4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012 by Delphix. All rights reserved.
24 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
25 * Copyright (c) 2011 Pawel Jakub Dawidek <pawel@dawidek.net>.
26 * All rights reserved.
27 * Portions Copyright (c) 2011 Martin Matuska <mm@FreeBSD.org>
30 #include <sys/dmu_objset.h>
31 #include <sys/dsl_dataset.h>
32 #include <sys/dsl_dir.h>
33 #include <sys/dsl_prop.h>
34 #include <sys/dsl_synctask.h>
35 #include <sys/dmu_traverse.h>
36 #include <sys/dmu_impl.h>
37 #include <sys/dmu_tx.h>
41 #include <sys/zfeature.h>
42 #include <sys/unique.h>
43 #include <sys/zfs_context.h>
44 #include <sys/zfs_ioctl.h>
46 #include <sys/zfs_znode.h>
47 #include <sys/zfs_onexit.h>
49 #include <sys/dsl_scan.h>
50 #include <sys/dsl_deadlist.h>
52 static char *dsl_reaper = "the grim reaper";
54 static dsl_checkfunc_t dsl_dataset_destroy_begin_check;
55 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync;
56 static dsl_syncfunc_t dsl_dataset_set_reservation_sync;
58 #define SWITCH64(x, y) \
60 uint64_t __tmp = (x); \
65 #define DS_REF_MAX (1ULL << 62)
67 #define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE
69 #define DSL_DATASET_IS_DESTROYED(ds) ((ds)->ds_owner == dsl_reaper)
73 * Figure out how much of this delta should be propogated to the dsl_dir
74 * layer. If there's a refreservation, that space has already been
75 * partially accounted for in our ancestors.
78 parent_delta(dsl_dataset_t *ds, int64_t delta)
80 uint64_t old_bytes, new_bytes;
82 if (ds->ds_reserved == 0)
85 old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
86 new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
88 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
89 return (new_bytes - old_bytes);
93 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
95 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
96 int compressed = BP_GET_PSIZE(bp);
97 int uncompressed = BP_GET_UCSIZE(bp);
100 dprintf_bp(bp, "ds=%p", ds);
102 ASSERT(dmu_tx_is_syncing(tx));
103 /* It could have been compressed away to nothing */
106 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
107 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
109 dsl_pool_mos_diduse_space(tx->tx_pool,
110 used, compressed, uncompressed);
113 dmu_buf_will_dirty(ds->ds_dbuf, tx);
115 mutex_enter(&ds->ds_dir->dd_lock);
116 mutex_enter(&ds->ds_lock);
117 delta = parent_delta(ds, used);
118 ds->ds_phys->ds_referenced_bytes += used;
119 ds->ds_phys->ds_compressed_bytes += compressed;
120 ds->ds_phys->ds_uncompressed_bytes += uncompressed;
121 ds->ds_phys->ds_unique_bytes += used;
122 mutex_exit(&ds->ds_lock);
123 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
124 compressed, uncompressed, tx);
125 dsl_dir_transfer_space(ds->ds_dir, used - delta,
126 DD_USED_REFRSRV, DD_USED_HEAD, tx);
127 mutex_exit(&ds->ds_dir->dd_lock);
131 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
137 ASSERT(dmu_tx_is_syncing(tx));
138 ASSERT(bp->blk_birth <= tx->tx_txg);
140 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
141 int compressed = BP_GET_PSIZE(bp);
142 int uncompressed = BP_GET_UCSIZE(bp);
146 dsl_free(tx->tx_pool, tx->tx_txg, bp);
147 dsl_pool_mos_diduse_space(tx->tx_pool,
148 -used, -compressed, -uncompressed);
151 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
153 ASSERT(!dsl_dataset_is_snapshot(ds));
154 dmu_buf_will_dirty(ds->ds_dbuf, tx);
156 if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
159 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
160 dsl_free(tx->tx_pool, tx->tx_txg, bp);
162 mutex_enter(&ds->ds_dir->dd_lock);
163 mutex_enter(&ds->ds_lock);
164 ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
165 !DS_UNIQUE_IS_ACCURATE(ds));
166 delta = parent_delta(ds, -used);
167 ds->ds_phys->ds_unique_bytes -= used;
168 mutex_exit(&ds->ds_lock);
169 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
170 delta, -compressed, -uncompressed, tx);
171 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
172 DD_USED_REFRSRV, DD_USED_HEAD, tx);
173 mutex_exit(&ds->ds_dir->dd_lock);
175 dprintf_bp(bp, "putting on dead list: %s", "");
178 * We are here as part of zio's write done callback,
179 * which means we're a zio interrupt thread. We can't
180 * call dsl_deadlist_insert() now because it may block
181 * waiting for I/O. Instead, put bp on the deferred
182 * queue and let dsl_pool_sync() finish the job.
184 bplist_append(&ds->ds_pending_deadlist, bp);
186 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
188 ASSERT3U(ds->ds_prev->ds_object, ==,
189 ds->ds_phys->ds_prev_snap_obj);
190 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
191 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
192 if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
193 ds->ds_object && bp->blk_birth >
194 ds->ds_prev->ds_phys->ds_prev_snap_txg) {
195 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
196 mutex_enter(&ds->ds_prev->ds_lock);
197 ds->ds_prev->ds_phys->ds_unique_bytes += used;
198 mutex_exit(&ds->ds_prev->ds_lock);
200 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
201 dsl_dir_transfer_space(ds->ds_dir, used,
202 DD_USED_HEAD, DD_USED_SNAP, tx);
205 mutex_enter(&ds->ds_lock);
206 ASSERT3U(ds->ds_phys->ds_referenced_bytes, >=, used);
207 ds->ds_phys->ds_referenced_bytes -= used;
208 ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
209 ds->ds_phys->ds_compressed_bytes -= compressed;
210 ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
211 ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
212 mutex_exit(&ds->ds_lock);
218 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
220 uint64_t trysnap = 0;
225 * The snapshot creation could fail, but that would cause an
226 * incorrect FALSE return, which would only result in an
227 * overestimation of the amount of space that an operation would
228 * consume, which is OK.
230 * There's also a small window where we could miss a pending
231 * snapshot, because we could set the sync task in the quiescing
232 * phase. So this should only be used as a guess.
234 if (ds->ds_trysnap_txg >
235 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
236 trysnap = ds->ds_trysnap_txg;
237 return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
241 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
244 if (blk_birth <= dsl_dataset_prev_snap_txg(ds))
247 ddt_prefetch(dsl_dataset_get_spa(ds), bp);
254 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
256 dsl_dataset_t *ds = dsv;
258 ASSERT(ds->ds_owner == NULL || DSL_DATASET_IS_DESTROYED(ds));
260 unique_remove(ds->ds_fsid_guid);
262 if (ds->ds_objset != NULL)
263 dmu_objset_evict(ds->ds_objset);
266 dsl_dataset_drop_ref(ds->ds_prev, ds);
270 bplist_destroy(&ds->ds_pending_deadlist);
272 dsl_deadlist_close(&ds->ds_deadlist);
274 ASSERT(ds->ds_deadlist.dl_dbuf == NULL);
275 ASSERT(!ds->ds_deadlist.dl_oldfmt);
278 dsl_dir_close(ds->ds_dir, ds);
280 ASSERT(!list_link_active(&ds->ds_synced_link));
282 if (mutex_owned(&ds->ds_lock))
283 mutex_exit(&ds->ds_lock);
284 mutex_destroy(&ds->ds_lock);
285 mutex_destroy(&ds->ds_recvlock);
286 if (mutex_owned(&ds->ds_opening_lock))
287 mutex_exit(&ds->ds_opening_lock);
288 mutex_destroy(&ds->ds_opening_lock);
289 rw_destroy(&ds->ds_rwlock);
290 cv_destroy(&ds->ds_exclusive_cv);
292 kmem_free(ds, sizeof (dsl_dataset_t));
296 dsl_dataset_get_snapname(dsl_dataset_t *ds)
298 dsl_dataset_phys_t *headphys;
301 dsl_pool_t *dp = ds->ds_dir->dd_pool;
302 objset_t *mos = dp->dp_meta_objset;
304 if (ds->ds_snapname[0])
306 if (ds->ds_phys->ds_next_snap_obj == 0)
309 err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
313 headphys = headdbuf->db_data;
314 err = zap_value_search(dp->dp_meta_objset,
315 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
316 dmu_buf_rele(headdbuf, FTAG);
321 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
323 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
324 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
328 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
333 err = zap_lookup_norm(mos, snapobj, name, 8, 1,
334 value, mt, NULL, 0, NULL);
335 if (err == ENOTSUP && mt == MT_FIRST)
336 err = zap_lookup(mos, snapobj, name, 8, 1, value);
341 dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx)
343 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
344 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
348 dsl_dir_snap_cmtime_update(ds->ds_dir);
350 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
355 err = zap_remove_norm(mos, snapobj, name, mt, tx);
356 if (err == ENOTSUP && mt == MT_FIRST)
357 err = zap_remove(mos, snapobj, name, tx);
362 dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag,
365 objset_t *mos = dp->dp_meta_objset;
369 dmu_object_info_t doi;
371 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
372 dsl_pool_sync_context(dp));
374 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
378 /* Make sure dsobj has the correct object type. */
379 dmu_object_info_from_db(dbuf, &doi);
380 if (doi.doi_type != DMU_OT_DSL_DATASET)
383 ds = dmu_buf_get_user(dbuf);
385 dsl_dataset_t *winner;
387 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
389 ds->ds_object = dsobj;
390 ds->ds_phys = dbuf->db_data;
392 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
393 mutex_init(&ds->ds_recvlock, NULL, MUTEX_DEFAULT, NULL);
394 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
395 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
397 rw_init(&ds->ds_rwlock, 0, 0, 0);
398 cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL);
400 bplist_create(&ds->ds_pending_deadlist);
401 dsl_deadlist_open(&ds->ds_deadlist,
402 mos, ds->ds_phys->ds_deadlist_obj);
404 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
405 offsetof(dmu_sendarg_t, dsa_link));
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();
802 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
803 sizeof (dsphys->ds_guid));
804 } while (dsphys->ds_guid == 0);
805 dsphys->ds_snapnames_zapobj =
806 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
808 dsphys->ds_creation_time = gethrestime_sec();
809 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
811 if (origin == NULL) {
812 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
816 dsphys->ds_prev_snap_obj = origin->ds_object;
817 dsphys->ds_prev_snap_txg =
818 origin->ds_phys->ds_creation_txg;
819 dsphys->ds_referenced_bytes =
820 origin->ds_phys->ds_referenced_bytes;
821 dsphys->ds_compressed_bytes =
822 origin->ds_phys->ds_compressed_bytes;
823 dsphys->ds_uncompressed_bytes =
824 origin->ds_phys->ds_uncompressed_bytes;
825 dsphys->ds_bp = origin->ds_phys->ds_bp;
826 dsphys->ds_flags |= origin->ds_phys->ds_flags;
828 dmu_buf_will_dirty(origin->ds_dbuf, tx);
829 origin->ds_phys->ds_num_children++;
831 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
832 origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds));
833 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
834 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
835 dsl_dataset_rele(ohds, FTAG);
837 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
838 if (origin->ds_phys->ds_next_clones_obj == 0) {
839 origin->ds_phys->ds_next_clones_obj =
841 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
843 VERIFY(0 == zap_add_int(mos,
844 origin->ds_phys->ds_next_clones_obj,
848 dmu_buf_will_dirty(dd->dd_dbuf, tx);
849 dd->dd_phys->dd_origin_obj = origin->ds_object;
850 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
851 if (origin->ds_dir->dd_phys->dd_clones == 0) {
852 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
853 origin->ds_dir->dd_phys->dd_clones =
855 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
857 VERIFY3U(0, ==, zap_add_int(mos,
858 origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
862 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
863 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
865 dmu_buf_rele(dbuf, FTAG);
867 dmu_buf_will_dirty(dd->dd_dbuf, tx);
868 dd->dd_phys->dd_head_dataset_obj = dsobj;
874 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
875 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
877 dsl_pool_t *dp = pdd->dd_pool;
878 uint64_t dsobj, ddobj;
881 ASSERT(lastname[0] != '@');
883 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
884 VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd));
886 dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx);
888 dsl_deleg_set_create_perms(dd, tx, cr);
890 dsl_dir_close(dd, FTAG);
893 * If we are creating a clone, make sure we zero out any stale
894 * data from the origin snapshots zil header.
896 if (origin != NULL) {
900 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
901 VERIFY3U(0, ==, dmu_objset_from_ds(ds, &os));
902 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
903 dsl_dataset_dirty(ds, tx);
904 dsl_dataset_rele(ds, FTAG);
911 /* FreeBSD ioctl compat begin */
914 const char *snapname;
918 dsl_check_snap_cb(const char *name, void *arg)
920 struct destroyarg *da = arg;
924 dsname = kmem_asprintf("%s@%s", name, da->snapname);
925 VERIFY(nvlist_add_boolean(da->nvl, dsname) == 0);
931 dmu_get_recursive_snaps_nvl(const char *fsname, const char *snapname,
934 struct destroyarg *da;
937 da = kmem_zalloc(sizeof (struct destroyarg), KM_SLEEP);
939 da->snapname = snapname;
940 err = dmu_objset_find(fsname, dsl_check_snap_cb, da,
942 kmem_free(da, sizeof (struct destroyarg));
946 /* FreeBSD ioctl compat end */
947 #endif /* __FreeBSD__ */
950 * The snapshots must all be in the same pool.
