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.
27 * Copyright (c) 2016 Actifio, Inc. All rights reserved.
30 #include <sys/dmu_objset.h>
31 #include <sys/dsl_dataset.h>
32 #include <sys/dsl_dir.h>
33 #include <sys/dsl_prop.h>
34 #include <sys/dsl_synctask.h>
35 #include <sys/dmu_traverse.h>
36 #include <sys/dmu_impl.h>
37 #include <sys/dmu_tx.h>
41 #include <sys/zfeature.h>
42 #include <sys/unique.h>
43 #include <sys/zfs_context.h>
44 #include <sys/zfs_ioctl.h>
46 #include <sys/zfs_znode.h>
47 #include <sys/zfs_onexit.h>
49 #include <sys/dsl_scan.h>
50 #include <sys/dsl_deadlist.h>
51 #include <sys/dsl_destroy.h>
52 #include <sys/dsl_userhold.h>
53 #include <sys/dsl_bookmark.h>
56 * The SPA supports block sizes up to 16MB. However, very large blocks
57 * can have an impact on i/o latency (e.g. tying up a spinning disk for
58 * ~300ms), and also potentially on the memory allocator. Therefore,
59 * we do not allow the recordsize to be set larger than zfs_max_recordsize
60 * (default 1MB). Larger blocks can be created by changing this tunable,
61 * and pools with larger blocks can always be imported and used, regardless
64 int zfs_max_recordsize
= 1 * 1024 * 1024;
66 #define SWITCH64(x, y) \
68 uint64_t __tmp = (x); \
73 #define DS_REF_MAX (1ULL << 62)
75 extern inline dsl_dataset_phys_t
*dsl_dataset_phys(dsl_dataset_t
*ds
);
78 * Figure out how much of this delta should be propogated to the dsl_dir
79 * layer. If there's a refreservation, that space has already been
80 * partially accounted for in our ancestors.
83 parent_delta(dsl_dataset_t
*ds
, int64_t delta
)
85 dsl_dataset_phys_t
*ds_phys
;
86 uint64_t old_bytes
, new_bytes
;
88 if (ds
->ds_reserved
== 0)
91 ds_phys
= dsl_dataset_phys(ds
);
92 old_bytes
= MAX(ds_phys
->ds_unique_bytes
, ds
->ds_reserved
);
93 new_bytes
= MAX(ds_phys
->ds_unique_bytes
+ delta
, ds
->ds_reserved
);
95 ASSERT3U(ABS((int64_t)(new_bytes
- old_bytes
)), <=, ABS(delta
));
96 return (new_bytes
- old_bytes
);
100 dsl_dataset_block_born(dsl_dataset_t
*ds
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
102 int used
, compressed
, uncompressed
;
105 used
= bp_get_dsize_sync(tx
->tx_pool
->dp_spa
, bp
);
106 compressed
= BP_GET_PSIZE(bp
);
107 uncompressed
= BP_GET_UCSIZE(bp
);
109 dprintf_bp(bp
, "ds=%p", ds
);
111 ASSERT(dmu_tx_is_syncing(tx
));
112 /* It could have been compressed away to nothing */
115 ASSERT(BP_GET_TYPE(bp
) != DMU_OT_NONE
);
116 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp
)));
118 dsl_pool_mos_diduse_space(tx
->tx_pool
,
119 used
, compressed
, uncompressed
);
123 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
124 mutex_enter(&ds
->ds_lock
);
125 delta
= parent_delta(ds
, used
);
126 dsl_dataset_phys(ds
)->ds_referenced_bytes
+= used
;
127 dsl_dataset_phys(ds
)->ds_compressed_bytes
+= compressed
;
128 dsl_dataset_phys(ds
)->ds_uncompressed_bytes
+= uncompressed
;
129 dsl_dataset_phys(ds
)->ds_unique_bytes
+= used
;
130 if (BP_GET_LSIZE(bp
) > SPA_OLD_MAXBLOCKSIZE
)
131 ds
->ds_need_large_blocks
= B_TRUE
;
132 mutex_exit(&ds
->ds_lock
);
133 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_HEAD
, delta
,
134 compressed
, uncompressed
, tx
);
135 dsl_dir_transfer_space(ds
->ds_dir
, used
- delta
,
136 DD_USED_REFRSRV
, DD_USED_HEAD
, tx
);
140 dsl_dataset_block_kill(dsl_dataset_t
*ds
, const blkptr_t
*bp
, dmu_tx_t
*tx
,
143 int used
= bp_get_dsize_sync(tx
->tx_pool
->dp_spa
, bp
);
144 int compressed
= BP_GET_PSIZE(bp
);
145 int uncompressed
= BP_GET_UCSIZE(bp
);
150 ASSERT(dmu_tx_is_syncing(tx
));
151 ASSERT(bp
->blk_birth
<= tx
->tx_txg
);
154 dsl_free(tx
->tx_pool
, tx
->tx_txg
, bp
);
155 dsl_pool_mos_diduse_space(tx
->tx_pool
,
156 -used
, -compressed
, -uncompressed
);
159 ASSERT3P(tx
->tx_pool
, ==, ds
->ds_dir
->dd_pool
);
161 ASSERT(!ds
->ds_is_snapshot
);
162 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
164 if (bp
->blk_birth
> dsl_dataset_phys(ds
)->ds_prev_snap_txg
) {
167 dprintf_bp(bp
, "freeing ds=%llu", ds
->ds_object
);
168 dsl_free(tx
->tx_pool
, tx
->tx_txg
, bp
);
170 mutex_enter(&ds
->ds_lock
);
171 ASSERT(dsl_dataset_phys(ds
)->ds_unique_bytes
>= used
||
172 !DS_UNIQUE_IS_ACCURATE(ds
));
173 delta
= parent_delta(ds
, -used
);
174 dsl_dataset_phys(ds
)->ds_unique_bytes
-= used
;
175 mutex_exit(&ds
->ds_lock
);
176 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_HEAD
,
177 delta
, -compressed
, -uncompressed
, tx
);
178 dsl_dir_transfer_space(ds
->ds_dir
, -used
- delta
,
179 DD_USED_REFRSRV
, DD_USED_HEAD
, tx
);
181 dprintf_bp(bp
, "putting on dead list: %s", "");
184 * We are here as part of zio's write done callback,
185 * which means we're a zio interrupt thread. We can't
186 * call dsl_deadlist_insert() now because it may block
187 * waiting for I/O. Instead, put bp on the deferred
188 * queue and let dsl_pool_sync() finish the job.
190 bplist_append(&ds
->ds_pending_deadlist
, bp
);
192 dsl_deadlist_insert(&ds
->ds_deadlist
, bp
, tx
);
194 ASSERT3U(ds
->ds_prev
->ds_object
, ==,
195 dsl_dataset_phys(ds
)->ds_prev_snap_obj
);
196 ASSERT(dsl_dataset_phys(ds
->ds_prev
)->ds_num_children
> 0);
197 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
198 if (dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
==
199 ds
->ds_object
&& bp
->blk_birth
>
200 dsl_dataset_phys(ds
->ds_prev
)->ds_prev_snap_txg
) {
201 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
202 mutex_enter(&ds
->ds_prev
->ds_lock
);
203 dsl_dataset_phys(ds
->ds_prev
)->ds_unique_bytes
+= used
;
204 mutex_exit(&ds
->ds_prev
->ds_lock
);
206 if (bp
->blk_birth
> ds
->ds_dir
->dd_origin_txg
) {
207 dsl_dir_transfer_space(ds
->ds_dir
, used
,
208 DD_USED_HEAD
, DD_USED_SNAP
, tx
);
211 mutex_enter(&ds
->ds_lock
);
212 ASSERT3U(dsl_dataset_phys(ds
)->ds_referenced_bytes
, >=, used
);
213 dsl_dataset_phys(ds
)->ds_referenced_bytes
-= used
;
214 ASSERT3U(dsl_dataset_phys(ds
)->ds_compressed_bytes
, >=, compressed
);
215 dsl_dataset_phys(ds
)->ds_compressed_bytes
-= compressed
;
216 ASSERT3U(dsl_dataset_phys(ds
)->ds_uncompressed_bytes
, >=, uncompressed
);
217 dsl_dataset_phys(ds
)->ds_uncompressed_bytes
-= uncompressed
;
218 mutex_exit(&ds
->ds_lock
);
224 dsl_dataset_prev_snap_txg(dsl_dataset_t
*ds
)
226 uint64_t trysnap
= 0;
231 * The snapshot creation could fail, but that would cause an
232 * incorrect FALSE return, which would only result in an
233 * overestimation of the amount of space that an operation would
234 * consume, which is OK.
236 * There's also a small window where we could miss a pending
237 * snapshot, because we could set the sync task in the quiescing
238 * phase. So this should only be used as a guess.
240 if (ds
->ds_trysnap_txg
>
241 spa_last_synced_txg(ds
->ds_dir
->dd_pool
->dp_spa
))
242 trysnap
= ds
->ds_trysnap_txg
;
243 return (MAX(dsl_dataset_phys(ds
)->ds_prev_snap_txg
, trysnap
));
247 dsl_dataset_block_freeable(dsl_dataset_t
*ds
, const blkptr_t
*bp
,
250 if (blk_birth
<= dsl_dataset_prev_snap_txg(ds
) ||
251 (bp
!= NULL
&& BP_IS_HOLE(bp
)))
254 ddt_prefetch(dsl_dataset_get_spa(ds
), bp
);
260 dsl_dataset_evict(void *dbu
)
262 dsl_dataset_t
*ds
= dbu
;
264 ASSERT(ds
->ds_owner
== NULL
);
268 unique_remove(ds
->ds_fsid_guid
);
270 if (ds
->ds_objset
!= NULL
)
271 dmu_objset_evict(ds
->ds_objset
);
274 dsl_dataset_rele(ds
->ds_prev
, ds
);
278 bplist_destroy(&ds
->ds_pending_deadlist
);
279 if (ds
->ds_deadlist
.dl_os
!= NULL
)
280 dsl_deadlist_close(&ds
->ds_deadlist
);
282 dsl_dir_async_rele(ds
->ds_dir
, ds
);
284 ASSERT(!list_link_active(&ds
->ds_synced_link
));
286 mutex_destroy(&ds
->ds_lock
);
287 mutex_destroy(&ds
->ds_opening_lock
);
288 mutex_destroy(&ds
->ds_sendstream_lock
);
289 refcount_destroy(&ds
->ds_longholds
);
291 kmem_free(ds
, sizeof (dsl_dataset_t
));
295 dsl_dataset_get_snapname(dsl_dataset_t
*ds
)
297 dsl_dataset_phys_t
*headphys
;
300 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
301 objset_t
*mos
= dp
->dp_meta_objset
;
303 if (ds
->ds_snapname
[0])
305 if (dsl_dataset_phys(ds
)->ds_next_snap_obj
== 0)
308 err
= dmu_bonus_hold(mos
, dsl_dir_phys(ds
->ds_dir
)->dd_head_dataset_obj
,
312 headphys
= headdbuf
->db_data
;
313 err
= zap_value_search(dp
->dp_meta_objset
,
314 headphys
->ds_snapnames_zapobj
, ds
->ds_object
, 0, ds
->ds_snapname
);
315 if (err
!= 0 && zfs_recover
== B_TRUE
) {
317 (void) snprintf(ds
->ds_snapname
, sizeof (ds
->ds_snapname
),
318 "SNAPOBJ=%llu-ERR=%d",
319 (unsigned long long)ds
->ds_object
, err
);
321 dmu_buf_rele(headdbuf
, FTAG
);
326 dsl_dataset_snap_lookup(dsl_dataset_t
*ds
, const char *name
, uint64_t *value
)
328 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
329 uint64_t snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
333 if (dsl_dataset_phys(ds
)->ds_flags
& DS_FLAG_CI_DATASET
)
338 err
= zap_lookup_norm(mos
, snapobj
, name
, 8, 1,
339 value
, mt
, NULL
, 0, NULL
);
340 if (err
== ENOTSUP
&& mt
== MT_FIRST
)
341 err
= zap_lookup(mos
, snapobj
, name
, 8, 1, value
);
346 dsl_dataset_snap_remove(dsl_dataset_t
*ds
, const char *name
, dmu_tx_t
*tx
,
349 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
350 uint64_t snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
354 dsl_dir_snap_cmtime_update(ds
->ds_dir
);
356 if (dsl_dataset_phys(ds
)->ds_flags
& DS_FLAG_CI_DATASET
)
361 err
= zap_remove_norm(mos
, snapobj
, name
, mt
, tx
);
362 if (err
== ENOTSUP
&& mt
== MT_FIRST
)
363 err
= zap_remove(mos
, snapobj
, name
, tx
);
365 if (err
== 0 && adj_cnt
)
366 dsl_fs_ss_count_adjust(ds
->ds_dir
, -1,
367 DD_FIELD_SNAPSHOT_COUNT
, tx
);
373 dsl_dataset_try_add_ref(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *tag
)
375 dmu_buf_t
*dbuf
= ds
->ds_dbuf
;
376 boolean_t result
= B_FALSE
;
378 if (dbuf
!= NULL
&& dmu_buf_try_add_ref(dbuf
, dp
->dp_meta_objset
,
379 ds
->ds_object
, DMU_BONUS_BLKID
, tag
)) {
381 if (ds
== dmu_buf_get_user(dbuf
))
384 dmu_buf_rele(dbuf
, tag
);
391 dsl_dataset_hold_obj(dsl_pool_t
*dp
, uint64_t dsobj
, void *tag
,
394 objset_t
*mos
= dp
->dp_meta_objset
;
398 dmu_object_info_t doi
;
400 ASSERT(dsl_pool_config_held(dp
));
402 err
= dmu_bonus_hold(mos
, dsobj
, tag
, &dbuf
);
406 /* Make sure dsobj has the correct object type. */
407 dmu_object_info_from_db(dbuf
, &doi
);
408 if (doi
.doi_bonus_type
!= DMU_OT_DSL_DATASET
) {
409 dmu_buf_rele(dbuf
, tag
);
410 return (SET_ERROR(EINVAL
));
413 ds
= dmu_buf_get_user(dbuf
);
415 dsl_dataset_t
*winner
= NULL
;
417 ds
= kmem_zalloc(sizeof (dsl_dataset_t
), KM_SLEEP
);
419 ds
->ds_object
= dsobj
;
420 ds
->ds_is_snapshot
= dsl_dataset_phys(ds
)->ds_num_children
!= 0;
421 list_link_init(&ds
->ds_synced_link
);
423 mutex_init(&ds
->ds_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
424 mutex_init(&ds
->ds_opening_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
425 mutex_init(&ds
->ds_sendstream_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
426 refcount_create(&ds
->ds_longholds
);
428 bplist_create(&ds
->ds_pending_deadlist
);
429 dsl_deadlist_open(&ds
->ds_deadlist
,
430 mos
, dsl_dataset_phys(ds
)->ds_deadlist_obj
);
432 list_create(&ds
->ds_sendstreams
, sizeof (dmu_sendarg_t
),
433 offsetof(dmu_sendarg_t
