddname, dsl_dir_set_reservation_check,
1869 dsl_dir_set_reservation_sync, &ddsqra, 0,
1870 ZFS_SPACE_CHECK_EXTRA_RESERVED));
1874 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1876 for (; ds1; ds1 = ds1->dd_parent) {
1878 for (dd = ds2; dd; dd = dd->dd_parent) {
1887 * If delta is applied to dd, how much of that delta would be applied to
1888 * ancestor? Syncing context only.
1891 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1896 mutex_enter(&dd->dd_lock);
1897 delta = parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, delta);
1898 mutex_exit(&dd->dd_lock);
1899 return (would_change(dd->dd_parent, delta, ancestor));
1902 typedef struct dsl_dir_rename_arg {
1903 const char *ddra_oldname;
1904 const char *ddra_newname;
1907 } dsl_dir_rename_arg_t;
1909 typedef struct dsl_valid_rename_arg {
1912 } dsl_valid_rename_arg_t;
1915 dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1918 dsl_valid_rename_arg_t *dvra = arg;
1919 char namebuf[ZFS_MAX_DATASET_NAME_LEN];
1921 dsl_dataset_name(ds, namebuf);
1923 ASSERT3U(strnlen(namebuf, ZFS_MAX_DATASET_NAME_LEN),
1924 <, ZFS_MAX_DATASET_NAME_LEN);
1925 int namelen = strlen(namebuf) + dvra->char_delta;
1926 int depth = get_dataset_depth(namebuf) + dvra->nest_delta;
1928 if (namelen >= ZFS_MAX_DATASET_NAME_LEN)
1929 return (SET_ERROR(ENAMETOOLONG));
1930 if (dvra->nest_delta > 0 && depth >= zfs_max_dataset_nesting)
1931 return (SET_ERROR(ENAMETOOLONG));
1936 dsl_dir_rename_check(void *arg, dmu_tx_t *tx)
1938 dsl_dir_rename_arg_t *ddra = arg;
1939 dsl_pool_t *dp = dmu_tx_pool(tx);
1940 dsl_dir_t *dd, *newparent;
1941 dsl_valid_rename_arg_t dvra;
1942 dsl_dataset_t *parentds;
1944 const char *mynewname;
1947 /* target dir should exist */
1948 error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL);
1952 /* new parent should exist */
1953 error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG,
1954 &newparent, &mynewname);
1956 dsl_dir_rele(dd, FTAG);
1960 /* can't rename to different pool */
1961 if (dd->dd_pool != newparent->dd_pool) {
1962 dsl_dir_rele(newparent, FTAG);
1963 dsl_dir_rele(dd, FTAG);
1964 return (SET_ERROR(EXDEV));
1967 /* new name should not already exist */
1968 if (mynewname == NULL) {
1969 dsl_dir_rele(newparent, FTAG);
1970 dsl_dir_rele(dd, FTAG);
1971 return (SET_ERROR(EEXIST));
1974 /* can't rename below anything but filesystems (eg. no ZVOLs) */
1975 error = dsl_dataset_hold_obj(newparent->dd_pool,
1976 dsl_dir_phys(newparent)->dd_head_dataset_obj, FTAG, &parentds);
1978 dsl_dir_rele(newparent, FTAG);
1979 dsl_dir_rele(dd, FTAG);
1982 error = dmu_objset_from_ds(parentds, &parentos);
1984 dsl_dataset_rele(parentds, FTAG);
1985 dsl_dir_rele(newparent, FTAG);
1986 dsl_dir_rele(dd, FTAG);
1989 if (dmu_objset_type(parentos) != DMU_OST_ZFS) {
1990 dsl_dataset_rele(parentds, FTAG);
1991 dsl_dir_rele(newparent, FTAG);
1992 dsl_dir_rele(dd, FTAG);
1993 return (SET_ERROR(ZFS_ERR_WRONG_PARENT));
1995 dsl_dataset_rele(parentds, FTAG);
1997 ASSERT3U(strnlen(ddra->ddra_newname, ZFS_MAX_DATASET_NAME_LEN),
1998 <, ZFS_MAX_DATASET_NAME_LEN);
1999 ASSERT3U(strnlen(ddra->ddra_oldname, ZFS_MAX_DATASET_NAME_LEN),
2000 <, ZFS_MAX_DATASET_NAME_LEN);
2001 dvra.char_delta = strlen(ddra->ddra_newname)
2002 - strlen(ddra->ddra_oldname);
2003 dvra.nest_delta = get_dataset_depth(ddra->ddra_newname)
2004 - get_dataset_depth(ddra->ddra_oldname);
2006 /* if the name length is growing, validate child name lengths */
2007 if (dvra.char_delta > 0 || dvra.nest_delta > 0) {
2008 error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename,
2009 &dvra, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2011 dsl_dir_rele(newparent, FTAG);
2012 dsl_dir_rele(dd, FTAG);
2017 if (dmu_tx_is_syncing(tx)) {
2018 if (spa_feature_is_active(dp->dp_spa,
2019 SPA_FEATURE_FS_SS_LIMIT)) {
2021 * Although this is the check function and we don't
2022 * normally make on-disk changes in check functions,
2023 * we need to do that here.
