4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011 Pawel Jakub Dawidek <pawel@dawidek.net>.
24 * All rights reserved.
25 * Copyright (c) 2012, 2016 by Delphix. All rights reserved.
26 * Copyright (c) 2014 Joyent, Inc. All rights reserved.
27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
32 #include <sys/dmu_objset.h>
33 #include <sys/dmu_tx.h>
34 #include <sys/dsl_dataset.h>
35 #include <sys/dsl_dir.h>
36 #include <sys/dsl_prop.h>
37 #include <sys/dsl_synctask.h>
38 #include <sys/dsl_deleg.h>
39 #include <sys/dmu_impl.h>
41 #include <sys/metaslab.h>
45 #include <sys/sunddi.h>
48 #include <sys/zfs_vfsops.h>
50 #include <sys/zfeature.h>
51 #include <sys/policy.h>
52 #include <sys/zfs_znode.h>
53 #include "zfs_namecheck.h"
57 * Filesystem and Snapshot Limits
58 * ------------------------------
60 * These limits are used to restrict the number of filesystems and/or snapshots
61 * that can be created at a given level in the tree or below. A typical
62 * use-case is with a delegated dataset where the administrator wants to ensure
63 * that a user within the zone is not creating too many additional filesystems
64 * or snapshots, even though they're not exceeding their space quota.
66 * The filesystem and snapshot counts are stored as extensible properties. This
67 * capability is controlled by a feature flag and must be enabled to be used.
68 * Once enabled, the feature is not active until the first limit is set. At
69 * that point, future operations to create/destroy filesystems or snapshots
70 * will validate and update the counts.
72 * Because the count properties will not exist before the feature is active,
73 * the counts are updated when a limit is first set on an uninitialized
74 * dsl_dir node in the tree (The filesystem/snapshot count on a node includes
75 * all of the nested filesystems/snapshots. Thus, a new leaf node has a
76 * filesystem count of 0 and a snapshot count of 0. Non-existent filesystem and
77 * snapshot count properties on a node indicate uninitialized counts on that
78 * node.) When first setting a limit on an uninitialized node, the code starts
79 * at the filesystem with the new limit and descends into all sub-filesystems
80 * to add the count properties.
82 * In practice this is lightweight since a limit is typically set when the
83 * filesystem is created and thus has no children. Once valid, changing the
84 * limit value won't require a re-traversal since the counts are already valid.
85 * When recursively fixing the counts, if a node with a limit is encountered
86 * during the descent, the counts are known to be valid and there is no need to
87 * descend into that filesystem's children. The counts on filesystems above the
88 * one with the new limit will still be uninitialized, unless a limit is
89 * eventually set on one of those filesystems. The counts are always recursively
90 * updated when a limit is set on a dataset, unless there is already a limit.
91 * When a new limit value is set on a filesystem with an existing limit, it is
92 * possible for the new limit to be less than the current count at that level
93 * since a user who can change the limit is also allowed to exceed the limit.
95 * Once the feature is active, then whenever a filesystem or snapshot is
96 * created, the code recurses up the tree, validating the new count against the
97 * limit at each initialized level. In practice, most levels will not have a
98 * limit set. If there is a limit at any initialized level up the tree, the
99 * check must pass or the creation will fail. Likewise, when a filesystem or
100 * snapshot is destroyed, the counts are recursively adjusted all the way up
101 * the initizized nodes in the tree. Renaming a filesystem into different point
102 * in the tree will first validate, then update the counts on each branch up to
103 * the common ancestor. A receive will also validate the counts and then update
106 * An exception to the above behavior is that the limit is not enforced if the
107 * user has permission to modify the limit. This is primarily so that
108 * recursive snapshots in the global zone always work. We want to prevent a
109 * denial-of-service in which a lower level delegated dataset could max out its
110 * limit and thus block recursive snapshots from being taken in the global zone.
111 * Because of this, it is possible for the snapshot count to be over the limit
112 * and snapshots taken in the global zone could cause a lower level dataset to
113 * hit or exceed its limit. The administrator taking the global zone recursive
114 * snapshot should be aware of this side-effect and behave accordingly.
115 * For consistency, the filesystem limit is also not enforced if the user can
118 * The filesystem and snapshot limits are validated by dsl_fs_ss_limit_check()
119 * and updated by dsl_fs_ss_count_adjust(). A new limit value is setup in
120 * dsl_dir_activate_fs_ss_limit() and the counts are adjusted, if necessary, by
121 * dsl_dir_init_fs_ss_count().
123 * There is a special case when we receive a filesystem that already exists. In
124 * this case a temporary clone name of %X is created (see dmu_recv_begin). We
125 * never update the filesystem counts for temporary clones.
127 * Likewise, we do not update the snapshot counts for temporary snapshots,
128 * such as those created by zfs diff.
131 extern inline dsl_dir_phys_t *dsl_dir_phys(dsl_dir_t *dd);
133 static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
135 typedef struct ddulrt_arg {
136 dsl_dir_t *ddulrta_dd;
141 dsl_dir_evict_async(void *dbu)
144 dsl_pool_t *dp = dd->dd_pool;
149 for (t = 0; t < TXG_SIZE; t++) {
150 ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t));
151 ASSERT(dd->dd_tempreserved[t] == 0);
152 ASSERT(dd->dd_space_towrite[t] == 0);
156 dsl_dir_async_rele(dd->dd_parent, dd);
158 spa_async_close(dd->dd_pool->dp_spa, dd);
161 mutex_destroy(&dd->dd_lock);
162 kmem_free(dd, sizeof (dsl_dir_t));
166 dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj,
167 const char *tail, void *tag, dsl_dir_t **ddp)
173 ASSERT(dsl_pool_config_held(dp));
175 err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
178 dd = dmu_buf_get_user(dbuf);
181 dmu_object_info_t doi;
182 dmu_object_info_from_db(dbuf, &doi);
183 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_DSL_DIR);
184 ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
190 dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
191 dd->dd_object = ddobj;
194 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
197 dsl_dir_snap_cmtime_update(dd);
199 if (dsl_dir_phys(dd)->dd_parent_obj) {
200 err = dsl_dir_hold_obj(dp,
201 dsl_dir_phys(dd)->dd_parent_obj, NULL, dd,
209 err = zap_lookup(dp->dp_meta_objset,
210 dsl_dir_phys(dd->dd_parent)->
211 dd_child_dir_zapobj, tail,
212 sizeof (foundobj), 1, &foundobj);
213 ASSERT(err || foundobj == ddobj);
215 (void) strcpy(dd->dd_myname, tail);
217 err = zap_value_search(dp->dp_meta_objset,
218 dsl_dir_phys(dd->dd_parent)->
220 ddobj, 0, dd->dd_myname);
225 (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
228 if (dsl_dir_is_clone(dd)) {
229 dmu_buf_t *origin_bonus;
230 dsl_dataset_phys_t *origin_phys;
233 * We can't open the origin dataset, because
234 * that would require opening this dsl_dir.
235 * Just look at its phys directly instead.
