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.
25 #include <sys/dmu_objset.h>
26 #include <sys/dsl_dataset.h>
27 #include <sys/dsl_dir.h>
28 #include <sys/dsl_prop.h>
29 #include <sys/dsl_synctask.h>
30 #include <sys/dmu_traverse.h>
31 #include <sys/dmu_tx.h>
35 #include <sys/unique.h>
36 #include <sys/zfs_context.h>
37 #include <sys/zfs_ioctl.h>
39 #include <sys/zfs_znode.h>
40 #include <sys/zfs_onexit.h>
42 #include <sys/dsl_scan.h>
43 #include <sys/dsl_deadlist.h>
45 static char *dsl_reaper = "the grim reaper";
47 static dsl_checkfunc_t dsl_dataset_destroy_begin_check;
48 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync;
49 static dsl_syncfunc_t dsl_dataset_set_reservation_sync;
51 #define SWITCH64(x, y) \
53 uint64_t __tmp = (x); \
58 #define DS_REF_MAX (1ULL << 62)
60 #define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE
62 #define DSL_DATASET_IS_DESTROYED(ds) ((ds)->ds_owner == dsl_reaper)
66 * Figure out how much of this delta should be propogated to the dsl_dir
67 * layer. If there's a refreservation, that space has already been
68 * partially accounted for in our ancestors.
71 parent_delta(dsl_dataset_t *ds, int64_t delta)
73 uint64_t old_bytes, new_bytes;
75 if (ds->ds_reserved == 0)
78 old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
79 new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
81 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
82 return (new_bytes - old_bytes);
86 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
88 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
89 int compressed = BP_GET_PSIZE(bp);
90 int uncompressed = BP_GET_UCSIZE(bp);
93 dprintf_bp(bp, "ds=%p", ds);
95 ASSERT(dmu_tx_is_syncing(tx));
96 /* It could have been compressed away to nothing */
99 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
100 ASSERT3U(BP_GET_TYPE(bp), <, DMU_OT_NUMTYPES);
103 * Account for the meta-objset space in its placeholder
106 ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */
107 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
108 used, compressed, uncompressed, tx);
109 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
112 dmu_buf_will_dirty(ds->ds_dbuf, tx);
114 mutex_enter(&ds->ds_dir->dd_lock);
115 mutex_enter(&ds->ds_lock);
116 delta = parent_delta(ds, used);
117 ds->ds_phys->ds_used_bytes += used;
118 ds->ds_phys->ds_compressed_bytes += compressed;
119 ds->ds_phys->ds_uncompressed_bytes += uncompressed;
120 ds->ds_phys->ds_unique_bytes += used;
121 mutex_exit(&ds->ds_lock);
122 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
123 compressed, uncompressed, tx);
124 dsl_dir_transfer_space(ds->ds_dir, used - delta,
125 DD_USED_REFRSRV, DD_USED_HEAD, tx);
126 mutex_exit(&ds->ds_dir->dd_lock);
130 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
136 ASSERT(dmu_tx_is_syncing(tx));
137 ASSERT(bp->blk_birth <= tx->tx_txg);
139 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
140 int compressed = BP_GET_PSIZE(bp);
141 int uncompressed = BP_GET_UCSIZE(bp);
146 * Account for the meta-objset space in its placeholder
149 dsl_free(tx->tx_pool, tx->tx_txg, bp);
151 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
152 -used, -compressed, -uncompressed, tx);
153 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
156 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
158 ASSERT(!dsl_dataset_is_snapshot(ds));
159 dmu_buf_will_dirty(ds->ds_dbuf, tx);
161 if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
164 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
165 dsl_free(tx->tx_pool, tx->tx_txg, bp);
167 mutex_enter(&ds->ds_dir->dd_lock);
168 mutex_enter(&ds->ds_lock);
169 ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
170 !DS_UNIQUE_IS_ACCURATE(ds));
171 delta = parent_delta(ds, -used);
172 ds->ds_phys->ds_unique_bytes -= used;
173 mutex_exit(&ds->ds_lock);
174 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
175 delta, -compressed, -uncompressed, tx);
176 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
177 DD_USED_REFRSRV, DD_USED_HEAD, tx);
178 mutex_exit(&ds->ds_dir->dd_lock);
180 dprintf_bp(bp, "putting on dead list: %s", "");
183 * We are here as part of zio's write done callback,
184 * which means we're a zio interrupt thread. We can't
185 * call dsl_deadlist_insert() now because it may block
186 * waiting for I/O. Instead, put bp on the deferred
187 * queue and let dsl_pool_sync() finish the job.
189 bplist_append(&ds->ds_pending_deadlist, bp);
191 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
193 ASSERT3U(ds->ds_prev->ds_object, ==,
194 ds->ds_phys->ds_prev_snap_obj);
195 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
196 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
197 if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
198 ds->ds_object && bp->blk_birth >
199 ds->ds_prev->ds_phys->ds_prev_snap_txg) {
200 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
201 mutex_enter(&ds->ds_prev->ds_lock);
202 ds->ds_prev->ds_phys->ds_unique_bytes += used;
203 mutex_exit(&ds->ds_prev->ds_lock);
205 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
206 dsl_dir_transfer_space(ds->ds_dir, used,
207 DD_USED_HEAD, DD_USED_SNAP, tx);
210 mutex_enter(&ds->ds_lock);
211 ASSERT3U(ds->ds_phys->ds_used_bytes, >=, used);
212 ds->ds_phys->ds_used_bytes -= used;
213 ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
214 ds->ds_phys->ds_compressed_bytes -= compressed;
215 ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
216 ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
217 mutex_exit(&ds->ds_lock);
223 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
225 uint64_t trysnap = 0;
230 * The snapshot creation could fail, but that would cause an
231 * incorrect FALSE return, which would only result in an
232 * overestimation of the amount of space that an operation would
233 * consume, which is OK.
235 * There's also a small window where we could miss a pending
236 * snapshot, because we could set the sync task in the quiescing
237 * phase. So this should only be used as a guess.
239 if (ds->ds_trysnap_txg >
240 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
241 trysnap = ds->ds_trysnap_txg;
242 return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
246 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
249 if (blk_birth <= dsl_dataset_prev_snap_txg(ds))
252 ddt_prefetch(dsl_dataset_get_spa(ds), bp);
259 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
261 dsl_dataset_t *ds = dsv;
263 ASSERT(ds->ds_owner == NULL || DSL_DATASET_IS_DESTROYED(ds));
265 unique_remove(ds->ds_fsid_guid);
267 if (ds->ds_objset != NULL)
268 dmu_objset_evict(ds->ds_objset);
271 dsl_dataset_drop_ref(ds->ds_prev, ds);
275 bplist_destroy(&ds->ds_pending_deadlist);
277 dsl_deadlist_close(&ds->ds_deadlist);
279 ASSERT(ds->ds_deadlist.dl_dbuf == NULL);
280 ASSERT(!ds->ds_deadlist.dl_oldfmt);
283 dsl_dir_close(ds->ds_dir, ds);
285 ASSERT(!list_link_active(&ds->ds_synced_link));
287 if (mutex_owned(&ds->ds_lock))
288 mutex_exit(&ds->ds_lock);
289 mutex_destroy(&ds->ds_lock);
290 mutex_destroy(&ds->ds_recvlock);
291 if (mutex_owned(&ds->ds_opening_lock))
292 mutex_exit(&ds->ds_opening_lock);
293 mutex_destroy(&ds->ds_opening_lock);
294 rw_destroy(&ds->ds_rwlock);
295 cv_destroy(&ds->ds_exclusive_cv);
297 kmem_free(ds, sizeof (dsl_dataset_t));
301 dsl_dataset_get_snapname(dsl_dataset_t *ds)
303 dsl_dataset_phys_t *headphys;
306 dsl_pool_t *dp = ds->ds_dir->dd_pool;
307 objset_t *mos = dp->dp_meta_objset;
309 if (ds->ds_snapname[0])
311 if (ds->ds_phys->ds_next_snap_obj == 0)
314 err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
318 headphys = headdbuf->db_data;
319 err = zap_value_search(dp->dp_meta_objset,
320 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
321 dmu_buf_rele(headdbuf, FTAG);
326 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
328 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
329 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
333 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
338 err = zap_lookup_norm(mos, snapobj, name, 8, 1,
339 value, mt, NULL, 0, NULL);
340 if (err == ENOTSUP && mt == MT_FIRST)
341 err = zap_lookup(mos, snapobj, name, 8, 1, value);
346 dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx)
348 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
349 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
353 dsl_dir_snap_cmtime_update(ds->ds_dir);
355 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
360 err = zap_remove_norm(mos, snapobj, name, mt, tx);
361 if (err == ENOTSUP && mt == MT_FIRST)
362 err = zap_remove(mos, snapobj, name, tx);
367 dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag,
370 objset_t *mos = dp->dp_meta_objset;
374 dmu_object_info_t doi;
376 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
377 dsl_pool_sync_context(dp));
379 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
383 /* Make sure dsobj has the correct object type. */
384 dmu_object_info_from_db(dbuf, &doi);
385 if (doi.doi_type != DMU_OT_DSL_DATASET)
388 ds = dmu_buf_get_user(dbuf);
390 dsl_dataset_t *winner;
392 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
394 ds->ds_object = dsobj;
395 ds->ds_phys = dbuf->db_data;
397 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
398 mutex_init(&ds->ds_recvlock, NULL, MUTEX_DEFAULT, NULL);
399 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
400 rw_init(&ds->ds_rwlock, 0, 0, 0);
401 cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL);
403 bplist_create(&ds->ds_pending_deadlist);
404 dsl_deadlist_open(&ds->ds_deadlist,
405 mos, ds->ds_phys->ds_deadlist_obj);
408 err = dsl_dir_open_obj(dp,
409 ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
412 mutex_destroy(&ds->ds_lock);
413 mutex_destroy(&ds->ds_recvlock);
414 mutex_destroy(&ds->ds_opening_lock);
415 rw_destroy(&ds->ds_rwlock);
416 cv_destroy(&ds->ds_exclusive_cv);
417 bplist_destroy(&ds->ds_pending_deadlist);
418 dsl_deadlist_close(&ds->ds_deadlist);
419 kmem_free(ds, sizeof (dsl_dataset_t));
420 dmu_buf_rele(dbuf, tag);
424 if (!dsl_dataset_is_snapshot(ds)) {
425 ds->ds_snapname[0] = '\0';
426 if (ds->ds_phys->ds_prev_snap_obj) {
427 err = dsl_dataset_get_ref(dp,
428 ds->ds_phys->ds_prev_snap_obj,
432 if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
433 err = dsl_dataset_get_snapname(ds);
434 if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) {
436 ds->ds_dir->dd_pool->dp_meta_objset,
437 ds->ds_phys->ds_userrefs_obj,
442 if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
444 * In sync context, we're called with either no lock
445 * or with the write lock. If we're not syncing,
446 * we're always called with the read lock held.
448 boolean_t need_lock =
449 !RW_WRITE_HELD(&dp->dp_config_rwlock) &&
450 dsl_pool_sync_context(dp);
453 rw_enter(&dp->dp_config_rwlock, RW_READER);
455 err = dsl_prop_get_ds(ds,
456 "refreservation", sizeof (uint64_t), 1,
457 &ds->ds_reserved, NULL);
459 err = dsl_prop_get_ds(ds,
460 "refquota", sizeof (uint64_t), 1,
461 &ds->ds_quota, NULL);
465 rw_exit(&dp->dp_config_rwlock);
467 ds->ds_reserved = ds->ds_quota = 0;
471 winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys,
475 bplist_destroy(&ds->ds_pending_deadlist);
476 dsl_deadlist_close(&ds->ds_deadlist);
478 dsl_dataset_drop_ref(ds->ds_prev, ds);
479 dsl_dir_close(ds->ds_dir, ds);
480 mutex_destroy(&ds->ds_lock);
481 mutex_destroy(&ds->ds_recvlock);
482 mutex_destroy(&ds->ds_opening_lock);
483 rw_destroy(&ds->ds_rwlock);
484 cv_destroy(&ds->ds_exclusive_cv);
485 kmem_free(ds, sizeof (dsl_dataset_t));
487 dmu_buf_rele(dbuf, tag);
493 unique_insert(ds->ds_phys->ds_fsid_guid);
496 ASSERT3P(ds->ds_dbuf, ==, dbuf);
497 ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
498 ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
499 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
500 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
501 mutex_enter(&ds->ds_lock);
502 if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) {
503 mutex_exit(&ds->ds_lock);
504 dmu_buf_rele(ds->ds_dbuf, tag);
507 mutex_exit(&ds->ds_lock);
513 dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag)
515 dsl_pool_t *dp = ds->ds_dir->dd_pool;
518 * In syncing context we don't want the rwlock lock: there
519 * may be an existing writer waiting for sync phase to
520 * finish. We don't need to worry about such writers, since
521 * sync phase is single-threaded, so the writer can't be
522 * doing anything while we are active.
524 if (dsl_pool_sync_context(dp)) {
525 ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
530 * Normal users will hold the ds_rwlock as a READER until they
531 * are finished (i.e., call dsl_dataset_rele()). "Owners" will
532 * drop their READER lock after they set the ds_owner field.
