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) 2014, Joyent, Inc. All rights reserved.
25 * Copyright (c) 2014 RackTop Systems.
26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
29 #include <sys/dmu_objset.h>
30 #include <sys/dsl_dataset.h>
31 #include <sys/dsl_dir.h>
32 #include <sys/dsl_prop.h>
33 #include <sys/dsl_synctask.h>
34 #include <sys/dmu_traverse.h>
35 #include <sys/dmu_impl.h>
36 #include <sys/dmu_tx.h>
40 #include <sys/zfeature.h>
41 #include <sys/unique.h>
42 #include <sys/zfs_context.h>
43 #include <sys/zfs_ioctl.h>
45 #include <sys/zfs_znode.h>
46 #include <sys/zfs_onexit.h>
48 #include <sys/dsl_scan.h>
49 #include <sys/dsl_deadlist.h>
50 #include <sys/dsl_destroy.h>
51 #include <sys/dsl_userhold.h>
52 #include <sys/dsl_bookmark.h>
55 * The SPA supports block sizes up to 16MB. However, very large blocks
56 * can have an impact on i/o latency (e.g. tying up a spinning disk for
57 * ~300ms), and also potentially on the memory allocator. Therefore,
58 * we do not allow the recordsize to be set larger than zfs_max_recordsize
59 * (default 1MB). Larger blocks can be created by changing this tunable,
60 * and pools with larger blocks can always be imported and used, regardless
63 int zfs_max_recordsize
= 1 * 1024 * 1024;
65 #define SWITCH64(x, y) \
67 uint64_t __tmp = (x); \
72 #define DS_REF_MAX (1ULL << 62)
74 extern inline dsl_dataset_phys_t
*dsl_dataset_phys(dsl_dataset_t
*ds
);
77 * Figure out how much of this delta should be propogated to the dsl_dir
78 * layer. If there's a refreservation, that space has already been
79 * partially accounted for in our ancestors.
82 parent_delta(dsl_dataset_t
*ds
, int64_t delta
)
84 dsl_dataset_phys_t
*ds_phys
;
85 uint64_t old_bytes
, new_bytes
;
87 if (ds
->ds_reserved
== 0)
90 ds_phys
= dsl_dataset_phys(ds
);
91 old_bytes
= MAX(ds_phys
->ds_unique_bytes
, ds
->ds_reserved
);
92 new_bytes
= MAX(ds_phys
->ds_unique_bytes
+ delta
, ds
->ds_reserved
);
94 ASSERT3U(ABS((int64_t)(new_bytes
- old_bytes
)), <=, ABS(delta
));
95 return (new_bytes
- old_bytes
);
99 dsl_dataset_block_born(dsl_dataset_t
*ds
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
101 int used
, compressed
, uncompressed
;
104 used
= bp_get_dsize_sync(tx
->tx_pool
->dp_spa
, bp
);
105 compressed
= BP_GET_PSIZE(bp
);
106 uncompressed
= BP_GET_UCSIZE(bp
);
108 dprintf_bp(bp
, "ds=%p", ds
);
110 ASSERT(dmu_tx_is_syncing(tx
));
111 /* It could have been compressed away to nothing */
114 ASSERT(BP_GET_TYPE(bp
) != DMU_OT_NONE
);
115 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp
)));
117 dsl_pool_mos_diduse_space(tx
->tx_pool
,
118 used
, compressed
, uncompressed
);
122 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
123 mutex_enter(&ds
->ds_lock
);
124 delta
= parent_delta(ds
, used
);
125 dsl_dataset_phys(ds
)->ds_referenced_bytes
+= used
;
126 dsl_dataset_phys(ds
)->ds_compressed_bytes
+= compressed
;
127 dsl_dataset_phys(ds
)->ds_uncompressed_bytes
+= uncompressed
;
128 dsl_dataset_phys(ds
)->ds_unique_bytes
+= used
;
129 if (BP_GET_LSIZE(bp
) > SPA_OLD_MAXBLOCKSIZE
)
130 ds
->ds_need_large_blocks
= B_TRUE
;
131 mutex_exit(&ds
->ds_lock
);
132 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_HEAD
, delta
,
133 compressed
, uncompressed
, tx
);
134 dsl_dir_transfer_space(ds
->ds_dir
, used
- delta
,
135 DD_USED_REFRSRV
, DD_USED_HEAD
, tx
);
139 dsl_dataset_block_kill(dsl_dataset_t
*ds
, const blkptr_t
*bp
, dmu_tx_t
*tx
,
142 int used
= bp_get_dsize_sync(tx
->tx_pool
->dp_spa
, bp
);
143 int compressed
= BP_GET_PSIZE(bp
);
144 int uncompressed
= BP_GET_UCSIZE(bp
);
149 ASSERT(dmu_tx_is_syncing(tx
));
150 ASSERT(bp
->blk_birth
<= tx
->tx_txg
);
153 dsl_free(tx
->tx_pool
, tx
->tx_txg
, bp
);
154 dsl_pool_mos_diduse_space(tx
->tx_pool
,
155 -used
, -compressed
, -uncompressed
);
158 ASSERT3P(tx
->tx_pool
, ==, ds
->ds_dir
->dd_pool
);
160 ASSERT(!ds
->ds_is_snapshot
);
161 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
163 if (bp
->blk_birth
> dsl_dataset_phys(ds
)->ds_prev_snap_txg
) {
166 dprintf_bp(bp
, "freeing ds=%llu", ds
->ds_object
);
167 dsl_free(tx
->tx_pool
, tx
->tx_txg
, bp
);
169 mutex_enter(&ds
->ds_lock
);
170 ASSERT(dsl_dataset_phys(ds
)->ds_unique_bytes
>= used
||
171 !DS_UNIQUE_IS_ACCURATE(ds
));
172 delta
= parent_delta(ds
, -used
);
173 dsl_dataset_phys(ds
)->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
);
180 dprintf_bp(bp
, "putting on dead list: %s", "");
183 * We are here as part of zio's write done callback,
184 * which means we're a zio interrupt thread. We can't
185 * call dsl_deadlist_insert() now because it may block
186 * waiting for I/O. Instead, put bp on the deferred
187 * queue and let dsl_pool_sync() finish the job.
189 bplist_append(&ds
->ds_pending_deadlist
, bp
);
191 dsl_deadlist_insert(&ds
->ds_deadlist
, bp
, tx
);
193 ASSERT3U(ds
->ds_prev
->ds_object
, ==,
194 dsl_dataset_phys(ds
)->ds_prev_snap_obj
);
195 ASSERT(dsl_dataset_phys(ds
->ds_prev
)->ds_num_children
> 0);
196 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
197 if (dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
==
198 ds
->ds_object
&& bp
->blk_birth
>
199 dsl_dataset_phys(ds
->ds_prev
)->ds_prev_snap_txg
) {
200 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
201 mutex_enter(&ds
->ds_prev
->ds_lock
);
202 dsl_dataset_phys(ds
->ds_prev
)->ds_unique_bytes
+= used
;
203 mutex_exit(&ds
->ds_prev
->ds_lock
);
205 if (bp
->blk_birth
> ds
->ds_dir
->dd_origin_txg
) {
206 dsl_dir_transfer_space(ds
->ds_dir
, used
,
207 DD_USED_HEAD
, DD_USED_SNAP
, tx
);
210 mutex_enter(&ds
->ds_lock
);
211 ASSERT3U(dsl_dataset_phys(ds
)->ds_referenced_bytes
, >=, used
);
212 dsl_dataset_phys(ds
)->ds_referenced_bytes
-= used
;
213 ASSERT3U(dsl_dataset_phys(ds
)->ds_compressed_bytes
, >=, compressed
);
214 dsl_dataset_phys(ds
)->ds_compressed_bytes
-= compressed
;
215 ASSERT3U(dsl_dataset_phys(ds
)->ds_uncompressed_bytes
, >=, uncompressed
);
216 dsl_dataset_phys(ds
)->ds_uncompressed_bytes
-= uncompressed
;
217 mutex_exit(&ds
->ds_lock
);
223 dsl_dataset_prev_snap_txg(dsl_dataset_t
*ds
)
225 uint64_t trysnap
= 0;
230 * The snapshot creation could fail, but that would cause an
231 * incorrect FALSE return, which would only result in an
232 * overestimation of the amount of space that an operation would
233 * consume, which is OK.
235 * There's also a small window where we could miss a pending
236 * snapshot, because we could set the sync task in the quiescing
237 * phase. So this should only be used as a guess.
239 if (ds
->ds_trysnap_txg
>
240 spa_last_synced_txg(ds
->ds_dir
->dd_pool
->dp_spa
))
241 trysnap
= ds
->ds_trysnap_txg
;
242 return (MAX(dsl_dataset_phys(ds
)->ds_prev_snap_txg
, trysnap
));
246 dsl_dataset_block_freeable(dsl_dataset_t
*ds
, const blkptr_t
*bp
,
249 if (blk_birth
<= dsl_dataset_prev_snap_txg(ds
) ||
250 (bp
!= NULL
&& BP_IS_HOLE(bp
)))
253 ddt_prefetch(dsl_dataset_get_spa(ds
), bp
);
259 dsl_dataset_evict(void *dbu
)
261 dsl_dataset_t
*ds
= dbu
;
263 ASSERT(ds
->ds_owner
== NULL
);
267 unique_remove(ds
->ds_fsid_guid
);
269 if (ds
->ds_objset
!= NULL
)
270 dmu_objset_evict(ds
->ds_objset
);
273 dsl_dataset_rele(ds
->ds_prev
, ds
);
277 bplist_destroy(&ds
->ds_pending_deadlist
);
278 if (ds
->ds_deadlist
.dl_os
!= NULL
)
279 dsl_deadlist_close(&ds
->ds_deadlist
);
281 dsl_dir_async_rele(ds
->ds_dir
, ds
);
283 ASSERT(!list_link_active(&ds
->ds_synced_link
));
285 mutex_destroy(&ds
->ds_lock
);
286 mutex_destroy(&ds
->ds_opening_lock
);
287 mutex_destroy(&ds
->ds_sendstream_lock
);
288 refcount_destroy(&ds
->ds_longholds
);
290 kmem_free(ds
, sizeof (dsl_dataset_t
));
294 dsl_dataset_get_snapname(dsl_dataset_t
*ds
)
296 dsl_dataset_phys_t
*headphys
;
299 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
300 objset_t
*mos
= dp
->dp_meta_objset
;
302 if (ds
->ds_snapname
[0])
304 if (dsl_dataset_phys(ds
)->ds_next_snap_obj
== 0)
307 err
= dmu_bonus_hold(mos
, dsl_dir_phys(ds
->ds_dir
)->dd_head_dataset_obj
,
311 headphys
= headdbuf
->db_data
;
312 err
= zap_value_search(dp
->dp_meta_objset
,
313 headphys
->ds_snapnames_zapobj
, ds
->ds_object
, 0, ds
->ds_snapname
);
314 dmu_buf_rele(headdbuf
, FTAG
);
319 dsl_dataset_snap_lookup(dsl_dataset_t
*ds
, const char *name
, uint64_t *value
)
321 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
322 uint64_t snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
326 if (dsl_dataset_phys(ds
)->ds_flags
& DS_FLAG_CI_DATASET
)
331 err
= zap_lookup_norm(mos
, snapobj
, name
, 8, 1,
332 value
, mt
, NULL
, 0, NULL
);
333 if (err
== ENOTSUP
&& mt
== MT_FIRST
)
334 err
= zap_lookup(mos
, snapobj
, name
, 8, 1, value
);
339 dsl_dataset_snap_remove(dsl_dataset_t
*ds
, const char *name
, dmu_tx_t
*tx
,
342 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
343 uint64_t snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
347 dsl_dir_snap_cmtime_update(ds
->ds_dir
);
349 if (dsl_dataset_phys(ds
)->ds_flags
& DS_FLAG_CI_DATASET
)
354 err
= zap_remove_norm(mos
, snapobj
, name
, mt
, tx
);
355 if (err
== ENOTSUP
&& mt
== MT_FIRST
)
356 err
= zap_remove(mos
, snapobj
, name
, tx
);
358 if (err
== 0 && adj_cnt
)
359 dsl_fs_ss_count_adjust(ds
->ds_dir
, -1,
360 DD_FIELD_SNAPSHOT_COUNT
, tx
);
366 dsl_dataset_try_add_ref(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *tag
)
368 dmu_buf_t
*dbuf
= ds
->ds_dbuf
;
369 boolean_t result
= B_FALSE
;
371 if (dbuf
!= NULL
&& dmu_buf_try_add_ref(dbuf
, dp
->dp_meta_objset
,
372 ds
->ds_object
, DMU_BONUS_BLKID
, tag
)) {
374 if (ds
== dmu_buf_get_user(dbuf
))
377 dmu_buf_rele(dbuf
, tag
);
384 dsl_dataset_hold_obj(dsl_pool_t
*dp
, uint64_t dsobj
, void *tag
,
387 objset_t
*mos
= dp
->dp_meta_objset
;
391 dmu_object_info_t doi
;
393 ASSERT(dsl_pool_config_held(dp
));
395 err
= dmu_bonus_hold(mos
, dsobj
, tag
, &dbuf
);
399 /* Make sure dsobj has the correct object type. */
400 dmu_object_info_from_db(dbuf
, &doi
);
401 if (doi
.doi_bonus_type
!= DMU_OT_DSL_DATASET
) {
402 dmu_buf_rele(dbuf
, tag
);
403 return (SET_ERROR(EINVAL
));
406 ds
= dmu_buf_get_user(dbuf
);
408 dsl_dataset_t
*winner
= NULL
;
410 ds
= kmem_zalloc(sizeof (dsl_dataset_t
), KM_SLEEP
);
412 ds
->ds_object
= dsobj
;
413 ds
->ds_is_snapshot
= dsl_dataset_phys(ds
)->ds_num_children
!= 0;
414 list_link_init(&ds
->ds_synced_link
);
416 mutex_init(&ds
->ds_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
417 mutex_init(&ds
->ds_opening_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
418 mutex_init(&ds
->ds_sendstream_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
419 refcount_create(&ds
->ds_longholds
);
421 bplist_create(&ds
->ds_pending_deadlist
);
422 dsl_deadlist_open(&ds
->ds_deadlist
,
423 mos
, dsl_dataset_phys(ds
)->ds_deadlist_obj
);
425 list_create(&ds
->ds_sendstreams
, sizeof (dmu_sendarg_t
),
426 offsetof(dmu_sendarg_t
, dsa_link
));
428 if (doi
.doi_type
== DMU_OTN_ZAP_METADATA
) {
429 int zaperr
= zap_contains(mos
, dsobj
,
430 DS_FIELD_LARGE_BLOCKS
);
431 if (zaperr
!= ENOENT
) {
433 ds
->ds_large_blocks
= B_TRUE
;
438 err
= dsl_dir_hold_obj(dp
,
439 dsl_dataset_phys(ds
)->ds_dir_obj
, NULL
, ds
,
443 mutex_destroy(&ds
->ds_lock
);
444 mutex_destroy(&ds
->ds_opening_lock
