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) 2013 by Delphix. All rights reserved.
24 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
25 * Copyright (c) 2014 RackTop Systems.
28 #include <sys/dmu_objset.h>
29 #include <sys/dsl_dataset.h>
30 #include <sys/dsl_dir.h>
31 #include <sys/dsl_prop.h>
32 #include <sys/dsl_synctask.h>
33 #include <sys/dmu_traverse.h>
34 #include <sys/dmu_impl.h>
35 #include <sys/dmu_tx.h>
39 #include <sys/zfeature.h>
40 #include <sys/unique.h>
41 #include <sys/zfs_context.h>
42 #include <sys/zfs_ioctl.h>
44 #include <sys/zfs_znode.h>
45 #include <sys/zfs_onexit.h>
47 #include <sys/dsl_scan.h>
48 #include <sys/dsl_deadlist.h>
49 #include <sys/dsl_destroy.h>
50 #include <sys/dsl_userhold.h>
51 #include <sys/dsl_bookmark.h>
53 #define SWITCH64(x, y) \
55 uint64_t __tmp = (x); \
60 #define DS_REF_MAX (1ULL << 62)
62 #define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE
65 * Figure out how much of this delta should be propogated to the dsl_dir
66 * layer. If there's a refreservation, that space has already been
67 * partially accounted for in our ancestors.
70 parent_delta(dsl_dataset_t
*ds
, int64_t delta
)
72 uint64_t old_bytes
, new_bytes
;
74 if (ds
->ds_reserved
== 0)
77 old_bytes
= MAX(ds
->ds_phys
->ds_unique_bytes
, ds
->ds_reserved
);
78 new_bytes
= MAX(ds
->ds_phys
->ds_unique_bytes
+ delta
, ds
->ds_reserved
);
80 ASSERT3U(ABS((int64_t)(new_bytes
- old_bytes
)), <=, ABS(delta
));
81 return (new_bytes
- old_bytes
);
85 dsl_dataset_block_born(dsl_dataset_t
*ds
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
87 int used
, compressed
, uncompressed
;
90 used
= bp_get_dsize_sync(tx
->tx_pool
->dp_spa
, bp
);
91 compressed
= BP_GET_PSIZE(bp
);
92 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 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp
)));
103 dsl_pool_mos_diduse_space(tx
->tx_pool
,
104 used
, compressed
, uncompressed
);
108 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
109 mutex_enter(&ds
->ds_lock
);
110 delta
= parent_delta(ds
, used
);
111 ds
->ds_phys
->ds_referenced_bytes
+= used
;
112 ds
->ds_phys
->ds_compressed_bytes
+= compressed
;
113 ds
->ds_phys
->ds_uncompressed_bytes
+= uncompressed
;
114 ds
->ds_phys
->ds_unique_bytes
+= used
;
115 mutex_exit(&ds
->ds_lock
);
116 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_HEAD
, delta
,
117 compressed
, uncompressed
, tx
);
118 dsl_dir_transfer_space(ds
->ds_dir
, used
- delta
,
119 DD_USED_REFRSRV
, DD_USED_HEAD
, tx
);
123 dsl_dataset_block_kill(dsl_dataset_t
*ds
, const blkptr_t
*bp
, dmu_tx_t
*tx
,
126 int used
= bp_get_dsize_sync(tx
->tx_pool
->dp_spa
, bp
);
127 int compressed
= BP_GET_PSIZE(bp
);
128 int uncompressed
= BP_GET_UCSIZE(bp
);
133 ASSERT(dmu_tx_is_syncing(tx
));
134 ASSERT(bp
->blk_birth
<= tx
->tx_txg
);
137 dsl_free(tx
->tx_pool
, tx
->tx_txg
, bp
);
138 dsl_pool_mos_diduse_space(tx
->tx_pool
,
139 -used
, -compressed
, -uncompressed
);
142 ASSERT3P(tx
->tx_pool
, ==, ds
->ds_dir
->dd_pool
);
144 ASSERT(!dsl_dataset_is_snapshot(ds
));
145 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
147 if (bp
->blk_birth
> ds
->ds_phys
->ds_prev_snap_txg
) {
150 dprintf_bp(bp
, "freeing ds=%llu", ds
->ds_object
);
151 dsl_free(tx
->tx_pool
, tx
->tx_txg
, bp
);
153 mutex_enter(&ds
->ds_lock
);
154 ASSERT(ds
->ds_phys
->ds_unique_bytes
>= used
||
155 !DS_UNIQUE_IS_ACCURATE(ds
));
156 delta
= parent_delta(ds
, -used
);
157 ds
->ds_phys
->ds_unique_bytes
-= used
;
158 mutex_exit(&ds
->ds_lock
);
159 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_HEAD
,
160 delta
, -compressed
, -uncompressed
, tx
);
161 dsl_dir_transfer_space(ds
->ds_dir
, -used
- delta
,
162 DD_USED_REFRSRV
, DD_USED_HEAD
, tx
);
164 dprintf_bp(bp
, "putting on dead list: %s", "");
167 * We are here as part of zio's write done callback,
168 * which means we're a zio interrupt thread. We can't
169 * call dsl_deadlist_insert() now because it may block
170 * waiting for I/O. Instead, put bp on the deferred
171 * queue and let dsl_pool_sync() finish the job.
173 bplist_append(&ds
->ds_pending_deadlist
, bp
);
175 dsl_deadlist_insert(&ds
->ds_deadlist
, bp
, tx
);
177 ASSERT3U(ds
->ds_prev
->ds_object
, ==,
178 ds
->ds_phys
->ds_prev_snap_obj
);
179 ASSERT(ds
->ds_prev
->ds_phys
->ds_num_children
> 0);
180 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
181 if (ds
->ds_prev
->ds_phys
->ds_next_snap_obj
==
182 ds
->ds_object
&& bp
->blk_birth
>
183 ds
->ds_prev
->ds_phys
->ds_prev_snap_txg
) {
184 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
185 mutex_enter(&ds
->ds_prev
->ds_lock
);
186 ds
->ds_prev
->ds_phys
->ds_unique_bytes
+= used
;
187 mutex_exit(&ds
->ds_prev
->ds_lock
);
189 if (bp
->blk_birth
> ds
->ds_dir
->dd_origin_txg
) {
190 dsl_dir_transfer_space(ds
->ds_dir
, used
,
191 DD_USED_HEAD
, DD_USED_SNAP
, tx
);
194 mutex_enter(&ds
->ds_lock
);
195 ASSERT3U(ds
->ds_phys
->ds_referenced_bytes
, >=, used
);
196 ds
->ds_phys
->ds_referenced_bytes
-= used
;
197 ASSERT3U(ds
->ds_phys
->ds_compressed_bytes
, >=, compressed
);
198 ds
->ds_phys
->ds_compressed_bytes
-= compressed
;
199 ASSERT3U(ds
->ds_phys
->ds_uncompressed_bytes
, >=, uncompressed
);
200 ds
->ds_phys
->ds_uncompressed_bytes
-= uncompressed
;
201 mutex_exit(&ds
->ds_lock
);
207 dsl_dataset_prev_snap_txg(dsl_dataset_t
*ds
)
209 uint64_t trysnap
= 0;
214 * The snapshot creation could fail, but that would cause an
215 * incorrect FALSE return, which would only result in an
216 * overestimation of the amount of space that an operation would
217 * consume, which is OK.
219 * There's also a small window where we could miss a pending
220 * snapshot, because we could set the sync task in the quiescing
221 * phase. So this should only be used as a guess.
223 if (ds
->ds_trysnap_txg
>
224 spa_last_synced_txg(ds
->ds_dir
->dd_pool
->dp_spa
))
225 trysnap
= ds
->ds_trysnap_txg
;
226 return (MAX(ds
->ds_phys
->ds_prev_snap_txg
, trysnap
));
230 dsl_dataset_block_freeable(dsl_dataset_t
*ds
, const blkptr_t
*bp
,
233 if (blk_birth
<= dsl_dataset_prev_snap_txg(ds
) ||
234 (bp
!= NULL
&& BP_IS_HOLE(bp
)))
237 ddt_prefetch(dsl_dataset_get_spa(ds
), bp
);
244 dsl_dataset_evict(dmu_buf_t
*db
, void *dsv
)
246 dsl_dataset_t
*ds
= dsv
;
248 ASSERT(ds
->ds_owner
== NULL
);
250 unique_remove(ds
->ds_fsid_guid
);
252 if (ds
->ds_objset
!= NULL
)
253 dmu_objset_evict(ds
->ds_objset
);
256 dsl_dataset_rele(ds
->ds_prev
, ds
);
260 bplist_destroy(&ds
->ds_pending_deadlist
);
261 if (ds
->ds_phys
->ds_deadlist_obj
!= 0)
262 dsl_deadlist_close(&ds
->ds_deadlist
);
264 dsl_dir_rele(ds
->ds_dir
, ds
);
266 ASSERT(!list_link_active(&ds
->ds_synced_link
));
268 mutex_destroy(&ds
->ds_lock
);
269 mutex_destroy(&ds
->ds_opening_lock
);
270 refcount_destroy(&ds
->ds_longholds
);
272 kmem_free(ds
, sizeof (dsl_dataset_t
));
276 dsl_dataset_get_snapname(dsl_dataset_t
*ds
)
278 dsl_dataset_phys_t
*headphys
;
281 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
282 objset_t
*mos
= dp
->dp_meta_objset
;
284 if (ds
->ds_snapname
[0])
286 if (ds
->ds_phys
->ds_next_snap_obj
== 0)
289 err
= dmu_bonus_hold(mos
, ds
->ds_dir
->dd_phys
->dd_head_dataset_obj
,
293 headphys
= headdbuf
->db_data
;
294 err
= zap_value_search(dp
->dp_meta_objset
,
295 headphys
->ds_snapnames_zapobj
, ds
->ds_object
, 0, ds
->ds_snapname
);
296 dmu_buf_rele(headdbuf
, FTAG
);
301 dsl_dataset_snap_lookup(dsl_dataset_t
*ds
, const char *name
, uint64_t *value
)
303 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
304 uint64_t snapobj
= ds
->ds_phys
->ds_snapnames_zapobj
;
308 if (ds
->ds_phys
->ds_flags
& DS_FLAG_CI_DATASET
)
313 err
= zap_lookup_norm(mos
, snapobj
, name
, 8, 1,
314 value
, mt
, NULL
, 0, NULL
);
315 if (err
== ENOTSUP
&& mt
== MT_FIRST
)
316 err
= zap_lookup(mos
, snapobj
, name
, 8, 1, value
);
321 dsl_dataset_snap_remove(dsl_dataset_t
*ds
, const char *name
, dmu_tx_t
*tx
)
323 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
324 uint64_t snapobj
= ds
->ds_phys
->ds_snapnames_zapobj
;
328 dsl_dir_snap_cmtime_update(ds
->ds_dir
);
330 if (ds
->ds_phys
->ds_flags
& DS_FLAG_CI_DATASET
)
335 err
= zap_remove_norm(mos
, snapobj
, name
, mt
, tx
);
336 if (err
== ENOTSUP
&& mt
== MT_FIRST
)
337 err
= zap_remove(mos
, snapobj
, name
, tx
);
342 dsl_dataset_hold_obj(dsl_pool_t
*dp
, uint64_t dsobj
, void *tag
,
345 objset_t
*mos
= dp
->dp_meta_objset
;
349 dmu_object_info_t doi
;
351 ASSERT(dsl_pool_config_held(dp
));
353 err
= dmu_bonus_hold(mos
, dsobj
, tag
, &dbuf
);
357 /* Make sure dsobj has the correct object type. */
358 dmu_object_info_from_db(dbuf
, &doi
);
359 if (doi
.doi_bonus_type
!= DMU_OT_DSL_DATASET
) {
360 dmu_buf_rele(dbuf
, tag
);
361 return (SET_ERROR(EINVAL
));
364 ds
= dmu_buf_get_user(dbuf
);
366 dsl_dataset_t
*winner
= NULL
;
368 ds
= kmem_zalloc(sizeof (dsl_dataset_t
), KM_PUSHPAGE
);
370 ds
->ds_object
= dsobj
;
371 ds
->ds_phys
= dbuf
->db_data
;
372 list_link_init(&ds
->ds_synced_link
);
374 mutex_init(&ds
->ds_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
375 mutex_init(&ds
->ds_opening_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
376 mutex_init(&ds
->ds_sendstream_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
377 refcount_create(&ds
->ds_longholds
);
379 bplist_create(&ds
->ds_pending_deadlist
);
380 dsl_deadlist_open(&ds
->ds_deadlist
,
381 mos
, ds
->ds_phys
->ds_deadlist_obj
);
383 list_create(&ds
->ds_sendstreams
, sizeof (dmu_sendarg_t
),
384 offsetof(dmu_sendarg_t
, dsa_link
));
387 err
= dsl_dir_hold_obj(dp
,
388 ds
->ds_phys
->ds_dir_obj
, NULL
, ds
, &ds
->ds_dir
);
391 mutex_destroy(&ds
->ds_lock
);
392 mutex_destroy(&ds
->ds_opening_lock
);
393 refcount_destroy(&ds
->ds_longholds
);
394 bplist_destroy(&ds
->ds_pending_deadlist
);
395 dsl_deadlist_close(&ds
->ds_deadlist
);
396 kmem_free(ds
, sizeof (dsl_dataset_t
));
397 dmu_buf_rele(dbuf
, tag
);
401 if (!dsl_dataset_is_snapshot(ds
)) {
402 ds
->ds_snapname
