4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011 by Delphix. All rights reserved.
26 #include <sys/dmu_objset.h>
27 #include <sys/dsl_dataset.h>
28 #include <sys/dsl_dir.h>
29 #include <sys/dsl_prop.h>
30 #include <sys/dsl_synctask.h>
31 #include <sys/dmu_traverse.h>
32 #include <sys/dmu_tx.h>
36 #include <sys/unique.h>
37 #include <sys/zfs_context.h>
38 #include <sys/zfs_ioctl.h>
40 #include <sys/zfs_znode.h>
41 #include <sys/zfs_onexit.h>
43 #include <sys/dsl_scan.h>
44 #include <sys/dsl_deadlist.h>
46 static char *dsl_reaper = "the grim reaper";
48 static dsl_checkfunc_t dsl_dataset_destroy_begin_check;
49 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync;
50 static dsl_syncfunc_t dsl_dataset_set_reservation_sync;
52 #define SWITCH64(x, y) \
54 uint64_t __tmp = (x); \
59 #define DS_REF_MAX (1ULL << 62)
61 #define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE
63 #define DSL_DATASET_IS_DESTROYED(ds) ((ds)->ds_owner == dsl_reaper)
67 * Figure out how much of this delta should be propogated to the dsl_dir
68 * layer. If there's a refreservation, that space has already been
69 * partially accounted for in our ancestors.
72 parent_delta(dsl_dataset_t *ds, int64_t delta)
74 uint64_t old_bytes, new_bytes;
76 if (ds->ds_reserved == 0)
79 old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
80 new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
82 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
83 return (new_bytes - old_bytes);
87 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
89 int used, compressed, uncompressed;
92 used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
93 compressed = BP_GET_PSIZE(bp);
94 uncompressed = BP_GET_UCSIZE(bp);
96 dprintf_bp(bp, "ds=%p", ds);
98 ASSERT(dmu_tx_is_syncing(tx));
99 /* It could have been compressed away to nothing */
102 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
103 ASSERT3U(BP_GET_TYPE(bp), <, DMU_OT_NUMTYPES);
106 * Account for the meta-objset space in its placeholder
109 ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */
110 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
111 used, compressed, uncompressed, tx);
112 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
115 dmu_buf_will_dirty(ds->ds_dbuf, tx);
117 mutex_enter(&ds->ds_dir->dd_lock);
118 mutex_enter(&ds->ds_lock);
119 delta = parent_delta(ds, used);
120 ds->ds_phys->ds_used_bytes += used;
121 ds->ds_phys->ds_compressed_bytes += compressed;
122 ds->ds_phys->ds_uncompressed_bytes += uncompressed;
123 ds->ds_phys->ds_unique_bytes += used;
124 mutex_exit(&ds->ds_lock);
125 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
126 compressed, uncompressed, tx);
127 dsl_dir_transfer_space(ds->ds_dir, used - delta,
128 DD_USED_REFRSRV, DD_USED_HEAD, tx);
129 mutex_exit(&ds->ds_dir->dd_lock);
133 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
136 int used, compressed, uncompressed;
141 ASSERT(dmu_tx_is_syncing(tx));
142 ASSERT(bp->blk_birth <= tx->tx_txg);
144 used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
145 compressed = BP_GET_PSIZE(bp);
146 uncompressed = BP_GET_UCSIZE(bp);
151 * Account for the meta-objset space in its placeholder
154 dsl_free(tx->tx_pool, tx->tx_txg, bp);
156 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
157 -used, -compressed, -uncompressed, tx);
158 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
161 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
163 ASSERT(!dsl_dataset_is_snapshot(ds));
164 dmu_buf_will_dirty(ds->ds_dbuf, tx);
166 if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
169 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
170 dsl_free(tx->tx_pool, tx->tx_txg, bp);
172 mutex_enter(&ds->ds_dir->dd_lock);
173 mutex_enter(&ds->ds_lock);
174 ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
175 !DS_UNIQUE_IS_ACCURATE(ds));
176 delta = parent_delta(ds, -used);
177 ds->ds_phys->ds_unique_bytes -= used;
178 mutex_exit(&ds->ds_lock);
179 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
180 delta, -compressed, -uncompressed, tx);
181 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
182 DD_USED_REFRSRV, DD_USED_HEAD, tx);
183 mutex_exit(&ds->ds_dir->dd_lock);
185 dprintf_bp(bp, "putting on dead list: %s", "");
188 * We are here as part of zio's write done callback,
189 * which means we're a zio interrupt thread. We can't
190 * call dsl_deadlist_insert() now because it may block
191 * waiting for I/O. Instead, put bp on the deferred
192 * queue and let dsl_pool_sync() finish the job.
194 bplist_append(&ds->ds_pending_deadlist, bp);
196 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
198 ASSERT3U(ds->ds_prev->ds_object, ==,
199 ds->ds_phys->ds_prev_snap_obj);
200 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
201 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
202 if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
203 ds->ds_object && bp->blk_birth >
204 ds->ds_prev->ds_phys->ds_prev_snap_txg) {
205 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
206 mutex_enter(&ds->ds_prev->ds_lock);
207 ds->ds_prev->ds_phys->ds_unique_bytes += used;
208 mutex_exit(&ds->ds_prev->ds_lock);
210 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
211 dsl_dir_transfer_space(ds->ds_dir, used,
212 DD_USED_HEAD, DD_USED_SNAP, tx);
215 mutex_enter(&ds->ds_lock);
216 ASSERT3U(ds->ds_phys->ds_used_bytes, >=, used);
217 ds->ds_phys->ds_used_bytes -= used;
218 ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
219 ds->ds_phys->ds_compressed_bytes -= compressed;
220 ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
221 ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
222 mutex_exit(&ds->ds_lock);
228 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
230 uint64_t trysnap = 0;
235 * The snapshot creation could fail, but that would cause an
236 * incorrect FALSE return, which would only result in an
237 * overestimation of the amount of space that an operation would
238 * consume, which is OK.
240 * There's also a small window where we could miss a pending
241 * snapshot, because we could set the sync task in the quiescing
242 * phase. So this should only be used as a guess.
244 if (ds->ds_trysnap_txg >
245 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
246 trysnap = ds->ds_trysnap_txg;
247 return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
251 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
254 if (blk_birth <= dsl_dataset_prev_snap_txg(ds))
257 ddt_prefetch(dsl_dataset_get_spa(ds), bp);
264 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
266 dsl_dataset_t *ds = dsv;
268 ASSERT(ds->ds_owner == NULL || DSL_DATASET_IS_DESTROYED(ds));
270 unique_remove(ds->ds_fsid_guid);
272 if (ds->ds_objset != NULL)
273 dmu_objset_evict(ds->ds_objset);
276 dsl_dataset_drop_ref(ds->ds_prev, ds);
280 bplist_destroy(&ds->ds_pending_deadlist);
282 dsl_deadlist_close(&ds->ds_deadlist);
284 ASSERT(ds->ds_deadlist.dl_dbuf == NULL);
285 ASSERT(!ds->ds_deadlist.dl_oldfmt);
288 dsl_dir_close(ds->ds_dir, ds);
290 ASSERT(!list_link_active(&ds->ds_synced_link));
292 mutex_destroy(&ds->ds_lock);
293 mutex_destroy(&ds->ds_recvlock);
294 mutex_destroy(&ds->ds_opening_lock);
295 rw_destroy(&ds->ds_rwlock);
296 cv_destroy(&ds->ds_exclusive_cv);
298 kmem_free(ds, sizeof (dsl_dataset_t));
302 dsl_dataset_get_snapname(dsl_dataset_t *ds)
304 dsl_dataset_phys_t *headphys;
307 dsl_pool_t *dp = ds->ds_dir->dd_pool;
308 objset_t *mos = dp->dp_meta_objset;
310 if (ds->ds_snapname[0])
312 if (ds->ds_phys->ds_next_snap_obj == 0)
315 err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
319 headphys = headdbuf->db_data;
320 err = zap_value_search(dp->dp_meta_objset,
321 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
322 dmu_buf_rele(headdbuf, FTAG);
327 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
329 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
330 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
334 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
339 err = zap_lookup_norm(mos, snapobj, name, 8, 1,
340 value, mt, NULL, 0, NULL);
341 if (err == ENOTSUP && mt == MT_FIRST)
342 err = zap_lookup(mos, snapobj, name, 8, 1, value);
347 dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx)
349 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
350 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
354 dsl_dir_snap_cmtime_update(ds->ds_dir);
356 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
361 err = zap_remove_norm(mos, snapobj, name, mt, tx);
362 if (err == ENOTSUP && mt == MT_FIRST)
363 err = zap_remove(mos, snapobj, name, tx);
368 dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag,
371 objset_t *mos = dp->dp_meta_objset;
375 dmu_object_info_t doi;
377 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
378 dsl_pool_sync_context(dp));
380 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
384 /* Make sure dsobj has the correct object type. */
385 dmu_object_info_from_db(dbuf, &doi);
386 if (doi.doi_type != DMU_OT_DSL_DATASET)
389 ds = dmu_buf_get_user(dbuf);
391 dsl_dataset_t *winner = NULL;
393 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
395 ds->ds_object = dsobj;
396 ds->ds_phys = dbuf->db_data;
397 list_link_init(&ds->ds_synced_link);
399 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
400 mutex_init(&ds->ds_recvlock, NULL, MUTEX_DEFAULT, NULL);
401 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
402 rw_init(&ds->ds_rwlock, NULL, RW_DEFAULT, NULL);
403 cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL);
405 bplist_create(&ds->ds_pending_deadlist);
406 dsl_deadlist_open(&ds->ds_deadlist,
407 mos, ds->ds_phys->ds_deadlist_obj);
410 err = dsl_dir_open_obj(dp,
411 ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
414 mutex_destroy(&ds->ds_lock);
415 mutex_destroy(&ds->ds_recvlock);
416 mutex_destroy(&ds->ds_opening_lock);
417 rw_destroy(&ds->ds_rwlock);
418 cv_destroy(&ds->ds_exclusive_cv);
419 bplist_destroy(&ds->ds_pending_deadlist);
420 dsl_deadlist_close(&ds->ds_deadlist);
421 kmem_free(ds, sizeof (dsl_dataset_t));
422 dmu_buf_rele(dbuf, tag);
426 if (!dsl_dataset_is_snapshot(ds)) {
427 ds->ds_snapname[0] = '\0';
428 if (ds->ds_phys->ds_prev_snap_obj) {
429 err = dsl_dataset_get_ref(dp,
430 ds->ds_phys->ds_prev_snap_obj,
434 if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
435 err = dsl_dataset_get_snapname(ds);
436 if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) {
438 ds->ds_dir->dd_pool->dp_meta_objset,
439 ds->ds_phys->ds_userrefs_obj,
444 if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
446 * In sync context, we're called with either no lock
447 * or with the write lock. If we're not syncing,
448 * we're always called with the read lock held.
450 boolean_t need_lock =
451 !RW_WRITE_HELD(&dp->dp_config_rwlock) &&
452 dsl_pool_sync_context(dp);
455 rw_enter(&dp->dp_config_rwlock, RW_READER);
457 err = dsl_prop_get_ds(ds,
458 "refreservation", sizeof (uint64_t), 1,
459 &ds->ds_reserved, NULL);
461 err = dsl_prop_get_ds(ds,
462 "refquota", sizeof (uint64_t), 1,
463 &ds->ds_quota, NULL);
467 rw_exit(&dp->dp_config_rwlock);
469 ds->ds_reserved = ds->ds_quota = 0;
473 winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys,
477 bplist_destroy(&ds->ds_pending_deadlist);
478 dsl_deadlist_close(&ds->ds_deadlist);
480 dsl_dataset_drop_ref(ds->ds_prev, ds);
481 dsl_dir_close(ds->ds_dir, ds);
482 mutex_destroy(&ds->ds_lock);
483 mutex_destroy(&ds->ds_recvlock);
484 mutex_destroy(&ds->ds_opening_lock);
485 rw_destroy(&ds->ds_rwlock);
486 cv_destroy(&ds->ds_exclusive_cv);
487 kmem_free(ds, sizeof (dsl_dataset_t));
489 dmu_buf_rele(dbuf, tag);
495 unique_insert(ds->ds_phys->ds_fsid_guid);
498 ASSERT3P(ds->ds_dbuf, ==, dbuf);
499 ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
500 ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
501 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
502 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
503 mutex_enter(&ds->ds_lock);
504 if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) {
505 mutex_exit(&ds->ds_lock);
506 dmu_buf_rele(ds->ds_dbuf, tag);
509 mutex_exit(&ds->ds_lock);
515 dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag)
517 dsl_pool_t *dp = ds->ds_dir->dd_pool;
520 * In syncing context we don't want the rwlock lock: there
521 * may be an existing writer waiting for sync phase to
522 * finish. We don't need to worry about such writers, since
523 * sync phase is single-threaded, so the writer can't be
524 * doing anything while we are active.
526 if (dsl_pool_sync_context(dp)) {
527 ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
532 * Normal users will hold the ds_rwlock as a READER until they
533 * are finished (i.e., call dsl_dataset_rele()). "Owners" will
534 * drop their READER lock after they set the ds_owner field.
