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 * Portions Copyright (c) 2011 Martin Matuska <mm@FreeBSD.org>
24 * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
25 * Copyright (c) 2014, Joyent, Inc. All rights reserved.
26 * Copyright (c) 2014 RackTop Systems.
29 #include <sys/dmu_objset.h>
30 #include <sys/dsl_dataset.h>
31 #include <sys/dsl_dir.h>
32 #include <sys/dsl_prop.h>
33 #include <sys/dsl_synctask.h>
34 #include <sys/dmu_traverse.h>
35 #include <sys/dmu_impl.h>
36 #include <sys/dmu_tx.h>
40 #include <sys/zfeature.h>
41 #include <sys/unique.h>
42 #include <sys/zfs_context.h>
43 #include <sys/zfs_ioctl.h>
45 #include <sys/zfs_znode.h>
46 #include <sys/zfs_onexit.h>
48 #include <sys/dsl_scan.h>
49 #include <sys/dsl_deadlist.h>
50 #include <sys/dsl_destroy.h>
51 #include <sys/dsl_userhold.h>
52 #include <sys/dsl_bookmark.h>
54 SYSCTL_DECL(_vfs_zfs);
57 * The SPA supports block sizes up to 16MB. However, very large blocks
58 * can have an impact on i/o latency (e.g. tying up a spinning disk for
59 * ~300ms), and also potentially on the memory allocator. Therefore,
60 * we do not allow the recordsize to be set larger than zfs_max_recordsize
61 * (default 1MB). Larger blocks can be created by changing this tunable,
62 * and pools with larger blocks can always be imported and used, regardless
65 int zfs_max_recordsize = 1 * 1024 * 1024;
66 SYSCTL_INT(_vfs_zfs, OID_AUTO, max_recordsize, CTLFLAG_RWTUN,
67 &zfs_max_recordsize, 0,
68 "Maximum block size. Expect dragons when tuning this.");
70 #define SWITCH64(x, y) \
72 uint64_t __tmp = (x); \
77 #define DS_REF_MAX (1ULL << 62)
79 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
80 extern inline boolean_t dsl_dataset_is_snapshot(dsl_dataset_t *ds);
83 * Figure out how much of this delta should be propogated to the dsl_dir
84 * layer. If there's a refreservation, that space has already been
85 * partially accounted for in our ancestors.
88 parent_delta(dsl_dataset_t *ds, int64_t delta)
90 dsl_dataset_phys_t *ds_phys;
91 uint64_t old_bytes, new_bytes;
93 if (ds->ds_reserved == 0)
96 ds_phys = dsl_dataset_phys(ds);
97 old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
98 new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
100 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
101 return (new_bytes - old_bytes);
105 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
107 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
108 int compressed = BP_GET_PSIZE(bp);
109 int uncompressed = BP_GET_UCSIZE(bp);
112 dprintf_bp(bp, "ds=%p", ds);
114 ASSERT(dmu_tx_is_syncing(tx));
115 /* It could have been compressed away to nothing */
118 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
119 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
121 dsl_pool_mos_diduse_space(tx->tx_pool,
122 used, compressed, uncompressed);
126 dmu_buf_will_dirty(ds->ds_dbuf, tx);
127 mutex_enter(&ds->ds_lock);
128 delta = parent_delta(ds, used);
129 dsl_dataset_phys(ds)->ds_referenced_bytes += used;
130 dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
131 dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
132 dsl_dataset_phys(ds)->ds_unique_bytes += used;
133 if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE)
134 ds->ds_need_large_blocks = B_TRUE;
135 mutex_exit(&ds->ds_lock);
136 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
137 compressed, uncompressed, tx);
138 dsl_dir_transfer_space(ds->ds_dir, used - delta,
139 DD_USED_REFRSRV, DD_USED_HEAD, NULL);
143 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
146 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
147 int compressed = BP_GET_PSIZE(bp);
148 int uncompressed = BP_GET_UCSIZE(bp);
153 ASSERT(dmu_tx_is_syncing(tx));
154 ASSERT(bp->blk_birth <= tx->tx_txg);
157 dsl_free(tx->tx_pool, tx->tx_txg, bp);
158 dsl_pool_mos_diduse_space(tx->tx_pool,
159 -used, -compressed, -uncompressed);
162 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
164 ASSERT(!dsl_dataset_is_snapshot(ds));
165 dmu_buf_will_dirty(ds->ds_dbuf, tx);
167 if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
170 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
171 dsl_free(tx->tx_pool, tx->tx_txg, bp);
173 mutex_enter(&ds->ds_lock);
174 ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
175 !DS_UNIQUE_IS_ACCURATE(ds));
176 delta = parent_delta(ds, -used);
177 dsl_dataset_phys(ds)->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, NULL);
184 dprintf_bp(bp, "putting on dead list: %s", "");
187 * We are here as part of zio's write done callback,
188 * which means we're a zio interrupt thread. We can't
189 * call dsl_deadlist_insert() now because it may block
190 * waiting for I/O. Instead, put bp on the deferred
191 * queue and let dsl_pool_sync() finish the job.
193 bplist_append(&ds->ds_pending_deadlist, bp);
195 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
197 ASSERT3U(ds->ds_prev->ds_object, ==,
198 dsl_dataset_phys(ds)->ds_prev_snap_obj);
199 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
200 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
201 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
202 ds->ds_object && bp->blk_birth >
203 dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
204 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
205 mutex_enter(&ds->ds_prev->ds_lock);
206 dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
207 mutex_exit(&ds->ds_prev->ds_lock);
209 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
210 dsl_dir_transfer_space(ds->ds_dir, used,
211 DD_USED_HEAD, DD_USED_SNAP, tx);
214 mutex_enter(&ds->ds_lock);
215 ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
216 dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
217 ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
218 dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
219 ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
220 dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
221 mutex_exit(&ds->ds_lock);
227 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
229 uint64_t trysnap = 0;
234 * The snapshot creation could fail, but that would cause an
235 * incorrect FALSE return, which would only result in an
236 * overestimation of the amount of space that an operation would
237 * consume, which is OK.
239 * There's also a small window where we could miss a pending
240 * snapshot, because we could set the sync task in the quiescing
241 * phase. So this should only be used as a guess.
243 if (ds->ds_trysnap_txg >
244 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
245 trysnap = ds->ds_trysnap_txg;
246 return (MAX(dsl_dataset_phys(ds)->ds_prev_snap_txg, trysnap));
250 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
253 if (blk_birth <= dsl_dataset_prev_snap_txg(ds) ||
254 (bp != NULL && BP_IS_HOLE(bp)))
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);
270 unique_remove(ds->ds_fsid_guid);
272 if (ds->ds_objset != NULL)
273 dmu_objset_evict(ds->ds_objset);
276 dsl_dataset_rele(ds->ds_prev, ds);
280 bplist_destroy(&ds->ds_pending_deadlist);
281 if (dsl_dataset_phys(ds)->ds_deadlist_obj != 0)
282 dsl_deadlist_close(&ds->ds_deadlist);
284 dsl_dir_rele(ds->ds_dir, ds);
286 ASSERT(!list_link_active(&ds->ds_synced_link));
288 if (mutex_owned(&ds->ds_lock))
289 mutex_exit(&ds->ds_lock);
290 mutex_destroy(&ds->ds_lock);
291 if (mutex_owned(&ds->ds_opening_lock))
292 mutex_exit(&ds->ds_opening_lock);
293 mutex_destroy(&ds->ds_opening_lock);
294 mutex_destroy(&ds->ds_sendstream_lock);
295 refcount_destroy(&ds->ds_longholds);
297 kmem_free(ds, sizeof (dsl_dataset_t));
301 dsl_dataset_get_snapname(dsl_dataset_t *ds)
303 dsl_dataset_phys_t *headphys;
306 dsl_pool_t *dp = ds->ds_dir->dd_pool;
307 objset_t *mos = dp->dp_meta_objset;
309 if (ds->ds_snapname[0])
311 if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
314 err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
318 headphys = headdbuf->db_data;
319 err = zap_value_search(dp->dp_meta_objset,
320 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
321 dmu_buf_rele(headdbuf, FTAG);
326 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
328 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
329 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
333 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
338 err = zap_lookup_norm(mos, snapobj, name, 8, 1,
339 value, mt, NULL, 0, NULL);
340 if (err == ENOTSUP && mt == MT_FIRST)
341 err = zap_lookup(mos, snapobj, name, 8, 1, value);
346 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
349 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
350 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
354 dsl_dir_snap_cmtime_update(ds->ds_dir);
356 if (dsl_dataset_phys(ds)->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);
365 if (err == 0 && adj_cnt)
366 dsl_fs_ss_count_adjust(ds->ds_dir, -1,
367 DD_FIELD_SNAPSHOT_COUNT, tx);
373 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
376 objset_t *mos = dp->dp_meta_objset;
380 dmu_object_info_t doi;
382 ASSERT(dsl_pool_config_held(dp));
384 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
388 /* Make sure dsobj has the correct object type. */
389 dmu_object_info_from_db(dbuf, &doi);
390 if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
391 dmu_buf_rele(dbuf, tag);
392 return (SET_ERROR(EINVAL));
395 ds = dmu_buf_get_user(dbuf);
397 dsl_dataset_t *winner = NULL;
399 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
401 ds->ds_object = dsobj;
403 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
404 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
405 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
406 refcount_create(&ds->ds_longholds);
408 bplist_create(&ds->ds_pending_deadlist);
409 dsl_deadlist_open(&ds->ds_deadlist,
410 mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
412 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
413 offsetof(dmu_sendarg_t, dsa_link));
415 if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
416 int zaperr = zap_contains(mos, dsobj, DS_FIELD_LARGE_BLOCKS);
417 if (zaperr != ENOENT) {
419 ds->ds_large_blocks = B_TRUE;
424 err = dsl_dir_hold_obj(dp,