953 dmu_snapshots_destroy_nvl(nvlist_t *snaps, boolean_t defer, char *failed)
956 dsl_sync_task_t *dst;
959 dsl_sync_task_group_t *dstg;
961 pair = nvlist_next_nvpair(snaps, NULL);
965 err = spa_open(nvpair_name(pair), &spa, FTAG);
968 dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
970 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
971 pair = nvlist_next_nvpair(snaps, pair)) {
974 err = dsl_dataset_own(nvpair_name(pair), B_TRUE, dstg, &ds);
976 struct dsl_ds_destroyarg *dsda;
978 dsl_dataset_make_exclusive(ds, dstg);
979 dsda = kmem_zalloc(sizeof (struct dsl_ds_destroyarg),
983 dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
984 dsl_dataset_destroy_sync, dsda, dstg, 0);
985 } else if (err == ENOENT) {
988 (void) strcpy(failed, nvpair_name(pair));
994 err = dsl_sync_task_group_wait(dstg);
996 for (dst = list_head(&dstg->dstg_tasks); dst;
997 dst = list_next(&dstg->dstg_tasks, dst)) {
998 struct dsl_ds_destroyarg *dsda = dst->dst_arg1;
999 dsl_dataset_t *ds = dsda->ds;
1002 * Return the file system name that triggered the error
1005 dsl_dataset_name(ds, failed);
1007 ASSERT3P(dsda->rm_origin, ==, NULL);
1008 dsl_dataset_disown(ds, dstg);
1009 kmem_free(dsda, sizeof (struct dsl_ds_destroyarg));
1012 dsl_sync_task_group_destroy(dstg);
1013 spa_close(spa, FTAG);
1019 dsl_dataset_might_destroy_origin(dsl_dataset_t *ds)
1021 boolean_t might_destroy = B_FALSE;
1023 mutex_enter(&ds->ds_lock);
1024 if (ds->ds_phys->ds_num_children == 2 && ds->ds_userrefs == 0 &&
1025 DS_IS_DEFER_DESTROY(ds))
1026 might_destroy = B_TRUE;
1027 mutex_exit(&ds->ds_lock);
1029 return (might_destroy);
1033 * If we're removing a clone, and these three conditions are true:
1034 * 1) the clone's origin has no other children
1035 * 2) the clone's origin has no user references
1036 * 3) the clone's origin has been marked for deferred destruction
1037 * Then, prepare to remove the origin as part of this sync task group.
1040 dsl_dataset_origin_rm_prep(struct dsl_ds_destroyarg *dsda, void *tag)
1042 dsl_dataset_t *ds = dsda->ds;
1043 dsl_dataset_t *origin = ds->ds_prev;
1045 if (dsl_dataset_might_destroy_origin(origin)) {
1050 namelen = dsl_dataset_namelen(origin) + 1;
1051 name = kmem_alloc(namelen, KM_SLEEP);
1052 dsl_dataset_name(origin, name);
1054 error = zfs_unmount_snap(name, NULL);
1056 kmem_free(name, namelen);
1060 error = dsl_dataset_own(name, B_TRUE, tag, &origin);
1061 kmem_free(name, namelen);
1064 dsda->rm_origin = origin;
1065 dsl_dataset_make_exclusive(origin, tag);
1072 * ds must be opened as OWNER. On return (whether successful or not),
1073 * ds will be closed and caller can no longer dereference it.
1076 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer)
1079 dsl_sync_task_group_t *dstg;
1083 struct dsl_ds_destroyarg dsda = { 0 };
1084 dsl_dataset_t dummy_ds = { 0 };
1088 if (dsl_dataset_is_snapshot(ds)) {
1089 /* Destroying a snapshot is simpler */
1090 dsl_dataset_make_exclusive(ds, tag);
1093 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1094 dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
1096 ASSERT3P(dsda.rm_origin, ==, NULL);
1104 dummy_ds.ds_dir = dd;
1105 dummy_ds.ds_object = ds->ds_object;
1107 if (!spa_feature_is_enabled(dsl_dataset_get_spa(ds),
1108 &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY])) {
1110 * Check for errors and mark this ds as inconsistent, in
1111 * case we crash while freeing the objects.
1113 err = dsl_sync_task_do(dd->dd_pool,
1114 dsl_dataset_destroy_begin_check,
1115 dsl_dataset_destroy_begin_sync, ds, NULL, 0);
1119 err = dmu_objset_from_ds(ds, &os);
1124 * Remove all objects while in the open context so that
1125 * there is less work to do in the syncing context.
1127 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
1128 ds->ds_phys->ds_prev_snap_txg)) {
1130 * Ignore errors, if there is not enough disk space
1131 * we will deal with it in dsl_dataset_destroy_sync().
1133 (void) dmu_free_object(os, obj);
1139 * Sync out all in-flight IO.
1141 txg_wait_synced(dd->dd_pool, 0);
1144 * If we managed to free all the objects in open
1145 * context, the user space accounting should be zero.
1147 if (ds->ds_phys->ds_bp.blk_fill == 0 &&
1148 dmu_objset_userused_enabled(os)) {
1151 ASSERT(zap_count(os, DMU_USERUSED_OBJECT,
1152 &count) != 0 || count == 0);
1153 ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT,
1154 &count) != 0 || count == 0);
1158 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
1159 err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd);
1160 rw_exit(&dd->dd_pool->dp_config_rwlock);
1166 * Blow away the dsl_dir + head dataset.
1168 dsl_dataset_make_exclusive(ds, tag);
1170 * If we're removing a clone, we might also need to remove its
1174 dsda.need_prep = B_FALSE;
1175 if (dsl_dir_is_clone(dd)) {
1176 err = dsl_dataset_origin_rm_prep(&dsda, tag);
1178 dsl_dir_close(dd, FTAG);
1183 dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
1184 dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
1185 dsl_dataset_destroy_sync, &dsda, tag, 0);
1186 dsl_sync_task_create(dstg, dsl_dir_destroy_check,
1187 dsl_dir_destroy_sync, &dummy_ds, FTAG, 0);
1188 err = dsl_sync_task_group_wait(dstg);
1189 dsl_sync_task_group_destroy(dstg);
1192 * We could be racing against 'zfs release' or 'zfs destroy -d'
1193 * on the origin snap, in which case we can get EBUSY if we
1194 * needed to destroy the origin snap but were not ready to
1197 if (dsda.need_prep) {
1198 ASSERT(err == EBUSY);
1199 ASSERT(dsl_dir_is_clone(dd));
1200 ASSERT(dsda.rm_origin == NULL);
1202 } while (dsda.need_prep);
1204 if (dsda.rm_origin != NULL)
1205 dsl_dataset_disown(dsda.rm_origin, tag);
1207 /* if it is successful, dsl_dir_destroy_sync will close the dd */
1209 dsl_dir_close(dd, FTAG);
1211 dsl_dataset_disown(ds, tag);
1216 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1218 return (&ds->ds_phys->ds_bp);
1222 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1224 ASSERT(dmu_tx_is_syncing(tx));
1225 /* If it's the meta-objset, set dp_meta_rootbp */
1227 tx->tx_pool->dp_meta_rootbp = *bp;
1229 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1230 ds->ds_phys->ds_bp = *bp;
1235 dsl_dataset_get_spa(dsl_dataset_t *ds)
1237 return (ds->ds_dir->dd_pool->dp_spa);
1241 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1245 if (ds == NULL) /* this is the meta-objset */
1248 ASSERT(ds->ds_objset != NULL);
1250 if (ds->ds_phys->ds_next_snap_obj != 0)
1251 panic("dirtying snapshot!");
1253 dp = ds->ds_dir->dd_pool;
1255 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) {
1256 /* up the hold count until we can be written out */
1257 dmu_buf_add_ref(ds->ds_dbuf, ds);
1262 dsl_dataset_is_dirty(dsl_dataset_t *ds)
1264 for (int t = 0; t < TXG_SIZE; t++) {
1265 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1273 * The unique space in the head dataset can be calculated by subtracting
1274 * the space used in the most recent snapshot, that is still being used
1275 * in this file system, from the space currently in use. To figure out
1276 * the space in the most recent snapshot still in use, we need to take
1277 * the total space used in the snapshot and subtract out the space that
1278 * has been freed up since the snapshot was taken.
1281 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1284 uint64_t dlused, dlcomp, dluncomp;
1286 ASSERT(!dsl_dataset_is_snapshot(ds));
1288 if (ds->ds_phys->ds_prev_snap_obj != 0)
1289 mrs_used = ds->ds_prev->ds_phys->ds_referenced_bytes;
1293 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
1295 ASSERT3U(dlused, <=, mrs_used);
1296 ds->ds_phys->ds_unique_bytes =
1297 ds->ds_phys->ds_referenced_bytes - (mrs_used - dlused);
1299 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1300 SPA_VERSION_UNIQUE_ACCURATE)
1301 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1311 kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf,
1312 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1314 struct killarg *ka = arg;
1315 dmu_tx_t *tx = ka->tx;
1320 if (zb->zb_level == ZB_ZIL_LEVEL) {
1321 ASSERT(zilog != NULL);
1323 * It's a block in the intent log. It has no
1324 * accounting, so just free it.
1326 dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp);
1328 ASSERT(zilog == NULL);
1329 ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg);
1330 (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE);
1338 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1340 dsl_dataset_t *ds = arg1;
1341 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1346 * Can't delete a head dataset if there are snapshots of it.
1347 * (Except if the only snapshots are from the branch we cloned
1350 if (ds->ds_prev != NULL &&
1351 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1355 * This is really a dsl_dir thing, but check it here so that
1356 * we'll be less likely to leave this dataset inconsistent &
1359 err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count);
1370 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1372 dsl_dataset_t *ds = arg1;
1373 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1375 /* Mark it as inconsistent on-disk, in case we crash */
1376 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1377 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1379 spa_history_log_internal(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx,
1380 "dataset = %llu", ds->ds_object);
1384 dsl_dataset_origin_check(struct dsl_ds_destroyarg *dsda, void *tag,
1387 dsl_dataset_t *ds = dsda->ds;
1388 dsl_dataset_t *ds_prev = ds->ds_prev;
1390 if (dsl_dataset_might_destroy_origin(ds_prev)) {
1391 struct dsl_ds_destroyarg ndsda = {0};
1394 * If we're not prepared to remove the origin, don't remove
1397 if (dsda->rm_origin == NULL) {
1398 dsda->need_prep = B_TRUE;
1403 ndsda.is_origin_rm = B_TRUE;
1404 return (dsl_dataset_destroy_check(&ndsda, tag, tx));
1408 * If we're not going to remove the origin after all,
1409 * undo the open context setup.
1411 if (dsda->rm_origin != NULL) {
1412 dsl_dataset_disown(dsda->rm_origin, tag);
1413 dsda->rm_origin = NULL;
1420 * If you add new checks here, you may need to add
1421 * additional checks to the "temporary" case in
1422 * snapshot_check() in dmu_objset.c.
1426 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1428 struct dsl_ds_destroyarg *dsda = arg1;
1429 dsl_dataset_t *ds = dsda->ds;
1431 /* we have an owner hold, so noone else can destroy us */
1432 ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
1435 * Only allow deferred destroy on pools that support it.
1436 * NOTE: deferred destroy is only supported on snapshots.
1439 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
1440 SPA_VERSION_USERREFS)
1442 ASSERT(dsl_dataset_is_snapshot(ds));
1447 * Can't delete a head dataset if there are snapshots of it.
1448 * (Except if the only snapshots are from the branch we cloned
1451 if (ds->ds_prev != NULL &&
1452 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1456 * If we made changes this txg, traverse_dsl_dataset won't find
1459 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1462 if (dsl_dataset_is_snapshot(ds)) {
1464 * If this snapshot has an elevated user reference count,
1465 * we can't destroy it yet.
1467 if (ds->ds_userrefs > 0 && !dsda->releasing)
1470 mutex_enter(&ds->ds_lock);
1472 * Can't delete a branch point. However, if we're destroying
1473 * a clone and removing its origin due to it having a user
1474 * hold count of 0 and having been marked for deferred destroy,
1475 * it's OK for the origin to have a single clone.
