4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Portions Copyright (c) 2011 Martin Matuska <mm@FreeBSD.org>
24 * Copyright (c) 2011, 2016 by Delphix. All rights reserved.
25 * Copyright (c) 2014, Joyent, Inc. All rights reserved.
26 * Copyright (c) 2014 RackTop Systems.
27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28 * Copyright (c) 2014 Integros [integros.com]
29 * Copyright 2016, OmniTI Computer Consulting, Inc. All rights reserved.
32 #include <sys/dmu_objset.h>
33 #include <sys/dsl_dataset.h>
34 #include <sys/dsl_dir.h>
35 #include <sys/dsl_prop.h>
36 #include <sys/dsl_synctask.h>
37 #include <sys/dmu_traverse.h>
38 #include <sys/dmu_impl.h>
39 #include <sys/dmu_send.h>
40 #include <sys/dmu_tx.h>
44 #include <sys/zfeature.h>
45 #include <sys/unique.h>
46 #include <sys/zfs_context.h>
47 #include <sys/zfs_ioctl.h>
49 #include <sys/zfs_znode.h>
50 #include <sys/zfs_onexit.h>
52 #include <sys/dsl_scan.h>
53 #include <sys/dsl_deadlist.h>
54 #include <sys/dsl_destroy.h>
55 #include <sys/dsl_userhold.h>
56 #include <sys/dsl_bookmark.h>
57 #include <sys/dmu_send.h>
58 #include <sys/zio_checksum.h>
59 #include <sys/zio_compress.h>
60 #include <zfs_fletcher.h>
62 SYSCTL_DECL(_vfs_zfs);
65 * The SPA supports block sizes up to 16MB. However, very large blocks
66 * can have an impact on i/o latency (e.g. tying up a spinning disk for
67 * ~300ms), and also potentially on the memory allocator. Therefore,
68 * we do not allow the recordsize to be set larger than zfs_max_recordsize
69 * (default 1MB). Larger blocks can be created by changing this tunable,
70 * and pools with larger blocks can always be imported and used, regardless
73 int zfs_max_recordsize = 1 * 1024 * 1024;
74 SYSCTL_INT(_vfs_zfs, OID_AUTO, max_recordsize, CTLFLAG_RWTUN,
75 &zfs_max_recordsize, 0,
76 "Maximum block size. Expect dragons when tuning this.");
78 #define SWITCH64(x, y) \
80 uint64_t __tmp = (x); \
85 #define DS_REF_MAX (1ULL << 62)
87 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
89 extern int spa_asize_inflation;
91 static zil_header_t zero_zil;
94 * Figure out how much of this delta should be propogated to the dsl_dir
95 * layer. If there's a refreservation, that space has already been
96 * partially accounted for in our ancestors.
99 parent_delta(dsl_dataset_t *ds, int64_t delta)
101 dsl_dataset_phys_t *ds_phys;
102 uint64_t old_bytes, new_bytes;
104 if (ds->ds_reserved == 0)
107 ds_phys = dsl_dataset_phys(ds);
108 old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
109 new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
111 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
112 return (new_bytes - old_bytes);
116 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
118 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
119 int compressed = BP_GET_PSIZE(bp);
120 int uncompressed = BP_GET_UCSIZE(bp);
123 dprintf_bp(bp, "ds=%p", ds);
125 ASSERT(dmu_tx_is_syncing(tx));
126 /* It could have been compressed away to nothing */
129 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
130 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
132 dsl_pool_mos_diduse_space(tx->tx_pool,
133 used, compressed, uncompressed);
137 ASSERT3U(bp->blk_birth, >, dsl_dataset_phys(ds)->ds_prev_snap_txg);
138 dmu_buf_will_dirty(ds->ds_dbuf, tx);
139 mutex_enter(&ds->ds_lock);
140 delta = parent_delta(ds, used);
141 dsl_dataset_phys(ds)->ds_referenced_bytes += used;
142 dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
143 dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
144 dsl_dataset_phys(ds)->ds_unique_bytes += used;
146 if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) {
147 ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_BLOCKS] =
151 spa_feature_t f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp));
152 if (f != SPA_FEATURE_NONE)
153 ds->ds_feature_activation_needed[f] = B_TRUE;
155 mutex_exit(&ds->ds_lock);
156 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
157 compressed, uncompressed, tx);
158 dsl_dir_transfer_space(ds->ds_dir, used - delta,
159 DD_USED_REFRSRV, DD_USED_HEAD, NULL);
163 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
166 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
167 int compressed = BP_GET_PSIZE(bp);
168 int uncompressed = BP_GET_UCSIZE(bp);
173 ASSERT(dmu_tx_is_syncing(tx));
174 ASSERT(bp->blk_birth <= tx->tx_txg);
177 dsl_free(tx->tx_pool, tx->tx_txg, bp);
178 dsl_pool_mos_diduse_space(tx->tx_pool,
179 -used, -compressed, -uncompressed);
182 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
184 ASSERT(!ds->ds_is_snapshot);
185 dmu_buf_will_dirty(ds->ds_dbuf, tx);
187 if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
190 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
191 dsl_free(tx->tx_pool, tx->tx_txg, bp);
193 mutex_enter(&ds->ds_lock);
194 ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
195 !DS_UNIQUE_IS_ACCURATE(ds));
196 delta = parent_delta(ds, -used);
197 dsl_dataset_phys(ds)->ds_unique_bytes -= used;
198 mutex_exit(&ds->ds_lock);
199 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
200 delta, -compressed, -uncompressed, tx);
201 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
202 DD_USED_REFRSRV, DD_USED_HEAD, NULL);
204 dprintf_bp(bp, "putting on dead list: %s", "");
207 * We are here as part of zio's write done callback,
208 * which means we're a zio interrupt thread. We can't
209 * call dsl_deadlist_insert() now because it may block
210 * waiting for I/O. Instead, put bp on the deferred
211 * queue and let dsl_pool_sync() finish the job.
213 bplist_append(&ds->ds_pending_deadlist, bp);
215 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
217 ASSERT3U(ds->ds_prev->ds_object, ==,
218 dsl_dataset_phys(ds)->ds_prev_snap_obj);
219 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
220 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
221 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
222 ds->ds_object && bp->blk_birth >
223 dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
224 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
225 mutex_enter(&ds->ds_prev->ds_lock);
226 dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
227 mutex_exit(&ds->ds_prev->ds_lock);
229 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
230 dsl_dir_transfer_space(ds->ds_dir, used,
231 DD_USED_HEAD, DD_USED_SNAP, tx);
234 mutex_enter(&ds->ds_lock);
235 ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
236 dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
237 ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
238 dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
239 ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
240 dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
241 mutex_exit(&ds->ds_lock);
247 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
249 uint64_t trysnap = 0;
254 * The snapshot creation could fail, but that would cause an
255 * incorrect FALSE return, which would only result in an
256 * overestimation of the amount of space that an operation would
257 * consume, which is OK.
259 * There's also a small window where we could miss a pending
260 * snapshot, because we could set the sync task in the quiescing
261 * phase. So this should only be used as a guess.
263 if (ds->ds_trysnap_txg >
264 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
265 trysnap = ds->ds_trysnap_txg;
266 return (MAX(dsl_dataset_phys(ds)->ds_prev_snap_txg, trysnap));
270 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
273 if (blk_birth <= dsl_dataset_prev_snap_txg(ds) ||
274 (bp != NULL && BP_IS_HOLE(bp)))
277 ddt_prefetch(dsl_dataset_get_spa(ds), bp);
283 * We have to release the fsid syncronously or we risk that a subsequent
284 * mount of the same dataset will fail to unique_insert the fsid. This
285 * failure would manifest itself as the fsid of this dataset changing
286 * between mounts which makes NFS clients quite unhappy.
289 dsl_dataset_evict_sync(void *dbu)
291 dsl_dataset_t *ds = dbu;
293 ASSERT(ds->ds_owner == NULL);
295 unique_remove(ds->ds_fsid_guid);
299 dsl_dataset_evict_async(void *dbu)
301 dsl_dataset_t *ds = dbu;
303 ASSERT(ds->ds_owner == NULL);
307 if (ds->ds_objset != NULL)
308 dmu_objset_evict(ds->ds_objset);
311 dsl_dataset_rele(ds->ds_prev, ds);
315 bplist_destroy(&ds->ds_pending_deadlist);
316 if (ds->ds_deadlist.dl_os != NULL)
317 dsl_deadlist_close(&ds->ds_deadlist);
319 dsl_dir_async_rele(ds->ds_dir, ds);
321 ASSERT(!list_link_active(&ds->ds_synced_link));
323 list_destroy(&ds->ds_prop_cbs);
324 if (mutex_owned(&ds->ds_lock))
325 mutex_exit(&ds->ds_lock);
326 mutex_destroy(&ds->ds_lock);
327 if (mutex_owned(&ds->ds_opening_lock))
328 mutex_exit(&ds->ds_opening_lock);
329 mutex_destroy(&ds->ds_opening_lock);
330 mutex_destroy(&ds->ds_sendstream_lock);
331 refcount_destroy(&ds->ds_longholds);
332 rrw_destroy(&ds->ds_bp_rwlock);
334 kmem_free(ds, sizeof (dsl_dataset_t));
338 dsl_dataset_get_snapname(dsl_dataset_t *ds)
340 dsl_dataset_phys_t *headphys;
343 dsl_pool_t *dp = ds->ds_dir->dd_pool;
344 objset_t *mos = dp->dp_meta_objset;
346 if (ds->ds_snapname[0])
348 if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
351 err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
355 headphys = headdbuf->db_data;
356 err = zap_value_search(dp->dp_meta_objset,
357 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
358 dmu_buf_rele(headdbuf, FTAG);
363 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
365 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
366 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
370 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
375 err = zap_lookup_norm(mos, snapobj, name, 8, 1,
376 value, mt, NULL, 0, NULL);
377 if (err == ENOTSUP && mt == MT_FIRST)
378 err = zap_lookup(mos, snapobj, name, 8, 1, value);
383 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
386 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
387 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
391 dsl_dir_snap_cmtime_update(ds->ds_dir);
393 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
398 err = zap_remove_norm(mos, snapobj, name, mt, tx);
399 if (err == ENOTSUP && mt == MT_FIRST)
400 err = zap_remove(mos, snapobj, name, tx);
402 if (err == 0 && adj_cnt)
403 dsl_fs_ss_count_adjust(ds->ds_dir, -1,
404 DD_FIELD_SNAPSHOT_COUNT, tx);
410 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag)
412 dmu_buf_t *dbuf = ds->ds_dbuf;
413 boolean_t result = B_FALSE;
415 if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset,
416 ds->ds_object, DMU_BONUS_BLKID, tag)) {
418 if (ds == dmu_buf_get_user(dbuf))
421 dmu_buf_rele(dbuf, tag);
428 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
431 objset_t *mos = dp->dp_meta_objset;
435 dmu_object_info_t doi;