2025 * Ensure this portion of the tree's counts have been
2026 * initialized in case the new parent has limits set.
2028 dsl_dir_init_fs_ss_count(dd, tx);
2032 if (newparent != dd->dd_parent) {
2033 /* is there enough space? */
2035 MAX(dsl_dir_phys(dd)->dd_used_bytes,
2036 dsl_dir_phys(dd)->dd_reserved);
2037 objset_t *os = dd->dd_pool->dp_meta_objset;
2038 uint64_t fs_cnt = 0;
2039 uint64_t ss_cnt = 0;
2041 if (dsl_dir_is_zapified(dd)) {
2044 err = zap_lookup(os, dd->dd_object,
2045 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
2047 if (err != ENOENT && err != 0) {
2048 dsl_dir_rele(newparent, FTAG);
2049 dsl_dir_rele(dd, FTAG);
2054 * have to add 1 for the filesystem itself that we're
2059 err = zap_lookup(os, dd->dd_object,
2060 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
2062 if (err != ENOENT && err != 0) {
2063 dsl_dir_rele(newparent, FTAG);
2064 dsl_dir_rele(dd, FTAG);
2069 /* check for encryption errors */
2070 error = dsl_dir_rename_crypt_check(dd, newparent);
2072 dsl_dir_rele(newparent, FTAG);
2073 dsl_dir_rele(dd, FTAG);
2074 return (SET_ERROR(EACCES));
2077 /* no rename into our descendant */
2078 if (closest_common_ancestor(dd, newparent) == dd) {
2079 dsl_dir_rele(newparent, FTAG);
2080 dsl_dir_rele(dd, FTAG);
2081 return (SET_ERROR(EINVAL));
2084 error = dsl_dir_transfer_possible(dd->dd_parent,
2085 newparent, fs_cnt, ss_cnt, myspace,
2086 ddra->ddra_cred, ddra->ddra_proc);
2088 dsl_dir_rele(newparent, FTAG);
2089 dsl_dir_rele(dd, FTAG);
2094 dsl_dir_rele(newparent, FTAG);
2095 dsl_dir_rele(dd, FTAG);
2100 dsl_dir_rename_sync(void *arg, dmu_tx_t *tx)
2102 dsl_dir_rename_arg_t *ddra = arg;
2103 dsl_pool_t *dp = dmu_tx_pool(tx);
2104 dsl_dir_t *dd, *newparent;
2105 const char *mynewname;
2106 objset_t *mos = dp->dp_meta_objset;
2108 VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL));
2109 VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
2112 /* Log this before we change the name. */
2113 spa_history_log_internal_dd(dd, "rename", tx,
2114 "-> %s", ddra->ddra_newname);
2116 if (newparent != dd->dd_parent) {
2117 objset_t *os = dd->dd_pool->dp_meta_objset;
2118 uint64_t fs_cnt = 0;
2119 uint64_t ss_cnt = 0;
2122 * We already made sure the dd counts were initialized in the
2125 if (spa_feature_is_active(dp->dp_spa,
2126 SPA_FEATURE_FS_SS_LIMIT)) {
2127 VERIFY0(zap_lookup(os, dd->dd_object,
2128 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
2130 /* add 1 for the filesystem itself that we're moving */
2133 VERIFY0(zap_lookup(os, dd->dd_object,
2134 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
2138 dsl_fs_ss_count_adjust(dd->dd_parent, -fs_cnt,
2139 DD_FIELD_FILESYSTEM_COUNT, tx);
2140 dsl_fs_ss_count_adjust(newparent, fs_cnt,