534 * If the dataset is being destroyed, the destroy thread will
535 * obtain a WRITER lock for exclusive access after it's done its
536 * open-context work and then change the ds_owner to
537 * dsl_reaper once destruction is assured. So threads
538 * may block here temporarily, until the "destructability" of
539 * the dataset is determined.
541 ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock));
542 mutex_enter(&ds->ds_lock);
543 while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) {
544 rw_exit(&dp->dp_config_rwlock);
545 cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock);
546 if (DSL_DATASET_IS_DESTROYED(ds)) {
547 mutex_exit(&ds->ds_lock);
548 dsl_dataset_drop_ref(ds, tag);
549 rw_enter(&dp->dp_config_rwlock, RW_READER);
553 * The dp_config_rwlock lives above the ds_lock. And
554 * we need to check DSL_DATASET_IS_DESTROYED() while
555 * holding the ds_lock, so we have to drop and reacquire
558 mutex_exit(&ds->ds_lock);
559 rw_enter(&dp->dp_config_rwlock, RW_READER);
560 mutex_enter(&ds->ds_lock);
562 mutex_exit(&ds->ds_lock);
567 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
570 int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp);
574 return (dsl_dataset_hold_ref(*dsp, tag));
578 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, boolean_t inconsistentok,
579 void *tag, dsl_dataset_t **dsp)
581 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
584 if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
585 dsl_dataset_rele(*dsp, tag);
593 dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp)
597 const char *snapname;
601 err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname);
606 obj = dd->dd_phys->dd_head_dataset_obj;
607 rw_enter(&dp->dp_config_rwlock, RW_READER);
609 err = dsl_dataset_get_ref(dp, obj, tag, dsp);
615 err = dsl_dataset_hold_ref(*dsp, tag);
617 /* we may be looking for a snapshot */
618 if (err == 0 && snapname != NULL) {
619 dsl_dataset_t *ds = NULL;
621 if (*snapname++ != '@') {
622 dsl_dataset_rele(*dsp, tag);
627 dprintf("looking for snapshot '%s'\n", snapname);
628 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
630 err = dsl_dataset_get_ref(dp, obj, tag, &ds);
631 dsl_dataset_rele(*dsp, tag);
633 ASSERT3U((err == 0), ==, (ds != NULL));
636 mutex_enter(&ds->ds_lock);
637 if (ds->ds_snapname[0] == 0)
638 (void) strlcpy(ds->ds_snapname, snapname,
639 sizeof (ds->ds_snapname));
640 mutex_exit(&ds->ds_lock);
641 err = dsl_dataset_hold_ref(ds, tag);
642 *dsp = err ? NULL : ds;
646 rw_exit(&dp->dp_config_rwlock);
647 dsl_dir_close(dd, FTAG);
652 dsl_dataset_own(const char *name, boolean_t inconsistentok,
653 void *tag, dsl_dataset_t **dsp)
655 int err = dsl_dataset_hold(name, tag, dsp);
658 if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
659 dsl_dataset_rele(*dsp, tag);
666 dsl_dataset_name(dsl_dataset_t *ds, char *name)
669 (void) strcpy(name, "mos");
671 dsl_dir_name(ds->ds_dir, name);
672 VERIFY(0 == dsl_dataset_get_snapname(ds));
673 if (ds->ds_snapname[0]) {
674 (void) strcat(name, "@");
676 * We use a "recursive" mutex so that we
677 * can call dprintf_ds() with ds_lock held.
679 if (!MUTEX_HELD(&ds->ds_lock)) {
680 mutex_enter(&ds->ds_lock);
681 (void) strcat(name, ds->ds_snapname);
682 mutex_exit(&ds->ds_lock);
684 (void) strcat(name, ds->ds_snapname);
691 dsl_dataset_namelen(dsl_dataset_t *ds)
696 result = 3; /* "mos" */
698 result = dsl_dir_namelen(ds->ds_dir);
699 VERIFY(0 == dsl_dataset_get_snapname(ds));
700 if (ds->ds_snapname[0]) {
701 ++result; /* adding one for the @-sign */
702 if (!MUTEX_HELD(&ds->ds_lock)) {
703 mutex_enter(&ds->ds_lock);
704 result += strlen(ds->ds_snapname);
705 mutex_exit(&ds->ds_lock);
707 result += strlen(ds->ds_snapname);
716 dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag)
718 dmu_buf_rele(ds->ds_dbuf, tag);
722 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
724 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) {
725 rw_exit(&ds->ds_rwlock);
727 dsl_dataset_drop_ref(ds, tag);
731 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
733 ASSERT((ds->ds_owner == tag && ds->ds_dbuf) ||
734 (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL));
736 mutex_enter(&ds->ds_lock);
738 if (RW_WRITE_HELD(&ds->ds_rwlock)) {
739 rw_exit(&ds->ds_rwlock);
740 cv_broadcast(&ds->ds_exclusive_cv);
742 mutex_exit(&ds->ds_lock);
744 dsl_dataset_drop_ref(ds, tag);
746 dsl_dataset_evict(NULL, ds);
750 dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *tag)
752 boolean_t gotit = FALSE;
754 mutex_enter(&ds->ds_lock);
755 if (ds->ds_owner == NULL &&
756 (!DS_IS_INCONSISTENT(ds) || inconsistentok)) {
758 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool))
759 rw_exit(&ds->ds_rwlock);
762 mutex_exit(&ds->ds_lock);
767 dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner)
769 ASSERT3P(owner, ==, ds->ds_owner);
770 if (!RW_WRITE_HELD(&ds->ds_rwlock))
771 rw_enter(&ds->ds_rwlock, RW_WRITER);
775 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
776 uint64_t flags, dmu_tx_t *tx)
778 dsl_pool_t *dp = dd->dd_pool;
780 dsl_dataset_phys_t *dsphys;
782 objset_t *mos = dp->dp_meta_objset;
785 origin = dp->dp_origin_snap;
787 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
788 ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
789 ASSERT(dmu_tx_is_syncing(tx));
790 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
792 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
793 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
794 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
795 dmu_buf_will_dirty(dbuf, tx);
796 dsphys = dbuf->db_data;
797 bzero(dsphys, sizeof (dsl_dataset_phys_t));
798 dsphys->ds_dir_obj = dd->dd_object;
799 dsphys->ds_flags = flags;
800 dsphys->ds_fsid_guid = unique_create();
801 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
802 sizeof (dsphys->ds_guid));
803 dsphys->ds_snapnames_zapobj =
804 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
806 dsphys->ds_creation_time = gethrestime_sec();
807 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
809 if (origin == NULL) {
810 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
814 dsphys->ds_prev_snap_obj = origin->ds_object;
815 dsphys->ds_prev_snap_txg =
816 origin->ds_phys->ds_creation_txg;
817 dsphys->ds_used_bytes =
818 origin->ds_phys->ds_used_bytes;
819 dsphys->ds_compressed_bytes =
820 origin->ds_phys->ds_compressed_bytes;
821 dsphys->ds_uncompressed_bytes =
822 origin->ds_phys->ds_uncompressed_bytes;
823 dsphys->ds_bp = origin->ds_phys->ds_bp;
824 dsphys->ds_flags |= origin->ds_phys->ds_flags;
826 dmu_buf_will_dirty(origin->ds_dbuf, tx);
827 origin->ds_phys->ds_num_children++;
829 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
830 origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds));
831 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
832 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
833 dsl_dataset_rele(ohds, FTAG);
835 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
836 if (origin->ds_phys->ds_next_clones_obj == 0) {
837 origin->ds_phys->ds_next_clones_obj =
839 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
841 VERIFY(0 == zap_add_int(mos,
842 origin->ds_phys->ds_next_clones_obj,
846 dmu_buf_will_dirty(dd->dd_dbuf, tx);
847 dd->dd_phys->dd_origin_obj = origin->ds_object;
848 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
849 if (origin->ds_dir->dd_phys->dd_clones == 0) {
850 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
851 origin->ds_dir->dd_phys->dd_clones =
853 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
855 VERIFY3U(0, ==, zap_add_int(mos,
856 origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
860 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
861 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
863 dmu_buf_rele(dbuf, FTAG);
865 dmu_buf_will_dirty(dd->dd_dbuf, tx);
866 dd->dd_phys->dd_head_dataset_obj = dsobj;
872 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
873 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
875 dsl_pool_t *dp = pdd->dd_pool;
876 uint64_t dsobj, ddobj;
879 ASSERT(lastname[0] != '@');
881 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
882 VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd));
884 dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx);
886 dsl_deleg_set_create_perms(dd, tx, cr);
888 dsl_dir_close(dd, FTAG);
891 * If we are creating a clone, make sure we zero out any stale
892 * data from the origin snapshots zil header.
894 if (origin != NULL) {
898 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
899 VERIFY3U(0, ==, dmu_objset_from_ds(ds, &os));
900 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
901 dsl_dataset_dirty(ds, tx);
902 dsl_dataset_rele(ds, FTAG);
909 dsl_sync_task_group_t *dstg;
916 dsl_snapshot_destroy_one(const char *name, void *arg)
918 struct destroyarg *da = arg;
923 dsname = kmem_asprintf("%s@%s", name, da->snapname);
924 err = dsl_dataset_own(dsname, B_TRUE, da->dstg, &ds);
927 struct dsl_ds_destroyarg *dsda;
929 dsl_dataset_make_exclusive(ds, da->dstg);
930 dsda = kmem_zalloc(sizeof (struct dsl_ds_destroyarg), KM_SLEEP);
932 dsda->defer = da->defer;
933 dsl_sync_task_create(da->dstg, dsl_dataset_destroy_check,
934 dsl_dataset_destroy_sync, dsda, da->dstg, 0);
935 } else if (err == ENOENT) {
938 (void) strcpy(da->failed, name);
944 * Destroy 'snapname' in all descendants of 'fsname'.
946 #pragma weak dmu_snapshots_destroy = dsl_snapshots_destroy
948 dsl_snapshots_destroy(char *fsname, char *snapname, boolean_t defer)
951 struct destroyarg da;
952 dsl_sync_task_t *dst;
955 err = spa_open(fsname, &spa, FTAG);
958 da.dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
959 da.snapname = snapname;
963 err = dmu_objset_find(fsname,
964 dsl_snapshot_destroy_one, &da, DS_FIND_CHILDREN);
967 err = dsl_sync_task_group_wait(da.dstg);
969 for (dst = list_head(&da.dstg->dstg_tasks); dst;
970 dst = list_next(&da.dstg->dstg_tasks, dst)) {
971 struct dsl_ds_destroyarg *dsda = dst->dst_arg1;
972 dsl_dataset_t *ds = dsda->ds;
975 * Return the file system name that triggered the error
978 dsl_dataset_name(ds, fsname);
979 *strchr(fsname, '@') = '\0';
981 ASSERT3P(dsda->rm_origin, ==, NULL);
982 dsl_dataset_disown(ds, da.dstg);
983 kmem_free(dsda, sizeof (struct dsl_ds_destroyarg));
986 dsl_sync_task_group_destroy(da.dstg);
987 spa_close(spa, FTAG);
992 dsl_dataset_might_destroy_origin(dsl_dataset_t *ds)
994 boolean_t might_destroy = B_FALSE;
996 mutex_enter(&ds->ds_lock);
997 if (ds->ds_phys->ds_num_children == 2 && ds->ds_userrefs == 0 &&
998 DS_IS_DEFER_DESTROY(ds))
999 might_destroy = B_TRUE;
1000 mutex_exit(&ds->ds_lock);
1002 return (might_destroy);
1006 * If we're removing a clone, and these three conditions are true:
1007 * 1) the clone's origin has no other children
1008 * 2) the clone's origin has no user references
1009 * 3) the clone's origin has been marked for deferred destruction
1010 * Then, prepare to remove the origin as part of this sync task group.
1013 dsl_dataset_origin_rm_prep(struct dsl_ds_destroyarg *dsda, void *tag)
1015 dsl_dataset_t *ds = dsda->ds;
1016 dsl_dataset_t *origin = ds->ds_prev;
1018 if (dsl_dataset_might_destroy_origin(origin)) {
1023 namelen = dsl_dataset_namelen(origin) + 1;
1024 name = kmem_alloc(namelen, KM_SLEEP);
1025 dsl_dataset_name(origin, name);
1027 error = zfs_unmount_snap(name, NULL);
1029 kmem_free(name, namelen);
1033 error = dsl_dataset_own(name, B_TRUE, tag, &origin);
1034 kmem_free(name, namelen);
1037 dsda->rm_origin = origin;
1038 dsl_dataset_make_exclusive(origin, tag);
1045 * ds must be opened as OWNER. On return (whether successful or not),
1046 * ds will be closed and caller can no longer dereference it.
1049 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer)
1052 dsl_sync_task_group_t *dstg;
1056 struct dsl_ds_destroyarg dsda = { 0 };
1057 dsl_dataset_t dummy_ds = { 0 };
1061 if (dsl_dataset_is_snapshot(ds)) {
1062 /* Destroying a snapshot is simpler */
1063 dsl_dataset_make_exclusive(ds, tag);
1066 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1067 dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
1069 ASSERT3P(dsda.rm_origin, ==, NULL);
1077 dummy_ds.ds_dir = dd;
1078 dummy_ds.ds_object = ds->ds_object;
1081 * Check for errors and mark this ds as inconsistent, in
1082 * case we crash while freeing the objects.