536 * If the dataset is being destroyed, the destroy thread will
537 * obtain a WRITER lock for exclusive access after it's done its
538 * open-context work and then change the ds_owner to
539 * dsl_reaper once destruction is assured. So threads
540 * may block here temporarily, until the "destructability" of
541 * the dataset is determined.
543 ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock));
544 mutex_enter(&ds->ds_lock);
545 while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) {
546 rw_exit(&dp->dp_config_rwlock);
547 cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock);
548 if (DSL_DATASET_IS_DESTROYED(ds)) {
549 mutex_exit(&ds->ds_lock);
550 dsl_dataset_drop_ref(ds, tag);
551 rw_enter(&dp->dp_config_rwlock, RW_READER);
555 * The dp_config_rwlock lives above the ds_lock. And
556 * we need to check DSL_DATASET_IS_DESTROYED() while
557 * holding the ds_lock, so we have to drop and reacquire
560 mutex_exit(&ds->ds_lock);
561 rw_enter(&dp->dp_config_rwlock, RW_READER);
562 mutex_enter(&ds->ds_lock);
564 mutex_exit(&ds->ds_lock);
569 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
572 int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp);
576 return (dsl_dataset_hold_ref(*dsp, tag));
580 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, boolean_t inconsistentok,
581 void *tag, dsl_dataset_t **dsp)
583 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
586 if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
587 dsl_dataset_rele(*dsp, tag);
595 dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp)
599 const char *snapname;
603 err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname);
608 obj = dd->dd_phys->dd_head_dataset_obj;
609 rw_enter(&dp->dp_config_rwlock, RW_READER);
611 err = dsl_dataset_get_ref(dp, obj, tag, dsp);
617 err = dsl_dataset_hold_ref(*dsp, tag);
619 /* we may be looking for a snapshot */
620 if (err == 0 && snapname != NULL) {
621 dsl_dataset_t *ds = NULL;
623 if (*snapname++ != '@') {
624 dsl_dataset_rele(*dsp, tag);
629 dprintf("looking for snapshot '%s'\n", snapname);
630 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
632 err = dsl_dataset_get_ref(dp, obj, tag, &ds);
633 dsl_dataset_rele(*dsp, tag);
635 ASSERT3U((err == 0), ==, (ds != NULL));
638 mutex_enter(&ds->ds_lock);
639 if (ds->ds_snapname[0] == 0)
640 (void) strlcpy(ds->ds_snapname, snapname,
641 sizeof (ds->ds_snapname));
642 mutex_exit(&ds->ds_lock);
643 err = dsl_dataset_hold_ref(ds, tag);
644 *dsp = err ? NULL : ds;
648 rw_exit(&dp->dp_config_rwlock);
649 dsl_dir_close(dd, FTAG);
654 dsl_dataset_own(const char *name, boolean_t inconsistentok,
655 void *tag, dsl_dataset_t **dsp)
657 int err = dsl_dataset_hold(name, tag, dsp);
660 if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
661 dsl_dataset_rele(*dsp, tag);
668 dsl_dataset_name(dsl_dataset_t *ds, char *name)
671 (void) strcpy(name, "mos");
673 dsl_dir_name(ds->ds_dir, name);
674 VERIFY(0 == dsl_dataset_get_snapname(ds));
675 if (ds->ds_snapname[0]) {
676 (void) strcat(name, "@");
678 * We use a "recursive" mutex so that we
679 * can call dprintf_ds() with ds_lock held.
681 if (!MUTEX_HELD(&ds->ds_lock)) {
682 mutex_enter(&ds->ds_lock);
683 (void) strcat(name, ds->ds_snapname);
684 mutex_exit(&ds->ds_lock);
686 (void) strcat(name, ds->ds_snapname);
693 dsl_dataset_namelen(dsl_dataset_t *ds)
698 result = 3; /* "mos" */
700 result = dsl_dir_namelen(ds->ds_dir);
701 VERIFY(0 == dsl_dataset_get_snapname(ds));
702 if (ds->ds_snapname[0]) {
703 ++result; /* adding one for the @-sign */
704 if (!MUTEX_HELD(&ds->ds_lock)) {
705 mutex_enter(&ds->ds_lock);
706 result += strlen(ds->ds_snapname);
707 mutex_exit(&ds->ds_lock);
709 result += strlen(ds->ds_snapname);
718 dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag)
720 dmu_buf_rele(ds->ds_dbuf, tag);
724 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
726 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) {
727 rw_exit(&ds->ds_rwlock);
729 dsl_dataset_drop_ref(ds, tag);
733 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
735 ASSERT((ds->ds_owner == tag && ds->ds_dbuf) ||
736 (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL));
738 mutex_enter(&ds->ds_lock);
740 if (RW_WRITE_HELD(&ds->ds_rwlock)) {
741 rw_exit(&ds->ds_rwlock);
742 cv_broadcast(&ds->ds_exclusive_cv);
744 mutex_exit(&ds->ds_lock);
746 dsl_dataset_drop_ref(ds, tag);
748 dsl_dataset_evict(NULL, ds);
752 dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *tag)
754 boolean_t gotit = FALSE;
756 mutex_enter(&ds->ds_lock);
757 if (ds->ds_owner == NULL &&
758 (!DS_IS_INCONSISTENT(ds) || inconsistentok)) {
760 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool))
761 rw_exit(&ds->ds_rwlock);
764 mutex_exit(&ds->ds_lock);
769 dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner)
771 ASSERT3P(owner, ==, ds->ds_owner);
772 if (!RW_WRITE_HELD(&ds->ds_rwlock))
773 rw_enter(&ds->ds_rwlock, RW_WRITER);
777 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
778 uint64_t flags, dmu_tx_t *tx)
780 dsl_pool_t *dp = dd->dd_pool;
782 dsl_dataset_phys_t *dsphys;
784 objset_t *mos = dp->dp_meta_objset;
787 origin = dp->dp_origin_snap;
789 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
790 ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
791 ASSERT(dmu_tx_is_syncing(tx));
792 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
794 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
795 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
796 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
797 dmu_buf_will_dirty(dbuf, tx);
798 dsphys = dbuf->db_data;
799 bzero(dsphys, sizeof (dsl_dataset_phys_t));
800 dsphys->ds_dir_obj = dd->dd_object;
801 dsphys->ds_flags = flags;
802 dsphys->ds_fsid_guid = unique_create();
803 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
804 sizeof (dsphys->ds_guid));
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_used_bytes =
820 origin->ds_phys->ds_used_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 dsl_sync_task_group_t *dstg;
918 dsl_snapshot_destroy_one(const char *name, void *arg)
920 struct destroyarg *da = arg;
925 dsname = kmem_asprintf("%s@%s", name, da->snapname);
926 err = dsl_dataset_own(dsname, B_TRUE, da->dstg, &ds);
929 struct dsl_ds_destroyarg *dsda;
931 dsl_dataset_make_exclusive(ds, da->dstg);
932 dsda = kmem_zalloc(sizeof (struct dsl_ds_destroyarg), KM_SLEEP);
934 dsda->defer = da->defer;
935 dsl_sync_task_create(da->dstg, dsl_dataset_destroy_check,
936 dsl_dataset_destroy_sync, dsda, da->dstg, 0);
937 } else if (err == ENOENT) {
940 (void) strcpy(da->failed, name);
946 * Destroy 'snapname' in all descendants of 'fsname'.
948 #pragma weak dmu_snapshots_destroy = dsl_snapshots_destroy
950 dsl_snapshots_destroy(char *fsname, char *snapname, boolean_t defer)
953 struct destroyarg da;
954 dsl_sync_task_t *dst;
957 err = spa_open(fsname, &spa, FTAG);
960 da.dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
961 da.snapname = snapname;
965 err = dmu_objset_find(fsname,
966 dsl_snapshot_destroy_one, &da, DS_FIND_CHILDREN);
969 err = dsl_sync_task_group_wait(da.dstg);
971 for (dst = list_head(&da.dstg->dstg_tasks); dst;
972 dst = list_next(&da.dstg->dstg_tasks, dst)) {
973 struct dsl_ds_destroyarg *dsda = dst->dst_arg1;
974 dsl_dataset_t *ds = dsda->ds;
977 * Return the file system name that triggered the error
980 dsl_dataset_name(ds, fsname);
981 *strchr(fsname, '@') = '\0';
983 ASSERT3P(dsda->rm_origin, ==, NULL);
984 dsl_dataset_disown(ds, da.dstg);
985 kmem_free(dsda, sizeof (struct dsl_ds_destroyarg));
988 dsl_sync_task_group_destroy(da.dstg);
989 spa_close(spa, FTAG);
994 dsl_dataset_might_destroy_origin(dsl_dataset_t *ds)
996 boolean_t might_destroy = B_FALSE;
998 mutex_enter(&ds->ds_lock);
999 if (ds->ds_phys->ds_num_children == 2 && ds->ds_userrefs == 0 &&
1000 DS_IS_DEFER_DESTROY(ds))
1001 might_destroy = B_TRUE;
1002 mutex_exit(&ds->ds_lock);
1004 return (might_destroy);
1008 * If we're removing a clone, and these three conditions are true:
1009 * 1) the clone's origin has no other children
1010 * 2) the clone's origin has no user references
1011 * 3) the clone's origin has been marked for deferred destruction
1012 * Then, prepare to remove the origin as part of this sync task group.
1015 dsl_dataset_origin_rm_prep(struct dsl_ds_destroyarg *dsda, void *tag)
1017 dsl_dataset_t *ds = dsda->ds;
1018 dsl_dataset_t *origin = ds->ds_prev;
1020 if (dsl_dataset_might_destroy_origin(origin)) {
1025 namelen = dsl_dataset_namelen(origin) + 1;
1026 name = kmem_alloc(namelen, KM_SLEEP);
1027 dsl_dataset_name(origin, name);
1029 error = zfs_unmount_snap(name, NULL);
1031 kmem_free(name, namelen);
1035 error = dsl_dataset_own(name, B_TRUE, tag, &origin);
1036 kmem_free(name, namelen);
1039 dsda->rm_origin = origin;
1040 dsl_dataset_make_exclusive(origin, tag);
1047 * ds must be opened as OWNER. On return (whether successful or not),
1048 * ds will be closed and caller can no longer dereference it.
1051 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer)
1054 dsl_sync_task_group_t *dstg;
1058 struct dsl_ds_destroyarg dsda = { 0 };
1059 dsl_dataset_t *dummy_ds;
1063 if (dsl_dataset_is_snapshot(ds)) {
1064 /* Destroying a snapshot is simpler */
1065 dsl_dataset_make_exclusive(ds, tag);
1068 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1069 dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
1071 ASSERT3P(dsda.rm_origin, ==, NULL);
1079 dummy_ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
1080 dummy_ds->ds_dir = dd;
1081 dummy_ds->ds_object = ds->ds_object;
1084 * Check for errors and mark this ds as inconsistent, in
1085 * case we crash while freeing the objects.
1087 err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check,
1088 dsl_dataset_destroy_begin_sync, ds, NULL, 0);
1092 err = dmu_objset_from_ds(ds, &os);
1097 * remove the objects in open context, so that we won't
1098 * have too much to do in syncing context.
1100 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
1101 ds->ds_phys->ds_prev_snap_txg)) {
1103 * Ignore errors, if there is not enough disk space
1104 * we will deal with it in dsl_dataset_destroy_sync().
1106 (void) dmu_free_object(os, obj);
1112 * Only the ZIL knows how to free log blocks.
1114 zil_destroy(dmu_objset_zil(os), B_FALSE);
1117 * Sync out all in-flight IO.
1119 txg_wait_synced(dd->dd_pool, 0);
1122 * If we managed to free all the objects in open
1123 * context, the user space accounting should be zero.
1125 if (ds->ds_phys->ds_bp.blk_fill == 0 &&
1126 dmu_objset_userused_enabled(os)) {
1127 ASSERTV(uint64_t count);
1128 ASSERT(zap_count(os, DMU_USERUSED_OBJECT, &count) != 0 ||
1130 ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT, &count) != 0 ||
1134 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
1135 err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd);
1136 rw_exit(&dd->dd_pool->dp_config_rwlock);
1142 * Blow away the dsl_dir + head dataset.