, dsa_link
));
435 if (doi
.doi_type
== DMU_OTN_ZAP_METADATA
) {
436 int zaperr
= zap_contains(mos
, dsobj
,
437 DS_FIELD_LARGE_BLOCKS
);
438 if (zaperr
!= ENOENT
) {
440 ds
->ds_large_blocks
= B_TRUE
;
445 err
= dsl_dir_hold_obj(dp
,
446 dsl_dataset_phys(ds
)->ds_dir_obj
, NULL
, ds
,
450 mutex_destroy(&ds
->ds_lock
);
451 mutex_destroy(&ds
->ds_opening_lock
);
452 mutex_destroy(&ds
->ds_sendstream_lock
);
453 refcount_destroy(&ds
->ds_longholds
);
454 bplist_destroy(&ds
->ds_pending_deadlist
);
455 dsl_deadlist_close(&ds
->ds_deadlist
);
456 kmem_free(ds
, sizeof (dsl_dataset_t
));
457 dmu_buf_rele(dbuf
, tag
);
461 if (!ds
->ds_is_snapshot
) {
462 ds
->ds_snapname
[0] = '\0';
463 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0) {
464 err
= dsl_dataset_hold_obj(dp
,
465 dsl_dataset_phys(ds
)->ds_prev_snap_obj
,
468 if (doi
.doi_type
== DMU_OTN_ZAP_METADATA
) {
469 int zaperr
= zap_lookup(mos
, ds
->ds_object
,
470 DS_FIELD_BOOKMARK_NAMES
,
471 sizeof (ds
->ds_bookmarks
), 1,
473 if (zaperr
!= ENOENT
)
477 if (zfs_flags
& ZFS_DEBUG_SNAPNAMES
)
478 err
= dsl_dataset_get_snapname(ds
);
480 dsl_dataset_phys(ds
)->ds_userrefs_obj
!= 0) {
482 ds
->ds_dir
->dd_pool
->dp_meta_objset
,
483 dsl_dataset_phys(ds
)->ds_userrefs_obj
,
488 if (err
== 0 && !ds
->ds_is_snapshot
) {
489 err
= dsl_prop_get_int_ds(ds
,
490 zfs_prop_to_name(ZFS_PROP_REFRESERVATION
),
493 err
= dsl_prop_get_int_ds(ds
,
494 zfs_prop_to_name(ZFS_PROP_REFQUOTA
),
498 ds
->ds_reserved
= ds
->ds_quota
= 0;
501 dmu_buf_init_user(&ds
->ds_dbu
, dsl_dataset_evict
, &ds
->ds_dbuf
);
503 winner
= dmu_buf_set_user_ie(dbuf
, &ds
->ds_dbu
);
505 if (err
!= 0 || winner
!= NULL
) {
506 bplist_destroy(&ds
->ds_pending_deadlist
);
507 dsl_deadlist_close(&ds
->ds_deadlist
);
509 dsl_dataset_rele(ds
->ds_prev
, ds
);
510 dsl_dir_rele(ds
->ds_dir
, ds
);
511 mutex_destroy(&ds
->ds_lock
);
512 mutex_destroy(&ds
->ds_opening_lock
);
513 mutex_destroy(&ds
->ds_sendstream_lock
);
514 refcount_destroy(&ds
->ds_longholds
);
515 kmem_free(ds
, sizeof (dsl_dataset_t
));
517 dmu_buf_rele(dbuf
, tag
);
523 unique_insert(dsl_dataset_phys(ds
)->ds_fsid_guid
);
526 ASSERT3P(ds
->ds_dbuf
, ==, dbuf
);
527 ASSERT3P(dsl_dataset_phys(ds
), ==, dbuf
->db_data
);
528 ASSERT(dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0 ||
529 spa_version(dp
->dp_spa
) < SPA_VERSION_ORIGIN
||
530 dp
->dp_origin_snap
== NULL
|| ds
== dp
->dp_origin_snap
);
536 dsl_dataset_hold(dsl_pool_t
*dp
, const char *name
,
537 void *tag
, dsl_dataset_t
**dsp
)
540 const char *snapname
;
544 err
= dsl_dir_hold(dp
, name
, FTAG
, &dd
, &snapname
);
548 ASSERT(dsl_pool_config_held(dp
));
549 obj
= dsl_dir_phys(dd
)->dd_head_dataset_obj
;
551 err
= dsl_dataset_hold_obj(dp
, obj
, tag
, dsp
);
553 err
= SET_ERROR(ENOENT
);
555 /* we may be looking for a snapshot */
556 if (err
== 0 && snapname
!= NULL
) {
559 if (*snapname
++ != '@') {
560 dsl_dataset_rele(*dsp
, tag
);
561 dsl_dir_rele(dd
, FTAG
);
562 return (SET_ERROR(ENOENT
));
565 dprintf("looking for snapshot '%s'\n", snapname
);
566 err
= dsl_dataset_snap_lookup(*dsp
, snapname
, &obj
);
568 err
= dsl_dataset_hold_obj(dp
, obj
, tag
, &ds
);
569 dsl_dataset_rele(*dsp
, tag
);
572 mutex_enter(&ds
->ds_lock
);
573 if (ds
->ds_snapname
[0] == 0)
574 (void) strlcpy(ds
->ds_snapname
, snapname
,
575 sizeof (ds
->ds_snapname
));
576 mutex_exit(&ds
->ds_lock
);
581 dsl_dir_rele(dd
, FTAG
);
586 dsl_dataset_own_obj(dsl_pool_t
*dp
, uint64_t dsobj
,
587 void *tag
, dsl_dataset_t
**dsp
)
589 int err
= dsl_dataset_hold_obj(dp
, dsobj
, tag
, dsp
);
592 if (!dsl_dataset_tryown(*dsp
, tag
)) {
593 dsl_dataset_rele(*dsp
, tag
);
595 return (SET_ERROR(EBUSY
));
601 dsl_dataset_own(dsl_pool_t
*dp
, const char *name
,
602 void *tag
, dsl_dataset_t
**dsp
)
604 int err
= dsl_dataset_hold(dp
, name
, tag
, dsp
);
607 if (!dsl_dataset_tryown(*dsp
, tag
)) {
608 dsl_dataset_rele(*dsp
, tag
);
609 return (SET_ERROR(EBUSY
));
615 * See the comment above dsl_pool_hold() for details. In summary, a long
616 * hold is used to prevent destruction of a dataset while the pool hold
617 * is dropped, allowing other concurrent operations (e.g. spa_sync()).
619 * The dataset and pool must be held when this function is called. After it
620 * is called, the pool hold may be released while the dataset is still held
624 dsl_dataset_long_hold(dsl_dataset_t
*ds
, void *tag
)
626 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
627 (void) refcount_add(&ds
->ds_longholds
, tag
);
631 dsl_dataset_long_rele(dsl_dataset_t
*ds
, void *tag
)
633 (void) refcount_remove(&ds
->ds_longholds
, tag
);
636 /* Return B_TRUE if there are any long holds on this dataset. */
638 dsl_dataset_long_held(dsl_dataset_t
*ds
)
640 return (!refcount_is_zero(&ds
->ds_longholds
));
644 dsl_dataset_name(dsl_dataset_t
*ds
, char *name
)
647 (void) strcpy(name
, "mos");
649 dsl_dir_name(ds
->ds_dir
, name
);
650 VERIFY0(dsl_dataset_get_snapname(ds
));
651 if (ds
->ds_snapname
[0]) {
652 (void) strcat(name
, "@");
654 * We use a "recursive" mutex so that we
655 * can call dprintf_ds() with ds_lock held.
657 if (!MUTEX_HELD(&ds
->ds_lock
)) {
658 mutex_enter(&ds
->ds_lock
);
659 (void) strcat(name
, ds
->ds_snapname
);
660 mutex_exit(&ds
->ds_lock
);
662 (void) strcat(name
, ds
->ds_snapname
);
669 dsl_dataset_namelen(dsl_dataset_t
*ds
)
672 VERIFY0(dsl_dataset_get_snapname(ds
));
673 mutex_enter(&ds
->ds_lock
);
674 len
= dsl_dir_namelen(ds
->ds_dir
) + 1 + strlen(ds
->ds_snapname
);
675 mutex_exit(&ds
->ds_lock
);
680 dsl_dataset_rele(dsl_dataset_t
*ds
, void *tag
)
682 dmu_buf_rele(ds
->ds_dbuf
, tag
);
686 dsl_dataset_disown(dsl_dataset_t
*ds
, void *tag
)
688 ASSERT3P(ds
->ds_owner
, ==, tag
);
689 ASSERT(ds
->ds_dbuf
!= NULL
);
691 mutex_enter(&ds
->ds_lock
);
693 mutex_exit(&ds
->ds_lock
);
694 dsl_dataset_long_rele(ds
, tag
);
695 dsl_dataset_rele(ds
, tag
);
699 dsl_dataset_tryown(dsl_dataset_t
*ds
, void *tag
)
701 boolean_t gotit
= FALSE
;
703 mutex_enter(&ds
->ds_lock
);
704 if (ds
->ds_owner
== NULL
&& !DS_IS_INCONSISTENT(ds
)) {
706 dsl_dataset_long_hold(ds
, tag
);
709 mutex_exit(&ds
->ds_lock
);
714 dsl_dataset_create_sync_dd(dsl_dir_t
*dd
, dsl_dataset_t
*origin
,
715 uint64_t flags
, dmu_tx_t
*tx
)
717 dsl_pool_t
*dp
= dd
->dd_pool
;
719 dsl_dataset_phys_t
*dsphys
;
721 objset_t
*mos
= dp
->dp_meta_objset
;
724 origin
= dp
->dp_origin_snap
;
726 ASSERT(origin
== NULL
|| origin
->ds_dir
->dd_pool
== dp
);
727 ASSERT(origin
== NULL
|| dsl_dataset_phys(origin
)->ds_num_children
> 0);
728 ASSERT(dmu_tx_is_syncing(tx
));
729 ASSERT(dsl_dir_phys(dd
)->dd_head_dataset_obj
== 0);
731 dsobj
= dmu_object_alloc(mos
, DMU_OT_DSL_DATASET
, 0,
732 DMU_OT_DSL_DATASET
, sizeof (dsl_dataset_phys_t
), tx
);
733 VERIFY0(dmu_bonus_hold(mos
, dsobj
, FTAG
, &dbuf
));
734 dmu_buf_will_dirty(dbuf
, tx
);
735 dsphys
= dbuf
->db_data
;
736 bzero(dsphys
, sizeof (dsl_dataset_phys_t
));
737 dsphys
->ds_dir_obj
= dd
->dd_object
;
738 dsphys
->ds_flags
= flags
;
739 dsphys
->ds_fsid_guid
= unique_create();
740 (void) random_get_pseudo_bytes((void*)&dsphys
->ds_guid
,
741 sizeof (dsphys
->ds_guid
));
742 dsphys
->ds_snapnames_zapobj
=
743 zap_create_norm(mos
, U8_TEXTPREP_TOUPPER
, DMU_OT_DSL_DS_SNAP_MAP
,
745 dsphys
->ds_creation_time
= gethrestime_sec();
746 dsphys
->ds_creation_txg
= tx
->tx_txg
== TXG_INITIAL
? 1 : tx
->tx_txg
;
748 if (origin
== NULL
) {
749 dsphys
->ds_deadlist_obj
= dsl_deadlist_alloc(mos
, tx
);
751 dsl_dataset_t
*ohds
; /* head of the origin snapshot */
753 dsphys
->ds_prev_snap_obj
= origin
->ds_object
;
754 dsphys
->ds_prev_snap_txg
=
755 dsl_dataset_phys(origin
)->ds_creation_txg
;
756 dsphys
->ds_referenced_bytes
=
757 dsl_dataset_phys(origin
)->ds_referenced_bytes
;
758 dsphys
->ds_compressed_bytes
=
759 dsl_dataset_phys(origin
)->ds_compressed_bytes
;
760 dsphys
->ds_uncompressed_bytes
=
761 dsl_dataset_phys(origin
)->ds_uncompressed_bytes
;
762 dsphys
->ds_bp
= dsl_dataset_phys(origin
)->ds_bp
;
765 * Inherit flags that describe the dataset's contents
766 * (INCONSISTENT) or properties (Case Insensitive).
768 dsphys
->ds_flags
|= dsl_dataset_phys(origin
)->ds_flags
&
769 (DS_FLAG_INCONSISTENT
| DS_FLAG_CI_DATASET
);
771 if (origin
->ds_large_blocks
)
772 dsl_dataset_activate_large_blocks_sync_impl(dsobj
, tx
);
774 dmu_buf_will_dirty(origin
->ds_dbuf
, tx
);
775 dsl_dataset_phys(origin
)->ds_num_children
++;
777 VERIFY0(dsl_dataset_hold_obj(dp
,
778 dsl_dir_phys(origin
->ds_dir
)->dd_head_dataset_obj
,
780 dsphys
->ds_deadlist_obj
= dsl_deadlist_clone(&ohds
->ds_deadlist
,
781 dsphys
->ds_prev_snap_txg
, dsphys
->ds_prev_snap_obj
, tx
);
782 dsl_dataset_rele(ohds
, FTAG
);
784 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_NEXT_CLONES
) {
785 if (dsl_dataset_phys(origin
)->ds_next_clones_obj
== 0) {
786 dsl_dataset_phys(origin
)->ds_next_clones_obj
=
788 DMU_OT_NEXT_CLONES
, DMU_OT_NONE
, 0, tx
);
790 VERIFY0(zap_add_int(mos
,
791 dsl_dataset_phys(origin
)->ds_next_clones_obj
,
795 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
796 dsl_dir_phys(dd
)->dd_origin_obj
= origin
->ds_object
;
797 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
798 if (dsl_dir_phys(origin
->ds_dir
)->dd_clones
== 0) {
799 dmu_buf_will_dirty(origin
->ds_dir
->dd_dbuf
, tx
);
800 dsl_dir_phys(origin
->ds_dir
)->dd_clones
=
802 DMU_OT_DSL_CLONES
, DMU_OT_NONE
, 0, tx
);
804 VERIFY0(zap_add_int(mos
,
805 dsl_dir_phys(origin
->ds_dir
)->dd_clones
,
810 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_UNIQUE_ACCURATE
)
811 dsphys
->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
813 dmu_buf_rele(dbuf
, FTAG
);
815 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
816 dsl_dir_phys(dd
)->dd_head_dataset_obj
= dsobj
;
822 dsl_dataset_zero_zil(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
826 VERIFY0(dmu_objset_from_ds(ds
, &os
));
827 bzero(&os
->os_zil_header
, sizeof (os
->os_zil_header
));
828 dsl_dataset_dirty(ds
, tx
);
832 dsl_dataset_create_sync(dsl_dir_t
*pdd
, const char *lastname
,
833 dsl_dataset_t
*origin
, uint64_t flags
, cred_t
*cr
, dmu_tx_t
*tx
)
835 dsl_pool_t
*dp
= pdd
->dd_pool
;
836 uint64_t dsobj
, ddobj
;
839 ASSERT(dmu_tx_is_syncing(tx
));
840 ASSERT(lastname
[0] != '@');
842 ddobj
= dsl_dir_create_sync(dp
, pdd
, lastname
, tx
);
843 VERIFY0(dsl_dir_hold_obj(dp
, ddobj
, lastname
, FTAG
, &dd
));
845 dsobj
= dsl_dataset_create_sync_dd(dd
, origin
,
846 flags
& ~DS_CREATE_FLAG_NODIRTY
, tx
);
848 dsl_deleg_set_create_perms(dd
, tx
, cr
);
851 * Since we're creating a new node we know it's a leaf, so we can
852 * initialize the counts if the limit feature is active.
854 if (spa_feature_is_active(dp
->dp_spa
, SPA_FEATURE_FS_SS_LIMIT
)) {
856 objset_t
*os
= dd
->dd_pool
->dp_meta_objset
;
858 dsl_dir_zapify(dd
, tx
);
859 VERIFY0(zap_add(os
, dd
->dd_object
, DD_FIELD_FILESYSTEM_COUNT
,
860 sizeof (cnt
), 1, &cnt
, tx
));
861 VERIFY0(zap_add(os
, dd
->dd_object
, DD_FIELD_SNAPSHOT_COUNT
,
862 sizeof (cnt
), 1, &cnt
, tx
));
865 dsl_dir_rele(dd
, FTAG
);
868 * If we are creating a clone, make sure we zero out any stale
869 * data from the origin snapshots zil header.