437 ASSERT(dsl_pool_config_held(dp));
439 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
443 /* Make sure dsobj has the correct object type. */
444 dmu_object_info_from_db(dbuf, &doi);
445 if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
446 dmu_buf_rele(dbuf, tag);
447 return (SET_ERROR(EINVAL));
450 ds = dmu_buf_get_user(dbuf);
452 dsl_dataset_t *winner = NULL;
454 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
456 ds->ds_object = dsobj;
457 ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0;
459 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
460 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
461 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
462 rrw_init(&ds->ds_bp_rwlock, B_FALSE);
463 refcount_create(&ds->ds_longholds);
465 bplist_create(&ds->ds_pending_deadlist);
466 dsl_deadlist_open(&ds->ds_deadlist,
467 mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
469 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
470 offsetof(dmu_sendarg_t, dsa_link));
472 list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t),
473 offsetof(dsl_prop_cb_record_t, cbr_ds_node));
475 if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
476 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
477 if (!(spa_feature_table[f].fi_flags &
478 ZFEATURE_FLAG_PER_DATASET))
480 err = zap_contains(mos, dsobj,
481 spa_feature_table[f].fi_guid);
483 ds->ds_feature_inuse[f] = B_TRUE;
485 ASSERT3U(err, ==, ENOENT);
491 err = dsl_dir_hold_obj(dp,
492 dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds, &ds->ds_dir);
494 mutex_destroy(&ds->ds_lock);
495 mutex_destroy(&ds->ds_opening_lock);
496 mutex_destroy(&ds->ds_sendstream_lock);
497 refcount_destroy(&ds->ds_longholds);
498 bplist_destroy(&ds->ds_pending_deadlist);
499 dsl_deadlist_close(&ds->ds_deadlist);
500 kmem_free(ds, sizeof (dsl_dataset_t));
501 dmu_buf_rele(dbuf, tag);
505 if (!ds->ds_is_snapshot) {
506 ds->ds_snapname[0] = '\0';
507 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
508 err = dsl_dataset_hold_obj(dp,
509 dsl_dataset_phys(ds)->ds_prev_snap_obj,
512 if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
513 int zaperr = zap_lookup(mos, ds->ds_object,
514 DS_FIELD_BOOKMARK_NAMES,
515 sizeof (ds->ds_bookmarks), 1,
517 if (zaperr != ENOENT)
521 if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
522 err = dsl_dataset_get_snapname(ds);
524 dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
526 ds->ds_dir->dd_pool->dp_meta_objset,
527 dsl_dataset_phys(ds)->ds_userrefs_obj,
532 if (err == 0 && !ds->ds_is_snapshot) {
533 err = dsl_prop_get_int_ds(ds,
534 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
537 err = dsl_prop_get_int_ds(ds,
538 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
542 ds->ds_reserved = ds->ds_quota = 0;
545 dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict_sync,
546 dsl_dataset_evict_async, &ds->ds_dbuf);
548 winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
550 if (err != 0 || winner != NULL) {
551 bplist_destroy(&ds->ds_pending_deadlist);
552 dsl_deadlist_close(&ds->ds_deadlist);
554 dsl_dataset_rele(ds->ds_prev, ds);
555 dsl_dir_rele(ds->ds_dir, ds);
556 mutex_destroy(&ds->ds_lock);
557 mutex_destroy(&ds->ds_opening_lock);
558 mutex_destroy(&ds->ds_sendstream_lock);
559 refcount_destroy(&ds->ds_longholds);
560 kmem_free(ds, sizeof (dsl_dataset_t));
562 dmu_buf_rele(dbuf, tag);
568 unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
569 if (ds->ds_fsid_guid !=
570 dsl_dataset_phys(ds)->ds_fsid_guid) {
571 zfs_dbgmsg("ds_fsid_guid changed from "
572 "%llx to %llx for pool %s dataset id %llu",
574 dsl_dataset_phys(ds)->ds_fsid_guid,
575 (long long)ds->ds_fsid_guid,
576 spa_name(dp->dp_spa),
581 ASSERT3P(ds->ds_dbuf, ==, dbuf);
582 ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
583 ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
584 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
585 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
591 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
592 void *tag, dsl_dataset_t **dsp)
595 const char *snapname;
600 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
604 ASSERT(dsl_pool_config_held(dp));
605 obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
607 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
609 err = SET_ERROR(ENOENT);
611 /* we may be looking for a snapshot */
612 if (err == 0 && snapname != NULL) {
613 dsl_dataset_t *snap_ds;
615 if (*snapname++ != '@') {
616 dsl_dataset_rele(ds, tag);
617 dsl_dir_rele(dd, FTAG);
618 return (SET_ERROR(ENOENT));
621 dprintf("looking for snapshot '%s'\n", snapname);
622 err = dsl_dataset_snap_lookup(ds, snapname, &obj);
624 err = dsl_dataset_hold_obj(dp, obj, tag, &snap_ds);
625 dsl_dataset_rele(ds, tag);
628 mutex_enter(&snap_ds->ds_lock);
629 if (snap_ds->ds_snapname[0] == 0)
630 (void) strlcpy(snap_ds->ds_snapname, snapname,
631 sizeof (snap_ds->ds_snapname));
632 mutex_exit(&snap_ds->ds_lock);
638 dsl_dir_rele(dd, FTAG);
643 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
644 void *tag, dsl_dataset_t **dsp)
646 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
649 if (!dsl_dataset_tryown(*dsp, tag)) {
650 dsl_dataset_rele(*dsp, tag);
652 return (SET_ERROR(EBUSY));
658 dsl_dataset_own(dsl_pool_t *dp, const char *name,
659 void *tag, dsl_dataset_t **dsp)
661 int err = dsl_dataset_hold(dp, name, tag, dsp);
664 if (!dsl_dataset_tryown(*dsp, tag)) {
665 dsl_dataset_rele(*dsp, tag);
666 return (SET_ERROR(EBUSY));
672 * See the comment above dsl_pool_hold() for details. In summary, a long
673 * hold is used to prevent destruction of a dataset while the pool hold
674 * is dropped, allowing other concurrent operations (e.g. spa_sync()).
676 * The dataset and pool must be held when this function is called. After it
677 * is called, the pool hold may be released while the dataset is still held
681 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
683 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
684 (void) refcount_add(&ds->ds_longholds, tag);
688 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
690 (void) refcount_remove(&ds->ds_longholds, tag);
693 /* Return B_TRUE if there are any long holds on this dataset. */
695 dsl_dataset_long_held(dsl_dataset_t *ds)
697 return (!refcount_is_zero(&ds->ds_longholds));
701 dsl_dataset_name(dsl_dataset_t *ds, char *name)
704 (void) strcpy(name, "mos");
706 dsl_dir_name(ds->ds_dir, name);
707 VERIFY0(dsl_dataset_get_snapname(ds));
708 if (ds->ds_snapname[0]) {
709 VERIFY3U(strlcat(name, "@", ZFS_MAX_DATASET_NAME_LEN),
710 <, ZFS_MAX_DATASET_NAME_LEN);
712 * We use a "recursive" mutex so that we
713 * can call dprintf_ds() with ds_lock held.
715 if (!MUTEX_HELD(&ds->ds_lock)) {
716 mutex_enter(&ds->ds_lock);
717 VERIFY3U(strlcat(name, ds->ds_snapname,
718 ZFS_MAX_DATASET_NAME_LEN), <,
719 ZFS_MAX_DATASET_NAME_LEN);
720 mutex_exit(&ds->ds_lock);
722 VERIFY3U(strlcat(name, ds->ds_snapname,
723 ZFS_MAX_DATASET_NAME_LEN), <,
724 ZFS_MAX_DATASET_NAME_LEN);
731 dsl_dataset_namelen(dsl_dataset_t *ds)
733 VERIFY0(dsl_dataset_get_snapname(ds));
734 mutex_enter(&ds->ds_lock);
735 int len = dsl_dir_namelen(ds->ds_dir) + 1 + strlen(ds->ds_snapname);
736 mutex_exit(&ds->ds_lock);
741 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
743 dmu_buf_rele(ds->ds_dbuf, tag);
747 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
749 ASSERT3P(ds->ds_owner, ==, tag);
750 ASSERT(ds->ds_dbuf != NULL);
752 mutex_enter(&ds->ds_lock);
754 mutex_exit(&ds->ds_lock);
755 dsl_dataset_long_rele(ds, tag);
756 dsl_dataset_rele(ds, tag);
760 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
762 boolean_t gotit = FALSE;
764 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
765 mutex_enter(&ds->ds_lock);
766 if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
768 dsl_dataset_long_hold(ds, tag);
771 mutex_exit(&ds->ds_lock);
776 dsl_dataset_has_owner(dsl_dataset_t *ds)
779 mutex_enter(&ds->ds_lock);
780 rv = (ds->ds_owner != NULL);
781 mutex_exit(&ds->ds_lock);
786 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
788 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
789 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
792 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
794 spa_feature_incr(spa, f, tx);
795 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
797 VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
798 sizeof (zero), 1, &zero, tx));
802 dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
804 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
805 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
807 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
809 VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx));
810 spa_feature_decr(spa, f, tx);
814 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
815 uint64_t flags, dmu_tx_t *tx)
817 dsl_pool_t *dp = dd->dd_pool;
819 dsl_dataset_phys_t *dsphys;
821 objset_t *mos = dp->dp_meta_objset;
824 origin = dp->dp_origin_snap;
826 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
827 ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
828 ASSERT(dmu_tx_is_syncing(tx));
829 ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
831 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
832 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
833 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
834 dmu_buf_will_dirty(dbuf, tx);
835 dsphys = dbuf->db_data;
836 bzero(dsphys, sizeof (dsl_dataset_phys_t));
837 dsphys->ds_dir_obj = dd->dd_object;
838 dsphys->ds_flags = flags;
839 dsphys->ds_fsid_guid = unique_create();
841 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
842 sizeof (dsphys->ds_guid));
843 } while (dsphys->ds_guid == 0);
844 dsphys->ds_snapnames_zapobj =
845 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
847 dsphys->ds_creation_time = gethrestime_sec();
848 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
850 if (origin == NULL) {
851 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
853 dsl_dataset_t *ohds; /* head of the origin snapshot */
855 dsphys->ds_prev_snap_obj = origin->ds_object;
856 dsphys->ds_prev_snap_txg =
857 dsl_dataset_phys(origin)->ds_creation_txg;
858 dsphys->ds_referenced_bytes =
859 dsl_dataset_phys(origin)->ds_referenced_bytes;
860 dsphys->ds_compressed_bytes =
861 dsl_dataset_phys(origin)->ds_compressed_bytes;
862 dsphys->ds_uncompressed_bytes =
863 dsl_dataset_phys(origin)->ds_uncompressed_bytes;
864 rrw_enter(&origin->ds_bp_rwlock, RW_READER, FTAG);
865 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
866 rrw_exit(&origin->ds_bp_rwlock, FTAG);
869 * Inherit flags that describe the dataset's contents
870 * (INCONSISTENT) or properties (Case Insensitive).