1084 err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check,
1085 dsl_dataset_destroy_begin_sync, ds, NULL, 0);
1089 err = dmu_objset_from_ds(ds, &os);
1094 * remove the objects in open context, so that we won't
1095 * have too much to do in syncing context.
1097 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
1098 ds->ds_phys->ds_prev_snap_txg)) {
1100 * Ignore errors, if there is not enough disk space
1101 * we will deal with it in dsl_dataset_destroy_sync().
1103 (void) dmu_free_object(os, obj);
1109 * Only the ZIL knows how to free log blocks.
1111 zil_destroy(dmu_objset_zil(os), B_FALSE);
1114 * Sync out all in-flight IO.
1116 txg_wait_synced(dd->dd_pool, 0);
1119 * If we managed to free all the objects in open
1120 * context, the user space accounting should be zero.
1122 if (ds->ds_phys->ds_bp.blk_fill == 0 &&
1123 dmu_objset_userused_enabled(os)) {
1126 ASSERT(zap_count(os, DMU_USERUSED_OBJECT, &count) != 0 ||
1128 ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT, &count) != 0 ||
1132 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
1133 err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd);
1134 rw_exit(&dd->dd_pool->dp_config_rwlock);
1140 * Blow away the dsl_dir + head dataset.
1142 dsl_dataset_make_exclusive(ds, tag);
1144 * If we're removing a clone, we might also need to remove its
1148 dsda.need_prep = B_FALSE;
1149 if (dsl_dir_is_clone(dd)) {
1150 err = dsl_dataset_origin_rm_prep(&dsda, tag);
1152 dsl_dir_close(dd, FTAG);
1157 dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
1158 dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
1159 dsl_dataset_destroy_sync, &dsda, tag, 0);
1160 dsl_sync_task_create(dstg, dsl_dir_destroy_check,
1161 dsl_dir_destroy_sync, &dummy_ds, FTAG, 0);
1162 err = dsl_sync_task_group_wait(dstg);
1163 dsl_sync_task_group_destroy(dstg);
1166 * We could be racing against 'zfs release' or 'zfs destroy -d'
1167 * on the origin snap, in which case we can get EBUSY if we
1168 * needed to destroy the origin snap but were not ready to
1171 if (dsda.need_prep) {
1172 ASSERT(err == EBUSY);
1173 ASSERT(dsl_dir_is_clone(dd));
1174 ASSERT(dsda.rm_origin == NULL);
1176 } while (dsda.need_prep);
1178 if (dsda.rm_origin != NULL)
1179 dsl_dataset_disown(dsda.rm_origin, tag);
1181 /* if it is successful, dsl_dir_destroy_sync will close the dd */
1183 dsl_dir_close(dd, FTAG);
1185 dsl_dataset_disown(ds, tag);
1190 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1192 return (&ds->ds_phys->ds_bp);
1196 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1198 ASSERT(dmu_tx_is_syncing(tx));
1199 /* If it's the meta-objset, set dp_meta_rootbp */
1201 tx->tx_pool->dp_meta_rootbp = *bp;
1203 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1204 ds->ds_phys->ds_bp = *bp;
1209 dsl_dataset_get_spa(dsl_dataset_t *ds)
1211 return (ds->ds_dir->dd_pool->dp_spa);
1215 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1219 if (ds == NULL) /* this is the meta-objset */
1222 ASSERT(ds->ds_objset != NULL);
1224 if (ds->ds_phys->ds_next_snap_obj != 0)
1225 panic("dirtying snapshot!");
1227 dp = ds->ds_dir->dd_pool;
1229 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) {
1230 /* up the hold count until we can be written out */
1231 dmu_buf_add_ref(ds->ds_dbuf, ds);
1236 * The unique space in the head dataset can be calculated by subtracting
1237 * the space used in the most recent snapshot, that is still being used
1238 * in this file system, from the space currently in use. To figure out
1239 * the space in the most recent snapshot still in use, we need to take
1240 * the total space used in the snapshot and subtract out the space that
1241 * has been freed up since the snapshot was taken.
1244 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1247 uint64_t dlused, dlcomp, dluncomp;
1249 ASSERT(!dsl_dataset_is_snapshot(ds));
1251 if (ds->ds_phys->ds_prev_snap_obj != 0)
1252 mrs_used = ds->ds_prev->ds_phys->ds_used_bytes;
1256 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
1258 ASSERT3U(dlused, <=, mrs_used);
1259 ds->ds_phys->ds_unique_bytes =
1260 ds->ds_phys->ds_used_bytes - (mrs_used - dlused);
1262 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1263 SPA_VERSION_UNIQUE_ACCURATE)
1264 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1274 kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf,
1275 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1277 struct killarg *ka = arg;
1278 dmu_tx_t *tx = ka->tx;
1283 if (zb->zb_level == ZB_ZIL_LEVEL) {
1284 ASSERT(zilog != NULL);
1286 * It's a block in the intent log. It has no
1287 * accounting, so just free it.
1289 dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp);
1291 ASSERT(zilog == NULL);
1292 ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg);
1293 (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE);
1301 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1303 dsl_dataset_t *ds = arg1;
1304 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1309 * Can't delete a head dataset if there are snapshots of it.
1310 * (Except if the only snapshots are from the branch we cloned
1313 if (ds->ds_prev != NULL &&
1314 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1318 * This is really a dsl_dir thing, but check it here so that
1319 * we'll be less likely to leave this dataset inconsistent &
1322 err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count);
1333 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1335 dsl_dataset_t *ds = arg1;
1336 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1338 /* Mark it as inconsistent on-disk, in case we crash */
1339 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1340 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1342 spa_history_log_internal(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx,
1343 "dataset = %llu", ds->ds_object);
1347 dsl_dataset_origin_check(struct dsl_ds_destroyarg *dsda, void *tag,
1350 dsl_dataset_t *ds = dsda->ds;
1351 dsl_dataset_t *ds_prev = ds->ds_prev;
1353 if (dsl_dataset_might_destroy_origin(ds_prev)) {
1354 struct dsl_ds_destroyarg ndsda = {0};
1357 * If we're not prepared to remove the origin, don't remove
1360 if (dsda->rm_origin == NULL) {
1361 dsda->need_prep = B_TRUE;
1366 ndsda.is_origin_rm = B_TRUE;
1367 return (dsl_dataset_destroy_check(&ndsda, tag, tx));
1371 * If we're not going to remove the origin after all,
1372 * undo the open context setup.
1374 if (dsda->rm_origin != NULL) {
1375 dsl_dataset_disown(dsda->rm_origin, tag);
1376 dsda->rm_origin = NULL;
1383 * If you add new checks here, you may need to add
1384 * additional checks to the "temporary" case in
1385 * snapshot_check() in dmu_objset.c.
1389 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1391 struct dsl_ds_destroyarg *dsda = arg1;
1392 dsl_dataset_t *ds = dsda->ds;
1394 /* we have an owner hold, so noone else can destroy us */
1395 ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
1398 * Only allow deferred destroy on pools that support it.
1399 * NOTE: deferred destroy is only supported on snapshots.
1402 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
1403 SPA_VERSION_USERREFS)
1405 ASSERT(dsl_dataset_is_snapshot(ds));
1410 * Can't delete a head dataset if there are snapshots of it.
1411 * (Except if the only snapshots are from the branch we cloned
1414 if (ds->ds_prev != NULL &&
1415 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1419 * If we made changes this txg, traverse_dsl_dataset won't find
1422 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1425 if (dsl_dataset_is_snapshot(ds)) {
1427 * If this snapshot has an elevated user reference count,
1428 * we can't destroy it yet.
1430 if (ds->ds_userrefs > 0 && !dsda->releasing)
1433 mutex_enter(&ds->ds_lock);
1435 * Can't delete a branch point. However, if we're destroying
1436 * a clone and removing its origin due to it having a user
1437 * hold count of 0 and having been marked for deferred destroy,
1438 * it's OK for the origin to have a single clone.
1440 if (ds->ds_phys->ds_num_children >
1441 (dsda->is_origin_rm ? 2 : 1)) {
1442 mutex_exit(&ds->ds_lock);
1445 mutex_exit(&ds->ds_lock);
1446 } else if (dsl_dir_is_clone(ds->ds_dir)) {
1447 return (dsl_dataset_origin_check(dsda, arg2, tx));
1450 /* XXX we should do some i/o error checking... */
1462 dsl_dataset_refs_gone(dmu_buf_t *db, void *argv)
1464 struct refsarg *arg = argv;
1466 mutex_enter(&arg->lock);
1468 cv_signal(&arg->cv);
1469 mutex_exit(&arg->lock);
1473 dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag)
1477 bzero(&arg, sizeof(arg));
1478 mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL);
1479 cv_init(&arg.cv, NULL, CV_DEFAULT, NULL);
1481 (void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys,
1482 dsl_dataset_refs_gone);
1483 dmu_buf_rele(ds->ds_dbuf, tag);
1484 mutex_enter(&arg.lock);
1486 cv_wait(&arg.cv, &arg.lock);
1488 mutex_exit(&arg.lock);
1491 mutex_destroy(&arg.lock);
1492 cv_destroy(&arg.cv);
1496 remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, dmu_tx_t *tx)
1498 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1502 ASSERT(ds->ds_phys->ds_num_children >= 2);
1503 err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx);
1505 * The err should not be ENOENT, but a bug in a previous version
1506 * of the code could cause upgrade_clones_cb() to not set
1507 * ds_next_snap_obj when it should, leading to a missing entry.
1508 * If we knew that the pool was created after
1509 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1510 * ENOENT. However, at least we can check that we don't have
1511 * too many entries in the next_clones_obj even after failing to
1514 if (err != ENOENT) {
1515 VERIFY3U(err, ==, 0);
1517 ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj,
1519 ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2);
1523 dsl_dataset_remove_clones_key(dsl_dataset_t *ds, uint64_t mintxg, dmu_tx_t *tx)
1525 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1530 * If it is the old version, dd_clones doesn't exist so we can't
1531 * find the clones, but deadlist_remove_key() is a no-op so it
1534 if (ds->ds_dir->dd_phys->dd_clones == 0)
1537 for (zap_cursor_init(&zc, mos, ds->ds_dir->dd_phys->dd_clones);
1538 zap_cursor_retrieve(&zc, &za) == 0;
1539 zap_cursor_advance(&zc)) {
1540 dsl_dataset_t *clone;
1542 VERIFY3U(0, ==, dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1543 za.za_first_integer, FTAG, &clone));
1544 if (clone->ds_dir->dd_origin_txg > mintxg) {
1545 dsl_deadlist_remove_key(&clone->ds_deadlist,
1547 dsl_dataset_remove_clones_key(clone, mintxg, tx);
1549 dsl_dataset_rele(clone, FTAG);
1551 zap_cursor_fini(&zc);
1554 struct process_old_arg {
1556 dsl_dataset_t *ds_prev;
1557 boolean_t after_branch_point;
1559 uint64_t used, comp, uncomp;
1563 process_old_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
1565 struct process_old_arg *poa = arg;
1566 dsl_pool_t *dp = poa->ds->ds_dir->dd_pool;
1568 if (bp->blk_birth <= poa->ds->ds_phys->ds_prev_snap_txg) {
1569 dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, tx);
1570 if (poa->ds_prev && !poa->after_branch_point &&
1572 poa->ds_prev->ds_phys->ds_prev_snap_txg) {
1573 poa->ds_prev->ds_phys->ds_unique_bytes +=
1574 bp_get_dsize_sync(dp->dp_spa, bp);
1577 poa->used += bp_get_dsize_sync(dp->dp_spa, bp);
1578 poa->comp += BP_GET_PSIZE(bp);
1579 poa->uncomp += BP_GET_UCSIZE(bp);
1580 dsl_free_sync(poa->pio, dp, tx->tx_txg, bp);
1586 process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev,
1587 dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx)
1589 struct process_old_arg poa = { 0 };
1590 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1591 objset_t *mos = dp->dp_meta_objset;
1593 ASSERT(ds->ds_deadlist.dl_oldfmt);
1594 ASSERT(ds_next->ds_deadlist.dl_oldfmt);
1597 poa.ds_prev = ds_prev;
1598 poa.after_branch_point = after_branch_point;
1599 poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1600 VERIFY3U(0, ==, bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj,
1601 process_old_cb, &poa, tx));
1602 VERIFY3U(zio_wait(poa.pio), ==, 0);
1603 ASSERT3U(poa.used, ==, ds->ds_phys->ds_unique_bytes);
1605 /* change snapused */
1606 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1607 -poa.used, -poa.comp, -poa.uncomp, tx);
1609 /* swap next's deadlist to our deadlist */
1610 dsl_deadlist_close(&ds->ds_deadlist);
1611 dsl_deadlist_close(&ds_next->ds_deadlist);
1612 SWITCH64(ds_next->ds_phys->ds_deadlist_obj,
1613 ds->ds_phys->ds_deadlist_obj);
1614 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
1615 dsl_deadlist_open(&ds_next->ds_deadlist, mos,
1616 ds_next->ds_phys->ds_deadlist_obj);
1620 dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
1622 struct dsl_ds_destroyarg *dsda = arg1;
1623 dsl_dataset_t *ds = dsda->ds;
1625 int after_branch_point = FALSE;
1626 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1627 objset_t *mos = dp->dp_meta_objset;
1628 dsl_dataset_t *ds_prev = NULL;
1629 boolean_t wont_destroy;
1632 wont_destroy = (dsda->defer &&
1633 (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1));
1635 ASSERT(ds->ds_owner || wont_destroy);
1636 ASSERT(dsda->defer || ds->ds_phys->ds_num_children <= 1);
1637 ASSERT(ds->ds_prev == NULL ||
1638 ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1639 ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1642 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
1643 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1644 ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY;
1648 /* signal any waiters that this dataset is going away */
1649 mutex_enter(&ds->ds_lock);
1650 ds->ds_owner = dsl_reaper;
1651 cv_broadcast(&ds->ds_exclusive_cv);
1652 mutex_exit(&ds->ds_lock);
1654 /* Remove our reservation */
1655 if (ds->ds_reserved != 0) {
1656 dsl_prop_setarg_t psa;
1659 dsl_prop_setarg_init_uint64(&psa, "refreservation",
1660 (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
1662 psa.psa_effective_value = 0; /* predict default value */
1664 dsl_dataset_set_reservation_sync(ds, &psa, tx);
1665 ASSERT3U(ds->ds_reserved, ==, 0);
1668 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1670 dsl_scan_ds_destroyed(ds, tx);
1672 obj = ds->ds_object;
1674 if (ds->ds_phys->ds_prev_snap_obj != 0) {
1676 ds_prev = ds->ds_prev;
1678 VERIFY(0 == dsl_dataset_hold_obj(dp,
1679 ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev));
1681 after_branch_point =
1682 (ds_prev->ds_phys->ds_next_snap_obj != obj);
1684 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
1685 if (after_branch_point &&
1686 ds_prev->ds_phys->ds_next_clones_obj != 0) {
1687 remove_from_next_clones(ds_prev, obj, tx);
1688 if (ds->ds_phys->ds_next_snap_obj != 0) {
1689 VERIFY(0 == zap_add_int(mos,
1690 ds_prev->ds_phys->ds_next_clones_obj,
1691 ds->ds_phys->ds_next_snap_obj, tx));
1694 if (after_branch_point &&
1695 ds->ds_phys->ds_next_snap_obj == 0) {
1696 /* This clone is toast. */
1697 ASSERT(ds_prev->ds_phys->ds_num_children > 1);
1698 ds_prev->ds_phys->ds_num_children--;
1701 * If the clone's origin has no other clones, no
1702 * user holds, and has been marked for deferred
1703 * deletion, then we should have done the necessary
1704 * destroy setup for it.
