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, 2017 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 VERIFY0(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 VERIFY0(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
938 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
940 VERIFY0(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_compressed(dsl_dir_t *dd)
981 return (dsl_dir_phys(dd)->dd_compressed_bytes);
985 dsl_dir_get_quota(dsl_dir_t *dd)
987 return (dsl_dir_phys(dd)->dd_quota);
991 dsl_dir_get_reservation(dsl_dir_t *dd)
993 return (dsl_dir_phys(dd)->dd_reserved);
997 dsl_dir_get_compressratio(dsl_dir_t *dd)
999 /* a fixed point number, 100x the ratio */
1000 return (dsl_dir_phys(dd)->dd_compressed_bytes == 0 ? 100 :
1001 (dsl_dir_phys(dd)->dd_uncompressed_bytes * 100 /
1002 dsl_dir_phys(dd)->dd_compressed_bytes));
1006 dsl_dir_get_logicalused(dsl_dir_t *dd)
1008 return (dsl_dir_phys(dd)->dd_uncompressed_bytes);
1012 dsl_dir_get_usedsnap(dsl_dir_t *dd)
1014 return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]);
1018 dsl_dir_get_usedds(dsl_dir_t *dd)
1020 return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_HEAD]);
1024 dsl_dir_get_usedrefreserv(dsl_dir_t *dd)
1026 return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_REFRSRV]);
1030 dsl_dir_get_usedchild(dsl_dir_t *dd)
1032 return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD] +
1033 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD_RSRV]);
1037 dsl_dir_get_origin(dsl_dir_t *dd, char *buf)
1040 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
1041 dsl_dir_phys(dd)->dd_origin_obj, FTAG, &ds));
1043 dsl_dataset_name(ds, buf);
1045 dsl_dataset_rele(ds, FTAG);
1049 dsl_dir_get_filesystem_count(dsl_dir_t *dd, uint64_t *count)
1051 if (dsl_dir_is_zapified(dd)) {
1052 objset_t *os = dd->dd_pool->dp_meta_objset;
1053 return (zap_lookup(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
1054 sizeof (*count), 1, count));
1061 dsl_dir_get_snapshot_count(dsl_dir_t *dd, uint64_t *count)
1063 if (dsl_dir_is_zapified(dd)) {
1064 objset_t *os = dd->dd_pool->dp_meta_objset;
1065 return (zap_lookup(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
1066 sizeof (*count), 1, count));
1073 dsl_dir_get_remaptxg(dsl_dir_t *dd, uint64_t *count)
1075 if (dsl_dir_is_zapified(dd)) {
1076 objset_t *os = dd->dd_pool->dp_meta_objset;
1077 return (zap_lookup(os, dd->dd_object, DD_FIELD_LAST_REMAP_TXG,
1078 sizeof (*count), 1, count));
1085 dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
1087 mutex_enter(&dd->dd_lock);
1088 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA,
1089 dsl_dir_get_quota(dd));
1090 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
1091 dsl_dir_get_reservation(dd));
1092 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED,
1093 dsl_dir_get_logicalused(dd));
1094 if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1095 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
1096 dsl_dir_get_usedsnap(dd));
1097 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
1098 dsl_dir_get_usedds(dd));
1099 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
1100 dsl_dir_get_usedrefreserv(dd));
1101 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
1102 dsl_dir_get_usedchild(dd));
1104 mutex_exit(&dd->dd_lock);
1107 if (dsl_dir_get_filesystem_count(dd, &count) == 0) {
1108 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_FILESYSTEM_COUNT,
1111 if (dsl_dir_get_snapshot_count(dd, &count) == 0) {
1112 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_SNAPSHOT_COUNT,
1115 if (dsl_dir_get_remaptxg(dd, &count) == 0) {
1116 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REMAPTXG,
1120 if (dsl_dir_is_clone(dd)) {
1121 char buf[ZFS_MAX_DATASET_NAME_LEN];
1122 dsl_dir_get_origin(dd, buf);
1123 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
1129 dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
1131 dsl_pool_t *dp = dd->dd_pool;
1133 ASSERT(dsl_dir_phys(dd));
1135 if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) {
1136 /* up the hold count until we can be written out */
1137 dmu_buf_add_ref(dd->dd_dbuf, dd);
1142 parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
1144 uint64_t old_accounted = MAX(used, dsl_dir_phys(dd)->dd_reserved);
1145 uint64_t new_accounted =
1146 MAX(used + delta, dsl_dir_phys(dd)->dd_reserved);
1147 return (new_accounted - old_accounted);
1151 dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
1153 ASSERT(dmu_tx_is_syncing(tx));
1155 mutex_enter(&dd->dd_lock);
1156 ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]);
1157 dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
1158 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
1159 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
1160 mutex_exit(&dd->dd_lock);
1162 /* release the hold from dsl_dir_dirty */
1163 dmu_buf_rele(dd->dd_dbuf, dd);
1167 dsl_dir_space_towrite(dsl_dir_t *dd)
1171 ASSERT(MUTEX_HELD(&dd->dd_lock));
1173 for (int i = 0; i < TXG_SIZE; i++) {
1174 space += dd->dd_space_towrite[i & TXG_MASK];
1175 ASSERT3U(dd->dd_space_towrite[i & TXG_MASK], >=, 0);
1181 * How much space would dd have available if ancestor had delta applied
1182 * to it? If ondiskonly is set, we're only interested in what's
1183 * on-disk, not estimated pending changes.