2141 DD_FIELD_FILESYSTEM_COUNT, tx);
2143 dsl_fs_ss_count_adjust(dd->dd_parent, -ss_cnt,
2144 DD_FIELD_SNAPSHOT_COUNT, tx);
2145 dsl_fs_ss_count_adjust(newparent, ss_cnt,
2146 DD_FIELD_SNAPSHOT_COUNT, tx);
2148 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
2149 -dsl_dir_phys(dd)->dd_used_bytes,
2150 -dsl_dir_phys(dd)->dd_compressed_bytes,
2151 -dsl_dir_phys(dd)->dd_uncompressed_bytes, tx);
2152 dsl_dir_diduse_space(newparent, DD_USED_CHILD,
2153 dsl_dir_phys(dd)->dd_used_bytes,
2154 dsl_dir_phys(dd)->dd_compressed_bytes,
2155 dsl_dir_phys(dd)->dd_uncompressed_bytes, tx);
2157 if (dsl_dir_phys(dd)->dd_reserved >
2158 dsl_dir_phys(dd)->dd_used_bytes) {
2159 uint64_t unused_rsrv = dsl_dir_phys(dd)->dd_reserved -
2160 dsl_dir_phys(dd)->dd_used_bytes;
2162 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
2163 -unused_rsrv, 0, 0, tx);
2164 dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV,
2165 unused_rsrv, 0, 0, tx);
2169 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2171 /* remove from old parent zapobj */
2172 VERIFY0(zap_remove(mos,
2173 dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj,
2174 dd->dd_myname, tx));
2176 (void) strlcpy(dd->dd_myname, mynewname,
2177 sizeof (dd->dd_myname));
2178 dsl_dir_rele(dd->dd_parent, dd);
2179 dsl_dir_phys(dd)->dd_parent_obj = newparent->dd_object;
2180 VERIFY0(dsl_dir_hold_obj(dp,
2181 newparent->dd_object, NULL, dd, &dd->dd_parent));
2183 /* add to new parent zapobj */
2184 VERIFY0(zap_add(mos, dsl_dir_phys(newparent)->dd_child_dir_zapobj,
2185 dd->dd_myname, 8, 1, &dd->dd_object, tx));
2187 /* TODO: A rename callback to avoid these layering violations. */
2188 zfsvfs_update_fromname(ddra->ddra_oldname, ddra->ddra_newname);
2189 zvol_rename_minors(dp->dp_spa, ddra->ddra_oldname,
2190 ddra->ddra_newname, B_TRUE);
2192 dsl_prop_notify_all(dd);
2194 dsl_dir_rele(newparent, FTAG);
2195 dsl_dir_rele(dd, FTAG);
2199 dsl_dir_rename(const char *oldname, const char *newname)
2201 dsl_dir_rename_arg_t ddra;
2203 ddra.ddra_oldname = oldname;
2204 ddra.ddra_newname = newname;
2205 ddra.ddra_cred = CRED();
2206 ddra.ddra_proc = curproc;
2208 return (dsl_sync_task(oldname,
2209 dsl_dir_rename_check, dsl_dir_rename_sync, &ddra,
2210 3, ZFS_SPACE_CHECK_RESERVED));
2214 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd,
2215 uint64_t fs_cnt, uint64_t ss_cnt, uint64_t space,
2216 cred_t *cr, proc_t *proc)
2218 dsl_dir_t *ancestor;
2223 ancestor = closest_common_ancestor(sdd, tdd);
2224 adelta = would_change(sdd, -space, ancestor);
2225 avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
2227 return (SET_ERROR(ENOSPC));
2229 err = dsl_fs_ss_limit_check(tdd, fs_cnt, ZFS_PROP_FILESYSTEM_LIMIT,
2230 ancestor, cr, proc);