);
445 mutex_destroy(&ds
->ds_sendstream_lock
);
446 refcount_destroy(&ds
->ds_longholds
);
447 bplist_destroy(&ds
->ds_pending_deadlist
);
448 dsl_deadlist_close(&ds
->ds_deadlist
);
449 kmem_free(ds
, sizeof (dsl_dataset_t
));
450 dmu_buf_rele(dbuf
, tag
);
454 if (!ds
->ds_is_snapshot
) {
455 ds
->ds_snapname
[0] = '\0';
456 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0) {
457 err
= dsl_dataset_hold_obj(dp
,
458 dsl_dataset_phys(ds
)->ds_prev_snap_obj
,
461 if (doi
.doi_type
== DMU_OTN_ZAP_METADATA
) {
462 int zaperr
= zap_lookup(mos
, ds
->ds_object
,
463 DS_FIELD_BOOKMARK_NAMES
,
464 sizeof (ds
->ds_bookmarks
), 1,
466 if (zaperr
!= ENOENT
)
470 if (zfs_flags
& ZFS_DEBUG_SNAPNAMES
)
471 err
= dsl_dataset_get_snapname(ds
);
473 dsl_dataset_phys(ds
)->ds_userrefs_obj
!= 0) {
475 ds
->ds_dir
->dd_pool
->dp_meta_objset
,
476 dsl_dataset_phys(ds
)->ds_userrefs_obj
,
481 if (err
== 0 && !ds
->ds_is_snapshot
) {
482 err
= dsl_prop_get_int_ds(ds
,
483 zfs_prop_to_name(ZFS_PROP_REFRESERVATION
),
486 err
= dsl_prop_get_int_ds(ds
,
487 zfs_prop_to_name(ZFS_PROP_REFQUOTA
),
491 ds
->ds_reserved
= ds
->ds_quota
= 0;
494 dmu_buf_init_user(&ds
->ds_dbu
, dsl_dataset_evict
, &ds
->ds_dbuf
);
496 winner
= dmu_buf_set_user_ie(dbuf
, &ds
->ds_dbu
);
498 if (err
!= 0 || winner
!= NULL
) {
499 bplist_destroy(&ds
->ds_pending_deadlist
);
500 dsl_deadlist_close(&ds
->ds_deadlist
);
502 dsl_dataset_rele(ds
->ds_prev
, ds
);
503 dsl_dir_rele(ds
->ds_dir
, ds
);
504 mutex_destroy(&ds
->ds_lock
);
505 mutex_destroy(&ds
->ds_opening_lock
);
506 mutex_destroy(&ds
->ds_sendstream_lock
);
507 refcount_destroy(&ds
->ds_longholds
);
508 kmem_free(ds
, sizeof (dsl_dataset_t
));
510 dmu_buf_rele(dbuf
, tag
);
516 unique_insert(dsl_dataset_phys(ds
)->ds_fsid_guid
);
519 ASSERT3P(ds
->ds_dbuf
, ==, dbuf
);
520 ASSERT3P(dsl_dataset_phys(ds
), ==, dbuf
->db_data
);
521 ASSERT(dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0 ||
522 spa_version(dp
->dp_spa
) < SPA_VERSION_ORIGIN
||
523 dp
->dp_origin_snap
== NULL
|| ds
== dp
->dp_origin_snap
);
529 dsl_dataset_hold(dsl_pool_t
*dp
, const char *name
,
530 void *tag
, dsl_dataset_t
**dsp
)
533 const char *snapname
;
537 err
= dsl_dir_hold(dp
, name
, FTAG
, &dd
, &snapname
);
541 ASSERT(dsl_pool_config_held(dp
));
542 obj
= dsl_dir_phys(dd
)->dd_head_dataset_obj
;
544 err
= dsl_dataset_hold_obj(dp
, obj
, tag
, dsp
);
546 err
= SET_ERROR(ENOENT
);
548 /* we may be looking for a snapshot */
549 if (err
== 0 && snapname
!= NULL
) {
552 if (*snapname
++ != '@') {
553 dsl_dataset_rele(*dsp
, tag
);
554 dsl_dir_rele(dd
, FTAG
);
555 return (SET_ERROR(ENOENT
));
558 dprintf("looking for snapshot '%s'\n", snapname
);
559 err
= dsl_dataset_snap_lookup(*dsp
, snapname
, &obj
);
561 err
= dsl_dataset_hold_obj(dp
, obj
, tag
, &ds
);
562 dsl_dataset_rele(*dsp
, tag
);
565 mutex_enter(&ds
->ds_lock
);
566 if (ds
->ds_snapname
[0] == 0)
567 (void) strlcpy(ds
->ds_snapname
, snapname
,
568 sizeof (ds
->ds_snapname
));
569 mutex_exit(&ds
->ds_lock
);
574 dsl_dir_rele(dd
, FTAG
);
579 dsl_dataset_own_obj(dsl_pool_t
*dp
, uint64_t dsobj
,
580 void *tag
, dsl_dataset_t
**dsp
)
582 int err
= dsl_dataset_hold_obj(dp
, dsobj
, tag
, dsp
);
585 if (!dsl_dataset_tryown(*dsp
, tag
)) {
586 dsl_dataset_rele(*dsp
, tag
);
588 return (SET_ERROR(EBUSY
));
594 dsl_dataset_own(dsl_pool_t
*dp
, const char *name
,
595 void *tag
, dsl_dataset_t
**dsp
)
597 int err
= dsl_dataset_hold(dp
, name
, tag
, dsp
);
600 if (!dsl_dataset_tryown(*dsp
, tag
)) {
601 dsl_dataset_rele(*dsp
, tag
);
602 return (SET_ERROR(EBUSY
));
608 * See the comment above dsl_pool_hold() for details. In summary, a long
609 * hold is used to prevent destruction of a dataset while the pool hold
610 * is dropped, allowing other concurrent operations (e.g. spa_sync()).
612 * The dataset and pool must be held when this function is called. After it
613 * is called, the pool hold may be released while the dataset is still held
617 dsl_dataset_long_hold(dsl_dataset_t
*ds
, void *tag
)
619 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
620 (void) refcount_add(&ds
->ds_longholds
, tag
);
624 dsl_dataset_long_rele(dsl_dataset_t
*ds
, void *tag
)
626 (void) refcount_remove(&ds
->ds_longholds
, tag
);
629 /* Return B_TRUE if there are any long holds on this dataset. */
631 dsl_dataset_long_held(dsl_dataset_t
*ds
)
633 return (!refcount_is_zero(&ds
->ds_longholds
));
637 dsl_dataset_name(dsl_dataset_t
*ds
, char *name
)
640 (void) strcpy(name
, "mos");
642 dsl_dir_name(ds
->ds_dir
, name
);
643 VERIFY0(dsl_dataset_get_snapname(ds
));
644 if (ds
->ds_snapname
[0]) {
645 (void) strcat(name
, "@");
647 * We use a "recursive" mutex so that we
648 * can call dprintf_ds() with ds_lock held.
650 if (!MUTEX_HELD(&ds
->ds_lock
)) {
651 mutex_enter(&ds
->ds_lock
);
652 (void) strcat(name
, ds
->ds_snapname
);
653 mutex_exit(&ds
->ds_lock
);
655 (void) strcat(name
, ds
->ds_snapname
);
662 dsl_dataset_rele(dsl_dataset_t
*ds
, void *tag
)
664 dmu_buf_rele(ds
->ds_dbuf
, tag
);
668 dsl_dataset_disown(dsl_dataset_t
*ds
, void *tag
)
670 ASSERT3P(ds
->ds_owner
, ==, tag
);
671 ASSERT(ds
->ds_dbuf
!= NULL
);
673 mutex_enter(&ds
->ds_lock
);
675 mutex_exit(&ds
->ds_lock
);
676 dsl_dataset_long_rele(ds
, tag
);
677 dsl_dataset_rele(ds
, tag
);
681 dsl_dataset_tryown(dsl_dataset_t
*ds
, void *tag
)
683 boolean_t gotit
= FALSE
;
685 mutex_enter(&ds
->ds_lock
);
686 if (ds
->ds_owner
== NULL
&& !DS_IS_INCONSISTENT(ds
)) {
688 dsl_dataset_long_hold(ds
, tag
);
691 mutex_exit(&ds
->ds_lock
);
696 dsl_dataset_create_sync_dd(dsl_dir_t
*dd
, dsl_dataset_t
*origin
,
697 uint64_t flags
, dmu_tx_t
*tx
)
699 dsl_pool_t
*dp
= dd
->dd_pool
;
701 dsl_dataset_phys_t
*dsphys
;
703 objset_t
*mos
= dp
->dp_meta_objset
;
706 origin
= dp
->dp_origin_snap
;
708 ASSERT(origin
== NULL
|| origin
->ds_dir
->dd_pool
== dp
);
709 ASSERT(origin
== NULL
|| dsl_dataset_phys(origin
)->ds_num_children
> 0);
710 ASSERT(dmu_tx_is_syncing(tx
));
711 ASSERT(dsl_dir_phys(dd
)->dd_head_dataset_obj
== 0);
713 dsobj
= dmu_object_alloc(mos
, DMU_OT_DSL_DATASET
, 0,
714 DMU_OT_DSL_DATASET
, sizeof (dsl_dataset_phys_t
), tx
);
715 VERIFY0(dmu_bonus_hold(mos
, dsobj
, FTAG
, &dbuf
));
716 dmu_buf_will_dirty(dbuf
, tx
);
717 dsphys
= dbuf
->db_data
;
718 bzero(dsphys
, sizeof (dsl_dataset_phys_t
));
719 dsphys
->ds_dir_obj
= dd
->dd_object
;
720 dsphys
->ds_flags
= flags
;
721 dsphys
->ds_fsid_guid
= unique_create();
722 (void) random_get_pseudo_bytes((void*)&dsphys
->ds_guid
,
723 sizeof (dsphys
->ds_guid
));
724 dsphys
->ds_snapnames_zapobj
=
725 zap_create_norm(mos
, U8_TEXTPREP_TOUPPER
, DMU_OT_DSL_DS_SNAP_MAP
,
727 dsphys
->ds_creation_time
= gethrestime_sec();
728 dsphys
->ds_creation_txg
= tx
->tx_txg
== TXG_INITIAL
? 1 : tx
->tx_txg
;
730 if (origin
== NULL
) {
731 dsphys
->ds_deadlist_obj
= dsl_deadlist_alloc(mos
, tx
);
733 dsl_dataset_t
*ohds
; /* head of the origin snapshot */
735 dsphys
->ds_prev_snap_obj
= origin
->ds_object
;
736 dsphys
->ds_prev_snap_txg
=
737 dsl_dataset_phys(origin
)->ds_creation_txg
;
738 dsphys
->ds_referenced_bytes
=
739 dsl_dataset_phys(origin
)->ds_referenced_bytes
;
740 dsphys
->ds_compressed_bytes
=
741 dsl_dataset_phys(origin
)->ds_compressed_bytes
;
742 dsphys
->ds_uncompressed_bytes
=
743 dsl_dataset_phys(origin
)->ds_uncompressed_bytes
;
744 dsphys
->ds_bp
= dsl_dataset_phys(origin
)->ds_bp
;
747 * Inherit flags that describe the dataset's contents
748 * (INCONSISTENT) or properties (Case Insensitive).
750 dsphys
->ds_flags
|= dsl_dataset_phys(origin
)->ds_flags
&
751 (DS_FLAG_INCONSISTENT
| DS_FLAG_CI_DATASET
);
753 if (origin
->ds_large_blocks
)
754 dsl_dataset_activate_large_blocks_sync_impl(dsobj
, tx
);
756 dmu_buf_will_dirty(origin
->ds_dbuf
, tx
);
757 dsl_dataset_phys(origin
)->ds_num_children
++;
759 VERIFY0(dsl_dataset_hold_obj(dp
,
760 dsl_dir_phys(origin
->ds_dir
)->dd_head_dataset_obj
,
762 dsphys
->ds_deadlist_obj
= dsl_deadlist_clone(&ohds
->ds_deadlist
,
763 dsphys
->ds_prev_snap_txg
, dsphys
->ds_prev_snap_obj
, tx
);
764 dsl_dataset_rele(ohds
, FTAG
);
766 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_NEXT_CLONES
) {
767 if (dsl_dataset_phys(origin
)->ds_next_clones_obj
== 0) {
768 dsl_dataset_phys(origin
)->ds_next_clones_obj
=
770 DMU_OT_NEXT_CLONES
, DMU_OT_NONE
, 0, tx
);
772 VERIFY0(zap_add_int(mos
,
773 dsl_dataset_phys(origin
)->ds_next_clones_obj
,
777 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
778 dsl_dir_phys(dd
)->dd_origin_obj
= origin
->ds_object
;
779 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
780 if (dsl_dir_phys(origin
->ds_dir
)->dd_clones
== 0) {
781 dmu_buf_will_dirty(origin
->ds_dir
->dd_dbuf
, tx
);
782 dsl_dir_phys(origin
->ds_dir
)->dd_clones
=
784 DMU_OT_DSL_CLONES
, DMU_OT_NONE
, 0, tx
);
786 VERIFY0(zap_add_int(mos
,
787 dsl_dir_phys(origin
->ds_dir
)->dd_clones
,
792 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_UNIQUE_ACCURATE
)
793 dsphys
->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
795 dmu_buf_rele(dbuf
, FTAG
);
797 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
798 dsl_dir_phys(dd
)->dd_head_dataset_obj
= dsobj
;
804 dsl_dataset_zero_zil(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
808 VERIFY0(dmu_objset_from_ds(ds
, &os
));
809 bzero(&os
->os_zil_header
, sizeof (os
->os_zil_header
));
810 dsl_dataset_dirty(ds
, tx
);
814 dsl_dataset_create_sync(dsl_dir_t
*pdd
, const char *lastname
,
815 dsl_dataset_t
*origin
, uint64_t flags
, cred_t
*cr
, dmu_tx_t
*tx
)
817 dsl_pool_t
*dp
= pdd
->dd_pool
;
818 uint64_t dsobj
, ddobj
;
821 ASSERT(dmu_tx_is_syncing(tx
));
822 ASSERT(lastname
[0] != '@');
824 ddobj
= dsl_dir_create_sync(dp
, pdd
, lastname
, tx
);
825 VERIFY0(dsl_dir_hold_obj(dp
, ddobj
, lastname
, FTAG
, &dd
));
827 dsobj
= dsl_dataset_create_sync_dd(dd
, origin
,
828 flags
& ~DS_CREATE_FLAG_NODIRTY
, tx
);
830 dsl_deleg_set_create_perms(dd
, tx
, cr
);
833 * Since we're creating a new node we know it's a leaf, so we can
834 * initialize the counts if the limit feature is active.
836 if (spa_feature_is_active(dp
->dp_spa
, SPA_FEATURE_FS_SS_LIMIT
)) {
838 objset_t
*os
= dd
->dd_pool
->dp_meta_objset
;
840 dsl_dir_zapify(dd
, tx
);
841 VERIFY0(zap_add(os
, dd
->dd_object
, DD_FIELD_FILESYSTEM_COUNT
,
842 sizeof (cnt
), 1, &cnt
, tx
));
843 VERIFY0(zap_add(os
, dd
->dd_object
, DD_FIELD_SNAPSHOT_COUNT
,
844 sizeof (cnt
), 1, &cnt
, tx
));
847 dsl_dir_rele(dd
, FTAG
);
850 * If we are creating a clone, make sure we zero out any stale
851 * data from the origin snapshots zil header.
853 if (origin
!= NULL
&& !(flags
& DS_CREATE_FLAG_NODIRTY
)) {
856 VERIFY0(dsl_dataset_hold_obj(dp
, dsobj
, FTAG
, &ds
));
857 dsl_dataset_zero_zil(ds
, tx
);
858 dsl_dataset_rele(ds
, FTAG
);
865 * The unique space in the head dataset can be calculated by subtracting
866 * the space used in the most recent snapshot, that is still being used
867 * in this file system, from the space currently in use. To figure out
868 * the space in the most recent snapshot still in use, we need to take
869 * the total space used in the snapshot and subtract out the space that
870 * has been freed up since the snapshot was taken.
