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 2007 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
26 #pragma ident "%Z%%M% %I% %E% 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 static dsl_checkfunc_t dsl_dataset_destroy_begin_check;
43 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync;
44 static dsl_checkfunc_t dsl_dataset_rollback_check;
45 static dsl_syncfunc_t dsl_dataset_rollback_sync;
46 static dsl_checkfunc_t dsl_dataset_destroy_check;
47 static dsl_syncfunc_t dsl_dataset_destroy_sync;
49 #define DS_REF_MAX (1ULL << 62)
51 #define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE
54 * We use weighted reference counts to express the various forms of exclusion
55 * between different open modes. A STANDARD open is 1 point, an EXCLUSIVE open
56 * is DS_REF_MAX, and a PRIMARY open is little more than half of an EXCLUSIVE.
57 * This makes the exclusion logic simple: the total refcnt for all opens cannot
58 * exceed DS_REF_MAX. For example, EXCLUSIVE opens are exclusive because their
59 * weight (DS_REF_MAX) consumes the entire refcnt space. PRIMARY opens consume
60 * just over half of the refcnt space, so there can't be more than one, but it
61 * can peacefully coexist with any number of STANDARD opens.
63 static uint64_t ds_refcnt_weight[DS_MODE_LEVELS] = {
64 0, /* DS_MODE_NONE - invalid */
65 1, /* DS_MODE_STANDARD - unlimited number */
66 (DS_REF_MAX >> 1) + 1, /* DS_MODE_PRIMARY - only one of these */
67 DS_REF_MAX /* DS_MODE_EXCLUSIVE - no other opens */
72 dsl_dataset_block_born(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
74 int used = bp_get_dasize(tx->tx_pool->dp_spa, bp);
75 int compressed = BP_GET_PSIZE(bp);
76 int uncompressed = BP_GET_UCSIZE(bp);
78 dprintf_bp(bp, "born, ds=%p\n", ds);
80 ASSERT(dmu_tx_is_syncing(tx));
81 /* It could have been compressed away to nothing */
84 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
85 ASSERT3U(BP_GET_TYPE(bp), <, DMU_OT_NUMTYPES);
88 * Account for the meta-objset space in its placeholder
91 ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */
92 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir,
93 used, compressed, uncompressed, tx);
94 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
97 dmu_buf_will_dirty(ds->ds_dbuf, tx);
98 mutex_enter(&ds->ds_lock);
99 ds->ds_phys->ds_used_bytes += used;
100 ds->ds_phys->ds_compressed_bytes += compressed;
101 ds->ds_phys->ds_uncompressed_bytes += uncompressed;
102 ds->ds_phys->ds_unique_bytes += used;
103 mutex_exit(&ds->ds_lock);
104 dsl_dir_diduse_space(ds->ds_dir,
105 used, compressed, uncompressed, tx);
109 dsl_dataset_block_kill(dsl_dataset_t *ds, blkptr_t *bp, zio_t *pio,
112 int used = bp_get_dasize(tx->tx_pool->dp_spa, bp);
113 int compressed = BP_GET_PSIZE(bp);
114 int uncompressed = BP_GET_UCSIZE(bp);
116 ASSERT(dmu_tx_is_syncing(tx));
117 /* No block pointer => nothing to free */
125 * Account for the meta-objset space in its placeholder
128 err = arc_free(pio, tx->tx_pool->dp_spa,
129 tx->tx_txg, bp, NULL, NULL, pio ? ARC_NOWAIT: ARC_WAIT);
132 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir,
133 -used, -compressed, -uncompressed, tx);
134 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
137 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
139 dmu_buf_will_dirty(ds->ds_dbuf, tx);
141 if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
144 dprintf_bp(bp, "freeing: %s", "");
145 err = arc_free(pio, tx->tx_pool->dp_spa,
146 tx->tx_txg, bp, NULL, NULL, pio ? ARC_NOWAIT: ARC_WAIT);
149 mutex_enter(&ds->ds_lock);
150 /* XXX unique_bytes is not accurate for head datasets */
151 /* ASSERT3U(ds->ds_phys->ds_unique_bytes, >=, used); */
152 ds->ds_phys->ds_unique_bytes -= used;
153 mutex_exit(&ds->ds_lock);
154 dsl_dir_diduse_space(ds->ds_dir,
155 -used, -compressed, -uncompressed, tx);
157 dprintf_bp(bp, "putting on dead list: %s", "");
158 VERIFY(0 == bplist_enqueue(&ds->ds_deadlist, bp, tx));
159 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
160 if (ds->ds_phys->ds_prev_snap_obj != 0) {
161 ASSERT3U(ds->ds_prev->ds_object, ==,
162 ds->ds_phys->ds_prev_snap_obj);
163 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
164 if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
165 ds->ds_object && bp->blk_birth >
166 ds->ds_prev->ds_phys->ds_prev_snap_txg) {
167 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
168 mutex_enter(&ds->ds_prev->ds_lock);
169 ds->ds_prev->ds_phys->ds_unique_bytes +=
171 mutex_exit(&ds->ds_prev->ds_lock);
175 mutex_enter(&ds->ds_lock);
176 ASSERT3U(ds->ds_phys->ds_used_bytes, >=, used);
177 ds->ds_phys->ds_used_bytes -= used;
178 ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
179 ds->ds_phys->ds_compressed_bytes -= compressed;
180 ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
181 ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
182 mutex_exit(&ds->ds_lock);
186 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
188 uint64_t trysnap = 0;
193 * The snapshot creation could fail, but that would cause an
194 * incorrect FALSE return, which would only result in an
195 * overestimation of the amount of space that an operation would
196 * consume, which is OK.
198 * There's also a small window where we could miss a pending
199 * snapshot, because we could set the sync task in the quiescing
200 * phase. So this should only be used as a guess.
