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 #pragma ident "@(#)dsl_dataset.c 1.42 08/04/28 SMI"
28 #include <sys/dmu_objset.h>
29 #include <sys/dsl_dataset.h>
30 #include <sys/dsl_dir.h>
31 #include <sys/dsl_prop.h>
32 #include <sys/dsl_synctask.h>
33 #include <sys/dmu_traverse.h>
34 #include <sys/dmu_tx.h>
38 #include <sys/unique.h>
39 #include <sys/zfs_context.h>
40 #include <sys/zfs_ioctl.h>
42 #include <sys/sunddi.h>
44 static dsl_checkfunc_t dsl_dataset_destroy_begin_check;
45 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync;
46 static dsl_checkfunc_t dsl_dataset_rollback_check;
47 static dsl_syncfunc_t dsl_dataset_rollback_sync;
48 static dsl_syncfunc_t dsl_dataset_set_reservation_sync;
50 #define DS_REF_MAX (1ULL << 62)
52 #define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE
55 * We use weighted reference counts to express the various forms of exclusion
56 * between different open modes. A STANDARD open is 1 point, an EXCLUSIVE open
57 * is DS_REF_MAX, and a PRIMARY open is little more than half of an EXCLUSIVE.
58 * This makes the exclusion logic simple: the total refcnt for all opens cannot
59 * exceed DS_REF_MAX. For example, EXCLUSIVE opens are exclusive because their
60 * weight (DS_REF_MAX) consumes the entire refcnt space. PRIMARY opens consume
61 * just over half of the refcnt space, so there can't be more than one, but it
62 * can peacefully coexist with any number of STANDARD opens.
64 static uint64_t ds_refcnt_weight[DS_MODE_LEVELS] = {
65 0, /* DS_MODE_NONE - invalid */
66 1, /* DS_MODE_STANDARD - unlimited number */
67 (DS_REF_MAX >> 1) + 1, /* DS_MODE_PRIMARY - only one of these */
68 DS_REF_MAX /* DS_MODE_EXCLUSIVE - no other opens */
72 * Figure out how much of this delta should be propogated to the dsl_dir
73 * layer. If there's a refreservation, that space has already been
74 * partially accounted for in our ancestors.
77 parent_delta(dsl_dataset_t *ds, int64_t delta)
79 uint64_t old_bytes, new_bytes;
81 if (ds->ds_reserved == 0)
84 old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
85 new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
87 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
88 return (new_bytes - old_bytes);
92 dsl_dataset_block_born(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
94 int used = bp_get_dasize(tx->tx_pool->dp_spa, bp);
95 int compressed = BP_GET_PSIZE(bp);
96 int uncompressed = BP_GET_UCSIZE(bp);
99 dprintf_bp(bp, "born, ds=%p\n", ds);
101 ASSERT(dmu_tx_is_syncing(tx));
102 /* It could have been compressed away to nothing */
105 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
106 ASSERT3U(BP_GET_TYPE(bp), <, DMU_OT_NUMTYPES);
109 * Account for the meta-objset space in its placeholder
112 ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */
113 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir,
114 used, compressed, uncompressed, tx);
115 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
118 dmu_buf_will_dirty(ds->ds_dbuf, tx);
119 mutex_enter(&ds->ds_lock);
120 delta = parent_delta(ds, used);
121 ds->ds_phys->ds_used_bytes += used;
122 ds->ds_phys->ds_compressed_bytes += compressed;
123 ds->ds_phys->ds_uncompressed_bytes += uncompressed;
124 ds->ds_phys->ds_unique_bytes += used;
125 mutex_exit(&ds->ds_lock);
126 dsl_dir_diduse_space(ds->ds_dir, delta, compressed, uncompressed, tx);
130 dsl_dataset_block_kill(dsl_dataset_t *ds, blkptr_t *bp, zio_t *pio,
133 int used = bp_get_dasize(tx->tx_pool->dp_spa, bp);
134 int compressed = BP_GET_PSIZE(bp);
135 int uncompressed = BP_GET_UCSIZE(bp);
137 ASSERT(dmu_tx_is_syncing(tx));
138 /* No block pointer => nothing to free */
146 * Account for the meta-objset space in its placeholder
149 err = arc_free(pio, tx->tx_pool->dp_spa,
150 tx->tx_txg, bp, NULL, NULL, pio ? ARC_NOWAIT: ARC_WAIT);
153 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir,
154 -used, -compressed, -uncompressed, tx);
155 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
158 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
160 dmu_buf_will_dirty(ds->ds_dbuf, tx);
162 if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
166 dprintf_bp(bp, "freeing: %s", "");
167 err = arc_free(pio, tx->tx_pool->dp_spa,
168 tx->tx_txg, bp, NULL, NULL, pio ? ARC_NOWAIT: ARC_WAIT);
171 mutex_enter(&ds->ds_lock);
172 ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
173 !DS_UNIQUE_IS_ACCURATE(ds));
174 delta = parent_delta(ds, -used);
175 ds->ds_phys->ds_unique_bytes -= used;
176 mutex_exit(&ds->ds_lock);
177 dsl_dir_diduse_space(ds->ds_dir,
178 delta, -compressed, -uncompressed, tx);
180 dprintf_bp(bp, "putting on dead list: %s", "");
181 VERIFY(0 == bplist_enqueue(&ds->ds_deadlist, bp, tx));
182 ASSERT3U(ds->ds_prev->ds_object, ==,
183 ds->ds_phys->ds_prev_snap_obj);
184 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
185 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
186 if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
187 ds->ds_object && bp->blk_birth >
188 ds->ds_prev->ds_phys->ds_prev_snap_txg) {
189 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
190 mutex_enter(&ds->ds_prev->ds_lock);
191 ds->ds_prev->ds_phys->ds_unique_bytes += used;
192 mutex_exit(&ds->ds_prev->ds_lock);
195 mutex_enter(&ds->ds_lock);
196 ASSERT3U(ds->ds_phys->ds_used_bytes, >=, used);
197 ds->ds_phys->ds_used_bytes -= used;
198 ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
199 ds->ds_phys->ds_compressed_bytes -= compressed;
200 ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
201 ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
202 mutex_exit(&ds->ds_lock);
206 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
208 uint64_t trysnap = 0;
213 * The snapshot creation could fail, but that would cause an
214 * incorrect FALSE return, which would only result in an
215 * overestimation of the amount of space that an operation would
216 * consume, which is OK.
218 * There's also a small window where we could miss a pending
219 * snapshot, because we could set the sync task in the quiescing
220 * phase. So this should only be used as a guess.
222 if (ds->ds_trysnap_txg >
223 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
224 trysnap = ds->ds_trysnap_txg;
225 return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
229 dsl_dataset_block_freeable(dsl_dataset_t *ds, uint64_t blk_birth)
231 return (blk_birth > dsl_dataset_prev_snap_txg(ds));
236 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
238 dsl_dataset_t *ds = dsv;
240 /* open_refcount == DS_REF_MAX when deleting */
241 ASSERT(ds->ds_open_refcount == 0 ||
242 ds->ds_open_refcount == DS_REF_MAX);
244 dprintf_ds(ds, "evicting %s\n", "");
246 unique_remove(ds->ds_fsid_guid);
248 if (ds->ds_user_ptr != NULL)
249 ds->ds_user_evict_func(ds, ds->ds_user_ptr);
252 dsl_dataset_close(ds->ds_prev, DS_MODE_NONE, ds);
256 bplist_close(&ds->ds_deadlist);
257 dsl_dir_close(ds->ds_dir, ds);
259 ASSERT(!list_link_active(&ds->ds_synced_link));
261 mutex_destroy(&ds->ds_lock);
262 mutex_destroy(&ds->ds_opening_lock);
263 mutex_destroy(&ds->ds_deadlist.bpl_lock);
265 kmem_free(ds, sizeof (dsl_dataset_t));
269 dsl_dataset_get_snapname(dsl_dataset_t *ds)
271 dsl_dataset_phys_t *headphys;
274 dsl_pool_t *dp = ds->ds_dir->dd_pool;
275 objset_t *mos = dp->dp_meta_objset;
277 if (ds->ds_snapname[0])
279 if (ds->ds_phys->ds_next_snap_obj == 0)
282 err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
286 headphys = headdbuf->db_data;
287 err = zap_value_search(dp->dp_meta_objset,
288 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
289 dmu_buf_rele(headdbuf, FTAG);
294 dsl_dataset_snap_lookup(objset_t *os, uint64_t flags,
295 uint64_t snapnames_zapobj, const char *name, uint64_t *value)
300 if (flags & DS_FLAG_CI_DATASET)
305 err = zap_lookup_norm(os, snapnames_zapobj, name, 8, 1,
306 value, mt, NULL, 0, NULL);
307 if (err == ENOTSUP && mt == MT_FIRST)
308 err = zap_lookup(os, snapnames_zapobj, name, 8, 1, value);
313 dsl_dataset_snap_remove(objset_t *os, uint64_t flags,
314 uint64_t snapnames_zapobj, char *name, dmu_tx_t *tx)
319 if (flags & DS_FLAG_CI_DATASET)
324 err = zap_remove_norm(os, snapnames_zapobj, name, mt, tx);
325 if (err == ENOTSUP && mt == MT_FIRST)
326 err = zap_remove(os, snapnames_zapobj, name, tx);
331 dsl_dataset_open_obj(dsl_pool_t *dp, uint64_t dsobj, const char *snapname,
332 int mode, void *tag, dsl_dataset_t **dsp)
334 uint64_t weight = ds_refcnt_weight[DS_MODE_LEVEL(mode)];
335 objset_t *mos = dp->dp_meta_objset;
340 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
341 dsl_pool_sync_context(dp));
343 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
346 ds = dmu_buf_get_user(dbuf);
348 dsl_dataset_t *winner;
350 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
352 ds->ds_object = dsobj;
353 ds->ds_phys = dbuf->db_data;
355 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
356 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
357 mutex_init(&ds->ds_deadlist.bpl_lock, NULL, MUTEX_DEFAULT,
360 err = bplist_open(&ds->ds_deadlist,
361 mos, ds->ds_phys->ds_deadlist_obj);
363 err = dsl_dir_open_obj(dp,
364 ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
368 * we don't really need to close the blist if we
371 mutex_destroy(&ds->ds_lock);
372 mutex_destroy(&ds->ds_opening_lock);
373 mutex_destroy(&ds->ds_deadlist.bpl_lock);
374 kmem_free(ds, sizeof (dsl_dataset_t));
375 dmu_buf_rele(dbuf, tag);
379 if (ds->ds_dir->dd_phys->dd_head_dataset_obj == dsobj) {
380 ds->ds_snapname[0] = '\0';
381 if (ds->ds_phys->ds_prev_snap_obj) {
382 err = dsl_dataset_open_obj(dp,
383 ds->ds_phys->ds_prev_snap_obj, NULL,
384 DS_MODE_NONE, ds, &ds->ds_prev);
389 dsl_dataset_phys_t *headphys;
391 err = dmu_bonus_hold(mos,
392 ds->ds_dir->dd_phys->dd_head_dataset_obj,
397 headphys = headdbuf->db_data;
398 err = dsl_dataset_snap_lookup(
401 headphys->ds_snapnames_zapobj,
402 snapname, &foundobj);
403 ASSERT3U(foundobj, ==, dsobj);
404 dmu_buf_rele(headdbuf, FTAG);
407 (void) strcat(ds->ds_snapname, snapname);
408 } else if (zfs_flags & ZFS_DEBUG_SNAPNAMES) {
409 err = dsl_dataset_get_snapname(ds);
413 if (!dsl_dataset_is_snapshot(ds)) {
415 * In sync context, we're called with either no lock
416 * or with the write lock. If we're not syncing,
417 * we're always called with the read lock held.