1144 dsl_dataset_make_exclusive(ds, tag);
1146 * If we're removing a clone, we might also need to remove its
1150 dsda.need_prep = B_FALSE;
1151 if (dsl_dir_is_clone(dd)) {
1152 err = dsl_dataset_origin_rm_prep(&dsda, tag);
1154 dsl_dir_close(dd, FTAG);
1159 dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
1160 dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
1161 dsl_dataset_destroy_sync, &dsda, tag, 0);
1162 dsl_sync_task_create(dstg, dsl_dir_destroy_check,
1163 dsl_dir_destroy_sync, dummy_ds, FTAG, 0);
1164 err = dsl_sync_task_group_wait(dstg);
1165 dsl_sync_task_group_destroy(dstg);
1168 * We could be racing against 'zfs release' or 'zfs destroy -d'
1169 * on the origin snap, in which case we can get EBUSY if we
1170 * needed to destroy the origin snap but were not ready to
1173 if (dsda.need_prep) {
1174 ASSERT(err == EBUSY);
1175 ASSERT(dsl_dir_is_clone(dd));
1176 ASSERT(dsda.rm_origin == NULL);
1178 } while (dsda.need_prep);
1180 if (dsda.rm_origin != NULL)
1181 dsl_dataset_disown(dsda.rm_origin, tag);
1183 /* if it is successful, dsl_dir_destroy_sync will close the dd */
1185 dsl_dir_close(dd, FTAG);
1188 kmem_free(dummy_ds, sizeof (dsl_dataset_t));
1190 dsl_dataset_disown(ds, tag);
1195 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1197 return (&ds->ds_phys->ds_bp);
1201 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1203 ASSERT(dmu_tx_is_syncing(tx));
1204 /* If it's the meta-objset, set dp_meta_rootbp */
1206 tx->tx_pool->dp_meta_rootbp = *bp;
1208 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1209 ds->ds_phys->ds_bp = *bp;
1214 dsl_dataset_get_spa(dsl_dataset_t *ds)
1216 return (ds->ds_dir->dd_pool->dp_spa);
1220 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1224 if (ds == NULL) /* this is the meta-objset */
1227 ASSERT(ds->ds_objset != NULL);
1229 if (ds->ds_phys->ds_next_snap_obj != 0)
1230 panic("dirtying snapshot!");
1232 dp = ds->ds_dir->dd_pool;
1234 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) {
1235 /* up the hold count until we can be written out */
1236 dmu_buf_add_ref(ds->ds_dbuf, ds);
1241 * The unique space in the head dataset can be calculated by subtracting
1242 * the space used in the most recent snapshot, that is still being used
1243 * in this file system, from the space currently in use. To figure out
1244 * the space in the most recent snapshot still in use, we need to take
1245 * the total space used in the snapshot and subtract out the space that
1246 * has been freed up since the snapshot was taken.
1249 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1252 uint64_t dlused, dlcomp, dluncomp;
1254 ASSERT(!dsl_dataset_is_snapshot(ds));
1256 if (ds->ds_phys->ds_prev_snap_obj != 0)
1257 mrs_used = ds->ds_prev->ds_phys->ds_used_bytes;
1261 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
1263 ASSERT3U(dlused, <=, mrs_used);
1264 ds->ds_phys->ds_unique_bytes =
1265 ds->ds_phys->ds_used_bytes - (mrs_used - dlused);
1267 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1268 SPA_VERSION_UNIQUE_ACCURATE)
1269 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1279 kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf,
1280 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1282 struct killarg *ka = arg;
1283 dmu_tx_t *tx = ka->tx;
1288 if (zb->zb_level == ZB_ZIL_LEVEL) {
1289 ASSERT(zilog != NULL);
1291 * It's a block in the intent log. It has no
1292 * accounting, so just free it.
1294 dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp);
1296 ASSERT(zilog == NULL);
1297 ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg);
1298 (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE);
1306 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1308 dsl_dataset_t *ds = arg1;
1309 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1314 * Can't delete a head dataset if there are snapshots of it.
1315 * (Except if the only snapshots are from the branch we cloned
1318 if (ds->ds_prev != NULL &&
1319 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1323 * This is really a dsl_dir thing, but check it here so that
1324 * we'll be less likely to leave this dataset inconsistent &
1327 err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count);
1338 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1340 dsl_dataset_t *ds = arg1;
1341 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1343 /* Mark it as inconsistent on-disk, in case we crash */
1344 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1345 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1347 spa_history_log_internal(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx,
1348 "dataset = %llu", ds->ds_object);
1352 dsl_dataset_origin_check(struct dsl_ds_destroyarg *dsda, void *tag,
1355 dsl_dataset_t *ds = dsda->ds;
1356 dsl_dataset_t *ds_prev = ds->ds_prev;
1358 if (dsl_dataset_might_destroy_origin(ds_prev)) {
1359 struct dsl_ds_destroyarg ndsda = {0};
1362 * If we're not prepared to remove the origin, don't remove
1365 if (dsda->rm_origin == NULL) {
1366 dsda->need_prep = B_TRUE;
1371 ndsda.is_origin_rm = B_TRUE;
1372 return (dsl_dataset_destroy_check(&ndsda, tag, tx));
1376 * If we're not going to remove the origin after all,
1377 * undo the open context setup.
1379 if (dsda->rm_origin != NULL) {
1380 dsl_dataset_disown(dsda->rm_origin, tag);
1381 dsda->rm_origin = NULL;
1388 * If you add new checks here, you may need to add
1389 * additional checks to the "temporary" case in
1390 * snapshot_check() in dmu_objset.c.
1394 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1396 struct dsl_ds_destroyarg *dsda = arg1;
1397 dsl_dataset_t *ds = dsda->ds;
1399 /* we have an owner hold, so noone else can destroy us */
1400 ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
1403 * Only allow deferred destroy on pools that support it.
1404 * NOTE: deferred destroy is only supported on snapshots.
1407 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
1408 SPA_VERSION_USERREFS)
1410 ASSERT(dsl_dataset_is_snapshot(ds));
1415 * Can't delete a head dataset if there are snapshots of it.
1416 * (Except if the only snapshots are from the branch we cloned
1419 if (ds->ds_prev != NULL &&
1420 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1424 * If we made changes this txg, traverse_dsl_dataset won't find
1427 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1430 if (dsl_dataset_is_snapshot(ds)) {
1432 * If this snapshot has an elevated user reference count,
1433 * we can't destroy it yet.
1435 if (ds->ds_userrefs > 0 && !dsda->releasing)
1438 mutex_enter(&ds->ds_lock);
1440 * Can't delete a branch point. However, if we're destroying
1441 * a clone and removing its origin due to it having a user
1442 * hold count of 0 and having been marked for deferred destroy,
1443 * it's OK for the origin to have a single clone.
1445 if (ds->ds_phys->ds_num_children >
1446 (dsda->is_origin_rm ? 2 : 1)) {
1447 mutex_exit(&ds->ds_lock);
1450 mutex_exit(&ds->ds_lock);
1451 } else if (dsl_dir_is_clone(ds->ds_dir)) {
1452 return (dsl_dataset_origin_check(dsda, arg2, tx));
1455 /* XXX we should do some i/o error checking... */
1467 dsl_dataset_refs_gone(dmu_buf_t *db, void *argv)
1469 struct refsarg *arg = argv;
1471 mutex_enter(&arg->lock);
1473 cv_signal(&arg->cv);
1474 mutex_exit(&arg->lock);
1478 dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag)
1482 mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL);
1483 cv_init(&arg.cv, NULL, CV_DEFAULT, NULL);
1485 (void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys,
1486 dsl_dataset_refs_gone);
1487 dmu_buf_rele(ds->ds_dbuf, tag);
1488 mutex_enter(&arg.lock);
1490 cv_wait(&arg.cv, &arg.lock);
1492 mutex_exit(&arg.lock);
1495 mutex_destroy(&arg.lock);
1496 cv_destroy(&arg.cv);
1500 remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, dmu_tx_t *tx)
1502 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1504 ASSERTV(uint64_t count);
1506 ASSERT(ds->ds_phys->ds_num_children >= 2);
1507 err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx);
1509 * The err should not be ENOENT, but a bug in a previous version
1510 * of the code could cause upgrade_clones_cb() to not set
1511 * ds_next_snap_obj when it should, leading to a missing entry.
1512 * If we knew that the pool was created after
1513 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1514 * ENOENT. However, at least we can check that we don't have
1515 * too many entries in the next_clones_obj even after failing to
1518 if (err != ENOENT) {
1519 VERIFY3U(err, ==, 0);
1521 ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj,
1523 ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2);
1527 dsl_dataset_remove_clones_key(dsl_dataset_t *ds, uint64_t mintxg, dmu_tx_t *tx)
1529 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1534 * If it is the old version, dd_clones doesn't exist so we can't
1535 * find the clones, but deadlist_remove_key() is a no-op so it
1538 if (ds->ds_dir->dd_phys->dd_clones == 0)
1541 for (zap_cursor_init(&zc, mos, ds->ds_dir->dd_phys->dd_clones);
1542 zap_cursor_retrieve(&zc, &za) == 0;
1543 zap_cursor_advance(&zc)) {
1544 dsl_dataset_t *clone;
1546 VERIFY3U(0, ==, dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1547 za.za_first_integer, FTAG, &clone));
1548 if (clone->ds_dir->dd_origin_txg > mintxg) {
1549 dsl_deadlist_remove_key(&clone->ds_deadlist,
1551 dsl_dataset_remove_clones_key(clone, mintxg, tx);
1553 dsl_dataset_rele(clone, FTAG);
1555 zap_cursor_fini(&zc);
1558 struct process_old_arg {
1560 dsl_dataset_t *ds_prev;
1561 boolean_t after_branch_point;
1563 uint64_t used, comp, uncomp;
1567 process_old_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
1569 struct process_old_arg *poa = arg;
1570 dsl_pool_t *dp = poa->ds->ds_dir->dd_pool;
1572 if (bp->blk_birth <= poa->ds->ds_phys->ds_prev_snap_txg) {
1573 dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, tx);
1574 if (poa->ds_prev && !poa->after_branch_point &&
1576 poa->ds_prev->ds_phys->ds_prev_snap_txg) {
1577 poa->ds_prev->ds_phys->ds_unique_bytes +=
1578 bp_get_dsize_sync(dp->dp_spa, bp);
1581 poa->used += bp_get_dsize_sync(dp->dp_spa, bp);
1582 poa->comp += BP_GET_PSIZE(bp);
1583 poa->uncomp += BP_GET_UCSIZE(bp);
1584 dsl_free_sync(poa->pio, dp, tx->tx_txg, bp);
1590 process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev,
1591 dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx)
1593 struct process_old_arg poa = { 0 };
1594 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1595 objset_t *mos = dp->dp_meta_objset;
1597 ASSERT(ds->ds_deadlist.dl_oldfmt);
1598 ASSERT(ds_next->ds_deadlist.dl_oldfmt);
1601 poa.ds_prev = ds_prev;
1602 poa.after_branch_point = after_branch_point;
1603 poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1604 VERIFY3U(0, ==, bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj,
1605 process_old_cb, &poa, tx));
1606 VERIFY3U(zio_wait(poa.pio), ==, 0);
1607 ASSERT3U(poa.used, ==, ds->ds_phys->ds_unique_bytes);
1609 /* change snapused */
1610 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1611 -poa.used, -poa.comp, -poa.uncomp, tx);
1613 /* swap next's deadlist to our deadlist */
1614 dsl_deadlist_close(&ds->ds_deadlist);
1615 dsl_deadlist_close(&ds_next->ds_deadlist);
1616 SWITCH64(ds_next->ds_phys->ds_deadlist_obj,
1617 ds->ds_phys->ds_deadlist_obj);
1618 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
1619 dsl_deadlist_open(&ds_next->ds_deadlist, mos,
1620 ds_next->ds_phys->ds_deadlist_obj);
1624 dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
1626 struct dsl_ds_destroyarg *dsda = arg1;
1627 dsl_dataset_t *ds = dsda->ds;
1629 int after_branch_point = FALSE;
1630 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1631 objset_t *mos = dp->dp_meta_objset;
1632 dsl_dataset_t *ds_prev = NULL;
1633 boolean_t wont_destroy;
1636 wont_destroy = (dsda->defer &&
1637 (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1));
1639 ASSERT(ds->ds_owner || wont_destroy);
1640 ASSERT(dsda->defer || ds->ds_phys->ds_num_children <= 1);
1641 ASSERT(ds->ds_prev == NULL ||
1642 ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1643 ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1646 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
1647 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1648 ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY;
1652 /* signal any waiters that this dataset is going away */
1653 mutex_enter(&ds->ds_lock);
1654 ds->ds_owner = dsl_reaper;
1655 cv_broadcast(&ds->ds_exclusive_cv);
1656 mutex_exit(&ds->ds_lock);
1658 /* Remove our reservation */
1659 if (ds->ds_reserved != 0) {
1660 dsl_prop_setarg_t psa;
1663 dsl_prop_setarg_init_uint64(&psa, "refreservation",
1664 (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
1666 psa.psa_effective_value = 0; /* predict default value */
1668 dsl_dataset_set_reservation_sync(ds, &psa, tx);
1669 ASSERT3U(ds->ds_reserved, ==, 0);
1672 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1674 dsl_scan_ds_destroyed(ds, tx);
1676 obj = ds->ds_object;
1678 if (ds->ds_phys->ds_prev_snap_obj != 0) {
1680 ds_prev = ds->ds_prev;
1682 VERIFY(0 == dsl_dataset_hold_obj(dp,
1683 ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev));
1685 after_branch_point =
1686 (ds_prev->ds_phys->ds_next_snap_obj != obj);
1688 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
1689 if (after_branch_point &&
1690 ds_prev->ds_phys->ds_next_clones_obj != 0) {
1691 remove_from_next_clones(ds_prev, obj, tx);
1692 if (ds->ds_phys->ds_next_snap_obj != 0) {
1693 VERIFY(0 == zap_add_int(mos,
1694 ds_prev->ds_phys->ds_next_clones_obj,
1695 ds->ds_phys->ds_next_snap_obj, tx));
1698 if (after_branch_point &&
1699 ds->ds_phys->ds_next_snap_obj == 0) {
1700 /* This clone is toast. */
1701 ASSERT(ds_prev->ds_phys->ds_num_children > 1);
1702 ds_prev->ds_phys->ds_num_children--;
1705 * If the clone's origin has no other clones, no
1706 * user holds, and has been marked for deferred
1707 * deletion, then we should have done the necessary
1708 * destroy setup for it.