202 if (ds->ds_trysnap_txg >
203 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
204 trysnap = ds->ds_trysnap_txg;
205 return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
209 dsl_dataset_block_freeable(dsl_dataset_t *ds, uint64_t blk_birth)
211 return (blk_birth > dsl_dataset_prev_snap_txg(ds));
216 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
218 dsl_dataset_t *ds = dsv;
219 dsl_pool_t *dp = ds->ds_dir->dd_pool;
221 /* open_refcount == DS_REF_MAX when deleting */
222 ASSERT(ds->ds_open_refcount == 0 ||
223 ds->ds_open_refcount == DS_REF_MAX);
225 dprintf_ds(ds, "evicting %s\n", "");
227 unique_remove(ds->ds_phys->ds_fsid_guid);
229 if (ds->ds_user_ptr != NULL)
230 ds->ds_user_evict_func(ds, ds->ds_user_ptr);
233 dsl_dataset_close(ds->ds_prev, DS_MODE_NONE, ds);
237 bplist_close(&ds->ds_deadlist);
238 dsl_dir_close(ds->ds_dir, ds);
240 if (list_link_active(&ds->ds_synced_link))
241 list_remove(&dp->dp_synced_objsets, ds);
243 mutex_destroy(&ds->ds_lock);
244 mutex_destroy(&ds->ds_deadlist.bpl_lock);
246 kmem_free(ds, sizeof (dsl_dataset_t));
250 dsl_dataset_get_snapname(dsl_dataset_t *ds)
252 dsl_dataset_phys_t *headphys;
255 dsl_pool_t *dp = ds->ds_dir->dd_pool;
256 objset_t *mos = dp->dp_meta_objset;
258 if (ds->ds_snapname[0])
260 if (ds->ds_phys->ds_next_snap_obj == 0)
263 err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
267 headphys = headdbuf->db_data;
268 err = zap_value_search(dp->dp_meta_objset,
269 headphys->ds_snapnames_zapobj, ds->ds_object, ds->ds_snapname);
270 dmu_buf_rele(headdbuf, FTAG);
275 dsl_dataset_open_obj(dsl_pool_t *dp, uint64_t dsobj, const char *snapname,
276 int mode, void *tag, dsl_dataset_t **dsp)
278 uint64_t weight = ds_refcnt_weight[DS_MODE_LEVEL(mode)];
279 objset_t *mos = dp->dp_meta_objset;
284 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
285 dsl_pool_sync_context(dp));
287 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
290 ds = dmu_buf_get_user(dbuf);
292 dsl_dataset_t *winner;
294 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
296 ds->ds_object = dsobj;
297 ds->ds_phys = dbuf->db_data;
299 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
300 mutex_init(&ds->ds_deadlist.bpl_lock, NULL, MUTEX_DEFAULT,
303 err = bplist_open(&ds->ds_deadlist,
304 mos, ds->ds_phys->ds_deadlist_obj);
306 err = dsl_dir_open_obj(dp,
307 ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
311 * we don't really need to close the blist if we
314 mutex_destroy(&ds->ds_lock);
315 mutex_destroy(&ds->ds_deadlist.bpl_lock);
316 kmem_free(ds, sizeof (dsl_dataset_t));
317 dmu_buf_rele(dbuf, tag);
321 if (ds->ds_dir->dd_phys->dd_head_dataset_obj == dsobj) {
322 ds->ds_snapname[0] = '\0';
323 if (ds->ds_phys->ds_prev_snap_obj) {
324 err = dsl_dataset_open_obj(dp,
325 ds->ds_phys->ds_prev_snap_obj, NULL,
326 DS_MODE_NONE, ds, &ds->ds_prev);
331 dsl_dataset_phys_t *headphys;
333 err = dmu_bonus_hold(mos,
334 ds->ds_dir->dd_phys->dd_head_dataset_obj,
337 headphys = headdbuf->db_data;
339 err = zap_lookup(dp->dp_meta_objset,
340 headphys->ds_snapnames_zapobj,
341 snapname, sizeof (foundobj), 1,
343 ASSERT3U(foundobj, ==, dsobj);
344 dmu_buf_rele(headdbuf, FTAG);
347 (void) strcat(ds->ds_snapname, snapname);
348 } else if (zfs_flags & ZFS_DEBUG_SNAPNAMES) {
349 err = dsl_dataset_get_snapname(ds);
354 winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys,
358 bplist_close(&ds->ds_deadlist);
360 dsl_dataset_close(ds->ds_prev,
363 dsl_dir_close(ds->ds_dir, ds);
364 mutex_destroy(&ds->ds_lock);
365 mutex_destroy(&ds->ds_deadlist.bpl_lock);
366 kmem_free(ds, sizeof (dsl_dataset_t));
368 dmu_buf_rele(dbuf, tag);
374 unique_insert(ds->ds_phys->ds_fsid_guid);
375 if (new != ds->ds_phys->ds_fsid_guid) {
376 /* XXX it won't necessarily be synced... */
377 ds->ds_phys->ds_fsid_guid = new;
381 ASSERT3P(ds->ds_dbuf, ==, dbuf);
382 ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
384 mutex_enter(&ds->ds_lock);
385 if ((DS_MODE_LEVEL(mode) == DS_MODE_PRIMARY &&
386 (ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT) &&
387 !DS_MODE_IS_INCONSISTENT(mode)) ||
388 (ds->ds_open_refcount + weight > DS_REF_MAX)) {
389 mutex_exit(&ds->ds_lock);
390 dsl_dataset_close(ds, DS_MODE_NONE, tag);
393 ds->ds_open_refcount += weight;
394 mutex_exit(&ds->ds_lock);
401 dsl_dataset_open_spa(spa_t *spa, const char *name, int mode,
402 void *tag, dsl_dataset_t **dsp)
408 dsl_dataset_t *ds = NULL;
411 err = dsl_dir_open_spa(spa, name, FTAG, &dd, &tail);
416 obj = dd->dd_phys->dd_head_dataset_obj;
417 rw_enter(&dp->dp_config_rwlock, RW_READER);
419 /* A dataset with no associated objset */
425 objset_t *mos = dp->dp_meta_objset;
427 err = dsl_dataset_open_obj(dp, obj, NULL,
428 DS_MODE_NONE, tag, &ds);
431 obj = ds->ds_phys->ds_snapnames_zapobj;
432 dsl_dataset_close(ds, DS_MODE_NONE, tag);
435 if (tail[0] != '@') {
441 /* Look for a snapshot */
442 if (!DS_MODE_IS_READONLY(mode)) {
446 dprintf("looking for snapshot '%s'\n", tail);
447 err = zap_lookup(mos, obj, tail, 8, 1, &obj);
451 err = dsl_dataset_open_obj(dp, obj, tail, mode, tag, &ds);
454 rw_exit(&dp->dp_config_rwlock);
455 dsl_dir_close(dd, FTAG);
457 ASSERT3U((err == 0), ==, (ds != NULL));
458 /* ASSERT(ds == NULL || strcmp(name, ds->ds_name) == 0); */
465 dsl_dataset_open(const char *name, int mode, void *tag, dsl_dataset_t **dsp)
467 return (dsl_dataset_open_spa(NULL, name, mode, tag, dsp));
471 dsl_dataset_name(dsl_dataset_t *ds, char *name)
474 (void) strcpy(name, "mos");
476 dsl_dir_name(ds->ds_dir, name);
477 VERIFY(0 == dsl_dataset_get_snapname(ds));
478 if (ds->ds_snapname[0]) {
479 (void) strcat(name, "@");
480 if (!MUTEX_HELD(&ds->ds_lock)) {
482 * We use a "recursive" mutex so that we
483 * can call dprintf_ds() with ds_lock held.