872 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
873 (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
875 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
876 if (origin->ds_feature_inuse[f])
877 dsl_dataset_activate_feature(dsobj, f, tx);
880 dmu_buf_will_dirty(origin->ds_dbuf, tx);
881 dsl_dataset_phys(origin)->ds_num_children++;
883 VERIFY0(dsl_dataset_hold_obj(dp,
884 dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
886 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
887 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
888 dsl_dataset_rele(ohds, FTAG);
890 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
891 if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
892 dsl_dataset_phys(origin)->ds_next_clones_obj =
894 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
896 VERIFY0(zap_add_int(mos,
897 dsl_dataset_phys(origin)->ds_next_clones_obj,
901 dmu_buf_will_dirty(dd->dd_dbuf, tx);
902 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
903 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
904 if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
905 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
906 dsl_dir_phys(origin->ds_dir)->dd_clones =
908 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
910 VERIFY0(zap_add_int(mos,
911 dsl_dir_phys(origin->ds_dir)->dd_clones,
916 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
917 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
919 dmu_buf_rele(dbuf, FTAG);
921 dmu_buf_will_dirty(dd->dd_dbuf, tx);
922 dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
928 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
932 VERIFY0(dmu_objset_from_ds(ds, &os));
933 if (bcmp(&os->os_zil_header, &zero_zil, sizeof (zero_zil)) != 0) {
934 dsl_pool_t *dp = ds->ds_dir->dd_pool;
937 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
939 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
940 dsl_dataset_sync(ds, zio, tx);
941 VERIFY0(zio_wait(zio));
943 /* dsl_dataset_sync_done will drop this reference. */
944 dmu_buf_add_ref(ds->ds_dbuf, ds);
945 dsl_dataset_sync_done(ds, tx);
950 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
951 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
953 dsl_pool_t *dp = pdd->dd_pool;
954 uint64_t dsobj, ddobj;
957 ASSERT(dmu_tx_is_syncing(tx));
958 ASSERT(lastname[0] != '@');
960 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
961 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
963 dsobj = dsl_dataset_create_sync_dd(dd, origin,
964 flags & ~DS_CREATE_FLAG_NODIRTY, tx);
966 dsl_deleg_set_create_perms(dd, tx, cr);
969 * Since we're creating a new node we know it's a leaf, so we can
970 * initialize the counts if the limit feature is active.
972 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
974 objset_t *os = dd->dd_pool->dp_meta_objset;
976 dsl_dir_zapify(dd, tx);
977 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
978 sizeof (cnt), 1, &cnt, tx));
979 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
980 sizeof (cnt), 1, &cnt, tx));
983 dsl_dir_rele(dd, FTAG);
986 * If we are creating a clone, make sure we zero out any stale
987 * data from the origin snapshots zil header.
989 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
992 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
993 dsl_dataset_zero_zil(ds, tx);
994 dsl_dataset_rele(ds, FTAG);
1001 /* FreeBSD ioctl compat begin */
1004 const char *snapname;
1008 dsl_check_snap_cb(const char *name, void *arg)
1010 struct destroyarg *da = arg;
1014 dsname = kmem_asprintf("%s@%s", name, da->snapname);
1015 fnvlist_add_boolean(da->nvl, dsname);
1016 kmem_free(dsname, strlen(dsname) + 1);
1022 dmu_get_recursive_snaps_nvl(char *fsname, const char *snapname,
1025 struct destroyarg *da;
1028 da = kmem_zalloc(sizeof (struct destroyarg), KM_SLEEP);
1030 da->snapname = snapname;
1031 err = dmu_objset_find(fsname, dsl_check_snap_cb, da,
1033 kmem_free(da, sizeof (struct destroyarg));
1037 /* FreeBSD ioctl compat end */
1038 #endif /* __FreeBSD__ */
1041 * The unique space in the head dataset can be calculated by subtracting
1042 * the space used in the most recent snapshot, that is still being used
1043 * in this file system, from the space currently in use. To figure out
1044 * the space in the most recent snapshot still in use, we need to take
1045 * the total space used in the snapshot and subtract out the space that
1046 * has been freed up since the snapshot was taken.
1049 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1052 uint64_t dlused, dlcomp, dluncomp;
1054 ASSERT(!ds->ds_is_snapshot);
1056 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
1057 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
1061 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
1063 ASSERT3U(dlused, <=, mrs_used);
1064 dsl_dataset_phys(ds)->ds_unique_bytes =
1065 dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
1067 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1068 SPA_VERSION_UNIQUE_ACCURATE)
1069 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1073 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
1076 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1080 ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
1081 err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1084 * The err should not be ENOENT, but a bug in a previous version
1085 * of the code could cause upgrade_clones_cb() to not set
1086 * ds_next_snap_obj when it should, leading to a missing entry.
1087 * If we knew that the pool was created after
1088 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1089 * ENOENT. However, at least we can check that we don't have
1090 * too many entries in the next_clones_obj even after failing to
1095 ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1097 ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
1102 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1104 return (&dsl_dataset_phys(ds)->ds_bp);
1108 dsl_dataset_get_spa(dsl_dataset_t *ds)
1110 return (ds->ds_dir->dd_pool->dp_spa);
1114 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1118 if (ds == NULL) /* this is the meta-objset */
1121 ASSERT(ds->ds_objset != NULL);
1123 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
1124 panic("dirtying snapshot!");
1126 /* Must not dirty a dataset in the same txg where it got snapshotted. */
1127 ASSERT3U(tx->tx_txg, >, dsl_dataset_phys(ds)->ds_prev_snap_txg);
1129 dp = ds->ds_dir->dd_pool;
1130 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
1131 /* up the hold count until we can be written out */
1132 dmu_buf_add_ref(ds->ds_dbuf, ds);
1137 dsl_dataset_is_dirty(dsl_dataset_t *ds)
1139 for (int t = 0; t < TXG_SIZE; t++) {
1140 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1148 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1152 if (!dmu_tx_is_syncing(tx))
1156 * If there's an fs-only reservation, any blocks that might become
1157 * owned by the snapshot dataset must be accommodated by space
1158 * outside of the reservation.
1160 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1161 asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
1162 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1163 return (SET_ERROR(ENOSPC));
1166 * Propagate any reserved space for this snapshot to other
1167 * snapshot checks in this sync group.
1170 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1175 typedef struct dsl_dataset_snapshot_arg {
1176 nvlist_t *ddsa_snaps;
1177 nvlist_t *ddsa_props;
1178 nvlist_t *ddsa_errors;
1180 } dsl_dataset_snapshot_arg_t;
1183 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1184 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
1189 ds->ds_trysnap_txg = tx->tx_txg;
1191 if (!dmu_tx_is_syncing(tx))
1195 * We don't allow multiple snapshots of the same txg. If there
1196 * is already one, try again.
1198 if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1199 return (SET_ERROR(EAGAIN));
1202 * Check for conflicting snapshot name.
1204 error = dsl_dataset_snap_lookup(ds, snapname, &value);
1206 return (SET_ERROR(EEXIST));
1207 if (error != ENOENT)
1211 * We don't allow taking snapshots of inconsistent datasets, such as
1212 * those into which we are currently receiving. However, if we are
1213 * creating this snapshot as part of a receive, this check will be
1214 * executed atomically with respect to the completion of the receive
1215 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1216 * case we ignore this, knowing it will be fixed up for us shortly in
1217 * dmu_recv_end_sync().
1219 if (!recv && DS_IS_INCONSISTENT(ds))
1220 return (SET_ERROR(EBUSY));
1223 * Skip the check for temporary snapshots or if we have already checked
1224 * the counts in dsl_dataset_snapshot_check. This means we really only
1225 * check the count here when we're receiving a stream.
1227 if (cnt != 0 && cr != NULL) {
1228 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1229 ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1234 error = dsl_dataset_snapshot_reserve_space(ds, tx);
1242 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1244 dsl_dataset_snapshot_arg_t *ddsa = arg;
1245 dsl_pool_t *dp = dmu_tx_pool(tx);
1250 * Pre-compute how many total new snapshots will be created for each
1251 * level in the tree and below. This is needed for validating the
1252 * snapshot limit when either taking a recursive snapshot or when
1253 * taking multiple snapshots.
1255 * The problem is that the counts are not actually adjusted when
1256 * we are checking, only when we finally sync. For a single snapshot,
1257 * this is easy, the count will increase by 1 at each node up the tree,
1258 * but its more complicated for the recursive/multiple snapshot case.
1260 * The dsl_fs_ss_limit_check function does recursively check the count
1261 * at each level up the tree but since it is validating each snapshot
1262 * independently we need to be sure that we are validating the complete
1263 * count for the entire set of snapshots. We do this by rolling up the
1264 * counts for each component of the name into an nvlist and then
1265 * checking each of those cases with the aggregated count.
1267 * This approach properly handles not only the recursive snapshot
1268 * case (where we get all of those on the ddsa_snaps list) but also
1269 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1270 * validate the limit on 'a' using a count of 2).
1272 * We validate the snapshot names in the third loop and only report
1275 if (dmu_tx_is_syncing(tx)) {
1276 nvlist_t *cnt_track = NULL;
1277 cnt_track = fnvlist_alloc();
1279 /* Rollup aggregated counts into the cnt_track list */
1280 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1282 pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1285 char nm[MAXPATHLEN];
1287 (void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1288 pdelim = strchr(nm, '@');
1294 if (nvlist_lookup_uint64(cnt_track, nm,
1296 /* update existing entry */
1297 fnvlist_add_uint64(cnt_track, nm,
1301 fnvlist_add_uint64(cnt_track, nm, 1);
1304 pdelim = strrchr(nm, '/');
1307 } while (pdelim != NULL);
1310 /* Check aggregated counts at each level */
1311 for (pair = nvlist_next_nvpair(cnt_track, NULL);
1312 pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1318 name = nvpair_name(pair);
1319 cnt = fnvpair_value_uint64(pair);
1322 error = dsl_dataset_hold(dp, name, FTAG, &ds);
1324 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1325 ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1327 dsl_dataset_rele(ds, FTAG);
1331 if (ddsa->ddsa_errors != NULL)
1332 fnvlist_add_int32(ddsa->ddsa_errors,
1335 /* only report one error for this check */
1339 nvlist_free(cnt_track);
1342 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1343 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1347 char dsname[ZFS_MAX_DATASET_NAME_LEN];
1349 name = nvpair_name(pair);
1350 if (strlen(name) >= ZFS_MAX_DATASET_NAME_LEN)
1351 error = SET_ERROR(ENAMETOOLONG);
1353 atp = strchr(name, '@');
1355 error = SET_ERROR(EINVAL);
1357 (void) strlcpy(dsname, name, atp - name + 1);
1360 error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1362 /* passing 0/NULL skips dsl_fs_ss_limit_check */
1363 error = dsl_dataset_snapshot_check_impl(ds,
1364 atp + 1, tx, B_FALSE, 0, NULL);
1365 dsl_dataset_rele(ds, FTAG);
1369 if (ddsa->ddsa_errors != NULL) {
1370 fnvlist_add_int32(ddsa->ddsa_errors,
1381 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1384 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1386 dsl_dataset_phys_t *dsphys;
1387 uint64_t dsobj, crtxg;
1388 objset_t *mos = dp->dp_meta_objset;
1391 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1394 * If we are on an old pool, the zil must not be active, in which
1395 * case it will be zeroed. Usually zil_suspend() accomplishes this.
