4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011 by Delphix. All rights reserved.
26 #include <sys/dmu_objset.h>
27 #include <sys/dsl_dataset.h>
28 #include <sys/dsl_dir.h>
29 #include <sys/dsl_prop.h>
30 #include <sys/dsl_synctask.h>
31 #include <sys/dmu_traverse.h>
32 #include <sys/dmu_tx.h>
36 #include <sys/unique.h>
37 #include <sys/zfs_context.h>
38 #include <sys/zfs_ioctl.h>
40 #include <sys/zfs_znode.h>
41 #include <sys/zfs_onexit.h>
43 #include <sys/dsl_scan.h>
44 #include <sys/dsl_deadlist.h>
46 static char *dsl_reaper
= "the grim reaper";
48 static dsl_checkfunc_t dsl_dataset_destroy_begin_check
;
49 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync
;
50 static dsl_syncfunc_t dsl_dataset_set_reservation_sync
;
52 #define SWITCH64(x, y) \
54 uint64_t __tmp = (x); \
59 #define DS_REF_MAX (1ULL << 62)
61 #define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE
63 #define DSL_DATASET_IS_DESTROYED(ds) ((ds)->ds_owner == dsl_reaper)
67 * Figure out how much of this delta should be propogated to the dsl_dir
68 * layer. If there's a refreservation, that space has already been
69 * partially accounted for in our ancestors.
72 parent_delta(dsl_dataset_t
*ds
, int64_t delta
)
74 uint64_t old_bytes
, new_bytes
;
76 if (ds
->ds_reserved
== 0)
79 old_bytes
= MAX(ds
->ds_phys
->ds_unique_bytes
, ds
->ds_reserved
);
80 new_bytes
= MAX(ds
->ds_phys
->ds_unique_bytes
+ delta
, ds
->ds_reserved
);
82 ASSERT3U(ABS((int64_t)(new_bytes
- old_bytes
)), <=, ABS(delta
));
83 return (new_bytes
- old_bytes
);
87 dsl_dataset_block_born(dsl_dataset_t
*ds
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
89 int used
= bp_get_dsize_sync(tx
->tx_pool
->dp_spa
, bp
);
90 int compressed
= BP_GET_PSIZE(bp
);
91 int uncompressed
= BP_GET_UCSIZE(bp
);
94 dprintf_bp(bp
, "ds=%p", ds
);
96 ASSERT(dmu_tx_is_syncing(tx
));
97 /* It could have been compressed away to nothing */
100 ASSERT(BP_GET_TYPE(bp
) != DMU_OT_NONE
);
101 ASSERT3U(BP_GET_TYPE(bp
), <, DMU_OT_NUMTYPES
);
104 * Account for the meta-objset space in its placeholder
107 ASSERT3U(compressed
, ==, uncompressed
); /* it's all metadata */
108 dsl_dir_diduse_space(tx
->tx_pool
->dp_mos_dir
, DD_USED_HEAD
,
109 used
, compressed
, uncompressed
, tx
);
110 dsl_dir_dirty(tx
->tx_pool
->dp_mos_dir
, tx
);
113 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
115 mutex_enter(&ds
->ds_dir
->dd_lock
);
116 mutex_enter(&ds
->ds_lock
);
117 delta
= parent_delta(ds
, used
);
118 ds
->ds_phys
->ds_used_bytes
+= used
;
119 ds
->ds_phys
->ds_compressed_bytes
+= compressed
;
120 ds
->ds_phys
->ds_uncompressed_bytes
+= uncompressed
;
121 ds
->ds_phys
->ds_unique_bytes
+= used
;
122 mutex_exit(&ds
->ds_lock
);
123 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_HEAD
, delta
,
124 compressed
, uncompressed
, tx
);
125 dsl_dir_transfer_space(ds
->ds_dir
, used
- delta
,
126 DD_USED_REFRSRV
, DD_USED_HEAD
, tx
);
127 mutex_exit(&ds
->ds_dir
->dd_lock
);
131 dsl_dataset_block_kill(dsl_dataset_t
*ds
, const blkptr_t
*bp
, dmu_tx_t
*tx
,
137 ASSERT(dmu_tx_is_syncing(tx
));
138 ASSERT(bp
->blk_birth
<= tx
->tx_txg
);
140 int used
= bp_get_dsize_sync(tx
->tx_pool
->dp_spa
, bp
);
141 int compressed
= BP_GET_PSIZE(bp
);
142 int uncompressed
= BP_GET_UCSIZE(bp
);
147 * Account for the meta-objset space in its placeholder
150 dsl_free(tx
->tx_pool
, tx
->tx_txg
, bp
);
152 dsl_dir_diduse_space(tx
->tx_pool
->dp_mos_dir
, DD_USED_HEAD
,
153 -used
, -compressed
, -uncompressed
, tx
);
154 dsl_dir_dirty(tx
->tx_pool
->dp_mos_dir
, tx
);
157 ASSERT3P(tx
->tx_pool
, ==, ds
->ds_dir
->dd_pool
);
159 ASSERT(!dsl_dataset_is_snapshot(ds
));
160 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
162 if (bp
->blk_birth
> ds
->ds_phys
->ds_prev_snap_txg
) {
165 dprintf_bp(bp
, "freeing ds=%llu", ds
->ds_object
);
166 dsl_free(tx
->tx_pool
, tx
->tx_txg
, bp
);
168 mutex_enter(&ds
->ds_dir
->dd_lock
);
169 mutex_enter(&ds
->ds_lock
);
170 ASSERT(ds
->ds_phys
->ds_unique_bytes
>= used
||
171 !DS_UNIQUE_IS_ACCURATE(ds
));
172 delta
= parent_delta(ds
, -used
);
173 ds
->ds_phys
->ds_unique_bytes
-= used
;
174 mutex_exit(&ds
->ds_lock
);
175 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_HEAD
,
176 delta
, -compressed
, -uncompressed
, tx
);
177 dsl_dir_transfer_space(ds
->ds_dir
, -used
- delta
,
178 DD_USED_REFRSRV
, DD_USED_HEAD
, tx
);
179 mutex_exit(&ds
->ds_dir
->dd_lock
);
181 dprintf_bp(bp
, "putting on dead list: %s", "");
184 * We are here as part of zio's write done callback,
185 * which means we're a zio interrupt thread. We can't
186 * call dsl_deadlist_insert() now because it may block
187 * waiting for I/O. Instead, put bp on the deferred
188 * queue and let dsl_pool_sync() finish the job.
190 bplist_append(&ds
->ds_pending_deadlist
, bp
);
192 dsl_deadlist_insert(&ds
->ds_deadlist
, bp
, tx
);
194 ASSERT3U(ds
->ds_prev
->ds_object
, ==,
195 ds
->ds_phys
->ds_prev_snap_obj
);
196 ASSERT(ds
->ds_prev
->ds_phys
->ds_num_children
> 0);
197 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
198 if (ds
->ds_prev
->ds_phys
->ds_next_snap_obj
==
199 ds
->ds_object
&& bp
->blk_birth
>
200 ds
->ds_prev
->ds_phys
->ds_prev_snap_txg
) {
201 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
202 mutex_enter(&ds
->ds_prev
->ds_lock
);
203 ds
->ds_prev
->ds_phys
->ds_unique_bytes
+= used
;
204 mutex_exit(&ds
->ds_prev
->ds_lock
);
206 if (bp
->blk_birth
> ds
->ds_dir
->dd_origin_txg
) {
207 dsl_dir_transfer_space(ds
->ds_dir
, used
,
208 DD_USED_HEAD
, DD_USED_SNAP
, tx
);
211 mutex_enter(&ds
->ds_lock
);
212 ASSERT3U(ds
->ds_phys
->ds_used_bytes
, >=, used
);
213 ds
->ds_phys
->ds_used_bytes
-= used
;
214 ASSERT3U(ds
->ds_phys
->ds_compressed_bytes
, >=, compressed
);
215 ds
->ds_phys
->ds_compressed_bytes
-= compressed
;
216 ASSERT3U(ds
->ds_phys
->ds_uncompressed_bytes
, >=, uncompressed
);
217 ds
->ds_phys
->ds_uncompressed_bytes
-= uncompressed
;
218 mutex_exit(&ds
->ds_lock
);
224 dsl_dataset_prev_snap_txg(dsl_dataset_t
*ds
)
226 uint64_t trysnap
= 0;
231 * The snapshot creation could fail, but that would cause an
232 * incorrect FALSE return, which would only result in an
233 * overestimation of the amount of space that an operation would
234 * consume, which is OK.
236 * There's also a small window where we could miss a pending
237 * snapshot, because we could set the sync task in the quiescing
238 * phase. So this should only be used as a guess.
240 if (ds
->ds_trysnap_txg
>
241 spa_last_synced_txg(ds
->ds_dir
->dd_pool
->dp_spa
))
242 trysnap
= ds
->ds_trysnap_txg
;
243 return (MAX(ds
->ds_phys
->ds_prev_snap_txg
, trysnap
));
247 dsl_dataset_block_freeable(dsl_dataset_t
*ds
, const blkptr_t
*bp
,
250 if (blk_birth
<= dsl_dataset_prev_snap_txg(ds
))
253 ddt_prefetch(dsl_dataset_get_spa(ds
), bp
);
260 dsl_dataset_evict(dmu_buf_t
*db
, void *dsv
)
262 dsl_dataset_t
*ds
= dsv
;
264 ASSERT(ds
->ds_owner
== NULL
|| DSL_DATASET_IS_DESTROYED(ds
));
266 unique_remove(ds
->ds_fsid_guid
);
268 if (ds
->ds_objset
!= NULL
)
269 dmu_objset_evict(ds
->ds_objset
);
272 dsl_dataset_drop_ref(ds
->ds_prev
, ds
);
276 bplist_destroy(&ds
->ds_pending_deadlist
);
278 dsl_deadlist_close(&ds
->ds_deadlist
);
280 ASSERT(ds
->ds_deadlist
.dl_dbuf
== NULL
);
281 ASSERT(!ds
->ds_deadlist
.dl_oldfmt
);
284 dsl_dir_close(ds
->ds_dir
, ds
);
286 ASSERT(!list_link_active(&ds
->ds_synced_link
));
288 mutex_destroy(&ds
->ds_lock
);
289 mutex_destroy(&ds
->ds_recvlock
);
290 mutex_destroy(&ds
->ds_opening_lock
);
291 rw_destroy(&ds
->ds_rwlock
);
292 cv_destroy(&ds
->ds_exclusive_cv
);
294 kmem_free(ds
, sizeof (dsl_dataset_t
));
298 dsl_dataset_get_snapname(dsl_dataset_t
*ds
)
300 dsl_dataset_phys_t
*headphys
;
303 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
304 objset_t
*mos
= dp
->dp_meta_objset
;
306 if (ds
->ds_snapname
[0])
308 if (ds
->ds_phys
->ds_next_snap_obj
== 0)
311 err
= dmu_bonus_hold(mos
, ds
->ds_dir
->dd_phys
->dd_head_dataset_obj
,
315 headphys
= headdbuf
->db_data
;
316 err
= zap_value_search(dp
->dp_meta_objset
,
317 headphys
->ds_snapnames_zapobj
, ds
->ds_object
, 0, ds
->ds_snapname
);
318 dmu_buf_rele(headdbuf
, FTAG
);
323 dsl_dataset_snap_lookup(dsl_dataset_t
*ds
, const char *name
, uint64_t *value
)
325 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
326 uint64_t snapobj
= ds
->ds_phys
->ds_snapnames_zapobj
;
330 if (ds
->ds_phys
->ds_flags
& DS_FLAG_CI_DATASET
)
335 err
= zap_lookup_norm(mos
, snapobj
, name
, 8, 1,
336 value
, mt
, NULL
, 0, NULL
);
337 if (err
== ENOTSUP
&& mt
== MT_FIRST
)
338 err
= zap_lookup(mos
, snapobj
, name
, 8, 1, value
);
343 dsl_dataset_snap_remove(dsl_dataset_t
*ds
, char *name
, dmu_tx_t
*tx
)
345 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
346 uint64_t snapobj
= ds
->ds_phys
->ds_snapnames_zapobj
;
350 dsl_dir_snap_cmtime_update(ds
->ds_dir
);
352 if (ds
->ds_phys
->ds_flags
& DS_FLAG_CI_DATASET
)
357 err
= zap_remove_norm(mos
, snapobj
, name
, mt
, tx
);
358 if (err
== ENOTSUP
&& mt
== MT_FIRST
)
359 err
= zap_remove(mos
, snapobj
, name
, tx
);
364 dsl_dataset_get_ref(dsl_pool_t
*dp
, uint64_t dsobj
, void *tag
,
367 objset_t
*mos
= dp
->dp_meta_objset
;
371 dmu_object_info_t doi
;
373 ASSERT(RW_LOCK_HELD(&dp
->dp_config_rwlock
) ||
374 dsl_pool_sync_context(dp
));
376 err
= dmu_bonus_hold(mos
, dsobj
, tag
, &dbuf
);
380 /* Make sure dsobj has the correct object type. */
381 dmu_object_info_from_db(dbuf
, &doi
);
382 if (doi
.doi_type
!= DMU_OT_DSL_DATASET
)
385 ds
= dmu_buf_get_user(dbuf
);
387 dsl_dataset_t
*winner
;
389 ds
= kmem_zalloc(sizeof (dsl_dataset_t
), KM_SLEEP
);
391 ds
->ds_object
= dsobj
;
392 ds
->ds_phys
= dbuf
->db_data
;
394 mutex_init(&ds
->ds_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
395 mutex_init(&ds
->ds_recvlock
, NULL
, MUTEX_DEFAULT
, NULL
);
396 mutex_init(&ds
->ds_opening_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
397 rw_init(&ds
->ds_rwlock
, 0, 0, 0);
398 cv_init(&ds
->ds_exclusive_cv
, NULL
, CV_DEFAULT
, NULL
);
400 bplist_create(&ds
->ds_pending_deadlist
);
401 dsl_deadlist_open(&ds
->ds_deadlist
,
402 mos
, ds
->ds_phys
->ds_deadlist_obj
);
405 err
= dsl_dir_open_obj(dp
,
406 ds
->ds_phys
->ds_dir_obj
, NULL
, ds
, &ds
->ds_dir
);
409 mutex_destroy(&ds
->ds_lock
);
410 mutex_destroy(&ds
->ds_recvlock
);
411 mutex_destroy(&ds
->ds_opening_lock
);
412 rw_destroy(&ds
->ds_rwlock
);
413 cv_destroy(&ds
->ds_exclusive_cv
);
414 bplist_destroy(&ds
->ds_pending_deadlist
);
415 dsl_deadlist_close(&ds
->ds_deadlist
);
416 kmem_free(ds
, sizeof (dsl_dataset_t
));
417 dmu_buf_rele(dbuf
, tag
);
421 if (!dsl_dataset_is_snapshot(ds
)) {
422 ds
->ds_snapname
[0] = '\0';
423 if (ds
->ds_phys
->ds_prev_snap_obj
) {
424 err
= dsl_dataset_get_ref(dp
,
425 ds
->ds_phys
->ds_prev_snap_obj
,
429 if (zfs_flags
& ZFS_DEBUG_SNAPNAMES
)
430 err
= dsl_dataset_get_snapname(ds
);
431 if (err
== 0 && ds
->ds_phys
->ds_userrefs_obj
!= 0) {
433 ds
->ds_dir
->dd_pool
->dp_meta_objset
,
434 ds
->ds_phys
->ds_userrefs_obj
,
439 if (err
== 0 && !dsl_dataset_is_snapshot(ds
)) {
441 * In sync context, we're called with either no lock
442 * or with the write lock. If we're not syncing,
443 * we're always called with the read lock held.
445 boolean_t need_lock
=
446 !RW_WRITE_HELD(&dp
->dp_config_rwlock
) &&
447 dsl_pool_sync_context(dp
);
450 rw_enter(&dp
->dp_config_rwlock
, RW_READER
);
452 err
= dsl_prop_get_ds(ds
,
453 "refreservation", sizeof (uint64_t), 1,
454 &ds
->ds_reserved
, NULL
);
456 err
= dsl_prop_get_ds(ds
,
457 "refquota", sizeof (uint64_t), 1,
458 &ds
->ds_quota
, NULL
);
462 rw_exit(&dp
->dp_config_rwlock
);
464 ds
->ds_reserved
= ds
->ds_quota
= 0;
468 winner
= dmu_buf_set_user_ie(dbuf
, ds
, &ds
->ds_phys
,
472 bplist_destroy(&ds
->ds_pending_deadlist
);
473 dsl_deadlist_close(&ds
->ds_deadlist
);
475 dsl_dataset_drop_ref(ds
->ds_prev
, ds
);
476 dsl_dir_close(ds
->ds_dir
, ds
);
477 mutex_destroy(&ds
->ds_lock
);
478 mutex_destroy(&ds
->ds_recvlock
);
479 mutex_destroy(&ds
->ds_opening_lock
);
480 rw_destroy(&ds
->ds_rwlock
);
481 cv_destroy(&ds
->ds_exclusive_cv
);
482 kmem_free(ds
, sizeof (dsl_dataset_t
));
484 dmu_buf_rele(dbuf
, tag
);
490 unique_insert(ds
->ds_phys
->ds_fsid_guid
);
493 ASSERT3P(ds
->ds_dbuf
, ==, dbuf
);
494 ASSERT3P(ds
->ds_phys
, ==, dbuf
->db_data
);
495 ASSERT(ds
->ds_phys
->ds_prev_snap_obj
!= 0 ||
496 spa_version(dp
->dp_spa
) < SPA_VERSION_ORIGIN
||
497 dp
->dp_origin_snap
== NULL
|| ds
== dp
->dp_origin_snap
);
498 mutex_enter(&ds
->ds_lock
);
499 if (!dsl_pool_sync_context(dp
) && DSL_DATASET_IS_DESTROYED(ds
)) {
500 mutex_exit(&ds
->ds_lock
);
501 dmu_buf_rele(ds
->ds_dbuf
, tag
);
504 mutex_exit(&ds
->ds_lock
);
510 dsl_dataset_hold_ref(dsl_dataset_t
*ds
, void *tag
)
512 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
515 * In syncing context we don't want the rwlock lock: there
516 * may be an existing writer waiting for sync phase to
517 * finish. We don't need to worry about such writers, since
518 * sync phase is single-threaded, so the writer can't be
519 * doing anything while we are active.
521 if (dsl_pool_sync_context(dp
)) {
522 ASSERT(!DSL_DATASET_IS_DESTROYED(ds
));
527 * Normal users will hold the ds_rwlock as a READER until they
528 * are finished (i.e., call dsl_dataset_rele()). "Owners" will
529 * drop their READER lock after they set the ds_owner field.
531 * If the dataset is being destroyed, the destroy thread will
532 * obtain a WRITER lock for exclusive access after it's done its
533 * open-context work and then change the ds_owner to
534 * dsl_reaper once destruction is assured. So threads
535 * may block here temporarily, until the "destructability" of
536 * the dataset is determined.