237 err = dmu_bonus_hold(dp->dp_meta_objset,
238 dsl_dir_phys(dd)->dd_origin_obj, FTAG,
242 origin_phys = origin_bonus->db_data;
244 origin_phys->ds_creation_txg;
245 dmu_buf_rele(origin_bonus, FTAG);
248 dmu_buf_init_user(&dd->dd_dbu, NULL, dsl_dir_evict_async,
250 winner = dmu_buf_set_user_ie(dbuf, &dd->dd_dbu);
251 if (winner != NULL) {
253 dsl_dir_rele(dd->dd_parent, dd);
255 mutex_destroy(&dd->dd_lock);
256 kmem_free(dd, sizeof (dsl_dir_t));
259 spa_open_ref(dp->dp_spa, dd);
264 * The dsl_dir_t has both open-to-close and instantiate-to-evict
265 * holds on the spa. We need the open-to-close holds because
266 * otherwise the spa_refcnt wouldn't change when we open a
267 * dir which the spa also has open, so we could incorrectly
268 * think it was OK to unload/export/destroy the pool. We need
269 * the instantiate-to-evict hold because the dsl_dir_t has a
270 * pointer to the dd_pool, which has a pointer to the spa_t.
272 spa_open_ref(dp->dp_spa, tag);
273 ASSERT3P(dd->dd_pool, ==, dp);
274 ASSERT3U(dd->dd_object, ==, ddobj);
275 ASSERT3P(dd->dd_dbuf, ==, dbuf);
281 dsl_dir_rele(dd->dd_parent, dd);
283 mutex_destroy(&dd->dd_lock);
284 kmem_free(dd, sizeof (dsl_dir_t));
285 dmu_buf_rele(dbuf, tag);
290 dsl_dir_rele(dsl_dir_t *dd, void *tag)
292 dprintf_dd(dd, "%s\n", "");
293 spa_close(dd->dd_pool->dp_spa, tag);
294 dmu_buf_rele(dd->dd_dbuf, tag);
298 * Remove a reference to the given dsl dir that is being asynchronously
299 * released. Async releases occur from a taskq performing eviction of
300 * dsl datasets and dirs. This process is identical to a normal release
301 * with the exception of using the async API for releasing the reference on
305 dsl_dir_async_rele(dsl_dir_t *dd, void *tag)
307 dprintf_dd(dd, "%s\n", "");
308 spa_async_close(dd->dd_pool->dp_spa, tag);
309 dmu_buf_rele(dd->dd_dbuf, tag);
312 /* buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes */
314 dsl_dir_name(dsl_dir_t *dd, char *buf)
317 dsl_dir_name(dd->dd_parent, buf);
318 VERIFY3U(strlcat(buf, "/", ZFS_MAX_DATASET_NAME_LEN), <,
319 ZFS_MAX_DATASET_NAME_LEN);
323 if (!MUTEX_HELD(&dd->dd_lock)) {
325 * recursive mutex so that we can use
326 * dprintf_dd() with dd_lock held
328 mutex_enter(&dd->dd_lock);
329 VERIFY3U(strlcat(buf, dd->dd_myname, ZFS_MAX_DATASET_NAME_LEN),
330 <, ZFS_MAX_DATASET_NAME_LEN);
331 mutex_exit(&dd->dd_lock);
333 VERIFY3U(strlcat(buf, dd->dd_myname, ZFS_MAX_DATASET_NAME_LEN),
334 <, ZFS_MAX_DATASET_NAME_LEN);
338 /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
340 dsl_dir_namelen(dsl_dir_t *dd)
345 /* parent's name + 1 for the "/" */
346 result = dsl_dir_namelen(dd->dd_parent) + 1;
349 if (!MUTEX_HELD(&dd->dd_lock)) {
350 /* see dsl_dir_name */
351 mutex_enter(&dd->dd_lock);
352 result += strlen(dd->dd_myname);
353 mutex_exit(&dd->dd_lock);
355 result += strlen(dd->dd_myname);
362 getcomponent(const char *path, char *component, const char **nextp)
366 if ((path == NULL) || (path[0] == '\0'))
367 return (SET_ERROR(ENOENT));
368 /* This would be a good place to reserve some namespace... */
369 p = strpbrk(path, "/@");
370 if (p && (p[1] == '/' || p[1] == '@')) {
371 /* two separators in a row */
372 return (SET_ERROR(EINVAL));
374 if (p == NULL || p == path) {
376 * if the first thing is an @ or /, it had better be an
377 * @ and it had better not have any more ats or slashes,
378 * and it had better have something after the @.
381 (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
382 return (SET_ERROR(EINVAL));
383 if (strlen(path) >= ZFS_MAX_DATASET_NAME_LEN)
384 return (SET_ERROR(ENAMETOOLONG));
385 (void) strcpy(component, path);
387 } else if (p[0] == '/') {
388 if (p - path >= ZFS_MAX_DATASET_NAME_LEN)
389 return (SET_ERROR(ENAMETOOLONG));
390 (void) strncpy(component, path, p - path);
391 component[p - path] = '\0';
393 } else if (p[0] == '@') {
395 * if the next separator is an @, there better not be
398 if (strchr(path, '/'))
399 return (SET_ERROR(EINVAL));
400 if (p - path >= ZFS_MAX_DATASET_NAME_LEN)
401 return (SET_ERROR(ENAMETOOLONG));
402 (void) strncpy(component, path, p - path);
403 component[p - path] = '\0';
405 panic("invalid p=%p", (void *)p);
412 * Return the dsl_dir_t, and possibly the last component which couldn't
413 * be found in *tail. The name must be in the specified dsl_pool_t. This
414 * thread must hold the dp_config_rwlock for the pool. Returns NULL if the
415 * path is bogus, or if tail==NULL and we couldn't parse the whole name.
416 * (*tail)[0] == '@' means that the last component is a snapshot.
419 dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag,
420 dsl_dir_t **ddp, const char **tailp)
422 char buf[ZFS_MAX_DATASET_NAME_LEN];
423 const char *spaname, *next, *nextnext = NULL;
428 err = getcomponent(name, buf, &next);
432 /* Make sure the name is in the specified pool. */
433 spaname = spa_name(dp->dp_spa);
434 if (strcmp(buf, spaname) != 0)
435 return (SET_ERROR(EXDEV));
437 ASSERT(dsl_pool_config_held(dp));
439 err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
444 while (next != NULL) {
446 err = getcomponent(next, buf, &nextnext);
449 ASSERT(next[0] != '\0');
452 dprintf("looking up %s in obj%lld\n",
453 buf, dsl_dir_phys(dd)->dd_child_dir_zapobj);
455 err = zap_lookup(dp->dp_meta_objset,
456 dsl_dir_phys(dd)->dd_child_dir_zapobj,
457 buf, sizeof (ddobj), 1, &ddobj);
464 err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_dd);
467 dsl_dir_rele(dd, tag);
473 dsl_dir_rele(dd, tag);
478 * It's an error if there's more than one component left, or
479 * tailp==NULL and there's any component left.
482 (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
484 dsl_dir_rele(dd, tag);
485 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
486 err = SET_ERROR(ENOENT);
495 * If the counts are already initialized for this filesystem and its
496 * descendants then do nothing, otherwise initialize the counts.
498 * The counts on this filesystem, and those below, may be uninitialized due to
499 * either the use of a pre-existing pool which did not support the
500 * filesystem/snapshot limit feature, or one in which the feature had not yet
503 * Recursively descend the filesystem tree and update the filesystem/snapshot
504 * counts on each filesystem below, then update the cumulative count on the
505 * current filesystem. If the filesystem already has a count set on it,
506 * then we know that its counts, and the counts on the filesystems below it,
507 * are already correct, so we don't have to update this filesystem.
510 dsl_dir_init_fs_ss_count(dsl_dir_t *dd, dmu_tx_t *tx)
512 uint64_t my_fs_cnt = 0;
513 uint64_t my_ss_cnt = 0;
514 dsl_pool_t *dp = dd->dd_pool;
515 objset_t *os = dp->dp_meta_objset;
520 ASSERT(spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT));
521 ASSERT(dsl_pool_config_held(dp));
522 ASSERT(dmu_tx_is_syncing(tx));
524 dsl_dir_zapify(dd, tx);
527 * If the filesystem count has already been initialized then we
528 * don't need to recurse down any further.