1397 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1398 dmu_objset_from_ds(ds, &os) != 0 ||
1399 bcmp(&os->os_phys->os_zil_header, &zero_zil,
1400 sizeof (zero_zil)) == 0);
1402 /* Should not snapshot a dirty dataset. */
1403 ASSERT(!txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1406 dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1409 * The origin's ds_creation_txg has to be < TXG_INITIAL
1411 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1416 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1417 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1418 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1419 dmu_buf_will_dirty(dbuf, tx);
1420 dsphys = dbuf->db_data;
1421 bzero(dsphys, sizeof (dsl_dataset_phys_t));
1422 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1423 dsphys->ds_fsid_guid = unique_create();
1425 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1426 sizeof (dsphys->ds_guid));
1427 } while (dsphys->ds_guid == 0);
1428 dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1429 dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1430 dsphys->ds_next_snap_obj = ds->ds_object;
1431 dsphys->ds_num_children = 1;
1432 dsphys->ds_creation_time = gethrestime_sec();
1433 dsphys->ds_creation_txg = crtxg;
1434 dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1435 dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1436 dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1437 dsphys->ds_uncompressed_bytes =
1438 dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1439 dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1440 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
1441 dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1442 rrw_exit(&ds->ds_bp_rwlock, FTAG);
1443 dmu_buf_rele(dbuf, FTAG);
1445 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1446 if (ds->ds_feature_inuse[f])
1447 dsl_dataset_activate_feature(dsobj, f, tx);
1450 ASSERT3U(ds->ds_prev != 0, ==,
1451 dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1453 uint64_t next_clones_obj =
1454 dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1455 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1457 dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1458 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1460 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1461 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1462 dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1463 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1464 } else if (next_clones_obj != 0) {
1465 dsl_dataset_remove_from_next_clones(ds->ds_prev,
1466 dsphys->ds_next_snap_obj, tx);
1467 VERIFY0(zap_add_int(mos,
1468 next_clones_obj, dsobj, tx));
1473 * If we have a reference-reservation on this dataset, we will
1474 * need to increase the amount of refreservation being charged
1475 * since our unique space is going to zero.
1477 if (ds->ds_reserved) {
1479 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1480 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1482 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1486 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1487 dsl_dataset_phys(ds)->ds_deadlist_obj =
1488 dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1489 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1490 dsl_deadlist_close(&ds->ds_deadlist);
1491 dsl_deadlist_open(&ds->ds_deadlist, mos,
1492 dsl_dataset_phys(ds)->ds_deadlist_obj);
1493 dsl_deadlist_add_key(&ds->ds_deadlist,
1494 dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1496 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1497 dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1498 dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1499 dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1500 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1501 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1503 VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1504 snapname, 8, 1, &dsobj, tx));
1507 dsl_dataset_rele(ds->ds_prev, ds);
1508 VERIFY0(dsl_dataset_hold_obj(dp,
1509 dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1511 dsl_scan_ds_snapshotted(ds, tx);
1513 dsl_dir_snap_cmtime_update(ds->ds_dir);
1515 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1519 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1521 dsl_dataset_snapshot_arg_t *ddsa = arg;
1522 dsl_pool_t *dp = dmu_tx_pool(tx);
1525 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1526 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1529 char dsname[ZFS_MAX_DATASET_NAME_LEN];
1531 name = nvpair_name(pair);
1532 atp = strchr(name, '@');
1533 (void) strlcpy(dsname, name, atp - name + 1);
1534 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1536 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1537 if (ddsa->ddsa_props != NULL) {
1538 dsl_props_set_sync_impl(ds->ds_prev,
1539 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1541 dsl_dataset_rele(ds, FTAG);
1546 * The snapshots must all be in the same pool.
1547 * All-or-nothing: if there are any failures, nothing will be modified.
1550 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1552 dsl_dataset_snapshot_arg_t ddsa;
1554 boolean_t needsuspend;
1558 nvlist_t *suspended = NULL;
1560 pair = nvlist_next_nvpair(snaps, NULL);
1563 firstname = nvpair_name(pair);
1565 error = spa_open(firstname, &spa, FTAG);
1568 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1569 spa_close(spa, FTAG);
1572 suspended = fnvlist_alloc();
1573 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1574 pair = nvlist_next_nvpair(snaps, pair)) {
1575 char fsname[ZFS_MAX_DATASET_NAME_LEN];
1576 char *snapname = nvpair_name(pair);
1580 atp = strchr(snapname, '@');
1582 error = SET_ERROR(EINVAL);
1585 (void) strlcpy(fsname, snapname, atp - snapname + 1);
1587 error = zil_suspend(fsname, &cookie);
1590 fnvlist_add_uint64(suspended, fsname,
1595 ddsa.ddsa_snaps = snaps;
1596 ddsa.ddsa_props = props;
1597 ddsa.ddsa_errors = errors;
1598 ddsa.ddsa_cr = CRED();
1601 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1602 dsl_dataset_snapshot_sync, &ddsa,
1603 fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1606 if (suspended != NULL) {
1607 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1608 pair = nvlist_next_nvpair(suspended, pair)) {
1609 zil_resume((void *)(uintptr_t)
1610 fnvpair_value_uint64(pair));
1612 fnvlist_free(suspended);
1618 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1619 pair = nvlist_next_nvpair(snaps, pair)) {
1620 char *snapname = nvpair_name(pair);
1621 zvol_create_minors(snapname);
1629 typedef struct dsl_dataset_snapshot_tmp_arg {
1630 const char *ddsta_fsname;
1631 const char *ddsta_snapname;
1632 minor_t ddsta_cleanup_minor;
1633 const char *ddsta_htag;
1634 } dsl_dataset_snapshot_tmp_arg_t;
1637 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1639 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1640 dsl_pool_t *dp = dmu_tx_pool(tx);
1644 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1648 /* NULL cred means no limit check for tmp snapshot */
1649 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1650 tx, B_FALSE, 0, NULL);
1652 dsl_dataset_rele(ds, FTAG);
1656 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1657 dsl_dataset_rele(ds, FTAG);
1658 return (SET_ERROR(ENOTSUP));
1660 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1663 dsl_dataset_rele(ds, FTAG);
1667 dsl_dataset_rele(ds, FTAG);
1672 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1674 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1675 dsl_pool_t *dp = dmu_tx_pool(tx);
1678 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1680 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1681 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1682 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1683 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1685 dsl_dataset_rele(ds, FTAG);
1689 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1690 minor_t cleanup_minor, const char *htag)
1692 dsl_dataset_snapshot_tmp_arg_t ddsta;
1695 boolean_t needsuspend;
1698 ddsta.ddsta_fsname = fsname;
1699 ddsta.ddsta_snapname = snapname;
1700 ddsta.ddsta_cleanup_minor = cleanup_minor;
1701 ddsta.ddsta_htag = htag;
1703 error = spa_open(fsname, &spa, FTAG);
1706 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1707 spa_close(spa, FTAG);
1710 error = zil_suspend(fsname, &cookie);
1715 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1716 dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1725 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1727 ASSERT(dmu_tx_is_syncing(tx));
1728 ASSERT(ds->ds_objset != NULL);
1729 ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1732 * in case we had to change ds_fsid_guid when we opened it,
1735 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1736 dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1738 if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) {
1739 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1740 ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1,
1741 &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx));
1742 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1743 ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1,
1744 &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx));
1745 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1746 ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1,
1747 &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx));
1748 ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0;
1749 ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0;
1750 ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0;
1753 dmu_objset_sync(ds->ds_objset, zio, tx);
1755 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1756 if (ds->ds_feature_activation_needed[f]) {
1757 if (ds->ds_feature_inuse[f])
1759 dsl_dataset_activate_feature(ds->ds_object, f, tx);
1760 ds->ds_feature_inuse[f] = B_TRUE;
1766 deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
1768 dsl_deadlist_t *dl = arg;
1769 dsl_deadlist_insert(dl, bp, tx);
1774 dsl_dataset_sync_done(dsl_dataset_t *ds, dmu_tx_t *tx)
1776 objset_t *os = ds->ds_objset;
1778 bplist_iterate(&ds->ds_pending_deadlist,
1779 deadlist_enqueue_cb, &ds->ds_deadlist, tx);
1781 ASSERT(!dmu_objset_is_dirty(os, dmu_tx_get_txg(tx)));
1783 dmu_buf_rele(ds->ds_dbuf, ds);
1787 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1790 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1793 nvlist_t *propval = fnvlist_alloc();
1794 nvlist_t *val = fnvlist_alloc();
1796 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1799 * There may be missing entries in ds_next_clones_obj
1800 * due to a bug in a previous version of the code.
1801 * Only trust it if it has the right number of entries.
1803 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1804 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1807 if (count != dsl_dataset_phys(ds)->ds_num_children - 1)
1809 for (zap_cursor_init(&zc, mos,
1810 dsl_dataset_phys(ds)->ds_next_clones_obj);
1811 zap_cursor_retrieve(&zc, &za) == 0;
1812 zap_cursor_advance(&zc)) {
1813 dsl_dataset_t *clone;
1814 char buf[ZFS_MAX_DATASET_NAME_LEN];
1815 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1816 za.za_first_integer, FTAG, &clone));
1817 dsl_dir_name(clone->ds_dir, buf);
1818 fnvlist_add_boolean(val, buf);
1819 dsl_dataset_rele(clone, FTAG);
1821 zap_cursor_fini(&zc);
1822 fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1823 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1826 nvlist_free(propval);
1830 get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv)
1832 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1834 if (dsl_dataset_has_resume_receive_state(ds)) {
1837 uint8_t *compressed;
1839 nvlist_t *token_nv = fnvlist_alloc();
1840 size_t packed_size, compressed_size;
1842 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1843 DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) {
1844 fnvlist_add_uint64(token_nv, "fromguid", val);
1846 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1847 DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) {
1848 fnvlist_add_uint64(token_nv, "object", val);
1850 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1851 DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) {
1852 fnvlist_add_uint64(token_nv, "offset", val);
1854 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1855 DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) {
1856 fnvlist_add_uint64(token_nv, "bytes", val);
1858 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1859 DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) {
1860 fnvlist_add_uint64(token_nv, "toguid", val);
1863 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1864 DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) {
1865 fnvlist_add_string(token_nv, "toname", buf);
1867 if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1868 DS_FIELD_RESUME_EMBEDOK) == 0) {
1869 fnvlist_add_boolean(token_nv, "embedok");
1871 packed = fnvlist_pack(token_nv, &packed_size);
1872 fnvlist_free(token_nv);
1873 compressed = kmem_alloc(packed_size, KM_SLEEP);
1875 compressed_size = gzip_compress(packed, compressed,
1876 packed_size, packed_size, 6);
1879 fletcher_4_native(compressed, compressed_size, NULL, &cksum);
1881 str = kmem_alloc(compressed_size * 2 + 1, KM_SLEEP);
1882 for (int i = 0; i < compressed_size; i++) {
1883 (void) sprintf(str + i * 2, "%02x", compressed[i]);
1885 str[compressed_size * 2] = '\0';
1886 char *propval = kmem_asprintf("%u-%llx-%llx-%s",
1887 ZFS_SEND_RESUME_TOKEN_VERSION,
1888 (longlong_t)cksum.zc_word[0],
1889 (longlong_t)packed_size, str);
1890 dsl_prop_nvlist_add_string(nv,
1891 ZFS_PROP_RECEIVE_RESUME_TOKEN, propval);
1892 kmem_free(packed, packed_size);
1893 kmem_free(str, compressed_size * 2 + 1);
1894 kmem_free(compressed, packed_size);
1900 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1902 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1903 uint64_t refd, avail, uobjs, aobjs, ratio;
1905 ASSERT(dsl_pool_config_held(dp));
1907 ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
1908 (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
1909 dsl_dataset_phys(ds)->ds_compressed_bytes);
1911 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1912 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1913 dsl_dataset_phys(ds)->ds_uncompressed_bytes);
1915 if (ds->ds_is_snapshot) {
1916 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1917 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1918 dsl_dataset_phys(ds)->ds_unique_bytes);
1919 get_clones_stat(ds, nv);
1921 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
1922 char buf[ZFS_MAX_DATASET_NAME_LEN];
1923 dsl_dataset_name(ds->ds_prev, buf);
1924 dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf);
1927 dsl_dir_stats(ds->ds_dir, nv);
1930 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1931 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1932 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1934 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1935 dsl_dataset_phys(ds)->ds_creation_time);
1936 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1937 dsl_dataset_phys(ds)->ds_creation_txg);
1938 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1940 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1942 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1943 dsl_dataset_phys(ds)->ds_guid);
1944 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1945 dsl_dataset_phys(ds)->ds_unique_bytes);
1946 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1948 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1950 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1951 DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1953 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1954 uint64_t written, comp, uncomp;
1955 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1956 dsl_dataset_t *prev;
1958 int err = dsl_dataset_hold_obj(dp,
1959 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1961 err = dsl_dataset_space_written(prev, ds, &written,
1963 dsl_dataset_rele(prev, FTAG);
1965 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1971 if (!dsl_dataset_is_snapshot(ds)) {
1973 * A failed "newfs" (e.g. full) resumable receive leaves
1974 * the stats set on this dataset. Check here for the prop.