2233 err = dsl_fs_ss_limit_check(tdd, ss_cnt, ZFS_PROP_SNAPSHOT_LIMIT,
2234 ancestor, cr, proc);
2242 dsl_dir_snap_cmtime(dsl_dir_t *dd)
2246 mutex_enter(&dd->dd_lock);
2247 t = dd->dd_snap_cmtime;
2248 mutex_exit(&dd->dd_lock);
2254 dsl_dir_snap_cmtime_update(dsl_dir_t *dd, dmu_tx_t *tx)
2256 dsl_pool_t *dp = dmu_tx_pool(tx);
2260 mutex_enter(&dd->dd_lock);
2261 dd->dd_snap_cmtime = t;
2262 if (spa_feature_is_enabled(dp->dp_spa,
2263 SPA_FEATURE_EXTENSIBLE_DATASET)) {
2264 objset_t *mos = dd->dd_pool->dp_meta_objset;
2265 uint64_t ddobj = dd->dd_object;
2266 dsl_dir_zapify(dd, tx);
2267 VERIFY0(zap_update(mos, ddobj,
2268 DD_FIELD_SNAPSHOTS_CHANGED,
2270 sizeof (inode_timespec_t) / sizeof (uint64_t),
2273 mutex_exit(&dd->dd_lock);
2277 dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx)
2279 objset_t *mos = dd->dd_pool->dp_meta_objset;
2280 dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx);
2284 dsl_dir_is_zapified(dsl_dir_t *dd)
2286 dmu_object_info_t doi;
2288 dmu_object_info_from_db(dd->dd_dbuf, &doi);
2289 return (doi.doi_type == DMU_OTN_ZAP_METADATA);
2293 dsl_dir_livelist_open(dsl_dir_t *dd, uint64_t obj)
2295 objset_t *mos = dd->dd_pool->dp_meta_objset;
2296 ASSERT(spa_feature_is_active(dd->dd_pool->dp_spa,
2297 SPA_FEATURE_LIVELIST));
2298 dsl_deadlist_open(&dd->dd_livelist, mos, obj);
2299 bplist_create(&dd->dd_pending_allocs);
2300 bplist_create(&dd->dd_pending_frees);
2304 dsl_dir_livelist_close(dsl_dir_t *dd)
2306 dsl_deadlist_close(&dd->dd_livelist);
2307 bplist_destroy(&dd->dd_pending_allocs);
2308 bplist_destroy(&dd->dd_pending_frees);
2312 dsl_dir_remove_livelist(dsl_dir_t *dd, dmu_tx_t *tx, boolean_t total)
2315 dsl_pool_t *dp = dmu_tx_pool(tx);
2316 spa_t *spa = dp->dp_spa;
2317 livelist_condense_entry_t to_condense = spa->spa_to_condense;
2319 if (!dsl_deadlist_is_open(&dd->dd_livelist))
2323 * If the livelist being removed is set to be condensed, stop the
2324 * condense zthr and indicate the cancellation in the spa_to_condense
2325 * struct in case the condense no-wait synctask has already started
2327 zthr_t *ll_condense_thread = spa->spa_livelist_condense_zthr;
2328 if (ll_condense_thread != NULL &&
2329 (to_condense.ds != NULL) && (to_condense.ds->ds_dir == dd)) {
2331 * We use zthr_wait_cycle_done instead of zthr_cancel
2332 * because we don't want to destroy the zthr, just have
2333 * it skip its current task.
2335 spa->spa_to_condense.cancelled = B_TRUE;
2336 zthr_wait_cycle_done(ll_condense_thread);
2338 * If we've returned from zthr_wait_cycle_done without
2339 * clearing the to_condense data structure it's either
2340 * because the no-wait synctask has started (which is
2341 * indicated by 'syncing' field of to_condense) and we
2342 * can expect it to clear to_condense on its own.