530 if (zap_contains(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT) == 0)
533 zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP);
534 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
536 /* Iterate my child dirs */
537 for (zap_cursor_init(zc, os, dsl_dir_phys(dd)->dd_child_dir_zapobj);
538 zap_cursor_retrieve(zc, za) == 0; zap_cursor_advance(zc)) {
542 VERIFY0(dsl_dir_hold_obj(dp, za->za_first_integer, NULL, FTAG,
546 * Ignore hidden ($FREE, $MOS & $ORIGIN) objsets and
547 * temporary datasets.
549 if (chld_dd->dd_myname[0] == '$' ||
550 chld_dd->dd_myname[0] == '%') {
551 dsl_dir_rele(chld_dd, FTAG);
555 my_fs_cnt++; /* count this child */
557 dsl_dir_init_fs_ss_count(chld_dd, tx);
559 VERIFY0(zap_lookup(os, chld_dd->dd_object,
560 DD_FIELD_FILESYSTEM_COUNT, sizeof (count), 1, &count));
562 VERIFY0(zap_lookup(os, chld_dd->dd_object,
563 DD_FIELD_SNAPSHOT_COUNT, sizeof (count), 1, &count));
566 dsl_dir_rele(chld_dd, FTAG);
569 /* Count my snapshots (we counted children's snapshots above) */
570 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
571 dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds));
573 for (zap_cursor_init(zc, os, dsl_dataset_phys(ds)->ds_snapnames_zapobj);
574 zap_cursor_retrieve(zc, za) == 0;
575 zap_cursor_advance(zc)) {
576 /* Don't count temporary snapshots */
577 if (za->za_name[0] != '%')
582 dsl_dataset_rele(ds, FTAG);
584 kmem_free(zc, sizeof (zap_cursor_t));
585 kmem_free(za, sizeof (zap_attribute_t));
587 /* we're in a sync task, update counts */
588 dmu_buf_will_dirty(dd->dd_dbuf, tx);
589 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
590 sizeof (my_fs_cnt), 1, &my_fs_cnt, tx));
591 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
592 sizeof (my_ss_cnt), 1, &my_ss_cnt, tx));
596 dsl_dir_actv_fs_ss_limit_check(void *arg, dmu_tx_t *tx)
598 char *ddname = (char *)arg;
599 dsl_pool_t *dp = dmu_tx_pool(tx);
604 error = dsl_dataset_hold(dp, ddname, FTAG, &ds);
608 if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
609 dsl_dataset_rele(ds, FTAG);
610 return (SET_ERROR(ENOTSUP));
614 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT) &&
615 dsl_dir_is_zapified(dd) &&
616 zap_contains(dp->dp_meta_objset, dd->dd_object,
617 DD_FIELD_FILESYSTEM_COUNT) == 0) {
618 dsl_dataset_rele(ds, FTAG);
619 return (SET_ERROR(EALREADY));
622 dsl_dataset_rele(ds, FTAG);
627 dsl_dir_actv_fs_ss_limit_sync(void *arg, dmu_tx_t *tx)
629 char *ddname = (char *)arg;
630 dsl_pool_t *dp = dmu_tx_pool(tx);
634 VERIFY0(dsl_dataset_hold(dp, ddname, FTAG, &ds));
636 spa = dsl_dataset_get_spa(ds);
638 if (!spa_feature_is_active(spa, SPA_FEATURE_FS_SS_LIMIT)) {
640 * Since the feature was not active and we're now setting a
641 * limit, increment the feature-active counter so that the
642 * feature becomes active for the first time.
644 * We are already in a sync task so we can update the MOS.
646 spa_feature_incr(spa, SPA_FEATURE_FS_SS_LIMIT, tx);
650 * Since we are now setting a non-UINT64_MAX limit on the filesystem,
651 * we need to ensure the counts are correct. Descend down the tree from
652 * this point and update all of the counts to be accurate.
654 dsl_dir_init_fs_ss_count(ds->ds_dir, tx);
656 dsl_dataset_rele(ds, FTAG);
660 * Make sure the feature is enabled and activate it if necessary.
661 * Since we're setting a limit, ensure the on-disk counts are valid.
662 * This is only called by the ioctl path when setting a limit value.
664 * We do not need to validate the new limit, since users who can change the
665 * limit are also allowed to exceed the limit.
668 dsl_dir_activate_fs_ss_limit(const char *ddname)
672 error = dsl_sync_task(ddname, dsl_dir_actv_fs_ss_limit_check,
673 dsl_dir_actv_fs_ss_limit_sync, (void *)ddname, 0,
674 ZFS_SPACE_CHECK_RESERVED);
676 if (error == EALREADY)
683 * Used to determine if the filesystem_limit or snapshot_limit should be
684 * enforced. We allow the limit to be exceeded if the user has permission to
685 * write the property value. We pass in the creds that we got in the open
686 * context since we will always be the GZ root in syncing context. We also have
687 * to handle the case where we are allowed to change the limit on the current
688 * dataset, but there may be another limit in the tree above.
690 * We can never modify these two properties within a non-global zone. In
691 * addition, the other checks are modeled on zfs_secpolicy_write_perms. We
692 * can't use that function since we are already holding the dp_config_rwlock.
693 * In addition, we already have the dd and dealing with snapshots is simplified
704 dsl_enforce_ds_ss_limits(dsl_dir_t *dd, zfs_prop_t prop, cred_t *cr)
706 enforce_res_t enforce = ENFORCE_ALWAYS;
711 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
712 prop == ZFS_PROP_SNAPSHOT_LIMIT);
718 if (crgetzoneid(cr) != GLOBAL_ZONEID)
720 return (ENFORCE_ALWAYS);
722 if (secpolicy_zfs(cr) == 0)
723 return (ENFORCE_NEVER);
726 if ((obj = dsl_dir_phys(dd)->dd_head_dataset_obj) == 0)
727 return (ENFORCE_ALWAYS);
729 ASSERT(dsl_pool_config_held(dd->dd_pool));
731 if (dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds) != 0)
732 return (ENFORCE_ALWAYS);
734 if (dsl_prop_get_ds(ds, "zoned", 8, 1, &zoned, NULL) || zoned) {
735 /* Only root can access zoned fs's from the GZ */
736 enforce = ENFORCE_ALWAYS;
738 if (dsl_deleg_access_impl(ds, zfs_prop_to_name(prop), cr) == 0)
739 enforce = ENFORCE_ABOVE;
742 dsl_dataset_rele(ds, FTAG);
747 dsl_dir_update_last_remap_txg_sync(void *varg, dmu_tx_t *tx)
749 ddulrt_arg_t *arg = varg;
750 uint64_t last_remap_txg;
751 dsl_dir_t *dd = arg->ddulrta_dd;
752 objset_t *mos = dd->dd_pool->dp_meta_objset;
754 dsl_dir_zapify(dd, tx);
755 if (zap_lookup(mos, dd->dd_object, DD_FIELD_LAST_REMAP_TXG,
756 sizeof (last_remap_txg), 1, &last_remap_txg) != 0 ||
757 last_remap_txg < arg->ddlrta_txg) {
758 VERIFY0(zap_update(mos, dd->dd_object, DD_FIELD_LAST_REMAP_TXG,
759 sizeof (arg->ddlrta_txg), 1, &arg->ddlrta_txg, tx));
764 dsl_dir_update_last_remap_txg(dsl_dir_t *dd, uint64_t txg)
768 arg.ddlrta_txg = txg;
770 return (dsl_sync_task(spa_name(dd->dd_pool->dp_spa),
771 NULL, dsl_dir_update_last_remap_txg_sync, &arg,
772 1, ZFS_SPACE_CHECK_RESERVED));
776 * Check if adding additional child filesystem(s) would exceed any filesystem
777 * limits or adding additional snapshot(s) would exceed any snapshot limits.