485 mutex_enter(&ds->ds_lock);
486 (void) strcat(name, ds->ds_snapname);
487 mutex_exit(&ds->ds_lock);
489 (void) strcat(name, ds->ds_snapname);
496 dsl_dataset_namelen(dsl_dataset_t *ds)
501 result = 3; /* "mos" */
503 result = dsl_dir_namelen(ds->ds_dir);
504 VERIFY(0 == dsl_dataset_get_snapname(ds));
505 if (ds->ds_snapname[0]) {
506 ++result; /* adding one for the @-sign */
507 if (!MUTEX_HELD(&ds->ds_lock)) {
508 /* see dsl_datset_name */
509 mutex_enter(&ds->ds_lock);
510 result += strlen(ds->ds_snapname);
511 mutex_exit(&ds->ds_lock);
513 result += strlen(ds->ds_snapname);
522 dsl_dataset_close(dsl_dataset_t *ds, int mode, void *tag)
524 uint64_t weight = ds_refcnt_weight[DS_MODE_LEVEL(mode)];
525 mutex_enter(&ds->ds_lock);
526 ASSERT3U(ds->ds_open_refcount, >=, weight);
527 ds->ds_open_refcount -= weight;
528 dprintf_ds(ds, "closing mode %u refcount now 0x%llx\n",
529 mode, ds->ds_open_refcount);
530 mutex_exit(&ds->ds_lock);
532 dmu_buf_rele(ds->ds_dbuf, tag);
536 dsl_dataset_create_root(dsl_pool_t *dp, uint64_t *ddobjp, dmu_tx_t *tx)
538 objset_t *mos = dp->dp_meta_objset;
540 dsl_dataset_phys_t *dsphys;
545 dsl_dir_create_root(mos, ddobjp, tx);
546 VERIFY(0 == dsl_dir_open_obj(dp, *ddobjp, NULL, FTAG, &dd));
548 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
549 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
550 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
551 dmu_buf_will_dirty(dbuf, tx);
552 dsphys = dbuf->db_data;
553 dsphys->ds_dir_obj = dd->dd_object;
554 dsphys->ds_fsid_guid = unique_create();
555 unique_remove(dsphys->ds_fsid_guid); /* it isn't open yet */
556 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
557 sizeof (dsphys->ds_guid));
558 dsphys->ds_snapnames_zapobj =
559 zap_create(mos, DMU_OT_DSL_DS_SNAP_MAP, DMU_OT_NONE, 0, tx);
560 dsphys->ds_creation_time = gethrestime_sec();
561 dsphys->ds_creation_txg = tx->tx_txg;
562 dsphys->ds_deadlist_obj =
563 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
564 dmu_buf_rele(dbuf, FTAG);
566 dmu_buf_will_dirty(dd->dd_dbuf, tx);
567 dd->dd_phys->dd_head_dataset_obj = dsobj;
568 dsl_dir_close(dd, FTAG);
571 dsl_dataset_open_obj(dp, dsobj, NULL, DS_MODE_NONE, FTAG, &ds));
572 (void) dmu_objset_create_impl(dp->dp_spa, ds,
573 &ds->ds_phys->ds_bp, DMU_OST_ZFS, tx);
574 dsl_dataset_close(ds, DS_MODE_NONE, FTAG);
578 dsl_dataset_create_sync(dsl_dir_t *pdd,
579 const char *lastname, dsl_dataset_t *clone_parent, dmu_tx_t *tx)
581 dsl_pool_t *dp = pdd->dd_pool;
583 dsl_dataset_phys_t *dsphys;
584 uint64_t dsobj, ddobj;
585 objset_t *mos = dp->dp_meta_objset;
588 ASSERT(clone_parent == NULL || clone_parent->ds_dir->dd_pool == dp);
589 ASSERT(clone_parent == NULL ||
590 clone_parent->ds_phys->ds_num_children > 0);
591 ASSERT(lastname[0] != '@');
592 ASSERT(dmu_tx_is_syncing(tx));
594 ddobj = dsl_dir_create_sync(pdd, lastname, tx);
595 VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd));
597 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
598 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
599 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
600 dmu_buf_will_dirty(dbuf, tx);
601 dsphys = dbuf->db_data;
602 dsphys->ds_dir_obj = dd->dd_object;
603 dsphys->ds_fsid_guid = unique_create();
604 unique_remove(dsphys->ds_fsid_guid); /* it isn't open yet */
605 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
606 sizeof (dsphys->ds_guid));
607 dsphys->ds_snapnames_zapobj =
608 zap_create(mos, DMU_OT_DSL_DS_SNAP_MAP, DMU_OT_NONE, 0, tx);
609 dsphys->ds_creation_time = gethrestime_sec();
610 dsphys->ds_creation_txg = tx->tx_txg;
611 dsphys->ds_deadlist_obj =
612 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
614 dsphys->ds_prev_snap_obj = clone_parent->ds_object;
615 dsphys->ds_prev_snap_txg =
616 clone_parent->ds_phys->ds_creation_txg;
617 dsphys->ds_used_bytes =
618 clone_parent->ds_phys->ds_used_bytes;
619 dsphys->ds_compressed_bytes =
620 clone_parent->ds_phys->ds_compressed_bytes;
621 dsphys->ds_uncompressed_bytes =
622 clone_parent->ds_phys->ds_uncompressed_bytes;
623 dsphys->ds_bp = clone_parent->ds_phys->ds_bp;
625 dmu_buf_will_dirty(clone_parent->ds_dbuf, tx);
626 clone_parent->ds_phys->ds_num_children++;
628 dmu_buf_will_dirty(dd->dd_dbuf, tx);
629 dd->dd_phys->dd_clone_parent_obj = clone_parent->ds_object;
631 dmu_buf_rele(dbuf, FTAG);
633 dmu_buf_will_dirty(dd->dd_dbuf, tx);
634 dd->dd_phys->dd_head_dataset_obj = dsobj;
635 dsl_dir_close(dd, FTAG);
641 dsl_sync_task_group_t *dstg;
647 dsl_snapshot_destroy_one(char *name, void *arg)
649 struct destroyarg *da = arg;
654 (void) strcat(name, "@");
655 (void) strcat(name, da->snapname);
656 err = dsl_dataset_open(name,
657 DS_MODE_EXCLUSIVE | DS_MODE_READONLY | DS_MODE_INCONSISTENT,
659 cp = strchr(name, '@');
664 (void) strcpy(da->failed, name);
668 dsl_sync_task_create(da->dstg, dsl_dataset_destroy_check,
669 dsl_dataset_destroy_sync, ds, da->dstg, 0);
674 * Destroy 'snapname' in all descendants of 'fsname'.
676 #pragma weak dmu_snapshots_destroy = dsl_snapshots_destroy
678 dsl_snapshots_destroy(char *fsname, char *snapname)
681 struct destroyarg da;
682 dsl_sync_task_t *dst;
686 cp = strchr(fsname, '/');
689 err = spa_open(fsname, &spa, FTAG);
692 err = spa_open(fsname, &spa, FTAG);
696 da.dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
697 da.snapname = snapname;
700 err = dmu_objset_find(fsname,
701 dsl_snapshot_destroy_one, &da, DS_FIND_CHILDREN);
704 err = dsl_sync_task_group_wait(da.dstg);
706 for (dst = list_head(&da.dstg->dstg_tasks); dst;
707 dst = list_next(&da.dstg->dstg_tasks, dst)) {
708 dsl_dataset_t *ds = dst->dst_arg1;
710 dsl_dataset_name(ds, fsname);
711 cp = strchr(fsname, '@');
715 * If it was successful, destroy_sync would have
719 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, da.dstg);
722 dsl_sync_task_group_destroy(da.dstg);
723 spa_close(spa, FTAG);
728 dsl_dataset_destroy(const char *name)
731 dsl_sync_task_group_t *dstg;
737 if (strchr(name, '@')) {
738 /* Destroying a snapshot is simpler */
739 err = dsl_dataset_open(name,
740 DS_MODE_EXCLUSIVE | DS_MODE_READONLY | DS_MODE_INCONSISTENT,
744 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
745 dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
748 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG);
752 err = dmu_objset_open(name, DMU_OST_ANY,
753 DS_MODE_EXCLUSIVE | DS_MODE_INCONSISTENT, &os);
756 ds = os->os->os_dsl_dataset;
760 * Check for errors and mark this ds as inconsistent, in
761 * case we crash while freeing the objects.
763 err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check,
764 dsl_dataset_destroy_begin_sync, ds, NULL, 0);
766 dmu_objset_close(os);
771 * remove the objects in open context, so that we won't
772 * have too much to do in syncing context.