871 if (origin
!= NULL
&& !(flags
& DS_CREATE_FLAG_NODIRTY
)) {
874 VERIFY0(dsl_dataset_hold_obj(dp
, dsobj
, FTAG
, &ds
));
875 dsl_dataset_zero_zil(ds
, tx
);
876 dsl_dataset_rele(ds
, FTAG
);
883 * The unique space in the head dataset can be calculated by subtracting
884 * the space used in the most recent snapshot, that is still being used
885 * in this file system, from the space currently in use. To figure out
886 * the space in the most recent snapshot still in use, we need to take
887 * the total space used in the snapshot and subtract out the space that
888 * has been freed up since the snapshot was taken.
891 dsl_dataset_recalc_head_uniq(dsl_dataset_t
*ds
)
894 uint64_t dlused
, dlcomp
, dluncomp
;
896 ASSERT(!ds
->ds_is_snapshot
);
898 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0)
899 mrs_used
= dsl_dataset_phys(ds
->ds_prev
)->ds_referenced_bytes
;
903 dsl_deadlist_space(&ds
->ds_deadlist
, &dlused
, &dlcomp
, &dluncomp
);
905 ASSERT3U(dlused
, <=, mrs_used
);
906 dsl_dataset_phys(ds
)->ds_unique_bytes
=
907 dsl_dataset_phys(ds
)->ds_referenced_bytes
- (mrs_used
- dlused
);
909 if (spa_version(ds
->ds_dir
->dd_pool
->dp_spa
) >=
910 SPA_VERSION_UNIQUE_ACCURATE
)
911 dsl_dataset_phys(ds
)->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
915 dsl_dataset_remove_from_next_clones(dsl_dataset_t
*ds
, uint64_t obj
,
918 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
920 ASSERTV(uint64_t count
);
922 ASSERT(dsl_dataset_phys(ds
)->ds_num_children
>= 2);
923 err
= zap_remove_int(mos
, dsl_dataset_phys(ds
)->ds_next_clones_obj
,
926 * The err should not be ENOENT, but a bug in a previous version
927 * of the code could cause upgrade_clones_cb() to not set
928 * ds_next_snap_obj when it should, leading to a missing entry.
929 * If we knew that the pool was created after
930 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
931 * ENOENT. However, at least we can check that we don't have
932 * too many entries in the next_clones_obj even after failing to
937 ASSERT0(zap_count(mos
, dsl_dataset_phys(ds
)->ds_next_clones_obj
,
939 ASSERT3U(count
, <=, dsl_dataset_phys(ds
)->ds_num_children
- 2);
944 dsl_dataset_get_blkptr(dsl_dataset_t
*ds
)
946 return (&dsl_dataset_phys(ds
)->ds_bp
);
950 dsl_dataset_set_blkptr(dsl_dataset_t
*ds
, blkptr_t
*bp
, dmu_tx_t
*tx
)
952 ASSERT(dmu_tx_is_syncing(tx
));
953 /* If it's the meta-objset, set dp_meta_rootbp */
955 tx
->tx_pool
->dp_meta_rootbp
= *bp
;
957 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
958 dsl_dataset_phys(ds
)->ds_bp
= *bp
;
963 dsl_dataset_get_spa(dsl_dataset_t
*ds
)
965 return (ds
->ds_dir
->dd_pool
->dp_spa
);
969 dsl_dataset_dirty(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
973 if (ds
== NULL
) /* this is the meta-objset */
976 ASSERT(ds
->ds_objset
!= NULL
);
978 if (dsl_dataset_phys(ds
)->ds_next_snap_obj
!= 0)
979 panic("dirtying snapshot!");
981 dp
= ds
->ds_dir
->dd_pool
;
983 if (txg_list_add(&dp
->dp_dirty_datasets
, ds
, tx
->tx_txg
)) {
984 /* up the hold count until we can be written out */
985 dmu_buf_add_ref(ds
->ds_dbuf
, ds
);
990 dsl_dataset_is_dirty(dsl_dataset_t
*ds
)
994 for (t
= 0; t
< TXG_SIZE
; t
++) {
995 if (txg_list_member(&ds
->ds_dir
->dd_pool
->dp_dirty_datasets
,
1003 dsl_dataset_snapshot_reserve_space(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
1007 if (!dmu_tx_is_syncing(tx
))
1011 * If there's an fs-only reservation, any blocks that might become
1012 * owned by the snapshot dataset must be accommodated by space
1013 * outside of the reservation.
1015 ASSERT(ds
->ds_reserved
== 0 || DS_UNIQUE_IS_ACCURATE(ds
));
1016 asize
= MIN(dsl_dataset_phys(ds
)->ds_unique_bytes
, ds
->ds_reserved
);
1017 if (asize
> dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
))
1018 return (SET_ERROR(ENOSPC
));
1021 * Propagate any reserved space for this snapshot to other
1022 * snapshot checks in this sync group.
1025 dsl_dir_willuse_space(ds
->ds_dir
, asize
, tx
);
1030 typedef struct dsl_dataset_snapshot_arg
{
1031 nvlist_t
*ddsa_snaps
;
1032 nvlist_t
*ddsa_props
;
1033 nvlist_t
*ddsa_errors
;
1035 } dsl_dataset_snapshot_arg_t
;
1038 dsl_dataset_snapshot_check_impl(dsl_dataset_t
*ds
, const char *snapname
,
1039 dmu_tx_t
*tx
, boolean_t recv
, uint64_t cnt
, cred_t
*cr
)
1044 ds
->ds_trysnap_txg
= tx
->tx_txg
;
1046 if (!dmu_tx_is_syncing(tx
))
1050 * We don't allow multiple snapshots of the same txg. If there
1051 * is already one, try again.
1053 if (dsl_dataset_phys(ds
)->ds_prev_snap_txg
>= tx
->tx_txg
)
1054 return (SET_ERROR(EAGAIN
));
1057 * Check for conflicting snapshot name.
1059 error
= dsl_dataset_snap_lookup(ds
, snapname
, &value
);
1061 return (SET_ERROR(EEXIST
));
1062 if (error
!= ENOENT
)
1066 * We don't allow taking snapshots of inconsistent datasets, such as
1067 * those into which we are currently receiving. However, if we are
1068 * creating this snapshot as part of a receive, this check will be
1069 * executed atomically with respect to the completion of the receive
1070 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1071 * case we ignore this, knowing it will be fixed up for us shortly in
1072 * dmu_recv_end_sync().
1074 if (!recv
&& DS_IS_INCONSISTENT(ds
))
1075 return (SET_ERROR(EBUSY
));
1078 * Skip the check for temporary snapshots or if we have already checked
1079 * the counts in dsl_dataset_snapshot_check. This means we really only
1080 * check the count here when we're receiving a stream.
1082 if (cnt
!= 0 && cr
!= NULL
) {
1083 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, cnt
,
1084 ZFS_PROP_SNAPSHOT_LIMIT
, NULL
, cr
);
1089 error
= dsl_dataset_snapshot_reserve_space(ds
, tx
);
1097 dsl_dataset_snapshot_check(void *arg
, dmu_tx_t
*tx
)
1099 dsl_dataset_snapshot_arg_t
*ddsa
= arg
;
1100 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1105 * Pre-compute how many total new snapshots will be created for each
1106 * level in the tree and below. This is needed for validating the
1107 * snapshot limit when either taking a recursive snapshot or when
1108 * taking multiple snapshots.
1110 * The problem is that the counts are not actually adjusted when
1111 * we are checking, only when we finally sync. For a single snapshot,
1112 * this is easy, the count will increase by 1 at each node up the tree,
1113 * but its more complicated for the recursive/multiple snapshot case.
1115 * The dsl_fs_ss_limit_check function does recursively check the count
1116 * at each level up the tree but since it is validating each snapshot
1117 * independently we need to be sure that we are validating the complete
1118 * count for the entire set of snapshots. We do this by rolling up the
1119 * counts for each component of the name into an nvlist and then
1120 * checking each of those cases with the aggregated count.
1122 * This approach properly handles not only the recursive snapshot
1123 * case (where we get all of those on the ddsa_snaps list) but also
1124 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1125 * validate the limit on 'a' using a count of 2).
1127 * We validate the snapshot names in the third loop and only report
1130 if (dmu_tx_is_syncing(tx
)) {
1132 nvlist_t
*cnt_track
= NULL
;
1133 cnt_track
= fnvlist_alloc();
1135 nm
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
1137 /* Rollup aggregated counts into the cnt_track list */
1138 for (pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, NULL
);
1140 pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, pair
)) {
1144 (void) strlcpy(nm
, nvpair_name(pair
), MAXPATHLEN
);
1145 pdelim
= strchr(nm
, '@');
1151 if (nvlist_lookup_uint64(cnt_track
, nm
,
1153 /* update existing entry */
1154 fnvlist_add_uint64(cnt_track
, nm
,
1158 fnvlist_add_uint64(cnt_track
, nm
, 1);
1161 pdelim
= strrchr(nm
, '/');
1164 } while (pdelim
!= NULL
);
1167 kmem_free(nm
, MAXPATHLEN
);
1169 /* Check aggregated counts at each level */
1170 for (pair
= nvlist_next_nvpair(cnt_track
, NULL
);
1171 pair
!= NULL
; pair
= nvlist_next_nvpair(cnt_track
, pair
)) {
1177 name
= nvpair_name(pair
);
1178 cnt
= fnvpair_value_uint64(pair
);
1181 error
= dsl_dataset_hold(dp
, name
, FTAG
, &ds
);
1183 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, cnt
,
1184 ZFS_PROP_SNAPSHOT_LIMIT
, NULL
,
1186 dsl_dataset_rele(ds
, FTAG
);
1190 if (ddsa
->ddsa_errors
!= NULL
)
1191 fnvlist_add_int32(ddsa
->ddsa_errors
,
1194 /* only report one error for this check */
1198 nvlist_free(cnt_track
);
1201 for (pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, NULL
);
1202 pair
!= NULL
; pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, pair
)) {
1206 char dsname
[MAXNAMELEN
];
1208 name
= nvpair_name(pair
);
1209 if (strlen(name
) >= MAXNAMELEN
)
1210 error
= SET_ERROR(ENAMETOOLONG
);
1212 atp
= strchr(name
, '@');
1214 error
= SET_ERROR(EINVAL
);
1216 (void) strlcpy(dsname
, name
, atp
- name
+ 1);
1219 error
= dsl_dataset_hold(dp
, dsname
, FTAG
, &ds
);
1221 /* passing 0/NULL skips dsl_fs_ss_limit_check */
1222 error
= dsl_dataset_snapshot_check_impl(ds
,
1223 atp
+ 1, tx
, B_FALSE
, 0, NULL
);
1224 dsl_dataset_rele(ds
, FTAG
);
1228 if (ddsa
->ddsa_errors
!= NULL
) {
1229 fnvlist_add_int32(ddsa
->ddsa_errors
,
1240 dsl_dataset_snapshot_sync_impl(dsl_dataset_t
*ds
, const char *snapname
,
1243 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1245 dsl_dataset_phys_t
*dsphys
;
1246 uint64_t dsobj
, crtxg
;
1247 objset_t
*mos
= dp
->dp_meta_objset
;
1248 ASSERTV(static zil_header_t zero_zil
);
1249 ASSERTV(objset_t
*os
);
1251 ASSERT(RRW_WRITE_HELD(&dp
->dp_config_rwlock
));
1254 * If we are on an old pool, the zil must not be active, in which
1255 * case it will be zeroed. Usually zil_suspend() accomplishes this.
1257 ASSERT(spa_version(dmu_tx_pool(tx
)->dp_spa
) >= SPA_VERSION_FAST_SNAP
||
1258 dmu_objset_from_ds(ds
, &os
) != 0 ||
1259 bcmp(&os
->os_phys
->os_zil_header
, &zero_zil
,
1260 sizeof (zero_zil
)) == 0);
1262 dsl_fs_ss_count_adjust(ds
->ds_dir
, 1, DD_FIELD_SNAPSHOT_COUNT
, tx
);
1265 * The origin's ds_creation_txg has to be < TXG_INITIAL
1267 if (strcmp(snapname
, ORIGIN_DIR_NAME
) == 0)
1272 dsobj
= dmu_object_alloc(mos
, DMU_OT_DSL_DATASET
, 0,
1273 DMU_OT_DSL_DATASET
, sizeof (dsl_dataset_phys_t
), tx
);
1274 VERIFY0(dmu_bonus_hold(mos
, dsobj
, FTAG
, &dbuf
));
1275 dmu_buf_will_dirty(dbuf
, tx
);
1276 dsphys
= dbuf
->db_data
;
1277 bzero(dsphys
, sizeof (dsl_dataset_phys_t
));
1278 dsphys
->ds_dir_obj
= ds
->ds_dir
->dd_object
;
1279 dsphys
->ds_fsid_guid
= unique_create();
1280 (void) random_get_pseudo_bytes((void*)&dsphys
->ds_guid
,
1281 sizeof (dsphys
->ds_guid
));
1282 dsphys
->ds_prev_snap_obj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
1283 dsphys
->ds_prev_snap_txg
= dsl_dataset_phys(ds
)->ds_prev_snap_txg
;
1284 dsphys
->ds_next_snap_obj
= ds
->ds_object
;
1285 dsphys
->ds_num_children
= 1;
1286 dsphys
->ds_creation_time
= gethrestime_sec();
1287 dsphys
->ds_creation_txg
= crtxg
;
1288 dsphys
->ds_deadlist_obj
= dsl_dataset_phys(ds
)->ds_deadlist_obj
;
1289 dsphys
->ds_referenced_bytes
= dsl_dataset_phys(ds
)->ds_referenced_bytes
;
1290 dsphys
->ds_compressed_bytes
= dsl_dataset_phys(ds
)->ds_compressed_bytes
;
1291 dsphys
->ds_uncompressed_bytes
=
1292 dsl_dataset_phys(ds
)->ds_uncompressed_bytes
;
1293 dsphys
->ds_flags
= dsl_dataset_phys(ds
)->ds_flags
;
1294 dsphys
->ds_bp
= dsl_dataset_phys(ds
)->ds_bp
;
1295 dmu_buf_rele(dbuf
, FTAG
);
1297 if (ds
->ds_large_blocks
)
1298 dsl_dataset_activate_large_blocks_sync_impl(dsobj
, tx
);
1300 ASSERT3U(ds
->ds_prev
!= 0, ==,
1301 dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0);
1303 uint64_t next_clones_obj
=
1304 dsl_dataset_phys(ds
->ds_prev
)->ds_next_clones_obj
;
1305 ASSERT(dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
==
1307 dsl_dataset_phys(ds
->ds_prev
)->ds_num_children
> 1);
1308 if (dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
==
1310 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
1311 ASSERT3U(dsl_dataset_phys(ds
)->ds_prev_snap_txg
, ==,
1312 dsl_dataset_phys(ds
->ds_prev
)->ds_creation_txg
);
1313 dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
= dsobj
;
1314 } else if (next_clones_obj
!= 0) {
1315 dsl_dataset_remove_from_next_clones(ds
->ds_prev
,
1316 dsphys
->ds_next_snap_obj
, tx
);
1317 VERIFY0(zap_add_int(mos
,
1318 next_clones_obj
, dsobj
, tx
));
1323 * If we have a reference-reservation on this dataset, we will
1324 * need to increase the amount of refreservation being charged
1325 * since our unique space is going to zero.