778 * The prop argument indicates which limit to check.
780 * Note that all filesystem limits up to the root (or the highest
781 * initialized) filesystem or the given ancestor must be satisfied.
784 dsl_fs_ss_limit_check(dsl_dir_t *dd, uint64_t delta, zfs_prop_t prop,
785 dsl_dir_t *ancestor, cred_t *cr)
787 objset_t *os = dd->dd_pool->dp_meta_objset;
788 uint64_t limit, count;
790 enforce_res_t enforce;
793 ASSERT(dsl_pool_config_held(dd->dd_pool));
794 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
795 prop == ZFS_PROP_SNAPSHOT_LIMIT);
798 * If we're allowed to change the limit, don't enforce the limit
799 * e.g. this can happen if a snapshot is taken by an administrative
800 * user in the global zone (i.e. a recursive snapshot by root).
801 * However, we must handle the case of delegated permissions where we
802 * are allowed to change the limit on the current dataset, but there
803 * is another limit in the tree above.
805 enforce = dsl_enforce_ds_ss_limits(dd, prop, cr);
806 if (enforce == ENFORCE_NEVER)
810 * e.g. if renaming a dataset with no snapshots, count adjustment
816 if (prop == ZFS_PROP_SNAPSHOT_LIMIT) {
818 * We don't enforce the limit for temporary snapshots. This is
819 * indicated by a NULL cred_t argument.
824 count_prop = DD_FIELD_SNAPSHOT_COUNT;
826 count_prop = DD_FIELD_FILESYSTEM_COUNT;
830 * If an ancestor has been provided, stop checking the limit once we
831 * hit that dir. We need this during rename so that we don't overcount
832 * the check once we recurse up to the common ancestor.
838 * If we hit an uninitialized node while recursing up the tree, we can
839 * stop since we know there is no limit here (or above). The counts are
840 * not valid on this node and we know we won't touch this node's counts.
842 if (!dsl_dir_is_zapified(dd) || zap_lookup(os, dd->dd_object,
843 count_prop, sizeof (count), 1, &count) == ENOENT)
846 err = dsl_prop_get_dd(dd, zfs_prop_to_name(prop), 8, 1, &limit, NULL,
851 /* Is there a limit which we've hit? */
852 if (enforce == ENFORCE_ALWAYS && (count + delta) > limit)
853 return (SET_ERROR(EDQUOT));
855 if (dd->dd_parent != NULL)
856 err = dsl_fs_ss_limit_check(dd->dd_parent, delta, prop,
863 * Adjust the filesystem or snapshot count for the specified dsl_dir_t and all
864 * parents. When a new filesystem/snapshot is created, increment the count on
865 * all parents, and when a filesystem/snapshot is destroyed, decrement the
869 dsl_fs_ss_count_adjust(dsl_dir_t *dd, int64_t delta, const char *prop,
873 objset_t *os = dd->dd_pool->dp_meta_objset;
876 ASSERT(dsl_pool_config_held(dd->dd_pool));
877 ASSERT(dmu_tx_is_syncing(tx));
878 ASSERT(strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0 ||
879 strcmp(prop, DD_FIELD_SNAPSHOT_COUNT) == 0);
882 * When we receive an incremental stream into a filesystem that already
883 * exists, a temporary clone is created. We don't count this temporary
884 * clone, whose name begins with a '%'. We also ignore hidden ($FREE,
885 * $MOS & $ORIGIN) objsets.
887 if ((dd->dd_myname[0] == '%' || dd->dd_myname[0] == '$') &&
888 strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0)
892 * e.g. if renaming a dataset with no snapshots, count adjustment is 0
898 * If we hit an uninitialized node while recursing up the tree, we can
899 * stop since we know the counts are not valid on this node and we
900 * know we shouldn't touch this node's counts. An uninitialized count
901 * on the node indicates that either the feature has not yet been
902 * activated or there are no limits on this part of the tree.
904 if (!dsl_dir_is_zapified(dd) || (err = zap_lookup(os, dd->dd_object,
905 prop, sizeof (count), 1, &count)) == ENOENT)
910 /* Use a signed verify to make sure we're not neg. */
911 VERIFY3S(count, >=, 0);
913 VERIFY0(zap_update(os, dd->dd_object, prop, sizeof (count), 1, &count,
916 /* Roll up this additional count into our ancestors */
917 if (dd->dd_parent != NULL)
918 dsl_fs_ss_count_adjust(dd->dd_parent, delta, prop, tx);
922 dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
925 objset_t *mos = dp->dp_meta_objset;
927 dsl_dir_phys_t *ddphys;
930 ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
931 DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
933 VERIFY(0 == zap_add(mos, dsl_dir_phys(pds)->dd_child_dir_zapobj,
934 name, sizeof (uint64_t), 1, &ddobj, tx));
936 /* it's the root dir */
937 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
938 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
940 VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
941 dmu_buf_will_dirty(dbuf, tx);
942 ddphys = dbuf->db_data;
944 ddphys->dd_creation_time = gethrestime_sec();
946 ddphys->dd_parent_obj = pds->dd_object;
948 /* update the filesystem counts */
949 dsl_fs_ss_count_adjust(pds, 1, DD_FIELD_FILESYSTEM_COUNT, tx);
951 ddphys->dd_props_zapobj = zap_create(mos,
952 DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
953 ddphys->dd_child_dir_zapobj = zap_create(mos,
954 DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
955 if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
956 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
957 dmu_buf_rele(dbuf, FTAG);
963 dsl_dir_is_clone(dsl_dir_t *dd)
965 return (dsl_dir_phys(dd)->dd_origin_obj &&
966 (dd->dd_pool->dp_origin_snap == NULL ||
967 dsl_dir_phys(dd)->dd_origin_obj !=
968 dd->dd_pool->dp_origin_snap->ds_object));
973 dsl_dir_get_used(dsl_dir_t *dd)
975 return (dsl_dir_phys(dd)->dd_used_bytes);
979 dsl_dir_get_quota(dsl_dir_t *dd)
981 return (dsl_dir_phys(dd)->dd_quota);
985 dsl_dir_get_reservation(dsl_dir_t *dd)
987 return (dsl_dir_phys(dd)->dd_reserved);
991 dsl_dir_get_compressratio(dsl_dir_t *dd)
993 /* a fixed point number, 100x the ratio */
994 return (dsl_dir_phys(dd)->dd_compressed_bytes == 0 ? 100 :
995 (dsl_dir_phys(dd)->dd_uncompressed_bytes * 100 /
996 dsl_dir_phys(dd)->dd_compressed_bytes));
1000 dsl_dir_get_logicalused(dsl_dir_t *dd)
1002 return (dsl_dir_phys(dd)->dd_uncompressed_bytes);
1006 dsl_dir_get_usedsnap(dsl_dir_t *dd)
1008 return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]);
1012 dsl_dir_get_usedds(dsl_dir_t *dd)
1014 return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_HEAD]);
1018 dsl_dir_get_usedrefreserv(dsl_dir_t *dd)
1020 return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_REFRSRV]);
1024 dsl_dir_get_usedchild(dsl_dir_t *dd)
1026 return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD] +
1027 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD_RSRV]);
1031 dsl_dir_get_origin(dsl_dir_t *dd, char *buf)
1034 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
1035 dsl_dir_phys(dd)->dd_origin_obj, FTAG, &ds));
1037 dsl_dataset_name(ds, buf);