419 boolean_t need_lock =
420 !RW_WRITE_HELD(&dp->dp_config_rwlock) &&
421 dsl_pool_sync_context(dp);
424 rw_enter(&dp->dp_config_rwlock, RW_READER);
426 err = dsl_prop_get_ds_locked(ds->ds_dir,
427 "refreservation", sizeof (uint64_t), 1,
428 &ds->ds_reserved, NULL);
430 err = dsl_prop_get_ds_locked(ds->ds_dir,
431 "refquota", sizeof (uint64_t), 1,
432 &ds->ds_quota, NULL);
436 rw_exit(&dp->dp_config_rwlock);
438 ds->ds_reserved = ds->ds_quota = 0;
442 winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys,
446 bplist_close(&ds->ds_deadlist);
448 dsl_dataset_close(ds->ds_prev,
451 dsl_dir_close(ds->ds_dir, ds);
452 mutex_destroy(&ds->ds_lock);
453 mutex_destroy(&ds->ds_opening_lock);
454 mutex_destroy(&ds->ds_deadlist.bpl_lock);
455 kmem_free(ds, sizeof (dsl_dataset_t));
457 dmu_buf_rele(dbuf, tag);
463 unique_insert(ds->ds_phys->ds_fsid_guid);
466 ASSERT3P(ds->ds_dbuf, ==, dbuf);
467 ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
469 mutex_enter(&ds->ds_lock);
470 if ((DS_MODE_LEVEL(mode) == DS_MODE_PRIMARY &&
471 (ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT) &&
472 !DS_MODE_IS_INCONSISTENT(mode)) ||
473 (ds->ds_open_refcount + weight > DS_REF_MAX)) {
474 mutex_exit(&ds->ds_lock);
475 dsl_dataset_close(ds, DS_MODE_NONE, tag);
478 ds->ds_open_refcount += weight;
479 mutex_exit(&ds->ds_lock);
486 dsl_dataset_open_spa(spa_t *spa, const char *name, int mode,
487 void *tag, dsl_dataset_t **dsp)
493 dsl_dataset_t *ds = NULL;
496 err = dsl_dir_open_spa(spa, name, FTAG, &dd, &tail);
501 obj = dd->dd_phys->dd_head_dataset_obj;
502 rw_enter(&dp->dp_config_rwlock, RW_READER);
504 /* A dataset with no associated objset */
510 objset_t *mos = dp->dp_meta_objset;
513 err = dsl_dataset_open_obj(dp, obj, NULL,
514 DS_MODE_NONE, tag, &ds);
517 flags = ds->ds_phys->ds_flags;
518 obj = ds->ds_phys->ds_snapnames_zapobj;
519 dsl_dataset_close(ds, DS_MODE_NONE, tag);
522 if (tail[0] != '@') {
528 /* Look for a snapshot */
529 if (!DS_MODE_IS_READONLY(mode)) {
533 dprintf("looking for snapshot '%s'\n", tail);
534 err = dsl_dataset_snap_lookup(mos, flags, obj, tail, &obj);
538 err = dsl_dataset_open_obj(dp, obj, tail, mode, tag, &ds);
541 rw_exit(&dp->dp_config_rwlock);
542 dsl_dir_close(dd, FTAG);
544 ASSERT3U((err == 0), ==, (ds != NULL));
545 /* ASSERT(ds == NULL || strcmp(name, ds->ds_name) == 0); */
552 dsl_dataset_open(const char *name, int mode, void *tag, dsl_dataset_t **dsp)
554 return (dsl_dataset_open_spa(NULL, name, mode, tag, dsp));
558 dsl_dataset_name(dsl_dataset_t *ds, char *name)
561 (void) strcpy(name, "mos");
563 dsl_dir_name(ds->ds_dir, name);
564 VERIFY(0 == dsl_dataset_get_snapname(ds));
565 if (ds->ds_snapname[0]) {
566 (void) strcat(name, "@");
567 if (!MUTEX_HELD(&ds->ds_lock)) {
569 * We use a "recursive" mutex so that we
570 * can call dprintf_ds() with ds_lock held.
572 mutex_enter(&ds->ds_lock);
573 (void) strcat(name, ds->ds_snapname);
574 mutex_exit(&ds->ds_lock);
576 (void) strcat(name, ds->ds_snapname);
583 dsl_dataset_namelen(dsl_dataset_t *ds)
588 result = 3; /* "mos" */
590 result = dsl_dir_namelen(ds->ds_dir);
591 VERIFY(0 == dsl_dataset_get_snapname(ds));
592 if (ds->ds_snapname[0]) {
593 ++result; /* adding one for the @-sign */
594 if (!MUTEX_HELD(&ds->ds_lock)) {
595 /* see dsl_datset_name */
596 mutex_enter(&ds->ds_lock);
597 result += strlen(ds->ds_snapname);
598 mutex_exit(&ds->ds_lock);
600 result += strlen(ds->ds_snapname);
609 dsl_dataset_close(dsl_dataset_t *ds, int mode, void *tag)
611 uint64_t weight = ds_refcnt_weight[DS_MODE_LEVEL(mode)];
612 mutex_enter(&ds->ds_lock);
613 ASSERT3U(ds->ds_open_refcount, >=, weight);
614 ds->ds_open_refcount -= weight;
615 mutex_exit(&ds->ds_lock);
617 dmu_buf_rele(ds->ds_dbuf, tag);
621 dsl_dataset_downgrade(dsl_dataset_t *ds, int oldmode, int newmode)
623 uint64_t oldweight = ds_refcnt_weight[DS_MODE_LEVEL(oldmode)];
624 uint64_t newweight = ds_refcnt_weight[DS_MODE_LEVEL(newmode)];
625 mutex_enter(&ds->ds_lock);
626 ASSERT3U(ds->ds_open_refcount, >=, oldweight);
627 ASSERT3U(oldweight, >=, newweight);
628 ds->ds_open_refcount -= oldweight;
629 ds->ds_open_refcount += newweight;
630 mutex_exit(&ds->ds_lock);
634 dsl_dataset_tryupgrade(dsl_dataset_t *ds, int oldmode, int newmode)
637 uint64_t oldweight = ds_refcnt_weight[DS_MODE_LEVEL(oldmode)];
638 uint64_t newweight = ds_refcnt_weight[DS_MODE_LEVEL(newmode)];
639 mutex_enter(&ds->ds_lock);
640 ASSERT3U(ds->ds_open_refcount, >=, oldweight);
641 ASSERT3U(newweight, >=, oldweight);
642 if (ds->ds_open_refcount - oldweight + newweight > DS_REF_MAX) {
645 ds->ds_open_refcount -= oldweight;
646 ds->ds_open_refcount += newweight;
649 mutex_exit(&ds->ds_lock);
654 dsl_dataset_create_root(dsl_pool_t *dp, uint64_t *ddobjp, dmu_tx_t *tx)
656 objset_t *mos = dp->dp_meta_objset;
658 dsl_dataset_phys_t *dsphys;
663 dsl_dir_create_root(mos, ddobjp, tx);
664 VERIFY(0 == dsl_dir_open_obj(dp, *ddobjp, NULL, FTAG, &dd));
666 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
667 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
668 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
669 dmu_buf_will_dirty(dbuf, tx);
670 dsphys = dbuf->db_data;
671 dsphys->ds_dir_obj = dd->dd_object;
672 dsphys->ds_fsid_guid = unique_create();
673 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
674 sizeof (dsphys->ds_guid));
675 dsphys->ds_snapnames_zapobj =
676 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
678 dsphys->ds_creation_time = gethrestime_sec();
679 dsphys->ds_creation_txg = tx->tx_txg;
680 dsphys->ds_deadlist_obj =
681 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
682 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
683 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
684 dmu_buf_rele(dbuf, FTAG);
686 dmu_buf_will_dirty(dd->dd_dbuf, tx);
687 dd->dd_phys->dd_head_dataset_obj = dsobj;
688 dsl_dir_close(dd, FTAG);
691 dsl_dataset_open_obj(dp, dsobj, NULL, DS_MODE_NONE, FTAG, &ds));
692 (void) dmu_objset_create_impl(dp->dp_spa, ds,
693 &ds->ds_phys->ds_bp, DMU_OST_ZFS, tx);
694 dsl_dataset_close(ds, DS_MODE_NONE, FTAG);
698 dsl_dataset_create_sync_impl(dsl_dir_t *dd, dsl_dataset_t *origin,
699 uint64_t flags, dmu_tx_t *tx)
701 dsl_pool_t *dp = dd->dd_pool;
703 dsl_dataset_phys_t *dsphys;
705 objset_t *mos = dp->dp_meta_objset;
707 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
708 ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
709 ASSERT(dmu_tx_is_syncing(tx));
710 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
712 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
713 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
714 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
715 dmu_buf_will_dirty(dbuf, tx);
716 dsphys = dbuf->db_data;
717 dsphys->ds_dir_obj = dd->dd_object;
718 dsphys->ds_flags = flags;
719 dsphys->ds_fsid_guid = unique_create();
720 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
721 sizeof (dsphys->ds_guid));
722 dsphys->ds_snapnames_zapobj =
723 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
725 dsphys->ds_creation_time = gethrestime_sec();
726 dsphys->ds_creation_txg = tx->tx_txg;
727 dsphys->ds_deadlist_obj =
728 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
731 dsphys->ds_prev_snap_obj = origin->ds_object;
732 dsphys->ds_prev_snap_txg =
733 origin->ds_phys->ds_creation_txg;
734 dsphys->ds_used_bytes =
735 origin->ds_phys->ds_used_bytes;
736 dsphys->ds_compressed_bytes =
737 origin->ds_phys->ds_compressed_bytes;
738 dsphys->ds_uncompressed_bytes =
739 origin->ds_phys->ds_uncompressed_bytes;
740 dsphys->ds_bp = origin->ds_phys->ds_bp;