1976 get_receive_resume_stats(ds, nv);
1979 * A failed incremental resumable receive leaves the
1980 * stats set on our child named "%recv". Check the child
1983 /* 6 extra bytes for /%recv */
1984 char recvname[ZFS_MAX_DATASET_NAME_LEN + 6];
1985 dsl_dataset_t *recv_ds;
1986 dsl_dataset_name(ds, recvname);
1987 if (strlcat(recvname, "/", sizeof (recvname)) <
1988 sizeof (recvname) &&
1989 strlcat(recvname, recv_clone_name, sizeof (recvname)) <
1990 sizeof (recvname) &&
1991 dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) {
1992 get_receive_resume_stats(recv_ds, nv);
1993 dsl_dataset_rele(recv_ds, FTAG);
1999 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
2001 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2002 ASSERT(dsl_pool_config_held(dp));
2004 stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg;
2005 stat->dds_inconsistent =
2006 dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT;
2007 stat->dds_guid = dsl_dataset_phys(ds)->ds_guid;
2008 stat->dds_origin[0] = '\0';
2009 if (ds->ds_is_snapshot) {
2010 stat->dds_is_snapshot = B_TRUE;
2011 stat->dds_num_clones =
2012 dsl_dataset_phys(ds)->ds_num_children - 1;
2014 stat->dds_is_snapshot = B_FALSE;
2015 stat->dds_num_clones = 0;
2017 if (dsl_dir_is_clone(ds->ds_dir)) {
2020 VERIFY0(dsl_dataset_hold_obj(dp,
2021 dsl_dir_phys(ds->ds_dir)->dd_origin_obj,
2023 dsl_dataset_name(ods, stat->dds_origin);
2024 dsl_dataset_rele(ods, FTAG);
2030 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2032 return (ds->ds_fsid_guid);
2036 dsl_dataset_space(dsl_dataset_t *ds,
2037 uint64_t *refdbytesp, uint64_t *availbytesp,
2038 uint64_t *usedobjsp, uint64_t *availobjsp)
2040 *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
2041 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2042 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
2044 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
2045 if (ds->ds_quota != 0) {
2047 * Adjust available bytes according to refquota
2049 if (*refdbytesp < ds->ds_quota)
2050 *availbytesp = MIN(*availbytesp,
2051 ds->ds_quota - *refdbytesp);
2055 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
2056 *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
2057 rrw_exit(&ds->ds_bp_rwlock, FTAG);
2058 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
2062 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
2064 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2067 ASSERT(dsl_pool_config_held(dp));
2070 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
2071 birth = dsl_dataset_get_blkptr(ds)->blk_birth;
2072 rrw_exit(&ds->ds_bp_rwlock, FTAG);
2073 if (birth > dsl_dataset_phys(snap)->ds_creation_txg) {
2074 objset_t *os, *os_snap;
2076 * It may be that only the ZIL differs, because it was
2077 * reset in the head. Don't count that as being
2080 if (dmu_objset_from_ds(ds, &os) != 0)
2082 if (dmu_objset_from_ds(snap, &os_snap) != 0)
2084 return (bcmp(&os->os_phys->os_meta_dnode,
2085 &os_snap->os_phys->os_meta_dnode,
2086 sizeof (os->os_phys->os_meta_dnode)) != 0);
2091 typedef struct dsl_dataset_rename_snapshot_arg {
2092 const char *ddrsa_fsname;
2093 const char *ddrsa_oldsnapname;
2094 const char *ddrsa_newsnapname;
2095 boolean_t ddrsa_recursive;
2097 } dsl_dataset_rename_snapshot_arg_t;
2101 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
2102 dsl_dataset_t *hds, void *arg)
2104 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2108 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2110 /* ignore nonexistent snapshots */
2111 return (error == ENOENT ? 0 : error);
2114 /* new name should not exist */
2115 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
2117 error = SET_ERROR(EEXIST);
2118 else if (error == ENOENT)
2121 /* dataset name + 1 for the "@" + the new snapshot name must fit */
2122 if (dsl_dir_namelen(hds->ds_dir) + 1 +
2123 strlen(ddrsa->ddrsa_newsnapname) >= ZFS_MAX_DATASET_NAME_LEN)
2124 error = SET_ERROR(ENAMETOOLONG);
2130 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
2132 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2133 dsl_pool_t *dp = dmu_tx_pool(tx);
2137 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
2141 if (ddrsa->ddrsa_recursive) {
2142 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2143 dsl_dataset_rename_snapshot_check_impl, ddrsa,
2146 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
2148 dsl_dataset_rele(hds, FTAG);
2153 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
2154 dsl_dataset_t *hds, void *arg)
2158 char *oldname, *newname;
2161 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2164 dmu_tx_t *tx = ddrsa->ddrsa_tx;
2167 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2168 ASSERT(error == 0 || error == ENOENT);
2169 if (error == ENOENT) {
2170 /* ignore nonexistent snapshots */
2174 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
2176 /* log before we change the name */
2177 spa_history_log_internal_ds(ds, "rename", tx,
2178 "-> @%s", ddrsa->ddrsa_newsnapname);
2180 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
2182 mutex_enter(&ds->ds_lock);
2183 (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
2184 mutex_exit(&ds->ds_lock);
2185 VERIFY0(zap_add(dp->dp_meta_objset,
2186 dsl_dataset_phys(hds)->ds_snapnames_zapobj,
2187 ds->ds_snapname, 8, 1, &ds->ds_object, tx));
2191 oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2192 newname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2193 snprintf(oldname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname,
2194 ddrsa->ddrsa_oldsnapname);
2195 snprintf(newname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname,
2196 ddrsa->ddrsa_newsnapname);
2197 zfsvfs_update_fromname(oldname, newname);
2198 zvol_rename_minors(oldname, newname);
2199 kmem_free(newname, MAXPATHLEN);
2200 kmem_free(oldname, MAXPATHLEN);
2203 dsl_dataset_rele(ds, FTAG);
2209 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
2211 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2212 dsl_pool_t *dp = dmu_tx_pool(tx);
2215 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
2216 ddrsa->ddrsa_tx = tx;
2217 if (ddrsa->ddrsa_recursive) {
2218 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2219 dsl_dataset_rename_snapshot_sync_impl, ddrsa,
2222 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
2224 dsl_dataset_rele(hds, FTAG);
2228 dsl_dataset_rename_snapshot(const char *fsname,
2229 const char *oldsnapname, const char *newsnapname, boolean_t recursive)
2231 dsl_dataset_rename_snapshot_arg_t ddrsa;
2233 ddrsa.ddrsa_fsname = fsname;
2234 ddrsa.ddrsa_oldsnapname = oldsnapname;
2235 ddrsa.ddrsa_newsnapname = newsnapname;
2236 ddrsa.ddrsa_recursive = recursive;
2238 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
2239 dsl_dataset_rename_snapshot_sync, &ddrsa,
2240 1, ZFS_SPACE_CHECK_RESERVED));
2244 * If we're doing an ownership handoff, we need to make sure that there is
2245 * only one long hold on the dataset. We're not allowed to change anything here
2246 * so we don't permanently release the long hold or regular hold here. We want
2247 * to do this only when syncing to avoid the dataset unexpectedly going away
2248 * when we release the long hold.
2251 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
2255 if (!dmu_tx_is_syncing(tx))
2258 if (owner != NULL) {
2259 VERIFY3P(ds->ds_owner, ==, owner);
2260 dsl_dataset_long_rele(ds, owner);
2263 held = dsl_dataset_long_held(ds);
2266 dsl_dataset_long_hold(ds, owner);
2269 return (SET_ERROR(EBUSY));
2274 typedef struct dsl_dataset_rollback_arg {
2275 const char *ddra_fsname;
2277 nvlist_t *ddra_result;
2278 } dsl_dataset_rollback_arg_t;
2281 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
2283 dsl_dataset_rollback_arg_t *ddra = arg;
2284 dsl_pool_t *dp = dmu_tx_pool(tx);
2286 int64_t unused_refres_delta;
2289 error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
2293 /* must not be a snapshot */
2294 if (ds->ds_is_snapshot) {
2295 dsl_dataset_rele(ds, FTAG);
2296 return (SET_ERROR(EINVAL));
2299 /* must have a most recent snapshot */
2300 if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
2301 dsl_dataset_rele(ds, FTAG);
2302 return (SET_ERROR(EINVAL));
2306 * No rollback to a snapshot created in the current txg, because
2307 * the rollback may dirty the dataset and create blocks that are
2308 * not reachable from the rootbp while having a birth txg that
2309 * falls into the snapshot's range.
2311 if (dmu_tx_is_syncing(tx) &&
2312 dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) {
2313 dsl_dataset_rele(ds, FTAG);
2314 return (SET_ERROR(EAGAIN));
2317 /* must not have any bookmarks after the most recent snapshot */
2318 nvlist_t *proprequest = fnvlist_alloc();
2319 fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
2320 nvlist_t *bookmarks = fnvlist_alloc();
2321 error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
2322 fnvlist_free(proprequest);
2325 for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
2326 pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
2328 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
2329 zfs_prop_to_name(ZFS_PROP_CREATETXG));
2330 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
2331 if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
2332 fnvlist_free(bookmarks);
2333 dsl_dataset_rele(ds, FTAG);
2334 return (SET_ERROR(EEXIST));
2337 fnvlist_free(bookmarks);
2339 error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
2341 dsl_dataset_rele(ds, FTAG);
2346 * Check if the snap we are rolling back to uses more than
2349 if (ds->ds_quota != 0 &&
2350 dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
2351 dsl_dataset_rele(ds, FTAG);
2352 return (SET_ERROR(EDQUOT));
2356 * When we do the clone swap, we will temporarily use more space
2357 * due to the refreservation (the head will no longer have any
2358 * unique space, so the entire amount of the refreservation will need
2359 * to be free). We will immediately destroy the clone, freeing
2360 * this space, but the freeing happens over many txg's.
