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]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Portions Copyright (c) 2011 Martin Matuska <mm@FreeBSD.org>
25 * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
26 * Copyright (c) 2014, Joyent, Inc. All rights reserved.
27 * Copyright (c) 2014 RackTop Systems.
28 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
29 * Copyright (c) 2014 Integros [integros.com]
30 * Copyright 2016, OmniTI Computer Consulting, Inc. All rights reserved.
31 * Copyright 2017 Nexenta Systems, Inc.
34 #include <sys/dmu_objset.h>
35 #include <sys/dsl_dataset.h>
36 #include <sys/dsl_dir.h>
37 #include <sys/dsl_prop.h>
38 #include <sys/dsl_synctask.h>
39 #include <sys/dmu_traverse.h>
40 #include <sys/dmu_impl.h>
41 #include <sys/dmu_send.h>
42 #include <sys/dmu_tx.h>
46 #include <sys/zfeature.h>
47 #include <sys/unique.h>
48 #include <sys/zfs_context.h>
49 #include <sys/zfs_ioctl.h>
51 #include <sys/spa_impl.h>
53 #include <sys/zfs_znode.h>
54 #include <sys/zfs_onexit.h>
56 #include <sys/dsl_scan.h>
57 #include <sys/dsl_deadlist.h>
58 #include <sys/dsl_destroy.h>
59 #include <sys/dsl_userhold.h>
60 #include <sys/dsl_bookmark.h>
61 #include <sys/dmu_send.h>
62 #include <sys/zio_checksum.h>
63 #include <sys/zio_compress.h>
64 #include <zfs_fletcher.h>
66 SYSCTL_DECL(_vfs_zfs);
69 * The SPA supports block sizes up to 16MB. However, very large blocks
70 * can have an impact on i/o latency (e.g. tying up a spinning disk for
71 * ~300ms), and also potentially on the memory allocator. Therefore,
72 * we do not allow the recordsize to be set larger than zfs_max_recordsize
73 * (default 1MB). Larger blocks can be created by changing this tunable,
74 * and pools with larger blocks can always be imported and used, regardless
77 int zfs_max_recordsize = 1 * 1024 * 1024;
78 SYSCTL_INT(_vfs_zfs, OID_AUTO, max_recordsize, CTLFLAG_RWTUN,
79 &zfs_max_recordsize, 0,
80 "Maximum block size. Expect dragons when tuning this.");
82 #define SWITCH64(x, y) \
84 uint64_t __tmp = (x); \
89 #define DS_REF_MAX (1ULL << 62)
91 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
93 static void dsl_dataset_set_remap_deadlist_object(dsl_dataset_t *ds,
94 uint64_t obj, dmu_tx_t *tx);
95 static void dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t *ds,
98 extern int spa_asize_inflation;
100 static zil_header_t zero_zil;
103 * Figure out how much of this delta should be propogated to the dsl_dir
104 * layer. If there's a refreservation, that space has already been
105 * partially accounted for in our ancestors.
108 parent_delta(dsl_dataset_t *ds, int64_t delta)
110 dsl_dataset_phys_t *ds_phys;
111 uint64_t old_bytes, new_bytes;
113 if (ds->ds_reserved == 0)
116 ds_phys = dsl_dataset_phys(ds);
117 old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
118 new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
120 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
121 return (new_bytes - old_bytes);
125 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
127 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
128 int compressed = BP_GET_PSIZE(bp);
129 int uncompressed = BP_GET_UCSIZE(bp);
132 dprintf_bp(bp, "ds=%p", ds);
134 ASSERT(dmu_tx_is_syncing(tx));
135 /* It could have been compressed away to nothing */
138 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
139 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
141 dsl_pool_mos_diduse_space(tx->tx_pool,
142 used, compressed, uncompressed);
146 ASSERT3U(bp->blk_birth, >, dsl_dataset_phys(ds)->ds_prev_snap_txg);
147 dmu_buf_will_dirty(ds->ds_dbuf, tx);
148 mutex_enter(&ds->ds_lock);
149 delta = parent_delta(ds, used);
150 dsl_dataset_phys(ds)->ds_referenced_bytes += used;
151 dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
152 dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
153 dsl_dataset_phys(ds)->ds_unique_bytes += used;
155 if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) {
156 ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_BLOCKS] =
160 spa_feature_t f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp));
161 if (f != SPA_FEATURE_NONE)
162 ds->ds_feature_activation_needed[f] = B_TRUE;
164 mutex_exit(&ds->ds_lock);
165 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
166 compressed, uncompressed, tx);
167 dsl_dir_transfer_space(ds->ds_dir, used - delta,
168 DD_USED_REFRSRV, DD_USED_HEAD, NULL);
172 * Called when the specified segment has been remapped, and is thus no
173 * longer referenced in the head dataset. The vdev must be indirect.
175 * If the segment is referenced by a snapshot, put it on the remap deadlist.
176 * Otherwise, add this segment to the obsolete spacemap.
179 dsl_dataset_block_remapped(dsl_dataset_t *ds, uint64_t vdev, uint64_t offset,
180 uint64_t size, uint64_t birth, dmu_tx_t *tx)
182 spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
184 ASSERT(dmu_tx_is_syncing(tx));
185 ASSERT(birth <= tx->tx_txg);
186 ASSERT(!ds->ds_is_snapshot);
188 if (birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
189 spa_vdev_indirect_mark_obsolete(spa, vdev, offset, size, tx);
192 dva_t *dva = &fakebp.blk_dva[0];
196 mutex_enter(&ds->ds_remap_deadlist_lock);
197 if (!dsl_dataset_remap_deadlist_exists(ds)) {
198 dsl_dataset_create_remap_deadlist(ds, tx);
200 mutex_exit(&ds->ds_remap_deadlist_lock);
203 fakebp.blk_birth = birth;
204 DVA_SET_VDEV(dva, vdev);
205 DVA_SET_OFFSET(dva, offset);
206 DVA_SET_ASIZE(dva, size);
208 dsl_deadlist_insert(&ds->ds_remap_deadlist, &fakebp, tx);
213 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
216 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
218 int used = bp_get_dsize_sync(spa, bp);
219 int compressed = BP_GET_PSIZE(bp);
220 int uncompressed = BP_GET_UCSIZE(bp);
225 ASSERT(dmu_tx_is_syncing(tx));
226 ASSERT(bp->blk_birth <= tx->tx_txg);
229 dsl_free(tx->tx_pool, tx->tx_txg, bp);
230 dsl_pool_mos_diduse_space(tx->tx_pool,
231 -used, -compressed, -uncompressed);
234 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
236 ASSERT(!ds->ds_is_snapshot);
237 dmu_buf_will_dirty(ds->ds_dbuf, tx);
239 if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
242 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
243 dsl_free(tx->tx_pool, tx->tx_txg, bp);
245 mutex_enter(&ds->ds_lock);
246 ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
247 !DS_UNIQUE_IS_ACCURATE(ds));
248 delta = parent_delta(ds, -used);
249 dsl_dataset_phys(ds)->ds_unique_bytes -= used;
250 mutex_exit(&ds->ds_lock);
251 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
252 delta, -compressed, -uncompressed, tx);
253 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
254 DD_USED_REFRSRV, DD_USED_HEAD, NULL);
256 dprintf_bp(bp, "putting on dead list: %s", "");
259 * We are here as part of zio's write done callback,
260 * which means we're a zio interrupt thread. We can't
261 * call dsl_deadlist_insert() now because it may block
262 * waiting for I/O. Instead, put bp on the deferred
263 * queue and let dsl_pool_sync() finish the job.
265 bplist_append(&ds->ds_pending_deadlist, bp);
267 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
269 ASSERT3U(ds->ds_prev->ds_object, ==,
270 dsl_dataset_phys(ds)->ds_prev_snap_obj);
271 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
272 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
273 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
274 ds->ds_object && bp->blk_birth >
275 dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
276 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
277 mutex_enter(&ds->ds_prev->ds_lock);
278 dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
279 mutex_exit(&ds->ds_prev->ds_lock);
281 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
282 dsl_dir_transfer_space(ds->ds_dir, used,
283 DD_USED_HEAD, DD_USED_SNAP, tx);
286 mutex_enter(&ds->ds_lock);
287 ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
288 dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
289 ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
290 dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
291 ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
292 dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
293 mutex_exit(&ds->ds_lock);
299 * We have to release the fsid syncronously or we risk that a subsequent
300 * mount of the same dataset will fail to unique_insert the fsid. This
301 * failure would manifest itself as the fsid of this dataset changing
302 * between mounts which makes NFS clients quite unhappy.
305 dsl_dataset_evict_sync(void *dbu)
307 dsl_dataset_t *ds = dbu;
309 ASSERT(ds->ds_owner == NULL);
311 unique_remove(ds->ds_fsid_guid);
315 dsl_dataset_evict_async(void *dbu)
317 dsl_dataset_t *ds = dbu;
319 ASSERT(ds->ds_owner == NULL);
323 if (ds->ds_objset != NULL)
324 dmu_objset_evict(ds->ds_objset);
327 dsl_dataset_rele(ds->ds_prev, ds);
331 bplist_destroy(&ds->ds_pending_deadlist);
332 if (dsl_deadlist_is_open(&ds->ds_deadlist))
333 dsl_deadlist_close(&ds->ds_deadlist);
334 if (dsl_deadlist_is_open(&ds->ds_remap_deadlist))
335 dsl_deadlist_close(&ds->ds_remap_deadlist);
337 dsl_dir_async_rele(ds->ds_dir, ds);
339 ASSERT(!list_link_active(&ds->ds_synced_link));
341 list_destroy(&ds->ds_prop_cbs);
342 if (mutex_owned(&ds->ds_lock))
343 mutex_exit(&ds->ds_lock);
344 mutex_destroy(&ds->ds_lock);
345 if (mutex_owned(&ds->ds_opening_lock))
346 mutex_exit(&ds->ds_opening_lock);
347 mutex_destroy(&ds->ds_opening_lock);
348 mutex_destroy(&ds->ds_sendstream_lock);
349 mutex_destroy(&ds->ds_remap_deadlist_lock);
350 refcount_destroy(&ds->ds_longholds);
351 rrw_destroy(&ds->ds_bp_rwlock);
353 kmem_free(ds, sizeof (dsl_dataset_t));
357 dsl_dataset_get_snapname(dsl_dataset_t *ds)
359 dsl_dataset_phys_t *headphys;
362 dsl_pool_t *dp = ds->ds_dir->dd_pool;
363 objset_t *mos = dp->dp_meta_objset;
365 if (ds->ds_snapname[0])
367 if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
370 err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
374 headphys = headdbuf->db_data;
375 err = zap_value_search(dp->dp_meta_objset,
376 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
377 dmu_buf_rele(headdbuf, FTAG);
382 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
384 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
385 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
389 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
392 err = zap_lookup_norm(mos, snapobj, name, 8, 1,
393 value, mt, NULL, 0, NULL);
394 if (err == ENOTSUP && (mt & MT_NORMALIZE))
395 err = zap_lookup(mos, snapobj, name, 8, 1, value);
400 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
403 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
404 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
408 dsl_dir_snap_cmtime_update(ds->ds_dir);
410 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
413 err = zap_remove_norm(mos, snapobj, name, mt, tx);
414 if (err == ENOTSUP && (mt & MT_NORMALIZE))
415 err = zap_remove(mos, snapobj, name, tx);
417 if (err == 0 && adj_cnt)
418 dsl_fs_ss_count_adjust(ds->ds_dir, -1,
419 DD_FIELD_SNAPSHOT_COUNT, tx);
425 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag)
427 dmu_buf_t *dbuf = ds->ds_dbuf;
428 boolean_t result = B_FALSE;
430 if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset,
431 ds->ds_object, DMU_BONUS_BLKID, tag)) {
433 if (ds == dmu_buf_get_user(dbuf))
436 dmu_buf_rele(dbuf, tag);
443 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
446 objset_t *mos = dp->dp_meta_objset;
450 dmu_object_info_t doi;
452 ASSERT(dsl_pool_config_held(dp));
454 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
458 /* Make sure dsobj has the correct object type. */
459 dmu_object_info_from_db(dbuf, &doi);
460 if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
461 dmu_buf_rele(dbuf, tag);
462 return (SET_ERROR(EINVAL));
465 ds = dmu_buf_get_user(dbuf);
467 dsl_dataset_t *winner = NULL;
469 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
471 ds->ds_object = dsobj;
472 ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0;
474 err = dsl_dir_hold_obj(dp, dsl_dataset_phys(ds)->ds_dir_obj,
475 NULL, ds, &ds->ds_dir);
477 kmem_free(ds, sizeof (dsl_dataset_t));
478 dmu_buf_rele(dbuf, tag);
482 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
483 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
484 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
485 mutex_init(&ds->ds_remap_deadlist_lock,
486 NULL, MUTEX_DEFAULT, NULL);
487 rrw_init(&ds->ds_bp_rwlock, B_FALSE);
488 refcount_create(&ds->ds_longholds);
490 bplist_create(&ds->ds_pending_deadlist);
492 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
493 offsetof(dmu_sendarg_t, dsa_link));
495 list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t),
496 offsetof(dsl_prop_cb_record_t, cbr_ds_node));
498 if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
499 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
500 if (!(spa_feature_table[f].fi_flags &
501 ZFEATURE_FLAG_PER_DATASET))
503 err = zap_contains(mos, dsobj,
504 spa_feature_table[f].fi_guid);
506 ds->ds_feature_inuse[f] = B_TRUE;
508 ASSERT3U(err, ==, ENOENT);
514 if (!ds->ds_is_snapshot) {
515 ds->ds_snapname[0] = '\0';
516 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
517 err = dsl_dataset_hold_obj(dp,
518 dsl_dataset_phys(ds)->ds_prev_snap_obj,
521 if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
522 int zaperr = zap_lookup(mos, ds->ds_object,