1710 if (ds_prev->ds_phys->ds_num_children == 1 &&
1711 ds_prev->ds_userrefs == 0 &&
1712 DS_IS_DEFER_DESTROY(ds_prev)) {
1713 ASSERT3P(dsda->rm_origin, !=, NULL);
1715 ASSERT3P(dsda->rm_origin, ==, NULL);
1717 } else if (!after_branch_point) {
1718 ds_prev->ds_phys->ds_next_snap_obj =
1719 ds->ds_phys->ds_next_snap_obj;
1723 if (dsl_dataset_is_snapshot(ds)) {
1724 dsl_dataset_t *ds_next;
1725 uint64_t old_unique;
1726 uint64_t used = 0, comp = 0, uncomp = 0;
1728 VERIFY(0 == dsl_dataset_hold_obj(dp,
1729 ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next));
1730 ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
1732 old_unique = ds_next->ds_phys->ds_unique_bytes;
1734 dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
1735 ds_next->ds_phys->ds_prev_snap_obj =
1736 ds->ds_phys->ds_prev_snap_obj;
1737 ds_next->ds_phys->ds_prev_snap_txg =
1738 ds->ds_phys->ds_prev_snap_txg;
1739 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1740 ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
1743 if (ds_next->ds_deadlist.dl_oldfmt) {
1744 process_old_deadlist(ds, ds_prev, ds_next,
1745 after_branch_point, tx);
1747 /* Adjust prev's unique space. */
1748 if (ds_prev && !after_branch_point) {
1749 dsl_deadlist_space_range(&ds_next->ds_deadlist,
1750 ds_prev->ds_phys->ds_prev_snap_txg,
1751 ds->ds_phys->ds_prev_snap_txg,
1752 &used, &comp, &uncomp);
1753 ds_prev->ds_phys->ds_unique_bytes += used;
1756 /* Adjust snapused. */
1757 dsl_deadlist_space_range(&ds_next->ds_deadlist,
1758 ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
1759 &used, &comp, &uncomp);
1760 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1761 -used, -comp, -uncomp, tx);
1763 /* Move blocks to be freed to pool's free list. */
1764 dsl_deadlist_move_bpobj(&ds_next->ds_deadlist,
1765 &dp->dp_free_bpobj, ds->ds_phys->ds_prev_snap_txg,
1767 dsl_dir_diduse_space(tx->tx_pool->dp_free_dir,
1768 DD_USED_HEAD, used, comp, uncomp, tx);
1769 dsl_dir_dirty(tx->tx_pool->dp_free_dir, tx);
1771 /* Merge our deadlist into next's and free it. */
1772 dsl_deadlist_merge(&ds_next->ds_deadlist,
1773 ds->ds_phys->ds_deadlist_obj, tx);
1775 dsl_deadlist_close(&ds->ds_deadlist);
1776 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
1778 /* Collapse range in clone heads */
1779 dsl_dataset_remove_clones_key(ds,
1780 ds->ds_phys->ds_creation_txg, tx);
1782 if (dsl_dataset_is_snapshot(ds_next)) {
1783 dsl_dataset_t *ds_nextnext;
1787 * Update next's unique to include blocks which
1788 * were previously shared by only this snapshot
1789 * and it. Those blocks will be born after the
1790 * prev snap and before this snap, and will have
1791 * died after the next snap and before the one
1792 * after that (ie. be on the snap after next's
1795 VERIFY(0 == dsl_dataset_hold_obj(dp,
1796 ds_next->ds_phys->ds_next_snap_obj,
1797 FTAG, &ds_nextnext));
1798 dsl_deadlist_space_range(&ds_nextnext->ds_deadlist,
1799 ds->ds_phys->ds_prev_snap_txg,
1800 ds->ds_phys->ds_creation_txg,
1801 &used, &comp, &uncomp);
1802 ds_next->ds_phys->ds_unique_bytes += used;
1803 dsl_dataset_rele(ds_nextnext, FTAG);
1804 ASSERT3P(ds_next->ds_prev, ==, NULL);
1806 /* Collapse range in this head. */
1807 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
1808 ds->ds_dir->dd_phys->dd_head_dataset_obj,
1810 dsl_deadlist_remove_key(&hds->ds_deadlist,
1811 ds->ds_phys->ds_creation_txg, tx);
1812 dsl_dataset_rele(hds, FTAG);
1815 ASSERT3P(ds_next->ds_prev, ==, ds);
1816 dsl_dataset_drop_ref(ds_next->ds_prev, ds_next);
1817 ds_next->ds_prev = NULL;
1819 VERIFY(0 == dsl_dataset_get_ref(dp,
1820 ds->ds_phys->ds_prev_snap_obj,
1821 ds_next, &ds_next->ds_prev));
1824 dsl_dataset_recalc_head_uniq(ds_next);
1827 * Reduce the amount of our unconsmed refreservation
1828 * being charged to our parent by the amount of
1829 * new unique data we have gained.
1831 if (old_unique < ds_next->ds_reserved) {
1833 uint64_t new_unique =
1834 ds_next->ds_phys->ds_unique_bytes;
1836 ASSERT(old_unique <= new_unique);
1837 mrsdelta = MIN(new_unique - old_unique,
1838 ds_next->ds_reserved - old_unique);
1839 dsl_dir_diduse_space(ds->ds_dir,
1840 DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
1843 dsl_dataset_rele(ds_next, FTAG);
1846 * There's no next snapshot, so this is a head dataset.
1847 * Destroy the deadlist. Unless it's a clone, the
1848 * deadlist should be empty. (If it's a clone, it's
1849 * safe to ignore the deadlist contents.)
1853 dsl_deadlist_close(&ds->ds_deadlist);
1854 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
1855 ds->ds_phys->ds_deadlist_obj = 0;
1858 * Free everything that we point to (that's born after
1859 * the previous snapshot, if we are a clone)
1861 * NB: this should be very quick, because we already
1862 * freed all the objects in open context.
1866 err = traverse_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1867 TRAVERSE_POST, kill_blkptr, &ka);
1868 ASSERT3U(err, ==, 0);
1869 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
1870 ds->ds_phys->ds_unique_bytes == 0);
1872 if (ds->ds_prev != NULL) {
1873 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
1874 VERIFY3U(0, ==, zap_remove_int(mos,
1875 ds->ds_prev->ds_dir->dd_phys->dd_clones,
1876 ds->ds_object, tx));
1878 dsl_dataset_rele(ds->ds_prev, ds);
1879 ds->ds_prev = ds_prev = NULL;
1884 * This must be done after the dsl_traverse(), because it will
1885 * re-open the objset.
1887 if (ds->ds_objset) {
1888 dmu_objset_evict(ds->ds_objset);
1889 ds->ds_objset = NULL;
1892 if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) {
1893 /* Erase the link in the dir */
1894 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1895 ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
1896 ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0);
1897 err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx);
1900 /* remove from snapshot namespace */
1901 dsl_dataset_t *ds_head;
1902 ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0);
1903 VERIFY(0 == dsl_dataset_hold_obj(dp,
1904 ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head));
1905 VERIFY(0 == dsl_dataset_get_snapname(ds));
1910 err = dsl_dataset_snap_lookup(ds_head,
1911 ds->ds_snapname, &val);
1912 ASSERT3U(err, ==, 0);
1913 ASSERT3U(val, ==, obj);
1916 err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx);
1918 dsl_dataset_rele(ds_head, FTAG);
1921 if (ds_prev && ds->ds_prev != ds_prev)
1922 dsl_dataset_rele(ds_prev, FTAG);
1924 spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
1925 spa_history_log_internal(LOG_DS_DESTROY, dp->dp_spa, tx,
1926 "dataset = %llu", ds->ds_object);
1928 if (ds->ds_phys->ds_next_clones_obj != 0) {
1929 ASSERTV(uint64_t count);
1930 ASSERT(0 == zap_count(mos,
1931 ds->ds_phys->ds_next_clones_obj, &count) && count == 0);
1932 VERIFY(0 == dmu_object_free(mos,
1933 ds->ds_phys->ds_next_clones_obj, tx));
1935 if (ds->ds_phys->ds_props_obj != 0)
1936 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx));
1937 if (ds->ds_phys->ds_userrefs_obj != 0)
1938 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx));
1939 dsl_dir_close(ds->ds_dir, ds);
1941 dsl_dataset_drain_refs(ds, tag);
1942 VERIFY(0 == dmu_object_free(mos, obj, tx));
1944 if (dsda->rm_origin) {
1946 * Remove the origin of the clone we just destroyed.
1948 struct dsl_ds_destroyarg ndsda = {0};
1950 ndsda.ds = dsda->rm_origin;
1951 dsl_dataset_destroy_sync(&ndsda, tag, tx);
1956 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1960 if (!dmu_tx_is_syncing(tx))
1964 * If there's an fs-only reservation, any blocks that might become
1965 * owned by the snapshot dataset must be accommodated by space
1966 * outside of the reservation.
1968 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1969 asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
1970 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1974 * Propogate any reserved space for this snapshot to other
1975 * snapshot checks in this sync group.
1978 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1984 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
1986 dsl_dataset_t *ds = arg1;
1987 const char *snapname = arg2;
1992 * We don't allow multiple snapshots of the same txg. If there
1993 * is already one, try again.
1995 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
1999 * Check for conflicting name snapshot name.
2001 err = dsl_dataset_snap_lookup(ds, snapname, &value);
2008 * Check that the dataset's name is not too long. Name consists
2009 * of the dataset's length + 1 for the @-sign + snapshot name's length
2011 if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
2012 return (ENAMETOOLONG);
2014 err = dsl_dataset_snapshot_reserve_space(ds, tx);
2018 ds->ds_trysnap_txg = tx->tx_txg;
2023 dsl_dataset_snapshot_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2025 dsl_dataset_t *ds = arg1;
2026 const char *snapname = arg2;
2027 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2029 dsl_dataset_phys_t *dsphys;
2030 uint64_t dsobj, crtxg;
2031 objset_t *mos = dp->dp_meta_objset;
2034 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
2037 * The origin's ds_creation_txg has to be < TXG_INITIAL
2039 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
2044 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
2045 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
2046 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
2047 dmu_buf_will_dirty(dbuf, tx);
2048 dsphys = dbuf->db_data;
2049 bzero(dsphys, sizeof (dsl_dataset_phys_t));
2050 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
2051 dsphys->ds_fsid_guid = unique_create();
2052 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
2053 sizeof (dsphys->ds_guid));
2054 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
2055 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
2056 dsphys->ds_next_snap_obj = ds->ds_object;
2057 dsphys->ds_num_children = 1;
2058 dsphys->ds_creation_time = gethrestime_sec();
2059 dsphys->ds_creation_txg = crtxg;
2060 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
2061 dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes;
2062 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
2063 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
2064 dsphys->ds_flags = ds->ds_phys->ds_flags;
2065 dsphys->ds_bp = ds->ds_phys->ds_bp;
2066 dmu_buf_rele(dbuf, FTAG);
2068 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
2070 uint64_t next_clones_obj =
2071 ds->ds_prev->ds_phys->ds_next_clones_obj;
2072 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
2074 ds->ds_prev->ds_phys->ds_num_children > 1);
2075 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
2076 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
2077 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
2078 ds->ds_prev->ds_phys->ds_creation_txg);
2079 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
2080 } else if (next_clones_obj != 0) {
2081 remove_from_next_clones(ds->ds_prev,
2082 dsphys->ds_next_snap_obj, tx);
2083 VERIFY3U(0, ==, zap_add_int(mos,
2084 next_clones_obj, dsobj, tx));
2089 * If we have a reference-reservation on this dataset, we will
2090 * need to increase the amount of refreservation being charged
2091 * since our unique space is going to zero.
2093 if (ds->ds_reserved) {
2095 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
2096 delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
2097 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
2101 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2102 zfs_dbgmsg("taking snapshot %s@%s/%llu; newkey=%llu",
2103 ds->ds_dir->dd_myname, snapname, dsobj,
2104 ds->ds_phys->ds_prev_snap_txg);
2105 ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist,
2106 UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx);
2107 dsl_deadlist_close(&ds->ds_deadlist);
2108 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
2109 dsl_deadlist_add_key(&ds->ds_deadlist,
2110 ds->ds_phys->ds_prev_snap_txg, tx);
2112 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
2113 ds->ds_phys->ds_prev_snap_obj = dsobj;
2114 ds->ds_phys->ds_prev_snap_txg = crtxg;
2115 ds->ds_phys->ds_unique_bytes = 0;
2116 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
2117 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
2119 err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
2120 snapname, 8, 1, &dsobj, tx);
2124 dsl_dataset_drop_ref(ds->ds_prev, ds);
2125 VERIFY(0 == dsl_dataset_get_ref(dp,
2126 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
2128 dsl_scan_ds_snapshotted(ds, tx);
2130 dsl_dir_snap_cmtime_update(ds->ds_dir);
2132 spa_history_log_internal(LOG_DS_SNAPSHOT, dp->dp_spa, tx,
2133 "dataset = %llu", dsobj);
2137 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
2139 ASSERT(dmu_tx_is_syncing(tx));
2140 ASSERT(ds->ds_objset != NULL);
2141 ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
2144 * in case we had to change ds_fsid_guid when we opened it,
2147 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2148 ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
2150 dsl_dir_dirty(ds->ds_dir, tx);
2151 dmu_objset_sync(ds->ds_objset, zio, tx);
2155 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
2157 uint64_t refd, avail, uobjs, aobjs, ratio;
2159 dsl_dir_stats(ds->ds_dir, nv);
2161 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
2162 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
2163 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
2165 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
2166 ds->ds_phys->ds_creation_time);
2167 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
2168 ds->ds_phys->ds_creation_txg);
2169 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
2171 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
2173 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
2174 ds->ds_phys->ds_guid);
2175 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
2176 ds->ds_phys->ds_unique_bytes);
2177 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
2179 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
2181 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
2182 DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
2184 ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
2185 (ds->ds_phys->ds_uncompressed_bytes * 100 /
2186 ds->ds_phys->ds_compressed_bytes);
2187 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
2189 if (ds->ds_phys->ds_next_snap_obj) {
2191 * This is a snapshot; override the dd's space used with
2192 * our unique space and compression ratio.