1186 dsl_dir_space_available(dsl_dir_t *dd,
1187 dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
1189 uint64_t parentspace, myspace, quota, used;
1192 * If there are no restrictions otherwise, assume we have
1193 * unlimited space available.
1196 parentspace = UINT64_MAX;
1198 if (dd->dd_parent != NULL) {
1199 parentspace = dsl_dir_space_available(dd->dd_parent,
1200 ancestor, delta, ondiskonly);
1203 mutex_enter(&dd->dd_lock);
1204 if (dsl_dir_phys(dd)->dd_quota != 0)
1205 quota = dsl_dir_phys(dd)->dd_quota;
1206 used = dsl_dir_phys(dd)->dd_used_bytes;
1208 used += dsl_dir_space_towrite(dd);
1210 if (dd->dd_parent == NULL) {
1211 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool,
1212 ZFS_SPACE_CHECK_NORMAL);
1213 quota = MIN(quota, poolsize);
1216 if (dsl_dir_phys(dd)->dd_reserved > used && parentspace != UINT64_MAX) {
1218 * We have some space reserved, in addition to what our
1221 parentspace += dsl_dir_phys(dd)->dd_reserved - used;
1224 if (dd == ancestor) {
1226 ASSERT(used >= -delta);
1228 if (parentspace != UINT64_MAX)
1229 parentspace -= delta;
1237 * the lesser of the space provided by our parent and
1238 * the space left in our quota
1240 myspace = MIN(parentspace, quota - used);
1243 mutex_exit(&dd->dd_lock);
1248 struct tempreserve {
1249 list_node_t tr_node;
1255 dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
1256 boolean_t ignorequota, list_t *tr_list,
1257 dmu_tx_t *tx, boolean_t first)
1259 uint64_t txg = tx->tx_txg;
1261 struct tempreserve *tr;
1262 int retval = EDQUOT;
1263 uint64_t ref_rsrv = 0;
1265 ASSERT3U(txg, !=, 0);
1266 ASSERT3S(asize, >, 0);
1268 mutex_enter(&dd->dd_lock);
1271 * Check against the dsl_dir's quota. We don't add in the delta
1272 * when checking for over-quota because they get one free hit.
1274 uint64_t est_inflight = dsl_dir_space_towrite(dd);
1275 for (int i = 0; i < TXG_SIZE; i++)
1276 est_inflight += dd->dd_tempreserved[i];
1277 uint64_t used_on_disk = dsl_dir_phys(dd)->dd_used_bytes;
1280 * On the first iteration, fetch the dataset's used-on-disk and
1281 * refreservation values. Also, if checkrefquota is set, test if
1282 * allocating this space would exceed the dataset's refquota.
1284 if (first && tx->tx_objset) {
1286 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
1288 error = dsl_dataset_check_quota(ds, !netfree,
1289 asize, est_inflight, &used_on_disk, &ref_rsrv);
1291 mutex_exit(&dd->dd_lock);
1297 * If this transaction will result in a net free of space,
1298 * we want to let it through.