[0] = '\0';
403 if (ds
->ds_phys
->ds_prev_snap_obj
!= 0) {
404 err
= dsl_dataset_hold_obj(dp
,
405 ds
->ds_phys
->ds_prev_snap_obj
,
408 if (doi
.doi_type
== DMU_OTN_ZAP_METADATA
) {
409 int zaperr
= zap_lookup(mos
, ds
->ds_object
,
410 DS_FIELD_BOOKMARK_NAMES
,
411 sizeof (ds
->ds_bookmarks
), 1,
413 if (zaperr
!= ENOENT
)
417 if (zfs_flags
& ZFS_DEBUG_SNAPNAMES
)
418 err
= dsl_dataset_get_snapname(ds
);
419 if (err
== 0 && ds
->ds_phys
->ds_userrefs_obj
!= 0) {
421 ds
->ds_dir
->dd_pool
->dp_meta_objset
,
422 ds
->ds_phys
->ds_userrefs_obj
,
427 if (err
== 0 && !dsl_dataset_is_snapshot(ds
)) {
428 err
= dsl_prop_get_int_ds(ds
,
429 zfs_prop_to_name(ZFS_PROP_REFRESERVATION
),
432 err
= dsl_prop_get_int_ds(ds
,
433 zfs_prop_to_name(ZFS_PROP_REFQUOTA
),
437 ds
->ds_reserved
= ds
->ds_quota
= 0;
440 if (err
!= 0 || (winner
= dmu_buf_set_user_ie(dbuf
, ds
,
441 &ds
->ds_phys
, dsl_dataset_evict
)) != NULL
) {
442 bplist_destroy(&ds
->ds_pending_deadlist
);
443 dsl_deadlist_close(&ds
->ds_deadlist
);
445 dsl_dataset_rele(ds
->ds_prev
, ds
);
446 dsl_dir_rele(ds
->ds_dir
, ds
);
447 mutex_destroy(&ds
->ds_lock
);
448 mutex_destroy(&ds
->ds_opening_lock
);
449 refcount_destroy(&ds
->ds_longholds
);
450 kmem_free(ds
, sizeof (dsl_dataset_t
));
452 dmu_buf_rele(dbuf
, tag
);
458 unique_insert(ds
->ds_phys
->ds_fsid_guid
);
461 ASSERT3P(ds
->ds_dbuf
, ==, dbuf
);
462 ASSERT3P(ds
->ds_phys
, ==, dbuf
->db_data
);
463 ASSERT(ds
->ds_phys
->ds_prev_snap_obj
!= 0 ||
464 spa_version(dp
->dp_spa
) < SPA_VERSION_ORIGIN
||
465 dp
->dp_origin_snap
== NULL
|| ds
== dp
->dp_origin_snap
);
471 dsl_dataset_hold(dsl_pool_t
*dp
, const char *name
,
472 void *tag
, dsl_dataset_t
**dsp
)
475 const char *snapname
;
479 err
= dsl_dir_hold(dp
, name
, FTAG
, &dd
, &snapname
);
483 ASSERT(dsl_pool_config_held(dp
));
484 obj
= dd
->dd_phys
->dd_head_dataset_obj
;
486 err
= dsl_dataset_hold_obj(dp
, obj
, tag
, dsp
);
488 err
= SET_ERROR(ENOENT
);
490 /* we may be looking for a snapshot */
491 if (err
== 0 && snapname
!= NULL
) {
494 if (*snapname
++ != '@') {
495 dsl_dataset_rele(*dsp
, tag
);
496 dsl_dir_rele(dd
, FTAG
);
497 return (SET_ERROR(ENOENT
));
500 dprintf("looking for snapshot '%s'\n", snapname
);
501 err
= dsl_dataset_snap_lookup(*dsp
, snapname
, &obj
);
503 err
= dsl_dataset_hold_obj(dp
, obj
, tag
, &ds
);
504 dsl_dataset_rele(*dsp
, tag
);
507 mutex_enter(&ds
->ds_lock
);
508 if (ds
->ds_snapname
[0] == 0)
509 (void) strlcpy(ds
->ds_snapname
, snapname
,
510 sizeof (ds
->ds_snapname
));
511 mutex_exit(&ds
->ds_lock
);
516 dsl_dir_rele(dd
, FTAG
);
521 dsl_dataset_own_obj(dsl_pool_t
*dp
, uint64_t dsobj
,
522 void *tag
, dsl_dataset_t
**dsp
)
524 int err
= dsl_dataset_hold_obj(dp
, dsobj
, tag
, dsp
);
527 if (!dsl_dataset_tryown(*dsp
, tag
)) {
528 dsl_dataset_rele(*dsp
, tag
);
530 return (SET_ERROR(EBUSY
));
536 dsl_dataset_own(dsl_pool_t
*dp
, const char *name
,
537 void *tag
, dsl_dataset_t
**dsp
)
539 int err
= dsl_dataset_hold(dp
, name
, tag
, dsp
);
542 if (!dsl_dataset_tryown(*dsp
, tag
)) {
543 dsl_dataset_rele(*dsp
, tag
);
544 return (SET_ERROR(EBUSY
));
550 * See the comment above dsl_pool_hold() for details. In summary, a long
551 * hold is used to prevent destruction of a dataset while the pool hold
552 * is dropped, allowing other concurrent operations (e.g. spa_sync()).
554 * The dataset and pool must be held when this function is called. After it
555 * is called, the pool hold may be released while the dataset is still held
559 dsl_dataset_long_hold(dsl_dataset_t
*ds
, void *tag
)
561 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
562 (void) refcount_add(&ds
->ds_longholds
, tag
);
566 dsl_dataset_long_rele(dsl_dataset_t
*ds
, void *tag
)
568 (void) refcount_remove(&ds
->ds_longholds
, tag
);
571 /* Return B_TRUE if there are any long holds on this dataset. */
573 dsl_dataset_long_held(dsl_dataset_t
*ds
)
575 return (!refcount_is_zero(&ds
->ds_longholds
));
579 dsl_dataset_name(dsl_dataset_t
*ds
, char *name
)
582 (void) strcpy(name
, "mos");
584 dsl_dir_name(ds
->ds_dir
, name
);
585 VERIFY0(dsl_dataset_get_snapname(ds
));
586 if (ds
->ds_snapname
[0]) {
587 (void) strcat(name
, "@");
589 * We use a "recursive" mutex so that we
590 * can call dprintf_ds() with ds_lock held.
592 if (!MUTEX_HELD(&ds
->ds_lock
)) {
593 mutex_enter(&ds
->ds_lock
);
594 (void) strcat(name
, ds
->ds_snapname
);
595 mutex_exit(&ds
->ds_lock
);
597 (void) strcat(name
, ds
->ds_snapname
);
604 dsl_dataset_rele(dsl_dataset_t
*ds
, void *tag
)
606 dmu_buf_rele(ds
->ds_dbuf
, tag
);
610 dsl_dataset_disown(dsl_dataset_t
*ds
, void *tag
)
612 ASSERT(ds
->ds_owner
== tag
&& ds
->ds_dbuf
!= NULL
);
614 mutex_enter(&ds
->ds_lock
);
616 mutex_exit(&ds
->ds_lock
);
617 dsl_dataset_long_rele(ds
, tag
);
618 if (ds
->ds_dbuf
!= NULL
)
619 dsl_dataset_rele(ds
, tag
);
621 dsl_dataset_evict(NULL
, ds
);
625 dsl_dataset_tryown(dsl_dataset_t
*ds
, void *tag
)
627 boolean_t gotit
= FALSE
;
629 mutex_enter(&ds
->ds_lock
);
630 if (ds
->ds_owner
== NULL
&& !DS_IS_INCONSISTENT(ds
)) {
632 dsl_dataset_long_hold(ds
, tag
);
635 mutex_exit(&ds
->ds_lock
);
640 dsl_dataset_create_sync_dd(dsl_dir_t
*dd
, dsl_dataset_t
*origin
,
641 uint64_t flags
, dmu_tx_t
*tx
)
643 dsl_pool_t
*dp
= dd
->dd_pool
;
645 dsl_dataset_phys_t
*dsphys
;
647 objset_t
*mos
= dp
->dp_meta_objset
;
650 origin
= dp
->dp_origin_snap
;
652 ASSERT(origin
== NULL
|| origin
->ds_dir
->dd_pool
== dp
);
653 ASSERT(origin
== NULL
|| origin
->ds_phys
->ds_num_children
> 0);
654 ASSERT(dmu_tx_is_syncing(tx
));
655 ASSERT(dd
->dd_phys
->dd_head_dataset_obj
== 0);
657 dsobj
= dmu_object_alloc(mos
, DMU_OT_DSL_DATASET
, 0,
658 DMU_OT_DSL_DATASET
, sizeof (dsl_dataset_phys_t
), tx
);
659 VERIFY0(dmu_bonus_hold(mos
, dsobj
, FTAG
, &dbuf
));
660 dmu_buf_will_dirty(dbuf
, tx
);
661 dsphys
= dbuf
->db_data
;
662 bzero(dsphys
, sizeof (dsl_dataset_phys_t
));
663 dsphys
->ds_dir_obj
= dd
->dd_object
;
664 dsphys
->ds_flags
= flags
;
665 dsphys
->ds_fsid_guid
= unique_create();
666 (void) random_get_pseudo_bytes((void*)&dsphys
->ds_guid
,
667 sizeof (dsphys
->ds_guid
));
668 dsphys
->ds_snapnames_zapobj
=
669 zap_create_norm(mos
, U8_TEXTPREP_TOUPPER
, DMU_OT_DSL_DS_SNAP_MAP
,
671 dsphys
->ds_creation_time
= gethrestime_sec();
672 dsphys
->ds_creation_txg
= tx
->tx_txg
== TXG_INITIAL
? 1 : tx
->tx_txg
;
674 if (origin
== NULL
) {
675 dsphys
->ds_deadlist_obj
= dsl_deadlist_alloc(mos
, tx
);
677 dsl_dataset_t
*ohds
; /* head of the origin snapshot */
679 dsphys
->ds_prev_snap_obj
= origin
->ds_object
;
680 dsphys
->ds_prev_snap_txg
=
681 origin
->ds_phys
->ds_creation_txg
;
682 dsphys
->ds_referenced_bytes
=
683 origin
->ds_phys
->ds_referenced_bytes
;
684 dsphys
->ds_compressed_bytes
=
685 origin
->ds_phys
->ds_compressed_bytes
;
686 dsphys
->ds_uncompressed_bytes
=
687 origin
->ds_phys
->ds_uncompressed_bytes
;
688 dsphys
->ds_bp
= origin
->ds_phys
->ds_bp
;
689 dsphys
->ds_flags
|= origin
->ds_phys
->ds_flags
;
691 dmu_buf_will_dirty(origin
->ds_dbuf
, tx
);
692 origin
->ds_phys
->ds_num_children
++;
694 VERIFY0(dsl_dataset_hold_obj(dp
,
695 origin
->ds_dir
->dd_phys
->dd_head_dataset_obj
, FTAG
, &ohds
));
696 dsphys
->ds_deadlist_obj
= dsl_deadlist_clone(&ohds
->ds_deadlist
,
697 dsphys
->ds_prev_snap_txg
, dsphys
->ds_prev_snap_obj
, tx
);
698 dsl_dataset_rele(ohds
, FTAG
);
700 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_NEXT_CLONES
) {
701 if (origin
->ds_phys
->ds_next_clones_obj
== 0) {
702 origin
->ds_phys
->ds_next_clones_obj
=
704 DMU_OT_NEXT_CLONES
, DMU_OT_NONE
, 0, tx
);
706 VERIFY0(zap_add_int(mos
,
707 origin
->ds_phys
->ds_next_clones_obj
, dsobj
, tx
));
710 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
711 dd
->dd_phys
->dd_origin_obj
= origin
->ds_object
;
712 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
713 if (origin
->ds_dir
->dd_phys
->dd_clones
== 0) {
714 dmu_buf_will_dirty(origin
->ds_dir
->dd_dbuf
, tx
);
715 origin
->ds_dir
->dd_phys
->dd_clones
=
717 DMU_OT_DSL_CLONES
, DMU_OT_NONE
, 0, tx
);
719 VERIFY0(zap_add_int(mos
,
720 origin
->ds_dir
->dd_phys
->dd_clones
, dsobj
, tx
));
724 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_UNIQUE_ACCURATE
)
725 dsphys
->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
727 dmu_buf_rele(dbuf
, FTAG
);
729 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
730 dd
->dd_phys
->dd_head_dataset_obj
= dsobj
;
736 dsl_dataset_zero_zil(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
740 VERIFY0(dmu_objset_from_ds(ds
, &os
));
741 bzero(&os
->os_zil_header
, sizeof (os
->os_zil_header
));
742 dsl_dataset_dirty(ds
, tx
);
746 dsl_dataset_create_sync(dsl_dir_t
*pdd
, const char *lastname
,
747 dsl_dataset_t
*origin
, uint64_t flags
, cred_t
*cr
, dmu_tx_t
*tx
)
749 dsl_pool_t
*dp
= pdd
->dd_pool
;
750 uint64_t dsobj
, ddobj
;
753 ASSERT(dmu_tx_is_syncing(tx
));
754 ASSERT(lastname
[0] != '@');
756 ddobj
= dsl_dir_create_sync(dp
, pdd
, lastname
, tx
);
757 VERIFY0(dsl_dir_hold_obj(dp
, ddobj
, lastname
, FTAG
, &dd
));
759 dsobj
= dsl_dataset_create_sync_dd(dd
, origin
,
760 flags
& ~DS_CREATE_FLAG_NODIRTY
, tx
);
762 dsl_deleg_set_create_perms(dd
, tx
, cr
);
764 dsl_dir_rele(dd
, FTAG
);
767 * If we are creating a clone, make sure we zero out any stale
768 * data from the origin snapshots zil header.
770 if (origin
!= NULL
&& !(flags
& DS_CREATE_FLAG_NODIRTY
)) {
773 VERIFY0(dsl_dataset_hold_obj(dp
, dsobj
, FTAG
, &ds
));
774 dsl_dataset_zero_zil(ds
, tx
);
775 dsl_dataset_rele(ds
, FTAG
);
782 * The unique space in the head dataset can be calculated by subtracting
783 * the space used in the most recent snapshot, that is still being used
784 * in this file system, from the space currently in use. To figure out
785 * the space in the most recent snapshot still in use, we need to take
786 * the total space used in the snapshot and subtract out the space that
787 * has been freed up since the snapshot was taken.