774 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
775 ds->ds_phys->ds_prev_snap_txg)) {
776 dmu_tx_t *tx = dmu_tx_create(os);
777 dmu_tx_hold_free(tx, obj, 0, DMU_OBJECT_END);
778 dmu_tx_hold_bonus(tx, obj);
779 err = dmu_tx_assign(tx, TXG_WAIT);
782 * Perhaps there is not enough disk
783 * space. Just deal with it from
784 * dsl_dataset_destroy_sync().
789 VERIFY(0 == dmu_object_free(os, obj, tx));
792 /* Make sure it's not dirty before we finish destroying it. */
793 txg_wait_synced(dd->dd_pool, 0);
795 dmu_objset_close(os);
799 err = dsl_dataset_open(name,
800 DS_MODE_EXCLUSIVE | DS_MODE_READONLY | DS_MODE_INCONSISTENT,
805 err = dsl_dir_open(name, FTAG, &dd, NULL);
807 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG);
812 * Blow away the dsl_dir + head dataset.
814 dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
815 dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
816 dsl_dataset_destroy_sync, ds, FTAG, 0);
817 dsl_sync_task_create(dstg, dsl_dir_destroy_check,
818 dsl_dir_destroy_sync, dd, FTAG, 0);
819 err = dsl_sync_task_group_wait(dstg);
820 dsl_sync_task_group_destroy(dstg);
821 /* if it is successful, *destroy_sync will close the ds+dd */
823 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG);
824 dsl_dir_close(dd, FTAG);
830 dsl_dataset_rollback(dsl_dataset_t *ds)
832 ASSERT3U(ds->ds_open_refcount, ==, DS_REF_MAX);
833 return (dsl_sync_task_do(ds->ds_dir->dd_pool,
834 dsl_dataset_rollback_check, dsl_dataset_rollback_sync,
839 dsl_dataset_set_user_ptr(dsl_dataset_t *ds,
840 void *p, dsl_dataset_evict_func_t func)
844 mutex_enter(&ds->ds_lock);
845 old = ds->ds_user_ptr;
848 ds->ds_user_evict_func = func;
850 mutex_exit(&ds->ds_lock);
855 dsl_dataset_get_user_ptr(dsl_dataset_t *ds)
857 return (ds->ds_user_ptr);
862 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
864 return (&ds->ds_phys->ds_bp);
868 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
870 ASSERT(dmu_tx_is_syncing(tx));
871 /* If it's the meta-objset, set dp_meta_rootbp */
873 tx->tx_pool->dp_meta_rootbp = *bp;
875 dmu_buf_will_dirty(ds->ds_dbuf, tx);
876 ds->ds_phys->ds_bp = *bp;
881 dsl_dataset_get_spa(dsl_dataset_t *ds)
883 return (ds->ds_dir->dd_pool->dp_spa);
887 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
891 if (ds == NULL) /* this is the meta-objset */
894 ASSERT(ds->ds_user_ptr != NULL);
896 if (ds->ds_phys->ds_next_snap_obj != 0)
897 panic("dirtying snapshot!");
899 dp = ds->ds_dir->dd_pool;
901 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) {
902 /* up the hold count until we can be written out */
903 dmu_buf_add_ref(ds->ds_dbuf, ds);
909 uint64_t *compressedp;
910 uint64_t *uncompressedp;
916 kill_blkptr(traverse_blk_cache_t *bc, spa_t *spa, void *arg)
918 struct killarg *ka = arg;
919 blkptr_t *bp = &bc->bc_blkptr;
921 ASSERT3U(bc->bc_errno, ==, 0);
924 * Since this callback is not called concurrently, no lock is
925 * needed on the accounting values.
927 *ka->usedp += bp_get_dasize(spa, bp);
928 *ka->compressedp += BP_GET_PSIZE(bp);
929 *ka->uncompressedp += BP_GET_UCSIZE(bp);
930 /* XXX check for EIO? */
931 (void) arc_free(ka->zio, spa, ka->tx->tx_txg, bp, NULL, NULL,
938 dsl_dataset_rollback_check(void *arg1, void *arg2, dmu_tx_t *tx)
940 dsl_dataset_t *ds = arg1;
943 * There must be a previous snapshot. I suppose we could roll
944 * it back to being empty (and re-initialize the upper (ZPL)
945 * layer). But for now there's no way to do this via the user
948 if (ds->ds_phys->ds_prev_snap_txg == 0)
952 * This must not be a snapshot.
954 if (ds->ds_phys->ds_next_snap_obj != 0)
958 * If we made changes this txg, traverse_dsl_dataset won't find
961 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
969 dsl_dataset_rollback_sync(void *arg1, void *arg2, dmu_tx_t *tx)
971 dsl_dataset_t *ds = arg1;
972 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
974 dmu_buf_will_dirty(ds->ds_dbuf, tx);
976 /* Zero out the deadlist. */
977 bplist_close(&ds->ds_deadlist);
978 bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx);
979 ds->ds_phys->ds_deadlist_obj =
980 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
981 VERIFY(0 == bplist_open(&ds->ds_deadlist, mos,
982 ds->ds_phys->ds_deadlist_obj));
985 /* Free blkptrs that we gave birth to */
987 uint64_t used = 0, compressed = 0, uncompressed = 0;
990 zio = zio_root(tx->tx_pool->dp_spa, NULL, NULL,
991 ZIO_FLAG_MUSTSUCCEED);
993 ka.compressedp = &compressed;
994 ka.uncompressedp = &uncompressed;
997 (void) traverse_dsl_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
998 ADVANCE_POST, kill_blkptr, &ka);
999 (void) zio_wait(zio);
1001 dsl_dir_diduse_space(ds->ds_dir,
1002 -used, -compressed, -uncompressed, tx);
1005 /* Change our contents to that of the prev snapshot */
1006 ASSERT3U(ds->ds_prev->ds_object, ==, ds->ds_phys->ds_prev_snap_obj);
1007 ds->ds_phys->ds_bp = ds->ds_prev->ds_phys->ds_bp;
1008 ds->ds_phys->ds_used_bytes = ds->ds_prev->ds_phys->ds_used_bytes;
1009 ds->ds_phys->ds_compressed_bytes =
1010 ds->ds_prev->ds_phys->ds_compressed_bytes;
1011 ds->ds_phys->ds_uncompressed_bytes =
1012 ds->ds_prev->ds_phys->ds_uncompressed_bytes;
1013 ds->ds_phys->ds_flags = ds->ds_prev->ds_phys->ds_flags;
1014 ds->ds_phys->ds_unique_bytes = 0;
1016 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1017 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1018 ds->ds_prev->ds_phys->ds_unique_bytes = 0;
1024 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1026 dsl_dataset_t *ds = arg1;
1029 * Can't delete a head dataset if there are snapshots of it.
1030 * (Except if the only snapshots are from the branch we cloned
1033 if (ds->ds_prev != NULL &&
1034 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1042 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1044 dsl_dataset_t *ds = arg1;
1046 /* Mark it as inconsistent on-disk, in case we crash */
1047 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1048 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1053 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1055 dsl_dataset_t *ds = arg1;
1057 /* Can't delete a branch point. */
1058 if (ds->ds_phys->ds_num_children > 1)
1062 * Can't delete a head dataset if there are snapshots of it.