1477 if (ds->ds_phys->ds_num_children >
1478 (dsda->is_origin_rm ? 2 : 1)) {
1479 mutex_exit(&ds->ds_lock);
1482 mutex_exit(&ds->ds_lock);
1483 } else if (dsl_dir_is_clone(ds->ds_dir)) {
1484 return (dsl_dataset_origin_check(dsda, arg2, tx));
1487 /* XXX we should do some i/o error checking... */
1499 dsl_dataset_refs_gone(dmu_buf_t *db, void *argv)
1501 struct refsarg *arg = argv;
1503 mutex_enter(&arg->lock);
1505 cv_signal(&arg->cv);
1506 mutex_exit(&arg->lock);
1510 dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag)
1514 bzero(&arg, sizeof(arg));
1515 mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL);
1516 cv_init(&arg.cv, NULL, CV_DEFAULT, NULL);
1518 (void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys,
1519 dsl_dataset_refs_gone);
1520 dmu_buf_rele(ds->ds_dbuf, tag);
1521 mutex_enter(&arg.lock);
1523 cv_wait(&arg.cv, &arg.lock);
1525 mutex_exit(&arg.lock);
1528 mutex_destroy(&arg.lock);
1529 cv_destroy(&arg.cv);
1533 remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, dmu_tx_t *tx)
1535 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1539 ASSERT(ds->ds_phys->ds_num_children >= 2);
1540 err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx);
1542 * The err should not be ENOENT, but a bug in a previous version
1543 * of the code could cause upgrade_clones_cb() to not set
1544 * ds_next_snap_obj when it should, leading to a missing entry.
1545 * If we knew that the pool was created after
1546 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1547 * ENOENT. However, at least we can check that we don't have
1548 * too many entries in the next_clones_obj even after failing to
1551 if (err != ENOENT) {
1554 ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj,
1556 ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2);
1560 dsl_dataset_remove_clones_key(dsl_dataset_t *ds, uint64_t mintxg, dmu_tx_t *tx)
1562 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1567 * If it is the old version, dd_clones doesn't exist so we can't
1568 * find the clones, but deadlist_remove_key() is a no-op so it
1571 if (ds->ds_dir->dd_phys->dd_clones == 0)
1574 for (zap_cursor_init(&zc, mos, ds->ds_dir->dd_phys->dd_clones);
1575 zap_cursor_retrieve(&zc, &za) == 0;
1576 zap_cursor_advance(&zc)) {
1577 dsl_dataset_t *clone;
1579 VERIFY3U(0, ==, dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1580 za.za_first_integer, FTAG, &clone));
1581 if (clone->ds_dir->dd_origin_txg > mintxg) {
1582 dsl_deadlist_remove_key(&clone->ds_deadlist,
1584 dsl_dataset_remove_clones_key(clone, mintxg, tx);
1586 dsl_dataset_rele(clone, FTAG);
1588 zap_cursor_fini(&zc);
1591 struct process_old_arg {
1593 dsl_dataset_t *ds_prev;
1594 boolean_t after_branch_point;
1596 uint64_t used, comp, uncomp;
1600 process_old_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
1602 struct process_old_arg *poa = arg;
1603 dsl_pool_t *dp = poa->ds->ds_dir->dd_pool;
1605 if (bp->blk_birth <= poa->ds->ds_phys->ds_prev_snap_txg) {
1606 dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, tx);
1607 if (poa->ds_prev && !poa->after_branch_point &&
1609 poa->ds_prev->ds_phys->ds_prev_snap_txg) {
1610 poa->ds_prev->ds_phys->ds_unique_bytes +=
1611 bp_get_dsize_sync(dp->dp_spa, bp);
1614 poa->used += bp_get_dsize_sync(dp->dp_spa, bp);
1615 poa->comp += BP_GET_PSIZE(bp);
1616 poa->uncomp += BP_GET_UCSIZE(bp);
1617 dsl_free_sync(poa->pio, dp, tx->tx_txg, bp);
1623 process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev,
1624 dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx)
1626 struct process_old_arg poa = { 0 };
1627 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1628 objset_t *mos = dp->dp_meta_objset;
1630 ASSERT(ds->ds_deadlist.dl_oldfmt);
1631 ASSERT(ds_next->ds_deadlist.dl_oldfmt);
1634 poa.ds_prev = ds_prev;
1635 poa.after_branch_point = after_branch_point;
1636 poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1637 VERIFY3U(0, ==, bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj,
1638 process_old_cb, &poa, tx));
1639 VERIFY0(zio_wait(poa.pio));
1640 ASSERT3U(poa.used, ==, ds->ds_phys->ds_unique_bytes);
1642 /* change snapused */
1643 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1644 -poa.used, -poa.comp, -poa.uncomp, tx);
1646 /* swap next's deadlist to our deadlist */
1647 dsl_deadlist_close(&ds->ds_deadlist);
1648 dsl_deadlist_close(&ds_next->ds_deadlist);
1649 SWITCH64(ds_next->ds_phys->ds_deadlist_obj,
1650 ds->ds_phys->ds_deadlist_obj);
1651 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
1652 dsl_deadlist_open(&ds_next->ds_deadlist, mos,
1653 ds_next->ds_phys->ds_deadlist_obj);
1657 old_synchronous_dataset_destroy(dsl_dataset_t *ds, dmu_tx_t *tx)
1663 * Free everything that we point to (that's born after
1664 * the previous snapshot, if we are a clone)
1666 * NB: this should be very quick, because we already
1667 * freed all the objects in open context.
1671 err = traverse_dataset(ds,
1672 ds->ds_phys->ds_prev_snap_txg, TRAVERSE_POST,
1675 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) || ds->ds_phys->ds_unique_bytes == 0);
1681 dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
1683 struct dsl_ds_destroyarg *dsda = arg1;
1684 dsl_dataset_t *ds = dsda->ds;
1686 int after_branch_point = FALSE;
1687 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1688 objset_t *mos = dp->dp_meta_objset;
1689 dsl_dataset_t *ds_prev = NULL;
1690 boolean_t wont_destroy;
1693 wont_destroy = (dsda->defer &&
1694 (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1));
1696 ASSERT(ds->ds_owner || wont_destroy);
1697 ASSERT(dsda->defer || ds->ds_phys->ds_num_children <= 1);
1698 ASSERT(ds->ds_prev == NULL ||
1699 ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1700 ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1703 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
1704 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1705 ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY;
1709 /* signal any waiters that this dataset is going away */
1710 mutex_enter(&ds->ds_lock);
1711 ds->ds_owner = dsl_reaper;
1712 cv_broadcast(&ds->ds_exclusive_cv);
1713 mutex_exit(&ds->ds_lock);
1715 /* Remove our reservation */
1716 if (ds->ds_reserved != 0) {
1717 dsl_prop_setarg_t psa;
1720 dsl_prop_setarg_init_uint64(&psa, "refreservation",
1721 (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
1723 psa.psa_effective_value = 0; /* predict default value */
1725 dsl_dataset_set_reservation_sync(ds, &psa, tx);
1726 ASSERT0(ds->ds_reserved);
1729 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1731 dsl_scan_ds_destroyed(ds, tx);
1733 obj = ds->ds_object;
1735 if (ds->ds_phys->ds_prev_snap_obj != 0) {
1737 ds_prev = ds->ds_prev;
1739 VERIFY(0 == dsl_dataset_hold_obj(dp,
1740 ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev));
1742 after_branch_point =
1743 (ds_prev->ds_phys->ds_next_snap_obj != obj);
1745 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
1746 if (after_branch_point &&
1747 ds_prev->ds_phys->ds_next_clones_obj != 0) {
1748 remove_from_next_clones(ds_prev, obj, tx);
1749 if (ds->ds_phys->ds_next_snap_obj != 0) {
1750 VERIFY(0 == zap_add_int(mos,
1751 ds_prev->ds_phys->ds_next_clones_obj,
1752 ds->ds_phys->ds_next_snap_obj, tx));
1755 if (after_branch_point &&
1756 ds->ds_phys->ds_next_snap_obj == 0) {
1757 /* This clone is toast. */
1758 ASSERT(ds_prev->ds_phys->ds_num_children > 1);
1759 ds_prev->ds_phys->ds_num_children--;
1762 * If the clone's origin has no other clones, no
1763 * user holds, and has been marked for deferred
1764 * deletion, then we should have done the necessary
1765 * destroy setup for it.
1767 if (ds_prev->ds_phys->ds_num_children == 1 &&
1768 ds_prev->ds_userrefs == 0 &&
1769 DS_IS_DEFER_DESTROY(ds_prev)) {
1770 ASSERT3P(dsda->rm_origin, !=, NULL);
1772 ASSERT3P(dsda->rm_origin, ==, NULL);
1774 } else if (!after_branch_point) {
1775 ds_prev->ds_phys->ds_next_snap_obj =
1776 ds->ds_phys->ds_next_snap_obj;
1780 if (dsl_dataset_is_snapshot(ds)) {
1781 dsl_dataset_t *ds_next;
1782 uint64_t old_unique;
1783 uint64_t used = 0, comp = 0, uncomp = 0;
1785 VERIFY(0 == dsl_dataset_hold_obj(dp,
1786 ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next));
1787 ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
1789 old_unique = ds_next->ds_phys->ds_unique_bytes;
1791 dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
1792 ds_next->ds_phys->ds_prev_snap_obj =
1793 ds->ds_phys->ds_prev_snap_obj;
1794 ds_next->ds_phys->ds_prev_snap_txg =
1795 ds->ds_phys->ds_prev_snap_txg;
1796 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1797 ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
1800 if (ds_next->ds_deadlist.dl_oldfmt) {
1801 process_old_deadlist(ds, ds_prev, ds_next,
1802 after_branch_point, tx);
1804 /* Adjust prev's unique space. */
1805 if (ds_prev && !after_branch_point) {
1806 dsl_deadlist_space_range(&ds_next->ds_deadlist,
1807 ds_prev->ds_phys->ds_prev_snap_txg,
1808 ds->ds_phys->ds_prev_snap_txg,
1809 &used, &comp, &uncomp);
1810 ds_prev->ds_phys->ds_unique_bytes += used;
1813 /* Adjust snapused. */
1814 dsl_deadlist_space_range(&ds_next->ds_deadlist,
1815 ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
1816 &used, &comp, &uncomp);
1817 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1818 -used, -comp, -uncomp, tx);
1820 /* Move blocks to be freed to pool's free list. */
1821 dsl_deadlist_move_bpobj(&ds_next->ds_deadlist,
1822 &dp->dp_free_bpobj, ds->ds_phys->ds_prev_snap_txg,
1824 dsl_dir_diduse_space(tx->tx_pool->dp_free_dir,
1825 DD_USED_HEAD, used, comp, uncomp, tx);
1827 /* Merge our deadlist into next's and free it. */
1828 dsl_deadlist_merge(&ds_next->ds_deadlist,
1829 ds->ds_phys->ds_deadlist_obj, tx);
1831 dsl_deadlist_close(&ds->ds_deadlist);
1832 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
1834 /* Collapse range in clone heads */
1835 dsl_dataset_remove_clones_key(ds,
1836 ds->ds_phys->ds_creation_txg, tx);
1838 if (dsl_dataset_is_snapshot(ds_next)) {
1839 dsl_dataset_t *ds_nextnext;
1842 * Update next's unique to include blocks which
1843 * were previously shared by only this snapshot
1844 * and it. Those blocks will be born after the
1845 * prev snap and before this snap, and will have
1846 * died after the next snap and before the one
1847 * after that (ie. be on the snap after next's
1850 VERIFY(0 == dsl_dataset_hold_obj(dp,
1851 ds_next->ds_phys->ds_next_snap_obj,
1852 FTAG, &ds_nextnext));
1853 dsl_deadlist_space_range(&ds_nextnext->ds_deadlist,
1854 ds->ds_phys->ds_prev_snap_txg,
1855 ds->ds_phys->ds_creation_txg,
1856 &used, &comp, &uncomp);
1857 ds_next->ds_phys->ds_unique_bytes += used;
1858 dsl_dataset_rele(ds_nextnext, FTAG);
1859 ASSERT3P(ds_next->ds_prev, ==, NULL);
1861 /* Collapse range in this head. */
1863 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
1864 ds->ds_dir->dd_phys->dd_head_dataset_obj,
1866 dsl_deadlist_remove_key(&hds->ds_deadlist,
1867 ds->ds_phys->ds_creation_txg, tx);
1868 dsl_dataset_rele(hds, FTAG);
1871 ASSERT3P(ds_next->ds_prev, ==, ds);
1872 dsl_dataset_drop_ref(ds_next->ds_prev, ds_next);
1873 ds_next->ds_prev = NULL;
1875 VERIFY(0 == dsl_dataset_get_ref(dp,
1876 ds->ds_phys->ds_prev_snap_obj,
1877 ds_next, &ds_next->ds_prev));
1880 dsl_dataset_recalc_head_uniq(ds_next);
1883 * Reduce the amount of our unconsmed refreservation
1884 * being charged to our parent by the amount of
1885 * new unique data we have gained.
1887 if (old_unique < ds_next->ds_reserved) {
1889 uint64_t new_unique =
1890 ds_next->ds_phys->ds_unique_bytes;
1892 ASSERT(old_unique <= new_unique);
1893 mrsdelta = MIN(new_unique - old_unique,
1894 ds_next->ds_reserved - old_unique);
1895 dsl_dir_diduse_space(ds->ds_dir,
1896 DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
1899 dsl_dataset_rele(ds_next, FTAG);
1901 zfeature_info_t *async_destroy =
1902 &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY];
1906 * There's no next snapshot, so this is a head dataset.