538 ASSERT(!RW_WRITE_HELD(&dp
->dp_config_rwlock
));
539 mutex_enter(&ds
->ds_lock
);
540 while (!rw_tryenter(&ds
->ds_rwlock
, RW_READER
)) {
541 rw_exit(&dp
->dp_config_rwlock
);
542 cv_wait(&ds
->ds_exclusive_cv
, &ds
->ds_lock
);
543 if (DSL_DATASET_IS_DESTROYED(ds
)) {
544 mutex_exit(&ds
->ds_lock
);
545 dsl_dataset_drop_ref(ds
, tag
);
546 rw_enter(&dp
->dp_config_rwlock
, RW_READER
);
550 * The dp_config_rwlock lives above the ds_lock. And
551 * we need to check DSL_DATASET_IS_DESTROYED() while
552 * holding the ds_lock, so we have to drop and reacquire
555 mutex_exit(&ds
->ds_lock
);
556 rw_enter(&dp
->dp_config_rwlock
, RW_READER
);
557 mutex_enter(&ds
->ds_lock
);
559 mutex_exit(&ds
->ds_lock
);
564 dsl_dataset_hold_obj(dsl_pool_t
*dp
, uint64_t dsobj
, void *tag
,
567 int err
= dsl_dataset_get_ref(dp
, dsobj
, tag
, dsp
);
571 return (dsl_dataset_hold_ref(*dsp
, tag
));
575 dsl_dataset_own_obj(dsl_pool_t
*dp
, uint64_t dsobj
, boolean_t inconsistentok
,
576 void *tag
, dsl_dataset_t
**dsp
)
578 int err
= dsl_dataset_hold_obj(dp
, dsobj
, tag
, dsp
);
581 if (!dsl_dataset_tryown(*dsp
, inconsistentok
, tag
)) {
582 dsl_dataset_rele(*dsp
, tag
);
590 dsl_dataset_hold(const char *name
, void *tag
, dsl_dataset_t
**dsp
)
594 const char *snapname
;
598 err
= dsl_dir_open_spa(NULL
, name
, FTAG
, &dd
, &snapname
);
603 obj
= dd
->dd_phys
->dd_head_dataset_obj
;
604 rw_enter(&dp
->dp_config_rwlock
, RW_READER
);
606 err
= dsl_dataset_get_ref(dp
, obj
, tag
, dsp
);
612 err
= dsl_dataset_hold_ref(*dsp
, tag
);
614 /* we may be looking for a snapshot */
615 if (err
== 0 && snapname
!= NULL
) {
616 dsl_dataset_t
*ds
= NULL
;
618 if (*snapname
++ != '@') {
619 dsl_dataset_rele(*dsp
, tag
);
624 dprintf("looking for snapshot '%s'\n", snapname
);
625 err
= dsl_dataset_snap_lookup(*dsp
, snapname
, &obj
);
627 err
= dsl_dataset_get_ref(dp
, obj
, tag
, &ds
);
628 dsl_dataset_rele(*dsp
, tag
);
630 ASSERT3U((err
== 0), ==, (ds
!= NULL
));
633 mutex_enter(&ds
->ds_lock
);
634 if (ds
->ds_snapname
[0] == 0)
635 (void) strlcpy(ds
->ds_snapname
, snapname
,
636 sizeof (ds
->ds_snapname
));
637 mutex_exit(&ds
->ds_lock
);
638 err
= dsl_dataset_hold_ref(ds
, tag
);
639 *dsp
= err
? NULL
: ds
;
643 rw_exit(&dp
->dp_config_rwlock
);
644 dsl_dir_close(dd
, FTAG
);
649 dsl_dataset_own(const char *name
, boolean_t inconsistentok
,
650 void *tag
, dsl_dataset_t
**dsp
)
652 int err
= dsl_dataset_hold(name
, tag
, dsp
);
655 if (!dsl_dataset_tryown(*dsp
, inconsistentok
, tag
)) {
656 dsl_dataset_rele(*dsp
, tag
);
663 dsl_dataset_name(dsl_dataset_t
*ds
, char *name
)
666 (void) strcpy(name
, "mos");
668 dsl_dir_name(ds
->ds_dir
, name
);
669 VERIFY(0 == dsl_dataset_get_snapname(ds
));
670 if (ds
->ds_snapname
[0]) {
671 (void) strcat(name
, "@");
673 * We use a "recursive" mutex so that we
674 * can call dprintf_ds() with ds_lock held.
676 if (!MUTEX_HELD(&ds
->ds_lock
)) {
677 mutex_enter(&ds
->ds_lock
);
678 (void) strcat(name
, ds
->ds_snapname
);
679 mutex_exit(&ds
->ds_lock
);
681 (void) strcat(name
, ds
->ds_snapname
);
688 dsl_dataset_namelen(dsl_dataset_t
*ds
)
693 result
= 3; /* "mos" */
695 result
= dsl_dir_namelen(ds
->ds_dir
);
696 VERIFY(0 == dsl_dataset_get_snapname(ds
));
697 if (ds
->ds_snapname
[0]) {
698 ++result
; /* adding one for the @-sign */
699 if (!MUTEX_HELD(&ds
->ds_lock
)) {
700 mutex_enter(&ds
->ds_lock
);
701 result
+= strlen(ds
->ds_snapname
);
702 mutex_exit(&ds
->ds_lock
);
704 result
+= strlen(ds
->ds_snapname
);
713 dsl_dataset_drop_ref(dsl_dataset_t
*ds
, void *tag
)
715 dmu_buf_rele(ds
->ds_dbuf
, tag
);
719 dsl_dataset_rele(dsl_dataset_t
*ds
, void *tag
)
721 if (!dsl_pool_sync_context(ds
->ds_dir
->dd_pool
)) {
722 rw_exit(&ds
->ds_rwlock
);
724 dsl_dataset_drop_ref(ds
, tag
);
728 dsl_dataset_disown(dsl_dataset_t
*ds
, void *tag
)
730 ASSERT((ds
->ds_owner
== tag
&& ds
->ds_dbuf
) ||
731 (DSL_DATASET_IS_DESTROYED(ds
) && ds
->ds_dbuf
== NULL
));
733 mutex_enter(&ds
->ds_lock
);
735 if (RW_WRITE_HELD(&ds
->ds_rwlock
)) {
736 rw_exit(&ds
->ds_rwlock
);
737 cv_broadcast(&ds
->ds_exclusive_cv
);
739 mutex_exit(&ds
->ds_lock
);
741 dsl_dataset_drop_ref(ds
, tag
);
743 dsl_dataset_evict(NULL
, ds
);
747 dsl_dataset_tryown(dsl_dataset_t
*ds
, boolean_t inconsistentok
, void *tag
)
749 boolean_t gotit
= FALSE
;
751 mutex_enter(&ds
->ds_lock
);
752 if (ds
->ds_owner
== NULL
&&
753 (!DS_IS_INCONSISTENT(ds
) || inconsistentok
)) {
755 if (!dsl_pool_sync_context(ds
->ds_dir
->dd_pool
))
756 rw_exit(&ds
->ds_rwlock
);
759 mutex_exit(&ds
->ds_lock
);
764 dsl_dataset_make_exclusive(dsl_dataset_t
*ds
, void *owner
)
766 ASSERT3P(owner
, ==, ds
->ds_owner
);
767 if (!RW_WRITE_HELD(&ds
->ds_rwlock
))
768 rw_enter(&ds
->ds_rwlock
, RW_WRITER
);
772 dsl_dataset_create_sync_dd(dsl_dir_t
*dd
, dsl_dataset_t
*origin
,
773 uint64_t flags
, dmu_tx_t
*tx
)
775 dsl_pool_t
*dp
= dd
->dd_pool
;
777 dsl_dataset_phys_t
*dsphys
;
779 objset_t
*mos
= dp
->dp_meta_objset
;
782 origin
= dp
->dp_origin_snap
;
784 ASSERT(origin
== NULL
|| origin
->ds_dir
->dd_pool
== dp
);
785 ASSERT(origin
== NULL
|| origin
->ds_phys
->ds_num_children
> 0);
786 ASSERT(dmu_tx_is_syncing(tx
));
787 ASSERT(dd
->dd_phys
->dd_head_dataset_obj
== 0);
789 dsobj
= dmu_object_alloc(mos
, DMU_OT_DSL_DATASET
, 0,
790 DMU_OT_DSL_DATASET
, sizeof (dsl_dataset_phys_t
), tx
);
791 VERIFY(0 == dmu_bonus_hold(mos
, dsobj
, FTAG
, &dbuf
));
792 dmu_buf_will_dirty(dbuf
, tx
);
793 dsphys
= dbuf
->db_data
;
794 bzero(dsphys
, sizeof (dsl_dataset_phys_t
));
795 dsphys
->ds_dir_obj
= dd
->dd_object
;
796 dsphys
->ds_flags
= flags
;
797 dsphys
->ds_fsid_guid
= unique_create();
798 (void) random_get_pseudo_bytes((void*)&dsphys
->ds_guid
,
799 sizeof (dsphys
->ds_guid
));
800 dsphys
->ds_snapnames_zapobj
=
801 zap_create_norm(mos
, U8_TEXTPREP_TOUPPER
, DMU_OT_DSL_DS_SNAP_MAP
,
803 dsphys
->ds_creation_time
= gethrestime_sec();
804 dsphys
->ds_creation_txg
= tx
->tx_txg
== TXG_INITIAL
? 1 : tx
->tx_txg
;
806 if (origin
== NULL
) {
807 dsphys
->ds_deadlist_obj
= dsl_deadlist_alloc(mos
, tx
);
811 dsphys
->ds_prev_snap_obj
= origin
->ds_object
;
812 dsphys
->ds_prev_snap_txg
=
813 origin
->ds_phys
->ds_creation_txg
;
814 dsphys
->ds_used_bytes
=
815 origin
->ds_phys
->ds_used_bytes
;
816 dsphys
->ds_compressed_bytes
=
817 origin
->ds_phys
->ds_compressed_bytes
;
818 dsphys
->ds_uncompressed_bytes
=
819 origin
->ds_phys
->ds_uncompressed_bytes
;
820 dsphys
->ds_bp
= origin
->ds_phys
->ds_bp
;
821 dsphys
->ds_flags
|= origin
->ds_phys
->ds_flags
;
823 dmu_buf_will_dirty(origin
->ds_dbuf
, tx
);
824 origin
->ds_phys
->ds_num_children
++;
826 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp
,
827 origin
->ds_dir
->dd_phys
->dd_head_dataset_obj
, FTAG
, &ohds
));
828 dsphys
->ds_deadlist_obj
= dsl_deadlist_clone(&ohds
->ds_deadlist
,
829 dsphys
->ds_prev_snap_txg
, dsphys
->ds_prev_snap_obj
, tx
);
830 dsl_dataset_rele(ohds
, FTAG
);
832 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_NEXT_CLONES
) {
833 if (origin
->ds_phys
->ds_next_clones_obj
== 0) {
834 origin
->ds_phys
->ds_next_clones_obj
=
836 DMU_OT_NEXT_CLONES
, DMU_OT_NONE
, 0, tx
);
838 VERIFY(0 == zap_add_int(mos
,
839 origin
->ds_phys
->ds_next_clones_obj
,
843 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
844 dd
->dd_phys
->dd_origin_obj
= origin
->ds_object
;
845 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
846 if (origin
->ds_dir
->dd_phys
->dd_clones
== 0) {
847 dmu_buf_will_dirty(origin
->ds_dir
->dd_dbuf
, tx
);
848 origin
->ds_dir
->dd_phys
->dd_clones
=
850 DMU_OT_DSL_CLONES
, DMU_OT_NONE
, 0, tx
);
852 VERIFY3U(0, ==, zap_add_int(mos
,
853 origin
->ds_dir
->dd_phys
->dd_clones
, dsobj
, tx
));
857 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_UNIQUE_ACCURATE
)
858 dsphys
->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
860 dmu_buf_rele(dbuf
, FTAG
);
862 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
863 dd
->dd_phys
->dd_head_dataset_obj
= dsobj
;
869 dsl_dataset_create_sync(dsl_dir_t
*pdd
, const char *lastname
,
870 dsl_dataset_t
*origin
, uint64_t flags
, cred_t
*cr
, dmu_tx_t
*tx
)
872 dsl_pool_t
*dp
= pdd
->dd_pool
;
873 uint64_t dsobj
, ddobj
;
876 ASSERT(lastname
[0] != '@');
878 ddobj
= dsl_dir_create_sync(dp
, pdd
, lastname
, tx
);
879 VERIFY(0 == dsl_dir_open_obj(dp
, ddobj
, lastname
, FTAG
, &dd
));
881 dsobj
= dsl_dataset_create_sync_dd(dd
, origin
, flags
, tx
);
883 dsl_deleg_set_create_perms(dd
, tx
, cr
);
885 dsl_dir_close(dd
, FTAG
);
888 * If we are creating a clone, make sure we zero out any stale
889 * data from the origin snapshots zil header.
891 if (origin
!= NULL
) {
895 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp
, dsobj
, FTAG
, &ds
));
896 VERIFY3U(0, ==, dmu_objset_from_ds(ds
, &os
));
897 bzero(&os
->os_zil_header
, sizeof (os
->os_zil_header
));
898 dsl_dataset_dirty(ds
, tx
);
899 dsl_dataset_rele(ds
, FTAG
);
906 * The snapshots must all be in the same pool.
909 dmu_snapshots_destroy_nvl(nvlist_t
*snaps
, boolean_t defer
, char *failed
)
912 dsl_sync_task_t
*dst
;
915 dsl_sync_task_group_t
*dstg
;
917 pair
= nvlist_next_nvpair(snaps
, NULL
);
921 err
= spa_open(nvpair_name(pair
), &spa
, FTAG
);
924 dstg
= dsl_sync_task_group_create(spa_get_dsl(spa
));
926 for (pair
= nvlist_next_nvpair(snaps
, NULL
); pair
!= NULL
;
927 pair
= nvlist_next_nvpair(snaps
, pair
)) {
931 err
= dsl_dataset_own(nvpair_name(pair
), B_TRUE
, dstg
, &ds
);
933 struct dsl_ds_destroyarg
*dsda
;
935 dsl_dataset_make_exclusive(ds
, dstg
);
936 dsda
= kmem_zalloc(sizeof (struct dsl_ds_destroyarg
),
940 dsl_sync_task_create(dstg
, dsl_dataset_destroy_check
,
941 dsl_dataset_destroy_sync
, dsda
, dstg
, 0);
942 } else if (err
== ENOENT
) {
945 (void) strcpy(failed
, nvpair_name(pair
));
951 err
= dsl_sync_task_group_wait(dstg
);
953 for (dst
= list_head(&dstg
->dstg_tasks
); dst
;
954 dst
= list_next(&dstg
->dstg_tasks
, dst
)) {
955 struct dsl_ds_destroyarg
*dsda
= dst
->dst_arg1
;
956 dsl_dataset_t
*ds
= dsda
->ds
;
959 * Return the file system name that triggered the error
962 dsl_dataset_name(ds
, failed
);
964 ASSERT3P(dsda
->rm_origin
, ==, NULL
);
965 dsl_dataset_disown(ds
, dstg
);
966 kmem_free(dsda
, sizeof (struct dsl_ds_destroyarg
));
969 dsl_sync_task_group_destroy(dstg
);
970 spa_close(spa
, FTAG
);
976 dsl_dataset_might_destroy_origin(dsl_dataset_t
*ds
)
978 boolean_t might_destroy
= B_FALSE
;
980 mutex_enter(&ds
->ds_lock
);
981 if (ds
->ds_phys
->ds_num_children
== 2 && ds
->ds_userrefs
== 0 &&
982 DS_IS_DEFER_DESTROY(ds
))
983 might_destroy
= B_TRUE
;
984 mutex_exit(&ds
->ds_lock
);
986 return (might_destroy
);
990 * If we're removing a clone, and these three conditions are true:
991 * 1) the clone's origin has no other children
992 * 2) the clone's origin has no user references
993 * 3) the clone's origin has been marked for deferred destruction
994 * Then, prepare to remove the origin as part of this sync task group.
997 dsl_dataset_origin_rm_prep(struct dsl_ds_destroyarg
*dsda
, void *tag
)
999 dsl_dataset_t
*ds
= dsda
->ds
;
1000 dsl_dataset_t
*origin
= ds
->ds_prev
;
1002 if (dsl_dataset_might_destroy_origin(origin
)) {
1007 namelen
= dsl_dataset_namelen(origin
) + 1;
1008 name
= kmem_alloc(namelen
, KM_SLEEP
);
1009 dsl_dataset_name(origin
, name
);
1011 error
= zfs_unmount_snap(name
, NULL
);
1013 kmem_free(name
, namelen
);
1017 error
= dsl_dataset_own(name
, B_TRUE
, tag
, &origin
);
1018 kmem_free(name
, namelen
);
1021 dsda
->rm_origin
= origin
;
1022 dsl_dataset_make_exclusive(origin
, tag
);
1029 * ds must be opened as OWNER. On return (whether successful or not),
1030 * ds will be closed and caller can no longer dereference it.
1033 dsl_dataset_destroy(dsl_dataset_t
*ds
, void *tag
, boolean_t defer
)
1036 dsl_sync_task_group_t
*dstg
;
1040 struct dsl_ds_destroyarg dsda
= { 0 };
1041 dsl_dataset_t dummy_ds
= { 0 };
1045 if (dsl_dataset_is_snapshot(ds
)) {
1046 /* Destroying a snapshot is simpler */
1047 dsl_dataset_make_exclusive(ds
, tag
);
1050 err
= dsl_sync_task_do(ds
->ds_dir
->dd_pool
,
1051 dsl_dataset_destroy_check
, dsl_dataset_destroy_sync
,
1053 ASSERT3P(dsda
.rm_origin
, ==, NULL
);
1061 dummy_ds
.ds_dir
= dd
;
1062 dummy_ds
.ds_object
= ds
->ds_object
;
1065 * Check for errors and mark this ds as inconsistent, in
1066 * case we crash while freeing the objects.
1068 err
= dsl_sync_task_do(dd
->dd_pool
, dsl_dataset_destroy_begin_check
,
1069 dsl_dataset_destroy_begin_sync
, ds
, NULL
, 0);
1073 err
= dmu_objset_from_ds(ds
, &os
);
1078 * remove the objects in open context, so that we won't
1079 * have too much to do in syncing context.
1081 for (obj
= 0; err
== 0; err
= dmu_object_next(os
, &obj
, FALSE
,
1082 ds
->ds_phys
->ds_prev_snap_txg
)) {
1084 * Ignore errors, if there is not enough disk space
1085 * we will deal with it in dsl_dataset_destroy_sync().
1087 (void) dmu_free_object(os
, obj
);
1093 * Only the ZIL knows how to free log blocks.
1095 zil_destroy(dmu_objset_zil(os
), B_FALSE
);
1098 * Sync out all in-flight IO.
1100 txg_wait_synced(dd
->dd_pool
, 0);
1103 * If we managed to free all the objects in open
1104 * context, the user space accounting should be zero.
1106 if (ds
->ds_phys
->ds_bp
.blk_fill
== 0 &&
1107 dmu_objset_userused_enabled(os
)) {
1110 ASSERT(zap_count(os
, DMU_USERUSED_OBJECT
, &count
) != 0 ||
1112 ASSERT(zap_count(os
, DMU_GROUPUSED_OBJECT
, &count
) != 0 ||
1116 rw_enter(&dd
->dd_pool
->dp_config_rwlock
, RW_READER
);
1117 err
= dsl_dir_open_obj(dd
->dd_pool
, dd
->dd_object
, NULL
, FTAG
, &dd
);
1118 rw_exit(&dd
->dd_pool
->dp_config_rwlock
);
1124 * Blow away the dsl_dir + head dataset.