1327 if (ds
->ds_reserved
) {
1329 ASSERT(DS_UNIQUE_IS_ACCURATE(ds
));
1330 delta
= MIN(dsl_dataset_phys(ds
)->ds_unique_bytes
,
1332 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_REFRSRV
,
1336 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1337 dsl_dataset_phys(ds
)->ds_deadlist_obj
=
1338 dsl_deadlist_clone(&ds
->ds_deadlist
, UINT64_MAX
,
1339 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, tx
);
1340 dsl_deadlist_close(&ds
->ds_deadlist
);
1341 dsl_deadlist_open(&ds
->ds_deadlist
, mos
,
1342 dsl_dataset_phys(ds
)->ds_deadlist_obj
);
1343 dsl_deadlist_add_key(&ds
->ds_deadlist
,
1344 dsl_dataset_phys(ds
)->ds_prev_snap_txg
, tx
);
1346 ASSERT3U(dsl_dataset_phys(ds
)->ds_prev_snap_txg
, <, tx
->tx_txg
);
1347 dsl_dataset_phys(ds
)->ds_prev_snap_obj
= dsobj
;
1348 dsl_dataset_phys(ds
)->ds_prev_snap_txg
= crtxg
;
1349 dsl_dataset_phys(ds
)->ds_unique_bytes
= 0;
1350 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_UNIQUE_ACCURATE
)
1351 dsl_dataset_phys(ds
)->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
1353 VERIFY0(zap_add(mos
, dsl_dataset_phys(ds
)->ds_snapnames_zapobj
,
1354 snapname
, 8, 1, &dsobj
, tx
));
1357 dsl_dataset_rele(ds
->ds_prev
, ds
);
1358 VERIFY0(dsl_dataset_hold_obj(dp
,
1359 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, ds
, &ds
->ds_prev
));
1361 dsl_scan_ds_snapshotted(ds
, tx
);
1363 dsl_dir_snap_cmtime_update(ds
->ds_dir
);
1365 spa_history_log_internal_ds(ds
->ds_prev
, "snapshot", tx
, "");
1369 dsl_dataset_snapshot_sync(void *arg
, dmu_tx_t
*tx
)
1371 dsl_dataset_snapshot_arg_t
*ddsa
= arg
;
1372 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1375 for (pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, NULL
);
1376 pair
!= NULL
; pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, pair
)) {
1379 char dsname
[MAXNAMELEN
];
1381 name
= nvpair_name(pair
);
1382 atp
= strchr(name
, '@');
1383 (void) strlcpy(dsname
, name
, atp
- name
+ 1);
1384 VERIFY0(dsl_dataset_hold(dp
, dsname
, FTAG
, &ds
));
1386 dsl_dataset_snapshot_sync_impl(ds
, atp
+ 1, tx
);
1387 if (ddsa
->ddsa_props
!= NULL
) {
1388 dsl_props_set_sync_impl(ds
->ds_prev
,
1389 ZPROP_SRC_LOCAL
, ddsa
->ddsa_props
, tx
);
1391 zvol_create_minors(dp
->dp_spa
, nvpair_name(pair
), B_TRUE
);
1392 dsl_dataset_rele(ds
, FTAG
);
1397 * The snapshots must all be in the same pool.
1398 * All-or-nothing: if there are any failures, nothing will be modified.
1401 dsl_dataset_snapshot(nvlist_t
*snaps
, nvlist_t
*props
, nvlist_t
*errors
)
1403 dsl_dataset_snapshot_arg_t ddsa
;
1405 boolean_t needsuspend
;
1409 nvlist_t
*suspended
= NULL
;
1411 pair
= nvlist_next_nvpair(snaps
, NULL
);
1414 firstname
= nvpair_name(pair
);
1416 error
= spa_open(firstname
, &spa
, FTAG
);
1419 needsuspend
= (spa_version(spa
) < SPA_VERSION_FAST_SNAP
);
1420 spa_close(spa
, FTAG
);
1423 suspended
= fnvlist_alloc();
1424 for (pair
= nvlist_next_nvpair(snaps
, NULL
); pair
!= NULL
;
1425 pair
= nvlist_next_nvpair(snaps
, pair
)) {
1426 char fsname
[MAXNAMELEN
];
1427 char *snapname
= nvpair_name(pair
);
1431 atp
= strchr(snapname
, '@');
1433 error
= SET_ERROR(EINVAL
);
1436 (void) strlcpy(fsname
, snapname
, atp
- snapname
+ 1);
1438 error
= zil_suspend(fsname
, &cookie
);
1441 fnvlist_add_uint64(suspended
, fsname
,
1446 ddsa
.ddsa_snaps
= snaps
;
1447 ddsa
.ddsa_props
= props
;
1448 ddsa
.ddsa_errors
= errors
;
1449 ddsa
.ddsa_cr
= CRED();
1452 error
= dsl_sync_task(firstname
, dsl_dataset_snapshot_check
,
1453 dsl_dataset_snapshot_sync
, &ddsa
,
1454 fnvlist_num_pairs(snaps
) * 3, ZFS_SPACE_CHECK_NORMAL
);
1457 if (suspended
!= NULL
) {
1458 for (pair
= nvlist_next_nvpair(suspended
, NULL
); pair
!= NULL
;
1459 pair
= nvlist_next_nvpair(suspended
, pair
)) {
1460 zil_resume((void *)(uintptr_t)
1461 fnvpair_value_uint64(pair
));
1463 fnvlist_free(suspended
);
1469 typedef struct dsl_dataset_snapshot_tmp_arg
{
1470 const char *ddsta_fsname
;
1471 const char *ddsta_snapname
;
1472 minor_t ddsta_cleanup_minor
;
1473 const char *ddsta_htag
;
1474 } dsl_dataset_snapshot_tmp_arg_t
;
1477 dsl_dataset_snapshot_tmp_check(void *arg
, dmu_tx_t
*tx
)
1479 dsl_dataset_snapshot_tmp_arg_t
*ddsta
= arg
;
1480 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1484 error
= dsl_dataset_hold(dp
, ddsta
->ddsta_fsname
, FTAG
, &ds
);
1488 /* NULL cred means no limit check for tmp snapshot */
1489 error
= dsl_dataset_snapshot_check_impl(ds
, ddsta
->ddsta_snapname
,
1490 tx
, B_FALSE
, 0, NULL
);
1492 dsl_dataset_rele(ds
, FTAG
);
1496 if (spa_version(dp
->dp_spa
) < SPA_VERSION_USERREFS
) {
1497 dsl_dataset_rele(ds
, FTAG
);
1498 return (SET_ERROR(ENOTSUP
));
1500 error
= dsl_dataset_user_hold_check_one(NULL
, ddsta
->ddsta_htag
,
1503 dsl_dataset_rele(ds
, FTAG
);
1507 dsl_dataset_rele(ds
, FTAG
);
1512 dsl_dataset_snapshot_tmp_sync(void *arg
, dmu_tx_t
*tx
)
1514 dsl_dataset_snapshot_tmp_arg_t
*ddsta
= arg
;
1515 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1518 VERIFY0(dsl_dataset_hold(dp
, ddsta
->ddsta_fsname
, FTAG
, &ds
));
1520 dsl_dataset_snapshot_sync_impl(ds
, ddsta
->ddsta_snapname
, tx
);
1521 dsl_dataset_user_hold_sync_one(ds
->ds_prev
, ddsta
->ddsta_htag
,
1522 ddsta
->ddsta_cleanup_minor
, gethrestime_sec(), tx
);
1523 dsl_destroy_snapshot_sync_impl(ds
->ds_prev
, B_TRUE
, tx
);
1525 dsl_dataset_rele(ds
, FTAG
);
1529 dsl_dataset_snapshot_tmp(const char *fsname
, const char *snapname
,
1530 minor_t cleanup_minor
, const char *htag
)
1532 dsl_dataset_snapshot_tmp_arg_t ddsta
;
1535 boolean_t needsuspend
;
1538 ddsta
.ddsta_fsname
= fsname
;
1539 ddsta
.ddsta_snapname
= snapname
;
1540 ddsta
.ddsta_cleanup_minor
= cleanup_minor
;
1541 ddsta
.ddsta_htag
= htag
;
1543 error
= spa_open(fsname
, &spa
, FTAG
);
1546 needsuspend
= (spa_version(spa
) < SPA_VERSION_FAST_SNAP
);
1547 spa_close(spa
, FTAG
);
1550 error
= zil_suspend(fsname
, &cookie
);
1555 error
= dsl_sync_task(fsname
, dsl_dataset_snapshot_tmp_check
,
1556 dsl_dataset_snapshot_tmp_sync
, &ddsta
, 3, ZFS_SPACE_CHECK_RESERVED
);
1565 dsl_dataset_sync(dsl_dataset_t
*ds
, zio_t
*zio
, dmu_tx_t
*tx
)
1567 ASSERT(dmu_tx_is_syncing(tx
));
1568 ASSERT(ds
->ds_objset
!= NULL
);
1569 ASSERT(dsl_dataset_phys(ds
)->ds_next_snap_obj
== 0);
1572 * in case we had to change ds_fsid_guid when we opened it,
1575 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1576 dsl_dataset_phys(ds
)->ds_fsid_guid
= ds
->ds_fsid_guid
;
1578 dmu_objset_sync(ds
->ds_objset
, zio
, tx
);
1580 if (ds
->ds_need_large_blocks
&& !ds
->ds_large_blocks
) {
1581 dsl_dataset_activate_large_blocks_sync_impl(ds
->ds_object
, tx
);
1582 ds
->ds_large_blocks
= B_TRUE
;
1587 get_clones_stat(dsl_dataset_t
*ds
, nvlist_t
*nv
)
1590 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
1593 nvlist_t
*propval
= fnvlist_alloc();
1594 nvlist_t
*val
= fnvlist_alloc();
1596 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
1599 * There may be missing entries in ds_next_clones_obj
1600 * due to a bug in a previous version of the code.
1601 * Only trust it if it has the right number of entries.
1603 if (dsl_dataset_phys(ds
)->ds_next_clones_obj
!= 0) {
1604 VERIFY0(zap_count(mos
, dsl_dataset_phys(ds
)->ds_next_clones_obj
,
1607 if (count
!= dsl_dataset_phys(ds
)->ds_num_children
- 1)
1609 for (zap_cursor_init(&zc
, mos
,
1610 dsl_dataset_phys(ds
)->ds_next_clones_obj
);
1611 zap_cursor_retrieve(&zc
, &za
) == 0;
1612 zap_cursor_advance(&zc
)) {
1613 dsl_dataset_t
*clone
;
1614 char buf
[ZFS_MAXNAMELEN
];
1615 VERIFY0(dsl_dataset_hold_obj(ds
->ds_dir
->dd_pool
,
1616 za
.za_first_integer
, FTAG
, &clone
));
1617 dsl_dir_name(clone
->ds_dir
, buf
);
1618 fnvlist_add_boolean(val
, buf
);
1619 dsl_dataset_rele(clone
, FTAG
);
1621 zap_cursor_fini(&zc
);
1622 fnvlist_add_nvlist(propval
, ZPROP_VALUE
, val
);
1623 fnvlist_add_nvlist(nv
, zfs_prop_to_name(ZFS_PROP_CLONES
), propval
);
1626 nvlist_free(propval
);
1630 dsl_dataset_stats(dsl_dataset_t
*ds
, nvlist_t
*nv
)
1632 uint64_t refd
, avail
, uobjs
, aobjs
, ratio
;
1633 ASSERTV(dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
);
1635 ASSERT(dsl_pool_config_held(dp
));
1637 ratio
= dsl_dataset_phys(ds
)->ds_compressed_bytes
== 0 ? 100 :
1638 (dsl_dataset_phys(ds
)->ds_uncompressed_bytes
* 100 /
1639 dsl_dataset_phys(ds
)->ds_compressed_bytes
);
1641 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFRATIO
, ratio
);
1642 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_LOGICALREFERENCED
,
1643 dsl_dataset_phys(ds
)->ds_uncompressed_bytes
);
1645 if (ds
->ds_is_snapshot
) {
1646 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_COMPRESSRATIO
, ratio
);
1647 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USED
,
1648 dsl_dataset_phys(ds
)->ds_unique_bytes
);
1649 get_clones_stat(ds
, nv
);
1651 dsl_dir_stats(ds
->ds_dir
, nv
);
1654 dsl_dataset_space(ds
, &refd
, &avail
, &uobjs
, &aobjs
);
1655 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_AVAILABLE
, avail
);
1656 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFERENCED
, refd
);
1658 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_CREATION
,
1659 dsl_dataset_phys(ds
)->ds_creation_time
);
1660 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_CREATETXG
,
1661 dsl_dataset_phys(ds
)->ds_creation_txg
);
1662 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFQUOTA
,
1664 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFRESERVATION
,
1666 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_GUID
,
1667 dsl_dataset_phys(ds
)->ds_guid
);
1668 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_UNIQUE
,
1669 dsl_dataset_phys(ds
)->ds_unique_bytes
);
1670 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_OBJSETID
,
1672 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USERREFS
,
1674 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_DEFER_DESTROY
,
1675 DS_IS_DEFER_DESTROY(ds
) ? 1 : 0);
1677 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0) {
1678 uint64_t written
, comp
, uncomp
;
1679 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1680 dsl_dataset_t
*prev
;
1683 err
= dsl_dataset_hold_obj(dp
,
1684 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, FTAG
, &prev
);
1686 err
= dsl_dataset_space_written(prev
, ds
, &written
,
1688 dsl_dataset_rele(prev
, FTAG
);
1690 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_WRITTEN
,
1699 dsl_dataset_fast_stat(dsl_dataset_t
*ds
, dmu_objset_stats_t
*stat
)
1701 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1702 ASSERT(dsl_pool_config_held(dp
));
1704 stat
->dds_creation_txg
= dsl_dataset_phys(ds
)->ds_creation_txg
;
1705 stat
->dds_inconsistent
=
1706 dsl_dataset_phys(ds
)->ds_flags
& DS_FLAG_INCONSISTENT
;
1707 stat
->dds_guid
= dsl_dataset_phys(ds
)->ds_guid
;
1708 stat
->dds_origin
[0] = '\0';
1709 if (ds
->ds_is_snapshot
) {
1710 stat
->dds_is_snapshot
= B_TRUE
;
1711 stat
->dds_num_clones
=
1712 dsl_dataset_phys(ds
)->ds_num_children
- 1;
1714 stat
->dds_is_snapshot
= B_FALSE
;
1715 stat
->dds_num_clones
= 0;
1717 if (dsl_dir_is_clone(ds
->ds_dir
)) {
1720 VERIFY0(dsl_dataset_hold_obj(dp
,
1721 dsl_dir_phys(ds
->ds_dir
)->dd_origin_obj
,
1723 dsl_dataset_name(ods
, stat
->dds_origin
);
1724 dsl_dataset_rele(ods
, FTAG
);
1730 dsl_dataset_fsid_guid(dsl_dataset_t
*ds
)
1732 return (ds
->ds_fsid_guid
);
1736 dsl_dataset_space(dsl_dataset_t
*ds
,
1737 uint64_t *refdbytesp
, uint64_t *availbytesp
,
1738 uint64_t *usedobjsp
, uint64_t *availobjsp
)
1740 *refdbytesp
= dsl_dataset_phys(ds
)->ds_referenced_bytes
;
1741 *availbytesp
= dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
);
1742 if (ds
->ds_reserved
> dsl_dataset_phys(ds
)->ds_unique_bytes
)
1744 ds
->ds_reserved
- dsl_dataset_phys(ds
)->ds_unique_bytes
;
1745 if (ds
->ds_quota
!= 0) {
1747 * Adjust available bytes according to refquota
1749 if (*refdbytesp
< ds
->ds_quota
)
1750 *availbytesp
= MIN(*availbytesp
,
1751 ds
->ds_quota
- *refdbytesp
);
1755 *usedobjsp
= BP_GET_FILL(&dsl_dataset_phys(ds
)->ds_bp
);
1756 *availobjsp
= DN_MAX_OBJECT
- *usedobjsp
;
1760 dsl_dataset_modified_since_snap(dsl_dataset_t
*ds
, dsl_dataset_t
*snap
)
1762 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
1765 if (dsl_dataset_phys(ds
)->ds_bp
.blk_birth
>
1766 dsl_dataset_phys(snap
)->ds_creation_txg
) {
1767 objset_t
*os
, *os_snap
;
1769 * It may be that only the ZIL differs, because it was
1770 * reset in the head. Don't count that as being
1773 if (dmu_objset_from_ds(ds
, &os
) != 0)