523 DS_FIELD_BOOKMARK_NAMES,
524 sizeof (ds->ds_bookmarks), 1,
526 if (zaperr != ENOENT)
530 if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
531 err = dsl_dataset_get_snapname(ds);
533 dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
535 ds->ds_dir->dd_pool->dp_meta_objset,
536 dsl_dataset_phys(ds)->ds_userrefs_obj,
541 if (err == 0 && !ds->ds_is_snapshot) {
542 err = dsl_prop_get_int_ds(ds,
543 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
546 err = dsl_prop_get_int_ds(ds,
547 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
551 ds->ds_reserved = ds->ds_quota = 0;
554 dsl_deadlist_open(&ds->ds_deadlist,
555 mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
556 uint64_t remap_deadlist_obj =
557 dsl_dataset_get_remap_deadlist_object(ds);
558 if (remap_deadlist_obj != 0) {
559 dsl_deadlist_open(&ds->ds_remap_deadlist, mos,
563 dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict_sync,
564 dsl_dataset_evict_async, &ds->ds_dbuf);
566 winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
568 if (err != 0 || winner != NULL) {
569 bplist_destroy(&ds->ds_pending_deadlist);
570 dsl_deadlist_close(&ds->ds_deadlist);
571 if (dsl_deadlist_is_open(&ds->ds_remap_deadlist))
572 dsl_deadlist_close(&ds->ds_remap_deadlist);
574 dsl_dataset_rele(ds->ds_prev, ds);
575 dsl_dir_rele(ds->ds_dir, ds);
576 mutex_destroy(&ds->ds_lock);
577 mutex_destroy(&ds->ds_opening_lock);
578 mutex_destroy(&ds->ds_sendstream_lock);
579 refcount_destroy(&ds->ds_longholds);
580 kmem_free(ds, sizeof (dsl_dataset_t));
582 dmu_buf_rele(dbuf, tag);
588 unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
589 if (ds->ds_fsid_guid !=
590 dsl_dataset_phys(ds)->ds_fsid_guid) {
591 zfs_dbgmsg("ds_fsid_guid changed from "
592 "%llx to %llx for pool %s dataset id %llu",
594 dsl_dataset_phys(ds)->ds_fsid_guid,
595 (long long)ds->ds_fsid_guid,
596 spa_name(dp->dp_spa),
601 ASSERT3P(ds->ds_dbuf, ==, dbuf);
602 ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
603 ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
604 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
605 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
611 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
612 void *tag, dsl_dataset_t **dsp)
615 const char *snapname;
620 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
624 ASSERT(dsl_pool_config_held(dp));
625 obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
627 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
629 err = SET_ERROR(ENOENT);
631 /* we may be looking for a snapshot */
632 if (err == 0 && snapname != NULL) {
633 dsl_dataset_t *snap_ds;
635 if (*snapname++ != '@') {
636 dsl_dataset_rele(ds, tag);
637 dsl_dir_rele(dd, FTAG);
638 return (SET_ERROR(ENOENT));
641 dprintf("looking for snapshot '%s'\n", snapname);
642 err = dsl_dataset_snap_lookup(ds, snapname, &obj);
644 err = dsl_dataset_hold_obj(dp, obj, tag, &snap_ds);
645 dsl_dataset_rele(ds, tag);
648 mutex_enter(&snap_ds->ds_lock);
649 if (snap_ds->ds_snapname[0] == 0)
650 (void) strlcpy(snap_ds->ds_snapname, snapname,
651 sizeof (snap_ds->ds_snapname));
652 mutex_exit(&snap_ds->ds_lock);
658 dsl_dir_rele(dd, FTAG);
663 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
664 void *tag, dsl_dataset_t **dsp)
666 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
669 if (!dsl_dataset_tryown(*dsp, tag)) {
670 dsl_dataset_rele(*dsp, tag);
672 return (SET_ERROR(EBUSY));
678 dsl_dataset_own(dsl_pool_t *dp, const char *name,
679 void *tag, dsl_dataset_t **dsp)
681 int err = dsl_dataset_hold(dp, name, tag, dsp);
684 if (!dsl_dataset_tryown(*dsp, tag)) {
685 dsl_dataset_rele(*dsp, tag);
686 return (SET_ERROR(EBUSY));
692 * See the comment above dsl_pool_hold() for details. In summary, a long
693 * hold is used to prevent destruction of a dataset while the pool hold
694 * is dropped, allowing other concurrent operations (e.g. spa_sync()).
696 * The dataset and pool must be held when this function is called. After it
697 * is called, the pool hold may be released while the dataset is still held
701 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
703 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
704 (void) refcount_add(&ds->ds_longholds, tag);
708 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
710 (void) refcount_remove(&ds->ds_longholds, tag);
713 /* Return B_TRUE if there are any long holds on this dataset. */
715 dsl_dataset_long_held(dsl_dataset_t *ds)
717 return (!refcount_is_zero(&ds->ds_longholds));
721 dsl_dataset_name(dsl_dataset_t *ds, char *name)
724 (void) strcpy(name, "mos");
726 dsl_dir_name(ds->ds_dir, name);
727 VERIFY0(dsl_dataset_get_snapname(ds));
728 if (ds->ds_snapname[0]) {
729 VERIFY3U(strlcat(name, "@", ZFS_MAX_DATASET_NAME_LEN),
730 <, ZFS_MAX_DATASET_NAME_LEN);
732 * We use a "recursive" mutex so that we
733 * can call dprintf_ds() with ds_lock held.
735 if (!MUTEX_HELD(&ds->ds_lock)) {
736 mutex_enter(&ds->ds_lock);
737 VERIFY3U(strlcat(name, ds->ds_snapname,
738 ZFS_MAX_DATASET_NAME_LEN), <,
739 ZFS_MAX_DATASET_NAME_LEN);
740 mutex_exit(&ds->ds_lock);
742 VERIFY3U(strlcat(name, ds->ds_snapname,
743 ZFS_MAX_DATASET_NAME_LEN), <,
744 ZFS_MAX_DATASET_NAME_LEN);
751 dsl_dataset_namelen(dsl_dataset_t *ds)
753 VERIFY0(dsl_dataset_get_snapname(ds));
754 mutex_enter(&ds->ds_lock);
755 int len = dsl_dir_namelen(ds->ds_dir) + 1 + strlen(ds->ds_snapname);
756 mutex_exit(&ds->ds_lock);
761 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
763 dmu_buf_rele(ds->ds_dbuf, tag);
767 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
769 ASSERT3P(ds->ds_owner, ==, tag);
770 ASSERT(ds->ds_dbuf != NULL);
772 mutex_enter(&ds->ds_lock);
774 mutex_exit(&ds->ds_lock);
775 dsl_dataset_long_rele(ds, tag);
776 dsl_dataset_rele(ds, tag);
780 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
782 boolean_t gotit = FALSE;
784 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
785 mutex_enter(&ds->ds_lock);
786 if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
788 dsl_dataset_long_hold(ds, tag);
791 mutex_exit(&ds->ds_lock);
796 dsl_dataset_has_owner(dsl_dataset_t *ds)
799 mutex_enter(&ds->ds_lock);
800 rv = (ds->ds_owner != NULL);
801 mutex_exit(&ds->ds_lock);
806 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
808 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
809 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
812 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
814 spa_feature_incr(spa, f, tx);
815 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
817 VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
818 sizeof (zero), 1, &zero, tx));
822 dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
824 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
825 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
827 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
829 VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx));
830 spa_feature_decr(spa, f, tx);
834 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
835 uint64_t flags, dmu_tx_t *tx)
837 dsl_pool_t *dp = dd->dd_pool;
839 dsl_dataset_phys_t *dsphys;
841 objset_t *mos = dp->dp_meta_objset;
844 origin = dp->dp_origin_snap;
846 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
847 ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
848 ASSERT(dmu_tx_is_syncing(tx));
849 ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
851 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
852 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
853 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
854 dmu_buf_will_dirty(dbuf, tx);
855 dsphys = dbuf->db_data;
856 bzero(dsphys, sizeof (dsl_dataset_phys_t));
857 dsphys->ds_dir_obj = dd->dd_object;
858 dsphys->ds_flags = flags;
859 dsphys->ds_fsid_guid = unique_create();
861 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
862 sizeof (dsphys->ds_guid));
863 } while (dsphys->ds_guid == 0);
864 dsphys->ds_snapnames_zapobj =
865 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
867 dsphys->ds_creation_time = gethrestime_sec();
868 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
870 if (origin == NULL) {
871 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
873 dsl_dataset_t *ohds; /* head of the origin snapshot */
875 dsphys->ds_prev_snap_obj = origin->ds_object;
876 dsphys->ds_prev_snap_txg =
877 dsl_dataset_phys(origin)->ds_creation_txg;
878 dsphys->ds_referenced_bytes =
879 dsl_dataset_phys(origin)->ds_referenced_bytes;
880 dsphys->ds_compressed_bytes =
881 dsl_dataset_phys(origin)->ds_compressed_bytes;
882 dsphys->ds_uncompressed_bytes =
883 dsl_dataset_phys(origin)->ds_uncompressed_bytes;
884 rrw_enter(&origin->ds_bp_rwlock, RW_READER, FTAG);
885 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
886 rrw_exit(&origin->ds_bp_rwlock, FTAG);
889 * Inherit flags that describe the dataset's contents
890 * (INCONSISTENT) or properties (Case Insensitive).
892 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
893 (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
895 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
896 if (origin->ds_feature_inuse[f])
897 dsl_dataset_activate_feature(dsobj, f, tx);
900 dmu_buf_will_dirty(origin->ds_dbuf, tx);
901 dsl_dataset_phys(origin)->ds_num_children++;
903 VERIFY0(dsl_dataset_hold_obj(dp,
904 dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
906 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
907 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
908 dsl_dataset_rele(ohds, FTAG);
910 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
911 if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
912 dsl_dataset_phys(origin)->ds_next_clones_obj =
914 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
916 VERIFY0(zap_add_int(mos,
917 dsl_dataset_phys(origin)->ds_next_clones_obj,
921 dmu_buf_will_dirty(dd->dd_dbuf, tx);
922 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
923 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
924 if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
925 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
926 dsl_dir_phys(origin->ds_dir)->dd_clones =
928 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
930 VERIFY0(zap_add_int(mos,
931 dsl_dir_phys(origin->ds_dir)->dd_clones,
936 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
937 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
939 dmu_buf_rele(dbuf, FTAG);
941 dmu_buf_will_dirty(dd->dd_dbuf, tx);
942 dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
948 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
952 VERIFY0(dmu_objset_from_ds(ds, &os));
953 if (bcmp(&os->os_zil_header, &zero_zil, sizeof (zero_zil)) != 0) {
954 dsl_pool_t *dp = ds->ds_dir->dd_pool;
957 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
959 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
960 dsl_dataset_sync(ds, zio, tx);
961 VERIFY0(zio_wait(zio));
963 /* dsl_dataset_sync_done will drop this reference. */
964 dmu_buf_add_ref(ds->ds_dbuf, ds);
965 dsl_dataset_sync_done(ds, tx);
970 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
971 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
973 dsl_pool_t *dp = pdd->dd_pool;
974 uint64_t dsobj, ddobj;
977 ASSERT(dmu_tx_is_syncing(tx));
978 ASSERT(lastname[0] != '@');
980 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
981 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
983 dsobj = dsl_dataset_create_sync_dd(dd, origin,
984 flags & ~DS_CREATE_FLAG_NODIRTY, tx);
986 dsl_deleg_set_create_perms(dd, tx, cr);
989 * Since we're creating a new node we know it's a leaf, so we can
990 * initialize the counts if the limit feature is active.
992 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
994 objset_t *os = dd->dd_pool->dp_meta_objset;
996 dsl_dir_zapify(dd, tx);
997 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
998 sizeof (cnt), 1, &cnt, tx));
999 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
1000 sizeof (cnt), 1, &cnt, tx));
1003 dsl_dir_rele(dd, FTAG);
1006 * If we are creating a clone, make sure we zero out any stale
1007 * data from the origin snapshots zil header.
1009 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
1012 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
1013 dsl_dataset_zero_zil(ds, tx);
1014 dsl_dataset_rele(ds, FTAG);
1021 /* FreeBSD ioctl compat begin */
1024 const char *snapname;
1028 dsl_check_snap_cb(const char *name, void *arg)
1030 struct destroyarg *da = arg;
1034 dsname = kmem_asprintf("%s@%s", name, da->snapname);
1035 fnvlist_add_boolean(da->nvl, dsname);
1036 kmem_free(dsname, strlen(dsname) + 1);
1042 dmu_get_recursive_snaps_nvl(char *fsname, const char *snapname,
1045 struct destroyarg *da;
1048 da = kmem_zalloc(sizeof (struct destroyarg), KM_SLEEP);
1050 da->snapname = snapname;
1051 err = dmu_objset_find(fsname, dsl_check_snap_cb, da,
1053 kmem_free(da, sizeof (struct destroyarg));
1057 /* FreeBSD ioctl compat end */
1058 #endif /* __FreeBSD__ */
1061 * The unique space in the head dataset can be calculated by subtracting
1062 * the space used in the most recent snapshot, that is still being used
1063 * in this file system, from the space currently in use. To figure out
1064 * the space in the most recent snapshot still in use, we need to take
1065 * the total space used in the snapshot and subtract out the space that
1066 * has been freed up since the snapshot was taken.