2362 unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2363 dsl_dataset_phys(ds)->ds_unique_bytes);
2365 if (unused_refres_delta > 0 &&
2366 unused_refres_delta >
2367 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2368 dsl_dataset_rele(ds, FTAG);
2369 return (SET_ERROR(ENOSPC));
2372 dsl_dataset_rele(ds, FTAG);
2377 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2379 dsl_dataset_rollback_arg_t *ddra = arg;
2380 dsl_pool_t *dp = dmu_tx_pool(tx);
2381 dsl_dataset_t *ds, *clone;
2383 char namebuf[ZFS_MAX_DATASET_NAME_LEN];
2385 VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2387 dsl_dataset_name(ds->ds_prev, namebuf);
2388 fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2390 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2391 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2393 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2395 dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2396 dsl_dataset_zero_zil(ds, tx);
2398 dsl_destroy_head_sync_impl(clone, tx);
2400 dsl_dataset_rele(clone, FTAG);
2401 dsl_dataset_rele(ds, FTAG);
2405 * Rolls back the given filesystem or volume to the most recent snapshot.
2406 * The name of the most recent snapshot will be returned under key "target"
2407 * in the result nvlist.
2410 * - The existing dataset MUST be owned by the specified owner at entry
2411 * - Upon return, dataset will still be held by the same owner, whether we
2414 * This mode is required any time the existing filesystem is mounted. See
2415 * notes above zfs_suspend_fs() for further details.
2418 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
2420 dsl_dataset_rollback_arg_t ddra;
2422 ddra.ddra_fsname = fsname;
2423 ddra.ddra_owner = owner;
2424 ddra.ddra_result = result;
2426 return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2427 dsl_dataset_rollback_sync, &ddra,
2428 1, ZFS_SPACE_CHECK_RESERVED));
2431 struct promotenode {
2436 typedef struct dsl_dataset_promote_arg {
2437 const char *ddpa_clonename;
2438 dsl_dataset_t *ddpa_clone;
2439 list_t shared_snaps, origin_snaps, clone_snaps;
2440 dsl_dataset_t *origin_origin; /* origin of the origin */
2441 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2444 } dsl_dataset_promote_arg_t;
2446 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2447 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2449 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2452 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2454 dsl_dataset_promote_arg_t *ddpa = arg;
2455 dsl_pool_t *dp = dmu_tx_pool(tx);
2457 struct promotenode *snap;
2458 dsl_dataset_t *origin_ds;
2462 size_t max_snap_len;
2464 err = promote_hold(ddpa, dp, FTAG);
2468 hds = ddpa->ddpa_clone;
2469 max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
2471 if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2472 promote_rele(ddpa, FTAG);
2473 return (SET_ERROR(EXDEV));
2477 * Compute and check the amount of space to transfer. Since this is
2478 * so expensive, don't do the preliminary check.
2480 if (!dmu_tx_is_syncing(tx)) {
2481 promote_rele(ddpa, FTAG);
2485 snap = list_head(&ddpa->shared_snaps);
2486 origin_ds = snap->ds;
2488 /* compute origin's new unique space */
2489 snap = list_tail(&ddpa->clone_snaps);
2490 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2491 origin_ds->ds_object);
2492 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2493 dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2494 &ddpa->unique, &unused, &unused);
2497 * Walk the snapshots that we are moving
2499 * Compute space to transfer. Consider the incremental changes
2500 * to used by each snapshot:
2501 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2502 * So each snapshot gave birth to:
2503 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2504 * So a sequence would look like:
2505 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2506 * Which simplifies to:
2507 * uN + kN + kN-1 + ... + k1 + k0
2508 * Note however, if we stop before we reach the ORIGIN we get:
2509 * uN + kN + kN-1 + ... + kM - uM-1
2512 ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2513 ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2514 ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2515 for (snap = list_head(&ddpa->shared_snaps); snap;
2516 snap = list_next(&ddpa->shared_snaps, snap)) {
2517 uint64_t val, dlused, dlcomp, dluncomp;
2518 dsl_dataset_t *ds = snap->ds;
2523 * If there are long holds, we won't be able to evict
2526 if (dsl_dataset_long_held(ds)) {
2527 err = SET_ERROR(EBUSY);
2531 /* Check that the snapshot name does not conflict */
2532 VERIFY0(dsl_dataset_get_snapname(ds));
2533 if (strlen(ds->ds_snapname) >= max_snap_len) {
2534 err = SET_ERROR(ENAMETOOLONG);
2537 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2539 (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2540 err = SET_ERROR(EEXIST);
2546 /* The very first snapshot does not have a deadlist */
2547 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2550 dsl_deadlist_space(&ds->ds_deadlist,
2551 &dlused, &dlcomp, &dluncomp);
2552 ddpa->used += dlused;
2553 ddpa->comp += dlcomp;
2554 ddpa->uncomp += dluncomp;
2558 * If we are a clone of a clone then we never reached ORIGIN,
2559 * so we need to subtract out the clone origin's used space.
2561 if (ddpa->origin_origin) {
2563 dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2565 dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2567 dsl_dataset_phys(ddpa->origin_origin)->
2568 ds_uncompressed_bytes;
2571 /* Check that there is enough space and limit headroom here */
2572 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2573 0, ss_mv_cnt, ddpa->used, ddpa->cr);
2578 * Compute the amounts of space that will be used by snapshots
2579 * after the promotion (for both origin and clone). For each,
2580 * it is the amount of space that will be on all of their
2581 * deadlists (that was not born before their new origin).
2583 if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2587 * Note, typically this will not be a clone of a clone,
2588 * so dd_origin_txg will be < TXG_INITIAL, so
2589 * these snaplist_space() -> dsl_deadlist_space_range()
2590 * calls will be fast because they do not have to
2591 * iterate over all bps.
2593 snap = list_head(&ddpa->origin_snaps);
2594 err = snaplist_space(&ddpa->shared_snaps,
2595 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2599 err = snaplist_space(&ddpa->clone_snaps,
2600 snap->ds->ds_dir->dd_origin_txg, &space);
2603 ddpa->cloneusedsnap += space;
2605 if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2606 DD_FLAG_USED_BREAKDOWN) {
2607 err = snaplist_space(&ddpa->origin_snaps,
2608 dsl_dataset_phys(origin_ds)->ds_creation_txg,
2609 &ddpa->originusedsnap);
2615 promote_rele(ddpa, FTAG);
2620 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2622 dsl_dataset_promote_arg_t *ddpa = arg;
2623 dsl_pool_t *dp = dmu_tx_pool(tx);
2625 struct promotenode *snap;
2626 dsl_dataset_t *origin_ds;
2627 dsl_dataset_t *origin_head;
2629 dsl_dir_t *odd = NULL;
2630 uint64_t oldnext_obj;
2632 #if defined(__FreeBSD__) && defined(_KERNEL)
2633 char *oldname, *newname;
2636 VERIFY0(promote_hold(ddpa, dp, FTAG));
2637 hds = ddpa->ddpa_clone;
2639 ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
2641 snap = list_head(&ddpa->shared_snaps);
2642 origin_ds = snap->ds;
2645 snap = list_head(&ddpa->origin_snaps);
2646 origin_head = snap->ds;
2649 * We need to explicitly open odd, since origin_ds's dd will be
2652 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2655 /* change origin's next snap */
2656 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2657 oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
2658 snap = list_tail(&ddpa->clone_snaps);
2659 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2660 origin_ds->ds_object);
2661 dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
2663 /* change the origin's next clone */
2664 if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
2665 dsl_dataset_remove_from_next_clones(origin_ds,
2666 snap->ds->ds_object, tx);
2667 VERIFY0(zap_add_int(dp->dp_meta_objset,
2668 dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
2673 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2674 ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
2675 dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
2676 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2677 dmu_buf_will_dirty(odd->dd_dbuf, tx);
2678 dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
2679 origin_head->ds_dir->dd_origin_txg =
2680 dsl_dataset_phys(origin_ds)->ds_creation_txg;
2682 /* change dd_clone entries */
2683 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2684 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2685 dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
2686 VERIFY0(zap_add_int(dp->dp_meta_objset,
2687 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2688 hds->ds_object, tx));
2690 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2691 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2692 origin_head->ds_object, tx));
2693 if (dsl_dir_phys(dd)->dd_clones == 0) {
2694 dsl_dir_phys(dd)->dd_clones =
2695 zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
2696 DMU_OT_NONE, 0, tx);
2698 VERIFY0(zap_add_int(dp->dp_meta_objset,
2699 dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
2702 #if defined(__FreeBSD__) && defined(_KERNEL)
2703 /* Take the spa_namespace_lock early so zvol renames don't deadlock. */
2704 mutex_enter(&spa_namespace_lock);
2706 oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2707 newname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2710 /* move snapshots to this dir */
2711 for (snap = list_head(&ddpa->shared_snaps); snap;
2712 snap = list_next(&ddpa->shared_snaps, snap)) {
2713 dsl_dataset_t *ds = snap->ds;
2716 * Property callbacks are registered to a particular
2717 * dsl_dir. Since ours is changing, evict the objset
2718 * so that they will be unregistered from the old dsl_dir.
2720 if (ds->ds_objset) {
2721 dmu_objset_evict(ds->ds_objset);
2722 ds->ds_objset = NULL;
2725 /* move snap name entry */
2726 VERIFY0(dsl_dataset_get_snapname(ds));
2727 VERIFY0(dsl_dataset_snap_remove(origin_head,
2728 ds->ds_snapname, tx, B_TRUE));
2729 VERIFY0(zap_add(dp->dp_meta_objset,
2730 dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
2731 8, 1, &ds->ds_object, tx));
2732 dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2733 DD_FIELD_SNAPSHOT_COUNT, tx);
2735 /* change containing dsl_dir */
2736 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2737 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
2738 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
2739 ASSERT3P(ds->ds_dir, ==, odd);
2740 dsl_dir_rele(ds->ds_dir, ds);
2741 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2742 NULL, ds, &ds->ds_dir));
2744 #if defined(__FreeBSD__) && defined(_KERNEL)
2745 dsl_dataset_name(ds, newname);
2746 zfsvfs_update_fromname(oldname, newname);
2747 zvol_rename_minors(oldname, newname);
2750 /* move any clone references */
2751 if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
2752 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2756 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2757 dsl_dataset_phys(ds)->ds_next_clones_obj);
2758 zap_cursor_retrieve(&zc, &za) == 0;
2759 zap_cursor_advance(&zc)) {
2760 dsl_dataset_t *cnds;
2763 if (za.za_first_integer == oldnext_obj) {
2765 * We've already moved the
2766 * origin's reference.
2771 VERIFY0(dsl_dataset_hold_obj(dp,
2772 za.za_first_integer, FTAG, &cnds));
2773 o = dsl_dir_phys(cnds->ds_dir)->
2774 dd_head_dataset_obj;
2776 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2777 dsl_dir_phys(odd)->dd_clones, o, tx));
2778 VERIFY0(zap_add_int(dp->dp_meta_objset,
2779 dsl_dir_phys(dd)->dd_clones, o, tx));
2780 dsl_dataset_rele(cnds, FTAG);
2782 zap_cursor_fini(&zc);
2785 ASSERT(!dsl_prop_hascb(ds));
2788 #if defined(__FreeBSD__) && defined(_KERNEL)
2789 mutex_exit(&spa_namespace_lock);
2791 kmem_free(newname, MAXPATHLEN);
2792 kmem_free(oldname, MAXPATHLEN);
2795 * Change space accounting.
2796 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2797 * both be valid, or both be 0 (resulting in delta == 0). This
2798 * is true for each of {clone,origin} independently.
