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 2008 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
26 #include <sys/dmu_objset.h>
27 #include <sys/dsl_dataset.h>
28 #include <sys/dsl_dir.h>
29 #include <sys/dsl_prop.h>
30 #include <sys/dsl_synctask.h>
31 #include <sys/dmu_traverse.h>
32 #include <sys/dmu_tx.h>
36 #include <sys/unique.h>
37 #include <sys/zfs_context.h>
38 #include <sys/zfs_ioctl.h>
40 #include <sys/zfs_znode.h>
41 #include <sys/sunddi.h>
43 static char *dsl_reaper = "the grim reaper";
45 static dsl_checkfunc_t dsl_dataset_destroy_begin_check;
46 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync;
47 static dsl_checkfunc_t dsl_dataset_rollback_check;
48 static dsl_syncfunc_t dsl_dataset_rollback_sync;
49 static dsl_syncfunc_t dsl_dataset_set_reservation_sync;
51 #define DS_REF_MAX (1ULL << 62)
53 #define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE
55 #define DSL_DATASET_IS_DESTROYED(ds) ((ds)->ds_owner == dsl_reaper)
59 * Figure out how much of this delta should be propogated to the dsl_dir
60 * layer. If there's a refreservation, that space has already been
61 * partially accounted for in our ancestors.
64 parent_delta(dsl_dataset_t *ds, int64_t delta)
66 uint64_t old_bytes, new_bytes;
68 if (ds->ds_reserved == 0)
71 old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
72 new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
74 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
75 return (new_bytes - old_bytes);
79 dsl_dataset_block_born(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
81 int used = bp_get_dasize(tx->tx_pool->dp_spa, bp);
82 int compressed = BP_GET_PSIZE(bp);
83 int uncompressed = BP_GET_UCSIZE(bp);
86 dprintf_bp(bp, "born, ds=%p\n", ds);
88 ASSERT(dmu_tx_is_syncing(tx));
89 /* It could have been compressed away to nothing */
92 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
93 ASSERT3U(BP_GET_TYPE(bp), <, DMU_OT_NUMTYPES);
96 * Account for the meta-objset space in its placeholder
99 ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */
100 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
101 used, compressed, uncompressed, tx);
102 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
105 dmu_buf_will_dirty(ds->ds_dbuf, tx);
106 mutex_enter(&ds->ds_dir->dd_lock);
107 mutex_enter(&ds->ds_lock);
108 delta = parent_delta(ds, used);
109 ds->ds_phys->ds_used_bytes += used;
110 ds->ds_phys->ds_compressed_bytes += compressed;
111 ds->ds_phys->ds_uncompressed_bytes += uncompressed;
112 ds->ds_phys->ds_unique_bytes += used;
113 mutex_exit(&ds->ds_lock);
114 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
115 compressed, uncompressed, tx);
116 dsl_dir_transfer_space(ds->ds_dir, used - delta,
117 DD_USED_REFRSRV, DD_USED_HEAD, tx);
118 mutex_exit(&ds->ds_dir->dd_lock);
122 dsl_dataset_block_kill(dsl_dataset_t *ds, blkptr_t *bp, zio_t *pio,
125 int used = bp_get_dasize(tx->tx_pool->dp_spa, bp);
126 int compressed = BP_GET_PSIZE(bp);
127 int uncompressed = BP_GET_UCSIZE(bp);
130 ASSERT(dmu_tx_is_syncing(tx));
131 /* No block pointer => nothing to free */
139 * Account for the meta-objset space in its placeholder
142 err = dsl_free(pio, tx->tx_pool,
143 tx->tx_txg, bp, NULL, NULL, ARC_NOWAIT);
146 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
147 -used, -compressed, -uncompressed, tx);
148 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
151 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
153 ASSERT(!dsl_dataset_is_snapshot(ds));
154 dmu_buf_will_dirty(ds->ds_dbuf, tx);
156 if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
160 dprintf_bp(bp, "freeing: %s", "");
161 err = dsl_free(pio, tx->tx_pool,
162 tx->tx_txg, bp, NULL, NULL, ARC_NOWAIT);
165 mutex_enter(&ds->ds_dir->dd_lock);
166 mutex_enter(&ds->ds_lock);
167 ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
168 !DS_UNIQUE_IS_ACCURATE(ds));
169 delta = parent_delta(ds, -used);
170 ds->ds_phys->ds_unique_bytes -= used;
171 mutex_exit(&ds->ds_lock);
172 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
173 delta, -compressed, -uncompressed, tx);
174 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
175 DD_USED_REFRSRV, DD_USED_HEAD, tx);
176 mutex_exit(&ds->ds_dir->dd_lock);
178 dprintf_bp(bp, "putting on dead list: %s", "");
179 VERIFY(0 == bplist_enqueue(&ds->ds_deadlist, bp, tx));
180 ASSERT3U(ds->ds_prev->ds_object, ==,
181 ds->ds_phys->ds_prev_snap_obj);
182 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
183 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
184 if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
185 ds->ds_object && bp->blk_birth >
186 ds->ds_prev->ds_phys->ds_prev_snap_txg) {
187 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
188 mutex_enter(&ds->ds_prev->ds_lock);
189 ds->ds_prev->ds_phys->ds_unique_bytes += used;
190 mutex_exit(&ds->ds_prev->ds_lock);
192 if (bp->blk_birth > ds->ds_origin_txg) {
193 dsl_dir_transfer_space(ds->ds_dir, used,
194 DD_USED_HEAD, DD_USED_SNAP, tx);
197 mutex_enter(&ds->ds_lock);
198 ASSERT3U(ds->ds_phys->ds_used_bytes, >=, used);
199 ds->ds_phys->ds_used_bytes -= used;
200 ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
201 ds->ds_phys->ds_compressed_bytes -= compressed;
202 ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
203 ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
204 mutex_exit(&ds->ds_lock);
210 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
212 uint64_t trysnap = 0;
217 * The snapshot creation could fail, but that would cause an
218 * incorrect FALSE return, which would only result in an
219 * overestimation of the amount of space that an operation would
220 * consume, which is OK.
222 * There's also a small window where we could miss a pending
223 * snapshot, because we could set the sync task in the quiescing
224 * phase. So this should only be used as a guess.
226 if (ds->ds_trysnap_txg >
227 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
228 trysnap = ds->ds_trysnap_txg;
229 return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
233 dsl_dataset_block_freeable(dsl_dataset_t *ds, uint64_t blk_birth)
235 return (blk_birth > dsl_dataset_prev_snap_txg(ds));
240 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
242 dsl_dataset_t *ds = dsv;
244 ASSERT(ds->ds_owner == NULL || DSL_DATASET_IS_DESTROYED(ds));
246 dprintf_ds(ds, "evicting %s\n", "");
248 unique_remove(ds->ds_fsid_guid);
250 if (ds->ds_user_ptr != NULL)
251 ds->ds_user_evict_func(ds, ds->ds_user_ptr);
254 dsl_dataset_drop_ref(ds->ds_prev, ds);
258 bplist_close(&ds->ds_deadlist);
260 dsl_dir_close(ds->ds_dir, ds);
262 ASSERT(!list_link_active(&ds->ds_synced_link));
264 if (mutex_owned(&ds->ds_lock))
265 mutex_exit(&ds->ds_lock);
266 mutex_destroy(&ds->ds_lock);
267 if (mutex_owned(&ds->ds_opening_lock))
268 mutex_exit(&ds->ds_opening_lock);
269 mutex_destroy(&ds->ds_opening_lock);
270 if (mutex_owned(&ds->ds_deadlist.bpl_lock))
271 mutex_exit(&ds->ds_deadlist.bpl_lock);
272 mutex_destroy(&ds->ds_deadlist.bpl_lock);
273 rw_destroy(&ds->ds_rwlock);
274 cv_destroy(&ds->ds_exclusive_cv);
276 kmem_free(ds, sizeof (dsl_dataset_t));
280 dsl_dataset_get_snapname(dsl_dataset_t *ds)
282 dsl_dataset_phys_t *headphys;
285 dsl_pool_t *dp = ds->ds_dir->dd_pool;
286 objset_t *mos = dp->dp_meta_objset;
288 if (ds->ds_snapname[0])
290 if (ds->ds_phys->ds_next_snap_obj == 0)
293 err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
297 headphys = headdbuf->db_data;
298 err = zap_value_search(dp->dp_meta_objset,
299 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
300 dmu_buf_rele(headdbuf, FTAG);
305 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
307 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
308 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
312 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
317 err = zap_lookup_norm(mos, snapobj, name, 8, 1,
318 value, mt, NULL, 0, NULL);
319 if (err == ENOTSUP && mt == MT_FIRST)
320 err = zap_lookup(mos, snapobj, name, 8, 1, value);
325 dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx)
327 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
328 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
332 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
337 err = zap_remove_norm(mos, snapobj, name, mt, tx);
338 if (err == ENOTSUP && mt == MT_FIRST)
339 err = zap_remove(mos, snapobj, name, tx);
344 dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag,
347 objset_t *mos = dp->dp_meta_objset;
352 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
353 dsl_pool_sync_context(dp));
355 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
358 ds = dmu_buf_get_user(dbuf);
360 dsl_dataset_t *winner;
362 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
364 ds->ds_object = dsobj;
365 ds->ds_phys = dbuf->db_data;
367 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
368 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
369 mutex_init(&ds->ds_deadlist.bpl_lock, NULL, MUTEX_DEFAULT,
371 rw_init(&ds->ds_rwlock, 0, 0, 0);
372 cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL);
374 err = bplist_open(&ds->ds_deadlist,
375 mos, ds->ds_phys->ds_deadlist_obj);
377 err = dsl_dir_open_obj(dp,
378 ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
382 * we don't really need to close the blist if we
385 mutex_destroy(&ds->ds_lock);
386 mutex_destroy(&ds->ds_opening_lock);
387 mutex_destroy(&ds->ds_deadlist.bpl_lock);
388 rw_destroy(&ds->ds_rwlock);
389 cv_destroy(&ds->ds_exclusive_cv);
390 kmem_free(ds, sizeof (dsl_dataset_t));
391 dmu_buf_rele(dbuf, tag);
395 if (!dsl_dataset_is_snapshot(ds)) {
396 ds->ds_snapname[0] = '\0';
397 if (ds->ds_phys->ds_prev_snap_obj) {
398 err = dsl_dataset_get_ref(dp,
399 ds->ds_phys->ds_prev_snap_obj,
403 if (err == 0 && dsl_dir_is_clone(ds->ds_dir)) {
404 dsl_dataset_t *origin;
406 err = dsl_dataset_hold_obj(dp,
407 ds->ds_dir->dd_phys->dd_origin_obj,
411 origin->ds_phys->ds_creation_txg;
412 dsl_dataset_rele(origin, FTAG);
415 } else if (zfs_flags & ZFS_DEBUG_SNAPNAMES) {
416 err = dsl_dataset_get_snapname(ds);
419 if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
421 * In sync context, we're called with either no lock
422 * or with the write lock. If we're not syncing,
423 * we're always called with the read lock held.