1063 * (Except if the only snapshots are from the branch we cloned
1066 if (ds->ds_prev != NULL &&
1067 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1071 * If we made changes this txg, traverse_dsl_dataset won't find
1074 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1077 /* XXX we should do some i/o error checking... */
1082 dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
1084 dsl_dataset_t *ds = arg1;
1085 uint64_t used = 0, compressed = 0, uncompressed = 0;
1088 int after_branch_point = FALSE;
1089 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1090 objset_t *mos = dp->dp_meta_objset;
1091 dsl_dataset_t *ds_prev = NULL;
1094 ASSERT3U(ds->ds_open_refcount, ==, DS_REF_MAX);
1095 ASSERT3U(ds->ds_phys->ds_num_children, <=, 1);
1096 ASSERT(ds->ds_prev == NULL ||
1097 ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1098 ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1100 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1102 obj = ds->ds_object;
1104 if (ds->ds_phys->ds_prev_snap_obj != 0) {
1106 ds_prev = ds->ds_prev;
1108 VERIFY(0 == dsl_dataset_open_obj(dp,
1109 ds->ds_phys->ds_prev_snap_obj, NULL,
1110 DS_MODE_NONE, FTAG, &ds_prev));
1112 after_branch_point =
1113 (ds_prev->ds_phys->ds_next_snap_obj != obj);
1115 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
1116 if (after_branch_point &&
1117 ds->ds_phys->ds_next_snap_obj == 0) {
1118 /* This clone is toast. */
1119 ASSERT(ds_prev->ds_phys->ds_num_children > 1);
1120 ds_prev->ds_phys->ds_num_children--;
1121 } else if (!after_branch_point) {
1122 ds_prev->ds_phys->ds_next_snap_obj =
1123 ds->ds_phys->ds_next_snap_obj;
1127 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1129 if (ds->ds_phys->ds_next_snap_obj != 0) {
1131 dsl_dataset_t *ds_next;
1134 spa_scrub_restart(dp->dp_spa, tx->tx_txg);
1136 VERIFY(0 == dsl_dataset_open_obj(dp,
1137 ds->ds_phys->ds_next_snap_obj, NULL,
1138 DS_MODE_NONE, FTAG, &ds_next));
1139 ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
1141 dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
1142 ds_next->ds_phys->ds_prev_snap_obj =
1143 ds->ds_phys->ds_prev_snap_obj;
1144 ds_next->ds_phys->ds_prev_snap_txg =
1145 ds->ds_phys->ds_prev_snap_txg;
1146 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1147 ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
1150 * Transfer to our deadlist (which will become next's
1151 * new deadlist) any entries from next's current
1152 * deadlist which were born before prev, and free the
1155 * XXX we're doing this long task with the config lock held
1157 while (bplist_iterate(&ds_next->ds_deadlist, &itor,
1159 if (bp.blk_birth <= ds->ds_phys->ds_prev_snap_txg) {
1160 VERIFY(0 == bplist_enqueue(&ds->ds_deadlist,
1162 if (ds_prev && !after_branch_point &&
1164 ds_prev->ds_phys->ds_prev_snap_txg) {
1165 ds_prev->ds_phys->ds_unique_bytes +=
1166 bp_get_dasize(dp->dp_spa, &bp);
1169 used += bp_get_dasize(dp->dp_spa, &bp);
1170 compressed += BP_GET_PSIZE(&bp);
1171 uncompressed += BP_GET_UCSIZE(&bp);
1172 /* XXX check return value? */
1173 (void) arc_free(zio, dp->dp_spa, tx->tx_txg,
1174 &bp, NULL, NULL, ARC_NOWAIT);
1178 /* free next's deadlist */
1179 bplist_close(&ds_next->ds_deadlist);
1180 bplist_destroy(mos, ds_next->ds_phys->ds_deadlist_obj, tx);
1182 /* set next's deadlist to our deadlist */
1183 ds_next->ds_phys->ds_deadlist_obj =
1184 ds->ds_phys->ds_deadlist_obj;
1185 VERIFY(0 == bplist_open(&ds_next->ds_deadlist, mos,
1186 ds_next->ds_phys->ds_deadlist_obj));
1187 ds->ds_phys->ds_deadlist_obj = 0;
1189 if (ds_next->ds_phys->ds_next_snap_obj != 0) {
1191 * Update next's unique to include blocks which
1192 * were previously shared by only this snapshot
1193 * and it. Those blocks will be born after the
1194 * prev snap and before this snap, and will have
1195 * died after the next snap and before the one
1196 * after that (ie. be on the snap after next's
1199 * XXX we're doing this long task with the
1202 dsl_dataset_t *ds_after_next;
1204 VERIFY(0 == dsl_dataset_open_obj(dp,
1205 ds_next->ds_phys->ds_next_snap_obj, NULL,
1206 DS_MODE_NONE, FTAG, &ds_after_next));
1208 while (bplist_iterate(&ds_after_next->ds_deadlist,
1211 ds->ds_phys->ds_prev_snap_txg &&
1213 ds->ds_phys->ds_creation_txg) {
1214 ds_next->ds_phys->ds_unique_bytes +=
1215 bp_get_dasize(dp->dp_spa, &bp);
1219 dsl_dataset_close(ds_after_next, DS_MODE_NONE, FTAG);
1220 ASSERT3P(ds_next->ds_prev, ==, NULL);
1223 * It would be nice to update the head dataset's
1224 * unique. To do so we would have to traverse
1225 * it for blocks born after ds_prev, which is
1226 * pretty expensive just to maintain something
1227 * for debugging purposes.
1229 ASSERT3P(ds_next->ds_prev, ==, ds);
1230 dsl_dataset_close(ds_next->ds_prev, DS_MODE_NONE,
1233 VERIFY(0 == dsl_dataset_open_obj(dp,
1234 ds->ds_phys->ds_prev_snap_obj, NULL,
1235 DS_MODE_NONE, ds_next, &ds_next->ds_prev));
1237 ds_next->ds_prev = NULL;
1240 dsl_dataset_close(ds_next, DS_MODE_NONE, FTAG);
1243 * NB: unique_bytes is not accurate for head objsets
1244 * because we don't update it when we delete the most
1245 * recent snapshot -- see above comment.
1247 ASSERT3U(used, ==, ds->ds_phys->ds_unique_bytes);
1250 * There's no next snapshot, so this is a head dataset.
1251 * Destroy the deadlist. Unless it's a clone, the
1252 * deadlist should be empty. (If it's a clone, it's
1253 * safe to ignore the deadlist contents.)