1069 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1072 uint64_t dlused, dlcomp, dluncomp;
1074 ASSERT(!ds->ds_is_snapshot);
1076 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
1077 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
1081 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
1083 ASSERT3U(dlused, <=, mrs_used);
1084 dsl_dataset_phys(ds)->ds_unique_bytes =
1085 dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
1087 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1088 SPA_VERSION_UNIQUE_ACCURATE)
1089 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1093 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
1096 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1100 ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
1101 err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1104 * The err should not be ENOENT, but a bug in a previous version
1105 * of the code could cause upgrade_clones_cb() to not set
1106 * ds_next_snap_obj when it should, leading to a missing entry.
1107 * If we knew that the pool was created after
1108 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1109 * ENOENT. However, at least we can check that we don't have
1110 * too many entries in the next_clones_obj even after failing to
1115 ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1117 ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
1122 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1124 return (&dsl_dataset_phys(ds)->ds_bp);
1128 dsl_dataset_get_spa(dsl_dataset_t *ds)
1130 return (ds->ds_dir->dd_pool->dp_spa);
1134 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1138 if (ds == NULL) /* this is the meta-objset */
1141 ASSERT(ds->ds_objset != NULL);
1143 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
1144 panic("dirtying snapshot!");
1146 /* Must not dirty a dataset in the same txg where it got snapshotted. */
1147 ASSERT3U(tx->tx_txg, >, dsl_dataset_phys(ds)->ds_prev_snap_txg);
1149 dp = ds->ds_dir->dd_pool;
1150 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
1151 /* up the hold count until we can be written out */
1152 dmu_buf_add_ref(ds->ds_dbuf, ds);
1157 dsl_dataset_is_dirty(dsl_dataset_t *ds)
1159 for (int t = 0; t < TXG_SIZE; t++) {
1160 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1168 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1172 if (!dmu_tx_is_syncing(tx))
1176 * If there's an fs-only reservation, any blocks that might become
1177 * owned by the snapshot dataset must be accommodated by space
1178 * outside of the reservation.
1180 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1181 asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
1182 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1183 return (SET_ERROR(ENOSPC));
1186 * Propagate any reserved space for this snapshot to other
1187 * snapshot checks in this sync group.
1190 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1196 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1197 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
1202 ds->ds_trysnap_txg = tx->tx_txg;
1204 if (!dmu_tx_is_syncing(tx))
1208 * We don't allow multiple snapshots of the same txg. If there
1209 * is already one, try again.
1211 if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1212 return (SET_ERROR(EAGAIN));
1215 * Check for conflicting snapshot name.
1217 error = dsl_dataset_snap_lookup(ds, snapname, &value);
1219 return (SET_ERROR(EEXIST));
1220 if (error != ENOENT)
1224 * We don't allow taking snapshots of inconsistent datasets, such as
1225 * those into which we are currently receiving. However, if we are
1226 * creating this snapshot as part of a receive, this check will be
1227 * executed atomically with respect to the completion of the receive
1228 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1229 * case we ignore this, knowing it will be fixed up for us shortly in
1230 * dmu_recv_end_sync().
1232 if (!recv && DS_IS_INCONSISTENT(ds))
1233 return (SET_ERROR(EBUSY));
1236 * Skip the check for temporary snapshots or if we have already checked
1237 * the counts in dsl_dataset_snapshot_check. This means we really only
1238 * check the count here when we're receiving a stream.
1240 if (cnt != 0 && cr != NULL) {
1241 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1242 ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1247 error = dsl_dataset_snapshot_reserve_space(ds, tx);
1255 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1257 dsl_dataset_snapshot_arg_t *ddsa = arg;
1258 dsl_pool_t *dp = dmu_tx_pool(tx);
1263 * Pre-compute how many total new snapshots will be created for each
1264 * level in the tree and below. This is needed for validating the
1265 * snapshot limit when either taking a recursive snapshot or when
1266 * taking multiple snapshots.
1268 * The problem is that the counts are not actually adjusted when
1269 * we are checking, only when we finally sync. For a single snapshot,
1270 * this is easy, the count will increase by 1 at each node up the tree,
1271 * but its more complicated for the recursive/multiple snapshot case.
1273 * The dsl_fs_ss_limit_check function does recursively check the count
1274 * at each level up the tree but since it is validating each snapshot
1275 * independently we need to be sure that we are validating the complete
1276 * count for the entire set of snapshots. We do this by rolling up the
1277 * counts for each component of the name into an nvlist and then
1278 * checking each of those cases with the aggregated count.
1280 * This approach properly handles not only the recursive snapshot
1281 * case (where we get all of those on the ddsa_snaps list) but also
1282 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1283 * validate the limit on 'a' using a count of 2).
1285 * We validate the snapshot names in the third loop and only report
1288 if (dmu_tx_is_syncing(tx)) {
1289 nvlist_t *cnt_track = NULL;
1290 cnt_track = fnvlist_alloc();
1292 /* Rollup aggregated counts into the cnt_track list */
1293 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1295 pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1298 char nm[MAXPATHLEN];
1300 (void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1301 pdelim = strchr(nm, '@');
1307 if (nvlist_lookup_uint64(cnt_track, nm,
1309 /* update existing entry */
1310 fnvlist_add_uint64(cnt_track, nm,
1314 fnvlist_add_uint64(cnt_track, nm, 1);
1317 pdelim = strrchr(nm, '/');
1320 } while (pdelim != NULL);
1323 /* Check aggregated counts at each level */
1324 for (pair = nvlist_next_nvpair(cnt_track, NULL);
1325 pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1331 name = nvpair_name(pair);
1332 cnt = fnvpair_value_uint64(pair);
1335 error = dsl_dataset_hold(dp, name, FTAG, &ds);
1337 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1338 ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1340 dsl_dataset_rele(ds, FTAG);
1344 if (ddsa->ddsa_errors != NULL)
1345 fnvlist_add_int32(ddsa->ddsa_errors,
1348 /* only report one error for this check */
1352 nvlist_free(cnt_track);
1355 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1356 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1360 char dsname[ZFS_MAX_DATASET_NAME_LEN];
1362 name = nvpair_name(pair);
1363 if (strlen(name) >= ZFS_MAX_DATASET_NAME_LEN)
1364 error = SET_ERROR(ENAMETOOLONG);
1366 atp = strchr(name, '@');
1368 error = SET_ERROR(EINVAL);
1370 (void) strlcpy(dsname, name, atp - name + 1);
1373 error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1375 /* passing 0/NULL skips dsl_fs_ss_limit_check */
1376 error = dsl_dataset_snapshot_check_impl(ds,
1377 atp + 1, tx, B_FALSE, 0, NULL);
1378 dsl_dataset_rele(ds, FTAG);
1382 if (ddsa->ddsa_errors != NULL) {
1383 fnvlist_add_int32(ddsa->ddsa_errors,
1394 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1397 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1399 dsl_dataset_phys_t *dsphys;
1400 uint64_t dsobj, crtxg;
1401 objset_t *mos = dp->dp_meta_objset;
1404 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1407 * If we are on an old pool, the zil must not be active, in which
1408 * case it will be zeroed. Usually zil_suspend() accomplishes this.
1410 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1411 dmu_objset_from_ds(ds, &os) != 0 ||
1412 bcmp(&os->os_phys->os_zil_header, &zero_zil,
1413 sizeof (zero_zil)) == 0);
1415 /* Should not snapshot a dirty dataset. */
1416 ASSERT(!txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1419 dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1422 * The origin's ds_creation_txg has to be < TXG_INITIAL
1424 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1429 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1430 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1431 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1432 dmu_buf_will_dirty(dbuf, tx);
1433 dsphys = dbuf->db_data;
1434 bzero(dsphys, sizeof (dsl_dataset_phys_t));
1435 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1436 dsphys->ds_fsid_guid = unique_create();
1438 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1439 sizeof (dsphys->ds_guid));
1440 } while (dsphys->ds_guid == 0);
1441 dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1442 dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1443 dsphys->ds_next_snap_obj = ds->ds_object;
1444 dsphys->ds_num_children = 1;
1445 dsphys->ds_creation_time = gethrestime_sec();
1446 dsphys->ds_creation_txg = crtxg;
1447 dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1448 dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1449 dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1450 dsphys->ds_uncompressed_bytes =
1451 dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1452 dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1453 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
1454 dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1455 rrw_exit(&ds->ds_bp_rwlock, FTAG);
1456 dmu_buf_rele(dbuf, FTAG);
1458 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1459 if (ds->ds_feature_inuse[f])
1460 dsl_dataset_activate_feature(dsobj, f, tx);
1463 ASSERT3U(ds->ds_prev != 0, ==,
1464 dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1466 uint64_t next_clones_obj =
1467 dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1468 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1470 dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1471 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1473 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1474 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1475 dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1476 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1477 } else if (next_clones_obj != 0) {
1478 dsl_dataset_remove_from_next_clones(ds->ds_prev,
1479 dsphys->ds_next_snap_obj, tx);
1480 VERIFY0(zap_add_int(mos,
1481 next_clones_obj, dsobj, tx));
1486 * If we have a reference-reservation on this dataset, we will
1487 * need to increase the amount of refreservation being charged
1488 * since our unique space is going to zero.
1490 if (ds->ds_reserved) {
1492 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1493 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1495 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1499 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1500 dsl_dataset_phys(ds)->ds_deadlist_obj =
1501 dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1502 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1503 dsl_deadlist_close(&ds->ds_deadlist);
1504 dsl_deadlist_open(&ds->ds_deadlist, mos,
1505 dsl_dataset_phys(ds)->ds_deadlist_obj);
1506 dsl_deadlist_add_key(&ds->ds_deadlist,
1507 dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1509 if (dsl_dataset_remap_deadlist_exists(ds)) {
1510 uint64_t remap_deadlist_obj =
1511 dsl_dataset_get_remap_deadlist_object(ds);
1513 * Move the remap_deadlist to the snapshot. The head
1514 * will create a new remap deadlist on demand, from
1515 * dsl_dataset_block_remapped().
1517 dsl_dataset_unset_remap_deadlist_object(ds, tx);
1518 dsl_deadlist_close(&ds->ds_remap_deadlist);
1520 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
1521 VERIFY0(zap_add(mos, dsobj, DS_FIELD_REMAP_DEADLIST,
1522 sizeof (remap_deadlist_obj), 1, &remap_deadlist_obj, tx));
1525 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1526 dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1527 dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1528 dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1530 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1531 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1533 VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1534 snapname, 8, 1, &dsobj, tx));
1537 dsl_dataset_rele(ds->ds_prev, ds);
1538 VERIFY0(dsl_dataset_hold_obj(dp,
1539 dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1541 dsl_scan_ds_snapshotted(ds, tx);
1543 dsl_dir_snap_cmtime_update(ds->ds_dir);
1545 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1549 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1551 dsl_dataset_snapshot_arg_t *ddsa = arg;
1552 dsl_pool_t *dp = dmu_tx_pool(tx);
1555 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1556 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1559 char dsname[ZFS_MAX_DATASET_NAME_LEN];
1561 name = nvpair_name(pair);
1562 atp = strchr(name, '@');
1563 (void) strlcpy(dsname, name, atp - name + 1);
1564 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1566 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1567 if (ddsa->ddsa_props != NULL) {
1568 dsl_props_set_sync_impl(ds->ds_prev,
1569 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1571 dsl_dataset_rele(ds, FTAG);
1576 * The snapshots must all be in the same pool.
1577 * All-or-nothing: if there are any failures, nothing will be modified.