425 boolean_t need_lock =
426 !RW_WRITE_HELD(&dp->dp_config_rwlock) &&
427 dsl_pool_sync_context(dp);
430 rw_enter(&dp->dp_config_rwlock, RW_READER);
432 err = dsl_prop_get_ds(ds,
433 "refreservation", sizeof (uint64_t), 1,
434 &ds->ds_reserved, NULL);
436 err = dsl_prop_get_ds(ds,
437 "refquota", sizeof (uint64_t), 1,
438 &ds->ds_quota, NULL);
442 rw_exit(&dp->dp_config_rwlock);
444 ds->ds_reserved = ds->ds_quota = 0;
448 winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys,
452 bplist_close(&ds->ds_deadlist);
454 dsl_dataset_drop_ref(ds->ds_prev, ds);
455 dsl_dir_close(ds->ds_dir, ds);
456 mutex_destroy(&ds->ds_lock);
457 mutex_destroy(&ds->ds_opening_lock);
458 mutex_destroy(&ds->ds_deadlist.bpl_lock);
459 rw_destroy(&ds->ds_rwlock);
460 cv_destroy(&ds->ds_exclusive_cv);
461 kmem_free(ds, sizeof (dsl_dataset_t));
463 dmu_buf_rele(dbuf, tag);
469 unique_insert(ds->ds_phys->ds_fsid_guid);
472 ASSERT3P(ds->ds_dbuf, ==, dbuf);
473 ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
474 ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
475 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
476 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
477 mutex_enter(&ds->ds_lock);
478 if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) {
479 mutex_exit(&ds->ds_lock);
480 dmu_buf_rele(ds->ds_dbuf, tag);
483 mutex_exit(&ds->ds_lock);
489 dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag)
491 dsl_pool_t *dp = ds->ds_dir->dd_pool;
494 * In syncing context we don't want the rwlock lock: there
495 * may be an existing writer waiting for sync phase to
496 * finish. We don't need to worry about such writers, since
497 * sync phase is single-threaded, so the writer can't be
498 * doing anything while we are active.
500 if (dsl_pool_sync_context(dp)) {
501 ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
506 * Normal users will hold the ds_rwlock as a READER until they
507 * are finished (i.e., call dsl_dataset_rele()). "Owners" will
508 * drop their READER lock after they set the ds_owner field.
510 * If the dataset is being destroyed, the destroy thread will
511 * obtain a WRITER lock for exclusive access after it's done its
512 * open-context work and then change the ds_owner to
513 * dsl_reaper once destruction is assured. So threads
514 * may block here temporarily, until the "destructability" of
515 * the dataset is determined.
517 ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock));
518 mutex_enter(&ds->ds_lock);
519 while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) {
520 rw_exit(&dp->dp_config_rwlock);
521 cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock);
522 if (DSL_DATASET_IS_DESTROYED(ds)) {
523 mutex_exit(&ds->ds_lock);
524 dsl_dataset_drop_ref(ds, tag);
525 rw_enter(&dp->dp_config_rwlock, RW_READER);
528 rw_enter(&dp->dp_config_rwlock, RW_READER);
530 mutex_exit(&ds->ds_lock);
535 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
538 int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp);
542 return (dsl_dataset_hold_ref(*dsp, tag));
546 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, int flags, void *owner,
549 int err = dsl_dataset_hold_obj(dp, dsobj, owner, dsp);
551 ASSERT(DS_MODE_TYPE(flags) != DS_MODE_USER);
555 if (!dsl_dataset_tryown(*dsp, DS_MODE_IS_INCONSISTENT(flags), owner)) {
556 dsl_dataset_rele(*dsp, owner);
563 dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp)
567 const char *snapname;
571 err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname);
576 obj = dd->dd_phys->dd_head_dataset_obj;
577 rw_enter(&dp->dp_config_rwlock, RW_READER);
579 err = dsl_dataset_get_ref(dp, obj, tag, dsp);
585 err = dsl_dataset_hold_ref(*dsp, tag);
587 /* we may be looking for a snapshot */
588 if (err == 0 && snapname != NULL) {
589 dsl_dataset_t *ds = NULL;
591 if (*snapname++ != '@') {
592 dsl_dataset_rele(*dsp, tag);
597 dprintf("looking for snapshot '%s'\n", snapname);
598 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
600 err = dsl_dataset_get_ref(dp, obj, tag, &ds);
601 dsl_dataset_rele(*dsp, tag);
603 ASSERT3U((err == 0), ==, (ds != NULL));
606 mutex_enter(&ds->ds_lock);
607 if (ds->ds_snapname[0] == 0)
608 (void) strlcpy(ds->ds_snapname, snapname,
609 sizeof (ds->ds_snapname));
610 mutex_exit(&ds->ds_lock);
611 err = dsl_dataset_hold_ref(ds, tag);
612 *dsp = err ? NULL : ds;
616 rw_exit(&dp->dp_config_rwlock);
617 dsl_dir_close(dd, FTAG);
622 dsl_dataset_own(const char *name, int flags, void *owner, dsl_dataset_t **dsp)
624 int err = dsl_dataset_hold(name, owner, dsp);
627 if ((*dsp)->ds_phys->ds_num_children > 0 &&
628 !DS_MODE_IS_READONLY(flags)) {
629 dsl_dataset_rele(*dsp, owner);
632 if (!dsl_dataset_tryown(*dsp, DS_MODE_IS_INCONSISTENT(flags), owner)) {
633 dsl_dataset_rele(*dsp, owner);
640 dsl_dataset_name(dsl_dataset_t *ds, char *name)
643 (void) strcpy(name, "mos");
645 dsl_dir_name(ds->ds_dir, name);
646 VERIFY(0 == dsl_dataset_get_snapname(ds));
647 if (ds->ds_snapname[0]) {
648 (void) strcat(name, "@");
650 * We use a "recursive" mutex so that we
651 * can call dprintf_ds() with ds_lock held.
653 if (!MUTEX_HELD(&ds->ds_lock)) {
654 mutex_enter(&ds->ds_lock);
655 (void) strcat(name, ds->ds_snapname);
656 mutex_exit(&ds->ds_lock);
658 (void) strcat(name, ds->ds_snapname);
665 dsl_dataset_namelen(dsl_dataset_t *ds)
670 result = 3; /* "mos" */
672 result = dsl_dir_namelen(ds->ds_dir);
673 VERIFY(0 == dsl_dataset_get_snapname(ds));
674 if (ds->ds_snapname[0]) {
675 ++result; /* adding one for the @-sign */
676 if (!MUTEX_HELD(&ds->ds_lock)) {
677 mutex_enter(&ds->ds_lock);
678 result += strlen(ds->ds_snapname);
679 mutex_exit(&ds->ds_lock);
681 result += strlen(ds->ds_snapname);
690 dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag)
692 dmu_buf_rele(ds->ds_dbuf, tag);
696 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
698 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) {
699 rw_exit(&ds->ds_rwlock);
701 dsl_dataset_drop_ref(ds, tag);
705 dsl_dataset_disown(dsl_dataset_t *ds, void *owner)
707 ASSERT((ds->ds_owner == owner && ds->ds_dbuf) ||
708 (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL));
710 mutex_enter(&ds->ds_lock);
712 if (RW_WRITE_HELD(&ds->ds_rwlock)) {
713 rw_exit(&ds->ds_rwlock);
714 cv_broadcast(&ds->ds_exclusive_cv);
716 mutex_exit(&ds->ds_lock);
718 dsl_dataset_drop_ref(ds, owner);
720 dsl_dataset_evict(ds->ds_dbuf, ds);
724 dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *owner)
726 boolean_t gotit = FALSE;
728 mutex_enter(&ds->ds_lock);
729 if (ds->ds_owner == NULL &&
730 (!DS_IS_INCONSISTENT(ds) || inconsistentok)) {
731 ds->ds_owner = owner;
732 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool))
733 rw_exit(&ds->ds_rwlock);
736 mutex_exit(&ds->ds_lock);
741 dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner)
743 ASSERT3P(owner, ==, ds->ds_owner);
744 if (!RW_WRITE_HELD(&ds->ds_rwlock))
745 rw_enter(&ds->ds_rwlock, RW_WRITER);
749 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
750 uint64_t flags, dmu_tx_t *tx)
752 dsl_pool_t *dp = dd->dd_pool;
754 dsl_dataset_phys_t *dsphys;
756 objset_t *mos = dp->dp_meta_objset;
759 origin = dp->dp_origin_snap;
761 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
762 ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
763 ASSERT(dmu_tx_is_syncing(tx));
764 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
766 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
767 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
768 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
769 dmu_buf_will_dirty(dbuf, tx);
770 dsphys = dbuf->db_data;
771 bzero(dsphys, sizeof (dsl_dataset_phys_t));
772 dsphys->ds_dir_obj = dd->dd_object;
773 dsphys->ds_flags = flags;
774 dsphys->ds_fsid_guid = unique_create();
775 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
776 sizeof (dsphys->ds_guid));
777 dsphys->ds_snapnames_zapobj =
778 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
780 dsphys->ds_creation_time = gethrestime_sec();
781 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
782 dsphys->ds_deadlist_obj =
783 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
786 dsphys->ds_prev_snap_obj = origin->ds_object;
787 dsphys->ds_prev_snap_txg =
788 origin->ds_phys->ds_creation_txg;
789 dsphys->ds_used_bytes =
790 origin->ds_phys->ds_used_bytes;
791 dsphys->ds_compressed_bytes =
792 origin->ds_phys->ds_compressed_bytes;
793 dsphys->ds_uncompressed_bytes =
794 origin->ds_phys->ds_uncompressed_bytes;
795 dsphys->ds_bp = origin->ds_phys->ds_bp;
796 dsphys->ds_flags |= origin->ds_phys->ds_flags;
798 dmu_buf_will_dirty(origin->ds_dbuf, tx);
799 origin->ds_phys->ds_num_children++;
801 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
802 if (origin->ds_phys->ds_next_clones_obj == 0) {
803 origin->ds_phys->ds_next_clones_obj =
805 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
807 VERIFY(0 == zap_add_int(mos,
808 origin->ds_phys->ds_next_clones_obj,
812 dmu_buf_will_dirty(dd->dd_dbuf, tx);
813 dd->dd_phys->dd_origin_obj = origin->ds_object;
816 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
817 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
819 dmu_buf_rele(dbuf, FTAG);
821 dmu_buf_will_dirty(dd->dd_dbuf, tx);
822 dd->dd_phys->dd_head_dataset_obj = dsobj;
828 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
829 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
831 dsl_pool_t *dp = pdd->dd_pool;
832 uint64_t dsobj, ddobj;
835 ASSERT(lastname[0] != '@');
837 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
838 VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd));
840 dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx);
842 dsl_deleg_set_create_perms(dd, tx, cr);
844 dsl_dir_close(dd, FTAG);
850 dsl_sync_task_group_t *dstg;
856 dsl_snapshot_destroy_one(char *name, void *arg)
858 struct destroyarg *da = arg;
863 (void) strcat(name, "@");
864 (void) strcat(name, da->snapname);
865 err = dsl_dataset_own(name, DS_MODE_READONLY | DS_MODE_INCONSISTENT,
867 cp = strchr(name, '@');
870 dsl_dataset_make_exclusive(ds, da->dstg);
871 if (ds->ds_user_ptr) {
872 ds->ds_user_evict_func(ds, ds->ds_user_ptr);
873 ds->ds_user_ptr = NULL;
875 dsl_sync_task_create(da->dstg, dsl_dataset_destroy_check,
876 dsl_dataset_destroy_sync, ds, da->dstg, 0);
877 } else if (err == ENOENT) {
880 (void) strcpy(da->failed, name);
886 * Destroy 'snapname' in all descendants of 'fsname'.