425 dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds,
429 mutex_destroy(&ds->ds_lock);
430 mutex_destroy(&ds->ds_opening_lock);
431 mutex_destroy(&ds->ds_sendstream_lock);
432 refcount_destroy(&ds->ds_longholds);
433 bplist_destroy(&ds->ds_pending_deadlist);
434 dsl_deadlist_close(&ds->ds_deadlist);
435 kmem_free(ds, sizeof (dsl_dataset_t));
436 dmu_buf_rele(dbuf, tag);
440 if (!dsl_dataset_is_snapshot(ds)) {
441 ds->ds_snapname[0] = '\0';
442 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
443 err = dsl_dataset_hold_obj(dp,
444 dsl_dataset_phys(ds)->ds_prev_snap_obj,
447 if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
448 int zaperr = zap_lookup(mos, ds->ds_object,
449 DS_FIELD_BOOKMARK_NAMES,
450 sizeof (ds->ds_bookmarks), 1,
452 if (zaperr != ENOENT)
456 if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
457 err = dsl_dataset_get_snapname(ds);
459 dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
461 ds->ds_dir->dd_pool->dp_meta_objset,
462 dsl_dataset_phys(ds)->ds_userrefs_obj,
467 if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
468 err = dsl_prop_get_int_ds(ds,
469 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
472 err = dsl_prop_get_int_ds(ds,
473 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
477 ds->ds_reserved = ds->ds_quota = 0;
480 if (err != 0 || (winner = dmu_buf_set_user_ie(dbuf, ds,
481 dsl_dataset_evict)) != NULL) {
482 bplist_destroy(&ds->ds_pending_deadlist);
483 dsl_deadlist_close(&ds->ds_deadlist);
485 dsl_dataset_rele(ds->ds_prev, ds);
486 dsl_dir_rele(ds->ds_dir, ds);
487 mutex_destroy(&ds->ds_lock);
488 mutex_destroy(&ds->ds_opening_lock);
489 mutex_destroy(&ds->ds_sendstream_lock);
490 refcount_destroy(&ds->ds_longholds);
491 kmem_free(ds, sizeof (dsl_dataset_t));
493 dmu_buf_rele(dbuf, tag);
499 unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
502 ASSERT3P(ds->ds_dbuf, ==, dbuf);
503 ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
504 ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
505 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
506 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
512 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
513 void *tag, dsl_dataset_t **dsp)
516 const char *snapname;
520 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
524 ASSERT(dsl_pool_config_held(dp));
525 obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
527 err = dsl_dataset_hold_obj(dp, obj, tag, dsp);
529 err = SET_ERROR(ENOENT);
531 /* we may be looking for a snapshot */
532 if (err == 0 && snapname != NULL) {
535 if (*snapname++ != '@') {
536 dsl_dataset_rele(*dsp, tag);
537 dsl_dir_rele(dd, FTAG);
538 return (SET_ERROR(ENOENT));
541 dprintf("looking for snapshot '%s'\n", snapname);
542 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
544 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
545 dsl_dataset_rele(*dsp, tag);
548 mutex_enter(&ds->ds_lock);
549 if (ds->ds_snapname[0] == 0)
550 (void) strlcpy(ds->ds_snapname, snapname,
551 sizeof (ds->ds_snapname));
552 mutex_exit(&ds->ds_lock);
557 dsl_dir_rele(dd, FTAG);
562 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
563 void *tag, dsl_dataset_t **dsp)
565 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
568 if (!dsl_dataset_tryown(*dsp, tag)) {
569 dsl_dataset_rele(*dsp, tag);
571 return (SET_ERROR(EBUSY));
577 dsl_dataset_own(dsl_pool_t *dp, const char *name,
578 void *tag, dsl_dataset_t **dsp)
580 int err = dsl_dataset_hold(dp, name, tag, dsp);
583 if (!dsl_dataset_tryown(*dsp, tag)) {
584 dsl_dataset_rele(*dsp, tag);
585 return (SET_ERROR(EBUSY));
591 * See the comment above dsl_pool_hold() for details. In summary, a long
592 * hold is used to prevent destruction of a dataset while the pool hold
593 * is dropped, allowing other concurrent operations (e.g. spa_sync()).
595 * The dataset and pool must be held when this function is called. After it
596 * is called, the pool hold may be released while the dataset is still held
600 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
602 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
603 (void) refcount_add(&ds->ds_longholds, tag);
607 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
609 (void) refcount_remove(&ds->ds_longholds, tag);
612 /* Return B_TRUE if there are any long holds on this dataset. */
614 dsl_dataset_long_held(dsl_dataset_t *ds)
616 return (!refcount_is_zero(&ds->ds_longholds));
620 dsl_dataset_name(dsl_dataset_t *ds, char *name)
623 (void) strcpy(name, "mos");
625 dsl_dir_name(ds->ds_dir, name);
626 VERIFY0(dsl_dataset_get_snapname(ds));
627 if (ds->ds_snapname[0]) {
628 (void) strcat(name, "@");
630 * We use a "recursive" mutex so that we
631 * can call dprintf_ds() with ds_lock held.
633 if (!MUTEX_HELD(&ds->ds_lock)) {
634 mutex_enter(&ds->ds_lock);
635 (void) strcat(name, ds->ds_snapname);
636 mutex_exit(&ds->ds_lock);
638 (void) strcat(name, ds->ds_snapname);
645 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
647 dmu_buf_rele(ds->ds_dbuf, tag);
651 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
653 ASSERT3P(ds->ds_owner, ==, tag);
654 ASSERT(ds->ds_dbuf != NULL);
656 mutex_enter(&ds->ds_lock);
658 mutex_exit(&ds->ds_lock);
659 dsl_dataset_long_rele(ds, tag);
660 dsl_dataset_rele(ds, tag);
664 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
666 boolean_t gotit = FALSE;
668 mutex_enter(&ds->ds_lock);
669 if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
671 dsl_dataset_long_hold(ds, tag);
674 mutex_exit(&ds->ds_lock);
679 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
680 uint64_t flags, dmu_tx_t *tx)
682 dsl_pool_t *dp = dd->dd_pool;
684 dsl_dataset_phys_t *dsphys;
686 objset_t *mos = dp->dp_meta_objset;
689 origin = dp->dp_origin_snap;
691 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
692 ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
693 ASSERT(dmu_tx_is_syncing(tx));
694 ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
696 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
697 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
698 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
699 dmu_buf_will_dirty(dbuf, tx);
700 dsphys = dbuf->db_data;
701 bzero(dsphys, sizeof (dsl_dataset_phys_t));
702 dsphys->ds_dir_obj = dd->dd_object;
703 dsphys->ds_flags = flags;
704 dsphys->ds_fsid_guid = unique_create();
706 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
707 sizeof (dsphys->ds_guid));
708 } while (dsphys->ds_guid == 0);
709 dsphys->ds_snapnames_zapobj =
710 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
712 dsphys->ds_creation_time = gethrestime_sec();
713 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
715 if (origin == NULL) {
716 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
718 dsl_dataset_t *ohds; /* head of the origin snapshot */
720 dsphys->ds_prev_snap_obj = origin->ds_object;
721 dsphys->ds_prev_snap_txg =
722 dsl_dataset_phys(origin)->ds_creation_txg;
723 dsphys->ds_referenced_bytes =
724 dsl_dataset_phys(origin)->ds_referenced_bytes;
725 dsphys->ds_compressed_bytes =
726 dsl_dataset_phys(origin)->ds_compressed_bytes;
727 dsphys->ds_uncompressed_bytes =
728 dsl_dataset_phys(origin)->ds_uncompressed_bytes;
729 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
732 * Inherit flags that describe the dataset's contents
733 * (INCONSISTENT) or properties (Case Insensitive).
735 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
736 (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
738 if (origin->ds_large_blocks)
739 dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx);
741 dmu_buf_will_dirty(origin->ds_dbuf, tx);
742 dsl_dataset_phys(origin)->ds_num_children++;
744 VERIFY0(dsl_dataset_hold_obj(dp,
745 dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
747 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
748 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
749 dsl_dataset_rele(ohds, FTAG);
751 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
752 if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
753 dsl_dataset_phys(origin)->ds_next_clones_obj =
755 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
757 VERIFY0(zap_add_int(mos,
758 dsl_dataset_phys(origin)->ds_next_clones_obj,
762 dmu_buf_will_dirty(dd->dd_dbuf, tx);
763 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
764 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
765 if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
766 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
767 dsl_dir_phys(origin->ds_dir)->dd_clones =
769 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
771 VERIFY0(zap_add_int(mos,
772 dsl_dir_phys(origin->ds_dir)->dd_clones,
777 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
778 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
780 dmu_buf_rele(dbuf, FTAG);
782 dmu_buf_will_dirty(dd->dd_dbuf, tx);
783 dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
789 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
793 VERIFY0(dmu_objset_from_ds(ds, &os));
794 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
795 dsl_dataset_dirty(ds, tx);
799 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
800 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
802 dsl_pool_t *dp = pdd->dd_pool;
803 uint64_t dsobj, ddobj;
806 ASSERT(dmu_tx_is_syncing(tx));
807 ASSERT(lastname[0] != '@');
809 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
810 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
812 dsobj = dsl_dataset_create_sync_dd(dd, origin,
813 flags & ~DS_CREATE_FLAG_NODIRTY, tx);
815 dsl_deleg_set_create_perms(dd, tx, cr);
818 * Since we're creating a new node we know it's a leaf, so we can
819 * initialize the counts if the limit feature is active.
821 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
823 objset_t *os = dd->dd_pool->dp_meta_objset;
825 dsl_dir_zapify(dd, tx);
826 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
827 sizeof (cnt), 1, &cnt, tx));
828 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
829 sizeof (cnt), 1, &cnt, tx));
832 dsl_dir_rele(dd, FTAG);
835 * If we are creating a clone, make sure we zero out any stale
836 * data from the origin snapshots zil header.