1706 if (ds_prev->ds_phys->ds_num_children == 1 &&
1707 ds_prev->ds_userrefs == 0 &&
1708 DS_IS_DEFER_DESTROY(ds_prev)) {
1709 ASSERT3P(dsda->rm_origin, !=, NULL);
1711 ASSERT3P(dsda->rm_origin, ==, NULL);
1713 } else if (!after_branch_point) {
1714 ds_prev->ds_phys->ds_next_snap_obj =
1715 ds->ds_phys->ds_next_snap_obj;
1719 if (dsl_dataset_is_snapshot(ds)) {
1720 dsl_dataset_t *ds_next;
1721 uint64_t old_unique;
1722 uint64_t used = 0, comp = 0, uncomp = 0;
1724 VERIFY(0 == dsl_dataset_hold_obj(dp,
1725 ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next));
1726 ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
1728 old_unique = ds_next->ds_phys->ds_unique_bytes;
1730 dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
1731 ds_next->ds_phys->ds_prev_snap_obj =
1732 ds->ds_phys->ds_prev_snap_obj;
1733 ds_next->ds_phys->ds_prev_snap_txg =
1734 ds->ds_phys->ds_prev_snap_txg;
1735 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1736 ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
1739 if (ds_next->ds_deadlist.dl_oldfmt) {
1740 process_old_deadlist(ds, ds_prev, ds_next,
1741 after_branch_point, tx);
1743 /* Adjust prev's unique space. */
1744 if (ds_prev && !after_branch_point) {
1745 dsl_deadlist_space_range(&ds_next->ds_deadlist,
1746 ds_prev->ds_phys->ds_prev_snap_txg,
1747 ds->ds_phys->ds_prev_snap_txg,
1748 &used, &comp, &uncomp);
1749 ds_prev->ds_phys->ds_unique_bytes += used;
1752 /* Adjust snapused. */
1753 dsl_deadlist_space_range(&ds_next->ds_deadlist,
1754 ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
1755 &used, &comp, &uncomp);
1756 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1757 -used, -comp, -uncomp, tx);
1759 /* Move blocks to be freed to pool's free list. */
1760 dsl_deadlist_move_bpobj(&ds_next->ds_deadlist,
1761 &dp->dp_free_bpobj, ds->ds_phys->ds_prev_snap_txg,
1763 dsl_dir_diduse_space(tx->tx_pool->dp_free_dir,
1764 DD_USED_HEAD, used, comp, uncomp, tx);
1765 dsl_dir_dirty(tx->tx_pool->dp_free_dir, tx);
1767 /* Merge our deadlist into next's and free it. */
1768 dsl_deadlist_merge(&ds_next->ds_deadlist,
1769 ds->ds_phys->ds_deadlist_obj, tx);
1771 dsl_deadlist_close(&ds->ds_deadlist);
1772 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
1774 /* Collapse range in clone heads */
1775 dsl_dataset_remove_clones_key(ds,
1776 ds->ds_phys->ds_creation_txg, tx);
1778 if (dsl_dataset_is_snapshot(ds_next)) {
1779 dsl_dataset_t *ds_nextnext;
1782 * Update next's unique to include blocks which
1783 * were previously shared by only this snapshot
1784 * and it. Those blocks will be born after the
1785 * prev snap and before this snap, and will have
1786 * died after the next snap and before the one
1787 * after that (ie. be on the snap after next's
1790 VERIFY(0 == dsl_dataset_hold_obj(dp,
1791 ds_next->ds_phys->ds_next_snap_obj,
1792 FTAG, &ds_nextnext));
1793 dsl_deadlist_space_range(&ds_nextnext->ds_deadlist,
1794 ds->ds_phys->ds_prev_snap_txg,
1795 ds->ds_phys->ds_creation_txg,
1796 &used, &comp, &uncomp);
1797 ds_next->ds_phys->ds_unique_bytes += used;
1798 dsl_dataset_rele(ds_nextnext, FTAG);
1799 ASSERT3P(ds_next->ds_prev, ==, NULL);
1801 /* Collapse range in this head. */
1803 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
1804 ds->ds_dir->dd_phys->dd_head_dataset_obj,
1806 dsl_deadlist_remove_key(&hds->ds_deadlist,
1807 ds->ds_phys->ds_creation_txg, tx);
1808 dsl_dataset_rele(hds, FTAG);
1811 ASSERT3P(ds_next->ds_prev, ==, ds);
1812 dsl_dataset_drop_ref(ds_next->ds_prev, ds_next);
1813 ds_next->ds_prev = NULL;
1815 VERIFY(0 == dsl_dataset_get_ref(dp,
1816 ds->ds_phys->ds_prev_snap_obj,
1817 ds_next, &ds_next->ds_prev));
1820 dsl_dataset_recalc_head_uniq(ds_next);
1823 * Reduce the amount of our unconsmed refreservation
1824 * being charged to our parent by the amount of
1825 * new unique data we have gained.
1827 if (old_unique < ds_next->ds_reserved) {
1829 uint64_t new_unique =
1830 ds_next->ds_phys->ds_unique_bytes;
1832 ASSERT(old_unique <= new_unique);
1833 mrsdelta = MIN(new_unique - old_unique,
1834 ds_next->ds_reserved - old_unique);
1835 dsl_dir_diduse_space(ds->ds_dir,
1836 DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
1839 dsl_dataset_rele(ds_next, FTAG);
1842 * There's no next snapshot, so this is a head dataset.
1843 * Destroy the deadlist. Unless it's a clone, the
1844 * deadlist should be empty. (If it's a clone, it's
1845 * safe to ignore the deadlist contents.)
1849 dsl_deadlist_close(&ds->ds_deadlist);
1850 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
1851 ds->ds_phys->ds_deadlist_obj = 0;
1854 * Free everything that we point to (that's born after
1855 * the previous snapshot, if we are a clone)
1857 * NB: this should be very quick, because we already
1858 * freed all the objects in open context.
1862 err = traverse_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1863 TRAVERSE_POST, kill_blkptr, &ka);
1864 ASSERT3U(err, ==, 0);
1865 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
1866 ds->ds_phys->ds_unique_bytes == 0);
1868 if (ds->ds_prev != NULL) {
1869 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
1870 VERIFY3U(0, ==, zap_remove_int(mos,
1871 ds->ds_prev->ds_dir->dd_phys->dd_clones,
1872 ds->ds_object, tx));
1874 dsl_dataset_rele(ds->ds_prev, ds);
1875 ds->ds_prev = ds_prev = NULL;
1880 * This must be done after the dsl_traverse(), because it will
1881 * re-open the objset.
1883 if (ds->ds_objset) {
1884 dmu_objset_evict(ds->ds_objset);
1885 ds->ds_objset = NULL;
1888 if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) {
1889 /* Erase the link in the dir */
1890 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1891 ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
1892 ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0);
1893 err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx);
1896 /* remove from snapshot namespace */
1897 dsl_dataset_t *ds_head;
1898 ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0);
1899 VERIFY(0 == dsl_dataset_hold_obj(dp,
1900 ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head));
1901 VERIFY(0 == dsl_dataset_get_snapname(ds));
1906 err = dsl_dataset_snap_lookup(ds_head,
1907 ds->ds_snapname, &val);
1908 ASSERT3U(err, ==, 0);
1909 ASSERT3U(val, ==, obj);
1912 err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx);
1914 dsl_dataset_rele(ds_head, FTAG);
1917 if (ds_prev && ds->ds_prev != ds_prev)
1918 dsl_dataset_rele(ds_prev, FTAG);
1920 spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
1921 spa_history_log_internal(LOG_DS_DESTROY, dp->dp_spa, tx,
1922 "dataset = %llu", ds->ds_object);
1924 if (ds->ds_phys->ds_next_clones_obj != 0) {
1926 ASSERT(0 == zap_count(mos,
1927 ds->ds_phys->ds_next_clones_obj, &count) && count == 0);
1928 VERIFY(0 == dmu_object_free(mos,
1929 ds->ds_phys->ds_next_clones_obj, tx));
1931 if (ds->ds_phys->ds_props_obj != 0)
1932 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx));
1933 if (ds->ds_phys->ds_userrefs_obj != 0)
1934 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx));
1935 dsl_dir_close(ds->ds_dir, ds);
1937 dsl_dataset_drain_refs(ds, tag);
1938 VERIFY(0 == dmu_object_free(mos, obj, tx));
1940 if (dsda->rm_origin) {
1942 * Remove the origin of the clone we just destroyed.
1944 struct dsl_ds_destroyarg ndsda = {0};
1946 ndsda.ds = dsda->rm_origin;
1947 dsl_dataset_destroy_sync(&ndsda, tag, tx);
1952 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1956 if (!dmu_tx_is_syncing(tx))
1960 * If there's an fs-only reservation, any blocks that might become
1961 * owned by the snapshot dataset must be accommodated by space
1962 * outside of the reservation.
1964 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1965 asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
1966 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1970 * Propogate any reserved space for this snapshot to other
1971 * snapshot checks in this sync group.
1974 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1980 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
1982 dsl_dataset_t *ds = arg1;
1983 const char *snapname = arg2;
1988 * We don't allow multiple snapshots of the same txg. If there
1989 * is already one, try again.
1991 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
1995 * Check for conflicting name snapshot name.