873 dsl_dataset_recalc_head_uniq(dsl_dataset_t
*ds
)
876 uint64_t dlused
, dlcomp
, dluncomp
;
878 ASSERT(!ds
->ds_is_snapshot
);
880 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0)
881 mrs_used
= dsl_dataset_phys(ds
->ds_prev
)->ds_referenced_bytes
;
885 dsl_deadlist_space(&ds
->ds_deadlist
, &dlused
, &dlcomp
, &dluncomp
);
887 ASSERT3U(dlused
, <=, mrs_used
);
888 dsl_dataset_phys(ds
)->ds_unique_bytes
=
889 dsl_dataset_phys(ds
)->ds_referenced_bytes
- (mrs_used
- dlused
);
891 if (spa_version(ds
->ds_dir
->dd_pool
->dp_spa
) >=
892 SPA_VERSION_UNIQUE_ACCURATE
)
893 dsl_dataset_phys(ds
)->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
897 dsl_dataset_remove_from_next_clones(dsl_dataset_t
*ds
, uint64_t obj
,
900 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
902 ASSERTV(uint64_t count
);
904 ASSERT(dsl_dataset_phys(ds
)->ds_num_children
>= 2);
905 err
= zap_remove_int(mos
, dsl_dataset_phys(ds
)->ds_next_clones_obj
,
908 * The err should not be ENOENT, but a bug in a previous version
909 * of the code could cause upgrade_clones_cb() to not set
910 * ds_next_snap_obj when it should, leading to a missing entry.
911 * If we knew that the pool was created after
912 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
913 * ENOENT. However, at least we can check that we don't have
914 * too many entries in the next_clones_obj even after failing to
919 ASSERT0(zap_count(mos
, dsl_dataset_phys(ds
)->ds_next_clones_obj
,
921 ASSERT3U(count
, <=, dsl_dataset_phys(ds
)->ds_num_children
- 2);
926 dsl_dataset_get_blkptr(dsl_dataset_t
*ds
)
928 return (&dsl_dataset_phys(ds
)->ds_bp
);
932 dsl_dataset_set_blkptr(dsl_dataset_t
*ds
, blkptr_t
*bp
, dmu_tx_t
*tx
)
934 ASSERT(dmu_tx_is_syncing(tx
));
935 /* If it's the meta-objset, set dp_meta_rootbp */
937 tx
->tx_pool
->dp_meta_rootbp
= *bp
;
939 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
940 dsl_dataset_phys(ds
)->ds_bp
= *bp
;
945 dsl_dataset_get_spa(dsl_dataset_t
*ds
)
947 return (ds
->ds_dir
->dd_pool
->dp_spa
);
951 dsl_dataset_dirty(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
955 if (ds
== NULL
) /* this is the meta-objset */
958 ASSERT(ds
->ds_objset
!= NULL
);
960 if (dsl_dataset_phys(ds
)->ds_next_snap_obj
!= 0)
961 panic("dirtying snapshot!");
963 dp
= ds
->ds_dir
->dd_pool
;
965 if (txg_list_add(&dp
->dp_dirty_datasets
, ds
, tx
->tx_txg
)) {
966 /* up the hold count until we can be written out */
967 dmu_buf_add_ref(ds
->ds_dbuf
, ds
);
972 dsl_dataset_is_dirty(dsl_dataset_t
*ds
)
976 for (t
= 0; t
< TXG_SIZE
; t
++) {
977 if (txg_list_member(&ds
->ds_dir
->dd_pool
->dp_dirty_datasets
,
985 dsl_dataset_snapshot_reserve_space(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
989 if (!dmu_tx_is_syncing(tx
))
993 * If there's an fs-only reservation, any blocks that might become
994 * owned by the snapshot dataset must be accommodated by space
995 * outside of the reservation.
997 ASSERT(ds
->ds_reserved
== 0 || DS_UNIQUE_IS_ACCURATE(ds
));
998 asize
= MIN(dsl_dataset_phys(ds
)->ds_unique_bytes
, ds
->ds_reserved
);
999 if (asize
> dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
))
1000 return (SET_ERROR(ENOSPC
));
1003 * Propagate any reserved space for this snapshot to other
1004 * snapshot checks in this sync group.
1007 dsl_dir_willuse_space(ds
->ds_dir
, asize
, tx
);
1012 typedef struct dsl_dataset_snapshot_arg
{
1013 nvlist_t
*ddsa_snaps
;
1014 nvlist_t
*ddsa_props
;
1015 nvlist_t
*ddsa_errors
;
1017 } dsl_dataset_snapshot_arg_t
;
1020 dsl_dataset_snapshot_check_impl(dsl_dataset_t
*ds
, const char *snapname
,
1021 dmu_tx_t
*tx
, boolean_t recv
, uint64_t cnt
, cred_t
*cr
)
1026 ds
->ds_trysnap_txg
= tx
->tx_txg
;
1028 if (!dmu_tx_is_syncing(tx
))
1032 * We don't allow multiple snapshots of the same txg. If there
1033 * is already one, try again.
1035 if (dsl_dataset_phys(ds
)->ds_prev_snap_txg
>= tx
->tx_txg
)
1036 return (SET_ERROR(EAGAIN
));
1039 * Check for conflicting snapshot name.
1041 error
= dsl_dataset_snap_lookup(ds
, snapname
, &value
);
1043 return (SET_ERROR(EEXIST
));
1044 if (error
!= ENOENT
)
1048 * We don't allow taking snapshots of inconsistent datasets, such as
1049 * those into which we are currently receiving. However, if we are
1050 * creating this snapshot as part of a receive, this check will be
1051 * executed atomically with respect to the completion of the receive
1052 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1053 * case we ignore this, knowing it will be fixed up for us shortly in
1054 * dmu_recv_end_sync().
1056 if (!recv
&& DS_IS_INCONSISTENT(ds
))
1057 return (SET_ERROR(EBUSY
));
1060 * Skip the check for temporary snapshots or if we have already checked
1061 * the counts in dsl_dataset_snapshot_check. This means we really only
1062 * check the count here when we're receiving a stream.
1064 if (cnt
!= 0 && cr
!= NULL
) {
1065 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, cnt
,
1066 ZFS_PROP_SNAPSHOT_LIMIT
, NULL
, cr
);
1071 error
= dsl_dataset_snapshot_reserve_space(ds
, tx
);
1079 dsl_dataset_snapshot_check(void *arg
, dmu_tx_t
*tx
)
1081 dsl_dataset_snapshot_arg_t
*ddsa
= arg
;
1082 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1087 * Pre-compute how many total new snapshots will be created for each
1088 * level in the tree and below. This is needed for validating the
1089 * snapshot limit when either taking a recursive snapshot or when
1090 * taking multiple snapshots.
1092 * The problem is that the counts are not actually adjusted when
1093 * we are checking, only when we finally sync. For a single snapshot,
1094 * this is easy, the count will increase by 1 at each node up the tree,
1095 * but its more complicated for the recursive/multiple snapshot case.
1097 * The dsl_fs_ss_limit_check function does recursively check the count
1098 * at each level up the tree but since it is validating each snapshot
1099 * independently we need to be sure that we are validating the complete
1100 * count for the entire set of snapshots. We do this by rolling up the
1101 * counts for each component of the name into an nvlist and then
1102 * checking each of those cases with the aggregated count.
1104 * This approach properly handles not only the recursive snapshot
1105 * case (where we get all of those on the ddsa_snaps list) but also
1106 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1107 * validate the limit on 'a' using a count of 2).
1109 * We validate the snapshot names in the third loop and only report
1112 if (dmu_tx_is_syncing(tx
)) {
1114 nvlist_t
*cnt_track
= NULL
;
1115 cnt_track
= fnvlist_alloc();
1117 nm
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
1119 /* Rollup aggregated counts into the cnt_track list */
1120 for (pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, NULL
);
1122 pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, pair
)) {
1126 (void) strlcpy(nm
, nvpair_name(pair
), MAXPATHLEN
);
1127 pdelim
= strchr(nm
, '@');
1133 if (nvlist_lookup_uint64(cnt_track
, nm
,
1135 /* update existing entry */
1136 fnvlist_add_uint64(cnt_track
, nm
,
1140 fnvlist_add_uint64(cnt_track
, nm
, 1);
1143 pdelim
= strrchr(nm
, '/');
1146 } while (pdelim
!= NULL
);
1149 kmem_free(nm
, MAXPATHLEN
);
1151 /* Check aggregated counts at each level */
1152 for (pair
= nvlist_next_nvpair(cnt_track
, NULL
);
1153 pair
!= NULL
; pair
= nvlist_next_nvpair(cnt_track
, pair
)) {
1159 name
= nvpair_name(pair
);
1160 cnt
= fnvpair_value_uint64(pair
);
1163 error
= dsl_dataset_hold(dp
, name
, FTAG
, &ds
);
1165 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, cnt
,
1166 ZFS_PROP_SNAPSHOT_LIMIT
, NULL
,
1168 dsl_dataset_rele(ds
, FTAG
);
1172 if (ddsa
->ddsa_errors
!= NULL
)
1173 fnvlist_add_int32(ddsa
->ddsa_errors
,
1176 /* only report one error for this check */
1180 nvlist_free(cnt_track
);
1183 for (pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, NULL
);
1184 pair
!= NULL
; pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, pair
)) {
1188 char dsname
[MAXNAMELEN
];
1190 name
= nvpair_name(pair
);
1191 if (strlen(name
) >= MAXNAMELEN
)
1192 error
= SET_ERROR(ENAMETOOLONG
);
1194 atp
= strchr(name
, '@');
1196 error
= SET_ERROR(EINVAL
);
1198 (void) strlcpy(dsname
, name
, atp
- name
+ 1);
1201 error
= dsl_dataset_hold(dp
, dsname
, FTAG
, &ds
);
1203 /* passing 0/NULL skips dsl_fs_ss_limit_check */
1204 error
= dsl_dataset_snapshot_check_impl(ds
,
1205 atp
+ 1, tx
, B_FALSE
, 0, NULL
);
1206 dsl_dataset_rele(ds
, FTAG
);
1210 if (ddsa
->ddsa_errors
!= NULL
) {
1211 fnvlist_add_int32(ddsa
->ddsa_errors
,
1222 dsl_dataset_snapshot_sync_impl(dsl_dataset_t
*ds
, const char *snapname
,
1225 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1227 dsl_dataset_phys_t
*dsphys
;
1228 uint64_t dsobj
, crtxg
;
1229 objset_t
*mos
= dp
->dp_meta_objset
;
1230 ASSERTV(static zil_header_t zero_zil
);
1231 ASSERTV(objset_t
*os
);
1233 ASSERT(RRW_WRITE_HELD(&dp
->dp_config_rwlock
));
1236 * If we are on an old pool, the zil must not be active, in which
1237 * case it will be zeroed. Usually zil_suspend() accomplishes this.
1239 ASSERT(spa_version(dmu_tx_pool(tx
)->dp_spa
) >= SPA_VERSION_FAST_SNAP
||
1240 dmu_objset_from_ds(ds
, &os
) != 0 ||
1241 bcmp(&os
->os_phys
->os_zil_header
, &zero_zil
,
1242 sizeof (zero_zil
)) == 0);
1244 dsl_fs_ss_count_adjust(ds
->ds_dir
, 1, DD_FIELD_SNAPSHOT_COUNT
, tx
);
1247 * The origin's ds_creation_txg has to be < TXG_INITIAL
1249 if (strcmp(snapname
, ORIGIN_DIR_NAME
) == 0)
1254 dsobj
= dmu_object_alloc(mos
, DMU_OT_DSL_DATASET
, 0,
1255 DMU_OT_DSL_DATASET
, sizeof (dsl_dataset_phys_t
), tx
);
1256 VERIFY0(dmu_bonus_hold(mos
, dsobj
, FTAG
, &dbuf
));
1257 dmu_buf_will_dirty(dbuf
, tx
);
1258 dsphys
= dbuf
->db_data
;
1259 bzero(dsphys
, sizeof (dsl_dataset_phys_t
));
1260 dsphys
->ds_dir_obj
= ds
->ds_dir
->dd_object
;
1261 dsphys
->ds_fsid_guid
= unique_create();
1262 (void) random_get_pseudo_bytes((void*)&dsphys
->ds_guid
,
1263 sizeof (dsphys
->ds_guid
));
1264 dsphys
->ds_prev_snap_obj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
1265 dsphys
->ds_prev_snap_txg
= dsl_dataset_phys(ds
)->ds_prev_snap_txg
;
1266 dsphys
->ds_next_snap_obj
= ds
->ds_object
;
1267 dsphys
->ds_num_children
= 1;
1268 dsphys
->ds_creation_time
= gethrestime_sec();
1269 dsphys
->ds_creation_txg
= crtxg
;
1270 dsphys
->ds_deadlist_obj
= dsl_dataset_phys(ds
)->ds_deadlist_obj
;
1271 dsphys
->ds_referenced_bytes
= dsl_dataset_phys(ds
)->ds_referenced_bytes
;
1272 dsphys
->ds_compressed_bytes
= dsl_dataset_phys(ds
)->ds_compressed_bytes
;
1273 dsphys
->ds_uncompressed_bytes
=
1274 dsl_dataset_phys(ds
)->ds_uncompressed_bytes
;
1275 dsphys
->ds_flags
= dsl_dataset_phys(ds
)->ds_flags
;
1276 dsphys
->ds_bp
= dsl_dataset_phys(ds
)->ds_bp
;
1277 dmu_buf_rele(dbuf
, FTAG
);
1279 if (ds
->ds_large_blocks
)
1280 dsl_dataset_activate_large_blocks_sync_impl(dsobj
, tx
);
1282 ASSERT3U(ds
->ds_prev
!= 0, ==,
1283 dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0);
1285 uint64_t next_clones_obj
=
1286 dsl_dataset_phys(ds
->ds_prev
)->ds_next_clones_obj
;
1287 ASSERT(dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
==
1289 dsl_dataset_phys(ds
->ds_prev
)->ds_num_children
> 1);
1290 if (dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
==
1292 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
1293 ASSERT3U(dsl_dataset_phys(ds
)->ds_prev_snap_txg
, ==,
1294 dsl_dataset_phys(ds
->ds_prev
)->ds_creation_txg
);
1295 dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
= dsobj
;
1296 } else if (next_clones_obj
!= 0) {
1297 dsl_dataset_remove_from_next_clones(ds
->ds_prev
,
1298 dsphys
->ds_next_snap_obj
, tx
);
1299 VERIFY0(zap_add_int(mos
,
1300 next_clones_obj
, dsobj
, tx
));
1305 * If we have a reference-reservation on this dataset, we will
1306 * need to increase the amount of refreservation being charged
1307 * since our unique space is going to zero.