741 dsphys->ds_flags |= origin->ds_phys->ds_flags;
743 dmu_buf_will_dirty(origin->ds_dbuf, tx);
744 origin->ds_phys->ds_num_children++;
746 dmu_buf_will_dirty(dd->dd_dbuf, tx);
747 dd->dd_phys->dd_origin_obj = origin->ds_object;
750 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
751 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
753 dmu_buf_rele(dbuf, FTAG);
755 dmu_buf_will_dirty(dd->dd_dbuf, tx);
756 dd->dd_phys->dd_head_dataset_obj = dsobj;
762 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
763 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
765 dsl_pool_t *dp = pdd->dd_pool;
766 uint64_t dsobj, ddobj;
769 ASSERT(lastname[0] != '@');
771 ddobj = dsl_dir_create_sync(pdd, lastname, tx);
772 VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd));
774 dsobj = dsl_dataset_create_sync_impl(dd, origin, flags, tx);
776 dsl_deleg_set_create_perms(dd, tx, cr);
778 dsl_dir_close(dd, FTAG);
784 dsl_sync_task_group_t *dstg;
790 dsl_snapshot_destroy_one(char *name, void *arg)
792 struct destroyarg *da = arg;
797 (void) strcat(name, "@");
798 (void) strcat(name, da->snapname);
799 err = dsl_dataset_open(name,
800 DS_MODE_EXCLUSIVE | DS_MODE_READONLY | DS_MODE_INCONSISTENT,
802 cp = strchr(name, '@');
807 (void) strcpy(da->failed, name);
811 dsl_sync_task_create(da->dstg, dsl_dataset_destroy_check,
812 dsl_dataset_destroy_sync, ds, da->dstg, 0);
817 * Destroy 'snapname' in all descendants of 'fsname'.
819 #pragma weak dmu_snapshots_destroy = dsl_snapshots_destroy
821 dsl_snapshots_destroy(char *fsname, char *snapname)
824 struct destroyarg da;
825 dsl_sync_task_t *dst;
828 err = spa_open(fsname, &spa, FTAG);
831 da.dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
832 da.snapname = snapname;
835 err = dmu_objset_find(fsname,
836 dsl_snapshot_destroy_one, &da, DS_FIND_CHILDREN);
839 err = dsl_sync_task_group_wait(da.dstg);
841 for (dst = list_head(&da.dstg->dstg_tasks); dst;
842 dst = list_next(&da.dstg->dstg_tasks, dst)) {
843 dsl_dataset_t *ds = dst->dst_arg1;
845 dsl_dataset_name(ds, fsname);
846 *strchr(fsname, '@') = '\0';
849 * If it was successful, destroy_sync would have
853 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, da.dstg);
856 dsl_sync_task_group_destroy(da.dstg);
857 spa_close(spa, FTAG);
862 * ds must be opened EXCLUSIVE or PRIMARY. on return (whether
863 * successful or not), ds will be closed and caller can no longer
867 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag)
870 dsl_sync_task_group_t *dstg;
875 if (ds->ds_open_refcount != DS_REF_MAX) {
876 if (dsl_dataset_tryupgrade(ds, DS_MODE_PRIMARY,
877 DS_MODE_EXCLUSIVE) == 0) {
878 dsl_dataset_close(ds, DS_MODE_PRIMARY, tag);
883 if (dsl_dataset_is_snapshot(ds)) {
884 /* Destroying a snapshot is simpler */
885 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
886 dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
894 * Check for errors and mark this ds as inconsistent, in
895 * case we crash while freeing the objects.
897 err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check,
898 dsl_dataset_destroy_begin_sync, ds, NULL, 0);
902 err = dmu_objset_open_ds(ds, DMU_OST_ANY, &os);
907 * remove the objects in open context, so that we won't
908 * have too much to do in syncing context.
910 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
911 ds->ds_phys->ds_prev_snap_txg)) {
912 dmu_tx_t *tx = dmu_tx_create(os);
913 dmu_tx_hold_free(tx, obj, 0, DMU_OBJECT_END);
914 dmu_tx_hold_bonus(tx, obj);
915 err = dmu_tx_assign(tx, TXG_WAIT);
918 * Perhaps there is not enough disk
919 * space. Just deal with it from
920 * dsl_dataset_destroy_sync().
925 VERIFY(0 == dmu_object_free(os, obj, tx));
928 /* Make sure it's not dirty before we finish destroying it. */
929 txg_wait_synced(dd->dd_pool, 0);
931 dmu_objset_close(os);
935 if (ds->ds_user_ptr) {
936 ds->ds_user_evict_func(ds, ds->ds_user_ptr);
937 ds->ds_user_ptr = NULL;
940 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
941 err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd);
942 rw_exit(&dd->dd_pool->dp_config_rwlock);
948 * Blow away the dsl_dir + head dataset.
950 dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
951 dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
952 dsl_dataset_destroy_sync, ds, tag, 0);
953 dsl_sync_task_create(dstg, dsl_dir_destroy_check,
954 dsl_dir_destroy_sync, dd, FTAG, 0);
955 err = dsl_sync_task_group_wait(dstg);
956 dsl_sync_task_group_destroy(dstg);
957 /* if it is successful, *destroy_sync will close the ds+dd */
959 dsl_dir_close(dd, FTAG);
962 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, tag);
967 dsl_dataset_rollback(dsl_dataset_t *ds, dmu_objset_type_t ost)
969 ASSERT3U(ds->ds_open_refcount, ==, DS_REF_MAX);
971 return (dsl_sync_task_do(ds->ds_dir->dd_pool,
972 dsl_dataset_rollback_check, dsl_dataset_rollback_sync,
977 dsl_dataset_set_user_ptr(dsl_dataset_t *ds,
978 void *p, dsl_dataset_evict_func_t func)
982 mutex_enter(&ds->ds_lock);
983 old = ds->ds_user_ptr;
986 ds->ds_user_evict_func = func;
988 mutex_exit(&ds->ds_lock);
993 dsl_dataset_get_user_ptr(dsl_dataset_t *ds)
995 return (ds->ds_user_ptr);
1000 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1002 return (&ds->ds_phys->ds_bp);
1006 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1008 ASSERT(dmu_tx_is_syncing(tx));
1009 /* If it's the meta-objset, set dp_meta_rootbp */
1011 tx->tx_pool->dp_meta_rootbp = *bp;
1013 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1014 ds->ds_phys->ds_bp = *bp;
1019 dsl_dataset_get_spa(dsl_dataset_t *ds)
1021 return (ds->ds_dir->dd_pool->dp_spa);
1025 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1029 if (ds == NULL) /* this is the meta-objset */
1032 ASSERT(ds->ds_user_ptr != NULL);
1034 if (ds->ds_phys->ds_next_snap_obj != 0)
1035 panic("dirtying snapshot!");
1037 dp = ds->ds_dir->dd_pool;
1039 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) {
1040 /* up the hold count until we can be written out */
1041 dmu_buf_add_ref(ds->ds_dbuf, ds);
1046 * The unique space in the head dataset can be calculated by subtracting
1047 * the space used in the most recent snapshot, that is still being used
1048 * in this file system, from the space currently in use. To figure out
1049 * the space in the most recent snapshot still in use, we need to take
1050 * the total space used in the snapshot and subtract out the space that
1051 * has been freed up since the snapshot was taken.
1054 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1057 uint64_t dlused, dlcomp, dluncomp;
1059 ASSERT(ds->ds_object == ds->ds_dir->dd_phys->dd_head_dataset_obj);
1061 if (ds->ds_phys->ds_prev_snap_obj != 0)
1062 mrs_used = ds->ds_prev->ds_phys->ds_used_bytes;
1066 VERIFY(0 == bplist_space(&ds->ds_deadlist, &dlused, &dlcomp,
1069 ASSERT3U(dlused, <=, mrs_used);
1070 ds->ds_phys->ds_unique_bytes =
1071 ds->ds_phys->ds_used_bytes - (mrs_used - dlused);
1073 if (!DS_UNIQUE_IS_ACCURATE(ds) &&
1074 spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1075 SPA_VERSION_UNIQUE_ACCURATE)
1076 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1080 dsl_dataset_unique(dsl_dataset_t *ds)
1082 if (!DS_UNIQUE_IS_ACCURATE(ds) && !dsl_dataset_is_snapshot(ds))
1083 dsl_dataset_recalc_head_uniq(ds);
1085 return (ds->ds_phys->ds_unique_bytes);
1090 int64_t *compressedp;
1091 int64_t *uncompressedp;
1097 kill_blkptr(traverse_blk_cache_t *bc, spa_t *spa, void *arg)
1099 struct killarg *ka = arg;
1100 blkptr_t *bp = &bc->bc_blkptr;
1102 ASSERT3U(bc->bc_errno, ==, 0);
1105 * Since this callback is not called concurrently, no lock is
1106 * needed on the accounting values.