1775 if (dmu_objset_from_ds(snap
, &os_snap
) != 0)
1777 return (bcmp(&os
->os_phys
->os_meta_dnode
,
1778 &os_snap
->os_phys
->os_meta_dnode
,
1779 sizeof (os
->os_phys
->os_meta_dnode
)) != 0);
1784 typedef struct dsl_dataset_rename_snapshot_arg
{
1785 const char *ddrsa_fsname
;
1786 const char *ddrsa_oldsnapname
;
1787 const char *ddrsa_newsnapname
;
1788 boolean_t ddrsa_recursive
;
1790 } dsl_dataset_rename_snapshot_arg_t
;
1794 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t
*dp
,
1795 dsl_dataset_t
*hds
, void *arg
)
1797 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
1801 error
= dsl_dataset_snap_lookup(hds
, ddrsa
->ddrsa_oldsnapname
, &val
);
1803 /* ignore nonexistent snapshots */
1804 return (error
== ENOENT
? 0 : error
);
1807 /* new name should not exist */
1808 error
= dsl_dataset_snap_lookup(hds
, ddrsa
->ddrsa_newsnapname
, &val
);
1810 error
= SET_ERROR(EEXIST
);
1811 else if (error
== ENOENT
)
1814 /* dataset name + 1 for the "@" + the new snapshot name must fit */
1815 if (dsl_dir_namelen(hds
->ds_dir
) + 1 +
1816 strlen(ddrsa
->ddrsa_newsnapname
) >= MAXNAMELEN
)
1817 error
= SET_ERROR(ENAMETOOLONG
);
1823 dsl_dataset_rename_snapshot_check(void *arg
, dmu_tx_t
*tx
)
1825 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
1826 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1830 error
= dsl_dataset_hold(dp
, ddrsa
->ddrsa_fsname
, FTAG
, &hds
);
1834 if (ddrsa
->ddrsa_recursive
) {
1835 error
= dmu_objset_find_dp(dp
, hds
->ds_dir
->dd_object
,
1836 dsl_dataset_rename_snapshot_check_impl
, ddrsa
,
1839 error
= dsl_dataset_rename_snapshot_check_impl(dp
, hds
, ddrsa
);
1841 dsl_dataset_rele(hds
, FTAG
);
1846 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t
*dp
,
1847 dsl_dataset_t
*hds
, void *arg
)
1849 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
1852 dmu_tx_t
*tx
= ddrsa
->ddrsa_tx
;
1855 error
= dsl_dataset_snap_lookup(hds
, ddrsa
->ddrsa_oldsnapname
, &val
);
1856 ASSERT(error
== 0 || error
== ENOENT
);
1857 if (error
== ENOENT
) {
1858 /* ignore nonexistent snapshots */
1862 VERIFY0(dsl_dataset_hold_obj(dp
, val
, FTAG
, &ds
));
1864 /* log before we change the name */
1865 spa_history_log_internal_ds(ds
, "rename", tx
,
1866 "-> @%s", ddrsa
->ddrsa_newsnapname
);
1868 VERIFY0(dsl_dataset_snap_remove(hds
, ddrsa
->ddrsa_oldsnapname
, tx
,
1870 mutex_enter(&ds
->ds_lock
);
1871 (void) strcpy(ds
->ds_snapname
, ddrsa
->ddrsa_newsnapname
);
1872 mutex_exit(&ds
->ds_lock
);
1873 VERIFY0(zap_add(dp
->dp_meta_objset
,
1874 dsl_dataset_phys(hds
)->ds_snapnames_zapobj
,
1875 ds
->ds_snapname
, 8, 1, &ds
->ds_object
, tx
));
1876 zvol_rename_minors(dp
->dp_spa
, ddrsa
->ddrsa_oldsnapname
,
1877 ddrsa
->ddrsa_newsnapname
, B_TRUE
);
1879 dsl_dataset_rele(ds
, FTAG
);
1884 dsl_dataset_rename_snapshot_sync(void *arg
, dmu_tx_t
*tx
)
1886 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
1887 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1890 VERIFY0(dsl_dataset_hold(dp
, ddrsa
->ddrsa_fsname
, FTAG
, &hds
));
1891 ddrsa
->ddrsa_tx
= tx
;
1892 if (ddrsa
->ddrsa_recursive
) {
1893 VERIFY0(dmu_objset_find_dp(dp
, hds
->ds_dir
->dd_object
,
1894 dsl_dataset_rename_snapshot_sync_impl
, ddrsa
,
1897 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp
, hds
, ddrsa
));
1899 dsl_dataset_rele(hds
, FTAG
);
1903 dsl_dataset_rename_snapshot(const char *fsname
,
1904 const char *oldsnapname
, const char *newsnapname
, boolean_t recursive
)
1906 dsl_dataset_rename_snapshot_arg_t ddrsa
;
1908 ddrsa
.ddrsa_fsname
= fsname
;
1909 ddrsa
.ddrsa_oldsnapname
= oldsnapname
;
1910 ddrsa
.ddrsa_newsnapname
= newsnapname
;
1911 ddrsa
.ddrsa_recursive
= recursive
;
1913 return (dsl_sync_task(fsname
, dsl_dataset_rename_snapshot_check
,
1914 dsl_dataset_rename_snapshot_sync
, &ddrsa
,
1915 1, ZFS_SPACE_CHECK_RESERVED
));
1919 * If we're doing an ownership handoff, we need to make sure that there is
1920 * only one long hold on the dataset. We're not allowed to change anything here
1921 * so we don't permanently release the long hold or regular hold here. We want
1922 * to do this only when syncing to avoid the dataset unexpectedly going away
1923 * when we release the long hold.
1926 dsl_dataset_handoff_check(dsl_dataset_t
*ds
, void *owner
, dmu_tx_t
*tx
)
1930 if (!dmu_tx_is_syncing(tx
))
1933 if (owner
!= NULL
) {
1934 VERIFY3P(ds
->ds_owner
, ==, owner
);
1935 dsl_dataset_long_rele(ds
, owner
);
1938 held
= dsl_dataset_long_held(ds
);
1941 dsl_dataset_long_hold(ds
, owner
);
1944 return (SET_ERROR(EBUSY
));
1949 typedef struct dsl_dataset_rollback_arg
{
1950 const char *ddra_fsname
;
1952 nvlist_t
*ddra_result
;
1953 } dsl_dataset_rollback_arg_t
;
1956 dsl_dataset_rollback_check(void *arg
, dmu_tx_t
*tx
)
1958 dsl_dataset_rollback_arg_t
*ddra
= arg
;
1959 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1961 int64_t unused_refres_delta
;
1964 nvlist_t
*proprequest
, *bookmarks
;
1966 error
= dsl_dataset_hold(dp
, ddra
->ddra_fsname
, FTAG
, &ds
);
1970 /* must not be a snapshot */
1971 if (ds
->ds_is_snapshot
) {
1972 dsl_dataset_rele(ds
, FTAG
);
1973 return (SET_ERROR(EINVAL
));
1976 /* must have a most recent snapshot */
1977 if (dsl_dataset_phys(ds
)->ds_prev_snap_txg
< TXG_INITIAL
) {
1978 dsl_dataset_rele(ds
, FTAG
);
1979 return (SET_ERROR(EINVAL
));
1982 /* must not have any bookmarks after the most recent snapshot */
1983 proprequest
= fnvlist_alloc();
1984 fnvlist_add_boolean(proprequest
, zfs_prop_to_name(ZFS_PROP_CREATETXG
));
1985 bookmarks
= fnvlist_alloc();
1986 error
= dsl_get_bookmarks_impl(ds
, proprequest
, bookmarks
);
1987 fnvlist_free(proprequest
);
1990 for (pair
= nvlist_next_nvpair(bookmarks
, NULL
);
1991 pair
!= NULL
; pair
= nvlist_next_nvpair(bookmarks
, pair
)) {
1993 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair
),
1994 zfs_prop_to_name(ZFS_PROP_CREATETXG
));
1995 uint64_t createtxg
= fnvlist_lookup_uint64(valuenv
, "value");
1996 if (createtxg
> dsl_dataset_phys(ds
)->ds_prev_snap_txg
) {
1997 fnvlist_free(bookmarks
);
1998 dsl_dataset_rele(ds
, FTAG
);
1999 return (SET_ERROR(EEXIST
));
2002 fnvlist_free(bookmarks
);
2004 error
= dsl_dataset_handoff_check(ds
, ddra
->ddra_owner
, tx
);
2006 dsl_dataset_rele(ds
, FTAG
);
2011 * Check if the snap we are rolling back to uses more than
2014 if (ds
->ds_quota
!= 0 &&
2015 dsl_dataset_phys(ds
->ds_prev
)->ds_referenced_bytes
> ds
->ds_quota
) {
2016 dsl_dataset_rele(ds
, FTAG
);
2017 return (SET_ERROR(EDQUOT
));
2021 * When we do the clone swap, we will temporarily use more space
2022 * due to the refreservation (the head will no longer have any
2023 * unique space, so the entire amount of the refreservation will need
2024 * to be free). We will immediately destroy the clone, freeing
2025 * this space, but the freeing happens over many txg's.
2027 unused_refres_delta
= (int64_t)MIN(ds
->ds_reserved
,
2028 dsl_dataset_phys(ds
)->ds_unique_bytes
);
2030 if (unused_refres_delta
> 0 &&
2031 unused_refres_delta
>
2032 dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
)) {
2033 dsl_dataset_rele(ds
, FTAG
);
2034 return (SET_ERROR(ENOSPC
));
2037 dsl_dataset_rele(ds
, FTAG
);
2042 dsl_dataset_rollback_sync(void *arg
, dmu_tx_t
*tx
)
2044 dsl_dataset_rollback_arg_t
*ddra
= arg
;
2045 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2046 dsl_dataset_t
*ds
, *clone
;
2048 char namebuf
[ZFS_MAXNAMELEN
];
2050 VERIFY0(dsl_dataset_hold(dp
, ddra
->ddra_fsname
, FTAG
, &ds
));
2052 dsl_dataset_name(ds
->ds_prev
, namebuf
);
2053 fnvlist_add_string(ddra
->ddra_result
, "target", namebuf
);
2055 cloneobj
= dsl_dataset_create_sync(ds
->ds_dir
, "%rollback",
2056 ds
->ds_prev
, DS_CREATE_FLAG_NODIRTY
, kcred
, tx
);
2058 VERIFY0(dsl_dataset_hold_obj(dp
, cloneobj
, FTAG
, &clone
));
2060 dsl_dataset_clone_swap_sync_impl(clone
, ds
, tx
);
2061 dsl_dataset_zero_zil(ds
, tx
);
2063 dsl_destroy_head_sync_impl(clone
, tx
);
2065 dsl_dataset_rele(clone
, FTAG
);
2066 dsl_dataset_rele(ds
, FTAG
);
2070 * Rolls back the given filesystem or volume to the most recent snapshot.
2071 * The name of the most recent snapshot will be returned under key "target"
2072 * in the result nvlist.
2075 * - The existing dataset MUST be owned by the specified owner at entry
2076 * - Upon return, dataset will still be held by the same owner, whether we
2079 * This mode is required any time the existing filesystem is mounted. See
2080 * notes above zfs_suspend_fs() for further details.
2083 dsl_dataset_rollback(const char *fsname
, void *owner
, nvlist_t
*result
)
2085 dsl_dataset_rollback_arg_t ddra
;
2087 ddra
.ddra_fsname
= fsname
;
2088 ddra
.ddra_owner
= owner
;
2089 ddra
.ddra_result
= result
;
2091 return (dsl_sync_task(fsname
, dsl_dataset_rollback_check
,
2092 dsl_dataset_rollback_sync
, &ddra
,
2093 1, ZFS_SPACE_CHECK_RESERVED
));
2096 struct promotenode
{
2101 typedef struct dsl_dataset_promote_arg
{
2102 const char *ddpa_clonename
;
2103 dsl_dataset_t
*ddpa_clone
;
2104 list_t shared_snaps
, origin_snaps
, clone_snaps
;
2105 dsl_dataset_t
*origin_origin
; /* origin of the origin */
2106 uint64_t used
, comp
, uncomp
, unique
, cloneusedsnap
, originusedsnap
;
2109 } dsl_dataset_promote_arg_t
;
2111 static int snaplist_space(list_t
*l
, uint64_t mintxg
, uint64_t *spacep
);
2112 static int promote_hold(dsl_dataset_promote_arg_t
*ddpa
, dsl_pool_t
*dp
,
2114 static void promote_rele(dsl_dataset_promote_arg_t
*ddpa
, void *tag
);
2117 dsl_dataset_promote_check(void *arg
, dmu_tx_t
*tx
)
2119 dsl_dataset_promote_arg_t
*ddpa
= arg
;
2120 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2122 struct promotenode
*snap
;
2123 dsl_dataset_t
*origin_ds
;
2127 size_t max_snap_len
;
2129 err
= promote_hold(ddpa
, dp
, FTAG
);
2133 hds
= ddpa
->ddpa_clone
;
2134 max_snap_len
= MAXNAMELEN
- strlen(ddpa
->ddpa_clonename
) - 1;
2136 if (dsl_dataset_phys(hds
)->ds_flags
& DS_FLAG_NOPROMOTE
) {
2137 promote_rele(ddpa
, FTAG
);
2138 return (SET_ERROR(EXDEV
));
2142 * Compute and check the amount of space to transfer. Since this is
2143 * so expensive, don't do the preliminary check.
2145 if (!dmu_tx_is_syncing(tx
)) {
2146 promote_rele(ddpa
, FTAG
);
2150 snap
= list_head(&ddpa
->shared_snaps
);
2151 origin_ds
= snap
->ds
;
2153 /* compute origin's new unique space */
2154 snap
= list_tail(&ddpa
->clone_snaps
);
2155 ASSERT3U(dsl_dataset_phys(snap
->ds
)->ds_prev_snap_obj
, ==,
2156 origin_ds
->ds_object
);
2157 dsl_deadlist_space_range(&snap
->ds
->ds_deadlist
,
2158 dsl_dataset_phys(origin_ds
)->ds_prev_snap_txg
, UINT64_MAX
,
2159 &ddpa
->unique
, &unused
, &unused
);
2162 * Walk the snapshots that we are moving
2164 * Compute space to transfer. Consider the incremental changes
2165 * to used by each snapshot:
2166 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2167 * So each snapshot gave birth to:
2168 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2169 * So a sequence would look like:
2170 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2171 * Which simplifies to:
2172 * uN + kN + kN-1 + ... + k1 + k0
2173 * Note however, if we stop before we reach the ORIGIN we get:
2174 * uN + kN + kN-1 + ... + kM - uM-1
2177 ddpa
->used
= dsl_dataset_phys(origin_ds
)->ds_referenced_bytes
;
2178 ddpa
->comp
= dsl_dataset_phys(origin_ds
)->ds_compressed_bytes
;
2179 ddpa
->uncomp
= dsl_dataset_phys(origin_ds
)->ds_uncompressed_bytes
;
2180 for (snap
= list_head(&ddpa
->shared_snaps
); snap
;
2181 snap
= list_next(&ddpa
->shared_snaps
, snap
)) {
2182 uint64_t val
, dlused
, dlcomp
, dluncomp
;
2183 dsl_dataset_t
*ds
= snap
->ds
;
2188 * If there are long holds, we won't be able to evict
2191 if (dsl_dataset_long_held(ds
)) {
2192 err
= SET_ERROR(EBUSY
);
2196 /* Check that the snapshot name does not conflict */
2197 VERIFY0(dsl_dataset_get_snapname(ds
));
2198 if (strlen(ds
->ds_snapname
) >= max_snap_len
) {
2199 err
= SET_ERROR(ENAMETOOLONG
);
2202 err
= dsl_dataset_snap_lookup(hds
, ds
->ds_snapname
, &val
);
2204 (void) strcpy(ddpa
->err_ds
, snap
->ds
->ds_snapname
);
2205 err
= SET_ERROR(EEXIST
);
2211 /* The very first snapshot does not have a deadlist */
2212 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
== 0)
2215 dsl_deadlist_space(&ds
->ds_deadlist
,
2216 &dlused
, &dlcomp
, &dluncomp
);
2217 ddpa
->used
+= dlused
;
2218 ddpa
->comp
+= dlcomp
;
2219 ddpa
->uncomp
+= dluncomp
;
2223 * If we are a clone of a clone then we never reached ORIGIN,
2224 * so we need to subtract out the clone origin's used space.