1309 if (ds
->ds_reserved
) {
1311 ASSERT(DS_UNIQUE_IS_ACCURATE(ds
));
1312 delta
= MIN(dsl_dataset_phys(ds
)->ds_unique_bytes
,
1314 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_REFRSRV
,
1318 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1319 dsl_dataset_phys(ds
)->ds_deadlist_obj
=
1320 dsl_deadlist_clone(&ds
->ds_deadlist
, UINT64_MAX
,
1321 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, tx
);
1322 dsl_deadlist_close(&ds
->ds_deadlist
);
1323 dsl_deadlist_open(&ds
->ds_deadlist
, mos
,
1324 dsl_dataset_phys(ds
)->ds_deadlist_obj
);
1325 dsl_deadlist_add_key(&ds
->ds_deadlist
,
1326 dsl_dataset_phys(ds
)->ds_prev_snap_txg
, tx
);
1328 ASSERT3U(dsl_dataset_phys(ds
)->ds_prev_snap_txg
, <, tx
->tx_txg
);
1329 dsl_dataset_phys(ds
)->ds_prev_snap_obj
= dsobj
;
1330 dsl_dataset_phys(ds
)->ds_prev_snap_txg
= crtxg
;
1331 dsl_dataset_phys(ds
)->ds_unique_bytes
= 0;
1332 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_UNIQUE_ACCURATE
)
1333 dsl_dataset_phys(ds
)->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
1335 VERIFY0(zap_add(mos
, dsl_dataset_phys(ds
)->ds_snapnames_zapobj
,
1336 snapname
, 8, 1, &dsobj
, tx
));
1339 dsl_dataset_rele(ds
->ds_prev
, ds
);
1340 VERIFY0(dsl_dataset_hold_obj(dp
,
1341 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, ds
, &ds
->ds_prev
));
1343 dsl_scan_ds_snapshotted(ds
, tx
);
1345 dsl_dir_snap_cmtime_update(ds
->ds_dir
);
1347 spa_history_log_internal_ds(ds
->ds_prev
, "snapshot", tx
, "");
1351 dsl_dataset_snapshot_sync(void *arg
, dmu_tx_t
*tx
)
1353 dsl_dataset_snapshot_arg_t
*ddsa
= arg
;
1354 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1357 for (pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, NULL
);
1358 pair
!= NULL
; pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, pair
)) {
1361 char dsname
[MAXNAMELEN
];
1363 name
= nvpair_name(pair
);
1364 atp
= strchr(name
, '@');
1365 (void) strlcpy(dsname
, name
, atp
- name
+ 1);
1366 VERIFY0(dsl_dataset_hold(dp
, dsname
, FTAG
, &ds
));
1368 dsl_dataset_snapshot_sync_impl(ds
, atp
+ 1, tx
);
1369 if (ddsa
->ddsa_props
!= NULL
) {
1370 dsl_props_set_sync_impl(ds
->ds_prev
,
1371 ZPROP_SRC_LOCAL
, ddsa
->ddsa_props
, tx
);
1373 dsl_dataset_rele(ds
, FTAG
);
1378 * The snapshots must all be in the same pool.
1379 * All-or-nothing: if there are any failures, nothing will be modified.
1382 dsl_dataset_snapshot(nvlist_t
*snaps
, nvlist_t
*props
, nvlist_t
*errors
)
1384 dsl_dataset_snapshot_arg_t ddsa
;
1386 boolean_t needsuspend
;
1390 nvlist_t
*suspended
= NULL
;
1392 pair
= nvlist_next_nvpair(snaps
, NULL
);
1395 firstname
= nvpair_name(pair
);
1397 error
= spa_open(firstname
, &spa
, FTAG
);
1400 needsuspend
= (spa_version(spa
) < SPA_VERSION_FAST_SNAP
);
1401 spa_close(spa
, FTAG
);
1404 suspended
= fnvlist_alloc();
1405 for (pair
= nvlist_next_nvpair(snaps
, NULL
); pair
!= NULL
;
1406 pair
= nvlist_next_nvpair(snaps
, pair
)) {
1407 char fsname
[MAXNAMELEN
];
1408 char *snapname
= nvpair_name(pair
);
1412 atp
= strchr(snapname
, '@');
1414 error
= SET_ERROR(EINVAL
);
1417 (void) strlcpy(fsname
, snapname
, atp
- snapname
+ 1);
1419 error
= zil_suspend(fsname
, &cookie
);
1422 fnvlist_add_uint64(suspended
, fsname
,
1427 ddsa
.ddsa_snaps
= snaps
;
1428 ddsa
.ddsa_props
= props
;
1429 ddsa
.ddsa_errors
= errors
;
1430 ddsa
.ddsa_cr
= CRED();
1433 error
= dsl_sync_task(firstname
, dsl_dataset_snapshot_check
,
1434 dsl_dataset_snapshot_sync
, &ddsa
,
1435 fnvlist_num_pairs(snaps
) * 3, ZFS_SPACE_CHECK_NORMAL
);
1438 if (suspended
!= NULL
) {
1439 for (pair
= nvlist_next_nvpair(suspended
, NULL
); pair
!= NULL
;
1440 pair
= nvlist_next_nvpair(suspended
, pair
)) {
1441 zil_resume((void *)(uintptr_t)
1442 fnvpair_value_uint64(pair
));
1444 fnvlist_free(suspended
);
1449 for (pair
= nvlist_next_nvpair(snaps
, NULL
); pair
!= NULL
;
1450 pair
= nvlist_next_nvpair(snaps
, pair
)) {
1451 char *snapname
= nvpair_name(pair
);
1452 zvol_create_minors(snapname
);
1460 typedef struct dsl_dataset_snapshot_tmp_arg
{
1461 const char *ddsta_fsname
;
1462 const char *ddsta_snapname
;
1463 minor_t ddsta_cleanup_minor
;
1464 const char *ddsta_htag
;
1465 } dsl_dataset_snapshot_tmp_arg_t
;
1468 dsl_dataset_snapshot_tmp_check(void *arg
, dmu_tx_t
*tx
)
1470 dsl_dataset_snapshot_tmp_arg_t
*ddsta
= arg
;
1471 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1475 error
= dsl_dataset_hold(dp
, ddsta
->ddsta_fsname
, FTAG
, &ds
);
1479 /* NULL cred means no limit check for tmp snapshot */
1480 error
= dsl_dataset_snapshot_check_impl(ds
, ddsta
->ddsta_snapname
,
1481 tx
, B_FALSE
, 0, NULL
);
1483 dsl_dataset_rele(ds
, FTAG
);
1487 if (spa_version(dp
->dp_spa
) < SPA_VERSION_USERREFS
) {
1488 dsl_dataset_rele(ds
, FTAG
);
1489 return (SET_ERROR(ENOTSUP
));
1491 error
= dsl_dataset_user_hold_check_one(NULL
, ddsta
->ddsta_htag
,
1494 dsl_dataset_rele(ds
, FTAG
);
1498 dsl_dataset_rele(ds
, FTAG
);
1503 dsl_dataset_snapshot_tmp_sync(void *arg
, dmu_tx_t
*tx
)
1505 dsl_dataset_snapshot_tmp_arg_t
*ddsta
= arg
;
1506 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1509 VERIFY0(dsl_dataset_hold(dp
, ddsta
->ddsta_fsname
, FTAG
, &ds
));
1511 dsl_dataset_snapshot_sync_impl(ds
, ddsta
->ddsta_snapname
, tx
);
1512 dsl_dataset_user_hold_sync_one(ds
->ds_prev
, ddsta
->ddsta_htag
,
1513 ddsta
->ddsta_cleanup_minor
, gethrestime_sec(), tx
);
1514 dsl_destroy_snapshot_sync_impl(ds
->ds_prev
, B_TRUE
, tx
);
1516 dsl_dataset_rele(ds
, FTAG
);
1520 dsl_dataset_snapshot_tmp(const char *fsname
, const char *snapname
,
1521 minor_t cleanup_minor
, const char *htag
)
1523 dsl_dataset_snapshot_tmp_arg_t ddsta
;
1526 boolean_t needsuspend
;
1529 ddsta
.ddsta_fsname
= fsname
;
1530 ddsta
.ddsta_snapname
= snapname
;
1531 ddsta
.ddsta_cleanup_minor
= cleanup_minor
;
1532 ddsta
.ddsta_htag
= htag
;
1534 error
= spa_open(fsname
, &spa
, FTAG
);
1537 needsuspend
= (spa_version(spa
) < SPA_VERSION_FAST_SNAP
);
1538 spa_close(spa
, FTAG
);
1541 error
= zil_suspend(fsname
, &cookie
);
1546 error
= dsl_sync_task(fsname
, dsl_dataset_snapshot_tmp_check
,
1547 dsl_dataset_snapshot_tmp_sync
, &ddsta
, 3, ZFS_SPACE_CHECK_RESERVED
);
1556 dsl_dataset_sync(dsl_dataset_t
*ds
, zio_t
*zio
, dmu_tx_t
*tx
)
1558 ASSERT(dmu_tx_is_syncing(tx
));
1559 ASSERT(ds
->ds_objset
!= NULL
);
1560 ASSERT(dsl_dataset_phys(ds
)->ds_next_snap_obj
== 0);
1563 * in case we had to change ds_fsid_guid when we opened it,
1566 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1567 dsl_dataset_phys(ds
)->ds_fsid_guid
= ds
->ds_fsid_guid
;
1569 dmu_objset_sync(ds
->ds_objset
, zio
, tx
);
1571 if (ds
->ds_need_large_blocks
&& !ds
->ds_large_blocks
) {
1572 dsl_dataset_activate_large_blocks_sync_impl(ds
->ds_object
, tx
);
1573 ds
->ds_large_blocks
= B_TRUE
;
1578 get_clones_stat(dsl_dataset_t
*ds
, nvlist_t
*nv
)
1581 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
1584 nvlist_t
*propval
= fnvlist_alloc();
1585 nvlist_t
*val
= fnvlist_alloc();
1587 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
1590 * There may be missing entries in ds_next_clones_obj
1591 * due to a bug in a previous version of the code.
1592 * Only trust it if it has the right number of entries.
1594 if (dsl_dataset_phys(ds
)->ds_next_clones_obj
!= 0) {
1595 VERIFY0(zap_count(mos
, dsl_dataset_phys(ds
)->ds_next_clones_obj
,
1598 if (count
!= dsl_dataset_phys(ds
)->ds_num_children
- 1)
1600 for (zap_cursor_init(&zc
, mos
,
1601 dsl_dataset_phys(ds
)->ds_next_clones_obj
);
1602 zap_cursor_retrieve(&zc
, &za
) == 0;
1603 zap_cursor_advance(&zc
)) {
1604 dsl_dataset_t
*clone
;
1605 char buf
[ZFS_MAXNAMELEN
];
1606 VERIFY0(dsl_dataset_hold_obj(ds
->ds_dir
->dd_pool
,
1607 za
.za_first_integer
, FTAG
, &clone
));
1608 dsl_dir_name(clone
->ds_dir
, buf
);
1609 fnvlist_add_boolean(val
, buf
);
1610 dsl_dataset_rele(clone
, FTAG
);
1612 zap_cursor_fini(&zc
);
1613 fnvlist_add_nvlist(propval
, ZPROP_VALUE
, val
);
1614 fnvlist_add_nvlist(nv
, zfs_prop_to_name(ZFS_PROP_CLONES
), propval
);
1617 nvlist_free(propval
);
1621 dsl_dataset_stats(dsl_dataset_t
*ds
, nvlist_t
*nv
)
1623 uint64_t refd
, avail
, uobjs
, aobjs
, ratio
;
1624 ASSERTV(dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
);
1626 ASSERT(dsl_pool_config_held(dp
));
1628 ratio
= dsl_dataset_phys(ds
)->ds_compressed_bytes
== 0 ? 100 :
1629 (dsl_dataset_phys(ds
)->ds_uncompressed_bytes
* 100 /
1630 dsl_dataset_phys(ds
)->ds_compressed_bytes
);
1632 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFRATIO
, ratio
);
1633 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_LOGICALREFERENCED
,
1634 dsl_dataset_phys(ds
)->ds_uncompressed_bytes
);
1636 if (ds
->ds_is_snapshot
) {
1637 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_COMPRESSRATIO
, ratio
);
1638 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USED
,
1639 dsl_dataset_phys(ds
)->ds_unique_bytes
);
1640 get_clones_stat(ds
, nv
);
1642 dsl_dir_stats(ds
->ds_dir
, nv
);
1645 dsl_dataset_space(ds
, &refd
, &avail
, &uobjs
, &aobjs
);
1646 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_AVAILABLE
, avail
);
1647 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFERENCED
, refd
);
1649 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_CREATION
,
1650 dsl_dataset_phys(ds
)->ds_creation_time
);
1651 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_CREATETXG
,
1652 dsl_dataset_phys(ds
)->ds_creation_txg
);
1653 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFQUOTA
,
1655 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFRESERVATION
,
1657 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_GUID
,
1658 dsl_dataset_phys(ds
)->ds_guid
);
1659 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_UNIQUE
,
1660 dsl_dataset_phys(ds
)->ds_unique_bytes
);
1661 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_OBJSETID
,
1663 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USERREFS
,
1665 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_DEFER_DESTROY
,
1666 DS_IS_DEFER_DESTROY(ds
) ? 1 : 0);
1668 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0) {
1669 uint64_t written
, comp
, uncomp
;
1670 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1671 dsl_dataset_t
*prev
;
1674 err
= dsl_dataset_hold_obj(dp
,
1675 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, FTAG
, &prev
);
1677 err
= dsl_dataset_space_written(prev
, ds
, &written
,
1679 dsl_dataset_rele(prev
, FTAG
);
1681 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_WRITTEN
,
1690 dsl_dataset_fast_stat(dsl_dataset_t
*ds
, dmu_objset_stats_t
*stat
)
1692 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1693 ASSERT(dsl_pool_config_held(dp
));
1695 stat
->dds_creation_txg
= dsl_dataset_phys(ds
)->ds_creation_txg
;
1696 stat
->dds_inconsistent
=
1697 dsl_dataset_phys(ds
)->ds_flags
& DS_FLAG_INCONSISTENT
;
1698 stat
->dds_guid
= dsl_dataset_phys(ds
)->ds_guid
;
1699 stat
->dds_origin
[0] = '\0';
1700 if (ds
->ds_is_snapshot
) {
1701 stat
->dds_is_snapshot
= B_TRUE
;
1702 stat
->dds_num_clones
=
1703 dsl_dataset_phys(ds
)->ds_num_children
- 1;
1705 stat
->dds_is_snapshot
= B_FALSE
;
1706 stat
->dds_num_clones
= 0;
1708 if (dsl_dir_is_clone(ds
->ds_dir
)) {
1711 VERIFY0(dsl_dataset_hold_obj(dp
,
1712 dsl_dir_phys(ds
->ds_dir
)->dd_origin_obj
,
1714 dsl_dataset_name(ods
, stat
->dds_origin
);
1715 dsl_dataset_rele(ods
, FTAG
);
1721 dsl_dataset_fsid_guid(dsl_dataset_t
*ds
)
1723 return (ds
->ds_fsid_guid
);
1727 dsl_dataset_space(dsl_dataset_t
*ds
,
1728 uint64_t *refdbytesp
, uint64_t *availbytesp
,
1729 uint64_t *usedobjsp
, uint64_t *availobjsp
)
1731 *refdbytesp
= dsl_dataset_phys(ds
)->ds_referenced_bytes
;
1732 *availbytesp
= dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
);
1733 if (ds
->ds_reserved
> dsl_dataset_phys(ds
)->ds_unique_bytes
)
1735 ds
->ds_reserved
- dsl_dataset_phys(ds
)->ds_unique_bytes
;
1736 if (ds
->ds_quota
!= 0) {
1738 * Adjust available bytes according to refquota
1740 if (*refdbytesp
< ds
->ds_quota
)
1741 *availbytesp
= MIN(*availbytesp
,
1742 ds
->ds_quota
- *refdbytesp
);
1746 *usedobjsp
= BP_GET_FILL(&dsl_dataset_phys(ds
)->ds_bp
);
1747 *availobjsp
= DN_MAX_OBJECT
- *usedobjsp
;
1751 dsl_dataset_modified_since_snap(dsl_dataset_t
*ds
, dsl_dataset_t
*snap
)
1753 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
1756 if (dsl_dataset_phys(ds
)->ds_bp
.blk_birth
>
1757 dsl_dataset_phys(snap
)->ds_creation_txg
) {
1758 objset_t
*os
, *os_snap
;
1760 * It may be that only the ZIL differs, because it was
1761 * reset in the head. Don't count that as being
1764 if (dmu_objset_from_ds(ds
, &os
) != 0)