2194 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
2195 ds->ds_phys->ds_unique_bytes);
2196 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
2201 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
2203 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
2204 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
2205 stat->dds_guid = ds->ds_phys->ds_guid;
2206 if (ds->ds_phys->ds_next_snap_obj) {
2207 stat->dds_is_snapshot = B_TRUE;
2208 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
2210 stat->dds_is_snapshot = B_FALSE;
2211 stat->dds_num_clones = 0;
2214 /* clone origin is really a dsl_dir thing... */
2215 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
2216 if (dsl_dir_is_clone(ds->ds_dir)) {
2219 VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool,
2220 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
2221 dsl_dataset_name(ods, stat->dds_origin);
2222 dsl_dataset_drop_ref(ods, FTAG);
2224 stat->dds_origin[0] = '\0';
2226 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2230 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2232 return (ds->ds_fsid_guid);
2236 dsl_dataset_space(dsl_dataset_t *ds,
2237 uint64_t *refdbytesp, uint64_t *availbytesp,
2238 uint64_t *usedobjsp, uint64_t *availobjsp)
2240 *refdbytesp = ds->ds_phys->ds_used_bytes;
2241 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2242 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
2243 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
2244 if (ds->ds_quota != 0) {
2246 * Adjust available bytes according to refquota
2248 if (*refdbytesp < ds->ds_quota)
2249 *availbytesp = MIN(*availbytesp,
2250 ds->ds_quota - *refdbytesp);
2254 *usedobjsp = ds->ds_phys->ds_bp.blk_fill;
2255 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
2259 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
2261 ASSERTV(dsl_pool_t *dp = ds->ds_dir->dd_pool);
2263 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
2264 dsl_pool_sync_context(dp));
2265 if (ds->ds_prev == NULL)
2267 if (ds->ds_phys->ds_bp.blk_birth >
2268 ds->ds_prev->ds_phys->ds_creation_txg) {
2269 objset_t *os, *os_prev;
2271 * It may be that only the ZIL differs, because it was
2272 * reset in the head. Don't count that as being
2275 if (dmu_objset_from_ds(ds, &os) != 0)
2277 if (dmu_objset_from_ds(ds->ds_prev, &os_prev) != 0)
2279 return (bcmp(&os->os_phys->os_meta_dnode,
2280 &os_prev->os_phys->os_meta_dnode,
2281 sizeof (os->os_phys->os_meta_dnode)) != 0);
2288 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
2290 dsl_dataset_t *ds = arg1;
2291 char *newsnapname = arg2;
2292 dsl_dir_t *dd = ds->ds_dir;
2297 err = dsl_dataset_hold_obj(dd->dd_pool,
2298 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds);
2302 /* new name better not be in use */
2303 err = dsl_dataset_snap_lookup(hds, newsnapname, &val);
2304 dsl_dataset_rele(hds, FTAG);
2308 else if (err == ENOENT)
2311 /* dataset name + 1 for the "@" + the new snapshot name must fit */
2312 if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
2319 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2321 dsl_dataset_t *ds = arg1;
2322 const char *newsnapname = arg2;
2323 dsl_dir_t *dd = ds->ds_dir;
2324 objset_t *mos = dd->dd_pool->dp_meta_objset;
2328 ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
2330 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
2331 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds));
2333 VERIFY(0 == dsl_dataset_get_snapname(ds));
2334 err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx);
2335 ASSERT3U(err, ==, 0);
2336 mutex_enter(&ds->ds_lock);
2337 (void) strcpy(ds->ds_snapname, newsnapname);
2338 mutex_exit(&ds->ds_lock);
2339 err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
2340 ds->ds_snapname, 8, 1, &ds->ds_object, tx);
2341 ASSERT3U(err, ==, 0);
2343 spa_history_log_internal(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx,
2344 "dataset = %llu", ds->ds_object);
2345 dsl_dataset_rele(hds, FTAG);
2348 struct renamesnaparg {
2349 dsl_sync_task_group_t *dstg;
2350 char failed[MAXPATHLEN];
2356 dsl_snapshot_rename_one(const char *name, void *arg)
2358 struct renamesnaparg *ra = arg;
2359 dsl_dataset_t *ds = NULL;
2363 snapname = kmem_asprintf("%s@%s", name, ra->oldsnap);
2364 (void) strlcpy(ra->failed, snapname, sizeof (ra->failed));
2367 * For recursive snapshot renames the parent won't be changing
2368 * so we just pass name for both the to/from argument.
2370 err = zfs_secpolicy_rename_perms(snapname, snapname, CRED());
2373 return (err == ENOENT ? 0 : err);
2378 * For all filesystems undergoing rename, we'll need to unmount it.
2380 (void) zfs_unmount_snap(snapname, NULL);
2382 err = dsl_dataset_hold(snapname, ra->dstg, &ds);
2385 return (err == ENOENT ? 0 : err);
2387 dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
2388 dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
2394 dsl_recursive_rename(char *oldname, const char *newname)
2397 struct renamesnaparg *ra;
2398 dsl_sync_task_t *dst;
2400 char *cp, *fsname = spa_strdup(oldname);
2401 int len = strlen(oldname) + 1;
2403 /* truncate the snapshot name to get the fsname */
2404 cp = strchr(fsname, '@');
2407 err = spa_open(fsname, &spa, FTAG);
2409 kmem_free(fsname, len);
2412 ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP);
2413 ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
2415 ra->oldsnap = strchr(oldname, '@') + 1;
2416 ra->newsnap = strchr(newname, '@') + 1;
2419 err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
2421 kmem_free(fsname, len);
2424 err = dsl_sync_task_group_wait(ra->dstg);
2427 for (dst = list_head(&ra->dstg->dstg_tasks); dst;
2428 dst = list_next(&ra->dstg->dstg_tasks, dst)) {
2429 dsl_dataset_t *ds = dst->dst_arg1;
2431 dsl_dir_name(ds->ds_dir, ra->failed);
2432 (void) strlcat(ra->failed, "@", sizeof (ra->failed));
2433 (void) strlcat(ra->failed, ra->newsnap,
2434 sizeof (ra->failed));
2436 dsl_dataset_rele(ds, ra->dstg);
2440 (void) strlcpy(oldname, ra->failed, sizeof (ra->failed));
2442 dsl_sync_task_group_destroy(ra->dstg);
2443 kmem_free(ra, sizeof (struct renamesnaparg));
2444 spa_close(spa, FTAG);
2449 dsl_valid_rename(const char *oldname, void *arg)
2451 int delta = *(int *)arg;
2453 if (strlen(oldname) + delta >= MAXNAMELEN)
2454 return (ENAMETOOLONG);
2459 #pragma weak dmu_objset_rename = dsl_dataset_rename
2461 dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive)
2468 err = dsl_dir_open(oldname, FTAG, &dd, &tail);
2473 int delta = strlen(newname) - strlen(oldname);
2475 /* if we're growing, validate child name lengths */
2477 err = dmu_objset_find(oldname, dsl_valid_rename,
2478 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2481 err = dsl_dir_rename(dd, newname);
2482 dsl_dir_close(dd, FTAG);
2486 if (tail[0] != '@') {
2487 /* the name ended in a nonexistent component */
2488 dsl_dir_close(dd, FTAG);
2492 dsl_dir_close(dd, FTAG);
2494 /* new name must be snapshot in same filesystem */
2495 tail = strchr(newname, '@');
2499 if (strncmp(oldname, newname, tail - newname) != 0)
2503 err = dsl_recursive_rename(oldname, newname);
2505 err = dsl_dataset_hold(oldname, FTAG, &ds);
2509 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2510 dsl_dataset_snapshot_rename_check,
2511 dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
2513 dsl_dataset_rele(ds, FTAG);
2519 struct promotenode {
2525 list_t shared_snaps, origin_snaps, clone_snaps;
2526 dsl_dataset_t *origin_origin;
2527 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2531 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2534 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
2536 dsl_dataset_t *hds = arg1;
2537 struct promotearg *pa = arg2;
2538 struct promotenode *snap = list_head(&pa->shared_snaps);
2539 dsl_dataset_t *origin_ds = snap->ds;
2543 /* Check that it is a real clone */
2544 if (!dsl_dir_is_clone(hds->ds_dir))
2547 /* Since this is so expensive, don't do the preliminary check */
2548 if (!dmu_tx_is_syncing(tx))
2551 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)
2554 /* compute origin's new unique space */
2555 snap = list_tail(&pa->clone_snaps);
2556 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2557 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2558 origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2559 &pa->unique, &unused, &unused);
2562 * Walk the snapshots that we are moving
2564 * Compute space to transfer. Consider the incremental changes
2565 * to used for each snapshot:
2566 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2567 * So each snapshot gave birth to:
2568 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2569 * So a sequence would look like:
2570 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2571 * Which simplifies to:
2572 * uN + kN + kN-1 + ... + k1 + k0
2573 * Note however, if we stop before we reach the ORIGIN we get:
2574 * uN + kN + kN-1 + ... + kM - uM-1
2576 pa->used = origin_ds->ds_phys->ds_used_bytes;
2577 pa->comp = origin_ds->ds_phys->ds_compressed_bytes;
2578 pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
2579 for (snap = list_head(&pa->shared_snaps); snap;
2580 snap = list_next(&pa->shared_snaps, snap)) {
2581 uint64_t val, dlused, dlcomp, dluncomp;
2582 dsl_dataset_t *ds = snap->ds;
2584 /* Check that the snapshot name does not conflict */
2585 VERIFY(0 == dsl_dataset_get_snapname(ds));
2586 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2594 /* The very first snapshot does not have a deadlist */
2595 if (ds->ds_phys->ds_prev_snap_obj == 0)
2598 dsl_deadlist_space(&ds->ds_deadlist,
2599 &dlused, &dlcomp, &dluncomp);
2602 pa->uncomp += dluncomp;
2606 * If we are a clone of a clone then we never reached ORIGIN,
2607 * so we need to subtract out the clone origin's used space.
2609 if (pa->origin_origin) {
2610 pa->used -= pa->origin_origin->ds_phys->ds_used_bytes;
2611 pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes;
2612 pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes;
2615 /* Check that there is enough space here */
2616 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2622 * Compute the amounts of space that will be used by snapshots
2623 * after the promotion (for both origin and clone). For each,
2624 * it is the amount of space that will be on all of their
2625 * deadlists (that was not born before their new origin).
2627 if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2631 * Note, typically this will not be a clone of a clone,
2632 * so dd_origin_txg will be < TXG_INITIAL, so
2633 * these snaplist_space() -> dsl_deadlist_space_range()
2634 * calls will be fast because they do not have to
2635 * iterate over all bps.