2801 delta = ddpa->cloneusedsnap -
2802 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
2803 ASSERT3S(delta, >=, 0);
2804 ASSERT3U(ddpa->used, >=, delta);
2805 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2806 dsl_dir_diduse_space(dd, DD_USED_HEAD,
2807 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2809 delta = ddpa->originusedsnap -
2810 dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
2811 ASSERT3S(delta, <=, 0);
2812 ASSERT3U(ddpa->used, >=, -delta);
2813 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2814 dsl_dir_diduse_space(odd, DD_USED_HEAD,
2815 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2817 dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
2819 /* log history record */
2820 spa_history_log_internal_ds(hds, "promote", tx, "");
2822 dsl_dir_rele(odd, FTAG);
2823 promote_rele(ddpa, FTAG);
2827 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2828 * (exclusive) and last_obj (inclusive). The list will be in reverse
2829 * order (last_obj will be the list_head()). If first_obj == 0, do all
2830 * snapshots back to this dataset's origin.
2833 snaplist_make(dsl_pool_t *dp,
2834 uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2836 uint64_t obj = last_obj;
2838 list_create(l, sizeof (struct promotenode),
2839 offsetof(struct promotenode, link));
2841 while (obj != first_obj) {
2843 struct promotenode *snap;
2846 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2847 ASSERT(err != ENOENT);
2852 first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
2854 snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2856 list_insert_tail(l, snap);
2857 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
2864 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2866 struct promotenode *snap;
2869 for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2870 uint64_t used, comp, uncomp;
2871 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2872 mintxg, UINT64_MAX, &used, &comp, &uncomp);
2879 snaplist_destroy(list_t *l, void *tag)
2881 struct promotenode *snap;
2883 if (l == NULL || !list_link_active(&l->list_head))
2886 while ((snap = list_tail(l)) != NULL) {
2887 list_remove(l, snap);
2888 dsl_dataset_rele(snap->ds, tag);
2889 kmem_free(snap, sizeof (*snap));
2895 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2899 struct promotenode *snap;
2901 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2905 dd = ddpa->ddpa_clone->ds_dir;
2907 if (ddpa->ddpa_clone->ds_is_snapshot ||
2908 !dsl_dir_is_clone(dd)) {
2909 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2910 return (SET_ERROR(EINVAL));
2913 error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
2914 &ddpa->shared_snaps, tag);
2918 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2919 &ddpa->clone_snaps, tag);
2923 snap = list_head(&ddpa->shared_snaps);
2924 ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
2925 error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
2926 dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
2927 &ddpa->origin_snaps, tag);
2931 if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
2932 error = dsl_dataset_hold_obj(dp,
2933 dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
2934 tag, &ddpa->origin_origin);
2940 promote_rele(ddpa, tag);
2945 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2947 snaplist_destroy(&ddpa->shared_snaps, tag);
2948 snaplist_destroy(&ddpa->clone_snaps, tag);
2949 snaplist_destroy(&ddpa->origin_snaps, tag);
2950 if (ddpa->origin_origin != NULL)
2951 dsl_dataset_rele(ddpa->origin_origin, tag);
2952 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2958 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2959 * in with the name. (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.)
2962 dsl_dataset_promote(const char *name, char *conflsnap)
2964 dsl_dataset_promote_arg_t ddpa = { 0 };
2970 * We will modify space proportional to the number of
2971 * snapshots. Compute numsnaps.
2973 error = dmu_objset_hold(name, FTAG, &os);
2976 error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2977 dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
2979 dmu_objset_rele(os, FTAG);
2983 ddpa.ddpa_clonename = name;
2984 ddpa.err_ds = conflsnap;
2987 return (dsl_sync_task(name, dsl_dataset_promote_check,
2988 dsl_dataset_promote_sync, &ddpa,
2989 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED));
2993 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2994 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2997 * "slack" factor for received datasets with refquota set on them.
2998 * See the bottom of this function for details on its use.
3000 uint64_t refquota_slack = DMU_MAX_ACCESS * spa_asize_inflation;
3001 int64_t unused_refres_delta;
3003 /* they should both be heads */
3004 if (clone->ds_is_snapshot ||
3005 origin_head->ds_is_snapshot)
3006 return (SET_ERROR(EINVAL));
3008 /* if we are not forcing, the branch point should be just before them */
3009 if (!force && clone->ds_prev != origin_head->ds_prev)
3010 return (SET_ERROR(EINVAL));
3012 /* clone should be the clone (unless they are unrelated) */
3013 if (clone->ds_prev != NULL &&
3014 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
3015 origin_head->ds_dir != clone->ds_prev->ds_dir)
3016 return (SET_ERROR(EINVAL));
3018 /* the clone should be a child of the origin */
3019 if (clone->ds_dir->dd_parent != origin_head->ds_dir)
3020 return (SET_ERROR(EINVAL));
3022 /* origin_head shouldn't be modified unless 'force' */
3024 dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
3025 return (SET_ERROR(ETXTBSY));
3027 /* origin_head should have no long holds (e.g. is not mounted) */
3028 if (dsl_dataset_handoff_check(origin_head, owner, tx))
3029 return (SET_ERROR(EBUSY));
3031 /* check amount of any unconsumed refreservation */
3032 unused_refres_delta =
3033 (int64_t)MIN(origin_head->ds_reserved,
3034 dsl_dataset_phys(origin_head)->ds_unique_bytes) -
3035 (int64_t)MIN(origin_head->ds_reserved,
3036 dsl_dataset_phys(clone)->ds_unique_bytes);
3038 if (unused_refres_delta > 0 &&
3039 unused_refres_delta >
3040 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
3041 return (SET_ERROR(ENOSPC));
3044 * The clone can't be too much over the head's refquota.
3046 * To ensure that the entire refquota can be used, we allow one
3047 * transaction to exceed the the refquota. Therefore, this check
3048 * needs to also allow for the space referenced to be more than the
3049 * refquota. The maximum amount of space that one transaction can use
3050 * on disk is DMU_MAX_ACCESS * spa_asize_inflation. Allowing this
3051 * overage ensures that we are able to receive a filesystem that
3052 * exceeds the refquota on the source system.
3054 * So that overage is the refquota_slack we use below.
3056 if (origin_head->ds_quota != 0 &&
3057 dsl_dataset_phys(clone)->ds_referenced_bytes >
3058 origin_head->ds_quota + refquota_slack)
3059 return (SET_ERROR(EDQUOT));
3065 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
3066 dsl_dataset_t *origin_head, dmu_tx_t *tx)
3068 dsl_pool_t *dp = dmu_tx_pool(tx);
3069 int64_t unused_refres_delta;
3071 ASSERT(clone->ds_reserved == 0);
3073 * NOTE: On DEBUG kernels there could be a race between this and
3074 * the check function if spa_asize_inflation is adjusted...
3076 ASSERT(origin_head->ds_quota == 0 ||
3077 dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota +
3078 DMU_MAX_ACCESS * spa_asize_inflation);
3079 ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
3082 * Swap per-dataset feature flags.
3084 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
3085 if (!(spa_feature_table[f].fi_flags &
3086 ZFEATURE_FLAG_PER_DATASET)) {
3087 ASSERT(!clone->ds_feature_inuse[f]);
3088 ASSERT(!origin_head->ds_feature_inuse[f]);
3092 boolean_t clone_inuse = clone->ds_feature_inuse[f];
3093 boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f];
3096 dsl_dataset_deactivate_feature(clone->ds_object, f, tx);
3097 clone->ds_feature_inuse[f] = B_FALSE;
3099 if (origin_head_inuse) {
3100 dsl_dataset_deactivate_feature(origin_head->ds_object,
3102 origin_head->ds_feature_inuse[f] = B_FALSE;
3105 dsl_dataset_activate_feature(origin_head->ds_object,
3107 origin_head->ds_feature_inuse[f] = B_TRUE;
3109 if (origin_head_inuse) {
3110 dsl_dataset_activate_feature(clone->ds_object, f, tx);
3111 clone->ds_feature_inuse[f] = B_TRUE;
3115 dmu_buf_will_dirty(clone->ds_dbuf, tx);
3116 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
3118 if (clone->ds_objset != NULL) {
3119 dmu_objset_evict(clone->ds_objset);
3120 clone->ds_objset = NULL;
3123 if (origin_head->ds_objset != NULL) {
3124 dmu_objset_evict(origin_head->ds_objset);
3125 origin_head->ds_objset = NULL;
3128 unused_refres_delta =
3129 (int64_t)MIN(origin_head->ds_reserved,
3130 dsl_dataset_phys(origin_head)->ds_unique_bytes) -
3131 (int64_t)MIN(origin_head->ds_reserved,
3132 dsl_dataset_phys(clone)->ds_unique_bytes);
3135 * Reset origin's unique bytes, if it exists.
3137 if (clone->ds_prev) {
3138 dsl_dataset_t *origin = clone->ds_prev;
3139 uint64_t comp, uncomp;
3141 dmu_buf_will_dirty(origin->ds_dbuf, tx);
3142 dsl_deadlist_space_range(&clone->ds_deadlist,
3143 dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
3144 &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
3149 rrw_enter(&clone->ds_bp_rwlock, RW_WRITER, FTAG);
3150 rrw_enter(&origin_head->ds_bp_rwlock, RW_WRITER, FTAG);
3152 tmp = dsl_dataset_phys(origin_head)->ds_bp;
3153 dsl_dataset_phys(origin_head)->ds_bp =
3154 dsl_dataset_phys(clone)->ds_bp;
3155 dsl_dataset_phys(clone)->ds_bp = tmp;
3156 rrw_exit(&origin_head->ds_bp_rwlock, FTAG);
3157 rrw_exit(&clone->ds_bp_rwlock, FTAG);
3160 /* set dd_*_bytes */
3162 int64_t dused, dcomp, duncomp;
3163 uint64_t cdl_used, cdl_comp, cdl_uncomp;
3164 uint64_t odl_used, odl_comp, odl_uncomp;
3166 ASSERT3U(dsl_dir_phys(clone->ds_dir)->
3167 dd_used_breakdown[DD_USED_SNAP], ==, 0);
3169 dsl_deadlist_space(&clone->ds_deadlist,
3170 &cdl_used, &cdl_comp, &cdl_uncomp);
3171 dsl_deadlist_space(&origin_head->ds_deadlist,
3172 &odl_used, &odl_comp, &odl_uncomp);
3174 dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
3176 (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
3178 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
3180 (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
3182 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
3184 (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
3187 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
3188 dused, dcomp, duncomp, tx);
3189 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
3190 -dused, -dcomp, -duncomp, tx);
3193 * The difference in the space used by snapshots is the
3194 * difference in snapshot space due to the head's
3195 * deadlist (since that's the only thing that's
3196 * changing that affects the snapused).