1108 *ka->usedp += bp_get_dasize(spa, bp);
1109 *ka->compressedp += BP_GET_PSIZE(bp);
1110 *ka->uncompressedp += BP_GET_UCSIZE(bp);
1111 /* XXX check for EIO? */
1112 (void) arc_free(ka->zio, spa, ka->tx->tx_txg, bp, NULL, NULL,
1119 dsl_dataset_rollback_check(void *arg1, void *arg2, dmu_tx_t *tx)
1121 dsl_dataset_t *ds = arg1;
1122 dmu_objset_type_t *ost = arg2;
1125 * We can only roll back to emptyness if it is a ZPL objset.
1127 if (*ost != DMU_OST_ZFS && ds->ds_phys->ds_prev_snap_txg == 0)
1131 * This must not be a snapshot.
1133 if (ds->ds_phys->ds_next_snap_obj != 0)
1137 * If we made changes this txg, traverse_dsl_dataset won't find
1140 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1148 dsl_dataset_rollback_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1150 dsl_dataset_t *ds = arg1;
1151 dmu_objset_type_t *ost = arg2;
1152 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1154 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1157 * Before the roll back destroy the zil.
1159 if (ds->ds_user_ptr != NULL) {
1160 zil_rollback_destroy(
1161 ((objset_impl_t *)ds->ds_user_ptr)->os_zil, tx);
1164 * We need to make sure that the objset_impl_t is reopened after
1165 * we do the rollback, otherwise it will have the wrong
1166 * objset_phys_t. Normally this would happen when this
1167 * DS_MODE_EXCLUSIVE dataset-open is closed, thus causing the
1168 * dataset to be immediately evicted. But when doing "zfs recv
1169 * -F", we reopen the objset before that, so that there is no
1170 * window where the dataset is closed and inconsistent.
1172 ds->ds_user_evict_func(ds, ds->ds_user_ptr);
1173 ds->ds_user_ptr = NULL;
1176 /* Zero out the deadlist. */
1177 bplist_close(&ds->ds_deadlist);
1178 bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx);
1179 ds->ds_phys->ds_deadlist_obj =
1180 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
1181 VERIFY(0 == bplist_open(&ds->ds_deadlist, mos,
1182 ds->ds_phys->ds_deadlist_obj));
1185 /* Free blkptrs that we gave birth to */
1187 int64_t used = 0, compressed = 0, uncompressed = 0;
1191 zio = zio_root(tx->tx_pool->dp_spa, NULL, NULL,
1192 ZIO_FLAG_MUSTSUCCEED);
1194 ka.compressedp = &compressed;
1195 ka.uncompressedp = &uncompressed;
1198 (void) traverse_dsl_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1199 ADVANCE_POST, kill_blkptr, &ka);
1200 (void) zio_wait(zio);
1202 /* only deduct space beyond any refreservation */
1203 delta = parent_delta(ds, -used);
1204 dsl_dir_diduse_space(ds->ds_dir,
1205 delta, -compressed, -uncompressed, tx);
1209 /* Change our contents to that of the prev snapshot */
1210 ASSERT3U(ds->ds_prev->ds_object, ==,
1211 ds->ds_phys->ds_prev_snap_obj);
1212 ds->ds_phys->ds_bp = ds->ds_prev->ds_phys->ds_bp;
1213 ds->ds_phys->ds_used_bytes =
1214 ds->ds_prev->ds_phys->ds_used_bytes;
1215 ds->ds_phys->ds_compressed_bytes =
1216 ds->ds_prev->ds_phys->ds_compressed_bytes;
1217 ds->ds_phys->ds_uncompressed_bytes =
1218 ds->ds_prev->ds_phys->ds_uncompressed_bytes;
1219 ds->ds_phys->ds_flags = ds->ds_prev->ds_phys->ds_flags;
1220 ds->ds_phys->ds_unique_bytes = 0;
1222 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1223 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1224 ds->ds_prev->ds_phys->ds_unique_bytes = 0;
1227 /* Zero out our contents, recreate objset */
1228 bzero(&ds->ds_phys->ds_bp, sizeof (blkptr_t));
1229 ds->ds_phys->ds_used_bytes = 0;
1230 ds->ds_phys->ds_compressed_bytes = 0;
1231 ds->ds_phys->ds_uncompressed_bytes = 0;
1232 ds->ds_phys->ds_flags = 0;
1233 ds->ds_phys->ds_unique_bytes = 0;
1234 (void) dmu_objset_create_impl(ds->ds_dir->dd_pool->dp_spa, ds,
1235 &ds->ds_phys->ds_bp, *ost, tx);
1238 spa_history_internal_log(LOG_DS_ROLLBACK, ds->ds_dir->dd_pool->dp_spa,
1239 tx, cr, "dataset = %llu", ds->ds_object);
1244 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1246 dsl_dataset_t *ds = arg1;
1247 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1252 * Can't delete a head dataset if there are snapshots of it.
1253 * (Except if the only snapshots are from the branch we cloned
1256 if (ds->ds_prev != NULL &&
1257 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1261 * This is really a dsl_dir thing, but check it here so that
1262 * we'll be less likely to leave this dataset inconsistent &
1265 err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count);
1276 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1278 dsl_dataset_t *ds = arg1;
1279 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1281 /* Mark it as inconsistent on-disk, in case we crash */
1282 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1283 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1285 spa_history_internal_log(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx,
1286 cr, "dataset = %llu", ds->ds_object);
1291 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1293 dsl_dataset_t *ds = arg1;
1295 /* Can't delete a branch point. */
1296 if (ds->ds_phys->ds_num_children > 1)
1300 * Can't delete a head dataset if there are snapshots of it.
1301 * (Except if the only snapshots are from the branch we cloned
1304 if (ds->ds_prev != NULL &&
1305 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1309 * If we made changes this txg, traverse_dsl_dataset won't find
1312 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1315 /* XXX we should do some i/o error checking... */
1320 dsl_dataset_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx)
1322 dsl_dataset_t *ds = arg1;
1323 int64_t used = 0, compressed = 0, uncompressed = 0;
1326 int after_branch_point = FALSE;
1327 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1328 objset_t *mos = dp->dp_meta_objset;
1329 dsl_dataset_t *ds_prev = NULL;
1332 ASSERT3U(ds->ds_open_refcount, ==, DS_REF_MAX);
1333 ASSERT3U(ds->ds_phys->ds_num_children, <=, 1);
1334 ASSERT(ds->ds_prev == NULL ||
1335 ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1336 ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1338 /* Remove our reservation */
1339 if (ds->ds_reserved != 0) {
1341 dsl_dataset_set_reservation_sync(ds, &val, cr, tx);
1342 ASSERT3U(ds->ds_reserved, ==, 0);
1345 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1347 obj = ds->ds_object;
1349 if (ds->ds_phys->ds_prev_snap_obj != 0) {
1351 ds_prev = ds->ds_prev;
1353 VERIFY(0 == dsl_dataset_open_obj(dp,
1354 ds->ds_phys->ds_prev_snap_obj, NULL,
1355 DS_MODE_NONE, FTAG, &ds_prev));
1357 after_branch_point =
1358 (ds_prev->ds_phys->ds_next_snap_obj != obj);
1360 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
1361 if (after_branch_point &&
1362 ds->ds_phys->ds_next_snap_obj == 0) {
1363 /* This clone is toast. */
1364 ASSERT(ds_prev->ds_phys->ds_num_children > 1);
1365 ds_prev->ds_phys->ds_num_children--;
1366 } else if (!after_branch_point) {
1367 ds_prev->ds_phys->ds_next_snap_obj =
1368 ds->ds_phys->ds_next_snap_obj;
1372 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1374 if (ds->ds_phys->ds_next_snap_obj != 0) {
1376 dsl_dataset_t *ds_next;
1378 uint64_t old_unique;
1380 spa_scrub_restart(dp->dp_spa, tx->tx_txg);
1382 VERIFY(0 == dsl_dataset_open_obj(dp,
1383 ds->ds_phys->ds_next_snap_obj, NULL,
1384 DS_MODE_NONE, FTAG, &ds_next));
1385 ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
1387 old_unique = dsl_dataset_unique(ds_next);
1389 dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
1390 ds_next->ds_phys->ds_prev_snap_obj =
1391 ds->ds_phys->ds_prev_snap_obj;
1392 ds_next->ds_phys->ds_prev_snap_txg =
1393 ds->ds_phys->ds_prev_snap_txg;
1394 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1395 ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
1398 * Transfer to our deadlist (which will become next's
1399 * new deadlist) any entries from next's current
1400 * deadlist which were born before prev, and free the
1403 * XXX we're doing this long task with the config lock held
1405 while (bplist_iterate(&ds_next->ds_deadlist, &itor,
1407 if (bp.blk_birth <= ds->ds_phys->ds_prev_snap_txg) {
1408 VERIFY(0 == bplist_enqueue(&ds->ds_deadlist,
1410 if (ds_prev && !after_branch_point &&
1412 ds_prev->ds_phys->ds_prev_snap_txg) {
1413 ds_prev->ds_phys->ds_unique_bytes +=
1414 bp_get_dasize(dp->dp_spa, &bp);
1417 used += bp_get_dasize(dp->dp_spa, &bp);
1418 compressed += BP_GET_PSIZE(&bp);
1419 uncompressed += BP_GET_UCSIZE(&bp);
1420 /* XXX check return value? */
1421 (void) arc_free(zio, dp->dp_spa, tx->tx_txg,
1422 &bp, NULL, NULL, ARC_NOWAIT);
1426 /* free next's deadlist */
1427 bplist_close(&ds_next->ds_deadlist);
1428 bplist_destroy(mos, ds_next->ds_phys->ds_deadlist_obj, tx);
1430 /* set next's deadlist to our deadlist */
1431 ds_next->ds_phys->ds_deadlist_obj =
1432 ds->ds_phys->ds_deadlist_obj;
1433 VERIFY(0 == bplist_open(&ds_next->ds_deadlist, mos,
1434 ds_next->ds_phys->ds_deadlist_obj));
1435 ds->ds_phys->ds_deadlist_obj = 0;
1437 if (ds_next->ds_phys->ds_next_snap_obj != 0) {
1439 * Update next's unique to include blocks which
1440 * were previously shared by only this snapshot
1441 * and it. Those blocks will be born after the
1442 * prev snap and before this snap, and will have
1443 * died after the next snap and before the one
1444 * after that (ie. be on the snap after next's
1447 * XXX we're doing this long task with the
1450 dsl_dataset_t *ds_after_next;
1452 VERIFY(0 == dsl_dataset_open_obj(dp,
1453 ds_next->ds_phys->ds_next_snap_obj, NULL,
1454 DS_MODE_NONE, FTAG, &ds_after_next));
1456 while (bplist_iterate(&ds_after_next->ds_deadlist,
1459 ds->ds_phys->ds_prev_snap_txg &&
1461 ds->ds_phys->ds_creation_txg) {
1462 ds_next->ds_phys->ds_unique_bytes +=
1463 bp_get_dasize(dp->dp_spa, &bp);
1467 dsl_dataset_close(ds_after_next, DS_MODE_NONE, FTAG);
1468 ASSERT3P(ds_next->ds_prev, ==, NULL);
1470 ASSERT3P(ds_next->ds_prev, ==, ds);
1471 dsl_dataset_close(ds_next->ds_prev, DS_MODE_NONE,
1474 VERIFY(0 == dsl_dataset_open_obj(dp,
1475 ds->ds_phys->ds_prev_snap_obj, NULL,
1476 DS_MODE_NONE, ds_next, &ds_next->ds_prev));
1478 ds_next->ds_prev = NULL;
1481 dsl_dataset_recalc_head_uniq(ds_next);
1484 * Reduce the amount of our unconsmed refreservation
1485 * being charged to our parent by the amount of
1486 * new unique data we have gained.