1300 if (ignorequota || netfree || dsl_dir_phys(dd)->dd_quota == 0)
1303 quota = dsl_dir_phys(dd)->dd_quota;
1306 * Adjust the quota against the actual pool size at the root
1307 * minus any outstanding deferred frees.
1308 * To ensure that it's possible to remove files from a full
1309 * pool without inducing transient overcommits, we throttle
1310 * netfree transactions against a quota that is slightly larger,
1311 * but still within the pool's allocation slop. In cases where
1312 * we're very close to full, this will allow a steady trickle of
1313 * removes to get through.
1315 uint64_t deferred = 0;
1316 if (dd->dd_parent == NULL) {
1317 uint64_t avail = dsl_pool_unreserved_space(dd->dd_pool,
1319 ZFS_SPACE_CHECK_RESERVED : ZFS_SPACE_CHECK_NORMAL);
1321 if (avail < quota) {
1328 * If they are requesting more space, and our current estimate
1329 * is over quota, they get to try again unless the actual
1330 * on-disk is over quota and there are no pending changes (which
1331 * may free up space for us).
1333 if (used_on_disk + est_inflight >= quota) {
1334 if (est_inflight > 0 || used_on_disk < quota ||
1335 (retval == ENOSPC && used_on_disk < quota + deferred))
1337 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
1338 "quota=%lluK tr=%lluK err=%d\n",
1339 used_on_disk>>10, est_inflight>>10,
1340 quota>>10, asize>>10, retval);
1341 mutex_exit(&dd->dd_lock);
1342 return (SET_ERROR(retval));
1345 /* We need to up our estimated delta before dropping dd_lock */
1346 dd->dd_tempreserved[txg & TXG_MASK] += asize;
1348 uint64_t parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
1350 mutex_exit(&dd->dd_lock);
1352 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1354 tr->tr_size = asize;
1355 list_insert_tail(tr_list, tr);
1357 /* see if it's OK with our parent */
1358 if (dd->dd_parent != NULL && parent_rsrv != 0) {
1359 boolean_t ismos = (dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
1361 return (dsl_dir_tempreserve_impl(dd->dd_parent,
1362 parent_rsrv, netfree, ismos, tr_list, tx, B_FALSE));
1369 * Reserve space in this dsl_dir, to be used in this tx's txg.
1370 * After the space has been dirtied (and dsl_dir_willuse_space()
1371 * has been called), the reservation should be canceled, using
1372 * dsl_dir_tempreserve_clear().
1375 dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
1376 boolean_t netfree, void **tr_cookiep, dmu_tx_t *tx)
1386 tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
1387 list_create(tr_list, sizeof (struct tempreserve),
1388 offsetof(struct tempreserve, tr_node));
1389 ASSERT3S(asize, >, 0);
1391 err = arc_tempreserve_space(dd->dd_pool->dp_spa, lsize, tx->tx_txg);
1393 struct tempreserve *tr;
1395 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1396 tr->tr_size = lsize;
1397 list_insert_tail(tr_list, tr);
1399 if (err == EAGAIN) {
1401 * If arc_memory_throttle() detected that pageout
1402 * is running and we are low on memory, we delay new
1403 * non-pageout transactions to give pageout an
1406 * It is unfortunate to be delaying while the caller's
1409 txg_delay(dd->dd_pool, tx->tx_txg,
1410 MSEC2NSEC(10), MSEC2NSEC(10));
1411 err = SET_ERROR(ERESTART);
1416 err = dsl_dir_tempreserve_impl(dd, asize, netfree,
1417 B_FALSE, tr_list, tx, B_TRUE);
1421 dsl_dir_tempreserve_clear(tr_list, tx);
1423 *tr_cookiep = tr_list;
1429 * Clear a temporary reservation that we previously made with
1430 * dsl_dir_tempreserve_space().
1433 dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
1435 int txgidx = tx->tx_txg & TXG_MASK;
1436 list_t *tr_list = tr_cookie;
1437 struct tempreserve *tr;
1439 ASSERT3U(tx->tx_txg, !=, 0);
1441 if (tr_cookie == NULL)
1444 while ((tr = list_head(tr_list)) != NULL) {
1446 mutex_enter(&tr->tr_ds->dd_lock);
1447 ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
1449 tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
1450 mutex_exit(&tr->tr_ds->dd_lock);
1452 arc_tempreserve_clear(tr->tr_size);
1454 list_remove(tr_list, tr);
1455 kmem_free(tr, sizeof (struct tempreserve));
1458 kmem_free(tr_list, sizeof (list_t));
1462 * This should be called from open context when we think we're going to write
1463 * or free space, for example when dirtying data. Be conservative; it's okay
1464 * to write less space or free more, but we don't want to write more or free
1465 * less than the amount specified.