888 #pragma weak dmu_snapshots_destroy = dsl_snapshots_destroy
890 dsl_snapshots_destroy(char *fsname, char *snapname)
893 struct destroyarg da;
894 dsl_sync_task_t *dst;
897 err = spa_open(fsname, &spa, FTAG);
900 da.dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
901 da.snapname = snapname;
904 err = dmu_objset_find(fsname,
905 dsl_snapshot_destroy_one, &da, DS_FIND_CHILDREN);
908 err = dsl_sync_task_group_wait(da.dstg);
910 for (dst = list_head(&da.dstg->dstg_tasks); dst;
911 dst = list_next(&da.dstg->dstg_tasks, dst)) {
912 dsl_dataset_t *ds = dst->dst_arg1;
914 * Return the file system name that triggered the error
917 dsl_dataset_name(ds, fsname);
918 *strchr(fsname, '@') = '\0';
920 dsl_dataset_disown(ds, da.dstg);
923 dsl_sync_task_group_destroy(da.dstg);
924 spa_close(spa, FTAG);
929 * ds must be opened as OWNER. On return (whether successful or not),
930 * ds will be closed and caller can no longer dereference it.
933 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag)
936 dsl_sync_task_group_t *dstg;
941 if (dsl_dataset_is_snapshot(ds)) {
942 /* Destroying a snapshot is simpler */
943 dsl_dataset_make_exclusive(ds, tag);
945 if (ds->ds_user_ptr) {
946 ds->ds_user_evict_func(ds, ds->ds_user_ptr);
947 ds->ds_user_ptr = NULL;
949 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
950 dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
958 * Check for errors and mark this ds as inconsistent, in
959 * case we crash while freeing the objects.
961 err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check,
962 dsl_dataset_destroy_begin_sync, ds, NULL, 0);
966 err = dmu_objset_open_ds(ds, DMU_OST_ANY, &os);
971 * remove the objects in open context, so that we won't
972 * have too much to do in syncing context.
974 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
975 ds->ds_phys->ds_prev_snap_txg)) {
977 * Ignore errors, if there is not enough disk space
978 * we will deal with it in dsl_dataset_destroy_sync().
980 (void) dmu_free_object(os, obj);
983 dmu_objset_close(os);
987 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
988 err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd);
989 rw_exit(&dd->dd_pool->dp_config_rwlock);
994 if (ds->ds_user_ptr) {
996 * We need to sync out all in-flight IO before we try
997 * to evict (the dataset evict func is trying to clear
998 * the cached entries for this dataset in the ARC).
1000 txg_wait_synced(dd->dd_pool, 0);
1004 * Blow away the dsl_dir + head dataset.
1006 dsl_dataset_make_exclusive(ds, tag);
1007 if (ds->ds_user_ptr) {
1008 ds->ds_user_evict_func(ds, ds->ds_user_ptr);
1009 ds->ds_user_ptr = NULL;
1011 dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
1012 dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
1013 dsl_dataset_destroy_sync, ds, tag, 0);
1014 dsl_sync_task_create(dstg, dsl_dir_destroy_check,
1015 dsl_dir_destroy_sync, dd, FTAG, 0);
1016 err = dsl_sync_task_group_wait(dstg);
1017 dsl_sync_task_group_destroy(dstg);
1018 /* if it is successful, dsl_dir_destroy_sync will close the dd */
1020 dsl_dir_close(dd, FTAG);
1022 dsl_dataset_disown(ds, tag);
1027 dsl_dataset_rollback(dsl_dataset_t *ds, dmu_objset_type_t ost)
1031 ASSERT(ds->ds_owner);
1033 dsl_dataset_make_exclusive(ds, ds->ds_owner);
1034 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1035 dsl_dataset_rollback_check, dsl_dataset_rollback_sync,
1037 /* drop exclusive access */
1038 mutex_enter(&ds->ds_lock);
1039 rw_exit(&ds->ds_rwlock);
1040 cv_broadcast(&ds->ds_exclusive_cv);
1041 mutex_exit(&ds->ds_lock);
1046 dsl_dataset_set_user_ptr(dsl_dataset_t *ds,
1047 void *p, dsl_dataset_evict_func_t func)
1051 mutex_enter(&ds->ds_lock);
1052 old = ds->ds_user_ptr;
1054 ds->ds_user_ptr = p;
1055 ds->ds_user_evict_func = func;
1057 mutex_exit(&ds->ds_lock);
1062 dsl_dataset_get_user_ptr(dsl_dataset_t *ds)
1064 return (ds->ds_user_ptr);
1069 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1071 return (&ds->ds_phys->ds_bp);
1075 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1077 ASSERT(dmu_tx_is_syncing(tx));
1078 /* If it's the meta-objset, set dp_meta_rootbp */
1080 tx->tx_pool->dp_meta_rootbp = *bp;
1082 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1083 ds->ds_phys->ds_bp = *bp;
1088 dsl_dataset_get_spa(dsl_dataset_t *ds)
1090 return (ds->ds_dir->dd_pool->dp_spa);
1094 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1098 if (ds == NULL) /* this is the meta-objset */
1101 ASSERT(ds->ds_user_ptr != NULL);
1103 if (ds->ds_phys->ds_next_snap_obj != 0)
1104 panic("dirtying snapshot!");
1106 dp = ds->ds_dir->dd_pool;
1108 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) {
1109 /* up the hold count until we can be written out */
1110 dmu_buf_add_ref(ds->ds_dbuf, ds);
1115 * The unique space in the head dataset can be calculated by subtracting
1116 * the space used in the most recent snapshot, that is still being used
1117 * in this file system, from the space currently in use. To figure out
1118 * the space in the most recent snapshot still in use, we need to take
1119 * the total space used in the snapshot and subtract out the space that
1120 * has been freed up since the snapshot was taken.
1123 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1126 uint64_t dlused, dlcomp, dluncomp;
1128 ASSERT(ds->ds_object == ds->ds_dir->dd_phys->dd_head_dataset_obj);
1130 if (ds->ds_phys->ds_prev_snap_obj != 0)
1131 mrs_used = ds->ds_prev->ds_phys->ds_used_bytes;
1135 VERIFY(0 == bplist_space(&ds->ds_deadlist, &dlused, &dlcomp,
1138 ASSERT3U(dlused, <=, mrs_used);
1139 ds->ds_phys->ds_unique_bytes =
1140 ds->ds_phys->ds_used_bytes - (mrs_used - dlused);
1142 if (!DS_UNIQUE_IS_ACCURATE(ds) &&
1143 spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1144 SPA_VERSION_UNIQUE_ACCURATE)
1145 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1149 dsl_dataset_unique(dsl_dataset_t *ds)
1151 if (!DS_UNIQUE_IS_ACCURATE(ds) && !dsl_dataset_is_snapshot(ds))
1152 dsl_dataset_recalc_head_uniq(ds);
1154 return (ds->ds_phys->ds_unique_bytes);
1165 kill_blkptr(traverse_blk_cache_t *bc, spa_t *spa, void *arg)
1167 struct killarg *ka = arg;
1168 blkptr_t *bp = &bc->bc_blkptr;
1170 ASSERT3U(bc->bc_errno, ==, 0);
1172 ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg);
1173 (void) dsl_dataset_block_kill(ka->ds, bp, ka->zio, ka->tx);
1180 dsl_dataset_rollback_check(void *arg1, void *arg2, dmu_tx_t *tx)
1182 dsl_dataset_t *ds = arg1;
1183 dmu_objset_type_t *ost = arg2;
1186 * We can only roll back to emptyness if it is a ZPL objset.
1188 if (*ost != DMU_OST_ZFS && ds->ds_phys->ds_prev_snap_txg == 0)
1192 * This must not be a snapshot.
1194 if (ds->ds_phys->ds_next_snap_obj != 0)
1198 * If we made changes this txg, traverse_dsl_dataset won't find
1201 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1209 dsl_dataset_rollback_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1211 dsl_dataset_t *ds = arg1;
1212 dmu_objset_type_t *ost = arg2;
1213 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1215 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1218 * Before the roll back destroy the zil.