1488 if (old_unique < ds_next->ds_reserved) {
1490 uint64_t new_unique =
1491 ds_next->ds_phys->ds_unique_bytes;
1493 ASSERT(old_unique <= new_unique);
1494 mrsdelta = MIN(new_unique - old_unique,
1495 ds_next->ds_reserved - old_unique);
1496 dsl_dir_diduse_space(ds->ds_dir, -mrsdelta,
1500 dsl_dataset_close(ds_next, DS_MODE_NONE, FTAG);
1503 * NB: unique_bytes might not be accurate for the head objset.
1504 * Before SPA_VERSION 9, we didn't update its value when we
1505 * deleted the most recent snapshot.
1507 ASSERT3U(used, ==, ds->ds_phys->ds_unique_bytes);
1510 * There's no next snapshot, so this is a head dataset.
1511 * Destroy the deadlist. Unless it's a clone, the
1512 * deadlist should be empty. (If it's a clone, it's
1513 * safe to ignore the deadlist contents.)
1517 ASSERT(after_branch_point || bplist_empty(&ds->ds_deadlist));
1518 bplist_close(&ds->ds_deadlist);
1519 bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx);
1520 ds->ds_phys->ds_deadlist_obj = 0;
1523 * Free everything that we point to (that's born after
1524 * the previous snapshot, if we are a clone)
1526 * XXX we're doing this long task with the config lock held
1529 ka.compressedp = &compressed;
1530 ka.uncompressedp = &uncompressed;
1533 err = traverse_dsl_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1534 ADVANCE_POST, kill_blkptr, &ka);
1535 ASSERT3U(err, ==, 0);
1536 ASSERT(spa_version(dp->dp_spa) <
1537 SPA_VERSION_UNIQUE_ACCURATE ||
1538 used == ds->ds_phys->ds_unique_bytes);
1541 err = zio_wait(zio);
1542 ASSERT3U(err, ==, 0);
1544 dsl_dir_diduse_space(ds->ds_dir, -used, -compressed, -uncompressed, tx);
1546 if (ds->ds_phys->ds_snapnames_zapobj) {
1547 err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx);
1551 if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) {
1552 /* Erase the link in the dataset */
1553 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1554 ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
1556 * dsl_dir_sync_destroy() called us, they'll destroy
1560 /* remove from snapshot namespace */
1561 dsl_dataset_t *ds_head;
1562 VERIFY(0 == dsl_dataset_open_obj(dp,
1563 ds->ds_dir->dd_phys->dd_head_dataset_obj, NULL,
1564 DS_MODE_NONE, FTAG, &ds_head));
1565 VERIFY(0 == dsl_dataset_get_snapname(ds));
1570 err = dsl_dataset_snap_lookup(mos,
1571 ds_head->ds_phys->ds_flags,
1572 ds_head->ds_phys->ds_snapnames_zapobj,
1573 ds->ds_snapname, &val);
1574 ASSERT3U(err, ==, 0);
1575 ASSERT3U(val, ==, obj);
1578 err = dsl_dataset_snap_remove(mos,
1579 ds_head->ds_phys->ds_flags,
1580 ds_head->ds_phys->ds_snapnames_zapobj,
1581 ds->ds_snapname, tx);
1583 dsl_dataset_close(ds_head, DS_MODE_NONE, FTAG);
1586 if (ds_prev && ds->ds_prev != ds_prev)
1587 dsl_dataset_close(ds_prev, DS_MODE_NONE, FTAG);
1589 spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
1590 spa_history_internal_log(LOG_DS_DESTROY, dp->dp_spa, tx,
1591 cr, "dataset = %llu", ds->ds_object);
1593 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, tag);
1594 VERIFY(0 == dmu_object_free(mos, obj, tx));
1599 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1603 if (!dmu_tx_is_syncing(tx))
1607 * If there's an fs-only reservation, any blocks that might become
1608 * owned by the snapshot dataset must be accommodated by space
1609 * outside of the reservation.
1611 asize = MIN(dsl_dataset_unique(ds), ds->ds_reserved);
1612 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, FALSE))
1616 * Propogate any reserved space for this snapshot to other
1617 * snapshot checks in this sync group.
1620 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1627 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
1629 dsl_dataset_t *ds = arg1;
1630 const char *snapname = arg2;
1631 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1636 * We don't allow multiple snapshots of the same txg. If there
1637 * is already one, try again.
1639 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
1643 * Check for conflicting name snapshot name.
1645 err = dsl_dataset_snap_lookup(mos, ds->ds_phys->ds_flags,
1646 ds->ds_phys->ds_snapnames_zapobj, snapname, &value);
1653 * Check that the dataset's name is not too long. Name consists
1654 * of the dataset's length + 1 for the @-sign + snapshot name's length
1656 if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
1657 return (ENAMETOOLONG);
1659 err = dsl_dataset_snapshot_reserve_space(ds, tx);
1663 ds->ds_trysnap_txg = tx->tx_txg;
1668 dsl_dataset_snapshot_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1670 dsl_dataset_t *ds = arg1;
1671 const char *snapname = arg2;
1672 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1674 dsl_dataset_phys_t *dsphys;
1676 objset_t *mos = dp->dp_meta_objset;
1679 spa_scrub_restart(dp->dp_spa, tx->tx_txg);
1680 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1682 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1683 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1684 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1685 dmu_buf_will_dirty(dbuf, tx);
1686 dsphys = dbuf->db_data;
1687 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1688 dsphys->ds_fsid_guid = unique_create();
1689 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1690 sizeof (dsphys->ds_guid));
1691 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
1692 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
1693 dsphys->ds_next_snap_obj = ds->ds_object;
1694 dsphys->ds_num_children = 1;
1695 dsphys->ds_creation_time = gethrestime_sec();
1696 dsphys->ds_creation_txg = tx->tx_txg;
1697 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
1698 dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes;
1699 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
1700 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
1701 dsphys->ds_flags = ds->ds_phys->ds_flags;
1702 dsphys->ds_bp = ds->ds_phys->ds_bp;
1703 dmu_buf_rele(dbuf, FTAG);
1705 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
1707 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
1709 ds->ds_prev->ds_phys->ds_num_children > 1);
1710 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1711 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1712 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1713 ds->ds_prev->ds_phys->ds_creation_txg);
1714 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
1719 * If we have a reference-reservation on this dataset, we will
1720 * need to increase the amount of refreservation being charged
1721 * since our unique space is going to zero.
1723 if (ds->ds_reserved) {
1724 int64_t add = MIN(dsl_dataset_unique(ds), ds->ds_reserved);
1725 dsl_dir_diduse_space(ds->ds_dir, add, 0, 0, tx);
1728 bplist_close(&ds->ds_deadlist);
1729 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1730 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
1731 ds->ds_phys->ds_prev_snap_obj = dsobj;
1732 ds->ds_phys->ds_prev_snap_txg = tx->tx_txg;
1733 ds->ds_phys->ds_unique_bytes = 0;
1734 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1735 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1736 ds->ds_phys->ds_deadlist_obj =
1737 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
1738 VERIFY(0 == bplist_open(&ds->ds_deadlist, mos,
1739 ds->ds_phys->ds_deadlist_obj));
1741 dprintf("snap '%s' -> obj %llu\n", snapname, dsobj);
1742 err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
1743 snapname, 8, 1, &dsobj, tx);
1747 dsl_dataset_close(ds->ds_prev, DS_MODE_NONE, ds);
1748 VERIFY(0 == dsl_dataset_open_obj(dp,
1749 ds->ds_phys->ds_prev_snap_obj, snapname,
1750 DS_MODE_NONE, ds, &ds->ds_prev));
1752 spa_history_internal_log(LOG_DS_SNAPSHOT, dp->dp_spa, tx, cr,
1753 "dataset = %llu", dsobj);
1757 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1759 ASSERT(dmu_tx_is_syncing(tx));
1760 ASSERT(ds->ds_user_ptr != NULL);
1761 ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
1764 * in case we had to change ds_fsid_guid when we opened it,
1767 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1768 ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
1770 dsl_dir_dirty(ds->ds_dir, tx);
1771 dmu_objset_sync(ds->ds_user_ptr, zio, tx);
1775 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1777 uint64_t refd, avail, uobjs, aobjs;
1779 dsl_dir_stats(ds->ds_dir, nv);
1781 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1782 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1783 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1785 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1786 ds->ds_phys->ds_creation_time);
1787 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1788 ds->ds_phys->ds_creation_txg);
1789 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1791 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1794 if (ds->ds_phys->ds_next_snap_obj) {
1796 * This is a snapshot; override the dd's space used with
1797 * our unique space and compression ratio.