790 dsl_dataset_recalc_head_uniq(dsl_dataset_t
*ds
)
793 uint64_t dlused
, dlcomp
, dluncomp
;
795 ASSERT(!dsl_dataset_is_snapshot(ds
));
797 if (ds
->ds_phys
->ds_prev_snap_obj
!= 0)
798 mrs_used
= ds
->ds_prev
->ds_phys
->ds_referenced_bytes
;
802 dsl_deadlist_space(&ds
->ds_deadlist
, &dlused
, &dlcomp
, &dluncomp
);
804 ASSERT3U(dlused
, <=, mrs_used
);
805 ds
->ds_phys
->ds_unique_bytes
=
806 ds
->ds_phys
->ds_referenced_bytes
- (mrs_used
- dlused
);
808 if (spa_version(ds
->ds_dir
->dd_pool
->dp_spa
) >=
809 SPA_VERSION_UNIQUE_ACCURATE
)
810 ds
->ds_phys
->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
814 dsl_dataset_remove_from_next_clones(dsl_dataset_t
*ds
, uint64_t obj
,
817 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
819 ASSERTV(uint64_t count
);
821 ASSERT(ds
->ds_phys
->ds_num_children
>= 2);
822 err
= zap_remove_int(mos
, ds
->ds_phys
->ds_next_clones_obj
, obj
, tx
);
824 * The err should not be ENOENT, but a bug in a previous version
825 * of the code could cause upgrade_clones_cb() to not set
826 * ds_next_snap_obj when it should, leading to a missing entry.
827 * If we knew that the pool was created after
828 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
829 * ENOENT. However, at least we can check that we don't have
830 * too many entries in the next_clones_obj even after failing to
835 ASSERT0(zap_count(mos
, ds
->ds_phys
->ds_next_clones_obj
,
837 ASSERT3U(count
, <=, ds
->ds_phys
->ds_num_children
- 2);
842 dsl_dataset_get_blkptr(dsl_dataset_t
*ds
)
844 return (&ds
->ds_phys
->ds_bp
);
848 dsl_dataset_set_blkptr(dsl_dataset_t
*ds
, blkptr_t
*bp
, dmu_tx_t
*tx
)
850 ASSERT(dmu_tx_is_syncing(tx
));
851 /* If it's the meta-objset, set dp_meta_rootbp */
853 tx
->tx_pool
->dp_meta_rootbp
= *bp
;
855 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
856 ds
->ds_phys
->ds_bp
= *bp
;
861 dsl_dataset_get_spa(dsl_dataset_t
*ds
)
863 return (ds
->ds_dir
->dd_pool
->dp_spa
);
867 dsl_dataset_dirty(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
871 if (ds
== NULL
) /* this is the meta-objset */
874 ASSERT(ds
->ds_objset
!= NULL
);
876 if (ds
->ds_phys
->ds_next_snap_obj
!= 0)
877 panic("dirtying snapshot!");
879 dp
= ds
->ds_dir
->dd_pool
;
881 if (txg_list_add(&dp
->dp_dirty_datasets
, ds
, tx
->tx_txg
)) {
882 /* up the hold count until we can be written out */
883 dmu_buf_add_ref(ds
->ds_dbuf
, ds
);
888 dsl_dataset_is_dirty(dsl_dataset_t
*ds
)
892 for (t
= 0; t
< TXG_SIZE
; t
++) {
893 if (txg_list_member(&ds
->ds_dir
->dd_pool
->dp_dirty_datasets
,
901 dsl_dataset_snapshot_reserve_space(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
905 if (!dmu_tx_is_syncing(tx
))
909 * If there's an fs-only reservation, any blocks that might become
910 * owned by the snapshot dataset must be accommodated by space
911 * outside of the reservation.
913 ASSERT(ds
->ds_reserved
== 0 || DS_UNIQUE_IS_ACCURATE(ds
));
914 asize
= MIN(ds
->ds_phys
->ds_unique_bytes
, ds
->ds_reserved
);
915 if (asize
> dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
))
916 return (SET_ERROR(ENOSPC
));
919 * Propagate any reserved space for this snapshot to other
920 * snapshot checks in this sync group.
923 dsl_dir_willuse_space(ds
->ds_dir
, asize
, tx
);
928 typedef struct dsl_dataset_snapshot_arg
{
929 nvlist_t
*ddsa_snaps
;
930 nvlist_t
*ddsa_props
;
931 nvlist_t
*ddsa_errors
;
932 } dsl_dataset_snapshot_arg_t
;
935 dsl_dataset_snapshot_check_impl(dsl_dataset_t
*ds
, const char *snapname
,
936 dmu_tx_t
*tx
, boolean_t recv
)
941 ds
->ds_trysnap_txg
= tx
->tx_txg
;
943 if (!dmu_tx_is_syncing(tx
))
947 * We don't allow multiple snapshots of the same txg. If there
948 * is already one, try again.
950 if (ds
->ds_phys
->ds_prev_snap_txg
>= tx
->tx_txg
)
951 return (SET_ERROR(EAGAIN
));
954 * Check for conflicting snapshot name.
956 error
= dsl_dataset_snap_lookup(ds
, snapname
, &value
);
958 return (SET_ERROR(EEXIST
));
963 * We don't allow taking snapshots of inconsistent datasets, such as
964 * those into which we are currently receiving. However, if we are
965 * creating this snapshot as part of a receive, this check will be
966 * executed atomically with respect to the completion of the receive
967 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
968 * case we ignore this, knowing it will be fixed up for us shortly in
969 * dmu_recv_end_sync().
971 if (!recv
&& DS_IS_INCONSISTENT(ds
))
972 return (SET_ERROR(EBUSY
));
974 error
= dsl_dataset_snapshot_reserve_space(ds
, tx
);
982 dsl_dataset_snapshot_check(void *arg
, dmu_tx_t
*tx
)
984 dsl_dataset_snapshot_arg_t
*ddsa
= arg
;
985 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
989 for (pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, NULL
);
990 pair
!= NULL
; pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, pair
)) {
994 char dsname
[MAXNAMELEN
];
996 name
= nvpair_name(pair
);
997 if (strlen(name
) >= MAXNAMELEN
)
998 error
= SET_ERROR(ENAMETOOLONG
);
1000 atp
= strchr(name
, '@');
1002 error
= SET_ERROR(EINVAL
);
1004 (void) strlcpy(dsname
, name
, atp
- name
+ 1);
1007 error
= dsl_dataset_hold(dp
, dsname
, FTAG
, &ds
);
1009 error
= dsl_dataset_snapshot_check_impl(ds
,
1010 atp
+ 1, tx
, B_FALSE
);
1011 dsl_dataset_rele(ds
, FTAG
);
1015 if (ddsa
->ddsa_errors
!= NULL
) {
1016 fnvlist_add_int32(ddsa
->ddsa_errors
,
1026 dsl_dataset_snapshot_sync_impl(dsl_dataset_t
*ds
, const char *snapname
,
1029 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1031 dsl_dataset_phys_t
*dsphys
;
1032 uint64_t dsobj
, crtxg
;
1033 objset_t
*mos
= dp
->dp_meta_objset
;
1034 ASSERTV(static zil_header_t zero_zil
);
1035 ASSERTV(objset_t
*os
);
1037 ASSERT(RRW_WRITE_HELD(&dp
->dp_config_rwlock
));
1040 * If we are on an old pool, the zil must not be active, in which
1041 * case it will be zeroed. Usually zil_suspend() accomplishes this.
1043 ASSERT(spa_version(dmu_tx_pool(tx
)->dp_spa
) >= SPA_VERSION_FAST_SNAP
||
1044 dmu_objset_from_ds(ds
, &os
) != 0 ||
1045 bcmp(&os
->os_phys
->os_zil_header
, &zero_zil
,
1046 sizeof (zero_zil
)) == 0);
1050 * The origin's ds_creation_txg has to be < TXG_INITIAL
1052 if (strcmp(snapname
, ORIGIN_DIR_NAME
) == 0)
1057 dsobj
= dmu_object_alloc(mos
, DMU_OT_DSL_DATASET
, 0,
1058 DMU_OT_DSL_DATASET
, sizeof (dsl_dataset_phys_t
), tx
);
1059 VERIFY0(dmu_bonus_hold(mos
, dsobj
, FTAG
, &dbuf
));
1060 dmu_buf_will_dirty(dbuf
, tx
);
1061 dsphys
= dbuf
->db_data
;
1062 bzero(dsphys
, sizeof (dsl_dataset_phys_t
));
1063 dsphys
->ds_dir_obj
= ds
->ds_dir
->dd_object
;
1064 dsphys
->ds_fsid_guid
= unique_create();
1065 (void) random_get_pseudo_bytes((void*)&dsphys
->ds_guid
,
1066 sizeof (dsphys
->ds_guid
));
1067 dsphys
->ds_prev_snap_obj
= ds
->ds_phys
->ds_prev_snap_obj
;
1068 dsphys
->ds_prev_snap_txg
= ds
->ds_phys
->ds_prev_snap_txg
;
1069 dsphys
->ds_next_snap_obj
= ds
->ds_object
;
1070 dsphys
->ds_num_children
= 1;
1071 dsphys
->ds_creation_time
= gethrestime_sec();
1072 dsphys
->ds_creation_txg
= crtxg
;
1073 dsphys
->ds_deadlist_obj
= ds
->ds_phys
->ds_deadlist_obj
;
1074 dsphys
->ds_referenced_bytes
= ds
->ds_phys
->ds_referenced_bytes
;
1075 dsphys
->ds_compressed_bytes
= ds
->ds_phys
->ds_compressed_bytes
;
1076 dsphys
->ds_uncompressed_bytes
= ds
->ds_phys
->ds_uncompressed_bytes
;
1077 dsphys
->ds_flags
= ds
->ds_phys
->ds_flags
;
1078 dsphys
->ds_bp
= ds
->ds_phys
->ds_bp
;
1079 dmu_buf_rele(dbuf
, FTAG
);
1081 ASSERT3U(ds
->ds_prev
!= 0, ==, ds
->ds_phys
->ds_prev_snap_obj
!= 0);
1083 uint64_t next_clones_obj
=
1084 ds
->ds_prev
->ds_phys
->ds_next_clones_obj
;
1085 ASSERT(ds
->ds_prev
->ds_phys
->ds_next_snap_obj
==
1087 ds
->ds_prev
->ds_phys
->ds_num_children
> 1);
1088 if (ds
->ds_prev
->ds_phys
->ds_next_snap_obj
== ds
->ds_object
) {
1089 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
1090 ASSERT3U(ds
->ds_phys
->ds_prev_snap_txg
, ==,
1091 ds
->ds_prev
->ds_phys
->ds_creation_txg
);
1092 ds
->ds_prev
->ds_phys
->ds_next_snap_obj
= dsobj
;
1093 } else if (next_clones_obj
!= 0) {
1094 dsl_dataset_remove_from_next_clones(ds
->ds_prev
,
1095 dsphys
->ds_next_snap_obj
, tx
);
1096 VERIFY0(zap_add_int(mos
,
1097 next_clones_obj
, dsobj
, tx
));
1102 * If we have a reference-reservation on this dataset, we will
1103 * need to increase the amount of refreservation being charged
1104 * since our unique space is going to zero.
1106 if (ds
->ds_reserved
) {
1108 ASSERT(DS_UNIQUE_IS_ACCURATE(ds
));
1109 delta
= MIN(ds
->ds_phys
->ds_unique_bytes
, ds
->ds_reserved
);
1110 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_REFRSRV
,
1114 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1115 ds
->ds_phys
->ds_deadlist_obj
= dsl_deadlist_clone(&ds
->ds_deadlist
,
1116 UINT64_MAX
, ds
->ds_phys
->ds_prev_snap_obj
, tx
);
1117 dsl_deadlist_close(&ds
->ds_deadlist
);
1118 dsl_deadlist_open(&ds
->ds_deadlist
, mos
, ds
->ds_phys
->ds_deadlist_obj
);
1119 dsl_deadlist_add_key(&ds
->ds_deadlist
,
1120 ds
->ds_phys
->ds_prev_snap_txg
, tx
);
1122 ASSERT3U(ds
->ds_phys
->ds_prev_snap_txg
, <, tx
->tx_txg
);
1123 ds
->ds_phys
->ds_prev_snap_obj
= dsobj
;
1124 ds
->ds_phys
->ds_prev_snap_txg
= crtxg
;
1125 ds
->ds_phys
->ds_unique_bytes
= 0;
1126 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_UNIQUE_ACCURATE
)
1127 ds
->ds_phys
->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
1129 VERIFY0(zap_add(mos
, ds
->ds_phys
->ds_snapnames_zapobj
,
1130 snapname
, 8, 1, &dsobj
, tx
));
1133 dsl_dataset_rele(ds
->ds_prev
, ds
);
1134 VERIFY0(dsl_dataset_hold_obj(dp
,
1135 ds
->ds_phys
->ds_prev_snap_obj
, ds
, &ds
->ds_prev
));
1137 dsl_scan_ds_snapshotted(ds
, tx
);
1139 dsl_dir_snap_cmtime_update(ds
->ds_dir
);
1141 spa_history_log_internal_ds(ds
->ds_prev
, "snapshot", tx
, "");
1145 dsl_dataset_snapshot_sync(void *arg
, dmu_tx_t
*tx
)
1147 dsl_dataset_snapshot_arg_t
*ddsa
= arg
;
1148 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1151 for (pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, NULL
);
1152 pair
!= NULL
; pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, pair
)) {
1155 char dsname
[MAXNAMELEN
];
1157 name
= nvpair_name(pair
);
1158 atp
= strchr(name
, '@');
1159 (void) strlcpy(dsname
, name
, atp
- name
+ 1);
1160 VERIFY0(dsl_dataset_hold(dp
, dsname
, FTAG
, &ds
));
1162 dsl_dataset_snapshot_sync_impl(ds
, atp
+ 1, tx
);
1163 if (ddsa
->ddsa_props
!= NULL
) {
1164 dsl_props_set_sync_impl(ds
->ds_prev
,
1165 ZPROP_SRC_LOCAL
, ddsa
->ddsa_props
, tx
);
1167 dsl_dataset_rele(ds
, FTAG
);
1172 * The snapshots must all be in the same pool.
1173 * All-or-nothing: if there are any failures, nothing will be modified.