1220 if (ds->ds_user_ptr != NULL) {
1221 zil_rollback_destroy(
1222 ((objset_impl_t *)ds->ds_user_ptr)->os_zil, tx);
1225 * We need to make sure that the objset_impl_t is reopened after
1226 * we do the rollback, otherwise it will have the wrong
1227 * objset_phys_t. Normally this would happen when this
1228 * dataset-open is closed, thus causing the
1229 * dataset to be immediately evicted. But when doing "zfs recv
1230 * -F", we reopen the objset before that, so that there is no
1231 * window where the dataset is closed and inconsistent.
1233 ds->ds_user_evict_func(ds, ds->ds_user_ptr);
1234 ds->ds_user_ptr = NULL;
1237 /* Transfer space that was freed since last snap back to the head. */
1241 VERIFY(0 == bplist_space_birthrange(&ds->ds_deadlist,
1242 ds->ds_origin_txg, UINT64_MAX, &used));
1243 dsl_dir_transfer_space(ds->ds_dir, used,
1244 DD_USED_SNAP, DD_USED_HEAD, tx);
1247 /* Zero out the deadlist. */
1248 bplist_close(&ds->ds_deadlist);
1249 bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx);
1250 ds->ds_phys->ds_deadlist_obj =
1251 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
1252 VERIFY(0 == bplist_open(&ds->ds_deadlist, mos,
1253 ds->ds_phys->ds_deadlist_obj));
1256 /* Free blkptrs that we gave birth to */
1260 zio = zio_root(tx->tx_pool->dp_spa, NULL, NULL,
1261 ZIO_FLAG_MUSTSUCCEED);
1265 (void) traverse_dsl_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1266 ADVANCE_POST, kill_blkptr, &ka);
1267 (void) zio_wait(zio);
1270 ASSERT(!(ds->ds_phys->ds_flags & DS_FLAG_UNIQUE_ACCURATE) ||
1271 ds->ds_phys->ds_unique_bytes == 0);
1273 if (ds->ds_prev && ds->ds_prev != ds->ds_dir->dd_pool->dp_origin_snap) {
1274 /* Change our contents to that of the prev snapshot */
1276 ASSERT3U(ds->ds_prev->ds_object, ==,
1277 ds->ds_phys->ds_prev_snap_obj);
1278 ASSERT3U(ds->ds_phys->ds_used_bytes, <=,
1279 ds->ds_prev->ds_phys->ds_used_bytes);
1281 ds->ds_phys->ds_bp = ds->ds_prev->ds_phys->ds_bp;
1282 ds->ds_phys->ds_used_bytes =
1283 ds->ds_prev->ds_phys->ds_used_bytes;
1284 ds->ds_phys->ds_compressed_bytes =
1285 ds->ds_prev->ds_phys->ds_compressed_bytes;
1286 ds->ds_phys->ds_uncompressed_bytes =
1287 ds->ds_prev->ds_phys->ds_uncompressed_bytes;
1288 ds->ds_phys->ds_flags = ds->ds_prev->ds_phys->ds_flags;
1290 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1291 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1292 ds->ds_prev->ds_phys->ds_unique_bytes = 0;
1297 ASSERT3U(ds->ds_phys->ds_used_bytes, ==, 0);
1298 ASSERT3U(ds->ds_phys->ds_compressed_bytes, ==, 0);
1299 ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, ==, 0);
1301 bzero(&ds->ds_phys->ds_bp, sizeof (blkptr_t));
1302 ds->ds_phys->ds_flags = 0;
1303 ds->ds_phys->ds_unique_bytes = 0;
1304 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1305 SPA_VERSION_UNIQUE_ACCURATE)
1306 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1308 osi = dmu_objset_create_impl(ds->ds_dir->dd_pool->dp_spa, ds,
1309 &ds->ds_phys->ds_bp, *ost, tx);
1311 zfs_create_fs(&osi->os, kcred, NULL, tx);
1315 spa_history_internal_log(LOG_DS_ROLLBACK, ds->ds_dir->dd_pool->dp_spa,
1316 tx, cr, "dataset = %llu", ds->ds_object);
1321 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1323 dsl_dataset_t *ds = arg1;
1324 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1329 * Can't delete a head dataset if there are snapshots of it.
1330 * (Except if the only snapshots are from the branch we cloned
1333 if (ds->ds_prev != NULL &&
1334 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1338 * This is really a dsl_dir thing, but check it here so that
1339 * we'll be less likely to leave this dataset inconsistent &
1342 err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count);
1353 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1355 dsl_dataset_t *ds = arg1;
1356 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1358 /* Mark it as inconsistent on-disk, in case we crash */
1359 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1360 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1362 spa_history_internal_log(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx,
1363 cr, "dataset = %llu", ds->ds_object);
1368 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1370 dsl_dataset_t *ds = arg1;
1372 /* we have an owner hold, so noone else can destroy us */
1373 ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
1375 /* Can't delete a branch point. */
1376 if (ds->ds_phys->ds_num_children > 1)
1380 * Can't delete a head dataset if there are snapshots of it.
1381 * (Except if the only snapshots are from the branch we cloned
1384 if (ds->ds_prev != NULL &&
1385 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1389 * If we made changes this txg, traverse_dsl_dataset won't find
1392 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1395 /* XXX we should do some i/o error checking... */
1407 dsl_dataset_refs_gone(dmu_buf_t *db, void *argv)
1409 struct refsarg *arg = argv;
1411 mutex_enter(&arg->lock);
1413 cv_signal(&arg->cv);
1414 mutex_exit(&arg->lock);
1418 dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag)
1422 mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL);
1423 cv_init(&arg.cv, NULL, CV_DEFAULT, NULL);
1425 (void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys,
1426 dsl_dataset_refs_gone);
1427 dmu_buf_rele(ds->ds_dbuf, tag);
1428 mutex_enter(&arg.lock);
1430 cv_wait(&arg.cv, &arg.lock);
1432 mutex_exit(&arg.lock);
1435 mutex_destroy(&arg.lock);
1436 cv_destroy(&arg.cv);
1440 dsl_dataset_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx)
1442 dsl_dataset_t *ds = arg1;
1445 int after_branch_point = FALSE;
1446 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1447 objset_t *mos = dp->dp_meta_objset;
1448 dsl_dataset_t *ds_prev = NULL;
1451 ASSERT(ds->ds_owner);
1452 ASSERT3U(ds->ds_phys->ds_num_children, <=, 1);
1453 ASSERT(ds->ds_prev == NULL ||
1454 ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1455 ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1457 /* signal any waiters that this dataset is going away */
1458 mutex_enter(&ds->ds_lock);
1459 ds->ds_owner = dsl_reaper;
1460 cv_broadcast(&ds->ds_exclusive_cv);
1461 mutex_exit(&ds->ds_lock);
1463 /* Remove our reservation */
1464 if (ds->ds_reserved != 0) {
1466 dsl_dataset_set_reservation_sync(ds, &val, cr, tx);
1467 ASSERT3U(ds->ds_reserved, ==, 0);
1470 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1472 dsl_pool_ds_destroyed(ds, tx);
1474 obj = ds->ds_object;
1476 if (ds->ds_phys->ds_prev_snap_obj != 0) {
1478 ds_prev = ds->ds_prev;
1480 VERIFY(0 == dsl_dataset_hold_obj(dp,
1481 ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev));
1483 after_branch_point =
1484 (ds_prev->ds_phys->ds_next_snap_obj != obj);
1486 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
1487 if (after_branch_point &&
1488 ds_prev->ds_phys->ds_next_clones_obj != 0) {
1489 VERIFY(0 == zap_remove_int(mos,
1490 ds_prev->ds_phys->ds_next_clones_obj, obj, tx));
1491 if (ds->ds_phys->ds_next_snap_obj != 0) {
1492 VERIFY(0 == zap_add_int(mos,
1493 ds_prev->ds_phys->ds_next_clones_obj,
1494 ds->ds_phys->ds_next_snap_obj, tx));
1497 if (after_branch_point &&
1498 ds->ds_phys->ds_next_snap_obj == 0) {
1499 /* This clone is toast. */
1500 ASSERT(ds_prev->ds_phys->ds_num_children > 1);
1501 ds_prev->ds_phys->ds_num_children--;
1502 } else if (!after_branch_point) {
1503 ds_prev->ds_phys->ds_next_snap_obj =
1504 ds->ds_phys->ds_next_snap_obj;
1508 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1510 if (ds->ds_phys->ds_next_snap_obj != 0) {
1512 dsl_dataset_t *ds_next;
1514 uint64_t old_unique;
1515 int64_t used = 0, compressed = 0, uncompressed = 0;
1517 VERIFY(0 == dsl_dataset_hold_obj(dp,
1518 ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next));
1519 ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
1521 old_unique = dsl_dataset_unique(ds_next);
1523 dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
1524 ds_next->ds_phys->ds_prev_snap_obj =
1525 ds->ds_phys->ds_prev_snap_obj;
1526 ds_next->ds_phys->ds_prev_snap_txg =
1527 ds->ds_phys->ds_prev_snap_txg;
1528 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1529 ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
1532 * Transfer to our deadlist (which will become next's
1533 * new deadlist) any entries from next's current
1534 * deadlist which were born before prev, and free the
1537 * XXX we're doing this long task with the config lock held
1539 while (bplist_iterate(&ds_next->ds_deadlist, &itor, &bp) == 0) {
1540 if (bp.blk_birth <= ds->ds_phys->ds_prev_snap_txg) {
1541 VERIFY(0 == bplist_enqueue(&ds->ds_deadlist,
1543 if (ds_prev && !after_branch_point &&
1545 ds_prev->ds_phys->ds_prev_snap_txg) {
1546 ds_prev->ds_phys->ds_unique_bytes +=
1547 bp_get_dasize(dp->dp_spa, &bp);
1550 used += bp_get_dasize(dp->dp_spa, &bp);
1551 compressed += BP_GET_PSIZE(&bp);
1552 uncompressed += BP_GET_UCSIZE(&bp);
1553 /* XXX check return value? */
1554 (void) dsl_free(zio, dp, tx->tx_txg,
1555 &bp, NULL, NULL, ARC_NOWAIT);
1559 ASSERT3U(used, ==, ds->ds_phys->ds_unique_bytes);
1561 /* change snapused */
1562 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1563 -used, -compressed, -uncompressed, tx);
1565 /* free next's deadlist */
1566 bplist_close(&ds_next->ds_deadlist);
1567 bplist_destroy(mos, ds_next->ds_phys->ds_deadlist_obj, tx);
1569 /* set next's deadlist to our deadlist */
1570 bplist_close(&ds->ds_deadlist);
1571 ds_next->ds_phys->ds_deadlist_obj =
1572 ds->ds_phys->ds_deadlist_obj;
1573 VERIFY(0 == bplist_open(&ds_next->ds_deadlist, mos,
1574 ds_next->ds_phys->ds_deadlist_obj));
1575 ds->ds_phys->ds_deadlist_obj = 0;
1577 if (ds_next->ds_phys->ds_next_snap_obj != 0) {
1579 * Update next's unique to include blocks which
1580 * were previously shared by only this snapshot
1581 * and it. Those blocks will be born after the
1582 * prev snap and before this snap, and will have
1583 * died after the next snap and before the one
1584 * after that (ie. be on the snap after next's
1587 * XXX we're doing this long task with the
1590 dsl_dataset_t *ds_after_next;
1593 VERIFY(0 == dsl_dataset_hold_obj(dp,
1594 ds_next->ds_phys->ds_next_snap_obj,
1595 FTAG, &ds_after_next));
1598 bplist_space_birthrange(&ds_after_next->ds_deadlist,
1599 ds->ds_phys->ds_prev_snap_txg,
1600 ds->ds_phys->ds_creation_txg, &space));
1601 ds_next->ds_phys->ds_unique_bytes += space;
1603 dsl_dataset_rele(ds_after_next, FTAG);
1604 ASSERT3P(ds_next->ds_prev, ==, NULL);
1606 ASSERT3P(ds_next->ds_prev, ==, ds);
1607 dsl_dataset_drop_ref(ds_next->ds_prev, ds_next);
1608 ds_next->ds_prev = NULL;
1610 VERIFY(0 == dsl_dataset_get_ref(dp,
1611 ds->ds_phys->ds_prev_snap_obj,
1612 ds_next, &ds_next->ds_prev));
1615 dsl_dataset_recalc_head_uniq(ds_next);
1618 * Reduce the amount of our unconsmed refreservation
1619 * being charged to our parent by the amount of
1620 * new unique data we have gained.