1580 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1582 dsl_dataset_snapshot_arg_t ddsa;
1584 boolean_t needsuspend;
1588 nvlist_t *suspended = NULL;
1590 pair = nvlist_next_nvpair(snaps, NULL);
1593 firstname = nvpair_name(pair);
1595 error = spa_open(firstname, &spa, FTAG);
1598 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1599 spa_close(spa, FTAG);
1602 suspended = fnvlist_alloc();
1603 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1604 pair = nvlist_next_nvpair(snaps, pair)) {
1605 char fsname[ZFS_MAX_DATASET_NAME_LEN];
1606 char *snapname = nvpair_name(pair);
1610 atp = strchr(snapname, '@');
1612 error = SET_ERROR(EINVAL);
1615 (void) strlcpy(fsname, snapname, atp - snapname + 1);
1617 error = zil_suspend(fsname, &cookie);
1620 fnvlist_add_uint64(suspended, fsname,
1625 ddsa.ddsa_snaps = snaps;
1626 ddsa.ddsa_props = props;
1627 ddsa.ddsa_errors = errors;
1628 ddsa.ddsa_cr = CRED();
1631 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1632 dsl_dataset_snapshot_sync, &ddsa,
1633 fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1636 if (suspended != NULL) {
1637 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1638 pair = nvlist_next_nvpair(suspended, pair)) {
1639 zil_resume((void *)(uintptr_t)
1640 fnvpair_value_uint64(pair));
1642 fnvlist_free(suspended);
1648 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1649 pair = nvlist_next_nvpair(snaps, pair)) {
1650 char *snapname = nvpair_name(pair);
1651 zvol_create_minors(snapname);
1659 typedef struct dsl_dataset_snapshot_tmp_arg {
1660 const char *ddsta_fsname;
1661 const char *ddsta_snapname;
1662 minor_t ddsta_cleanup_minor;
1663 const char *ddsta_htag;
1664 } dsl_dataset_snapshot_tmp_arg_t;
1667 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1669 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1670 dsl_pool_t *dp = dmu_tx_pool(tx);
1674 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1678 /* NULL cred means no limit check for tmp snapshot */
1679 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1680 tx, B_FALSE, 0, NULL);
1682 dsl_dataset_rele(ds, FTAG);
1686 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1687 dsl_dataset_rele(ds, FTAG);
1688 return (SET_ERROR(ENOTSUP));
1690 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1693 dsl_dataset_rele(ds, FTAG);
1697 dsl_dataset_rele(ds, FTAG);
1702 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1704 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1705 dsl_pool_t *dp = dmu_tx_pool(tx);
1708 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1710 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1711 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1712 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1713 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1715 dsl_dataset_rele(ds, FTAG);
1719 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1720 minor_t cleanup_minor, const char *htag)
1722 dsl_dataset_snapshot_tmp_arg_t ddsta;
1725 boolean_t needsuspend;
1728 ddsta.ddsta_fsname = fsname;
1729 ddsta.ddsta_snapname = snapname;
1730 ddsta.ddsta_cleanup_minor = cleanup_minor;
1731 ddsta.ddsta_htag = htag;
1733 error = spa_open(fsname, &spa, FTAG);
1736 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1737 spa_close(spa, FTAG);
1740 error = zil_suspend(fsname, &cookie);
1745 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1746 dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1754 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1756 ASSERT(dmu_tx_is_syncing(tx));
1757 ASSERT(ds->ds_objset != NULL);
1758 ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1761 * in case we had to change ds_fsid_guid when we opened it,
1764 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1765 dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1767 if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) {
1768 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1769 ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1,
1770 &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx));
1771 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1772 ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1,
1773 &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx));
1774 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1775 ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1,
1776 &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx));
1777 ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0;
1778 ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0;
1779 ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0;
1782 dmu_objset_sync(ds->ds_objset, zio, tx);
1784 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1785 if (ds->ds_feature_activation_needed[f]) {
1786 if (ds->ds_feature_inuse[f])
1788 dsl_dataset_activate_feature(ds->ds_object, f, tx);
1789 ds->ds_feature_inuse[f] = B_TRUE;
1795 deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
1797 dsl_deadlist_t *dl = arg;
1798 dsl_deadlist_insert(dl, bp, tx);
1803 dsl_dataset_sync_done(dsl_dataset_t *ds, dmu_tx_t *tx)
1805 objset_t *os = ds->ds_objset;
1807 bplist_iterate(&ds->ds_pending_deadlist,
1808 deadlist_enqueue_cb, &ds->ds_deadlist, tx);
1810 if (os->os_synced_dnodes != NULL) {
1811 multilist_destroy(os->os_synced_dnodes);
1812 os->os_synced_dnodes = NULL;
1815 ASSERT(!dmu_objset_is_dirty(os, dmu_tx_get_txg(tx)));
1817 dmu_buf_rele(ds->ds_dbuf, ds);
1821 get_clones_stat_impl(dsl_dataset_t *ds, nvlist_t *val)
1824 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1828 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1831 * There may be missing entries in ds_next_clones_obj
1832 * due to a bug in a previous version of the code.
1833 * Only trust it if it has the right number of entries.
1835 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1836 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1839 if (count != dsl_dataset_phys(ds)->ds_num_children - 1) {
1842 for (zap_cursor_init(&zc, mos,
1843 dsl_dataset_phys(ds)->ds_next_clones_obj);
1844 zap_cursor_retrieve(&zc, &za) == 0;
1845 zap_cursor_advance(&zc)) {
1846 dsl_dataset_t *clone;
1847 char buf[ZFS_MAX_DATASET_NAME_LEN];
1848 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1849 za.za_first_integer, FTAG, &clone));
1850 dsl_dir_name(clone->ds_dir, buf);
1851 fnvlist_add_boolean(val, buf);
1852 dsl_dataset_rele(clone, FTAG);
1854 zap_cursor_fini(&zc);
1859 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1861 nvlist_t *propval = fnvlist_alloc();
1865 * We use nvlist_alloc() instead of fnvlist_alloc() because the
1866 * latter would allocate the list with NV_UNIQUE_NAME flag.
1867 * As a result, every time a clone name is appended to the list
1868 * it would be (linearly) searched for for a duplicate name.
1869 * We already know that all clone names must be unique and we
1870 * want avoid the quadratic complexity of double-checking that
1871 * because we can have a large number of clones.
1873 VERIFY0(nvlist_alloc(&val, 0, KM_SLEEP));
1875 if (get_clones_stat_impl(ds, val) == 0) {
1876 fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1877 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES),
1882 nvlist_free(propval);
1886 * Returns a string that represents the receive resume stats token. It should
1887 * be freed with strfree().
1890 get_receive_resume_stats_impl(dsl_dataset_t *ds)
1892 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1894 if (dsl_dataset_has_resume_receive_state(ds)) {
1897 uint8_t *compressed;
1899 nvlist_t *token_nv = fnvlist_alloc();
1900 size_t packed_size, compressed_size;
1902 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1903 DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) {
1904 fnvlist_add_uint64(token_nv, "fromguid", val);
1906 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1907 DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) {
1908 fnvlist_add_uint64(token_nv, "object", val);
1910 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1911 DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) {
1912 fnvlist_add_uint64(token_nv, "offset", val);
1914 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1915 DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) {
1916 fnvlist_add_uint64(token_nv, "bytes", val);
1918 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1919 DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) {
1920 fnvlist_add_uint64(token_nv, "toguid", val);
1923 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1924 DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) {
1925 fnvlist_add_string(token_nv, "toname", buf);
1927 if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1928 DS_FIELD_RESUME_LARGEBLOCK) == 0) {
1929 fnvlist_add_boolean(token_nv, "largeblockok");
1931 if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1932 DS_FIELD_RESUME_EMBEDOK) == 0) {
1933 fnvlist_add_boolean(token_nv, "embedok");
1935 if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1936 DS_FIELD_RESUME_COMPRESSOK) == 0) {
1937 fnvlist_add_boolean(token_nv, "compressok");
1939 packed = fnvlist_pack(token_nv, &packed_size);
1940 fnvlist_free(token_nv);
1941 compressed = kmem_alloc(packed_size, KM_SLEEP);
1943 compressed_size = gzip_compress(packed, compressed,
1944 packed_size, packed_size, 6);
1947 fletcher_4_native(compressed, compressed_size, NULL, &cksum);
1949 str = kmem_alloc(compressed_size * 2 + 1, KM_SLEEP);
1950 for (int i = 0; i < compressed_size; i++) {
1951 (void) sprintf(str + i * 2, "%02x", compressed[i]);
1953 str[compressed_size * 2] = '\0';
1954 char *propval = kmem_asprintf("%u-%llx-%llx-%s",
1955 ZFS_SEND_RESUME_TOKEN_VERSION,
1956 (longlong_t)cksum.zc_word[0],
1957 (longlong_t)packed_size, str);
1958 kmem_free(packed, packed_size);
1959 kmem_free(str, compressed_size * 2 + 1);
1960 kmem_free(compressed, packed_size);
1963 return (spa_strdup(""));
1967 * Returns a string that represents the receive resume stats token of the
1968 * dataset's child. It should be freed with strfree().
1971 get_child_receive_stats(dsl_dataset_t *ds)
1973 char recvname[ZFS_MAX_DATASET_NAME_LEN + 6];
1974 dsl_dataset_t *recv_ds;
1975 dsl_dataset_name(ds, recvname);
1976 if (strlcat(recvname, "/", sizeof (recvname)) <
1977 sizeof (recvname) &&
1978 strlcat(recvname, recv_clone_name, sizeof (recvname)) <
1979 sizeof (recvname) &&
1980 dsl_dataset_hold(ds->ds_dir->dd_pool, recvname, FTAG,
1982 char *propval = get_receive_resume_stats_impl(recv_ds);
1983 dsl_dataset_rele(recv_ds, FTAG);
1986 return (spa_strdup(""));
1990 get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv)
1992 char *propval = get_receive_resume_stats_impl(ds);
1993 if (strcmp(propval, "") != 0) {
1994 dsl_prop_nvlist_add_string(nv,
1995 ZFS_PROP_RECEIVE_RESUME_TOKEN, propval);
1997 char *childval = get_child_receive_stats(ds);
1998 if (strcmp(childval, "") != 0) {
1999 dsl_prop_nvlist_add_string(nv,
2000 ZFS_PROP_RECEIVE_RESUME_TOKEN, childval);
2008 dsl_get_refratio(dsl_dataset_t *ds)
2010 uint64_t ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
2011 (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
2012 dsl_dataset_phys(ds)->ds_compressed_bytes);
2017 dsl_get_logicalreferenced(dsl_dataset_t *ds)
2019 return (dsl_dataset_phys(ds)->ds_uncompressed_bytes);
2023 dsl_get_compressratio(dsl_dataset_t *ds)
2025 if (ds->ds_is_snapshot) {
2026 return (dsl_get_refratio(ds));
2028 dsl_dir_t *dd = ds->ds_dir;
2029 mutex_enter(&dd->dd_lock);
2030 uint64_t val = dsl_dir_get_compressratio(dd);
2031 mutex_exit(&dd->dd_lock);
2037 dsl_get_used(dsl_dataset_t *ds)
2039 if (ds->ds_is_snapshot) {
2040 return (dsl_dataset_phys(ds)->ds_unique_bytes);
2042 dsl_dir_t *dd = ds->ds_dir;
2043 mutex_enter(&dd->dd_lock);
2044 uint64_t val = dsl_dir_get_used(dd);
2045 mutex_exit(&dd->dd_lock);
2051 dsl_get_creation(dsl_dataset_t *ds)
2053 return (dsl_dataset_phys(ds)->ds_creation_time);
2057 dsl_get_creationtxg(dsl_dataset_t *ds)
2059 return (dsl_dataset_phys(ds)->ds_creation_txg);
2063 dsl_get_refquota(dsl_dataset_t *ds)
2065 return (ds->ds_quota);
2069 dsl_get_refreservation(dsl_dataset_t *ds)
2071 return (ds->ds_reserved);
2075 dsl_get_guid(dsl_dataset_t *ds)
2077 return (dsl_dataset_phys(ds)->ds_guid);
2081 dsl_get_unique(dsl_dataset_t *ds)
2083 return (dsl_dataset_phys(ds)->ds_unique_bytes);
2087 dsl_get_objsetid(dsl_dataset_t *ds)
2089 return (ds->ds_object);
2093 dsl_get_userrefs(dsl_dataset_t *ds)
2095 return (ds->ds_userrefs);
2099 dsl_get_defer_destroy(dsl_dataset_t *ds)
2101 return (DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
2105 dsl_get_referenced(dsl_dataset_t *ds)
2107 return (dsl_dataset_phys(ds)->ds_referenced_bytes);
2111 dsl_get_numclones(dsl_dataset_t *ds)
2113 ASSERT(ds->ds_is_snapshot);
2114 return (dsl_dataset_phys(ds)->ds_num_children - 1);
2118 dsl_get_inconsistent(dsl_dataset_t *ds)
2120 return ((dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT) ?
2125 dsl_get_available(dsl_dataset_t *ds)
2127 uint64_t refdbytes = dsl_get_referenced(ds);
2128 uint64_t availbytes = dsl_dir_space_available(ds->ds_dir,
2130 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
2132 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
2134 if (ds->ds_quota != 0) {
2136 * Adjust available bytes according to refquota
2138 if (refdbytes < ds->ds_quota) {
2139 availbytes = MIN(availbytes,
2140 ds->ds_quota - refdbytes);
2145 return (availbytes);
2149 dsl_get_written(dsl_dataset_t *ds, uint64_t *written)
2151 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2152 dsl_dataset_t *prev;
2153 int err = dsl_dataset_hold_obj(dp,
2154 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
2156 uint64_t comp, uncomp;
2157 err = dsl_dataset_space_written(prev, ds, written,
2159 dsl_dataset_rele(prev, FTAG);
2165 * 'snap' should be a buffer of size ZFS_MAX_DATASET_NAME_LEN.
2168 dsl_get_prev_snap(dsl_dataset_t *ds, char *snap)
2170 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2171 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
2172 dsl_dataset_name(ds->ds_prev, snap);
2180 * Returns the mountpoint property and source for the given dataset in the value
2181 * and source buffers. The value buffer must be at least as large as MAXPATHLEN
2182 * and the source buffer as least as large a ZFS_MAX_DATASET_NAME_LEN.
2183 * Returns 0 on success and an error on failure.
2186 dsl_get_mountpoint(dsl_dataset_t *ds, const char *dsname, char *value,
2190 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2192 /* Retrieve the mountpoint value stored in the zap opbject */
2193 error = dsl_prop_get_ds(ds, zfs_prop_to_name(ZFS_PROP_MOUNTPOINT), 1,
2194 ZAP_MAXVALUELEN, value, source);
2199 /* Process the dsname and source to find the full mountpoint string */
2200 if (value[0] == '/') {
2201 char *buf = kmem_alloc(ZAP_MAXVALUELEN, KM_SLEEP);
2203 const char *relpath;
2206 * If we inherit the mountpoint, even from a dataset
2207 * with a received value, the source will be the path of
2208 * the dataset we inherit from. If source is
2209 * ZPROP_SOURCE_VAL_RECVD, the received value is not
2212 if (strcmp(source, ZPROP_SOURCE_VAL_RECVD) == 0) {
2215 ASSERT0(strncmp(dsname, source, strlen(source)));
2216 relpath = dsname + strlen(source);
2217 if (relpath[0] == '/')
2221 spa_altroot(dp->dp_spa, root, ZAP_MAXVALUELEN);
2224 * Special case an alternate root of '/'. This will
2225 * avoid having multiple leading slashes in the
2228 if (strcmp(root, "/") == 0)
2232 * If the mountpoint is '/' then skip over this
2233 * if we are obtaining either an alternate root or
2234 * an inherited mountpoint.