838 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
841 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
842 dsl_dataset_zero_zil(ds, tx);
843 dsl_dataset_rele(ds, FTAG);
850 /* FreeBSD ioctl compat begin */
853 const char *snapname;
857 dsl_check_snap_cb(const char *name, void *arg)
859 struct destroyarg *da = arg;
863 dsname = kmem_asprintf("%s@%s", name, da->snapname);
864 fnvlist_add_boolean(da->nvl, dsname);
865 kmem_free(dsname, strlen(dsname) + 1);
871 dmu_get_recursive_snaps_nvl(char *fsname, const char *snapname,
874 struct destroyarg *da;
877 da = kmem_zalloc(sizeof (struct destroyarg), KM_SLEEP);
879 da->snapname = snapname;
880 err = dmu_objset_find(fsname, dsl_check_snap_cb, da,
882 kmem_free(da, sizeof (struct destroyarg));
886 /* FreeBSD ioctl compat end */
887 #endif /* __FreeBSD__ */
890 * The unique space in the head dataset can be calculated by subtracting
891 * the space used in the most recent snapshot, that is still being used
892 * in this file system, from the space currently in use. To figure out
893 * the space in the most recent snapshot still in use, we need to take
894 * the total space used in the snapshot and subtract out the space that
895 * has been freed up since the snapshot was taken.
898 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
901 uint64_t dlused, dlcomp, dluncomp;
903 ASSERT(!dsl_dataset_is_snapshot(ds));
905 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
906 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
910 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
912 ASSERT3U(dlused, <=, mrs_used);
913 dsl_dataset_phys(ds)->ds_unique_bytes =
914 dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
916 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
917 SPA_VERSION_UNIQUE_ACCURATE)
918 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
922 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
925 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
929 ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
930 err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
933 * The err should not be ENOENT, but a bug in a previous version
934 * of the code could cause upgrade_clones_cb() to not set
935 * ds_next_snap_obj when it should, leading to a missing entry.
936 * If we knew that the pool was created after
937 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
938 * ENOENT. However, at least we can check that we don't have
939 * too many entries in the next_clones_obj even after failing to
944 ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
946 ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
951 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
953 return (&dsl_dataset_phys(ds)->ds_bp);
957 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
959 ASSERT(dmu_tx_is_syncing(tx));
960 /* If it's the meta-objset, set dp_meta_rootbp */
962 tx->tx_pool->dp_meta_rootbp = *bp;
964 dmu_buf_will_dirty(ds->ds_dbuf, tx);
965 dsl_dataset_phys(ds)->ds_bp = *bp;
970 dsl_dataset_get_spa(dsl_dataset_t *ds)
972 return (ds->ds_dir->dd_pool->dp_spa);
976 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
980 if (ds == NULL) /* this is the meta-objset */
983 ASSERT(ds->ds_objset != NULL);
985 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
986 panic("dirtying snapshot!");
988 dp = ds->ds_dir->dd_pool;
990 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
991 /* up the hold count until we can be written out */
992 dmu_buf_add_ref(ds->ds_dbuf, ds);
997 dsl_dataset_is_dirty(dsl_dataset_t *ds)
999 for (int t = 0; t < TXG_SIZE; t++) {
1000 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1008 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1012 if (!dmu_tx_is_syncing(tx))
1016 * If there's an fs-only reservation, any blocks that might become
1017 * owned by the snapshot dataset must be accommodated by space
1018 * outside of the reservation.
1020 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1021 asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
1022 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1023 return (SET_ERROR(ENOSPC));
1026 * Propagate any reserved space for this snapshot to other
1027 * snapshot checks in this sync group.
1030 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1035 typedef struct dsl_dataset_snapshot_arg {
1036 nvlist_t *ddsa_snaps;
1037 nvlist_t *ddsa_props;
1038 nvlist_t *ddsa_errors;
1040 } dsl_dataset_snapshot_arg_t;
1043 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1044 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
1049 ds->ds_trysnap_txg = tx->tx_txg;
1051 if (!dmu_tx_is_syncing(tx))
1055 * We don't allow multiple snapshots of the same txg. If there
1056 * is already one, try again.
1058 if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1059 return (SET_ERROR(EAGAIN));
1062 * Check for conflicting snapshot name.
1064 error = dsl_dataset_snap_lookup(ds, snapname, &value);
1066 return (SET_ERROR(EEXIST));
1067 if (error != ENOENT)
1071 * We don't allow taking snapshots of inconsistent datasets, such as
1072 * those into which we are currently receiving. However, if we are
1073 * creating this snapshot as part of a receive, this check will be
1074 * executed atomically with respect to the completion of the receive
1075 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1076 * case we ignore this, knowing it will be fixed up for us shortly in
1077 * dmu_recv_end_sync().
1079 if (!recv && DS_IS_INCONSISTENT(ds))
1080 return (SET_ERROR(EBUSY));
1083 * Skip the check for temporary snapshots or if we have already checked
1084 * the counts in dsl_dataset_snapshot_check. This means we really only
1085 * check the count here when we're receiving a stream.
1087 if (cnt != 0 && cr != NULL) {
1088 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1089 ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1094 error = dsl_dataset_snapshot_reserve_space(ds, tx);
1102 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1104 dsl_dataset_snapshot_arg_t *ddsa = arg;
1105 dsl_pool_t *dp = dmu_tx_pool(tx);
1110 * Pre-compute how many total new snapshots will be created for each
1111 * level in the tree and below. This is needed for validating the
1112 * snapshot limit when either taking a recursive snapshot or when
1113 * taking multiple snapshots.
1115 * The problem is that the counts are not actually adjusted when
1116 * we are checking, only when we finally sync. For a single snapshot,
1117 * this is easy, the count will increase by 1 at each node up the tree,
1118 * but its more complicated for the recursive/multiple snapshot case.
1120 * The dsl_fs_ss_limit_check function does recursively check the count
1121 * at each level up the tree but since it is validating each snapshot
1122 * independently we need to be sure that we are validating the complete
1123 * count for the entire set of snapshots. We do this by rolling up the
1124 * counts for each component of the name into an nvlist and then
1125 * checking each of those cases with the aggregated count.
1127 * This approach properly handles not only the recursive snapshot
1128 * case (where we get all of those on the ddsa_snaps list) but also
1129 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1130 * validate the limit on 'a' using a count of 2).
1132 * We validate the snapshot names in the third loop and only report
1135 if (dmu_tx_is_syncing(tx)) {
1136 nvlist_t *cnt_track = NULL;
1137 cnt_track = fnvlist_alloc();
1139 /* Rollup aggregated counts into the cnt_track list */
1140 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1142 pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1145 char nm[MAXPATHLEN];
1147 (void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1148 pdelim = strchr(nm, '@');
1154 if (nvlist_lookup_uint64(cnt_track, nm,
1156 /* update existing entry */
1157 fnvlist_add_uint64(cnt_track, nm,
1161 fnvlist_add_uint64(cnt_track, nm, 1);
1164 pdelim = strrchr(nm, '/');
1167 } while (pdelim != NULL);
1170 /* Check aggregated counts at each level */
1171 for (pair = nvlist_next_nvpair(cnt_track, NULL);
1172 pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1178 name = nvpair_name(pair);
1179 cnt = fnvpair_value_uint64(pair);
1182 error = dsl_dataset_hold(dp, name, FTAG, &ds);
1184 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1185 ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1187 dsl_dataset_rele(ds, FTAG);
1191 if (ddsa->ddsa_errors != NULL)
1192 fnvlist_add_int32(ddsa->ddsa_errors,
1195 /* only report one error for this check */
1199 nvlist_free(cnt_track);
1202 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1203 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1207 char dsname[MAXNAMELEN];
1209 name = nvpair_name(pair);
1210 if (strlen(name) >= MAXNAMELEN)
1211 error = SET_ERROR(ENAMETOOLONG);
1213 atp = strchr(name, '@');
1215 error = SET_ERROR(EINVAL);
1217 (void) strlcpy(dsname, name, atp - name + 1);
1220 error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1222 /* passing 0/NULL skips dsl_fs_ss_limit_check */
1223 error = dsl_dataset_snapshot_check_impl(ds,
1224 atp + 1, tx, B_FALSE, 0, NULL);
1225 dsl_dataset_rele(ds, FTAG);
1229 if (ddsa->ddsa_errors != NULL) {
1230 fnvlist_add_int32(ddsa->ddsa_errors,
1241 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1244 static zil_header_t zero_zil;
1246 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1248 dsl_dataset_phys_t *dsphys;
1249 uint64_t dsobj, crtxg;
1250 objset_t *mos = dp->dp_meta_objset;
1253 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1256 * If we are on an old pool, the zil must not be active, in which
1257 * case it will be zeroed. Usually zil_suspend() accomplishes this.
1259 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1260 dmu_objset_from_ds(ds, &os) != 0 ||
1261 bcmp(&os->os_phys->os_zil_header, &zero_zil,
1262 sizeof (zero_zil)) == 0);
1264 dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1267 * The origin's ds_creation_txg has to be < TXG_INITIAL
1269 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1274 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1275 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1276 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1277 dmu_buf_will_dirty(dbuf, tx);
1278 dsphys = dbuf->db_data;
1279 bzero(dsphys, sizeof (dsl_dataset_phys_t));
1280 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1281 dsphys->ds_fsid_guid = unique_create();
1283 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1284 sizeof (dsphys->ds_guid));
1285 } while (dsphys->ds_guid == 0);
1286 dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1287 dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1288 dsphys->ds_next_snap_obj = ds->ds_object;
1289 dsphys->ds_num_children = 1;
1290 dsphys->ds_creation_time = gethrestime_sec();
1291 dsphys->ds_creation_txg = crtxg;
1292 dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1293 dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1294 dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1295 dsphys->ds_uncompressed_bytes =
1296 dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1297 dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1298 dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1299 dmu_buf_rele(dbuf, FTAG);
1301 if (ds->ds_large_blocks)
1302 dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx);
1304 ASSERT3U(ds->ds_prev != 0, ==,
1305 dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1307 uint64_t next_clones_obj =
1308 dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1309 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1311 dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1312 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1314 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1315 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1316 dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1317 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1318 } else if (next_clones_obj != 0) {
1319 dsl_dataset_remove_from_next_clones(ds->ds_prev,
1320 dsphys->ds_next_snap_obj, tx);
1321 VERIFY0(zap_add_int(mos,
1322 next_clones_obj, dsobj, tx));
1327 * If we have a reference-reservation on this dataset, we will
1328 * need to increase the amount of refreservation being charged
1329 * since our unique space is going to zero.