1176 dsl_dataset_snapshot(nvlist_t
*snaps
, nvlist_t
*props
, nvlist_t
*errors
)
1178 dsl_dataset_snapshot_arg_t ddsa
;
1180 boolean_t needsuspend
;
1184 nvlist_t
*suspended
= NULL
;
1186 pair
= nvlist_next_nvpair(snaps
, NULL
);
1189 firstname
= nvpair_name(pair
);
1191 error
= spa_open(firstname
, &spa
, FTAG
);
1194 needsuspend
= (spa_version(spa
) < SPA_VERSION_FAST_SNAP
);
1195 spa_close(spa
, FTAG
);
1198 suspended
= fnvlist_alloc();
1199 for (pair
= nvlist_next_nvpair(snaps
, NULL
); pair
!= NULL
;
1200 pair
= nvlist_next_nvpair(snaps
, pair
)) {
1201 char fsname
[MAXNAMELEN
];
1202 char *snapname
= nvpair_name(pair
);
1206 atp
= strchr(snapname
, '@');
1208 error
= SET_ERROR(EINVAL
);
1211 (void) strlcpy(fsname
, snapname
, atp
- snapname
+ 1);
1213 error
= zil_suspend(fsname
, &cookie
);
1216 fnvlist_add_uint64(suspended
, fsname
,
1221 ddsa
.ddsa_snaps
= snaps
;
1222 ddsa
.ddsa_props
= props
;
1223 ddsa
.ddsa_errors
= errors
;
1226 error
= dsl_sync_task(firstname
, dsl_dataset_snapshot_check
,
1227 dsl_dataset_snapshot_sync
, &ddsa
,
1228 fnvlist_num_pairs(snaps
) * 3);
1231 if (suspended
!= NULL
) {
1232 for (pair
= nvlist_next_nvpair(suspended
, NULL
); pair
!= NULL
;
1233 pair
= nvlist_next_nvpair(suspended
, pair
)) {
1234 zil_resume((void *)(uintptr_t)
1235 fnvpair_value_uint64(pair
));
1237 fnvlist_free(suspended
);
1242 for (pair
= nvlist_next_nvpair(snaps
, NULL
); pair
!= NULL
;
1243 pair
= nvlist_next_nvpair(snaps
, pair
)) {
1244 char *snapname
= nvpair_name(pair
);
1245 zvol_create_minors(snapname
);
1253 typedef struct dsl_dataset_snapshot_tmp_arg
{
1254 const char *ddsta_fsname
;
1255 const char *ddsta_snapname
;
1256 minor_t ddsta_cleanup_minor
;
1257 const char *ddsta_htag
;
1258 } dsl_dataset_snapshot_tmp_arg_t
;
1261 dsl_dataset_snapshot_tmp_check(void *arg
, dmu_tx_t
*tx
)
1263 dsl_dataset_snapshot_tmp_arg_t
*ddsta
= arg
;
1264 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1268 error
= dsl_dataset_hold(dp
, ddsta
->ddsta_fsname
, FTAG
, &ds
);
1272 error
= dsl_dataset_snapshot_check_impl(ds
, ddsta
->ddsta_snapname
,
1275 dsl_dataset_rele(ds
, FTAG
);
1279 if (spa_version(dp
->dp_spa
) < SPA_VERSION_USERREFS
) {
1280 dsl_dataset_rele(ds
, FTAG
);
1281 return (SET_ERROR(ENOTSUP
));
1283 error
= dsl_dataset_user_hold_check_one(NULL
, ddsta
->ddsta_htag
,
1286 dsl_dataset_rele(ds
, FTAG
);
1290 dsl_dataset_rele(ds
, FTAG
);
1295 dsl_dataset_snapshot_tmp_sync(void *arg
, dmu_tx_t
*tx
)
1297 dsl_dataset_snapshot_tmp_arg_t
*ddsta
= arg
;
1298 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1301 VERIFY0(dsl_dataset_hold(dp
, ddsta
->ddsta_fsname
, FTAG
, &ds
));
1303 dsl_dataset_snapshot_sync_impl(ds
, ddsta
->ddsta_snapname
, tx
);
1304 dsl_dataset_user_hold_sync_one(ds
->ds_prev
, ddsta
->ddsta_htag
,
1305 ddsta
->ddsta_cleanup_minor
, gethrestime_sec(), tx
);
1306 dsl_destroy_snapshot_sync_impl(ds
->ds_prev
, B_TRUE
, tx
);
1308 dsl_dataset_rele(ds
, FTAG
);
1312 dsl_dataset_snapshot_tmp(const char *fsname
, const char *snapname
,
1313 minor_t cleanup_minor
, const char *htag
)
1315 dsl_dataset_snapshot_tmp_arg_t ddsta
;
1318 boolean_t needsuspend
;
1321 ddsta
.ddsta_fsname
= fsname
;
1322 ddsta
.ddsta_snapname
= snapname
;
1323 ddsta
.ddsta_cleanup_minor
= cleanup_minor
;
1324 ddsta
.ddsta_htag
= htag
;
1326 error
= spa_open(fsname
, &spa
, FTAG
);
1329 needsuspend
= (spa_version(spa
) < SPA_VERSION_FAST_SNAP
);
1330 spa_close(spa
, FTAG
);
1333 error
= zil_suspend(fsname
, &cookie
);
1338 error
= dsl_sync_task(fsname
, dsl_dataset_snapshot_tmp_check
,
1339 dsl_dataset_snapshot_tmp_sync
, &ddsta
, 3);
1348 dsl_dataset_sync(dsl_dataset_t
*ds
, zio_t
*zio
, dmu_tx_t
*tx
)
1350 ASSERT(dmu_tx_is_syncing(tx
));
1351 ASSERT(ds
->ds_objset
!= NULL
);
1352 ASSERT(ds
->ds_phys
->ds_next_snap_obj
== 0);
1355 * in case we had to change ds_fsid_guid when we opened it,
1358 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1359 ds
->ds_phys
->ds_fsid_guid
= ds
->ds_fsid_guid
;
1361 dmu_objset_sync(ds
->ds_objset
, zio
, tx
);
1365 get_clones_stat(dsl_dataset_t
*ds
, nvlist_t
*nv
)
1368 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
1371 nvlist_t
*propval
= fnvlist_alloc();
1372 nvlist_t
*val
= fnvlist_alloc();
1374 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
1377 * There may be missing entries in ds_next_clones_obj
1378 * due to a bug in a previous version of the code.
1379 * Only trust it if it has the right number of entries.
1381 if (ds
->ds_phys
->ds_next_clones_obj
!= 0) {
1382 VERIFY0(zap_count(mos
, ds
->ds_phys
->ds_next_clones_obj
,
1385 if (count
!= ds
->ds_phys
->ds_num_children
- 1)
1387 for (zap_cursor_init(&zc
, mos
, ds
->ds_phys
->ds_next_clones_obj
);
1388 zap_cursor_retrieve(&zc
, &za
) == 0;
1389 zap_cursor_advance(&zc
)) {
1390 dsl_dataset_t
*clone
;
1391 char buf
[ZFS_MAXNAMELEN
];
1392 VERIFY0(dsl_dataset_hold_obj(ds
->ds_dir
->dd_pool
,
1393 za
.za_first_integer
, FTAG
, &clone
));
1394 dsl_dir_name(clone
->ds_dir
, buf
);
1395 fnvlist_add_boolean(val
, buf
);
1396 dsl_dataset_rele(clone
, FTAG
);
1398 zap_cursor_fini(&zc
);
1399 fnvlist_add_nvlist(propval
, ZPROP_VALUE
, val
);
1400 fnvlist_add_nvlist(nv
, zfs_prop_to_name(ZFS_PROP_CLONES
), propval
);
1403 nvlist_free(propval
);
1407 dsl_dataset_stats(dsl_dataset_t
*ds
, nvlist_t
*nv
)
1409 uint64_t refd
, avail
, uobjs
, aobjs
, ratio
;
1410 ASSERTV(dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
);
1412 ASSERT(dsl_pool_config_held(dp
));
1414 ratio
= ds
->ds_phys
->ds_compressed_bytes
== 0 ? 100 :
1415 (ds
->ds_phys
->ds_uncompressed_bytes
* 100 /
1416 ds
->ds_phys
->ds_compressed_bytes
);
1418 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFRATIO
, ratio
);
1419 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_LOGICALREFERENCED
,
1420 ds
->ds_phys
->ds_uncompressed_bytes
);
1422 if (dsl_dataset_is_snapshot(ds
)) {
1423 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_COMPRESSRATIO
, ratio
);
1424 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USED
,
1425 ds
->ds_phys
->ds_unique_bytes
);
1426 get_clones_stat(ds
, nv
);
1428 dsl_dir_stats(ds
->ds_dir
, nv
);
1431 dsl_dataset_space(ds
, &refd
, &avail
, &uobjs
, &aobjs
);
1432 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_AVAILABLE
, avail
);
1433 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFERENCED
, refd
);
1435 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_CREATION
,
1436 ds
->ds_phys
->ds_creation_time
);
1437 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_CREATETXG
,
1438 ds
->ds_phys
->ds_creation_txg
);
1439 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFQUOTA
,
1441 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFRESERVATION
,
1443 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_GUID
,
1444 ds
->ds_phys
->ds_guid
);
1445 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_UNIQUE
,
1446 ds
->ds_phys
->ds_unique_bytes
);
1447 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_OBJSETID
,
1449 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USERREFS
,
1451 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_DEFER_DESTROY
,
1452 DS_IS_DEFER_DESTROY(ds
) ? 1 : 0);
1454 if (ds
->ds_phys
->ds_prev_snap_obj
!= 0) {
1455 uint64_t written
, comp
, uncomp
;
1456 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1457 dsl_dataset_t
*prev
;
1460 err
= dsl_dataset_hold_obj(dp
,
1461 ds
->ds_phys
->ds_prev_snap_obj
, FTAG
, &prev
);
1463 err
= dsl_dataset_space_written(prev
, ds
, &written
,
1465 dsl_dataset_rele(prev
, FTAG
);
1467 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_WRITTEN
,
1476 dsl_dataset_fast_stat(dsl_dataset_t
*ds
, dmu_objset_stats_t
*stat
)
1478 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1479 ASSERT(dsl_pool_config_held(dp
));
1481 stat
->dds_creation_txg
= ds
->ds_phys
->ds_creation_txg
;
1482 stat
->dds_inconsistent
= ds
->ds_phys
->ds_flags
& DS_FLAG_INCONSISTENT
;
1483 stat
->dds_guid
= ds
->ds_phys
->ds_guid
;
1484 stat
->dds_origin
[0] = '\0';
1485 if (dsl_dataset_is_snapshot(ds
)) {
1486 stat
->dds_is_snapshot
= B_TRUE
;
1487 stat
->dds_num_clones
= ds
->ds_phys
->ds_num_children
- 1;
1489 stat
->dds_is_snapshot
= B_FALSE
;
1490 stat
->dds_num_clones
= 0;
1492 if (dsl_dir_is_clone(ds
->ds_dir
)) {
1495 VERIFY0(dsl_dataset_hold_obj(dp
,
1496 ds
->ds_dir
->dd_phys
->dd_origin_obj
, FTAG
, &ods
));
1497 dsl_dataset_name(ods
, stat
->dds_origin
);
1498 dsl_dataset_rele(ods
, FTAG
);
1504 dsl_dataset_fsid_guid(dsl_dataset_t
*ds
)
1506 return (ds
->ds_fsid_guid
);
1510 dsl_dataset_space(dsl_dataset_t
*ds
,
1511 uint64_t *refdbytesp
, uint64_t *availbytesp
,
1512 uint64_t *usedobjsp
, uint64_t *availobjsp
)
1514 *refdbytesp
= ds
->ds_phys
->ds_referenced_bytes
;
1515 *availbytesp
= dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
);
1516 if (ds
->ds_reserved
> ds
->ds_phys
->ds_unique_bytes
)
1517 *availbytesp
+= ds
->ds_reserved
- ds
->ds_phys
->ds_unique_bytes
;
1518 if (ds
->ds_quota
!= 0) {
1520 * Adjust available bytes according to refquota
1522 if (*refdbytesp
< ds
->ds_quota
)
1523 *availbytesp
= MIN(*availbytesp
,
1524 ds
->ds_quota
- *refdbytesp
);
1528 *usedobjsp
= BP_GET_FILL(&ds
->ds_phys
->ds_bp
);
1529 *availobjsp
= DN_MAX_OBJECT
- *usedobjsp
;
1533 dsl_dataset_modified_since_snap(dsl_dataset_t
*ds
, dsl_dataset_t
*snap
)
1535 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
1538 if (ds
->ds_phys
->ds_bp
.blk_birth
>
1539 snap
->ds_phys
->ds_creation_txg
) {
1540 objset_t
*os
, *os_snap
;
1542 * It may be that only the ZIL differs, because it was
1543 * reset in the head. Don't count that as being
1546 if (dmu_objset_from_ds(ds
, &os
) != 0)
1548 if (dmu_objset_from_ds(snap
, &os_snap
) != 0)
1550 return (bcmp(&os
->os_phys
->os_meta_dnode
,
1551 &os_snap
->os_phys
->os_meta_dnode
,
1552 sizeof (os
->os_phys
->os_meta_dnode
)) != 0);
1557 typedef struct dsl_dataset_rename_snapshot_arg
{
1558 const char *ddrsa_fsname
;
1559 const char *ddrsa_oldsnapname
;
1560 const char *ddrsa_newsnapname
;
1561 boolean_t ddrsa_recursive
;
1563 } dsl_dataset_rename_snapshot_arg_t
;
1567 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t
*dp
,
1568 dsl_dataset_t
*hds
, void *arg
)
1570 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
1574 error
= dsl_dataset_snap_lookup(hds
, ddrsa
->ddrsa_oldsnapname
, &val
);
1576 /* ignore nonexistent snapshots */
1577 return (error
== ENOENT
? 0 : error
);
1580 /* new name should not exist */
1581 error
= dsl_dataset_snap_lookup(hds
, ddrsa
->ddrsa_newsnapname
, &val
);
1583 error
= SET_ERROR(EEXIST
);
1584 else if (error
== ENOENT
)
1587 /* dataset name + 1 for the "@" + the new snapshot name must fit */
1588 if (dsl_dir_namelen(hds
->ds_dir
) + 1 +
1589 strlen(ddrsa
->ddrsa_newsnapname
) >= MAXNAMELEN
)
1590 error
= SET_ERROR(ENAMETOOLONG
);
1596 dsl_dataset_rename_snapshot_check(void *arg
, dmu_tx_t
*tx
)
1598 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
1599 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1603 error
= dsl_dataset_hold(dp
, ddrsa
->ddrsa_fsname
, FTAG
, &hds
);
1607 if (ddrsa
->ddrsa_recursive
) {
1608 error
= dmu_objset_find_dp(dp
, hds
->ds_dir
->dd_object
,
1609 dsl_dataset_rename_snapshot_check_impl
, ddrsa
,
1612 error
= dsl_dataset_rename_snapshot_check_impl(dp
, hds
, ddrsa
);
1614 dsl_dataset_rele(hds
, FTAG
);
1619 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t
*dp
,
1620 dsl_dataset_t
*hds
, void *arg
)
1623 char *oldname
, *newname
;
1625 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
1628 dmu_tx_t
*tx
= ddrsa
->ddrsa_tx
;
1631 error
= dsl_dataset_snap_lookup(hds
, ddrsa
->ddrsa_oldsnapname
, &val
);
1632 ASSERT(error
== 0 || error
== ENOENT
);
1633 if (error
== ENOENT
) {
1634 /* ignore nonexistent snapshots */
1638 VERIFY0(dsl_dataset_hold_obj(dp
, val
, FTAG
, &ds
));
1640 /* log before we change the name */
1641 spa_history_log_internal_ds(ds
, "rename", tx
,
1642 "-> @%s", ddrsa
->ddrsa_newsnapname
);
1644 VERIFY0(dsl_dataset_snap_remove(hds
, ddrsa
->ddrsa_oldsnapname
, tx
));
1645 mutex_enter(&ds
->ds_lock
);
1646 (void) strcpy(ds
->ds_snapname
, ddrsa
->ddrsa_newsnapname
);
1647 mutex_exit(&ds
->ds_lock
);
1648 VERIFY0(zap_add(dp
->dp_meta_objset
, hds
->ds_phys
->ds_snapnames_zapobj
,
1649 ds
->ds_snapname
, 8, 1, &ds
->ds_object
, tx
));
1652 oldname
= kmem_alloc(MAXPATHLEN
, KM_PUSHPAGE
);
1653 newname
= kmem_alloc(MAXPATHLEN
, KM_PUSHPAGE
);
1654 snprintf(oldname
, MAXPATHLEN
, "%s@%s", ddrsa
->ddrsa_fsname
,
1655 ddrsa
->ddrsa_oldsnapname
);
1656 snprintf(newname
, MAXPATHLEN
, "%s@%s", ddrsa
->ddrsa_fsname
,
1657 ddrsa
->ddrsa_newsnapname
);
1658 zvol_rename_minors(oldname
, newname
);
1659 kmem_free(newname
, MAXPATHLEN
);
1660 kmem_free(oldname
, MAXPATHLEN
);
1663 dsl_dataset_rele(ds
, FTAG
);
1668 dsl_dataset_rename_snapshot_sync(void *arg
, dmu_tx_t
*tx
)
1670 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
1671 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1674 VERIFY0(dsl_dataset_hold(dp
, ddrsa
->ddrsa_fsname
, FTAG
, &hds
));
1675 ddrsa
->ddrsa_tx
= tx
;
1676 if (ddrsa
->ddrsa_recursive
) {
1677 VERIFY0(dmu_objset_find_dp(dp
, hds
->ds_dir
->dd_object
,
1678 dsl_dataset_rename_snapshot_sync_impl
, ddrsa
,
1681 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp
, hds
, ddrsa
));
1683 dsl_dataset_rele(hds
, FTAG
);
1687 dsl_dataset_rename_snapshot(const char *fsname
,
1688 const char *oldsnapname
, const char *newsnapname
, boolean_t recursive
)
1690 dsl_dataset_rename_snapshot_arg_t ddrsa
;
1692 ddrsa
.ddrsa_fsname
= fsname
;
1693 ddrsa
.ddrsa_oldsnapname
= oldsnapname
;
1694 ddrsa
.ddrsa_newsnapname
= newsnapname
;
1695 ddrsa
.ddrsa_recursive
= recursive
;
1697 return (dsl_sync_task(fsname
, dsl_dataset_rename_snapshot_check
,
1698 dsl_dataset_rename_snapshot_sync
, &ddrsa
, 1));
1702 * If we're doing an ownership handoff, we need to make sure that there is
1703 * only one long hold on the dataset. We're not allowed to change anything here
1704 * so we don't permanently release the long hold or regular hold here. We want
1705 * to do this only when syncing to avoid the dataset unexpectedly going away
1706 * when we release the long hold.