2237 if (value[1] == '\0' && (root[0] != '\0' ||
2238 relpath[0] != '\0'))
2241 if (relpath[0] == '\0') {
2242 (void) snprintf(value, ZAP_MAXVALUELEN, "%s%s",
2245 (void) snprintf(value, ZAP_MAXVALUELEN, "%s%s%s%s",
2246 root, mnt, relpath[0] == '@' ? "" : "/",
2249 kmem_free(buf, ZAP_MAXVALUELEN);
2251 /* 'legacy' or 'none' */
2252 (void) snprintf(value, ZAP_MAXVALUELEN, "%s", value);
2258 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
2260 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2262 ASSERT(dsl_pool_config_held(dp));
2264 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO,
2265 dsl_get_refratio(ds));
2266 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
2267 dsl_get_logicalreferenced(ds));
2268 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
2269 dsl_get_compressratio(ds));
2270 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
2273 if (ds->ds_is_snapshot) {
2274 get_clones_stat(ds, nv);
2276 char buf[ZFS_MAX_DATASET_NAME_LEN];
2277 if (dsl_get_prev_snap(ds, buf) == 0)
2278 dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP,
2280 dsl_dir_stats(ds->ds_dir, nv);
2283 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE,
2284 dsl_get_available(ds));
2285 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED,
2286 dsl_get_referenced(ds));
2287 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
2288 dsl_get_creation(ds));
2289 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
2290 dsl_get_creationtxg(ds));
2291 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
2292 dsl_get_refquota(ds));
2293 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
2294 dsl_get_refreservation(ds));
2295 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
2297 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
2298 dsl_get_unique(ds));
2299 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
2300 dsl_get_objsetid(ds));
2301 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
2302 dsl_get_userrefs(ds));
2303 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
2304 dsl_get_defer_destroy(ds));
2306 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
2308 if (dsl_get_written(ds, &written) == 0) {
2309 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
2314 if (!dsl_dataset_is_snapshot(ds)) {
2316 * A failed "newfs" (e.g. full) resumable receive leaves
2317 * the stats set on this dataset. Check here for the prop.
2319 get_receive_resume_stats(ds, nv);
2322 * A failed incremental resumable receive leaves the
2323 * stats set on our child named "%recv". Check the child
2326 /* 6 extra bytes for /%recv */
2327 char recvname[ZFS_MAX_DATASET_NAME_LEN + 6];
2328 dsl_dataset_t *recv_ds;
2329 dsl_dataset_name(ds, recvname);
2330 if (strlcat(recvname, "/", sizeof (recvname)) <
2331 sizeof (recvname) &&
2332 strlcat(recvname, recv_clone_name, sizeof (recvname)) <
2333 sizeof (recvname) &&
2334 dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) {
2335 get_receive_resume_stats(recv_ds, nv);
2336 dsl_dataset_rele(recv_ds, FTAG);
2342 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
2344 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2345 ASSERT(dsl_pool_config_held(dp));
2347 stat->dds_creation_txg = dsl_get_creationtxg(ds);
2348 stat->dds_inconsistent = dsl_get_inconsistent(ds);
2349 stat->dds_guid = dsl_get_guid(ds);
2350 stat->dds_origin[0] = '\0';
2351 if (ds->ds_is_snapshot) {
2352 stat->dds_is_snapshot = B_TRUE;
2353 stat->dds_num_clones = dsl_get_numclones(ds);
2355 stat->dds_is_snapshot = B_FALSE;
2356 stat->dds_num_clones = 0;
2358 if (dsl_dir_is_clone(ds->ds_dir)) {
2359 dsl_dir_get_origin(ds->ds_dir, stat->dds_origin);
2365 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2367 return (ds->ds_fsid_guid);
2371 dsl_dataset_space(dsl_dataset_t *ds,
2372 uint64_t *refdbytesp, uint64_t *availbytesp,
2373 uint64_t *usedobjsp, uint64_t *availobjsp)
2375 *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
2376 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2377 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
2379 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
2380 if (ds->ds_quota != 0) {
2382 * Adjust available bytes according to refquota
2384 if (*refdbytesp < ds->ds_quota)
2385 *availbytesp = MIN(*availbytesp,
2386 ds->ds_quota - *refdbytesp);
2390 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
2391 *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
2392 rrw_exit(&ds->ds_bp_rwlock, FTAG);
2393 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
2397 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
2399 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2402 ASSERT(dsl_pool_config_held(dp));
2405 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
2406 birth = dsl_dataset_get_blkptr(ds)->blk_birth;
2407 rrw_exit(&ds->ds_bp_rwlock, FTAG);
2408 if (birth > dsl_dataset_phys(snap)->ds_creation_txg) {
2409 objset_t *os, *os_snap;
2411 * It may be that only the ZIL differs, because it was
2412 * reset in the head. Don't count that as being
2415 if (dmu_objset_from_ds(ds, &os) != 0)
2417 if (dmu_objset_from_ds(snap, &os_snap) != 0)
2419 return (bcmp(&os->os_phys->os_meta_dnode,
2420 &os_snap->os_phys->os_meta_dnode,
2421 sizeof (os->os_phys->os_meta_dnode)) != 0);
2426 typedef struct dsl_dataset_rename_snapshot_arg {
2427 const char *ddrsa_fsname;
2428 const char *ddrsa_oldsnapname;
2429 const char *ddrsa_newsnapname;
2430 boolean_t ddrsa_recursive;
2432 } dsl_dataset_rename_snapshot_arg_t;
2436 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
2437 dsl_dataset_t *hds, void *arg)
2439 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2443 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2445 /* ignore nonexistent snapshots */
2446 return (error == ENOENT ? 0 : error);
2449 /* new name should not exist */
2450 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
2452 error = SET_ERROR(EEXIST);
2453 else if (error == ENOENT)
2456 /* dataset name + 1 for the "@" + the new snapshot name must fit */
2457 if (dsl_dir_namelen(hds->ds_dir) + 1 +
2458 strlen(ddrsa->ddrsa_newsnapname) >= ZFS_MAX_DATASET_NAME_LEN)
2459 error = SET_ERROR(ENAMETOOLONG);
2465 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
2467 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2468 dsl_pool_t *dp = dmu_tx_pool(tx);
2472 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
2476 if (ddrsa->ddrsa_recursive) {
2477 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2478 dsl_dataset_rename_snapshot_check_impl, ddrsa,
2481 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
2483 dsl_dataset_rele(hds, FTAG);
2488 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
2489 dsl_dataset_t *hds, void *arg)
2493 char *oldname, *newname;
2496 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2499 dmu_tx_t *tx = ddrsa->ddrsa_tx;
2502 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2503 ASSERT(error == 0 || error == ENOENT);
2504 if (error == ENOENT) {
2505 /* ignore nonexistent snapshots */
2509 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
2511 /* log before we change the name */
2512 spa_history_log_internal_ds(ds, "rename", tx,
2513 "-> @%s", ddrsa->ddrsa_newsnapname);
2515 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
2517 mutex_enter(&ds->ds_lock);
2518 (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
2519 mutex_exit(&ds->ds_lock);
2520 VERIFY0(zap_add(dp->dp_meta_objset,
2521 dsl_dataset_phys(hds)->ds_snapnames_zapobj,
2522 ds->ds_snapname, 8, 1, &ds->ds_object, tx));
2526 oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2527 newname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2528 snprintf(oldname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname,
2529 ddrsa->ddrsa_oldsnapname);
2530 snprintf(newname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname,
2531 ddrsa->ddrsa_newsnapname);
2532 zfsvfs_update_fromname(oldname, newname);
2533 zvol_rename_minors(oldname, newname);
2534 kmem_free(newname, MAXPATHLEN);
2535 kmem_free(oldname, MAXPATHLEN);
2538 dsl_dataset_rele(ds, FTAG);
2544 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
2546 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2547 dsl_pool_t *dp = dmu_tx_pool(tx);
2550 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
2551 ddrsa->ddrsa_tx = tx;
2552 if (ddrsa->ddrsa_recursive) {
2553 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2554 dsl_dataset_rename_snapshot_sync_impl, ddrsa,
2557 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
2559 dsl_dataset_rele(hds, FTAG);
2563 dsl_dataset_rename_snapshot(const char *fsname,
2564 const char *oldsnapname, const char *newsnapname, boolean_t recursive)
2566 dsl_dataset_rename_snapshot_arg_t ddrsa;
2568 ddrsa.ddrsa_fsname = fsname;
2569 ddrsa.ddrsa_oldsnapname = oldsnapname;
2570 ddrsa.ddrsa_newsnapname = newsnapname;
2571 ddrsa.ddrsa_recursive = recursive;
2573 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
2574 dsl_dataset_rename_snapshot_sync, &ddrsa,
2575 1, ZFS_SPACE_CHECK_RESERVED));
2579 * If we're doing an ownership handoff, we need to make sure that there is
2580 * only one long hold on the dataset. We're not allowed to change anything here
2581 * so we don't permanently release the long hold or regular hold here. We want
2582 * to do this only when syncing to avoid the dataset unexpectedly going away
2583 * when we release the long hold.
2586 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
2590 if (!dmu_tx_is_syncing(tx))
2593 if (owner != NULL) {
2594 VERIFY3P(ds->ds_owner, ==, owner);
2595 dsl_dataset_long_rele(ds, owner);
2598 held = dsl_dataset_long_held(ds);
2601 dsl_dataset_long_hold(ds, owner);
2604 return (SET_ERROR(EBUSY));
2610 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
2612 dsl_dataset_rollback_arg_t *ddra = arg;
2613 dsl_pool_t *dp = dmu_tx_pool(tx);
2615 int64_t unused_refres_delta;
2618 error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
2622 /* must not be a snapshot */
2623 if (ds->ds_is_snapshot) {
2624 dsl_dataset_rele(ds, FTAG);
2625 return (SET_ERROR(EINVAL));
2628 /* must have a most recent snapshot */
2629 if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
2630 dsl_dataset_rele(ds, FTAG);
2631 return (SET_ERROR(ESRCH));
2635 * No rollback to a snapshot created in the current txg, because
2636 * the rollback may dirty the dataset and create blocks that are
2637 * not reachable from the rootbp while having a birth txg that
2638 * falls into the snapshot's range.
2640 if (dmu_tx_is_syncing(tx) &&
2641 dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) {
2642 dsl_dataset_rele(ds, FTAG);
2643 return (SET_ERROR(EAGAIN));
2647 * If the expected target snapshot is specified, then check that
2648 * the latest snapshot is it.
2650 if (ddra->ddra_tosnap != NULL) {
2651 dsl_dataset_t *snapds;
2653 /* Check if the target snapshot exists at all. */
2654 error = dsl_dataset_hold(dp, ddra->ddra_tosnap, FTAG, &snapds);
2657 * ESRCH is used to signal that the target snapshot does
2658 * not exist, while ENOENT is used to report that
2659 * the rolled back dataset does not exist.
2660 * ESRCH is also used to cover other cases where the
2661 * target snapshot is not related to the dataset being
2662 * rolled back such as being in a different pool.
2664 if (error == ENOENT || error == EXDEV)
2665 error = SET_ERROR(ESRCH);
2666 dsl_dataset_rele(ds, FTAG);
2669 ASSERT(snapds->ds_is_snapshot);
2671 /* Check if the snapshot is the latest snapshot indeed. */
2672 if (snapds != ds->ds_prev) {
2674 * Distinguish between the case where the only problem
2675 * is intervening snapshots (EEXIST) vs the snapshot
2676 * not being a valid target for rollback (ESRCH).
2678 if (snapds->ds_dir == ds->ds_dir ||
2679 (dsl_dir_is_clone(ds->ds_dir) &&
2680 dsl_dir_phys(ds->ds_dir)->dd_origin_obj ==
2681 snapds->ds_object)) {
2682 error = SET_ERROR(EEXIST);
2684 error = SET_ERROR(ESRCH);
2686 dsl_dataset_rele(snapds, FTAG);
2687 dsl_dataset_rele(ds, FTAG);
2690 dsl_dataset_rele(snapds, FTAG);
2693 /* must not have any bookmarks after the most recent snapshot */
2694 nvlist_t *proprequest = fnvlist_alloc();
2695 fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
2696 nvlist_t *bookmarks = fnvlist_alloc();
2697 error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
2698 fnvlist_free(proprequest);
2700 dsl_dataset_rele(ds, FTAG);
2703 for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
2704 pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
2706 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
2707 zfs_prop_to_name(ZFS_PROP_CREATETXG));
2708 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
2709 if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
2710 fnvlist_free(bookmarks);
2711 dsl_dataset_rele(ds, FTAG);
2712 return (SET_ERROR(EEXIST));
2715 fnvlist_free(bookmarks);
2717 error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
2719 dsl_dataset_rele(ds, FTAG);
2724 * Check if the snap we are rolling back to uses more than
2727 if (ds->ds_quota != 0 &&
2728 dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
2729 dsl_dataset_rele(ds, FTAG);
2730 return (SET_ERROR(EDQUOT));
2734 * When we do the clone swap, we will temporarily use more space
2735 * due to the refreservation (the head will no longer have any
2736 * unique space, so the entire amount of the refreservation will need
2737 * to be free). We will immediately destroy the clone, freeing
2738 * this space, but the freeing happens over many txg's.