1622 if (old_unique < ds_next->ds_reserved) {
1624 uint64_t new_unique =
1625 ds_next->ds_phys->ds_unique_bytes;
1627 ASSERT(old_unique <= new_unique);
1628 mrsdelta = MIN(new_unique - old_unique,
1629 ds_next->ds_reserved - old_unique);
1630 dsl_dir_diduse_space(ds->ds_dir,
1631 DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
1634 dsl_dataset_rele(ds_next, FTAG);
1637 * There's no next snapshot, so this is a head dataset.
1638 * Destroy the deadlist. Unless it's a clone, the
1639 * deadlist should be empty. (If it's a clone, it's
1640 * safe to ignore the deadlist contents.)
1644 ASSERT(after_branch_point || bplist_empty(&ds->ds_deadlist));
1645 bplist_close(&ds->ds_deadlist);
1646 bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx);
1647 ds->ds_phys->ds_deadlist_obj = 0;
1650 * Free everything that we point to (that's born after
1651 * the previous snapshot, if we are a clone)
1653 * NB: this should be very quick, because we already
1654 * freed all the objects in open context.
1659 err = traverse_dsl_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1660 ADVANCE_POST, kill_blkptr, &ka);
1661 ASSERT3U(err, ==, 0);
1662 ASSERT(spa_version(dp->dp_spa) < SPA_VERSION_UNIQUE_ACCURATE ||
1663 ds->ds_phys->ds_unique_bytes == 0);
1666 err = zio_wait(zio);
1667 ASSERT3U(err, ==, 0);
1669 if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) {
1670 /* Erase the link in the dir */
1671 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1672 ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
1673 ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0);
1674 err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx);
1677 /* remove from snapshot namespace */
1678 dsl_dataset_t *ds_head;
1679 ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0);
1680 VERIFY(0 == dsl_dataset_hold_obj(dp,
1681 ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head));
1682 VERIFY(0 == dsl_dataset_get_snapname(ds));
1687 err = dsl_dataset_snap_lookup(ds_head,
1688 ds->ds_snapname, &val);
1689 ASSERT3U(err, ==, 0);
1690 ASSERT3U(val, ==, obj);
1693 err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx);
1695 dsl_dataset_rele(ds_head, FTAG);
1698 if (ds_prev && ds->ds_prev != ds_prev)
1699 dsl_dataset_rele(ds_prev, FTAG);
1701 spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
1702 spa_history_internal_log(LOG_DS_DESTROY, dp->dp_spa, tx,
1703 cr, "dataset = %llu", ds->ds_object);
1705 if (ds->ds_phys->ds_next_clones_obj != 0) {
1707 ASSERT(0 == zap_count(mos,
1708 ds->ds_phys->ds_next_clones_obj, &count) && count == 0);
1709 VERIFY(0 == dmu_object_free(mos,
1710 ds->ds_phys->ds_next_clones_obj, tx));
1712 if (ds->ds_phys->ds_props_obj != 0)
1713 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx));
1714 dsl_dir_close(ds->ds_dir, ds);
1716 dsl_dataset_drain_refs(ds, tag);
1717 VERIFY(0 == dmu_object_free(mos, obj, tx));
1721 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1725 if (!dmu_tx_is_syncing(tx))
1729 * If there's an fs-only reservation, any blocks that might become
1730 * owned by the snapshot dataset must be accommodated by space
1731 * outside of the reservation.
1733 asize = MIN(dsl_dataset_unique(ds), ds->ds_reserved);
1734 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, FALSE))
1738 * Propogate any reserved space for this snapshot to other
1739 * snapshot checks in this sync group.
1742 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1749 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
1751 dsl_dataset_t *ds = arg1;
1752 const char *snapname = arg2;
1757 * We don't allow multiple snapshots of the same txg. If there
1758 * is already one, try again.
1760 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
1764 * Check for conflicting name snapshot name.
1766 err = dsl_dataset_snap_lookup(ds, snapname, &value);
1773 * Check that the dataset's name is not too long. Name consists
1774 * of the dataset's length + 1 for the @-sign + snapshot name's length
1776 if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
1777 return (ENAMETOOLONG);
1779 err = dsl_dataset_snapshot_reserve_space(ds, tx);
1783 ds->ds_trysnap_txg = tx->tx_txg;
1788 dsl_dataset_snapshot_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1790 dsl_dataset_t *ds = arg1;
1791 const char *snapname = arg2;
1792 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1794 dsl_dataset_phys_t *dsphys;
1795 uint64_t dsobj, crtxg;
1796 objset_t *mos = dp->dp_meta_objset;
1799 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1802 * The origin's ds_creation_txg has to be < TXG_INITIAL
1804 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1809 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1810 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1811 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1812 dmu_buf_will_dirty(dbuf, tx);
1813 dsphys = dbuf->db_data;
1814 bzero(dsphys, sizeof (dsl_dataset_phys_t));
1815 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1816 dsphys->ds_fsid_guid = unique_create();
1817 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1818 sizeof (dsphys->ds_guid));
1819 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
1820 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
1821 dsphys->ds_next_snap_obj = ds->ds_object;
1822 dsphys->ds_num_children = 1;
1823 dsphys->ds_creation_time = gethrestime_sec();
1824 dsphys->ds_creation_txg = crtxg;
1825 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
1826 dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes;
1827 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
1828 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
1829 dsphys->ds_flags = ds->ds_phys->ds_flags;
1830 dsphys->ds_bp = ds->ds_phys->ds_bp;
1831 dmu_buf_rele(dbuf, FTAG);
1833 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
1835 uint64_t next_clones_obj =
1836 ds->ds_prev->ds_phys->ds_next_clones_obj;
1837 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
1839 ds->ds_prev->ds_phys->ds_num_children > 1);
1840 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1841 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1842 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1843 ds->ds_prev->ds_phys->ds_creation_txg);
1844 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
1845 } else if (next_clones_obj != 0) {
1846 VERIFY3U(0, ==, zap_remove_int(mos,
1847 next_clones_obj, dsphys->ds_next_snap_obj, tx));
1848 VERIFY3U(0, ==, zap_add_int(mos,
1849 next_clones_obj, dsobj, tx));
1854 * If we have a reference-reservation on this dataset, we will
1855 * need to increase the amount of refreservation being charged
1856 * since our unique space is going to zero.
1858 if (ds->ds_reserved) {
1859 int64_t add = MIN(dsl_dataset_unique(ds), ds->ds_reserved);
1860 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1864 bplist_close(&ds->ds_deadlist);
1865 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1866 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
1867 ds->ds_phys->ds_prev_snap_obj = dsobj;
1868 ds->ds_phys->ds_prev_snap_txg = crtxg;
1869 ds->ds_phys->ds_unique_bytes = 0;
1870 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1871 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1872 ds->ds_phys->ds_deadlist_obj =
1873 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
1874 VERIFY(0 == bplist_open(&ds->ds_deadlist, mos,
1875 ds->ds_phys->ds_deadlist_obj));
1877 dprintf("snap '%s' -> obj %llu\n", snapname, dsobj);
1878 err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
1879 snapname, 8, 1, &dsobj, tx);
1883 dsl_dataset_drop_ref(ds->ds_prev, ds);
1884 VERIFY(0 == dsl_dataset_get_ref(dp,
1885 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
1887 dsl_pool_ds_snapshotted(ds, tx);
1889 spa_history_internal_log(LOG_DS_SNAPSHOT, dp->dp_spa, tx, cr,
1890 "dataset = %llu", dsobj);
1894 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1896 ASSERT(dmu_tx_is_syncing(tx));
1897 ASSERT(ds->ds_user_ptr != NULL);
1898 ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
1901 * in case we had to change ds_fsid_guid when we opened it,
1904 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1905 ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
1907 dsl_dir_dirty(ds->ds_dir, tx);
1908 dmu_objset_sync(ds->ds_user_ptr, zio, tx);
1912 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1914 uint64_t refd, avail, uobjs, aobjs;
1916 dsl_dir_stats(ds->ds_dir, nv);
1918 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1919 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1920 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1922 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1923 ds->ds_phys->ds_creation_time);
1924 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1925 ds->ds_phys->ds_creation_txg);
1926 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1928 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1930 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1931 ds->ds_phys->ds_guid);
1933 if (ds->ds_phys->ds_next_snap_obj) {
1935 * This is a snapshot; override the dd's space used with
1936 * our unique space and compression ratio.