1907 * Destroy the deadlist. Unless it's a clone, the
1908 * deadlist should be empty. (If it's a clone, it's
1909 * safe to ignore the deadlist contents.)
1911 dsl_deadlist_close(&ds->ds_deadlist);
1912 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
1913 ds->ds_phys->ds_deadlist_obj = 0;
1915 VERIFY3U(0, ==, dmu_objset_from_ds(ds, &os));
1917 if (!spa_feature_is_enabled(dp->dp_spa, async_destroy)) {
1918 err = old_synchronous_dataset_destroy(ds, tx);
1921 * Move the bptree into the pool's list of trees to
1922 * clean up and update space accounting information.
1924 uint64_t used, comp, uncomp;
1926 zil_destroy_sync(dmu_objset_zil(os), tx);
1928 if (!spa_feature_is_active(dp->dp_spa, async_destroy)) {
1929 spa_feature_incr(dp->dp_spa, async_destroy, tx);
1930 dp->dp_bptree_obj = bptree_alloc(mos, tx);
1932 DMU_POOL_DIRECTORY_OBJECT,
1933 DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1,
1934 &dp->dp_bptree_obj, tx) == 0);
1937 used = ds->ds_dir->dd_phys->dd_used_bytes;
1938 comp = ds->ds_dir->dd_phys->dd_compressed_bytes;
1939 uncomp = ds->ds_dir->dd_phys->dd_uncompressed_bytes;
1941 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
1942 ds->ds_phys->ds_unique_bytes == used);
1944 bptree_add(mos, dp->dp_bptree_obj,
1945 &ds->ds_phys->ds_bp, ds->ds_phys->ds_prev_snap_txg,
1946 used, comp, uncomp, tx);
1947 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
1948 -used, -comp, -uncomp, tx);
1949 dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD,
1950 used, comp, uncomp, tx);
1953 if (ds->ds_prev != NULL) {
1954 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
1955 VERIFY3U(0, ==, zap_remove_int(mos,
1956 ds->ds_prev->ds_dir->dd_phys->dd_clones,
1957 ds->ds_object, tx));
1959 dsl_dataset_rele(ds->ds_prev, ds);
1960 ds->ds_prev = ds_prev = NULL;
1965 * This must be done after the dsl_traverse(), because it will
1966 * re-open the objset.
1968 if (ds->ds_objset) {
1969 dmu_objset_evict(ds->ds_objset);
1970 ds->ds_objset = NULL;
1973 if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) {
1974 /* Erase the link in the dir */
1975 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1976 ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
1977 ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0);
1978 err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx);
1981 /* remove from snapshot namespace */
1982 dsl_dataset_t *ds_head;
1983 ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0);
1984 VERIFY(0 == dsl_dataset_hold_obj(dp,
1985 ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head));
1986 VERIFY(0 == dsl_dataset_get_snapname(ds));
1991 err = dsl_dataset_snap_lookup(ds_head,
1992 ds->ds_snapname, &val);
1994 ASSERT3U(val, ==, obj);
1997 err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx);
1999 dsl_dataset_rele(ds_head, FTAG);
2002 if (ds_prev && ds->ds_prev != ds_prev)
2003 dsl_dataset_rele(ds_prev, FTAG);
2005 spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
2006 spa_history_log_internal(LOG_DS_DESTROY, dp->dp_spa, tx,
2007 "dataset = %llu", ds->ds_object);
2009 if (ds->ds_phys->ds_next_clones_obj != 0) {
2011 ASSERT(0 == zap_count(mos,
2012 ds->ds_phys->ds_next_clones_obj, &count) && count == 0);
2013 VERIFY(0 == dmu_object_free(mos,
2014 ds->ds_phys->ds_next_clones_obj, tx));
2016 if (ds->ds_phys->ds_props_obj != 0)
2017 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx));
2018 if (ds->ds_phys->ds_userrefs_obj != 0)
2019 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx));
2020 dsl_dir_close(ds->ds_dir, ds);
2022 dsl_dataset_drain_refs(ds, tag);
2023 VERIFY(0 == dmu_object_free(mos, obj, tx));
2025 if (dsda->rm_origin) {
2027 * Remove the origin of the clone we just destroyed.
2029 struct dsl_ds_destroyarg ndsda = {0};
2031 ndsda.ds = dsda->rm_origin;
2032 dsl_dataset_destroy_sync(&ndsda, tag, tx);
2037 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
2041 if (!dmu_tx_is_syncing(tx))
2045 * If there's an fs-only reservation, any blocks that might become
2046 * owned by the snapshot dataset must be accommodated by space
2047 * outside of the reservation.
2049 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
2050 asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
2051 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
2055 * Propogate any reserved space for this snapshot to other
2056 * snapshot checks in this sync group.
2059 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
2065 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
2067 dsl_dataset_t *ds = arg1;
2068 const char *snapname = arg2;
2073 * We don't allow multiple snapshots of the same txg. If there
2074 * is already one, try again.
2076 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
2080 * Check for conflicting name snapshot name.
2082 err = dsl_dataset_snap_lookup(ds, snapname, &value);
2089 * Check that the dataset's name is not too long. Name consists
2090 * of the dataset's length + 1 for the @-sign + snapshot name's length
2092 if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
2093 return (ENAMETOOLONG);
2095 err = dsl_dataset_snapshot_reserve_space(ds, tx);
2099 ds->ds_trysnap_txg = tx->tx_txg;
2104 dsl_dataset_snapshot_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2106 dsl_dataset_t *ds = arg1;
2107 const char *snapname = arg2;
2108 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2110 dsl_dataset_phys_t *dsphys;
2111 uint64_t dsobj, crtxg;
2112 objset_t *mos = dp->dp_meta_objset;
2115 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
2118 * The origin's ds_creation_txg has to be < TXG_INITIAL
2120 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
2125 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
2126 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
2127 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
2128 dmu_buf_will_dirty(dbuf, tx);
2129 dsphys = dbuf->db_data;
2130 bzero(dsphys, sizeof (dsl_dataset_phys_t));
2131 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
2132 dsphys->ds_fsid_guid = unique_create();
2134 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
2135 sizeof (dsphys->ds_guid));
2136 } while (dsphys->ds_guid == 0);
2137 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
2138 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
2139 dsphys->ds_next_snap_obj = ds->ds_object;
2140 dsphys->ds_num_children = 1;
2141 dsphys->ds_creation_time = gethrestime_sec();
2142 dsphys->ds_creation_txg = crtxg;
2143 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
2144 dsphys->ds_referenced_bytes = ds->ds_phys->ds_referenced_bytes;
2145 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
2146 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
2147 dsphys->ds_flags = ds->ds_phys->ds_flags;
2148 dsphys->ds_bp = ds->ds_phys->ds_bp;
2149 dmu_buf_rele(dbuf, FTAG);
2151 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
2153 uint64_t next_clones_obj =
2154 ds->ds_prev->ds_phys->ds_next_clones_obj;
2155 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
2157 ds->ds_prev->ds_phys->ds_num_children > 1);
2158 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
2159 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
2160 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
2161 ds->ds_prev->ds_phys->ds_creation_txg);
2162 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
2163 } else if (next_clones_obj != 0) {
2164 remove_from_next_clones(ds->ds_prev,
2165 dsphys->ds_next_snap_obj, tx);
2166 VERIFY3U(0, ==, zap_add_int(mos,
2167 next_clones_obj, dsobj, tx));
2172 * If we have a reference-reservation on this dataset, we will
2173 * need to increase the amount of refreservation being charged
2174 * since our unique space is going to zero.
2176 if (ds->ds_reserved) {
2178 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
2179 delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
2180 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
2184 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2185 zfs_dbgmsg("taking snapshot %s@%s/%llu; newkey=%llu",
2186 ds->ds_dir->dd_myname, snapname, dsobj,
2187 ds->ds_phys->ds_prev_snap_txg);
2188 ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist,
2189 UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx);
2190 dsl_deadlist_close(&ds->ds_deadlist);
2191 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
2192 dsl_deadlist_add_key(&ds->ds_deadlist,
2193 ds->ds_phys->ds_prev_snap_txg, tx);
2195 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
2196 ds->ds_phys->ds_prev_snap_obj = dsobj;
2197 ds->ds_phys->ds_prev_snap_txg = crtxg;
2198 ds->ds_phys->ds_unique_bytes = 0;
2199 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
2200 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
2202 err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
2203 snapname, 8, 1, &dsobj, tx);
2207 dsl_dataset_drop_ref(ds->ds_prev, ds);
2208 VERIFY(0 == dsl_dataset_get_ref(dp,
2209 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
2211 dsl_scan_ds_snapshotted(ds, tx);
2213 dsl_dir_snap_cmtime_update(ds->ds_dir);
2215 spa_history_log_internal(LOG_DS_SNAPSHOT, dp->dp_spa, tx,
2216 "dataset = %llu", dsobj);
2220 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
2222 ASSERT(dmu_tx_is_syncing(tx));
2223 ASSERT(ds->ds_objset != NULL);
2224 ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
2227 * in case we had to change ds_fsid_guid when we opened it,
2230 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2231 ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
2233 dmu_objset_sync(ds->ds_objset, zio, tx);
2237 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
2240 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
2246 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
2247 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2248 VERIFY(nvlist_alloc(&val, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2251 * There may me missing entries in ds_next_clones_obj
2252 * due to a bug in a previous version of the code.
2253 * Only trust it if it has the right number of entries.
2255 if (ds->ds_phys->ds_next_clones_obj != 0) {
2256 ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj,
2259 if (count != ds->ds_phys->ds_num_children - 1) {
2262 for (zap_cursor_init(&zc, mos, ds->ds_phys->ds_next_clones_obj);
2263 zap_cursor_retrieve(&zc, &za) == 0;
2264 zap_cursor_advance(&zc)) {
2265 dsl_dataset_t *clone;
2266 char buf[ZFS_MAXNAMELEN];
2268 * Even though we hold the dp_config_rwlock, the dataset
2269 * may fail to open, returning ENOENT. If there is a
2270 * thread concurrently attempting to destroy this
2271 * dataset, it will have the ds_rwlock held for
2272 * RW_WRITER. Our call to dsl_dataset_hold_obj() ->
2273 * dsl_dataset_hold_ref() will fail its
2274 * rw_tryenter(&ds->ds_rwlock, RW_READER), drop the
2275 * dp_config_rwlock, and wait for the destroy progress
2276 * and signal ds_exclusive_cv. If the destroy was
2277 * successful, we will see that
2278 * DSL_DATASET_IS_DESTROYED(), and return ENOENT.
2280 if (dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
2281 za.za_first_integer, FTAG, &clone) != 0)
2283 dsl_dir_name(clone->ds_dir, buf);
2284 VERIFY(nvlist_add_boolean(val, buf) == 0);
2285 dsl_dataset_rele(clone, FTAG);
2287 zap_cursor_fini(&zc);
2288 VERIFY(nvlist_add_nvlist(propval, ZPROP_VALUE, val) == 0);
2289 VERIFY(nvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES),
2293 nvlist_free(propval);
2294 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2298 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
2300 uint64_t refd, avail, uobjs, aobjs, ratio;
2302 dsl_dir_stats(ds->ds_dir, nv);
2304 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
2305 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
2306 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
2308 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
2309 ds->ds_phys->ds_creation_time);
2310 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
2311 ds->ds_phys->ds_creation_txg);
2312 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
2314 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
2316 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
2317 ds->ds_phys->ds_guid);
2318 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
2319 ds->ds_phys->ds_unique_bytes);
2320 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
2322 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
2324 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
2325 DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
2327 if (ds->ds_phys->ds_prev_snap_obj != 0) {
2328 uint64_t written, comp, uncomp;
2329 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2330 dsl_dataset_t *prev;
2332 rw_enter(&dp->dp_config_rwlock, RW_READER);
2333 int err = dsl_dataset_hold_obj(dp,
2334 ds->ds_phys->ds_prev_snap_obj, FTAG, &prev);
2335 rw_exit(&dp->dp_config_rwlock);
2337 err = dsl_dataset_space_written(prev, ds, &written,
2339 dsl_dataset_rele(prev, FTAG);
2341 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
2346 ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
2347 (ds->ds_phys->ds_uncompressed_bytes * 100 /
2348 ds->ds_phys->ds_compressed_bytes);
2349 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
2351 if (ds->ds_phys->ds_next_snap_obj) {
2353 * This is a snapshot; override the dd's space used with
2354 * our unique space and compression ratio.