1799 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1800 ds->ds_phys->ds_unique_bytes);
1801 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
1802 ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
1803 (ds->ds_phys->ds_uncompressed_bytes * 100 /
1804 ds->ds_phys->ds_compressed_bytes));
1809 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1811 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
1812 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
1813 stat->dds_guid = ds->ds_phys->ds_guid;
1814 if (ds->ds_phys->ds_next_snap_obj) {
1815 stat->dds_is_snapshot = B_TRUE;
1816 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
1819 /* clone origin is really a dsl_dir thing... */
1820 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
1821 if (ds->ds_dir->dd_phys->dd_origin_obj) {
1824 VERIFY(0 == dsl_dataset_open_obj(ds->ds_dir->dd_pool,
1825 ds->ds_dir->dd_phys->dd_origin_obj,
1826 NULL, DS_MODE_NONE, FTAG, &ods));
1827 dsl_dataset_name(ods, stat->dds_origin);
1828 dsl_dataset_close(ods, DS_MODE_NONE, FTAG);
1830 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
1834 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1836 return (ds->ds_fsid_guid);
1840 dsl_dataset_space(dsl_dataset_t *ds,
1841 uint64_t *refdbytesp, uint64_t *availbytesp,
1842 uint64_t *usedobjsp, uint64_t *availobjsp)
1844 *refdbytesp = ds->ds_phys->ds_used_bytes;
1845 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1846 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
1847 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
1848 if (ds->ds_quota != 0) {
1850 * Adjust available bytes according to refquota
1852 if (*refdbytesp < ds->ds_quota)
1853 *availbytesp = MIN(*availbytesp,
1854 ds->ds_quota - *refdbytesp);
1858 *usedobjsp = ds->ds_phys->ds_bp.blk_fill;
1859 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
1863 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
1865 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1867 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
1868 dsl_pool_sync_context(dp));
1869 if (ds->ds_prev == NULL)
1871 if (ds->ds_phys->ds_bp.blk_birth >
1872 ds->ds_prev->ds_phys->ds_creation_txg)
1879 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
1881 dsl_dataset_t *ds = arg1;
1882 char *newsnapname = arg2;
1883 dsl_dir_t *dd = ds->ds_dir;
1884 objset_t *mos = dd->dd_pool->dp_meta_objset;
1889 err = dsl_dataset_open_obj(dd->dd_pool,
1890 dd->dd_phys->dd_head_dataset_obj, NULL, DS_MODE_NONE, FTAG, &hds);
1894 /* new name better not be in use */
1895 err = dsl_dataset_snap_lookup(mos, hds->ds_phys->ds_flags,
1896 hds->ds_phys->ds_snapnames_zapobj, newsnapname, &val);
1897 dsl_dataset_close(hds, DS_MODE_NONE, FTAG);
1901 else if (err == ENOENT)
1904 /* dataset name + 1 for the "@" + the new snapshot name must fit */
1905 if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
1912 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2,
1913 cred_t *cr, dmu_tx_t *tx)
1915 dsl_dataset_t *ds = arg1;
1916 const char *newsnapname = arg2;
1917 dsl_dir_t *dd = ds->ds_dir;
1918 objset_t *mos = dd->dd_pool->dp_meta_objset;
1922 ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
1924 VERIFY(0 == dsl_dataset_open_obj(dd->dd_pool,
1925 dd->dd_phys->dd_head_dataset_obj, NULL, DS_MODE_NONE, FTAG, &hds));
1927 VERIFY(0 == dsl_dataset_get_snapname(ds));
1928 err = dsl_dataset_snap_remove(mos, hds->ds_phys->ds_flags,
1929 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname, tx);
1930 ASSERT3U(err, ==, 0);
1931 mutex_enter(&ds->ds_lock);
1932 (void) strcpy(ds->ds_snapname, newsnapname);
1933 mutex_exit(&ds->ds_lock);
1934 err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
1935 ds->ds_snapname, 8, 1, &ds->ds_object, tx);
1936 ASSERT3U(err, ==, 0);
1938 spa_history_internal_log(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx,
1939 cr, "dataset = %llu", ds->ds_object);
1940 dsl_dataset_close(hds, DS_MODE_NONE, FTAG);
1943 struct renamesnaparg {
1944 dsl_sync_task_group_t *dstg;
1945 char failed[MAXPATHLEN];
1951 dsl_snapshot_rename_one(char *name, void *arg)
1953 struct renamesnaparg *ra = arg;
1954 dsl_dataset_t *ds = NULL;
1958 cp = name + strlen(name);
1960 (void) strcpy(cp + 1, ra->oldsnap);
1963 * For recursive snapshot renames the parent won't be changing
1964 * so we just pass name for both the to/from argument.
1966 if (err = zfs_secpolicy_rename_perms(name, name, CRED())) {
1967 (void) strcpy(ra->failed, name);
1971 err = dsl_dataset_open(name, DS_MODE_READONLY | DS_MODE_STANDARD,
1973 if (err == ENOENT) {
1978 (void) strcpy(ra->failed, name);
1980 dsl_dataset_close(ds, DS_MODE_STANDARD, ra->dstg);
1985 /* for all filesystems undergoing rename, we'll need to unmount it */
1986 (void) zfs_unmount_snap(name, NULL);
1991 dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
1992 dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
1998 dsl_recursive_rename(char *oldname, const char *newname)
2001 struct renamesnaparg *ra;
2002 dsl_sync_task_t *dst;
2004 char *cp, *fsname = spa_strdup(oldname);
2005 int len = strlen(oldname);
2007 /* truncate the snapshot name to get the fsname */
2008 cp = strchr(fsname, '@');
2011 err = spa_open(fsname, &spa, FTAG);
2013 kmem_free(fsname, len + 1);
2016 ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP);
2017 ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
2019 ra->oldsnap = strchr(oldname, '@') + 1;
2020 ra->newsnap = strchr(newname, '@') + 1;
2023 err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
2025 kmem_free(fsname, len + 1);
2028 err = dsl_sync_task_group_wait(ra->dstg);
2031 for (dst = list_head(&ra->dstg->dstg_tasks); dst;
2032 dst = list_next(&ra->dstg->dstg_tasks, dst)) {
2033 dsl_dataset_t *ds = dst->dst_arg1;
2035 dsl_dir_name(ds->ds_dir, ra->failed);
2036 (void) strcat(ra->failed, "@");
2037 (void) strcat(ra->failed, ra->newsnap);
2039 dsl_dataset_close(ds, DS_MODE_STANDARD, ra->dstg);
2043 (void) strcpy(oldname, ra->failed);
2045 dsl_sync_task_group_destroy(ra->dstg);
2046 kmem_free(ra, sizeof (struct renamesnaparg));
2047 spa_close(spa, FTAG);
2052 dsl_valid_rename(char *oldname, void *arg)
2054 int delta = *(int *)arg;
2056 if (strlen(oldname) + delta >= MAXNAMELEN)
2057 return (ENAMETOOLONG);
2062 #pragma weak dmu_objset_rename = dsl_dataset_rename
2064 dsl_dataset_rename(char *oldname, const char *newname,
2065 boolean_t recursive)
2072 err = dsl_dir_open(oldname, FTAG, &dd, &tail);
2076 int delta = strlen(newname) - strlen(oldname);
2078 /* if we're growing, validate child size lengths */
2080 err = dmu_objset_find(oldname, dsl_valid_rename,
2081 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2084 err = dsl_dir_rename(dd, newname);
2085 dsl_dir_close(dd, FTAG);
2088 if (tail[0] != '@') {
2089 /* the name ended in a nonexistant component */
2090 dsl_dir_close(dd, FTAG);
2094 dsl_dir_close(dd, FTAG);
2096 /* new name must be snapshot in same filesystem */
2097 tail = strchr(newname, '@');
2101 if (strncmp(oldname, newname, tail - newname) != 0)
2105 err = dsl_recursive_rename(oldname, newname);
2107 err = dsl_dataset_open(oldname,
2108 DS_MODE_READONLY | DS_MODE_STANDARD, FTAG, &ds);
2112 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2113 dsl_dataset_snapshot_rename_check,
2114 dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
2116 dsl_dataset_close(ds, DS_MODE_STANDARD, FTAG);
2123 uint64_t used, comp, uncomp, unique;
2124 uint64_t ds_flags, newnext_obj, snapnames_obj;
2129 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
2131 dsl_dataset_t *hds = arg1;
2132 struct promotearg *pa = arg2;
2133 dsl_dir_t *dd = hds->ds_dir;
2134 dsl_pool_t *dp = hds->ds_dir->dd_pool;
2135 dsl_dir_t *odd = NULL;
2136 dsl_dataset_t *ds = NULL;
2137 dsl_dataset_t *origin_ds = NULL;
2138 dsl_dataset_t *newnext_ds = NULL;
2144 bzero(pa, sizeof (*pa));
2146 /* Check that it is a clone */
2147 if (dd->dd_phys->dd_origin_obj == 0)
2150 /* Since this is so expensive, don't do the preliminary check */
2151 if (!dmu_tx_is_syncing(tx))
2154 if (err = dsl_dataset_open_obj(dp, dd->dd_phys->dd_origin_obj,
2155 NULL, DS_MODE_EXCLUSIVE, FTAG, &origin_ds))
2157 odd = origin_ds->ds_dir;
2160 dsl_dataset_t *phds;