1126 dsl_dataset_make_exclusive(ds
, tag
);
1128 * If we're removing a clone, we might also need to remove its
1132 dsda
.need_prep
= B_FALSE
;
1133 if (dsl_dir_is_clone(dd
)) {
1134 err
= dsl_dataset_origin_rm_prep(&dsda
, tag
);
1136 dsl_dir_close(dd
, FTAG
);
1141 dstg
= dsl_sync_task_group_create(ds
->ds_dir
->dd_pool
);
1142 dsl_sync_task_create(dstg
, dsl_dataset_destroy_check
,
1143 dsl_dataset_destroy_sync
, &dsda
, tag
, 0);
1144 dsl_sync_task_create(dstg
, dsl_dir_destroy_check
,
1145 dsl_dir_destroy_sync
, &dummy_ds
, FTAG
, 0);
1146 err
= dsl_sync_task_group_wait(dstg
);
1147 dsl_sync_task_group_destroy(dstg
);
1150 * We could be racing against 'zfs release' or 'zfs destroy -d'
1151 * on the origin snap, in which case we can get EBUSY if we
1152 * needed to destroy the origin snap but were not ready to
1155 if (dsda
.need_prep
) {
1156 ASSERT(err
== EBUSY
);
1157 ASSERT(dsl_dir_is_clone(dd
));
1158 ASSERT(dsda
.rm_origin
== NULL
);
1160 } while (dsda
.need_prep
);
1162 if (dsda
.rm_origin
!= NULL
)
1163 dsl_dataset_disown(dsda
.rm_origin
, tag
);
1165 /* if it is successful, dsl_dir_destroy_sync will close the dd */
1167 dsl_dir_close(dd
, FTAG
);
1169 dsl_dataset_disown(ds
, tag
);
1174 dsl_dataset_get_blkptr(dsl_dataset_t
*ds
)
1176 return (&ds
->ds_phys
->ds_bp
);
1180 dsl_dataset_set_blkptr(dsl_dataset_t
*ds
, blkptr_t
*bp
, dmu_tx_t
*tx
)
1182 ASSERT(dmu_tx_is_syncing(tx
));
1183 /* If it's the meta-objset, set dp_meta_rootbp */
1185 tx
->tx_pool
->dp_meta_rootbp
= *bp
;
1187 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1188 ds
->ds_phys
->ds_bp
= *bp
;
1193 dsl_dataset_get_spa(dsl_dataset_t
*ds
)
1195 return (ds
->ds_dir
->dd_pool
->dp_spa
);
1199 dsl_dataset_dirty(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
1203 if (ds
== NULL
) /* this is the meta-objset */
1206 ASSERT(ds
->ds_objset
!= NULL
);
1208 if (ds
->ds_phys
->ds_next_snap_obj
!= 0)
1209 panic("dirtying snapshot!");
1211 dp
= ds
->ds_dir
->dd_pool
;
1213 if (txg_list_add(&dp
->dp_dirty_datasets
, ds
, tx
->tx_txg
) == 0) {
1214 /* up the hold count until we can be written out */
1215 dmu_buf_add_ref(ds
->ds_dbuf
, ds
);
1220 * The unique space in the head dataset can be calculated by subtracting
1221 * the space used in the most recent snapshot, that is still being used
1222 * in this file system, from the space currently in use. To figure out
1223 * the space in the most recent snapshot still in use, we need to take
1224 * the total space used in the snapshot and subtract out the space that
1225 * has been freed up since the snapshot was taken.
1228 dsl_dataset_recalc_head_uniq(dsl_dataset_t
*ds
)
1231 uint64_t dlused
, dlcomp
, dluncomp
;
1233 ASSERT(!dsl_dataset_is_snapshot(ds
));
1235 if (ds
->ds_phys
->ds_prev_snap_obj
!= 0)
1236 mrs_used
= ds
->ds_prev
->ds_phys
->ds_used_bytes
;
1240 dsl_deadlist_space(&ds
->ds_deadlist
, &dlused
, &dlcomp
, &dluncomp
);
1242 ASSERT3U(dlused
, <=, mrs_used
);
1243 ds
->ds_phys
->ds_unique_bytes
=
1244 ds
->ds_phys
->ds_used_bytes
- (mrs_used
- dlused
);
1246 if (spa_version(ds
->ds_dir
->dd_pool
->dp_spa
) >=
1247 SPA_VERSION_UNIQUE_ACCURATE
)
1248 ds
->ds_phys
->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
1258 kill_blkptr(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
, arc_buf_t
*pbuf
,
1259 const zbookmark_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
1261 struct killarg
*ka
= arg
;
1262 dmu_tx_t
*tx
= ka
->tx
;
1267 if (zb
->zb_level
== ZB_ZIL_LEVEL
) {
1268 ASSERT(zilog
!= NULL
);
1270 * It's a block in the intent log. It has no
1271 * accounting, so just free it.
1273 dsl_free(ka
->tx
->tx_pool
, ka
->tx
->tx_txg
, bp
);
1275 ASSERT(zilog
== NULL
);
1276 ASSERT3U(bp
->blk_birth
, >, ka
->ds
->ds_phys
->ds_prev_snap_txg
);
1277 (void) dsl_dataset_block_kill(ka
->ds
, bp
, tx
, B_FALSE
);
1285 dsl_dataset_destroy_begin_check(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
1287 dsl_dataset_t
*ds
= arg1
;
1288 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
1293 * Can't delete a head dataset if there are snapshots of it.
1294 * (Except if the only snapshots are from the branch we cloned
1297 if (ds
->ds_prev
!= NULL
&&
1298 ds
->ds_prev
->ds_phys
->ds_next_snap_obj
== ds
->ds_object
)
1302 * This is really a dsl_dir thing, but check it here so that
1303 * we'll be less likely to leave this dataset inconsistent &
1306 err
= zap_count(mos
, ds
->ds_dir
->dd_phys
->dd_child_dir_zapobj
, &count
);
1317 dsl_dataset_destroy_begin_sync(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
1319 dsl_dataset_t
*ds
= arg1
;
1320 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1322 /* Mark it as inconsistent on-disk, in case we crash */
1323 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1324 ds
->ds_phys
->ds_flags
|= DS_FLAG_INCONSISTENT
;
1326 spa_history_log_internal(LOG_DS_DESTROY_BEGIN
, dp
->dp_spa
, tx
,
1327 "dataset = %llu", ds
->ds_object
);
1331 dsl_dataset_origin_check(struct dsl_ds_destroyarg
*dsda
, void *tag
,
1334 dsl_dataset_t
*ds
= dsda
->ds
;
1335 dsl_dataset_t
*ds_prev
= ds
->ds_prev
;
1337 if (dsl_dataset_might_destroy_origin(ds_prev
)) {
1338 struct dsl_ds_destroyarg ndsda
= {0};
1341 * If we're not prepared to remove the origin, don't remove
1344 if (dsda
->rm_origin
== NULL
) {
1345 dsda
->need_prep
= B_TRUE
;
1350 ndsda
.is_origin_rm
= B_TRUE
;
1351 return (dsl_dataset_destroy_check(&ndsda
, tag
, tx
));
1355 * If we're not going to remove the origin after all,
1356 * undo the open context setup.
1358 if (dsda
->rm_origin
!= NULL
) {
1359 dsl_dataset_disown(dsda
->rm_origin
, tag
);
1360 dsda
->rm_origin
= NULL
;
1367 * If you add new checks here, you may need to add
1368 * additional checks to the "temporary" case in
1369 * snapshot_check() in dmu_objset.c.
1373 dsl_dataset_destroy_check(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
1375 struct dsl_ds_destroyarg
*dsda
= arg1
;
1376 dsl_dataset_t
*ds
= dsda
->ds
;
1378 /* we have an owner hold, so noone else can destroy us */
1379 ASSERT(!DSL_DATASET_IS_DESTROYED(ds
));
1382 * Only allow deferred destroy on pools that support it.
1383 * NOTE: deferred destroy is only supported on snapshots.
1386 if (spa_version(ds
->ds_dir
->dd_pool
->dp_spa
) <
1387 SPA_VERSION_USERREFS
)
1389 ASSERT(dsl_dataset_is_snapshot(ds
));
1394 * Can't delete a head dataset if there are snapshots of it.
1395 * (Except if the only snapshots are from the branch we cloned
1398 if (ds
->ds_prev
!= NULL
&&
1399 ds
->ds_prev
->ds_phys
->ds_next_snap_obj
== ds
->ds_object
)
1403 * If we made changes this txg, traverse_dsl_dataset won't find
1406 if (ds
->ds_phys
->ds_bp
.blk_birth
>= tx
->tx_txg
)
1409 if (dsl_dataset_is_snapshot(ds
)) {
1411 * If this snapshot has an elevated user reference count,
1412 * we can't destroy it yet.
1414 if (ds
->ds_userrefs
> 0 && !dsda
->releasing
)
1417 mutex_enter(&ds
->ds_lock
);
1419 * Can't delete a branch point. However, if we're destroying
1420 * a clone and removing its origin due to it having a user
1421 * hold count of 0 and having been marked for deferred destroy,
1422 * it's OK for the origin to have a single clone.
1424 if (ds
->ds_phys
->ds_num_children
>
1425 (dsda
->is_origin_rm
? 2 : 1)) {
1426 mutex_exit(&ds
->ds_lock
);
1429 mutex_exit(&ds
->ds_lock
);
1430 } else if (dsl_dir_is_clone(ds
->ds_dir
)) {
1431 return (dsl_dataset_origin_check(dsda
, arg2
, tx
));
1434 /* XXX we should do some i/o error checking... */
1446 dsl_dataset_refs_gone(dmu_buf_t
*db
, void *argv
)
1448 struct refsarg
*arg
= argv
;
1450 mutex_enter(&arg
->lock
);
1452 cv_signal(&arg
->cv
);
1453 mutex_exit(&arg
->lock
);
1457 dsl_dataset_drain_refs(dsl_dataset_t
*ds
, void *tag
)
1461 mutex_init(&arg
.lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1462 cv_init(&arg
.cv
, NULL
, CV_DEFAULT
, NULL
);
1464 (void) dmu_buf_update_user(ds
->ds_dbuf
, ds
, &arg
, &ds
->ds_phys
,
1465 dsl_dataset_refs_gone
);
1466 dmu_buf_rele(ds
->ds_dbuf
, tag
);
1467 mutex_enter(&arg
.lock
);
1469 cv_wait(&arg
.cv
, &arg
.lock
);
1471 mutex_exit(&arg
.lock
);
1474 mutex_destroy(&arg
.lock
);
1475 cv_destroy(&arg
.cv
);
1479 remove_from_next_clones(dsl_dataset_t
*ds
, uint64_t obj
, dmu_tx_t
*tx
)
1481 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
1485 ASSERT(ds
->ds_phys
->ds_num_children
>= 2);
1486 err
= zap_remove_int(mos
, ds
->ds_phys
->ds_next_clones_obj
, obj
, tx
);
1488 * The err should not be ENOENT, but a bug in a previous version
1489 * of the code could cause upgrade_clones_cb() to not set
1490 * ds_next_snap_obj when it should, leading to a missing entry.
1491 * If we knew that the pool was created after
1492 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1493 * ENOENT. However, at least we can check that we don't have
1494 * too many entries in the next_clones_obj even after failing to
1497 if (err
!= ENOENT
) {
1498 VERIFY3U(err
, ==, 0);
1500 ASSERT3U(0, ==, zap_count(mos
, ds
->ds_phys
->ds_next_clones_obj
,
1502 ASSERT3U(count
, <=, ds
->ds_phys
->ds_num_children
- 2);
1506 dsl_dataset_remove_clones_key(dsl_dataset_t
*ds
, uint64_t mintxg
, dmu_tx_t
*tx
)
1508 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
1513 * If it is the old version, dd_clones doesn't exist so we can't
1514 * find the clones, but deadlist_remove_key() is a no-op so it
1517 if (ds
->ds_dir
->dd_phys
->dd_clones
== 0)
1520 for (zap_cursor_init(&zc
, mos
, ds
->ds_dir
->dd_phys
->dd_clones
);
1521 zap_cursor_retrieve(&zc
, &za
) == 0;
1522 zap_cursor_advance(&zc
)) {
1523 dsl_dataset_t
*clone
;
1525 VERIFY3U(0, ==, dsl_dataset_hold_obj(ds
->ds_dir
->dd_pool
,
1526 za
.za_first_integer
, FTAG
, &clone
));
1527 if (clone
->ds_dir
->dd_origin_txg
> mintxg
) {
1528 dsl_deadlist_remove_key(&clone
->ds_deadlist
,
1530 dsl_dataset_remove_clones_key(clone
, mintxg
, tx
);
1532 dsl_dataset_rele(clone
, FTAG
);
1534 zap_cursor_fini(&zc
);
1537 struct process_old_arg
{
1539 dsl_dataset_t
*ds_prev
;
1540 boolean_t after_branch_point
;
1542 uint64_t used
, comp
, uncomp
;
1546 process_old_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
1548 struct process_old_arg
*poa
= arg
;
1549 dsl_pool_t
*dp
= poa
->ds
->ds_dir
->dd_pool
;
1551 if (bp
->blk_birth
<= poa
->ds
->ds_phys
->ds_prev_snap_txg
) {
1552 dsl_deadlist_insert(&poa
->ds
->ds_deadlist
, bp
, tx
);
1553 if (poa
->ds_prev
&& !poa
->after_branch_point
&&
1555 poa
->ds_prev
->ds_phys
->ds_prev_snap_txg
) {
1556 poa
->ds_prev
->ds_phys
->ds_unique_bytes
+=
1557 bp_get_dsize_sync(dp
->dp_spa
, bp
);
1560 poa
->used
+= bp_get_dsize_sync(dp
->dp_spa
, bp
);
1561 poa
->comp
+= BP_GET_PSIZE(bp
);
1562 poa
->uncomp
+= BP_GET_UCSIZE(bp
);
1563 dsl_free_sync(poa
->pio
, dp
, tx
->tx_txg
, bp
);
1569 process_old_deadlist(dsl_dataset_t
*ds
, dsl_dataset_t
*ds_prev
,
1570 dsl_dataset_t
*ds_next
, boolean_t after_branch_point
, dmu_tx_t
*tx
)
1572 struct process_old_arg poa
= { 0 };
1573 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1574 objset_t
*mos
= dp
->dp_meta_objset
;
1576 ASSERT(ds
->ds_deadlist
.dl_oldfmt
);
1577 ASSERT(ds_next
->ds_deadlist
.dl_oldfmt
);
1580 poa
.ds_prev
= ds_prev
;
1581 poa
.after_branch_point
= after_branch_point
;
1582 poa
.pio
= zio_root(dp
->dp_spa
, NULL
, NULL
, ZIO_FLAG_MUSTSUCCEED
);
1583 VERIFY3U(0, ==, bpobj_iterate(&ds_next
->ds_deadlist
.dl_bpobj
,
1584 process_old_cb
, &poa
, tx
));
1585 VERIFY3U(zio_wait(poa
.pio
), ==, 0);
1586 ASSERT3U(poa
.used
, ==, ds
->ds_phys
->ds_unique_bytes
);
1588 /* change snapused */
1589 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_SNAP
,
1590 -poa
.used
, -poa
.comp
, -poa
.uncomp
, tx
);
1592 /* swap next's deadlist to our deadlist */
1593 dsl_deadlist_close(&ds
->ds_deadlist
);
1594 dsl_deadlist_close(&ds_next
->ds_deadlist
);
1595 SWITCH64(ds_next
->ds_phys
->ds_deadlist_obj
,
1596 ds
->ds_phys
->ds_deadlist_obj
);
1597 dsl_deadlist_open(&ds
->ds_deadlist
, mos
, ds
->ds_phys
->ds_deadlist_obj
);
1598 dsl_deadlist_open(&ds_next
->ds_deadlist
, mos
,
1599 ds_next
->ds_phys
->ds_deadlist_obj
);
1603 dsl_dataset_destroy_sync(void *arg1
, void *tag
, dmu_tx_t
*tx
)
1605 struct dsl_ds_destroyarg
*dsda
= arg1
;
1606 dsl_dataset_t
*ds
= dsda
->ds
;
1608 int after_branch_point
= FALSE
;
1609 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1610 objset_t
*mos
= dp
->dp_meta_objset
;
1611 dsl_dataset_t
*ds_prev
= NULL
;
1612 boolean_t wont_destroy
;
1615 wont_destroy
= (dsda
->defer
&&
1616 (ds
->ds_userrefs
> 0 || ds
->ds_phys
->ds_num_children
> 1));
1618 ASSERT(ds
->ds_owner
|| wont_destroy
);
1619 ASSERT(dsda
->defer
|| ds
->ds_phys
->ds_num_children
<= 1);
1620 ASSERT(ds
->ds_prev
== NULL
||
1621 ds
->ds_prev
->ds_phys
->ds_next_snap_obj
!= ds
->ds_object
);
1622 ASSERT3U(ds
->ds_phys
->ds_bp
.blk_birth
, <=, tx
->tx_txg
);
1625 ASSERT(spa_version(dp
->dp_spa
) >= SPA_VERSION_USERREFS
);
1626 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1627 ds
->ds_phys
->ds_flags
|= DS_FLAG_DEFER_DESTROY
;
1631 /* signal any waiters that this dataset is going away */
1632 mutex_enter(&ds
->ds_lock
);
1633 ds
->ds_owner
= dsl_reaper
;
1634 cv_broadcast(&ds
->ds_exclusive_cv
);
1635 mutex_exit(&ds
->ds_lock
);
1637 /* Remove our reservation */
1638 if (ds
->ds_reserved
!= 0) {
1639 dsl_prop_setarg_t psa
;
1642 dsl_prop_setarg_init_uint64(&psa
, "refreservation",
1643 (ZPROP_SRC_NONE
| ZPROP_SRC_LOCAL
| ZPROP_SRC_RECEIVED
),
1645 psa
.psa_effective_value
= 0; /* predict default value */
1647 dsl_dataset_set_reservation_sync(ds
, &psa
, tx
);
1648 ASSERT3U(ds
->ds_reserved
, ==, 0);
1651 ASSERT(RW_WRITE_HELD(&dp
->dp_config_rwlock
));
1653 dsl_scan_ds_destroyed(ds
, tx
);
1655 obj
= ds
->ds_object
;
1657 if (ds
->ds_phys
->ds_prev_snap_obj
!= 0) {
1659 ds_prev
= ds
->ds_prev
;
1661 VERIFY(0 == dsl_dataset_hold_obj(dp
,
1662 ds
->ds_phys
->ds_prev_snap_obj
, FTAG
, &ds_prev
));
1664 after_branch_point
=
1665 (ds_prev
->ds_phys
->ds_next_snap_obj
!= obj
);
1667 dmu_buf_will_dirty(ds_prev
->ds_dbuf
, tx
);
1668 if (after_branch_point
&&
1669 ds_prev
->ds_phys
->ds_next_clones_obj
!= 0) {
1670 remove_from_next_clones(ds_prev
, obj
, tx
);
1671 if (ds
->ds_phys
->ds_next_snap_obj
!= 0) {
1672 VERIFY(0 == zap_add_int(mos
,
1673 ds_prev
->ds_phys
->ds_next_clones_obj
,
1674 ds
->ds_phys
->ds_next_snap_obj
, tx
));
1677 if (after_branch_point
&&
1678 ds
->ds_phys
->ds_next_snap_obj
== 0) {
1679 /* This clone is toast. */
1680 ASSERT(ds_prev
->ds_phys
->ds_num_children
> 1);
1681 ds_prev
->ds_phys
->ds_num_children
--;
1684 * If the clone's origin has no other clones, no
1685 * user holds, and has been marked for deferred
1686 * deletion, then we should have done the necessary
1687 * destroy setup for it.