1331 if (ds->ds_reserved) {
1333 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1334 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1336 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1340 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1341 dsl_dataset_phys(ds)->ds_deadlist_obj =
1342 dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1343 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1344 dsl_deadlist_close(&ds->ds_deadlist);
1345 dsl_deadlist_open(&ds->ds_deadlist, mos,
1346 dsl_dataset_phys(ds)->ds_deadlist_obj);
1347 dsl_deadlist_add_key(&ds->ds_deadlist,
1348 dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1350 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1351 dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1352 dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1353 dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1354 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1355 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1357 VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1358 snapname, 8, 1, &dsobj, tx));
1361 dsl_dataset_rele(ds->ds_prev, ds);
1362 VERIFY0(dsl_dataset_hold_obj(dp,
1363 dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1365 dsl_scan_ds_snapshotted(ds, tx);
1367 dsl_dir_snap_cmtime_update(ds->ds_dir);
1369 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1373 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1375 dsl_dataset_snapshot_arg_t *ddsa = arg;
1376 dsl_pool_t *dp = dmu_tx_pool(tx);
1379 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1380 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1383 char dsname[MAXNAMELEN];
1385 name = nvpair_name(pair);
1386 atp = strchr(name, '@');
1387 (void) strlcpy(dsname, name, atp - name + 1);
1388 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1390 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1391 if (ddsa->ddsa_props != NULL) {
1392 dsl_props_set_sync_impl(ds->ds_prev,
1393 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1395 dsl_dataset_rele(ds, FTAG);
1400 * The snapshots must all be in the same pool.
1401 * All-or-nothing: if there are any failures, nothing will be modified.
1404 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1406 dsl_dataset_snapshot_arg_t ddsa;
1408 boolean_t needsuspend;
1412 nvlist_t *suspended = NULL;
1414 pair = nvlist_next_nvpair(snaps, NULL);
1417 firstname = nvpair_name(pair);
1419 error = spa_open(firstname, &spa, FTAG);
1422 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1423 spa_close(spa, FTAG);
1426 suspended = fnvlist_alloc();
1427 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1428 pair = nvlist_next_nvpair(snaps, pair)) {
1429 char fsname[MAXNAMELEN];
1430 char *snapname = nvpair_name(pair);
1434 atp = strchr(snapname, '@');
1436 error = SET_ERROR(EINVAL);
1439 (void) strlcpy(fsname, snapname, atp - snapname + 1);
1441 error = zil_suspend(fsname, &cookie);
1444 fnvlist_add_uint64(suspended, fsname,
1449 ddsa.ddsa_snaps = snaps;
1450 ddsa.ddsa_props = props;
1451 ddsa.ddsa_errors = errors;
1452 ddsa.ddsa_cr = CRED();
1455 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1456 dsl_dataset_snapshot_sync, &ddsa,
1457 fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1460 if (suspended != NULL) {
1461 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1462 pair = nvlist_next_nvpair(suspended, pair)) {
1463 zil_resume((void *)(uintptr_t)
1464 fnvpair_value_uint64(pair));
1466 fnvlist_free(suspended);
1472 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1473 pair = nvlist_next_nvpair(snaps, pair)) {
1474 char *snapname = nvpair_name(pair);
1475 zvol_create_minors(snapname);
1483 typedef struct dsl_dataset_snapshot_tmp_arg {
1484 const char *ddsta_fsname;
1485 const char *ddsta_snapname;
1486 minor_t ddsta_cleanup_minor;
1487 const char *ddsta_htag;
1488 } dsl_dataset_snapshot_tmp_arg_t;
1491 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1493 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1494 dsl_pool_t *dp = dmu_tx_pool(tx);
1498 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1502 /* NULL cred means no limit check for tmp snapshot */
1503 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1504 tx, B_FALSE, 0, NULL);
1506 dsl_dataset_rele(ds, FTAG);
1510 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1511 dsl_dataset_rele(ds, FTAG);
1512 return (SET_ERROR(ENOTSUP));
1514 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1517 dsl_dataset_rele(ds, FTAG);
1521 dsl_dataset_rele(ds, FTAG);
1526 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1528 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1529 dsl_pool_t *dp = dmu_tx_pool(tx);
1532 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1534 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1535 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1536 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1537 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1539 dsl_dataset_rele(ds, FTAG);
1543 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1544 minor_t cleanup_minor, const char *htag)
1546 dsl_dataset_snapshot_tmp_arg_t ddsta;
1549 boolean_t needsuspend;
1552 ddsta.ddsta_fsname = fsname;
1553 ddsta.ddsta_snapname = snapname;
1554 ddsta.ddsta_cleanup_minor = cleanup_minor;
1555 ddsta.ddsta_htag = htag;
1557 error = spa_open(fsname, &spa, FTAG);
1560 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1561 spa_close(spa, FTAG);
1564 error = zil_suspend(fsname, &cookie);
1569 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1570 dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1579 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1581 ASSERT(dmu_tx_is_syncing(tx));
1582 ASSERT(ds->ds_objset != NULL);
1583 ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1586 * in case we had to change ds_fsid_guid when we opened it,
1589 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1590 dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1592 dmu_objset_sync(ds->ds_objset, zio, tx);
1594 if (ds->ds_need_large_blocks && !ds->ds_large_blocks) {
1595 dsl_dataset_activate_large_blocks_sync_impl(ds->ds_object, tx);
1596 ds->ds_large_blocks = B_TRUE;
1601 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1604 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1607 nvlist_t *propval = fnvlist_alloc();
1608 nvlist_t *val = fnvlist_alloc();
1610 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1613 * There may be missing entries in ds_next_clones_obj
1614 * due to a bug in a previous version of the code.
1615 * Only trust it if it has the right number of entries.
1617 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1618 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1621 if (count != dsl_dataset_phys(ds)->ds_num_children - 1)
1623 for (zap_cursor_init(&zc, mos,
1624 dsl_dataset_phys(ds)->ds_next_clones_obj);
1625 zap_cursor_retrieve(&zc, &za) == 0;
1626 zap_cursor_advance(&zc)) {
1627 dsl_dataset_t *clone;
1628 char buf[ZFS_MAXNAMELEN];
1629 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1630 za.za_first_integer, FTAG, &clone));
1631 dsl_dir_name(clone->ds_dir, buf);
1632 fnvlist_add_boolean(val, buf);
1633 dsl_dataset_rele(clone, FTAG);
1635 zap_cursor_fini(&zc);
1636 fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1637 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1640 nvlist_free(propval);
1644 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1646 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1647 uint64_t refd, avail, uobjs, aobjs, ratio;
1649 ASSERT(dsl_pool_config_held(dp));
1651 ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
1652 (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
1653 dsl_dataset_phys(ds)->ds_compressed_bytes);
1655 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1656 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1657 dsl_dataset_phys(ds)->ds_uncompressed_bytes);
1659 if (dsl_dataset_is_snapshot(ds)) {
1660 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1661 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1662 dsl_dataset_phys(ds)->ds_unique_bytes);
1663 get_clones_stat(ds, nv);
1665 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
1666 char buf[MAXNAMELEN];
1667 dsl_dataset_name(ds->ds_prev, buf);
1668 dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf);
1671 dsl_dir_stats(ds->ds_dir, nv);
1674 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1675 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1676 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1678 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1679 dsl_dataset_phys(ds)->ds_creation_time);
1680 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1681 dsl_dataset_phys(ds)->ds_creation_txg);
1682 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1684 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1686 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1687 dsl_dataset_phys(ds)->ds_guid);
1688 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1689 dsl_dataset_phys(ds)->ds_unique_bytes);
1690 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1692 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1694 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1695 DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1697 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1698 uint64_t written, comp, uncomp;
1699 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1700 dsl_dataset_t *prev;
1702 int err = dsl_dataset_hold_obj(dp,
1703 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1705 err = dsl_dataset_space_written(prev, ds, &written,
1707 dsl_dataset_rele(prev, FTAG);
1709 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1717 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1719 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1720 ASSERT(dsl_pool_config_held(dp));
1722 stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg;
1723 stat->dds_inconsistent =
1724 dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT;
1725 stat->dds_guid = dsl_dataset_phys(ds)->ds_guid;
1726 stat->dds_origin[0] = '\0';
1727 if (dsl_dataset_is_snapshot(ds)) {
1728 stat->dds_is_snapshot = B_TRUE;
1729 stat->dds_num_clones =
1730 dsl_dataset_phys(ds)->ds_num_children - 1;
1732 stat->dds_is_snapshot = B_FALSE;
1733 stat->dds_num_clones = 0;
1735 if (dsl_dir_is_clone(ds->ds_dir)) {
1738 VERIFY0(dsl_dataset_hold_obj(dp,
1739 dsl_dir_phys(ds->ds_dir)->dd_origin_obj,
1741 dsl_dataset_name(ods, stat->dds_origin);
1742 dsl_dataset_rele(ods, FTAG);
1748 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1750 return (ds->ds_fsid_guid);
1754 dsl_dataset_space(dsl_dataset_t *ds,