1938 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1939 ds->ds_phys->ds_unique_bytes);
1940 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
1941 ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
1942 (ds->ds_phys->ds_uncompressed_bytes * 100 /
1943 ds->ds_phys->ds_compressed_bytes));
1948 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1950 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
1951 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
1952 stat->dds_guid = ds->ds_phys->ds_guid;
1953 if (ds->ds_phys->ds_next_snap_obj) {
1954 stat->dds_is_snapshot = B_TRUE;
1955 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
1958 /* clone origin is really a dsl_dir thing... */
1959 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
1960 if (dsl_dir_is_clone(ds->ds_dir)) {
1963 VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool,
1964 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
1965 dsl_dataset_name(ods, stat->dds_origin);
1966 dsl_dataset_drop_ref(ods, FTAG);
1968 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
1972 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1974 return (ds->ds_fsid_guid);
1978 dsl_dataset_space(dsl_dataset_t *ds,
1979 uint64_t *refdbytesp, uint64_t *availbytesp,
1980 uint64_t *usedobjsp, uint64_t *availobjsp)
1982 *refdbytesp = ds->ds_phys->ds_used_bytes;
1983 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1984 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
1985 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
1986 if (ds->ds_quota != 0) {
1988 * Adjust available bytes according to refquota
1990 if (*refdbytesp < ds->ds_quota)
1991 *availbytesp = MIN(*availbytesp,
1992 ds->ds_quota - *refdbytesp);
1996 *usedobjsp = ds->ds_phys->ds_bp.blk_fill;
1997 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
2001 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
2003 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2005 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
2006 dsl_pool_sync_context(dp));
2007 if (ds->ds_prev == NULL)
2009 if (ds->ds_phys->ds_bp.blk_birth >
2010 ds->ds_prev->ds_phys->ds_creation_txg)
2017 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
2019 dsl_dataset_t *ds = arg1;
2020 char *newsnapname = arg2;
2021 dsl_dir_t *dd = ds->ds_dir;
2026 err = dsl_dataset_hold_obj(dd->dd_pool,
2027 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds);
2031 /* new name better not be in use */
2032 err = dsl_dataset_snap_lookup(hds, newsnapname, &val);
2033 dsl_dataset_rele(hds, FTAG);
2037 else if (err == ENOENT)
2040 /* dataset name + 1 for the "@" + the new snapshot name must fit */
2041 if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
2048 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2,
2049 cred_t *cr, dmu_tx_t *tx)
2051 dsl_dataset_t *ds = arg1;
2052 const char *newsnapname = arg2;
2053 dsl_dir_t *dd = ds->ds_dir;
2054 objset_t *mos = dd->dd_pool->dp_meta_objset;
2058 ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
2060 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
2061 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds));
2063 VERIFY(0 == dsl_dataset_get_snapname(ds));
2064 err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx);
2065 ASSERT3U(err, ==, 0);
2066 mutex_enter(&ds->ds_lock);
2067 (void) strcpy(ds->ds_snapname, newsnapname);
2068 mutex_exit(&ds->ds_lock);
2069 err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
2070 ds->ds_snapname, 8, 1, &ds->ds_object, tx);
2071 ASSERT3U(err, ==, 0);
2073 spa_history_internal_log(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx,
2074 cr, "dataset = %llu", ds->ds_object);
2075 dsl_dataset_rele(hds, FTAG);
2078 struct renamesnaparg {
2079 dsl_sync_task_group_t *dstg;
2080 char failed[MAXPATHLEN];
2086 dsl_snapshot_rename_one(char *name, void *arg)
2088 struct renamesnaparg *ra = arg;
2089 dsl_dataset_t *ds = NULL;
2093 cp = name + strlen(name);
2095 (void) strcpy(cp + 1, ra->oldsnap);
2098 * For recursive snapshot renames the parent won't be changing
2099 * so we just pass name for both the to/from argument.
2101 err = zfs_secpolicy_rename_perms(name, name, CRED());
2102 if (err == ENOENT) {
2105 (void) strcpy(ra->failed, name);
2111 * For all filesystems undergoing rename, we'll need to unmount it.
2113 (void) zfs_unmount_snap(name, NULL);
2115 err = dsl_dataset_hold(name, ra->dstg, &ds);
2117 if (err == ENOENT) {
2120 (void) strcpy(ra->failed, name);
2124 dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
2125 dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
2131 dsl_recursive_rename(char *oldname, const char *newname)
2134 struct renamesnaparg *ra;
2135 dsl_sync_task_t *dst;
2137 char *cp, *fsname = spa_strdup(oldname);
2138 int len = strlen(oldname);
2140 /* truncate the snapshot name to get the fsname */
2141 cp = strchr(fsname, '@');
2144 err = spa_open(fsname, &spa, FTAG);
2146 kmem_free(fsname, len + 1);
2149 ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP);
2150 ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
2152 ra->oldsnap = strchr(oldname, '@') + 1;
2153 ra->newsnap = strchr(newname, '@') + 1;
2156 err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
2158 kmem_free(fsname, len + 1);
2161 err = dsl_sync_task_group_wait(ra->dstg);
2164 for (dst = list_head(&ra->dstg->dstg_tasks); dst;
2165 dst = list_next(&ra->dstg->dstg_tasks, dst)) {
2166 dsl_dataset_t *ds = dst->dst_arg1;
2168 dsl_dir_name(ds->ds_dir, ra->failed);
2169 (void) strcat(ra->failed, "@");
2170 (void) strcat(ra->failed, ra->newsnap);
2172 dsl_dataset_rele(ds, ra->dstg);
2176 (void) strcpy(oldname, ra->failed);
2178 dsl_sync_task_group_destroy(ra->dstg);
2179 kmem_free(ra, sizeof (struct renamesnaparg));
2180 spa_close(spa, FTAG);
2185 dsl_valid_rename(char *oldname, void *arg)
2187 int delta = *(int *)arg;
2189 if (strlen(oldname) + delta >= MAXNAMELEN)
2190 return (ENAMETOOLONG);
2195 #pragma weak dmu_objset_rename = dsl_dataset_rename
2197 dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive)
2204 err = dsl_dir_open(oldname, FTAG, &dd, &tail);
2208 int delta = strlen(newname) - strlen(oldname);
2210 /* if we're growing, validate child name lengths */
2212 err = dmu_objset_find(oldname, dsl_valid_rename,
2213 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2216 err = dsl_dir_rename(dd, newname);
2217 dsl_dir_close(dd, FTAG);
2220 if (tail[0] != '@') {
2221 /* the name ended in a nonexistant component */
2222 dsl_dir_close(dd, FTAG);
2226 dsl_dir_close(dd, FTAG);
2228 /* new name must be snapshot in same filesystem */
2229 tail = strchr(newname, '@');
2233 if (strncmp(oldname, newname, tail - newname) != 0)
2237 err = dsl_recursive_rename(oldname, newname);
2239 err = dsl_dataset_hold(oldname, FTAG, &ds);
2243 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2244 dsl_dataset_snapshot_rename_check,
2245 dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
2247 dsl_dataset_rele(ds, FTAG);
2253 struct promotenode {
2259 list_t shared_snaps, origin_snaps, clone_snaps;
2260 dsl_dataset_t *origin_origin, *origin_head;
2261 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2264 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2268 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
2270 dsl_dataset_t *hds = arg1;
2271 struct promotearg *pa = arg2;
2272 struct promotenode *snap = list_head(&pa->shared_snaps);
2273 dsl_dataset_t *origin_ds = snap->ds;
2276 /* Check that it is a real clone */
2277 if (!dsl_dir_is_clone(hds->ds_dir))
2280 /* Since this is so expensive, don't do the preliminary check */
2281 if (!dmu_tx_is_syncing(tx))
2284 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)
2287 /* compute origin's new unique space */
2288 snap = list_tail(&pa->clone_snaps);
2289 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2290 err = bplist_space_birthrange(&snap->ds->ds_deadlist,
2291 origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX, &pa->unique);
2296 * Walk the snapshots that we are moving
2298 * Compute space to transfer. Consider the incremental changes
2299 * to used for each snapshot:
2300 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2301 * So each snapshot gave birth to:
2302 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2303 * So a sequence would look like:
2304 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2305 * Which simplifies to:
2306 * uN + kN + kN-1 + ... + k1 + k0
2307 * Note however, if we stop before we reach the ORIGIN we get:
2308 * uN + kN + kN-1 + ... + kM - uM-1
2310 pa->used = origin_ds->ds_phys->ds_used_bytes;
2311 pa->comp = origin_ds->ds_phys->ds_compressed_bytes;
2312 pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
2313 for (snap = list_head(&pa->shared_snaps); snap;
2314 snap = list_next(&pa->shared_snaps, snap)) {
2315 uint64_t val, dlused, dlcomp, dluncomp;
2316 dsl_dataset_t *ds = snap->ds;
2318 /* Check that the snapshot name does not conflict */
2319 VERIFY(0 == dsl_dataset_get_snapname(ds));
2320 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2326 /* The very first snapshot does not have a deadlist */
2327 if (ds->ds_phys->ds_prev_snap_obj == 0)
2330 if (err = bplist_space(&ds->ds_deadlist,
2331 &dlused, &dlcomp, &dluncomp))
2335 pa->uncomp += dluncomp;
2339 * If we are a clone of a clone then we never reached ORIGIN,
2340 * so we need to subtract out the clone origin's used space.
2342 if (pa->origin_origin) {
2343 pa->used -= pa->origin_origin->ds_phys->ds_used_bytes;
2344 pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes;
2345 pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes;
2348 /* Check that there is enough space here */
2349 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2355 * Compute the amounts of space that will be used by snapshots
2356 * after the promotion (for both origin and clone). For each,
2357 * it is the amount of space that will be on all of their
2358 * deadlists (that was not born before their new origin).