1689 if (ds_prev
->ds_phys
->ds_num_children
== 1 &&
1690 ds_prev
->ds_userrefs
== 0 &&
1691 DS_IS_DEFER_DESTROY(ds_prev
)) {
1692 ASSERT3P(dsda
->rm_origin
, !=, NULL
);
1694 ASSERT3P(dsda
->rm_origin
, ==, NULL
);
1696 } else if (!after_branch_point
) {
1697 ds_prev
->ds_phys
->ds_next_snap_obj
=
1698 ds
->ds_phys
->ds_next_snap_obj
;
1702 if (dsl_dataset_is_snapshot(ds
)) {
1703 dsl_dataset_t
*ds_next
;
1704 uint64_t old_unique
;
1705 uint64_t used
= 0, comp
= 0, uncomp
= 0;
1707 VERIFY(0 == dsl_dataset_hold_obj(dp
,
1708 ds
->ds_phys
->ds_next_snap_obj
, FTAG
, &ds_next
));
1709 ASSERT3U(ds_next
->ds_phys
->ds_prev_snap_obj
, ==, obj
);
1711 old_unique
= ds_next
->ds_phys
->ds_unique_bytes
;
1713 dmu_buf_will_dirty(ds_next
->ds_dbuf
, tx
);
1714 ds_next
->ds_phys
->ds_prev_snap_obj
=
1715 ds
->ds_phys
->ds_prev_snap_obj
;
1716 ds_next
->ds_phys
->ds_prev_snap_txg
=
1717 ds
->ds_phys
->ds_prev_snap_txg
;
1718 ASSERT3U(ds
->ds_phys
->ds_prev_snap_txg
, ==,
1719 ds_prev
? ds_prev
->ds_phys
->ds_creation_txg
: 0);
1722 if (ds_next
->ds_deadlist
.dl_oldfmt
) {
1723 process_old_deadlist(ds
, ds_prev
, ds_next
,
1724 after_branch_point
, tx
);
1726 /* Adjust prev's unique space. */
1727 if (ds_prev
&& !after_branch_point
) {
1728 dsl_deadlist_space_range(&ds_next
->ds_deadlist
,
1729 ds_prev
->ds_phys
->ds_prev_snap_txg
,
1730 ds
->ds_phys
->ds_prev_snap_txg
,
1731 &used
, &comp
, &uncomp
);
1732 ds_prev
->ds_phys
->ds_unique_bytes
+= used
;
1735 /* Adjust snapused. */
1736 dsl_deadlist_space_range(&ds_next
->ds_deadlist
,
1737 ds
->ds_phys
->ds_prev_snap_txg
, UINT64_MAX
,
1738 &used
, &comp
, &uncomp
);
1739 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_SNAP
,
1740 -used
, -comp
, -uncomp
, tx
);
1742 /* Move blocks to be freed to pool's free list. */
1743 dsl_deadlist_move_bpobj(&ds_next
->ds_deadlist
,
1744 &dp
->dp_free_bpobj
, ds
->ds_phys
->ds_prev_snap_txg
,
1746 dsl_dir_diduse_space(tx
->tx_pool
->dp_free_dir
,
1747 DD_USED_HEAD
, used
, comp
, uncomp
, tx
);
1748 dsl_dir_dirty(tx
->tx_pool
->dp_free_dir
, tx
);
1750 /* Merge our deadlist into next's and free it. */
1751 dsl_deadlist_merge(&ds_next
->ds_deadlist
,
1752 ds
->ds_phys
->ds_deadlist_obj
, tx
);
1754 dsl_deadlist_close(&ds
->ds_deadlist
);
1755 dsl_deadlist_free(mos
, ds
->ds_phys
->ds_deadlist_obj
, tx
);
1757 /* Collapse range in clone heads */
1758 dsl_dataset_remove_clones_key(ds
,
1759 ds
->ds_phys
->ds_creation_txg
, tx
);
1761 if (dsl_dataset_is_snapshot(ds_next
)) {
1762 dsl_dataset_t
*ds_nextnext
;
1765 * Update next's unique to include blocks which
1766 * were previously shared by only this snapshot
1767 * and it. Those blocks will be born after the
1768 * prev snap and before this snap, and will have
1769 * died after the next snap and before the one
1770 * after that (ie. be on the snap after next's
1773 VERIFY(0 == dsl_dataset_hold_obj(dp
,
1774 ds_next
->ds_phys
->ds_next_snap_obj
,
1775 FTAG
, &ds_nextnext
));
1776 dsl_deadlist_space_range(&ds_nextnext
->ds_deadlist
,
1777 ds
->ds_phys
->ds_prev_snap_txg
,
1778 ds
->ds_phys
->ds_creation_txg
,
1779 &used
, &comp
, &uncomp
);
1780 ds_next
->ds_phys
->ds_unique_bytes
+= used
;
1781 dsl_dataset_rele(ds_nextnext
, FTAG
);
1782 ASSERT3P(ds_next
->ds_prev
, ==, NULL
);
1784 /* Collapse range in this head. */
1786 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp
,
1787 ds
->ds_dir
->dd_phys
->dd_head_dataset_obj
,
1789 dsl_deadlist_remove_key(&hds
->ds_deadlist
,
1790 ds
->ds_phys
->ds_creation_txg
, tx
);
1791 dsl_dataset_rele(hds
, FTAG
);
1794 ASSERT3P(ds_next
->ds_prev
, ==, ds
);
1795 dsl_dataset_drop_ref(ds_next
->ds_prev
, ds_next
);
1796 ds_next
->ds_prev
= NULL
;
1798 VERIFY(0 == dsl_dataset_get_ref(dp
,
1799 ds
->ds_phys
->ds_prev_snap_obj
,
1800 ds_next
, &ds_next
->ds_prev
));
1803 dsl_dataset_recalc_head_uniq(ds_next
);
1806 * Reduce the amount of our unconsmed refreservation
1807 * being charged to our parent by the amount of
1808 * new unique data we have gained.
1810 if (old_unique
< ds_next
->ds_reserved
) {
1812 uint64_t new_unique
=
1813 ds_next
->ds_phys
->ds_unique_bytes
;
1815 ASSERT(old_unique
<= new_unique
);
1816 mrsdelta
= MIN(new_unique
- old_unique
,
1817 ds_next
->ds_reserved
- old_unique
);
1818 dsl_dir_diduse_space(ds
->ds_dir
,
1819 DD_USED_REFRSRV
, -mrsdelta
, 0, 0, tx
);
1822 dsl_dataset_rele(ds_next
, FTAG
);
1825 * There's no next snapshot, so this is a head dataset.
1826 * Destroy the deadlist. Unless it's a clone, the
1827 * deadlist should be empty. (If it's a clone, it's
1828 * safe to ignore the deadlist contents.)
1832 dsl_deadlist_close(&ds
->ds_deadlist
);
1833 dsl_deadlist_free(mos
, ds
->ds_phys
->ds_deadlist_obj
, tx
);
1834 ds
->ds_phys
->ds_deadlist_obj
= 0;
1837 * Free everything that we point to (that's born after
1838 * the previous snapshot, if we are a clone)
1840 * NB: this should be very quick, because we already
1841 * freed all the objects in open context.
1845 err
= traverse_dataset(ds
, ds
->ds_phys
->ds_prev_snap_txg
,
1846 TRAVERSE_POST
, kill_blkptr
, &ka
);
1847 ASSERT3U(err
, ==, 0);
1848 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds
) ||
1849 ds
->ds_phys
->ds_unique_bytes
== 0);
1851 if (ds
->ds_prev
!= NULL
) {
1852 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
1853 VERIFY3U(0, ==, zap_remove_int(mos
,
1854 ds
->ds_prev
->ds_dir
->dd_phys
->dd_clones
,
1855 ds
->ds_object
, tx
));
1857 dsl_dataset_rele(ds
->ds_prev
, ds
);
1858 ds
->ds_prev
= ds_prev
= NULL
;
1863 * This must be done after the dsl_traverse(), because it will
1864 * re-open the objset.
1866 if (ds
->ds_objset
) {
1867 dmu_objset_evict(ds
->ds_objset
);
1868 ds
->ds_objset
= NULL
;
1871 if (ds
->ds_dir
->dd_phys
->dd_head_dataset_obj
== ds
->ds_object
) {
1872 /* Erase the link in the dir */
1873 dmu_buf_will_dirty(ds
->ds_dir
->dd_dbuf
, tx
);
1874 ds
->ds_dir
->dd_phys
->dd_head_dataset_obj
= 0;
1875 ASSERT(ds
->ds_phys
->ds_snapnames_zapobj
!= 0);
1876 err
= zap_destroy(mos
, ds
->ds_phys
->ds_snapnames_zapobj
, tx
);
1879 /* remove from snapshot namespace */
1880 dsl_dataset_t
*ds_head
;
1881 ASSERT(ds
->ds_phys
->ds_snapnames_zapobj
== 0);
1882 VERIFY(0 == dsl_dataset_hold_obj(dp
,
1883 ds
->ds_dir
->dd_phys
->dd_head_dataset_obj
, FTAG
, &ds_head
));
1884 VERIFY(0 == dsl_dataset_get_snapname(ds
));
1889 err
= dsl_dataset_snap_lookup(ds_head
,
1890 ds
->ds_snapname
, &val
);
1891 ASSERT3U(err
, ==, 0);
1892 ASSERT3U(val
, ==, obj
);
1895 err
= dsl_dataset_snap_remove(ds_head
, ds
->ds_snapname
, tx
);
1897 dsl_dataset_rele(ds_head
, FTAG
);
1900 if (ds_prev
&& ds
->ds_prev
!= ds_prev
)
1901 dsl_dataset_rele(ds_prev
, FTAG
);
1903 spa_prop_clear_bootfs(dp
->dp_spa
, ds
->ds_object
, tx
);
1904 spa_history_log_internal(LOG_DS_DESTROY
, dp
->dp_spa
, tx
,
1905 "dataset = %llu", ds
->ds_object
);
1907 if (ds
->ds_phys
->ds_next_clones_obj
!= 0) {
1909 ASSERT(0 == zap_count(mos
,
1910 ds
->ds_phys
->ds_next_clones_obj
, &count
) && count
== 0);
1911 VERIFY(0 == dmu_object_free(mos
,
1912 ds
->ds_phys
->ds_next_clones_obj
, tx
));
1914 if (ds
->ds_phys
->ds_props_obj
!= 0)
1915 VERIFY(0 == zap_destroy(mos
, ds
->ds_phys
->ds_props_obj
, tx
));
1916 if (ds
->ds_phys
->ds_userrefs_obj
!= 0)
1917 VERIFY(0 == zap_destroy(mos
, ds
->ds_phys
->ds_userrefs_obj
, tx
));
1918 dsl_dir_close(ds
->ds_dir
, ds
);
1920 dsl_dataset_drain_refs(ds
, tag
);
1921 VERIFY(0 == dmu_object_free(mos
, obj
, tx
));
1923 if (dsda
->rm_origin
) {
1925 * Remove the origin of the clone we just destroyed.
1927 struct dsl_ds_destroyarg ndsda
= {0};
1929 ndsda
.ds
= dsda
->rm_origin
;
1930 dsl_dataset_destroy_sync(&ndsda
, tag
, tx
);
1935 dsl_dataset_snapshot_reserve_space(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
1939 if (!dmu_tx_is_syncing(tx
))
1943 * If there's an fs-only reservation, any blocks that might become
1944 * owned by the snapshot dataset must be accommodated by space
1945 * outside of the reservation.
1947 ASSERT(ds
->ds_reserved
== 0 || DS_UNIQUE_IS_ACCURATE(ds
));
1948 asize
= MIN(ds
->ds_phys
->ds_unique_bytes
, ds
->ds_reserved
);
1949 if (asize
> dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
))
1953 * Propogate any reserved space for this snapshot to other
1954 * snapshot checks in this sync group.
1957 dsl_dir_willuse_space(ds
->ds_dir
, asize
, tx
);
1963 dsl_dataset_snapshot_check(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
1965 dsl_dataset_t
*ds
= arg1
;
1966 const char *snapname
= arg2
;
1971 * We don't allow multiple snapshots of the same txg. If there
1972 * is already one, try again.
1974 if (ds
->ds_phys
->ds_prev_snap_txg
>= tx
->tx_txg
)
1978 * Check for conflicting name snapshot name.
1980 err
= dsl_dataset_snap_lookup(ds
, snapname
, &value
);
1987 * Check that the dataset's name is not too long. Name consists
1988 * of the dataset's length + 1 for the @-sign + snapshot name's length
1990 if (dsl_dataset_namelen(ds
) + 1 + strlen(snapname
) >= MAXNAMELEN
)
1991 return (ENAMETOOLONG
);
1993 err
= dsl_dataset_snapshot_reserve_space(ds
, tx
);
1997 ds
->ds_trysnap_txg
= tx
->tx_txg
;
2002 dsl_dataset_snapshot_sync(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
2004 dsl_dataset_t
*ds
= arg1
;
2005 const char *snapname
= arg2
;
2006 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
2008 dsl_dataset_phys_t
*dsphys
;
2009 uint64_t dsobj
, crtxg
;
2010 objset_t
*mos
= dp
->dp_meta_objset
;
2013 ASSERT(RW_WRITE_HELD(&dp
->dp_config_rwlock
));
2016 * The origin's ds_creation_txg has to be < TXG_INITIAL
2018 if (strcmp(snapname
, ORIGIN_DIR_NAME
) == 0)
2023 dsobj
= dmu_object_alloc(mos
, DMU_OT_DSL_DATASET
, 0,
2024 DMU_OT_DSL_DATASET
, sizeof (dsl_dataset_phys_t
), tx
);
2025 VERIFY(0 == dmu_bonus_hold(mos
, dsobj
, FTAG
, &dbuf
));
2026 dmu_buf_will_dirty(dbuf
, tx
);
2027 dsphys
= dbuf
->db_data
;
2028 bzero(dsphys
, sizeof (dsl_dataset_phys_t
));
2029 dsphys
->ds_dir_obj
= ds
->ds_dir
->dd_object
;
2030 dsphys
->ds_fsid_guid
= unique_create();
2031 (void) random_get_pseudo_bytes((void*)&dsphys
->ds_guid
,
2032 sizeof (dsphys
->ds_guid
));
2033 dsphys
->ds_prev_snap_obj
= ds
->ds_phys
->ds_prev_snap_obj
;
2034 dsphys
->ds_prev_snap_txg
= ds
->ds_phys
->ds_prev_snap_txg
;
2035 dsphys
->ds_next_snap_obj
= ds
->ds_object
;
2036 dsphys
->ds_num_children
= 1;
2037 dsphys
->ds_creation_time
= gethrestime_sec();
2038 dsphys
->ds_creation_txg
= crtxg
;
2039 dsphys
->ds_deadlist_obj
= ds
->ds_phys
->ds_deadlist_obj
;
2040 dsphys
->ds_used_bytes
= ds
->ds_phys
->ds_used_bytes
;
2041 dsphys
->ds_compressed_bytes
= ds
->ds_phys
->ds_compressed_bytes
;
2042 dsphys
->ds_uncompressed_bytes
= ds
->ds_phys
->ds_uncompressed_bytes
;
2043 dsphys
->ds_flags
= ds
->ds_phys
->ds_flags
;
2044 dsphys
->ds_bp
= ds
->ds_phys
->ds_bp
;
2045 dmu_buf_rele(dbuf
, FTAG
);
2047 ASSERT3U(ds
->ds_prev
!= 0, ==, ds
->ds_phys
->ds_prev_snap_obj
!= 0);
2049 uint64_t next_clones_obj
=
2050 ds
->ds_prev
->ds_phys
->ds_next_clones_obj
;
2051 ASSERT(ds
->ds_prev
->ds_phys
->ds_next_snap_obj
==
2053 ds
->ds_prev
->ds_phys
->ds_num_children
> 1);
2054 if (ds
->ds_prev
->ds_phys
->ds_next_snap_obj
== ds
->ds_object
) {
2055 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
2056 ASSERT3U(ds
->ds_phys
->ds_prev_snap_txg
, ==,
2057 ds
->ds_prev
->ds_phys
->ds_creation_txg
);
2058 ds
->ds_prev
->ds_phys
->ds_next_snap_obj
= dsobj
;
2059 } else if (next_clones_obj
!= 0) {
2060 remove_from_next_clones(ds
->ds_prev
,
2061 dsphys
->ds_next_snap_obj
, tx
);
2062 VERIFY3U(0, ==, zap_add_int(mos
,
2063 next_clones_obj
, dsobj
, tx
));
2068 * If we have a reference-reservation on this dataset, we will
2069 * need to increase the amount of refreservation being charged
2070 * since our unique space is going to zero.
2072 if (ds
->ds_reserved
) {
2074 ASSERT(DS_UNIQUE_IS_ACCURATE(ds
));
2075 delta
= MIN(ds
->ds_phys
->ds_unique_bytes
, ds
->ds_reserved
);
2076 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_REFRSRV
,
2080 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
2081 zfs_dbgmsg("taking snapshot %s@%s/%llu; newkey=%llu",
2082 ds
->ds_dir
->dd_myname
, snapname
, dsobj
,
2083 ds
->ds_phys
->ds_prev_snap_txg
);
2084 ds
->ds_phys
->ds_deadlist_obj
= dsl_deadlist_clone(&ds
->ds_deadlist
,
2085 UINT64_MAX
, ds
->ds_phys
->ds_prev_snap_obj
, tx
);
2086 dsl_deadlist_close(&ds
->ds_deadlist
);
2087 dsl_deadlist_open(&ds
->ds_deadlist
, mos
, ds
->ds_phys
->ds_deadlist_obj
);
2088 dsl_deadlist_add_key(&ds
->ds_deadlist
,
2089 ds
->ds_phys
->ds_prev_snap_txg
, tx
);
2091 ASSERT3U(ds
->ds_phys
->ds_prev_snap_txg
, <, tx
->tx_txg
);
2092 ds
->ds_phys
->ds_prev_snap_obj
= dsobj
;
2093 ds
->ds_phys
->ds_prev_snap_txg
= crtxg
;
2094 ds
->ds_phys
->ds_unique_bytes
= 0;
2095 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_UNIQUE_ACCURATE
)
2096 ds
->ds_phys
->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
2098 err
= zap_add(mos
, ds
->ds_phys
->ds_snapnames_zapobj
,
2099 snapname
, 8, 1, &dsobj
, tx
);
2103 dsl_dataset_drop_ref(ds
->ds_prev
, ds
);
2104 VERIFY(0 == dsl_dataset_get_ref(dp
,
2105 ds
->ds_phys
->ds_prev_snap_obj
, ds
, &ds
->ds_prev
));
2107 dsl_scan_ds_snapshotted(ds
, tx
);
2109 dsl_dir_snap_cmtime_update(ds
->ds_dir
);
2111 spa_history_log_internal(LOG_DS_SNAPSHOT
, dp
->dp_spa
, tx
,
2112 "dataset = %llu", dsobj
);
2116 dsl_dataset_sync(dsl_dataset_t
*ds
, zio_t
*zio
, dmu_tx_t
*tx
)
2118 ASSERT(dmu_tx_is_syncing(tx
));
2119 ASSERT(ds
->ds_objset
!= NULL
);
2120 ASSERT(ds
->ds_phys
->ds_next_snap_obj
== 0);
2123 * in case we had to change ds_fsid_guid when we opened it,
2126 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
2127 ds
->ds_phys
->ds_fsid_guid
= ds
->ds_fsid_guid
;
2129 dsl_dir_dirty(ds
->ds_dir
, tx
);
2130 dmu_objset_sync(ds
->ds_objset
, zio
, tx
);
2134 get_clones_stat(dsl_dataset_t
*ds
, nvlist_t
*nv
)
2137 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
2143 rw_enter(&ds
->ds_dir
->dd_pool
->dp_config_rwlock
, RW_READER
);
2144 VERIFY(nvlist_alloc(&propval
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
2145 VERIFY(nvlist_alloc(&val
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
2148 * There may me missing entries in ds_next_clones_obj
2149 * due to a bug in a previous version of the code.