1709 dsl_dataset_handoff_check(dsl_dataset_t
*ds
, void *owner
, dmu_tx_t
*tx
)
1713 if (!dmu_tx_is_syncing(tx
))
1716 if (owner
!= NULL
) {
1717 VERIFY3P(ds
->ds_owner
, ==, owner
);
1718 dsl_dataset_long_rele(ds
, owner
);
1721 held
= dsl_dataset_long_held(ds
);
1724 dsl_dataset_long_hold(ds
, owner
);
1727 return (SET_ERROR(EBUSY
));
1732 typedef struct dsl_dataset_rollback_arg
{
1733 const char *ddra_fsname
;
1735 nvlist_t
*ddra_result
;
1736 } dsl_dataset_rollback_arg_t
;
1739 dsl_dataset_rollback_check(void *arg
, dmu_tx_t
*tx
)
1741 dsl_dataset_rollback_arg_t
*ddra
= arg
;
1742 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1744 int64_t unused_refres_delta
;
1747 nvlist_t
*proprequest
, *bookmarks
;
1749 error
= dsl_dataset_hold(dp
, ddra
->ddra_fsname
, FTAG
, &ds
);
1753 /* must not be a snapshot */
1754 if (dsl_dataset_is_snapshot(ds
)) {
1755 dsl_dataset_rele(ds
, FTAG
);
1756 return (SET_ERROR(EINVAL
));
1759 /* must have a most recent snapshot */
1760 if (ds
->ds_phys
->ds_prev_snap_txg
< TXG_INITIAL
) {
1761 dsl_dataset_rele(ds
, FTAG
);
1762 return (SET_ERROR(EINVAL
));
1765 /* must not have any bookmarks after the most recent snapshot */
1766 proprequest
= fnvlist_alloc();
1767 fnvlist_add_boolean(proprequest
, zfs_prop_to_name(ZFS_PROP_CREATETXG
));
1768 bookmarks
= fnvlist_alloc();
1769 error
= dsl_get_bookmarks_impl(ds
, proprequest
, bookmarks
);
1770 fnvlist_free(proprequest
);
1773 for (pair
= nvlist_next_nvpair(bookmarks
, NULL
);
1774 pair
!= NULL
; pair
= nvlist_next_nvpair(bookmarks
, pair
)) {
1776 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair
),
1777 zfs_prop_to_name(ZFS_PROP_CREATETXG
));
1778 uint64_t createtxg
= fnvlist_lookup_uint64(valuenv
, "value");
1779 if (createtxg
> ds
->ds_phys
->ds_prev_snap_txg
) {
1780 fnvlist_free(bookmarks
);
1781 dsl_dataset_rele(ds
, FTAG
);
1782 return (SET_ERROR(EEXIST
));
1785 fnvlist_free(bookmarks
);
1787 error
= dsl_dataset_handoff_check(ds
, ddra
->ddra_owner
, tx
);
1789 dsl_dataset_rele(ds
, FTAG
);
1794 * Check if the snap we are rolling back to uses more than
1797 if (ds
->ds_quota
!= 0 &&
1798 ds
->ds_prev
->ds_phys
->ds_referenced_bytes
> ds
->ds_quota
) {
1799 dsl_dataset_rele(ds
, FTAG
);
1800 return (SET_ERROR(EDQUOT
));
1804 * When we do the clone swap, we will temporarily use more space
1805 * due to the refreservation (the head will no longer have any
1806 * unique space, so the entire amount of the refreservation will need
1807 * to be free). We will immediately destroy the clone, freeing
1808 * this space, but the freeing happens over many txg's.
1810 unused_refres_delta
= (int64_t)MIN(ds
->ds_reserved
,
1811 ds
->ds_phys
->ds_unique_bytes
);
1813 if (unused_refres_delta
> 0 &&
1814 unused_refres_delta
>
1815 dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
)) {
1816 dsl_dataset_rele(ds
, FTAG
);
1817 return (SET_ERROR(ENOSPC
));
1820 dsl_dataset_rele(ds
, FTAG
);
1825 dsl_dataset_rollback_sync(void *arg
, dmu_tx_t
*tx
)
1827 dsl_dataset_rollback_arg_t
*ddra
= arg
;
1828 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1829 dsl_dataset_t
*ds
, *clone
;
1831 char namebuf
[ZFS_MAXNAMELEN
];
1833 VERIFY0(dsl_dataset_hold(dp
, ddra
->ddra_fsname
, FTAG
, &ds
));
1835 dsl_dataset_name(ds
->ds_prev
, namebuf
);
1836 fnvlist_add_string(ddra
->ddra_result
, "target", namebuf
);
1838 cloneobj
= dsl_dataset_create_sync(ds
->ds_dir
, "%rollback",
1839 ds
->ds_prev
, DS_CREATE_FLAG_NODIRTY
, kcred
, tx
);
1841 VERIFY0(dsl_dataset_hold_obj(dp
, cloneobj
, FTAG
, &clone
));
1843 dsl_dataset_clone_swap_sync_impl(clone
, ds
, tx
);
1844 dsl_dataset_zero_zil(ds
, tx
);
1846 dsl_destroy_head_sync_impl(clone
, tx
);
1848 dsl_dataset_rele(clone
, FTAG
);
1849 dsl_dataset_rele(ds
, FTAG
);
1853 * Rolls back the given filesystem or volume to the most recent snapshot.
1854 * The name of the most recent snapshot will be returned under key "target"
1855 * in the result nvlist.
1858 * - The existing dataset MUST be owned by the specified owner at entry
1859 * - Upon return, dataset will still be held by the same owner, whether we
1862 * This mode is required any time the existing filesystem is mounted. See
1863 * notes above zfs_suspend_fs() for further details.
1866 dsl_dataset_rollback(const char *fsname
, void *owner
, nvlist_t
*result
)
1868 dsl_dataset_rollback_arg_t ddra
;
1870 ddra
.ddra_fsname
= fsname
;
1871 ddra
.ddra_owner
= owner
;
1872 ddra
.ddra_result
= result
;
1874 return (dsl_sync_task(fsname
, dsl_dataset_rollback_check
,
1875 dsl_dataset_rollback_sync
, &ddra
, 1));
1878 struct promotenode
{
1883 typedef struct dsl_dataset_promote_arg
{
1884 const char *ddpa_clonename
;
1885 dsl_dataset_t
*ddpa_clone
;
1886 list_t shared_snaps
, origin_snaps
, clone_snaps
;
1887 dsl_dataset_t
*origin_origin
; /* origin of the origin */
1888 uint64_t used
, comp
, uncomp
, unique
, cloneusedsnap
, originusedsnap
;
1890 } dsl_dataset_promote_arg_t
;
1892 static int snaplist_space(list_t
*l
, uint64_t mintxg
, uint64_t *spacep
);
1893 static int promote_hold(dsl_dataset_promote_arg_t
*ddpa
, dsl_pool_t
*dp
,
1895 static void promote_rele(dsl_dataset_promote_arg_t
*ddpa
, void *tag
);
1898 dsl_dataset_promote_check(void *arg
, dmu_tx_t
*tx
)
1900 dsl_dataset_promote_arg_t
*ddpa
= arg
;
1901 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1903 struct promotenode
*snap
;
1904 dsl_dataset_t
*origin_ds
;
1908 err
= promote_hold(ddpa
, dp
, FTAG
);
1912 hds
= ddpa
->ddpa_clone
;
1914 if (hds
->ds_phys
->ds_flags
& DS_FLAG_NOPROMOTE
) {
1915 promote_rele(ddpa
, FTAG
);
1916 return (SET_ERROR(EXDEV
));
1920 * Compute and check the amount of space to transfer. Since this is
1921 * so expensive, don't do the preliminary check.
1923 if (!dmu_tx_is_syncing(tx
)) {
1924 promote_rele(ddpa
, FTAG
);
1928 snap
= list_head(&ddpa
->shared_snaps
);
1929 origin_ds
= snap
->ds
;
1931 /* compute origin's new unique space */
1932 snap
= list_tail(&ddpa
->clone_snaps
);
1933 ASSERT3U(snap
->ds
->ds_phys
->ds_prev_snap_obj
, ==, origin_ds
->ds_object
);
1934 dsl_deadlist_space_range(&snap
->ds
->ds_deadlist
,
1935 origin_ds
->ds_phys
->ds_prev_snap_txg
, UINT64_MAX
,
1936 &ddpa
->unique
, &unused
, &unused
);
1939 * Walk the snapshots that we are moving
1941 * Compute space to transfer. Consider the incremental changes
1942 * to used by each snapshot:
1943 * (my used) = (prev's used) + (blocks born) - (blocks killed)
1944 * So each snapshot gave birth to:
1945 * (blocks born) = (my used) - (prev's used) + (blocks killed)
1946 * So a sequence would look like:
1947 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
1948 * Which simplifies to:
1949 * uN + kN + kN-1 + ... + k1 + k0
1950 * Note however, if we stop before we reach the ORIGIN we get:
1951 * uN + kN + kN-1 + ... + kM - uM-1
1953 ddpa
->used
= origin_ds
->ds_phys
->ds_referenced_bytes
;
1954 ddpa
->comp
= origin_ds
->ds_phys
->ds_compressed_bytes
;
1955 ddpa
->uncomp
= origin_ds
->ds_phys
->ds_uncompressed_bytes
;
1956 for (snap
= list_head(&ddpa
->shared_snaps
); snap
;
1957 snap
= list_next(&ddpa
->shared_snaps
, snap
)) {
1958 uint64_t val
, dlused
, dlcomp
, dluncomp
;
1959 dsl_dataset_t
*ds
= snap
->ds
;
1962 * If there are long holds, we won't be able to evict
1965 if (dsl_dataset_long_held(ds
)) {
1966 err
= SET_ERROR(EBUSY
);
1970 /* Check that the snapshot name does not conflict */
1971 VERIFY0(dsl_dataset_get_snapname(ds
));
1972 err
= dsl_dataset_snap_lookup(hds
, ds
->ds_snapname
, &val
);
1974 (void) strcpy(ddpa
->err_ds
, snap
->ds
->ds_snapname
);
1975 err
= SET_ERROR(EEXIST
);
1981 /* The very first snapshot does not have a deadlist */
1982 if (ds
->ds_phys
->ds_prev_snap_obj
== 0)
1985 dsl_deadlist_space(&ds
->ds_deadlist
,
1986 &dlused
, &dlcomp
, &dluncomp
);
1987 ddpa
->used
+= dlused
;
1988 ddpa
->comp
+= dlcomp
;
1989 ddpa
->uncomp
+= dluncomp
;
1993 * If we are a clone of a clone then we never reached ORIGIN,
1994 * so we need to subtract out the clone origin's used space.