2161 if (err = dsl_dataset_open_obj(dd->dd_pool,
2162 odd->dd_phys->dd_head_dataset_obj,
2163 NULL, DS_MODE_NONE, FTAG, &phds))
2165 pa->ds_flags = phds->ds_phys->ds_flags;
2166 pa->snapnames_obj = phds->ds_phys->ds_snapnames_zapobj;
2167 dsl_dataset_close(phds, DS_MODE_NONE, FTAG);
2170 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) {
2175 /* find origin's new next ds */
2176 VERIFY(0 == dsl_dataset_open_obj(dd->dd_pool, hds->ds_object,
2177 NULL, DS_MODE_NONE, FTAG, &newnext_ds));
2178 while (newnext_ds->ds_phys->ds_prev_snap_obj != origin_ds->ds_object) {
2179 dsl_dataset_t *prev;
2181 if (err = dsl_dataset_open_obj(dd->dd_pool,
2182 newnext_ds->ds_phys->ds_prev_snap_obj,
2183 NULL, DS_MODE_NONE, FTAG, &prev))
2185 dsl_dataset_close(newnext_ds, DS_MODE_NONE, FTAG);
2188 pa->newnext_obj = newnext_ds->ds_object;
2190 /* compute origin's new unique space */
2191 while ((err = bplist_iterate(&newnext_ds->ds_deadlist,
2192 &itor, &bp)) == 0) {
2193 if (bp.blk_birth > origin_ds->ds_phys->ds_prev_snap_txg)
2194 pa->unique += bp_get_dasize(dd->dd_pool->dp_spa, &bp);
2199 /* Walk the snapshots that we are moving */
2200 name = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2204 uint64_t val, dlused, dlcomp, dluncomp;
2205 dsl_dataset_t *prev;
2207 /* Check that the snapshot name does not conflict */
2208 dsl_dataset_name(ds, name);
2209 err = dsl_dataset_snap_lookup(dd->dd_pool->dp_meta_objset,
2210 hds->ds_phys->ds_flags, hds->ds_phys->ds_snapnames_zapobj,
2211 ds->ds_snapname, &val);
2212 if (err != ENOENT) {
2219 * compute space to transfer. Each snapshot gave birth to:
2220 * (my used) - (prev's used) + (deadlist's used)
2222 pa->used += ds->ds_phys->ds_used_bytes;
2223 pa->comp += ds->ds_phys->ds_compressed_bytes;
2224 pa->uncomp += ds->ds_phys->ds_uncompressed_bytes;
2226 /* If we reach the first snapshot, we're done. */
2227 if (ds->ds_phys->ds_prev_snap_obj == 0)
2230 if (err = bplist_space(&ds->ds_deadlist,
2231 &dlused, &dlcomp, &dluncomp))
2233 if (err = dsl_dataset_open_obj(dd->dd_pool,
2234 ds->ds_phys->ds_prev_snap_obj, NULL, DS_MODE_EXCLUSIVE,
2237 pa->used += dlused - prev->ds_phys->ds_used_bytes;
2238 pa->comp += dlcomp - prev->ds_phys->ds_compressed_bytes;
2239 pa->uncomp += dluncomp - prev->ds_phys->ds_uncompressed_bytes;
2242 * We could be a clone of a clone. If we reach our
2243 * parent's branch point, we're done.
2245 if (prev->ds_phys->ds_next_snap_obj != ds->ds_object) {
2246 dsl_dataset_close(prev, DS_MODE_EXCLUSIVE, FTAG);
2249 if (ds != origin_ds)
2250 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG);
2254 /* Check that there is enough space here */
2255 err = dsl_dir_transfer_possible(odd, dd, pa->used);
2258 if (ds && ds != origin_ds)
2259 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG);
2261 dsl_dataset_close(origin_ds, DS_MODE_EXCLUSIVE, FTAG);
2263 dsl_dataset_close(newnext_ds, DS_MODE_NONE, FTAG);
2265 kmem_free(name, MAXPATHLEN);
2270 dsl_dataset_promote_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2272 dsl_dataset_t *hds = arg1;
2273 struct promotearg *pa = arg2;
2274 dsl_dir_t *dd = hds->ds_dir;
2275 dsl_pool_t *dp = hds->ds_dir->dd_pool;
2276 dsl_dir_t *odd = NULL;
2277 dsl_dataset_t *ds, *origin_ds;
2280 ASSERT(dd->dd_phys->dd_origin_obj != 0);
2281 ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
2283 VERIFY(0 == dsl_dataset_open_obj(dp, dd->dd_phys->dd_origin_obj,
2284 NULL, DS_MODE_EXCLUSIVE, FTAG, &origin_ds));
2286 * We need to explicitly open odd, since origin_ds's dd will be
2289 VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object,
2292 /* move snapshots to this dir */
2293 name = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2297 dsl_dataset_t *prev;
2299 /* move snap name entry */
2300 dsl_dataset_name(ds, name);
2301 VERIFY(0 == dsl_dataset_snap_remove(dp->dp_meta_objset,
2302 pa->ds_flags, pa->snapnames_obj, ds->ds_snapname, tx));
2303 VERIFY(0 == zap_add(dp->dp_meta_objset,
2304 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2305 8, 1, &ds->ds_object, tx));
2307 /* change containing dsl_dir */
2308 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2309 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2310 ds->ds_phys->ds_dir_obj = dd->dd_object;
2311 ASSERT3P(ds->ds_dir, ==, odd);
2312 dsl_dir_close(ds->ds_dir, ds);
2313 VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
2314 NULL, ds, &ds->ds_dir));
2316 ASSERT3U(dsl_prop_numcb(ds), ==, 0);
2318 if (ds->ds_phys->ds_prev_snap_obj == 0)
2321 VERIFY(0 == dsl_dataset_open_obj(dp,
2322 ds->ds_phys->ds_prev_snap_obj, NULL, DS_MODE_EXCLUSIVE,
2325 if (prev->ds_phys->ds_next_snap_obj != ds->ds_object) {
2326 dsl_dataset_close(prev, DS_MODE_EXCLUSIVE, FTAG);
2329 if (ds != origin_ds)
2330 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG);
2333 if (ds != origin_ds)
2334 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG);
2336 /* change origin's next snap */
2337 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2338 origin_ds->ds_phys->ds_next_snap_obj = pa->newnext_obj;
2341 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2342 ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2343 dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2344 dmu_buf_will_dirty(odd->dd_dbuf, tx);
2345 odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2347 /* change space accounting */
2348 dsl_dir_diduse_space(odd, -pa->used, -pa->comp, -pa->uncomp, tx);
2349 dsl_dir_diduse_space(dd, pa->used, pa->comp, pa->uncomp, tx);
2350 origin_ds->ds_phys->ds_unique_bytes = pa->unique;
2352 /* log history record */
2353 spa_history_internal_log(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx,
2354 cr, "dataset = %llu", ds->ds_object);
2356 dsl_dir_close(odd, FTAG);
2357 dsl_dataset_close(origin_ds, DS_MODE_EXCLUSIVE, FTAG);
2358 kmem_free(name, MAXPATHLEN);
2362 dsl_dataset_promote(const char *name)
2366 dmu_object_info_t doi;
2367 struct promotearg pa;
2369 err = dsl_dataset_open(name, DS_MODE_NONE, FTAG, &ds);
2373 err = dmu_object_info(ds->ds_dir->dd_pool->dp_meta_objset,
2374 ds->ds_phys->ds_snapnames_zapobj, &doi);
2376 dsl_dataset_close(ds, DS_MODE_NONE, FTAG);
2381 * Add in 128x the snapnames zapobj size, since we will be moving
2382 * a bunch of snapnames to the promoted ds, and dirtying their
2385 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2386 dsl_dataset_promote_check,
2387 dsl_dataset_promote_sync, ds, &pa, 2 + 2 * doi.doi_physical_blks);
2388 dsl_dataset_close(ds, DS_MODE_NONE, FTAG);
2392 struct cloneswaparg {
2393 dsl_dataset_t *cds; /* clone dataset */
2394 dsl_dataset_t *ohds; /* origin's head dataset */
2396 int64_t unused_refres_delta; /* change in unconsumed refreservation */
2401 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx)
2403 struct cloneswaparg *csa = arg1;
2405 /* they should both be heads */
2406 if (dsl_dataset_is_snapshot(csa->cds) ||
2407 dsl_dataset_is_snapshot(csa->ohds))
2410 /* the branch point should be just before them */
2411 if (csa->cds->ds_prev != csa->ohds->ds_prev)
2414 /* cds should be the clone */
2415 if (csa->cds->ds_prev->ds_phys->ds_next_snap_obj !=
2416 csa->ohds->ds_object)
2419 /* the clone should be a child of the origin */
2420 if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir)
2423 /* ohds shouldn't be modified unless 'force' */
2424 if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds))
2427 /* adjust amount of any unconsumed refreservation */
2428 csa->unused_refres_delta =
2429 (int64_t)MIN(csa->ohds->ds_reserved,
2430 csa->ohds->ds_phys->ds_unique_bytes) -
2431 (int64_t)MIN(csa->ohds->ds_reserved,
2432 csa->cds->ds_phys->ds_unique_bytes);
2434 if (csa->unused_refres_delta > 0 &&
2435 csa->unused_refres_delta >
2436 dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE))
2444 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2446 struct cloneswaparg *csa = arg1;
2447 dsl_pool_t *dp = csa->cds->ds_dir->dd_pool;
2450 uint64_t unique = 0;
2453 ASSERT(csa->cds->ds_reserved == 0);
2454 ASSERT(csa->cds->ds_quota == csa->ohds->ds_quota);
2456 dmu_buf_will_dirty(csa->cds->ds_dbuf, tx);
2457 dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx);
2458 dmu_buf_will_dirty(csa->cds->ds_prev->ds_dbuf, tx);