2150 * Only trust it if it has the right number of entries.
2152 if (ds
->ds_phys
->ds_next_clones_obj
!= 0) {
2153 ASSERT3U(0, ==, zap_count(mos
, ds
->ds_phys
->ds_next_clones_obj
,
2156 if (count
!= ds
->ds_phys
->ds_num_children
- 1) {
2159 for (zap_cursor_init(&zc
, mos
, ds
->ds_phys
->ds_next_clones_obj
);
2160 zap_cursor_retrieve(&zc
, &za
) == 0;
2161 zap_cursor_advance(&zc
)) {
2162 dsl_dataset_t
*clone
;
2163 char buf
[ZFS_MAXNAMELEN
];
2164 if (dsl_dataset_hold_obj(ds
->ds_dir
->dd_pool
,
2165 za
.za_first_integer
, FTAG
, &clone
) != 0) {
2168 dsl_dir_name(clone
->ds_dir
, buf
);
2169 VERIFY(nvlist_add_boolean(val
, buf
) == 0);
2170 dsl_dataset_rele(clone
, FTAG
);
2172 zap_cursor_fini(&zc
);
2173 VERIFY(nvlist_add_nvlist(propval
, ZPROP_VALUE
, val
) == 0);
2174 VERIFY(nvlist_add_nvlist(nv
, zfs_prop_to_name(ZFS_PROP_CLONES
),
2178 nvlist_free(propval
);
2179 rw_exit(&ds
->ds_dir
->dd_pool
->dp_config_rwlock
);
2183 dsl_dataset_stats(dsl_dataset_t
*ds
, nvlist_t
*nv
)
2185 uint64_t refd
, avail
, uobjs
, aobjs
, ratio
;
2187 dsl_dir_stats(ds
->ds_dir
, nv
);
2189 dsl_dataset_space(ds
, &refd
, &avail
, &uobjs
, &aobjs
);
2190 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_AVAILABLE
, avail
);
2191 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFERENCED
, refd
);
2193 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_CREATION
,
2194 ds
->ds_phys
->ds_creation_time
);
2195 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_CREATETXG
,
2196 ds
->ds_phys
->ds_creation_txg
);
2197 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFQUOTA
,
2199 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFRESERVATION
,
2201 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_GUID
,
2202 ds
->ds_phys
->ds_guid
);
2203 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_UNIQUE
,
2204 ds
->ds_phys
->ds_unique_bytes
);
2205 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_OBJSETID
,
2207 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USERREFS
,
2209 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_DEFER_DESTROY
,
2210 DS_IS_DEFER_DESTROY(ds
) ? 1 : 0);
2212 if (ds
->ds_phys
->ds_prev_snap_obj
!= 0) {
2213 uint64_t written
, comp
, uncomp
;
2214 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
2215 dsl_dataset_t
*prev
;
2217 rw_enter(&dp
->dp_config_rwlock
, RW_READER
);
2218 int err
= dsl_dataset_hold_obj(dp
,
2219 ds
->ds_phys
->ds_prev_snap_obj
, FTAG
, &prev
);
2220 rw_exit(&dp
->dp_config_rwlock
);
2222 err
= dsl_dataset_space_written(prev
, ds
, &written
,
2224 dsl_dataset_rele(prev
, FTAG
);
2226 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_WRITTEN
,
2232 ratio
= ds
->ds_phys
->ds_compressed_bytes
== 0 ? 100 :
2233 (ds
->ds_phys
->ds_uncompressed_bytes
* 100 /
2234 ds
->ds_phys
->ds_compressed_bytes
);
2235 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFRATIO
, ratio
);
2237 if (ds
->ds_phys
->ds_next_snap_obj
) {
2239 * This is a snapshot; override the dd's space used with
2240 * our unique space and compression ratio.
2242 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USED
,
2243 ds
->ds_phys
->ds_unique_bytes
);
2244 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_COMPRESSRATIO
, ratio
);
2246 get_clones_stat(ds
, nv
);
2251 dsl_dataset_fast_stat(dsl_dataset_t
*ds
, dmu_objset_stats_t
*stat
)
2253 stat
->dds_creation_txg
= ds
->ds_phys
->ds_creation_txg
;
2254 stat
->dds_inconsistent
= ds
->ds_phys
->ds_flags
& DS_FLAG_INCONSISTENT
;
2255 stat
->dds_guid
= ds
->ds_phys
->ds_guid
;
2256 if (ds
->ds_phys
->ds_next_snap_obj
) {
2257 stat
->dds_is_snapshot
= B_TRUE
;
2258 stat
->dds_num_clones
= ds
->ds_phys
->ds_num_children
- 1;
2260 stat
->dds_is_snapshot
= B_FALSE
;
2261 stat
->dds_num_clones
= 0;
2264 /* clone origin is really a dsl_dir thing... */
2265 rw_enter(&ds
->ds_dir
->dd_pool
->dp_config_rwlock
, RW_READER
);
2266 if (dsl_dir_is_clone(ds
->ds_dir
)) {
2269 VERIFY(0 == dsl_dataset_get_ref(ds
->ds_dir
->dd_pool
,
2270 ds
->ds_dir
->dd_phys
->dd_origin_obj
, FTAG
, &ods
));
2271 dsl_dataset_name(ods
, stat
->dds_origin
);
2272 dsl_dataset_drop_ref(ods
, FTAG
);
2274 stat
->dds_origin
[0] = '\0';
2276 rw_exit(&ds
->ds_dir
->dd_pool
->dp_config_rwlock
);
2280 dsl_dataset_fsid_guid(dsl_dataset_t
*ds
)
2282 return (ds
->ds_fsid_guid
);
2286 dsl_dataset_space(dsl_dataset_t
*ds
,
2287 uint64_t *refdbytesp
, uint64_t *availbytesp
,
2288 uint64_t *usedobjsp
, uint64_t *availobjsp
)
2290 *refdbytesp
= ds
->ds_phys
->ds_used_bytes
;
2291 *availbytesp
= dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
);
2292 if (ds
->ds_reserved
> ds
->ds_phys
->ds_unique_bytes
)
2293 *availbytesp
+= ds
->ds_reserved
- ds
->ds_phys
->ds_unique_bytes
;
2294 if (ds
->ds_quota
!= 0) {
2296 * Adjust available bytes according to refquota
2298 if (*refdbytesp
< ds
->ds_quota
)
2299 *availbytesp
= MIN(*availbytesp
,
2300 ds
->ds_quota
- *refdbytesp
);
2304 *usedobjsp
= ds
->ds_phys
->ds_bp
.blk_fill
;
2305 *availobjsp
= DN_MAX_OBJECT
- *usedobjsp
;
2309 dsl_dataset_modified_since_lastsnap(dsl_dataset_t
*ds
)
2311 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
2313 ASSERT(RW_LOCK_HELD(&dp
->dp_config_rwlock
) ||
2314 dsl_pool_sync_context(dp
));
2315 if (ds
->ds_prev
== NULL
)
2317 if (ds
->ds_phys
->ds_bp
.blk_birth
>
2318 ds
->ds_prev
->ds_phys
->ds_creation_txg
) {
2319 objset_t
*os
, *os_prev
;
2321 * It may be that only the ZIL differs, because it was
2322 * reset in the head. Don't count that as being
2325 if (dmu_objset_from_ds(ds
, &os
) != 0)
2327 if (dmu_objset_from_ds(ds
->ds_prev
, &os_prev
) != 0)
2329 return (bcmp(&os
->os_phys
->os_meta_dnode
,
2330 &os_prev
->os_phys
->os_meta_dnode
,
2331 sizeof (os
->os_phys
->os_meta_dnode
)) != 0);
2338 dsl_dataset_snapshot_rename_check(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
2340 dsl_dataset_t
*ds
= arg1
;
2341 char *newsnapname
= arg2
;
2342 dsl_dir_t
*dd
= ds
->ds_dir
;
2347 err
= dsl_dataset_hold_obj(dd
->dd_pool
,
2348 dd
->dd_phys
->dd_head_dataset_obj
, FTAG
, &hds
);
2352 /* new name better not be in use */
2353 err
= dsl_dataset_snap_lookup(hds
, newsnapname
, &val
);
2354 dsl_dataset_rele(hds
, FTAG
);
2358 else if (err
== ENOENT
)
2361 /* dataset name + 1 for the "@" + the new snapshot name must fit */
2362 if (dsl_dir_namelen(ds
->ds_dir
) + 1 + strlen(newsnapname
) >= MAXNAMELEN
)
2369 dsl_dataset_snapshot_rename_sync(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
2371 dsl_dataset_t
*ds
= arg1
;
2372 const char *newsnapname
= arg2
;
2373 dsl_dir_t
*dd
= ds
->ds_dir
;
2374 objset_t
*mos
= dd
->dd_pool
->dp_meta_objset
;
2378 ASSERT(ds
->ds_phys
->ds_next_snap_obj
!= 0);
2380 VERIFY(0 == dsl_dataset_hold_obj(dd
->dd_pool
,
2381 dd
->dd_phys
->dd_head_dataset_obj
, FTAG
, &hds
));
2383 VERIFY(0 == dsl_dataset_get_snapname(ds
));
2384 err
= dsl_dataset_snap_remove(hds
, ds
->ds_snapname
, tx
);
2385 ASSERT3U(err
, ==, 0);
2386 mutex_enter(&ds
->ds_lock
);
2387 (void) strcpy(ds
->ds_snapname
, newsnapname
);
2388 mutex_exit(&ds
->ds_lock
);
2389 err
= zap_add(mos
, hds
->ds_phys
->ds_snapnames_zapobj
,
2390 ds
->ds_snapname
, 8, 1, &ds
->ds_object
, tx
);
2391 ASSERT3U(err
, ==, 0);
2393 spa_history_log_internal(LOG_DS_RENAME
, dd
->dd_pool
->dp_spa
, tx
,
2394 "dataset = %llu", ds
->ds_object
);
2395 dsl_dataset_rele(hds
, FTAG
);
2398 struct renamesnaparg
{
2399 dsl_sync_task_group_t
*dstg
;
2400 char failed
[MAXPATHLEN
];
2406 dsl_snapshot_rename_one(const char *name
, void *arg
)
2408 struct renamesnaparg
*ra
= arg
;
2409 dsl_dataset_t
*ds
= NULL
;
2413 snapname
= kmem_asprintf("%s@%s", name
, ra
->oldsnap
);
2414 (void) strlcpy(ra
->failed
, snapname
, sizeof (ra
->failed
));
2417 * For recursive snapshot renames the parent won't be changing
2418 * so we just pass name for both the to/from argument.
2420 err
= zfs_secpolicy_rename_perms(snapname
, snapname
, CRED());
2423 return (err
== ENOENT
? 0 : err
);
2428 * For all filesystems undergoing rename, we'll need to unmount it.
2430 (void) zfs_unmount_snap(snapname
, NULL
);
2432 err
= dsl_dataset_hold(snapname
, ra
->dstg
, &ds
);
2435 return (err
== ENOENT
? 0 : err
);
2437 dsl_sync_task_create(ra
->dstg
, dsl_dataset_snapshot_rename_check
,
2438 dsl_dataset_snapshot_rename_sync
, ds
, ra
->newsnap
, 0);
2444 dsl_recursive_rename(char *oldname
, const char *newname
)
2447 struct renamesnaparg
*ra
;
2448 dsl_sync_task_t
*dst
;
2450 char *cp
, *fsname
= spa_strdup(oldname
);
2451 int len
= strlen(oldname
) + 1;
2453 /* truncate the snapshot name to get the fsname */
2454 cp
= strchr(fsname
, '@');
2457 err
= spa_open(fsname
, &spa
, FTAG
);
2459 kmem_free(fsname
, len
);
2462 ra
= kmem_alloc(sizeof (struct renamesnaparg
), KM_SLEEP
);
2463 ra
->dstg
= dsl_sync_task_group_create(spa_get_dsl(spa
));
2465 ra
->oldsnap
= strchr(oldname
, '@') + 1;
2466 ra
->newsnap
= strchr(newname
, '@') + 1;
2469 err
= dmu_objset_find(fsname
, dsl_snapshot_rename_one
, ra
,
2471 kmem_free(fsname
, len
);
2474 err
= dsl_sync_task_group_wait(ra
->dstg
);
2477 for (dst
= list_head(&ra
->dstg
->dstg_tasks
); dst
;
2478 dst
= list_next(&ra
->dstg
->dstg_tasks
, dst
)) {
2479 dsl_dataset_t
*ds
= dst
->dst_arg1
;
2481 dsl_dir_name(ds
->ds_dir
, ra
->failed
);
2482 (void) strlcat(ra
->failed
, "@", sizeof (ra
->failed
));
2483 (void) strlcat(ra
->failed
, ra
->newsnap
,
2484 sizeof (ra
->failed
));
2486 dsl_dataset_rele(ds
, ra
->dstg
);
2490 (void) strlcpy(oldname
, ra
->failed
, sizeof (ra
->failed
));
2492 dsl_sync_task_group_destroy(ra
->dstg
);
2493 kmem_free(ra
, sizeof (struct renamesnaparg
));
2494 spa_close(spa
, FTAG
);
2499 dsl_valid_rename(const char *oldname
, void *arg
)
2501 int delta
= *(int *)arg
;
2503 if (strlen(oldname
) + delta
>= MAXNAMELEN
)
2504 return (ENAMETOOLONG
);
2509 #pragma weak dmu_objset_rename = dsl_dataset_rename
2511 dsl_dataset_rename(char *oldname
, const char *newname
, boolean_t recursive
)
2518 err
= dsl_dir_open(oldname
, FTAG
, &dd
, &tail
);
2523 int delta
= strlen(newname
) - strlen(oldname
);
2525 /* if we're growing, validate child name lengths */
2527 err
= dmu_objset_find(oldname
, dsl_valid_rename
,
2528 &delta
, DS_FIND_CHILDREN
| DS_FIND_SNAPSHOTS
);
2531 err
= dsl_dir_rename(dd
, newname
);
2532 dsl_dir_close(dd
, FTAG
);
2536 if (tail
[0] != '@') {
2537 /* the name ended in a nonexistent component */
2538 dsl_dir_close(dd
, FTAG
);
2542 dsl_dir_close(dd
, FTAG
);
2544 /* new name must be snapshot in same filesystem */
2545 tail
= strchr(newname
, '@');
2549 if (strncmp(oldname
, newname
, tail
- newname
) != 0)
2553 err
= dsl_recursive_rename(oldname
, newname
);
2555 err
= dsl_dataset_hold(oldname
, FTAG
, &ds
);
2559 err
= dsl_sync_task_do(ds
->ds_dir
->dd_pool
,
2560 dsl_dataset_snapshot_rename_check
,
2561 dsl_dataset_snapshot_rename_sync
, ds
, (char *)tail
, 1);
2563 dsl_dataset_rele(ds
, FTAG
);
2569 struct promotenode
{
2575 list_t shared_snaps
, origin_snaps
, clone_snaps
;
2576 dsl_dataset_t
*origin_origin
;
2577 uint64_t used
, comp
, uncomp
, unique
, cloneusedsnap
, originusedsnap
;
2581 static int snaplist_space(list_t
*l
, uint64_t mintxg
, uint64_t *spacep
);
2582 static boolean_t
snaplist_unstable(list_t
*l
);
2585 dsl_dataset_promote_check(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
2587 dsl_dataset_t
*hds
= arg1
;
2588 struct promotearg
*pa
= arg2
;
2589 struct promotenode
*snap
= list_head(&pa
->shared_snaps
);
2590 dsl_dataset_t
*origin_ds
= snap
->ds
;
2594 /* Check that it is a real clone */
2595 if (!dsl_dir_is_clone(hds
->ds_dir
))
2598 /* Since this is so expensive, don't do the preliminary check */
2599 if (!dmu_tx_is_syncing(tx
))
2602 if (hds
->ds_phys
->ds_flags
& DS_FLAG_NOPROMOTE
)
2605 /* compute origin's new unique space */
2606 snap
= list_tail(&pa
->clone_snaps
);
2607 ASSERT3U(snap
->ds
->ds_phys
->ds_prev_snap_obj
, ==, origin_ds
->ds_object
);
2608 dsl_deadlist_space_range(&snap
->ds
->ds_deadlist
,
2609 origin_ds
->ds_phys
->ds_prev_snap_txg
, UINT64_MAX
,
2610 &pa
->unique
, &unused
, &unused
);
2613 * Walk the snapshots that we are moving
2615 * Compute space to transfer. Consider the incremental changes
2616 * to used for each snapshot:
2617 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2618 * So each snapshot gave birth to:
2619 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2620 * So a sequence would look like:
2621 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2622 * Which simplifies to:
2623 * uN + kN + kN-1 + ... + k1 + k0
2624 * Note however, if we stop before we reach the ORIGIN we get:
2625 * uN + kN + kN-1 + ... + kM - uM-1
2627 pa
->used
= origin_ds
->ds_phys
->ds_used_bytes
;
2628 pa
->comp
= origin_ds
->ds_phys
->ds_compressed_bytes
;
2629 pa
->uncomp
= origin_ds
->ds_phys
->ds_uncompressed_bytes
;
2630 for (snap
= list_head(&pa
->shared_snaps
); snap
;
2631 snap
= list_next(&pa
->shared_snaps
, snap
)) {
2632 uint64_t val
, dlused
, dlcomp
, dluncomp
;
2633 dsl_dataset_t
*ds
= snap
->ds
;
2635 /* Check that the snapshot name does not conflict */
2636 VERIFY(0 == dsl_dataset_get_snapname(ds
));
2637 err
= dsl_dataset_snap_lookup(hds
, ds
->ds_snapname
, &val
);
2645 /* The very first snapshot does not have a deadlist */
2646 if (ds
->ds_phys
->ds_prev_snap_obj
== 0)
2649 dsl_deadlist_space(&ds
->ds_deadlist
,
2650 &dlused
, &dlcomp
, &dluncomp
);
2653 pa
->uncomp
+= dluncomp
;
2657 * If we are a clone of a clone then we never reached ORIGIN,
2658 * so we need to subtract out the clone origin's used space.