1468 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
1470 int64_t parent_space;
1473 mutex_enter(&dd->dd_lock);
1475 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
1477 est_used = dsl_dir_space_towrite(dd) + dsl_dir_phys(dd)->dd_used_bytes;
1478 parent_space = parent_delta(dd, est_used, space);
1479 mutex_exit(&dd->dd_lock);
1481 /* Make sure that we clean up dd_space_to* */
1482 dsl_dir_dirty(dd, tx);
1484 /* XXX this is potentially expensive and unnecessary... */
1485 if (parent_space && dd->dd_parent)
1486 dsl_dir_willuse_space(dd->dd_parent, parent_space, tx);
1489 /* call from syncing context when we actually write/free space for this dd */
1491 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
1492 int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
1494 int64_t accounted_delta;
1497 * dsl_dataset_set_refreservation_sync_impl() calls this with
1498 * dd_lock held, so that it can atomically update
1499 * ds->ds_reserved and the dsl_dir accounting, so that
1500 * dsl_dataset_check_quota() can see dataset and dir accounting
1503 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
1505 ASSERT(dmu_tx_is_syncing(tx));
1506 ASSERT(type < DD_USED_NUM);
1508 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1511 mutex_enter(&dd->dd_lock);
1513 parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, used);
1514 ASSERT(used >= 0 || dsl_dir_phys(dd)->dd_used_bytes >= -used);
1515 ASSERT(compressed >= 0 ||
1516 dsl_dir_phys(dd)->dd_compressed_bytes >= -compressed);
1517 ASSERT(uncompressed >= 0 ||
1518 dsl_dir_phys(dd)->dd_uncompressed_bytes >= -uncompressed);
1519 dsl_dir_phys(dd)->dd_used_bytes += used;
1520 dsl_dir_phys(dd)->dd_uncompressed_bytes += uncompressed;
1521 dsl_dir_phys(dd)->dd_compressed_bytes += compressed;
1523 if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1525 dsl_dir_phys(dd)->dd_used_breakdown[type] >= -used);
1526 dsl_dir_phys(dd)->dd_used_breakdown[type] += used;
1530 for (t = 0; t < DD_USED_NUM; t++)
1531 u += dsl_dir_phys(dd)->dd_used_breakdown[t];
1532 ASSERT3U(u, ==, dsl_dir_phys(dd)->dd_used_bytes);
1536 mutex_exit(&dd->dd_lock);
1538 if (dd->dd_parent != NULL) {
1539 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1540 accounted_delta, compressed, uncompressed, tx);
1541 dsl_dir_transfer_space(dd->dd_parent,
1542 used - accounted_delta,
1543 DD_USED_CHILD_RSRV, DD_USED_CHILD, NULL);
1548 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
1549 dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
1551 ASSERT(tx == NULL || dmu_tx_is_syncing(tx));
1552 ASSERT(oldtype < DD_USED_NUM);
1553 ASSERT(newtype < DD_USED_NUM);
1556 !(dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN))
1560 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1561 mutex_enter(&dd->dd_lock);
1563 dsl_dir_phys(dd)->dd_used_breakdown[oldtype] >= delta :
1564 dsl_dir_phys(dd)->dd_used_breakdown[newtype] >= -delta);
1565 ASSERT(dsl_dir_phys(dd)->dd_used_bytes >= ABS(delta));
1566 dsl_dir_phys(dd)->dd_used_breakdown[oldtype] -= delta;
1567 dsl_dir_phys(dd)->dd_used_breakdown[newtype] += delta;
1568 mutex_exit(&dd->dd_lock);
1571 typedef struct dsl_dir_set_qr_arg {
1572 const char *ddsqra_name;
1573 zprop_source_t ddsqra_source;
1574 uint64_t ddsqra_value;
1575 } dsl_dir_set_qr_arg_t;
1578 dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx)
1580 dsl_dir_set_qr_arg_t *ddsqra = arg;
1581 dsl_pool_t *dp = dmu_tx_pool(tx);
1584 uint64_t towrite, newval;
1586 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1590 error = dsl_prop_predict(ds->ds_dir, "quota",
1591 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1593 dsl_dataset_rele(ds, FTAG);
1598 dsl_dataset_rele(ds, FTAG);
1602 mutex_enter(&ds->ds_dir->dd_lock);
1604 * If we are doing the preliminary check in open context, and
1605 * there are pending changes, then don't fail it, since the