1996 if (ddpa
->origin_origin
) {
1997 ddpa
->used
-= ddpa
->origin_origin
->ds_phys
->ds_referenced_bytes
;
1998 ddpa
->comp
-= ddpa
->origin_origin
->ds_phys
->ds_compressed_bytes
;
2000 ddpa
->origin_origin
->ds_phys
->ds_uncompressed_bytes
;
2003 /* Check that there is enough space here */
2004 err
= dsl_dir_transfer_possible(origin_ds
->ds_dir
, hds
->ds_dir
,
2010 * Compute the amounts of space that will be used by snapshots
2011 * after the promotion (for both origin and clone). For each,
2012 * it is the amount of space that will be on all of their
2013 * deadlists (that was not born before their new origin).
2015 if (hds
->ds_dir
->dd_phys
->dd_flags
& DD_FLAG_USED_BREAKDOWN
) {
2019 * Note, typically this will not be a clone of a clone,
2020 * so dd_origin_txg will be < TXG_INITIAL, so
2021 * these snaplist_space() -> dsl_deadlist_space_range()
2022 * calls will be fast because they do not have to
2023 * iterate over all bps.
2025 snap
= list_head(&ddpa
->origin_snaps
);
2026 err
= snaplist_space(&ddpa
->shared_snaps
,
2027 snap
->ds
->ds_dir
->dd_origin_txg
, &ddpa
->cloneusedsnap
);
2031 err
= snaplist_space(&ddpa
->clone_snaps
,
2032 snap
->ds
->ds_dir
->dd_origin_txg
, &space
);
2035 ddpa
->cloneusedsnap
+= space
;
2037 if (origin_ds
->ds_dir
->dd_phys
->dd_flags
& DD_FLAG_USED_BREAKDOWN
) {
2038 err
= snaplist_space(&ddpa
->origin_snaps
,
2039 origin_ds
->ds_phys
->ds_creation_txg
, &ddpa
->originusedsnap
);
2045 promote_rele(ddpa
, FTAG
);
2050 dsl_dataset_promote_sync(void *arg
, dmu_tx_t
*tx
)
2052 dsl_dataset_promote_arg_t
*ddpa
= arg
;
2053 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2055 struct promotenode
*snap
;
2056 dsl_dataset_t
*origin_ds
;
2057 dsl_dataset_t
*origin_head
;
2059 dsl_dir_t
*odd
= NULL
;
2060 uint64_t oldnext_obj
;
2063 VERIFY0(promote_hold(ddpa
, dp
, FTAG
));
2064 hds
= ddpa
->ddpa_clone
;
2066 ASSERT0(hds
->ds_phys
->ds_flags
& DS_FLAG_NOPROMOTE
);
2068 snap
= list_head(&ddpa
->shared_snaps
);
2069 origin_ds
= snap
->ds
;
2072 snap
= list_head(&ddpa
->origin_snaps
);
2073 origin_head
= snap
->ds
;
2076 * We need to explicitly open odd, since origin_ds's dd will be
2079 VERIFY0(dsl_dir_hold_obj(dp
, origin_ds
->ds_dir
->dd_object
,
2082 /* change origin's next snap */
2083 dmu_buf_will_dirty(origin_ds
->ds_dbuf
, tx
);
2084 oldnext_obj
= origin_ds
->ds_phys
->ds_next_snap_obj
;
2085 snap
= list_tail(&ddpa
->clone_snaps
);
2086 ASSERT3U(snap
->ds
->ds_phys
->ds_prev_snap_obj
, ==, origin_ds
->ds_object
);
2087 origin_ds
->ds_phys
->ds_next_snap_obj
= snap
->ds
->ds_object
;
2089 /* change the origin's next clone */
2090 if (origin_ds
->ds_phys
->ds_next_clones_obj
) {
2091 dsl_dataset_remove_from_next_clones(origin_ds
,
2092 snap
->ds
->ds_object
, tx
);
2093 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2094 origin_ds
->ds_phys
->ds_next_clones_obj
,
2099 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
2100 ASSERT3U(dd
->dd_phys
->dd_origin_obj
, ==, origin_ds
->ds_object
);
2101 dd
->dd_phys
->dd_origin_obj
= odd
->dd_phys
->dd_origin_obj
;
2102 dd
->dd_origin_txg
= origin_head
->ds_dir
->dd_origin_txg
;
2103 dmu_buf_will_dirty(odd
->dd_dbuf
, tx
);
2104 odd
->dd_phys
->dd_origin_obj
= origin_ds
->ds_object
;
2105 origin_head
->ds_dir
->dd_origin_txg
=
2106 origin_ds
->ds_phys
->ds_creation_txg
;
2108 /* change dd_clone entries */
2109 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
2110 VERIFY0(zap_remove_int(dp
->dp_meta_objset
,
2111 odd
->dd_phys
->dd_clones
, hds
->ds_object
, tx
));
2112 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2113 ddpa
->origin_origin
->ds_dir
->dd_phys
->dd_clones
,
2114 hds
->ds_object
, tx
));
2116 VERIFY0(zap_remove_int(dp
->dp_meta_objset
,
2117 ddpa
->origin_origin
->ds_dir
->dd_phys
->dd_clones
,
2118 origin_head
->ds_object
, tx
));
2119 if (dd
->dd_phys
->dd_clones
== 0) {
2120 dd
->dd_phys
->dd_clones
= zap_create(dp
->dp_meta_objset
,
2121 DMU_OT_DSL_CLONES
, DMU_OT_NONE
, 0, tx
);
2123 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2124 dd
->dd_phys
->dd_clones
, origin_head
->ds_object
, tx
));
2127 /* move snapshots to this dir */
2128 for (snap
= list_head(&ddpa
->shared_snaps
); snap
;
2129 snap
= list_next(&ddpa
->shared_snaps
, snap
)) {
2130 dsl_dataset_t
*ds
= snap
->ds
;
2133 * Property callbacks are registered to a particular
2134 * dsl_dir. Since ours is changing, evict the objset
2135 * so that they will be unregistered from the old dsl_dir.
2137 if (ds
->ds_objset
) {
2138 dmu_objset_evict(ds
->ds_objset
);
2139 ds
->ds_objset
= NULL
;
2142 /* move snap name entry */
2143 VERIFY0(dsl_dataset_get_snapname(ds
));
2144 VERIFY0(dsl_dataset_snap_remove(origin_head
,
2145 ds
->ds_snapname
, tx
));
2146 VERIFY0(zap_add(dp
->dp_meta_objset
,
2147 hds
->ds_phys
->ds_snapnames_zapobj
, ds
->ds_snapname
,
2148 8, 1, &ds
->ds_object
, tx
));
2150 /* change containing dsl_dir */
2151 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
2152 ASSERT3U(ds
->ds_phys
->ds_dir_obj
, ==, odd
->dd_object
);
2153 ds
->ds_phys
->ds_dir_obj
= dd
->dd_object
;
2154 ASSERT3P(ds
->ds_dir
, ==, odd
);
2155 dsl_dir_rele(ds
->ds_dir
, ds
);
2156 VERIFY0(dsl_dir_hold_obj(dp
, dd
->dd_object
,
2157 NULL
, ds
, &ds
->ds_dir
));
2159 /* move any clone references */
2160 if (ds
->ds_phys
->ds_next_clones_obj
&&
2161 spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
2165 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
2166 ds
->ds_phys
->ds_next_clones_obj
);
2167 zap_cursor_retrieve(&zc
, &za
) == 0;
2168 zap_cursor_advance(&zc
)) {
2169 dsl_dataset_t
*cnds
;
2172 if (za
.za_first_integer
== oldnext_obj
) {
2174 * We've already moved the
2175 * origin's reference.
2180 VERIFY0(dsl_dataset_hold_obj(dp
,
2181 za
.za_first_integer
, FTAG
, &cnds
));
2182 o
= cnds
->ds_dir
->dd_phys
->dd_head_dataset_obj
;
2184 VERIFY0(zap_remove_int(dp
->dp_meta_objset
,
2185 odd
->dd_phys
->dd_clones
, o
, tx
));
2186 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2187 dd
->dd_phys
->dd_clones
, o
, tx
));
2188 dsl_dataset_rele(cnds
, FTAG
);
2190 zap_cursor_fini(&zc
);
2193 ASSERT(!dsl_prop_hascb(ds
));
2197 * Change space accounting.
2198 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2199 * both be valid, or both be 0 (resulting in delta == 0). This
2200 * is true for each of {clone,origin} independently.
2203 delta
= ddpa
->cloneusedsnap
-
2204 dd
->dd_phys
->dd_used_breakdown
[DD_USED_SNAP
];
2205 ASSERT3S(delta
, >=, 0);
2206 ASSERT3U(ddpa
->used
, >=, delta
);
2207 dsl_dir_diduse_space(dd
, DD_USED_SNAP
, delta
, 0, 0, tx
);
2208 dsl_dir_diduse_space(dd
, DD_USED_HEAD
,
2209 ddpa
->used
- delta
, ddpa
->comp
, ddpa
->uncomp
, tx
);
2211 delta
= ddpa
->originusedsnap
-
2212 odd
->dd_phys
->dd_used_breakdown
[DD_USED_SNAP
];
2213 ASSERT3S(delta
, <=, 0);
2214 ASSERT3U(ddpa
->used
, >=, -delta
);
2215 dsl_dir_diduse_space(odd
, DD_USED_SNAP
, delta
, 0, 0, tx
);
2216 dsl_dir_diduse_space(odd
, DD_USED_HEAD
,
2217 -ddpa
->used
- delta
, -ddpa
->comp
, -ddpa
->uncomp
, tx
);
2219 origin_ds
->ds_phys
->ds_unique_bytes
= ddpa
->unique
;
2221 /* log history record */
2222 spa_history_log_internal_ds(hds
, "promote", tx
, "");
2224 dsl_dir_rele(odd
, FTAG
);
2225 promote_rele(ddpa
, FTAG
);
2229 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2230 * (exclusive) and last_obj (inclusive). The list will be in reverse
2231 * order (last_obj will be the list_head()). If first_obj == 0, do all
2232 * snapshots back to this dataset's origin.
2235 snaplist_make(dsl_pool_t
*dp
,
2236 uint64_t first_obj
, uint64_t last_obj
, list_t
*l
, void *tag
)
2238 uint64_t obj
= last_obj
;
2240 list_create(l
, sizeof (struct promotenode
),
2241 offsetof(struct promotenode
, link
));
2243 while (obj
!= first_obj
) {
2245 struct promotenode
*snap
;
2248 err
= dsl_dataset_hold_obj(dp
, obj
, tag
, &ds
);
2249 ASSERT(err
!= ENOENT
);
2254 first_obj
= ds
->ds_dir
->dd_phys
->dd_origin_obj
;
2256 snap
= kmem_alloc(sizeof (*snap
), KM_PUSHPAGE
);
2258 list_insert_tail(l
, snap
);
2259 obj
= ds
->ds_phys
->ds_prev_snap_obj
;
2266 snaplist_space(list_t
*l
, uint64_t mintxg
, uint64_t *spacep
)
2268 struct promotenode
*snap
;
2271 for (snap
= list_head(l
); snap
; snap
= list_next(l
, snap
)) {
2272 uint64_t used
, comp
, uncomp
;
2273 dsl_deadlist_space_range(&snap
->ds
->ds_deadlist
,
2274 mintxg
, UINT64_MAX
, &used
, &comp
, &uncomp
);
2281 snaplist_destroy(list_t
*l
, void *tag
)
2283 struct promotenode
*snap
;
2285 if (l
== NULL
|| !list_link_active(&l
->list_head
))
2288 while ((snap
= list_tail(l
)) != NULL
) {
2289 list_remove(l
, snap
);
2290 dsl_dataset_rele(snap
->ds
, tag
);
2291 kmem_free(snap
, sizeof (*snap
));
2297 promote_hold(dsl_dataset_promote_arg_t
*ddpa
, dsl_pool_t
*dp
, void *tag
)
2301 struct promotenode
*snap
;
2303 error
= dsl_dataset_hold(dp
, ddpa
->ddpa_clonename
, tag
,
2307 dd
= ddpa
->ddpa_clone
->ds_dir
;
2309 if (dsl_dataset_is_snapshot(ddpa
->ddpa_clone
) ||
2310 !dsl_dir_is_clone(dd
)) {
2311 dsl_dataset_rele(ddpa
->ddpa_clone
, tag
);
2312 return (SET_ERROR(EINVAL
));
2315 error
= snaplist_make(dp
, 0, dd
->dd_phys
->dd_origin_obj
,
2316 &ddpa
->shared_snaps
, tag
);
2320 error
= snaplist_make(dp
, 0, ddpa
->ddpa_clone
->ds_object
,
2321 &ddpa
->clone_snaps
, tag
);
2325 snap
= list_head(&ddpa
->shared_snaps
);
2326 ASSERT3U(snap
->ds
->ds_object
, ==, dd
->dd_phys
->dd_origin_obj
);
2327 error
= snaplist_make(dp
, dd
->dd_phys
->dd_origin_obj
,
2328 snap
->ds
->ds_dir
->dd_phys
->dd_head_dataset_obj
,
2329 &ddpa
->origin_snaps
, tag
);
2333 if (snap
->ds
->ds_dir
->dd_phys
->dd_origin_obj
!= 0) {
2334 error
= dsl_dataset_hold_obj(dp
,
2335 snap
->ds
->ds_dir
->dd_phys
->dd_origin_obj
,
2336 tag
, &ddpa
->origin_origin
);
2342 promote_rele(ddpa
, tag
);
2347 promote_rele(dsl_dataset_promote_arg_t
*ddpa
, void *tag
)
2349 snaplist_destroy(&ddpa
->shared_snaps
, tag
);
2350 snaplist_destroy(&ddpa
->clone_snaps
, tag
);
2351 snaplist_destroy(&ddpa
->origin_snaps
, tag
);
2352 if (ddpa
->origin_origin
!= NULL
)
2353 dsl_dataset_rele(ddpa
->origin_origin
, tag
);
2354 dsl_dataset_rele(ddpa
->ddpa_clone
, tag
);
2360 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2361 * in with the name. (It must be at least MAXNAMELEN bytes long.)