2740 unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2741 dsl_dataset_phys(ds)->ds_unique_bytes);
2743 if (unused_refres_delta > 0 &&
2744 unused_refres_delta >
2745 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2746 dsl_dataset_rele(ds, FTAG);
2747 return (SET_ERROR(ENOSPC));
2750 dsl_dataset_rele(ds, FTAG);
2755 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2757 dsl_dataset_rollback_arg_t *ddra = arg;
2758 dsl_pool_t *dp = dmu_tx_pool(tx);
2759 dsl_dataset_t *ds, *clone;
2761 char namebuf[ZFS_MAX_DATASET_NAME_LEN];
2763 VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2765 dsl_dataset_name(ds->ds_prev, namebuf);
2766 fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2768 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2769 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2771 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2773 dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2774 dsl_dataset_zero_zil(ds, tx);
2776 dsl_destroy_head_sync_impl(clone, tx);
2778 dsl_dataset_rele(clone, FTAG);
2779 dsl_dataset_rele(ds, FTAG);
2783 * Rolls back the given filesystem or volume to the most recent snapshot.
2784 * The name of the most recent snapshot will be returned under key "target"
2785 * in the result nvlist.
2788 * - The existing dataset MUST be owned by the specified owner at entry
2789 * - Upon return, dataset will still be held by the same owner, whether we
2792 * This mode is required any time the existing filesystem is mounted. See
2793 * notes above zfs_suspend_fs() for further details.
2796 dsl_dataset_rollback(const char *fsname, const char *tosnap, void *owner,
2799 dsl_dataset_rollback_arg_t ddra;
2801 ddra.ddra_fsname = fsname;
2802 ddra.ddra_tosnap = tosnap;
2803 ddra.ddra_owner = owner;
2804 ddra.ddra_result = result;
2806 return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2807 dsl_dataset_rollback_sync, &ddra,
2808 1, ZFS_SPACE_CHECK_RESERVED));
2811 struct promotenode {
2816 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2817 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2819 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2822 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2824 dsl_dataset_promote_arg_t *ddpa = arg;
2825 dsl_pool_t *dp = dmu_tx_pool(tx);
2827 struct promotenode *snap;
2828 dsl_dataset_t *origin_ds;
2832 size_t max_snap_len;
2833 boolean_t conflicting_snaps;
2835 err = promote_hold(ddpa, dp, FTAG);
2839 hds = ddpa->ddpa_clone;
2840 snap = list_head(&ddpa->shared_snaps);
2841 origin_ds = snap->ds;
2842 max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
2844 snap = list_head(&ddpa->origin_snaps);
2846 if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2847 promote_rele(ddpa, FTAG);
2848 return (SET_ERROR(EXDEV));
2852 * Compute and check the amount of space to transfer. Since this is
2853 * so expensive, don't do the preliminary check.
2855 if (!dmu_tx_is_syncing(tx)) {
2856 promote_rele(ddpa, FTAG);
2860 /* compute origin's new unique space */
2861 snap = list_tail(&ddpa->clone_snaps);
2862 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2863 origin_ds->ds_object);
2864 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2865 dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2866 &ddpa->unique, &unused, &unused);
2869 * Walk the snapshots that we are moving
2871 * Compute space to transfer. Consider the incremental changes
2872 * to used by each snapshot:
2873 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2874 * So each snapshot gave birth to:
2875 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2876 * So a sequence would look like:
2877 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2878 * Which simplifies to:
2879 * uN + kN + kN-1 + ... + k1 + k0
2880 * Note however, if we stop before we reach the ORIGIN we get:
2881 * uN + kN + kN-1 + ... + kM - uM-1
2883 conflicting_snaps = B_FALSE;
2885 ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2886 ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2887 ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2888 for (snap = list_head(&ddpa->shared_snaps); snap;
2889 snap = list_next(&ddpa->shared_snaps, snap)) {
2890 uint64_t val, dlused, dlcomp, dluncomp;
2891 dsl_dataset_t *ds = snap->ds;
2896 * If there are long holds, we won't be able to evict
2899 if (dsl_dataset_long_held(ds)) {
2900 err = SET_ERROR(EBUSY);
2904 /* Check that the snapshot name does not conflict */
2905 VERIFY0(dsl_dataset_get_snapname(ds));
2906 if (strlen(ds->ds_snapname) >= max_snap_len) {
2907 err = SET_ERROR(ENAMETOOLONG);
2910 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2912 fnvlist_add_boolean(ddpa->err_ds,
2913 snap->ds->ds_snapname);
2914 conflicting_snaps = B_TRUE;
2915 } else if (err != ENOENT) {
2919 /* The very first snapshot does not have a deadlist */
2920 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2923 dsl_deadlist_space(&ds->ds_deadlist,
2924 &dlused, &dlcomp, &dluncomp);
2925 ddpa->used += dlused;
2926 ddpa->comp += dlcomp;
2927 ddpa->uncomp += dluncomp;
2931 * In order to return the full list of conflicting snapshots, we check
2932 * whether there was a conflict after traversing all of them.
2934 if (conflicting_snaps) {
2935 err = SET_ERROR(EEXIST);
2940 * If we are a clone of a clone then we never reached ORIGIN,
2941 * so we need to subtract out the clone origin's used space.
2943 if (ddpa->origin_origin) {
2945 dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2947 dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2949 dsl_dataset_phys(ddpa->origin_origin)->
2950 ds_uncompressed_bytes;
2953 /* Check that there is enough space and limit headroom here */
2954 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2955 0, ss_mv_cnt, ddpa->used, ddpa->cr);
2960 * Compute the amounts of space that will be used by snapshots
2961 * after the promotion (for both origin and clone). For each,
2962 * it is the amount of space that will be on all of their
2963 * deadlists (that was not born before their new origin).
2965 if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2969 * Note, typically this will not be a clone of a clone,
2970 * so dd_origin_txg will be < TXG_INITIAL, so
2971 * these snaplist_space() -> dsl_deadlist_space_range()
2972 * calls will be fast because they do not have to
2973 * iterate over all bps.
2975 snap = list_head(&ddpa->origin_snaps);
2976 err = snaplist_space(&ddpa->shared_snaps,
2977 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2981 err = snaplist_space(&ddpa->clone_snaps,
2982 snap->ds->ds_dir->dd_origin_txg, &space);
2985 ddpa->cloneusedsnap += space;
2987 if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2988 DD_FLAG_USED_BREAKDOWN) {
2989 err = snaplist_space(&ddpa->origin_snaps,
2990 dsl_dataset_phys(origin_ds)->ds_creation_txg,
2991 &ddpa->originusedsnap);
2997 promote_rele(ddpa, FTAG);
3002 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
3004 dsl_dataset_promote_arg_t *ddpa = arg;
3005 dsl_pool_t *dp = dmu_tx_pool(tx);
3007 struct promotenode *snap;
3008 dsl_dataset_t *origin_ds;
3009 dsl_dataset_t *origin_head;
3011 dsl_dir_t *odd = NULL;
3012 uint64_t oldnext_obj;
3014 #if defined(__FreeBSD__) && defined(_KERNEL)
3015 char *oldname, *newname;
3018 VERIFY0(promote_hold(ddpa, dp, FTAG));
3019 hds = ddpa->ddpa_clone;
3021 ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
3023 snap = list_head(&ddpa->shared_snaps);
3024 origin_ds = snap->ds;
3027 snap = list_head(&ddpa->origin_snaps);
3028 origin_head = snap->ds;
3031 * We need to explicitly open odd, since origin_ds's dd will be
3034 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
3037 /* change origin's next snap */
3038 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
3039 oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
3040 snap = list_tail(&ddpa->clone_snaps);
3041 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
3042 origin_ds->ds_object);
3043 dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
3045 /* change the origin's next clone */
3046 if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
3047 dsl_dataset_remove_from_next_clones(origin_ds,
3048 snap->ds->ds_object, tx);
3049 VERIFY0(zap_add_int(dp->dp_meta_objset,
3050 dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
3055 dmu_buf_will_dirty(dd->dd_dbuf, tx);
3056 ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
3057 dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
3058 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
3059 dmu_buf_will_dirty(odd->dd_dbuf, tx);
3060 dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
3061 origin_head->ds_dir->dd_origin_txg =
3062 dsl_dataset_phys(origin_ds)->ds_creation_txg;
3064 /* change dd_clone entries */
3065 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
3066 VERIFY0(zap_remove_int(dp->dp_meta_objset,
3067 dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
3068 VERIFY0(zap_add_int(dp->dp_meta_objset,
3069 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
3070 hds->ds_object, tx));
3072 VERIFY0(zap_remove_int(dp->dp_meta_objset,
3073 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
3074 origin_head->ds_object, tx));
3075 if (dsl_dir_phys(dd)->dd_clones == 0) {
3076 dsl_dir_phys(dd)->dd_clones =
3077 zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
3078 DMU_OT_NONE, 0, tx);
3080 VERIFY0(zap_add_int(dp->dp_meta_objset,
3081 dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
3084 #if defined(__FreeBSD__) && defined(_KERNEL)
3085 /* Take the spa_namespace_lock early so zvol renames don't deadlock. */
3086 mutex_enter(&spa_namespace_lock);
3088 oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3089 newname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3092 /* move snapshots to this dir */
3093 for (snap = list_head(&ddpa->shared_snaps); snap;
3094 snap = list_next(&ddpa->shared_snaps, snap)) {
3095 dsl_dataset_t *ds = snap->ds;
3098 * Property callbacks are registered to a particular
3099 * dsl_dir. Since ours is changing, evict the objset
3100 * so that they will be unregistered from the old dsl_dir.
3102 if (ds->ds_objset) {
3103 dmu_objset_evict(ds->ds_objset);
3104 ds->ds_objset = NULL;
3107 /* move snap name entry */
3108 VERIFY0(dsl_dataset_get_snapname(ds));
3109 VERIFY0(dsl_dataset_snap_remove(origin_head,
3110 ds->ds_snapname, tx, B_TRUE));
3111 VERIFY0(zap_add(dp->dp_meta_objset,
3112 dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
3113 8, 1, &ds->ds_object, tx));
3114 dsl_fs_ss_count_adjust(hds->ds_dir, 1,
3115 DD_FIELD_SNAPSHOT_COUNT, tx);
3117 /* change containing dsl_dir */
3118 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3119 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
3120 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
3121 ASSERT3P(ds->ds_dir, ==, odd);
3122 dsl_dir_rele(ds->ds_dir, ds);
3123 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
3124 NULL, ds, &ds->ds_dir));
3126 #if defined(__FreeBSD__) && defined(_KERNEL)
3127 dsl_dataset_name(ds, newname);
3128 zfsvfs_update_fromname(oldname, newname);
3129 zvol_rename_minors(oldname, newname);
3132 /* move any clone references */
3133 if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
3134 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
3138 for (zap_cursor_init(&zc, dp->dp_meta_objset,
3139 dsl_dataset_phys(ds)->ds_next_clones_obj);
3140 zap_cursor_retrieve(&zc, &za) == 0;
3141 zap_cursor_advance(&zc)) {
3142 dsl_dataset_t *cnds;
3145 if (za.za_first_integer == oldnext_obj) {
3147 * We've already moved the
3148 * origin's reference.
3153 VERIFY0(dsl_dataset_hold_obj(dp,
3154 za.za_first_integer, FTAG, &cnds));
3155 o = dsl_dir_phys(cnds->ds_dir)->
3156 dd_head_dataset_obj;
3158 VERIFY0(zap_remove_int(dp->dp_meta_objset,
3159 dsl_dir_phys(odd)->dd_clones, o, tx));
3160 VERIFY0(zap_add_int(dp->dp_meta_objset,
3161 dsl_dir_phys(dd)->dd_clones, o, tx));
3162 dsl_dataset_rele(cnds, FTAG);
3164 zap_cursor_fini(&zc);
3167 ASSERT(!dsl_prop_hascb(ds));
3170 #if defined(__FreeBSD__) && defined(_KERNEL)
3171 mutex_exit(&spa_namespace_lock);
3173 kmem_free(newname, MAXPATHLEN);
3174 kmem_free(oldname, MAXPATHLEN);
3177 * Change space accounting.
3178 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
3179 * both be valid, or both be 0 (resulting in delta == 0). This
3180 * is true for each of {clone,origin} independently.
3183 delta = ddpa->cloneusedsnap -
3184 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
3185 ASSERT3S(delta, >=, 0);
3186 ASSERT3U(ddpa->used, >=, delta);
3187 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
3188 dsl_dir_diduse_space(dd, DD_USED_HEAD,
3189 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
3191 delta = ddpa->originusedsnap -
3192 dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
3193 ASSERT3S(delta, <=, 0);
3194 ASSERT3U(ddpa->used, >=, -delta);
3195 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
3196 dsl_dir_diduse_space(odd, DD_USED_HEAD,
3197 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
3199 dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
3201 /* log history record */
3202 spa_history_log_internal_ds(hds, "promote", tx, "");
3204 dsl_dir_rele(odd, FTAG);
3205 promote_rele(ddpa, FTAG);
3209 * Make a list of dsl_dataset_t's for the snapshots between first_obj
3210 * (exclusive) and last_obj (inclusive). The list will be in reverse
3211 * order (last_obj will be the list_head()). If first_obj == 0, do all
3212 * snapshots back to this dataset's origin.