2637 snap = list_head(&pa->origin_snaps);
2638 err = snaplist_space(&pa->shared_snaps,
2639 snap->ds->ds_dir->dd_origin_txg, &pa->cloneusedsnap);
2643 err = snaplist_space(&pa->clone_snaps,
2644 snap->ds->ds_dir->dd_origin_txg, &space);
2647 pa->cloneusedsnap += space;
2649 if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2650 err = snaplist_space(&pa->origin_snaps,
2651 origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap);
2658 pa->err_ds = snap->ds->ds_snapname;
2663 dsl_dataset_promote_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2665 dsl_dataset_t *hds = arg1;
2666 struct promotearg *pa = arg2;
2667 struct promotenode *snap = list_head(&pa->shared_snaps);
2668 dsl_dataset_t *origin_ds = snap->ds;
2669 dsl_dataset_t *origin_head;
2670 dsl_dir_t *dd = hds->ds_dir;
2671 dsl_pool_t *dp = hds->ds_dir->dd_pool;
2672 dsl_dir_t *odd = NULL;
2673 uint64_t oldnext_obj;
2676 ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
2678 snap = list_head(&pa->origin_snaps);
2679 origin_head = snap->ds;
2682 * We need to explicitly open odd, since origin_ds's dd will be
2685 VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object,
2688 /* change origin's next snap */
2689 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2690 oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2691 snap = list_tail(&pa->clone_snaps);
2692 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2693 origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
2695 /* change the origin's next clone */
2696 if (origin_ds->ds_phys->ds_next_clones_obj) {
2697 remove_from_next_clones(origin_ds, snap->ds->ds_object, tx);
2698 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2699 origin_ds->ds_phys->ds_next_clones_obj,
2704 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2705 ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2706 dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2707 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2708 dmu_buf_will_dirty(odd->dd_dbuf, tx);
2709 odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2710 origin_head->ds_dir->dd_origin_txg =
2711 origin_ds->ds_phys->ds_creation_txg;
2713 /* change dd_clone entries */
2714 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2715 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2716 odd->dd_phys->dd_clones, hds->ds_object, tx));
2717 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2718 pa->origin_origin->ds_dir->dd_phys->dd_clones,
2719 hds->ds_object, tx));
2721 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2722 pa->origin_origin->ds_dir->dd_phys->dd_clones,
2723 origin_head->ds_object, tx));
2724 if (dd->dd_phys->dd_clones == 0) {
2725 dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset,
2726 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
2728 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2729 dd->dd_phys->dd_clones, origin_head->ds_object, tx));
2733 /* move snapshots to this dir */
2734 for (snap = list_head(&pa->shared_snaps); snap;
2735 snap = list_next(&pa->shared_snaps, snap)) {
2736 dsl_dataset_t *ds = snap->ds;
2738 /* unregister props as dsl_dir is changing */
2739 if (ds->ds_objset) {
2740 dmu_objset_evict(ds->ds_objset);
2741 ds->ds_objset = NULL;
2743 /* move snap name entry */
2744 VERIFY(0 == dsl_dataset_get_snapname(ds));
2745 VERIFY(0 == dsl_dataset_snap_remove(origin_head,
2746 ds->ds_snapname, tx));
2747 VERIFY(0 == zap_add(dp->dp_meta_objset,
2748 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2749 8, 1, &ds->ds_object, tx));
2751 /* change containing dsl_dir */
2752 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2753 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2754 ds->ds_phys->ds_dir_obj = dd->dd_object;
2755 ASSERT3P(ds->ds_dir, ==, odd);
2756 dsl_dir_close(ds->ds_dir, ds);
2757 VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
2758 NULL, ds, &ds->ds_dir));
2760 /* move any clone references */
2761 if (ds->ds_phys->ds_next_clones_obj &&
2762 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2766 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2767 ds->ds_phys->ds_next_clones_obj);
2768 zap_cursor_retrieve(&zc, &za) == 0;
2769 zap_cursor_advance(&zc)) {
2770 dsl_dataset_t *cnds;
2773 if (za.za_first_integer == oldnext_obj) {
2775 * We've already moved the
2776 * origin's reference.
2781 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
2782 za.za_first_integer, FTAG, &cnds));
2783 o = cnds->ds_dir->dd_phys->dd_head_dataset_obj;
2785 VERIFY3U(zap_remove_int(dp->dp_meta_objset,
2786 odd->dd_phys->dd_clones, o, tx), ==, 0);
2787 VERIFY3U(zap_add_int(dp->dp_meta_objset,
2788 dd->dd_phys->dd_clones, o, tx), ==, 0);
2789 dsl_dataset_rele(cnds, FTAG);
2791 zap_cursor_fini(&zc);
2794 ASSERT3U(dsl_prop_numcb(ds), ==, 0);
2798 * Change space accounting.
2799 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2800 * both be valid, or both be 0 (resulting in delta == 0). This
2801 * is true for each of {clone,origin} independently.
2804 delta = pa->cloneusedsnap -
2805 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2806 ASSERT3S(delta, >=, 0);
2807 ASSERT3U(pa->used, >=, delta);
2808 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2809 dsl_dir_diduse_space(dd, DD_USED_HEAD,
2810 pa->used - delta, pa->comp, pa->uncomp, tx);
2812 delta = pa->originusedsnap -
2813 odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2814 ASSERT3S(delta, <=, 0);
2815 ASSERT3U(pa->used, >=, -delta);
2816 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2817 dsl_dir_diduse_space(odd, DD_USED_HEAD,
2818 -pa->used - delta, -pa->comp, -pa->uncomp, tx);
2820 origin_ds->ds_phys->ds_unique_bytes = pa->unique;
2822 /* log history record */
2823 spa_history_log_internal(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx,
2824 "dataset = %llu", hds->ds_object);
2826 dsl_dir_close(odd, FTAG);
2829 static char *snaplist_tag = "snaplist";
2831 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2832 * (exclusive) and last_obj (inclusive). The list will be in reverse
2833 * order (last_obj will be the list_head()). If first_obj == 0, do all
2834 * snapshots back to this dataset's origin.
2837 snaplist_make(dsl_pool_t *dp, boolean_t own,
2838 uint64_t first_obj, uint64_t last_obj, list_t *l)
2840 uint64_t obj = last_obj;
2842 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock));
2844 list_create(l, sizeof (struct promotenode),
2845 offsetof(struct promotenode, link));
2847 while (obj != first_obj) {
2849 struct promotenode *snap;
2853 err = dsl_dataset_own_obj(dp, obj,
2854 0, snaplist_tag, &ds);
2856 dsl_dataset_make_exclusive(ds, snaplist_tag);
2858 err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds);
2860 if (err == ENOENT) {
2861 /* lost race with snapshot destroy */
2862 struct promotenode *last = list_tail(l);
2863 ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj);
2864 obj = last->ds->ds_phys->ds_prev_snap_obj;
2871 first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2873 snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP);
2875 list_insert_tail(l, snap);
2876 obj = ds->ds_phys->ds_prev_snap_obj;
2883 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2885 struct promotenode *snap;
2888 for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2889 uint64_t used, comp, uncomp;
2890 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2891 mintxg, UINT64_MAX, &used, &comp, &uncomp);
2898 snaplist_destroy(list_t *l, boolean_t own)
2900 struct promotenode *snap;
2902 if (!l || !list_link_active(&l->list_head))
2905 while ((snap = list_tail(l)) != NULL) {
2906 list_remove(l, snap);
2908 dsl_dataset_disown(snap->ds, snaplist_tag);
2910 dsl_dataset_rele(snap->ds, snaplist_tag);
2911 kmem_free(snap, sizeof (struct promotenode));
2917 * Promote a clone. Nomenclature note:
2918 * "clone" or "cds": the original clone which is being promoted
2919 * "origin" or "ods": the snapshot which is originally clone's origin
2920 * "origin head" or "ohds": the dataset which is the head
2921 * (filesystem/volume) for the origin
2922 * "origin origin": the origin of the origin's filesystem (typically
2923 * NULL, indicating that the clone is not a clone of a clone).
2926 dsl_dataset_promote(const char *name, char *conflsnap)
2931 dmu_object_info_t doi;
2932 struct promotearg pa;
2933 struct promotenode *snap;
2936 bzero(&pa, sizeof(struct promotearg));
2937 err = dsl_dataset_hold(name, FTAG, &ds);
2943 err = dmu_object_info(dp->dp_meta_objset,
2944 ds->ds_phys->ds_snapnames_zapobj, &doi);
2946 dsl_dataset_rele(ds, FTAG);
2950 if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) {
2951 dsl_dataset_rele(ds, FTAG);
2956 * We are going to inherit all the snapshots taken before our
2957 * origin (i.e., our new origin will be our parent's origin).
2958 * Take ownership of them so that we can rename them into our
2961 rw_enter(&dp->dp_config_rwlock, RW_READER);
2963 err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj,
2968 err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps);
2972 snap = list_head(&pa.shared_snaps);
2973 ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
2974 err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj,
2975 snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps);
2979 if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) {
2980 err = dsl_dataset_hold_obj(dp,
2981 snap->ds->ds_dir->dd_phys->dd_origin_obj,
2982 FTAG, &pa.origin_origin);
2988 rw_exit(&dp->dp_config_rwlock);
2991 * Add in 128x the snapnames zapobj size, since we will be moving
2992 * a bunch of snapnames to the promoted ds, and dirtying their
2996 err = dsl_sync_task_do(dp, dsl_dataset_promote_check,
2997 dsl_dataset_promote_sync, ds, &pa,
2998 2 + 2 * doi.doi_physical_blocks_512);
2999 if (err && pa.err_ds && conflsnap)
3000 (void) strncpy(conflsnap, pa.err_ds, MAXNAMELEN);
3003 snaplist_destroy(&pa.shared_snaps, B_TRUE);
3004 snaplist_destroy(&pa.clone_snaps, B_FALSE);
3005 snaplist_destroy(&pa.origin_snaps, B_FALSE);
3006 if (pa.origin_origin)
3007 dsl_dataset_rele(pa.origin_origin, FTAG);
3008 dsl_dataset_rele(ds, FTAG);
3012 struct cloneswaparg {
3013 dsl_dataset_t *cds; /* clone dataset */
3014 dsl_dataset_t *ohds; /* origin's head dataset */
3016 int64_t unused_refres_delta; /* change in unconsumed refreservation */
3021 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx)
3023 struct cloneswaparg *csa = arg1;
3025 /* they should both be heads */
3026 if (dsl_dataset_is_snapshot(csa->cds) ||
3027 dsl_dataset_is_snapshot(csa->ohds))
3030 /* the branch point should be just before them */
3031 if (csa->cds->ds_prev != csa->ohds->ds_prev)
3034 /* cds should be the clone (unless they are unrelated) */
3035 if (csa->cds->ds_prev != NULL &&
3036 csa->cds->ds_prev != csa->cds->ds_dir->dd_pool->dp_origin_snap &&
3037 csa->ohds->ds_object !=
3038 csa->cds->ds_prev->ds_phys->ds_next_snap_obj)
3041 /* the clone should be a child of the origin */
3042 if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir)
3045 /* ohds shouldn't be modified unless 'force' */
3046 if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds))
3049 /* adjust amount of any unconsumed refreservation */
3050 csa->unused_refres_delta =
3051 (int64_t)MIN(csa->ohds->ds_reserved,
3052 csa->ohds->ds_phys->ds_unique_bytes) -
3053 (int64_t)MIN(csa->ohds->ds_reserved,
3054 csa->cds->ds_phys->ds_unique_bytes);
3056 if (csa->unused_refres_delta > 0 &&
3057 csa->unused_refres_delta >
3058 dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE))
3061 if (csa->ohds->ds_quota != 0 &&
3062 csa->cds->ds_phys->ds_unique_bytes > csa->ohds->ds_quota)
3070 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3072 struct cloneswaparg *csa = arg1;
3073 dsl_pool_t *dp = csa->cds->ds_dir->dd_pool;
3075 ASSERT(csa->cds->ds_reserved == 0);
3076 ASSERT(csa->ohds->ds_quota == 0 ||
3077 csa->cds->ds_phys->ds_unique_bytes <= csa->ohds->ds_quota);
3079 dmu_buf_will_dirty(csa->cds->ds_dbuf, tx);
3080 dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx);
3082 if (csa->cds->ds_objset != NULL) {
3083 dmu_objset_evict(csa->cds->ds_objset);
3084 csa->cds->ds_objset = NULL;
3087 if (csa->ohds->ds_objset != NULL) {
3088 dmu_objset_evict(csa->ohds->ds_objset);
3089 csa->ohds->ds_objset = NULL;
3093 * Reset origin's unique bytes, if it exists.
3095 if (csa->cds->ds_prev) {
3096 dsl_dataset_t *origin = csa->cds->ds_prev;
3097 uint64_t comp, uncomp;
3099 dmu_buf_will_dirty(origin->ds_dbuf, tx);
3100 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3101 origin->ds_phys->ds_prev_snap_txg, UINT64_MAX,
3102 &origin->ds_phys->ds_unique_bytes, &comp, &uncomp);
3108 tmp = csa->ohds->ds_phys->ds_bp;
3109 csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp;
3110 csa->cds->ds_phys->ds_bp = tmp;
3113 /* set dd_*_bytes */
3115 int64_t dused, dcomp, duncomp;
3116 uint64_t cdl_used, cdl_comp, cdl_uncomp;
3117 uint64_t odl_used, odl_comp, odl_uncomp;
3119 ASSERT3U(csa->cds->ds_dir->dd_phys->
3120 dd_used_breakdown[DD_USED_SNAP], ==, 0);
3122 dsl_deadlist_space(&csa->cds->ds_deadlist,
3123 &cdl_used, &cdl_comp, &cdl_uncomp);
3124 dsl_deadlist_space(&csa->ohds->ds_deadlist,
3125 &odl_used, &odl_comp, &odl_uncomp);
3127 dused = csa->cds->ds_phys->ds_used_bytes + cdl_used -
3128 (csa->ohds->ds_phys->ds_used_bytes + odl_used);
3129 dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp -
3130 (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp);
3131 duncomp = csa->cds->ds_phys->ds_uncompressed_bytes +
3133 (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp);
3135 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD,
3136 dused, dcomp, duncomp, tx);
3137 dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD,
3138 -dused, -dcomp, -duncomp, tx);
3141 * The difference in the space used by snapshots is the
3142 * difference in snapshot space due to the head's
3143 * deadlist (since that's the only thing that's
3144 * changing that affects the snapused).