2356 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
2357 ds->ds_phys->ds_unique_bytes);
2358 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
2360 get_clones_stat(ds, nv);
2365 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
2367 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
2368 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
2369 stat->dds_guid = ds->ds_phys->ds_guid;
2370 if (ds->ds_phys->ds_next_snap_obj) {
2371 stat->dds_is_snapshot = B_TRUE;
2372 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
2374 stat->dds_is_snapshot = B_FALSE;
2375 stat->dds_num_clones = 0;
2378 /* clone origin is really a dsl_dir thing... */
2379 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
2380 if (dsl_dir_is_clone(ds->ds_dir)) {
2383 VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool,
2384 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
2385 dsl_dataset_name(ods, stat->dds_origin);
2386 dsl_dataset_drop_ref(ods, FTAG);
2388 stat->dds_origin[0] = '\0';
2390 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2394 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2396 return (ds->ds_fsid_guid);
2400 dsl_dataset_space(dsl_dataset_t *ds,
2401 uint64_t *refdbytesp, uint64_t *availbytesp,
2402 uint64_t *usedobjsp, uint64_t *availobjsp)
2404 *refdbytesp = ds->ds_phys->ds_referenced_bytes;
2405 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2406 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
2407 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
2408 if (ds->ds_quota != 0) {
2410 * Adjust available bytes according to refquota
2412 if (*refdbytesp < ds->ds_quota)
2413 *availbytesp = MIN(*availbytesp,
2414 ds->ds_quota - *refdbytesp);
2418 *usedobjsp = ds->ds_phys->ds_bp.blk_fill;
2419 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
2423 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
2425 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2427 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
2428 dsl_pool_sync_context(dp));
2429 if (ds->ds_prev == NULL)
2431 if (ds->ds_phys->ds_bp.blk_birth >
2432 ds->ds_prev->ds_phys->ds_creation_txg) {
2433 objset_t *os, *os_prev;
2435 * It may be that only the ZIL differs, because it was
2436 * reset in the head. Don't count that as being
2439 if (dmu_objset_from_ds(ds, &os) != 0)
2441 if (dmu_objset_from_ds(ds->ds_prev, &os_prev) != 0)
2443 return (bcmp(&os->os_phys->os_meta_dnode,
2444 &os_prev->os_phys->os_meta_dnode,
2445 sizeof (os->os_phys->os_meta_dnode)) != 0);
2452 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
2454 dsl_dataset_t *ds = arg1;
2455 char *newsnapname = arg2;
2456 dsl_dir_t *dd = ds->ds_dir;
2461 err = dsl_dataset_hold_obj(dd->dd_pool,
2462 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds);
2466 /* new name better not be in use */
2467 err = dsl_dataset_snap_lookup(hds, newsnapname, &val);
2468 dsl_dataset_rele(hds, FTAG);
2472 else if (err == ENOENT)
2475 /* dataset name + 1 for the "@" + the new snapshot name must fit */
2476 if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
2483 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2485 char oldname[MAXPATHLEN], newname[MAXPATHLEN];
2486 dsl_dataset_t *ds = arg1;
2487 const char *newsnapname = arg2;
2488 dsl_dir_t *dd = ds->ds_dir;
2489 objset_t *mos = dd->dd_pool->dp_meta_objset;
2493 ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
2495 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
2496 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds));
2498 VERIFY(0 == dsl_dataset_get_snapname(ds));
2499 err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx);
2501 dsl_dataset_name(ds, oldname);
2502 mutex_enter(&ds->ds_lock);
2503 (void) strcpy(ds->ds_snapname, newsnapname);
2504 mutex_exit(&ds->ds_lock);
2505 err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
2506 ds->ds_snapname, 8, 1, &ds->ds_object, tx);
2508 dsl_dataset_name(ds, newname);
2510 zvol_rename_minors(oldname, newname);
2513 spa_history_log_internal(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx,
2514 "dataset = %llu", ds->ds_object);
2515 dsl_dataset_rele(hds, FTAG);
2518 struct renamesnaparg {
2519 dsl_sync_task_group_t *dstg;
2520 char failed[MAXPATHLEN];
2526 dsl_snapshot_rename_one(const char *name, void *arg)
2528 struct renamesnaparg *ra = arg;
2529 dsl_dataset_t *ds = NULL;
2533 snapname = kmem_asprintf("%s@%s", name, ra->oldsnap);
2534 (void) strlcpy(ra->failed, snapname, sizeof (ra->failed));
2537 * For recursive snapshot renames the parent won't be changing
2538 * so we just pass name for both the to/from argument.
2540 err = zfs_secpolicy_rename_perms(snapname, snapname, CRED());
2543 return (err == ENOENT ? 0 : err);
2548 * For all filesystems undergoing rename, we'll need to unmount it.
2550 (void) zfs_unmount_snap(snapname, NULL);
2552 err = dsl_dataset_hold(snapname, ra->dstg, &ds);
2555 return (err == ENOENT ? 0 : err);
2557 dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
2558 dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
2564 dsl_recursive_rename(char *oldname, const char *newname)
2567 struct renamesnaparg *ra;
2568 dsl_sync_task_t *dst;
2570 char *cp, *fsname = spa_strdup(oldname);
2571 int len = strlen(oldname) + 1;
2573 /* truncate the snapshot name to get the fsname */
2574 cp = strchr(fsname, '@');
2577 err = spa_open(fsname, &spa, FTAG);
2579 kmem_free(fsname, len);
2582 ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP);
2583 ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
2585 ra->oldsnap = strchr(oldname, '@') + 1;
2586 ra->newsnap = strchr(newname, '@') + 1;
2589 err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
2591 kmem_free(fsname, len);
2594 err = dsl_sync_task_group_wait(ra->dstg);
2597 for (dst = list_head(&ra->dstg->dstg_tasks); dst;
2598 dst = list_next(&ra->dstg->dstg_tasks, dst)) {
2599 dsl_dataset_t *ds = dst->dst_arg1;
2601 dsl_dir_name(ds->ds_dir, ra->failed);
2602 (void) strlcat(ra->failed, "@", sizeof (ra->failed));
2603 (void) strlcat(ra->failed, ra->newsnap,
2604 sizeof (ra->failed));
2606 dsl_dataset_rele(ds, ra->dstg);
2610 (void) strlcpy(oldname, ra->failed, sizeof (ra->failed));
2612 dsl_sync_task_group_destroy(ra->dstg);
2613 kmem_free(ra, sizeof (struct renamesnaparg));
2614 spa_close(spa, FTAG);
2619 dsl_valid_rename(const char *oldname, void *arg)
2621 int delta = *(int *)arg;
2623 if (strlen(oldname) + delta >= MAXNAMELEN)
2624 return (ENAMETOOLONG);
2629 #pragma weak dmu_objset_rename = dsl_dataset_rename
2631 dsl_dataset_rename(char *oldname, const char *newname, int flags)
2638 err = dsl_dir_open(oldname, FTAG, &dd, &tail);
2643 int delta = strlen(newname) - strlen(oldname);
2645 /* if we're growing, validate child name lengths */
2647 err = dmu_objset_find(oldname, dsl_valid_rename,
2648 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2651 err = dsl_dir_rename(dd, newname, flags);
2652 dsl_dir_close(dd, FTAG);
2656 if (tail[0] != '@') {
2657 /* the name ended in a nonexistent component */
2658 dsl_dir_close(dd, FTAG);
2662 dsl_dir_close(dd, FTAG);
2664 /* new name must be snapshot in same filesystem */
2665 tail = strchr(newname, '@');
2669 if (strncmp(oldname, newname, tail - newname) != 0)
2672 if (flags & ZFS_RENAME_RECURSIVE) {
2673 err = dsl_recursive_rename(oldname, newname);
2675 err = dsl_dataset_hold(oldname, FTAG, &ds);
2679 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2680 dsl_dataset_snapshot_rename_check,
2681 dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
2683 dsl_dataset_rele(ds, FTAG);
2689 struct promotenode {
2695 list_t shared_snaps, origin_snaps, clone_snaps;
2696 dsl_dataset_t *origin_origin;
2697 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2701 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2702 static boolean_t snaplist_unstable(list_t *l);
2705 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
2707 dsl_dataset_t *hds = arg1;
2708 struct promotearg *pa = arg2;
2709 struct promotenode *snap = list_head(&pa->shared_snaps);
2710 dsl_dataset_t *origin_ds = snap->ds;
2714 /* Check that it is a real clone */
2715 if (!dsl_dir_is_clone(hds->ds_dir))
2718 /* Since this is so expensive, don't do the preliminary check */
2719 if (!dmu_tx_is_syncing(tx))
2722 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)
2725 /* compute origin's new unique space */
2726 snap = list_tail(&pa->clone_snaps);
2727 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2728 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2729 origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2730 &pa->unique, &unused, &unused);
2733 * Walk the snapshots that we are moving
2735 * Compute space to transfer. Consider the incremental changes
2736 * to used for each snapshot:
2737 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2738 * So each snapshot gave birth to:
2739 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2740 * So a sequence would look like:
2741 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2742 * Which simplifies to:
2743 * uN + kN + kN-1 + ... + k1 + k0
2744 * Note however, if we stop before we reach the ORIGIN we get:
2745 * uN + kN + kN-1 + ... + kM - uM-1
2747 pa->used = origin_ds->ds_phys->ds_referenced_bytes;
2748 pa->comp = origin_ds->ds_phys->ds_compressed_bytes;
2749 pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
2750 for (snap = list_head(&pa->shared_snaps); snap;
2751 snap = list_next(&pa->shared_snaps, snap)) {
2752 uint64_t val, dlused, dlcomp, dluncomp;
2753 dsl_dataset_t *ds = snap->ds;
2755 /* Check that the snapshot name does not conflict */
2756 VERIFY(0 == dsl_dataset_get_snapname(ds));
2757 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2765 /* The very first snapshot does not have a deadlist */
2766 if (ds->ds_phys->ds_prev_snap_obj == 0)
2769 dsl_deadlist_space(&ds->ds_deadlist,
2770 &dlused, &dlcomp, &dluncomp);
2773 pa->uncomp += dluncomp;
2777 * If we are a clone of a clone then we never reached ORIGIN,
2778 * so we need to subtract out the clone origin's used space.
2780 if (pa->origin_origin) {
2781 pa->used -= pa->origin_origin->ds_phys->ds_referenced_bytes;
2782 pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes;
2783 pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes;
2786 /* Check that there is enough space here */
2787 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2793 * Compute the amounts of space that will be used by snapshots
2794 * after the promotion (for both origin and clone). For each,
2795 * it is the amount of space that will be on all of their
2796 * deadlists (that was not born before their new origin).
2798 if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2802 * Note, typically this will not be a clone of a clone,
2803 * so dd_origin_txg will be < TXG_INITIAL, so
2804 * these snaplist_space() -> dsl_deadlist_space_range()
2805 * calls will be fast because they do not have to
2806 * iterate over all bps.
2808 snap = list_head(&pa->origin_snaps);
2809 err = snaplist_space(&pa->shared_snaps,
2810 snap->ds->ds_dir->dd_origin_txg, &pa->cloneusedsnap);
2814 err = snaplist_space(&pa->clone_snaps,
2815 snap->ds->ds_dir->dd_origin_txg, &space);
2818 pa->cloneusedsnap += space;
2820 if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2821 err = snaplist_space(&pa->origin_snaps,
2822 origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap);
2829 pa->err_ds = snap->ds->ds_snapname;
2834 dsl_dataset_promote_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2836 dsl_dataset_t *hds = arg1;
2837 struct promotearg *pa = arg2;
2838 struct promotenode *snap = list_head(&pa->shared_snaps);
2839 dsl_dataset_t *origin_ds = snap->ds;
2840 dsl_dataset_t *origin_head;
2841 dsl_dir_t *dd = hds->ds_dir;
2842 dsl_pool_t *dp = hds->ds_dir->dd_pool;
2843 dsl_dir_t *odd = NULL;
2844 uint64_t oldnext_obj;
2847 ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
2849 snap = list_head(&pa->origin_snaps);
2850 origin_head = snap->ds;
2853 * We need to explicitly open odd, since origin_ds's dd will be
2856 VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object,
2859 /* change origin's next snap */
2860 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2861 oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2862 snap = list_tail(&pa->clone_snaps);
2863 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2864 origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
2866 /* change the origin's next clone */
2867 if (origin_ds->ds_phys->ds_next_clones_obj) {
2868 remove_from_next_clones(origin_ds, snap->ds->ds_object, tx);
2869 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2870 origin_ds->ds_phys->ds_next_clones_obj,
2875 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2876 ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2877 dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2878 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2879 dmu_buf_will_dirty(odd->dd_dbuf, tx);
2880 odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2881 origin_head->ds_dir->dd_origin_txg =
2882 origin_ds->ds_phys->ds_creation_txg;
2884 /* change dd_clone entries */
2885 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2886 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2887 odd->dd_phys->dd_clones, hds->ds_object, tx));
2888 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2889 pa->origin_origin->ds_dir->dd_phys->dd_clones,
2890 hds->ds_object, tx));
2892 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2893 pa->origin_origin->ds_dir->dd_phys->dd_clones,
2894 origin_head->ds_object, tx));
2895 if (dd->dd_phys->dd_clones == 0) {
2896 dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset,
2897 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
2899 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2900 dd->dd_phys->dd_clones, origin_head->ds_object, tx));
2904 /* move snapshots to this dir */
2905 for (snap = list_head(&pa->shared_snaps); snap;
2906 snap = list_next(&pa->shared_snaps, snap)) {
2907 dsl_dataset_t *ds = snap->ds;
2909 /* unregister props as dsl_dir is changing */
2910 if (ds->ds_objset) {
2911 dmu_objset_evict(ds->ds_objset);
2912 ds->ds_objset = NULL;
2914 /* move snap name entry */
2915 VERIFY(0 == dsl_dataset_get_snapname(ds));
2916 VERIFY(0 == dsl_dataset_snap_remove(origin_head,
2917 ds->ds_snapname, tx));
2918 VERIFY(0 == zap_add(dp->dp_meta_objset,
2919 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2920 8, 1, &ds->ds_object, tx));
2922 /* change containing dsl_dir */
2923 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2924 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2925 ds->ds_phys->ds_dir_obj = dd->dd_object;
2926 ASSERT3P(ds->ds_dir, ==, odd);
2927 dsl_dir_close(ds->ds_dir, ds);
2928 VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
2929 NULL, ds, &ds->ds_dir));
2931 /* move any clone references */
2932 if (ds->ds_phys->ds_next_clones_obj &&
2933 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2937 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2938 ds->ds_phys->ds_next_clones_obj);
2939 zap_cursor_retrieve(&zc, &za) == 0;
2940 zap_cursor_advance(&zc)) {
2941 dsl_dataset_t *cnds;
2944 if (za.za_first_integer == oldnext_obj) {
2946 * We've already moved the
2947 * origin's reference.