1606 * pending changes could under-estimate the amount of space to be
1609 towrite = dsl_dir_space_towrite(ds->ds_dir);
1610 if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
1611 (newval < dsl_dir_phys(ds->ds_dir)->dd_reserved ||
1612 newval < dsl_dir_phys(ds->ds_dir)->dd_used_bytes + towrite)) {
1613 error = SET_ERROR(ENOSPC);
1615 mutex_exit(&ds->ds_dir->dd_lock);
1616 dsl_dataset_rele(ds, FTAG);
1621 dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx)
1623 dsl_dir_set_qr_arg_t *ddsqra = arg;
1624 dsl_pool_t *dp = dmu_tx_pool(tx);
1628 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1630 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1631 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA),
1632 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1633 &ddsqra->ddsqra_value, tx);
1635 VERIFY0(dsl_prop_get_int_ds(ds,
1636 zfs_prop_to_name(ZFS_PROP_QUOTA), &newval));
1638 newval = ddsqra->ddsqra_value;
1639 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1640 zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval);
1643 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1644 mutex_enter(&ds->ds_dir->dd_lock);
1645 dsl_dir_phys(ds->ds_dir)->dd_quota = newval;
1646 mutex_exit(&ds->ds_dir->dd_lock);
1647 dsl_dataset_rele(ds, FTAG);
1651 dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
1653 dsl_dir_set_qr_arg_t ddsqra;
1655 ddsqra.ddsqra_name = ddname;
1656 ddsqra.ddsqra_source = source;
1657 ddsqra.ddsqra_value = quota;
1659 return (dsl_sync_task(ddname, dsl_dir_set_quota_check,
1660 dsl_dir_set_quota_sync, &ddsqra, 0,
1661 ZFS_SPACE_CHECK_EXTRA_RESERVED));
1665 dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx)
1667 dsl_dir_set_qr_arg_t *ddsqra = arg;
1668 dsl_pool_t *dp = dmu_tx_pool(tx);
1671 uint64_t newval, used, avail;
1674 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1680 * If we are doing the preliminary check in open context, the
1681 * space estimates may be inaccurate.
1683 if (!dmu_tx_is_syncing(tx)) {
1684 dsl_dataset_rele(ds, FTAG);
1688 error = dsl_prop_predict(ds->ds_dir,
1689 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1690 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1692 dsl_dataset_rele(ds, FTAG);
1696 mutex_enter(&dd->dd_lock);
1697 used = dsl_dir_phys(dd)->dd_used_bytes;
1698 mutex_exit(&dd->dd_lock);
1700 if (dd->dd_parent) {
1701 avail = dsl_dir_space_available(dd->dd_parent,
1704 avail = dsl_pool_adjustedsize(dd->dd_pool,
1705 ZFS_SPACE_CHECK_NORMAL) - used;
1708 if (MAX(used, newval) > MAX(used, dsl_dir_phys(dd)->dd_reserved)) {
1709 uint64_t delta = MAX(used, newval) -
1710 MAX(used, dsl_dir_phys(dd)->dd_reserved);
1712 if (delta > avail ||
1713 (dsl_dir_phys(dd)->dd_quota > 0 &&
1714 newval > dsl_dir_phys(dd)->dd_quota))
1715 error = SET_ERROR(ENOSPC);
1718 dsl_dataset_rele(ds, FTAG);
1723 dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
1728 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1730 mutex_enter(&dd->dd_lock);
1731 used = dsl_dir_phys(dd)->dd_used_bytes;
1732 delta = MAX(used, value) - MAX(used, dsl_dir_phys(dd)->dd_reserved);
1733 dsl_dir_phys(dd)->dd_reserved = value;
1735 if (dd->dd_parent != NULL) {
1736 /* Roll up this additional usage into our ancestors */
1737 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1740 mutex_exit(&dd->dd_lock);
1744 dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx)
1746 dsl_dir_set_qr_arg_t *ddsqra = arg;
1747 dsl_pool_t *dp = dmu_tx_pool(tx);
1751 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1753 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1754 dsl_prop_set_sync_impl(ds,
1755 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1756 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1757 &ddsqra->ddsqra_value, tx);
1759 VERIFY0(dsl_prop_get_int_ds(ds,