2660 if (pa
->origin_origin
) {
2661 pa
->used
-= pa
->origin_origin
->ds_phys
->ds_used_bytes
;
2662 pa
->comp
-= pa
->origin_origin
->ds_phys
->ds_compressed_bytes
;
2663 pa
->uncomp
-= pa
->origin_origin
->ds_phys
->ds_uncompressed_bytes
;
2666 /* Check that there is enough space here */
2667 err
= dsl_dir_transfer_possible(origin_ds
->ds_dir
, hds
->ds_dir
,
2673 * Compute the amounts of space that will be used by snapshots
2674 * after the promotion (for both origin and clone). For each,
2675 * it is the amount of space that will be on all of their
2676 * deadlists (that was not born before their new origin).
2678 if (hds
->ds_dir
->dd_phys
->dd_flags
& DD_FLAG_USED_BREAKDOWN
) {
2682 * Note, typically this will not be a clone of a clone,
2683 * so dd_origin_txg will be < TXG_INITIAL, so
2684 * these snaplist_space() -> dsl_deadlist_space_range()
2685 * calls will be fast because they do not have to
2686 * iterate over all bps.
2688 snap
= list_head(&pa
->origin_snaps
);
2689 err
= snaplist_space(&pa
->shared_snaps
,
2690 snap
->ds
->ds_dir
->dd_origin_txg
, &pa
->cloneusedsnap
);
2694 err
= snaplist_space(&pa
->clone_snaps
,
2695 snap
->ds
->ds_dir
->dd_origin_txg
, &space
);
2698 pa
->cloneusedsnap
+= space
;
2700 if (origin_ds
->ds_dir
->dd_phys
->dd_flags
& DD_FLAG_USED_BREAKDOWN
) {
2701 err
= snaplist_space(&pa
->origin_snaps
,
2702 origin_ds
->ds_phys
->ds_creation_txg
, &pa
->originusedsnap
);
2709 pa
->err_ds
= snap
->ds
->ds_snapname
;
2714 dsl_dataset_promote_sync(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
2716 dsl_dataset_t
*hds
= arg1
;
2717 struct promotearg
*pa
= arg2
;
2718 struct promotenode
*snap
= list_head(&pa
->shared_snaps
);
2719 dsl_dataset_t
*origin_ds
= snap
->ds
;
2720 dsl_dataset_t
*origin_head
;
2721 dsl_dir_t
*dd
= hds
->ds_dir
;
2722 dsl_pool_t
*dp
= hds
->ds_dir
->dd_pool
;
2723 dsl_dir_t
*odd
= NULL
;
2724 uint64_t oldnext_obj
;
2727 ASSERT(0 == (hds
->ds_phys
->ds_flags
& DS_FLAG_NOPROMOTE
));
2729 snap
= list_head(&pa
->origin_snaps
);
2730 origin_head
= snap
->ds
;
2733 * We need to explicitly open odd, since origin_ds's dd will be
2736 VERIFY(0 == dsl_dir_open_obj(dp
, origin_ds
->ds_dir
->dd_object
,
2739 /* change origin's next snap */
2740 dmu_buf_will_dirty(origin_ds
->ds_dbuf
, tx
);
2741 oldnext_obj
= origin_ds
->ds_phys
->ds_next_snap_obj
;
2742 snap
= list_tail(&pa
->clone_snaps
);
2743 ASSERT3U(snap
->ds
->ds_phys
->ds_prev_snap_obj
, ==, origin_ds
->ds_object
);
2744 origin_ds
->ds_phys
->ds_next_snap_obj
= snap
->ds
->ds_object
;
2746 /* change the origin's next clone */
2747 if (origin_ds
->ds_phys
->ds_next_clones_obj
) {
2748 remove_from_next_clones(origin_ds
, snap
->ds
->ds_object
, tx
);
2749 VERIFY3U(0, ==, zap_add_int(dp
->dp_meta_objset
,
2750 origin_ds
->ds_phys
->ds_next_clones_obj
,
2755 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
2756 ASSERT3U(dd
->dd_phys
->dd_origin_obj
, ==, origin_ds
->ds_object
);
2757 dd
->dd_phys
->dd_origin_obj
= odd
->dd_phys
->dd_origin_obj
;
2758 dd
->dd_origin_txg
= origin_head
->ds_dir
->dd_origin_txg
;
2759 dmu_buf_will_dirty(odd
->dd_dbuf
, tx
);
2760 odd
->dd_phys
->dd_origin_obj
= origin_ds
->ds_object
;
2761 origin_head
->ds_dir
->dd_origin_txg
=
2762 origin_ds
->ds_phys
->ds_creation_txg
;
2764 /* change dd_clone entries */
2765 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
2766 VERIFY3U(0, ==, zap_remove_int(dp
->dp_meta_objset
,
2767 odd
->dd_phys
->dd_clones
, hds
->ds_object
, tx
));
2768 VERIFY3U(0, ==, zap_add_int(dp
->dp_meta_objset
,
2769 pa
->origin_origin
->ds_dir
->dd_phys
->dd_clones
,
2770 hds
->ds_object
, tx
));
2772 VERIFY3U(0, ==, zap_remove_int(dp
->dp_meta_objset
,
2773 pa
->origin_origin
->ds_dir
->dd_phys
->dd_clones
,
2774 origin_head
->ds_object
, tx
));
2775 if (dd
->dd_phys
->dd_clones
== 0) {
2776 dd
->dd_phys
->dd_clones
= zap_create(dp
->dp_meta_objset
,
2777 DMU_OT_DSL_CLONES
, DMU_OT_NONE
, 0, tx
);
2779 VERIFY3U(0, ==, zap_add_int(dp
->dp_meta_objset
,
2780 dd
->dd_phys
->dd_clones
, origin_head
->ds_object
, tx
));
2784 /* move snapshots to this dir */
2785 for (snap
= list_head(&pa
->shared_snaps
); snap
;
2786 snap
= list_next(&pa
->shared_snaps
, snap
)) {
2787 dsl_dataset_t
*ds
= snap
->ds
;
2789 /* unregister props as dsl_dir is changing */
2790 if (ds
->ds_objset
) {
2791 dmu_objset_evict(ds
->ds_objset
);
2792 ds
->ds_objset
= NULL
;
2794 /* move snap name entry */
2795 VERIFY(0 == dsl_dataset_get_snapname(ds
));
2796 VERIFY(0 == dsl_dataset_snap_remove(origin_head
,
2797 ds
->ds_snapname
, tx
));
2798 VERIFY(0 == zap_add(dp
->dp_meta_objset
,
2799 hds
->ds_phys
->ds_snapnames_zapobj
, ds
->ds_snapname
,
2800 8, 1, &ds
->ds_object
, tx
));
2802 /* change containing dsl_dir */
2803 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
2804 ASSERT3U(ds
->ds_phys
->ds_dir_obj
, ==, odd
->dd_object
);
2805 ds
->ds_phys
->ds_dir_obj
= dd
->dd_object
;
2806 ASSERT3P(ds
->ds_dir
, ==, odd
);
2807 dsl_dir_close(ds
->ds_dir
, ds
);
2808 VERIFY(0 == dsl_dir_open_obj(dp
, dd
->dd_object
,
2809 NULL
, ds
, &ds
->ds_dir
));
2811 /* move any clone references */
2812 if (ds
->ds_phys
->ds_next_clones_obj
&&
2813 spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
2817 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
2818 ds
->ds_phys
->ds_next_clones_obj
);
2819 zap_cursor_retrieve(&zc
, &za
) == 0;
2820 zap_cursor_advance(&zc
)) {
2821 dsl_dataset_t
*cnds
;
2824 if (za
.za_first_integer
== oldnext_obj
) {
2826 * We've already moved the
2827 * origin's reference.
2832 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp
,
2833 za
.za_first_integer
, FTAG
, &cnds
));
2834 o
= cnds
->ds_dir
->dd_phys
->dd_head_dataset_obj
;
2836 VERIFY3U(zap_remove_int(dp
->dp_meta_objset
,
2837 odd
->dd_phys
->dd_clones
, o
, tx
), ==, 0);
2838 VERIFY3U(zap_add_int(dp
->dp_meta_objset
,
2839 dd
->dd_phys
->dd_clones
, o
, tx
), ==, 0);
2840 dsl_dataset_rele(cnds
, FTAG
);
2842 zap_cursor_fini(&zc
);
2845 ASSERT3U(dsl_prop_numcb(ds
), ==, 0);
2849 * Change space accounting.
2850 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2851 * both be valid, or both be 0 (resulting in delta == 0). This
2852 * is true for each of {clone,origin} independently.
2855 delta
= pa
->cloneusedsnap
-
2856 dd
->dd_phys
->dd_used_breakdown
[DD_USED_SNAP
];
2857 ASSERT3S(delta
, >=, 0);
2858 ASSERT3U(pa
->used
, >=, delta
);
2859 dsl_dir_diduse_space(dd
, DD_USED_SNAP
, delta
, 0, 0, tx
);
2860 dsl_dir_diduse_space(dd
, DD_USED_HEAD
,
2861 pa
->used
- delta
, pa
->comp
, pa
->uncomp
, tx
);
2863 delta
= pa
->originusedsnap
-
2864 odd
->dd_phys
->dd_used_breakdown
[DD_USED_SNAP
];
2865 ASSERT3S(delta
, <=, 0);
2866 ASSERT3U(pa
->used
, >=, -delta
);
2867 dsl_dir_diduse_space(odd
, DD_USED_SNAP
, delta
, 0, 0, tx
);
2868 dsl_dir_diduse_space(odd
, DD_USED_HEAD
,
2869 -pa
->used
- delta
, -pa
->comp
, -pa
->uncomp
, tx
);
2871 origin_ds
->ds_phys
->ds_unique_bytes
= pa
->unique
;
2873 /* log history record */
2874 spa_history_log_internal(LOG_DS_PROMOTE
, dd
->dd_pool
->dp_spa
, tx
,
2875 "dataset = %llu", hds
->ds_object
);
2877 dsl_dir_close(odd
, FTAG
);
2880 static char *snaplist_tag
= "snaplist";
2882 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2883 * (exclusive) and last_obj (inclusive). The list will be in reverse
2884 * order (last_obj will be the list_head()). If first_obj == 0, do all
2885 * snapshots back to this dataset's origin.
2888 snaplist_make(dsl_pool_t
*dp
, boolean_t own
,
2889 uint64_t first_obj
, uint64_t last_obj
, list_t
*l
)
2891 uint64_t obj
= last_obj
;
2893 ASSERT(RW_LOCK_HELD(&dp
->dp_config_rwlock
));
2895 list_create(l
, sizeof (struct promotenode
),
2896 offsetof(struct promotenode
, link
));
2898 while (obj
!= first_obj
) {
2900 struct promotenode
*snap
;
2904 err
= dsl_dataset_own_obj(dp
, obj
,
2905 0, snaplist_tag
, &ds
);
2907 dsl_dataset_make_exclusive(ds
, snaplist_tag
);
2909 err
= dsl_dataset_hold_obj(dp
, obj
, snaplist_tag
, &ds
);
2911 if (err
== ENOENT
) {
2912 /* lost race with snapshot destroy */
2913 struct promotenode
*last
= list_tail(l
);
2914 ASSERT(obj
!= last
->ds
->ds_phys
->ds_prev_snap_obj
);
2915 obj
= last
->ds
->ds_phys
->ds_prev_snap_obj
;
2922 first_obj
= ds
->ds_dir
->dd_phys
->dd_origin_obj
;
2924 snap
= kmem_alloc(sizeof (struct promotenode
), KM_SLEEP
);
2926 list_insert_tail(l
, snap
);
2927 obj
= ds
->ds_phys
->ds_prev_snap_obj
;
2934 snaplist_space(list_t
*l
, uint64_t mintxg
, uint64_t *spacep
)
2936 struct promotenode
*snap
;
2939 for (snap
= list_head(l
); snap
; snap
= list_next(l
, snap
)) {
2940 uint64_t used
, comp
, uncomp
;
2941 dsl_deadlist_space_range(&snap
->ds
->ds_deadlist
,
2942 mintxg
, UINT64_MAX
, &used
, &comp
, &uncomp
);
2949 snaplist_destroy(list_t
*l
, boolean_t own
)
2951 struct promotenode
*snap
;
2953 if (!l
|| !list_link_active(&l
->list_head
))
2956 while ((snap
= list_tail(l
)) != NULL
) {
2957 list_remove(l
, snap
);
2959 dsl_dataset_disown(snap
->ds
, snaplist_tag
);
2961 dsl_dataset_rele(snap
->ds
, snaplist_tag
);
2962 kmem_free(snap
, sizeof (struct promotenode
));
2968 * Promote a clone. Nomenclature note:
2969 * "clone" or "cds": the original clone which is being promoted
2970 * "origin" or "ods": the snapshot which is originally clone's origin
2971 * "origin head" or "ohds": the dataset which is the head
2972 * (filesystem/volume) for the origin
2973 * "origin origin": the origin of the origin's filesystem (typically
2974 * NULL, indicating that the clone is not a clone of a clone).
2977 dsl_dataset_promote(const char *name
, char *conflsnap
)
2982 dmu_object_info_t doi
;
2983 struct promotearg pa
= { 0 };
2984 struct promotenode
*snap
;
2987 err
= dsl_dataset_hold(name
, FTAG
, &ds
);
2993 err
= dmu_object_info(dp
->dp_meta_objset
,
2994 ds
->ds_phys
->ds_snapnames_zapobj
, &doi
);
2996 dsl_dataset_rele(ds
, FTAG
);
3000 if (dsl_dataset_is_snapshot(ds
) || dd
->dd_phys
->dd_origin_obj
== 0) {
3001 dsl_dataset_rele(ds
, FTAG
);
3006 * We are going to inherit all the snapshots taken before our
3007 * origin (i.e., our new origin will be our parent's origin).
3008 * Take ownership of them so that we can rename them into our
3011 rw_enter(&dp
->dp_config_rwlock
, RW_READER
);
3013 err
= snaplist_make(dp
, B_TRUE
, 0, dd
->dd_phys
->dd_origin_obj
,
3018 err
= snaplist_make(dp
, B_FALSE
, 0, ds
->ds_object
, &pa
.clone_snaps
);
3022 snap
= list_head(&pa
.shared_snaps
);
3023 ASSERT3U(snap
->ds
->ds_object
, ==, dd
->dd_phys
->dd_origin_obj
);
3024 err
= snaplist_make(dp
, B_FALSE
, dd
->dd_phys
->dd_origin_obj
,
3025 snap
->ds
->ds_dir
->dd_phys
->dd_head_dataset_obj
, &pa
.origin_snaps
);
3029 if (snap
->ds
->ds_dir
->dd_phys
->dd_origin_obj
!= 0) {
3030 err
= dsl_dataset_hold_obj(dp
,
3031 snap
->ds
->ds_dir
->dd_phys
->dd_origin_obj
,
3032 FTAG
, &pa
.origin_origin
);
3038 rw_exit(&dp
->dp_config_rwlock
);
3041 * Add in 128x the snapnames zapobj size, since we will be moving
3042 * a bunch of snapnames to the promoted ds, and dirtying their
3046 err
= dsl_sync_task_do(dp
, dsl_dataset_promote_check
,
3047 dsl_dataset_promote_sync
, ds
, &pa
,
3048 2 + 2 * doi
.doi_physical_blocks_512
);
3049 if (err
&& pa
.err_ds
&& conflsnap
)
3050 (void) strncpy(conflsnap
, pa
.err_ds
, MAXNAMELEN
);
3053 snaplist_destroy(&pa
.shared_snaps
, B_TRUE
);
3054 snaplist_destroy(&pa
.clone_snaps
, B_FALSE
);
3055 snaplist_destroy(&pa
.origin_snaps
, B_FALSE
);
3056 if (pa
.origin_origin
)
3057 dsl_dataset_rele(pa
.origin_origin
, FTAG
);
3058 dsl_dataset_rele(ds
, FTAG
);
3062 struct cloneswaparg
{
3063 dsl_dataset_t
*cds
; /* clone dataset */
3064 dsl_dataset_t
*ohds
; /* origin's head dataset */
3066 int64_t unused_refres_delta
; /* change in unconsumed refreservation */
3071 dsl_dataset_clone_swap_check(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
3073 struct cloneswaparg
*csa
= arg1
;
3075 /* they should both be heads */
3076 if (dsl_dataset_is_snapshot(csa
->cds
) ||
3077 dsl_dataset_is_snapshot(csa
->ohds
))
3080 /* the branch point should be just before them */
3081 if (csa
->cds
->ds_prev
!= csa
->ohds
->ds_prev
)
3084 /* cds should be the clone (unless they are unrelated) */
3085 if (csa
->cds
->ds_prev
!= NULL
&&
3086 csa
->cds
->ds_prev
!= csa
->cds
->ds_dir
->dd_pool
->dp_origin_snap
&&
3087 csa
->ohds
->ds_object
!=
3088 csa
->cds
->ds_prev
->ds_phys
->ds_next_snap_obj
)
3091 /* the clone should be a child of the origin */
3092 if (csa
->cds
->ds_dir
->dd_parent
!= csa
->ohds
->ds_dir
)
3095 /* ohds shouldn't be modified unless 'force' */
3096 if (!csa
->force
&& dsl_dataset_modified_since_lastsnap(csa
->ohds
))
3099 /* adjust amount of any unconsumed refreservation */
3100 csa
->unused_refres_delta
=
3101 (int64_t)MIN(csa
->ohds
->ds_reserved
,
3102 csa
->ohds
->ds_phys
->ds_unique_bytes
) -
3103 (int64_t)MIN(csa
->ohds
->ds_reserved
,
3104 csa
->cds
->ds_phys
->ds_unique_bytes
);
3106 if (csa
->unused_refres_delta
> 0 &&
3107 csa
->unused_refres_delta
>
3108 dsl_dir_space_available(csa
->ohds
->ds_dir
, NULL
, 0, TRUE
))
3111 if (csa
->ohds
->ds_quota
!= 0 &&
3112 csa
->cds
->ds_phys
->ds_unique_bytes
> csa
->ohds
->ds_quota
)
3120 dsl_dataset_clone_swap_sync(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
3122 struct cloneswaparg
*csa
= arg1
;
3123 dsl_pool_t
*dp
= csa
->cds
->ds_dir
->dd_pool
;
3125 ASSERT(csa
->cds
->ds_reserved
== 0);
3126 ASSERT(csa
->ohds
->ds_quota
== 0 ||
3127 csa
->cds
->ds_phys
->ds_unique_bytes
<= csa
->ohds
->ds_quota
);
3129 dmu_buf_will_dirty(csa
->cds
->ds_dbuf
, tx
);
3130 dmu_buf_will_dirty(csa
->ohds
->ds_dbuf
, tx
);
3132 if (csa
->cds
->ds_objset
!= NULL
) {
3133 dmu_objset_evict(csa
->cds
->ds_objset
);
3134 csa
->cds
->ds_objset
= NULL
;
3137 if (csa
->ohds
->ds_objset
!= NULL
) {
3138 dmu_objset_evict(csa
->ohds
->ds_objset
);
3139 csa
->ohds
->ds_objset
= NULL
;
3143 * Reset origin's unique bytes, if it exists.