1755 uint64_t *refdbytesp, uint64_t *availbytesp,
1756 uint64_t *usedobjsp, uint64_t *availobjsp)
1758 *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
1759 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1760 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
1762 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
1763 if (ds->ds_quota != 0) {
1765 * Adjust available bytes according to refquota
1767 if (*refdbytesp < ds->ds_quota)
1768 *availbytesp = MIN(*availbytesp,
1769 ds->ds_quota - *refdbytesp);
1773 *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
1774 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
1778 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
1780 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1782 ASSERT(dsl_pool_config_held(dp));
1785 if (dsl_dataset_phys(ds)->ds_bp.blk_birth >
1786 dsl_dataset_phys(snap)->ds_creation_txg) {
1787 objset_t *os, *os_snap;
1789 * It may be that only the ZIL differs, because it was
1790 * reset in the head. Don't count that as being
1793 if (dmu_objset_from_ds(ds, &os) != 0)
1795 if (dmu_objset_from_ds(snap, &os_snap) != 0)
1797 return (bcmp(&os->os_phys->os_meta_dnode,
1798 &os_snap->os_phys->os_meta_dnode,
1799 sizeof (os->os_phys->os_meta_dnode)) != 0);
1804 typedef struct dsl_dataset_rename_snapshot_arg {
1805 const char *ddrsa_fsname;
1806 const char *ddrsa_oldsnapname;
1807 const char *ddrsa_newsnapname;
1808 boolean_t ddrsa_recursive;
1810 } dsl_dataset_rename_snapshot_arg_t;
1814 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1815 dsl_dataset_t *hds, void *arg)
1817 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1821 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1823 /* ignore nonexistent snapshots */
1824 return (error == ENOENT ? 0 : error);
1827 /* new name should not exist */
1828 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
1830 error = SET_ERROR(EEXIST);
1831 else if (error == ENOENT)
1834 /* dataset name + 1 for the "@" + the new snapshot name must fit */
1835 if (dsl_dir_namelen(hds->ds_dir) + 1 +
1836 strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN)
1837 error = SET_ERROR(ENAMETOOLONG);
1843 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
1845 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1846 dsl_pool_t *dp = dmu_tx_pool(tx);
1850 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
1854 if (ddrsa->ddrsa_recursive) {
1855 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1856 dsl_dataset_rename_snapshot_check_impl, ddrsa,
1859 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
1861 dsl_dataset_rele(hds, FTAG);
1866 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
1867 dsl_dataset_t *hds, void *arg)
1871 char *oldname, *newname;
1874 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1877 dmu_tx_t *tx = ddrsa->ddrsa_tx;
1880 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1881 ASSERT(error == 0 || error == ENOENT);
1882 if (error == ENOENT) {
1883 /* ignore nonexistent snapshots */
1887 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
1889 /* log before we change the name */
1890 spa_history_log_internal_ds(ds, "rename", tx,
1891 "-> @%s", ddrsa->ddrsa_newsnapname);
1893 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
1895 mutex_enter(&ds->ds_lock);
1896 (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
1897 mutex_exit(&ds->ds_lock);
1898 VERIFY0(zap_add(dp->dp_meta_objset,
1899 dsl_dataset_phys(hds)->ds_snapnames_zapobj,
1900 ds->ds_snapname, 8, 1, &ds->ds_object, tx));
1904 oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1905 newname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1906 snprintf(oldname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname,
1907 ddrsa->ddrsa_oldsnapname);
1908 snprintf(newname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname,
1909 ddrsa->ddrsa_newsnapname);
1910 zfsvfs_update_fromname(oldname, newname);
1911 zvol_rename_minors(oldname, newname);
1912 kmem_free(newname, MAXPATHLEN);
1913 kmem_free(oldname, MAXPATHLEN);
1916 dsl_dataset_rele(ds, FTAG);
1922 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
1924 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1925 dsl_pool_t *dp = dmu_tx_pool(tx);
1928 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
1929 ddrsa->ddrsa_tx = tx;
1930 if (ddrsa->ddrsa_recursive) {
1931 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1932 dsl_dataset_rename_snapshot_sync_impl, ddrsa,
1935 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
1937 dsl_dataset_rele(hds, FTAG);
1941 dsl_dataset_rename_snapshot(const char *fsname,
1942 const char *oldsnapname, const char *newsnapname, boolean_t recursive)
1944 dsl_dataset_rename_snapshot_arg_t ddrsa;
1946 ddrsa.ddrsa_fsname = fsname;
1947 ddrsa.ddrsa_oldsnapname = oldsnapname;
1948 ddrsa.ddrsa_newsnapname = newsnapname;
1949 ddrsa.ddrsa_recursive = recursive;
1951 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
1952 dsl_dataset_rename_snapshot_sync, &ddrsa,
1953 1, ZFS_SPACE_CHECK_RESERVED));
1957 * If we're doing an ownership handoff, we need to make sure that there is
1958 * only one long hold on the dataset. We're not allowed to change anything here
1959 * so we don't permanently release the long hold or regular hold here. We want
1960 * to do this only when syncing to avoid the dataset unexpectedly going away
1961 * when we release the long hold.
1964 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
1968 if (!dmu_tx_is_syncing(tx))
1971 if (owner != NULL) {
1972 VERIFY3P(ds->ds_owner, ==, owner);
1973 dsl_dataset_long_rele(ds, owner);
1976 held = dsl_dataset_long_held(ds);
1979 dsl_dataset_long_hold(ds, owner);
1982 return (SET_ERROR(EBUSY));
1987 typedef struct dsl_dataset_rollback_arg {
1988 const char *ddra_fsname;
1990 nvlist_t *ddra_result;
1991 } dsl_dataset_rollback_arg_t;
1994 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
1996 dsl_dataset_rollback_arg_t *ddra = arg;
1997 dsl_pool_t *dp = dmu_tx_pool(tx);
1999 int64_t unused_refres_delta;
2002 error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
2006 /* must not be a snapshot */
2007 if (dsl_dataset_is_snapshot(ds)) {
2008 dsl_dataset_rele(ds, FTAG);
2009 return (SET_ERROR(EINVAL));
2012 /* must have a most recent snapshot */
2013 if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
2014 dsl_dataset_rele(ds, FTAG);
2015 return (SET_ERROR(EINVAL));
2018 /* must not have any bookmarks after the most recent snapshot */
2019 nvlist_t *proprequest = fnvlist_alloc();
2020 fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
2021 nvlist_t *bookmarks = fnvlist_alloc();
2022 error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
2023 fnvlist_free(proprequest);
2026 for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
2027 pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
2029 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
2030 zfs_prop_to_name(ZFS_PROP_CREATETXG));
2031 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
2032 if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
2033 fnvlist_free(bookmarks);
2034 dsl_dataset_rele(ds, FTAG);
2035 return (SET_ERROR(EEXIST));
2038 fnvlist_free(bookmarks);
2040 error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
2042 dsl_dataset_rele(ds, FTAG);
2047 * Check if the snap we are rolling back to uses more than
2050 if (ds->ds_quota != 0 &&
2051 dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
2052 dsl_dataset_rele(ds, FTAG);
2053 return (SET_ERROR(EDQUOT));
2057 * When we do the clone swap, we will temporarily use more space
2058 * due to the refreservation (the head will no longer have any
2059 * unique space, so the entire amount of the refreservation will need
2060 * to be free). We will immediately destroy the clone, freeing
2061 * this space, but the freeing happens over many txg's.
2063 unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2064 dsl_dataset_phys(ds)->ds_unique_bytes);
2066 if (unused_refres_delta > 0 &&
2067 unused_refres_delta >
2068 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2069 dsl_dataset_rele(ds, FTAG);
2070 return (SET_ERROR(ENOSPC));
2073 dsl_dataset_rele(ds, FTAG);
2078 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2080 dsl_dataset_rollback_arg_t *ddra = arg;
2081 dsl_pool_t *dp = dmu_tx_pool(tx);
2082 dsl_dataset_t *ds, *clone;
2084 char namebuf[ZFS_MAXNAMELEN];
2086 VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2088 dsl_dataset_name(ds->ds_prev, namebuf);
2089 fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2091 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2092 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2094 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2096 dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2097 dsl_dataset_zero_zil(ds, tx);
2099 dsl_destroy_head_sync_impl(clone, tx);
2101 dsl_dataset_rele(clone, FTAG);
2102 dsl_dataset_rele(ds, FTAG);
2106 * Rolls back the given filesystem or volume to the most recent snapshot.
2107 * The name of the most recent snapshot will be returned under key "target"
2108 * in the result nvlist.
2111 * - The existing dataset MUST be owned by the specified owner at entry
2112 * - Upon return, dataset will still be held by the same owner, whether we
2115 * This mode is required any time the existing filesystem is mounted. See
2116 * notes above zfs_suspend_fs() for further details.
2119 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
2121 dsl_dataset_rollback_arg_t ddra;
2123 ddra.ddra_fsname = fsname;
2124 ddra.ddra_owner = owner;
2125 ddra.ddra_result = result;
2127 return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2128 dsl_dataset_rollback_sync, &ddra,
2129 1, ZFS_SPACE_CHECK_RESERVED));
2132 struct promotenode {
2137 typedef struct dsl_dataset_promote_arg {
2138 const char *ddpa_clonename;
2139 dsl_dataset_t *ddpa_clone;
2140 list_t shared_snaps, origin_snaps, clone_snaps;
2141 dsl_dataset_t *origin_origin; /* origin of the origin */
2142 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2145 } dsl_dataset_promote_arg_t;
2147 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2148 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2150 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2153 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2155 dsl_dataset_promote_arg_t *ddpa = arg;
2156 dsl_pool_t *dp = dmu_tx_pool(tx);
2158 struct promotenode *snap;
2159 dsl_dataset_t *origin_ds;
2164 err = promote_hold(ddpa, dp, FTAG);
2168 hds = ddpa->ddpa_clone;
2170 if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2171 promote_rele(ddpa, FTAG);
2172 return (SET_ERROR(EXDEV));
2176 * Compute and check the amount of space to transfer. Since this is
2177 * so expensive, don't do the preliminary check.