1039 dsl_dataset_rele(ds, FTAG);
1043 dsl_dir_get_filesystem_count(dsl_dir_t *dd, uint64_t *count)
1045 if (dsl_dir_is_zapified(dd)) {
1046 objset_t *os = dd->dd_pool->dp_meta_objset;
1047 return (zap_lookup(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
1048 sizeof (*count), 1, count));
1055 dsl_dir_get_snapshot_count(dsl_dir_t *dd, uint64_t *count)
1057 if (dsl_dir_is_zapified(dd)) {
1058 objset_t *os = dd->dd_pool->dp_meta_objset;
1059 return (zap_lookup(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
1060 sizeof (*count), 1, count));
1067 dsl_dir_get_remaptxg(dsl_dir_t *dd, uint64_t *count)
1069 if (dsl_dir_is_zapified(dd)) {
1070 objset_t *os = dd->dd_pool->dp_meta_objset;
1071 return (zap_lookup(os, dd->dd_object, DD_FIELD_LAST_REMAP_TXG,
1072 sizeof (*count), 1, count));
1079 dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
1081 mutex_enter(&dd->dd_lock);
1082 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA,
1083 dsl_dir_get_quota(dd));
1084 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
1085 dsl_dir_get_reservation(dd));
1086 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED,
1087 dsl_dir_get_logicalused(dd));
1088 if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1089 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
1090 dsl_dir_get_usedsnap(dd));
1091 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
1092 dsl_dir_get_usedds(dd));
1093 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
1094 dsl_dir_get_usedrefreserv(dd));
1095 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
1096 dsl_dir_get_usedchild(dd));
1098 mutex_exit(&dd->dd_lock);
1101 if (dsl_dir_get_filesystem_count(dd, &count) == 0) {
1102 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_FILESYSTEM_COUNT,
1105 if (dsl_dir_get_snapshot_count(dd, &count) == 0) {
1106 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_SNAPSHOT_COUNT,
1109 if (dsl_dir_get_remaptxg(dd, &count) == 0) {
1110 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REMAPTXG,
1114 if (dsl_dir_is_clone(dd)) {
1115 char buf[ZFS_MAX_DATASET_NAME_LEN];
1116 dsl_dir_get_origin(dd, buf);
1117 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
1123 dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
1125 dsl_pool_t *dp = dd->dd_pool;
1127 ASSERT(dsl_dir_phys(dd));
1129 if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) {
1130 /* up the hold count until we can be written out */
1131 dmu_buf_add_ref(dd->dd_dbuf, dd);
1136 parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
1138 uint64_t old_accounted = MAX(used, dsl_dir_phys(dd)->dd_reserved);
1139 uint64_t new_accounted =
1140 MAX(used + delta, dsl_dir_phys(dd)->dd_reserved);
1141 return (new_accounted - old_accounted);
1145 dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
1147 ASSERT(dmu_tx_is_syncing(tx));
1149 mutex_enter(&dd->dd_lock);
1150 ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]);
1151 dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
1152 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
1153 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
1154 mutex_exit(&dd->dd_lock);
1156 /* release the hold from dsl_dir_dirty */
1157 dmu_buf_rele(dd->dd_dbuf, dd);
1161 dsl_dir_space_towrite(dsl_dir_t *dd)
1165 ASSERT(MUTEX_HELD(&dd->dd_lock));
1167 for (int i = 0; i < TXG_SIZE; i++) {
1168 space += dd->dd_space_towrite[i & TXG_MASK];
1169 ASSERT3U(dd->dd_space_towrite[i & TXG_MASK], >=, 0);
1175 * How much space would dd have available if ancestor had delta applied
1176 * to it? If ondiskonly is set, we're only interested in what's
1177 * on-disk, not estimated pending changes.
1180 dsl_dir_space_available(dsl_dir_t *dd,
1181 dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
1183 uint64_t parentspace, myspace, quota, used;
1186 * If there are no restrictions otherwise, assume we have
1187 * unlimited space available.
1190 parentspace = UINT64_MAX;
1192 if (dd->dd_parent != NULL) {
1193 parentspace = dsl_dir_space_available(dd->dd_parent,
1194 ancestor, delta, ondiskonly);
1197 mutex_enter(&dd->dd_lock);
1198 if (dsl_dir_phys(dd)->dd_quota != 0)
1199 quota = dsl_dir_phys(dd)->dd_quota;
1200 used = dsl_dir_phys(dd)->dd_used_bytes;
1202 used += dsl_dir_space_towrite(dd);
1204 if (dd->dd_parent == NULL) {
1205 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
1206 quota = MIN(quota, poolsize);
1209 if (dsl_dir_phys(dd)->dd_reserved > used && parentspace != UINT64_MAX) {
1211 * We have some space reserved, in addition to what our
1214 parentspace += dsl_dir_phys(dd)->dd_reserved - used;
1217 if (dd == ancestor) {
1219 ASSERT(used >= -delta);
1221 if (parentspace != UINT64_MAX)
1222 parentspace -= delta;
1230 * the lesser of the space provided by our parent and
1231 * the space left in our quota
1233 myspace = MIN(parentspace, quota - used);
1236 mutex_exit(&dd->dd_lock);
1241 struct tempreserve {
1242 list_node_t tr_node;
1248 dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
1249 boolean_t ignorequota, list_t *tr_list,
1250 dmu_tx_t *tx, boolean_t first)
1252 uint64_t txg = tx->tx_txg;
1254 struct tempreserve *tr;
1255 int retval = EDQUOT;
1256 uint64_t ref_rsrv = 0;
1258 ASSERT3U(txg, !=, 0);
1259 ASSERT3S(asize, >, 0);
1261 mutex_enter(&dd->dd_lock);
1264 * Check against the dsl_dir's quota. We don't add in the delta
1265 * when checking for over-quota because they get one free hit.
1267 uint64_t est_inflight = dsl_dir_space_towrite(dd);
1268 for (int i = 0; i < TXG_SIZE; i++)
1269 est_inflight += dd->dd_tempreserved[i];
1270 uint64_t used_on_disk = dsl_dir_phys(dd)->dd_used_bytes;
1273 * On the first iteration, fetch the dataset's used-on-disk and
1274 * refreservation values. Also, if checkrefquota is set, test if
1275 * allocating this space would exceed the dataset's refquota.
1277 if (first && tx->tx_objset) {
1279 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
1281 error = dsl_dataset_check_quota(ds, !netfree,
1282 asize, est_inflight, &used_on_disk, &ref_rsrv);
1284 mutex_exit(&dd->dd_lock);
1290 * If this transaction will result in a net free of space,
1291 * we want to let it through.
1293 if (ignorequota || netfree || dsl_dir_phys(dd)->dd_quota == 0)
1296 quota = dsl_dir_phys(dd)->dd_quota;
1299 * Adjust the quota against the actual pool size at the root
1300 * minus any outstanding deferred frees.
1301 * To ensure that it's possible to remove files from a full
1302 * pool without inducing transient overcommits, we throttle
1303 * netfree transactions against a quota that is slightly larger,
1304 * but still within the pool's allocation slop. In cases where
1305 * we're very close to full, this will allow a steady trickle of
1306 * removes to get through.