1257 ASSERT(after_branch_point || bplist_empty(&ds->ds_deadlist));
1258 bplist_close(&ds->ds_deadlist);
1259 bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx);
1260 ds->ds_phys->ds_deadlist_obj = 0;
1263 * Free everything that we point to (that's born after
1264 * the previous snapshot, if we are a clone)
1266 * XXX we're doing this long task with the config lock held
1269 ka.compressedp = &compressed;
1270 ka.uncompressedp = &uncompressed;
1273 err = traverse_dsl_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1274 ADVANCE_POST, kill_blkptr, &ka);
1275 ASSERT3U(err, ==, 0);
1278 err = zio_wait(zio);
1279 ASSERT3U(err, ==, 0);
1281 dsl_dir_diduse_space(ds->ds_dir, -used, -compressed, -uncompressed, tx);
1283 if (ds->ds_phys->ds_snapnames_zapobj) {
1284 err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx);
1288 if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) {
1289 /* Erase the link in the dataset */
1290 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1291 ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
1293 * dsl_dir_sync_destroy() called us, they'll destroy
1297 /* remove from snapshot namespace */
1298 dsl_dataset_t *ds_head;
1299 VERIFY(0 == dsl_dataset_open_obj(dp,
1300 ds->ds_dir->dd_phys->dd_head_dataset_obj, NULL,
1301 DS_MODE_NONE, FTAG, &ds_head));
1302 VERIFY(0 == dsl_dataset_get_snapname(ds));
1306 err = zap_lookup(mos,
1307 ds_head->ds_phys->ds_snapnames_zapobj,
1308 ds->ds_snapname, 8, 1, &val);
1309 ASSERT3U(err, ==, 0);
1310 ASSERT3U(val, ==, obj);
1313 err = zap_remove(mos, ds_head->ds_phys->ds_snapnames_zapobj,
1314 ds->ds_snapname, tx);
1316 dsl_dataset_close(ds_head, DS_MODE_NONE, FTAG);
1319 if (ds_prev && ds->ds_prev != ds_prev)
1320 dsl_dataset_close(ds_prev, DS_MODE_NONE, FTAG);
1322 spa_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
1323 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, tag);
1324 VERIFY(0 == dmu_object_free(mos, obj, tx));
1330 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
1332 objset_t *os = arg1;
1333 dsl_dataset_t *ds = os->os->os_dsl_dataset;
1334 const char *snapname = arg2;
1335 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1340 * We don't allow multiple snapshots of the same txg. If there
1341 * is already one, try again.
1343 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
1347 * Check for conflicting name snapshot name.
1349 err = zap_lookup(mos, ds->ds_phys->ds_snapnames_zapobj,
1350 snapname, 8, 1, &value);
1357 * Check that the dataset's name is not too long. Name consists
1358 * of the dataset's length + 1 for the @-sign + snapshot name's length
1360 if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
1361 return (ENAMETOOLONG);
1363 ds->ds_trysnap_txg = tx->tx_txg;
1368 dsl_dataset_snapshot_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1370 objset_t *os = arg1;
1371 dsl_dataset_t *ds = os->os->os_dsl_dataset;
1372 const char *snapname = arg2;
1373 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1375 dsl_dataset_phys_t *dsphys;
1377 objset_t *mos = dp->dp_meta_objset;
1380 spa_scrub_restart(dp->dp_spa, tx->tx_txg);
1381 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1383 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1384 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1385 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1386 dmu_buf_will_dirty(dbuf, tx);
1387 dsphys = dbuf->db_data;
1388 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1389 dsphys->ds_fsid_guid = unique_create();
1390 unique_remove(dsphys->ds_fsid_guid); /* it isn't open yet */
1391 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1392 sizeof (dsphys->ds_guid));
1393 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
1394 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
1395 dsphys->ds_next_snap_obj = ds->ds_object;
1396 dsphys->ds_num_children = 1;
1397 dsphys->ds_creation_time = gethrestime_sec();
1398 dsphys->ds_creation_txg = tx->tx_txg;
1399 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
1400 dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes;
1401 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
1402 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
1403 dsphys->ds_flags = ds->ds_phys->ds_flags;
1404 dsphys->ds_bp = ds->ds_phys->ds_bp;
1405 dmu_buf_rele(dbuf, FTAG);
1407 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
1409 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
1411 ds->ds_prev->ds_phys->ds_num_children > 1);
1412 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1413 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1414 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1415 ds->ds_prev->ds_phys->ds_creation_txg);
1416 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
1420 bplist_close(&ds->ds_deadlist);
1421 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1422 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, dsphys->ds_creation_txg);
1423 ds->ds_phys->ds_prev_snap_obj = dsobj;
1424 ds->ds_phys->ds_prev_snap_txg = dsphys->ds_creation_txg;
1425 ds->ds_phys->ds_unique_bytes = 0;
1426 ds->ds_phys->ds_deadlist_obj =
1427 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
1428 VERIFY(0 == bplist_open(&ds->ds_deadlist, mos,
1429 ds->ds_phys->ds_deadlist_obj));
1431 dprintf("snap '%s' -> obj %llu\n", snapname, dsobj);
1432 err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
1433 snapname, 8, 1, &dsobj, tx);
1437 dsl_dataset_close(ds->ds_prev, DS_MODE_NONE, ds);
1438 VERIFY(0 == dsl_dataset_open_obj(dp,
1439 ds->ds_phys->ds_prev_snap_obj, snapname,
1440 DS_MODE_NONE, ds, &ds->ds_prev));
1444 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1446 ASSERT(dmu_tx_is_syncing(tx));
1447 ASSERT(ds->ds_user_ptr != NULL);
1448 ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
1450 dsl_dir_dirty(ds->ds_dir, tx);
1451 dmu_objset_sync(ds->ds_user_ptr, zio, tx);
1452 /* Unneeded? bplist_close(&ds->ds_deadlist); */
1456 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1458 dsl_dir_stats(ds->ds_dir, nv);
1460 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1461 ds->ds_phys->ds_creation_time);
1462 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1463 ds->ds_phys->ds_creation_txg);
1464 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED,
1465 ds->ds_phys->ds_used_bytes);
1467 if (ds->ds_phys->ds_next_snap_obj) {
1469 * This is a snapshot; override the dd's space used with
1470 * our unique space and compression ratio.