3198 dsl_deadlist_space_range(&clone->ds_deadlist,
3199 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3200 &cdl_used, &cdl_comp, &cdl_uncomp);
3201 dsl_deadlist_space_range(&origin_head->ds_deadlist,
3202 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3203 &odl_used, &odl_comp, &odl_uncomp);
3204 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
3205 DD_USED_HEAD, DD_USED_SNAP, NULL);
3208 /* swap ds_*_bytes */
3209 SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
3210 dsl_dataset_phys(clone)->ds_referenced_bytes);
3211 SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
3212 dsl_dataset_phys(clone)->ds_compressed_bytes);
3213 SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
3214 dsl_dataset_phys(clone)->ds_uncompressed_bytes);
3215 SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
3216 dsl_dataset_phys(clone)->ds_unique_bytes);
3218 /* apply any parent delta for change in unconsumed refreservation */
3219 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
3220 unused_refres_delta, 0, 0, tx);
3225 dsl_deadlist_close(&clone->ds_deadlist);
3226 dsl_deadlist_close(&origin_head->ds_deadlist);
3227 SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
3228 dsl_dataset_phys(clone)->ds_deadlist_obj);
3229 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
3230 dsl_dataset_phys(clone)->ds_deadlist_obj);
3231 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
3232 dsl_dataset_phys(origin_head)->ds_deadlist_obj);
3234 dsl_scan_ds_clone_swapped(origin_head, clone, tx);
3236 spa_history_log_internal_ds(clone, "clone swap", tx,
3237 "parent=%s", origin_head->ds_dir->dd_myname);
3241 * Given a pool name and a dataset object number in that pool,
3242 * return the name of that dataset.
3245 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3251 error = dsl_pool_hold(pname, FTAG, &dp);
3255 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
3257 dsl_dataset_name(ds, buf);
3258 dsl_dataset_rele(ds, FTAG);
3260 dsl_pool_rele(dp, FTAG);
3266 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3267 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3271 ASSERT3S(asize, >, 0);
3274 * *ref_rsrv is the portion of asize that will come from any
3275 * unconsumed refreservation space.
3279 mutex_enter(&ds->ds_lock);
3281 * Make a space adjustment for reserved bytes.
3283 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
3285 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3287 (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3289 asize - MIN(asize, parent_delta(ds, asize + inflight));
3292 if (!check_quota || ds->ds_quota == 0) {
3293 mutex_exit(&ds->ds_lock);
3297 * If they are requesting more space, and our current estimate
3298 * is over quota, they get to try again unless the actual
3299 * on-disk is over quota and there are no pending changes (which
3300 * may free up space for us).
3302 if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
3305 dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
3306 error = SET_ERROR(ERESTART);
3308 error = SET_ERROR(EDQUOT);
3310 mutex_exit(&ds->ds_lock);
3315 typedef struct dsl_dataset_set_qr_arg {
3316 const char *ddsqra_name;
3317 zprop_source_t ddsqra_source;
3318 uint64_t ddsqra_value;
3319 } dsl_dataset_set_qr_arg_t;
3324 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
3326 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3327 dsl_pool_t *dp = dmu_tx_pool(tx);
3332 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
3333 return (SET_ERROR(ENOTSUP));
3335 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3339 if (ds->ds_is_snapshot) {
3340 dsl_dataset_rele(ds, FTAG);
3341 return (SET_ERROR(EINVAL));
3344 error = dsl_prop_predict(ds->ds_dir,
3345 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3346 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3348 dsl_dataset_rele(ds, FTAG);
3353 dsl_dataset_rele(ds, FTAG);
3357 if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
3358 newval < ds->ds_reserved) {
3359 dsl_dataset_rele(ds, FTAG);
3360 return (SET_ERROR(ENOSPC));
3363 dsl_dataset_rele(ds, FTAG);
3368 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
3370 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3371 dsl_pool_t *dp = dmu_tx_pool(tx);
3375 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3377 dsl_prop_set_sync_impl(ds,
3378 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3379 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
3380 &ddsqra->ddsqra_value, tx);
3382 VERIFY0(dsl_prop_get_int_ds(ds,
3383 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
3385 if (ds->ds_quota != newval) {
3386 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3387 ds->ds_quota = newval;
3389 dsl_dataset_rele(ds, FTAG);
3393 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
3396 dsl_dataset_set_qr_arg_t ddsqra;
3398 ddsqra.ddsqra_name = dsname;
3399 ddsqra.ddsqra_source = source;
3400 ddsqra.ddsqra_value = refquota;
3402 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
3403 dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
3407 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
3409 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3410 dsl_pool_t *dp = dmu_tx_pool(tx);
3413 uint64_t newval, unique;
3415 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
3416 return (SET_ERROR(ENOTSUP));
3418 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3422 if (ds->ds_is_snapshot) {
3423 dsl_dataset_rele(ds, FTAG);
3424 return (SET_ERROR(EINVAL));
3427 error = dsl_prop_predict(ds->ds_dir,
3428 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3429 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3431 dsl_dataset_rele(ds, FTAG);
3436 * If we are doing the preliminary check in open context, the
3437 * space estimates may be inaccurate.
3439 if (!dmu_tx_is_syncing(tx)) {
3440 dsl_dataset_rele(ds, FTAG);
3444 mutex_enter(&ds->ds_lock);
3445 if (!DS_UNIQUE_IS_ACCURATE(ds))
3446 dsl_dataset_recalc_head_uniq(ds);
3447 unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3448 mutex_exit(&ds->ds_lock);
3450 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3451 uint64_t delta = MAX(unique, newval) -
3452 MAX(unique, ds->ds_reserved);
3455 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3456 (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3457 dsl_dataset_rele(ds, FTAG);
3458 return (SET_ERROR(ENOSPC));
3462 dsl_dataset_rele(ds, FTAG);
3467 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3468 zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3474 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3475 source, sizeof (value), 1, &value, tx);
3477 VERIFY0(dsl_prop_get_int_ds(ds,
3478 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3480 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3481 mutex_enter(&ds->ds_dir->dd_lock);
3482 mutex_enter(&ds->ds_lock);
3483 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3484 unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3485 delta = MAX(0, (int64_t)(newval - unique)) -
3486 MAX(0, (int64_t)(ds->ds_reserved - unique));
3487 ds->ds_reserved = newval;
3488 mutex_exit(&ds->ds_lock);
3490 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3491 mutex_exit(&ds->ds_dir->dd_lock);
3495 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3497 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3498 dsl_pool_t *dp = dmu_tx_pool(tx);
3501 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3502 dsl_dataset_set_refreservation_sync_impl(ds,
3503 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3504 dsl_dataset_rele(ds, FTAG);
3508 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3509 uint64_t refreservation)
3511 dsl_dataset_set_qr_arg_t ddsqra;
3513 ddsqra.ddsqra_name = dsname;
3514 ddsqra.ddsqra_source = source;
3515 ddsqra.ddsqra_value = refreservation;
3517 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3518 dsl_dataset_set_refreservation_sync, &ddsqra,
3519 0, ZFS_SPACE_CHECK_NONE));
3523 * Return (in *usedp) the amount of space written in new that is not
3524 * present in oldsnap. New may be a snapshot or the head. Old must be
3525 * a snapshot before new, in new's filesystem (or its origin). If not then
3526 * fail and return EINVAL.
3528 * The written space is calculated by considering two components: First, we
3529 * ignore any freed space, and calculate the written as new's used space
3530 * minus old's used space. Next, we add in the amount of space that was freed
3531 * between the two snapshots, thus reducing new's used space relative to old's.
3532 * Specifically, this is the space that was born before old->ds_creation_txg,
3533 * and freed before new (ie. on new's deadlist or a previous deadlist).
3535 * space freed [---------------------]
3536 * snapshots ---O-------O--------O-------O------
3540 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3541 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3545 dsl_pool_t *dp = new->ds_dir->dd_pool;
3547 ASSERT(dsl_pool_config_held(dp));
3550 *usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3551 *usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3554 *compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3555 *compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3558 *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3559 *uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3561 snapobj = new->ds_object;
3562 while (snapobj != oldsnap->ds_object) {
3563 dsl_dataset_t *snap;
3564 uint64_t used, comp, uncomp;
3566 if (snapobj == new->ds_object) {
3569 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3574 if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
3575 dsl_dataset_phys(oldsnap)->ds_creation_txg) {
3577 * The blocks in the deadlist can not be born after
3578 * ds_prev_snap_txg, so get the whole deadlist space,
3579 * which is more efficient (especially for old-format
3580 * deadlists). Unfortunately the deadlist code
3581 * doesn't have enough information to make this
3582 * optimization itself.
3584 dsl_deadlist_space(&snap->ds_deadlist,
3585 &used, &comp, &uncomp);
3587 dsl_deadlist_space_range(&snap->ds_deadlist,
3588 0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
3589 &used, &comp, &uncomp);
3596 * If we get to the beginning of the chain of snapshots
3597 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3598 * was not a snapshot of/before new.
3600 snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
3602 dsl_dataset_rele(snap, FTAG);
3604 err = SET_ERROR(EINVAL);
3613 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3614 * lastsnap, and all snapshots in between are deleted.
3616 * blocks that would be freed [---------------------------]
3617 * snapshots ---O-------O--------O-------O--------O
3618 * firstsnap lastsnap
3620 * This is the set of blocks that were born after the snap before firstsnap,
3621 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3622 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3623 * We calculate this by iterating over the relevant deadlists (from the snap
3624 * after lastsnap, backward to the snap after firstsnap), summing up the
3625 * space on the deadlist that was born after the snap before firstsnap.
3628 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3629 dsl_dataset_t *lastsnap,
3630 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3634 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3636 ASSERT(firstsnap->ds_is_snapshot);
3637 ASSERT(lastsnap->ds_is_snapshot);
3640 * Check that the snapshots are in the same dsl_dir, and firstsnap
3641 * is before lastsnap.
3643 if (firstsnap->ds_dir != lastsnap->ds_dir ||
3644 dsl_dataset_phys(firstsnap)->ds_creation_txg >
3645 dsl_dataset_phys(lastsnap)->ds_creation_txg)
3646 return (SET_ERROR(EINVAL));
3648 *usedp = *compp = *uncompp = 0;
3650 snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
3651 while (snapobj != firstsnap->ds_object) {
3653 uint64_t used, comp, uncomp;
3655 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3659 dsl_deadlist_space_range(&ds->ds_deadlist,
3660 dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
3661 &used, &comp, &uncomp);
3666 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3667 ASSERT3U(snapobj, !=, 0);
3668 dsl_dataset_rele(ds, FTAG);
3674 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3675 * For example, they could both be snapshots of the same filesystem, and
3676 * 'earlier' is before 'later'. Or 'earlier' could be the origin of
3677 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's
3678 * filesystem. Or 'earlier' could be the origin's origin.
3680 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3683 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3684 uint64_t earlier_txg)
3686 dsl_pool_t *dp = later->ds_dir->dd_pool;
3690 ASSERT(dsl_pool_config_held(dp));
3691 ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
3693 if (earlier_txg == 0)
3694 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
3696 if (later->ds_is_snapshot &&
3697 earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
3700 if (later->ds_dir == earlier->ds_dir)
3702 if (!dsl_dir_is_clone(later->ds_dir))
3705 if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
3707 dsl_dataset_t *origin;
3708 error = dsl_dataset_hold_obj(dp,
3709 dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
3712 ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3713 dsl_dataset_rele(origin, FTAG);
3718 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3720 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3721 dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3725 dsl_dataset_is_zapified(dsl_dataset_t *ds)
3727 dmu_object_info_t doi;
3729 dmu_object_info_from_db(ds->ds_dbuf, &doi);
3730 return (doi.doi_type == DMU_OTN_ZAP_METADATA);
3734 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds)
3736 return (dsl_dataset_is_zapified(ds) &&
3737 zap_contains(ds->ds_dir->dd_pool->dp_meta_objset,
3738 ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0);