1308 uint64_t deferred = 0;
1309 if (dd->dd_parent == NULL) {
1310 spa_t *spa = dd->dd_pool->dp_spa;
1311 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
1312 deferred = metaslab_class_get_deferred(spa_normal_class(spa));
1313 if (poolsize - deferred < quota) {
1314 quota = poolsize - deferred;
1320 * If they are requesting more space, and our current estimate
1321 * is over quota, they get to try again unless the actual
1322 * on-disk is over quota and there are no pending changes (which
1323 * may free up space for us).
1325 if (used_on_disk + est_inflight >= quota) {
1326 if (est_inflight > 0 || used_on_disk < quota ||
1327 (retval == ENOSPC && used_on_disk < quota + deferred))
1329 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
1330 "quota=%lluK tr=%lluK err=%d\n",
1331 used_on_disk>>10, est_inflight>>10,
1332 quota>>10, asize>>10, retval);
1333 mutex_exit(&dd->dd_lock);
1334 return (SET_ERROR(retval));
1337 /* We need to up our estimated delta before dropping dd_lock */
1338 dd->dd_tempreserved[txg & TXG_MASK] += asize;
1340 uint64_t parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
1342 mutex_exit(&dd->dd_lock);
1344 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1346 tr->tr_size = asize;
1347 list_insert_tail(tr_list, tr);
1349 /* see if it's OK with our parent */
1350 if (dd->dd_parent != NULL && parent_rsrv != 0) {
1351 boolean_t ismos = (dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
1353 return (dsl_dir_tempreserve_impl(dd->dd_parent,
1354 parent_rsrv, netfree, ismos, tr_list, tx, B_FALSE));
1361 * Reserve space in this dsl_dir, to be used in this tx's txg.
1362 * After the space has been dirtied (and dsl_dir_willuse_space()
1363 * has been called), the reservation should be canceled, using
1364 * dsl_dir_tempreserve_clear().
1367 dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
1368 boolean_t netfree, void **tr_cookiep, dmu_tx_t *tx)
1378 tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
1379 list_create(tr_list, sizeof (struct tempreserve),
1380 offsetof(struct tempreserve, tr_node));
1381 ASSERT3S(asize, >, 0);
1383 err = arc_tempreserve_space(lsize, tx->tx_txg);
1385 struct tempreserve *tr;
1387 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1388 tr->tr_size = lsize;
1389 list_insert_tail(tr_list, tr);
1391 if (err == EAGAIN) {
1393 * If arc_memory_throttle() detected that pageout
1394 * is running and we are low on memory, we delay new
1395 * non-pageout transactions to give pageout an
1398 * It is unfortunate to be delaying while the caller's
1401 txg_delay(dd->dd_pool, tx->tx_txg,
1402 MSEC2NSEC(10), MSEC2NSEC(10));
1403 err = SET_ERROR(ERESTART);
1408 err = dsl_dir_tempreserve_impl(dd, asize, netfree,
1409 B_FALSE, tr_list, tx, B_TRUE);
1413 dsl_dir_tempreserve_clear(tr_list, tx);
1415 *tr_cookiep = tr_list;
1421 * Clear a temporary reservation that we previously made with
1422 * dsl_dir_tempreserve_space().
1425 dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
1427 int txgidx = tx->tx_txg & TXG_MASK;
1428 list_t *tr_list = tr_cookie;
1429 struct tempreserve *tr;
1431 ASSERT3U(tx->tx_txg, !=, 0);
1433 if (tr_cookie == NULL)
1436 while ((tr = list_head(tr_list)) != NULL) {
1438 mutex_enter(&tr->tr_ds->dd_lock);
1439 ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
1441 tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
1442 mutex_exit(&tr->tr_ds->dd_lock);
1444 arc_tempreserve_clear(tr->tr_size);
1446 list_remove(tr_list, tr);
1447 kmem_free(tr, sizeof (struct tempreserve));
1450 kmem_free(tr_list, sizeof (list_t));
1454 * This should be called from open context when we think we're going to write
1455 * or free space, for example when dirtying data. Be conservative; it's okay
1456 * to write less space or free more, but we don't want to write more or free
1457 * less than the amount specified.
1460 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
1462 int64_t parent_space;
1465 mutex_enter(&dd->dd_lock);
1467 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
1469 est_used = dsl_dir_space_towrite(dd) + dsl_dir_phys(dd)->dd_used_bytes;
1470 parent_space = parent_delta(dd, est_used, space);
1471 mutex_exit(&dd->dd_lock);
1473 /* Make sure that we clean up dd_space_to* */
1474 dsl_dir_dirty(dd, tx);
1476 /* XXX this is potentially expensive and unnecessary... */
1477 if (parent_space && dd->dd_parent)
1478 dsl_dir_willuse_space(dd->dd_parent, parent_space, tx);
1481 /* call from syncing context when we actually write/free space for this dd */
1483 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
1484 int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
1486 int64_t accounted_delta;
1489 * dsl_dataset_set_refreservation_sync_impl() calls this with
1490 * dd_lock held, so that it can atomically update
1491 * ds->ds_reserved and the dsl_dir accounting, so that
1492 * dsl_dataset_check_quota() can see dataset and dir accounting
1495 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
1497 ASSERT(dmu_tx_is_syncing(tx));
1498 ASSERT(type < DD_USED_NUM);
1500 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1503 mutex_enter(&dd->dd_lock);
1505 parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, used);
1506 ASSERT(used >= 0 || dsl_dir_phys(dd)->dd_used_bytes >= -used);
1507 ASSERT(compressed >= 0 ||
1508 dsl_dir_phys(dd)->dd_compressed_bytes >= -compressed);
1509 ASSERT(uncompressed >= 0 ||
1510 dsl_dir_phys(dd)->dd_uncompressed_bytes >= -uncompressed);
1511 dsl_dir_phys(dd)->dd_used_bytes += used;
1512 dsl_dir_phys(dd)->dd_uncompressed_bytes += uncompressed;
1513 dsl_dir_phys(dd)->dd_compressed_bytes += compressed;
1515 if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1517 dsl_dir_phys(dd)->dd_used_breakdown[type] >= -used);
1518 dsl_dir_phys(dd)->dd_used_breakdown[type] += used;
1522 for (t = 0; t < DD_USED_NUM; t++)
1523 u += dsl_dir_phys(dd)->dd_used_breakdown[t];
1524 ASSERT3U(u, ==, dsl_dir_phys(dd)->dd_used_bytes);
1528 mutex_exit(&dd->dd_lock);
1530 if (dd->dd_parent != NULL) {
1531 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1532 accounted_delta, compressed, uncompressed, tx);
1533 dsl_dir_transfer_space(dd->dd_parent,
1534 used - accounted_delta,
1535 DD_USED_CHILD_RSRV, DD_USED_CHILD, NULL);
1540 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
1541 dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
1543 ASSERT(tx == NULL || dmu_tx_is_syncing(tx));
1544 ASSERT(oldtype < DD_USED_NUM);
1545 ASSERT(newtype < DD_USED_NUM);
1548 !(dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN))
1552 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1553 mutex_enter(&dd->dd_lock);
1555 dsl_dir_phys(dd)->dd_used_breakdown[oldtype] >= delta :
1556 dsl_dir_phys(dd)->dd_used_breakdown[newtype] >= -delta);
1557 ASSERT(dsl_dir_phys(dd)->dd_used_bytes >= ABS(delta));
1558 dsl_dir_phys(dd)->dd_used_breakdown[oldtype] -= delta;
1559 dsl_dir_phys(dd)->dd_used_breakdown[newtype] += delta;
1560 mutex_exit(&dd->dd_lock);
1563 typedef struct dsl_dir_set_qr_arg {
1564 const char *ddsqra_name;
1565 zprop_source_t ddsqra_source;
1566 uint64_t ddsqra_value;
1567 } dsl_dir_set_qr_arg_t;
1570 dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx)
1572 dsl_dir_set_qr_arg_t *ddsqra = arg;