2952 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
2953 za.za_first_integer, FTAG, &cnds));
2954 o = cnds->ds_dir->dd_phys->dd_head_dataset_obj;
2956 VERIFY3U(zap_remove_int(dp->dp_meta_objset,
2957 odd->dd_phys->dd_clones, o, tx), ==, 0);
2958 VERIFY3U(zap_add_int(dp->dp_meta_objset,
2959 dd->dd_phys->dd_clones, o, tx), ==, 0);
2960 dsl_dataset_rele(cnds, FTAG);
2962 zap_cursor_fini(&zc);
2965 ASSERT0(dsl_prop_numcb(ds));
2969 * Change space accounting.
2970 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2971 * both be valid, or both be 0 (resulting in delta == 0). This
2972 * is true for each of {clone,origin} independently.
2975 delta = pa->cloneusedsnap -
2976 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2977 ASSERT3S(delta, >=, 0);
2978 ASSERT3U(pa->used, >=, delta);
2979 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2980 dsl_dir_diduse_space(dd, DD_USED_HEAD,
2981 pa->used - delta, pa->comp, pa->uncomp, tx);
2983 delta = pa->originusedsnap -
2984 odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2985 ASSERT3S(delta, <=, 0);
2986 ASSERT3U(pa->used, >=, -delta);
2987 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2988 dsl_dir_diduse_space(odd, DD_USED_HEAD,
2989 -pa->used - delta, -pa->comp, -pa->uncomp, tx);
2991 origin_ds->ds_phys->ds_unique_bytes = pa->unique;
2993 /* log history record */
2994 spa_history_log_internal(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx,
2995 "dataset = %llu", hds->ds_object);
2997 dsl_dir_close(odd, FTAG);
3000 static char *snaplist_tag = "snaplist";
3002 * Make a list of dsl_dataset_t's for the snapshots between first_obj
3003 * (exclusive) and last_obj (inclusive). The list will be in reverse
3004 * order (last_obj will be the list_head()). If first_obj == 0, do all
3005 * snapshots back to this dataset's origin.
3008 snaplist_make(dsl_pool_t *dp, boolean_t own,
3009 uint64_t first_obj, uint64_t last_obj, list_t *l)
3011 uint64_t obj = last_obj;
3013 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock));
3015 list_create(l, sizeof (struct promotenode),
3016 offsetof(struct promotenode, link));
3018 while (obj != first_obj) {
3020 struct promotenode *snap;
3024 err = dsl_dataset_own_obj(dp, obj,
3025 0, snaplist_tag, &ds);
3027 dsl_dataset_make_exclusive(ds, snaplist_tag);
3029 err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds);
3031 if (err == ENOENT) {
3032 /* lost race with snapshot destroy */
3033 struct promotenode *last = list_tail(l);
3034 ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj);
3035 obj = last->ds->ds_phys->ds_prev_snap_obj;
3042 first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
3044 snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP);
3046 list_insert_tail(l, snap);
3047 obj = ds->ds_phys->ds_prev_snap_obj;
3054 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
3056 struct promotenode *snap;
3059 for (snap = list_head(l); snap; snap = list_next(l, snap)) {
3060 uint64_t used, comp, uncomp;
3061 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
3062 mintxg, UINT64_MAX, &used, &comp, &uncomp);
3069 snaplist_destroy(list_t *l, boolean_t own)
3071 struct promotenode *snap;
3073 if (!l || !list_link_active(&l->list_head))
3076 while ((snap = list_tail(l)) != NULL) {
3077 list_remove(l, snap);
3079 dsl_dataset_disown(snap->ds, snaplist_tag);
3081 dsl_dataset_rele(snap->ds, snaplist_tag);
3082 kmem_free(snap, sizeof (struct promotenode));
3088 * Promote a clone. Nomenclature note:
3089 * "clone" or "cds": the original clone which is being promoted
3090 * "origin" or "ods": the snapshot which is originally clone's origin
3091 * "origin head" or "ohds": the dataset which is the head
3092 * (filesystem/volume) for the origin
3093 * "origin origin": the origin of the origin's filesystem (typically
3094 * NULL, indicating that the clone is not a clone of a clone).
3097 dsl_dataset_promote(const char *name, char *conflsnap)
3102 dmu_object_info_t doi;
3103 struct promotearg pa = { 0 };
3104 struct promotenode *snap;
3107 err = dsl_dataset_hold(name, FTAG, &ds);
3113 err = dmu_object_info(dp->dp_meta_objset,
3114 ds->ds_phys->ds_snapnames_zapobj, &doi);
3116 dsl_dataset_rele(ds, FTAG);
3120 if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) {
3121 dsl_dataset_rele(ds, FTAG);
3126 * We are going to inherit all the snapshots taken before our
3127 * origin (i.e., our new origin will be our parent's origin).
3128 * Take ownership of them so that we can rename them into our
3131 rw_enter(&dp->dp_config_rwlock, RW_READER);
3133 err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj,
3138 err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps);
3142 snap = list_head(&pa.shared_snaps);
3143 ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
3144 err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj,
3145 snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps);
3149 if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) {
3150 err = dsl_dataset_hold_obj(dp,
3151 snap->ds->ds_dir->dd_phys->dd_origin_obj,
3152 FTAG, &pa.origin_origin);
3158 rw_exit(&dp->dp_config_rwlock);
3161 * Add in 128x the snapnames zapobj size, since we will be moving
3162 * a bunch of snapnames to the promoted ds, and dirtying their
3166 err = dsl_sync_task_do(dp, dsl_dataset_promote_check,
3167 dsl_dataset_promote_sync, ds, &pa,
3168 2 + 2 * doi.doi_physical_blocks_512);
3169 if (err && pa.err_ds && conflsnap)
3170 (void) strncpy(conflsnap, pa.err_ds, MAXNAMELEN);
3173 snaplist_destroy(&pa.shared_snaps, B_TRUE);
3174 snaplist_destroy(&pa.clone_snaps, B_FALSE);
3175 snaplist_destroy(&pa.origin_snaps, B_FALSE);
3176 if (pa.origin_origin)
3177 dsl_dataset_rele(pa.origin_origin, FTAG);
3178 dsl_dataset_rele(ds, FTAG);
3182 struct cloneswaparg {
3183 dsl_dataset_t *cds; /* clone dataset */
3184 dsl_dataset_t *ohds; /* origin's head dataset */
3186 int64_t unused_refres_delta; /* change in unconsumed refreservation */
3191 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx)
3193 struct cloneswaparg *csa = arg1;
3195 /* they should both be heads */
3196 if (dsl_dataset_is_snapshot(csa->cds) ||
3197 dsl_dataset_is_snapshot(csa->ohds))
3200 /* the branch point should be just before them */
3201 if (csa->cds->ds_prev != csa->ohds->ds_prev)
3204 /* cds should be the clone (unless they are unrelated) */
3205 if (csa->cds->ds_prev != NULL &&
3206 csa->cds->ds_prev != csa->cds->ds_dir->dd_pool->dp_origin_snap &&
3207 csa->ohds->ds_object !=
3208 csa->cds->ds_prev->ds_phys->ds_next_snap_obj)
3211 /* the clone should be a child of the origin */
3212 if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir)
3215 /* ohds shouldn't be modified unless 'force' */
3216 if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds))
3219 /* adjust amount of any unconsumed refreservation */
3220 csa->unused_refres_delta =
3221 (int64_t)MIN(csa->ohds->ds_reserved,
3222 csa->ohds->ds_phys->ds_unique_bytes) -
3223 (int64_t)MIN(csa->ohds->ds_reserved,
3224 csa->cds->ds_phys->ds_unique_bytes);
3226 if (csa->unused_refres_delta > 0 &&
3227 csa->unused_refres_delta >
3228 dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE))
3231 if (csa->ohds->ds_quota != 0 &&
3232 csa->cds->ds_phys->ds_unique_bytes > csa->ohds->ds_quota)
3240 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3242 struct cloneswaparg *csa = arg1;
3243 dsl_pool_t *dp = csa->cds->ds_dir->dd_pool;
3245 ASSERT(csa->cds->ds_reserved == 0);
3246 ASSERT(csa->ohds->ds_quota == 0 ||
3247 csa->cds->ds_phys->ds_unique_bytes <= csa->ohds->ds_quota);
3249 dmu_buf_will_dirty(csa->cds->ds_dbuf, tx);
3250 dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx);
3252 if (csa->cds->ds_objset != NULL) {
3253 dmu_objset_evict(csa->cds->ds_objset);
3254 csa->cds->ds_objset = NULL;
3257 if (csa->ohds->ds_objset != NULL) {
3258 dmu_objset_evict(csa->ohds->ds_objset);
3259 csa->ohds->ds_objset = NULL;
3263 * Reset origin's unique bytes, if it exists.
3265 if (csa->cds->ds_prev) {
3266 dsl_dataset_t *origin = csa->cds->ds_prev;
3267 uint64_t comp, uncomp;
3269 dmu_buf_will_dirty(origin->ds_dbuf, tx);
3270 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3271 origin->ds_phys->ds_prev_snap_txg, UINT64_MAX,
3272 &origin->ds_phys->ds_unique_bytes, &comp, &uncomp);
3278 tmp = csa->ohds->ds_phys->ds_bp;
3279 csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp;
3280 csa->cds->ds_phys->ds_bp = tmp;
3283 /* set dd_*_bytes */
3285 int64_t dused, dcomp, duncomp;
3286 uint64_t cdl_used, cdl_comp, cdl_uncomp;
3287 uint64_t odl_used, odl_comp, odl_uncomp;
3289 ASSERT3U(csa->cds->ds_dir->dd_phys->
3290 dd_used_breakdown[DD_USED_SNAP], ==, 0);
3292 dsl_deadlist_space(&csa->cds->ds_deadlist,
3293 &cdl_used, &cdl_comp, &cdl_uncomp);
3294 dsl_deadlist_space(&csa->ohds->ds_deadlist,
3295 &odl_used, &odl_comp, &odl_uncomp);
3297 dused = csa->cds->ds_phys->ds_referenced_bytes + cdl_used -
3298 (csa->ohds->ds_phys->ds_referenced_bytes + odl_used);
3299 dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp -
3300 (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp);
3301 duncomp = csa->cds->ds_phys->ds_uncompressed_bytes +
3303 (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp);
3305 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD,
3306 dused, dcomp, duncomp, tx);
3307 dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD,
3308 -dused, -dcomp, -duncomp, tx);
3311 * The difference in the space used by snapshots is the
3312 * difference in snapshot space due to the head's
3313 * deadlist (since that's the only thing that's
3314 * changing that affects the snapused).
3316 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3317 csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3318 &cdl_used, &cdl_comp, &cdl_uncomp);
3319 dsl_deadlist_space_range(&csa->ohds->ds_deadlist,
3320 csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3321 &odl_used, &odl_comp, &odl_uncomp);
3322 dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used,
3323 DD_USED_HEAD, DD_USED_SNAP, tx);
3326 /* swap ds_*_bytes */
3327 SWITCH64(csa->ohds->ds_phys->ds_referenced_bytes,
3328 csa->cds->ds_phys->ds_referenced_bytes);
3329 SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes,
3330 csa->cds->ds_phys->ds_compressed_bytes);
3331 SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes,
3332 csa->cds->ds_phys->ds_uncompressed_bytes);
3333 SWITCH64(csa->ohds->ds_phys->ds_unique_bytes,
3334 csa->cds->ds_phys->ds_unique_bytes);
3336 /* apply any parent delta for change in unconsumed refreservation */
3337 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV,
3338 csa->unused_refres_delta, 0, 0, tx);
3343 dsl_deadlist_close(&csa->cds->ds_deadlist);
3344 dsl_deadlist_close(&csa->ohds->ds_deadlist);
3345 SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj,
3346 csa->cds->ds_phys->ds_deadlist_obj);
3347 dsl_deadlist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset,
3348 csa->cds->ds_phys->ds_deadlist_obj);
3349 dsl_deadlist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset,
3350 csa->ohds->ds_phys->ds_deadlist_obj);
3352 dsl_scan_ds_clone_swapped(csa->ohds, csa->cds, tx);
3356 * Swap 'clone' with its origin head datasets. Used at the end of "zfs
3357 * recv" into an existing fs to swizzle the file system to the new
3358 * version, and by "zfs rollback". Can also be used to swap two
3359 * independent head datasets if neither has any snapshots.