3146 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3147 csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3148 &cdl_used, &cdl_comp, &cdl_uncomp);
3149 dsl_deadlist_space_range(&csa->ohds->ds_deadlist,
3150 csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3151 &odl_used, &odl_comp, &odl_uncomp);
3152 dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used,
3153 DD_USED_HEAD, DD_USED_SNAP, tx);
3156 /* swap ds_*_bytes */
3157 SWITCH64(csa->ohds->ds_phys->ds_used_bytes,
3158 csa->cds->ds_phys->ds_used_bytes);
3159 SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes,
3160 csa->cds->ds_phys->ds_compressed_bytes);
3161 SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes,
3162 csa->cds->ds_phys->ds_uncompressed_bytes);
3163 SWITCH64(csa->ohds->ds_phys->ds_unique_bytes,
3164 csa->cds->ds_phys->ds_unique_bytes);
3166 /* apply any parent delta for change in unconsumed refreservation */
3167 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV,
3168 csa->unused_refres_delta, 0, 0, tx);
3173 dsl_deadlist_close(&csa->cds->ds_deadlist);
3174 dsl_deadlist_close(&csa->ohds->ds_deadlist);
3175 SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj,
3176 csa->cds->ds_phys->ds_deadlist_obj);
3177 dsl_deadlist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset,
3178 csa->cds->ds_phys->ds_deadlist_obj);
3179 dsl_deadlist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset,
3180 csa->ohds->ds_phys->ds_deadlist_obj);
3182 dsl_scan_ds_clone_swapped(csa->ohds, csa->cds, tx);
3186 * Swap 'clone' with its origin head datasets. Used at the end of "zfs
3187 * recv" into an existing fs to swizzle the file system to the new
3188 * version, and by "zfs rollback". Can also be used to swap two
3189 * independent head datasets if neither has any snapshots.
3192 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
3195 struct cloneswaparg csa;
3198 ASSERT(clone->ds_owner);
3199 ASSERT(origin_head->ds_owner);
3202 * Need exclusive access for the swap. If we're swapping these
3203 * datasets back after an error, we already hold the locks.
3205 if (!RW_WRITE_HELD(&clone->ds_rwlock))
3206 rw_enter(&clone->ds_rwlock, RW_WRITER);
3207 if (!RW_WRITE_HELD(&origin_head->ds_rwlock) &&
3208 !rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) {
3209 rw_exit(&clone->ds_rwlock);
3210 rw_enter(&origin_head->ds_rwlock, RW_WRITER);
3211 if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) {
3212 rw_exit(&origin_head->ds_rwlock);
3217 csa.ohds = origin_head;
3219 error = dsl_sync_task_do(clone->ds_dir->dd_pool,
3220 dsl_dataset_clone_swap_check,
3221 dsl_dataset_clone_swap_sync, &csa, NULL, 9);
3226 * Given a pool name and a dataset object number in that pool,
3227 * return the name of that dataset.
3230 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3237 if ((error = spa_open(pname, &spa, FTAG)) != 0)
3239 dp = spa_get_dsl(spa);
3240 rw_enter(&dp->dp_config_rwlock, RW_READER);
3241 if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) {
3242 dsl_dataset_name(ds, buf);
3243 dsl_dataset_rele(ds, FTAG);
3245 rw_exit(&dp->dp_config_rwlock);
3246 spa_close(spa, FTAG);
3252 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3253 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3257 ASSERT3S(asize, >, 0);
3260 * *ref_rsrv is the portion of asize that will come from any
3261 * unconsumed refreservation space.
3265 mutex_enter(&ds->ds_lock);
3267 * Make a space adjustment for reserved bytes.
3269 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
3271 ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3272 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3274 asize - MIN(asize, parent_delta(ds, asize + inflight));
3277 if (!check_quota || ds->ds_quota == 0) {
3278 mutex_exit(&ds->ds_lock);
3282 * If they are requesting more space, and our current estimate
3283 * is over quota, they get to try again unless the actual
3284 * on-disk is over quota and there are no pending changes (which
3285 * may free up space for us).
3287 if (ds->ds_phys->ds_used_bytes + inflight >= ds->ds_quota) {
3288 if (inflight > 0 || ds->ds_phys->ds_used_bytes < ds->ds_quota)
3293 DMU_TX_STAT_BUMP(dmu_tx_quota);
3295 mutex_exit(&ds->ds_lock);
3302 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
3304 dsl_dataset_t *ds = arg1;
3305 dsl_prop_setarg_t *psa = arg2;
3308 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA)
3311 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3314 if (psa->psa_effective_value == 0)
3317 if (psa->psa_effective_value < ds->ds_phys->ds_used_bytes ||
3318 psa->psa_effective_value < ds->ds_reserved)
3324 extern void dsl_prop_set_sync(void *, void *, dmu_tx_t *);
3327 dsl_dataset_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3329 dsl_dataset_t *ds = arg1;
3330 dsl_prop_setarg_t *psa = arg2;
3331 uint64_t effective_value = psa->psa_effective_value;
3333 dsl_prop_set_sync(ds, psa, tx);
3334 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3336 if (ds->ds_quota != effective_value) {
3337 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3338 ds->ds_quota = effective_value;
3340 spa_history_log_internal(LOG_DS_REFQUOTA,
3341 ds->ds_dir->dd_pool->dp_spa, tx, "%lld dataset = %llu ",
3342 (longlong_t)ds->ds_quota, ds->ds_object);
3347 dsl_dataset_set_quota(const char *dsname, zprop_source_t source, uint64_t quota)
3350 dsl_prop_setarg_t psa;
3353 dsl_prop_setarg_init_uint64(&psa, "refquota", source, "a);
3355 err = dsl_dataset_hold(dsname, FTAG, &ds);
3360 * If someone removes a file, then tries to set the quota, we
3361 * want to make sure the file freeing takes effect.
3363 txg_wait_open(ds->ds_dir->dd_pool, 0);
3365 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3366 dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync,
3369 dsl_dataset_rele(ds, FTAG);
3374 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
3376 dsl_dataset_t *ds = arg1;
3377 dsl_prop_setarg_t *psa = arg2;
3378 uint64_t effective_value;
3382 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
3383 SPA_VERSION_REFRESERVATION)
3386 if (dsl_dataset_is_snapshot(ds))
3389 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3392 effective_value = psa->psa_effective_value;
3395 * If we are doing the preliminary check in open context, the
3396 * space estimates may be inaccurate.
3398 if (!dmu_tx_is_syncing(tx))
3401 mutex_enter(&ds->ds_lock);
3402 if (!DS_UNIQUE_IS_ACCURATE(ds))
3403 dsl_dataset_recalc_head_uniq(ds);
3404 unique = ds->ds_phys->ds_unique_bytes;
3405 mutex_exit(&ds->ds_lock);
3407 if (MAX(unique, effective_value) > MAX(unique, ds->ds_reserved)) {
3408 uint64_t delta = MAX(unique, effective_value) -
3409 MAX(unique, ds->ds_reserved);
3411 if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
3413 if (ds->ds_quota > 0 &&
3414 effective_value > ds->ds_quota)
3422 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3424 dsl_dataset_t *ds = arg1;
3425 dsl_prop_setarg_t *psa = arg2;
3426 uint64_t effective_value = psa->psa_effective_value;
3430 dsl_prop_set_sync(ds, psa, tx);
3431 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3433 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3435 mutex_enter(&ds->ds_dir->dd_lock);
3436 mutex_enter(&ds->ds_lock);
3437 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3438 unique = ds->ds_phys->ds_unique_bytes;
3439 delta = MAX(0, (int64_t)(effective_value - unique)) -
3440 MAX(0, (int64_t)(ds->ds_reserved - unique));
3441 ds->ds_reserved = effective_value;
3442 mutex_exit(&ds->ds_lock);
3444 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3445 mutex_exit(&ds->ds_dir->dd_lock);
3447 spa_history_log_internal(LOG_DS_REFRESERV,
3448 ds->ds_dir->dd_pool->dp_spa, tx, "%lld dataset = %llu",
3449 (longlong_t)effective_value, ds->ds_object);
3453 dsl_dataset_set_reservation(const char *dsname, zprop_source_t source,
3454 uint64_t reservation)
3457 dsl_prop_setarg_t psa;
3460 dsl_prop_setarg_init_uint64(&psa, "refreservation", source,
3463 err = dsl_dataset_hold(dsname, FTAG, &ds);
3467 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3468 dsl_dataset_set_reservation_check,
3469 dsl_dataset_set_reservation_sync, ds, &psa, 0);
3471 dsl_dataset_rele(ds, FTAG);
3475 typedef struct zfs_hold_cleanup_arg {
3478 char htag[MAXNAMELEN];
3479 } zfs_hold_cleanup_arg_t;
3482 dsl_dataset_user_release_onexit(void *arg)
3484 zfs_hold_cleanup_arg_t *ca = arg;
3486 (void) dsl_dataset_user_release_tmp(ca->dp, ca->dsobj, ca->htag,
3488 kmem_free(ca, sizeof (zfs_hold_cleanup_arg_t));
3492 dsl_register_onexit_hold_cleanup(dsl_dataset_t *ds, const char *htag,
3495 zfs_hold_cleanup_arg_t *ca;
3497 ca = kmem_alloc(sizeof (zfs_hold_cleanup_arg_t), KM_SLEEP);
3498 ca->dp = ds->ds_dir->dd_pool;
3499 ca->dsobj = ds->ds_object;
3500 (void) strlcpy(ca->htag, htag, sizeof (ca->htag));
3501 VERIFY3U(0, ==, zfs_onexit_add_cb(minor,
3502 dsl_dataset_user_release_onexit, ca, NULL));
3506 * If you add new checks here, you may need to add
3507 * additional checks to the "temporary" case in
3508 * snapshot_check() in dmu_objset.c.
3511 dsl_dataset_user_hold_check(void *arg1, void *arg2, dmu_tx_t *tx)
3513 dsl_dataset_t *ds = arg1;
3514 struct dsl_ds_holdarg *ha = arg2;
3515 char *htag = ha->htag;
3516 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3519 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3522 if (!dsl_dataset_is_snapshot(ds))
3525 /* tags must be unique */
3526 mutex_enter(&ds->ds_lock);
3527 if (ds->ds_phys->ds_userrefs_obj) {
3528 error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj, htag,
3532 else if (error == ENOENT)
3535 mutex_exit(&ds->ds_lock);
3537 if (error == 0 && ha->temphold &&
3538 strlen(htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN)
3545 dsl_dataset_user_hold_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3547 dsl_dataset_t *ds = arg1;
3548 struct dsl_ds_holdarg *ha = arg2;
3549 char *htag = ha->htag;
3550 dsl_pool_t *dp = ds->ds_dir->dd_pool;
3551 objset_t *mos = dp->dp_meta_objset;
3552 uint64_t now = gethrestime_sec();
3555 mutex_enter(&ds->ds_lock);
3556 if (ds->ds_phys->ds_userrefs_obj == 0) {
3558 * This is the first user hold for this dataset. Create
3559 * the userrefs zap object.
3561 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3562 zapobj = ds->ds_phys->ds_userrefs_obj =
3563 zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx);
3565 zapobj = ds->ds_phys->ds_userrefs_obj;
3568 mutex_exit(&ds->ds_lock);
3570 VERIFY(0 == zap_add(mos, zapobj, htag, 8, 1, &now, tx));
3573 VERIFY(0 == dsl_pool_user_hold(dp, ds->ds_object,
3577 spa_history_log_internal(LOG_DS_USER_HOLD,
3578 dp->dp_spa, tx, "<%s> temp = %d dataset = %llu", htag,
3579 (int)ha->temphold, ds->ds_object);
3583 dsl_dataset_user_hold_one(const char *dsname, void *arg)
3585 struct dsl_ds_holdarg *ha = arg;
3590 /* alloc a buffer to hold dsname@snapname plus terminating NULL */
3591 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3592 error = dsl_dataset_hold(name, ha->dstg, &ds);
3595 ha->gotone = B_TRUE;
3596 dsl_sync_task_create(ha->dstg, dsl_dataset_user_hold_check,
3597 dsl_dataset_user_hold_sync, ds, ha, 0);
3598 } else if (error == ENOENT && ha->recursive) {
3601 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3607 dsl_dataset_user_hold_for_send(dsl_dataset_t *ds, char *htag,
3610 struct dsl_ds_holdarg *ha;
3613 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3615 ha->temphold = temphold;
3616 error = dsl_sync_task_do(ds->ds_dir->dd_pool,
3617 dsl_dataset_user_hold_check, dsl_dataset_user_hold_sync,
3619 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3625 dsl_dataset_user_hold(char *dsname, char *snapname, char *htag,
3626 boolean_t recursive, boolean_t temphold, int cleanup_fd)
3628 struct dsl_ds_holdarg *ha;
3629 dsl_sync_task_t *dst;
3634 if (cleanup_fd != -1) {
3635 /* Currently we only support cleanup-on-exit of tempholds. */
3638 error = zfs_onexit_fd_hold(cleanup_fd, &minor);
3643 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3645 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3647 error = spa_open(dsname, &spa, FTAG);
3649 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3650 if (cleanup_fd != -1)
3651 zfs_onexit_fd_rele(cleanup_fd);
3655 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3657 ha->snapname = snapname;
3658 ha->recursive = recursive;
3659 ha->temphold = temphold;
3662 error = dmu_objset_find(dsname, dsl_dataset_user_hold_one,
3663 ha, DS_FIND_CHILDREN);
3665 error = dsl_dataset_user_hold_one(dsname, ha);
3668 error = dsl_sync_task_group_wait(ha->dstg);
3670 for (dst = list_head(&ha->dstg->dstg_tasks); dst;
3671 dst = list_next(&ha->dstg->dstg_tasks, dst)) {
3672 dsl_dataset_t *ds = dst->dst_arg1;
3675 dsl_dataset_name(ds, ha->failed);
3676 *strchr(ha->failed, '@') = '\0';
3677 } else if (error == 0 && minor != 0 && temphold) {
3679 * If this hold is to be released upon process exit,
3680 * register that action now.