2364 dsl_dataset_promote(const char *name
, char *conflsnap
)
2366 dsl_dataset_promote_arg_t ddpa
= { 0 };
2372 * We will modify space proportional to the number of
2373 * snapshots. Compute numsnaps.
2375 error
= dmu_objset_hold(name
, FTAG
, &os
);
2378 error
= zap_count(dmu_objset_pool(os
)->dp_meta_objset
,
2379 dmu_objset_ds(os
)->ds_phys
->ds_snapnames_zapobj
, &numsnaps
);
2380 dmu_objset_rele(os
, FTAG
);
2384 ddpa
.ddpa_clonename
= name
;
2385 ddpa
.err_ds
= conflsnap
;
2387 return (dsl_sync_task(name
, dsl_dataset_promote_check
,
2388 dsl_dataset_promote_sync
, &ddpa
, 2 + numsnaps
));
2392 dsl_dataset_clone_swap_check_impl(dsl_dataset_t
*clone
,
2393 dsl_dataset_t
*origin_head
, boolean_t force
, void *owner
, dmu_tx_t
*tx
)
2395 int64_t unused_refres_delta
;
2397 /* they should both be heads */
2398 if (dsl_dataset_is_snapshot(clone
) ||
2399 dsl_dataset_is_snapshot(origin_head
))
2400 return (SET_ERROR(EINVAL
));
2402 /* if we are not forcing, the branch point should be just before them */
2403 if (!force
&& clone
->ds_prev
!= origin_head
->ds_prev
)
2404 return (SET_ERROR(EINVAL
));
2406 /* clone should be the clone (unless they are unrelated) */
2407 if (clone
->ds_prev
!= NULL
&&
2408 clone
->ds_prev
!= clone
->ds_dir
->dd_pool
->dp_origin_snap
&&
2409 origin_head
->ds_dir
!= clone
->ds_prev
->ds_dir
)
2410 return (SET_ERROR(EINVAL
));
2412 /* the clone should be a child of the origin */
2413 if (clone
->ds_dir
->dd_parent
!= origin_head
->ds_dir
)
2414 return (SET_ERROR(EINVAL
));
2416 /* origin_head shouldn't be modified unless 'force' */
2418 dsl_dataset_modified_since_snap(origin_head
, origin_head
->ds_prev
))
2419 return (SET_ERROR(ETXTBSY
));
2421 /* origin_head should have no long holds (e.g. is not mounted) */
2422 if (dsl_dataset_handoff_check(origin_head
, owner
, tx
))
2423 return (SET_ERROR(EBUSY
));
2425 /* check amount of any unconsumed refreservation */
2426 unused_refres_delta
=
2427 (int64_t)MIN(origin_head
->ds_reserved
,
2428 origin_head
->ds_phys
->ds_unique_bytes
) -
2429 (int64_t)MIN(origin_head
->ds_reserved
,
2430 clone
->ds_phys
->ds_unique_bytes
);
2432 if (unused_refres_delta
> 0 &&
2433 unused_refres_delta
>
2434 dsl_dir_space_available(origin_head
->ds_dir
, NULL
, 0, TRUE
))
2435 return (SET_ERROR(ENOSPC
));
2437 /* clone can't be over the head's refquota */
2438 if (origin_head
->ds_quota
!= 0 &&
2439 clone
->ds_phys
->ds_referenced_bytes
> origin_head
->ds_quota
)
2440 return (SET_ERROR(EDQUOT
));
2446 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t
*clone
,
2447 dsl_dataset_t
*origin_head
, dmu_tx_t
*tx
)
2449 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2450 int64_t unused_refres_delta
;
2452 ASSERT(clone
->ds_reserved
== 0);
2453 ASSERT(origin_head
->ds_quota
== 0 ||
2454 clone
->ds_phys
->ds_unique_bytes
<= origin_head
->ds_quota
);
2455 ASSERT3P(clone
->ds_prev
, ==, origin_head
->ds_prev
);
2457 dmu_buf_will_dirty(clone
->ds_dbuf
, tx
);
2458 dmu_buf_will_dirty(origin_head
->ds_dbuf
, tx
);
2460 if (clone
->ds_objset
!= NULL
) {
2461 dmu_objset_evict(clone
->ds_objset
);
2462 clone
->ds_objset
= NULL
;
2465 if (origin_head
->ds_objset
!= NULL
) {
2466 dmu_objset_evict(origin_head
->ds_objset
);
2467 origin_head
->ds_objset
= NULL
;
2470 unused_refres_delta
=
2471 (int64_t)MIN(origin_head
->ds_reserved
,
2472 origin_head
->ds_phys
->ds_unique_bytes
) -
2473 (int64_t)MIN(origin_head
->ds_reserved
,
2474 clone
->ds_phys
->ds_unique_bytes
);
2477 * Reset origin's unique bytes, if it exists.
2479 if (clone
->ds_prev
) {
2480 dsl_dataset_t
*origin
= clone
->ds_prev
;
2481 uint64_t comp
, uncomp
;
2483 dmu_buf_will_dirty(origin
->ds_dbuf
, tx
);
2484 dsl_deadlist_space_range(&clone
->ds_deadlist
,
2485 origin
->ds_phys
->ds_prev_snap_txg
, UINT64_MAX
,
2486 &origin
->ds_phys
->ds_unique_bytes
, &comp
, &uncomp
);
2492 tmp
= origin_head
->ds_phys
->ds_bp
;
2493 origin_head
->ds_phys
->ds_bp
= clone
->ds_phys
->ds_bp
;
2494 clone
->ds_phys
->ds_bp
= tmp
;
2497 /* set dd_*_bytes */
2499 int64_t dused
, dcomp
, duncomp
;
2500 uint64_t cdl_used
, cdl_comp
, cdl_uncomp
;
2501 uint64_t odl_used
, odl_comp
, odl_uncomp
;
2503 ASSERT3U(clone
->ds_dir
->dd_phys
->
2504 dd_used_breakdown
[DD_USED_SNAP
], ==, 0);
2506 dsl_deadlist_space(&clone
->ds_deadlist
,
2507 &cdl_used
, &cdl_comp
, &cdl_uncomp
);
2508 dsl_deadlist_space(&origin_head
->ds_deadlist
,
2509 &odl_used
, &odl_comp
, &odl_uncomp
);
2511 dused
= clone
->ds_phys
->ds_referenced_bytes
+ cdl_used
-
2512 (origin_head
->ds_phys
->ds_referenced_bytes
+ odl_used
);
2513 dcomp
= clone
->ds_phys
->ds_compressed_bytes
+ cdl_comp
-
2514 (origin_head
->ds_phys
->ds_compressed_bytes
+ odl_comp
);
2515 duncomp
= clone
->ds_phys
->ds_uncompressed_bytes
+
2517 (origin_head
->ds_phys
->ds_uncompressed_bytes
+ odl_uncomp
);
2519 dsl_dir_diduse_space(origin_head
->ds_dir
, DD_USED_HEAD
,
2520 dused
, dcomp
, duncomp
, tx
);
2521 dsl_dir_diduse_space(clone
->ds_dir
, DD_USED_HEAD
,
2522 -dused
, -dcomp
, -duncomp
, tx
);
2525 * The difference in the space used by snapshots is the
2526 * difference in snapshot space due to the head's
2527 * deadlist (since that's the only thing that's
2528 * changing that affects the snapused).
2530 dsl_deadlist_space_range(&clone
->ds_deadlist
,
2531 origin_head
->ds_dir
->dd_origin_txg
, UINT64_MAX
,
2532 &cdl_used
, &cdl_comp
, &cdl_uncomp
);
2533 dsl_deadlist_space_range(&origin_head
->ds_deadlist
,
2534 origin_head
->ds_dir
->dd_origin_txg
, UINT64_MAX
,
2535 &odl_used
, &odl_comp
, &odl_uncomp
);
2536 dsl_dir_transfer_space(origin_head
->ds_dir
, cdl_used
- odl_used
,
2537 DD_USED_HEAD
, DD_USED_SNAP
, tx
);
2540 /* swap ds_*_bytes */
2541 SWITCH64(origin_head
->ds_phys
->ds_referenced_bytes
,
2542 clone
->ds_phys
->ds_referenced_bytes
);
2543 SWITCH64(origin_head
->ds_phys
->ds_compressed_bytes
,
2544 clone
->ds_phys
->ds_compressed_bytes
);
2545 SWITCH64(origin_head
->ds_phys
->ds_uncompressed_bytes
,
2546 clone
->ds_phys
->ds_uncompressed_bytes
);
2547 SWITCH64(origin_head
->ds_phys
->ds_unique_bytes
,
2548 clone
->ds_phys
->ds_unique_bytes
);
2550 /* apply any parent delta for change in unconsumed refreservation */
2551 dsl_dir_diduse_space(origin_head
->ds_dir
, DD_USED_REFRSRV
,
2552 unused_refres_delta
, 0, 0, tx
);
2557 dsl_deadlist_close(&clone
->ds_deadlist
);
2558 dsl_deadlist_close(&origin_head
->ds_deadlist
);
2559 SWITCH64(origin_head
->ds_phys
->ds_deadlist_obj
,
2560 clone
->ds_phys
->ds_deadlist_obj
);
2561 dsl_deadlist_open(&clone
->ds_deadlist
, dp
->dp_meta_objset
,
2562 clone
->ds_phys
->ds_deadlist_obj
);
2563 dsl_deadlist_open(&origin_head
->ds_deadlist
, dp
->dp_meta_objset
,
2564 origin_head
->ds_phys
->ds_deadlist_obj
);
2566 dsl_scan_ds_clone_swapped(origin_head
, clone
, tx
);
2568 spa_history_log_internal_ds(clone
, "clone swap", tx
,
2569 "parent=%s", origin_head
->ds_dir
->dd_myname
);
2573 * Given a pool name and a dataset object number in that pool,
2574 * return the name of that dataset.
2577 dsl_dsobj_to_dsname(char *pname
, uint64_t obj
, char *buf
)
2583 error
= dsl_pool_hold(pname
, FTAG
, &dp
);
2587 error
= dsl_dataset_hold_obj(dp
, obj
, FTAG
, &ds
);
2589 dsl_dataset_name(ds
, buf
);
2590 dsl_dataset_rele(ds
, FTAG
);
2592 dsl_pool_rele(dp
, FTAG
);
2598 dsl_dataset_check_quota(dsl_dataset_t
*ds
, boolean_t check_quota
,
2599 uint64_t asize
, uint64_t inflight
, uint64_t *used
, uint64_t *ref_rsrv
)
2603 ASSERT3S(asize
, >, 0);
2606 * *ref_rsrv is the portion of asize that will come from any
2607 * unconsumed refreservation space.
2611 mutex_enter(&ds
->ds_lock
);
2613 * Make a space adjustment for reserved bytes.
2615 if (ds
->ds_reserved
> ds
->ds_phys
->ds_unique_bytes
) {
2617 ds
->ds_reserved
- ds
->ds_phys
->ds_unique_bytes
);
2618 *used
-= (ds
->ds_reserved
- ds
->ds_phys
->ds_unique_bytes
);
2620 asize
- MIN(asize
, parent_delta(ds
, asize
+ inflight
));
2623 if (!check_quota
|| ds
->ds_quota
== 0) {
2624 mutex_exit(&ds
->ds_lock
);
2628 * If they are requesting more space, and our current estimate
2629 * is over quota, they get to try again unless the actual
2630 * on-disk is over quota and there are no pending changes (which
2631 * may free up space for us).
2633 if (ds
->ds_phys
->ds_referenced_bytes
+ inflight
>= ds
->ds_quota
) {
2635 ds
->ds_phys
->ds_referenced_bytes
< ds
->ds_quota
)
2636 error
= SET_ERROR(ERESTART
);
2638 error
= SET_ERROR(EDQUOT
);
2640 mutex_exit(&ds
->ds_lock
);
2645 typedef struct dsl_dataset_set_qr_arg
{
2646 const char *ddsqra_name
;
2647 zprop_source_t ddsqra_source
;
2648 uint64_t ddsqra_value
;
2649 } dsl_dataset_set_qr_arg_t
;
2654 dsl_dataset_set_refquota_check(void *arg
, dmu_tx_t
*tx
)
2656 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
2657 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2662 if (spa_version(dp
->dp_spa
) < SPA_VERSION_REFQUOTA
)
2663 return (SET_ERROR(ENOTSUP
));
2665 error
= dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
);
2669 if (dsl_dataset_is_snapshot(ds
)) {
2670 dsl_dataset_rele(ds
, FTAG
);
2671 return (SET_ERROR(EINVAL
));
2674 error
= dsl_prop_predict(ds
->ds_dir
,
2675 zfs_prop_to_name(ZFS_PROP_REFQUOTA
),
2676 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, &newval
);
2678 dsl_dataset_rele(ds
, FTAG
);
2683 dsl_dataset_rele(ds
, FTAG
);
2687 if (newval
< ds
->ds_phys
->ds_referenced_bytes
||
2688 newval
< ds
->ds_reserved
) {
2689 dsl_dataset_rele(ds
, FTAG
);
2690 return (SET_ERROR(ENOSPC
));
2693 dsl_dataset_rele(ds
, FTAG
);
2698 dsl_dataset_set_refquota_sync(void *arg
, dmu_tx_t
*tx
)
2700 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
2701 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2705 VERIFY0(dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
));
2707 dsl_prop_set_sync_impl(ds
,
2708 zfs_prop_to_name(ZFS_PROP_REFQUOTA
),
2709 ddsqra
->ddsqra_source
, sizeof (ddsqra
->ddsqra_value
), 1,
2710 &ddsqra
->ddsqra_value
, tx
);
2712 VERIFY0(dsl_prop_get_int_ds(ds
,
2713 zfs_prop_to_name(ZFS_PROP_REFQUOTA
), &newval
));
2715 if (ds
->ds_quota
!= newval
) {
2716 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
2717 ds
->ds_quota
= newval
;
2719 dsl_dataset_rele(ds
, FTAG
);
2723 dsl_dataset_set_refquota(const char *dsname
, zprop_source_t source
,
2726 dsl_dataset_set_qr_arg_t ddsqra
;
2728 ddsqra
.ddsqra_name
= dsname
;
2729 ddsqra
.ddsqra_source
= source
;
2730 ddsqra
.ddsqra_value
= refquota
;
2732 return (dsl_sync_task(dsname
, dsl_dataset_set_refquota_check
,
2733 dsl_dataset_set_refquota_sync
, &ddsqra
, 0));
2737 dsl_dataset_set_refreservation_check(void *arg
, dmu_tx_t
*tx
)
2739 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
2740 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2743 uint64_t newval
, unique
;
2745 if (spa_version(dp
->dp_spa
) < SPA_VERSION_REFRESERVATION
)
2746 return (SET_ERROR(ENOTSUP
));
2748 error
= dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
);
2752 if (dsl_dataset_is_snapshot(ds
)) {
2753 dsl_dataset_rele(ds
, FTAG
);
2754 return (SET_ERROR(EINVAL
));
2757 error
= dsl_prop_predict(ds
->ds_dir
,
2758 zfs_prop_to_name(ZFS_PROP_REFRESERVATION
),
2759 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, &newval
);
2761 dsl_dataset_rele(ds
, FTAG
);
2766 * If we are doing the preliminary check in open context, the
2767 * space estimates may be inaccurate.