1766 if (dmu_objset_from_ds(snap
, &os_snap
) != 0)
1768 return (bcmp(&os
->os_phys
->os_meta_dnode
,
1769 &os_snap
->os_phys
->os_meta_dnode
,
1770 sizeof (os
->os_phys
->os_meta_dnode
)) != 0);
1775 typedef struct dsl_dataset_rename_snapshot_arg
{
1776 const char *ddrsa_fsname
;
1777 const char *ddrsa_oldsnapname
;
1778 const char *ddrsa_newsnapname
;
1779 boolean_t ddrsa_recursive
;
1781 } dsl_dataset_rename_snapshot_arg_t
;
1785 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t
*dp
,
1786 dsl_dataset_t
*hds
, void *arg
)
1788 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
1792 error
= dsl_dataset_snap_lookup(hds
, ddrsa
->ddrsa_oldsnapname
, &val
);
1794 /* ignore nonexistent snapshots */
1795 return (error
== ENOENT
? 0 : error
);
1798 /* new name should not exist */
1799 error
= dsl_dataset_snap_lookup(hds
, ddrsa
->ddrsa_newsnapname
, &val
);
1801 error
= SET_ERROR(EEXIST
);
1802 else if (error
== ENOENT
)
1805 /* dataset name + 1 for the "@" + the new snapshot name must fit */
1806 if (dsl_dir_namelen(hds
->ds_dir
) + 1 +
1807 strlen(ddrsa
->ddrsa_newsnapname
) >= MAXNAMELEN
)
1808 error
= SET_ERROR(ENAMETOOLONG
);
1814 dsl_dataset_rename_snapshot_check(void *arg
, dmu_tx_t
*tx
)
1816 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
1817 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1821 error
= dsl_dataset_hold(dp
, ddrsa
->ddrsa_fsname
, FTAG
, &hds
);
1825 if (ddrsa
->ddrsa_recursive
) {
1826 error
= dmu_objset_find_dp(dp
, hds
->ds_dir
->dd_object
,
1827 dsl_dataset_rename_snapshot_check_impl
, ddrsa
,
1830 error
= dsl_dataset_rename_snapshot_check_impl(dp
, hds
, ddrsa
);
1832 dsl_dataset_rele(hds
, FTAG
);
1837 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t
*dp
,
1838 dsl_dataset_t
*hds
, void *arg
)
1840 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
1843 dmu_tx_t
*tx
= ddrsa
->ddrsa_tx
;
1846 error
= dsl_dataset_snap_lookup(hds
, ddrsa
->ddrsa_oldsnapname
, &val
);
1847 ASSERT(error
== 0 || error
== ENOENT
);
1848 if (error
== ENOENT
) {
1849 /* ignore nonexistent snapshots */
1853 VERIFY0(dsl_dataset_hold_obj(dp
, val
, FTAG
, &ds
));
1855 /* log before we change the name */
1856 spa_history_log_internal_ds(ds
, "rename", tx
,
1857 "-> @%s", ddrsa
->ddrsa_newsnapname
);
1859 VERIFY0(dsl_dataset_snap_remove(hds
, ddrsa
->ddrsa_oldsnapname
, tx
,
1861 mutex_enter(&ds
->ds_lock
);
1862 (void) strcpy(ds
->ds_snapname
, ddrsa
->ddrsa_newsnapname
);
1863 mutex_exit(&ds
->ds_lock
);
1864 VERIFY0(zap_add(dp
->dp_meta_objset
,
1865 dsl_dataset_phys(hds
)->ds_snapnames_zapobj
,
1866 ds
->ds_snapname
, 8, 1, &ds
->ds_object
, tx
));
1868 dsl_dataset_rele(ds
, FTAG
);
1873 dsl_dataset_rename_snapshot_sync(void *arg
, dmu_tx_t
*tx
)
1875 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
1876 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1879 VERIFY0(dsl_dataset_hold(dp
, ddrsa
->ddrsa_fsname
, FTAG
, &hds
));
1880 ddrsa
->ddrsa_tx
= tx
;
1881 if (ddrsa
->ddrsa_recursive
) {
1882 VERIFY0(dmu_objset_find_dp(dp
, hds
->ds_dir
->dd_object
,
1883 dsl_dataset_rename_snapshot_sync_impl
, ddrsa
,
1886 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp
, hds
, ddrsa
));
1888 dsl_dataset_rele(hds
, FTAG
);
1892 dsl_dataset_rename_snapshot(const char *fsname
,
1893 const char *oldsnapname
, const char *newsnapname
, boolean_t recursive
)
1896 char *oldname
, *newname
;
1900 dsl_dataset_rename_snapshot_arg_t ddrsa
;
1902 ddrsa
.ddrsa_fsname
= fsname
;
1903 ddrsa
.ddrsa_oldsnapname
= oldsnapname
;
1904 ddrsa
.ddrsa_newsnapname
= newsnapname
;
1905 ddrsa
.ddrsa_recursive
= recursive
;
1907 error
= dsl_sync_task(fsname
, dsl_dataset_rename_snapshot_check
,
1908 dsl_dataset_rename_snapshot_sync
, &ddrsa
,
1909 1, ZFS_SPACE_CHECK_RESERVED
);
1912 return (SET_ERROR(error
));
1915 oldname
= kmem_asprintf("%s@%s", fsname
, oldsnapname
);
1916 newname
= kmem_asprintf("%s@%s", fsname
, newsnapname
);
1917 zvol_rename_minors(oldname
, newname
);
1926 * If we're doing an ownership handoff, we need to make sure that there is
1927 * only one long hold on the dataset. We're not allowed to change anything here
1928 * so we don't permanently release the long hold or regular hold here. We want
1929 * to do this only when syncing to avoid the dataset unexpectedly going away
1930 * when we release the long hold.
1933 dsl_dataset_handoff_check(dsl_dataset_t
*ds
, void *owner
, dmu_tx_t
*tx
)
1937 if (!dmu_tx_is_syncing(tx
))
1940 if (owner
!= NULL
) {
1941 VERIFY3P(ds
->ds_owner
, ==, owner
);
1942 dsl_dataset_long_rele(ds
, owner
);
1945 held
= dsl_dataset_long_held(ds
);
1948 dsl_dataset_long_hold(ds
, owner
);
1951 return (SET_ERROR(EBUSY
));
1956 typedef struct dsl_dataset_rollback_arg
{
1957 const char *ddra_fsname
;
1959 nvlist_t
*ddra_result
;
1960 } dsl_dataset_rollback_arg_t
;
1963 dsl_dataset_rollback_check(void *arg
, dmu_tx_t
*tx
)
1965 dsl_dataset_rollback_arg_t
*ddra
= arg
;
1966 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1968 int64_t unused_refres_delta
;
1971 nvlist_t
*proprequest
, *bookmarks
;
1973 error
= dsl_dataset_hold(dp
, ddra
->ddra_fsname
, FTAG
, &ds
);
1977 /* must not be a snapshot */
1978 if (ds
->ds_is_snapshot
) {
1979 dsl_dataset_rele(ds
, FTAG
);
1980 return (SET_ERROR(EINVAL
));
1983 /* must have a most recent snapshot */
1984 if (dsl_dataset_phys(ds
)->ds_prev_snap_txg
< TXG_INITIAL
) {
1985 dsl_dataset_rele(ds
, FTAG
);
1986 return (SET_ERROR(EINVAL
));
1989 /* must not have any bookmarks after the most recent snapshot */
1990 proprequest
= fnvlist_alloc();
1991 fnvlist_add_boolean(proprequest
, zfs_prop_to_name(ZFS_PROP_CREATETXG
));
1992 bookmarks
= fnvlist_alloc();
1993 error
= dsl_get_bookmarks_impl(ds
, proprequest
, bookmarks
);
1994 fnvlist_free(proprequest
);
1997 for (pair
= nvlist_next_nvpair(bookmarks
, NULL
);
1998 pair
!= NULL
; pair
= nvlist_next_nvpair(bookmarks
, pair
)) {
2000 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair
),
2001 zfs_prop_to_name(ZFS_PROP_CREATETXG
));
2002 uint64_t createtxg
= fnvlist_lookup_uint64(valuenv
, "value");
2003 if (createtxg
> dsl_dataset_phys(ds
)->ds_prev_snap_txg
) {
2004 fnvlist_free(bookmarks
);
2005 dsl_dataset_rele(ds
, FTAG
);
2006 return (SET_ERROR(EEXIST
));
2009 fnvlist_free(bookmarks
);
2011 error
= dsl_dataset_handoff_check(ds
, ddra
->ddra_owner
, tx
);
2013 dsl_dataset_rele(ds
, FTAG
);
2018 * Check if the snap we are rolling back to uses more than
2021 if (ds
->ds_quota
!= 0 &&
2022 dsl_dataset_phys(ds
->ds_prev
)->ds_referenced_bytes
> ds
->ds_quota
) {
2023 dsl_dataset_rele(ds
, FTAG
);
2024 return (SET_ERROR(EDQUOT
));
2028 * When we do the clone swap, we will temporarily use more space
2029 * due to the refreservation (the head will no longer have any
2030 * unique space, so the entire amount of the refreservation will need
2031 * to be free). We will immediately destroy the clone, freeing
2032 * this space, but the freeing happens over many txg's.
2034 unused_refres_delta
= (int64_t)MIN(ds
->ds_reserved
,
2035 dsl_dataset_phys(ds
)->ds_unique_bytes
);
2037 if (unused_refres_delta
> 0 &&
2038 unused_refres_delta
>
2039 dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
)) {
2040 dsl_dataset_rele(ds
, FTAG
);
2041 return (SET_ERROR(ENOSPC
));
2044 dsl_dataset_rele(ds
, FTAG
);
2049 dsl_dataset_rollback_sync(void *arg
, dmu_tx_t
*tx
)
2051 dsl_dataset_rollback_arg_t
*ddra
= arg
;
2052 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2053 dsl_dataset_t
*ds
, *clone
;
2055 char namebuf
[ZFS_MAXNAMELEN
];
2057 VERIFY0(dsl_dataset_hold(dp
, ddra
->ddra_fsname
, FTAG
, &ds
));
2059 dsl_dataset_name(ds
->ds_prev
, namebuf
);
2060 fnvlist_add_string(ddra
->ddra_result
, "target", namebuf
);
2062 cloneobj
= dsl_dataset_create_sync(ds
->ds_dir
, "%rollback",
2063 ds
->ds_prev
, DS_CREATE_FLAG_NODIRTY
, kcred
, tx
);
2065 VERIFY0(dsl_dataset_hold_obj(dp
, cloneobj
, FTAG
, &clone
));
2067 dsl_dataset_clone_swap_sync_impl(clone
, ds
, tx
);
2068 dsl_dataset_zero_zil(ds
, tx
);
2070 dsl_destroy_head_sync_impl(clone
, tx
);
2072 dsl_dataset_rele(clone
, FTAG
);
2073 dsl_dataset_rele(ds
, FTAG
);
2077 * Rolls back the given filesystem or volume to the most recent snapshot.
2078 * The name of the most recent snapshot will be returned under key "target"
2079 * in the result nvlist.
2082 * - The existing dataset MUST be owned by the specified owner at entry
2083 * - Upon return, dataset will still be held by the same owner, whether we
2086 * This mode is required any time the existing filesystem is mounted. See
2087 * notes above zfs_suspend_fs() for further details.
2090 dsl_dataset_rollback(const char *fsname
, void *owner
, nvlist_t
*result
)
2092 dsl_dataset_rollback_arg_t ddra
;
2094 ddra
.ddra_fsname
= fsname
;
2095 ddra
.ddra_owner
= owner
;
2096 ddra
.ddra_result
= result
;
2098 return (dsl_sync_task(fsname
, dsl_dataset_rollback_check
,
2099 dsl_dataset_rollback_sync
, &ddra
,
2100 1, ZFS_SPACE_CHECK_RESERVED
));
2103 struct promotenode
{
2108 typedef struct dsl_dataset_promote_arg
{
2109 const char *ddpa_clonename
;
2110 dsl_dataset_t
*ddpa_clone
;
2111 list_t shared_snaps
, origin_snaps
, clone_snaps
;
2112 dsl_dataset_t
*origin_origin
; /* origin of the origin */
2113 uint64_t used
, comp
, uncomp
, unique
, cloneusedsnap
, originusedsnap
;
2116 } dsl_dataset_promote_arg_t
;
2118 static int snaplist_space(list_t
*l
, uint64_t mintxg
, uint64_t *spacep
);
2119 static int promote_hold(dsl_dataset_promote_arg_t
*ddpa
, dsl_pool_t
*dp
,
2121 static void promote_rele(dsl_dataset_promote_arg_t
*ddpa
, void *tag
);
2124 dsl_dataset_promote_check(void *arg
, dmu_tx_t
*tx
)
2126 dsl_dataset_promote_arg_t
*ddpa
= arg
;
2127 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2129 struct promotenode
*snap
;
2130 dsl_dataset_t
*origin_ds
;
2134 size_t max_snap_len
;
2136 err
= promote_hold(ddpa
, dp
, FTAG
);
2140 hds
= ddpa
->ddpa_clone
;
2141 max_snap_len
= MAXNAMELEN
- strlen(ddpa
->ddpa_clonename
) - 1;
2143 if (dsl_dataset_phys(hds
)->ds_flags
& DS_FLAG_NOPROMOTE
) {
2144 promote_rele(ddpa
, FTAG
);
2145 return (SET_ERROR(EXDEV
));
2149 * Compute and check the amount of space to transfer. Since this is
2150 * so expensive, don't do the preliminary check.
2152 if (!dmu_tx_is_syncing(tx
)) {
2153 promote_rele(ddpa
, FTAG
);
2157 snap
= list_head(&ddpa
->shared_snaps
);
2158 origin_ds
= snap
->ds
;
2160 /* compute origin's new unique space */
2161 snap
= list_tail(&ddpa
->clone_snaps
);
2162 ASSERT3U(dsl_dataset_phys(snap
->ds
)->ds_prev_snap_obj
, ==,
2163 origin_ds
->ds_object
);
2164 dsl_deadlist_space_range(&snap
->ds
->ds_deadlist
,
2165 dsl_dataset_phys(origin_ds
)->ds_prev_snap_txg
, UINT64_MAX
,
2166 &ddpa
->unique
, &unused
, &unused
);
2169 * Walk the snapshots that we are moving
2171 * Compute space to transfer. Consider the incremental changes
2172 * to used by each snapshot:
2173 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2174 * So each snapshot gave birth to:
2175 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2176 * So a sequence would look like:
2177 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2178 * Which simplifies to:
2179 * uN + kN + kN-1 + ... + k1 + k0
2180 * Note however, if we stop before we reach the ORIGIN we get:
2181 * uN + kN + kN-1 + ... + kM - uM-1
2184 ddpa
->used
= dsl_dataset_phys(origin_ds
)->ds_referenced_bytes
;
2185 ddpa
->comp
= dsl_dataset_phys(origin_ds
)->ds_compressed_bytes
;
2186 ddpa
->uncomp
= dsl_dataset_phys(origin_ds
)->ds_uncompressed_bytes
;
2187 for (snap
= list_head(&ddpa
->shared_snaps
); snap
;
2188 snap
= list_next(&ddpa
->shared_snaps
, snap
)) {
2189 uint64_t val
, dlused
, dlcomp
, dluncomp
;
2190 dsl_dataset_t
*ds
= snap
->ds
;
2195 * If there are long holds, we won't be able to evict
2198 if (dsl_dataset_long_held(ds
)) {
2199 err
= SET_ERROR(EBUSY
);
2203 /* Check that the snapshot name does not conflict */
2204 VERIFY0(dsl_dataset_get_snapname(ds
));
2205 if (strlen(ds
->ds_snapname
) >= max_snap_len
) {
2206 err
= SET_ERROR(ENAMETOOLONG
);
2209 err
= dsl_dataset_snap_lookup(hds
, ds
->ds_snapname
, &val
);
2211 (void) strcpy(ddpa
->err_ds
, snap
->ds
->ds_snapname
);
2212 err
= SET_ERROR(EEXIST
);
2218 /* The very first snapshot does not have a deadlist */
2219 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
== 0)
2222 dsl_deadlist_space(&ds
->ds_deadlist
,
2223 &dlused
, &dlcomp
, &dluncomp
);
2224 ddpa
->used
+= dlused
;
2225 ddpa
->comp
+= dlcomp
;
2226 ddpa
->uncomp
+= dluncomp
;
2230 * If we are a clone of a clone then we never reached ORIGIN,
2231 * so we need to subtract out the clone origin's used space.