2460 if (csa->cds->ds_user_ptr != NULL) {
2461 csa->cds->ds_user_evict_func(csa->cds, csa->cds->ds_user_ptr);
2462 csa->cds->ds_user_ptr = NULL;
2465 if (csa->ohds->ds_user_ptr != NULL) {
2466 csa->ohds->ds_user_evict_func(csa->ohds,
2467 csa->ohds->ds_user_ptr);
2468 csa->ohds->ds_user_ptr = NULL;
2471 /* compute unique space */
2472 while ((err = bplist_iterate(&csa->cds->ds_deadlist,
2473 &itor, &bp)) == 0) {
2474 if (bp.blk_birth > csa->cds->ds_prev->ds_phys->ds_prev_snap_txg)
2475 unique += bp_get_dasize(dp->dp_spa, &bp);
2477 VERIFY(err == ENOENT);
2479 /* reset origin's unique bytes */
2480 csa->cds->ds_prev->ds_phys->ds_unique_bytes = unique;
2485 tmp = csa->ohds->ds_phys->ds_bp;
2486 csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp;
2487 csa->cds->ds_phys->ds_bp = tmp;
2490 /* set dd_*_bytes */
2492 int64_t dused, dcomp, duncomp;
2493 uint64_t cdl_used, cdl_comp, cdl_uncomp;
2494 uint64_t odl_used, odl_comp, odl_uncomp;
2496 VERIFY(0 == bplist_space(&csa->cds->ds_deadlist, &cdl_used,
2497 &cdl_comp, &cdl_uncomp));
2498 VERIFY(0 == bplist_space(&csa->ohds->ds_deadlist, &odl_used,
2499 &odl_comp, &odl_uncomp));
2500 dused = csa->cds->ds_phys->ds_used_bytes + cdl_used -
2501 (csa->ohds->ds_phys->ds_used_bytes + odl_used);
2502 dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp -
2503 (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp);
2504 duncomp = csa->cds->ds_phys->ds_uncompressed_bytes +
2506 (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp);
2508 dsl_dir_diduse_space(csa->ohds->ds_dir,
2509 dused, dcomp, duncomp, tx);
2510 dsl_dir_diduse_space(csa->cds->ds_dir,
2511 -dused, -dcomp, -duncomp, tx);
2514 #define SWITCH64(x, y) \
2516 uint64_t __tmp = (x); \
2521 /* swap ds_*_bytes */
2522 SWITCH64(csa->ohds->ds_phys->ds_used_bytes,
2523 csa->cds->ds_phys->ds_used_bytes);
2524 SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes,
2525 csa->cds->ds_phys->ds_compressed_bytes);
2526 SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes,
2527 csa->cds->ds_phys->ds_uncompressed_bytes);
2528 SWITCH64(csa->ohds->ds_phys->ds_unique_bytes,
2529 csa->cds->ds_phys->ds_unique_bytes);
2531 /* apply any parent delta for change in unconsumed refreservation */
2532 dsl_dir_diduse_space(csa->ohds->ds_dir, csa->unused_refres_delta,
2535 /* swap deadlists */
2536 bplist_close(&csa->cds->ds_deadlist);
2537 bplist_close(&csa->ohds->ds_deadlist);
2538 SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj,
2539 csa->cds->ds_phys->ds_deadlist_obj);
2540 VERIFY(0 == bplist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset,
2541 csa->cds->ds_phys->ds_deadlist_obj));
2542 VERIFY(0 == bplist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset,
2543 csa->ohds->ds_phys->ds_deadlist_obj));
2547 * Swap 'clone' with its origin head file system.
2550 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
2553 struct cloneswaparg csa;
2555 ASSERT(clone->ds_open_refcount == DS_REF_MAX);
2556 ASSERT(origin_head->ds_open_refcount == DS_REF_MAX);
2559 csa.ohds = origin_head;
2561 return (dsl_sync_task_do(clone->ds_dir->dd_pool,
2562 dsl_dataset_clone_swap_check,
2563 dsl_dataset_clone_swap_sync, &csa, NULL, 9));
2567 * Given a pool name and a dataset object number in that pool,
2568 * return the name of that dataset.
2571 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2575 dsl_dataset_t *ds = NULL;
2578 if ((error = spa_open(pname, &spa, FTAG)) != 0)
2580 dp = spa_get_dsl(spa);
2581 rw_enter(&dp->dp_config_rwlock, RW_READER);
2582 if ((error = dsl_dataset_open_obj(dp, obj,
2583 NULL, DS_MODE_NONE, FTAG, &ds)) != 0) {
2584 rw_exit(&dp->dp_config_rwlock);
2585 spa_close(spa, FTAG);
2588 dsl_dataset_name(ds, buf);
2589 dsl_dataset_close(ds, DS_MODE_NONE, FTAG);
2590 rw_exit(&dp->dp_config_rwlock);
2591 spa_close(spa, FTAG);
2597 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
2598 uint64_t asize, uint64_t inflight, uint64_t *used,
2603 ASSERT3S(asize, >, 0);
2606 * *ref_rsrv is the portion of asize that will come from any
2607 * unconsumed refreservation space.
2611 mutex_enter(&ds->ds_lock);
2613 * Make a space adjustment for reserved bytes.
2615 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
2617 ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2618 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2620 asize - MIN(asize, parent_delta(ds, asize + inflight));
2623 if (!check_quota || ds->ds_quota == 0) {
2624 mutex_exit(&ds->ds_lock);
2628 * If they are requesting more space, and our current estimate
2629 * is over quota, they get to try again unless the actual
2630 * on-disk is over quota and there are no pending changes (which
2631 * may free up space for us).
2633 if (ds->ds_phys->ds_used_bytes + inflight >= ds->ds_quota) {
2634 if (inflight > 0 || ds->ds_phys->ds_used_bytes < ds->ds_quota)
2639 mutex_exit(&ds->ds_lock);
2646 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
2648 dsl_dataset_t *ds = arg1;
2649 uint64_t *quotap = arg2;
2650 uint64_t new_quota = *quotap;
2652 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA)
2658 if (new_quota < ds->ds_phys->ds_used_bytes ||
2659 new_quota < ds->ds_reserved)
2667 dsl_dataset_set_quota_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2669 dsl_dataset_t *ds = arg1;
2670 uint64_t *quotap = arg2;
2671 uint64_t new_quota = *quotap;
2673 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2675 mutex_enter(&ds->ds_lock);
2676 ds->ds_quota = new_quota;
2677 mutex_exit(&ds->ds_lock);
2679 dsl_prop_set_uint64_sync(ds->ds_dir, "refquota", new_quota, cr, tx);
2681 spa_history_internal_log(LOG_DS_REFQUOTA, ds->ds_dir->dd_pool->dp_spa,
2682 tx, cr, "%lld dataset = %llu ",
2683 (longlong_t)new_quota, ds->ds_dir->dd_phys->dd_head_dataset_obj);
2687 dsl_dataset_set_quota(const char *dsname, uint64_t quota)
2692 err = dsl_dataset_open(dsname, DS_MODE_STANDARD, FTAG, &ds);
2696 if (quota != ds->ds_quota) {
2698 * If someone removes a file, then tries to set the quota, we
2699 * want to make sure the file freeing takes effect.
2701 txg_wait_open(ds->ds_dir->dd_pool, 0);
2703 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2704 dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync,
2707 dsl_dataset_close(ds, DS_MODE_STANDARD, FTAG);
2712 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
2714 dsl_dataset_t *ds = arg1;
2715 uint64_t *reservationp = arg2;
2716 uint64_t new_reservation = *reservationp;
2720 if (new_reservation > INT64_MAX)
2723 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
2724 SPA_VERSION_REFRESERVATION)
2727 if (dsl_dataset_is_snapshot(ds))
2731 * If we are doing the preliminary check in open context, the
2732 * space estimates may be inaccurate.
2734 if (!dmu_tx_is_syncing(tx))
2737 mutex_enter(&ds->ds_lock);
2738 unique = dsl_dataset_unique(ds);
2739 delta = MAX(unique, new_reservation) - MAX(unique, ds->ds_reserved);
2740 mutex_exit(&ds->ds_lock);
2743 delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
2745 if (delta > 0 && ds->ds_quota > 0 &&
2746 new_reservation > ds->ds_quota)
2754 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, cred_t *cr,
2757 dsl_dataset_t *ds = arg1;
2758 uint64_t *reservationp = arg2;
2759 uint64_t new_reservation = *reservationp;
2763 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2765 mutex_enter(&ds->ds_lock);
2766 unique = dsl_dataset_unique(ds);
2767 delta = MAX(0, (int64_t)(new_reservation - unique)) -
2768 MAX(0, (int64_t)(ds->ds_reserved - unique));
2769 ds->ds_reserved = new_reservation;
2770 mutex_exit(&ds->ds_lock);
2772 dsl_prop_set_uint64_sync(ds->ds_dir, "refreservation",
2773 new_reservation, cr, tx);
2775 dsl_dir_diduse_space(ds->ds_dir, delta, 0, 0, tx);
2777 spa_history_internal_log(LOG_DS_REFRESERV,
2778 ds->ds_dir->dd_pool->dp_spa, tx, cr, "%lld dataset = %llu",
2779 (longlong_t)new_reservation,
2780 ds->ds_dir->dd_phys->dd_head_dataset_obj);
2784 dsl_dataset_set_reservation(const char *dsname, uint64_t reservation)
2789 err = dsl_dataset_open(dsname, DS_MODE_STANDARD, FTAG, &ds);
2793 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2794 dsl_dataset_set_reservation_check,
2795 dsl_dataset_set_reservation_sync, ds, &reservation, 0);
2796 dsl_dataset_close(ds, DS_MODE_STANDARD, FTAG);