2179 if (!dmu_tx_is_syncing(tx)) {
2180 promote_rele(ddpa, FTAG);
2184 snap = list_head(&ddpa->shared_snaps);
2185 origin_ds = snap->ds;
2187 /* compute origin's new unique space */
2188 snap = list_tail(&ddpa->clone_snaps);
2189 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2190 origin_ds->ds_object);
2191 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2192 dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2193 &ddpa->unique, &unused, &unused);
2196 * Walk the snapshots that we are moving
2198 * Compute space to transfer. Consider the incremental changes
2199 * to used by each snapshot:
2200 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2201 * So each snapshot gave birth to:
2202 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2203 * So a sequence would look like:
2204 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2205 * Which simplifies to:
2206 * uN + kN + kN-1 + ... + k1 + k0
2207 * Note however, if we stop before we reach the ORIGIN we get:
2208 * uN + kN + kN-1 + ... + kM - uM-1
2211 ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2212 ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2213 ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2214 for (snap = list_head(&ddpa->shared_snaps); snap;
2215 snap = list_next(&ddpa->shared_snaps, snap)) {
2216 uint64_t val, dlused, dlcomp, dluncomp;
2217 dsl_dataset_t *ds = snap->ds;
2222 * If there are long holds, we won't be able to evict
2225 if (dsl_dataset_long_held(ds)) {
2226 err = SET_ERROR(EBUSY);
2230 /* Check that the snapshot name does not conflict */
2231 VERIFY0(dsl_dataset_get_snapname(ds));
2232 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2234 (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2235 err = SET_ERROR(EEXIST);
2241 /* The very first snapshot does not have a deadlist */
2242 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2245 dsl_deadlist_space(&ds->ds_deadlist,
2246 &dlused, &dlcomp, &dluncomp);
2247 ddpa->used += dlused;
2248 ddpa->comp += dlcomp;
2249 ddpa->uncomp += dluncomp;
2253 * If we are a clone of a clone then we never reached ORIGIN,
2254 * so we need to subtract out the clone origin's used space.
2256 if (ddpa->origin_origin) {
2258 dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2260 dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2262 dsl_dataset_phys(ddpa->origin_origin)->
2263 ds_uncompressed_bytes;
2266 /* Check that there is enough space and limit headroom here */
2267 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2268 0, ss_mv_cnt, ddpa->used, ddpa->cr);
2273 * Compute the amounts of space that will be used by snapshots
2274 * after the promotion (for both origin and clone). For each,
2275 * it is the amount of space that will be on all of their
2276 * deadlists (that was not born before their new origin).
2278 if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2282 * Note, typically this will not be a clone of a clone,
2283 * so dd_origin_txg will be < TXG_INITIAL, so
2284 * these snaplist_space() -> dsl_deadlist_space_range()
2285 * calls will be fast because they do not have to
2286 * iterate over all bps.
2288 snap = list_head(&ddpa->origin_snaps);
2289 err = snaplist_space(&ddpa->shared_snaps,
2290 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2294 err = snaplist_space(&ddpa->clone_snaps,
2295 snap->ds->ds_dir->dd_origin_txg, &space);
2298 ddpa->cloneusedsnap += space;
2300 if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2301 DD_FLAG_USED_BREAKDOWN) {
2302 err = snaplist_space(&ddpa->origin_snaps,
2303 dsl_dataset_phys(origin_ds)->ds_creation_txg,
2304 &ddpa->originusedsnap);
2310 promote_rele(ddpa, FTAG);
2315 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2317 dsl_dataset_promote_arg_t *ddpa = arg;
2318 dsl_pool_t *dp = dmu_tx_pool(tx);
2320 struct promotenode *snap;
2321 dsl_dataset_t *origin_ds;
2322 dsl_dataset_t *origin_head;
2324 dsl_dir_t *odd = NULL;
2325 uint64_t oldnext_obj;
2327 #if defined(__FreeBSD__) && defined(_KERNEL)
2328 char *oldname, *newname;
2331 VERIFY0(promote_hold(ddpa, dp, FTAG));
2332 hds = ddpa->ddpa_clone;
2334 ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
2336 snap = list_head(&ddpa->shared_snaps);
2337 origin_ds = snap->ds;
2340 snap = list_head(&ddpa->origin_snaps);
2341 origin_head = snap->ds;
2344 * We need to explicitly open odd, since origin_ds's dd will be
2347 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2350 /* change origin's next snap */
2351 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2352 oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
2353 snap = list_tail(&ddpa->clone_snaps);
2354 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2355 origin_ds->ds_object);
2356 dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
2358 /* change the origin's next clone */
2359 if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
2360 dsl_dataset_remove_from_next_clones(origin_ds,
2361 snap->ds->ds_object, tx);
2362 VERIFY0(zap_add_int(dp->dp_meta_objset,
2363 dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
2368 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2369 ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
2370 dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
2371 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2372 dmu_buf_will_dirty(odd->dd_dbuf, tx);
2373 dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
2374 origin_head->ds_dir->dd_origin_txg =
2375 dsl_dataset_phys(origin_ds)->ds_creation_txg;
2377 /* change dd_clone entries */
2378 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2379 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2380 dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
2381 VERIFY0(zap_add_int(dp->dp_meta_objset,
2382 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2383 hds->ds_object, tx));
2385 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2386 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2387 origin_head->ds_object, tx));
2388 if (dsl_dir_phys(dd)->dd_clones == 0) {
2389 dsl_dir_phys(dd)->dd_clones =
2390 zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
2391 DMU_OT_NONE, 0, tx);
2393 VERIFY0(zap_add_int(dp->dp_meta_objset,
2394 dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
2397 #if defined(__FreeBSD__) && defined(_KERNEL)
2398 /* Take the spa_namespace_lock early so zvol renames don't deadlock. */
2399 mutex_enter(&spa_namespace_lock);
2401 oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2402 newname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2405 /* move snapshots to this dir */
2406 for (snap = list_head(&ddpa->shared_snaps); snap;
2407 snap = list_next(&ddpa->shared_snaps, snap)) {
2408 dsl_dataset_t *ds = snap->ds;
2411 * Property callbacks are registered to a particular
2412 * dsl_dir. Since ours is changing, evict the objset
2413 * so that they will be unregistered from the old dsl_dir.
2415 if (ds->ds_objset) {
2416 dmu_objset_evict(ds->ds_objset);
2417 ds->ds_objset = NULL;
2420 /* move snap name entry */
2421 VERIFY0(dsl_dataset_get_snapname(ds));
2422 VERIFY0(dsl_dataset_snap_remove(origin_head,
2423 ds->ds_snapname, tx, B_TRUE));
2424 VERIFY0(zap_add(dp->dp_meta_objset,
2425 dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
2426 8, 1, &ds->ds_object, tx));
2427 dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2428 DD_FIELD_SNAPSHOT_COUNT, tx);
2430 /* change containing dsl_dir */
2431 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2432 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
2433 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
2434 ASSERT3P(ds->ds_dir, ==, odd);
2435 dsl_dir_rele(ds->ds_dir, ds);
2436 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2437 NULL, ds, &ds->ds_dir));
2439 #if defined(__FreeBSD__) && defined(_KERNEL)
2440 dsl_dataset_name(ds, newname);
2441 zfsvfs_update_fromname(oldname, newname);
2442 zvol_rename_minors(oldname, newname);
2445 /* move any clone references */
2446 if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
2447 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2451 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2452 dsl_dataset_phys(ds)->ds_next_clones_obj);
2453 zap_cursor_retrieve(&zc, &za) == 0;
2454 zap_cursor_advance(&zc)) {
2455 dsl_dataset_t *cnds;
2458 if (za.za_first_integer == oldnext_obj) {
2460 * We've already moved the
2461 * origin's reference.
2466 VERIFY0(dsl_dataset_hold_obj(dp,
2467 za.za_first_integer, FTAG, &cnds));
2468 o = dsl_dir_phys(cnds->ds_dir)->
2469 dd_head_dataset_obj;
2471 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2472 dsl_dir_phys(odd)->dd_clones, o, tx));
2473 VERIFY0(zap_add_int(dp->dp_meta_objset,
2474 dsl_dir_phys(dd)->dd_clones, o, tx));
2475 dsl_dataset_rele(cnds, FTAG);
2477 zap_cursor_fini(&zc);
2480 ASSERT(!dsl_prop_hascb(ds));
2483 #if defined(__FreeBSD__) && defined(_KERNEL)
2484 mutex_exit(&spa_namespace_lock);
2486 kmem_free(newname, MAXPATHLEN);
2487 kmem_free(oldname, MAXPATHLEN);
2490 * Change space accounting.
2491 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2492 * both be valid, or both be 0 (resulting in delta == 0). This
2493 * is true for each of {clone,origin} independently.
2496 delta = ddpa->cloneusedsnap -
2497 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
2498 ASSERT3S(delta, >=, 0);
2499 ASSERT3U(ddpa->used, >=, delta);
2500 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2501 dsl_dir_diduse_space(dd, DD_USED_HEAD,
2502 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2504 delta = ddpa->originusedsnap -
2505 dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
2506 ASSERT3S(delta, <=, 0);
2507 ASSERT3U(ddpa->used, >=, -delta);
2508 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2509 dsl_dir_diduse_space(odd, DD_USED_HEAD,
2510 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2512 dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
2514 /* log history record */
2515 spa_history_log_internal_ds(hds, "promote", tx, "");
2517 dsl_dir_rele(odd, FTAG);
2518 promote_rele(ddpa, FTAG);
2522 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2523 * (exclusive) and last_obj (inclusive). The list will be in reverse
2524 * order (last_obj will be the list_head()). If first_obj == 0, do all
2525 * snapshots back to this dataset's origin.