2233 if (ddpa
->origin_origin
) {
2235 dsl_dataset_phys(ddpa
->origin_origin
)->ds_referenced_bytes
;
2237 dsl_dataset_phys(ddpa
->origin_origin
)->ds_compressed_bytes
;
2239 dsl_dataset_phys(ddpa
->origin_origin
)->
2240 ds_uncompressed_bytes
;
2243 /* Check that there is enough space and limit headroom here */
2244 err
= dsl_dir_transfer_possible(origin_ds
->ds_dir
, hds
->ds_dir
,
2245 0, ss_mv_cnt
, ddpa
->used
, ddpa
->cr
);
2250 * Compute the amounts of space that will be used by snapshots
2251 * after the promotion (for both origin and clone). For each,
2252 * it is the amount of space that will be on all of their
2253 * deadlists (that was not born before their new origin).
2255 if (dsl_dir_phys(hds
->ds_dir
)->dd_flags
& DD_FLAG_USED_BREAKDOWN
) {
2259 * Note, typically this will not be a clone of a clone,
2260 * so dd_origin_txg will be < TXG_INITIAL, so
2261 * these snaplist_space() -> dsl_deadlist_space_range()
2262 * calls will be fast because they do not have to
2263 * iterate over all bps.
2265 snap
= list_head(&ddpa
->origin_snaps
);
2266 err
= snaplist_space(&ddpa
->shared_snaps
,
2267 snap
->ds
->ds_dir
->dd_origin_txg
, &ddpa
->cloneusedsnap
);
2271 err
= snaplist_space(&ddpa
->clone_snaps
,
2272 snap
->ds
->ds_dir
->dd_origin_txg
, &space
);
2275 ddpa
->cloneusedsnap
+= space
;
2277 if (dsl_dir_phys(origin_ds
->ds_dir
)->dd_flags
&
2278 DD_FLAG_USED_BREAKDOWN
) {
2279 err
= snaplist_space(&ddpa
->origin_snaps
,
2280 dsl_dataset_phys(origin_ds
)->ds_creation_txg
,
2281 &ddpa
->originusedsnap
);
2287 promote_rele(ddpa
, FTAG
);
2292 dsl_dataset_promote_sync(void *arg
, dmu_tx_t
*tx
)
2294 dsl_dataset_promote_arg_t
*ddpa
= arg
;
2295 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2297 struct promotenode
*snap
;
2298 dsl_dataset_t
*origin_ds
;
2299 dsl_dataset_t
*origin_head
;
2301 dsl_dir_t
*odd
= NULL
;
2302 uint64_t oldnext_obj
;
2305 VERIFY0(promote_hold(ddpa
, dp
, FTAG
));
2306 hds
= ddpa
->ddpa_clone
;
2308 ASSERT0(dsl_dataset_phys(hds
)->ds_flags
& DS_FLAG_NOPROMOTE
);
2310 snap
= list_head(&ddpa
->shared_snaps
);
2311 origin_ds
= snap
->ds
;
2314 snap
= list_head(&ddpa
->origin_snaps
);
2315 origin_head
= snap
->ds
;
2318 * We need to explicitly open odd, since origin_ds's dd will be
2321 VERIFY0(dsl_dir_hold_obj(dp
, origin_ds
->ds_dir
->dd_object
,
2324 /* change origin's next snap */
2325 dmu_buf_will_dirty(origin_ds
->ds_dbuf
, tx
);
2326 oldnext_obj
= dsl_dataset_phys(origin_ds
)->ds_next_snap_obj
;
2327 snap
= list_tail(&ddpa
->clone_snaps
);
2328 ASSERT3U(dsl_dataset_phys(snap
->ds
)->ds_prev_snap_obj
, ==,
2329 origin_ds
->ds_object
);
2330 dsl_dataset_phys(origin_ds
)->ds_next_snap_obj
= snap
->ds
->ds_object
;
2332 /* change the origin's next clone */
2333 if (dsl_dataset_phys(origin_ds
)->ds_next_clones_obj
) {
2334 dsl_dataset_remove_from_next_clones(origin_ds
,
2335 snap
->ds
->ds_object
, tx
);
2336 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2337 dsl_dataset_phys(origin_ds
)->ds_next_clones_obj
,
2342 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
2343 ASSERT3U(dsl_dir_phys(dd
)->dd_origin_obj
, ==, origin_ds
->ds_object
);
2344 dsl_dir_phys(dd
)->dd_origin_obj
= dsl_dir_phys(odd
)->dd_origin_obj
;
2345 dd
->dd_origin_txg
= origin_head
->ds_dir
->dd_origin_txg
;
2346 dmu_buf_will_dirty(odd
->dd_dbuf
, tx
);
2347 dsl_dir_phys(odd
)->dd_origin_obj
= origin_ds
->ds_object
;
2348 origin_head
->ds_dir
->dd_origin_txg
=
2349 dsl_dataset_phys(origin_ds
)->ds_creation_txg
;
2351 /* change dd_clone entries */
2352 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
2353 VERIFY0(zap_remove_int(dp
->dp_meta_objset
,
2354 dsl_dir_phys(odd
)->dd_clones
, hds
->ds_object
, tx
));
2355 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2356 dsl_dir_phys(ddpa
->origin_origin
->ds_dir
)->dd_clones
,
2357 hds
->ds_object
, tx
));
2359 VERIFY0(zap_remove_int(dp
->dp_meta_objset
,
2360 dsl_dir_phys(ddpa
->origin_origin
->ds_dir
)->dd_clones
,
2361 origin_head
->ds_object
, tx
));
2362 if (dsl_dir_phys(dd
)->dd_clones
== 0) {
2363 dsl_dir_phys(dd
)->dd_clones
=
2364 zap_create(dp
->dp_meta_objset
, DMU_OT_DSL_CLONES
,
2365 DMU_OT_NONE
, 0, tx
);
2367 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2368 dsl_dir_phys(dd
)->dd_clones
, origin_head
->ds_object
, tx
));
2371 /* move snapshots to this dir */
2372 for (snap
= list_head(&ddpa
->shared_snaps
); snap
;
2373 snap
= list_next(&ddpa
->shared_snaps
, snap
)) {
2374 dsl_dataset_t
*ds
= snap
->ds
;
2377 * Property callbacks are registered to a particular
2378 * dsl_dir. Since ours is changing, evict the objset
2379 * so that they will be unregistered from the old dsl_dir.
2381 if (ds
->ds_objset
) {
2382 dmu_objset_evict(ds
->ds_objset
);
2383 ds
->ds_objset
= NULL
;
2386 /* move snap name entry */
2387 VERIFY0(dsl_dataset_get_snapname(ds
));
2388 VERIFY0(dsl_dataset_snap_remove(origin_head
,
2389 ds
->ds_snapname
, tx
, B_TRUE
));
2390 VERIFY0(zap_add(dp
->dp_meta_objset
,
2391 dsl_dataset_phys(hds
)->ds_snapnames_zapobj
, ds
->ds_snapname
,
2392 8, 1, &ds
->ds_object
, tx
));
2393 dsl_fs_ss_count_adjust(hds
->ds_dir
, 1,
2394 DD_FIELD_SNAPSHOT_COUNT
, tx
);
2396 /* change containing dsl_dir */
2397 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
2398 ASSERT3U(dsl_dataset_phys(ds
)->ds_dir_obj
, ==, odd
->dd_object
);
2399 dsl_dataset_phys(ds
)->ds_dir_obj
= dd
->dd_object
;
2400 ASSERT3P(ds
->ds_dir
, ==, odd
);
2401 dsl_dir_rele(ds
->ds_dir
, ds
);
2402 VERIFY0(dsl_dir_hold_obj(dp
, dd
->dd_object
,
2403 NULL
, ds
, &ds
->ds_dir
));
2405 /* move any clone references */
2406 if (dsl_dataset_phys(ds
)->ds_next_clones_obj
&&
2407 spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
2411 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
2412 dsl_dataset_phys(ds
)->ds_next_clones_obj
);
2413 zap_cursor_retrieve(&zc
, &za
) == 0;
2414 zap_cursor_advance(&zc
)) {
2415 dsl_dataset_t
*cnds
;
2418 if (za
.za_first_integer
== oldnext_obj
) {
2420 * We've already moved the
2421 * origin's reference.
2426 VERIFY0(dsl_dataset_hold_obj(dp
,
2427 za
.za_first_integer
, FTAG
, &cnds
));
2428 o
= dsl_dir_phys(cnds
->ds_dir
)->
2429 dd_head_dataset_obj
;
2431 VERIFY0(zap_remove_int(dp
->dp_meta_objset
,
2432 dsl_dir_phys(odd
)->dd_clones
, o
, tx
));
2433 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2434 dsl_dir_phys(dd
)->dd_clones
, o
, tx
));
2435 dsl_dataset_rele(cnds
, FTAG
);
2437 zap_cursor_fini(&zc
);
2440 ASSERT(!dsl_prop_hascb(ds
));
2444 * Change space accounting.
2445 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2446 * both be valid, or both be 0 (resulting in delta == 0). This
2447 * is true for each of {clone,origin} independently.
2450 delta
= ddpa
->cloneusedsnap
-
2451 dsl_dir_phys(dd
)->dd_used_breakdown
[DD_USED_SNAP
];
2452 ASSERT3S(delta
, >=, 0);
2453 ASSERT3U(ddpa
->used
, >=, delta
);
2454 dsl_dir_diduse_space(dd
, DD_USED_SNAP
, delta
, 0, 0, tx
);
2455 dsl_dir_diduse_space(dd
, DD_USED_HEAD
,
2456 ddpa
->used
- delta
, ddpa
->comp
, ddpa
->uncomp
, tx
);
2458 delta
= ddpa
->originusedsnap
-
2459 dsl_dir_phys(odd
)->dd_used_breakdown
[DD_USED_SNAP
];
2460 ASSERT3S(delta
, <=, 0);
2461 ASSERT3U(ddpa
->used
, >=, -delta
);
2462 dsl_dir_diduse_space(odd
, DD_USED_SNAP
, delta
, 0, 0, tx
);
2463 dsl_dir_diduse_space(odd
, DD_USED_HEAD
,
2464 -ddpa
->used
- delta
, -ddpa
->comp
, -ddpa
->uncomp
, tx
);
2466 dsl_dataset_phys(origin_ds
)->ds_unique_bytes
= ddpa
->unique
;
2468 /* log history record */
2469 spa_history_log_internal_ds(hds
, "promote", tx
, "");
2471 dsl_dir_rele(odd
, FTAG
);
2472 promote_rele(ddpa
, FTAG
);
2476 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2477 * (exclusive) and last_obj (inclusive). The list will be in reverse
2478 * order (last_obj will be the list_head()). If first_obj == 0, do all
2479 * snapshots back to this dataset's origin.
2482 snaplist_make(dsl_pool_t
*dp
,
2483 uint64_t first_obj
, uint64_t last_obj
, list_t
*l
, void *tag
)
2485 uint64_t obj
= last_obj
;
2487 list_create(l
, sizeof (struct promotenode
),
2488 offsetof(struct promotenode
, link
));
2490 while (obj
!= first_obj
) {
2492 struct promotenode
*snap
;
2495 err
= dsl_dataset_hold_obj(dp
, obj
, tag
, &ds
);
2496 ASSERT(err
!= ENOENT
);
2501 first_obj
= dsl_dir_phys(ds
->ds_dir
)->dd_origin_obj
;
2503 snap
= kmem_alloc(sizeof (*snap
), KM_SLEEP
);
2505 list_insert_tail(l
, snap
);
2506 obj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
2513 snaplist_space(list_t
*l
, uint64_t mintxg
, uint64_t *spacep
)
2515 struct promotenode
*snap
;
2518 for (snap
= list_head(l
); snap
; snap
= list_next(l
, snap
)) {
2519 uint64_t used
, comp
, uncomp
;
2520 dsl_deadlist_space_range(&snap
->ds
->ds_deadlist
,
2521 mintxg
, UINT64_MAX
, &used
, &comp
, &uncomp
);
2528 snaplist_destroy(list_t
*l
, void *tag
)
2530 struct promotenode
*snap
;
2532 if (l
== NULL
|| !list_link_active(&l
->list_head
))
2535 while ((snap
= list_tail(l
)) != NULL
) {
2536 list_remove(l
, snap
);
2537 dsl_dataset_rele(snap
->ds
, tag
);
2538 kmem_free(snap
, sizeof (*snap
));
2544 promote_hold(dsl_dataset_promote_arg_t
*ddpa
, dsl_pool_t
*dp
, void *tag
)
2548 struct promotenode
*snap
;
2550 error
= dsl_dataset_hold(dp
, ddpa
->ddpa_clonename
, tag
,
2554 dd
= ddpa
->ddpa_clone
->ds_dir
;
2556 if (ddpa
->ddpa_clone
->ds_is_snapshot
||
2557 !dsl_dir_is_clone(dd
)) {
2558 dsl_dataset_rele(ddpa
->ddpa_clone
, tag
);
2559 return (SET_ERROR(EINVAL
));
2562 error
= snaplist_make(dp
, 0, dsl_dir_phys(dd
)->dd_origin_obj
,
2563 &ddpa
->shared_snaps
, tag
);
2567 error
= snaplist_make(dp
, 0, ddpa
->ddpa_clone
->ds_object
,
2568 &ddpa
->clone_snaps
, tag
);
2572 snap
= list_head(&ddpa
->shared_snaps
);
2573 ASSERT3U(snap
->ds
->ds_object
, ==, dsl_dir_phys(dd
)->dd_origin_obj
);
2574 error
= snaplist_make(dp
, dsl_dir_phys(dd
)->dd_origin_obj
,
2575 dsl_dir_phys(snap
->ds
->ds_dir
)->dd_head_dataset_obj
,
2576 &ddpa
->origin_snaps
, tag
);
2580 if (dsl_dir_phys(snap
->ds
->ds_dir
)->dd_origin_obj
!= 0) {
2581 error
= dsl_dataset_hold_obj(dp
,
2582 dsl_dir_phys(snap
->ds
->ds_dir
)->dd_origin_obj
,
2583 tag
, &ddpa
->origin_origin
);
2589 promote_rele(ddpa
, tag
);
2594 promote_rele(dsl_dataset_promote_arg_t
*ddpa
, void *tag
)
2596 snaplist_destroy(&ddpa
->shared_snaps
, tag
);
2597 snaplist_destroy(&ddpa
->clone_snaps
, tag
);
2598 snaplist_destroy(&ddpa
->origin_snaps
, tag
);
2599 if (ddpa
->origin_origin
!= NULL
)
2600 dsl_dataset_rele(ddpa
->origin_origin
, tag
);
2601 dsl_dataset_rele(ddpa
->ddpa_clone
, tag
);
2607 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2608 * in with the name. (It must be at least MAXNAMELEN bytes long.)