3145 if (csa
->cds
->ds_prev
) {
3146 dsl_dataset_t
*origin
= csa
->cds
->ds_prev
;
3147 uint64_t comp
, uncomp
;
3149 dmu_buf_will_dirty(origin
->ds_dbuf
, tx
);
3150 dsl_deadlist_space_range(&csa
->cds
->ds_deadlist
,
3151 origin
->ds_phys
->ds_prev_snap_txg
, UINT64_MAX
,
3152 &origin
->ds_phys
->ds_unique_bytes
, &comp
, &uncomp
);
3158 tmp
= csa
->ohds
->ds_phys
->ds_bp
;
3159 csa
->ohds
->ds_phys
->ds_bp
= csa
->cds
->ds_phys
->ds_bp
;
3160 csa
->cds
->ds_phys
->ds_bp
= tmp
;
3163 /* set dd_*_bytes */
3165 int64_t dused
, dcomp
, duncomp
;
3166 uint64_t cdl_used
, cdl_comp
, cdl_uncomp
;
3167 uint64_t odl_used
, odl_comp
, odl_uncomp
;
3169 ASSERT3U(csa
->cds
->ds_dir
->dd_phys
->
3170 dd_used_breakdown
[DD_USED_SNAP
], ==, 0);
3172 dsl_deadlist_space(&csa
->cds
->ds_deadlist
,
3173 &cdl_used
, &cdl_comp
, &cdl_uncomp
);
3174 dsl_deadlist_space(&csa
->ohds
->ds_deadlist
,
3175 &odl_used
, &odl_comp
, &odl_uncomp
);
3177 dused
= csa
->cds
->ds_phys
->ds_used_bytes
+ cdl_used
-
3178 (csa
->ohds
->ds_phys
->ds_used_bytes
+ odl_used
);
3179 dcomp
= csa
->cds
->ds_phys
->ds_compressed_bytes
+ cdl_comp
-
3180 (csa
->ohds
->ds_phys
->ds_compressed_bytes
+ odl_comp
);
3181 duncomp
= csa
->cds
->ds_phys
->ds_uncompressed_bytes
+
3183 (csa
->ohds
->ds_phys
->ds_uncompressed_bytes
+ odl_uncomp
);
3185 dsl_dir_diduse_space(csa
->ohds
->ds_dir
, DD_USED_HEAD
,
3186 dused
, dcomp
, duncomp
, tx
);
3187 dsl_dir_diduse_space(csa
->cds
->ds_dir
, DD_USED_HEAD
,
3188 -dused
, -dcomp
, -duncomp
, tx
);
3191 * The difference in the space used by snapshots is the
3192 * difference in snapshot space due to the head's
3193 * deadlist (since that's the only thing that's
3194 * changing that affects the snapused).
3196 dsl_deadlist_space_range(&csa
->cds
->ds_deadlist
,
3197 csa
->ohds
->ds_dir
->dd_origin_txg
, UINT64_MAX
,
3198 &cdl_used
, &cdl_comp
, &cdl_uncomp
);
3199 dsl_deadlist_space_range(&csa
->ohds
->ds_deadlist
,
3200 csa
->ohds
->ds_dir
->dd_origin_txg
, UINT64_MAX
,
3201 &odl_used
, &odl_comp
, &odl_uncomp
);
3202 dsl_dir_transfer_space(csa
->ohds
->ds_dir
, cdl_used
- odl_used
,
3203 DD_USED_HEAD
, DD_USED_SNAP
, tx
);
3206 /* swap ds_*_bytes */
3207 SWITCH64(csa
->ohds
->ds_phys
->ds_used_bytes
,
3208 csa
->cds
->ds_phys
->ds_used_bytes
);
3209 SWITCH64(csa
->ohds
->ds_phys
->ds_compressed_bytes
,
3210 csa
->cds
->ds_phys
->ds_compressed_bytes
);
3211 SWITCH64(csa
->ohds
->ds_phys
->ds_uncompressed_bytes
,
3212 csa
->cds
->ds_phys
->ds_uncompressed_bytes
);
3213 SWITCH64(csa
->ohds
->ds_phys
->ds_unique_bytes
,
3214 csa
->cds
->ds_phys
->ds_unique_bytes
);
3216 /* apply any parent delta for change in unconsumed refreservation */
3217 dsl_dir_diduse_space(csa
->ohds
->ds_dir
, DD_USED_REFRSRV
,
3218 csa
->unused_refres_delta
, 0, 0, tx
);
3223 dsl_deadlist_close(&csa
->cds
->ds_deadlist
);
3224 dsl_deadlist_close(&csa
->ohds
->ds_deadlist
);
3225 SWITCH64(csa
->ohds
->ds_phys
->ds_deadlist_obj
,
3226 csa
->cds
->ds_phys
->ds_deadlist_obj
);
3227 dsl_deadlist_open(&csa
->cds
->ds_deadlist
, dp
->dp_meta_objset
,
3228 csa
->cds
->ds_phys
->ds_deadlist_obj
);
3229 dsl_deadlist_open(&csa
->ohds
->ds_deadlist
, dp
->dp_meta_objset
,
3230 csa
->ohds
->ds_phys
->ds_deadlist_obj
);
3232 dsl_scan_ds_clone_swapped(csa
->ohds
, csa
->cds
, tx
);
3236 * Swap 'clone' with its origin head datasets. Used at the end of "zfs
3237 * recv" into an existing fs to swizzle the file system to the new
3238 * version, and by "zfs rollback". Can also be used to swap two
3239 * independent head datasets if neither has any snapshots.
3242 dsl_dataset_clone_swap(dsl_dataset_t
*clone
, dsl_dataset_t
*origin_head
,
3245 struct cloneswaparg csa
;
3248 ASSERT(clone
->ds_owner
);
3249 ASSERT(origin_head
->ds_owner
);
3252 * Need exclusive access for the swap. If we're swapping these
3253 * datasets back after an error, we already hold the locks.
3255 if (!RW_WRITE_HELD(&clone
->ds_rwlock
))
3256 rw_enter(&clone
->ds_rwlock
, RW_WRITER
);
3257 if (!RW_WRITE_HELD(&origin_head
->ds_rwlock
) &&
3258 !rw_tryenter(&origin_head
->ds_rwlock
, RW_WRITER
)) {
3259 rw_exit(&clone
->ds_rwlock
);
3260 rw_enter(&origin_head
->ds_rwlock
, RW_WRITER
);
3261 if (!rw_tryenter(&clone
->ds_rwlock
, RW_WRITER
)) {
3262 rw_exit(&origin_head
->ds_rwlock
);
3267 csa
.ohds
= origin_head
;
3269 error
= dsl_sync_task_do(clone
->ds_dir
->dd_pool
,
3270 dsl_dataset_clone_swap_check
,
3271 dsl_dataset_clone_swap_sync
, &csa
, NULL
, 9);
3276 * Given a pool name and a dataset object number in that pool,
3277 * return the name of that dataset.
3280 dsl_dsobj_to_dsname(char *pname
, uint64_t obj
, char *buf
)
3287 if ((error
= spa_open(pname
, &spa
, FTAG
)) != 0)
3289 dp
= spa_get_dsl(spa
);
3290 rw_enter(&dp
->dp_config_rwlock
, RW_READER
);
3291 if ((error
= dsl_dataset_hold_obj(dp
, obj
, FTAG
, &ds
)) == 0) {
3292 dsl_dataset_name(ds
, buf
);
3293 dsl_dataset_rele(ds
, FTAG
);
3295 rw_exit(&dp
->dp_config_rwlock
);
3296 spa_close(spa
, FTAG
);
3302 dsl_dataset_check_quota(dsl_dataset_t
*ds
, boolean_t check_quota
,
3303 uint64_t asize
, uint64_t inflight
, uint64_t *used
, uint64_t *ref_rsrv
)
3307 ASSERT3S(asize
, >, 0);
3310 * *ref_rsrv is the portion of asize that will come from any
3311 * unconsumed refreservation space.
3315 mutex_enter(&ds
->ds_lock
);
3317 * Make a space adjustment for reserved bytes.
3319 if (ds
->ds_reserved
> ds
->ds_phys
->ds_unique_bytes
) {
3321 ds
->ds_reserved
- ds
->ds_phys
->ds_unique_bytes
);
3322 *used
-= (ds
->ds_reserved
- ds
->ds_phys
->ds_unique_bytes
);
3324 asize
- MIN(asize
, parent_delta(ds
, asize
+ inflight
));
3327 if (!check_quota
|| ds
->ds_quota
== 0) {
3328 mutex_exit(&ds
->ds_lock
);
3332 * If they are requesting more space, and our current estimate
3333 * is over quota, they get to try again unless the actual
3334 * on-disk is over quota and there are no pending changes (which
3335 * may free up space for us).
3337 if (ds
->ds_phys
->ds_used_bytes
+ inflight
>= ds
->ds_quota
) {
3338 if (inflight
> 0 || ds
->ds_phys
->ds_used_bytes
< ds
->ds_quota
)
3343 mutex_exit(&ds
->ds_lock
);
3350 dsl_dataset_set_quota_check(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
3352 dsl_dataset_t
*ds
= arg1
;
3353 dsl_prop_setarg_t
*psa
= arg2
;
3356 if (spa_version(ds
->ds_dir
->dd_pool
->dp_spa
) < SPA_VERSION_REFQUOTA
)
3359 if ((err
= dsl_prop_predict_sync(ds
->ds_dir
, psa
)) != 0)
3362 if (psa
->psa_effective_value
== 0)
3365 if (psa
->psa_effective_value
< ds
->ds_phys
->ds_used_bytes
||
3366 psa
->psa_effective_value
< ds
->ds_reserved
)
3372 extern void dsl_prop_set_sync(void *, void *, dmu_tx_t
*);
3375 dsl_dataset_set_quota_sync(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
3377 dsl_dataset_t
*ds
= arg1
;
3378 dsl_prop_setarg_t
*psa
= arg2
;
3379 uint64_t effective_value
= psa
->psa_effective_value
;
3381 dsl_prop_set_sync(ds
, psa
, tx
);
3382 DSL_PROP_CHECK_PREDICTION(ds
->ds_dir
, psa
);
3384 if (ds
->ds_quota
!= effective_value
) {
3385 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
3386 ds
->ds_quota
= effective_value
;
3388 spa_history_log_internal(LOG_DS_REFQUOTA
,
3389 ds
->ds_dir
->dd_pool
->dp_spa
, tx
, "%lld dataset = %llu ",
3390 (longlong_t
)ds
->ds_quota
, ds
->ds_object
);
3395 dsl_dataset_set_quota(const char *dsname
, zprop_source_t source
, uint64_t quota
)
3398 dsl_prop_setarg_t psa
;
3401 dsl_prop_setarg_init_uint64(&psa
, "refquota", source
, "a
);
3403 err
= dsl_dataset_hold(dsname
, FTAG
, &ds
);
3408 * If someone removes a file, then tries to set the quota, we
3409 * want to make sure the file freeing takes effect.
3411 txg_wait_open(ds
->ds_dir
->dd_pool
, 0);
3413 err
= dsl_sync_task_do(ds
->ds_dir
->dd_pool
,
3414 dsl_dataset_set_quota_check
, dsl_dataset_set_quota_sync
,
3417 dsl_dataset_rele(ds
, FTAG
);
3422 dsl_dataset_set_reservation_check(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
3424 dsl_dataset_t
*ds
= arg1
;
3425 dsl_prop_setarg_t
*psa
= arg2
;
3426 uint64_t effective_value
;
3430 if (spa_version(ds
->ds_dir
->dd_pool
->dp_spa
) <
3431 SPA_VERSION_REFRESERVATION
)
3434 if (dsl_dataset_is_snapshot(ds
))
3437 if ((err
= dsl_prop_predict_sync(ds
->ds_dir
, psa
)) != 0)
3440 effective_value
= psa
->psa_effective_value
;
3443 * If we are doing the preliminary check in open context, the
3444 * space estimates may be inaccurate.
3446 if (!dmu_tx_is_syncing(tx
))
3449 mutex_enter(&ds
->ds_lock
);
3450 if (!DS_UNIQUE_IS_ACCURATE(ds
))
3451 dsl_dataset_recalc_head_uniq(ds
);
3452 unique
= ds
->ds_phys
->ds_unique_bytes
;
3453 mutex_exit(&ds
->ds_lock
);
3455 if (MAX(unique
, effective_value
) > MAX(unique
, ds
->ds_reserved
)) {
3456 uint64_t delta
= MAX(unique
, effective_value
) -
3457 MAX(unique
, ds
->ds_reserved
);
3459 if (delta
> dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
))
3461 if (ds
->ds_quota
> 0 &&
3462 effective_value
> ds
->ds_quota
)
3470 dsl_dataset_set_reservation_sync(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
3472 dsl_dataset_t
*ds
= arg1
;
3473 dsl_prop_setarg_t
*psa
= arg2
;
3474 uint64_t effective_value
= psa
->psa_effective_value
;
3478 dsl_prop_set_sync(ds
, psa
, tx
);
3479 DSL_PROP_CHECK_PREDICTION(ds
->ds_dir
, psa
);
3481 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
3483 mutex_enter(&ds
->ds_dir
->dd_lock
);
3484 mutex_enter(&ds
->ds_lock
);
3485 ASSERT(DS_UNIQUE_IS_ACCURATE(ds
));
3486 unique
= ds
->ds_phys
->ds_unique_bytes
;
3487 delta
= MAX(0, (int64_t)(effective_value
- unique
)) -
3488 MAX(0, (int64_t)(ds
->ds_reserved
- unique
));
3489 ds
->ds_reserved
= effective_value
;
3490 mutex_exit(&ds
->ds_lock
);
3492 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_REFRSRV
, delta
, 0, 0, tx
);
3493 mutex_exit(&ds
->ds_dir
->dd_lock
);
3495 spa_history_log_internal(LOG_DS_REFRESERV
,
3496 ds
->ds_dir
->dd_pool
->dp_spa
, tx
, "%lld dataset = %llu",
3497 (longlong_t
)effective_value
, ds
->ds_object
);
3501 dsl_dataset_set_reservation(const char *dsname
, zprop_source_t source
,
3502 uint64_t reservation
)
3505 dsl_prop_setarg_t psa
;
3508 dsl_prop_setarg_init_uint64(&psa
, "refreservation", source
,
3511 err
= dsl_dataset_hold(dsname
, FTAG
, &ds
);
3515 err
= dsl_sync_task_do(ds
->ds_dir
->dd_pool
,
3516 dsl_dataset_set_reservation_check
,
3517 dsl_dataset_set_reservation_sync
, ds
, &psa
, 0);
3519 dsl_dataset_rele(ds
, FTAG
);
3523 typedef struct zfs_hold_cleanup_arg
{
3526 char htag
[MAXNAMELEN
];
3527 } zfs_hold_cleanup_arg_t
;
3530 dsl_dataset_user_release_onexit(void *arg
)
3532 zfs_hold_cleanup_arg_t
*ca
= arg
;
3534 (void) dsl_dataset_user_release_tmp(ca
->dp
, ca
->dsobj
, ca
->htag
,
3536 kmem_free(ca
, sizeof (zfs_hold_cleanup_arg_t
));
3540 dsl_register_onexit_hold_cleanup(dsl_dataset_t
*ds
, const char *htag
,
3543 zfs_hold_cleanup_arg_t
*ca
;
3545 ca
= kmem_alloc(sizeof (zfs_hold_cleanup_arg_t
), KM_SLEEP
);
3546 ca
->dp
= ds
->ds_dir
->dd_pool
;
3547 ca
->dsobj
= ds
->ds_object
;
3548 (void) strlcpy(ca
->htag
, htag
, sizeof (ca
->htag
));
3549 VERIFY3U(0, ==, zfs_onexit_add_cb(minor
,
3550 dsl_dataset_user_release_onexit
, ca
, NULL
));
3554 * If you add new checks here, you may need to add
3555 * additional checks to the "temporary" case in
3556 * snapshot_check() in dmu_objset.c.
3559 dsl_dataset_user_hold_check(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
3561 dsl_dataset_t
*ds
= arg1
;
3562 struct dsl_ds_holdarg
*ha
= arg2
;
3563 char *htag
= ha
->htag
;
3564 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
3567 if (spa_version(ds
->ds_dir
->dd_pool
->dp_spa
) < SPA_VERSION_USERREFS
)
3570 if (!dsl_dataset_is_snapshot(ds
))
3573 /* tags must be unique */
3574 mutex_enter(&ds
->ds_lock
);
3575 if (ds
->ds_phys
->ds_userrefs_obj
) {
3576 error
= zap_lookup(mos
, ds
->ds_phys
->ds_userrefs_obj
, htag
,
3580 else if (error
== ENOENT
)
3583 mutex_exit(&ds
->ds_lock
);
3585 if (error
== 0 && ha
->temphold
&&
3586 strlen(htag
) + MAX_TAG_PREFIX_LEN
>= MAXNAMELEN
)
3593 dsl_dataset_user_hold_sync(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
3595 dsl_dataset_t
*ds
= arg1
;
3596 struct dsl_ds_holdarg
*ha
= arg2
;
3597 char *htag
= ha
->htag
;
3598 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
3599 objset_t
*mos
= dp
->dp_meta_objset
;
3600 uint64_t now
= gethrestime_sec();
3603 mutex_enter(&ds
->ds_lock
);
3604 if (ds
->ds_phys
->ds_userrefs_obj
== 0) {
3606 * This is the first user hold for this dataset. Create
3607 * the userrefs zap object.