2528 snaplist_make(dsl_pool_t *dp,
2529 uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2531 uint64_t obj = last_obj;
2533 list_create(l, sizeof (struct promotenode),
2534 offsetof(struct promotenode, link));
2536 while (obj != first_obj) {
2538 struct promotenode *snap;
2541 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2542 ASSERT(err != ENOENT);
2547 first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
2549 snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2551 list_insert_tail(l, snap);
2552 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
2559 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2561 struct promotenode *snap;
2564 for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2565 uint64_t used, comp, uncomp;
2566 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2567 mintxg, UINT64_MAX, &used, &comp, &uncomp);
2574 snaplist_destroy(list_t *l, void *tag)
2576 struct promotenode *snap;
2578 if (l == NULL || !list_link_active(&l->list_head))
2581 while ((snap = list_tail(l)) != NULL) {
2582 list_remove(l, snap);
2583 dsl_dataset_rele(snap->ds, tag);
2584 kmem_free(snap, sizeof (*snap));
2590 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2594 struct promotenode *snap;
2596 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2600 dd = ddpa->ddpa_clone->ds_dir;
2602 if (dsl_dataset_is_snapshot(ddpa->ddpa_clone) ||
2603 !dsl_dir_is_clone(dd)) {
2604 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2605 return (SET_ERROR(EINVAL));
2608 error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
2609 &ddpa->shared_snaps, tag);
2613 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2614 &ddpa->clone_snaps, tag);
2618 snap = list_head(&ddpa->shared_snaps);
2619 ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
2620 error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
2621 dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
2622 &ddpa->origin_snaps, tag);
2626 if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
2627 error = dsl_dataset_hold_obj(dp,
2628 dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
2629 tag, &ddpa->origin_origin);
2635 promote_rele(ddpa, tag);
2640 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2642 snaplist_destroy(&ddpa->shared_snaps, tag);
2643 snaplist_destroy(&ddpa->clone_snaps, tag);
2644 snaplist_destroy(&ddpa->origin_snaps, tag);
2645 if (ddpa->origin_origin != NULL)
2646 dsl_dataset_rele(ddpa->origin_origin, tag);
2647 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2653 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2654 * in with the name. (It must be at least MAXNAMELEN bytes long.)
2657 dsl_dataset_promote(const char *name, char *conflsnap)
2659 dsl_dataset_promote_arg_t ddpa = { 0 };
2665 * We will modify space proportional to the number of
2666 * snapshots. Compute numsnaps.
2668 error = dmu_objset_hold(name, FTAG, &os);
2671 error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2672 dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
2674 dmu_objset_rele(os, FTAG);
2678 ddpa.ddpa_clonename = name;
2679 ddpa.err_ds = conflsnap;
2682 return (dsl_sync_task(name, dsl_dataset_promote_check,
2683 dsl_dataset_promote_sync, &ddpa,
2684 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED));
2688 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2689 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2691 int64_t unused_refres_delta;
2693 /* they should both be heads */
2694 if (dsl_dataset_is_snapshot(clone) ||
2695 dsl_dataset_is_snapshot(origin_head))
2696 return (SET_ERROR(EINVAL));
2698 /* if we are not forcing, the branch point should be just before them */
2699 if (!force && clone->ds_prev != origin_head->ds_prev)
2700 return (SET_ERROR(EINVAL));
2702 /* clone should be the clone (unless they are unrelated) */
2703 if (clone->ds_prev != NULL &&
2704 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2705 origin_head->ds_dir != clone->ds_prev->ds_dir)
2706 return (SET_ERROR(EINVAL));
2708 /* the clone should be a child of the origin */
2709 if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2710 return (SET_ERROR(EINVAL));
2712 /* origin_head shouldn't be modified unless 'force' */
2714 dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2715 return (SET_ERROR(ETXTBSY));
2717 /* origin_head should have no long holds (e.g. is not mounted) */
2718 if (dsl_dataset_handoff_check(origin_head, owner, tx))
2719 return (SET_ERROR(EBUSY));
2721 /* check amount of any unconsumed refreservation */
2722 unused_refres_delta =
2723 (int64_t)MIN(origin_head->ds_reserved,
2724 dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2725 (int64_t)MIN(origin_head->ds_reserved,
2726 dsl_dataset_phys(clone)->ds_unique_bytes);
2728 if (unused_refres_delta > 0 &&
2729 unused_refres_delta >
2730 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2731 return (SET_ERROR(ENOSPC));
2733 /* clone can't be over the head's refquota */
2734 if (origin_head->ds_quota != 0 &&
2735 dsl_dataset_phys(clone)->ds_referenced_bytes >
2736 origin_head->ds_quota)
2737 return (SET_ERROR(EDQUOT));
2743 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2744 dsl_dataset_t *origin_head, dmu_tx_t *tx)
2746 dsl_pool_t *dp = dmu_tx_pool(tx);
2747 int64_t unused_refres_delta;
2749 ASSERT(clone->ds_reserved == 0);
2750 ASSERT(origin_head->ds_quota == 0 ||
2751 dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota);
2752 ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
2754 dmu_buf_will_dirty(clone->ds_dbuf, tx);
2755 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2757 if (clone->ds_objset != NULL) {
2758 dmu_objset_evict(clone->ds_objset);
2759 clone->ds_objset = NULL;
2762 if (origin_head->ds_objset != NULL) {
2763 dmu_objset_evict(origin_head->ds_objset);
2764 origin_head->ds_objset = NULL;
2767 unused_refres_delta =
2768 (int64_t)MIN(origin_head->ds_reserved,
2769 dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2770 (int64_t)MIN(origin_head->ds_reserved,
2771 dsl_dataset_phys(clone)->ds_unique_bytes);
2774 * Reset origin's unique bytes, if it exists.
2776 if (clone->ds_prev) {
2777 dsl_dataset_t *origin = clone->ds_prev;
2778 uint64_t comp, uncomp;
2780 dmu_buf_will_dirty(origin->ds_dbuf, tx);
2781 dsl_deadlist_space_range(&clone->ds_deadlist,
2782 dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
2783 &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
2789 tmp = dsl_dataset_phys(origin_head)->ds_bp;
2790 dsl_dataset_phys(origin_head)->ds_bp =
2791 dsl_dataset_phys(clone)->ds_bp;
2792 dsl_dataset_phys(clone)->ds_bp = tmp;
2795 /* set dd_*_bytes */
2797 int64_t dused, dcomp, duncomp;
2798 uint64_t cdl_used, cdl_comp, cdl_uncomp;
2799 uint64_t odl_used, odl_comp, odl_uncomp;
2801 ASSERT3U(dsl_dir_phys(clone->ds_dir)->
2802 dd_used_breakdown[DD_USED_SNAP], ==, 0);
2804 dsl_deadlist_space(&clone->ds_deadlist,
2805 &cdl_used, &cdl_comp, &cdl_uncomp);
2806 dsl_deadlist_space(&origin_head->ds_deadlist,
2807 &odl_used, &odl_comp, &odl_uncomp);
2809 dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
2811 (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
2813 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
2815 (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
2817 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
2819 (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
2822 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
2823 dused, dcomp, duncomp, tx);
2824 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
2825 -dused, -dcomp, -duncomp, tx);
2828 * The difference in the space used by snapshots is the
2829 * difference in snapshot space due to the head's
2830 * deadlist (since that's the only thing that's
2831 * changing that affects the snapused).
2833 dsl_deadlist_space_range(&clone->ds_deadlist,
2834 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2835 &cdl_used, &cdl_comp, &cdl_uncomp);
2836 dsl_deadlist_space_range(&origin_head->ds_deadlist,
2837 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2838 &odl_used, &odl_comp, &odl_uncomp);
2839 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
2840 DD_USED_HEAD, DD_USED_SNAP, NULL);
2843 /* swap ds_*_bytes */
2844 SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
2845 dsl_dataset_phys(clone)->ds_referenced_bytes);
2846 SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
2847 dsl_dataset_phys(clone)->ds_compressed_bytes);
2848 SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
2849 dsl_dataset_phys(clone)->ds_uncompressed_bytes);
2850 SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
2851 dsl_dataset_phys(clone)->ds_unique_bytes);
2853 /* apply any parent delta for change in unconsumed refreservation */
2854 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
2855 unused_refres_delta, 0, 0, tx);
2860 dsl_deadlist_close(&clone->ds_deadlist);
2861 dsl_deadlist_close(&origin_head->ds_deadlist);
2862 SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
2863 dsl_dataset_phys(clone)->ds_deadlist_obj);
2864 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
2865 dsl_dataset_phys(clone)->ds_deadlist_obj);
2866 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
2867 dsl_dataset_phys(origin_head)->ds_deadlist_obj);
2869 dsl_scan_ds_clone_swapped(origin_head, clone, tx);
2871 spa_history_log_internal_ds(clone, "clone swap", tx,
2872 "parent=%s", origin_head->ds_dir->dd_myname);
2876 * Given a pool name and a dataset object number in that pool,
2877 * return the name of that dataset.
2880 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2886 error = dsl_pool_hold(pname, FTAG, &dp);
2890 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
2892 dsl_dataset_name(ds, buf);
2893 dsl_dataset_rele(ds, FTAG);
2895 dsl_pool_rele(dp, FTAG);
2901 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
2902 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
2906 ASSERT3S(asize, >, 0);
2909 * *ref_rsrv is the portion of asize that will come from any
2910 * unconsumed refreservation space.
2914 mutex_enter(&ds->ds_lock);
2916 * Make a space adjustment for reserved bytes.
2918 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
2920 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
2922 (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
2924 asize - MIN(asize, parent_delta(ds, asize + inflight));
2927 if (!check_quota || ds->ds_quota == 0) {
2928 mutex_exit(&ds->ds_lock);
2932 * If they are requesting more space, and our current estimate
2933 * is over quota, they get to try again unless the actual
2934 * on-disk is over quota and there are no pending changes (which
2935 * may free up space for us).