1997 err = dsl_dataset_snap_lookup(ds, snapname, &value);
2004 * Check that the dataset's name is not too long. Name consists
2005 * of the dataset's length + 1 for the @-sign + snapshot name's length
2007 if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
2008 return (ENAMETOOLONG);
2010 err = dsl_dataset_snapshot_reserve_space(ds, tx);
2014 ds->ds_trysnap_txg = tx->tx_txg;
2019 dsl_dataset_snapshot_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2021 dsl_dataset_t *ds = arg1;
2022 const char *snapname = arg2;
2023 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2025 dsl_dataset_phys_t *dsphys;
2026 uint64_t dsobj, crtxg;
2027 objset_t *mos = dp->dp_meta_objset;
2030 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
2033 * The origin's ds_creation_txg has to be < TXG_INITIAL
2035 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
2040 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
2041 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
2042 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
2043 dmu_buf_will_dirty(dbuf, tx);
2044 dsphys = dbuf->db_data;
2045 bzero(dsphys, sizeof (dsl_dataset_phys_t));
2046 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
2047 dsphys->ds_fsid_guid = unique_create();
2048 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
2049 sizeof (dsphys->ds_guid));
2050 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
2051 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
2052 dsphys->ds_next_snap_obj = ds->ds_object;
2053 dsphys->ds_num_children = 1;
2054 dsphys->ds_creation_time = gethrestime_sec();
2055 dsphys->ds_creation_txg = crtxg;
2056 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
2057 dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes;
2058 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
2059 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
2060 dsphys->ds_flags = ds->ds_phys->ds_flags;
2061 dsphys->ds_bp = ds->ds_phys->ds_bp;
2062 dmu_buf_rele(dbuf, FTAG);
2064 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
2066 uint64_t next_clones_obj =
2067 ds->ds_prev->ds_phys->ds_next_clones_obj;
2068 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
2070 ds->ds_prev->ds_phys->ds_num_children > 1);
2071 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
2072 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
2073 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
2074 ds->ds_prev->ds_phys->ds_creation_txg);
2075 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
2076 } else if (next_clones_obj != 0) {
2077 remove_from_next_clones(ds->ds_prev,
2078 dsphys->ds_next_snap_obj, tx);
2079 VERIFY3U(0, ==, zap_add_int(mos,
2080 next_clones_obj, dsobj, tx));
2085 * If we have a reference-reservation on this dataset, we will
2086 * need to increase the amount of refreservation being charged
2087 * since our unique space is going to zero.
2089 if (ds->ds_reserved) {
2091 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
2092 delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
2093 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
2097 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2098 zfs_dbgmsg("taking snapshot %s@%s/%llu; newkey=%llu",
2099 ds->ds_dir->dd_myname, snapname, dsobj,
2100 ds->ds_phys->ds_prev_snap_txg);
2101 ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist,
2102 UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx);
2103 dsl_deadlist_close(&ds->ds_deadlist);
2104 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
2105 dsl_deadlist_add_key(&ds->ds_deadlist,
2106 ds->ds_phys->ds_prev_snap_txg, tx);
2108 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
2109 ds->ds_phys->ds_prev_snap_obj = dsobj;
2110 ds->ds_phys->ds_prev_snap_txg = crtxg;
2111 ds->ds_phys->ds_unique_bytes = 0;
2112 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
2113 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
2115 err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
2116 snapname, 8, 1, &dsobj, tx);
2120 dsl_dataset_drop_ref(ds->ds_prev, ds);
2121 VERIFY(0 == dsl_dataset_get_ref(dp,
2122 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
2124 dsl_scan_ds_snapshotted(ds, tx);
2126 dsl_dir_snap_cmtime_update(ds->ds_dir);
2128 spa_history_log_internal(LOG_DS_SNAPSHOT, dp->dp_spa, tx,
2129 "dataset = %llu", dsobj);
2133 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
2135 ASSERT(dmu_tx_is_syncing(tx));
2136 ASSERT(ds->ds_objset != NULL);
2137 ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
2140 * in case we had to change ds_fsid_guid when we opened it,
2143 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2144 ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
2146 dsl_dir_dirty(ds->ds_dir, tx);
2147 dmu_objset_sync(ds->ds_objset, zio, tx);
2151 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
2153 uint64_t refd, avail, uobjs, aobjs;
2155 dsl_dir_stats(ds->ds_dir, nv);
2157 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
2158 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
2159 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
2161 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
2162 ds->ds_phys->ds_creation_time);
2163 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
2164 ds->ds_phys->ds_creation_txg);
2165 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
2167 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
2169 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
2170 ds->ds_phys->ds_guid);
2171 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
2172 ds->ds_phys->ds_unique_bytes);
2173 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
2175 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
2177 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
2178 DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
2180 if (ds->ds_phys->ds_next_snap_obj) {
2182 * This is a snapshot; override the dd's space used with
2183 * our unique space and compression ratio.
2185 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
2186 ds->ds_phys->ds_unique_bytes);
2187 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
2188 ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
2189 (ds->ds_phys->ds_uncompressed_bytes * 100 /
2190 ds->ds_phys->ds_compressed_bytes));
2195 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
2197 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
2198 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
2199 stat->dds_guid = ds->ds_phys->ds_guid;
2200 if (ds->ds_phys->ds_next_snap_obj) {
2201 stat->dds_is_snapshot = B_TRUE;
2202 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
2204 stat->dds_is_snapshot = B_FALSE;
2205 stat->dds_num_clones = 0;
2208 /* clone origin is really a dsl_dir thing... */
2209 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
2210 if (dsl_dir_is_clone(ds->ds_dir)) {
2213 VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool,
2214 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
2215 dsl_dataset_name(ods, stat->dds_origin);
2216 dsl_dataset_drop_ref(ods, FTAG);
2218 stat->dds_origin[0] = '\0';
2220 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2224 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2226 return (ds->ds_fsid_guid);
2230 dsl_dataset_space(dsl_dataset_t *ds,
2231 uint64_t *refdbytesp, uint64_t *availbytesp,
2232 uint64_t *usedobjsp, uint64_t *availobjsp)
2234 *refdbytesp = ds->ds_phys->ds_used_bytes;
2235 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2236 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
2237 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
2238 if (ds->ds_quota != 0) {
2240 * Adjust available bytes according to refquota
2242 if (*refdbytesp < ds->ds_quota)
2243 *availbytesp = MIN(*availbytesp,
2244 ds->ds_quota - *refdbytesp);
2248 *usedobjsp = ds->ds_phys->ds_bp.blk_fill;
2249 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
2253 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
2255 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2257 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
2258 dsl_pool_sync_context(dp));
2259 if (ds->ds_prev == NULL)
2261 if (ds->ds_phys->ds_bp.blk_birth >
2262 ds->ds_prev->ds_phys->ds_creation_txg) {
2263 objset_t *os, *os_prev;
2265 * It may be that only the ZIL differs, because it was
2266 * reset in the head. Don't count that as being
2269 if (dmu_objset_from_ds(ds, &os) != 0)
2271 if (dmu_objset_from_ds(ds->ds_prev, &os_prev) != 0)
2273 return (bcmp(&os->os_phys->os_meta_dnode,
2274 &os_prev->os_phys->os_meta_dnode,
2275 sizeof (os->os_phys->os_meta_dnode)) != 0);
2282 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
2284 dsl_dataset_t *ds = arg1;
2285 char *newsnapname = arg2;
2286 dsl_dir_t *dd = ds->ds_dir;
2291 err = dsl_dataset_hold_obj(dd->dd_pool,
2292 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds);
2296 /* new name better not be in use */
2297 err = dsl_dataset_snap_lookup(hds, newsnapname, &val);
2298 dsl_dataset_rele(hds, FTAG);
2302 else if (err == ENOENT)
2305 /* dataset name + 1 for the "@" + the new snapshot name must fit */
2306 if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
2313 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2315 char oldname[MAXPATHLEN], newname[MAXPATHLEN];
2316 dsl_dataset_t *ds = arg1;
2317 const char *newsnapname = arg2;
2318 dsl_dir_t *dd = ds->ds_dir;
2319 objset_t *mos = dd->dd_pool->dp_meta_objset;
2323 ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
2325 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
2326 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds));
2328 VERIFY(0 == dsl_dataset_get_snapname(ds));
2329 err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx);
2330 ASSERT3U(err, ==, 0);
2331 dsl_dataset_name(ds, oldname);
2332 mutex_enter(&ds->ds_lock);
2333 (void) strcpy(ds->ds_snapname, newsnapname);
2334 mutex_exit(&ds->ds_lock);
2335 err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
2336 ds->ds_snapname, 8, 1, &ds->ds_object, tx);
2337 ASSERT3U(err, ==, 0);
2338 dsl_dataset_name(ds, newname);
2340 zvol_rename_minors(oldname, newname);
2343 spa_history_log_internal(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx,
2344 "dataset = %llu", ds->ds_object);
2345 dsl_dataset_rele(hds, FTAG);
2348 struct renamesnaparg {
2349 dsl_sync_task_group_t *dstg;
2350 char failed[MAXPATHLEN];
2356 dsl_snapshot_rename_one(const char *name, void *arg)
2358 struct renamesnaparg *ra = arg;
2359 dsl_dataset_t *ds = NULL;
2363 snapname = kmem_asprintf("%s@%s", name, ra->oldsnap);
2364 (void) strlcpy(ra->failed, snapname, sizeof (ra->failed));
2367 * For recursive snapshot renames the parent won't be changing
2368 * so we just pass name for both the to/from argument.
2370 err = zfs_secpolicy_rename_perms(snapname, snapname, CRED());
2373 return (err == ENOENT ? 0 : err);
2378 * For all filesystems undergoing rename, we'll need to unmount it.
2380 (void) zfs_unmount_snap(snapname, NULL);
2382 err = dsl_dataset_hold(snapname, ra->dstg, &ds);
2385 return (err == ENOENT ? 0 : err);
2387 dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
2388 dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
2394 dsl_recursive_rename(char *oldname, const char *newname)
2397 struct renamesnaparg *ra;
2398 dsl_sync_task_t *dst;
2400 char *cp, *fsname = spa_strdup(oldname);
2401 int len = strlen(oldname) + 1;
2403 /* truncate the snapshot name to get the fsname */
2404 cp = strchr(fsname, '@');
2407 err = spa_open(fsname, &spa, FTAG);
2409 kmem_free(fsname, len);
2412 ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP);
2413 ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
2415 ra->oldsnap = strchr(oldname, '@') + 1;
2416 ra->newsnap = strchr(newname, '@') + 1;
2419 err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
2421 kmem_free(fsname, len);
2424 err = dsl_sync_task_group_wait(ra->dstg);
2427 for (dst = list_head(&ra->dstg->dstg_tasks); dst;
2428 dst = list_next(&ra->dstg->dstg_tasks, dst)) {
2429 dsl_dataset_t *ds = dst->dst_arg1;
2431 dsl_dir_name(ds->ds_dir, ra->failed);
2432 (void) strlcat(ra->failed, "@", sizeof (ra->failed));
2433 (void) strlcat(ra->failed, ra->newsnap,
2434 sizeof (ra->failed));
2436 dsl_dataset_rele(ds, ra->dstg);
2440 (void) strlcpy(oldname, ra->failed, sizeof (ra->failed));
2442 dsl_sync_task_group_destroy(ra->dstg);
2443 kmem_free(ra, sizeof (struct renamesnaparg));
2444 spa_close(spa, FTAG);
2449 dsl_valid_rename(const char *oldname, void *arg)
2451 int delta = *(int *)arg;
2453 if (strlen(oldname) + delta >= MAXNAMELEN)
2454 return (ENAMETOOLONG);
2459 #pragma weak dmu_objset_rename = dsl_dataset_rename
2461 dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive)
2468 err = dsl_dir_open(oldname, FTAG, &dd, &tail);
2473 int delta = strlen(newname) - strlen(oldname);
2475 /* if we're growing, validate child name lengths */
2477 err = dmu_objset_find(oldname, dsl_valid_rename,
2478 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2481 err = dsl_dir_rename(dd, newname);
2482 dsl_dir_close(dd, FTAG);
2486 if (tail[0] != '@') {
2487 /* the name ended in a nonexistent component */
2488 dsl_dir_close(dd, FTAG);
2492 dsl_dir_close(dd, FTAG);
2494 /* new name must be snapshot in same filesystem */
2495 tail = strchr(newname, '@');
2499 if (strncmp(oldname, newname, tail - newname) != 0)
2503 err = dsl_recursive_rename(oldname, newname);
2505 err = dsl_dataset_hold(oldname, FTAG, &ds);
2509 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2510 dsl_dataset_snapshot_rename_check,
2511 dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
2513 dsl_dataset_rele(ds, FTAG);
2519 struct promotenode {
2525 list_t shared_snaps, origin_snaps, clone_snaps;
2526 dsl_dataset_t *origin_origin;
2527 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2531 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2532 static boolean_t snaplist_unstable(list_t *l);
2535 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
2537 dsl_dataset_t *hds = arg1;
2538 struct promotearg *pa = arg2;
2539 struct promotenode *snap = list_head(&pa->shared_snaps);
2540 dsl_dataset_t *origin_ds = snap->ds;
2544 /* Check that it is a real clone */
2545 if (!dsl_dir_is_clone(hds->ds_dir))
2548 /* Since this is so expensive, don't do the preliminary check */
2549 if (!dmu_tx_is_syncing(tx))
2552 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)
2555 /* compute origin's new unique space */
2556 snap = list_tail(&pa->clone_snaps);
2557 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2558 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2559 origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2560 &pa->unique, &unused, &unused);
2563 * Walk the snapshots that we are moving
2565 * Compute space to transfer. Consider the incremental changes
2566 * to used for each snapshot:
2567 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2568 * So each snapshot gave birth to:
2569 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2570 * So a sequence would look like:
2571 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2572 * Which simplifies to:
2573 * uN + kN + kN-1 + ... + k1 + k0
2574 * Note however, if we stop before we reach the ORIGIN we get:
2575 * uN + kN + kN-1 + ... + kM - uM-1
2577 pa->used = origin_ds->ds_phys->ds_used_bytes;
2578 pa->comp = origin_ds->ds_phys->ds_compressed_bytes;
2579 pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
2580 for (snap = list_head(&pa->shared_snaps); snap;
2581 snap = list_next(&pa->shared_snaps, snap)) {
2582 uint64_t val, dlused, dlcomp, dluncomp;
2583 dsl_dataset_t *ds = snap->ds;
2585 /* Check that the snapshot name does not conflict */
2586 VERIFY(0 == dsl_dataset_get_snapname(ds));
2587 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2595 /* The very first snapshot does not have a deadlist */
2596 if (ds->ds_phys->ds_prev_snap_obj == 0)
2599 dsl_deadlist_space(&ds->ds_deadlist,
2600 &dlused, &dlcomp, &dluncomp);
2603 pa->uncomp += dluncomp;
2607 * If we are a clone of a clone then we never reached ORIGIN,
2608 * so we need to subtract out the clone origin's used space.