2226 if (ddpa
->origin_origin
) {
2228 dsl_dataset_phys(ddpa
->origin_origin
)->ds_referenced_bytes
;
2230 dsl_dataset_phys(ddpa
->origin_origin
)->ds_compressed_bytes
;
2232 dsl_dataset_phys(ddpa
->origin_origin
)->
2233 ds_uncompressed_bytes
;
2236 /* Check that there is enough space and limit headroom here */
2237 err
= dsl_dir_transfer_possible(origin_ds
->ds_dir
, hds
->ds_dir
,
2238 0, ss_mv_cnt
, ddpa
->used
, ddpa
->cr
);
2243 * Compute the amounts of space that will be used by snapshots
2244 * after the promotion (for both origin and clone). For each,
2245 * it is the amount of space that will be on all of their
2246 * deadlists (that was not born before their new origin).
2248 if (dsl_dir_phys(hds
->ds_dir
)->dd_flags
& DD_FLAG_USED_BREAKDOWN
) {
2252 * Note, typically this will not be a clone of a clone,
2253 * so dd_origin_txg will be < TXG_INITIAL, so
2254 * these snaplist_space() -> dsl_deadlist_space_range()
2255 * calls will be fast because they do not have to
2256 * iterate over all bps.
2258 snap
= list_head(&ddpa
->origin_snaps
);
2259 err
= snaplist_space(&ddpa
->shared_snaps
,
2260 snap
->ds
->ds_dir
->dd_origin_txg
, &ddpa
->cloneusedsnap
);
2264 err
= snaplist_space(&ddpa
->clone_snaps
,
2265 snap
->ds
->ds_dir
->dd_origin_txg
, &space
);
2268 ddpa
->cloneusedsnap
+= space
;
2270 if (dsl_dir_phys(origin_ds
->ds_dir
)->dd_flags
&
2271 DD_FLAG_USED_BREAKDOWN
) {
2272 err
= snaplist_space(&ddpa
->origin_snaps
,
2273 dsl_dataset_phys(origin_ds
)->ds_creation_txg
,
2274 &ddpa
->originusedsnap
);
2280 promote_rele(ddpa
, FTAG
);
2285 dsl_dataset_promote_sync(void *arg
, dmu_tx_t
*tx
)
2287 dsl_dataset_promote_arg_t
*ddpa
= arg
;
2288 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2290 struct promotenode
*snap
;
2291 dsl_dataset_t
*origin_ds
;
2292 dsl_dataset_t
*origin_head
;
2294 dsl_dir_t
*odd
= NULL
;
2295 uint64_t oldnext_obj
;
2298 VERIFY0(promote_hold(ddpa
, dp
, FTAG
));
2299 hds
= ddpa
->ddpa_clone
;
2301 ASSERT0(dsl_dataset_phys(hds
)->ds_flags
& DS_FLAG_NOPROMOTE
);
2303 snap
= list_head(&ddpa
->shared_snaps
);
2304 origin_ds
= snap
->ds
;
2307 snap
= list_head(&ddpa
->origin_snaps
);
2308 origin_head
= snap
->ds
;
2311 * We need to explicitly open odd, since origin_ds's dd will be
2314 VERIFY0(dsl_dir_hold_obj(dp
, origin_ds
->ds_dir
->dd_object
,
2317 /* change origin's next snap */
2318 dmu_buf_will_dirty(origin_ds
->ds_dbuf
, tx
);
2319 oldnext_obj
= dsl_dataset_phys(origin_ds
)->ds_next_snap_obj
;
2320 snap
= list_tail(&ddpa
->clone_snaps
);
2321 ASSERT3U(dsl_dataset_phys(snap
->ds
)->ds_prev_snap_obj
, ==,
2322 origin_ds
->ds_object
);
2323 dsl_dataset_phys(origin_ds
)->ds_next_snap_obj
= snap
->ds
->ds_object
;
2325 /* change the origin's next clone */
2326 if (dsl_dataset_phys(origin_ds
)->ds_next_clones_obj
) {
2327 dsl_dataset_remove_from_next_clones(origin_ds
,
2328 snap
->ds
->ds_object
, tx
);
2329 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2330 dsl_dataset_phys(origin_ds
)->ds_next_clones_obj
,
2335 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
2336 ASSERT3U(dsl_dir_phys(dd
)->dd_origin_obj
, ==, origin_ds
->ds_object
);
2337 dsl_dir_phys(dd
)->dd_origin_obj
= dsl_dir_phys(odd
)->dd_origin_obj
;
2338 dd
->dd_origin_txg
= origin_head
->ds_dir
->dd_origin_txg
;
2339 dmu_buf_will_dirty(odd
->dd_dbuf
, tx
);
2340 dsl_dir_phys(odd
)->dd_origin_obj
= origin_ds
->ds_object
;
2341 origin_head
->ds_dir
->dd_origin_txg
=
2342 dsl_dataset_phys(origin_ds
)->ds_creation_txg
;
2344 /* change dd_clone entries */
2345 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
2346 VERIFY0(zap_remove_int(dp
->dp_meta_objset
,
2347 dsl_dir_phys(odd
)->dd_clones
, hds
->ds_object
, tx
));
2348 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2349 dsl_dir_phys(ddpa
->origin_origin
->ds_dir
)->dd_clones
,
2350 hds
->ds_object
, tx
));
2352 VERIFY0(zap_remove_int(dp
->dp_meta_objset
,
2353 dsl_dir_phys(ddpa
->origin_origin
->ds_dir
)->dd_clones
,
2354 origin_head
->ds_object
, tx
));
2355 if (dsl_dir_phys(dd
)->dd_clones
== 0) {
2356 dsl_dir_phys(dd
)->dd_clones
=
2357 zap_create(dp
->dp_meta_objset
, DMU_OT_DSL_CLONES
,
2358 DMU_OT_NONE
, 0, tx
);
2360 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2361 dsl_dir_phys(dd
)->dd_clones
, origin_head
->ds_object
, tx
));
2364 /* move snapshots to this dir */
2365 for (snap
= list_head(&ddpa
->shared_snaps
); snap
;
2366 snap
= list_next(&ddpa
->shared_snaps
, snap
)) {
2367 dsl_dataset_t
*ds
= snap
->ds
;
2370 * Property callbacks are registered to a particular
2371 * dsl_dir. Since ours is changing, evict the objset
2372 * so that they will be unregistered from the old dsl_dir.
2374 if (ds
->ds_objset
) {
2375 dmu_objset_evict(ds
->ds_objset
);
2376 ds
->ds_objset
= NULL
;
2379 /* move snap name entry */
2380 VERIFY0(dsl_dataset_get_snapname(ds
));
2381 VERIFY0(dsl_dataset_snap_remove(origin_head
,
2382 ds
->ds_snapname
, tx
, B_TRUE
));
2383 VERIFY0(zap_add(dp
->dp_meta_objset
,
2384 dsl_dataset_phys(hds
)->ds_snapnames_zapobj
, ds
->ds_snapname
,
2385 8, 1, &ds
->ds_object
, tx
));
2386 dsl_fs_ss_count_adjust(hds
->ds_dir
, 1,
2387 DD_FIELD_SNAPSHOT_COUNT
, tx
);
2389 /* change containing dsl_dir */
2390 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
2391 ASSERT3U(dsl_dataset_phys(ds
)->ds_dir_obj
, ==, odd
->dd_object
);
2392 dsl_dataset_phys(ds
)->ds_dir_obj
= dd
->dd_object
;
2393 ASSERT3P(ds
->ds_dir
, ==, odd
);
2394 dsl_dir_rele(ds
->ds_dir
, ds
);
2395 VERIFY0(dsl_dir_hold_obj(dp
, dd
->dd_object
,
2396 NULL
, ds
, &ds
->ds_dir
));
2398 /* move any clone references */
2399 if (dsl_dataset_phys(ds
)->ds_next_clones_obj
&&
2400 spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
2404 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
2405 dsl_dataset_phys(ds
)->ds_next_clones_obj
);
2406 zap_cursor_retrieve(&zc
, &za
) == 0;
2407 zap_cursor_advance(&zc
)) {
2408 dsl_dataset_t
*cnds
;
2411 if (za
.za_first_integer
== oldnext_obj
) {
2413 * We've already moved the
2414 * origin's reference.
2419 VERIFY0(dsl_dataset_hold_obj(dp
,
2420 za
.za_first_integer
, FTAG
, &cnds
));
2421 o
= dsl_dir_phys(cnds
->ds_dir
)->
2422 dd_head_dataset_obj
;
2424 VERIFY0(zap_remove_int(dp
->dp_meta_objset
,
2425 dsl_dir_phys(odd
)->dd_clones
, o
, tx
));
2426 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2427 dsl_dir_phys(dd
)->dd_clones
, o
, tx
));
2428 dsl_dataset_rele(cnds
, FTAG
);
2430 zap_cursor_fini(&zc
);
2433 ASSERT(!dsl_prop_hascb(ds
));
2437 * Change space accounting.
2438 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2439 * both be valid, or both be 0 (resulting in delta == 0). This
2440 * is true for each of {clone,origin} independently.
2443 delta
= ddpa
->cloneusedsnap
-
2444 dsl_dir_phys(dd
)->dd_used_breakdown
[DD_USED_SNAP
];
2445 ASSERT3S(delta
, >=, 0);
2446 ASSERT3U(ddpa
->used
, >=, delta
);
2447 dsl_dir_diduse_space(dd
, DD_USED_SNAP
, delta
, 0, 0, tx
);
2448 dsl_dir_diduse_space(dd
, DD_USED_HEAD
,
2449 ddpa
->used
- delta
, ddpa
->comp
, ddpa
->uncomp
, tx
);
2451 delta
= ddpa
->originusedsnap
-
2452 dsl_dir_phys(odd
)->dd_used_breakdown
[DD_USED_SNAP
];
2453 ASSERT3S(delta
, <=, 0);
2454 ASSERT3U(ddpa
->used
, >=, -delta
);
2455 dsl_dir_diduse_space(odd
, DD_USED_SNAP
, delta
, 0, 0, tx
);
2456 dsl_dir_diduse_space(odd
, DD_USED_HEAD
,
2457 -ddpa
->used
- delta
, -ddpa
->comp
, -ddpa
->uncomp
, tx
);
2459 dsl_dataset_phys(origin_ds
)->ds_unique_bytes
= ddpa
->unique
;
2461 /* log history record */
2462 spa_history_log_internal_ds(hds
, "promote", tx
, "");
2464 dsl_dir_rele(odd
, FTAG
);
2465 promote_rele(ddpa
, FTAG
);
2469 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2470 * (exclusive) and last_obj (inclusive). The list will be in reverse
2471 * order (last_obj will be the list_head()). If first_obj == 0, do all
2472 * snapshots back to this dataset's origin.
2475 snaplist_make(dsl_pool_t
*dp
,
2476 uint64_t first_obj
, uint64_t last_obj
, list_t
*l
, void *tag
)
2478 uint64_t obj
= last_obj
;
2480 list_create(l
, sizeof (struct promotenode
),
2481 offsetof(struct promotenode
, link
));
2483 while (obj
!= first_obj
) {
2485 struct promotenode
*snap
;
2488 err
= dsl_dataset_hold_obj(dp
, obj
, tag
, &ds
);
2489 ASSERT(err
!= ENOENT
);
2494 first_obj
= dsl_dir_phys(ds
->ds_dir
)->dd_origin_obj
;
2496 snap
= kmem_alloc(sizeof (*snap
), KM_SLEEP
);
2498 list_insert_tail(l
, snap
);
2499 obj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
2506 snaplist_space(list_t
*l
, uint64_t mintxg
, uint64_t *spacep
)
2508 struct promotenode
*snap
;
2511 for (snap
= list_head(l
); snap
; snap
= list_next(l
, snap
)) {
2512 uint64_t used
, comp
, uncomp
;
2513 dsl_deadlist_space_range(&snap
->ds
->ds_deadlist
,
2514 mintxg
, UINT64_MAX
, &used
, &comp
, &uncomp
);
2521 snaplist_destroy(list_t
*l
, void *tag
)
2523 struct promotenode
*snap
;
2525 if (l
== NULL
|| !list_link_active(&l
->list_head
))
2528 while ((snap
= list_tail(l
)) != NULL
) {
2529 list_remove(l
, snap
);
2530 dsl_dataset_rele(snap
->ds
, tag
);
2531 kmem_free(snap
, sizeof (*snap
));
2537 promote_hold(dsl_dataset_promote_arg_t
*ddpa
, dsl_pool_t
*dp
, void *tag
)
2541 struct promotenode
*snap
;
2543 error
= dsl_dataset_hold(dp
, ddpa
->ddpa_clonename
, tag
,
2547 dd
= ddpa
->ddpa_clone
->ds_dir
;
2549 if (ddpa
->ddpa_clone
->ds_is_snapshot
||
2550 !dsl_dir_is_clone(dd
)) {
2551 dsl_dataset_rele(ddpa
->ddpa_clone
, tag
);
2552 return (SET_ERROR(EINVAL
));
2555 error
= snaplist_make(dp
, 0, dsl_dir_phys(dd
)->dd_origin_obj
,
2556 &ddpa
->shared_snaps
, tag
);
2560 error
= snaplist_make(dp
, 0, ddpa
->ddpa_clone
->ds_object
,
2561 &ddpa
->clone_snaps
, tag
);
2565 snap
= list_head(&ddpa
->shared_snaps
);
2566 ASSERT3U(snap
->ds
->ds_object
, ==, dsl_dir_phys(dd
)->dd_origin_obj
);
2567 error
= snaplist_make(dp
, dsl_dir_phys(dd
)->dd_origin_obj
,
2568 dsl_dir_phys(snap
->ds
->ds_dir
)->dd_head_dataset_obj
,
2569 &ddpa
->origin_snaps
, tag
);
2573 if (dsl_dir_phys(snap
->ds
->ds_dir
)->dd_origin_obj
!= 0) {
2574 error
= dsl_dataset_hold_obj(dp
,
2575 dsl_dir_phys(snap
->ds
->ds_dir
)->dd_origin_obj
,
2576 tag
, &ddpa
->origin_origin
);
2582 promote_rele(ddpa
, tag
);
2587 promote_rele(dsl_dataset_promote_arg_t
*ddpa
, void *tag
)
2589 snaplist_destroy(&ddpa
->shared_snaps
, tag
);
2590 snaplist_destroy(&ddpa
->clone_snaps
, tag
);
2591 snaplist_destroy(&ddpa
->origin_snaps
, tag
);
2592 if (ddpa
->origin_origin
!= NULL
)
2593 dsl_dataset_rele(ddpa
->origin_origin
, tag
);
2594 dsl_dataset_rele(ddpa
->ddpa_clone
, tag
);
2600 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2601 * in with the name. (It must be at least MAXNAMELEN bytes long.)