2769 if (!dmu_tx_is_syncing(tx
)) {
2770 dsl_dataset_rele(ds
, FTAG
);
2774 mutex_enter(&ds
->ds_lock
);
2775 if (!DS_UNIQUE_IS_ACCURATE(ds
))
2776 dsl_dataset_recalc_head_uniq(ds
);
2777 unique
= ds
->ds_phys
->ds_unique_bytes
;
2778 mutex_exit(&ds
->ds_lock
);
2780 if (MAX(unique
, newval
) > MAX(unique
, ds
->ds_reserved
)) {
2781 uint64_t delta
= MAX(unique
, newval
) -
2782 MAX(unique
, ds
->ds_reserved
);
2785 dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, B_TRUE
) ||
2786 (ds
->ds_quota
> 0 && newval
> ds
->ds_quota
)) {
2787 dsl_dataset_rele(ds
, FTAG
);
2788 return (SET_ERROR(ENOSPC
));
2792 dsl_dataset_rele(ds
, FTAG
);
2797 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t
*ds
,
2798 zprop_source_t source
, uint64_t value
, dmu_tx_t
*tx
)
2804 dsl_prop_set_sync_impl(ds
, zfs_prop_to_name(ZFS_PROP_REFRESERVATION
),
2805 source
, sizeof (value
), 1, &value
, tx
);
2807 VERIFY0(dsl_prop_get_int_ds(ds
,
2808 zfs_prop_to_name(ZFS_PROP_REFRESERVATION
), &newval
));
2810 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
2811 mutex_enter(&ds
->ds_dir
->dd_lock
);
2812 mutex_enter(&ds
->ds_lock
);
2813 ASSERT(DS_UNIQUE_IS_ACCURATE(ds
));
2814 unique
= ds
->ds_phys
->ds_unique_bytes
;
2815 delta
= MAX(0, (int64_t)(newval
- unique
)) -
2816 MAX(0, (int64_t)(ds
->ds_reserved
- unique
));
2817 ds
->ds_reserved
= newval
;
2818 mutex_exit(&ds
->ds_lock
);
2820 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_REFRSRV
, delta
, 0, 0, tx
);
2821 mutex_exit(&ds
->ds_dir
->dd_lock
);
2825 dsl_dataset_set_refreservation_sync(void *arg
, dmu_tx_t
*tx
)
2827 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
2828 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2831 VERIFY0(dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
));
2832 dsl_dataset_set_refreservation_sync_impl(ds
,
2833 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, tx
);
2834 dsl_dataset_rele(ds
, FTAG
);
2838 dsl_dataset_set_refreservation(const char *dsname
, zprop_source_t source
,
2839 uint64_t refreservation
)
2841 dsl_dataset_set_qr_arg_t ddsqra
;
2843 ddsqra
.ddsqra_name
= dsname
;
2844 ddsqra
.ddsqra_source
= source
;
2845 ddsqra
.ddsqra_value
= refreservation
;
2847 return (dsl_sync_task(dsname
, dsl_dataset_set_refreservation_check
,
2848 dsl_dataset_set_refreservation_sync
, &ddsqra
, 0));
2852 * Return (in *usedp) the amount of space written in new that is not
2853 * present in oldsnap. New may be a snapshot or the head. Old must be
2854 * a snapshot before new, in new's filesystem (or its origin). If not then
2855 * fail and return EINVAL.
2857 * The written space is calculated by considering two components: First, we
2858 * ignore any freed space, and calculate the written as new's used space
2859 * minus old's used space. Next, we add in the amount of space that was freed
2860 * between the two snapshots, thus reducing new's used space relative to old's.
2861 * Specifically, this is the space that was born before old->ds_creation_txg,
2862 * and freed before new (ie. on new's deadlist or a previous deadlist).
2864 * space freed [---------------------]
2865 * snapshots ---O-------O--------O-------O------
2869 dsl_dataset_space_written(dsl_dataset_t
*oldsnap
, dsl_dataset_t
*new,
2870 uint64_t *usedp
, uint64_t *compp
, uint64_t *uncompp
)
2874 dsl_pool_t
*dp
= new->ds_dir
->dd_pool
;
2876 ASSERT(dsl_pool_config_held(dp
));
2879 *usedp
+= new->ds_phys
->ds_referenced_bytes
;
2880 *usedp
-= oldsnap
->ds_phys
->ds_referenced_bytes
;
2883 *compp
+= new->ds_phys
->ds_compressed_bytes
;
2884 *compp
-= oldsnap
->ds_phys
->ds_compressed_bytes
;
2887 *uncompp
+= new->ds_phys
->ds_uncompressed_bytes
;
2888 *uncompp
-= oldsnap
->ds_phys
->ds_uncompressed_bytes
;
2890 snapobj
= new->ds_object
;
2891 while (snapobj
!= oldsnap
->ds_object
) {
2892 dsl_dataset_t
*snap
;
2893 uint64_t used
, comp
, uncomp
;
2895 if (snapobj
== new->ds_object
) {
2898 err
= dsl_dataset_hold_obj(dp
, snapobj
, FTAG
, &snap
);
2903 if (snap
->ds_phys
->ds_prev_snap_txg
==
2904 oldsnap
->ds_phys
->ds_creation_txg
) {
2906 * The blocks in the deadlist can not be born after
2907 * ds_prev_snap_txg, so get the whole deadlist space,
2908 * which is more efficient (especially for old-format
2909 * deadlists). Unfortunately the deadlist code
2910 * doesn't have enough information to make this
2911 * optimization itself.
2913 dsl_deadlist_space(&snap
->ds_deadlist
,
2914 &used
, &comp
, &uncomp
);
2916 dsl_deadlist_space_range(&snap
->ds_deadlist
,
2917 0, oldsnap
->ds_phys
->ds_creation_txg
,
2918 &used
, &comp
, &uncomp
);
2925 * If we get to the beginning of the chain of snapshots
2926 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
2927 * was not a snapshot of/before new.
2929 snapobj
= snap
->ds_phys
->ds_prev_snap_obj
;
2931 dsl_dataset_rele(snap
, FTAG
);
2933 err
= SET_ERROR(EINVAL
);
2942 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
2943 * lastsnap, and all snapshots in between are deleted.
2945 * blocks that would be freed [---------------------------]
2946 * snapshots ---O-------O--------O-------O--------O
2947 * firstsnap lastsnap
2949 * This is the set of blocks that were born after the snap before firstsnap,
2950 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
2951 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
2952 * We calculate this by iterating over the relevant deadlists (from the snap
2953 * after lastsnap, backward to the snap after firstsnap), summing up the
2954 * space on the deadlist that was born after the snap before firstsnap.
2957 dsl_dataset_space_wouldfree(dsl_dataset_t
*firstsnap
,
2958 dsl_dataset_t
*lastsnap
,
2959 uint64_t *usedp
, uint64_t *compp
, uint64_t *uncompp
)
2963 dsl_pool_t
*dp
= firstsnap
->ds_dir
->dd_pool
;
2965 ASSERT(dsl_dataset_is_snapshot(firstsnap
));
2966 ASSERT(dsl_dataset_is_snapshot(lastsnap
));
2969 * Check that the snapshots are in the same dsl_dir, and firstsnap
2970 * is before lastsnap.
2972 if (firstsnap
->ds_dir
!= lastsnap
->ds_dir
||
2973 firstsnap
->ds_phys
->ds_creation_txg
>
2974 lastsnap
->ds_phys
->ds_creation_txg
)
2975 return (SET_ERROR(EINVAL
));
2977 *usedp
= *compp
= *uncompp
= 0;
2979 snapobj
= lastsnap
->ds_phys
->ds_next_snap_obj
;
2980 while (snapobj
!= firstsnap
->ds_object
) {
2982 uint64_t used
, comp
, uncomp
;
2984 err
= dsl_dataset_hold_obj(dp
, snapobj
, FTAG
, &ds
);
2988 dsl_deadlist_space_range(&ds
->ds_deadlist
,
2989 firstsnap
->ds_phys
->ds_prev_snap_txg
, UINT64_MAX
,
2990 &used
, &comp
, &uncomp
);
2995 snapobj
= ds
->ds_phys
->ds_prev_snap_obj
;
2996 ASSERT3U(snapobj
, !=, 0);
2997 dsl_dataset_rele(ds
, FTAG
);
3003 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3004 * For example, they could both be snapshots of the same filesystem, and
3005 * 'earlier' is before 'later'. Or 'earlier' could be the origin of
3006 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's
3007 * filesystem. Or 'earlier' could be the origin's origin.
3009 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3012 dsl_dataset_is_before(dsl_dataset_t
*later
, dsl_dataset_t
*earlier
,
3013 uint64_t earlier_txg
)
3015 dsl_pool_t
*dp
= later
->ds_dir
->dd_pool
;
3018 dsl_dataset_t
*origin
;
3020 ASSERT(dsl_pool_config_held(dp
));
3021 ASSERT(dsl_dataset_is_snapshot(earlier
) || earlier_txg
!= 0);
3023 if (earlier_txg
== 0)
3024 earlier_txg
= earlier
->ds_phys
->ds_creation_txg
;
3026 if (dsl_dataset_is_snapshot(later
) &&
3027 earlier_txg
>= later
->ds_phys
->ds_creation_txg
)
3030 if (later
->ds_dir
== earlier
->ds_dir
)
3032 if (!dsl_dir_is_clone(later
->ds_dir
))
3035 if (later
->ds_dir
->dd_phys
->dd_origin_obj
== earlier
->ds_object
)
3037 error
= dsl_dataset_hold_obj(dp
,
3038 later
->ds_dir
->dd_phys
->dd_origin_obj
, FTAG
, &origin
);
3041 ret
= dsl_dataset_is_before(origin
, earlier
, earlier_txg
);
3042 dsl_dataset_rele(origin
, FTAG
);
3048 dsl_dataset_zapify(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
3050 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
3051 dmu_object_zapify(mos
, ds
->ds_object
, DMU_OT_DSL_DATASET
, tx
);
3054 #if defined(_KERNEL) && defined(HAVE_SPL)
3055 EXPORT_SYMBOL(dsl_dataset_hold
);
3056 EXPORT_SYMBOL(dsl_dataset_hold_obj
);
3057 EXPORT_SYMBOL(dsl_dataset_own
);
3058 EXPORT_SYMBOL(dsl_dataset_own_obj
);
3059 EXPORT_SYMBOL(dsl_dataset_name
);
3060 EXPORT_SYMBOL(dsl_dataset_rele
);
3061 EXPORT_SYMBOL(dsl_dataset_disown
);
3062 EXPORT_SYMBOL(dsl_dataset_tryown
);
3063 EXPORT_SYMBOL(dsl_dataset_create_sync
);
3064 EXPORT_SYMBOL(dsl_dataset_create_sync_dd
);
3065 EXPORT_SYMBOL(dsl_dataset_snapshot_check
);
3066 EXPORT_SYMBOL(dsl_dataset_snapshot_sync
);
3067 EXPORT_SYMBOL(dsl_dataset_promote
);
3068 EXPORT_SYMBOL(dsl_dataset_user_hold
);
3069 EXPORT_SYMBOL(dsl_dataset_user_release
);
3070 EXPORT_SYMBOL(dsl_dataset_get_holds
);
3071 EXPORT_SYMBOL(dsl_dataset_get_blkptr
);
3072 EXPORT_SYMBOL(dsl_dataset_set_blkptr
);
3073 EXPORT_SYMBOL(dsl_dataset_get_spa
);
3074 EXPORT_SYMBOL(dsl_dataset_modified_since_snap
);
3075 EXPORT_SYMBOL(dsl_dataset_space_written
);
3076 EXPORT_SYMBOL(dsl_dataset_space_wouldfree
);
3077 EXPORT_SYMBOL(dsl_dataset_sync
);
3078 EXPORT_SYMBOL(dsl_dataset_block_born
);
3079 EXPORT_SYMBOL(dsl_dataset_block_kill
);
3080 EXPORT_SYMBOL(dsl_dataset_block_freeable
);
3081 EXPORT_SYMBOL(dsl_dataset_prev_snap_txg
);
3082 EXPORT_SYMBOL(dsl_dataset_dirty
);
3083 EXPORT_SYMBOL(dsl_dataset_stats
);
3084 EXPORT_SYMBOL(dsl_dataset_fast_stat
);
3085 EXPORT_SYMBOL(dsl_dataset_space
);
3086 EXPORT_SYMBOL(dsl_dataset_fsid_guid
);
3087 EXPORT_SYMBOL(dsl_dsobj_to_dsname
);
3088 EXPORT_SYMBOL(dsl_dataset_check_quota
);
3089 EXPORT_SYMBOL(dsl_dataset_clone_swap_check_impl
);
3090 EXPORT_SYMBOL(dsl_dataset_clone_swap_sync_impl
);