2610 if (pa->origin_origin) {
2611 pa->used -= pa->origin_origin->ds_phys->ds_used_bytes;
2612 pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes;
2613 pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes;
2616 /* Check that there is enough space here */
2617 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2623 * Compute the amounts of space that will be used by snapshots
2624 * after the promotion (for both origin and clone). For each,
2625 * it is the amount of space that will be on all of their
2626 * deadlists (that was not born before their new origin).
2628 if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2632 * Note, typically this will not be a clone of a clone,
2633 * so dd_origin_txg will be < TXG_INITIAL, so
2634 * these snaplist_space() -> dsl_deadlist_space_range()
2635 * calls will be fast because they do not have to
2636 * iterate over all bps.
2638 snap = list_head(&pa->origin_snaps);
2639 err = snaplist_space(&pa->shared_snaps,
2640 snap->ds->ds_dir->dd_origin_txg, &pa->cloneusedsnap);
2644 err = snaplist_space(&pa->clone_snaps,
2645 snap->ds->ds_dir->dd_origin_txg, &space);
2648 pa->cloneusedsnap += space;
2650 if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2651 err = snaplist_space(&pa->origin_snaps,
2652 origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap);
2659 pa->err_ds = snap->ds->ds_snapname;
2664 dsl_dataset_promote_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2666 dsl_dataset_t *hds = arg1;
2667 struct promotearg *pa = arg2;
2668 struct promotenode *snap = list_head(&pa->shared_snaps);
2669 dsl_dataset_t *origin_ds = snap->ds;
2670 dsl_dataset_t *origin_head;
2671 dsl_dir_t *dd = hds->ds_dir;
2672 dsl_pool_t *dp = hds->ds_dir->dd_pool;
2673 dsl_dir_t *odd = NULL;
2674 uint64_t oldnext_obj;
2677 ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
2679 snap = list_head(&pa->origin_snaps);
2680 origin_head = snap->ds;
2683 * We need to explicitly open odd, since origin_ds's dd will be
2686 VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object,
2689 /* change origin's next snap */
2690 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2691 oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2692 snap = list_tail(&pa->clone_snaps);
2693 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2694 origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
2696 /* change the origin's next clone */
2697 if (origin_ds->ds_phys->ds_next_clones_obj) {
2698 remove_from_next_clones(origin_ds, snap->ds->ds_object, tx);
2699 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2700 origin_ds->ds_phys->ds_next_clones_obj,
2705 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2706 ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2707 dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2708 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2709 dmu_buf_will_dirty(odd->dd_dbuf, tx);
2710 odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2711 origin_head->ds_dir->dd_origin_txg =
2712 origin_ds->ds_phys->ds_creation_txg;
2714 /* change dd_clone entries */
2715 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2716 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2717 odd->dd_phys->dd_clones, hds->ds_object, tx));
2718 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2719 pa->origin_origin->ds_dir->dd_phys->dd_clones,
2720 hds->ds_object, tx));
2722 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2723 pa->origin_origin->ds_dir->dd_phys->dd_clones,
2724 origin_head->ds_object, tx));
2725 if (dd->dd_phys->dd_clones == 0) {
2726 dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset,
2727 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
2729 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2730 dd->dd_phys->dd_clones, origin_head->ds_object, tx));
2734 /* move snapshots to this dir */
2735 for (snap = list_head(&pa->shared_snaps); snap;
2736 snap = list_next(&pa->shared_snaps, snap)) {
2737 dsl_dataset_t *ds = snap->ds;
2739 /* unregister props as dsl_dir is changing */
2740 if (ds->ds_objset) {
2741 dmu_objset_evict(ds->ds_objset);
2742 ds->ds_objset = NULL;
2744 /* move snap name entry */
2745 VERIFY(0 == dsl_dataset_get_snapname(ds));
2746 VERIFY(0 == dsl_dataset_snap_remove(origin_head,
2747 ds->ds_snapname, tx));
2748 VERIFY(0 == zap_add(dp->dp_meta_objset,
2749 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2750 8, 1, &ds->ds_object, tx));
2752 /* change containing dsl_dir */
2753 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2754 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2755 ds->ds_phys->ds_dir_obj = dd->dd_object;
2756 ASSERT3P(ds->ds_dir, ==, odd);
2757 dsl_dir_close(ds->ds_dir, ds);
2758 VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
2759 NULL, ds, &ds->ds_dir));
2761 /* move any clone references */
2762 if (ds->ds_phys->ds_next_clones_obj &&
2763 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2767 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2768 ds->ds_phys->ds_next_clones_obj);
2769 zap_cursor_retrieve(&zc, &za) == 0;
2770 zap_cursor_advance(&zc)) {
2771 dsl_dataset_t *cnds;
2774 if (za.za_first_integer == oldnext_obj) {
2776 * We've already moved the
2777 * origin's reference.
2782 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
2783 za.za_first_integer, FTAG, &cnds));
2784 o = cnds->ds_dir->dd_phys->dd_head_dataset_obj;
2786 VERIFY3U(zap_remove_int(dp->dp_meta_objset,
2787 odd->dd_phys->dd_clones, o, tx), ==, 0);
2788 VERIFY3U(zap_add_int(dp->dp_meta_objset,
2789 dd->dd_phys->dd_clones, o, tx), ==, 0);
2790 dsl_dataset_rele(cnds, FTAG);
2792 zap_cursor_fini(&zc);
2795 ASSERT3U(dsl_prop_numcb(ds), ==, 0);
2799 * Change space accounting.
2800 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2801 * both be valid, or both be 0 (resulting in delta == 0). This
2802 * is true for each of {clone,origin} independently.
2805 delta = pa->cloneusedsnap -
2806 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2807 ASSERT3S(delta, >=, 0);
2808 ASSERT3U(pa->used, >=, delta);
2809 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2810 dsl_dir_diduse_space(dd, DD_USED_HEAD,
2811 pa->used - delta, pa->comp, pa->uncomp, tx);
2813 delta = pa->originusedsnap -
2814 odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2815 ASSERT3S(delta, <=, 0);
2816 ASSERT3U(pa->used, >=, -delta);
2817 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2818 dsl_dir_diduse_space(odd, DD_USED_HEAD,
2819 -pa->used - delta, -pa->comp, -pa->uncomp, tx);
2821 origin_ds->ds_phys->ds_unique_bytes = pa->unique;
2823 /* log history record */
2824 spa_history_log_internal(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx,
2825 "dataset = %llu", hds->ds_object);
2827 dsl_dir_close(odd, FTAG);
2830 static char *snaplist_tag = "snaplist";
2832 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2833 * (exclusive) and last_obj (inclusive). The list will be in reverse
2834 * order (last_obj will be the list_head()). If first_obj == 0, do all
2835 * snapshots back to this dataset's origin.
2838 snaplist_make(dsl_pool_t *dp, boolean_t own,
2839 uint64_t first_obj, uint64_t last_obj, list_t *l)
2841 uint64_t obj = last_obj;
2843 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock));
2845 list_create(l, sizeof (struct promotenode),
2846 offsetof(struct promotenode, link));
2848 while (obj != first_obj) {
2850 struct promotenode *snap;
2854 err = dsl_dataset_own_obj(dp, obj,
2855 0, snaplist_tag, &ds);
2857 dsl_dataset_make_exclusive(ds, snaplist_tag);
2859 err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds);
2861 if (err == ENOENT) {
2862 /* lost race with snapshot destroy */
2863 struct promotenode *last = list_tail(l);
2864 ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj);
2865 obj = last->ds->ds_phys->ds_prev_snap_obj;
2872 first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2874 snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP);
2876 list_insert_tail(l, snap);
2877 obj = ds->ds_phys->ds_prev_snap_obj;
2884 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2886 struct promotenode *snap;
2889 for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2890 uint64_t used, comp, uncomp;
2891 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2892 mintxg, UINT64_MAX, &used, &comp, &uncomp);
2899 snaplist_destroy(list_t *l, boolean_t own)
2901 struct promotenode *snap;
2903 if (!l || !list_link_active(&l->list_head))
2906 while ((snap = list_tail(l)) != NULL) {
2907 list_remove(l, snap);
2909 dsl_dataset_disown(snap->ds, snaplist_tag);
2911 dsl_dataset_rele(snap->ds, snaplist_tag);
2912 kmem_free(snap, sizeof (struct promotenode));
2918 * Promote a clone. Nomenclature note:
2919 * "clone" or "cds": the original clone which is being promoted
2920 * "origin" or "ods": the snapshot which is originally clone's origin
2921 * "origin head" or "ohds": the dataset which is the head
2922 * (filesystem/volume) for the origin
2923 * "origin origin": the origin of the origin's filesystem (typically
2924 * NULL, indicating that the clone is not a clone of a clone).
2927 dsl_dataset_promote(const char *name, char *conflsnap)
2932 dmu_object_info_t doi;
2933 struct promotearg pa = { 0 };
2934 struct promotenode *snap;
2937 err = dsl_dataset_hold(name, FTAG, &ds);
2943 err = dmu_object_info(dp->dp_meta_objset,
2944 ds->ds_phys->ds_snapnames_zapobj, &doi);
2946 dsl_dataset_rele(ds, FTAG);
2950 if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) {
2951 dsl_dataset_rele(ds, FTAG);
2956 * We are going to inherit all the snapshots taken before our
2957 * origin (i.e., our new origin will be our parent's origin).
2958 * Take ownership of them so that we can rename them into our
2961 rw_enter(&dp->dp_config_rwlock, RW_READER);
2963 err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj,
2968 err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps);
2972 snap = list_head(&pa.shared_snaps);
2973 ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
2974 err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj,
2975 snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps);
2979 if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) {
2980 err = dsl_dataset_hold_obj(dp,
2981 snap->ds->ds_dir->dd_phys->dd_origin_obj,
2982 FTAG, &pa.origin_origin);
2988 rw_exit(&dp->dp_config_rwlock);
2991 * Add in 128x the snapnames zapobj size, since we will be moving
2992 * a bunch of snapnames to the promoted ds, and dirtying their
2996 err = dsl_sync_task_do(dp, dsl_dataset_promote_check,
2997 dsl_dataset_promote_sync, ds, &pa,
2998 2 + 2 * doi.doi_physical_blocks_512);
2999 if (err && pa.err_ds && conflsnap)
3000 (void) strncpy(conflsnap, pa.err_ds, MAXNAMELEN);
3003 snaplist_destroy(&pa.shared_snaps, B_TRUE);
3004 snaplist_destroy(&pa.clone_snaps, B_FALSE);
3005 snaplist_destroy(&pa.origin_snaps, B_FALSE);
3006 if (pa.origin_origin)
3007 dsl_dataset_rele(pa.origin_origin, FTAG);
3008 dsl_dataset_rele(ds, FTAG);
3012 struct cloneswaparg {
3013 dsl_dataset_t *cds; /* clone dataset */
3014 dsl_dataset_t *ohds; /* origin's head dataset */
3016 int64_t unused_refres_delta; /* change in unconsumed refreservation */
3021 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx)
3023 struct cloneswaparg *csa = arg1;
3025 /* they should both be heads */
3026 if (dsl_dataset_is_snapshot(csa->cds) ||
3027 dsl_dataset_is_snapshot(csa->ohds))
3030 /* the branch point should be just before them */
3031 if (csa->cds->ds_prev != csa->ohds->ds_prev)
3034 /* cds should be the clone (unless they are unrelated) */
3035 if (csa->cds->ds_prev != NULL &&
3036 csa->cds->ds_prev != csa->cds->ds_dir->dd_pool->dp_origin_snap &&
3037 csa->ohds->ds_object !=
3038 csa->cds->ds_prev->ds_phys->ds_next_snap_obj)
3041 /* the clone should be a child of the origin */
3042 if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir)
3045 /* ohds shouldn't be modified unless 'force' */
3046 if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds))
3049 /* adjust amount of any unconsumed refreservation */
3050 csa->unused_refres_delta =
3051 (int64_t)MIN(csa->ohds->ds_reserved,
3052 csa->ohds->ds_phys->ds_unique_bytes) -
3053 (int64_t)MIN(csa->ohds->ds_reserved,
3054 csa->cds->ds_phys->ds_unique_bytes);
3056 if (csa->unused_refres_delta > 0 &&
3057 csa->unused_refres_delta >
3058 dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE))
3061 if (csa->ohds->ds_quota != 0 &&
3062 csa->cds->ds_phys->ds_unique_bytes > csa->ohds->ds_quota)
3070 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3072 struct cloneswaparg *csa = arg1;
3073 dsl_pool_t *dp = csa->cds->ds_dir->dd_pool;
3075 ASSERT(csa->cds->ds_reserved == 0);
3076 ASSERT(csa->ohds->ds_quota == 0 ||
3077 csa->cds->ds_phys->ds_unique_bytes <= csa->ohds->ds_quota);
3079 dmu_buf_will_dirty(csa->cds->ds_dbuf, tx);
3080 dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx);
3082 if (csa->cds->ds_objset != NULL) {
3083 dmu_objset_evict(csa->cds->ds_objset);
3084 csa->cds->ds_objset = NULL;
3087 if (csa->ohds->ds_objset != NULL) {
3088 dmu_objset_evict(csa->ohds->ds_objset);
3089 csa->ohds->ds_objset = NULL;
3093 * Reset origin's unique bytes, if it exists.