1573 dsl_pool_t *dp = dmu_tx_pool(tx);
1576 uint64_t towrite, newval;
1578 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1582 error = dsl_prop_predict(ds->ds_dir, "quota",
1583 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1585 dsl_dataset_rele(ds, FTAG);
1590 dsl_dataset_rele(ds, FTAG);
1594 mutex_enter(&ds->ds_dir->dd_lock);
1596 * If we are doing the preliminary check in open context, and
1597 * there are pending changes, then don't fail it, since the
1598 * pending changes could under-estimate the amount of space to be
1601 towrite = dsl_dir_space_towrite(ds->ds_dir);
1602 if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
1603 (newval < dsl_dir_phys(ds->ds_dir)->dd_reserved ||
1604 newval < dsl_dir_phys(ds->ds_dir)->dd_used_bytes + towrite)) {
1605 error = SET_ERROR(ENOSPC);
1607 mutex_exit(&ds->ds_dir->dd_lock);
1608 dsl_dataset_rele(ds, FTAG);
1613 dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx)
1615 dsl_dir_set_qr_arg_t *ddsqra = arg;
1616 dsl_pool_t *dp = dmu_tx_pool(tx);
1620 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1622 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1623 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA),
1624 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1625 &ddsqra->ddsqra_value, tx);
1627 VERIFY0(dsl_prop_get_int_ds(ds,
1628 zfs_prop_to_name(ZFS_PROP_QUOTA), &newval));
1630 newval = ddsqra->ddsqra_value;
1631 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1632 zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval);
1635 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1636 mutex_enter(&ds->ds_dir->dd_lock);
1637 dsl_dir_phys(ds->ds_dir)->dd_quota = newval;
1638 mutex_exit(&ds->ds_dir->dd_lock);
1639 dsl_dataset_rele(ds, FTAG);
1643 dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
1645 dsl_dir_set_qr_arg_t ddsqra;
1647 ddsqra.ddsqra_name = ddname;
1648 ddsqra.ddsqra_source = source;
1649 ddsqra.ddsqra_value = quota;
1651 return (dsl_sync_task(ddname, dsl_dir_set_quota_check,
1652 dsl_dir_set_quota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
1656 dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx)
1658 dsl_dir_set_qr_arg_t *ddsqra = arg;
1659 dsl_pool_t *dp = dmu_tx_pool(tx);
1662 uint64_t newval, used, avail;
1665 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1671 * If we are doing the preliminary check in open context, the
1672 * space estimates may be inaccurate.
1674 if (!dmu_tx_is_syncing(tx)) {
1675 dsl_dataset_rele(ds, FTAG);
1679 error = dsl_prop_predict(ds->ds_dir,
1680 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1681 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1683 dsl_dataset_rele(ds, FTAG);
1687 mutex_enter(&dd->dd_lock);
1688 used = dsl_dir_phys(dd)->dd_used_bytes;
1689 mutex_exit(&dd->dd_lock);
1691 if (dd->dd_parent) {
1692 avail = dsl_dir_space_available(dd->dd_parent,
1695 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
1698 if (MAX(used, newval) > MAX(used, dsl_dir_phys(dd)->dd_reserved)) {
1699 uint64_t delta = MAX(used, newval) -
1700 MAX(used, dsl_dir_phys(dd)->dd_reserved);
1702 if (delta > avail ||
1703 (dsl_dir_phys(dd)->dd_quota > 0 &&
1704 newval > dsl_dir_phys(dd)->dd_quota))
1705 error = SET_ERROR(ENOSPC);
1708 dsl_dataset_rele(ds, FTAG);
1713 dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
1718 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1720 mutex_enter(&dd->dd_lock);
1721 used = dsl_dir_phys(dd)->dd_used_bytes;
1722 delta = MAX(used, value) - MAX(used, dsl_dir_phys(dd)->dd_reserved);
1723 dsl_dir_phys(dd)->dd_reserved = value;
1725 if (dd->dd_parent != NULL) {
1726 /* Roll up this additional usage into our ancestors */
1727 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1730 mutex_exit(&dd->dd_lock);
1734 dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx)
1736 dsl_dir_set_qr_arg_t *ddsqra = arg;
1737 dsl_pool_t *dp = dmu_tx_pool(tx);
1741 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1743 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1744 dsl_prop_set_sync_impl(ds,
1745 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1746 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1747 &ddsqra->ddsqra_value, tx);
1749 VERIFY0(dsl_prop_get_int_ds(ds,
1750 zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval));
1752 newval = ddsqra->ddsqra_value;
1753 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1754 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1755 (longlong_t)newval);
1758 dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx);
1759 dsl_dataset_rele(ds, FTAG);
1763 dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1764 uint64_t reservation)
1766 dsl_dir_set_qr_arg_t ddsqra;
1768 ddsqra.ddsqra_name = ddname;
1769 ddsqra.ddsqra_source = source;
1770 ddsqra.ddsqra_value = reservation;
1772 return (dsl_sync_task(ddname, dsl_dir_set_reservation_check,
1773 dsl_dir_set_reservation_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
1777 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1779 for (; ds1; ds1 = ds1->dd_parent) {
1781 for (dd = ds2; dd; dd = dd->dd_parent) {
1790 * If delta is applied to dd, how much of that delta would be applied to
1791 * ancestor? Syncing context only.
1794 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1799 mutex_enter(&dd->dd_lock);
1800 delta = parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, delta);
1801 mutex_exit(&dd->dd_lock);
1802 return (would_change(dd->dd_parent, delta, ancestor));
1805 typedef struct dsl_dir_rename_arg {
1806 const char *ddra_oldname;
1807 const char *ddra_newname;
1809 } dsl_dir_rename_arg_t;
1813 dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1816 char namebuf[ZFS_MAX_DATASET_NAME_LEN];
1818 dsl_dataset_name(ds, namebuf);
1820 if (strlen(namebuf) + *deltap >= ZFS_MAX_DATASET_NAME_LEN)
1821 return (SET_ERROR(ENAMETOOLONG));
1826 dsl_dir_rename_check(void *arg, dmu_tx_t *tx)
1828 dsl_dir_rename_arg_t *ddra = arg;
1829 dsl_pool_t *dp = dmu_tx_pool(tx);
1830 dsl_dir_t *dd, *newparent;
1831 const char *mynewname;
1833 int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname);
1835 /* target dir should exist */
1836 error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL);
1840 /* new parent should exist */
1841 error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG,
1842 &newparent, &mynewname);
1844 dsl_dir_rele(dd, FTAG);
1848 /* can't rename to different pool */
1849 if (dd->dd_pool != newparent->dd_pool) {
1850 dsl_dir_rele(newparent, FTAG);
1851 dsl_dir_rele(dd, FTAG);
1852 return (SET_ERROR(EXDEV));
1855 /* new name should not already exist */
1856 if (mynewname == NULL) {
1857 dsl_dir_rele(newparent, FTAG);
1858 dsl_dir_rele(dd, FTAG);
1859 return (SET_ERROR(EEXIST));
1862 /* if the name length is growing, validate child name lengths */
1864 error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename,
1865 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1867 dsl_dir_rele(newparent, FTAG);
1868 dsl_dir_rele(dd, FTAG);
1873 if (dmu_tx_is_syncing(tx)) {
1874 if (spa_feature_is_active(dp->dp_spa,
1875 SPA_FEATURE_FS_SS_LIMIT)) {
1877 * Although this is the check function and we don't
1878 * normally make on-disk changes in check functions,
1879 * we need to do that here.