2343 * Otherwise, we returned before the zthr ran. The
2344 * checkfunc will now fail as cancelled == B_TRUE so we
2345 * can safely NULL out ds, allowing a different dir's
2346 * livelist to be condensed.
2348 * We can be sure that the to_condense struct will not
2349 * be repopulated at this stage because both this
2350 * function and dsl_livelist_try_condense execute in
2353 if ((spa->spa_to_condense.ds != NULL) &&
2354 !spa->spa_to_condense.syncing) {
2355 dmu_buf_rele(spa->spa_to_condense.ds->ds_dbuf,
2357 spa->spa_to_condense.ds = NULL;
2361 dsl_dir_livelist_close(dd);
2362 VERIFY0(zap_lookup(dp->dp_meta_objset, dd->dd_object,
2363 DD_FIELD_LIVELIST, sizeof (uint64_t), 1, &obj));
2364 VERIFY0(zap_remove(dp->dp_meta_objset, dd->dd_object,
2365 DD_FIELD_LIVELIST, tx));
2367 dsl_deadlist_free(dp->dp_meta_objset, obj, tx);
2368 spa_feature_decr(spa, SPA_FEATURE_LIVELIST, tx);
2373 dsl_dir_activity_in_progress(dsl_dir_t *dd, dsl_dataset_t *ds,
2374 zfs_wait_activity_t activity, boolean_t *in_progress)
2378 ASSERT(MUTEX_HELD(&dd->dd_activity_lock));
2381 case ZFS_WAIT_DELETEQ: {
2384 error = dmu_objset_from_ds(ds, &os);
2388 mutex_enter(&os->os_user_ptr_lock);
2389 void *user = dmu_objset_get_user(os);
2390 mutex_exit(&os->os_user_ptr_lock);
2391 if (dmu_objset_type(os) != DMU_OST_ZFS ||
2392 user == NULL || zfs_get_vfs_flag_unmounted(os)) {
2393 *in_progress = B_FALSE;
2397 uint64_t readonly = B_FALSE;
2398 error = zfs_get_temporary_prop(ds, ZFS_PROP_READONLY, &readonly,
2404 if (readonly || !spa_writeable(dd->dd_pool->dp_spa)) {
2405 *in_progress = B_FALSE;
2409 uint64_t count, unlinked_obj;
2410 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
2413 dsl_dataset_rele(ds, FTAG);
2416 error = zap_count(os, unlinked_obj, &count);
2419 *in_progress = (count != 0);
2423 * The delete queue is ZPL specific, and libzpool doesn't have
2424 * it. It doesn't make sense to wait for it.
2427 *in_progress = B_FALSE;
2432 panic("unrecognized value for activity %d", activity);
2439 dsl_dir_wait(dsl_dir_t *dd, dsl_dataset_t *ds, zfs_wait_activity_t activity,
2443 boolean_t in_progress;
2444 dsl_pool_t *dp = dd->dd_pool;
2446 dsl_pool_config_enter(dp, FTAG);
2447 error = dsl_dir_activity_in_progress(dd, ds, activity,
2449 dsl_pool_config_exit(dp, FTAG);
2450 if (error != 0 || !in_progress)
2455 if (cv_wait_sig(&dd->dd_activity_cv, &dd->dd_activity_lock) ==
2456 0 || dd->dd_activity_cancelled) {
2457 error = SET_ERROR(EINTR);
2465 dsl_dir_cancel_waiters(dsl_dir_t *dd)
2467 mutex_enter(&dd->dd_activity_lock);
2468 dd->dd_activity_cancelled = B_TRUE;
2469 cv_broadcast(&dd->dd_activity_cv);
2470 while (dd->dd_activity_waiters > 0)
2471 cv_wait(&dd->dd_activity_cv, &dd->dd_activity_lock);
2472 mutex_exit(&dd->dd_activity_lock);
2475 #if defined(_KERNEL)
2476 EXPORT_SYMBOL(dsl_dir_set_quota);
2477 EXPORT_SYMBOL(dsl_dir_set_reservation);