2604 dsl_dataset_promote(const char *name
, char *conflsnap
)
2606 dsl_dataset_promote_arg_t ddpa
= { 0 };
2612 * We will modify space proportional to the number of
2613 * snapshots. Compute numsnaps.
2615 error
= dmu_objset_hold(name
, FTAG
, &os
);
2618 error
= zap_count(dmu_objset_pool(os
)->dp_meta_objset
,
2619 dsl_dataset_phys(dmu_objset_ds(os
))->ds_snapnames_zapobj
,
2621 dmu_objset_rele(os
, FTAG
);
2625 ddpa
.ddpa_clonename
= name
;
2626 ddpa
.err_ds
= conflsnap
;
2629 return (dsl_sync_task(name
, dsl_dataset_promote_check
,
2630 dsl_dataset_promote_sync
, &ddpa
,
2631 2 + numsnaps
, ZFS_SPACE_CHECK_RESERVED
));
2635 dsl_dataset_clone_swap_check_impl(dsl_dataset_t
*clone
,
2636 dsl_dataset_t
*origin_head
, boolean_t force
, void *owner
, dmu_tx_t
*tx
)
2638 int64_t unused_refres_delta
;
2640 /* they should both be heads */
2641 if (clone
->ds_is_snapshot
||
2642 origin_head
->ds_is_snapshot
)
2643 return (SET_ERROR(EINVAL
));
2645 /* if we are not forcing, the branch point should be just before them */
2646 if (!force
&& clone
->ds_prev
!= origin_head
->ds_prev
)
2647 return (SET_ERROR(EINVAL
));
2649 /* clone should be the clone (unless they are unrelated) */
2650 if (clone
->ds_prev
!= NULL
&&
2651 clone
->ds_prev
!= clone
->ds_dir
->dd_pool
->dp_origin_snap
&&
2652 origin_head
->ds_dir
!= clone
->ds_prev
->ds_dir
)
2653 return (SET_ERROR(EINVAL
));
2655 /* the clone should be a child of the origin */
2656 if (clone
->ds_dir
->dd_parent
!= origin_head
->ds_dir
)
2657 return (SET_ERROR(EINVAL
));
2659 /* origin_head shouldn't be modified unless 'force' */
2661 dsl_dataset_modified_since_snap(origin_head
, origin_head
->ds_prev
))
2662 return (SET_ERROR(ETXTBSY
));
2664 /* origin_head should have no long holds (e.g. is not mounted) */
2665 if (dsl_dataset_handoff_check(origin_head
, owner
, tx
))
2666 return (SET_ERROR(EBUSY
));
2668 /* check amount of any unconsumed refreservation */
2669 unused_refres_delta
=
2670 (int64_t)MIN(origin_head
->ds_reserved
,
2671 dsl_dataset_phys(origin_head
)->ds_unique_bytes
) -
2672 (int64_t)MIN(origin_head
->ds_reserved
,
2673 dsl_dataset_phys(clone
)->ds_unique_bytes
);
2675 if (unused_refres_delta
> 0 &&
2676 unused_refres_delta
>
2677 dsl_dir_space_available(origin_head
->ds_dir
, NULL
, 0, TRUE
))
2678 return (SET_ERROR(ENOSPC
));
2680 /* clone can't be over the head's refquota */
2681 if (origin_head
->ds_quota
!= 0 &&
2682 dsl_dataset_phys(clone
)->ds_referenced_bytes
>
2683 origin_head
->ds_quota
)
2684 return (SET_ERROR(EDQUOT
));
2690 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t
*clone
,
2691 dsl_dataset_t
*origin_head
, dmu_tx_t
*tx
)
2693 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2694 int64_t unused_refres_delta
;
2696 ASSERT(clone
->ds_reserved
== 0);
2697 ASSERT(origin_head
->ds_quota
== 0 ||
2698 dsl_dataset_phys(clone
)->ds_unique_bytes
<= origin_head
->ds_quota
);
2699 ASSERT3P(clone
->ds_prev
, ==, origin_head
->ds_prev
);
2701 dmu_buf_will_dirty(clone
->ds_dbuf
, tx
);
2702 dmu_buf_will_dirty(origin_head
->ds_dbuf
, tx
);
2704 if (clone
->ds_objset
!= NULL
) {
2705 dmu_objset_evict(clone
->ds_objset
);
2706 clone
->ds_objset
= NULL
;
2709 if (origin_head
->ds_objset
!= NULL
) {
2710 dmu_objset_evict(origin_head
->ds_objset
);
2711 origin_head
->ds_objset
= NULL
;
2714 unused_refres_delta
=
2715 (int64_t)MIN(origin_head
->ds_reserved
,
2716 dsl_dataset_phys(origin_head
)->ds_unique_bytes
) -
2717 (int64_t)MIN(origin_head
->ds_reserved
,
2718 dsl_dataset_phys(clone
)->ds_unique_bytes
);
2721 * Reset origin's unique bytes, if it exists.
2723 if (clone
->ds_prev
) {
2724 dsl_dataset_t
*origin
= clone
->ds_prev
;
2725 uint64_t comp
, uncomp
;
2727 dmu_buf_will_dirty(origin
->ds_dbuf
, tx
);
2728 dsl_deadlist_space_range(&clone
->ds_deadlist
,
2729 dsl_dataset_phys(origin
)->ds_prev_snap_txg
, UINT64_MAX
,
2730 &dsl_dataset_phys(origin
)->ds_unique_bytes
, &comp
, &uncomp
);
2736 tmp
= dsl_dataset_phys(origin_head
)->ds_bp
;
2737 dsl_dataset_phys(origin_head
)->ds_bp
=
2738 dsl_dataset_phys(clone
)->ds_bp
;
2739 dsl_dataset_phys(clone
)->ds_bp
= tmp
;
2742 /* set dd_*_bytes */
2744 int64_t dused
, dcomp
, duncomp
;
2745 uint64_t cdl_used
, cdl_comp
, cdl_uncomp
;
2746 uint64_t odl_used
, odl_comp
, odl_uncomp
;
2748 ASSERT3U(dsl_dir_phys(clone
->ds_dir
)->
2749 dd_used_breakdown
[DD_USED_SNAP
], ==, 0);
2751 dsl_deadlist_space(&clone
->ds_deadlist
,
2752 &cdl_used
, &cdl_comp
, &cdl_uncomp
);
2753 dsl_deadlist_space(&origin_head
->ds_deadlist
,
2754 &odl_used
, &odl_comp
, &odl_uncomp
);
2756 dused
= dsl_dataset_phys(clone
)->ds_referenced_bytes
+
2758 (dsl_dataset_phys(origin_head
)->ds_referenced_bytes
+
2760 dcomp
= dsl_dataset_phys(clone
)->ds_compressed_bytes
+
2762 (dsl_dataset_phys(origin_head
)->ds_compressed_bytes
+
2764 duncomp
= dsl_dataset_phys(clone
)->ds_uncompressed_bytes
+
2766 (dsl_dataset_phys(origin_head
)->ds_uncompressed_bytes
+
2769 dsl_dir_diduse_space(origin_head
->ds_dir
, DD_USED_HEAD
,
2770 dused
, dcomp
, duncomp
, tx
);
2771 dsl_dir_diduse_space(clone
->ds_dir
, DD_USED_HEAD
,
2772 -dused
, -dcomp
, -duncomp
, tx
);
2775 * The difference in the space used by snapshots is the
2776 * difference in snapshot space due to the head's
2777 * deadlist (since that's the only thing that's
2778 * changing that affects the snapused).
2780 dsl_deadlist_space_range(&clone
->ds_deadlist
,
2781 origin_head
->ds_dir
->dd_origin_txg
, UINT64_MAX
,
2782 &cdl_used
, &cdl_comp
, &cdl_uncomp
);
2783 dsl_deadlist_space_range(&origin_head
->ds_deadlist
,
2784 origin_head
->ds_dir
->dd_origin_txg
, UINT64_MAX
,
2785 &odl_used
, &odl_comp
, &odl_uncomp
);
2786 dsl_dir_transfer_space(origin_head
->ds_dir
, cdl_used
- odl_used
,
2787 DD_USED_HEAD
, DD_USED_SNAP
, tx
);
2790 /* swap ds_*_bytes */
2791 SWITCH64(dsl_dataset_phys(origin_head
)->ds_referenced_bytes
,
2792 dsl_dataset_phys(clone
)->ds_referenced_bytes
);
2793 SWITCH64(dsl_dataset_phys(origin_head
)->ds_compressed_bytes
,
2794 dsl_dataset_phys(clone
)->ds_compressed_bytes
);
2795 SWITCH64(dsl_dataset_phys(origin_head
)->ds_uncompressed_bytes
,
2796 dsl_dataset_phys(clone
)->ds_uncompressed_bytes
);
2797 SWITCH64(dsl_dataset_phys(origin_head
)->ds_unique_bytes
,
2798 dsl_dataset_phys(clone
)->ds_unique_bytes
);
2800 /* apply any parent delta for change in unconsumed refreservation */
2801 dsl_dir_diduse_space(origin_head
->ds_dir
, DD_USED_REFRSRV
,
2802 unused_refres_delta
, 0, 0, tx
);
2807 dsl_deadlist_close(&clone
->ds_deadlist
);
2808 dsl_deadlist_close(&origin_head
->ds_deadlist
);
2809 SWITCH64(dsl_dataset_phys(origin_head
)->ds_deadlist_obj
,
2810 dsl_dataset_phys(clone
)->ds_deadlist_obj
);
2811 dsl_deadlist_open(&clone
->ds_deadlist
, dp
->dp_meta_objset
,
2812 dsl_dataset_phys(clone
)->ds_deadlist_obj
);
2813 dsl_deadlist_open(&origin_head
->ds_deadlist
, dp
->dp_meta_objset
,
2814 dsl_dataset_phys(origin_head
)->ds_deadlist_obj
);
2816 dsl_scan_ds_clone_swapped(origin_head
, clone
, tx
);
2818 spa_history_log_internal_ds(clone
, "clone swap", tx
,
2819 "parent=%s", origin_head
->ds_dir
->dd_myname
);
2823 * Given a pool name and a dataset object number in that pool,
2824 * return the name of that dataset.
2827 dsl_dsobj_to_dsname(char *pname
, uint64_t obj
, char *buf
)
2833 error
= dsl_pool_hold(pname
, FTAG
, &dp
);
2837 error
= dsl_dataset_hold_obj(dp
, obj
, FTAG
, &ds
);
2839 dsl_dataset_name(ds
, buf
);
2840 dsl_dataset_rele(ds
, FTAG
);
2842 dsl_pool_rele(dp
, FTAG
);
2848 dsl_dataset_check_quota(dsl_dataset_t
*ds
, boolean_t check_quota
,
2849 uint64_t asize
, uint64_t inflight
, uint64_t *used
, uint64_t *ref_rsrv
)
2853 ASSERT3S(asize
, >, 0);
2856 * *ref_rsrv is the portion of asize that will come from any
2857 * unconsumed refreservation space.
2861 mutex_enter(&ds
->ds_lock
);
2863 * Make a space adjustment for reserved bytes.
2865 if (ds
->ds_reserved
> dsl_dataset_phys(ds
)->ds_unique_bytes
) {
2867 ds
->ds_reserved
- dsl_dataset_phys(ds
)->ds_unique_bytes
);
2869 (ds
->ds_reserved
- dsl_dataset_phys(ds
)->ds_unique_bytes
);
2871 asize
- MIN(asize
, parent_delta(ds
, asize
+ inflight
));
2874 if (!check_quota
|| ds
->ds_quota
== 0) {
2875 mutex_exit(&ds
->ds_lock
);
2879 * If they are requesting more space, and our current estimate
2880 * is over quota, they get to try again unless the actual
2881 * on-disk is over quota and there are no pending changes (which
2882 * may free up space for us).
2884 if (dsl_dataset_phys(ds
)->ds_referenced_bytes
+ inflight
>=
2887 dsl_dataset_phys(ds
)->ds_referenced_bytes
< ds
->ds_quota
)
2888 error
= SET_ERROR(ERESTART
);
2890 error
= SET_ERROR(EDQUOT
);
2892 mutex_exit(&ds
->ds_lock
);
2897 typedef struct dsl_dataset_set_qr_arg
{
2898 const char *ddsqra_name
;
2899 zprop_source_t ddsqra_source
;
2900 uint64_t ddsqra_value
;
2901 } dsl_dataset_set_qr_arg_t
;
2906 dsl_dataset_set_refquota_check(void *arg
, dmu_tx_t
*tx
)
2908 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
2909 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2914 if (spa_version(dp
->dp_spa
) < SPA_VERSION_REFQUOTA
)
2915 return (SET_ERROR(ENOTSUP
));
2917 error
= dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
);
2921 if (ds
->ds_is_snapshot
) {
2922 dsl_dataset_rele(ds
, FTAG
);
2923 return (SET_ERROR(EINVAL
));
2926 error
= dsl_prop_predict(ds
->ds_dir
,
2927 zfs_prop_to_name(ZFS_PROP_REFQUOTA
),
2928 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, &newval
);
2930 dsl_dataset_rele(ds
, FTAG
);
2935 dsl_dataset_rele(ds
, FTAG
);
2939 if (newval
< dsl_dataset_phys(ds
)->ds_referenced_bytes
||
2940 newval
< ds
->ds_reserved
) {
2941 dsl_dataset_rele(ds
, FTAG
);
2942 return (SET_ERROR(ENOSPC
));
2945 dsl_dataset_rele(ds
, FTAG
);
2950 dsl_dataset_set_refquota_sync(void *arg
, dmu_tx_t
*tx
)
2952 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
2953 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2957 VERIFY0(dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
));
2959 dsl_prop_set_sync_impl(ds
,
2960 zfs_prop_to_name(ZFS_PROP_REFQUOTA
),
2961 ddsqra
->ddsqra_source
, sizeof (ddsqra
->ddsqra_value
), 1,
2962 &ddsqra
->ddsqra_value
, tx
);
2964 VERIFY0(dsl_prop_get_int_ds(ds
,
2965 zfs_prop_to_name(ZFS_PROP_REFQUOTA
), &newval
));
2967 if (ds
->ds_quota
!= newval
) {
2968 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
2969 ds
->ds_quota
= newval
;
2971 dsl_dataset_rele(ds
, FTAG
);
2975 dsl_dataset_set_refquota(const char *dsname
, zprop_source_t source
,
2978 dsl_dataset_set_qr_arg_t ddsqra
;
2980 ddsqra
.ddsqra_name
= dsname
;
2981 ddsqra
.ddsqra_source
= source
;
2982 ddsqra
.ddsqra_value
= refquota
;
2984 return (dsl_sync_task(dsname
, dsl_dataset_set_refquota_check
,
2985 dsl_dataset_set_refquota_sync
, &ddsqra
, 0, ZFS_SPACE_CHECK_NONE
));
2989 dsl_dataset_set_refreservation_check(void *arg
, dmu_tx_t
*tx
)
2991 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
2992 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2995 uint64_t newval
, unique
;
2997 if (spa_version(dp
->dp_spa
) < SPA_VERSION_REFRESERVATION
)
2998 return (SET_ERROR(ENOTSUP
));
3000 error
= dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
);
3004 if (ds
->ds_is_snapshot
) {
3005 dsl_dataset_rele(ds
, FTAG
);
3006 return (SET_ERROR(EINVAL
));
3009 error
= dsl_prop_predict(ds
->ds_dir
,
3010 zfs_prop_to_name(ZFS_PROP_REFRESERVATION
),
3011 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, &newval
);
3013 dsl_dataset_rele(ds
, FTAG
);
3018 * If we are doing the preliminary check in open context, the
3019 * space estimates may be inaccurate.