3362 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
3365 struct cloneswaparg csa;
3368 ASSERT(clone->ds_owner);
3369 ASSERT(origin_head->ds_owner);
3372 * Need exclusive access for the swap. If we're swapping these
3373 * datasets back after an error, we already hold the locks.
3375 if (!RW_WRITE_HELD(&clone->ds_rwlock))
3376 rw_enter(&clone->ds_rwlock, RW_WRITER);
3377 if (!RW_WRITE_HELD(&origin_head->ds_rwlock) &&
3378 !rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) {
3379 rw_exit(&clone->ds_rwlock);
3380 rw_enter(&origin_head->ds_rwlock, RW_WRITER);
3381 if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) {
3382 rw_exit(&origin_head->ds_rwlock);
3387 csa.ohds = origin_head;
3389 error = dsl_sync_task_do(clone->ds_dir->dd_pool,
3390 dsl_dataset_clone_swap_check,
3391 dsl_dataset_clone_swap_sync, &csa, NULL, 9);
3396 * Given a pool name and a dataset object number in that pool,
3397 * return the name of that dataset.
3400 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3407 if ((error = spa_open(pname, &spa, FTAG)) != 0)
3409 dp = spa_get_dsl(spa);
3410 rw_enter(&dp->dp_config_rwlock, RW_READER);
3411 if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) {
3412 dsl_dataset_name(ds, buf);
3413 dsl_dataset_rele(ds, FTAG);
3415 rw_exit(&dp->dp_config_rwlock);
3416 spa_close(spa, FTAG);
3422 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3423 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3427 ASSERT3S(asize, >, 0);
3430 * *ref_rsrv is the portion of asize that will come from any
3431 * unconsumed refreservation space.
3435 mutex_enter(&ds->ds_lock);
3437 * Make a space adjustment for reserved bytes.
3439 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
3441 ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3442 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3444 asize - MIN(asize, parent_delta(ds, asize + inflight));
3447 if (!check_quota || ds->ds_quota == 0) {
3448 mutex_exit(&ds->ds_lock);
3452 * If they are requesting more space, and our current estimate
3453 * is over quota, they get to try again unless the actual
3454 * on-disk is over quota and there are no pending changes (which
3455 * may free up space for us).
3457 if (ds->ds_phys->ds_referenced_bytes + inflight >= ds->ds_quota) {
3459 ds->ds_phys->ds_referenced_bytes < ds->ds_quota)
3464 mutex_exit(&ds->ds_lock);
3471 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
3473 dsl_dataset_t *ds = arg1;
3474 dsl_prop_setarg_t *psa = arg2;
3477 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA)
3480 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3483 if (psa->psa_effective_value == 0)
3486 if (psa->psa_effective_value < ds->ds_phys->ds_referenced_bytes ||
3487 psa->psa_effective_value < ds->ds_reserved)
3493 extern void dsl_prop_set_sync(void *, void *, dmu_tx_t *);
3496 dsl_dataset_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3498 dsl_dataset_t *ds = arg1;
3499 dsl_prop_setarg_t *psa = arg2;
3500 uint64_t effective_value = psa->psa_effective_value;
3502 dsl_prop_set_sync(ds, psa, tx);
3503 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3505 if (ds->ds_quota != effective_value) {
3506 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3507 ds->ds_quota = effective_value;
3512 dsl_dataset_set_quota(const char *dsname, zprop_source_t source, uint64_t quota)
3515 dsl_prop_setarg_t psa;
3518 dsl_prop_setarg_init_uint64(&psa, "refquota", source, "a);
3520 err = dsl_dataset_hold(dsname, FTAG, &ds);
3525 * If someone removes a file, then tries to set the quota, we
3526 * want to make sure the file freeing takes effect.
3528 txg_wait_open(ds->ds_dir->dd_pool, 0);
3530 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3531 dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync,
3534 dsl_dataset_rele(ds, FTAG);
3539 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
3541 dsl_dataset_t *ds = arg1;
3542 dsl_prop_setarg_t *psa = arg2;
3543 uint64_t effective_value;
3547 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
3548 SPA_VERSION_REFRESERVATION)
3551 if (dsl_dataset_is_snapshot(ds))
3554 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3557 effective_value = psa->psa_effective_value;
3560 * If we are doing the preliminary check in open context, the
3561 * space estimates may be inaccurate.
3563 if (!dmu_tx_is_syncing(tx))
3566 mutex_enter(&ds->ds_lock);
3567 if (!DS_UNIQUE_IS_ACCURATE(ds))
3568 dsl_dataset_recalc_head_uniq(ds);
3569 unique = ds->ds_phys->ds_unique_bytes;
3570 mutex_exit(&ds->ds_lock);
3572 if (MAX(unique, effective_value) > MAX(unique, ds->ds_reserved)) {
3573 uint64_t delta = MAX(unique, effective_value) -
3574 MAX(unique, ds->ds_reserved);
3576 if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
3578 if (ds->ds_quota > 0 &&
3579 effective_value > ds->ds_quota)
3587 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3589 dsl_dataset_t *ds = arg1;
3590 dsl_prop_setarg_t *psa = arg2;
3591 uint64_t effective_value = psa->psa_effective_value;
3595 dsl_prop_set_sync(ds, psa, tx);
3596 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3598 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3600 mutex_enter(&ds->ds_dir->dd_lock);
3601 mutex_enter(&ds->ds_lock);
3602 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3603 unique = ds->ds_phys->ds_unique_bytes;
3604 delta = MAX(0, (int64_t)(effective_value - unique)) -
3605 MAX(0, (int64_t)(ds->ds_reserved - unique));
3606 ds->ds_reserved = effective_value;
3607 mutex_exit(&ds->ds_lock);
3609 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3610 mutex_exit(&ds->ds_dir->dd_lock);
3614 dsl_dataset_set_reservation(const char *dsname, zprop_source_t source,
3615 uint64_t reservation)
3618 dsl_prop_setarg_t psa;
3621 dsl_prop_setarg_init_uint64(&psa, "refreservation", source,
3624 err = dsl_dataset_hold(dsname, FTAG, &ds);
3628 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3629 dsl_dataset_set_reservation_check,
3630 dsl_dataset_set_reservation_sync, ds, &psa, 0);
3632 dsl_dataset_rele(ds, FTAG);
3636 typedef struct zfs_hold_cleanup_arg {
3639 char htag[MAXNAMELEN];
3640 } zfs_hold_cleanup_arg_t;
3643 dsl_dataset_user_release_onexit(void *arg)
3645 zfs_hold_cleanup_arg_t *ca = arg;
3647 (void) dsl_dataset_user_release_tmp(ca->dp, ca->dsobj, ca->htag,
3649 kmem_free(ca, sizeof (zfs_hold_cleanup_arg_t));
3653 dsl_register_onexit_hold_cleanup(dsl_dataset_t *ds, const char *htag,
3656 zfs_hold_cleanup_arg_t *ca;
3658 ca = kmem_alloc(sizeof (zfs_hold_cleanup_arg_t), KM_SLEEP);
3659 ca->dp = ds->ds_dir->dd_pool;
3660 ca->dsobj = ds->ds_object;
3661 (void) strlcpy(ca->htag, htag, sizeof (ca->htag));
3662 VERIFY3U(0, ==, zfs_onexit_add_cb(minor,
3663 dsl_dataset_user_release_onexit, ca, NULL));
3667 * If you add new checks here, you may need to add
3668 * additional checks to the "temporary" case in
3669 * snapshot_check() in dmu_objset.c.
3672 dsl_dataset_user_hold_check(void *arg1, void *arg2, dmu_tx_t *tx)
3674 dsl_dataset_t *ds = arg1;
3675 struct dsl_ds_holdarg *ha = arg2;
3676 char *htag = ha->htag;
3677 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3680 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3683 if (!dsl_dataset_is_snapshot(ds))
3686 /* tags must be unique */
3687 mutex_enter(&ds->ds_lock);
3688 if (ds->ds_phys->ds_userrefs_obj) {
3689 error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj, htag,
3693 else if (error == ENOENT)
3696 mutex_exit(&ds->ds_lock);
3698 if (error == 0 && ha->temphold &&
3699 strlen(htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN)
3706 dsl_dataset_user_hold_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3708 dsl_dataset_t *ds = arg1;
3709 struct dsl_ds_holdarg *ha = arg2;
3710 char *htag = ha->htag;
3711 dsl_pool_t *dp = ds->ds_dir->dd_pool;
3712 objset_t *mos = dp->dp_meta_objset;
3713 uint64_t now = gethrestime_sec();
3716 mutex_enter(&ds->ds_lock);
3717 if (ds->ds_phys->ds_userrefs_obj == 0) {
3719 * This is the first user hold for this dataset. Create
3720 * the userrefs zap object.
3722 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3723 zapobj = ds->ds_phys->ds_userrefs_obj =
3724 zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx);
3726 zapobj = ds->ds_phys->ds_userrefs_obj;
3729 mutex_exit(&ds->ds_lock);
3731 VERIFY(0 == zap_add(mos, zapobj, htag, 8, 1, &now, tx));
3734 VERIFY(0 == dsl_pool_user_hold(dp, ds->ds_object,
3738 spa_history_log_internal(LOG_DS_USER_HOLD,
3739 dp->dp_spa, tx, "<%s> temp = %d dataset = %llu", htag,
3740 (int)ha->temphold, ds->ds_object);
3744 dsl_dataset_user_hold_one(const char *dsname, void *arg)
3746 struct dsl_ds_holdarg *ha = arg;
3751 /* alloc a buffer to hold dsname@snapname plus terminating NULL */
3752 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3753 error = dsl_dataset_hold(name, ha->dstg, &ds);
3756 ha->gotone = B_TRUE;
3757 dsl_sync_task_create(ha->dstg, dsl_dataset_user_hold_check,
3758 dsl_dataset_user_hold_sync, ds, ha, 0);
3759 } else if (error == ENOENT && ha->recursive) {
3762 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3768 dsl_dataset_user_hold_for_send(dsl_dataset_t *ds, char *htag,
3771 struct dsl_ds_holdarg *ha;
3774 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3776 ha->temphold = temphold;
3777 error = dsl_sync_task_do(ds->ds_dir->dd_pool,
3778 dsl_dataset_user_hold_check, dsl_dataset_user_hold_sync,
3780 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3786 dsl_dataset_user_hold(char *dsname, char *snapname, char *htag,
3787 boolean_t recursive, boolean_t temphold, int cleanup_fd)
3789 struct dsl_ds_holdarg *ha;
3790 dsl_sync_task_t *dst;
3795 if (cleanup_fd != -1) {
3796 /* Currently we only support cleanup-on-exit of tempholds. */
3799 error = zfs_onexit_fd_hold(cleanup_fd, &minor);
3804 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3806 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3808 error = spa_open(dsname, &spa, FTAG);
3810 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3811 if (cleanup_fd != -1)
3812 zfs_onexit_fd_rele(cleanup_fd);
3816 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3818 ha->snapname = snapname;
3819 ha->recursive = recursive;
3820 ha->temphold = temphold;
3823 error = dmu_objset_find(dsname, dsl_dataset_user_hold_one,
3824 ha, DS_FIND_CHILDREN);
3826 error = dsl_dataset_user_hold_one(dsname, ha);
3829 error = dsl_sync_task_group_wait(ha->dstg);
3831 for (dst = list_head(&ha->dstg->dstg_tasks); dst;
3832 dst = list_next(&ha->dstg->dstg_tasks, dst)) {
3833 dsl_dataset_t *ds = dst->dst_arg1;
3836 dsl_dataset_name(ds, ha->failed);
3837 *strchr(ha->failed, '@') = '\0';
3838 } else if (error == 0 && minor != 0 && temphold) {
3840 * If this hold is to be released upon process exit,
3841 * register that action now.