2611 dsl_dataset_promote(const char *name
, char *conflsnap
)
2613 dsl_dataset_promote_arg_t ddpa
= { 0 };
2619 * We will modify space proportional to the number of
2620 * snapshots. Compute numsnaps.
2622 error
= dmu_objset_hold(name
, FTAG
, &os
);
2625 error
= zap_count(dmu_objset_pool(os
)->dp_meta_objset
,
2626 dsl_dataset_phys(dmu_objset_ds(os
))->ds_snapnames_zapobj
,
2628 dmu_objset_rele(os
, FTAG
);
2632 ddpa
.ddpa_clonename
= name
;
2633 ddpa
.err_ds
= conflsnap
;
2636 return (dsl_sync_task(name
, dsl_dataset_promote_check
,
2637 dsl_dataset_promote_sync
, &ddpa
,
2638 2 + numsnaps
, ZFS_SPACE_CHECK_RESERVED
));
2642 dsl_dataset_clone_swap_check_impl(dsl_dataset_t
*clone
,
2643 dsl_dataset_t
*origin_head
, boolean_t force
, void *owner
, dmu_tx_t
*tx
)
2645 int64_t unused_refres_delta
;
2647 /* they should both be heads */
2648 if (clone
->ds_is_snapshot
||
2649 origin_head
->ds_is_snapshot
)
2650 return (SET_ERROR(EINVAL
));
2652 /* if we are not forcing, the branch point should be just before them */
2653 if (!force
&& clone
->ds_prev
!= origin_head
->ds_prev
)
2654 return (SET_ERROR(EINVAL
));
2656 /* clone should be the clone (unless they are unrelated) */
2657 if (clone
->ds_prev
!= NULL
&&
2658 clone
->ds_prev
!= clone
->ds_dir
->dd_pool
->dp_origin_snap
&&
2659 origin_head
->ds_dir
!= clone
->ds_prev
->ds_dir
)
2660 return (SET_ERROR(EINVAL
));
2662 /* the clone should be a child of the origin */
2663 if (clone
->ds_dir
->dd_parent
!= origin_head
->ds_dir
)
2664 return (SET_ERROR(EINVAL
));
2666 /* origin_head shouldn't be modified unless 'force' */
2668 dsl_dataset_modified_since_snap(origin_head
, origin_head
->ds_prev
))
2669 return (SET_ERROR(ETXTBSY
));
2671 /* origin_head should have no long holds (e.g. is not mounted) */
2672 if (dsl_dataset_handoff_check(origin_head
, owner
, tx
))
2673 return (SET_ERROR(EBUSY
));
2675 /* check amount of any unconsumed refreservation */
2676 unused_refres_delta
=
2677 (int64_t)MIN(origin_head
->ds_reserved
,
2678 dsl_dataset_phys(origin_head
)->ds_unique_bytes
) -
2679 (int64_t)MIN(origin_head
->ds_reserved
,
2680 dsl_dataset_phys(clone
)->ds_unique_bytes
);
2682 if (unused_refres_delta
> 0 &&
2683 unused_refres_delta
>
2684 dsl_dir_space_available(origin_head
->ds_dir
, NULL
, 0, TRUE
))
2685 return (SET_ERROR(ENOSPC
));
2687 /* clone can't be over the head's refquota */
2688 if (origin_head
->ds_quota
!= 0 &&
2689 dsl_dataset_phys(clone
)->ds_referenced_bytes
>
2690 origin_head
->ds_quota
)
2691 return (SET_ERROR(EDQUOT
));
2697 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t
*clone
,
2698 dsl_dataset_t
*origin_head
, dmu_tx_t
*tx
)
2700 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2701 int64_t unused_refres_delta
;
2703 ASSERT(clone
->ds_reserved
== 0);
2704 ASSERT(origin_head
->ds_quota
== 0 ||
2705 dsl_dataset_phys(clone
)->ds_unique_bytes
<= origin_head
->ds_quota
);
2706 ASSERT3P(clone
->ds_prev
, ==, origin_head
->ds_prev
);
2708 dmu_buf_will_dirty(clone
->ds_dbuf
, tx
);
2709 dmu_buf_will_dirty(origin_head
->ds_dbuf
, tx
);
2711 if (clone
->ds_objset
!= NULL
) {
2712 dmu_objset_evict(clone
->ds_objset
);
2713 clone
->ds_objset
= NULL
;
2716 if (origin_head
->ds_objset
!= NULL
) {
2717 dmu_objset_evict(origin_head
->ds_objset
);
2718 origin_head
->ds_objset
= NULL
;
2721 unused_refres_delta
=
2722 (int64_t)MIN(origin_head
->ds_reserved
,
2723 dsl_dataset_phys(origin_head
)->ds_unique_bytes
) -
2724 (int64_t)MIN(origin_head
->ds_reserved
,
2725 dsl_dataset_phys(clone
)->ds_unique_bytes
);
2728 * Reset origin's unique bytes, if it exists.
2730 if (clone
->ds_prev
) {
2731 dsl_dataset_t
*origin
= clone
->ds_prev
;
2732 uint64_t comp
, uncomp
;
2734 dmu_buf_will_dirty(origin
->ds_dbuf
, tx
);
2735 dsl_deadlist_space_range(&clone
->ds_deadlist
,
2736 dsl_dataset_phys(origin
)->ds_prev_snap_txg
, UINT64_MAX
,
2737 &dsl_dataset_phys(origin
)->ds_unique_bytes
, &comp
, &uncomp
);
2743 tmp
= dsl_dataset_phys(origin_head
)->ds_bp
;
2744 dsl_dataset_phys(origin_head
)->ds_bp
=
2745 dsl_dataset_phys(clone
)->ds_bp
;
2746 dsl_dataset_phys(clone
)->ds_bp
= tmp
;
2749 /* set dd_*_bytes */
2751 int64_t dused
, dcomp
, duncomp
;
2752 uint64_t cdl_used
, cdl_comp
, cdl_uncomp
;
2753 uint64_t odl_used
, odl_comp
, odl_uncomp
;
2755 ASSERT3U(dsl_dir_phys(clone
->ds_dir
)->
2756 dd_used_breakdown
[DD_USED_SNAP
], ==, 0);
2758 dsl_deadlist_space(&clone
->ds_deadlist
,
2759 &cdl_used
, &cdl_comp
, &cdl_uncomp
);
2760 dsl_deadlist_space(&origin_head
->ds_deadlist
,
2761 &odl_used
, &odl_comp
, &odl_uncomp
);
2763 dused
= dsl_dataset_phys(clone
)->ds_referenced_bytes
+
2765 (dsl_dataset_phys(origin_head
)->ds_referenced_bytes
+
2767 dcomp
= dsl_dataset_phys(clone
)->ds_compressed_bytes
+
2769 (dsl_dataset_phys(origin_head
)->ds_compressed_bytes
+
2771 duncomp
= dsl_dataset_phys(clone
)->ds_uncompressed_bytes
+
2773 (dsl_dataset_phys(origin_head
)->ds_uncompressed_bytes
+
2776 dsl_dir_diduse_space(origin_head
->ds_dir
, DD_USED_HEAD
,
2777 dused
, dcomp
, duncomp
, tx
);
2778 dsl_dir_diduse_space(clone
->ds_dir
, DD_USED_HEAD
,
2779 -dused
, -dcomp
, -duncomp
, tx
);
2782 * The difference in the space used by snapshots is the
2783 * difference in snapshot space due to the head's
2784 * deadlist (since that's the only thing that's
2785 * changing that affects the snapused).
2787 dsl_deadlist_space_range(&clone
->ds_deadlist
,
2788 origin_head
->ds_dir
->dd_origin_txg
, UINT64_MAX
,
2789 &cdl_used
, &cdl_comp
, &cdl_uncomp
);
2790 dsl_deadlist_space_range(&origin_head
->ds_deadlist
,
2791 origin_head
->ds_dir
->dd_origin_txg
, UINT64_MAX
,
2792 &odl_used
, &odl_comp
, &odl_uncomp
);
2793 dsl_dir_transfer_space(origin_head
->ds_dir
, cdl_used
- odl_used
,
2794 DD_USED_HEAD
, DD_USED_SNAP
, tx
);
2797 /* swap ds_*_bytes */
2798 SWITCH64(dsl_dataset_phys(origin_head
)->ds_referenced_bytes
,
2799 dsl_dataset_phys(clone
)->ds_referenced_bytes
);
2800 SWITCH64(dsl_dataset_phys(origin_head
)->ds_compressed_bytes
,
2801 dsl_dataset_phys(clone
)->ds_compressed_bytes
);
2802 SWITCH64(dsl_dataset_phys(origin_head
)->ds_uncompressed_bytes
,
2803 dsl_dataset_phys(clone
)->ds_uncompressed_bytes
);
2804 SWITCH64(dsl_dataset_phys(origin_head
)->ds_unique_bytes
,
2805 dsl_dataset_phys(clone
)->ds_unique_bytes
);
2807 /* apply any parent delta for change in unconsumed refreservation */
2808 dsl_dir_diduse_space(origin_head
->ds_dir
, DD_USED_REFRSRV
,
2809 unused_refres_delta
, 0, 0, tx
);
2814 dsl_deadlist_close(&clone
->ds_deadlist
);
2815 dsl_deadlist_close(&origin_head
->ds_deadlist
);
2816 SWITCH64(dsl_dataset_phys(origin_head
)->ds_deadlist_obj
,
2817 dsl_dataset_phys(clone
)->ds_deadlist_obj
);
2818 dsl_deadlist_open(&clone
->ds_deadlist
, dp
->dp_meta_objset
,
2819 dsl_dataset_phys(clone
)->ds_deadlist_obj
);
2820 dsl_deadlist_open(&origin_head
->ds_deadlist
, dp
->dp_meta_objset
,
2821 dsl_dataset_phys(origin_head
)->ds_deadlist_obj
);
2823 dsl_scan_ds_clone_swapped(origin_head
, clone
, tx
);
2825 spa_history_log_internal_ds(clone
, "clone swap", tx
,
2826 "parent=%s", origin_head
->ds_dir
->dd_myname
);
2830 * Given a pool name and a dataset object number in that pool,
2831 * return the name of that dataset.
2834 dsl_dsobj_to_dsname(char *pname
, uint64_t obj
, char *buf
)
2840 error
= dsl_pool_hold(pname
, FTAG
, &dp
);
2844 error
= dsl_dataset_hold_obj(dp
, obj
, FTAG
, &ds
);
2846 dsl_dataset_name(ds
, buf
);
2847 dsl_dataset_rele(ds
, FTAG
);
2849 dsl_pool_rele(dp
, FTAG
);
2855 dsl_dataset_check_quota(dsl_dataset_t
*ds
, boolean_t check_quota
,
2856 uint64_t asize
, uint64_t inflight
, uint64_t *used
, uint64_t *ref_rsrv
)
2860 ASSERT3S(asize
, >, 0);
2863 * *ref_rsrv is the portion of asize that will come from any
2864 * unconsumed refreservation space.
2868 mutex_enter(&ds
->ds_lock
);
2870 * Make a space adjustment for reserved bytes.
2872 if (ds
->ds_reserved
> dsl_dataset_phys(ds
)->ds_unique_bytes
) {
2874 ds
->ds_reserved
- dsl_dataset_phys(ds
)->ds_unique_bytes
);
2876 (ds
->ds_reserved
- dsl_dataset_phys(ds
)->ds_unique_bytes
);
2878 asize
- MIN(asize
, parent_delta(ds
, asize
+ inflight
));
2881 if (!check_quota
|| ds
->ds_quota
== 0) {
2882 mutex_exit(&ds
->ds_lock
);
2886 * If they are requesting more space, and our current estimate
2887 * is over quota, they get to try again unless the actual
2888 * on-disk is over quota and there are no pending changes (which
2889 * may free up space for us).
2891 if (dsl_dataset_phys(ds
)->ds_referenced_bytes
+ inflight
>=
2894 dsl_dataset_phys(ds
)->ds_referenced_bytes
< ds
->ds_quota
)
2895 error
= SET_ERROR(ERESTART
);
2897 error
= SET_ERROR(EDQUOT
);
2899 mutex_exit(&ds
->ds_lock
);
2904 typedef struct dsl_dataset_set_qr_arg
{
2905 const char *ddsqra_name
;
2906 zprop_source_t ddsqra_source
;
2907 uint64_t ddsqra_value
;
2908 } dsl_dataset_set_qr_arg_t
;
2913 dsl_dataset_set_refquota_check(void *arg
, dmu_tx_t
*tx
)
2915 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
2916 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2921 if (spa_version(dp
->dp_spa
) < SPA_VERSION_REFQUOTA
)
2922 return (SET_ERROR(ENOTSUP
));
2924 error
= dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
);
2928 if (ds
->ds_is_snapshot
) {
2929 dsl_dataset_rele(ds
, FTAG
);
2930 return (SET_ERROR(EINVAL
));
2933 error
= dsl_prop_predict(ds
->ds_dir
,
2934 zfs_prop_to_name(ZFS_PROP_REFQUOTA
),
2935 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, &newval
);
2937 dsl_dataset_rele(ds
, FTAG
);
2942 dsl_dataset_rele(ds
, FTAG
);
2946 if (newval
< dsl_dataset_phys(ds
)->ds_referenced_bytes
||
2947 newval
< ds
->ds_reserved
) {
2948 dsl_dataset_rele(ds
, FTAG
);
2949 return (SET_ERROR(ENOSPC
));
2952 dsl_dataset_rele(ds
, FTAG
);
2957 dsl_dataset_set_refquota_sync(void *arg
, dmu_tx_t
*tx
)
2959 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
2960 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2964 VERIFY0(dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
));
2966 dsl_prop_set_sync_impl(ds
,
2967 zfs_prop_to_name(ZFS_PROP_REFQUOTA
),
2968 ddsqra
->ddsqra_source
, sizeof (ddsqra
->ddsqra_value
), 1,
2969 &ddsqra
->ddsqra_value
, tx
);
2971 VERIFY0(dsl_prop_get_int_ds(ds
,
2972 zfs_prop_to_name(ZFS_PROP_REFQUOTA
), &newval
));
2974 if (ds
->ds_quota
!= newval
) {
2975 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
2976 ds
->ds_quota
= newval
;
2978 dsl_dataset_rele(ds
, FTAG
);
2982 dsl_dataset_set_refquota(const char *dsname
, zprop_source_t source
,
2985 dsl_dataset_set_qr_arg_t ddsqra
;
2987 ddsqra
.ddsqra_name
= dsname
;
2988 ddsqra
.ddsqra_source
= source
;
2989 ddsqra
.ddsqra_value
= refquota
;
2991 return (dsl_sync_task(dsname
, dsl_dataset_set_refquota_check
,
2992 dsl_dataset_set_refquota_sync
, &ddsqra
, 0, ZFS_SPACE_CHECK_NONE
));
2996 dsl_dataset_set_refreservation_check(void *arg
, dmu_tx_t
*tx
)
2998 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
2999 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3002 uint64_t newval
, unique
;
3004 if (spa_version(dp
->dp_spa
) < SPA_VERSION_REFRESERVATION
)
3005 return (SET_ERROR(ENOTSUP
));
3007 error
= dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
);
3011 if (ds
->ds_is_snapshot
) {
3012 dsl_dataset_rele(ds
, FTAG
);
3013 return (SET_ERROR(EINVAL
));
3016 error
= dsl_prop_predict(ds
->ds_dir
,
3017 zfs_prop_to_name(ZFS_PROP_REFRESERVATION
),
3018 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, &newval
);
3020 dsl_dataset_rele(ds
, FTAG
);
3025 * If we are doing the preliminary check in open context, the
3026 * space estimates may be inaccurate.