1760 zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval));
1762 newval = ddsqra->ddsqra_value;
1763 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1764 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1765 (longlong_t)newval);
1768 dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx);
1769 dsl_dataset_rele(ds, FTAG);
1773 dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1774 uint64_t reservation)
1776 dsl_dir_set_qr_arg_t ddsqra;
1778 ddsqra.ddsqra_name = ddname;
1779 ddsqra.ddsqra_source = source;
1780 ddsqra.ddsqra_value = reservation;
1782 return (dsl_sync_task(ddname, dsl_dir_set_reservation_check,
1783 dsl_dir_set_reservation_sync, &ddsqra, 0,
1784 ZFS_SPACE_CHECK_EXTRA_RESERVED));
1788 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1790 for (; ds1; ds1 = ds1->dd_parent) {
1792 for (dd = ds2; dd; dd = dd->dd_parent) {
1801 * If delta is applied to dd, how much of that delta would be applied to
1802 * ancestor? Syncing context only.
1805 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1810 mutex_enter(&dd->dd_lock);
1811 delta = parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, delta);
1812 mutex_exit(&dd->dd_lock);
1813 return (would_change(dd->dd_parent, delta, ancestor));
1816 typedef struct dsl_dir_rename_arg {
1817 const char *ddra_oldname;
1818 const char *ddra_newname;
1820 } dsl_dir_rename_arg_t;
1822 typedef struct dsl_valid_rename_arg {
1825 } dsl_valid_rename_arg_t;
1829 dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1831 dsl_valid_rename_arg_t *dvra = arg;
1832 char namebuf[ZFS_MAX_DATASET_NAME_LEN];
1834 dsl_dataset_name(ds, namebuf);
1836 ASSERT3U(strnlen(namebuf, ZFS_MAX_DATASET_NAME_LEN),
1837 <, ZFS_MAX_DATASET_NAME_LEN);
1838 int namelen = strlen(namebuf) + dvra->char_delta;
1839 int depth = get_dataset_depth(namebuf) + dvra->nest_delta;
1841 if (namelen >= ZFS_MAX_DATASET_NAME_LEN)
1842 return (SET_ERROR(ENAMETOOLONG));
1843 if (dvra->nest_delta > 0 && depth >= zfs_max_dataset_nesting)
1844 return (SET_ERROR(ENAMETOOLONG));
1849 dsl_dir_rename_check(void *arg, dmu_tx_t *tx)
1851 dsl_dir_rename_arg_t *ddra = arg;
1852 dsl_pool_t *dp = dmu_tx_pool(tx);
1853 dsl_dir_t *dd, *newparent;
1854 dsl_valid_rename_arg_t dvra;
1855 const char *mynewname;
1858 /* target dir should exist */
1859 error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL);
1863 /* new parent should exist */
1864 error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG,
1865 &newparent, &mynewname);
1867 dsl_dir_rele(dd, FTAG);
1871 /* can't rename to different pool */
1872 if (dd->dd_pool != newparent->dd_pool) {
1873 dsl_dir_rele(newparent, FTAG);
1874 dsl_dir_rele(dd, FTAG);
1875 return (SET_ERROR(EXDEV));
1878 /* new name should not already exist */
1879 if (mynewname == NULL) {
1880 dsl_dir_rele(newparent, FTAG);
1881 dsl_dir_rele(dd, FTAG);
1882 return (SET_ERROR(EEXIST));
1885 ASSERT3U(strnlen(ddra->ddra_newname, ZFS_MAX_DATASET_NAME_LEN),
1886 <, ZFS_MAX_DATASET_NAME_LEN);
1887 ASSERT3U(strnlen(ddra->ddra_oldname, ZFS_MAX_DATASET_NAME_LEN),
1888 <, ZFS_MAX_DATASET_NAME_LEN);
1889 dvra.char_delta = strlen(ddra->ddra_newname)
1890 - strlen(ddra->ddra_oldname);
1891 dvra.nest_delta = get_dataset_depth(ddra->ddra_newname)
1892 - get_dataset_depth(ddra->ddra_oldname);
1894 /* if the name length is growing, validate child name lengths */
1895 if (dvra.char_delta > 0 || dvra.nest_delta > 0) {
1896 error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename,
1897 &dvra, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1899 dsl_dir_rele(newparent, FTAG);
1900 dsl_dir_rele(dd, FTAG);
1905 if (dmu_tx_is_syncing(tx)) {
1906 if (spa_feature_is_active(dp->dp_spa,
1907 SPA_FEATURE_FS_SS_LIMIT)) {
1909 * Although this is the check function and we don't
1910 * normally make on-disk changes in check functions,
1911 * we need to do that here.