3021 if (!dmu_tx_is_syncing(tx
)) {
3022 dsl_dataset_rele(ds
, FTAG
);
3026 mutex_enter(&ds
->ds_lock
);
3027 if (!DS_UNIQUE_IS_ACCURATE(ds
))
3028 dsl_dataset_recalc_head_uniq(ds
);
3029 unique
= dsl_dataset_phys(ds
)->ds_unique_bytes
;
3030 mutex_exit(&ds
->ds_lock
);
3032 if (MAX(unique
, newval
) > MAX(unique
, ds
->ds_reserved
)) {
3033 uint64_t delta
= MAX(unique
, newval
) -
3034 MAX(unique
, ds
->ds_reserved
);
3037 dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, B_TRUE
) ||
3038 (ds
->ds_quota
> 0 && newval
> ds
->ds_quota
)) {
3039 dsl_dataset_rele(ds
, FTAG
);
3040 return (SET_ERROR(ENOSPC
));
3044 dsl_dataset_rele(ds
, FTAG
);
3049 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t
*ds
,
3050 zprop_source_t source
, uint64_t value
, dmu_tx_t
*tx
)
3056 dsl_prop_set_sync_impl(ds
, zfs_prop_to_name(ZFS_PROP_REFRESERVATION
),
3057 source
, sizeof (value
), 1, &value
, tx
);
3059 VERIFY0(dsl_prop_get_int_ds(ds
,
3060 zfs_prop_to_name(ZFS_PROP_REFRESERVATION
), &newval
));
3062 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
3063 mutex_enter(&ds
->ds_dir
->dd_lock
);
3064 mutex_enter(&ds
->ds_lock
);
3065 ASSERT(DS_UNIQUE_IS_ACCURATE(ds
));
3066 unique
= dsl_dataset_phys(ds
)->ds_unique_bytes
;
3067 delta
= MAX(0, (int64_t)(newval
- unique
)) -
3068 MAX(0, (int64_t)(ds
->ds_reserved
- unique
));
3069 ds
->ds_reserved
= newval
;
3070 mutex_exit(&ds
->ds_lock
);
3072 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_REFRSRV
, delta
, 0, 0, tx
);
3073 mutex_exit(&ds
->ds_dir
->dd_lock
);
3077 dsl_dataset_set_refreservation_sync(void *arg
, dmu_tx_t
*tx
)
3079 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
3080 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3083 VERIFY0(dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
));
3084 dsl_dataset_set_refreservation_sync_impl(ds
,
3085 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, tx
);
3086 dsl_dataset_rele(ds
, FTAG
);
3090 dsl_dataset_set_refreservation(const char *dsname
, zprop_source_t source
,
3091 uint64_t refreservation
)
3093 dsl_dataset_set_qr_arg_t ddsqra
;
3095 ddsqra
.ddsqra_name
= dsname
;
3096 ddsqra
.ddsqra_source
= source
;
3097 ddsqra
.ddsqra_value
= refreservation
;
3099 return (dsl_sync_task(dsname
, dsl_dataset_set_refreservation_check
,
3100 dsl_dataset_set_refreservation_sync
, &ddsqra
,
3101 0, ZFS_SPACE_CHECK_NONE
));
3105 * Return (in *usedp) the amount of space written in new that is not
3106 * present in oldsnap. New may be a snapshot or the head. Old must be
3107 * a snapshot before new, in new's filesystem (or its origin). If not then
3108 * fail and return EINVAL.
3110 * The written space is calculated by considering two components: First, we
3111 * ignore any freed space, and calculate the written as new's used space
3112 * minus old's used space. Next, we add in the amount of space that was freed
3113 * between the two snapshots, thus reducing new's used space relative to old's.
3114 * Specifically, this is the space that was born before old->ds_creation_txg,
3115 * and freed before new (ie. on new's deadlist or a previous deadlist).
3117 * space freed [---------------------]
3118 * snapshots ---O-------O--------O-------O------
3122 dsl_dataset_space_written(dsl_dataset_t
*oldsnap
, dsl_dataset_t
*new,
3123 uint64_t *usedp
, uint64_t *compp
, uint64_t *uncompp
)
3127 dsl_pool_t
*dp
= new->ds_dir
->dd_pool
;
3129 ASSERT(dsl_pool_config_held(dp
));
3132 *usedp
+= dsl_dataset_phys(new)->ds_referenced_bytes
;
3133 *usedp
-= dsl_dataset_phys(oldsnap
)->ds_referenced_bytes
;
3136 *compp
+= dsl_dataset_phys(new)->ds_compressed_bytes
;
3137 *compp
-= dsl_dataset_phys(oldsnap
)->ds_compressed_bytes
;
3140 *uncompp
+= dsl_dataset_phys(new)->ds_uncompressed_bytes
;
3141 *uncompp
-= dsl_dataset_phys(oldsnap
)->ds_uncompressed_bytes
;
3143 snapobj
= new->ds_object
;
3144 while (snapobj
!= oldsnap
->ds_object
) {
3145 dsl_dataset_t
*snap
;
3146 uint64_t used
, comp
, uncomp
;
3148 if (snapobj
== new->ds_object
) {
3151 err
= dsl_dataset_hold_obj(dp
, snapobj
, FTAG
, &snap
);
3156 if (dsl_dataset_phys(snap
)->ds_prev_snap_txg
==
3157 dsl_dataset_phys(oldsnap
)->ds_creation_txg
) {
3159 * The blocks in the deadlist can not be born after
3160 * ds_prev_snap_txg, so get the whole deadlist space,
3161 * which is more efficient (especially for old-format
3162 * deadlists). Unfortunately the deadlist code
3163 * doesn't have enough information to make this
3164 * optimization itself.
3166 dsl_deadlist_space(&snap
->ds_deadlist
,
3167 &used
, &comp
, &uncomp
);
3169 dsl_deadlist_space_range(&snap
->ds_deadlist
,
3170 0, dsl_dataset_phys(oldsnap
)->ds_creation_txg
,
3171 &used
, &comp
, &uncomp
);
3178 * If we get to the beginning of the chain of snapshots
3179 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3180 * was not a snapshot of/before new.
3182 snapobj
= dsl_dataset_phys(snap
)->ds_prev_snap_obj
;
3184 dsl_dataset_rele(snap
, FTAG
);
3186 err
= SET_ERROR(EINVAL
);
3195 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3196 * lastsnap, and all snapshots in between are deleted.
3198 * blocks that would be freed [---------------------------]
3199 * snapshots ---O-------O--------O-------O--------O
3200 * firstsnap lastsnap
3202 * This is the set of blocks that were born after the snap before firstsnap,
3203 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3204 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3205 * We calculate this by iterating over the relevant deadlists (from the snap
3206 * after lastsnap, backward to the snap after firstsnap), summing up the
3207 * space on the deadlist that was born after the snap before firstsnap.
3210 dsl_dataset_space_wouldfree(dsl_dataset_t
*firstsnap
,
3211 dsl_dataset_t
*lastsnap
,
3212 uint64_t *usedp
, uint64_t *compp
, uint64_t *uncompp
)
3216 dsl_pool_t
*dp
= firstsnap
->ds_dir
->dd_pool
;
3218 ASSERT(firstsnap
->ds_is_snapshot
);
3219 ASSERT(lastsnap
->ds_is_snapshot
);
3222 * Check that the snapshots are in the same dsl_dir, and firstsnap
3223 * is before lastsnap.
3225 if (firstsnap
->ds_dir
!= lastsnap
->ds_dir
||
3226 dsl_dataset_phys(firstsnap
)->ds_creation_txg
>
3227 dsl_dataset_phys(lastsnap
)->ds_creation_txg
)
3228 return (SET_ERROR(EINVAL
));
3230 *usedp
= *compp
= *uncompp
= 0;
3232 snapobj
= dsl_dataset_phys(lastsnap
)->ds_next_snap_obj
;
3233 while (snapobj
!= firstsnap
->ds_object
) {
3235 uint64_t used
, comp
, uncomp
;
3237 err
= dsl_dataset_hold_obj(dp
, snapobj
, FTAG
, &ds
);
3241 dsl_deadlist_space_range(&ds
->ds_deadlist
,
3242 dsl_dataset_phys(firstsnap
)->ds_prev_snap_txg
, UINT64_MAX
,
3243 &used
, &comp
, &uncomp
);
3248 snapobj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
3249 ASSERT3U(snapobj
, !=, 0);
3250 dsl_dataset_rele(ds
, FTAG
);
3256 dsl_dataset_activate_large_blocks_check(void *arg
, dmu_tx_t
*tx
)
3258 const char *dsname
= arg
;
3260 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3263 if (!spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_LARGE_BLOCKS
))
3264 return (SET_ERROR(ENOTSUP
));
3266 ASSERT(spa_feature_is_enabled(dp
->dp_spa
,
3267 SPA_FEATURE_EXTENSIBLE_DATASET
));
3269 error
= dsl_dataset_hold(dp
, dsname
, FTAG
, &ds
);
3273 if (ds
->ds_large_blocks
)
3275 dsl_dataset_rele(ds
, FTAG
);
3281 dsl_dataset_activate_large_blocks_sync_impl(uint64_t dsobj
, dmu_tx_t
*tx
)
3283 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
3284 objset_t
*mos
= dmu_tx_pool(tx
)->dp_meta_objset
;
3287 spa_feature_incr(spa
, SPA_FEATURE_LARGE_BLOCKS
, tx
);
3288 dmu_object_zapify(mos
, dsobj
, DMU_OT_DSL_DATASET
, tx
);
3290 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_LARGE_BLOCKS
,
3291 sizeof (zero
), 1, &zero
, tx
));
3295 dsl_dataset_activate_large_blocks_sync(void *arg
, dmu_tx_t
*tx
)
3297 const char *dsname
= arg
;
3300 VERIFY0(dsl_dataset_hold(dmu_tx_pool(tx
), dsname
, FTAG
, &ds
));
3302 dsl_dataset_activate_large_blocks_sync_impl(ds
->ds_object
, tx
);
3303 ASSERT(!ds
->ds_large_blocks
);
3304 ds
->ds_large_blocks
= B_TRUE
;
3305 dsl_dataset_rele(ds
, FTAG
);
3309 dsl_dataset_activate_large_blocks(const char *dsname
)
3313 error
= dsl_sync_task(dsname
,
3314 dsl_dataset_activate_large_blocks_check
,
3315 dsl_dataset_activate_large_blocks_sync
, (void *)dsname
,
3316 1, ZFS_SPACE_CHECK_RESERVED
);
3319 * EALREADY indicates that this dataset already supports large blocks.
3321 if (error
== EALREADY
)
3327 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3328 * For example, they could both be snapshots of the same filesystem, and
3329 * 'earlier' is before 'later'. Or 'earlier' could be the origin of
3330 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's
3331 * filesystem. Or 'earlier' could be the origin's origin.
3333 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3336 dsl_dataset_is_before(dsl_dataset_t
*later
, dsl_dataset_t
*earlier
,
3337 uint64_t earlier_txg
)
3339 dsl_pool_t
*dp
= later
->ds_dir
->dd_pool
;
3342 dsl_dataset_t
*origin
;
3344 ASSERT(dsl_pool_config_held(dp
));
3345 ASSERT(earlier
->ds_is_snapshot
|| earlier_txg
!= 0);
3347 if (earlier_txg
== 0)
3348 earlier_txg
= dsl_dataset_phys(earlier
)->ds_creation_txg
;
3350 if (later
->ds_is_snapshot
&&
3351 earlier_txg
>= dsl_dataset_phys(later
)->ds_creation_txg
)
3354 if (later
->ds_dir
== earlier
->ds_dir
)
3356 if (!dsl_dir_is_clone(later
->ds_dir
))
3359 if (dsl_dir_phys(later
->ds_dir
)->dd_origin_obj
== earlier
->ds_object
)
3361 error
= dsl_dataset_hold_obj(dp
,
3362 dsl_dir_phys(later
->ds_dir
)->dd_origin_obj
, FTAG
, &origin
);
3365 ret
= dsl_dataset_is_before(origin
, earlier
, earlier_txg
);
3366 dsl_dataset_rele(origin
, FTAG
);
3372 dsl_dataset_zapify(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
3374 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
3375 dmu_object_zapify(mos
, ds
->ds_object
, DMU_OT_DSL_DATASET
, tx
);
3378 #if defined(_KERNEL) && defined(HAVE_SPL)
3380 module_param(zfs_max_recordsize
, int, 0644);
3381 MODULE_PARM_DESC(zfs_max_recordsize
, "Max allowed record size");
3383 /* Limited to 1M on 32-bit platforms due to lack of virtual address space */
3384 module_param(zfs_max_recordsize
, int, 0444);
3385 MODULE_PARM_DESC(zfs_max_recordsize
, "Max allowed record size");
3388 EXPORT_SYMBOL(dsl_dataset_hold
);
3389 EXPORT_SYMBOL(dsl_dataset_hold_obj
);
3390 EXPORT_SYMBOL(dsl_dataset_own
);
3391 EXPORT_SYMBOL(dsl_dataset_own_obj
);
3392 EXPORT_SYMBOL(dsl_dataset_name
);
3393 EXPORT_SYMBOL(dsl_dataset_rele
);
3394 EXPORT_SYMBOL(dsl_dataset_disown
);
3395 EXPORT_SYMBOL(dsl_dataset_tryown
);
3396 EXPORT_SYMBOL(dsl_dataset_create_sync
);
3397 EXPORT_SYMBOL(dsl_dataset_create_sync_dd
);
3398 EXPORT_SYMBOL(dsl_dataset_snapshot_check
);
3399 EXPORT_SYMBOL(dsl_dataset_snapshot_sync
);
3400 EXPORT_SYMBOL(dsl_dataset_promote
);
3401 EXPORT_SYMBOL(dsl_dataset_user_hold
);
3402 EXPORT_SYMBOL(dsl_dataset_user_release
);
3403 EXPORT_SYMBOL(dsl_dataset_get_holds
);
3404 EXPORT_SYMBOL(dsl_dataset_get_blkptr
);
3405 EXPORT_SYMBOL(dsl_dataset_set_blkptr
);
3406 EXPORT_SYMBOL(dsl_dataset_get_spa
);
3407 EXPORT_SYMBOL(dsl_dataset_modified_since_snap
);
3408 EXPORT_SYMBOL(dsl_dataset_space_written
);
3409 EXPORT_SYMBOL(dsl_dataset_space_wouldfree
);
3410 EXPORT_SYMBOL(dsl_dataset_sync
);
3411 EXPORT_SYMBOL(dsl_dataset_block_born
);
3412 EXPORT_SYMBOL(dsl_dataset_block_kill
);
3413 EXPORT_SYMBOL(dsl_dataset_block_freeable
);
3414 EXPORT_SYMBOL(dsl_dataset_prev_snap_txg
);
3415 EXPORT_SYMBOL(dsl_dataset_dirty
);
3416 EXPORT_SYMBOL(dsl_dataset_stats
);
3417 EXPORT_SYMBOL(dsl_dataset_fast_stat
);
3418 EXPORT_SYMBOL(dsl_dataset_space
);
3419 EXPORT_SYMBOL(dsl_dataset_fsid_guid
);
3420 EXPORT_SYMBOL(dsl_dsobj_to_dsname
);
3421 EXPORT_SYMBOL(dsl_dataset_check_quota
);
3422 EXPORT_SYMBOL(dsl_dataset_clone_swap_check_impl
);
3423 EXPORT_SYMBOL(dsl_dataset_clone_swap_sync_impl
);