3682 dsl_register_onexit_hold_cleanup(ds, htag, minor);
3684 dsl_dataset_rele(ds, ha->dstg);
3687 if (error == 0 && recursive && !ha->gotone)
3691 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
3693 dsl_sync_task_group_destroy(ha->dstg);
3695 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3696 spa_close(spa, FTAG);
3697 if (cleanup_fd != -1)
3698 zfs_onexit_fd_rele(cleanup_fd);
3702 struct dsl_ds_releasearg {
3705 boolean_t own; /* do we own or just hold ds? */
3709 dsl_dataset_release_might_destroy(dsl_dataset_t *ds, const char *htag,
3710 boolean_t *might_destroy)
3712 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3717 *might_destroy = B_FALSE;
3719 mutex_enter(&ds->ds_lock);
3720 zapobj = ds->ds_phys->ds_userrefs_obj;
3722 /* The tag can't possibly exist */
3723 mutex_exit(&ds->ds_lock);
3727 /* Make sure the tag exists */
3728 error = zap_lookup(mos, zapobj, htag, 8, 1, &tmp);
3730 mutex_exit(&ds->ds_lock);
3731 if (error == ENOENT)
3736 if (ds->ds_userrefs == 1 && ds->ds_phys->ds_num_children == 1 &&
3737 DS_IS_DEFER_DESTROY(ds))
3738 *might_destroy = B_TRUE;
3740 mutex_exit(&ds->ds_lock);
3745 dsl_dataset_user_release_check(void *arg1, void *tag, dmu_tx_t *tx)
3747 struct dsl_ds_releasearg *ra = arg1;
3748 dsl_dataset_t *ds = ra->ds;
3749 boolean_t might_destroy;
3752 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3755 error = dsl_dataset_release_might_destroy(ds, ra->htag, &might_destroy);
3759 if (might_destroy) {
3760 struct dsl_ds_destroyarg dsda = {0};
3762 if (dmu_tx_is_syncing(tx)) {
3764 * If we're not prepared to remove the snapshot,
3765 * we can't allow the release to happen right now.
3771 dsda.releasing = B_TRUE;
3772 return (dsl_dataset_destroy_check(&dsda, tag, tx));
3779 dsl_dataset_user_release_sync(void *arg1, void *tag, dmu_tx_t *tx)
3781 struct dsl_ds_releasearg *ra = arg1;
3782 dsl_dataset_t *ds = ra->ds;
3783 dsl_pool_t *dp = ds->ds_dir->dd_pool;
3784 objset_t *mos = dp->dp_meta_objset;
3786 uint64_t dsobj = ds->ds_object;
3790 mutex_enter(&ds->ds_lock);
3792 refs = ds->ds_userrefs;
3793 mutex_exit(&ds->ds_lock);
3794 error = dsl_pool_user_release(dp, ds->ds_object, ra->htag, tx);
3795 VERIFY(error == 0 || error == ENOENT);
3796 zapobj = ds->ds_phys->ds_userrefs_obj;
3797 VERIFY(0 == zap_remove(mos, zapobj, ra->htag, tx));
3798 if (ds->ds_userrefs == 0 && ds->ds_phys->ds_num_children == 1 &&
3799 DS_IS_DEFER_DESTROY(ds)) {
3800 struct dsl_ds_destroyarg dsda = {0};
3804 dsda.releasing = B_TRUE;
3805 /* We already did the destroy_check */
3806 dsl_dataset_destroy_sync(&dsda, tag, tx);
3809 spa_history_log_internal(LOG_DS_USER_RELEASE,
3810 dp->dp_spa, tx, "<%s> %lld dataset = %llu",
3811 ra->htag, (longlong_t)refs, dsobj);
3815 dsl_dataset_user_release_one(const char *dsname, void *arg)
3817 struct dsl_ds_holdarg *ha = arg;
3818 struct dsl_ds_releasearg *ra;
3821 void *dtag = ha->dstg;
3823 boolean_t own = B_FALSE;
3824 boolean_t might_destroy;
3826 /* alloc a buffer to hold dsname@snapname, plus the terminating NULL */
3827 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3828 error = dsl_dataset_hold(name, dtag, &ds);
3830 if (error == ENOENT && ha->recursive)
3832 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3836 ha->gotone = B_TRUE;
3838 ASSERT(dsl_dataset_is_snapshot(ds));
3840 error = dsl_dataset_release_might_destroy(ds, ha->htag, &might_destroy);
3842 dsl_dataset_rele(ds, dtag);
3846 if (might_destroy) {
3848 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3849 error = zfs_unmount_snap(name, NULL);
3852 dsl_dataset_rele(ds, dtag);
3856 if (!dsl_dataset_tryown(ds, B_TRUE, dtag)) {
3857 dsl_dataset_rele(ds, dtag);
3861 dsl_dataset_make_exclusive(ds, dtag);
3865 ra = kmem_alloc(sizeof (struct dsl_ds_releasearg), KM_SLEEP);
3867 ra->htag = ha->htag;
3869 dsl_sync_task_create(ha->dstg, dsl_dataset_user_release_check,
3870 dsl_dataset_user_release_sync, ra, dtag, 0);
3876 dsl_dataset_user_release(char *dsname, char *snapname, char *htag,
3877 boolean_t recursive)
3879 struct dsl_ds_holdarg *ha;
3880 dsl_sync_task_t *dst;
3885 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3887 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3889 error = spa_open(dsname, &spa, FTAG);
3891 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3895 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3897 ha->snapname = snapname;
3898 ha->recursive = recursive;
3900 error = dmu_objset_find(dsname, dsl_dataset_user_release_one,
3901 ha, DS_FIND_CHILDREN);
3903 error = dsl_dataset_user_release_one(dsname, ha);
3906 error = dsl_sync_task_group_wait(ha->dstg);
3908 for (dst = list_head(&ha->dstg->dstg_tasks); dst;
3909 dst = list_next(&ha->dstg->dstg_tasks, dst)) {
3910 struct dsl_ds_releasearg *ra = dst->dst_arg1;
3911 dsl_dataset_t *ds = ra->ds;
3914 dsl_dataset_name(ds, ha->failed);
3917 dsl_dataset_disown(ds, ha->dstg);
3919 dsl_dataset_rele(ds, ha->dstg);
3921 kmem_free(ra, sizeof (struct dsl_ds_releasearg));
3924 if (error == 0 && recursive && !ha->gotone)
3927 if (error && error != EBUSY)
3928 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
3930 dsl_sync_task_group_destroy(ha->dstg);
3931 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3932 spa_close(spa, FTAG);
3935 * We can get EBUSY if we were racing with deferred destroy and
3936 * dsl_dataset_user_release_check() hadn't done the necessary
3937 * open context setup. We can also get EBUSY if we're racing
3938 * with destroy and that thread is the ds_owner. Either way
3939 * the busy condition should be transient, and we should retry
3940 * the release operation.
3949 * Called at spa_load time (with retry == B_FALSE) to release a stale
3950 * temporary user hold. Also called by the onexit code (with retry == B_TRUE).
3953 dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, char *htag,
3963 rw_enter(&dp->dp_config_rwlock, RW_READER);
3964 error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
3965 rw_exit(&dp->dp_config_rwlock);
3968 namelen = dsl_dataset_namelen(ds)+1;
3969 name = kmem_alloc(namelen, KM_SLEEP);
3970 dsl_dataset_name(ds, name);
3971 dsl_dataset_rele(ds, FTAG);
3973 snap = strchr(name, '@');
3976 error = dsl_dataset_user_release(name, snap, htag, B_FALSE);
3977 kmem_free(name, namelen);
3980 * The object can't have been destroyed because we have a hold,
3981 * but it might have been renamed, resulting in ENOENT. Retry
3982 * if we've been requested to do so.
3984 * It would be nice if we could use the dsobj all the way
3985 * through and avoid ENOENT entirely. But we might need to
3986 * unmount the snapshot, and there's currently no way to lookup
3987 * a vfsp using a ZFS object id.
3989 } while ((error == ENOENT) && retry);
3995 dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp)
4000 err = dsl_dataset_hold(dsname, FTAG, &ds);
4004 VERIFY(0 == nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP));
4005 if (ds->ds_phys->ds_userrefs_obj != 0) {
4006 zap_attribute_t *za;
4009 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
4010 for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset,
4011 ds->ds_phys->ds_userrefs_obj);
4012 zap_cursor_retrieve(&zc, za) == 0;
4013 zap_cursor_advance(&zc)) {
4014 VERIFY(0 == nvlist_add_uint64(*nvp, za->za_name,
4015 za->za_first_integer));
4017 zap_cursor_fini(&zc);
4018 kmem_free(za, sizeof (zap_attribute_t));
4020 dsl_dataset_rele(ds, FTAG);
4025 * Note, this fuction is used as the callback for dmu_objset_find(). We
4026 * always return 0 so that we will continue to find and process
4027 * inconsistent datasets, even if we encounter an error trying to
4028 * process one of them.
4032 dsl_destroy_inconsistent(const char *dsname, void *arg)
4036 if (dsl_dataset_own(dsname, B_TRUE, FTAG, &ds) == 0) {
4037 if (DS_IS_INCONSISTENT(ds))
4038 (void) dsl_dataset_destroy(ds, FTAG, B_FALSE);
4040 dsl_dataset_disown(ds, FTAG);
4045 #if defined(_KERNEL) && defined(HAVE_SPL)
4046 EXPORT_SYMBOL(dsl_dataset_hold);
4047 EXPORT_SYMBOL(dsl_dataset_hold_obj);
4048 EXPORT_SYMBOL(dsl_dataset_own);
4049 EXPORT_SYMBOL(dsl_dataset_own_obj);
4050 EXPORT_SYMBOL(dsl_dataset_name);
4051 EXPORT_SYMBOL(dsl_dataset_rele);
4052 EXPORT_SYMBOL(dsl_dataset_disown);
4053 EXPORT_SYMBOL(dsl_dataset_drop_ref);
4054 EXPORT_SYMBOL(dsl_dataset_tryown);
4055 EXPORT_SYMBOL(dsl_dataset_make_exclusive);
4056 EXPORT_SYMBOL(dsl_dataset_create_sync);
4057 EXPORT_SYMBOL(dsl_dataset_create_sync_dd);
4058 EXPORT_SYMBOL(dsl_dataset_destroy);
4059 EXPORT_SYMBOL(dsl_snapshots_destroy);
4060 EXPORT_SYMBOL(dsl_dataset_destroy_check);
4061 EXPORT_SYMBOL(dsl_dataset_destroy_sync);
4062 EXPORT_SYMBOL(dsl_dataset_snapshot_check);
4063 EXPORT_SYMBOL(dsl_dataset_snapshot_sync);
4064 EXPORT_SYMBOL(dsl_dataset_rename);
4065 EXPORT_SYMBOL(dsl_dataset_promote);
4066 EXPORT_SYMBOL(dsl_dataset_clone_swap);
4067 EXPORT_SYMBOL(dsl_dataset_user_hold);
4068 EXPORT_SYMBOL(dsl_dataset_user_release);
4069 EXPORT_SYMBOL(dsl_dataset_user_release_tmp);
4070 EXPORT_SYMBOL(dsl_dataset_get_holds);
4071 EXPORT_SYMBOL(dsl_dataset_get_blkptr);
4072 EXPORT_SYMBOL(dsl_dataset_set_blkptr);
4073 EXPORT_SYMBOL(dsl_dataset_get_spa);
4074 EXPORT_SYMBOL(dsl_dataset_modified_since_lastsnap);
4075 EXPORT_SYMBOL(dsl_dataset_sync);
4076 EXPORT_SYMBOL(dsl_dataset_block_born);
4077 EXPORT_SYMBOL(dsl_dataset_block_kill);
4078 EXPORT_SYMBOL(dsl_dataset_block_freeable);
4079 EXPORT_SYMBOL(dsl_dataset_prev_snap_txg);
4080 EXPORT_SYMBOL(dsl_dataset_dirty);
4081 EXPORT_SYMBOL(dsl_dataset_stats);
4082 EXPORT_SYMBOL(dsl_dataset_fast_stat);
4083 EXPORT_SYMBOL(dsl_dataset_space);
4084 EXPORT_SYMBOL(dsl_dataset_fsid_guid);
4085 EXPORT_SYMBOL(dsl_dsobj_to_dsname);
4086 EXPORT_SYMBOL(dsl_dataset_check_quota);
4087 EXPORT_SYMBOL(dsl_dataset_set_quota);
4088 EXPORT_SYMBOL(dsl_dataset_set_quota_sync);
4089 EXPORT_SYMBOL(dsl_dataset_set_reservation);
4090 EXPORT_SYMBOL(dsl_destroy_inconsistent);