2360 if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2364 * Note, typically this will not be a clone of a clone,
2365 * so snap->ds->ds_origin_txg will be < TXG_INITIAL, so
2366 * these snaplist_space() -> bplist_space_birthrange()
2367 * calls will be fast because they do not have to
2368 * iterate over all bps.
2370 snap = list_head(&pa->origin_snaps);
2371 err = snaplist_space(&pa->shared_snaps,
2372 snap->ds->ds_origin_txg, &pa->cloneusedsnap);
2376 err = snaplist_space(&pa->clone_snaps,
2377 snap->ds->ds_origin_txg, &space);
2380 pa->cloneusedsnap += space;
2382 if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2383 err = snaplist_space(&pa->origin_snaps,
2384 origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap);
2393 dsl_dataset_promote_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2395 dsl_dataset_t *hds = arg1;
2396 struct promotearg *pa = arg2;
2397 struct promotenode *snap = list_head(&pa->shared_snaps);
2398 dsl_dataset_t *origin_ds = snap->ds;
2399 dsl_dataset_t *origin_head;
2400 dsl_dir_t *dd = hds->ds_dir;
2401 dsl_pool_t *dp = hds->ds_dir->dd_pool;
2402 dsl_dir_t *odd = NULL;
2403 uint64_t oldnext_obj;
2406 ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
2408 snap = list_head(&pa->origin_snaps);
2409 origin_head = snap->ds;
2412 * We need to explicitly open odd, since origin_ds's dd will be
2415 VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object,
2418 /* change origin's next snap */
2419 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2420 oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2421 snap = list_tail(&pa->clone_snaps);
2422 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2423 origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
2425 /* change the origin's next clone */
2426 if (origin_ds->ds_phys->ds_next_clones_obj) {
2427 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2428 origin_ds->ds_phys->ds_next_clones_obj,
2429 origin_ds->ds_phys->ds_next_snap_obj, tx));
2430 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2431 origin_ds->ds_phys->ds_next_clones_obj,
2436 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2437 ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2438 dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2439 hds->ds_origin_txg = origin_head->ds_origin_txg;
2440 dmu_buf_will_dirty(odd->dd_dbuf, tx);
2441 odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2442 origin_head->ds_origin_txg = origin_ds->ds_phys->ds_creation_txg;
2444 /* move snapshots to this dir */
2445 for (snap = list_head(&pa->shared_snaps); snap;
2446 snap = list_next(&pa->shared_snaps, snap)) {
2447 dsl_dataset_t *ds = snap->ds;
2449 /* unregister props as dsl_dir is changing */
2450 if (ds->ds_user_ptr) {
2451 ds->ds_user_evict_func(ds, ds->ds_user_ptr);
2452 ds->ds_user_ptr = NULL;
2454 /* move snap name entry */
2455 VERIFY(0 == dsl_dataset_get_snapname(ds));
2456 VERIFY(0 == dsl_dataset_snap_remove(origin_head,
2457 ds->ds_snapname, tx));
2458 VERIFY(0 == zap_add(dp->dp_meta_objset,
2459 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2460 8, 1, &ds->ds_object, tx));
2461 /* change containing dsl_dir */
2462 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2463 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2464 ds->ds_phys->ds_dir_obj = dd->dd_object;
2465 ASSERT3P(ds->ds_dir, ==, odd);
2466 dsl_dir_close(ds->ds_dir, ds);
2467 VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
2468 NULL, ds, &ds->ds_dir));
2470 ASSERT3U(dsl_prop_numcb(ds), ==, 0);
2474 * Change space accounting.
2475 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2476 * both be valid, or both be 0 (resulting in delta == 0). This
2477 * is true for each of {clone,origin} independently.
2480 delta = pa->cloneusedsnap -
2481 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2482 ASSERT3S(delta, >=, 0);
2483 ASSERT3U(pa->used, >=, delta);
2484 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2485 dsl_dir_diduse_space(dd, DD_USED_HEAD,
2486 pa->used - delta, pa->comp, pa->uncomp, tx);
2488 delta = pa->originusedsnap -
2489 odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2490 ASSERT3S(delta, <=, 0);
2491 ASSERT3U(pa->used, >=, -delta);
2492 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2493 dsl_dir_diduse_space(odd, DD_USED_HEAD,
2494 -pa->used - delta, -pa->comp, -pa->uncomp, tx);
2496 origin_ds->ds_phys->ds_unique_bytes = pa->unique;
2498 /* log history record */
2499 spa_history_internal_log(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx,
2500 cr, "dataset = %llu", hds->ds_object);
2502 dsl_dir_close(odd, FTAG);
2505 static char *snaplist_tag = "snaplist";
2507 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2508 * (exclusive) and last_obj (inclusive). The list will be in reverse
2509 * order (last_obj will be the list_head()). If first_obj == 0, do all
2510 * snapshots back to this dataset's origin.
2513 snaplist_make(dsl_pool_t *dp, boolean_t own,
2514 uint64_t first_obj, uint64_t last_obj, list_t *l)
2516 uint64_t obj = last_obj;
2518 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock));
2520 list_create(l, sizeof (struct promotenode),
2521 offsetof(struct promotenode, link));
2523 while (obj != first_obj) {
2525 struct promotenode *snap;
2529 err = dsl_dataset_own_obj(dp, obj,
2530 0, snaplist_tag, &ds);
2532 dsl_dataset_make_exclusive(ds, snaplist_tag);
2534 err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds);
2536 if (err == ENOENT) {
2537 /* lost race with snapshot destroy */
2538 struct promotenode *last = list_tail(l);
2539 ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj);
2540 obj = last->ds->ds_phys->ds_prev_snap_obj;
2547 first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2549 snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP);
2551 list_insert_tail(l, snap);
2552 obj = ds->ds_phys->ds_prev_snap_obj;
2559 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2561 struct promotenode *snap;
2564 for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2566 int err = bplist_space_birthrange(&snap->ds->ds_deadlist,
2567 mintxg, UINT64_MAX, &used);
2576 snaplist_destroy(list_t *l, boolean_t own)
2578 struct promotenode *snap;
2580 if (!list_link_active(&l->list_head))
2583 while ((snap = list_tail(l)) != NULL) {
2584 list_remove(l, snap);
2586 dsl_dataset_disown(snap->ds, snaplist_tag);
2588 dsl_dataset_rele(snap->ds, snaplist_tag);
2589 kmem_free(snap, sizeof (struct promotenode));
2595 * Promote a clone. Nomenclature note:
2596 * "clone" or "cds": the original clone which is being promoted
2597 * "origin" or "ods": the snapshot which is originally clone's origin
2598 * "origin head" or "ohds": the dataset which is the head
2599 * (filesystem/volume) for the origin
2600 * "origin origin": the origin of the origin's filesystem (typically
2601 * NULL, indicating that the clone is not a clone of a clone).
2604 dsl_dataset_promote(const char *name)
2609 dmu_object_info_t doi;
2610 struct promotearg pa = { 0 };
2611 struct promotenode *snap;
2614 err = dsl_dataset_hold(name, FTAG, &ds);
2620 err = dmu_object_info(dp->dp_meta_objset,
2621 ds->ds_phys->ds_snapnames_zapobj, &doi);
2623 dsl_dataset_rele(ds, FTAG);
2627 if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) {
2628 dsl_dataset_rele(ds, FTAG);
2633 * We are going to inherit all the snapshots taken before our
2634 * origin (i.e., our new origin will be our parent's origin).
2635 * Take ownership of them so that we can rename them into our
2638 rw_enter(&dp->dp_config_rwlock, RW_READER);
2640 err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj,
2645 err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps);
2649 snap = list_head(&pa.shared_snaps);
2650 ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
2651 err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj,
2652 snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps);
2656 if (dsl_dir_is_clone(snap->ds->ds_dir)) {
2657 err = dsl_dataset_own_obj(dp,
2658 snap->ds->ds_dir->dd_phys->dd_origin_obj,
2659 0, FTAG, &pa.origin_origin);
2665 rw_exit(&dp->dp_config_rwlock);
2668 * Add in 128x the snapnames zapobj size, since we will be moving
2669 * a bunch of snapnames to the promoted ds, and dirtying their
2673 err = dsl_sync_task_do(dp, dsl_dataset_promote_check,
2674 dsl_dataset_promote_sync, ds, &pa,
2675 2 + 2 * doi.doi_physical_blks);
2678 snaplist_destroy(&pa.shared_snaps, B_TRUE);
2679 snaplist_destroy(&pa.clone_snaps, B_FALSE);
2680 snaplist_destroy(&pa.origin_snaps, B_FALSE);
2681 if (pa.origin_origin)
2682 dsl_dataset_disown(pa.origin_origin, FTAG);
2683 dsl_dataset_rele(ds, FTAG);
2687 struct cloneswaparg {
2688 dsl_dataset_t *cds; /* clone dataset */
2689 dsl_dataset_t *ohds; /* origin's head dataset */
2691 int64_t unused_refres_delta; /* change in unconsumed refreservation */
2696 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx)
2698 struct cloneswaparg *csa = arg1;
2700 /* they should both be heads */
2701 if (dsl_dataset_is_snapshot(csa->cds) ||
2702 dsl_dataset_is_snapshot(csa->ohds))
2705 /* the branch point should be just before them */
2706 if (csa->cds->ds_prev != csa->ohds->ds_prev)
2709 /* cds should be the clone */
2710 if (csa->cds->ds_prev->ds_phys->ds_next_snap_obj !=
2711 csa->ohds->ds_object)
2714 /* the clone should be a child of the origin */
2715 if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir)
2718 /* ohds shouldn't be modified unless 'force' */
2719 if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds))
2722 /* adjust amount of any unconsumed refreservation */
2723 csa->unused_refres_delta =
2724 (int64_t)MIN(csa->ohds->ds_reserved,
2725 csa->ohds->ds_phys->ds_unique_bytes) -
2726 (int64_t)MIN(csa->ohds->ds_reserved,
2727 csa->cds->ds_phys->ds_unique_bytes);
2729 if (csa->unused_refres_delta > 0 &&
2730 csa->unused_refres_delta >
2731 dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE))
2739 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2741 struct cloneswaparg *csa = arg1;
2742 dsl_pool_t *dp = csa->cds->ds_dir->dd_pool;
2744 ASSERT(csa->cds->ds_reserved == 0);
2745 ASSERT(csa->cds->ds_quota == csa->ohds->ds_quota);
2747 dmu_buf_will_dirty(csa->cds->ds_dbuf, tx);
2748 dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx);
2749 dmu_buf_will_dirty(csa->cds->ds_prev->ds_dbuf, tx);
2751 if (csa->cds->ds_user_ptr != NULL) {
2752 csa->cds->ds_user_evict_func(csa->cds, csa->cds->ds_user_ptr);
2753 csa->cds->ds_user_ptr = NULL;
2756 if (csa->ohds->ds_user_ptr != NULL) {
2757 csa->ohds->ds_user_evict_func(csa->ohds,
2758 csa->ohds->ds_user_ptr);
2759 csa->ohds->ds_user_ptr = NULL;
2762 /* reset origin's unique bytes */
2763 VERIFY(0 == bplist_space_birthrange(&csa->cds->ds_deadlist,
2764 csa->cds->ds_prev->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2765 &csa->cds->ds_prev->ds_phys->ds_unique_bytes));
2770 tmp = csa->ohds->ds_phys->ds_bp;
2771 csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp;
2772 csa->cds->ds_phys->ds_bp = tmp;
2775 /* set dd_*_bytes */
2777 int64_t dused, dcomp, duncomp;
2778 uint64_t cdl_used, cdl_comp, cdl_uncomp;
2779 uint64_t odl_used, odl_comp, odl_uncomp;
2781 ASSERT3U(csa->cds->ds_dir->dd_phys->
2782 dd_used_breakdown[DD_USED_SNAP], ==, 0);
2784 VERIFY(0 == bplist_space(&csa->cds->ds_deadlist, &cdl_used,
2785 &cdl_comp, &cdl_uncomp));
2786 VERIFY(0 == bplist_space(&csa->ohds->ds_deadlist, &odl_used,
2787 &odl_comp, &odl_uncomp));
2789 dused = csa->cds->ds_phys->ds_used_bytes + cdl_used -
2790 (csa->ohds->ds_phys->ds_used_bytes + odl_used);
2791 dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp -
2792 (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp);
2793 duncomp = csa->cds->ds_phys->ds_uncompressed_bytes +
2795 (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp);
2797 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD,
2798 dused, dcomp, duncomp, tx);
2799 dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD,
2800 -dused, -dcomp, -duncomp, tx);
2803 * The difference in the space used by snapshots is the
2804 * difference in snapshot space due to the head's
2805 * deadlist (since that's the only thing that's
2806 * changing that affects the snapused).