2937 if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
2940 dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
2941 error = SET_ERROR(ERESTART);
2943 error = SET_ERROR(EDQUOT);
2945 mutex_exit(&ds->ds_lock);
2950 typedef struct dsl_dataset_set_qr_arg {
2951 const char *ddsqra_name;
2952 zprop_source_t ddsqra_source;
2953 uint64_t ddsqra_value;
2954 } dsl_dataset_set_qr_arg_t;
2959 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
2961 dsl_dataset_set_qr_arg_t *ddsqra = arg;
2962 dsl_pool_t *dp = dmu_tx_pool(tx);
2967 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
2968 return (SET_ERROR(ENOTSUP));
2970 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2974 if (dsl_dataset_is_snapshot(ds)) {
2975 dsl_dataset_rele(ds, FTAG);
2976 return (SET_ERROR(EINVAL));
2979 error = dsl_prop_predict(ds->ds_dir,
2980 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2981 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2983 dsl_dataset_rele(ds, FTAG);
2988 dsl_dataset_rele(ds, FTAG);
2992 if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
2993 newval < ds->ds_reserved) {
2994 dsl_dataset_rele(ds, FTAG);
2995 return (SET_ERROR(ENOSPC));
2998 dsl_dataset_rele(ds, FTAG);
3003 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
3005 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3006 dsl_pool_t *dp = dmu_tx_pool(tx);
3010 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3012 dsl_prop_set_sync_impl(ds,
3013 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3014 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
3015 &ddsqra->ddsqra_value, tx);
3017 VERIFY0(dsl_prop_get_int_ds(ds,
3018 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
3020 if (ds->ds_quota != newval) {
3021 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3022 ds->ds_quota = newval;
3024 dsl_dataset_rele(ds, FTAG);
3028 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
3031 dsl_dataset_set_qr_arg_t ddsqra;
3033 ddsqra.ddsqra_name = dsname;
3034 ddsqra.ddsqra_source = source;
3035 ddsqra.ddsqra_value = refquota;
3037 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
3038 dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
3042 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
3044 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3045 dsl_pool_t *dp = dmu_tx_pool(tx);
3048 uint64_t newval, unique;
3050 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
3051 return (SET_ERROR(ENOTSUP));
3053 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3057 if (dsl_dataset_is_snapshot(ds)) {
3058 dsl_dataset_rele(ds, FTAG);
3059 return (SET_ERROR(EINVAL));
3062 error = dsl_prop_predict(ds->ds_dir,
3063 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3064 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3066 dsl_dataset_rele(ds, FTAG);
3071 * If we are doing the preliminary check in open context, the
3072 * space estimates may be inaccurate.
3074 if (!dmu_tx_is_syncing(tx)) {
3075 dsl_dataset_rele(ds, FTAG);
3079 mutex_enter(&ds->ds_lock);
3080 if (!DS_UNIQUE_IS_ACCURATE(ds))
3081 dsl_dataset_recalc_head_uniq(ds);
3082 unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3083 mutex_exit(&ds->ds_lock);
3085 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3086 uint64_t delta = MAX(unique, newval) -
3087 MAX(unique, ds->ds_reserved);
3090 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3091 (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3092 dsl_dataset_rele(ds, FTAG);
3093 return (SET_ERROR(ENOSPC));
3097 dsl_dataset_rele(ds, FTAG);
3102 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3103 zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3109 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3110 source, sizeof (value), 1, &value, tx);
3112 VERIFY0(dsl_prop_get_int_ds(ds,
3113 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3115 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3116 mutex_enter(&ds->ds_dir->dd_lock);
3117 mutex_enter(&ds->ds_lock);
3118 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3119 unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3120 delta = MAX(0, (int64_t)(newval - unique)) -
3121 MAX(0, (int64_t)(ds->ds_reserved - unique));
3122 ds->ds_reserved = newval;
3123 mutex_exit(&ds->ds_lock);
3125 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3126 mutex_exit(&ds->ds_dir->dd_lock);
3130 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3132 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3133 dsl_pool_t *dp = dmu_tx_pool(tx);
3136 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3137 dsl_dataset_set_refreservation_sync_impl(ds,
3138 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3139 dsl_dataset_rele(ds, FTAG);
3143 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3144 uint64_t refreservation)
3146 dsl_dataset_set_qr_arg_t ddsqra;
3148 ddsqra.ddsqra_name = dsname;
3149 ddsqra.ddsqra_source = source;
3150 ddsqra.ddsqra_value = refreservation;
3152 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3153 dsl_dataset_set_refreservation_sync, &ddsqra,
3154 0, ZFS_SPACE_CHECK_NONE));
3158 * Return (in *usedp) the amount of space written in new that is not
3159 * present in oldsnap. New may be a snapshot or the head. Old must be
3160 * a snapshot before new, in new's filesystem (or its origin). If not then
3161 * fail and return EINVAL.
3163 * The written space is calculated by considering two components: First, we
3164 * ignore any freed space, and calculate the written as new's used space
3165 * minus old's used space. Next, we add in the amount of space that was freed
3166 * between the two snapshots, thus reducing new's used space relative to old's.
3167 * Specifically, this is the space that was born before old->ds_creation_txg,
3168 * and freed before new (ie. on new's deadlist or a previous deadlist).
3170 * space freed [---------------------]
3171 * snapshots ---O-------O--------O-------O------
3175 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3176 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3180 dsl_pool_t *dp = new->ds_dir->dd_pool;
3182 ASSERT(dsl_pool_config_held(dp));
3185 *usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3186 *usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3189 *compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3190 *compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3193 *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3194 *uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3196 snapobj = new->ds_object;
3197 while (snapobj != oldsnap->ds_object) {
3198 dsl_dataset_t *snap;
3199 uint64_t used, comp, uncomp;
3201 if (snapobj == new->ds_object) {
3204 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3209 if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
3210 dsl_dataset_phys(oldsnap)->ds_creation_txg) {
3212 * The blocks in the deadlist can not be born after
3213 * ds_prev_snap_txg, so get the whole deadlist space,
3214 * which is more efficient (especially for old-format
3215 * deadlists). Unfortunately the deadlist code
3216 * doesn't have enough information to make this
3217 * optimization itself.
3219 dsl_deadlist_space(&snap->ds_deadlist,
3220 &used, &comp, &uncomp);
3222 dsl_deadlist_space_range(&snap->ds_deadlist,
3223 0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
3224 &used, &comp, &uncomp);
3231 * If we get to the beginning of the chain of snapshots
3232 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3233 * was not a snapshot of/before new.
3235 snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
3237 dsl_dataset_rele(snap, FTAG);
3239 err = SET_ERROR(EINVAL);
3248 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3249 * lastsnap, and all snapshots in between are deleted.
3251 * blocks that would be freed [---------------------------]
3252 * snapshots ---O-------O--------O-------O--------O
3253 * firstsnap lastsnap
3255 * This is the set of blocks that were born after the snap before firstsnap,
3256 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3257 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3258 * We calculate this by iterating over the relevant deadlists (from the snap
3259 * after lastsnap, backward to the snap after firstsnap), summing up the
3260 * space on the deadlist that was born after the snap before firstsnap.
3263 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3264 dsl_dataset_t *lastsnap,
3265 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3269 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3271 ASSERT(dsl_dataset_is_snapshot(firstsnap));
3272 ASSERT(dsl_dataset_is_snapshot(lastsnap));
3275 * Check that the snapshots are in the same dsl_dir, and firstsnap
3276 * is before lastsnap.
3278 if (firstsnap->ds_dir != lastsnap->ds_dir ||
3279 dsl_dataset_phys(firstsnap)->ds_creation_txg >
3280 dsl_dataset_phys(lastsnap)->ds_creation_txg)
3281 return (SET_ERROR(EINVAL));
3283 *usedp = *compp = *uncompp = 0;
3285 snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
3286 while (snapobj != firstsnap->ds_object) {
3288 uint64_t used, comp, uncomp;
3290 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3294 dsl_deadlist_space_range(&ds->ds_deadlist,
3295 dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
3296 &used, &comp, &uncomp);
3301 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3302 ASSERT3U(snapobj, !=, 0);
3303 dsl_dataset_rele(ds, FTAG);
3309 dsl_dataset_activate_large_blocks_check(void *arg, dmu_tx_t *tx)
3311 const char *dsname = arg;
3313 dsl_pool_t *dp = dmu_tx_pool(tx);
3316 if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS))
3317 return (SET_ERROR(ENOTSUP));
3319 ASSERT(spa_feature_is_enabled(dp->dp_spa,
3320 SPA_FEATURE_EXTENSIBLE_DATASET));
3322 error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
3326 if (ds->ds_large_blocks)
3328 dsl_dataset_rele(ds, FTAG);
3334 dsl_dataset_activate_large_blocks_sync_impl(uint64_t dsobj, dmu_tx_t *tx)
3336 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
3337 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
3340 spa_feature_incr(spa, SPA_FEATURE_LARGE_BLOCKS, tx);
3341 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
3343 VERIFY0(zap_add(mos, dsobj, DS_FIELD_LARGE_BLOCKS,
3344 sizeof (zero), 1, &zero, tx));
3348 dsl_dataset_activate_large_blocks_sync(void *arg, dmu_tx_t *tx)
3350 const char *dsname = arg;
3353 VERIFY0(dsl_dataset_hold(dmu_tx_pool(tx), dsname, FTAG, &ds));
3355 dsl_dataset_activate_large_blocks_sync_impl(ds->ds_object, tx);
3356 ASSERT(!ds->ds_large_blocks);
3357 ds->ds_large_blocks = B_TRUE;
3358 dsl_dataset_rele(ds, FTAG);
3362 dsl_dataset_activate_large_blocks(const char *dsname)
3366 error = dsl_sync_task(dsname,
3367 dsl_dataset_activate_large_blocks_check,
3368 dsl_dataset_activate_large_blocks_sync, (void *)dsname,
3369 1, ZFS_SPACE_CHECK_RESERVED);
3372 * EALREADY indicates that this dataset already supports large blocks.
3374 if (error == EALREADY)
3380 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3381 * For example, they could both be snapshots of the same filesystem, and
3382 * 'earlier' is before 'later'. Or 'earlier' could be the origin of
3383 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's
3384 * filesystem. Or 'earlier' could be the origin's origin.
3386 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3389 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3390 uint64_t earlier_txg)
3392 dsl_pool_t *dp = later->ds_dir->dd_pool;
3396 ASSERT(dsl_pool_config_held(dp));
3397 ASSERT(dsl_dataset_is_snapshot(earlier) || earlier_txg != 0);
3399 if (earlier_txg == 0)
3400 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
3402 if (dsl_dataset_is_snapshot(later) &&
3403 earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
3406 if (later->ds_dir == earlier->ds_dir)
3408 if (!dsl_dir_is_clone(later->ds_dir))
3411 if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
3413 dsl_dataset_t *origin;
3414 error = dsl_dataset_hold_obj(dp,
3415 dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
3418 ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3419 dsl_dataset_rele(origin, FTAG);
3425 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3427 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3428 dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);