1472 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1473 ds->ds_phys->ds_unique_bytes);
1474 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
1475 ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
1476 (ds->ds_phys->ds_uncompressed_bytes * 100 /
1477 ds->ds_phys->ds_compressed_bytes));
1482 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1484 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
1485 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
1486 if (ds->ds_phys->ds_next_snap_obj) {
1487 stat->dds_is_snapshot = B_TRUE;
1488 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
1491 /* clone origin is really a dsl_dir thing... */
1492 if (ds->ds_dir->dd_phys->dd_clone_parent_obj) {
1495 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
1496 VERIFY(0 == dsl_dataset_open_obj(ds->ds_dir->dd_pool,
1497 ds->ds_dir->dd_phys->dd_clone_parent_obj,
1498 NULL, DS_MODE_NONE, FTAG, &ods));
1499 dsl_dataset_name(ods, stat->dds_clone_of);
1500 dsl_dataset_close(ods, DS_MODE_NONE, FTAG);
1501 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
1506 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1508 return (ds->ds_phys->ds_fsid_guid);
1512 dsl_dataset_space(dsl_dataset_t *ds,
1513 uint64_t *refdbytesp, uint64_t *availbytesp,
1514 uint64_t *usedobjsp, uint64_t *availobjsp)
1516 *refdbytesp = ds->ds_phys->ds_used_bytes;
1517 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1518 *usedobjsp = ds->ds_phys->ds_bp.blk_fill;
1519 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
1524 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
1526 dsl_dataset_t *ds = arg1;
1527 char *newsnapname = arg2;
1528 dsl_dir_t *dd = ds->ds_dir;
1529 objset_t *mos = dd->dd_pool->dp_meta_objset;
1534 err = dsl_dataset_open_obj(dd->dd_pool,
1535 dd->dd_phys->dd_head_dataset_obj, NULL, DS_MODE_NONE, FTAG, &hds);
1539 /* new name better not be in use */
1540 err = zap_lookup(mos, hds->ds_phys->ds_snapnames_zapobj,
1541 newsnapname, 8, 1, &val);
1542 dsl_dataset_close(hds, DS_MODE_NONE, FTAG);
1546 else if (err == ENOENT)
1549 /* dataset name + 1 for the "@" + the new snapshot name must fit */
1550 if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
1557 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1559 dsl_dataset_t *ds = arg1;
1560 char *newsnapname = arg2;
1561 dsl_dir_t *dd = ds->ds_dir;
1562 objset_t *mos = dd->dd_pool->dp_meta_objset;
1566 ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
1568 VERIFY(0 == dsl_dataset_open_obj(dd->dd_pool,
1569 dd->dd_phys->dd_head_dataset_obj, NULL, DS_MODE_NONE, FTAG, &hds));
1571 VERIFY(0 == dsl_dataset_get_snapname(ds));
1572 err = zap_remove(mos, hds->ds_phys->ds_snapnames_zapobj,
1573 ds->ds_snapname, tx);
1574 ASSERT3U(err, ==, 0);
1575 mutex_enter(&ds->ds_lock);
1576 (void) strcpy(ds->ds_snapname, newsnapname);
1577 mutex_exit(&ds->ds_lock);
1578 err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
1579 ds->ds_snapname, 8, 1, &ds->ds_object, tx);
1580 ASSERT3U(err, ==, 0);
1582 dsl_dataset_close(hds, DS_MODE_NONE, FTAG);
1586 dsl_sync_task_group_t *dstg;
1587 char failed[MAXPATHLEN];
1593 dsl_snapshot_rename_one(char *name, void *arg)
1595 struct renamearg *ra = arg;
1596 dsl_dataset_t *ds = NULL;
1600 cp = name + strlen(name);
1602 (void) strcpy(cp + 1, ra->oldsnap);
1603 err = dsl_dataset_open(name, DS_MODE_READONLY | DS_MODE_STANDARD,
1605 if (err == ENOENT) {
1610 (void) strcpy(ra->failed, name);
1612 dsl_dataset_close(ds, DS_MODE_STANDARD, ra->dstg);
1617 /* for all filesystems undergoing rename, we'll need to unmount it */
1618 (void) zfs_unmount_snap(name, NULL);
1623 dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
1624 dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
1630 dsl_recursive_rename(char *oldname, const char *newname)
1633 struct renamearg *ra;
1634 dsl_sync_task_t *dst;
1636 char *cp, *fsname = spa_strdup(oldname);
1637 int len = strlen(oldname);
1639 /* truncate the snapshot name to get the fsname */
1640 cp = strchr(fsname, '@');
1643 cp = strchr(fsname, '/');
1646 err = spa_open(fsname, &spa, FTAG);
1649 err = spa_open(fsname, &spa, FTAG);
1652 kmem_free(fsname, len + 1);
1655 ra = kmem_alloc(sizeof (struct renamearg), KM_SLEEP);
1656 ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
1658 ra->oldsnap = strchr(oldname, '@') + 1;
1659 ra->newsnap = strchr(newname, '@') + 1;
1662 err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
1664 kmem_free(fsname, len + 1);
1667 err = dsl_sync_task_group_wait(ra->dstg);
1670 for (dst = list_head(&ra->dstg->dstg_tasks); dst;
1671 dst = list_next(&ra->dstg->dstg_tasks, dst)) {
1672 dsl_dataset_t *ds = dst->dst_arg1;
1674 dsl_dir_name(ds->ds_dir, ra->failed);
1675 (void) strcat(ra->failed, "@");
1676 (void) strcat(ra->failed, ra->newsnap);
1678 dsl_dataset_close(ds, DS_MODE_STANDARD, ra->dstg);
1681 (void) strcpy(oldname, ra->failed);
1683 dsl_sync_task_group_destroy(ra->dstg);
1684 kmem_free(ra, sizeof (struct renamearg));
1685 spa_close(spa, FTAG);
1689 #pragma weak dmu_objset_rename = dsl_dataset_rename
1691 dsl_dataset_rename(char *oldname, const char *newname,
1692 boolean_t recursive)
1699 err = dsl_dir_open(oldname, FTAG, &dd, &tail);
1703 err = dsl_dir_rename(dd, newname);
1704 dsl_dir_close(dd, FTAG);
1707 if (tail[0] != '@') {
1708 /* the name ended in a nonexistant component */
1709 dsl_dir_close(dd, FTAG);
1713 dsl_dir_close(dd, FTAG);
1715 /* new name must be snapshot in same filesystem */
1716 tail = strchr(newname, '@');
1720 if (strncmp(oldname, newname, tail - newname) != 0)
1724 err = dsl_recursive_rename(oldname, newname);
1726 err = dsl_dataset_open(oldname,
1727 DS_MODE_READONLY | DS_MODE_STANDARD, FTAG, &ds);
1731 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1732 dsl_dataset_snapshot_rename_check,
1733 dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
1735 dsl_dataset_close(ds, DS_MODE_STANDARD, FTAG);
1742 uint64_t used, comp, uncomp, unique;
1743 uint64_t newnext_obj, snapnames_obj;
1747 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
1749 dsl_dataset_t *hds = arg1;
1750 struct promotearg *pa = arg2;
1751 dsl_dir_t *dd = hds->ds_dir;
1752 dsl_pool_t *dp = hds->ds_dir->dd_pool;
1753 dsl_dir_t *pdd = NULL;
1754 dsl_dataset_t *ds = NULL;
1755 dsl_dataset_t *pivot_ds = NULL;
1756 dsl_dataset_t *newnext_ds = NULL;
1762 bzero(pa, sizeof (*pa));
1764 /* Check that it is a clone */
1765 if (dd->dd_phys->dd_clone_parent_obj == 0)
1768 /* Since this is so expensive, don't do the preliminary check */
1769 if (!dmu_tx_is_syncing(tx))
1772 if (err = dsl_dataset_open_obj(dp,
1773 dd->dd_phys->dd_clone_parent_obj,
1774 NULL, DS_MODE_EXCLUSIVE, FTAG, &pivot_ds))
1776 pdd = pivot_ds->ds_dir;
1779 dsl_dataset_t *phds;
1780 if (err = dsl_dataset_open_obj(dd->dd_pool,
1781 pdd->dd_phys->dd_head_dataset_obj,
1782 NULL, DS_MODE_NONE, FTAG, &phds))
1784 pa->snapnames_obj = phds->ds_phys->ds_snapnames_zapobj;
1785 dsl_dataset_close(phds, DS_MODE_NONE, FTAG);
1788 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) {
1793 /* find pivot point's new next ds */
1794 VERIFY(0 == dsl_dataset_open_obj(dd->dd_pool, hds->ds_object,
1795 NULL, DS_MODE_NONE, FTAG, &newnext_ds));
1796 while (newnext_ds->ds_phys->ds_prev_snap_obj != pivot_ds->ds_object) {
1797 dsl_dataset_t *prev;
1799 if (err = dsl_dataset_open_obj(dd->dd_pool,
1800 newnext_ds->ds_phys->ds_prev_snap_obj,
1801 NULL, DS_MODE_NONE, FTAG, &prev))
1803 dsl_dataset_close(newnext_ds, DS_MODE_NONE, FTAG);
1806 pa->newnext_obj = newnext_ds->ds_object;
1808 /* compute pivot point's new unique space */
1809 while ((err = bplist_iterate(&newnext_ds->ds_deadlist,
1810 &itor, &bp)) == 0) {
1811 if (bp.blk_birth > pivot_ds->ds_phys->ds_prev_snap_txg)
1812 pa->unique += bp_get_dasize(dd->dd_pool->dp_spa, &bp);
1817 /* Walk the snapshots that we are moving */
1818 name = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1822 uint64_t val, dlused, dlcomp, dluncomp;
1823 dsl_dataset_t *prev;
1825 /* Check that the snapshot name does not conflict */
1826 dsl_dataset_name(ds, name);
1827 err = zap_lookup(dd->dd_pool->dp_meta_objset,
1828 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
1830 if (err != ENOENT) {
1837 * compute space to transfer. Each snapshot gave birth to:
1838 * (my used) - (prev's used) + (deadlist's used)
1840 pa->used += ds->ds_phys->ds_used_bytes;
1841 pa->comp += ds->ds_phys->ds_compressed_bytes;
1842 pa->uncomp += ds->ds_phys->ds_uncompressed_bytes;
1844 /* If we reach the first snapshot, we're done. */
1845 if (ds->ds_phys->ds_prev_snap_obj == 0)
1848 if (err = bplist_space(&ds->ds_deadlist,
1849 &dlused, &dlcomp, &dluncomp))
1851 if (err = dsl_dataset_open_obj(dd->dd_pool,
1852 ds->ds_phys->ds_prev_snap_obj, NULL, DS_MODE_EXCLUSIVE,
1855 pa->used += dlused - prev->ds_phys->ds_used_bytes;
1856 pa->comp += dlcomp - prev->ds_phys->ds_compressed_bytes;
1857 pa->uncomp += dluncomp - prev->ds_phys->ds_uncompressed_bytes;
1860 * We could be a clone of a clone. If we reach our
1861 * parent's branch point, we're done.