3095 if (csa->cds->ds_prev) {
3096 dsl_dataset_t *origin = csa->cds->ds_prev;
3097 uint64_t comp, uncomp;
3099 dmu_buf_will_dirty(origin->ds_dbuf, tx);
3100 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3101 origin->ds_phys->ds_prev_snap_txg, UINT64_MAX,
3102 &origin->ds_phys->ds_unique_bytes, &comp, &uncomp);
3108 tmp = csa->ohds->ds_phys->ds_bp;
3109 csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp;
3110 csa->cds->ds_phys->ds_bp = tmp;
3113 /* set dd_*_bytes */
3115 int64_t dused, dcomp, duncomp;
3116 uint64_t cdl_used, cdl_comp, cdl_uncomp;
3117 uint64_t odl_used, odl_comp, odl_uncomp;
3119 ASSERT3U(csa->cds->ds_dir->dd_phys->
3120 dd_used_breakdown[DD_USED_SNAP], ==, 0);
3122 dsl_deadlist_space(&csa->cds->ds_deadlist,
3123 &cdl_used, &cdl_comp, &cdl_uncomp);
3124 dsl_deadlist_space(&csa->ohds->ds_deadlist,
3125 &odl_used, &odl_comp, &odl_uncomp);
3127 dused = csa->cds->ds_phys->ds_used_bytes + cdl_used -
3128 (csa->ohds->ds_phys->ds_used_bytes + odl_used);
3129 dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp -
3130 (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp);
3131 duncomp = csa->cds->ds_phys->ds_uncompressed_bytes +
3133 (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp);
3135 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD,
3136 dused, dcomp, duncomp, tx);
3137 dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD,
3138 -dused, -dcomp, -duncomp, tx);
3141 * The difference in the space used by snapshots is the
3142 * difference in snapshot space due to the head's
3143 * deadlist (since that's the only thing that's
3144 * changing that affects the snapused).
3146 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3147 csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3148 &cdl_used, &cdl_comp, &cdl_uncomp);
3149 dsl_deadlist_space_range(&csa->ohds->ds_deadlist,
3150 csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3151 &odl_used, &odl_comp, &odl_uncomp);
3152 dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used,
3153 DD_USED_HEAD, DD_USED_SNAP, tx);
3156 /* swap ds_*_bytes */
3157 SWITCH64(csa->ohds->ds_phys->ds_used_bytes,
3158 csa->cds->ds_phys->ds_used_bytes);
3159 SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes,
3160 csa->cds->ds_phys->ds_compressed_bytes);
3161 SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes,
3162 csa->cds->ds_phys->ds_uncompressed_bytes);
3163 SWITCH64(csa->ohds->ds_phys->ds_unique_bytes,
3164 csa->cds->ds_phys->ds_unique_bytes);
3166 /* apply any parent delta for change in unconsumed refreservation */
3167 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV,
3168 csa->unused_refres_delta, 0, 0, tx);
3173 dsl_deadlist_close(&csa->cds->ds_deadlist);
3174 dsl_deadlist_close(&csa->ohds->ds_deadlist);
3175 SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj,
3176 csa->cds->ds_phys->ds_deadlist_obj);
3177 dsl_deadlist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset,
3178 csa->cds->ds_phys->ds_deadlist_obj);
3179 dsl_deadlist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset,
3180 csa->ohds->ds_phys->ds_deadlist_obj);
3182 dsl_scan_ds_clone_swapped(csa->ohds, csa->cds, tx);
3186 * Swap 'clone' with its origin head datasets. Used at the end of "zfs
3187 * recv" into an existing fs to swizzle the file system to the new
3188 * version, and by "zfs rollback". Can also be used to swap two
3189 * independent head datasets if neither has any snapshots.
3192 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
3195 struct cloneswaparg csa;
3198 ASSERT(clone->ds_owner);
3199 ASSERT(origin_head->ds_owner);
3202 * Need exclusive access for the swap. If we're swapping these
3203 * datasets back after an error, we already hold the locks.
3205 if (!RW_WRITE_HELD(&clone->ds_rwlock))
3206 rw_enter(&clone->ds_rwlock, RW_WRITER);
3207 if (!RW_WRITE_HELD(&origin_head->ds_rwlock) &&
3208 !rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) {
3209 rw_exit(&clone->ds_rwlock);
3210 rw_enter(&origin_head->ds_rwlock, RW_WRITER);
3211 if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) {
3212 rw_exit(&origin_head->ds_rwlock);
3217 csa.ohds = origin_head;
3219 error = dsl_sync_task_do(clone->ds_dir->dd_pool,
3220 dsl_dataset_clone_swap_check,
3221 dsl_dataset_clone_swap_sync, &csa, NULL, 9);
3226 * Given a pool name and a dataset object number in that pool,
3227 * return the name of that dataset.
3230 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3237 if ((error = spa_open(pname, &spa, FTAG)) != 0)
3239 dp = spa_get_dsl(spa);
3240 rw_enter(&dp->dp_config_rwlock, RW_READER);
3241 if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) {
3242 dsl_dataset_name(ds, buf);
3243 dsl_dataset_rele(ds, FTAG);
3245 rw_exit(&dp->dp_config_rwlock);
3246 spa_close(spa, FTAG);
3252 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3253 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3257 ASSERT3S(asize, >, 0);
3260 * *ref_rsrv is the portion of asize that will come from any
3261 * unconsumed refreservation space.
3265 mutex_enter(&ds->ds_lock);
3267 * Make a space adjustment for reserved bytes.
3269 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
3271 ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3272 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3274 asize - MIN(asize, parent_delta(ds, asize + inflight));
3277 if (!check_quota || ds->ds_quota == 0) {
3278 mutex_exit(&ds->ds_lock);
3282 * If they are requesting more space, and our current estimate
3283 * is over quota, they get to try again unless the actual
3284 * on-disk is over quota and there are no pending changes (which
3285 * may free up space for us).
3287 if (ds->ds_phys->ds_used_bytes + inflight >= ds->ds_quota) {
3288 if (inflight > 0 || ds->ds_phys->ds_used_bytes < ds->ds_quota)
3293 mutex_exit(&ds->ds_lock);
3300 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
3302 dsl_dataset_t *ds = arg1;
3303 dsl_prop_setarg_t *psa = arg2;
3306 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA)
3309 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3312 if (psa->psa_effective_value == 0)
3315 if (psa->psa_effective_value < ds->ds_phys->ds_used_bytes ||
3316 psa->psa_effective_value < ds->ds_reserved)
3322 extern void dsl_prop_set_sync(void *, void *, dmu_tx_t *);
3325 dsl_dataset_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3327 dsl_dataset_t *ds = arg1;
3328 dsl_prop_setarg_t *psa = arg2;
3329 uint64_t effective_value = psa->psa_effective_value;
3331 dsl_prop_set_sync(ds, psa, tx);
3332 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3334 if (ds->ds_quota != effective_value) {
3335 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3336 ds->ds_quota = effective_value;
3338 spa_history_log_internal(LOG_DS_REFQUOTA,
3339 ds->ds_dir->dd_pool->dp_spa, tx, "%lld dataset = %llu ",
3340 (longlong_t)ds->ds_quota, ds->ds_object);
3345 dsl_dataset_set_quota(const char *dsname, zprop_source_t source, uint64_t quota)
3348 dsl_prop_setarg_t psa;
3351 dsl_prop_setarg_init_uint64(&psa, "refquota", source, "a);
3353 err = dsl_dataset_hold(dsname, FTAG, &ds);
3358 * If someone removes a file, then tries to set the quota, we
3359 * want to make sure the file freeing takes effect.
3361 txg_wait_open(ds->ds_dir->dd_pool, 0);
3363 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3364 dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync,
3367 dsl_dataset_rele(ds, FTAG);
3372 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
3374 dsl_dataset_t *ds = arg1;
3375 dsl_prop_setarg_t *psa = arg2;
3376 uint64_t effective_value;
3380 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
3381 SPA_VERSION_REFRESERVATION)
3384 if (dsl_dataset_is_snapshot(ds))
3387 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3390 effective_value = psa->psa_effective_value;
3393 * If we are doing the preliminary check in open context, the
3394 * space estimates may be inaccurate.
3396 if (!dmu_tx_is_syncing(tx))
3399 mutex_enter(&ds->ds_lock);
3400 if (!DS_UNIQUE_IS_ACCURATE(ds))
3401 dsl_dataset_recalc_head_uniq(ds);
3402 unique = ds->ds_phys->ds_unique_bytes;
3403 mutex_exit(&ds->ds_lock);
3405 if (MAX(unique, effective_value) > MAX(unique, ds->ds_reserved)) {
3406 uint64_t delta = MAX(unique, effective_value) -
3407 MAX(unique, ds->ds_reserved);
3409 if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
3411 if (ds->ds_quota > 0 &&
3412 effective_value > ds->ds_quota)
3420 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3422 dsl_dataset_t *ds = arg1;
3423 dsl_prop_setarg_t *psa = arg2;
3424 uint64_t effective_value = psa->psa_effective_value;
3428 dsl_prop_set_sync(ds, psa, tx);
3429 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3431 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3433 mutex_enter(&ds->ds_dir->dd_lock);
3434 mutex_enter(&ds->ds_lock);
3435 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3436 unique = ds->ds_phys->ds_unique_bytes;
3437 delta = MAX(0, (int64_t)(effective_value - unique)) -
3438 MAX(0, (int64_t)(ds->ds_reserved - unique));
3439 ds->ds_reserved = effective_value;
3440 mutex_exit(&ds->ds_lock);
3442 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3443 mutex_exit(&ds->ds_dir->dd_lock);
3445 spa_history_log_internal(LOG_DS_REFRESERV,
3446 ds->ds_dir->dd_pool->dp_spa, tx, "%lld dataset = %llu",
3447 (longlong_t)effective_value, ds->ds_object);
3451 dsl_dataset_set_reservation(const char *dsname, zprop_source_t source,
3452 uint64_t reservation)
3455 dsl_prop_setarg_t psa;
3458 dsl_prop_setarg_init_uint64(&psa, "refreservation", source,
3461 err = dsl_dataset_hold(dsname, FTAG, &ds);
3465 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3466 dsl_dataset_set_reservation_check,
3467 dsl_dataset_set_reservation_sync, ds, &psa, 0);
3469 dsl_dataset_rele(ds, FTAG);
3473 typedef struct zfs_hold_cleanup_arg {
3476 char htag[MAXNAMELEN];
3477 } zfs_hold_cleanup_arg_t;
3480 dsl_dataset_user_release_onexit(void *arg)
3482 zfs_hold_cleanup_arg_t *ca = arg;
3484 (void) dsl_dataset_user_release_tmp(ca->dp, ca->dsobj, ca->htag,
3486 kmem_free(ca, sizeof (zfs_hold_cleanup_arg_t));
3490 dsl_register_onexit_hold_cleanup(dsl_dataset_t *ds, const char *htag,
3493 zfs_hold_cleanup_arg_t *ca;
3495 ca = kmem_alloc(sizeof (zfs_hold_cleanup_arg_t), KM_SLEEP);
3496 ca->dp = ds->ds_dir->dd_pool;
3497 ca->dsobj = ds->ds_object;
3498 (void) strlcpy(ca->htag, htag, sizeof (ca->htag));
3499 VERIFY3U(0, ==, zfs_onexit_add_cb(minor,
3500 dsl_dataset_user_release_onexit, ca, NULL));
3504 * If you add new checks here, you may need to add
3505 * additional checks to the "temporary" case in
3506 * snapshot_check() in dmu_objset.c.