3215 snaplist_make(dsl_pool_t *dp,
3216 uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
3218 uint64_t obj = last_obj;
3220 list_create(l, sizeof (struct promotenode),
3221 offsetof(struct promotenode, link));
3223 while (obj != first_obj) {
3225 struct promotenode *snap;
3228 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
3229 ASSERT(err != ENOENT);
3234 first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
3236 snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
3238 list_insert_tail(l, snap);
3239 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3246 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
3248 struct promotenode *snap;
3251 for (snap = list_head(l); snap; snap = list_next(l, snap)) {
3252 uint64_t used, comp, uncomp;
3253 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
3254 mintxg, UINT64_MAX, &used, &comp, &uncomp);
3261 snaplist_destroy(list_t *l, void *tag)
3263 struct promotenode *snap;
3265 if (l == NULL || !list_link_active(&l->list_head))
3268 while ((snap = list_tail(l)) != NULL) {
3269 list_remove(l, snap);
3270 dsl_dataset_rele(snap->ds, tag);
3271 kmem_free(snap, sizeof (*snap));
3277 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
3281 struct promotenode *snap;
3283 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
3287 dd = ddpa->ddpa_clone->ds_dir;
3289 if (ddpa->ddpa_clone->ds_is_snapshot ||
3290 !dsl_dir_is_clone(dd)) {
3291 dsl_dataset_rele(ddpa->ddpa_clone, tag);
3292 return (SET_ERROR(EINVAL));
3295 error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
3296 &ddpa->shared_snaps, tag);
3300 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
3301 &ddpa->clone_snaps, tag);
3305 snap = list_head(&ddpa->shared_snaps);
3306 ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
3307 error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
3308 dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
3309 &ddpa->origin_snaps, tag);
3313 if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
3314 error = dsl_dataset_hold_obj(dp,
3315 dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
3316 tag, &ddpa->origin_origin);
3322 promote_rele(ddpa, tag);
3327 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
3329 snaplist_destroy(&ddpa->shared_snaps, tag);
3330 snaplist_destroy(&ddpa->clone_snaps, tag);
3331 snaplist_destroy(&ddpa->origin_snaps, tag);
3332 if (ddpa->origin_origin != NULL)
3333 dsl_dataset_rele(ddpa->origin_origin, tag);
3334 dsl_dataset_rele(ddpa->ddpa_clone, tag);
3340 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
3341 * in with the name. (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.)
3344 dsl_dataset_promote(const char *name, char *conflsnap)
3346 dsl_dataset_promote_arg_t ddpa = { 0 };
3349 nvpair_t *snap_pair;
3353 * We will modify space proportional to the number of
3354 * snapshots. Compute numsnaps.
3356 error = dmu_objset_hold(name, FTAG, &os);
3359 error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
3360 dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
3362 dmu_objset_rele(os, FTAG);
3366 ddpa.ddpa_clonename = name;
3367 ddpa.err_ds = fnvlist_alloc();
3370 error = dsl_sync_task(name, dsl_dataset_promote_check,
3371 dsl_dataset_promote_sync, &ddpa,
3372 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED);
3375 * Return the first conflicting snapshot found.
3377 snap_pair = nvlist_next_nvpair(ddpa.err_ds, NULL);
3378 if (snap_pair != NULL && conflsnap != NULL)
3379 (void) strcpy(conflsnap, nvpair_name(snap_pair));
3381 fnvlist_free(ddpa.err_ds);
3386 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
3387 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
3390 * "slack" factor for received datasets with refquota set on them.
3391 * See the bottom of this function for details on its use.
3393 uint64_t refquota_slack = DMU_MAX_ACCESS * spa_asize_inflation;
3394 int64_t unused_refres_delta;
3396 /* they should both be heads */
3397 if (clone->ds_is_snapshot ||
3398 origin_head->ds_is_snapshot)
3399 return (SET_ERROR(EINVAL));
3401 /* if we are not forcing, the branch point should be just before them */
3402 if (!force && clone->ds_prev != origin_head->ds_prev)
3403 return (SET_ERROR(EINVAL));
3405 /* clone should be the clone (unless they are unrelated) */
3406 if (clone->ds_prev != NULL &&
3407 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
3408 origin_head->ds_dir != clone->ds_prev->ds_dir)
3409 return (SET_ERROR(EINVAL));
3411 /* the clone should be a child of the origin */
3412 if (clone->ds_dir->dd_parent != origin_head->ds_dir)
3413 return (SET_ERROR(EINVAL));
3415 /* origin_head shouldn't be modified unless 'force' */
3417 dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
3418 return (SET_ERROR(ETXTBSY));
3420 /* origin_head should have no long holds (e.g. is not mounted) */
3421 if (dsl_dataset_handoff_check(origin_head, owner, tx))
3422 return (SET_ERROR(EBUSY));
3424 /* check amount of any unconsumed refreservation */
3425 unused_refres_delta =
3426 (int64_t)MIN(origin_head->ds_reserved,
3427 dsl_dataset_phys(origin_head)->ds_unique_bytes) -
3428 (int64_t)MIN(origin_head->ds_reserved,
3429 dsl_dataset_phys(clone)->ds_unique_bytes);
3431 if (unused_refres_delta > 0 &&
3432 unused_refres_delta >
3433 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
3434 return (SET_ERROR(ENOSPC));
3437 * The clone can't be too much over the head's refquota.
3439 * To ensure that the entire refquota can be used, we allow one
3440 * transaction to exceed the the refquota. Therefore, this check
3441 * needs to also allow for the space referenced to be more than the
3442 * refquota. The maximum amount of space that one transaction can use
3443 * on disk is DMU_MAX_ACCESS * spa_asize_inflation. Allowing this
3444 * overage ensures that we are able to receive a filesystem that
3445 * exceeds the refquota on the source system.
3447 * So that overage is the refquota_slack we use below.
3449 if (origin_head->ds_quota != 0 &&
3450 dsl_dataset_phys(clone)->ds_referenced_bytes >
3451 origin_head->ds_quota + refquota_slack)
3452 return (SET_ERROR(EDQUOT));
3458 dsl_dataset_swap_remap_deadlists(dsl_dataset_t *clone,
3459 dsl_dataset_t *origin, dmu_tx_t *tx)
3461 uint64_t clone_remap_dl_obj, origin_remap_dl_obj;
3462 dsl_pool_t *dp = dmu_tx_pool(tx);
3464 ASSERT(dsl_pool_sync_context(dp));
3466 clone_remap_dl_obj = dsl_dataset_get_remap_deadlist_object(clone);
3467 origin_remap_dl_obj = dsl_dataset_get_remap_deadlist_object(origin);
3469 if (clone_remap_dl_obj != 0) {
3470 dsl_deadlist_close(&clone->ds_remap_deadlist);
3471 dsl_dataset_unset_remap_deadlist_object(clone, tx);
3473 if (origin_remap_dl_obj != 0) {
3474 dsl_deadlist_close(&origin->ds_remap_deadlist);
3475 dsl_dataset_unset_remap_deadlist_object(origin, tx);
3478 if (clone_remap_dl_obj != 0) {
3479 dsl_dataset_set_remap_deadlist_object(origin,
3480 clone_remap_dl_obj, tx);
3481 dsl_deadlist_open(&origin->ds_remap_deadlist,
3482 dp->dp_meta_objset, clone_remap_dl_obj);
3484 if (origin_remap_dl_obj != 0) {
3485 dsl_dataset_set_remap_deadlist_object(clone,
3486 origin_remap_dl_obj, tx);
3487 dsl_deadlist_open(&clone->ds_remap_deadlist,
3488 dp->dp_meta_objset, origin_remap_dl_obj);
3493 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
3494 dsl_dataset_t *origin_head, dmu_tx_t *tx)
3496 dsl_pool_t *dp = dmu_tx_pool(tx);
3497 int64_t unused_refres_delta;
3499 ASSERT(clone->ds_reserved == 0);
3501 * NOTE: On DEBUG kernels there could be a race between this and
3502 * the check function if spa_asize_inflation is adjusted...
3504 ASSERT(origin_head->ds_quota == 0 ||
3505 dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota +
3506 DMU_MAX_ACCESS * spa_asize_inflation);
3507 ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
3510 * Swap per-dataset feature flags.
3512 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
3513 if (!(spa_feature_table[f].fi_flags &
3514 ZFEATURE_FLAG_PER_DATASET)) {
3515 ASSERT(!clone->ds_feature_inuse[f]);
3516 ASSERT(!origin_head->ds_feature_inuse[f]);
3520 boolean_t clone_inuse = clone->ds_feature_inuse[f];
3521 boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f];
3524 dsl_dataset_deactivate_feature(clone->ds_object, f, tx);
3525 clone->ds_feature_inuse[f] = B_FALSE;
3527 if (origin_head_inuse) {
3528 dsl_dataset_deactivate_feature(origin_head->ds_object,
3530 origin_head->ds_feature_inuse[f] = B_FALSE;
3533 dsl_dataset_activate_feature(origin_head->ds_object,
3535 origin_head->ds_feature_inuse[f] = B_TRUE;
3537 if (origin_head_inuse) {
3538 dsl_dataset_activate_feature(clone->ds_object, f, tx);
3539 clone->ds_feature_inuse[f] = B_TRUE;
3543 dmu_buf_will_dirty(clone->ds_dbuf, tx);
3544 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
3546 if (clone->ds_objset != NULL) {
3547 dmu_objset_evict(clone->ds_objset);
3548 clone->ds_objset = NULL;
3551 if (origin_head->ds_objset != NULL) {
3552 dmu_objset_evict(origin_head->ds_objset);
3553 origin_head->ds_objset = NULL;
3556 unused_refres_delta =
3557 (int64_t)MIN(origin_head->ds_reserved,
3558 dsl_dataset_phys(origin_head)->ds_unique_bytes) -
3559 (int64_t)MIN(origin_head->ds_reserved,
3560 dsl_dataset_phys(clone)->ds_unique_bytes);
3563 * Reset origin's unique bytes, if it exists.
3565 if (clone->ds_prev) {
3566 dsl_dataset_t *origin = clone->ds_prev;
3567 uint64_t comp, uncomp;
3569 dmu_buf_will_dirty(origin->ds_dbuf, tx);
3570 dsl_deadlist_space_range(&clone->ds_deadlist,
3571 dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
3572 &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
3577 rrw_enter(&clone->ds_bp_rwlock, RW_WRITER, FTAG);
3578 rrw_enter(&origin_head->ds_bp_rwlock, RW_WRITER, FTAG);
3580 tmp = dsl_dataset_phys(origin_head)->ds_bp;
3581 dsl_dataset_phys(origin_head)->ds_bp =
3582 dsl_dataset_phys(clone)->ds_bp;
3583 dsl_dataset_phys(clone)->ds_bp = tmp;
3584 rrw_exit(&origin_head->ds_bp_rwlock, FTAG);
3585 rrw_exit(&clone->ds_bp_rwlock, FTAG);
3588 /* set dd_*_bytes */
3590 int64_t dused, dcomp, duncomp;
3591 uint64_t cdl_used, cdl_comp, cdl_uncomp;
3592 uint64_t odl_used, odl_comp, odl_uncomp;
3594 ASSERT3U(dsl_dir_phys(clone->ds_dir)->
3595 dd_used_breakdown[DD_USED_SNAP], ==, 0);
3597 dsl_deadlist_space(&clone->ds_deadlist,
3598 &cdl_used, &cdl_comp, &cdl_uncomp);
3599 dsl_deadlist_space(&origin_head->ds_deadlist,
3600 &odl_used, &odl_comp, &odl_uncomp);
3602 dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
3604 (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
3606 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
3608 (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
3610 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
3612 (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
3615 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
3616 dused, dcomp, duncomp, tx);
3617 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
3618 -dused, -dcomp, -duncomp, tx);
3621 * The difference in the space used by snapshots is the
3622 * difference in snapshot space due to the head's
3623 * deadlist (since that's the only thing that's
3624 * changing that affects the snapused).
3626 dsl_deadlist_space_range(&clone->ds_deadlist,
3627 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3628 &cdl_used, &cdl_comp, &cdl_uncomp);
3629 dsl_deadlist_space_range(&origin_head->ds_deadlist,
3630 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3631 &odl_used, &odl_comp, &odl_uncomp);
3632 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
3633 DD_USED_HEAD, DD_USED_SNAP, NULL);
3636 /* swap ds_*_bytes */
3637 SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
3638 dsl_dataset_phys(clone)->ds_referenced_bytes);
3639 SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
3640 dsl_dataset_phys(clone)->ds_compressed_bytes);
3641 SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
3642 dsl_dataset_phys(clone)->ds_uncompressed_bytes);
3643 SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
3644 dsl_dataset_phys(clone)->ds_unique_bytes);
3646 /* apply any parent delta for change in unconsumed refreservation */
3647 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
3648 unused_refres_delta, 0, 0, tx);
3653 dsl_deadlist_close(&clone->ds_deadlist);
3654 dsl_deadlist_close(&origin_head->ds_deadlist);
3655 SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
3656 dsl_dataset_phys(clone)->ds_deadlist_obj);
3657 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
3658 dsl_dataset_phys(clone)->ds_deadlist_obj);
3659 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
3660 dsl_dataset_phys(origin_head)->ds_deadlist_obj);
3661 dsl_dataset_swap_remap_deadlists(clone, origin_head, tx);
3663 dsl_scan_ds_clone_swapped(origin_head, clone, tx);
3665 spa_history_log_internal_ds(clone, "clone swap", tx,
3666 "parent=%s", origin_head->ds_dir->dd_myname);
3670 * Given a pool name and a dataset object number in that pool,
3671 * return the name of that dataset.