1913 * Ensure this portion of the tree's counts have been
1914 * initialized in case the new parent has limits set.
1916 dsl_dir_init_fs_ss_count(dd, tx);
1920 if (newparent != dd->dd_parent) {
1921 /* is there enough space? */
1923 MAX(dsl_dir_phys(dd)->dd_used_bytes,
1924 dsl_dir_phys(dd)->dd_reserved);
1925 objset_t *os = dd->dd_pool->dp_meta_objset;
1926 uint64_t fs_cnt = 0;
1927 uint64_t ss_cnt = 0;
1929 if (dsl_dir_is_zapified(dd)) {
1932 err = zap_lookup(os, dd->dd_object,
1933 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1935 if (err != ENOENT && err != 0) {
1936 dsl_dir_rele(newparent, FTAG);
1937 dsl_dir_rele(dd, FTAG);
1942 * have to add 1 for the filesystem itself that we're
1947 err = zap_lookup(os, dd->dd_object,
1948 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1950 if (err != ENOENT && err != 0) {
1951 dsl_dir_rele(newparent, FTAG);
1952 dsl_dir_rele(dd, FTAG);
1957 /* no rename into our descendant */
1958 if (closest_common_ancestor(dd, newparent) == dd) {
1959 dsl_dir_rele(newparent, FTAG);
1960 dsl_dir_rele(dd, FTAG);
1961 return (SET_ERROR(EINVAL));
1964 error = dsl_dir_transfer_possible(dd->dd_parent,
1965 newparent, fs_cnt, ss_cnt, myspace, ddra->ddra_cred);
1967 dsl_dir_rele(newparent, FTAG);
1968 dsl_dir_rele(dd, FTAG);
1973 dsl_dir_rele(newparent, FTAG);
1974 dsl_dir_rele(dd, FTAG);
1979 dsl_dir_rename_sync(void *arg, dmu_tx_t *tx)
1981 dsl_dir_rename_arg_t *ddra = arg;
1982 dsl_pool_t *dp = dmu_tx_pool(tx);
1983 dsl_dir_t *dd, *newparent;
1984 const char *mynewname;
1986 objset_t *mos = dp->dp_meta_objset;
1988 VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL));
1989 VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
1992 /* Log this before we change the name. */
1993 spa_history_log_internal_dd(dd, "rename", tx,
1994 "-> %s", ddra->ddra_newname);
1996 if (newparent != dd->dd_parent) {
1997 objset_t *os = dd->dd_pool->dp_meta_objset;
1998 uint64_t fs_cnt = 0;
1999 uint64_t ss_cnt = 0;
2002 * We already made sure the dd counts were initialized in the
2005 if (spa_feature_is_active(dp->dp_spa,
2006 SPA_FEATURE_FS_SS_LIMIT)) {
2007 VERIFY0(zap_lookup(os, dd->dd_object,
2008 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
2010 /* add 1 for the filesystem itself that we're moving */
2013 VERIFY0(zap_lookup(os, dd->dd_object,
2014 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
2018 dsl_fs_ss_count_adjust(dd->dd_parent, -fs_cnt,
2019 DD_FIELD_FILESYSTEM_COUNT, tx);
2020 dsl_fs_ss_count_adjust(newparent, fs_cnt,
2021 DD_FIELD_FILESYSTEM_COUNT, tx);
2023 dsl_fs_ss_count_adjust(dd->dd_parent, -ss_cnt,
2024 DD_FIELD_SNAPSHOT_COUNT, tx);
2025 dsl_fs_ss_count_adjust(newparent, ss_cnt,
2026 DD_FIELD_SNAPSHOT_COUNT, tx);
2028 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
2029 -dsl_dir_phys(dd)->dd_used_bytes,
2030 -dsl_dir_phys(dd)->dd_compressed_bytes,