3609 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
3610 zapobj
= ds
->ds_phys
->ds_userrefs_obj
=
3611 zap_create(mos
, DMU_OT_USERREFS
, DMU_OT_NONE
, 0, tx
);
3613 zapobj
= ds
->ds_phys
->ds_userrefs_obj
;
3616 mutex_exit(&ds
->ds_lock
);
3618 VERIFY(0 == zap_add(mos
, zapobj
, htag
, 8, 1, &now
, tx
));
3621 VERIFY(0 == dsl_pool_user_hold(dp
, ds
->ds_object
,
3625 spa_history_log_internal(LOG_DS_USER_HOLD
,
3626 dp
->dp_spa
, tx
, "<%s> temp = %d dataset = %llu", htag
,
3627 (int)ha
->temphold
, ds
->ds_object
);
3631 dsl_dataset_user_hold_one(const char *dsname
, void *arg
)
3633 struct dsl_ds_holdarg
*ha
= arg
;
3638 /* alloc a buffer to hold dsname@snapname plus terminating NULL */
3639 name
= kmem_asprintf("%s@%s", dsname
, ha
->snapname
);
3640 error
= dsl_dataset_hold(name
, ha
->dstg
, &ds
);
3643 ha
->gotone
= B_TRUE
;
3644 dsl_sync_task_create(ha
->dstg
, dsl_dataset_user_hold_check
,
3645 dsl_dataset_user_hold_sync
, ds
, ha
, 0);
3646 } else if (error
== ENOENT
&& ha
->recursive
) {
3649 (void) strlcpy(ha
->failed
, dsname
, sizeof (ha
->failed
));
3655 dsl_dataset_user_hold_for_send(dsl_dataset_t
*ds
, char *htag
,
3658 struct dsl_ds_holdarg
*ha
;
3661 ha
= kmem_zalloc(sizeof (struct dsl_ds_holdarg
), KM_SLEEP
);
3663 ha
->temphold
= temphold
;
3664 error
= dsl_sync_task_do(ds
->ds_dir
->dd_pool
,
3665 dsl_dataset_user_hold_check
, dsl_dataset_user_hold_sync
,
3667 kmem_free(ha
, sizeof (struct dsl_ds_holdarg
));
3673 dsl_dataset_user_hold(char *dsname
, char *snapname
, char *htag
,
3674 boolean_t recursive
, boolean_t temphold
, int cleanup_fd
)
3676 struct dsl_ds_holdarg
*ha
;
3677 dsl_sync_task_t
*dst
;
3682 if (cleanup_fd
!= -1) {
3683 /* Currently we only support cleanup-on-exit of tempholds. */
3686 error
= zfs_onexit_fd_hold(cleanup_fd
, &minor
);
3691 ha
= kmem_zalloc(sizeof (struct dsl_ds_holdarg
), KM_SLEEP
);
3693 (void) strlcpy(ha
->failed
, dsname
, sizeof (ha
->failed
));
3695 error
= spa_open(dsname
, &spa
, FTAG
);
3697 kmem_free(ha
, sizeof (struct dsl_ds_holdarg
));
3698 if (cleanup_fd
!= -1)
3699 zfs_onexit_fd_rele(cleanup_fd
);
3703 ha
->dstg
= dsl_sync_task_group_create(spa_get_dsl(spa
));
3705 ha
->snapname
= snapname
;
3706 ha
->recursive
= recursive
;
3707 ha
->temphold
= temphold
;
3710 error
= dmu_objset_find(dsname
, dsl_dataset_user_hold_one
,
3711 ha
, DS_FIND_CHILDREN
);
3713 error
= dsl_dataset_user_hold_one(dsname
, ha
);
3716 error
= dsl_sync_task_group_wait(ha
->dstg
);
3718 for (dst
= list_head(&ha
->dstg
->dstg_tasks
); dst
;
3719 dst
= list_next(&ha
->dstg
->dstg_tasks
, dst
)) {
3720 dsl_dataset_t
*ds
= dst
->dst_arg1
;
3723 dsl_dataset_name(ds
, ha
->failed
);
3724 *strchr(ha
->failed
, '@') = '\0';
3725 } else if (error
== 0 && minor
!= 0 && temphold
) {
3727 * If this hold is to be released upon process exit,
3728 * register that action now.
3730 dsl_register_onexit_hold_cleanup(ds
, htag
, minor
);
3732 dsl_dataset_rele(ds
, ha
->dstg
);
3735 if (error
== 0 && recursive
&& !ha
->gotone
)
3739 (void) strlcpy(dsname
, ha
->failed
, sizeof (ha
->failed
));
3741 dsl_sync_task_group_destroy(ha
->dstg
);
3743 kmem_free(ha
, sizeof (struct dsl_ds_holdarg
));
3744 spa_close(spa
, FTAG
);
3745 if (cleanup_fd
!= -1)
3746 zfs_onexit_fd_rele(cleanup_fd
);
3750 struct dsl_ds_releasearg
{
3753 boolean_t own
; /* do we own or just hold ds? */
3757 dsl_dataset_release_might_destroy(dsl_dataset_t
*ds
, const char *htag
,
3758 boolean_t
*might_destroy
)
3760 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
3765 *might_destroy
= B_FALSE
;
3767 mutex_enter(&ds
->ds_lock
);
3768 zapobj
= ds
->ds_phys
->ds_userrefs_obj
;
3770 /* The tag can't possibly exist */
3771 mutex_exit(&ds
->ds_lock
);
3775 /* Make sure the tag exists */
3776 error
= zap_lookup(mos
, zapobj
, htag
, 8, 1, &tmp
);
3778 mutex_exit(&ds
->ds_lock
);
3779 if (error
== ENOENT
)
3784 if (ds
->ds_userrefs
== 1 && ds
->ds_phys
->ds_num_children
== 1 &&
3785 DS_IS_DEFER_DESTROY(ds
))
3786 *might_destroy
= B_TRUE
;
3788 mutex_exit(&ds
->ds_lock
);
3793 dsl_dataset_user_release_check(void *arg1
, void *tag
, dmu_tx_t
*tx
)
3795 struct dsl_ds_releasearg
*ra
= arg1
;
3796 dsl_dataset_t
*ds
= ra
->ds
;
3797 boolean_t might_destroy
;
3800 if (spa_version(ds
->ds_dir
->dd_pool
->dp_spa
) < SPA_VERSION_USERREFS
)
3803 error
= dsl_dataset_release_might_destroy(ds
, ra
->htag
, &might_destroy
);
3807 if (might_destroy
) {
3808 struct dsl_ds_destroyarg dsda
= {0};
3810 if (dmu_tx_is_syncing(tx
)) {
3812 * If we're not prepared to remove the snapshot,
3813 * we can't allow the release to happen right now.
3819 dsda
.releasing
= B_TRUE
;
3820 return (dsl_dataset_destroy_check(&dsda
, tag
, tx
));
3827 dsl_dataset_user_release_sync(void *arg1
, void *tag
, dmu_tx_t
*tx
)
3829 struct dsl_ds_releasearg
*ra
= arg1
;
3830 dsl_dataset_t
*ds
= ra
->ds
;
3831 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
3832 objset_t
*mos
= dp
->dp_meta_objset
;
3834 uint64_t dsobj
= ds
->ds_object
;
3838 mutex_enter(&ds
->ds_lock
);
3840 refs
= ds
->ds_userrefs
;
3841 mutex_exit(&ds
->ds_lock
);
3842 error
= dsl_pool_user_release(dp
, ds
->ds_object
, ra
->htag
, tx
);
3843 VERIFY(error
== 0 || error
== ENOENT
);
3844 zapobj
= ds
->ds_phys
->ds_userrefs_obj
;
3845 VERIFY(0 == zap_remove(mos
, zapobj
, ra
->htag
, tx
));
3846 if (ds
->ds_userrefs
== 0 && ds
->ds_phys
->ds_num_children
== 1 &&
3847 DS_IS_DEFER_DESTROY(ds
)) {
3848 struct dsl_ds_destroyarg dsda
= {0};
3852 dsda
.releasing
= B_TRUE
;
3853 /* We already did the destroy_check */
3854 dsl_dataset_destroy_sync(&dsda
, tag
, tx
);
3857 spa_history_log_internal(LOG_DS_USER_RELEASE
,
3858 dp
->dp_spa
, tx
, "<%s> %lld dataset = %llu",
3859 ra
->htag
, (longlong_t
)refs
, dsobj
);
3863 dsl_dataset_user_release_one(const char *dsname
, void *arg
)
3865 struct dsl_ds_holdarg
*ha
= arg
;
3866 struct dsl_ds_releasearg
*ra
;
3869 void *dtag
= ha
->dstg
;
3871 boolean_t own
= B_FALSE
;
3872 boolean_t might_destroy
;
3874 /* alloc a buffer to hold dsname@snapname, plus the terminating NULL */
3875 name
= kmem_asprintf("%s@%s", dsname
, ha
->snapname
);
3876 error
= dsl_dataset_hold(name
, dtag
, &ds
);
3878 if (error
== ENOENT
&& ha
->recursive
)
3880 (void) strlcpy(ha
->failed
, dsname
, sizeof (ha
->failed
));
3884 ha
->gotone
= B_TRUE
;
3886 ASSERT(dsl_dataset_is_snapshot(ds
));
3888 error
= dsl_dataset_release_might_destroy(ds
, ha
->htag
, &might_destroy
);
3890 dsl_dataset_rele(ds
, dtag
);
3894 if (might_destroy
) {
3896 name
= kmem_asprintf("%s@%s", dsname
, ha
->snapname
);
3897 error
= zfs_unmount_snap(name
, NULL
);
3900 dsl_dataset_rele(ds
, dtag
);
3904 if (!dsl_dataset_tryown(ds
, B_TRUE
, dtag
)) {
3905 dsl_dataset_rele(ds
, dtag
);
3909 dsl_dataset_make_exclusive(ds
, dtag
);
3913 ra
= kmem_alloc(sizeof (struct dsl_ds_releasearg
), KM_SLEEP
);
3915 ra
->htag
= ha
->htag
;
3917 dsl_sync_task_create(ha
->dstg
, dsl_dataset_user_release_check
,
3918 dsl_dataset_user_release_sync
, ra
, dtag
, 0);
3924 dsl_dataset_user_release(char *dsname
, char *snapname
, char *htag
,
3925 boolean_t recursive
)
3927 struct dsl_ds_holdarg
*ha
;
3928 dsl_sync_task_t
*dst
;
3933 ha
= kmem_zalloc(sizeof (struct dsl_ds_holdarg
), KM_SLEEP
);
3935 (void) strlcpy(ha
->failed
, dsname
, sizeof (ha
->failed
));
3937 error
= spa_open(dsname
, &spa
, FTAG
);
3939 kmem_free(ha
, sizeof (struct dsl_ds_holdarg
));
3943 ha
->dstg
= dsl_sync_task_group_create(spa_get_dsl(spa
));
3945 ha
->snapname
= snapname
;
3946 ha
->recursive
= recursive
;
3948 error
= dmu_objset_find(dsname
, dsl_dataset_user_release_one
,
3949 ha
, DS_FIND_CHILDREN
);
3951 error
= dsl_dataset_user_release_one(dsname
, ha
);
3954 error
= dsl_sync_task_group_wait(ha
->dstg
);
3956 for (dst
= list_head(&ha
->dstg
->dstg_tasks
); dst
;
3957 dst
= list_next(&ha
->dstg
->dstg_tasks
, dst
)) {
3958 struct dsl_ds_releasearg
*ra
= dst
->dst_arg1
;
3959 dsl_dataset_t
*ds
= ra
->ds
;
3962 dsl_dataset_name(ds
, ha
->failed
);
3965 dsl_dataset_disown(ds
, ha
->dstg
);
3967 dsl_dataset_rele(ds
, ha
->dstg
);
3969 kmem_free(ra
, sizeof (struct dsl_ds_releasearg
));
3972 if (error
== 0 && recursive
&& !ha
->gotone
)
3975 if (error
&& error
!= EBUSY
)
3976 (void) strlcpy(dsname
, ha
->failed
, sizeof (ha
->failed
));
3978 dsl_sync_task_group_destroy(ha
->dstg
);
3979 kmem_free(ha
, sizeof (struct dsl_ds_holdarg
));
3980 spa_close(spa
, FTAG
);
3983 * We can get EBUSY if we were racing with deferred destroy and
3984 * dsl_dataset_user_release_check() hadn't done the necessary
3985 * open context setup. We can also get EBUSY if we're racing
3986 * with destroy and that thread is the ds_owner. Either way
3987 * the busy condition should be transient, and we should retry
3988 * the release operation.
3997 * Called at spa_load time (with retry == B_FALSE) to release a stale
3998 * temporary user hold. Also called by the onexit code (with retry == B_TRUE).
4001 dsl_dataset_user_release_tmp(dsl_pool_t
*dp
, uint64_t dsobj
, char *htag
,
4011 rw_enter(&dp
->dp_config_rwlock
, RW_READER
);
4012 error
= dsl_dataset_hold_obj(dp
, dsobj
, FTAG
, &ds
);
4013 rw_exit(&dp
->dp_config_rwlock
);
4016 namelen
= dsl_dataset_namelen(ds
)+1;
4017 name
= kmem_alloc(namelen
, KM_SLEEP
);
4018 dsl_dataset_name(ds
, name
);
4019 dsl_dataset_rele(ds
, FTAG
);
4021 snap
= strchr(name
, '@');
4024 error
= dsl_dataset_user_release(name
, snap
, htag
, B_FALSE
);
4025 kmem_free(name
, namelen
);
4028 * The object can't have been destroyed because we have a hold,
4029 * but it might have been renamed, resulting in ENOENT. Retry
4030 * if we've been requested to do so.
4032 * It would be nice if we could use the dsobj all the way
4033 * through and avoid ENOENT entirely. But we might need to
4034 * unmount the snapshot, and there's currently no way to lookup
4035 * a vfsp using a ZFS object id.
4037 } while ((error
== ENOENT
) && retry
);
4043 dsl_dataset_get_holds(const char *dsname
, nvlist_t
**nvp
)
4048 err
= dsl_dataset_hold(dsname
, FTAG
, &ds
);
4052 VERIFY(0 == nvlist_alloc(nvp
, NV_UNIQUE_NAME
, KM_SLEEP
));
4053 if (ds
->ds_phys
->ds_userrefs_obj
!= 0) {
4054 zap_attribute_t
*za
;
4057 za
= kmem_alloc(sizeof (zap_attribute_t
), KM_SLEEP
);
4058 for (zap_cursor_init(&zc
, ds
->ds_dir
->dd_pool
->dp_meta_objset
,
4059 ds
->ds_phys
->ds_userrefs_obj
);
4060 zap_cursor_retrieve(&zc
, za
) == 0;
4061 zap_cursor_advance(&zc
)) {
4062 VERIFY(0 == nvlist_add_uint64(*nvp
, za
->za_name
,
4063 za
->za_first_integer
));
4065 zap_cursor_fini(&zc
);
4066 kmem_free(za
, sizeof (zap_attribute_t
));
4068 dsl_dataset_rele(ds
, FTAG
);
4073 * Note, this function is used as the callback for dmu_objset_find(). We
4074 * always return 0 so that we will continue to find and process
4075 * inconsistent datasets, even if we encounter an error trying to
4076 * process one of them.
4080 dsl_destroy_inconsistent(const char *dsname
, void *arg
)
4084 if (dsl_dataset_own(dsname
, B_TRUE
, FTAG
, &ds
) == 0) {
4085 if (DS_IS_INCONSISTENT(ds
))
4086 (void) dsl_dataset_destroy(ds
, FTAG
, B_FALSE
);
4088 dsl_dataset_disown(ds
, FTAG
);
4094 * Return (in *usedp) the amount of space written in new that is not
4095 * present in oldsnap. New may be a snapshot or the head. Old must be
4096 * a snapshot before new, in new's filesystem (or its origin). If not then
4097 * fail and return EINVAL.
4099 * The written space is calculated by considering two components: First, we
4100 * ignore any freed space, and calculate the written as new's used space
4101 * minus old's used space. Next, we add in the amount of space that was freed
4102 * between the two snapshots, thus reducing new's used space relative to old's.
4103 * Specifically, this is the space that was born before old->ds_creation_txg,
4104 * and freed before new (ie. on new's deadlist or a previous deadlist).
4106 * space freed [---------------------]
4107 * snapshots ---O-------O--------O-------O------
4111 dsl_dataset_space_written(dsl_dataset_t
*oldsnap
, dsl_dataset_t
*new,
4112 uint64_t *usedp
, uint64_t *compp
, uint64_t *uncompp
)
4116 dsl_pool_t
*dp
= new->ds_dir
->dd_pool
;
4119 *usedp
+= new->ds_phys
->ds_used_bytes
;
4120 *usedp
-= oldsnap
->ds_phys
->ds_used_bytes
;
4123 *compp
+= new->ds_phys
->ds_compressed_bytes
;
4124 *compp
-= oldsnap
->ds_phys
->ds_compressed_bytes
;
4127 *uncompp
+= new->ds_phys
->ds_uncompressed_bytes
;
4128 *uncompp
-= oldsnap
->ds_phys
->ds_uncompressed_bytes
;
4130 rw_enter(&dp
->dp_config_rwlock
, RW_READER
);
4131 snapobj
= new->ds_object
;
4132 while (snapobj
!= oldsnap
->ds_object
) {
4133 dsl_dataset_t
*snap
;
4134 uint64_t used
, comp
, uncomp
;
4136 err
= dsl_dataset_hold_obj(dp
, snapobj
, FTAG
, &snap
);
4140 if (snap
->ds_phys
->ds_prev_snap_txg
==
4141 oldsnap
->ds_phys
->ds_creation_txg
) {
4143 * The blocks in the deadlist can not be born after
4144 * ds_prev_snap_txg, so get the whole deadlist space,
4145 * which is more efficient (especially for old-format
4146 * deadlists). Unfortunately the deadlist code
4147 * doesn't have enough information to make this
4148 * optimization itself.
4150 dsl_deadlist_space(&snap
->ds_deadlist
,
4151 &used
, &comp
, &uncomp
);
4153 dsl_deadlist_space_range(&snap
->ds_deadlist
,
4154 0, oldsnap
->ds_phys
->ds_creation_txg
,
4155 &used
, &comp
, &uncomp
);
4162 * If we get to the beginning of the chain of snapshots
4163 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
4164 * was not a snapshot of/before new.
4166 snapobj
= snap
->ds_phys
->ds_prev_snap_obj
;
4167 dsl_dataset_rele(snap
, FTAG
);
4174 rw_exit(&dp
->dp_config_rwlock
);
4179 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
4180 * lastsnap, and all snapshots in between are deleted.
4182 * blocks that would be freed [---------------------------]
4183 * snapshots ---O-------O--------O-------O--------O
4184 * firstsnap lastsnap
4186 * This is the set of blocks that were born after the snap before firstsnap,
4187 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
4188 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
4189 * We calculate this by iterating over the relevant deadlists (from the snap
4190 * after lastsnap, backward to the snap after firstsnap), summing up the
4191 * space on the deadlist that was born after the snap before firstsnap.
4194 dsl_dataset_space_wouldfree(dsl_dataset_t
*firstsnap
,
4195 dsl_dataset_t
*lastsnap
,
4196 uint64_t *usedp
, uint64_t *compp
, uint64_t *uncompp
)
4200 dsl_pool_t
*dp
= firstsnap
->ds_dir
->dd_pool
;
4202 ASSERT(dsl_dataset_is_snapshot(firstsnap
));
4203 ASSERT(dsl_dataset_is_snapshot(lastsnap
));
4206 * Check that the snapshots are in the same dsl_dir, and firstsnap
4207 * is before lastsnap.
4209 if (firstsnap
->ds_dir
!= lastsnap
->ds_dir
||
4210 firstsnap
->ds_phys
->ds_creation_txg
>
4211 lastsnap
->ds_phys
->ds_creation_txg
)
4214 *usedp
= *compp
= *uncompp
= 0;
4216 rw_enter(&dp
->dp_config_rwlock
, RW_READER
);
4217 snapobj
= lastsnap
->ds_phys
->ds_next_snap_obj
;
4218 while (snapobj
!= firstsnap
->ds_object
) {
4220 uint64_t used
, comp
, uncomp
;
4222 err
= dsl_dataset_hold_obj(dp
, snapobj
, FTAG
, &ds
);
4226 dsl_deadlist_space_range(&ds
->ds_deadlist
,
4227 firstsnap
->ds_phys
->ds_prev_snap_txg
, UINT64_MAX
,
4228 &used
, &comp
, &uncomp
);
4233 snapobj
= ds
->ds_phys
->ds_prev_snap_obj
;
4234 ASSERT3U(snapobj
, !=, 0);
4235 dsl_dataset_rele(ds
, FTAG
);
4237 rw_exit(&dp
->dp_config_rwlock
);