3509 dsl_dataset_user_hold_check(void *arg1, void *arg2, dmu_tx_t *tx)
3511 dsl_dataset_t *ds = arg1;
3512 struct dsl_ds_holdarg *ha = arg2;
3513 char *htag = ha->htag;
3514 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3517 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3520 if (!dsl_dataset_is_snapshot(ds))
3523 /* tags must be unique */
3524 mutex_enter(&ds->ds_lock);
3525 if (ds->ds_phys->ds_userrefs_obj) {
3526 error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj, htag,
3530 else if (error == ENOENT)
3533 mutex_exit(&ds->ds_lock);
3535 if (error == 0 && ha->temphold &&
3536 strlen(htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN)
3543 dsl_dataset_user_hold_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3545 dsl_dataset_t *ds = arg1;
3546 struct dsl_ds_holdarg *ha = arg2;
3547 char *htag = ha->htag;
3548 dsl_pool_t *dp = ds->ds_dir->dd_pool;
3549 objset_t *mos = dp->dp_meta_objset;
3550 uint64_t now = gethrestime_sec();
3553 mutex_enter(&ds->ds_lock);
3554 if (ds->ds_phys->ds_userrefs_obj == 0) {
3556 * This is the first user hold for this dataset. Create
3557 * the userrefs zap object.
3559 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3560 zapobj = ds->ds_phys->ds_userrefs_obj =
3561 zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx);
3563 zapobj = ds->ds_phys->ds_userrefs_obj;
3566 mutex_exit(&ds->ds_lock);
3568 VERIFY(0 == zap_add(mos, zapobj, htag, 8, 1, &now, tx));
3571 VERIFY(0 == dsl_pool_user_hold(dp, ds->ds_object,
3575 spa_history_log_internal(LOG_DS_USER_HOLD,
3576 dp->dp_spa, tx, "<%s> temp = %d dataset = %llu", htag,
3577 (int)ha->temphold, ds->ds_object);
3581 dsl_dataset_user_hold_one(const char *dsname, void *arg)
3583 struct dsl_ds_holdarg *ha = arg;
3588 /* alloc a buffer to hold dsname@snapname plus terminating NULL */
3589 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3590 error = dsl_dataset_hold(name, ha->dstg, &ds);
3593 ha->gotone = B_TRUE;
3594 dsl_sync_task_create(ha->dstg, dsl_dataset_user_hold_check,
3595 dsl_dataset_user_hold_sync, ds, ha, 0);
3596 } else if (error == ENOENT && ha->recursive) {
3599 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3605 dsl_dataset_user_hold_for_send(dsl_dataset_t *ds, char *htag,
3608 struct dsl_ds_holdarg *ha;
3611 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3613 ha->temphold = temphold;
3614 error = dsl_sync_task_do(ds->ds_dir->dd_pool,
3615 dsl_dataset_user_hold_check, dsl_dataset_user_hold_sync,
3617 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3623 dsl_dataset_user_hold(char *dsname, char *snapname, char *htag,
3624 boolean_t recursive, boolean_t temphold, int cleanup_fd)
3626 struct dsl_ds_holdarg *ha;
3627 dsl_sync_task_t *dst;
3632 if (cleanup_fd != -1) {
3633 /* Currently we only support cleanup-on-exit of tempholds. */
3636 error = zfs_onexit_fd_hold(cleanup_fd, &minor);
3641 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3643 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3645 error = spa_open(dsname, &spa, FTAG);
3647 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3648 if (cleanup_fd != -1)
3649 zfs_onexit_fd_rele(cleanup_fd);
3653 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3655 ha->snapname = snapname;
3656 ha->recursive = recursive;
3657 ha->temphold = temphold;
3660 error = dmu_objset_find(dsname, dsl_dataset_user_hold_one,
3661 ha, DS_FIND_CHILDREN);
3663 error = dsl_dataset_user_hold_one(dsname, ha);
3666 error = dsl_sync_task_group_wait(ha->dstg);
3668 for (dst = list_head(&ha->dstg->dstg_tasks); dst;
3669 dst = list_next(&ha->dstg->dstg_tasks, dst)) {
3670 dsl_dataset_t *ds = dst->dst_arg1;
3673 dsl_dataset_name(ds, ha->failed);
3674 *strchr(ha->failed, '@') = '\0';
3675 } else if (error == 0 && minor != 0 && temphold) {
3677 * If this hold is to be released upon process exit,
3678 * register that action now.
3680 dsl_register_onexit_hold_cleanup(ds, htag, minor);
3682 dsl_dataset_rele(ds, ha->dstg);
3685 if (error == 0 && recursive && !ha->gotone)
3689 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
3691 dsl_sync_task_group_destroy(ha->dstg);
3693 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3694 spa_close(spa, FTAG);
3695 if (cleanup_fd != -1)
3696 zfs_onexit_fd_rele(cleanup_fd);
3700 struct dsl_ds_releasearg {
3703 boolean_t own; /* do we own or just hold ds? */
3707 dsl_dataset_release_might_destroy(dsl_dataset_t *ds, const char *htag,
3708 boolean_t *might_destroy)
3710 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3715 *might_destroy = B_FALSE;
3717 mutex_enter(&ds->ds_lock);
3718 zapobj = ds->ds_phys->ds_userrefs_obj;
3720 /* The tag can't possibly exist */
3721 mutex_exit(&ds->ds_lock);
3725 /* Make sure the tag exists */
3726 error = zap_lookup(mos, zapobj, htag, 8, 1, &tmp);
3728 mutex_exit(&ds->ds_lock);
3729 if (error == ENOENT)
3734 if (ds->ds_userrefs == 1 && ds->ds_phys->ds_num_children == 1 &&
3735 DS_IS_DEFER_DESTROY(ds))
3736 *might_destroy = B_TRUE;
3738 mutex_exit(&ds->ds_lock);
3743 dsl_dataset_user_release_check(void *arg1, void *tag, dmu_tx_t *tx)
3745 struct dsl_ds_releasearg *ra = arg1;
3746 dsl_dataset_t *ds = ra->ds;
3747 boolean_t might_destroy;
3750 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3753 error = dsl_dataset_release_might_destroy(ds, ra->htag, &might_destroy);
3757 if (might_destroy) {
3758 struct dsl_ds_destroyarg dsda = {0};
3760 if (dmu_tx_is_syncing(tx)) {
3762 * If we're not prepared to remove the snapshot,
3763 * we can't allow the release to happen right now.
3769 dsda.releasing = B_TRUE;
3770 return (dsl_dataset_destroy_check(&dsda, tag, tx));
3777 dsl_dataset_user_release_sync(void *arg1, void *tag, dmu_tx_t *tx)
3779 struct dsl_ds_releasearg *ra = arg1;
3780 dsl_dataset_t *ds = ra->ds;
3781 dsl_pool_t *dp = ds->ds_dir->dd_pool;
3782 objset_t *mos = dp->dp_meta_objset;
3784 uint64_t dsobj = ds->ds_object;
3788 mutex_enter(&ds->ds_lock);
3790 refs = ds->ds_userrefs;
3791 mutex_exit(&ds->ds_lock);
3792 error = dsl_pool_user_release(dp, ds->ds_object, ra->htag, tx);
3793 VERIFY(error == 0 || error == ENOENT);
3794 zapobj = ds->ds_phys->ds_userrefs_obj;
3795 VERIFY(0 == zap_remove(mos, zapobj, ra->htag, tx));
3796 if (ds->ds_userrefs == 0 && ds->ds_phys->ds_num_children == 1 &&
3797 DS_IS_DEFER_DESTROY(ds)) {
3798 struct dsl_ds_destroyarg dsda = {0};
3802 dsda.releasing = B_TRUE;
3803 /* We already did the destroy_check */
3804 dsl_dataset_destroy_sync(&dsda, tag, tx);
3807 spa_history_log_internal(LOG_DS_USER_RELEASE,
3808 dp->dp_spa, tx, "<%s> %lld dataset = %llu",
3809 ra->htag, (longlong_t)refs, dsobj);
3813 dsl_dataset_user_release_one(const char *dsname, void *arg)
3815 struct dsl_ds_holdarg *ha = arg;
3816 struct dsl_ds_releasearg *ra;
3819 void *dtag = ha->dstg;
3821 boolean_t own = B_FALSE;
3822 boolean_t might_destroy;
3824 /* alloc a buffer to hold dsname@snapname, plus the terminating NULL */
3825 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3826 error = dsl_dataset_hold(name, dtag, &ds);
3828 if (error == ENOENT && ha->recursive)
3830 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3834 ha->gotone = B_TRUE;
3836 ASSERT(dsl_dataset_is_snapshot(ds));
3838 error = dsl_dataset_release_might_destroy(ds, ha->htag, &might_destroy);
3840 dsl_dataset_rele(ds, dtag);
3844 if (might_destroy) {
3846 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3847 error = zfs_unmount_snap(name, NULL);
3850 dsl_dataset_rele(ds, dtag);
3854 if (!dsl_dataset_tryown(ds, B_TRUE, dtag)) {
3855 dsl_dataset_rele(ds, dtag);
3859 dsl_dataset_make_exclusive(ds, dtag);
3863 ra = kmem_alloc(sizeof (struct dsl_ds_releasearg), KM_SLEEP);
3865 ra->htag = ha->htag;
3867 dsl_sync_task_create(ha->dstg, dsl_dataset_user_release_check,
3868 dsl_dataset_user_release_sync, ra, dtag, 0);
3874 dsl_dataset_user_release(char *dsname, char *snapname, char *htag,
3875 boolean_t recursive)
3877 struct dsl_ds_holdarg *ha;
3878 dsl_sync_task_t *dst;
3883 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3885 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3887 error = spa_open(dsname, &spa, FTAG);
3889 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3893 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3895 ha->snapname = snapname;
3896 ha->recursive = recursive;
3898 error = dmu_objset_find(dsname, dsl_dataset_user_release_one,
3899 ha, DS_FIND_CHILDREN);
3901 error = dsl_dataset_user_release_one(dsname, ha);
3904 error = dsl_sync_task_group_wait(ha->dstg);
3906 for (dst = list_head(&ha->dstg->dstg_tasks); dst;
3907 dst = list_next(&ha->dstg->dstg_tasks, dst)) {
3908 struct dsl_ds_releasearg *ra = dst->dst_arg1;
3909 dsl_dataset_t *ds = ra->ds;
3912 dsl_dataset_name(ds, ha->failed);
3915 dsl_dataset_disown(ds, ha->dstg);
3917 dsl_dataset_rele(ds, ha->dstg);
3919 kmem_free(ra, sizeof (struct dsl_ds_releasearg));
3922 if (error == 0 && recursive && !ha->gotone)
3925 if (error && error != EBUSY)
3926 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
3928 dsl_sync_task_group_destroy(ha->dstg);
3929 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3930 spa_close(spa, FTAG);
3933 * We can get EBUSY if we were racing with deferred destroy and
3934 * dsl_dataset_user_release_check() hadn't done the necessary
3935 * open context setup. We can also get EBUSY if we're racing
3936 * with destroy and that thread is the ds_owner. Either way
3937 * the busy condition should be transient, and we should retry
3938 * the release operation.
3947 * Called at spa_load time (with retry == B_FALSE) to release a stale
3948 * temporary user hold. Also called by the onexit code (with retry == B_TRUE).
3951 dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, char *htag,
3961 rw_enter(&dp->dp_config_rwlock, RW_READER);
3962 error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
3963 rw_exit(&dp->dp_config_rwlock);
3966 namelen = dsl_dataset_namelen(ds)+1;
3967 name = kmem_alloc(namelen, KM_SLEEP);
3968 dsl_dataset_name(ds, name);
3969 dsl_dataset_rele(ds, FTAG);
3971 snap = strchr(name, '@');
3974 error = dsl_dataset_user_release(name, snap, htag, B_FALSE);
3975 kmem_free(name, namelen);
3978 * The object can't have been destroyed because we have a hold,
3979 * but it might have been renamed, resulting in ENOENT. Retry
3980 * if we've been requested to do so.
3982 * It would be nice if we could use the dsobj all the way
3983 * through and avoid ENOENT entirely. But we might need to
3984 * unmount the snapshot, and there's currently no way to lookup
3985 * a vfsp using a ZFS object id.
3987 } while ((error == ENOENT) && retry);
3993 dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp)
3998 err = dsl_dataset_hold(dsname, FTAG, &ds);
4002 VERIFY(0 == nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP));
4003 if (ds->ds_phys->ds_userrefs_obj != 0) {
4004 zap_attribute_t *za;
4007 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
4008 for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset,
4009 ds->ds_phys->ds_userrefs_obj);
4010 zap_cursor_retrieve(&zc, za) == 0;
4011 zap_cursor_advance(&zc)) {
4012 VERIFY(0 == nvlist_add_uint64(*nvp, za->za_name,
4013 za->za_first_integer));
4015 zap_cursor_fini(&zc);
4016 kmem_free(za, sizeof (zap_attribute_t));
4018 dsl_dataset_rele(ds, FTAG);
4023 * Note, this fuction is used as the callback for dmu_objset_find(). We
4024 * always return 0 so that we will continue to find and process
4025 * inconsistent datasets, even if we encounter an error trying to
4026 * process one of them.
4030 dsl_destroy_inconsistent(const char *dsname, void *arg)
4034 if (dsl_dataset_own(dsname, B_TRUE, FTAG, &ds) == 0) {
4035 if (DS_IS_INCONSISTENT(ds))
4036 (void) dsl_dataset_destroy(ds, FTAG, B_FALSE);
4038 dsl_dataset_disown(ds, FTAG);