1863 if (prev->ds_phys->ds_next_snap_obj != ds->ds_object) {
1864 dsl_dataset_close(prev, DS_MODE_EXCLUSIVE, FTAG);
1868 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG);
1872 /* Check that there is enough space here */
1873 err = dsl_dir_transfer_possible(pdd, dd, pa->used);
1876 if (ds && ds != pivot_ds)
1877 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG);
1879 dsl_dataset_close(pivot_ds, DS_MODE_EXCLUSIVE, FTAG);
1881 dsl_dataset_close(newnext_ds, DS_MODE_NONE, FTAG);
1883 kmem_free(name, MAXPATHLEN);
1888 dsl_dataset_promote_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1890 dsl_dataset_t *hds = arg1;
1891 struct promotearg *pa = arg2;
1892 dsl_dir_t *dd = hds->ds_dir;
1893 dsl_pool_t *dp = hds->ds_dir->dd_pool;
1894 dsl_dir_t *pdd = NULL;
1895 dsl_dataset_t *ds, *pivot_ds;
1898 ASSERT(dd->dd_phys->dd_clone_parent_obj != 0);
1899 ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
1901 VERIFY(0 == dsl_dataset_open_obj(dp,
1902 dd->dd_phys->dd_clone_parent_obj,
1903 NULL, DS_MODE_EXCLUSIVE, FTAG, &pivot_ds));
1905 * We need to explicitly open pdd, since pivot_ds's pdd will be
1908 VERIFY(0 == dsl_dir_open_obj(dp, pivot_ds->ds_dir->dd_object,
1911 /* move snapshots to this dir */
1912 name = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1916 dsl_dataset_t *prev;
1918 /* move snap name entry */
1919 dsl_dataset_name(ds, name);
1920 VERIFY(0 == zap_remove(dp->dp_meta_objset,
1921 pa->snapnames_obj, ds->ds_snapname, tx));
1922 VERIFY(0 == zap_add(dp->dp_meta_objset,
1923 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
1924 8, 1, &ds->ds_object, tx));
1926 /* change containing dsl_dir */
1927 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1928 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, pdd->dd_object);
1929 ds->ds_phys->ds_dir_obj = dd->dd_object;
1930 ASSERT3P(ds->ds_dir, ==, pdd);
1931 dsl_dir_close(ds->ds_dir, ds);
1932 VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
1933 NULL, ds, &ds->ds_dir));
1935 ASSERT3U(dsl_prop_numcb(ds), ==, 0);
1937 if (ds->ds_phys->ds_prev_snap_obj == 0)
1940 VERIFY(0 == dsl_dataset_open_obj(dp,
1941 ds->ds_phys->ds_prev_snap_obj, NULL, DS_MODE_EXCLUSIVE,
1944 if (prev->ds_phys->ds_next_snap_obj != ds->ds_object) {
1945 dsl_dataset_close(prev, DS_MODE_EXCLUSIVE, FTAG);
1949 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG);
1953 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG);
1955 /* change pivot point's next snap */
1956 dmu_buf_will_dirty(pivot_ds->ds_dbuf, tx);
1957 pivot_ds->ds_phys->ds_next_snap_obj = pa->newnext_obj;
1959 /* change clone_parent-age */
1960 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1961 ASSERT3U(dd->dd_phys->dd_clone_parent_obj, ==, pivot_ds->ds_object);
1962 dd->dd_phys->dd_clone_parent_obj = pdd->dd_phys->dd_clone_parent_obj;
1963 dmu_buf_will_dirty(pdd->dd_dbuf, tx);
1964 pdd->dd_phys->dd_clone_parent_obj = pivot_ds->ds_object;
1966 /* change space accounting */
1967 dsl_dir_diduse_space(pdd, -pa->used, -pa->comp, -pa->uncomp, tx);
1968 dsl_dir_diduse_space(dd, pa->used, pa->comp, pa->uncomp, tx);
1969 pivot_ds->ds_phys->ds_unique_bytes = pa->unique;
1971 dsl_dir_close(pdd, FTAG);
1972 dsl_dataset_close(pivot_ds, DS_MODE_EXCLUSIVE, FTAG);
1973 kmem_free(name, MAXPATHLEN);
1977 dsl_dataset_promote(const char *name)
1981 dmu_object_info_t doi;
1982 struct promotearg pa;
1984 err = dsl_dataset_open(name, DS_MODE_NONE, FTAG, &ds);
1988 err = dmu_object_info(ds->ds_dir->dd_pool->dp_meta_objset,
1989 ds->ds_phys->ds_snapnames_zapobj, &doi);
1991 dsl_dataset_close(ds, DS_MODE_NONE, FTAG);
1996 * Add in 128x the snapnames zapobj size, since we will be moving
1997 * a bunch of snapnames to the promoted ds, and dirtying their
2000 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2001 dsl_dataset_promote_check,
2002 dsl_dataset_promote_sync, ds, &pa, 2 + 2 * doi.doi_physical_blks);
2003 dsl_dataset_close(ds, DS_MODE_NONE, FTAG);
2008 * Given a pool name and a dataset object number in that pool,
2009 * return the name of that dataset.
2012 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2016 dsl_dataset_t *ds = NULL;
2019 if ((error = spa_open(pname, &spa, FTAG)) != 0)
2021 dp = spa_get_dsl(spa);
2022 rw_enter(&dp->dp_config_rwlock, RW_READER);
2023 if ((error = dsl_dataset_open_obj(dp, obj,
2024 NULL, DS_MODE_NONE, FTAG, &ds)) != 0) {
2025 rw_exit(&dp->dp_config_rwlock);
2026 spa_close(spa, FTAG);
2029 dsl_dataset_name(ds, buf);
2030 dsl_dataset_close(ds, DS_MODE_NONE, FTAG);
2031 rw_exit(&dp->dp_config_rwlock);
2032 spa_close(spa, FTAG);