2808 VERIFY(0 == bplist_space_birthrange(&csa->cds->ds_deadlist,
2809 csa->ohds->ds_origin_txg, UINT64_MAX, &cdl_used));
2810 VERIFY(0 == bplist_space_birthrange(&csa->ohds->ds_deadlist,
2811 csa->ohds->ds_origin_txg, UINT64_MAX, &odl_used));
2812 dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used,
2813 DD_USED_HEAD, DD_USED_SNAP, tx);
2816 #define SWITCH64(x, y) \
2818 uint64_t __tmp = (x); \
2823 /* swap ds_*_bytes */
2824 SWITCH64(csa->ohds->ds_phys->ds_used_bytes,
2825 csa->cds->ds_phys->ds_used_bytes);
2826 SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes,
2827 csa->cds->ds_phys->ds_compressed_bytes);
2828 SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes,
2829 csa->cds->ds_phys->ds_uncompressed_bytes);
2830 SWITCH64(csa->ohds->ds_phys->ds_unique_bytes,
2831 csa->cds->ds_phys->ds_unique_bytes);
2833 /* apply any parent delta for change in unconsumed refreservation */
2834 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV,
2835 csa->unused_refres_delta, 0, 0, tx);
2837 /* swap deadlists */
2838 bplist_close(&csa->cds->ds_deadlist);
2839 bplist_close(&csa->ohds->ds_deadlist);
2840 SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj,
2841 csa->cds->ds_phys->ds_deadlist_obj);
2842 VERIFY(0 == bplist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset,
2843 csa->cds->ds_phys->ds_deadlist_obj));
2844 VERIFY(0 == bplist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset,
2845 csa->ohds->ds_phys->ds_deadlist_obj));
2849 * Swap 'clone' with its origin head file system. Used at the end
2850 * of "online recv" to swizzle the file system to the new version.
2853 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
2856 struct cloneswaparg csa;
2859 ASSERT(clone->ds_owner);
2860 ASSERT(origin_head->ds_owner);
2862 /* Need exclusive access for the swap */
2863 rw_enter(&clone->ds_rwlock, RW_WRITER);
2864 if (!rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) {
2865 rw_exit(&clone->ds_rwlock);
2866 rw_enter(&origin_head->ds_rwlock, RW_WRITER);
2867 if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) {
2868 rw_exit(&origin_head->ds_rwlock);
2873 csa.ohds = origin_head;
2875 error = dsl_sync_task_do(clone->ds_dir->dd_pool,
2876 dsl_dataset_clone_swap_check,
2877 dsl_dataset_clone_swap_sync, &csa, NULL, 9);
2882 * Given a pool name and a dataset object number in that pool,
2883 * return the name of that dataset.
2886 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2893 if ((error = spa_open(pname, &spa, FTAG)) != 0)
2895 dp = spa_get_dsl(spa);
2896 rw_enter(&dp->dp_config_rwlock, RW_READER);
2897 if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) {
2898 dsl_dataset_name(ds, buf);
2899 dsl_dataset_rele(ds, FTAG);
2901 rw_exit(&dp->dp_config_rwlock);
2902 spa_close(spa, FTAG);
2908 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
2909 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
2913 ASSERT3S(asize, >, 0);
2916 * *ref_rsrv is the portion of asize that will come from any
2917 * unconsumed refreservation space.
2921 mutex_enter(&ds->ds_lock);
2923 * Make a space adjustment for reserved bytes.
2925 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
2927 ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2928 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2930 asize - MIN(asize, parent_delta(ds, asize + inflight));
2933 if (!check_quota || ds->ds_quota == 0) {
2934 mutex_exit(&ds->ds_lock);
2938 * If they are requesting more space, and our current estimate
2939 * is over quota, they get to try again unless the actual
2940 * on-disk is over quota and there are no pending changes (which
2941 * may free up space for us).
2943 if (ds->ds_phys->ds_used_bytes + inflight >= ds->ds_quota) {
2944 if (inflight > 0 || ds->ds_phys->ds_used_bytes < ds->ds_quota)
2949 mutex_exit(&ds->ds_lock);
2956 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
2958 dsl_dataset_t *ds = arg1;
2959 uint64_t *quotap = arg2;
2960 uint64_t new_quota = *quotap;
2962 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA)
2968 if (new_quota < ds->ds_phys->ds_used_bytes ||
2969 new_quota < ds->ds_reserved)
2977 dsl_dataset_set_quota_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2979 dsl_dataset_t *ds = arg1;
2980 uint64_t *quotap = arg2;
2981 uint64_t new_quota = *quotap;
2983 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2985 ds->ds_quota = new_quota;
2987 dsl_prop_set_uint64_sync(ds->ds_dir, "refquota", new_quota, cr, tx);
2989 spa_history_internal_log(LOG_DS_REFQUOTA, ds->ds_dir->dd_pool->dp_spa,
2990 tx, cr, "%lld dataset = %llu ",
2991 (longlong_t)new_quota, ds->ds_object);
2995 dsl_dataset_set_quota(const char *dsname, uint64_t quota)
3000 err = dsl_dataset_hold(dsname, FTAG, &ds);
3004 if (quota != ds->ds_quota) {
3006 * If someone removes a file, then tries to set the quota, we
3007 * want to make sure the file freeing takes effect.
3009 txg_wait_open(ds->ds_dir->dd_pool, 0);
3011 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3012 dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync,
3015 dsl_dataset_rele(ds, FTAG);
3020 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
3022 dsl_dataset_t *ds = arg1;
3023 uint64_t *reservationp = arg2;
3024 uint64_t new_reservation = *reservationp;
3028 if (new_reservation > INT64_MAX)
3031 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
3032 SPA_VERSION_REFRESERVATION)
3035 if (dsl_dataset_is_snapshot(ds))
3039 * If we are doing the preliminary check in open context, the
3040 * space estimates may be inaccurate.
3042 if (!dmu_tx_is_syncing(tx))
3045 mutex_enter(&ds->ds_lock);
3046 unique = dsl_dataset_unique(ds);
3047 delta = MAX(unique, new_reservation) - MAX(unique, ds->ds_reserved);
3048 mutex_exit(&ds->ds_lock);
3051 delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
3053 if (delta > 0 && ds->ds_quota > 0 &&
3054 new_reservation > ds->ds_quota)
3062 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, cred_t *cr,
3065 dsl_dataset_t *ds = arg1;
3066 uint64_t *reservationp = arg2;
3067 uint64_t new_reservation = *reservationp;
3071 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3073 mutex_enter(&ds->ds_dir->dd_lock);
3074 mutex_enter(&ds->ds_lock);
3075 unique = dsl_dataset_unique(ds);
3076 delta = MAX(0, (int64_t)(new_reservation - unique)) -
3077 MAX(0, (int64_t)(ds->ds_reserved - unique));
3078 ds->ds_reserved = new_reservation;
3079 mutex_exit(&ds->ds_lock);
3081 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3082 mutex_exit(&ds->ds_dir->dd_lock);
3083 dsl_prop_set_uint64_sync(ds->ds_dir, "refreservation",
3084 new_reservation, cr, tx);
3086 spa_history_internal_log(LOG_DS_REFRESERV,
3087 ds->ds_dir->dd_pool->dp_spa, tx, cr, "%lld dataset = %llu",
3088 (longlong_t)new_reservation, ds->ds_object);
3092 dsl_dataset_set_reservation(const char *dsname, uint64_t reservation)
3097 err = dsl_dataset_hold(dsname, FTAG, &ds);
3101 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3102 dsl_dataset_set_reservation_check,
3103 dsl_dataset_set_reservation_sync, ds, &reservation, 0);
3104 dsl_dataset_rele(ds, FTAG);