1881 * Ensure this portion of the tree's counts have been
1882 * initialized in case the new parent has limits set.
1884 dsl_dir_init_fs_ss_count(dd, tx);
1888 if (newparent != dd->dd_parent) {
1889 /* is there enough space? */
1891 MAX(dsl_dir_phys(dd)->dd_used_bytes,
1892 dsl_dir_phys(dd)->dd_reserved);
1893 objset_t *os = dd->dd_pool->dp_meta_objset;
1894 uint64_t fs_cnt = 0;
1895 uint64_t ss_cnt = 0;
1897 if (dsl_dir_is_zapified(dd)) {
1900 err = zap_lookup(os, dd->dd_object,
1901 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1903 if (err != ENOENT && err != 0) {
1904 dsl_dir_rele(newparent, FTAG);
1905 dsl_dir_rele(dd, FTAG);
1910 * have to add 1 for the filesystem itself that we're
1915 err = zap_lookup(os, dd->dd_object,
1916 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1918 if (err != ENOENT && err != 0) {
1919 dsl_dir_rele(newparent, FTAG);
1920 dsl_dir_rele(dd, FTAG);
1925 /* no rename into our descendant */
1926 if (closest_common_ancestor(dd, newparent) == dd) {
1927 dsl_dir_rele(newparent, FTAG);
1928 dsl_dir_rele(dd, FTAG);
1929 return (SET_ERROR(EINVAL));
1932 error = dsl_dir_transfer_possible(dd->dd_parent,
1933 newparent, fs_cnt, ss_cnt, myspace, ddra->ddra_cred);
1935 dsl_dir_rele(newparent, FTAG);
1936 dsl_dir_rele(dd, FTAG);
1941 dsl_dir_rele(newparent, FTAG);
1942 dsl_dir_rele(dd, FTAG);
1947 dsl_dir_rename_sync(void *arg, dmu_tx_t *tx)
1949 dsl_dir_rename_arg_t *ddra = arg;
1950 dsl_pool_t *dp = dmu_tx_pool(tx);
1951 dsl_dir_t *dd, *newparent;
1952 const char *mynewname;
1954 objset_t *mos = dp->dp_meta_objset;
1956 VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL));
1957 VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
1960 /* Log this before we change the name. */
1961 spa_history_log_internal_dd(dd, "rename", tx,
1962 "-> %s", ddra->ddra_newname);
1964 if (newparent != dd->dd_parent) {
1965 objset_t *os = dd->dd_pool->dp_meta_objset;
1966 uint64_t fs_cnt = 0;
1967 uint64_t ss_cnt = 0;
1970 * We already made sure the dd counts were initialized in the
1973 if (spa_feature_is_active(dp->dp_spa,
1974 SPA_FEATURE_FS_SS_LIMIT)) {
1975 VERIFY0(zap_lookup(os, dd->dd_object,
1976 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1978 /* add 1 for the filesystem itself that we're moving */
1981 VERIFY0(zap_lookup(os, dd->dd_object,
1982 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1986 dsl_fs_ss_count_adjust(dd->dd_parent, -fs_cnt,
1987 DD_FIELD_FILESYSTEM_COUNT, tx);
1988 dsl_fs_ss_count_adjust(newparent, fs_cnt,
1989 DD_FIELD_FILESYSTEM_COUNT, tx);
1991 dsl_fs_ss_count_adjust(dd->dd_parent, -ss_cnt,
1992 DD_FIELD_SNAPSHOT_COUNT, tx);
1993 dsl_fs_ss_count_adjust(newparent, ss_cnt,
1994 DD_FIELD_SNAPSHOT_COUNT, tx);
1996 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1997 -dsl_dir_phys(dd)->dd_used_bytes,
1998 -dsl_dir_phys(dd)->dd_compressed_bytes,
1999 -dsl_dir_phys(dd)->dd_uncompressed_bytes, tx);
2000 dsl_dir_diduse_space(newparent, DD_USED_CHILD,
2001 dsl_dir_phys(dd)->dd_used_bytes,
2002 dsl_dir_phys(dd)->dd_compressed_bytes,
2003 dsl_dir_phys(dd)->dd_uncompressed_bytes, tx);
2005 if (dsl_dir_phys(dd)->dd_reserved >
2006 dsl_dir_phys(dd)->dd_used_bytes) {
2007 uint64_t unused_rsrv = dsl_dir_phys(dd)->dd_reserved -
2008 dsl_dir_phys(dd)->dd_used_bytes;
2010 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
2011 -unused_rsrv, 0, 0, tx);
2012 dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV,
2013 unused_rsrv, 0, 0, tx);
2017 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2019 /* remove from old parent zapobj */
2020 error = zap_remove(mos,
2021 dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj,
2025 (void) strcpy(dd->dd_myname, mynewname);
2026 dsl_dir_rele(dd->dd_parent, dd);
2027 dsl_dir_phys(dd)->dd_parent_obj = newparent->dd_object;
2028 VERIFY0(dsl_dir_hold_obj(dp,
2029 newparent->dd_object, NULL, dd, &dd->dd_parent));
2031 /* add to new parent zapobj */
2032 VERIFY0(zap_add(mos, dsl_dir_phys(newparent)->dd_child_dir_zapobj,
2033 dd->dd_myname, 8, 1, &dd->dd_object, tx));
2037 zfsvfs_update_fromname(ddra->ddra_oldname, ddra->ddra_newname);
2038 zvol_rename_minors(ddra->ddra_oldname, ddra->ddra_newname);
2042 dsl_prop_notify_all(dd);
2044 dsl_dir_rele(newparent, FTAG);
2045 dsl_dir_rele(dd, FTAG);
2049 dsl_dir_rename(const char *oldname, const char *newname)
2051 dsl_dir_rename_arg_t ddra;
2053 ddra.ddra_oldname = oldname;
2054 ddra.ddra_newname = newname;
2055 ddra.ddra_cred = CRED();
2057 return (dsl_sync_task(oldname,
2058 dsl_dir_rename_check, dsl_dir_rename_sync, &ddra,
2059 3, ZFS_SPACE_CHECK_RESERVED));
2063 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd,
2064 uint64_t fs_cnt, uint64_t ss_cnt, uint64_t space, cred_t *cr)
2066 dsl_dir_t *ancestor;
2071 ancestor = closest_common_ancestor(sdd, tdd);
2072 adelta = would_change(sdd, -space, ancestor);
2073 avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
2075 return (SET_ERROR(ENOSPC));
2077 err = dsl_fs_ss_limit_check(tdd, fs_cnt, ZFS_PROP_FILESYSTEM_LIMIT,
2081 err = dsl_fs_ss_limit_check(tdd, ss_cnt, ZFS_PROP_SNAPSHOT_LIMIT,
2090 dsl_dir_snap_cmtime(dsl_dir_t *dd)
2094 mutex_enter(&dd->dd_lock);
2095 t = dd->dd_snap_cmtime;
2096 mutex_exit(&dd->dd_lock);
2102 dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
2107 mutex_enter(&dd->dd_lock);
2108 dd->dd_snap_cmtime = t;
2109 mutex_exit(&dd->dd_lock);
2113 dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx)
2115 objset_t *mos = dd->dd_pool->dp_meta_objset;
2116 dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx);
2120 dsl_dir_is_zapified(dsl_dir_t *dd)
2122 dmu_object_info_t doi;
2124 dmu_object_info_from_db(dd->dd_dbuf, &doi);
2125 return (doi.doi_type == DMU_OTN_ZAP_METADATA);