3028 if (!dmu_tx_is_syncing(tx
)) {
3029 dsl_dataset_rele(ds
, FTAG
);
3033 mutex_enter(&ds
->ds_lock
);
3034 if (!DS_UNIQUE_IS_ACCURATE(ds
))
3035 dsl_dataset_recalc_head_uniq(ds
);
3036 unique
= dsl_dataset_phys(ds
)->ds_unique_bytes
;
3037 mutex_exit(&ds
->ds_lock
);
3039 if (MAX(unique
, newval
) > MAX(unique
, ds
->ds_reserved
)) {
3040 uint64_t delta
= MAX(unique
, newval
) -
3041 MAX(unique
, ds
->ds_reserved
);
3044 dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, B_TRUE
) ||
3045 (ds
->ds_quota
> 0 && newval
> ds
->ds_quota
)) {
3046 dsl_dataset_rele(ds
, FTAG
);
3047 return (SET_ERROR(ENOSPC
));
3051 dsl_dataset_rele(ds
, FTAG
);
3056 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t
*ds
,
3057 zprop_source_t source
, uint64_t value
, dmu_tx_t
*tx
)
3063 dsl_prop_set_sync_impl(ds
, zfs_prop_to_name(ZFS_PROP_REFRESERVATION
),
3064 source
, sizeof (value
), 1, &value
, tx
);
3066 VERIFY0(dsl_prop_get_int_ds(ds
,
3067 zfs_prop_to_name(ZFS_PROP_REFRESERVATION
), &newval
));
3069 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
3070 mutex_enter(&ds
->ds_dir
->dd_lock
);
3071 mutex_enter(&ds
->ds_lock
);
3072 ASSERT(DS_UNIQUE_IS_ACCURATE(ds
));
3073 unique
= dsl_dataset_phys(ds
)->ds_unique_bytes
;
3074 delta
= MAX(0, (int64_t)(newval
- unique
)) -
3075 MAX(0, (int64_t)(ds
->ds_reserved
- unique
));
3076 ds
->ds_reserved
= newval
;
3077 mutex_exit(&ds
->ds_lock
);
3079 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_REFRSRV
, delta
, 0, 0, tx
);
3080 mutex_exit(&ds
->ds_dir
->dd_lock
);
3084 dsl_dataset_set_refreservation_sync(void *arg
, dmu_tx_t
*tx
)
3086 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
3087 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3090 VERIFY0(dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
));
3091 dsl_dataset_set_refreservation_sync_impl(ds
,
3092 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, tx
);
3093 dsl_dataset_rele(ds
, FTAG
);
3097 dsl_dataset_set_refreservation(const char *dsname
, zprop_source_t source
,
3098 uint64_t refreservation
)
3100 dsl_dataset_set_qr_arg_t ddsqra
;
3102 ddsqra
.ddsqra_name
= dsname
;
3103 ddsqra
.ddsqra_source
= source
;
3104 ddsqra
.ddsqra_value
= refreservation
;
3106 return (dsl_sync_task(dsname
, dsl_dataset_set_refreservation_check
,
3107 dsl_dataset_set_refreservation_sync
, &ddsqra
,
3108 0, ZFS_SPACE_CHECK_NONE
));
3112 * Return (in *usedp) the amount of space written in new that is not
3113 * present in oldsnap. New may be a snapshot or the head. Old must be
3114 * a snapshot before new, in new's filesystem (or its origin). If not then
3115 * fail and return EINVAL.
3117 * The written space is calculated by considering two components: First, we
3118 * ignore any freed space, and calculate the written as new's used space
3119 * minus old's used space. Next, we add in the amount of space that was freed
3120 * between the two snapshots, thus reducing new's used space relative to old's.
3121 * Specifically, this is the space that was born before old->ds_creation_txg,
3122 * and freed before new (ie. on new's deadlist or a previous deadlist).
3124 * space freed [---------------------]
3125 * snapshots ---O-------O--------O-------O------
3129 dsl_dataset_space_written(dsl_dataset_t
*oldsnap
, dsl_dataset_t
*new,
3130 uint64_t *usedp
, uint64_t *compp
, uint64_t *uncompp
)
3134 dsl_pool_t
*dp
= new->ds_dir
->dd_pool
;
3136 ASSERT(dsl_pool_config_held(dp
));
3139 *usedp
+= dsl_dataset_phys(new)->ds_referenced_bytes
;
3140 *usedp
-= dsl_dataset_phys(oldsnap
)->ds_referenced_bytes
;
3143 *compp
+= dsl_dataset_phys(new)->ds_compressed_bytes
;
3144 *compp
-= dsl_dataset_phys(oldsnap
)->ds_compressed_bytes
;
3147 *uncompp
+= dsl_dataset_phys(new)->ds_uncompressed_bytes
;
3148 *uncompp
-= dsl_dataset_phys(oldsnap
)->ds_uncompressed_bytes
;
3150 snapobj
= new->ds_object
;
3151 while (snapobj
!= oldsnap
->ds_object
) {
3152 dsl_dataset_t
*snap
;
3153 uint64_t used
, comp
, uncomp
;
3155 if (snapobj
== new->ds_object
) {
3158 err
= dsl_dataset_hold_obj(dp
, snapobj
, FTAG
, &snap
);
3163 if (dsl_dataset_phys(snap
)->ds_prev_snap_txg
==
3164 dsl_dataset_phys(oldsnap
)->ds_creation_txg
) {
3166 * The blocks in the deadlist can not be born after
3167 * ds_prev_snap_txg, so get the whole deadlist space,
3168 * which is more efficient (especially for old-format
3169 * deadlists). Unfortunately the deadlist code
3170 * doesn't have enough information to make this
3171 * optimization itself.
3173 dsl_deadlist_space(&snap
->ds_deadlist
,
3174 &used
, &comp
, &uncomp
);
3176 dsl_deadlist_space_range(&snap
->ds_deadlist
,
3177 0, dsl_dataset_phys(oldsnap
)->ds_creation_txg
,
3178 &used
, &comp
, &uncomp
);
3185 * If we get to the beginning of the chain of snapshots
3186 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3187 * was not a snapshot of/before new.
3189 snapobj
= dsl_dataset_phys(snap
)->ds_prev_snap_obj
;
3191 dsl_dataset_rele(snap
, FTAG
);
3193 err
= SET_ERROR(EINVAL
);
3202 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3203 * lastsnap, and all snapshots in between are deleted.
3205 * blocks that would be freed [---------------------------]
3206 * snapshots ---O-------O--------O-------O--------O
3207 * firstsnap lastsnap
3209 * This is the set of blocks that were born after the snap before firstsnap,
3210 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3211 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3212 * We calculate this by iterating over the relevant deadlists (from the snap
3213 * after lastsnap, backward to the snap after firstsnap), summing up the
3214 * space on the deadlist that was born after the snap before firstsnap.
3217 dsl_dataset_space_wouldfree(dsl_dataset_t
*firstsnap
,
3218 dsl_dataset_t
*lastsnap
,
3219 uint64_t *usedp
, uint64_t *compp
, uint64_t *uncompp
)
3223 dsl_pool_t
*dp
= firstsnap
->ds_dir
->dd_pool
;
3225 ASSERT(firstsnap
->ds_is_snapshot
);
3226 ASSERT(lastsnap
->ds_is_snapshot
);
3229 * Check that the snapshots are in the same dsl_dir, and firstsnap
3230 * is before lastsnap.
3232 if (firstsnap
->ds_dir
!= lastsnap
->ds_dir
||
3233 dsl_dataset_phys(firstsnap
)->ds_creation_txg
>
3234 dsl_dataset_phys(lastsnap
)->ds_creation_txg
)
3235 return (SET_ERROR(EINVAL
));
3237 *usedp
= *compp
= *uncompp
= 0;
3239 snapobj
= dsl_dataset_phys(lastsnap
)->ds_next_snap_obj
;
3240 while (snapobj
!= firstsnap
->ds_object
) {
3242 uint64_t used
, comp
, uncomp
;
3244 err
= dsl_dataset_hold_obj(dp
, snapobj
, FTAG
, &ds
);
3248 dsl_deadlist_space_range(&ds
->ds_deadlist
,
3249 dsl_dataset_phys(firstsnap
)->ds_prev_snap_txg
, UINT64_MAX
,
3250 &used
, &comp
, &uncomp
);
3255 snapobj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
3256 ASSERT3U(snapobj
, !=, 0);
3257 dsl_dataset_rele(ds
, FTAG
);
3263 dsl_dataset_activate_large_blocks_check(void *arg
, dmu_tx_t
*tx
)
3265 const char *dsname
= arg
;
3267 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3270 if (!spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_LARGE_BLOCKS
))
3271 return (SET_ERROR(ENOTSUP
));
3273 ASSERT(spa_feature_is_enabled(dp
->dp_spa
,
3274 SPA_FEATURE_EXTENSIBLE_DATASET
));
3276 error
= dsl_dataset_hold(dp
, dsname
, FTAG
, &ds
);
3280 if (ds
->ds_large_blocks
)
3282 dsl_dataset_rele(ds
, FTAG
);
3288 dsl_dataset_activate_large_blocks_sync_impl(uint64_t dsobj
, dmu_tx_t
*tx
)
3290 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
3291 objset_t
*mos
= dmu_tx_pool(tx
)->dp_meta_objset
;
3294 spa_feature_incr(spa
, SPA_FEATURE_LARGE_BLOCKS
, tx
);
3295 dmu_object_zapify(mos
, dsobj
, DMU_OT_DSL_DATASET
, tx
);
3297 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_LARGE_BLOCKS
,
3298 sizeof (zero
), 1, &zero
, tx
));
3302 dsl_dataset_activate_large_blocks_sync(void *arg
, dmu_tx_t
*tx
)
3304 const char *dsname
= arg
;
3307 VERIFY0(dsl_dataset_hold(dmu_tx_pool(tx
), dsname
, FTAG
, &ds
));
3309 dsl_dataset_activate_large_blocks_sync_impl(ds
->ds_object
, tx
);
3310 ASSERT(!ds
->ds_large_blocks
);
3311 ds
->ds_large_blocks
= B_TRUE
;
3312 dsl_dataset_rele(ds
, FTAG
);
3316 dsl_dataset_activate_large_blocks(const char *dsname
)
3320 error
= dsl_sync_task(dsname
,
3321 dsl_dataset_activate_large_blocks_check
,
3322 dsl_dataset_activate_large_blocks_sync
, (void *)dsname
,
3323 1, ZFS_SPACE_CHECK_RESERVED
);
3326 * EALREADY indicates that this dataset already supports large blocks.
3328 if (error
== EALREADY
)
3334 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3335 * For example, they could both be snapshots of the same filesystem, and
3336 * 'earlier' is before 'later'. Or 'earlier' could be the origin of
3337 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's
3338 * filesystem. Or 'earlier' could be the origin's origin.
3340 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3343 dsl_dataset_is_before(dsl_dataset_t
*later
, dsl_dataset_t
*earlier
,
3344 uint64_t earlier_txg
)
3346 dsl_pool_t
*dp
= later
->ds_dir
->dd_pool
;
3349 dsl_dataset_t
*origin
;
3351 ASSERT(dsl_pool_config_held(dp
));
3352 ASSERT(earlier
->ds_is_snapshot
|| earlier_txg
!= 0);
3354 if (earlier_txg
== 0)
3355 earlier_txg
= dsl_dataset_phys(earlier
)->ds_creation_txg
;
3357 if (later
->ds_is_snapshot
&&
3358 earlier_txg
>= dsl_dataset_phys(later
)->ds_creation_txg
)
3361 if (later
->ds_dir
== earlier
->ds_dir
)
3363 if (!dsl_dir_is_clone(later
->ds_dir
))
3366 if (dsl_dir_phys(later
->ds_dir
)->dd_origin_obj
== earlier
->ds_object
)
3368 error
= dsl_dataset_hold_obj(dp
,
3369 dsl_dir_phys(later
->ds_dir
)->dd_origin_obj
, FTAG
, &origin
);
3372 ret
= dsl_dataset_is_before(origin
, earlier
, earlier_txg
);
3373 dsl_dataset_rele(origin
, FTAG
);
3379 dsl_dataset_zapify(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
3381 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
3382 dmu_object_zapify(mos
, ds
->ds_object
, DMU_OT_DSL_DATASET
, tx
);
3385 #if defined(_KERNEL) && defined(HAVE_SPL)
3387 module_param(zfs_max_recordsize
, int, 0644);
3388 MODULE_PARM_DESC(zfs_max_recordsize
, "Max allowed record size");
3390 /* Limited to 1M on 32-bit platforms due to lack of virtual address space */
3391 module_param(zfs_max_recordsize
, int, 0444);
3392 MODULE_PARM_DESC(zfs_max_recordsize
, "Max allowed record size");
3395 EXPORT_SYMBOL(dsl_dataset_hold
);
3396 EXPORT_SYMBOL(dsl_dataset_hold_obj
);
3397 EXPORT_SYMBOL(dsl_dataset_own
);
3398 EXPORT_SYMBOL(dsl_dataset_own_obj
);
3399 EXPORT_SYMBOL(dsl_dataset_name
);
3400 EXPORT_SYMBOL(dsl_dataset_rele
);
3401 EXPORT_SYMBOL(dsl_dataset_disown
);
3402 EXPORT_SYMBOL(dsl_dataset_tryown
);
3403 EXPORT_SYMBOL(dsl_dataset_create_sync
);
3404 EXPORT_SYMBOL(dsl_dataset_create_sync_dd
);
3405 EXPORT_SYMBOL(dsl_dataset_snapshot_check
);
3406 EXPORT_SYMBOL(dsl_dataset_snapshot_sync
);
3407 EXPORT_SYMBOL(dsl_dataset_promote
);
3408 EXPORT_SYMBOL(dsl_dataset_user_hold
);
3409 EXPORT_SYMBOL(dsl_dataset_user_release
);
3410 EXPORT_SYMBOL(dsl_dataset_get_holds
);
3411 EXPORT_SYMBOL(dsl_dataset_get_blkptr
);
3412 EXPORT_SYMBOL(dsl_dataset_set_blkptr
);
3413 EXPORT_SYMBOL(dsl_dataset_get_spa
);
3414 EXPORT_SYMBOL(dsl_dataset_modified_since_snap
);
3415 EXPORT_SYMBOL(dsl_dataset_space_written
);
3416 EXPORT_SYMBOL(dsl_dataset_space_wouldfree
);
3417 EXPORT_SYMBOL(dsl_dataset_sync
);
3418 EXPORT_SYMBOL(dsl_dataset_block_born
);
3419 EXPORT_SYMBOL(dsl_dataset_block_kill
);
3420 EXPORT_SYMBOL(dsl_dataset_block_freeable
);
3421 EXPORT_SYMBOL(dsl_dataset_prev_snap_txg
);
3422 EXPORT_SYMBOL(dsl_dataset_dirty
);
3423 EXPORT_SYMBOL(dsl_dataset_stats
);
3424 EXPORT_SYMBOL(dsl_dataset_fast_stat
);
3425 EXPORT_SYMBOL(dsl_dataset_space
);
3426 EXPORT_SYMBOL(dsl_dataset_fsid_guid
);
3427 EXPORT_SYMBOL(dsl_dsobj_to_dsname
);
3428 EXPORT_SYMBOL(dsl_dataset_check_quota
);
3429 EXPORT_SYMBOL(dsl_dataset_clone_swap_check_impl
);
3430 EXPORT_SYMBOL(dsl_dataset_clone_swap_sync_impl
);