3843 dsl_register_onexit_hold_cleanup(ds, htag, minor);
3845 dsl_dataset_rele(ds, ha->dstg);
3848 if (error == 0 && recursive && !ha->gotone)
3852 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
3854 dsl_sync_task_group_destroy(ha->dstg);
3856 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3857 spa_close(spa, FTAG);
3858 if (cleanup_fd != -1)
3859 zfs_onexit_fd_rele(cleanup_fd);
3863 struct dsl_ds_releasearg {
3866 boolean_t own; /* do we own or just hold ds? */
3870 dsl_dataset_release_might_destroy(dsl_dataset_t *ds, const char *htag,
3871 boolean_t *might_destroy)
3873 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3878 *might_destroy = B_FALSE;
3880 mutex_enter(&ds->ds_lock);
3881 zapobj = ds->ds_phys->ds_userrefs_obj;
3883 /* The tag can't possibly exist */
3884 mutex_exit(&ds->ds_lock);
3888 /* Make sure the tag exists */
3889 error = zap_lookup(mos, zapobj, htag, 8, 1, &tmp);
3891 mutex_exit(&ds->ds_lock);
3892 if (error == ENOENT)
3897 if (ds->ds_userrefs == 1 && ds->ds_phys->ds_num_children == 1 &&
3898 DS_IS_DEFER_DESTROY(ds))
3899 *might_destroy = B_TRUE;
3901 mutex_exit(&ds->ds_lock);
3906 dsl_dataset_user_release_check(void *arg1, void *tag, dmu_tx_t *tx)
3908 struct dsl_ds_releasearg *ra = arg1;
3909 dsl_dataset_t *ds = ra->ds;
3910 boolean_t might_destroy;
3913 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3916 error = dsl_dataset_release_might_destroy(ds, ra->htag, &might_destroy);
3920 if (might_destroy) {
3921 struct dsl_ds_destroyarg dsda = {0};
3923 if (dmu_tx_is_syncing(tx)) {
3925 * If we're not prepared to remove the snapshot,
3926 * we can't allow the release to happen right now.
3932 dsda.releasing = B_TRUE;
3933 return (dsl_dataset_destroy_check(&dsda, tag, tx));
3940 dsl_dataset_user_release_sync(void *arg1, void *tag, dmu_tx_t *tx)
3942 struct dsl_ds_releasearg *ra = arg1;
3943 dsl_dataset_t *ds = ra->ds;
3944 dsl_pool_t *dp = ds->ds_dir->dd_pool;
3945 objset_t *mos = dp->dp_meta_objset;
3947 uint64_t dsobj = ds->ds_object;
3951 mutex_enter(&ds->ds_lock);
3953 refs = ds->ds_userrefs;
3954 mutex_exit(&ds->ds_lock);
3955 error = dsl_pool_user_release(dp, ds->ds_object, ra->htag, tx);
3956 VERIFY(error == 0 || error == ENOENT);
3957 zapobj = ds->ds_phys->ds_userrefs_obj;
3958 VERIFY(0 == zap_remove(mos, zapobj, ra->htag, tx));
3960 spa_history_log_internal(LOG_DS_USER_RELEASE,
3961 dp->dp_spa, tx, "<%s> %lld dataset = %llu",
3962 ra->htag, (longlong_t)refs, dsobj);
3964 if (ds->ds_userrefs == 0 && ds->ds_phys->ds_num_children == 1 &&
3965 DS_IS_DEFER_DESTROY(ds)) {
3966 struct dsl_ds_destroyarg dsda = {0};
3970 dsda.releasing = B_TRUE;
3971 /* We already did the destroy_check */
3972 dsl_dataset_destroy_sync(&dsda, tag, tx);
3977 dsl_dataset_user_release_one(const char *dsname, void *arg)
3979 struct dsl_ds_holdarg *ha = arg;
3980 struct dsl_ds_releasearg *ra;
3983 void *dtag = ha->dstg;
3985 boolean_t own = B_FALSE;
3986 boolean_t might_destroy;
3988 /* alloc a buffer to hold dsname@snapname, plus the terminating NULL */
3989 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3990 error = dsl_dataset_hold(name, dtag, &ds);
3992 if (error == ENOENT && ha->recursive)
3994 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3998 ha->gotone = B_TRUE;
4000 ASSERT(dsl_dataset_is_snapshot(ds));
4002 error = dsl_dataset_release_might_destroy(ds, ha->htag, &might_destroy);
4004 dsl_dataset_rele(ds, dtag);
4008 if (might_destroy) {
4010 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
4011 error = zfs_unmount_snap(name, NULL);
4014 dsl_dataset_rele(ds, dtag);
4018 if (!dsl_dataset_tryown(ds, B_TRUE, dtag)) {
4019 dsl_dataset_rele(ds, dtag);
4023 dsl_dataset_make_exclusive(ds, dtag);
4027 ra = kmem_alloc(sizeof (struct dsl_ds_releasearg), KM_SLEEP);
4029 ra->htag = ha->htag;
4031 dsl_sync_task_create(ha->dstg, dsl_dataset_user_release_check,
4032 dsl_dataset_user_release_sync, ra, dtag, 0);
4038 dsl_dataset_user_release(char *dsname, char *snapname, char *htag,
4039 boolean_t recursive)
4041 struct dsl_ds_holdarg *ha;
4042 dsl_sync_task_t *dst;
4047 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
4049 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
4051 error = spa_open(dsname, &spa, FTAG);
4053 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
4057 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
4059 ha->snapname = snapname;
4060 ha->recursive = recursive;
4062 error = dmu_objset_find(dsname, dsl_dataset_user_release_one,
4063 ha, DS_FIND_CHILDREN);
4065 error = dsl_dataset_user_release_one(dsname, ha);
4068 error = dsl_sync_task_group_wait(ha->dstg);
4070 for (dst = list_head(&ha->dstg->dstg_tasks); dst;
4071 dst = list_next(&ha->dstg->dstg_tasks, dst)) {
4072 struct dsl_ds_releasearg *ra = dst->dst_arg1;
4073 dsl_dataset_t *ds = ra->ds;
4076 dsl_dataset_name(ds, ha->failed);
4079 dsl_dataset_disown(ds, ha->dstg);
4081 dsl_dataset_rele(ds, ha->dstg);
4083 kmem_free(ra, sizeof (struct dsl_ds_releasearg));
4086 if (error == 0 && recursive && !ha->gotone)
4089 if (error && error != EBUSY)
4090 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
4092 dsl_sync_task_group_destroy(ha->dstg);
4093 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
4094 spa_close(spa, FTAG);
4097 * We can get EBUSY if we were racing with deferred destroy and
4098 * dsl_dataset_user_release_check() hadn't done the necessary
4099 * open context setup. We can also get EBUSY if we're racing
4100 * with destroy and that thread is the ds_owner. Either way
4101 * the busy condition should be transient, and we should retry
4102 * the release operation.
4111 * Called at spa_load time (with retry == B_FALSE) to release a stale
4112 * temporary user hold. Also called by the onexit code (with retry == B_TRUE).
4115 dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, char *htag,
4125 rw_enter(&dp->dp_config_rwlock, RW_READER);
4126 error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
4127 rw_exit(&dp->dp_config_rwlock);
4130 namelen = dsl_dataset_namelen(ds)+1;
4131 name = kmem_alloc(namelen, KM_SLEEP);
4132 dsl_dataset_name(ds, name);
4133 dsl_dataset_rele(ds, FTAG);
4135 snap = strchr(name, '@');
4138 error = dsl_dataset_user_release(name, snap, htag, B_FALSE);
4139 kmem_free(name, namelen);
4142 * The object can't have been destroyed because we have a hold,
4143 * but it might have been renamed, resulting in ENOENT. Retry
4144 * if we've been requested to do so.
4146 * It would be nice if we could use the dsobj all the way
4147 * through and avoid ENOENT entirely. But we might need to
4148 * unmount the snapshot, and there's currently no way to lookup
4149 * a vfsp using a ZFS object id.
4151 } while ((error == ENOENT) && retry);
4157 dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp)
4162 err = dsl_dataset_hold(dsname, FTAG, &ds);
4166 VERIFY(0 == nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP));
4167 if (ds->ds_phys->ds_userrefs_obj != 0) {
4168 zap_attribute_t *za;
4171 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
4172 for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset,
4173 ds->ds_phys->ds_userrefs_obj);
4174 zap_cursor_retrieve(&zc, za) == 0;
4175 zap_cursor_advance(&zc)) {
4176 VERIFY(0 == nvlist_add_uint64(*nvp, za->za_name,
4177 za->za_first_integer));
4179 zap_cursor_fini(&zc);
4180 kmem_free(za, sizeof (zap_attribute_t));
4182 dsl_dataset_rele(ds, FTAG);
4187 * Note, this function is used as the callback for dmu_objset_find(). We
4188 * always return 0 so that we will continue to find and process
4189 * inconsistent datasets, even if we encounter an error trying to
4190 * process one of them.
4194 dsl_destroy_inconsistent(const char *dsname, void *arg)
4198 if (dsl_dataset_own(dsname, B_TRUE, FTAG, &ds) == 0) {
4199 if (DS_IS_INCONSISTENT(ds))
4200 (void) dsl_dataset_destroy(ds, FTAG, B_FALSE);
4202 dsl_dataset_disown(ds, FTAG);
4208 * Return (in *usedp) the amount of space written in new that is not
4209 * present in oldsnap. New may be a snapshot or the head. Old must be
4210 * a snapshot before new, in new's filesystem (or its origin). If not then
4211 * fail and return EINVAL.
4213 * The written space is calculated by considering two components: First, we
4214 * ignore any freed space, and calculate the written as new's used space
4215 * minus old's used space. Next, we add in the amount of space that was freed
4216 * between the two snapshots, thus reducing new's used space relative to old's.
4217 * Specifically, this is the space that was born before old->ds_creation_txg,
4218 * and freed before new (ie. on new's deadlist or a previous deadlist).
4220 * space freed [---------------------]
4221 * snapshots ---O-------O--------O-------O------
4225 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
4226 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4230 dsl_pool_t *dp = new->ds_dir->dd_pool;
4233 *usedp += new->ds_phys->ds_referenced_bytes;
4234 *usedp -= oldsnap->ds_phys->ds_referenced_bytes;
4237 *compp += new->ds_phys->ds_compressed_bytes;
4238 *compp -= oldsnap->ds_phys->ds_compressed_bytes;
4241 *uncompp += new->ds_phys->ds_uncompressed_bytes;
4242 *uncompp -= oldsnap->ds_phys->ds_uncompressed_bytes;
4244 rw_enter(&dp->dp_config_rwlock, RW_READER);
4245 snapobj = new->ds_object;
4246 while (snapobj != oldsnap->ds_object) {
4247 dsl_dataset_t *snap;
4248 uint64_t used, comp, uncomp;
4250 if (snapobj == new->ds_object) {
4253 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
4258 if (snap->ds_phys->ds_prev_snap_txg ==
4259 oldsnap->ds_phys->ds_creation_txg) {
4261 * The blocks in the deadlist can not be born after
4262 * ds_prev_snap_txg, so get the whole deadlist space,
4263 * which is more efficient (especially for old-format
4264 * deadlists). Unfortunately the deadlist code
4265 * doesn't have enough information to make this
4266 * optimization itself.
4268 dsl_deadlist_space(&snap->ds_deadlist,
4269 &used, &comp, &uncomp);
4271 dsl_deadlist_space_range(&snap->ds_deadlist,
4272 0, oldsnap->ds_phys->ds_creation_txg,
4273 &used, &comp, &uncomp);
4280 * If we get to the beginning of the chain of snapshots
4281 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
4282 * was not a snapshot of/before new.
4284 snapobj = snap->ds_phys->ds_prev_snap_obj;
4286 dsl_dataset_rele(snap, FTAG);
4293 rw_exit(&dp->dp_config_rwlock);
4298 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
4299 * lastsnap, and all snapshots in between are deleted.
4301 * blocks that would be freed [---------------------------]
4302 * snapshots ---O-------O--------O-------O--------O
4303 * firstsnap lastsnap
4305 * This is the set of blocks that were born after the snap before firstsnap,
4306 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
4307 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
4308 * We calculate this by iterating over the relevant deadlists (from the snap
4309 * after lastsnap, backward to the snap after firstsnap), summing up the
4310 * space on the deadlist that was born after the snap before firstsnap.
4313 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
4314 dsl_dataset_t *lastsnap,
4315 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4319 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
4321 ASSERT(dsl_dataset_is_snapshot(firstsnap));
4322 ASSERT(dsl_dataset_is_snapshot(lastsnap));
4325 * Check that the snapshots are in the same dsl_dir, and firstsnap
4326 * is before lastsnap.
4328 if (firstsnap->ds_dir != lastsnap->ds_dir ||
4329 firstsnap->ds_phys->ds_creation_txg >
4330 lastsnap->ds_phys->ds_creation_txg)
4333 *usedp = *compp = *uncompp = 0;
4335 rw_enter(&dp->dp_config_rwlock, RW_READER);
4336 snapobj = lastsnap->ds_phys->ds_next_snap_obj;
4337 while (snapobj != firstsnap->ds_object) {
4339 uint64_t used, comp, uncomp;
4341 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
4345 dsl_deadlist_space_range(&ds->ds_deadlist,
4346 firstsnap->ds_phys->ds_prev_snap_txg, UINT64_MAX,
4347 &used, &comp, &uncomp);
4352 snapobj = ds->ds_phys->ds_prev_snap_obj;
4353 ASSERT3U(snapobj, !=, 0);
4354 dsl_dataset_rele(ds, FTAG);
4356 rw_exit(&dp->dp_config_rwlock);