2031 -dsl_dir_phys(dd)->dd_uncompressed_bytes, tx);
2032 dsl_dir_diduse_space(newparent, DD_USED_CHILD,
2033 dsl_dir_phys(dd)->dd_used_bytes,
2034 dsl_dir_phys(dd)->dd_compressed_bytes,
2035 dsl_dir_phys(dd)->dd_uncompressed_bytes, tx);
2037 if (dsl_dir_phys(dd)->dd_reserved >
2038 dsl_dir_phys(dd)->dd_used_bytes) {
2039 uint64_t unused_rsrv = dsl_dir_phys(dd)->dd_reserved -
2040 dsl_dir_phys(dd)->dd_used_bytes;
2042 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
2043 -unused_rsrv, 0, 0, tx);
2044 dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV,
2045 unused_rsrv, 0, 0, tx);
2049 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2051 /* remove from old parent zapobj */
2052 error = zap_remove(mos,
2053 dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj,
2057 (void) strcpy(dd->dd_myname, mynewname);
2058 dsl_dir_rele(dd->dd_parent, dd);
2059 dsl_dir_phys(dd)->dd_parent_obj = newparent->dd_object;
2060 VERIFY0(dsl_dir_hold_obj(dp,
2061 newparent->dd_object, NULL, dd, &dd->dd_parent));
2063 /* add to new parent zapobj */
2064 VERIFY0(zap_add(mos, dsl_dir_phys(newparent)->dd_child_dir_zapobj,
2065 dd->dd_myname, 8, 1, &dd->dd_object, tx));
2069 zfsvfs_update_fromname(ddra->ddra_oldname, ddra->ddra_newname);
2070 zvol_rename_minors(ddra->ddra_oldname, ddra->ddra_newname);
2074 dsl_prop_notify_all(dd);
2076 dsl_dir_rele(newparent, FTAG);
2077 dsl_dir_rele(dd, FTAG);
2081 dsl_dir_rename(const char *oldname, const char *newname)
2083 dsl_dir_rename_arg_t ddra;
2085 ddra.ddra_oldname = oldname;
2086 ddra.ddra_newname = newname;
2087 ddra.ddra_cred = CRED();
2089 return (dsl_sync_task(oldname,
2090 dsl_dir_rename_check, dsl_dir_rename_sync, &ddra,
2091 3, ZFS_SPACE_CHECK_RESERVED));
2095 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd,
2096 uint64_t fs_cnt, uint64_t ss_cnt, uint64_t space, cred_t *cr)
2098 dsl_dir_t *ancestor;
2103 ancestor = closest_common_ancestor(sdd, tdd);
2104 adelta = would_change(sdd, -space, ancestor);
2105 avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
2107 return (SET_ERROR(ENOSPC));
2109 err = dsl_fs_ss_limit_check(tdd, fs_cnt, ZFS_PROP_FILESYSTEM_LIMIT,
2113 err = dsl_fs_ss_limit_check(tdd, ss_cnt, ZFS_PROP_SNAPSHOT_LIMIT,
2122 dsl_dir_snap_cmtime(dsl_dir_t *dd)
2126 mutex_enter(&dd->dd_lock);
2127 t = dd->dd_snap_cmtime;
2128 mutex_exit(&dd->dd_lock);
2134 dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
2139 mutex_enter(&dd->dd_lock);
2140 dd->dd_snap_cmtime = t;
2141 mutex_exit(&dd->dd_lock);
2145 dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx)
2147 objset_t *mos = dd->dd_pool->dp_meta_objset;
2148 dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx);
2152 dsl_dir_is_zapified(dsl_dir_t *dd)
2154 dmu_object_info_t doi;
2156 dmu_object_info_from_db(dd->dd_dbuf, &doi);
2157 return (doi.doi_type == DMU_OTN_ZAP_METADATA);