3674 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3680 error = dsl_pool_hold(pname, FTAG, &dp);
3684 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
3686 dsl_dataset_name(ds, buf);
3687 dsl_dataset_rele(ds, FTAG);
3689 dsl_pool_rele(dp, FTAG);
3695 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3696 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3700 ASSERT3S(asize, >, 0);
3703 * *ref_rsrv is the portion of asize that will come from any
3704 * unconsumed refreservation space.
3708 mutex_enter(&ds->ds_lock);
3710 * Make a space adjustment for reserved bytes.
3712 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
3714 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3716 (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3718 asize - MIN(asize, parent_delta(ds, asize + inflight));
3721 if (!check_quota || ds->ds_quota == 0) {
3722 mutex_exit(&ds->ds_lock);
3726 * If they are requesting more space, and our current estimate
3727 * is over quota, they get to try again unless the actual
3728 * on-disk is over quota and there are no pending changes (which
3729 * may free up space for us).
3731 if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
3734 dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
3735 error = SET_ERROR(ERESTART);
3737 error = SET_ERROR(EDQUOT);
3739 mutex_exit(&ds->ds_lock);
3744 typedef struct dsl_dataset_set_qr_arg {
3745 const char *ddsqra_name;
3746 zprop_source_t ddsqra_source;
3747 uint64_t ddsqra_value;
3748 } dsl_dataset_set_qr_arg_t;
3753 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
3755 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3756 dsl_pool_t *dp = dmu_tx_pool(tx);
3761 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
3762 return (SET_ERROR(ENOTSUP));
3764 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3768 if (ds->ds_is_snapshot) {
3769 dsl_dataset_rele(ds, FTAG);
3770 return (SET_ERROR(EINVAL));
3773 error = dsl_prop_predict(ds->ds_dir,
3774 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3775 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3777 dsl_dataset_rele(ds, FTAG);
3782 dsl_dataset_rele(ds, FTAG);
3786 if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
3787 newval < ds->ds_reserved) {
3788 dsl_dataset_rele(ds, FTAG);
3789 return (SET_ERROR(ENOSPC));
3792 dsl_dataset_rele(ds, FTAG);
3797 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
3799 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3800 dsl_pool_t *dp = dmu_tx_pool(tx);
3804 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3806 dsl_prop_set_sync_impl(ds,
3807 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3808 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
3809 &ddsqra->ddsqra_value, tx);
3811 VERIFY0(dsl_prop_get_int_ds(ds,
3812 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
3814 if (ds->ds_quota != newval) {
3815 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3816 ds->ds_quota = newval;
3818 dsl_dataset_rele(ds, FTAG);
3822 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
3825 dsl_dataset_set_qr_arg_t ddsqra;
3827 ddsqra.ddsqra_name = dsname;
3828 ddsqra.ddsqra_source = source;
3829 ddsqra.ddsqra_value = refquota;
3831 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
3832 dsl_dataset_set_refquota_sync, &ddsqra, 0,
3833 ZFS_SPACE_CHECK_EXTRA_RESERVED));
3837 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
3839 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3840 dsl_pool_t *dp = dmu_tx_pool(tx);
3843 uint64_t newval, unique;
3845 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
3846 return (SET_ERROR(ENOTSUP));
3848 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3852 if (ds->ds_is_snapshot) {
3853 dsl_dataset_rele(ds, FTAG);
3854 return (SET_ERROR(EINVAL));
3857 error = dsl_prop_predict(ds->ds_dir,
3858 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3859 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3861 dsl_dataset_rele(ds, FTAG);
3866 * If we are doing the preliminary check in open context, the
3867 * space estimates may be inaccurate.
3869 if (!dmu_tx_is_syncing(tx)) {
3870 dsl_dataset_rele(ds, FTAG);
3874 mutex_enter(&ds->ds_lock);
3875 if (!DS_UNIQUE_IS_ACCURATE(ds))
3876 dsl_dataset_recalc_head_uniq(ds);
3877 unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3878 mutex_exit(&ds->ds_lock);
3880 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3881 uint64_t delta = MAX(unique, newval) -
3882 MAX(unique, ds->ds_reserved);
3885 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3886 (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3887 dsl_dataset_rele(ds, FTAG);
3888 return (SET_ERROR(ENOSPC));
3892 dsl_dataset_rele(ds, FTAG);
3897 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3898 zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3904 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3905 source, sizeof (value), 1, &value, tx);
3907 VERIFY0(dsl_prop_get_int_ds(ds,
3908 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3910 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3911 mutex_enter(&ds->ds_dir->dd_lock);
3912 mutex_enter(&ds->ds_lock);
3913 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3914 unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3915 delta = MAX(0, (int64_t)(newval - unique)) -
3916 MAX(0, (int64_t)(ds->ds_reserved - unique));
3917 ds->ds_reserved = newval;
3918 mutex_exit(&ds->ds_lock);
3920 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3921 mutex_exit(&ds->ds_dir->dd_lock);
3925 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3927 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3928 dsl_pool_t *dp = dmu_tx_pool(tx);
3931 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3932 dsl_dataset_set_refreservation_sync_impl(ds,
3933 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3934 dsl_dataset_rele(ds, FTAG);
3938 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3939 uint64_t refreservation)
3941 dsl_dataset_set_qr_arg_t ddsqra;
3943 ddsqra.ddsqra_name = dsname;
3944 ddsqra.ddsqra_source = source;
3945 ddsqra.ddsqra_value = refreservation;
3947 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3948 dsl_dataset_set_refreservation_sync, &ddsqra, 0,
3949 ZFS_SPACE_CHECK_EXTRA_RESERVED));
3953 * Return (in *usedp) the amount of space written in new that is not
3954 * present in oldsnap. New may be a snapshot or the head. Old must be
3955 * a snapshot before new, in new's filesystem (or its origin). If not then
3956 * fail and return EINVAL.
3958 * The written space is calculated by considering two components: First, we
3959 * ignore any freed space, and calculate the written as new's used space
3960 * minus old's used space. Next, we add in the amount of space that was freed
3961 * between the two snapshots, thus reducing new's used space relative to old's.
3962 * Specifically, this is the space that was born before old->ds_creation_txg,
3963 * and freed before new (ie. on new's deadlist or a previous deadlist).
3965 * space freed [---------------------]
3966 * snapshots ---O-------O--------O-------O------
3970 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3971 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3975 dsl_pool_t *dp = new->ds_dir->dd_pool;
3977 ASSERT(dsl_pool_config_held(dp));
3980 *usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3981 *usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3984 *compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3985 *compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3988 *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3989 *uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3991 snapobj = new->ds_object;
3992 while (snapobj != oldsnap->ds_object) {
3993 dsl_dataset_t *snap;
3994 uint64_t used, comp, uncomp;
3996 if (snapobj == new->ds_object) {
3999 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
4004 if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
4005 dsl_dataset_phys(oldsnap)->ds_creation_txg) {
4007 * The blocks in the deadlist can not be born after
4008 * ds_prev_snap_txg, so get the whole deadlist space,
4009 * which is more efficient (especially for old-format
4010 * deadlists). Unfortunately the deadlist code
4011 * doesn't have enough information to make this
4012 * optimization itself.
4014 dsl_deadlist_space(&snap->ds_deadlist,
4015 &used, &comp, &uncomp);
4017 dsl_deadlist_space_range(&snap->ds_deadlist,
4018 0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
4019 &used, &comp, &uncomp);
4026 * If we get to the beginning of the chain of snapshots
4027 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
4028 * was not a snapshot of/before new.
4030 snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
4032 dsl_dataset_rele(snap, FTAG);
4034 err = SET_ERROR(EINVAL);
4043 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
4044 * lastsnap, and all snapshots in between are deleted.
4046 * blocks that would be freed [---------------------------]
4047 * snapshots ---O-------O--------O-------O--------O
4048 * firstsnap lastsnap
4050 * This is the set of blocks that were born after the snap before firstsnap,
4051 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
4052 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
4053 * We calculate this by iterating over the relevant deadlists (from the snap
4054 * after lastsnap, backward to the snap after firstsnap), summing up the
4055 * space on the deadlist that was born after the snap before firstsnap.
4058 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
4059 dsl_dataset_t *lastsnap,
4060 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4064 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
4066 ASSERT(firstsnap->ds_is_snapshot);
4067 ASSERT(lastsnap->ds_is_snapshot);
4070 * Check that the snapshots are in the same dsl_dir, and firstsnap
4071 * is before lastsnap.
4073 if (firstsnap->ds_dir != lastsnap->ds_dir ||
4074 dsl_dataset_phys(firstsnap)->ds_creation_txg >
4075 dsl_dataset_phys(lastsnap)->ds_creation_txg)
4076 return (SET_ERROR(EINVAL));
4078 *usedp = *compp = *uncompp = 0;
4080 snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
4081 while (snapobj != firstsnap->ds_object) {
4083 uint64_t used, comp, uncomp;
4085 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
4089 dsl_deadlist_space_range(&ds->ds_deadlist,
4090 dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
4091 &used, &comp, &uncomp);
4096 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
4097 ASSERT3U(snapobj, !=, 0);
4098 dsl_dataset_rele(ds, FTAG);
4104 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
4105 * For example, they could both be snapshots of the same filesystem, and
4106 * 'earlier' is before 'later'. Or 'earlier' could be the origin of
4107 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's
4108 * filesystem. Or 'earlier' could be the origin's origin.
4110 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
4113 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
4114 uint64_t earlier_txg)
4116 dsl_pool_t *dp = later->ds_dir->dd_pool;
4120 ASSERT(dsl_pool_config_held(dp));
4121 ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
4123 if (earlier_txg == 0)
4124 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
4126 if (later->ds_is_snapshot &&
4127 earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
4130 if (later->ds_dir == earlier->ds_dir)
4132 if (!dsl_dir_is_clone(later->ds_dir))
4135 if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
4137 dsl_dataset_t *origin;
4138 error = dsl_dataset_hold_obj(dp,
4139 dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
4142 ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
4143 dsl_dataset_rele(origin, FTAG);
4148 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
4150 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
4151 dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
4155 dsl_dataset_is_zapified(dsl_dataset_t *ds)
4157 dmu_object_info_t doi;
4159 dmu_object_info_from_db(ds->ds_dbuf, &doi);
4160 return (doi.doi_type == DMU_OTN_ZAP_METADATA);
4164 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds)
4166 return (dsl_dataset_is_zapified(ds) &&
4167 zap_contains(ds->ds_dir->dd_pool->dp_meta_objset,
4168 ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0);
4172 dsl_dataset_get_remap_deadlist_object(dsl_dataset_t *ds)
4174 uint64_t remap_deadlist_obj;
4177 if (!dsl_dataset_is_zapified(ds))
4180 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object,
4181 DS_FIELD_REMAP_DEADLIST, sizeof (remap_deadlist_obj), 1,
4182 &remap_deadlist_obj);
4185 VERIFY3S(err, ==, ENOENT);
4189 ASSERT(remap_deadlist_obj != 0);
4190 return (remap_deadlist_obj);
4194 dsl_dataset_remap_deadlist_exists(dsl_dataset_t *ds)
4196 EQUIV(dsl_deadlist_is_open(&ds->ds_remap_deadlist),
4197 dsl_dataset_get_remap_deadlist_object(ds) != 0);
4198 return (dsl_deadlist_is_open(&ds->ds_remap_deadlist));
4202 dsl_dataset_set_remap_deadlist_object(dsl_dataset_t *ds, uint64_t obj,
4206 dsl_dataset_zapify(ds, tx);
4207 VERIFY0(zap_add(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object,
4208 DS_FIELD_REMAP_DEADLIST, sizeof (obj), 1, &obj, tx));
4212 dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t *ds, dmu_tx_t *tx)
4214 VERIFY0(zap_remove(ds->ds_dir->dd_pool->dp_meta_objset,
4215 ds->ds_object, DS_FIELD_REMAP_DEADLIST, tx));
4219 dsl_dataset_destroy_remap_deadlist(dsl_dataset_t *ds, dmu_tx_t *tx)
4221 uint64_t remap_deadlist_object;
4222 spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
4224 ASSERT(dmu_tx_is_syncing(tx));
4225 ASSERT(dsl_dataset_remap_deadlist_exists(ds));
4227 remap_deadlist_object = ds->ds_remap_deadlist.dl_object;
4228 dsl_deadlist_close(&ds->ds_remap_deadlist);
4229 dsl_deadlist_free(spa_meta_objset(spa), remap_deadlist_object, tx);
4230 dsl_dataset_unset_remap_deadlist_object(ds, tx);
4231 spa_feature_decr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx);
4235 dsl_dataset_create_remap_deadlist(dsl_dataset_t *ds, dmu_tx_t *tx)
4237 uint64_t remap_deadlist_obj;
4238 spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
4240 ASSERT(dmu_tx_is_syncing(tx));
4241 ASSERT(MUTEX_HELD(&ds->ds_remap_deadlist_lock));
4243 * Currently we only create remap deadlists when there are indirect
4244 * vdevs with referenced mappings.
4246 ASSERT(spa_feature_is_active(spa, SPA_FEATURE_DEVICE_REMOVAL));
4248 remap_deadlist_obj = dsl_deadlist_clone(
4249 &ds->ds_deadlist, UINT64_MAX,
4250 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
4251 dsl_dataset_set_remap_deadlist_object(ds,
4252 remap_deadlist_obj, tx);
4253 dsl_deadlist_open(&ds->ds_remap_deadlist, spa_meta_objset(spa),
4254 remap_deadlist_obj);
4255 spa_feature_incr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx);